From patchwork Thu Oct 15 02:06:00 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269812 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 95491C433DF for ; Thu, 15 Oct 2020 02:06:54 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 42D6922257 for ; Thu, 15 Oct 2020 02:06:54 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1727833AbgJOCGx (ORCPT ); Wed, 14 Oct 2020 22:06:53 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40051 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726493AbgJOCGx (ORCPT ); Wed, 14 Oct 2020 22:06:53 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id AD35D204259; Thu, 15 Oct 2020 04:06:49 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id rFTrSCH0ljre; Thu, 15 Oct 2020 04:06:46 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 86E6120423F; Thu, 15 Oct 2020 04:06:45 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 01/44] sg: move functions around Date: Wed, 14 Oct 2020 22:06:00 -0400 Message-Id: <20201015020643.432908-2-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Move main entry point functions around so submission code comes before completion code. Prior to this, the driver used the traditional open(), close(), read(), write(), ioctl() ordering however in this case that places completion code (i.e. sg_read()) before submission code (i.e. sg_write()). The main driver entry points are considered to be those named in struct file_operations sg_fops' definition. Helper functions are placed above their caller to reduce the number of forward function declarations needed. *** Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 496 ++++++++++++++++++++++++---------------------- 1 file changed, 261 insertions(+), 235 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 20472aaaf630..1e020298e939 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -8,11 +8,12 @@ * Original driver (sg.c): * Copyright (C) 1992 Lawrence Foard * Version 2 and 3 extensions to driver: - * Copyright (C) 1998 - 2014 Douglas Gilbert + * Copyright (C) 1998 - 2019 Douglas Gilbert */ -static int sg_version_num = 30536; /* 2 digits for each component */ -#define SG_VERSION_STR "3.5.36" +static int sg_version_num = 30901; /* [x]xyyzz where [x] empty when x=0 */ +#define SG_VERSION_STR "3.9.01" /* [x]x.[y]y.zz */ +static char *sg_version_date = "20190606"; /* * D. P. Gilbert (dgilbert@interlog.com), notes: @@ -47,6 +48,7 @@ static int sg_version_num = 30536; /* 2 digits for each component */ #include #include #include /* for sg_check_file_access() */ +#include #include "scsi.h" #include @@ -57,12 +59,6 @@ static int sg_version_num = 30536; /* 2 digits for each component */ #include "scsi_logging.h" -#ifdef CONFIG_SCSI_PROC_FS -#include -static char *sg_version_date = "20140603"; - -static int sg_proc_init(void); -#endif #define SG_ALLOW_DIO_DEF 0 @@ -173,11 +169,11 @@ typedef struct sg_device { /* holds the state of each scsi generic device */ /* tasklet or soft irq callback */ static void sg_rq_end_io(struct request *rq, blk_status_t status); +/* Declarations of other static functions used before they are defined */ +static int sg_proc_init(void); static int sg_start_req(Sg_request *srp, unsigned char *cmd); static int sg_finish_rem_req(Sg_request * srp); static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size); -static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, - Sg_request * srp); static ssize_t sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf, size_t count, int blocking, int read_only, int sg_io_owned, Sg_request **o_srp); @@ -190,7 +186,6 @@ static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size); static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp); static Sg_fd *sg_add_sfp(Sg_device * sdp); static void sg_remove_sfp(struct kref *); -static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id); static Sg_request *sg_add_request(Sg_fd * sfp); static int sg_remove_request(Sg_fd * sfp, Sg_request * srp); static Sg_device *sg_get_dev(int dev); @@ -232,16 +227,6 @@ static int sg_check_file_access(struct file *filp, const char *caller) return 0; } -static int sg_allow_access(struct file *filp, unsigned char *cmd) -{ - struct sg_fd *sfp = filp->private_data; - - if (sfp->parentdp->device->type == TYPE_SCANNER) - return 0; - - return blk_verify_command(cmd, filp->f_mode); -} - static int open_wait(Sg_device *sdp, int flags) { @@ -405,196 +390,6 @@ sg_release(struct inode *inode, struct file *filp) return 0; } -static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count) -{ - struct sg_header __user *old_hdr = buf; - int reply_len; - - if (count >= SZ_SG_HEADER) { - /* negative reply_len means v3 format, otherwise v1/v2 */ - if (get_user(reply_len, &old_hdr->reply_len)) - return -EFAULT; - - if (reply_len >= 0) - return get_user(*pack_id, &old_hdr->pack_id); - - if (in_compat_syscall() && - count >= sizeof(struct compat_sg_io_hdr)) { - struct compat_sg_io_hdr __user *hp = buf; - - return get_user(*pack_id, &hp->pack_id); - } - - if (count >= sizeof(struct sg_io_hdr)) { - struct sg_io_hdr __user *hp = buf; - - return get_user(*pack_id, &hp->pack_id); - } - } - - /* no valid header was passed, so ignore the pack_id */ - *pack_id = -1; - return 0; -} - -static ssize_t -sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos) -{ - Sg_device *sdp; - Sg_fd *sfp; - Sg_request *srp; - int req_pack_id = -1; - sg_io_hdr_t *hp; - struct sg_header *old_hdr; - int retval; - - /* - * This could cause a response to be stranded. Close the associated - * file descriptor to free up any resources being held. - */ - retval = sg_check_file_access(filp, __func__); - if (retval) - return retval; - - if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp))) - return -ENXIO; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "sg_read: count=%d\n", (int) count)); - - if (sfp->force_packid) - retval = get_sg_io_pack_id(&req_pack_id, buf, count); - if (retval) - return retval; - - srp = sg_get_rq_mark(sfp, req_pack_id); - if (!srp) { /* now wait on packet to arrive */ - if (atomic_read(&sdp->detaching)) - return -ENODEV; - if (filp->f_flags & O_NONBLOCK) - return -EAGAIN; - retval = wait_event_interruptible(sfp->read_wait, - (atomic_read(&sdp->detaching) || - (srp = sg_get_rq_mark(sfp, req_pack_id)))); - if (atomic_read(&sdp->detaching)) - return -ENODEV; - if (retval) - /* -ERESTARTSYS as signal hit process */ - return retval; - } - if (srp->header.interface_id != '\0') - return sg_new_read(sfp, buf, count, srp); - - hp = &srp->header; - old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL); - if (!old_hdr) - return -ENOMEM; - - old_hdr->reply_len = (int) hp->timeout; - old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */ - old_hdr->pack_id = hp->pack_id; - old_hdr->twelve_byte = - ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0; - old_hdr->target_status = hp->masked_status; - old_hdr->host_status = hp->host_status; - old_hdr->driver_status = hp->driver_status; - if ((CHECK_CONDITION & hp->masked_status) || - (DRIVER_SENSE & hp->driver_status)) - memcpy(old_hdr->sense_buffer, srp->sense_b, - sizeof (old_hdr->sense_buffer)); - switch (hp->host_status) { - /* This setup of 'result' is for backward compatibility and is best - ignored by the user who should use target, host + driver status */ - case DID_OK: - case DID_PASSTHROUGH: - case DID_SOFT_ERROR: - old_hdr->result = 0; - break; - case DID_NO_CONNECT: - case DID_BUS_BUSY: - case DID_TIME_OUT: - old_hdr->result = EBUSY; - break; - case DID_BAD_TARGET: - case DID_ABORT: - case DID_PARITY: - case DID_RESET: - case DID_BAD_INTR: - old_hdr->result = EIO; - break; - case DID_ERROR: - old_hdr->result = (srp->sense_b[0] == 0 && - hp->masked_status == GOOD) ? 0 : EIO; - break; - default: - old_hdr->result = EIO; - break; - } - - /* Now copy the result back to the user buffer. */ - if (count >= SZ_SG_HEADER) { - if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) { - retval = -EFAULT; - goto free_old_hdr; - } - buf += SZ_SG_HEADER; - if (count > old_hdr->reply_len) - count = old_hdr->reply_len; - if (count > SZ_SG_HEADER) { - if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) { - retval = -EFAULT; - goto free_old_hdr; - } - } - } else - count = (old_hdr->result == 0) ? 0 : -EIO; - sg_finish_rem_req(srp); - sg_remove_request(sfp, srp); - retval = count; -free_old_hdr: - kfree(old_hdr); - return retval; -} - -static ssize_t -sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp) -{ - sg_io_hdr_t *hp = &srp->header; - int err = 0, err2; - int len; - - if (in_compat_syscall()) { - if (count < sizeof(struct compat_sg_io_hdr)) { - err = -EINVAL; - goto err_out; - } - } else if (count < SZ_SG_IO_HDR) { - err = -EINVAL; - goto err_out; - } - hp->sb_len_wr = 0; - if ((hp->mx_sb_len > 0) && hp->sbp) { - if ((CHECK_CONDITION & hp->masked_status) || - (DRIVER_SENSE & hp->driver_status)) { - int sb_len = SCSI_SENSE_BUFFERSIZE; - sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len; - len = 8 + (int) srp->sense_b[7]; /* Additional sense length field */ - len = (len > sb_len) ? sb_len : len; - if (copy_to_user(hp->sbp, srp->sense_b, len)) { - err = -EFAULT; - goto err_out; - } - hp->sb_len_wr = len; - } - } - if (hp->masked_status || hp->host_status || hp->driver_status) - hp->info |= SG_INFO_CHECK; - err = put_sg_io_hdr(hp, buf); -err_out: - err2 = sg_finish_rem_req(srp); - sg_remove_request(sfp, srp); - return err ? : err2 ? : count; -} - static ssize_t sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) { @@ -708,6 +503,16 @@ sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) return (k < 0) ? k : count; } +static int sg_allow_access(struct file *filp, unsigned char *cmd) +{ + struct sg_fd *sfp = filp->private_data; + + if (sfp->parentdp->device->type == TYPE_SCANNER) + return 0; + + return blk_verify_command(cmd, filp->f_mode); +} + static ssize_t sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf, size_t count, int blocking, int read_only, int sg_io_owned, @@ -834,6 +639,75 @@ sg_common_write(Sg_fd * sfp, Sg_request * srp, return 0; } +/* + * read(2) related functions follow. They are shown after write(2) related + * functions. Apart from read(2) itself, ioctl(SG_IORECEIVE) and the second + * half of the ioctl(SG_IO) share code with read(2). + */ + +static Sg_request * +sg_get_rq_mark(Sg_fd *sfp, int pack_id) +{ + Sg_request *resp; + unsigned long iflags; + + write_lock_irqsave(&sfp->rq_list_lock, iflags); + list_for_each_entry(resp, &sfp->rq_list, entry) { + /* look for requests that are ready + not SG_IO owned */ + if (resp->done == 1 && !resp->sg_io_owned && + (-1 == pack_id || resp->header.pack_id == pack_id)) { + resp->done = 2; /* guard against other readers */ + write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + return resp; + } + } + write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + return NULL; +} + +static ssize_t +sg_new_read(Sg_fd *sfp, char __user *buf, size_t count, Sg_request *srp) +{ + sg_io_hdr_t *hp = &srp->header; + int err = 0, err2; + int len; + + if (in_compat_syscall()) { + if (count < sizeof(struct compat_sg_io_hdr)) { + err = -EINVAL; + goto err_out; + } + } else if (count < SZ_SG_IO_HDR) { + err = -EINVAL; + goto err_out; + } + hp->sb_len_wr = 0; + if (hp->mx_sb_len > 0 && hp->sbp) { + if ((CHECK_CONDITION & hp->masked_status) || + (DRIVER_SENSE & hp->driver_status)) { + int sb_len = SCSI_SENSE_BUFFERSIZE; + + sb_len = (hp->mx_sb_len > sb_len) ? sb_len : + hp->mx_sb_len; + /* Additional sense length field */ + len = 8 + (int)srp->sense_b[7]; + len = (len > sb_len) ? sb_len : len; + if (copy_to_user(hp->sbp, srp->sense_b, len)) { + err = -EFAULT; + goto err_out; + } + hp->sb_len_wr = len; + } + } + if (hp->masked_status || hp->host_status || hp->driver_status) + hp->info |= SG_INFO_CHECK; + err = put_sg_io_hdr(hp, buf); +err_out: + err2 = sg_finish_rem_req(srp); + sg_remove_request(sfp, srp); + return err ? : err2 ? : count; +} + static int srp_done(Sg_fd *sfp, Sg_request *srp) { unsigned long flags; @@ -845,6 +719,171 @@ static int srp_done(Sg_fd *sfp, Sg_request *srp) return ret; } +static ssize_t +sg_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) +{ + Sg_device *sdp; + Sg_fd *sfp; + Sg_request *srp; + int req_pack_id = -1; + sg_io_hdr_t *hp; + struct sg_header *old_hdr = NULL; + int retval = 0; + + /* + * This could cause a response to be stranded. Close the associated + * file descriptor to free up any resources being held. + */ + retval = sg_check_file_access(filp, __func__); + if (retval) + return retval; + + sfp = filp->private_data; + sdp = sfp->parentdp; + SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, + "%s: count=%d\n", __func__, + (int)count)); + if (!sdp) + return -ENXIO; + + if (!access_ok(buf, count)) + return -EFAULT; + if (sfp->force_packid && count >= SZ_SG_HEADER) { + old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL); + if (!old_hdr) + return -ENOMEM; + if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) { + retval = -EFAULT; + goto free_old_hdr; + } + if (old_hdr->reply_len < 0) { + if (count >= SZ_SG_IO_HDR) { + sg_io_hdr_t *new_hdr; + + new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL); + if (!new_hdr) { + retval = -ENOMEM; + goto free_old_hdr; + } + retval = __copy_from_user + (new_hdr, buf, SZ_SG_IO_HDR); + req_pack_id = new_hdr->pack_id; + kfree(new_hdr); + if (retval) { + retval = -EFAULT; + goto free_old_hdr; + } + } + } else { + req_pack_id = old_hdr->pack_id; + } + } + srp = sg_get_rq_mark(sfp, req_pack_id); + if (!srp) { /* now wait on packet to arrive */ + if (atomic_read(&sdp->detaching)) { + retval = -ENODEV; + goto free_old_hdr; + } + if (filp->f_flags & O_NONBLOCK) { + retval = -EAGAIN; + goto free_old_hdr; + } + retval = wait_event_interruptible + (sfp->read_wait, + (atomic_read(&sdp->detaching) || + (srp = sg_get_rq_mark(sfp, req_pack_id)))); + if (atomic_read(&sdp->detaching)) { + retval = -ENODEV; + goto free_old_hdr; + } + if (retval) { + /* -ERESTARTSYS as signal hit process */ + goto free_old_hdr; + } + } + if (srp->header.interface_id != '\0') { + retval = sg_new_read(sfp, buf, count, srp); + goto free_old_hdr; + } + + hp = &srp->header; + if (!old_hdr) { + old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL); + if (!old_hdr) { + retval = -ENOMEM; + goto free_old_hdr; + } + } + memset(old_hdr, 0, SZ_SG_HEADER); + old_hdr->reply_len = (int)hp->timeout; + old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */ + old_hdr->pack_id = hp->pack_id; + old_hdr->twelve_byte = + ((srp->data.cmd_opcode >= 0xc0) && (hp->cmd_len == 12)) ? 1 : 0; + old_hdr->target_status = hp->masked_status; + old_hdr->host_status = hp->host_status; + old_hdr->driver_status = hp->driver_status; + if ((hp->masked_status & CHECK_CONDITION) || + (hp->driver_status & DRIVER_SENSE)) + memcpy(old_hdr->sense_buffer, srp->sense_b, + sizeof(old_hdr->sense_buffer)); + switch (hp->host_status) { + /* + * This setup of 'result' is for backward compatibility and is best + * ignored by the user who should use target, host + driver status + */ + case DID_OK: + case DID_PASSTHROUGH: + case DID_SOFT_ERROR: + old_hdr->result = 0; + break; + case DID_NO_CONNECT: + case DID_BUS_BUSY: + case DID_TIME_OUT: + old_hdr->result = EBUSY; + break; + case DID_BAD_TARGET: + case DID_ABORT: + case DID_PARITY: + case DID_RESET: + case DID_BAD_INTR: + old_hdr->result = EIO; + break; + case DID_ERROR: + old_hdr->result = (srp->sense_b[0] == 0 && + hp->masked_status == GOOD) ? 0 : EIO; + break; + default: + old_hdr->result = EIO; + break; + } + + /* Now copy the result back to the user buffer. */ + if (count >= SZ_SG_HEADER) { + if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) { + retval = -EFAULT; + goto free_old_hdr; + } + buf += SZ_SG_HEADER; + if (count > old_hdr->reply_len) + count = old_hdr->reply_len; + if (count > SZ_SG_HEADER) { + if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) { + retval = -EFAULT; + goto free_old_hdr; + } + } + } else { + count = (old_hdr->result == 0) ? 0 : -EIO; + } + sg_finish_rem_req(srp); + sg_remove_request(sfp, srp); + retval = count; +free_old_hdr: + kfree(old_hdr); + return retval; +} + static int max_sectors_bytes(struct request_queue *q) { unsigned int max_sectors = queue_max_sectors(q); @@ -1692,9 +1731,7 @@ init_sg(void) sg_sysfs_valid = 1; rc = scsi_register_interface(&sg_interface); if (0 == rc) { -#ifdef CONFIG_SCSI_PROC_FS sg_proc_init(); -#endif /* CONFIG_SCSI_PROC_FS */ return 0; } class_destroy(sg_sysfs_class); @@ -1703,6 +1740,14 @@ init_sg(void) return rc; } +#ifndef CONFIG_SCSI_PROC_FS +static int +sg_proc_init(void) +{ + return 0; +} +#endif + static void __exit exit_sg(void) { @@ -2076,9 +2121,10 @@ sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size) rem -= num; } - if (k >= rsv_schp->k_use_sg) + if (k >= rsv_schp->k_use_sg) { SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp, "sg_link_reserve: BAD size\n")); + } } static void @@ -2099,26 +2145,6 @@ sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp) sfp->res_in_use = 0; } -static Sg_request * -sg_get_rq_mark(Sg_fd * sfp, int pack_id) -{ - Sg_request *resp; - unsigned long iflags; - - write_lock_irqsave(&sfp->rq_list_lock, iflags); - list_for_each_entry(resp, &sfp->rq_list, entry) { - /* look for requests that are ready + not SG_IO owned */ - if ((1 == resp->done) && (!resp->sg_io_owned) && - ((-1 == pack_id) || (resp->header.pack_id == pack_id))) { - resp->done = 2; /* guard against other readers */ - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); - return resp; - } - } - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); - return NULL; -} - /* always adds to end of list */ static Sg_request * sg_add_request(Sg_fd * sfp) From patchwork Thu Oct 15 02:06:02 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269811 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id CD480C433E7 for ; Thu, 15 Oct 2020 02:06:57 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 81C1322257 for ; Thu, 15 Oct 2020 02:06:57 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1728405AbgJOCG4 (ORCPT ); Wed, 14 Oct 2020 22:06:56 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40076 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728050AbgJOCG4 (ORCPT ); Wed, 14 Oct 2020 22:06:56 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 0BF9F204248; Thu, 15 Oct 2020 04:06:54 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id KbqM1wL3+siJ; Thu, 15 Oct 2020 04:06:52 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id EBBCB204255; Thu, 15 Oct 2020 04:06:47 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 03/44] sg: sg_log and is_enabled Date: Wed, 14 Oct 2020 22:06:02 -0400 Message-Id: <20201015020643.432908-4-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Replace SCSI_LOG_TIMEOUT macros with SG_LOG macros across the driver. The definition of SG_LOG calls SCSI_LOG_TIMEOUT if given and derived pointers are non-zero, calls pr_info otherwise. SG_LOGS additionally prints the sg device name and the thread id. The thread id is very useful, even in single threaded invocations because the driver not only uses the invocer's thread but also uses work queues and the main callback (i.e. sg_rq_end_io()) may hit any thread. Some interesting cases arise when the callback hits its invocer's thread. SG_LOGS takes 48 bytes on the stack to build this printf format string: "sg%u: tid=%d" whose size is clearly bounded above by the maximum size of those two integers. Protecting against the 'current' pointer being zero is for safety and the case where the boot device is SCSI and the sg driver is built into the kernel. Also when debugging, getting a message from a compromised kernel can be very useful in pinpointing the location of the failure. The simple fact that the SG_LOG macro is shorter than SCSI_LOG_TIMEOUT macro allow more error message "payload" per line. Also replace #if and #ifdef conditional compilations with the IS_ENABLED macro. *** Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 260 +++++++++++++++++++++++----------------------- 1 file changed, 130 insertions(+), 130 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index e5de7d78f4dd..d27c726981b9 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -57,6 +57,15 @@ static char *sg_version_date = "20190606"; #define SG_MAX_DEVS 32768 +/* Comment out the following line to compile out SCSI_LOGGING stuff */ +#define SG_DEBUG 1 + +#if !IS_ENABLED(SG_DEBUG) +#if IS_ENABLED(DEBUG) /* If SG_DEBUG not defined, check for DEBUG */ +#define SG_DEBUG DEBUG +#endif +#endif + /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater * than 16 bytes are "variable length" whose length is a multiple of 4 @@ -174,7 +183,7 @@ static ssize_t sg_new_write(struct sg_fd *sfp, struct file *file, static int sg_common_write(struct sg_fd *sfp, struct sg_request *srp, u8 *cmnd, int timeout, int blocking); static int sg_read_oxfer(struct sg_request *srp, char __user *outp, - int num_read_xfer); + int num_xfer); static void sg_remove_scat(struct sg_fd *sfp, struct sg_scatter_hold *schp); static void sg_build_reserve(struct sg_fd *sfp, int req_size); static void sg_link_reserve(struct sg_fd *sfp, struct sg_request *srp, @@ -187,14 +196,45 @@ static int sg_remove_request(struct sg_fd *sfp, struct sg_request *srp); static struct sg_device *sg_get_dev(int dev); static void sg_device_destroy(struct kref *kref); -#define SZ_SG_HEADER sizeof(struct sg_header) -#define SZ_SG_IO_HDR sizeof(sg_io_hdr_t) -#define SZ_SG_IOVEC sizeof(sg_iovec_t) -#define SZ_SG_REQ_INFO sizeof(sg_req_info_t) +#define SZ_SG_HEADER ((int)sizeof(struct sg_header)) /* v1 and v2 header */ +#define SZ_SG_IO_HDR ((int)sizeof(struct sg_io_hdr)) /* v3 header */ +#define SZ_SG_REQ_INFO ((int)sizeof(struct sg_req_info)) + +/* + * Kernel needs to be built with CONFIG_SCSI_LOGGING to see log messages. + * 'depth' is a number between 1 (most severe) and 7 (most noisy, most + * information). All messages are logged as informational (KERN_INFO). In + * the unexpected situation where sfp or sdp is NULL the macro reverts to + * a pr_info and ignores SCSI_LOG_TIMEOUT and always prints to the log. + * Example: this invocation: 'scsi_logging_level -s -T 3' will print + * depth (aka level) 1 and 2 SG_LOG() messages. + */ + +#define SG_PROC_DEBUG_SZ 8192 + +#if IS_ENABLED(CONFIG_SCSI_LOGGING) && IS_ENABLED(SG_DEBUG) +#define SG_LOG_BUFF_SZ 48 + +#define SG_LOG(depth, sfp, fmt, a...) \ + do { \ + char _b[SG_LOG_BUFF_SZ]; \ + int _tid = (current ? current->pid : -1); \ + struct sg_fd *_fp = sfp; \ + struct sg_device *_sdp = _fp ? _fp->parentdp : NULL; \ + \ + if (likely(_sdp && _sdp->disk)) { \ + snprintf(_b, sizeof(_b), "sg%u: tid=%d", \ + _sdp->index, _tid); \ + SCSI_LOG_TIMEOUT(depth, \ + sdev_prefix_printk(KERN_INFO, \ + _sdp->device, _b, fmt, ##a)); \ + } else \ + pr_info("sg: sdp or sfp NULL, " fmt, ##a); \ + } while (0) +#else +#define SG_LOG(depth, sfp, fmt, a...) +#endif /* end of CONFIG_SCSI_LOGGING && SG_DEBUG conditional */ -#define sg_printk(prefix, sdp, fmt, a...) \ - sdev_prefix_printk(prefix, (sdp)->device, \ - (sdp)->disk->disk_name, fmt, ##a) /* * The SCSI interfaces that use read() and write() as an asynchronous variant of @@ -286,9 +326,6 @@ sg_open(struct inode *inode, struct file *filp) if (IS_ERR(sdp)) return PTR_ERR(sdp); - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "sg_open: flags=0x%x\n", op_flags)); - /* This driver's module count bumped by fops_get in */ /* Prevent the device driver from vanishing while we sleep */ retval = scsi_device_get(sdp->device); @@ -346,6 +383,9 @@ sg_open(struct inode *inode, struct file *filp) filp->private_data = sfp; sdp->open_cnt++; mutex_unlock(&sdp->open_rel_lock); + SG_LOG(3, sfp, "%s: minor=%d, op_flags=0x%x; %s count prior=%d%s\n", + __func__, min_dev, op_flags, "device open", sdp->open_cnt, + ((op_flags & O_NONBLOCK) ? " O_NONBLOCK" : "")); retval = 0; sg_put: @@ -376,9 +416,10 @@ sg_release(struct inode *inode, struct file *filp) sfp = filp->private_data; sdp = sfp->parentdp; + SG_LOG(3, sfp, "%s: device open count prior=%d\n", __func__, + sdp->open_cnt); if (!sdp) return -ENXIO; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n")); mutex_lock(&sdp->open_rel_lock); scsi_autopm_put_device(sdp->device); @@ -417,10 +458,9 @@ sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) sfp = filp->private_data; sdp = sfp->parentdp; + SG_LOG(3, sfp, "%s: write(3rd arg) count=%d\n", __func__, (int)count); if (!sdp) return -ENXIO; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "sg_write: count=%d\n", (int) count)); if (atomic_read(&sdp->detaching)) return -ENODEV; if (!((filp->f_flags & O_NONBLOCK) || @@ -443,8 +483,7 @@ sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) return -EFAULT; if (!(srp = sg_add_request(sfp))) { - SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp, - "sg_write: queue full\n")); + SG_LOG(1, sfp, "%s: queue full\n", __func__); return -EDOM; } mutex_lock(&sfp->f_mutex); @@ -457,9 +496,8 @@ sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) cmd_size = 12; } mutex_unlock(&sfp->f_mutex); - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp, - "sg_write: scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size)); -/* Determine buffer size. */ + SG_LOG(4, sfp, "%s: scsi opcode=0x%02x, cmd_size=%d\n", __func__, + (unsigned int)opcode, cmd_size); input_size = count - cmd_size; mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len; mxsize -= SZ_SG_HEADER; @@ -540,8 +578,7 @@ sg_new_write(struct sg_fd *sfp, struct file *file, const char __user *buf, sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */ if (!(srp = sg_add_request(sfp))) { - SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp, - "sg_new_write: queue full\n")); + SG_LOG(1, sfp, "%s: queue full\n", __func__); return -EDOM; } srp->sg_io_owned = sg_io_owned; @@ -606,9 +643,8 @@ sg_common_write(struct sg_fd *sfp, struct sg_request *srp, hp->host_status = 0; hp->driver_status = 0; hp->resid = 0; - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp, - "sg_common_write: scsi opcode=0x%02x, cmd_size=%d\n", - (int) cmnd[0], (int) hp->cmd_len)); + SG_LOG(4, sfp, "%s: opcode=0x%02x, cmd_sz=%d\n", __func__, + (int)cmnd[0], hp->cmd_len); if (hp->dxfer_len >= SZ_256M) { sg_remove_request(sfp, srp); @@ -617,8 +653,7 @@ sg_common_write(struct sg_fd *sfp, struct sg_request *srp, k = sg_start_req(srp, cmnd); if (k) { - SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp, - "sg_common_write: start_req err=%d\n", k)); + SG_LOG(1, sfp, "%s: start_req err=%d\n", __func__, k); sg_finish_rem_req(srp); sg_remove_request(sfp, srp); return k; /* probably out of space --> ENOMEM */ @@ -752,9 +787,7 @@ sg_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) sfp = filp->private_data; sdp = sfp->parentdp; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "%s: count=%d\n", __func__, - (int)count)); + SG_LOG(3, sfp, "%s: read() count=%d\n", __func__, (int)count); if (!sdp) return -ENXIO; @@ -975,8 +1008,8 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, struct sg_request *srp; unsigned long iflags; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "sg_ioctl: cmd=0x%x\n", (int) cmd_in)); + SG_LOG(6, sfp, "%s: cmd=0x%x, O_NONBLOCK=%d\n", __func__, cmd_in, + !!(filp->f_flags & O_NONBLOCK)); read_only = (O_RDWR != (filp->f_flags & O_ACCMODE)); switch (cmd_in) { @@ -1221,8 +1254,9 @@ sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) return scsi_ioctl(sdp->device, cmd_in, p); } -#ifdef CONFIG_COMPAT -static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) +#if IS_ENABLED(CONFIG_COMPAT) +static long +sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) { void __user *p = compat_ptr(arg); struct sg_device *sdp; @@ -1275,24 +1309,16 @@ sg_poll(struct file *filp, poll_table * wait) res |= EPOLLOUT | EPOLLWRNORM; } else if (count < SG_MAX_QUEUE) res |= EPOLLOUT | EPOLLWRNORM; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "sg_poll: res=0x%x\n", (__force u32) res)); + SG_LOG(3, sfp, "%s: res=0x%x\n", __func__, (__force u32)res); return res; } static int sg_fasync(int fd, struct file *filp, int mode) { - struct sg_device *sdp; - struct sg_fd *sfp; - - sfp = filp->private_data; - sdp = sfp->parentdp; - if (!sdp) - return -ENXIO; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "sg_fasync: mode=%d\n", mode)); + struct sg_fd *sfp = filp->private_data; + SG_LOG(3, sfp, "%s: mode(%s)\n", __func__, (mode ? "add" : "remove")); return fasync_helper(fd, filp, mode, &sfp->async_qp); } @@ -1319,10 +1345,8 @@ sg_vma_fault(struct vm_fault *vmf) offset = vmf->pgoff << PAGE_SHIFT; if (offset >= rsv_schp->bufflen) return VM_FAULT_SIGBUS; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp, - "sg_vma_fault: offset=%lu, scatg=%d\n", - offset, rsv_schp->k_use_sg)); sa = vma->vm_start; + SG_LOG(3, sfp, "%s: vm_start=0x%lx, off=%lu\n", __func__, sa, offset); length = 1 << (PAGE_SHIFT + rsv_schp->page_order); for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) { len = vma->vm_end - sa; @@ -1362,9 +1386,8 @@ sg_mmap(struct file *filp, struct vm_area_struct *vma) return -ENXIO; } req_sz = vma->vm_end - vma->vm_start; - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp, - "sg_mmap starting, vm_start=%p, len=%d\n", - (void *) vma->vm_start, (int) req_sz)); + SG_LOG(3, sfp, "%s: vm_start=%p, len=%d\n", __func__, + (void *)vma->vm_start, (int)req_sz); if (vma->vm_pgoff) return -EINVAL; /* want no offset */ rsv_schp = &sfp->reserve; @@ -1433,10 +1456,9 @@ sg_rq_end_io(struct request *rq, blk_status_t status) result = req->result; resid = req->resid_len; - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp, - "sg_cmd_done: pack_id=%d, res=0x%x\n", - srp->header.pack_id, result)); srp->header.resid = resid; + SG_LOG(6, sfp, "%s: pack_id=%d, res=0x%x\n", __func__, + srp->header.pack_id, result); ms = jiffies_to_msecs(jiffies); srp->header.duration = (ms > srp->header.duration) ? (ms - srp->header.duration) : 0; @@ -1510,7 +1532,7 @@ static const struct file_operations sg_fops = { .write = sg_write, .poll = sg_poll, .unlocked_ioctl = sg_ioctl, -#ifdef CONFIG_COMPAT +#if IS_ENABLED(CONFIG_COMPAT) .compat_ioctl = sg_compat_ioctl, #endif .open = sg_open, @@ -1557,7 +1579,7 @@ sg_alloc(struct gendisk *disk, struct scsi_device *scsidp) k = error; SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp, - "sg_alloc: dev=%d \n", k)); + "%s: dev=%d, sdp=0x%p ++\n", __func__, k, sdp)); sprintf(disk->disk_name, "sg%d", k); disk->first_minor = k; sdp->disk = disk; @@ -1667,7 +1689,11 @@ sg_device_destroy(struct kref *kref) struct sg_device *sdp = container_of(kref, struct sg_device, d_ref); unsigned long flags; - /* CAUTION! Note that the device can still be found via idr_find() + SCSI_LOG_TIMEOUT(1, pr_info("[tid=%d] %s: sdp idx=%d, sdp=0x%p --\n", + (current ? current->pid : -1), __func__, + sdp->index, sdp)); + /* + * CAUTION! Note that the device can still be found via idr_find() * even though the refcount is 0. Therefore, do idr_remove() BEFORE * any other cleanup. */ @@ -1676,9 +1702,6 @@ sg_device_destroy(struct kref *kref) idr_remove(&sg_index_idr, sdp->index); write_unlock_irqrestore(&sg_index_lock, flags); - SCSI_LOG_TIMEOUT(3, - sg_printk(KERN_INFO, sdp, "sg_device_destroy\n")); - put_disk(sdp->disk); kfree(sdp); } @@ -1699,8 +1722,8 @@ sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf) if (val > 1) return; /* only want to do following once per device */ - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "%s\n", __func__)); + SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, sdp->device, + "%s: 0x%p\n", __func__, sdp)); read_lock_irqsave(&sdp->sfd_lock, iflags); list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) { @@ -1769,7 +1792,7 @@ init_sg(void) return rc; } -#ifndef CONFIG_SCSI_PROC_FS +#if !IS_ENABLED(CONFIG_SCSI_PROC_FS) static int sg_proc_init(void) { @@ -1780,9 +1803,8 @@ sg_proc_init(void) static void __exit exit_sg(void) { -#ifdef CONFIG_SCSI_PROC_FS - remove_proc_subtree("scsi/sg", NULL); -#endif /* CONFIG_SCSI_PROC_FS */ + if (IS_ENABLED(CONFIG_SCSI_PROC_FS)) + remove_proc_subtree("scsi/sg", NULL); scsi_unregister_interface(&sg_interface); class_destroy(sg_sysfs_class); sg_sysfs_valid = 0; @@ -1809,15 +1831,14 @@ sg_start_req(struct sg_request *srp, u8 *cmd) int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ; u8 *long_cmdp = NULL; - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp, - "sg_start_req: dxfer_len=%d\n", - dxfer_len)); - if (hp->cmd_len > BLK_MAX_CDB) { long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL); if (!long_cmdp) return -ENOMEM; + SG_LOG(5, sfp, "%s: long_cmdp=0x%p ++\n", __func__, long_cmdp); } + SG_LOG(4, sfp, "%s: dxfer_len=%d, data-%s\n", __func__, dxfer_len, + (rw ? "OUT" : "IN")); /* * NOTE @@ -1936,9 +1957,8 @@ sg_finish_rem_req(struct sg_request *srp) struct sg_fd *sfp = srp->parentfp; struct sg_scatter_hold *req_schp = &srp->data; - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp, - "sg_finish_rem_req: res_used=%d\n", - (int) srp->res_used)); + SG_LOG(4, sfp, "%s: srp=0x%p%s\n", __func__, srp, + (srp->res_used) ? " rsv" : ""); if (srp->bio) ret = blk_rq_unmap_user(srp->bio); @@ -1985,9 +2005,8 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, ++blk_size; /* don't know why */ /* round request up to next highest SG_SECTOR_SZ byte boundary */ blk_size = ALIGN(blk_size, SG_SECTOR_SZ); - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp, - "sg_build_indirect: buff_size=%d, blk_size=%d\n", - buff_size, blk_size)); + SG_LOG(4, sfp, "%s: buff_size=%d, blk_size=%d\n", __func__, buff_size, + blk_size); /* N.B. ret_sz carried into this block ... */ mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize); @@ -2026,18 +2045,13 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, scatter_elem_sz_prev = ret_sz; } } - - SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp, - "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n", - k, num, ret_sz)); + SG_LOG(5, sfp, "%s: k=%d, num=%d, ret_sz=%d\n", __func__, k, + num, ret_sz); } /* end of for loop */ schp->page_order = order; schp->k_use_sg = k; - SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp, - "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n", - k, rem_sz)); - + SG_LOG(5, sfp, "%s: k_use_sg=%d, order=%d\n", __func__, k, order); schp->bufflen = blk_size; if (rem_sz > 0) /* must have failed */ return -ENOMEM; @@ -2055,35 +2069,34 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, static void sg_remove_scat(struct sg_fd *sfp, struct sg_scatter_hold *schp) { - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp, - "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg)); + SG_LOG(4, sfp, "%s: num_sgat=%d\n", __func__, schp->k_use_sg); if (schp->pages && schp->sglist_len > 0) { if (!schp->dio_in_use) { int k; for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) { - SCSI_LOG_TIMEOUT(5, - sg_printk(KERN_INFO, sfp->parentdp, - "sg_remove_scat: k=%d, pg=0x%p\n", - k, schp->pages[k])); + SG_LOG(5, sfp, "%s: pg[%d]=0x%p --\n", + __func__, k, schp->pages[k]); __free_pages(schp->pages[k], schp->page_order); } - kfree(schp->pages); } } memset(schp, 0, sizeof (*schp)); } +/* + * For sg v1 and v2 interface: with a command yielding a data-in buffer, after + * it has arrived in kernel memory, this function copies it to the user space, + * appended to given struct sg_header object. + */ static int sg_read_oxfer(struct sg_request *srp, char __user *outp, int num_read_xfer) { struct sg_scatter_hold *schp = &srp->data; int k, num; - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp, - "sg_read_oxfer: num_read_xfer=%d\n", - num_read_xfer)); + SG_LOG(4, srp->parentfp, "%s: num_xfer=%d\n", __func__, num_read_xfer); if ((!outp) || (num_read_xfer <= 0)) return 0; @@ -2113,8 +2126,7 @@ sg_build_reserve(struct sg_fd *sfp, int req_size) { struct sg_scatter_hold *schp = &sfp->reserve; - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp, - "sg_build_reserve: req_size=%d\n", req_size)); + SG_LOG(3, sfp, "%s: buflen=%d\n", __func__, req_size); do { if (req_size < PAGE_SIZE) req_size = PAGE_SIZE; @@ -2134,8 +2146,7 @@ sg_link_reserve(struct sg_fd *sfp, struct sg_request *srp, int size) int k, num, rem; srp->res_used = 1; - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp, - "sg_link_reserve: size=%d\n", size)); + SG_LOG(4, sfp, "%s: size=%d\n", __func__, size); rem = size; num = 1 << (PAGE_SHIFT + rsv_schp->page_order); @@ -2153,8 +2164,7 @@ sg_link_reserve(struct sg_fd *sfp, struct sg_request *srp, int size) } if (k >= rsv_schp->k_use_sg) { - SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp, - "sg_link_reserve: BAD size\n")); + SG_LOG(1, sfp, "%s: BAD size\n", __func__); } } @@ -2163,9 +2173,8 @@ sg_unlink_reserve(struct sg_fd *sfp, struct sg_request *srp) { struct sg_scatter_hold *req_schp = &srp->data; - SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp, - "sg_unlink_reserve: req->k_use_sg=%d\n", - (int) req_schp->k_use_sg)); + SG_LOG(4, srp->parentfp, "%s: req->k_use_sg=%d\n", __func__, + (int)req_schp->k_use_sg); req_schp->k_use_sg = 0; req_schp->bufflen = 0; req_schp->pages = NULL; @@ -2256,18 +2265,15 @@ sg_add_sfp(struct sg_device *sdp) } list_add_tail(&sfp->sfd_siblings, &sdp->sfds); write_unlock_irqrestore(&sdp->sfd_lock, iflags); - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "sg_add_sfp: sfp=0x%p\n", sfp)); + SG_LOG(3, sfp, "%s: sfp=0x%p\n", __func__, sfp); if (unlikely(sg_big_buff != def_reserved_size)) sg_big_buff = def_reserved_size; bufflen = min_t(int, sg_big_buff, max_sectors_bytes(sdp->device->request_queue)); sg_build_reserve(sfp, bufflen); - SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, - "sg_add_sfp: bufflen=%d, k_use_sg=%d\n", - sfp->reserve.bufflen, - sfp->reserve.k_use_sg)); + SG_LOG(3, sfp, "%s: bufflen=%d, k_use_sg=%d\n", __func__, + sfp->reserve.bufflen, sfp->reserve.k_use_sg); kref_get(&sdp->d_ref); __module_get(THIS_MODULE); @@ -2293,15 +2299,12 @@ sg_remove_sfp_usercontext(struct work_struct *work) write_unlock_irqrestore(&sfp->rq_list_lock, iflags); if (sfp->reserve.bufflen > 0) { - SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp, - "sg_remove_sfp: bufflen=%d, k_use_sg=%d\n", - (int) sfp->reserve.bufflen, - (int) sfp->reserve.k_use_sg)); + SG_LOG(6, sfp, "%s: bufflen=%d, k_use_sg=%d\n", __func__, + (int)sfp->reserve.bufflen, (int)sfp->reserve.k_use_sg); sg_remove_scat(sfp, &sfp->reserve); } - SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp, - "sg_remove_sfp: sfp=0x%p\n", sfp)); + SG_LOG(6, sfp, "%s: sfp=0x%p\n", __func__, sfp); kfree(sfp); scsi_device_put(sdp->device); @@ -2324,7 +2327,6 @@ sg_remove_sfp(struct kref *kref) schedule_work(&sfp->ew.work); } -#ifdef CONFIG_SCSI_PROC_FS static int sg_idr_max_id(int id, void *p, void *data) { @@ -2336,19 +2338,6 @@ sg_idr_max_id(int id, void *p, void *data) return 0; } -static int -sg_last_dev(void) -{ - int k = -1; - unsigned long iflags; - - read_lock_irqsave(&sg_index_lock, iflags); - idr_for_each(&sg_index_idr, sg_idr_max_id, &k); - read_unlock_irqrestore(&sg_index_lock, iflags); - return k + 1; /* origin 1 */ -} -#endif - /* must be called with sg_index_lock held */ static struct sg_device * sg_lookup_dev(int dev) @@ -2378,7 +2367,7 @@ sg_get_dev(int dev) return sdp; } -#ifdef CONFIG_SCSI_PROC_FS +#if IS_ENABLED(CONFIG_SCSI_PROC_FS) /* long, almost to end of file */ static int sg_proc_seq_show_int(struct seq_file *s, void *v); static int sg_proc_single_open_adio(struct inode *inode, struct file *file); @@ -2451,6 +2440,17 @@ sg_proc_init(void) return 0; } +static int +sg_last_dev(void) +{ + int k = -1; + unsigned long iflags; + + read_lock_irqsave(&sg_index_lock, iflags); + idr_for_each(&sg_index_idr, sg_idr_max_id, &k); + read_unlock_irqrestore(&sg_index_lock, iflags); + return k + 1; /* origin 1 */ +} static int sg_proc_seq_show_int(struct seq_file *s, void *v) @@ -2709,7 +2709,7 @@ sg_proc_seq_show_debug(struct seq_file *s, void *v) return 0; } -#endif /* CONFIG_SCSI_PROC_FS */ +#endif /* CONFIG_SCSI_PROC_FS (~300 lines back) */ module_init(init_sg); module_exit(exit_sg); From patchwork Thu Oct 15 02:06:05 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269810 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id A4A45C43457 for ; Thu, 15 Oct 2020 02:07:02 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 553EF22257 for ; Thu, 15 Oct 2020 02:07:02 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729033AbgJOCHB (ORCPT ); Wed, 14 Oct 2020 22:07:01 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40092 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728418AbgJOCG7 (ORCPT ); Wed, 14 Oct 2020 22:06:59 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 9B760204248; Thu, 15 Oct 2020 04:06:57 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id Jjm1dCVI-FFo; Thu, 15 Oct 2020 04:06:55 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 4E66320425A; Thu, 15 Oct 2020 04:06:51 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 06/44] sg: make open count an atomic Date: Wed, 14 Oct 2020 22:06:05 -0400 Message-Id: <20201015020643.432908-7-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Convert sg_device::open_cnt into an atomic. Also rename sg_tablesize into the more descriptive max_sgat_elems. **** Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 44 +++++++++++++++++++++++--------------------- 1 file changed, 23 insertions(+), 21 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 19714ccdda57..b07fa2fc24a0 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -162,9 +162,9 @@ struct sg_device { /* holds the state of each scsi generic device */ struct mutex open_rel_lock; /* held when in open() or release() */ struct list_head sfds; rwlock_t sfd_lock; /* protect access to sfd list */ - int sg_tablesize; /* adapter's max scatter-gather table size */ + int max_sgat_elems; /* adapter's max sgat number of elements */ u32 index; /* device index number */ - int open_cnt; /* count of opens (perhaps < num(sfds) ) */ + atomic_t open_cnt; /* count of opens (perhaps < num(sfds) ) */ unsigned long fdev_bm[1]; /* see SG_FDEV_* defines above */ struct gendisk *disk; struct cdev *cdev; @@ -276,11 +276,11 @@ sg_wait_open_event(struct sg_device *sdp, bool o_excl) int retval = 0; if (o_excl) { - while (sdp->open_cnt > 0) { + while (atomic_read(&sdp->open_cnt) > 0) { mutex_unlock(&sdp->open_rel_lock); retval = wait_event_interruptible(sdp->open_wait, (SG_IS_DETACHING(sdp) || - !sdp->open_cnt)); + atomic_read(&sdp->open_cnt) == 0)); mutex_lock(&sdp->open_rel_lock); if (retval) /* -ERESTARTSYS */ @@ -328,7 +328,7 @@ sg_open(struct inode *inode, struct file *filp) o_excl = !!(op_flags & O_EXCL); if (o_excl && ((op_flags & O_ACCMODE) == O_RDONLY)) return -EPERM; /* Can't lock it with read only access */ - sdp = sg_get_dev(min_dev); + sdp = sg_get_dev(min_dev); /* increments sdp->d_ref */ if (IS_ERR(sdp)) return PTR_ERR(sdp); @@ -355,7 +355,7 @@ sg_open(struct inode *inode, struct file *filp) mutex_lock(&sdp->open_rel_lock); if (op_flags & O_NONBLOCK) { if (o_excl) { - if (sdp->open_cnt > 0) { + if (atomic_read(&sdp->open_cnt) > 0) { retval = -EBUSY; goto error_mutex_locked; } @@ -375,27 +375,29 @@ sg_open(struct inode *inode, struct file *filp) if (o_excl) set_bit(SG_FDEV_EXCLUDE, sdp->fdev_bm); - if (sdp->open_cnt < 1) { /* no existing opens */ + if (atomic_read(&sdp->open_cnt) < 1) { /* no existing opens */ clear_bit(SG_FDEV_LOG_SENSE, sdp->fdev_bm); q = sdp->device->request_queue; - sdp->sg_tablesize = queue_max_segments(q); + sdp->max_sgat_elems = queue_max_segments(q); } - sfp = sg_add_sfp(sdp); + sfp = sg_add_sfp(sdp); /* increments sdp->d_ref */ if (IS_ERR(sfp)) { retval = PTR_ERR(sfp); goto out_undo; } filp->private_data = sfp; - sdp->open_cnt++; + atomic_inc(&sdp->open_cnt); mutex_unlock(&sdp->open_rel_lock); SG_LOG(3, sfp, "%s: minor=%d, op_flags=0x%x; %s count prior=%d%s\n", - __func__, min_dev, op_flags, "device open", sdp->open_cnt, + __func__, min_dev, op_flags, "device open", + atomic_read(&sdp->open_cnt), ((op_flags & O_NONBLOCK) ? " O_NONBLOCK" : "")); retval = 0; sg_put: kref_put(&sdp->d_ref, sg_device_destroy); + /* if success, sdp->d_ref is incremented twice, decremented once */ return retval; out_undo: @@ -423,20 +425,20 @@ sg_release(struct inode *inode, struct file *filp) sfp = filp->private_data; sdp = sfp->parentdp; SG_LOG(3, sfp, "%s: device open count prior=%d\n", __func__, - sdp->open_cnt); + atomic_read(&sdp->open_cnt)); if (!sdp) return -ENXIO; mutex_lock(&sdp->open_rel_lock); scsi_autopm_put_device(sdp->device); kref_put(&sfp->f_ref, sg_remove_sfp); - sdp->open_cnt--; + atomic_dec(&sdp->open_cnt); /* possibly many open()s waiting on exlude clearing, start many; * only open(O_EXCL)s wait on 0==open_cnt so only start one */ if (test_and_clear_bit(SG_FDEV_EXCLUDE, sdp->fdev_bm)) wake_up_interruptible_all(&sdp->open_wait); - else if (sdp->open_cnt == 0) + else if (atomic_read(&sdp->open_cnt) == 0) wake_up_interruptible(&sdp->open_wait); mutex_unlock(&sdp->open_rel_lock); return 0; @@ -1110,7 +1112,7 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, read_unlock_irqrestore(&sfp->rq_list_lock, iflags); return put_user(val, ip); case SG_GET_SG_TABLESIZE: - return put_user(sdp->sg_tablesize, ip); + return put_user(sdp->max_sgat_elems, ip); case SG_SET_RESERVED_SIZE: result = get_user(val, ip); if (result) @@ -1589,7 +1591,7 @@ sg_alloc(struct gendisk *disk, struct scsi_device *scsidp) init_waitqueue_head(&sdp->open_wait); clear_bit(SG_FDEV_DETACHING, sdp->fdev_bm); rwlock_init(&sdp->sfd_lock); - sdp->sg_tablesize = queue_max_segments(q); + sdp->max_sgat_elems = queue_max_segments(q); sdp->index = k; kref_init(&sdp->d_ref); error = 0; @@ -1992,7 +1994,7 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, int buff_size) { int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems; - int sg_tablesize = sfp->parentdp->sg_tablesize; + int max_sgat_elems = sfp->parentdp->max_sgat_elems; int blk_size = buff_size, order; gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO; struct sg_device *sdp = sfp->parentdp; @@ -2007,7 +2009,7 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, blk_size); /* N.B. ret_sz carried into this block ... */ - mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize); + mx_sc_elems = sg_build_sgat(schp, sfp, max_sgat_elems); if (mx_sc_elems < 0) return mx_sc_elems; /* most likely -ENOMEM */ @@ -2695,9 +2697,9 @@ sg_proc_seq_show_debug(struct seq_file *s, void *v) scsidp->lun, scsidp->host->hostt->emulated); } - seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n", - sdp->sg_tablesize, SG_HAVE_EXCLUDE(sdp), - sdp->open_cnt); + seq_printf(s, " max_sgat_elems=%d excl=%d open_cnt=%d\n", + sdp->max_sgat_elems, SG_HAVE_EXCLUDE(sdp), + atomic_read(&sdp->open_cnt)); sg_proc_debug_helper(s, sdp); } read_unlock(&sdp->sfd_lock); From patchwork Thu Oct 15 02:06:07 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269809 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 787B7C4363D for ; Thu, 15 Oct 2020 02:07:03 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 3B73C22258 for ; Thu, 15 Oct 2020 02:07:03 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729034AbgJOCHD (ORCPT ); Wed, 14 Oct 2020 22:07:03 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40099 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728528AbgJOCHB (ORCPT ); Wed, 14 Oct 2020 22:07:01 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 527BB20425B; Thu, 15 Oct 2020 04:06:59 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id zc6XBj0gRXlo; Thu, 15 Oct 2020 04:06:57 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 87C2D204269; Thu, 15 Oct 2020 04:06:53 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de, Hannes Reinecke Subject: [PATCH v11 08/44] sg: speed sg_poll and sg_get_num_waiting Date: Wed, 14 Oct 2020 22:06:07 -0400 Message-Id: <20201015020643.432908-9-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Track the number of submitted and waiting (for read/receive) requests on each file descriptor with two atomic integers. This speeds sg_poll() and ioctl(SG_GET_NUM_WAITING) which are oft used with the asynchronous (non-blocking) interfaces. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 54 +++++++++++++++++++++++------------------------ 1 file changed, 27 insertions(+), 27 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index b07fa2fc24a0..0ace3e2e464a 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -142,6 +142,8 @@ struct sg_fd { /* holds the state of a file descriptor */ struct mutex f_mutex; /* protect against changes in this fd */ int timeout; /* defaults to SG_DEFAULT_TIMEOUT */ int timeout_user; /* defaults to SG_DEFAULT_TIMEOUT_USER */ + atomic_t submitted; /* number inflight or awaiting read */ + atomic_t waiting; /* number of requests awaiting read */ struct sg_scatter_hold reserve; /* buffer for this file descriptor */ struct list_head rq_list; /* head of request list */ struct fasync_struct *async_qp; /* used by asynchronous notification */ @@ -683,6 +685,8 @@ sg_common_write(struct sg_fd *sfp, struct sg_request *srp, else at_head = 1; + if (!blocking) + atomic_inc(&sfp->submitted); srp->rq->timeout = timeout; kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */ blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk, @@ -1103,14 +1107,7 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, read_unlock_irqrestore(&sfp->rq_list_lock, iflags); return put_user(-1, ip); case SG_GET_NUM_WAITING: - read_lock_irqsave(&sfp->rq_list_lock, iflags); - val = 0; - list_for_each_entry(srp, &sfp->rq_list, entry) { - if ((1 == srp->done) && (!srp->sg_io_owned)) - ++val; - } - read_unlock_irqrestore(&sfp->rq_list_lock, iflags); - return put_user(val, ip); + return put_user(atomic_read(&sfp->waiting), ip); case SG_GET_SG_TABLESIZE: return put_user(sdp->max_sgat_elems, ip); case SG_SET_RESERVED_SIZE: @@ -1282,35 +1279,26 @@ sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) } #endif +/* + * Implements the poll(2) system call for this driver. Returns various EPOLL* + * flags OR-ed together. + */ static __poll_t sg_poll(struct file *filp, poll_table * wait) { __poll_t p_res = 0; struct sg_fd *sfp = filp->private_data; - struct sg_request *srp; - int count = 0; - unsigned long iflags; - if (!sfp) - return EPOLLERR; poll_wait(filp, &sfp->read_wait, wait); - read_lock_irqsave(&sfp->rq_list_lock, iflags); - list_for_each_entry(srp, &sfp->rq_list, entry) { - /* if any read waiting, flag it */ - if (p_res == 0 && srp->done == 1 && !srp->sg_io_owned) - p_res = EPOLLIN | EPOLLRDNORM; - ++count; - } - read_unlock_irqrestore(&sfp->rq_list_lock, iflags); + if (atomic_read(&sfp->waiting) > 0) + p_res = EPOLLIN | EPOLLRDNORM; - if (sfp->parentdp && SG_IS_DETACHING(sfp->parentdp)) { + if (unlikely(SG_IS_DETACHING(sfp->parentdp))) p_res |= EPOLLHUP; - } else if (!sfp->cmd_q) { - if (count == 0) - p_res |= EPOLLOUT | EPOLLWRNORM; - } else if (count < SG_MAX_QUEUE) { + else if (likely(sfp->cmd_q)) + p_res |= EPOLLOUT | EPOLLWRNORM; + else if (atomic_read(&sfp->submitted) == 0) p_res |= EPOLLOUT | EPOLLWRNORM; - } SG_LOG(3, sfp, "%s: p_res=0x%x\n", __func__, (__force u32)p_res); return p_res; } @@ -1495,6 +1483,8 @@ sg_rq_end_io(struct request *rq, blk_status_t status) /* Rely on write phase to clean out srp status values, so no "else" */ + if (!srp->sg_io_owned) + atomic_inc(&sfp->waiting); /* * Free the request as soon as it is complete so that its resources * can be reused without waiting for userspace to read() the @@ -1959,6 +1949,10 @@ sg_finish_rem_req(struct sg_request *srp) SG_LOG(4, sfp, "%s: srp=0x%p%s\n", __func__, srp, (srp->res_used) ? " rsv" : ""); + if (!srp->sg_io_owned) { + atomic_dec(&sfp->submitted); + atomic_dec(&sfp->waiting); + } if (srp->bio) ret = blk_rq_unmap_user(srp->bio); @@ -2256,6 +2250,9 @@ sg_add_sfp(struct sg_device *sdp) sfp->cmd_q = SG_DEF_COMMAND_Q; sfp->keep_orphan = SG_DEF_KEEP_ORPHAN; sfp->parentdp = sdp; + atomic_set(&sfp->submitted, 0); + atomic_set(&sfp->waiting, 0); + write_lock_irqsave(&sdp->sfd_lock, iflags); if (SG_IS_DETACHING(sdp)) { write_unlock_irqrestore(&sdp->sfd_lock, iflags); @@ -2627,6 +2624,9 @@ sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) seq_printf(s, " cmd_q=%d f_packid=%d k_orphan=%d closed=0\n", (int) fp->cmd_q, (int) fp->force_packid, (int) fp->keep_orphan); + seq_printf(s, " submitted=%d waiting=%d\n", + atomic_read(&fp->submitted), + atomic_read(&fp->waiting)); list_for_each_entry(srp, &fp->rq_list, entry) { hp = &srp->header; new_interface = (hp->interface_id == '\0') ? 0 : 1; From patchwork Thu Oct 15 02:06:09 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269807 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id EF6ECC433DF for ; Thu, 15 Oct 2020 02:07:12 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id A487622257 for ; Thu, 15 Oct 2020 02:07:12 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729635AbgJOCHM (ORCPT ); Wed, 14 Oct 2020 22:07:12 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40122 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728528AbgJOCHH (ORCPT ); Wed, 14 Oct 2020 22:07:07 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 53383204269; Thu, 15 Oct 2020 04:07:04 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id fYocAKyeI1DB; Thu, 15 Oct 2020 04:07:00 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id E8ACD20424C; Thu, 15 Oct 2020 04:06:55 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 10/44] sg: improve naming Date: Wed, 14 Oct 2020 22:06:09 -0400 Message-Id: <20201015020643.432908-11-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Remove use of typedef sg_io_hdr_t and replace with struct sg_io_hdr. Change some names on driver wide structure fields and comment them. Rename sg_alloc() to sg_add_device_helper() to reflect its current role. Rename sg_add_request() to sg_setup_req() to imply that it precedes sg_start_req(). Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 248 ++++++++++++++++++++++++---------------------- 1 file changed, 131 insertions(+), 117 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 131fec52df29..7aae7ce6cebf 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -92,7 +92,7 @@ static int sg_allow_dio = SG_ALLOW_DIO_DEF; static int scatter_elem_sz = SG_SCATTER_SZ; static int scatter_elem_sz_prev = SG_SCATTER_SZ; -#define SG_SECTOR_SZ 512 +#define SG_DEF_SECTOR_SZ 512 static int sg_add_device(struct device *, struct class_interface *); static void sg_remove_device(struct device *, struct class_interface *); @@ -105,12 +105,13 @@ static struct class_interface sg_interface = { .remove_dev = sg_remove_device, }; -struct sg_scatter_hold { /* holding area for scsi scatter gather info */ - u16 k_use_sg; /* Count of kernel scatter-gather pieces */ +struct sg_scatter_hold { /* holding area for scsi scatter gather info */ + struct page **pages; /* num_sgat element array of struct page* */ + int buflen; /* capacity in bytes (dlen<=buflen) */ + int dlen; /* current valid data length of this req */ + u16 page_order; /* byte_len = (page_size*(2**page_order)) */ + u16 num_sgat; /* actual number of scatter-gather segments */ unsigned int sglist_len; /* size of malloc'd scatter-gather list ++ */ - unsigned int bufflen; /* Size of (aggregate) data buffer */ - struct page **pages; - int page_order; char dio_in_use; /* 0->indirect IO (or mmap), 1->dio */ u8 cmd_opcode; /* first byte of command */ }; @@ -122,20 +123,20 @@ struct sg_request { /* SG_MAX_QUEUE requests outstanding per file */ struct list_head entry; /* list entry */ struct sg_fd *parentfp; /* NULL -> not in use */ struct sg_scatter_hold data; /* hold buffer, perhaps scatter list */ - sg_io_hdr_t header; /* scsi command+info, see */ + struct sg_io_hdr header; /* scsi command+info, see */ u8 sense_b[SCSI_SENSE_BUFFERSIZE]; char res_used; /* 1 -> using reserve buffer, 0 -> not ... */ char orphan; /* 1 -> drop on sight, 0 -> normal */ char sg_io_owned; /* 1 -> packet belongs to SG_IO */ /* done protected by rq_list_lock */ char done; /* 0->before bh, 1->before read, 2->read */ - struct request *rq; - struct bio *bio; - struct execute_work ew; + struct request *rq; /* released in sg_rq_end_io(), bio kept */ + struct bio *bio; /* kept until this req -->SG_RS_INACTIVE */ + struct execute_work ew_orph; /* harvest orphan request */ }; struct sg_fd { /* holds the state of a file descriptor */ - struct list_head sfd_siblings; /* protected by device's sfd_lock */ + struct list_head sfd_entry; /* member sg_device::sfds list */ struct sg_device *parentdp; /* owning device */ wait_queue_head_t read_wait; /* queue read until command done */ rwlock_t rq_list_lock; /* protect access to list in req_arr */ @@ -155,7 +156,7 @@ struct sg_fd { /* holds the state of a file descriptor */ char mmap_called; /* 0 -> mmap() never called on this fd */ char res_in_use; /* 1 -> 'reserve' array in use */ struct kref f_ref; - struct execute_work ew; + struct execute_work ew_fd; /* harvest all fd resources and lists */ }; struct sg_device { /* holds the state of each scsi generic device */ @@ -164,7 +165,7 @@ struct sg_device { /* holds the state of each scsi generic device */ struct mutex open_rel_lock; /* held when in open() or release() */ struct list_head sfds; rwlock_t sfd_lock; /* protect access to sfd list */ - int max_sgat_elems; /* adapter's max sgat number of elements */ + int max_sgat_sz; /* max number of bytes in sgat list */ u32 index; /* device index number */ atomic_t open_cnt; /* count of opens (perhaps < num(sfds) ) */ unsigned long fdev_bm[1]; /* see SG_FDEV_* defines above */ @@ -196,7 +197,7 @@ static void sg_link_reserve(struct sg_fd *sfp, struct sg_request *srp, static void sg_unlink_reserve(struct sg_fd *sfp, struct sg_request *srp); static struct sg_fd *sg_add_sfp(struct sg_device *sdp); static void sg_remove_sfp(struct kref *); -static struct sg_request *sg_add_request(struct sg_fd *sfp); +static struct sg_request *sg_setup_req(struct sg_fd *sfp); static int sg_remove_request(struct sg_fd *sfp, struct sg_request *srp); static struct sg_device *sg_get_dev(int dev); static void sg_device_destroy(struct kref *kref); @@ -407,7 +408,7 @@ sg_open(struct inode *inode, struct file *filp) if (atomic_read(&sdp->open_cnt) < 1) { /* no existing opens */ clear_bit(SG_FDEV_LOG_SENSE, sdp->fdev_bm); q = sdp->device->request_queue; - sdp->max_sgat_elems = queue_max_segments(q); + sdp->max_sgat_sz = queue_max_segments(q); } sfp = sg_add_sfp(sdp); /* increments sdp->d_ref */ if (IS_ERR(sfp)) { @@ -474,16 +475,18 @@ sg_release(struct inode *inode, struct file *filp) } static ssize_t -sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) +sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) { bool blocking = !(filp->f_flags & O_NONBLOCK); - u8 opcode; int mxsize, cmd_size, input_size, res; + u8 opcode; struct sg_device *sdp; struct sg_fd *sfp; struct sg_request *srp; - struct sg_header old_hdr; - sg_io_hdr_t *hp; + struct sg_header ov2hdr; + struct sg_io_hdr v3hdr; + struct sg_header *ohp = &ov2hdr; + struct sg_io_hdr *h3p = &v3hdr; u8 cmnd[SG_MAX_CDB_SIZE]; res = sg_check_file_access(filp, __func__); @@ -493,25 +496,36 @@ sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) sfp = filp->private_data; sdp = sfp->parentdp; SG_LOG(3, sfp, "%s: write(3rd arg) count=%d\n", __func__, (int)count); - res = sg_allow_if_err_recovery(sdp, !blocking); + res = sg_allow_if_err_recovery(sdp, !!(filp->f_flags & O_NONBLOCK)); if (res) return res; if (count < SZ_SG_HEADER) return -EIO; - if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER)) + if (copy_from_user(ohp, p, SZ_SG_HEADER)) return -EFAULT; - if (old_hdr.reply_len < 0) - return sg_submit(sfp, filp, buf, count, blocking, false, false, + if (ohp->reply_len < 0) { /* assume this is v3 */ + struct sg_io_hdr *reinter_2p = (struct sg_io_hdr *)ohp; + + if (count < SZ_SG_IO_HDR) + return -EIO; + if (reinter_2p->interface_id != 'S') { + pr_info_once("sg: %s: v3 interface only here\n", + __func__); + return -EPERM; + } + return sg_submit(sfp, filp, p, count, + !(filp->f_flags & O_NONBLOCK), false, false, NULL); + } if (count < (SZ_SG_HEADER + 6)) return -EIO; /* The minimum scsi command length is 6 bytes. */ - buf += SZ_SG_HEADER; - if (get_user(opcode, buf)) + p += SZ_SG_HEADER; + if (get_user(opcode, p)) return -EFAULT; - if (!(srp = sg_add_request(sfp))) { + if (!(srp = sg_setup_req(sfp))) { SG_LOG(1, sfp, "%s: queue full\n", __func__); return -EDOM; } @@ -520,43 +534,44 @@ sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) cmd_size = sfp->next_cmd_len; sfp->next_cmd_len = 0; /* reset so only this write() effected */ } else { - cmd_size = COMMAND_SIZE(opcode); /* based on SCSI command group */ - if ((opcode >= 0xc0) && old_hdr.twelve_byte) + cmd_size = COMMAND_SIZE(opcode); /* old: SCSI command group */ + if (opcode >= 0xc0 && ohp->twelve_byte) cmd_size = 12; } mutex_unlock(&sfp->f_mutex); SG_LOG(4, sfp, "%s: scsi opcode=0x%02x, cmd_size=%d\n", __func__, (unsigned int)opcode, cmd_size); input_size = count - cmd_size; - mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len; + mxsize = max_t(int, input_size, ohp->reply_len); mxsize -= SZ_SG_HEADER; input_size -= SZ_SG_HEADER; if (input_size < 0) { sg_remove_request(sfp, srp); return -EIO; /* User did not pass enough bytes for this command. */ } - hp = &srp->header; - hp->interface_id = '\0'; /* indicator of old interface tunnelled */ - hp->cmd_len = (u8)cmd_size; - hp->iovec_count = 0; - hp->mx_sb_len = 0; + h3p = &srp->header; + h3p->interface_id = '\0'; /* indicator of old interface tunnelled */ + h3p->cmd_len = (u8)cmd_size; + h3p->iovec_count = 0; + h3p->mx_sb_len = 0; if (input_size > 0) - hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ? + h3p->dxfer_direction = (ohp->reply_len > SZ_SG_HEADER) ? SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV; else - hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE; - hp->dxfer_len = mxsize; - if ((hp->dxfer_direction == SG_DXFER_TO_DEV) || - (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV)) - hp->dxferp = (char __user *)buf + cmd_size; + h3p->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : + SG_DXFER_NONE; + h3p->dxfer_len = mxsize; + if (h3p->dxfer_direction == SG_DXFER_TO_DEV || + h3p->dxfer_direction == SG_DXFER_TO_FROM_DEV) + h3p->dxferp = (char __user *)p + cmd_size; else - hp->dxferp = NULL; - hp->sbp = NULL; - hp->timeout = old_hdr.reply_len; /* structure abuse ... */ - hp->flags = input_size; /* structure abuse ... */ - hp->pack_id = old_hdr.pack_id; - hp->usr_ptr = NULL; - if (copy_from_user(cmnd, buf, cmd_size)) { + h3p->dxferp = NULL; + h3p->sbp = NULL; + h3p->timeout = ohp->reply_len; /* structure abuse ... */ + h3p->flags = input_size; /* structure abuse ... */ + h3p->pack_id = ohp->pack_id; + h3p->usr_ptr = NULL; + if (copy_from_user(cmnd, p, cmd_size)) { sg_remove_request(sfp, srp); return -EFAULT; } @@ -565,13 +580,13 @@ sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos) * but is is possible that the app intended SG_DXFER_TO_DEV, because there * is a non-zero input_size, so emit a warning. */ - if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) { + if (h3p->dxfer_direction == SG_DXFER_TO_FROM_DEV) { printk_ratelimited(KERN_WARNING "sg_write: data in/out %d/%d bytes " "for SCSI command 0x%x-- guessing " "data in;\n program %s not setting " "count and/or reply_len properly\n", - old_hdr.reply_len - (int)SZ_SG_HEADER, + ohp->reply_len - (int)SZ_SG_HEADER, input_size, (unsigned int) cmnd[0], current->comm); } @@ -597,7 +612,7 @@ sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, { int k; struct sg_request *srp; - sg_io_hdr_t *hp; + struct sg_io_hdr *hp; u8 cmnd[SG_MAX_CDB_SIZE]; int timeout; unsigned long ul_timeout; @@ -606,7 +621,7 @@ sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, return -EINVAL; sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */ - if (!(srp = sg_add_request(sfp))) { + if (!(srp = sg_setup_req(sfp))) { SG_LOG(1, sfp, "%s: queue full\n", __func__); return -EDOM; } @@ -621,7 +636,7 @@ sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, return -ENOSYS; } if (hp->flags & SG_FLAG_MMAP_IO) { - if (hp->dxfer_len > sfp->reserve.bufflen) { + if (hp->dxfer_len > sfp->reserve.buflen) { sg_remove_request(sfp, srp); return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */ } @@ -662,7 +677,7 @@ sg_common_write(struct sg_fd *sfp, struct sg_request *srp, { int k, at_head; struct sg_device *sdp = sfp->parentdp; - sg_io_hdr_t *hp = &srp->header; + struct sg_io_hdr *hp = &srp->header; srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */ hp->status = 0; @@ -745,7 +760,7 @@ static ssize_t sg_new_read(struct sg_fd *sfp, char __user *buf, size_t count, struct sg_request *srp) { - sg_io_hdr_t *hp = &srp->header; + struct sg_io_hdr *hp = &srp->header; int err = 0, err2; int len; @@ -805,7 +820,7 @@ sg_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) struct sg_request *srp; int req_pack_id = -1; int ret = 0; - sg_io_hdr_t *hp; + struct sg_io_hdr *hp; struct sg_header *old_hdr = NULL; /* @@ -835,7 +850,7 @@ sg_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) } if (old_hdr->reply_len < 0) { if (count >= SZ_SG_IO_HDR) { - sg_io_hdr_t *new_hdr; + struct sg_io_hdr *new_hdr; new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL); if (!new_hdr) { @@ -1130,7 +1145,7 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, case SG_GET_NUM_WAITING: return put_user(atomic_read(&sfp->waiting), ip); case SG_GET_SG_TABLESIZE: - return put_user(sdp->max_sgat_elems, ip); + return put_user(sdp->max_sgat_sz, ip); case SG_SET_RESERVED_SIZE: result = get_user(val, ip); if (result) @@ -1140,7 +1155,7 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, val = min_t(int, val, max_sectors_bytes(sdp->device->request_queue)); mutex_lock(&sfp->f_mutex); - if (val != sfp->reserve.bufflen) { + if (val != sfp->reserve.buflen) { if (sfp->mmap_called || sfp->res_in_use) { mutex_unlock(&sfp->f_mutex); @@ -1153,7 +1168,7 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, mutex_unlock(&sfp->f_mutex); return 0; case SG_GET_RESERVED_SIZE: - val = min_t(int, sfp->reserve.bufflen, + val = min_t(int, sfp->reserve.buflen, max_sectors_bytes(sdp->device->request_queue)); return put_user(val, ip); case SG_SET_COMMAND_Q: @@ -1353,12 +1368,12 @@ sg_vma_fault(struct vm_fault *vmf) } rsv_schp = &sfp->reserve; offset = vmf->pgoff << PAGE_SHIFT; - if (offset >= rsv_schp->bufflen) + if (offset >= rsv_schp->buflen) return VM_FAULT_SIGBUS; sa = vma->vm_start; SG_LOG(3, sfp, "%s: vm_start=0x%lx, off=%lu\n", __func__, sa, offset); length = 1 << (PAGE_SHIFT + rsv_schp->page_order); - for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) { + for (k = 0; k < rsv_schp->num_sgat && sa < vma->vm_end; k++) { len = vma->vm_end - sa; len = (len < length) ? len : length; if (offset < len) { @@ -1402,14 +1417,14 @@ sg_mmap(struct file *filp, struct vm_area_struct *vma) return -EINVAL; /* want no offset */ rsv_schp = &sfp->reserve; mutex_lock(&sfp->f_mutex); - if (req_sz > rsv_schp->bufflen) { + if (req_sz > rsv_schp->buflen) { ret = -ENOMEM; /* cannot map more than reserved buffer */ goto out; } sa = vma->vm_start; length = 1 << (PAGE_SHIFT + rsv_schp->page_order); - for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) { + for (k = 0; k < rsv_schp->num_sgat && sa < vma->vm_end; k++) { len = vma->vm_end - sa; len = (len < length) ? len : length; sa += len; @@ -1427,7 +1442,8 @@ sg_mmap(struct file *filp, struct vm_area_struct *vma) static void sg_rq_end_io_usercontext(struct work_struct *work) { - struct sg_request *srp = container_of(work, struct sg_request, ew.work); + struct sg_request *srp = container_of(work, struct sg_request, + ew_orph.work); struct sg_fd *sfp = srp->parentfp; sg_finish_scsi_blk_rq(srp); @@ -1533,8 +1549,8 @@ sg_rq_end_io(struct request *rq, blk_status_t status) kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN); kref_put(&sfp->f_ref, sg_remove_sfp); } else { - INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext); - schedule_work(&srp->ew.work); + INIT_WORK(&srp->ew_orph.work, sg_rq_end_io_usercontext); + schedule_work(&srp->ew_orph.work); } } @@ -1559,7 +1575,7 @@ static struct class *sg_sysfs_class; static int sg_sysfs_valid = 0; static struct sg_device * -sg_alloc(struct gendisk *disk, struct scsi_device *scsidp) +sg_add_device_helper(struct gendisk *disk, struct scsi_device *scsidp) { struct request_queue *q = scsidp->request_queue; struct sg_device *sdp; @@ -1601,7 +1617,7 @@ sg_alloc(struct gendisk *disk, struct scsi_device *scsidp) init_waitqueue_head(&sdp->open_wait); clear_bit(SG_FDEV_DETACHING, sdp->fdev_bm); rwlock_init(&sdp->sfd_lock); - sdp->max_sgat_elems = queue_max_segments(q); + sdp->max_sgat_sz = queue_max_segments(q); sdp->index = k; kref_init(&sdp->d_ref); error = 0; @@ -1643,9 +1659,8 @@ sg_add_device(struct device *cl_dev, struct class_interface *cl_intf) cdev->owner = THIS_MODULE; cdev->ops = &sg_fops; - sdp = sg_alloc(disk, scsidp); + sdp = sg_add_device_helper(disk, scsidp); if (IS_ERR(sdp)) { - pr_warn("%s: sg_alloc failed\n", __func__); error = PTR_ERR(sdp); goto out; } @@ -1736,7 +1751,7 @@ sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf) "%s: 0x%p\n", __func__, sdp)); read_lock_irqsave(&sdp->sfd_lock, iflags); - list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) { + list_for_each_entry(sfp, &sdp->sfds, sfd_entry) { wake_up_interruptible_all(&sfp->read_wait); kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP); } @@ -1830,7 +1845,7 @@ sg_start_req(struct sg_request *srp, u8 *cmd) struct request *rq; struct scsi_request *req; struct sg_fd *sfp = srp->parentfp; - sg_io_hdr_t *hp = &srp->header; + struct sg_io_hdr *hp = &srp->header; int dxfer_len = (int) hp->dxfer_len; int dxfer_dir = hp->dxfer_direction; unsigned int iov_count = hp->iovec_count; @@ -1891,13 +1906,13 @@ sg_start_req(struct sg_request *srp, u8 *cmd) if (md) { mutex_lock(&sfp->f_mutex); - if (dxfer_len <= rsv_schp->bufflen && + if (dxfer_len <= rsv_schp->buflen && !sfp->res_in_use) { sfp->res_in_use = 1; sg_link_reserve(sfp, srp, dxfer_len); } else if (hp->flags & SG_FLAG_MMAP_IO) { res = -EBUSY; /* sfp->res_in_use == 1 */ - if (dxfer_len > rsv_schp->bufflen) + if (dxfer_len > rsv_schp->buflen) res = -ENOMEM; mutex_unlock(&sfp->f_mutex); return res; @@ -1912,7 +1927,7 @@ sg_start_req(struct sg_request *srp, u8 *cmd) md->pages = req_schp->pages; md->page_order = req_schp->page_order; - md->nr_entries = req_schp->k_use_sg; + md->nr_entries = req_schp->num_sgat; md->offset = 0; md->null_mapped = hp->dxferp ? 0 : 1; if (dxfer_dir == SG_DXFER_TO_FROM_DEV) @@ -1993,13 +2008,13 @@ static int sg_build_sgat(struct sg_scatter_hold *schp, const struct sg_fd *sfp, int tablesize) { - int sg_bufflen = tablesize * sizeof(struct page *); + int sg_buflen = tablesize * sizeof(struct page *); gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN; - schp->pages = kzalloc(sg_bufflen, gfp_flags); + schp->pages = kzalloc(sg_buflen, gfp_flags); if (!schp->pages) return -ENOMEM; - schp->sglist_len = sg_bufflen; + schp->sglist_len = sg_buflen; return tablesize; /* number of scat_gath elements allocated */ } @@ -2008,7 +2023,7 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, int buff_size) { int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems; - int max_sgat_elems = sfp->parentdp->max_sgat_elems; + int max_sgat_sz = sfp->parentdp->max_sgat_sz; int blk_size = buff_size, order; gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO; struct sg_device *sdp = sfp->parentdp; @@ -2017,13 +2032,13 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, return -EFAULT; if (0 == blk_size) ++blk_size; /* don't know why */ - /* round request up to next highest SG_SECTOR_SZ byte boundary */ - blk_size = ALIGN(blk_size, SG_SECTOR_SZ); + /* round request up to next highest SG_DEF_SECTOR_SZ byte boundary */ + blk_size = ALIGN(blk_size, SG_DEF_SECTOR_SZ); SG_LOG(4, sfp, "%s: buff_size=%d, blk_size=%d\n", __func__, buff_size, blk_size); /* N.B. ret_sz carried into this block ... */ - mx_sc_elems = sg_build_sgat(schp, sfp, max_sgat_elems); + mx_sc_elems = sg_build_sgat(schp, sfp, max_sgat_sz); if (mx_sc_elems < 0) return mx_sc_elems; /* most likely -ENOMEM */ @@ -2064,9 +2079,9 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, } /* end of for loop */ schp->page_order = order; - schp->k_use_sg = k; - SG_LOG(5, sfp, "%s: k_use_sg=%d, order=%d\n", __func__, k, order); - schp->bufflen = blk_size; + schp->num_sgat = k; + SG_LOG(5, sfp, "%s: num_sgat=%d, order=%d\n", __func__, k, order); + schp->buflen = blk_size; if (rem_sz > 0) /* must have failed */ return -ENOMEM; return 0; @@ -2083,12 +2098,12 @@ sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, static void sg_remove_scat(struct sg_fd *sfp, struct sg_scatter_hold *schp) { - SG_LOG(4, sfp, "%s: num_sgat=%d\n", __func__, schp->k_use_sg); + SG_LOG(4, sfp, "%s: num_sgat=%d\n", __func__, schp->num_sgat); if (schp->pages && schp->sglist_len > 0) { if (!schp->dio_in_use) { int k; - for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) { + for (k = 0; k < schp->num_sgat && schp->pages[k]; k++) { SG_LOG(5, sfp, "%s: pg[%d]=0x%p --\n", __func__, k, schp->pages[k]); __free_pages(schp->pages[k], schp->page_order); @@ -2115,7 +2130,7 @@ sg_rd_append(struct sg_request *srp, void __user *outp, int num_xfer) return 0; num = 1 << (PAGE_SHIFT + schp->page_order); - for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) { + for (k = 0; k < schp->num_sgat && schp->pages[k]; k++) { if (num > num_xfer) { if (copy_to_user(outp, page_address(schp->pages[k]), num_xfer)) @@ -2163,22 +2178,21 @@ sg_link_reserve(struct sg_fd *sfp, struct sg_request *srp, int size) rem = size; num = 1 << (PAGE_SHIFT + rsv_schp->page_order); - for (k = 0; k < rsv_schp->k_use_sg; k++) { + for (k = 0; k < rsv_schp->num_sgat; k++) { if (rem <= num) { - req_schp->k_use_sg = k + 1; + req_schp->num_sgat = k + 1; req_schp->sglist_len = rsv_schp->sglist_len; req_schp->pages = rsv_schp->pages; - req_schp->bufflen = size; + req_schp->buflen = size; req_schp->page_order = rsv_schp->page_order; break; } else rem -= num; } - if (k >= rsv_schp->k_use_sg) { + if (k >= rsv_schp->num_sgat) SG_LOG(1, sfp, "%s: BAD size\n", __func__); - } } static void @@ -2186,10 +2200,10 @@ sg_unlink_reserve(struct sg_fd *sfp, struct sg_request *srp) { struct sg_scatter_hold *req_schp = &srp->data; - SG_LOG(4, srp->parentfp, "%s: req->k_use_sg=%d\n", __func__, - (int)req_schp->k_use_sg); - req_schp->k_use_sg = 0; - req_schp->bufflen = 0; + SG_LOG(4, srp->parentfp, "%s: req->num_sgat=%d\n", __func__, + (int)req_schp->num_sgat); + req_schp->num_sgat = 0; + req_schp->buflen = 0; req_schp->pages = NULL; req_schp->page_order = 0; req_schp->sglist_len = 0; @@ -2200,7 +2214,7 @@ sg_unlink_reserve(struct sg_fd *sfp, struct sg_request *srp) /* always adds to end of list */ static struct sg_request * -sg_add_request(struct sg_fd *sfp) +sg_setup_req(struct sg_fd *sfp) { int k; unsigned long iflags; @@ -2279,7 +2293,7 @@ sg_add_sfp(struct sg_device *sdp) kfree(sfp); return ERR_PTR(-ENODEV); } - list_add_tail(&sfp->sfd_siblings, &sdp->sfds); + list_add_tail(&sfp->sfd_entry, &sdp->sfds); write_unlock_irqrestore(&sdp->sfd_lock, iflags); SG_LOG(3, sfp, "%s: sfp=0x%p\n", __func__, sfp); if (unlikely(sg_big_buff != def_reserved_size)) @@ -2288,8 +2302,8 @@ sg_add_sfp(struct sg_device *sdp) bufflen = min_t(int, sg_big_buff, max_sectors_bytes(sdp->device->request_queue)); sg_build_reserve(sfp, bufflen); - SG_LOG(3, sfp, "%s: bufflen=%d, k_use_sg=%d\n", __func__, - sfp->reserve.bufflen, sfp->reserve.k_use_sg); + SG_LOG(3, sfp, "%s: bufflen=%d, num_sgat=%d\n", __func__, + sfp->reserve.buflen, sfp->reserve.num_sgat); kref_get(&sdp->d_ref); __module_get(THIS_MODULE); @@ -2299,7 +2313,7 @@ sg_add_sfp(struct sg_device *sdp) static void sg_remove_sfp_usercontext(struct work_struct *work) { - struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work); + struct sg_fd *sfp = container_of(work, struct sg_fd, ew_fd.work); struct sg_device *sdp = sfp->parentdp; struct sg_request *srp; unsigned long iflags; @@ -2314,9 +2328,9 @@ sg_remove_sfp_usercontext(struct work_struct *work) } write_unlock_irqrestore(&sfp->rq_list_lock, iflags); - if (sfp->reserve.bufflen > 0) { - SG_LOG(6, sfp, "%s: bufflen=%d, k_use_sg=%d\n", __func__, - (int)sfp->reserve.bufflen, (int)sfp->reserve.k_use_sg); + if (sfp->reserve.buflen > 0) { + SG_LOG(6, sfp, "%s: buflen=%d, num_sgat=%d\n", __func__, + (int)sfp->reserve.buflen, (int)sfp->reserve.num_sgat); sg_remove_scat(sfp, &sfp->reserve); } @@ -2336,11 +2350,11 @@ sg_remove_sfp(struct kref *kref) unsigned long iflags; write_lock_irqsave(&sdp->sfd_lock, iflags); - list_del(&sfp->sfd_siblings); + list_del(&sfp->sfd_entry); write_unlock_irqrestore(&sdp->sfd_lock, iflags); - INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext); - schedule_work(&sfp->ew.work); + INIT_WORK(&sfp->ew_fd.work, sg_remove_sfp_usercontext); + schedule_work(&sfp->ew_fd.work); } static int @@ -2627,19 +2641,19 @@ sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) int k, new_interface, blen, usg; struct sg_request *srp; struct sg_fd *fp; - const sg_io_hdr_t *hp; + const struct sg_io_hdr *hp; const char * cp; unsigned int ms; k = 0; - list_for_each_entry(fp, &sdp->sfds, sfd_siblings) { + list_for_each_entry(fp, &sdp->sfds, sfd_entry) { k++; read_lock(&fp->rq_list_lock); /* irqs already disabled */ - seq_printf(s, " FD(%d): timeout=%dms bufflen=%d " + seq_printf(s, " FD(%d): timeout=%dms buflen=%d " "(res)sgat=%d low_dma=%d\n", k, jiffies_to_msecs(fp->timeout), - fp->reserve.bufflen, - (int) fp->reserve.k_use_sg, + fp->reserve.buflen, + (int)fp->reserve.num_sgat, (int) sdp->device->host->unchecked_isa_dma); seq_printf(s, " cmd_q=%d f_packid=%d k_orphan=%d closed=0\n", (int) fp->cmd_q, (int) fp->force_packid, @@ -2663,8 +2677,8 @@ sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) cp = " "; } seq_puts(s, cp); - blen = srp->data.bufflen; - usg = srp->data.k_use_sg; + blen = srp->data.buflen; + usg = srp->data.num_sgat; seq_puts(s, srp->done ? ((1 == srp->done) ? "rcv:" : "fin:") : "act:"); @@ -2717,8 +2731,8 @@ sg_proc_seq_show_debug(struct seq_file *s, void *v) scsidp->lun, scsidp->host->hostt->emulated); } - seq_printf(s, " max_sgat_elems=%d excl=%d open_cnt=%d\n", - sdp->max_sgat_elems, SG_HAVE_EXCLUDE(sdp), + seq_printf(s, " max_sgat_sz=%d excl=%d open_cnt=%d\n", + sdp->max_sgat_sz, SG_HAVE_EXCLUDE(sdp), atomic_read(&sdp->open_cnt)); sg_proc_debug_helper(s, sdp); } From patchwork Thu Oct 15 02:06:10 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269808 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 83855C41604 for ; Thu, 15 Oct 2020 02:07:07 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 43E5622258 for ; Thu, 15 Oct 2020 02:07:07 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729244AbgJOCHG (ORCPT ); Wed, 14 Oct 2020 22:07:06 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40115 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726111AbgJOCHF (ORCPT ); Wed, 14 Oct 2020 22:07:05 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 48EC720425B; Thu, 15 Oct 2020 04:07:03 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id ISoqQOKSwqDd; Thu, 15 Oct 2020 04:07:01 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 144C920423F; Thu, 15 Oct 2020 04:06:56 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 11/44] sg: change rwlock to spinlock Date: Wed, 14 Oct 2020 22:06:10 -0400 Message-Id: <20201015020643.432908-12-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org A reviewer suggested that the extra overhead associated with a rw lock compared to a spinlock was not worth it for short, oft-used critcal sections. So the rwlock on the request list/array is changed to a spinlock. The head of that list is in the owning sf file descriptor object. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 52 +++++++++++++++++++++++------------------------ 1 file changed, 26 insertions(+), 26 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 7aae7ce6cebf..fd1e278d5b9e 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -139,7 +139,7 @@ struct sg_fd { /* holds the state of a file descriptor */ struct list_head sfd_entry; /* member sg_device::sfds list */ struct sg_device *parentdp; /* owning device */ wait_queue_head_t read_wait; /* queue read until command done */ - rwlock_t rq_list_lock; /* protect access to list in req_arr */ + spinlock_t rq_list_lock; /* protect access to list in req_arr */ struct mutex f_mutex; /* protect against changes in this fd */ int timeout; /* defaults to SG_DEFAULT_TIMEOUT */ int timeout_user; /* defaults to SG_DEFAULT_TIMEOUT_USER */ @@ -742,17 +742,17 @@ sg_get_rq_mark(struct sg_fd *sfp, int pack_id) struct sg_request *resp; unsigned long iflags; - write_lock_irqsave(&sfp->rq_list_lock, iflags); + spin_lock_irqsave(&sfp->rq_list_lock, iflags); list_for_each_entry(resp, &sfp->rq_list, entry) { /* look for requests that are ready + not SG_IO owned */ if (resp->done == 1 && !resp->sg_io_owned && (-1 == pack_id || resp->header.pack_id == pack_id)) { resp->done = 2; /* guard against other readers */ - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return resp; } } - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return NULL; } @@ -806,9 +806,9 @@ srp_done(struct sg_fd *sfp, struct sg_request *srp) unsigned long flags; int ret; - read_lock_irqsave(&sfp->rq_list_lock, flags); + spin_lock_irqsave(&sfp->rq_list_lock, flags); ret = srp->done; - read_unlock_irqrestore(&sfp->rq_list_lock, flags); + spin_unlock_irqrestore(&sfp->rq_list_lock, flags); return ret; } @@ -1072,15 +1072,15 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, (srp_done(sfp, srp) || SG_IS_DETACHING(sdp))); if (SG_IS_DETACHING(sdp)) return -ENODEV; - write_lock_irq(&sfp->rq_list_lock); + spin_lock_irq(&sfp->rq_list_lock); if (srp->done) { srp->done = 2; - write_unlock_irq(&sfp->rq_list_lock); + spin_unlock_irq(&sfp->rq_list_lock); result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp); return (result < 0) ? result : 0; } srp->orphan = 1; - write_unlock_irq(&sfp->rq_list_lock); + spin_unlock_irq(&sfp->rq_list_lock); return result; /* -ERESTARTSYS because signal hit process */ case SG_SET_TIMEOUT: result = get_user(val, ip); @@ -1132,15 +1132,15 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, sfp->force_packid = val ? 1 : 0; return 0; case SG_GET_PACK_ID: - read_lock_irqsave(&sfp->rq_list_lock, iflags); + spin_lock_irqsave(&sfp->rq_list_lock, iflags); list_for_each_entry(srp, &sfp->rq_list, entry) { if ((1 == srp->done) && (!srp->sg_io_owned)) { - read_unlock_irqrestore(&sfp->rq_list_lock, + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return put_user(srp->header.pack_id, ip); } } - read_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return put_user(-1, ip); case SG_GET_NUM_WAITING: return put_user(atomic_read(&sfp->waiting), ip); @@ -1209,9 +1209,9 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, GFP_KERNEL); if (!rinfo) return -ENOMEM; - read_lock_irqsave(&sfp->rq_list_lock, iflags); + spin_lock_irqsave(&sfp->rq_list_lock, iflags); sg_fill_request_table(sfp, rinfo); - read_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); #ifdef CONFIG_COMPAT if (in_compat_syscall()) result = put_compat_request_table(p, rinfo); @@ -1531,7 +1531,7 @@ sg_rq_end_io(struct request *rq, blk_status_t status) scsi_req_free_cmd(scsi_req(rq)); blk_put_request(rq); - write_lock_irqsave(&sfp->rq_list_lock, iflags); + spin_lock_irqsave(&sfp->rq_list_lock, iflags); if (unlikely(srp->orphan)) { if (sfp->keep_orphan) srp->sg_io_owned = 0; @@ -1539,7 +1539,7 @@ sg_rq_end_io(struct request *rq, blk_status_t status) done = 0; } srp->done = done; - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); if (likely(done)) { /* Now wake up any sg_read() that is waiting for this @@ -2220,7 +2220,7 @@ sg_setup_req(struct sg_fd *sfp) unsigned long iflags; struct sg_request *rp = sfp->req_arr; - write_lock_irqsave(&sfp->rq_list_lock, iflags); + spin_lock_irqsave(&sfp->rq_list_lock, iflags); if (!list_empty(&sfp->rq_list)) { if (!sfp->cmd_q) goto out_unlock; @@ -2236,10 +2236,10 @@ sg_setup_req(struct sg_fd *sfp) rp->parentfp = sfp; rp->header.duration = jiffies_to_msecs(jiffies); list_add_tail(&rp->entry, &sfp->rq_list); - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return rp; out_unlock: - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return NULL; } @@ -2252,13 +2252,13 @@ sg_remove_request(struct sg_fd *sfp, struct sg_request *srp) if (!sfp || !srp || list_empty(&sfp->rq_list)) return res; - write_lock_irqsave(&sfp->rq_list_lock, iflags); + spin_lock_irqsave(&sfp->rq_list_lock, iflags); if (!list_empty(&srp->entry)) { list_del(&srp->entry); srp->parentfp = NULL; res = 1; } - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return res; } @@ -2274,7 +2274,7 @@ sg_add_sfp(struct sg_device *sdp) return ERR_PTR(-ENOMEM); init_waitqueue_head(&sfp->read_wait); - rwlock_init(&sfp->rq_list_lock); + spin_lock_init(&sfp->rq_list_lock); INIT_LIST_HEAD(&sfp->rq_list); kref_init(&sfp->f_ref); mutex_init(&sfp->f_mutex); @@ -2319,14 +2319,14 @@ sg_remove_sfp_usercontext(struct work_struct *work) unsigned long iflags; /* Cleanup any responses which were never read(). */ - write_lock_irqsave(&sfp->rq_list_lock, iflags); + spin_lock_irqsave(&sfp->rq_list_lock, iflags); while (!list_empty(&sfp->rq_list)) { srp = list_first_entry(&sfp->rq_list, struct sg_request, entry); sg_finish_scsi_blk_rq(srp); list_del(&srp->entry); srp->parentfp = NULL; } - write_unlock_irqrestore(&sfp->rq_list_lock, iflags); + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); if (sfp->reserve.buflen > 0) { SG_LOG(6, sfp, "%s: buflen=%d, num_sgat=%d\n", __func__, @@ -2648,7 +2648,7 @@ sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) k = 0; list_for_each_entry(fp, &sdp->sfds, sfd_entry) { k++; - read_lock(&fp->rq_list_lock); /* irqs already disabled */ + spin_lock(&fp->rq_list_lock); /* irqs already disabled */ seq_printf(s, " FD(%d): timeout=%dms buflen=%d " "(res)sgat=%d low_dma=%d\n", k, jiffies_to_msecs(fp->timeout), @@ -2698,7 +2698,7 @@ sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) } if (list_empty(&fp->rq_list)) seq_puts(s, " No requests active\n"); - read_unlock(&fp->rq_list_lock); + spin_unlock(&fp->rq_list_lock); } } From patchwork Thu Oct 15 02:06:13 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269805 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 2288FC4363A for ; Thu, 15 Oct 2020 02:07:15 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id DBA1222257 for ; Thu, 15 Oct 2020 02:07:14 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1727833AbgJOCHO (ORCPT ); Wed, 14 Oct 2020 22:07:14 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40109 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1729251AbgJOCHI (ORCPT ); Wed, 14 Oct 2020 22:07:08 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id C985920425B; Thu, 15 Oct 2020 04:07:06 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id fowT4rWhP8I9; Thu, 15 Oct 2020 04:07:05 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 70C0E20425A; Thu, 15 Oct 2020 04:07:00 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 14/44] sg: sg_common_write add structure for arguments Date: Wed, 14 Oct 2020 22:06:13 -0400 Message-Id: <20201015020643.432908-15-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org As the number of arguments to sg_common_write() starts to grow (more in later patches) add a structure to hold most of these arguments. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 47 ++++++++++++++++++++++++++++++++--------------- 1 file changed, 32 insertions(+), 15 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index dcc35145c079..eeaf246b8754 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -174,6 +174,13 @@ struct sg_device { /* holds the state of each scsi generic device */ struct kref d_ref; }; +struct sg_comm_wr_t { /* arguments to sg_common_write() */ + int timeout; + int blocking; + struct sg_request *srp; + u8 *cmnd; +}; + /* tasklet or soft irq callback */ static void sg_rq_end_io(struct request *rq, blk_status_t status); /* Declarations of other static functions used before they are defined */ @@ -186,8 +193,7 @@ static ssize_t sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, size_t count, bool blocking, bool read_only, bool sg_io_owned, struct sg_request **o_srp); -static int sg_common_write(struct sg_fd *sfp, struct sg_request *srp, - u8 *cmnd, int timeout, int blocking); +static int sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwp); static int sg_read_append(struct sg_request *srp, void __user *outp, int num_xfer); static void sg_remove_scat(struct sg_fd *sfp, struct sg_scatter_hold *schp); @@ -487,6 +493,7 @@ sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) struct sg_io_hdr v3hdr; struct sg_header *ohp = &ov2hdr; struct sg_io_hdr *h3p = &v3hdr; + struct sg_comm_wr_t cwr; u8 cmnd[SG_MAX_CDB_SIZE]; res = sg_check_file_access(filp, __func__); @@ -590,7 +597,11 @@ sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) input_size, (unsigned int) cmnd[0], current->comm); } - res = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking); + cwr.timeout = sfp->timeout; + cwr.blocking = blocking; + cwr.srp = srp; + cwr.cmnd = cmnd; + res = sg_common_write(sfp, &cwr); return (res < 0) ? res : count; } @@ -613,6 +624,7 @@ sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, int k; struct sg_request *srp; struct sg_io_hdr *hp; + struct sg_comm_wr_t cwr; u8 cmnd[SG_MAX_CDB_SIZE]; int timeout; unsigned long ul_timeout; @@ -663,23 +675,28 @@ sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, sg_remove_request(sfp, srp); return -EPERM; } - k = sg_common_write(sfp, srp, cmnd, timeout, blocking); + cwr.timeout = timeout; + cwr.blocking = blocking; + cwr.srp = srp; + cwr.cmnd = cmnd; + k = sg_common_write(sfp, &cwr); if (k < 0) return k; if (o_srp) - *o_srp = srp; + *o_srp = cwr.srp; return count; } static int -sg_common_write(struct sg_fd *sfp, struct sg_request *srp, - u8 *cmnd, int timeout, int blocking) +sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) { - int k, at_head; + bool at_head; + int k; struct sg_device *sdp = sfp->parentdp; + struct sg_request *srp = cwrp->srp; struct sg_io_hdr *hp = &srp->header; - srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */ + srp->data.cmd_opcode = cwrp->cmnd[0]; /* hold opcode of command */ hp->status = 0; hp->masked_status = 0; hp->msg_status = 0; @@ -688,14 +705,14 @@ sg_common_write(struct sg_fd *sfp, struct sg_request *srp, hp->driver_status = 0; hp->resid = 0; SG_LOG(4, sfp, "%s: opcode=0x%02x, cmd_sz=%d\n", __func__, - (int)cmnd[0], hp->cmd_len); + (int)cwrp->cmnd[0], hp->cmd_len); if (hp->dxfer_len >= SZ_256M) { sg_remove_request(sfp, srp); return -EINVAL; } - k = sg_start_req(srp, cmnd); + k = sg_start_req(srp, cwrp->cmnd); if (k) { SG_LOG(1, sfp, "%s: start_req err=%d\n", __func__, k); sg_finish_scsi_blk_rq(srp); @@ -717,13 +734,13 @@ sg_common_write(struct sg_fd *sfp, struct sg_request *srp, hp->duration = jiffies_to_msecs(jiffies); if (hp->interface_id != '\0' && /* v3 (or later) interface */ (SG_FLAG_Q_AT_TAIL & hp->flags)) - at_head = 0; + at_head = false; else - at_head = 1; + at_head = true; - if (!blocking) + if (!srp->sg_io_owned) atomic_inc(&sfp->submitted); - srp->rq->timeout = timeout; + srp->rq->timeout = cwrp->timeout; kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */ blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk, srp->rq, at_head, sg_rq_end_io); From patchwork Thu Oct 15 02:06:14 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269806 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id A7372C43457 for ; Thu, 15 Oct 2020 02:07:14 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 5FF3C22258 for ; Thu, 15 Oct 2020 02:07:14 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729770AbgJOCHN (ORCPT ); Wed, 14 Oct 2020 22:07:13 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40149 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1729351AbgJOCHJ (ORCPT ); Wed, 14 Oct 2020 22:07:09 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 6B77D204259; Thu, 15 Oct 2020 04:07:08 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id DIJV81XKAuDh; Thu, 15 Oct 2020 04:07:06 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 901DC204247; Thu, 15 Oct 2020 04:07:01 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 15/44] sg: rework sg_vma_fault Date: Wed, 14 Oct 2020 22:06:14 -0400 Message-Id: <20201015020643.432908-16-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Simple refactoring of the sg_vma_fault() function. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 33 +++++++++++++++++++++++---------- 1 file changed, 23 insertions(+), 10 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index eeaf246b8754..adfd4abe312f 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -1429,14 +1429,16 @@ sg_fasync(int fd, struct file *filp, int mode) return fasync_helper(fd, filp, mode, &sfp->async_qp); } +/* Note: the error return: VM_FAULT_SIGBUS causes a "bus error" */ static vm_fault_t sg_vma_fault(struct vm_fault *vmf) { - struct vm_area_struct *vma = vmf->vma; - struct sg_fd *sfp; + int k, length; unsigned long offset, len, sa; + struct vm_area_struct *vma = vmf->vma; struct sg_scatter_hold *rsv_schp; - int k, length; + struct sg_device *sdp; + struct sg_fd *sfp; const char *nbp = "==NULL, bad"; if (!vma) { @@ -1448,20 +1450,31 @@ sg_vma_fault(struct vm_fault *vmf) pr_warn("%s: sfp%s\n", __func__, nbp); goto out_err; } + sdp = sfp->parentdp; + if (sdp && unlikely(SG_IS_DETACHING(sdp))) { + SG_LOG(1, sfp, "%s: device detaching\n", __func__); + goto out_err; + } rsv_schp = &sfp->reserve; offset = vmf->pgoff << PAGE_SHIFT; - if (offset >= rsv_schp->buflen) - return VM_FAULT_SIGBUS; + if (offset >= (unsigned int)rsv_schp->buflen) { + SG_LOG(1, sfp, "%s: offset[%lu] >= rsv.buflen\n", __func__, + offset); + goto out_err; + } sa = vma->vm_start; SG_LOG(3, sfp, "%s: vm_start=0x%lx, off=%lu\n", __func__, sa, offset); length = 1 << (PAGE_SHIFT + rsv_schp->page_order); - for (k = 0; k < rsv_schp->num_sgat && sa < vma->vm_end; k++) { + for (k = 0; k < rsv_schp->num_sgat && sa < vma->vm_end; ++k) { len = vma->vm_end - sa; - len = (len < length) ? len : length; + len = min_t(int, len, (int)length); if (offset < len) { - struct page *page = nth_page(rsv_schp->pages[k], - offset >> PAGE_SHIFT); - get_page(page); /* increment page count */ + struct page *page; + struct page *pp; + + pp = rsv_schp->pages[k]; + page = nth_page(pp, offset >> PAGE_SHIFT); + get_page(page); /* increment page count */ vmf->page = page; return 0; /* success */ } From patchwork Thu Oct 15 02:06:17 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269800 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id B84BCC2D0A3 for ; Thu, 15 Oct 2020 02:07:27 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 6F4D422258 for ; Thu, 15 Oct 2020 02:07:27 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1732557AbgJOCH1 (ORCPT ); Wed, 14 Oct 2020 22:07:27 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40163 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1729585AbgJOCHN (ORCPT ); Wed, 14 Oct 2020 22:07:13 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 4F95E20423F; Thu, 15 Oct 2020 04:07:11 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 4DZ87H8j5eUq; Thu, 15 Oct 2020 04:07:08 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id EDD5B20426C; Thu, 15 Oct 2020 04:07:04 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 18/44] sg: rework scatter gather handling Date: Wed, 14 Oct 2020 22:06:17 -0400 Message-Id: <20201015020643.432908-19-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Rename sg_build_indirect() to sg_mk_sgat() and sg_remove_scat() to sg_remove_sgat(). Re-implement those functions. Add sg_calc_sgat_param() to calculate various scatter gather list parameters. Some other minor clean-ups. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 288 +++++++++++++++++++++++++--------------------- 1 file changed, 160 insertions(+), 128 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index cd1ae1e38acb..6f7093baa3ff 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -90,7 +90,6 @@ static int def_reserved_size = -1; /* picks up init parameter */ static int sg_allow_dio = SG_ALLOW_DIO_DEF; static int scatter_elem_sz = SG_SCATTER_SZ; -static int scatter_elem_sz_prev = SG_SCATTER_SZ; #define SG_DEF_SECTOR_SZ 512 @@ -145,6 +144,7 @@ struct sg_fd { /* holds the state of a file descriptor */ int timeout_user; /* defaults to SG_DEFAULT_TIMEOUT_USER */ atomic_t submitted; /* number inflight or awaiting read */ atomic_t waiting; /* number of requests awaiting read */ + int sgat_elem_sz; /* initialized to scatter_elem_sz */ struct sg_scatter_hold reserve; /* buffer for this file descriptor */ struct list_head rq_list; /* head of request list */ struct fasync_struct *async_qp; /* used by asynchronous notification */ @@ -165,6 +165,7 @@ struct sg_device { /* holds the state of each scsi generic device */ struct mutex open_rel_lock; /* held when in open() or release() */ struct list_head sfds; rwlock_t sfd_lock; /* protect access to sfd list */ + int max_sgat_elems; /* adapter's max number of elements in sgat */ int max_sgat_sz; /* max number of bytes in sgat list */ u32 index; /* device index number */ atomic_t open_cnt; /* count of opens (perhaps < num(sfds) ) */ @@ -187,8 +188,8 @@ static void sg_rq_end_io(struct request *rq, blk_status_t status); static int sg_proc_init(void); static int sg_start_req(struct sg_request *srp, u8 *cmd); static void sg_finish_scsi_blk_rq(struct sg_request *srp); -static int sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, - int buff_size); +static int sg_mk_sgat(struct sg_scatter_hold *schp, struct sg_fd *sfp, + int minlen); static ssize_t sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, size_t count, bool blocking, bool read_only, bool sg_io_owned, @@ -196,7 +197,7 @@ static ssize_t sg_submit(struct sg_fd *sfp, struct file *filp, static int sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwp); static int sg_read_append(struct sg_request *srp, void __user *outp, int num_xfer); -static void sg_remove_scat(struct sg_fd *sfp, struct sg_scatter_hold *schp); +static void sg_remove_sgat(struct sg_fd *sfp, struct sg_scatter_hold *schp); static void sg_build_reserve(struct sg_fd *sfp, int req_size); static void sg_link_reserve(struct sg_fd *sfp, struct sg_request *srp, int size); @@ -207,6 +208,7 @@ static struct sg_request *sg_setup_req(struct sg_fd *sfp); static int sg_deact_request(struct sg_fd *sfp, struct sg_request *srp); static struct sg_device *sg_get_dev(int dev); static void sg_device_destroy(struct kref *kref); +static void sg_calc_sgat_param(struct sg_device *sdp); #define SZ_SG_HEADER ((int)sizeof(struct sg_header)) /* v1 and v2 header */ #define SZ_SG_IO_HDR ((int)sizeof(struct sg_io_hdr)) /* v3 header */ @@ -352,7 +354,6 @@ sg_open(struct inode *inode, struct file *filp) int min_dev = iminor(inode); int op_flags = filp->f_flags; int res; - struct request_queue *q; struct sg_device *sdp; struct sg_fd *sfp; @@ -411,16 +412,12 @@ sg_open(struct inode *inode, struct file *filp) if (o_excl) set_bit(SG_FDEV_EXCLUDE, sdp->fdev_bm); - if (atomic_read(&sdp->open_cnt) < 1) { /* no existing opens */ - clear_bit(SG_FDEV_LOG_SENSE, sdp->fdev_bm); - q = sdp->device->request_queue; - sdp->max_sgat_sz = queue_max_segments(q); - } + if (atomic_read(&sdp->open_cnt) < 1) /* no existing opens */ + sg_calc_sgat_param(sdp); sfp = sg_add_sfp(sdp); /* increments sdp->d_ref */ if (IS_ERR(sfp)) { res = PTR_ERR(sfp); - goto out_undo; - } + goto out_undo; } filp->private_data = sfp; atomic_inc(&sdp->open_cnt); @@ -997,10 +994,43 @@ max_sectors_bytes(struct request_queue *q) unsigned int max_sectors = queue_max_sectors(q); max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9); - return max_sectors << 9; } +/* + * Calculates sg_device::max_sgat_elems and sg_device::max_sgat_sz. It uses + * the device's request queue. If q not available sets max_sgat_elems to 1 + * and max_sgat_sz to PAGE_SIZE. If potential max_sgat_sz is greater than + * 2^30 scales down the implied max_segment_size so the product of the + * max_segment_size and max_sgat_elems is less than or equal to 2^30 . + */ +static void +sg_calc_sgat_param(struct sg_device *sdp) +{ + int sz; + u64 m; + struct scsi_device *sdev = sdp->device; + struct request_queue *q = sdev ? sdev->request_queue : NULL; + + clear_bit(SG_FDEV_LOG_SENSE, sdp->fdev_bm); + if (!q) { + sdp->max_sgat_elems = 1; + sdp->max_sgat_sz = PAGE_SIZE; + return; + } + sdp->max_sgat_elems = queue_max_segments(q); + m = (u64)queue_max_segment_size(q) * queue_max_segments(q); + if (m < PAGE_SIZE) { + sdp->max_sgat_elems = 1; + sdp->max_sgat_sz = PAGE_SIZE; + return; + } + sz = (int)min_t(u64, m, 1 << 30); + if (sz == (1 << 30)) /* round down so: sz = elems * elem_sz */ + sz = ((1 << 30) / sdp->max_sgat_elems) * sdp->max_sgat_elems; + sdp->max_sgat_sz = sz; +} + static void sg_fill_request_table(struct sg_fd *sfp, struct sg_req_info *rinfo) { @@ -1066,7 +1096,7 @@ sg_ctl_sg_io(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, } srp->orphan = 1; spin_unlock_irq(&sfp->rq_list_lock); - return res; /* -ERESTARTSYS because signal hit process */ + return res; } static int @@ -1077,8 +1107,7 @@ sg_set_reserved_sz(struct sg_fd *sfp, int want_rsv_sz) sfp->res_in_use) { return -EBUSY; } - - sg_remove_scat(sfp, &sfp->reserve); + sg_remove_sgat(sfp, &sfp->reserve); sg_build_reserve(sfp, want_rsv_sz); } return 0; @@ -1547,8 +1576,18 @@ sg_rq_end_io_usercontext(struct work_struct *work) { struct sg_request *srp = container_of(work, struct sg_request, ew_orph.work); - struct sg_fd *sfp = srp->parentfp; + struct sg_fd *sfp; + if (!srp) { + WARN_ONCE("%s: srp unexpectedly NULL\n", __func__); + return; + } + sfp = srp->parentfp; + if (!sfp) { + WARN_ONCE(1, "%s: sfp unexpectedly NULL\n", __func__); + return; + } + SG_LOG(3, sfp, "%s: srp=0x%p\n", __func__, srp); sg_finish_scsi_blk_rq(srp); sg_deact_request(sfp, srp); kref_put(&sfp->f_ref, sg_remove_sfp); @@ -1680,7 +1719,6 @@ static bool sg_sysfs_valid; static struct sg_device * sg_add_device_helper(struct gendisk *disk, struct scsi_device *scsidp) { - struct request_queue *q = scsidp->request_queue; struct sg_device *sdp; unsigned long iflags; int error; @@ -1720,7 +1758,7 @@ sg_add_device_helper(struct gendisk *disk, struct scsi_device *scsidp) init_waitqueue_head(&sdp->open_wait); clear_bit(SG_FDEV_DETACHING, sdp->fdev_bm); rwlock_init(&sdp->sfd_lock); - sdp->max_sgat_sz = queue_max_segments(q); + sg_calc_sgat_param(sdp); sdp->index = k; kref_init(&sdp->d_ref); error = 0; @@ -1890,24 +1928,24 @@ init_sg(void) { int rc; - if (scatter_elem_sz < PAGE_SIZE) { + if (scatter_elem_sz < (int)PAGE_SIZE) scatter_elem_sz = PAGE_SIZE; - scatter_elem_sz_prev = scatter_elem_sz; - } + else if (!is_power_of_2(scatter_elem_sz)) + scatter_elem_sz = roundup_pow_of_two(scatter_elem_sz); if (def_reserved_size >= 0) sg_big_buff = def_reserved_size; else def_reserved_size = sg_big_buff; - rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), + rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS, "sg"); if (rc) return rc; sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic"); if ( IS_ERR(sg_sysfs_class) ) { rc = PTR_ERR(sg_sysfs_class); - goto err_out; - } + goto err_out_unreg; + } sg_sysfs_valid = true; rc = scsi_register_interface(&sg_interface); if (0 == rc) { @@ -1915,7 +1953,7 @@ init_sg(void) return 0; } class_destroy(sg_sysfs_class); -err_out: +err_out_unreg: unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS); return rc; } @@ -2020,7 +2058,7 @@ sg_start_req(struct sg_request *srp, u8 *cmd) mutex_unlock(&sfp->f_mutex); return res; } else { - res = sg_build_indirect(req_schp, sfp, dxfer_len); + res = sg_mk_sgat(req_schp, sfp, dxfer_len); if (res) { mutex_unlock(&sfp->f_mutex); return res; @@ -2107,117 +2145,104 @@ sg_finish_scsi_blk_rq(struct sg_request *srp) if (srp->res_used) sg_unlink_reserve(sfp, srp); else - sg_remove_scat(sfp, req_schp); -} - -static int -sg_build_sgat(struct sg_scatter_hold *schp, const struct sg_fd *sfp, - int tablesize) -{ - int sg_buflen = tablesize * sizeof(struct page *); - gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN; - - schp->pages = kzalloc(sg_buflen, gfp_flags); - if (!schp->pages) - return -ENOMEM; - schp->sglist_len = sg_buflen; - return tablesize; /* number of scat_gath elements allocated */ + sg_remove_sgat(sfp, req_schp); } static int -sg_build_indirect(struct sg_scatter_hold *schp, struct sg_fd *sfp, - int buff_size) -{ - int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems; - int max_sgat_sz = sfp->parentdp->max_sgat_sz; - int blk_size = buff_size, order; - gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO; +sg_mk_sgat(struct sg_scatter_hold *schp, struct sg_fd *sfp, int minlen) +{ + int j, k, rem_sz, align_sz; + int mx_sgat_elems = sfp->parentdp->max_sgat_elems; + unsigned int elem_sz, order, o_order; + const size_t ptr_sz = sizeof(struct page *); + gfp_t mask_ap = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO; + gfp_t mask_kz = GFP_ATOMIC | __GFP_NOWARN; struct sg_device *sdp = sfp->parentdp; - if (blk_size < 0) - return -EFAULT; - if (0 == blk_size) - ++blk_size; /* don't know why */ - /* round request up to next highest SG_DEF_SECTOR_SZ byte boundary */ - blk_size = ALIGN(blk_size, SG_DEF_SECTOR_SZ); - SG_LOG(4, sfp, "%s: buff_size=%d, blk_size=%d\n", __func__, buff_size, - blk_size); - - /* N.B. ret_sz carried into this block ... */ - mx_sc_elems = sg_build_sgat(schp, sfp, max_sgat_sz); - if (mx_sc_elems < 0) - return mx_sc_elems; /* most likely -ENOMEM */ - - num = scatter_elem_sz; - if (unlikely(num != scatter_elem_sz_prev)) { - if (num < PAGE_SIZE) { - scatter_elem_sz = PAGE_SIZE; - scatter_elem_sz_prev = PAGE_SIZE; - } else - scatter_elem_sz_prev = num; + if (unlikely(minlen <= 0)) { + if (minlen < 0) + return -EFAULT; + ++minlen; /* don't remember why */ } + /* round request up to next highest SG_DEF_SECTOR_SZ byte boundary */ + align_sz = ALIGN(minlen, SG_DEF_SECTOR_SZ); - if (sdp->device->host->unchecked_isa_dma) - gfp_mask |= GFP_DMA; - - order = get_order(num); -retry: - ret_sz = 1 << (PAGE_SHIFT + order); - - for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems; - k++, rem_sz -= ret_sz) { + schp->pages = kcalloc(mx_sgat_elems, ptr_sz, mask_kz); + SG_LOG(4, sfp, "%s: minlen=%d, align_sz=%d [sz=%zu, 0x%p ++]\n", + __func__, minlen, align_sz, mx_sgat_elems * ptr_sz, + schp->pages); + if (unlikely(!schp->pages)) + return -ENOMEM; - num = (rem_sz > scatter_elem_sz_prev) ? - scatter_elem_sz_prev : rem_sz; + elem_sz = sfp->sgat_elem_sz; /* power of 2 and >= PAGE_SIZE */ + if (sdp && unlikely(sdp->device->host->unchecked_isa_dma)) + mask_ap |= GFP_DMA; + o_order = get_order(elem_sz); + order = o_order; - schp->pages[k] = alloc_pages(gfp_mask, order); +again: + for (k = 0, rem_sz = align_sz; rem_sz > 0 && k < mx_sgat_elems; + ++k, rem_sz -= elem_sz) { + schp->pages[k] = alloc_pages(mask_ap, order); if (!schp->pages[k]) - goto out; - - if (num == scatter_elem_sz_prev) { - if (unlikely(ret_sz > scatter_elem_sz_prev)) { - scatter_elem_sz = ret_sz; - scatter_elem_sz_prev = ret_sz; - } - } - SG_LOG(5, sfp, "%s: k=%d, num=%d, ret_sz=%d\n", __func__, k, - num, ret_sz); - } /* end of for loop */ - + goto err_out; + SG_LOG(5, sfp, "%s: k=%d, order=%d [0x%p ++]\n", __func__, k, + order, schp->pages[k]); + } schp->page_order = order; schp->num_sgat = k; - SG_LOG(5, sfp, "%s: num_sgat=%d, order=%d\n", __func__, k, order); - schp->buflen = blk_size; - if (rem_sz > 0) /* must have failed */ - return -ENOMEM; + SG_LOG(((order != o_order || rem_sz > 0) ? 2 : 5), sfp, + "%s: num_sgat=%d, order=%d,%d\n", __func__, k, o_order, order); + if (unlikely(rem_sz > 0)) { /* hit mx_sgat_elems */ + order = 0; /* force exit */ + goto err_out; + } + schp->buflen = align_sz; return 0; -out: - for (i = 0; i < k; i++) - __free_pages(schp->pages[i], order); - - if (--order >= 0) - goto retry; +err_out: + for (j = 0; j < k; ++j) + __free_pages(schp->pages[j], order); + if (--order >= 0) { + elem_sz >>= 1; + goto again; + } + kfree(schp->pages); + schp->pages = NULL; return -ENOMEM; } static void -sg_remove_scat(struct sg_fd *sfp, struct sg_scatter_hold *schp) +sg_remove_sgat_helper(struct sg_fd *sfp, struct sg_scatter_hold *schp) { - SG_LOG(4, sfp, "%s: num_sgat=%d\n", __func__, schp->num_sgat); - if (schp->pages && schp->sglist_len > 0) { - if (!schp->dio_in_use) { - int k; + int k; + void *p; - for (k = 0; k < schp->num_sgat && schp->pages[k]; k++) { - SG_LOG(5, sfp, "%s: pg[%d]=0x%p --\n", - __func__, k, schp->pages[k]); - __free_pages(schp->pages[k], schp->page_order); - } - kfree(schp->pages); - } + if (!schp->pages) + return; + for (k = 0; k < schp->num_sgat; ++k) { + p = schp->pages[k]; + SG_LOG(5, sfp, "%s: pg[%d]=0x%p --\n", __func__, k, p); + if (unlikely(!p)) + continue; + __free_pages(p, schp->page_order); } - memset(schp, 0, sizeof (*schp)); + SG_LOG(5, sfp, "%s: pg_order=%u, free pgs=0x%p --\n", __func__, + schp->page_order, schp->pages); + kfree(schp->pages); +} + +/* Remove the data (possibly a sgat list) held by srp, not srp itself */ +static void +sg_remove_sgat(struct sg_fd *sfp, struct sg_scatter_hold *schp) +{ + SG_LOG(4, sfp, "%s: num_sgat=%d%s\n", __func__, schp->num_sgat, + ((sfp ? (&sfp->reserve == schp) : false) ? + " [rsv]" : "")); + if (!schp->dio_in_use) + sg_remove_sgat_helper(sfp, schp); + + memset(schp, 0, sizeof(*schp)); /* zeros buflen and dlen */ } /* @@ -2239,12 +2264,12 @@ sg_read_append(struct sg_request *srp, void __user *outp, int num_xfer) for (k = 0; k < schp->num_sgat && schp->pages[k]; k++) { if (num > num_xfer) { if (copy_to_user(outp, page_address(schp->pages[k]), - num_xfer)) + num_xfer)) return -EFAULT; break; } else { if (copy_to_user(outp, page_address(schp->pages[k]), - num)) + num)) return -EFAULT; num_xfer -= num; if (num_xfer <= 0) @@ -2264,10 +2289,10 @@ sg_build_reserve(struct sg_fd *sfp, int req_size) do { if (req_size < PAGE_SIZE) req_size = PAGE_SIZE; - if (0 == sg_build_indirect(schp, sfp, req_size)) + if (sg_mk_sgat(schp, sfp, req_size) == 0) return; else - sg_remove_scat(sfp, schp); + sg_remove_sgat(sfp, schp); req_size >>= 1; /* divide by 2 */ } while (req_size > (PAGE_SIZE / 2)); } @@ -2371,8 +2396,8 @@ sg_deact_request(struct sg_fd *sfp, struct sg_request *srp) static struct sg_fd * sg_add_sfp(struct sg_device *sdp) { + int rbuf_len; unsigned long iflags; - int bufflen; struct sg_fd *sfp; sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN); @@ -2389,6 +2414,14 @@ sg_add_sfp(struct sg_device *sdp) sfp->force_packid = SG_DEF_FORCE_PACK_ID; sfp->cmd_q = SG_DEF_COMMAND_Q; sfp->keep_orphan = SG_DEF_KEEP_ORPHAN; + /* + * SG_SCATTER_SZ initializes scatter_elem_sz but different value may + * be given as driver/module parameter (e.g. 'scatter_elem_sz=8192'). + * Any user provided number will be changed to be PAGE_SIZE as a + * minimum, otherwise it will be rounded down (if required) to a + * power of 2. So it will always be a power of 2. + */ + sfp->sgat_elem_sz = scatter_elem_sz; sfp->parentdp = sdp; atomic_set(&sfp->submitted, 0); atomic_set(&sfp->waiting, 0); @@ -2405,14 +2438,13 @@ sg_add_sfp(struct sg_device *sdp) if (unlikely(sg_big_buff != def_reserved_size)) sg_big_buff = def_reserved_size; - bufflen = min_t(int, sg_big_buff, - max_sectors_bytes(sdp->device->request_queue)); - sg_build_reserve(sfp, bufflen); - SG_LOG(3, sfp, "%s: bufflen=%d, num_sgat=%d\n", __func__, - sfp->reserve.buflen, sfp->reserve.num_sgat); + rbuf_len = min_t(int, sg_big_buff, sdp->max_sgat_sz); + if (rbuf_len > 0) + sg_build_reserve(sfp, rbuf_len); kref_get(&sdp->d_ref); __module_get(THIS_MODULE); + SG_LOG(3, sfp, "%s: success, sfp=0x%p ++\n", __func__, sfp); return sfp; } @@ -2443,7 +2475,7 @@ sg_remove_sfp_usercontext(struct work_struct *work) if (sfp->reserve.buflen > 0) { SG_LOG(6, sfp, "%s: buflen=%d, num_sgat=%d\n", __func__, (int)sfp->reserve.buflen, (int)sfp->reserve.num_sgat); - sg_remove_scat(sfp, &sfp->reserve); + sg_remove_sgat(sfp, &sfp->reserve); } SG_LOG(6, sfp, "%s: sfp=0x%p\n", __func__, sfp); From patchwork Thu Oct 15 02:06:19 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269801 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id DDD9EC43457 for ; Thu, 15 Oct 2020 02:07:24 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 8BB3E22257 for ; Thu, 15 Oct 2020 02:07:24 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1732548AbgJOCHY (ORCPT ); Wed, 14 Oct 2020 22:07:24 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40172 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728185AbgJOCHQ (ORCPT ); Wed, 14 Oct 2020 22:07:16 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 46E1E204259; Thu, 15 Oct 2020 04:07:14 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id R20Q8Xbu2y-J; Thu, 15 Oct 2020 04:07:11 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 3AA08204255; Thu, 15 Oct 2020 04:07:07 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 20/44] sg: sg_find_srp_by_id Date: Wed, 14 Oct 2020 22:06:19 -0400 Message-Id: <20201015020643.432908-21-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Replace sg_get_rq_mark() with sg_find_srp_by_id() and sg_get_ready_srp(). Add sg_chk_mmap() to check flags and reserve buffer available for mmap() based requests. Add sg_copy_sense() and sg_rec_state_v3() which is just refactoring. Add sg_calc_rq_dur() and sg_get_dur() in preparation for optional nanosecond duration timing. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 286 +++++++++++++++++++++++++++++++--------------- 1 file changed, 197 insertions(+), 89 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 591852609ba5..c453c3d79587 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -154,16 +154,19 @@ struct sg_fd; struct sg_request { /* SG_MAX_QUEUE requests outstanding per file */ struct list_head entry; /* list entry */ - struct sg_fd *parentfp; /* NULL -> not in use */ struct sg_scatter_hold data; /* hold buffer, perhaps scatter list */ struct sg_io_hdr header; /* scsi command+info, see */ u8 sense_b[SCSI_SENSE_BUFFERSIZE]; + u32 duration; /* cmd duration in milliseconds */ char res_used; /* 1 -> using reserve buffer, 0 -> not ... */ char orphan; /* 1 -> drop on sight, 0 -> normal */ char sg_io_owned; /* 1 -> packet belongs to SG_IO */ /* done protected by rq_list_lock */ char done; /* 0->before bh, 1->before read, 2->read */ atomic_t rq_st; /* request state, holds a enum sg_rq_state */ + u64 start_ns; /* starting point of command duration calc */ + unsigned long frq_bm[1]; /* see SG_FRQ_* defines above */ + struct sg_fd *parentfp; /* pointer to owning fd, even when on fl */ struct request *rq; /* released in sg_rq_end_io(), bio kept */ struct bio *bio; /* kept until this req -->SG_RS_INACTIVE */ struct execute_work ew_orph; /* harvest orphan request */ @@ -229,7 +232,7 @@ static ssize_t sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, size_t count, bool blocking, bool read_only, bool sg_io_owned, struct sg_request **o_srp); -static int sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwp); +static int sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp); static int sg_read_append(struct sg_request *srp, void __user *outp, int num_xfer); static void sg_remove_sgat(struct sg_fd *sfp, struct sg_scatter_hold *schp); @@ -239,6 +242,7 @@ static void sg_link_reserve(struct sg_fd *sfp, struct sg_request *srp, static void sg_unlink_reserve(struct sg_fd *sfp, struct sg_request *srp); static struct sg_fd *sg_add_sfp(struct sg_device *sdp); static void sg_remove_sfp(struct kref *); +static struct sg_request *sg_find_srp_by_id(struct sg_fd *sfp, int pack_id); static struct sg_request *sg_setup_req(struct sg_fd *sfp); static void sg_deact_request(struct sg_fd *sfp, struct sg_request *srp); static struct sg_device *sg_get_dev(int dev); @@ -450,7 +454,8 @@ sg_open(struct inode *inode, struct file *filp) sfp = sg_add_sfp(sdp); /* increments sdp->d_ref */ if (IS_ERR(sfp)) { res = PTR_ERR(sfp); - goto out_undo; } + goto out_undo; + } filp->private_data = sfp; atomic_inc(&sdp->open_cnt); @@ -513,7 +518,6 @@ sg_release(struct inode *inode, struct file *filp) static ssize_t sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) { - bool blocking = !(filp->f_flags & O_NONBLOCK); int mxsize, cmd_size, input_size, res; u8 opcode; struct sg_device *sdp; @@ -614,21 +618,19 @@ sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) } /* * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV, - * but is is possible that the app intended SG_DXFER_TO_DEV, because there - * is a non-zero input_size, so emit a warning. + * but it is possible that the app intended SG_DXFER_TO_DEV, because + * there is a non-zero input_size, so emit a warning. */ if (h3p->dxfer_direction == SG_DXFER_TO_FROM_DEV) { - printk_ratelimited(KERN_WARNING - "sg_write: data in/out %d/%d bytes " - "for SCSI command 0x%x-- guessing " - "data in;\n program %s not setting " - "count and/or reply_len properly\n", - ohp->reply_len - (int)SZ_SG_HEADER, - input_size, (unsigned int) cmnd[0], - current->comm); + printk_ratelimited + (KERN_WARNING + "%s: data in/out %d/%d bytes for SCSI command 0x%x-- guessing data in;\n" + " program %s not setting count and/or reply_len properly\n", + __func__, ohp->reply_len - (int)SZ_SG_HEADER, + input_size, (unsigned int)cmnd[0], current->comm); } cwr.timeout = sfp->timeout; - cwr.blocking = blocking; + cwr.blocking = !(filp->f_flags & O_NONBLOCK); cwr.srp = srp; cwr.cmnd = cmnd; res = sg_common_write(sfp, &cwr); @@ -656,6 +658,18 @@ sg_fetch_cmnd(struct file *filp, struct sg_fd *sfp, const u8 __user *u_cdbp, return 0; } +static inline int +sg_chk_mmap(struct sg_fd *sfp, int rq_flags, int len) +{ + if (len > sfp->reserve.buflen) + return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */ + if (rq_flags & SG_FLAG_DIRECT_IO) + return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */ + if (sfp->res_in_use) + return -EBUSY; /* reserve buffer already being used */ + return 0; +} + static ssize_t sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, size_t count, bool blocking, bool read_only, bool sg_io_owned, @@ -688,17 +702,10 @@ sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, return -ENOSYS; } if (hp->flags & SG_FLAG_MMAP_IO) { - if (hp->dxfer_len > sfp->reserve.buflen) { - sg_deact_request(sfp, srp); - return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */ - } - if (hp->flags & SG_FLAG_DIRECT_IO) { + res = sg_chk_mmap(sfp, hp->flags, hp->dxfer_len); + if (res) { sg_deact_request(sfp, srp); - return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */ - } - if (sfp->res_in_use) { - sg_deact_request(sfp, srp); - return -EBUSY; /* reserve buffer already being used */ + return res; } } ul_timeout = msecs_to_jiffies(srp->header.timeout); @@ -720,6 +727,12 @@ sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, return count; } +/* + * All writes and submits converge on this function to launch the SCSI + * command/request (via blk_execute_rq_nowait). Returns a pointer to a + * sg_request object holding the request just issued or a negated errno + * value twisted by ERR_PTR. + */ static int sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) { @@ -801,36 +814,58 @@ sg_rstate_chg(struct sg_request *srp, enum sg_rq_state old_st, * returns true (or an event like a signal (e.g. control-C) occurs). */ -static struct sg_request * -sg_get_rq_mark(struct sg_fd *sfp, int pack_id) +static inline bool +sg_get_ready_srp(struct sg_fd *sfp, struct sg_request **srpp, int pack_id) { - struct sg_request *resp; - unsigned long iflags; + struct sg_request *srp; - spin_lock_irqsave(&sfp->rq_list_lock, iflags); - list_for_each_entry(resp, &sfp->rq_list, entry) { - /* look for requests that are ready + not SG_IO owned */ - if (resp->done == 1 && !resp->sg_io_owned && - (-1 == pack_id || resp->header.pack_id == pack_id)) { - resp->done = 2; /* guard against other readers */ - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); - return resp; - } + if (unlikely(SG_IS_DETACHING(sfp->parentdp))) { + *srpp = NULL; + return true; } - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); - return NULL; + srp = sg_find_srp_by_id(sfp, pack_id); + *srpp = srp; + return !!srp; +} + +/* + * Returns number of bytes copied to user space provided sense buffer or + * negated errno value. + */ +static int +sg_copy_sense(struct sg_request *srp) +{ + int sb_len_ret = 0; + struct sg_io_hdr *hp = &srp->header; + + /* If need be, copy the sense buffer to the user space */ + if ((CHECK_CONDITION & hp->masked_status) || + (DRIVER_SENSE & hp->driver_status)) { + int sb_len = SCSI_SENSE_BUFFERSIZE; + void __user *up = hp->sbp; + + sb_len = min_t(int, hp->mx_sb_len, sb_len); + /* Additional sense length field */ + sb_len_ret = 8 + (int)srp->sense_b[7]; + sb_len_ret = min_t(int, sb_len_ret, sb_len); + if (copy_to_user(up, srp->sense_b, sb_len_ret)) + return -EFAULT; + hp->sb_len_wr = sb_len_ret; + } + return sb_len_ret; } static int -srp_done(struct sg_fd *sfp, struct sg_request *srp) +sg_rec_state_v3(struct sg_fd *sfp, struct sg_request *srp) { - unsigned long flags; - int ret; + int sb_len_wr; - spin_lock_irqsave(&sfp->rq_list_lock, flags); - ret = srp->done; - spin_unlock_irqrestore(&sfp->rq_list_lock, flags); - return ret; + sb_len_wr = sg_copy_sense(srp); + if (sb_len_wr < 0) + return sb_len_wr; + if (unlikely(SG_IS_DETACHING(sfp->parentdp))) + return -ENODEV; + return 0; } #if IS_ENABLED(CONFIG_SCSI_LOGGING) @@ -860,12 +895,11 @@ sg_rep_rq_state_fail(struct sg_fd *sfp, enum sg_rq_state exp_old_st, #endif static ssize_t -sg_receive_v3(struct sg_fd *sfp, char __user *buf, size_t count, - struct sg_request *srp) +sg_receive_v3(struct sg_fd *sfp, struct sg_request *srp, size_t count, + void __user *p) { - struct sg_io_hdr *hp = &srp->header; int err = 0; - int len; + struct sg_io_hdr *hp = &srp->header; if (in_compat_syscall()) { if (count < sizeof(struct compat_sg_io_hdr)) { @@ -876,27 +910,11 @@ sg_receive_v3(struct sg_fd *sfp, char __user *buf, size_t count, err = -EINVAL; goto err_out; } - hp->sb_len_wr = 0; - if (hp->mx_sb_len > 0 && hp->sbp) { - if ((CHECK_CONDITION & hp->masked_status) || - (DRIVER_SENSE & hp->driver_status)) { - int sb_len = SCSI_SENSE_BUFFERSIZE; - - sb_len = (hp->mx_sb_len > sb_len) ? sb_len : - hp->mx_sb_len; - /* Additional sense length field */ - len = 8 + (int)srp->sense_b[7]; - len = (len > sb_len) ? sb_len : len; - if (copy_to_user(hp->sbp, srp->sense_b, len)) { - err = -EFAULT; - goto err_out; - } - hp->sb_len_wr = len; - } - } + SG_LOG(3, sfp, "%s: srp=0x%p\n", __func__, srp); + err = sg_rec_state_v3(sfp, srp); if (hp->masked_status || hp->host_status || hp->driver_status) hp->info |= SG_INFO_CHECK; - err = put_sg_io_hdr(hp, buf); + err = put_sg_io_hdr(hp, p); err_out: sg_finish_scsi_blk_rq(srp); sg_deact_request(sfp, srp); @@ -977,16 +995,22 @@ sg_read_v1v2(void __user *buf, int count, struct sg_fd *sfp, return res; } +/* + * This is the read(2) system call entry point (see sg_fops) for this driver. + * Accepts v1, v2 or v3 type headers (not v4). Returns count or negated + * errno; if count is 0 then v3: returns -EINVAL; v1+v2: 0 when no other + * error detected or -EIO. + */ static ssize_t sg_read(struct file *filp, char __user *p, size_t count, loff_t *ppos) { bool could_be_v3; bool non_block = !!(filp->f_flags & O_NONBLOCK); - int want_id = -1; + int want_id = SG_PACK_ID_WILDCARD; int hlen, ret; - struct sg_device *sdp; + struct sg_device *sdp = NULL; struct sg_fd *sfp; - struct sg_request *srp; + struct sg_request *srp = NULL; struct sg_header *h2p = NULL; struct sg_io_hdr a_sg_io_hdr; @@ -1001,7 +1025,7 @@ sg_read(struct file *filp, char __user *p, size_t count, loff_t *ppos) sfp = filp->private_data; sdp = sfp->parentdp; SG_LOG(3, sfp, "%s: read() count=%d\n", __func__, (int)count); - ret = sg_allow_if_err_recovery(sdp, false); + ret = sg_allow_if_err_recovery(sdp, non_block); if (ret) return ret; @@ -1020,17 +1044,13 @@ sg_read(struct file *filp, char __user *p, size_t count, loff_t *ppos) if (h2p->reply_len < 0 && could_be_v3) { struct sg_io_hdr *v3_hdr = (struct sg_io_hdr *)h2p; - if (likely(v3_hdr->interface_id == 'S')) { + if (v3_hdr->interface_id == 'S') { struct sg_io_hdr __user *h3_up; h3_up = (struct sg_io_hdr __user *)p; ret = get_user(want_id, &h3_up->pack_id); - if (unlikely(ret)) + if (ret) return ret; - } else if (v3_hdr->interface_id == 'Q') { - pr_info_once("sg: %s: v4 interface%s here\n", - __func__, " disallowed"); - return -EPERM; } else { return -EPERM; } @@ -1038,25 +1058,25 @@ sg_read(struct file *filp, char __user *p, size_t count, loff_t *ppos) want_id = h2p->pack_id; } } - srp = sg_get_rq_mark(sfp, want_id); - if (!srp) { /* now wait on packet to arrive */ + srp = sg_find_srp_by_id(sfp, want_id); + if (!srp) { /* nothing available so wait on packet to arrive or */ if (SG_IS_DETACHING(sdp)) return -ENODEV; if (non_block) /* O_NONBLOCK or v3::flags & SGV4_FLAG_IMMED */ return -EAGAIN; - ret = wait_event_interruptible - (sfp->read_wait, - (SG_IS_DETACHING(sdp) || - (srp = sg_get_rq_mark(sfp, want_id)))); + ret = wait_event_interruptible(sfp->read_wait, + sg_get_ready_srp(sfp, &srp, + want_id)); if (SG_IS_DETACHING(sdp)) return -ENODEV; if (ret) /* -ERESTARTSYS as signal hit process */ return ret; + /* otherwise srp should be valid */ } if (srp->header.interface_id == '\0') ret = sg_read_v1v2(p, (int)count, sfp, srp); else - ret = sg_receive_v3(sfp, p, count, srp); + ret = sg_receive_v3(sfp, srp, count, p); if (ret < 0) SG_LOG(1, sfp, "%s: negated errno: %d\n", __func__, ret); return ret < 0 ? ret : (int)count; @@ -1105,6 +1125,52 @@ sg_calc_sgat_param(struct sg_device *sdp) sdp->max_sgat_sz = sz; } +static u32 +sg_calc_rq_dur(const struct sg_request *srp) +{ + ktime_t ts0 = ns_to_ktime(srp->start_ns); + ktime_t now_ts; + s64 diff; + + if (ts0 == 0) + return 0; + if (unlikely(ts0 == S64_MAX)) /* _prior_ to issuing req */ + return 999999999; /* eye catching */ + now_ts = ktime_get_boottime(); + if (unlikely(ts0 > now_ts)) + return 999999998; + /* unlikely req duration will exceed 2**32 milliseconds */ + diff = ktime_ms_delta(now_ts, ts0); + return (diff > (s64)U32_MAX) ? 3999999999U : (u32)diff; +} + +/* Return of U32_MAX means srp is inactive */ +static u32 +sg_get_dur(struct sg_request *srp, const enum sg_rq_state *sr_stp, + bool *is_durp) +{ + bool is_dur = false; + u32 res = U32_MAX; + + switch (sr_stp ? *sr_stp : atomic_read(&srp->rq_st)) { + case SG_RS_INFLIGHT: + case SG_RS_BUSY: + res = sg_calc_rq_dur(srp); + break; + case SG_RS_AWAIT_RCV: + case SG_RS_RCV_DONE: + case SG_RS_INACTIVE: + res = srp->duration; + is_dur = true; /* completion has occurred, timing finished */ + break; + default: + break; + } + if (is_durp) + *is_durp = is_dur; + return res; +} + static void sg_fill_request_table(struct sg_fd *sfp, struct sg_req_info *rinfo) { @@ -1121,6 +1187,7 @@ sg_fill_request_table(struct sg_fd *sfp, struct sg_req_info *rinfo) srp->header.masked_status & srp->header.host_status & srp->header.driver_status; + rinfo[val].duration = sg_get_dur(srp, NULL, NULL); /* dummy */ if (srp->done) rinfo[val].duration = srp->header.duration; @@ -1138,6 +1205,18 @@ sg_fill_request_table(struct sg_fd *sfp, struct sg_req_info *rinfo) } } +static int +srp_done(struct sg_fd *sfp, struct sg_request *srp) +{ + unsigned long flags; + int ret; + + spin_lock_irqsave(&sfp->rq_list_lock, flags); + ret = srp->done; + spin_unlock_irqrestore(&sfp->rq_list_lock, flags); + return ret; +} + /* * Handles ioctl(SG_IO) for blocking (sync) usage of v3 or v4 interface. * Returns 0 on success else a negated errno. @@ -1165,7 +1244,7 @@ sg_ctl_sg_io(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, if (srp->done) { srp->done = 2; spin_unlock_irq(&sfp->rq_list_lock); - res = sg_receive_v3(sfp, p, SZ_SG_IO_HDR, srp); + res = sg_receive_v3(sfp, srp, SZ_SG_IO_HDR, p); return (res < 0) ? res : 0; } srp->orphan = 1; @@ -1393,7 +1472,9 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, return result; if (val > SG_MAX_CDB_SIZE) return -ENOMEM; - sfp->next_cmd_len = (val > 0) ? val : 0; + mutex_lock(&sfp->f_mutex); + sfp->next_cmd_len = max_t(int, val, 0); + mutex_unlock(&sfp->f_mutex); return 0; case SG_GET_ACCESS_COUNT: SG_LOG(3, sfp, "%s: SG_GET_ACCESS_COUNT\n", __func__); @@ -2360,6 +2441,33 @@ sg_read_append(struct sg_request *srp, void __user *outp, int num_xfer) return res; } +/* + * If there are multiple requests outstanding, the speed of this function is + * important. SG_PACK_ID_WILDCARD is -1 and that case is typically + * the fast path. This function is only used in the non-blocking cases. + * Returns pointer to (first) matching sg_request or NULL. If found, + * sg_request state is moved from SG_RS_AWAIT_RCV to SG_RS_BUSY. + */ +static struct sg_request * +sg_find_srp_by_id(struct sg_fd *sfp, int pack_id) +{ + unsigned long iflags; + struct sg_request *resp; + + spin_lock_irqsave(&sfp->rq_list_lock, iflags); + list_for_each_entry(resp, &sfp->rq_list, entry) { + /* look for requests that are ready + not SG_IO owned */ + if (resp->done == 1 && !resp->sg_io_owned && + (-1 == pack_id || resp->header.pack_id == pack_id)) { + resp->done = 2; /* guard against other readers */ + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); + return resp; + } + } + spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); + return NULL; +} + static void sg_link_reserve(struct sg_fd *sfp, struct sg_request *srp, int size) { From patchwork Thu Oct 15 02:06:21 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269803 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 4634FC43467 for ; Thu, 15 Oct 2020 02:07:25 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id F245822258 for ; Thu, 15 Oct 2020 02:07:24 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729854AbgJOCHW (ORCPT ); Wed, 14 Oct 2020 22:07:22 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40183 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732513AbgJOCHS (ORCPT ); Wed, 14 Oct 2020 22:07:18 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 3B5E8204269; Thu, 15 Oct 2020 04:07:16 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id UvpOfgz-uG1R; Thu, 15 Oct 2020 04:07:14 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 773F3204258; Thu, 15 Oct 2020 04:07:09 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 22/44] sg: printk change %p to %pK Date: Wed, 14 Oct 2020 22:06:21 -0400 Message-Id: <20201015020643.432908-23-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org This driver does a lot of buffer juggling in an attempt to take some of that chore away from its users. When debugging problems associated with that buffer juggling getting sensible pointer values is a major aid. So change %p to %pK. The system administrator can choose to obfuscate %pK pointers. The "pK" is also easier to search for in the code if further changes are required. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index d0377bf95046..5fa750a9db9b 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -911,7 +911,7 @@ sg_receive_v3(struct sg_fd *sfp, struct sg_request *srp, size_t count, err = -EINVAL; goto err_out; } - SG_LOG(3, sfp, "%s: srp=0x%p\n", __func__, srp); + SG_LOG(3, sfp, "%s: srp=0x%pK\n", __func__, srp); err = sg_rec_state_v3(sfp, srp); if (hp->masked_status || hp->host_status || hp->driver_status) hp->info |= SG_INFO_CHECK; @@ -1698,7 +1698,7 @@ sg_mmap(struct file *filp, struct vm_area_struct *vma) return -ENXIO; } req_sz = vma->vm_end - vma->vm_start; - SG_LOG(3, sfp, "%s: vm_start=%p, len=%d\n", __func__, + SG_LOG(3, sfp, "%s: vm_start=%pK, len=%d\n", __func__, (void *)vma->vm_start, (int)req_sz); if (vma->vm_pgoff) return -EINVAL; /* only an offset of 0 accepted */ @@ -1746,7 +1746,7 @@ sg_rq_end_io_usercontext(struct work_struct *work) WARN_ONCE(1, "%s: sfp unexpectedly NULL\n", __func__); return; } - SG_LOG(3, sfp, "%s: srp=0x%p\n", __func__, srp); + SG_LOG(3, sfp, "%s: srp=0x%pK\n", __func__, srp); sg_finish_scsi_blk_rq(srp); sg_deact_request(sfp, srp); kref_put(&sfp->f_ref, sg_remove_sfp); @@ -1919,7 +1919,7 @@ sg_add_device_helper(struct gendisk *disk, struct scsi_device *scsidp) k = error; SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp, - "%s: dev=%d, sdp=0x%p ++\n", __func__, k, sdp)); + "%s: dev=%d, sdp=0x%pK ++\n", __func__, k, sdp)); sprintf(disk->disk_name, "sg%d", k); disk->first_minor = k; sdp->disk = disk; @@ -2028,7 +2028,7 @@ sg_device_destroy(struct kref *kref) struct sg_device *sdp = container_of(kref, struct sg_device, d_ref); unsigned long flags; - SCSI_LOG_TIMEOUT(1, pr_info("[tid=%d] %s: sdp idx=%d, sdp=0x%p --\n", + SCSI_LOG_TIMEOUT(1, pr_info("[tid=%d] %s: sdp idx=%d, sdp=0x%pK --\n", (current ? current->pid : -1), __func__, sdp->index, sdp)); /* @@ -2060,7 +2060,7 @@ sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf) return; /* only want to do following once per device */ SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, sdp->device, - "%s: 0x%p\n", __func__, sdp)); + "%s: 0x%pK\n", __func__, sdp)); read_lock_irqsave(&sdp->sfd_lock, iflags); list_for_each_entry(sfp, &sdp->sfds, sfd_entry) { @@ -2188,7 +2188,7 @@ sg_start_req(struct sg_request *srp, u8 *cmd) long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL); if (!long_cmdp) return -ENOMEM; - SG_LOG(5, sfp, "%s: long_cmdp=0x%p ++\n", __func__, long_cmdp); + SG_LOG(5, sfp, "%s: long_cmdp=0x%pK ++\n", __func__, long_cmdp); } SG_LOG(4, sfp, "%s: dxfer_len=%d, data-%s\n", __func__, dxfer_len, (r0w ? "OUT" : "IN")); @@ -2303,7 +2303,7 @@ sg_finish_scsi_blk_rq(struct sg_request *srp) struct sg_fd *sfp = srp->parentfp; struct sg_scatter_hold *req_schp = &srp->data; - SG_LOG(4, sfp, "%s: srp=0x%p%s\n", __func__, srp, + SG_LOG(4, sfp, "%s: srp=0x%pK%s\n", __func__, srp, (srp->res_used) ? " rsv" : ""); if (!srp->sg_io_owned) { atomic_dec(&sfp->submitted); @@ -2348,7 +2348,7 @@ sg_mk_sgat(struct sg_scatter_hold *schp, struct sg_fd *sfp, int minlen) align_sz = ALIGN(minlen, SG_DEF_SECTOR_SZ); schp->pages = kcalloc(mx_sgat_elems, ptr_sz, mask_kz); - SG_LOG(4, sfp, "%s: minlen=%d, align_sz=%d [sz=%zu, 0x%p ++]\n", + SG_LOG(4, sfp, "%s: minlen=%d, align_sz=%d [sz=%zu, 0x%pK ++]\n", __func__, minlen, align_sz, mx_sgat_elems * ptr_sz, schp->pages); if (unlikely(!schp->pages)) @@ -2366,7 +2366,7 @@ sg_mk_sgat(struct sg_scatter_hold *schp, struct sg_fd *sfp, int minlen) schp->pages[k] = alloc_pages(mask_ap, order); if (!schp->pages[k]) goto err_out; - SG_LOG(5, sfp, "%s: k=%d, order=%d [0x%p ++]\n", __func__, k, + SG_LOG(5, sfp, "%s: k=%d, order=%d [0x%pK ++]\n", __func__, k, order, schp->pages[k]); } schp->page_order = order; @@ -2402,12 +2402,12 @@ sg_remove_sgat_helper(struct sg_fd *sfp, struct sg_scatter_hold *schp) return; for (k = 0; k < schp->num_sgat; ++k) { p = schp->pages[k]; - SG_LOG(5, sfp, "%s: pg[%d]=0x%p --\n", __func__, k, p); + SG_LOG(5, sfp, "%s: pg[%d]=0x%pK --\n", __func__, k, p); if (unlikely(!p)) continue; __free_pages(p, schp->page_order); } - SG_LOG(5, sfp, "%s: pg_order=%u, free pgs=0x%p --\n", __func__, + SG_LOG(5, sfp, "%s: pg_order=%u, free pgs=0x%pK --\n", __func__, schp->page_order, schp->pages); kfree(schp->pages); } @@ -2644,7 +2644,7 @@ sg_add_sfp(struct sg_device *sdp) } list_add_tail(&sfp->sfd_entry, &sdp->sfds); write_unlock_irqrestore(&sdp->sfd_lock, iflags); - SG_LOG(3, sfp, "%s: sfp=0x%p\n", __func__, sfp); + SG_LOG(3, sfp, "%s: sfp=0x%pK\n", __func__, sfp); if (unlikely(sg_big_buff != def_reserved_size)) sg_big_buff = def_reserved_size; @@ -2654,7 +2654,7 @@ sg_add_sfp(struct sg_device *sdp) kref_get(&sdp->d_ref); __module_get(THIS_MODULE); - SG_LOG(3, sfp, "%s: success, sfp=0x%p ++\n", __func__, sfp); + SG_LOG(3, sfp, "%s: success, sfp=0x%pK ++\n", __func__, sfp); return sfp; } @@ -2697,7 +2697,7 @@ sg_remove_sfp_usercontext(struct work_struct *work) sg_remove_sgat(sfp, &sfp->reserve); } - SG_LOG(6, sfp, "%s: sfp=0x%p\n", __func__, sfp); + SG_LOG(6, sfp, "%s: sfp=0x%pK\n", __func__, sfp); kfree(sfp); if (sdp) { From patchwork Thu Oct 15 02:06:23 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269804 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 511B5C433DF for ; Thu, 15 Oct 2020 02:07:23 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 06FA722258 for ; Thu, 15 Oct 2020 02:07:22 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1730026AbgJOCHW (ORCPT ); Wed, 14 Oct 2020 22:07:22 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40195 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732540AbgJOCHV (ORCPT ); Wed, 14 Oct 2020 22:07:21 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 0CF10204272; Thu, 15 Oct 2020 04:07:19 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id AMb6ycxA0zf7; Thu, 15 Oct 2020 04:07:16 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id B4BC720425A; Thu, 15 Oct 2020 04:07:11 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 24/44] sg: xarray for reqs in fd Date: Wed, 14 Oct 2020 22:06:23 -0400 Message-Id: <20201015020643.432908-25-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Replace the linked list and the fixed array of requests (max 16) with an xarray. The xarray (srp_arr) has two marks: one for INACTIVE state (i.e. available for re-use) requests; the other is AWAIT state which is after the internal completion point of a request and before the user space has fetched the response. Of the five states in sg_request::rq_st, two are marked. They are SG_RS_INACTIVE and SG_RS_AWAIT_RCV. This allows the request xarray (sg_fd::srp_arr) to be searched (with xa_for_each_mark) on two embedded sub-lists. The SG_RS_INACTIVE sub-list replaces the free list. The SG_RS_AWAIT_RCV sub-list contains requests that have reached their internal completion point but have not been read/ received by the user space. Add support functions for this and partially wire them up. Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 321 +++++++++++++++++++++++++++++++++------------- 1 file changed, 234 insertions(+), 87 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 898175d5867d..6043d25519ab 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -73,7 +73,7 @@ static char *sg_version_date = "20190606"; #define SG_MAX_CDB_SIZE 252 /* Following enum contains the states of sg_request::rq_st */ -enum sg_rq_state { +enum sg_rq_state { /* N.B. sg_rq_state_arr assumes SG_RS_AWAIT_RCV==2 */ SG_RS_INACTIVE = 0, /* request not in use (e.g. on fl) */ SG_RS_INFLIGHT, /* active: cmd/req issued, no response yet */ SG_RS_AWAIT_RCV, /* have response from LLD, awaiting receive */ @@ -114,6 +114,11 @@ enum sg_rq_state { #define SG_FDEV_DETACHING 1 /* may be unexpected device removal */ #define SG_FDEV_LOG_SENSE 2 /* set by ioctl(SG_SET_DEBUG) */ +/* xarray 'mark's allow sub-lists within main array/list. */ +#define SG_XA_RQ_FREE XA_MARK_0 /* xarray sets+clears */ +#define SG_XA_RQ_INACTIVE XA_MARK_1 +#define SG_XA_RQ_AWAIT XA_MARK_2 + int sg_big_buff = SG_DEF_RESERVED_SIZE; /* * This variable is accessible via /proc/scsi/sg/def_reserved_size . Each @@ -153,11 +158,11 @@ struct sg_device; /* forward declarations */ struct sg_fd; struct sg_request { /* SG_MAX_QUEUE requests outstanding per file */ - struct list_head entry; /* list entry */ struct sg_scatter_hold data; /* hold buffer, perhaps scatter list */ struct sg_io_hdr header; /* scsi command+info, see */ u8 sense_b[SCSI_SENSE_BUFFERSIZE]; u32 duration; /* cmd duration in milliseconds */ + u32 rq_idx; /* my index within parent's srp_arr */ char res_used; /* 1 -> using reserve buffer, 0 -> not ... */ char orphan; /* 1 -> drop on sight, 0 -> normal */ u32 rq_result; /* packed scsi request result from LLD */ @@ -176,24 +181,23 @@ struct sg_request { /* SG_MAX_QUEUE requests outstanding per file */ struct sg_fd { /* holds the state of a file descriptor */ struct sg_device *parentdp; /* owning device */ wait_queue_head_t read_wait; /* queue read until command done */ - spinlock_t rq_list_lock; /* protect access to list in req_arr */ struct mutex f_mutex; /* protect against changes in this fd */ int timeout; /* defaults to SG_DEFAULT_TIMEOUT */ int timeout_user; /* defaults to SG_DEFAULT_TIMEOUT_USER */ u32 idx; /* my index within parent's sfp_arr */ - atomic_t submitted; /* number inflight or awaiting read */ - atomic_t waiting; /* number of requests awaiting read */ + atomic_t submitted; /* number inflight or awaiting receive */ + atomic_t waiting; /* number of requests awaiting receive */ + atomic_t req_cnt; /* number of requests */ int sgat_elem_sz; /* initialized to scatter_elem_sz */ struct sg_scatter_hold reserve; /* buffer for this file descriptor */ - struct list_head rq_list; /* head of request list */ - struct fasync_struct *async_qp; /* used by asynchronous notification */ - struct sg_request req_arr[SG_MAX_QUEUE];/* use as singly-linked list */ char force_packid; /* 1 -> pack_id input to read(), 0 -> ignored */ char cmd_q; /* 1 -> allow command queuing, 0 -> don't */ u8 next_cmd_len; /* 0: automatic, >0: use on next write() */ char keep_orphan; /* 0 -> drop orphan (def), 1 -> keep for read() */ char mmap_called; /* 0 -> mmap() never called on this fd */ char res_in_use; /* 1 -> 'reserve' array in use */ + struct fasync_struct *async_qp; /* used by asynchronous notification */ + struct xarray srp_arr; struct kref f_ref; struct execute_work ew_fd; /* harvest all fd resources and lists */ }; @@ -272,6 +276,7 @@ static const char *sg_rq_st_str(enum sg_rq_state rq_st, bool long_str); #if IS_ENABLED(CONFIG_SCSI_LOGGING) && IS_ENABLED(SG_DEBUG) #define SG_LOG_BUFF_SZ 48 +#define SG_LOG_ACTIVE 1 #define SG_LOG(depth, sfp, fmt, a...) \ do { \ @@ -725,6 +730,115 @@ sg_submit(struct sg_fd *sfp, struct file *filp, const char __user *buf, return count; } +#if IS_ENABLED(SG_LOG_ACTIVE) +static void +sg_rq_state_fail_msg(struct sg_fd *sfp, enum sg_rq_state exp_old_st, + enum sg_rq_state want_st, enum sg_rq_state act_old_st, + const char *fromp) +{ + const char *eaw_rs = "expected_old,actual_old,wanted rq_st"; + + if (IS_ENABLED(CONFIG_SCSI_PROC_FS)) + SG_LOG(1, sfp, "%s: %s: %s: %s,%s,%s\n", + __func__, fromp, eaw_rs, + sg_rq_st_str(exp_old_st, false), + sg_rq_st_str(act_old_st, false), + sg_rq_st_str(want_st, false)); + else + pr_info("sg: %s: %s: %s: %d,%d,%d\n", __func__, fromp, eaw_rs, + (int)exp_old_st, (int)act_old_st, (int)want_st); +} +#endif + +static void +sg_rq_state_force(struct sg_request *srp, enum sg_rq_state new_st) +{ + bool prev, want; + struct xarray *xafp = &srp->parentfp->srp_arr; + + atomic_set(&srp->rq_st, new_st); + want = (new_st == SG_RS_AWAIT_RCV); + prev = xa_get_mark(xafp, srp->rq_idx, SG_XA_RQ_AWAIT); + if (prev != want) { + if (want) + __xa_set_mark(xafp, srp->rq_idx, SG_XA_RQ_AWAIT); + else + __xa_clear_mark(xafp, srp->rq_idx, SG_XA_RQ_AWAIT); + } + want = (new_st == SG_RS_INACTIVE); + prev = xa_get_mark(xafp, srp->rq_idx, SG_XA_RQ_INACTIVE); + if (prev != want) { + if (want) + __xa_set_mark(xafp, srp->rq_idx, SG_XA_RQ_INACTIVE); + else + __xa_clear_mark(xafp, srp->rq_idx, SG_XA_RQ_INACTIVE); + } +} + +static void +sg_rq_state_helper(struct xarray *xafp, struct sg_request *srp, int indic) +{ + if (indic & 1) /* from inactive state */ + __xa_clear_mark(xafp, srp->rq_idx, SG_XA_RQ_INACTIVE); + else if (indic & 2) /* to inactive state */ + __xa_set_mark(xafp, srp->rq_idx, SG_XA_RQ_INACTIVE); + + if (indic & 4) /* from await state */ + __xa_clear_mark(xafp, srp->rq_idx, SG_XA_RQ_AWAIT); + else if (indic & 8) /* to await state */ + __xa_set_mark(xafp, srp->rq_idx, SG_XA_RQ_AWAIT); +} + +/* Following array indexed by enum sg_rq_state, 0 means no xa mark change */ +static const int sg_rq_state_arr[] = {1, 0, 4, 0, 0}; +static const int sg_rq_state_mul2arr[] = {2, 0, 8, 0, 0}; + +/* + * This function keeps the srp->rq_st state and associated marks on the + * owning xarray's element in sync. If force is true then new_st is stored + * in srp->rq_st and xarray marks are set accordingly (and old_st is + * ignored); and 0 is returned. + * If force is false, then atomic_cmpxchg() is called. If the actual + * srp->rq_st is not old_st, then -EPROTOTYPE is returned. If the actual + * srp->rq_st is old_st then it is replaced by new_st and the xarray marks + * are setup accordingly and 0 is returned. This assumes srp_arr xarray + * spinlock is held. + */ +static int +sg_rq_state_chg(struct sg_request *srp, enum sg_rq_state old_st, + enum sg_rq_state new_st, bool force, const char *fromp) +{ + enum sg_rq_state act_old_st; + int indic; + unsigned long iflags; + struct xarray *xafp = &srp->parentfp->srp_arr; + + if (force) { + xa_lock_irqsave(xafp, iflags); + sg_rq_state_force(srp, new_st); + xa_unlock_irqrestore(xafp, iflags); + return 0; + } + indic = sg_rq_state_arr[(int)old_st] + + sg_rq_state_mul2arr[(int)new_st]; + act_old_st = (enum sg_rq_state)atomic_cmpxchg(&srp->rq_st, old_st, + new_st); + if (act_old_st != old_st) { +#if IS_ENABLED(SG_LOG_ACTIVE) + if (fromp) + sg_rq_state_fail_msg(srp->parentfp, old_st, new_st, + act_old_st, fromp); +#endif + return -EPROTOTYPE; /* only used for this error type */ + } + if (indic) { + xa_lock_irqsave(xafp, iflags); + sg_rq_state_helper(xafp, srp, indic); + xa_unlock_irqrestore(xafp, iflags); + } + return 0; +} + /* * All writes and submits converge on this function to launch the SCSI * command/request (via blk_execute_rq_nowait). Returns a pointer to a @@ -763,17 +877,8 @@ sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) sg_deact_request(sfp, srp); return k; /* probably out of space --> ENOMEM */ } - if (SG_IS_DETACHING(sdp)) { - if (srp->bio) { - scsi_req_free_cmd(scsi_req(srp->rq)); - blk_put_request(srp->rq); - srp->rq = NULL; - } - - sg_finish_scsi_blk_rq(srp); - sg_deact_request(sfp, srp); - return -ENODEV; - } + if (SG_IS_DETACHING(sdp)) + goto err_out; hp->duration = jiffies_to_msecs(jiffies); if (hp->interface_id != '\0' && /* v3 (or later) interface */ @@ -782,6 +887,10 @@ sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) else at_head = true; + if (!xa_get_mark(&sfp->srp_arr, srp->rq_idx, SG_XA_RQ_FREE)) { + SG_LOG(1, sfp, "%s: ahhh, request erased!!!\n", __func__); + goto err_out; + } if (!srp->sg_io_owned) atomic_inc(&sfp->submitted); srp->rq->timeout = cwrp->timeout; @@ -789,6 +898,22 @@ sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk, srp->rq, at_head, sg_rq_end_io); return 0; +err_out: + if (srp->bio) { + scsi_req_free_cmd(scsi_req(srp->rq)); + blk_put_request(srp->rq); + srp->rq = NULL; + } + sg_finish_scsi_blk_rq(srp); + sg_deact_request(sfp, srp); + return -ENODEV; +} + +static inline int +sg_rstate_chg(struct sg_request *srp, enum sg_rq_state old_st, + enum sg_rq_state new_st) +{ + return sg_rq_state_chg(srp, old_st, new_st, false, __func__); } /* @@ -1161,12 +1286,9 @@ sg_fill_request_element(struct sg_fd *sfp, struct sg_request *srp, static int srp_done(struct sg_fd *sfp, struct sg_request *srp) { - unsigned long flags; int ret; - spin_lock_irqsave(&sfp->rq_list_lock, flags); ret = srp->done; - spin_unlock_irqrestore(&sfp->rq_list_lock, flags); return ret; } @@ -1193,15 +1315,12 @@ sg_ctl_sg_io(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, (sfp->read_wait, (srp_done(sfp, srp) || SG_IS_DETACHING(sdp))); if (SG_IS_DETACHING(sdp)) return -ENODEV; - spin_lock_irq(&sfp->rq_list_lock); if (srp->done) { srp->done = 2; - spin_unlock_irq(&sfp->rq_list_lock); res = sg_receive_v3(sfp, srp, SZ_SG_IO_HDR, p); return (res < 0) ? res : 0; } srp->orphan = 1; - spin_unlock_irq(&sfp->rq_list_lock); return res; } @@ -1250,7 +1369,7 @@ static int sg_ctl_req_tbl(struct sg_fd *sfp, void __user *p) { int result, val; - unsigned long iflags; + unsigned long idx; struct sg_request *srp; sg_req_info_t *rinfop; @@ -1258,15 +1377,17 @@ sg_ctl_req_tbl(struct sg_fd *sfp, void __user *p) GFP_KERNEL); if (!rinfop) return -ENOMEM; - spin_lock_irqsave(&sfp->rq_list_lock, iflags); val = 0; - list_for_each_entry(srp, &sfp->rq_list, entry) { + xa_for_each(&sfp->srp_arr, idx, srp) { + if (!srp) + continue; + if (xa_get_mark(&sfp->srp_arr, idx, SG_XA_RQ_AWAIT)) + continue; if (val >= SG_MAX_QUEUE) break; sg_fill_request_element(sfp, srp, rinfop + val); val++; } - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); #ifdef CONFIG_COMPAT if (in_compat_syscall()) result = put_compat_request_table(p, rinfop); @@ -1311,7 +1432,7 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, int __user *ip = p; struct sg_request *srp; struct scsi_device *sdev; - unsigned long iflags; + unsigned long idx; __maybe_unused const char *pmlp = ", pass to mid-level"; SG_LOG(6, sfp, "%s: cmd=0x%x, O_NONBLOCK=%d\n", __func__, cmd_in, @@ -1334,14 +1455,15 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, return 0; case SG_GET_PACK_ID: /* or tag of oldest "read"-able, -1 if none */ val = -1; - spin_lock_irqsave(&sfp->rq_list_lock, iflags); - list_for_each_entry(srp, &sfp->rq_list, entry) { + srp = NULL; + xa_for_each_marked(&sfp->srp_arr, idx, srp, SG_XA_RQ_AWAIT) { + if (!srp) + continue; if ((1 == srp->done) && (!srp->sg_io_owned)) { val = srp->header.pack_id; break; } } - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); SG_LOG(3, sfp, "%s: SG_GET_PACK_ID=%d\n", __func__, val); return put_user(val, ip); case SG_GET_NUM_WAITING: @@ -1754,10 +1876,10 @@ sg_rq_end_io(struct request *rq, blk_status_t status) struct scsi_request *scsi_rp = scsi_req(rq); struct sg_device *sdp; struct sg_fd *sfp; - unsigned long iflags; unsigned int ms; int resid, slen; int done = 1; + unsigned long iflags; if (WARN_ON(srp->done != 0)) return; @@ -1801,7 +1923,6 @@ sg_rq_end_io(struct request *rq, blk_status_t status) scsi_req_free_cmd(scsi_req(rq)); blk_put_request(rq); - spin_lock_irqsave(&sfp->rq_list_lock, iflags); if (unlikely(srp->orphan)) { if (sfp->keep_orphan) srp->sg_io_owned = 0; @@ -1809,12 +1930,14 @@ sg_rq_end_io(struct request *rq, blk_status_t status) done = 0; } srp->done = done; - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); if (likely(done)) { /* Now wake up any sg_read() that is waiting for this * packet. */ + xa_lock_irqsave(&sfp->srp_arr, iflags); + __xa_set_mark(&sfp->srp_arr, srp->rq_idx, SG_XA_RQ_AWAIT); + xa_unlock_irqrestore(&sfp->srp_arr, iflags); wake_up_interruptible(&sfp->read_wait); kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN); kref_put(&sfp->f_ref, sg_remove_sfp); @@ -2434,20 +2557,19 @@ sg_read_append(struct sg_request *srp, void __user *outp, int num_xfer) static struct sg_request * sg_find_srp_by_id(struct sg_fd *sfp, int pack_id) { - unsigned long iflags; + unsigned long idx; struct sg_request *resp; - spin_lock_irqsave(&sfp->rq_list_lock, iflags); - list_for_each_entry(resp, &sfp->rq_list, entry) { + xa_for_each_marked(&sfp->srp_arr, idx, resp, SG_XA_RQ_AWAIT) { + if (!resp) + continue; /* look for requests that are ready + not SG_IO owned */ if (resp->done == 1 && !resp->sg_io_owned && (-1 == pack_id || resp->header.pack_id == pack_id)) { resp->done = 2; /* guard against other readers */ - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return resp; } } - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); return NULL; } @@ -2517,31 +2639,51 @@ sg_build_reserve(struct sg_fd *sfp, int req_size) static struct sg_request * sg_setup_req(struct sg_fd *sfp) { - int k; - unsigned long iflags; - struct sg_request *rp = sfp->req_arr; - - spin_lock_irqsave(&sfp->rq_list_lock, iflags); - if (!list_empty(&sfp->rq_list)) { - if (!sfp->cmd_q) - goto out_unlock; - - for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) { - if (!rp->parentfp) - break; + bool found = false; + int res; + unsigned long idx, iflags; + struct sg_request *rp; + struct xarray *xafp = &sfp->srp_arr; + + if (!xa_empty(xafp)) { + xa_for_each_marked(xafp, idx, rp, SG_XA_RQ_INACTIVE) { + if (!rp) + continue; + if (sg_rstate_chg(rp, SG_RS_INACTIVE, SG_RS_BUSY)) + continue; + memset(rp, 0, sizeof(*rp)); + rp->rq_idx = idx; + xa_lock_irqsave(xafp, iflags); + __xa_clear_mark(xafp, idx, SG_XA_RQ_INACTIVE); + xa_unlock_irqrestore(xafp, iflags); + found = true; + break; } - if (k >= SG_MAX_QUEUE) - goto out_unlock; } - memset(rp, 0, sizeof(struct sg_request)); + if (!found) { + rp = kzalloc(sizeof(*rp), GFP_KERNEL); + if (!rp) + return NULL; + } rp->parentfp = sfp; rp->header.duration = jiffies_to_msecs(jiffies); - list_add_tail(&rp->entry, &sfp->rq_list); - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); + if (!found) { + u32 n_idx; + struct xa_limit xal = { .max = 0, .min = 0 }; + + atomic_set(&rp->rq_st, SG_RS_BUSY); + xa_lock_irqsave(xafp, iflags); + xal.max = atomic_inc_return(&sfp->req_cnt); + res = __xa_alloc(xafp, &n_idx, rp, xal, GFP_KERNEL); + xa_unlock_irqrestore(xafp, iflags); + if (res < 0) { + pr_warn("%s: don't expect xa_alloc() to fail, errno=%d\n", + __func__, -res); + return NULL; + } + rp->rq_idx = n_idx; + } return rp; -out_unlock: - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); - return NULL; } static void @@ -2551,14 +2693,10 @@ sg_deact_request(struct sg_fd *sfp, struct sg_request *srp) if (WARN_ON(!sfp || !srp)) return; - if (list_empty(&sfp->rq_list)) - return; - spin_lock_irqsave(&sfp->rq_list_lock, iflags); - if (!list_empty(&srp->entry)) { - list_del(&srp->entry); - srp->parentfp = NULL; - } - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); + xa_lock_irqsave(&sfp->srp_arr, iflags); + __xa_set_mark(&sfp->srp_arr, srp->rq_idx, SG_XA_RQ_INACTIVE); + xa_unlock_irqrestore(&sfp->srp_arr, iflags); + atomic_set(&srp->rq_st, SG_RS_INACTIVE); } static struct sg_fd * @@ -2575,8 +2713,7 @@ sg_add_sfp(struct sg_device *sdp) return ERR_PTR(-ENOMEM); init_waitqueue_head(&sfp->read_wait); - spin_lock_init(&sfp->rq_list_lock); - INIT_LIST_HEAD(&sfp->rq_list); + xa_init_flags(&sfp->srp_arr, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_IRQ); kref_init(&sfp->f_ref); mutex_init(&sfp->f_mutex); sfp->timeout = SG_DEFAULT_TIMEOUT; @@ -2595,6 +2732,7 @@ sg_add_sfp(struct sg_device *sdp) sfp->parentdp = sdp; atomic_set(&sfp->submitted, 0); atomic_set(&sfp->waiting, 0); + atomic_set(&sfp->req_cnt, 0); if (SG_IS_DETACHING(sdp)) { kfree(sfp); @@ -2641,11 +2779,13 @@ static void sg_remove_sfp_usercontext(struct work_struct *work) { __maybe_unused int o_count; - unsigned long iflags; + unsigned long idx, iflags; struct sg_device *sdp; struct sg_fd *sfp = container_of(work, struct sg_fd, ew_fd.work); struct sg_fd *e_sfp; struct sg_request *srp; + struct sg_request *e_srp; + struct xarray *xafp = &sfp->srp_arr; if (!sfp) { pr_warn("sg: %s: sfp is NULL\n", __func__); @@ -2654,15 +2794,20 @@ sg_remove_sfp_usercontext(struct work_struct *work) sdp = sfp->parentdp; /* Cleanup any responses which were never read(). */ - spin_lock_irqsave(&sfp->rq_list_lock, iflags); - while (!list_empty(&sfp->rq_list)) { - srp = list_first_entry(&sfp->rq_list, struct sg_request, entry); - sg_finish_scsi_blk_rq(srp); - list_del(&srp->entry); - srp->parentfp = NULL; - } - spin_unlock_irqrestore(&sfp->rq_list_lock, iflags); - + xa_for_each(xafp, idx, srp) { + if (!srp) + continue; + if (!xa_get_mark(xafp, srp->rq_idx, SG_XA_RQ_INACTIVE)) + sg_finish_scsi_blk_rq(srp); + xa_lock_irqsave(xafp, iflags); + e_srp = __xa_erase(xafp, srp->rq_idx); + xa_unlock_irqrestore(xafp, iflags); + if (srp != e_srp) + SG_LOG(1, sfp, "%s: xa_erase() return unexpected\n", + __func__); + kfree(srp); + } + xa_destroy(xafp); if (sfp->reserve.buflen > 0) { SG_LOG(6, sfp, "%s: buflen=%d, num_sgat=%d\n", __func__, (int)sfp->reserve.buflen, (int)sfp->reserve.num_sgat); @@ -2755,7 +2900,9 @@ sg_rq_st_str(enum sg_rq_state rq_st, bool long_str) return long_str ? "unknown" : "unk"; } } -#else + +#elif IS_ENABLED(SG_LOG_ACTIVE) + static const char * sg_rq_st_str(enum sg_rq_state rq_st, bool long_str) { @@ -3005,7 +3152,7 @@ static void sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) { int k, new_interface, blen, usg; - unsigned long idx; + unsigned long idx, idx2; struct sg_request *srp; struct sg_fd *fp; const struct sg_io_hdr *hp; @@ -3017,7 +3164,6 @@ sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) if (!fp) continue; k++; - spin_lock(&fp->rq_list_lock); /* irqs already disabled */ seq_printf(s, " FD(%d): timeout=%dms buflen=%d (res)sgat=%d low_dma=%d idx=%lu\n", k, jiffies_to_msecs(fp->timeout), fp->reserve.buflen, (int)fp->reserve.num_sgat, @@ -3028,7 +3174,9 @@ sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) seq_printf(s, " submitted=%d waiting=%d\n", atomic_read(&fp->submitted), atomic_read(&fp->waiting)); - list_for_each_entry(srp, &fp->rq_list, entry) { + xa_for_each(&fp->srp_arr, idx2, srp) { + if (!srp) + continue; hp = &srp->header; new_interface = (hp->interface_id == '\0') ? 0 : 1; if (srp->res_used) { @@ -3064,9 +3212,8 @@ sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) (int)srp->data.cmd_opcode, sg_rq_st_str(SG_RS_INACTIVE, false)); } - if (list_empty(&fp->rq_list)) + if (xa_empty(&fp->srp_arr)) seq_puts(s, " No requests active\n"); - spin_unlock(&fp->rq_list_lock); } } From patchwork Thu Oct 15 02:06:25 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269802 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 7517EC4363D for ; Thu, 15 Oct 2020 02:07:26 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 2FA6222257 for ; Thu, 15 Oct 2020 02:07:26 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1730103AbgJOCHZ (ORCPT ); Wed, 14 Oct 2020 22:07:25 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40200 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732543AbgJOCHW (ORCPT ); Wed, 14 Oct 2020 22:07:22 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 3D010204274; Thu, 15 Oct 2020 04:07:21 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id tkzEiTWceCD9; Thu, 15 Oct 2020 04:07:19 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 2635220414F; Thu, 15 Oct 2020 04:07:14 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 26/44] sg: sense buffer rework Date: Wed, 14 Oct 2020 22:06:25 -0400 Message-Id: <20201015020643.432908-27-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org The biggest single item in the sg_request object is the sense buffer array which is SCSI_SENSE_BUFFERSIZE bytes long. That constant started out at 18 bytes 20 years ago and is 96 bytes now and might grow in the future. On the other hand the sense buffer is only used by a small number of SCSI commands: those that fail and those that want to return more information other than a SCSI status of GOOD. Set up a small mempool called "sg_sense" that is only used as required and released back to the mempool as soon as practical. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 114 +++++++++++++++++++++++++++++++++++----------- 1 file changed, 88 insertions(+), 26 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index eb33cb203cb2..e30fac7e1d86 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -72,6 +72,10 @@ static char *sg_version_date = "20190606"; */ #define SG_MAX_CDB_SIZE 252 +static struct kmem_cache *sg_sense_cache; +#define SG_MEMPOOL_MIN_NR 4 +static mempool_t *sg_sense_pool; + /* Following enum contains the states of sg_request::rq_st */ enum sg_rq_state { /* N.B. sg_rq_state_arr assumes SG_RS_AWAIT_RCV==2 */ SG_RS_INACTIVE = 0, /* request not in use (e.g. on fl) */ @@ -174,7 +178,6 @@ struct sg_fd; struct sg_request { /* active SCSI command or inactive request */ struct sg_scatter_hold sgat_h; /* hold buffer, perhaps scatter list */ struct sg_slice_hdr3 s_hdr3; /* subset of sg_io_hdr */ - u8 sense_b[SCSI_SENSE_BUFFERSIZE]; u32 duration; /* cmd duration in milliseconds */ u32 rq_flags; /* hold user supplied flags */ u32 rq_idx; /* my index within parent's srp_arr */ @@ -187,6 +190,7 @@ struct sg_request { /* active SCSI command or inactive request */ u8 cmd_opcode; /* first byte of SCSI cdb */ u64 start_ns; /* starting point of command duration calc */ unsigned long frq_bm[1]; /* see SG_FRQ_* defines above */ + u8 *sense_bp; /* mempool alloc-ed sense buffer, as needed */ struct sg_fd *parentfp; /* pointer to owning fd, even when on fl */ struct request *rq; /* released in sg_rq_end_io(), bio kept */ struct bio *bio; /* kept until this req -->SG_RS_INACTIVE */ @@ -971,18 +975,21 @@ sg_copy_sense(struct sg_request *srp) (driver_byte(srp->rq_result) & DRIVER_SENSE)) { int sb_len = min_t(int, SCSI_SENSE_BUFFERSIZE, srp->sense_len); int mx_sb_len = srp->s_hdr3.mx_sb_len; + u8 *sbp = srp->sense_bp; void __user *up = srp->s_hdr3.sbp; - if (up && mx_sb_len > 0) { + srp->sense_bp = NULL; + if (up && mx_sb_len > 0 && sbp) { sb_len = min_t(int, mx_sb_len, sb_len); /* Additional sense length field */ - sb_len_ret = 8 + (int)srp->sense_b[7]; + sb_len_ret = 8 + (int)sbp[7]; sb_len_ret = min_t(int, sb_len_ret, sb_len); - if (copy_to_user(up, srp->sense_b, sb_len_ret)) + if (copy_to_user(up, sbp, sb_len_ret)) sb_len_ret = -EFAULT; } else { sb_len_ret = 0; } + mempool_free(sbp, sg_sense_pool); } return sb_len_ret; } @@ -1073,8 +1080,14 @@ sg_read_v1v2(void __user *buf, int count, struct sg_fd *sfp, h2p->driver_status = driver_byte(rq_result); if ((CHECK_CONDITION & status_byte(rq_result)) || (DRIVER_SENSE & driver_byte(rq_result))) { - memcpy(h2p->sense_buffer, srp->sense_b, - sizeof(h2p->sense_buffer)); + if (srp->sense_bp) { + u8 *sbp = srp->sense_bp; + + srp->sense_bp = NULL; + memcpy(h2p->sense_buffer, sbp, + sizeof(h2p->sense_buffer)); + mempool_free(sbp, sg_sense_pool); + } } switch (host_byte(rq_result)) { /* @@ -1109,18 +1122,22 @@ sg_read_v1v2(void __user *buf, int count, struct sg_fd *sfp, /* Now copy the result back to the user buffer. */ if (count >= SZ_SG_HEADER) { - if (copy_to_user(buf, h2p, SZ_SG_HEADER)) - return -EFAULT; + if (copy_to_user(buf, h2p, SZ_SG_HEADER)) { + res = -EFAULT; + goto fini; + } buf += SZ_SG_HEADER; if (count > h2p->reply_len) count = h2p->reply_len; if (count > SZ_SG_HEADER) { - if (sg_read_append(srp, buf, count - SZ_SG_HEADER)) - return -EFAULT; + res = sg_read_append(srp, buf, count - SZ_SG_HEADER); + if (res) + goto fini; } } else { res = (h2p->result == 0) ? 0 : -EIO; } +fini: atomic_set(&srp->rq_st, SG_RS_RCV_DONE); sg_finish_scsi_blk_rq(srp); sg_deact_request(sfp, srp); @@ -2107,8 +2124,25 @@ sg_rq_end_io(struct request *rq, blk_status_t status) srp->duration = sg_calc_rq_dur(srp); if (unlikely((srp->rq_result & SG_ML_RESULT_MSK) && slen > 0)) sg_check_sense(sdp, srp, slen); - if (slen > 0) - memcpy(srp->sense_b, scsi_rp->sense, slen); + if (slen > 0) { + if (scsi_rp->sense) { + srp->sense_bp = mempool_alloc(sg_sense_pool, + GFP_ATOMIC); + if (srp->sense_bp) { + memcpy(srp->sense_bp, scsi_rp->sense, slen); + if (slen < SCSI_SENSE_BUFFERSIZE) + memset(srp->sense_bp + slen, 0, + SCSI_SENSE_BUFFERSIZE - slen); + } else { + slen = 0; + pr_warn("%s: sense but can't alloc buffer\n", + __func__); + } + } else { + slen = 0; + pr_warn("%s: sense_len>0 but sense==NULL\n", __func__); + } + } srp->sense_len = slen; if (unlikely(test_bit(SG_FRQ_IS_ORPHAN, srp->frq_bm))) { if (test_bit(SG_FFD_KEEP_ORPHAN, sfp->ffd_bm)) { @@ -2397,13 +2431,30 @@ init_sg(void) if (rc) return rc; + sg_sense_cache = kmem_cache_create_usercopy + ("sg_sense", SCSI_SENSE_BUFFERSIZE, 0, + SLAB_HWCACHE_ALIGN, 0, + SCSI_SENSE_BUFFERSIZE, NULL); + if (!sg_sense_cache) { + pr_err("sg: can't init sense cache\n"); + rc = -ENOMEM; + goto err_out_unreg; + } + sg_sense_pool = mempool_create_slab_pool(SG_MEMPOOL_MIN_NR, + sg_sense_cache); + if (!sg_sense_pool) { + pr_err("sg: can't init sense pool\n"); + rc = -ENOMEM; + goto err_out_cache; + } + pr_info("Registered %s[char major=0x%x], version: %s, date: %s\n", "sg device ", SCSI_GENERIC_MAJOR, SG_VERSION_STR, sg_version_date); sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic"); if (IS_ERR(sg_sysfs_class)) { rc = PTR_ERR(sg_sysfs_class); - goto err_out_unreg; + goto err_out_pool; } sg_sysfs_valid = true; rc = scsi_register_interface(&sg_interface); @@ -2413,6 +2464,10 @@ init_sg(void) } class_destroy(sg_sysfs_class); +err_out_pool: + mempool_destroy(sg_sense_pool); +err_out_cache: + kmem_cache_destroy(sg_sense_cache); err_out_unreg: unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS); return rc; @@ -2432,6 +2487,8 @@ exit_sg(void) if (IS_ENABLED(CONFIG_SCSI_PROC_FS)) remove_proc_subtree("scsi/sg", NULL); scsi_unregister_interface(&sg_interface); + mempool_destroy(sg_sense_pool); + kmem_cache_destroy(sg_sense_cache); class_destroy(sg_sysfs_class); sg_sysfs_valid = false; unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), @@ -2966,6 +3023,7 @@ sg_setup_req(struct sg_fd *sfp, int dxfr_len, struct sg_comm_wr_t *cwrp) int num_inactive = 0; unsigned long idx, last_idx, iflags; struct sg_request *r_srp = NULL; /* request to return */ + struct sg_request *last_srp = NULL; struct xarray *xafp = &sfp->srp_arr; __maybe_unused const char *cp; @@ -2982,19 +3040,17 @@ sg_setup_req(struct sg_fd *sfp, int dxfr_len, struct sg_comm_wr_t *cwrp) ++num_inactive; if (dxfr_len < SG_DEF_SECTOR_SZ) { last_idx = idx; + last_srp = r_srp; continue; } } /* If dxfr_len is small, use last inactive request */ - if (last_idx != ~0UL) { - idx = last_idx; - r_srp = xa_load(xafp, idx); - if (!r_srp) - goto start_again; + if (last_idx != ~0UL && last_srp) { + r_srp = last_srp; if (sg_rq_state_chg(r_srp, SG_RS_INACTIVE, SG_RS_BUSY, false, __func__)) goto start_again; /* gone to another thread */ - cp = "toward back of srp_arr"; + cp = "toward end of srp_arr"; found = true; } } else { @@ -3079,15 +3135,16 @@ sg_setup_req(struct sg_fd *sfp, int dxfr_len, struct sg_comm_wr_t *cwrp) static void sg_deact_request(struct sg_fd *sfp, struct sg_request *srp) { - unsigned long iflags; + u8 *sbp; if (WARN_ON(!sfp || !srp)) return; - atomic_set(&srp->rq_st, SG_RS_INACTIVE); - xa_lock_irqsave(&sfp->srp_arr, iflags); - __xa_set_mark(&sfp->srp_arr, srp->rq_idx, SG_XA_RQ_INACTIVE); - __xa_clear_mark(&sfp->srp_arr, srp->rq_idx, SG_XA_RQ_AWAIT); - xa_unlock_irqrestore(&sfp->srp_arr, iflags); + sbp = srp->sense_bp; + srp->sense_bp = NULL; + sg_rq_state_chg(srp, 0, SG_RS_INACTIVE, true /* force */, __func__); + /* maybe orphaned req, thus never read */ + if (sbp) + mempool_free(sbp, sg_sense_pool); } /* Returns pointer to sg_fd object or negated errno twisted by ERR_PTR */ @@ -3236,7 +3293,12 @@ sg_remove_sfp_usercontext(struct work_struct *work) continue; if (!xa_get_mark(xafp, srp->rq_idx, SG_XA_RQ_INACTIVE)) sg_finish_scsi_blk_rq(srp); - sg_remove_sgat(srp); + if (srp->sgat_h.buflen > 0) + sg_remove_sgat(srp); + if (srp->sense_bp) { + mempool_free(srp->sense_bp, sg_sense_pool); + srp->sense_bp = NULL; + } xa_lock_irqsave(xafp, iflags); e_srp = __xa_erase(xafp, srp->rq_idx); xa_unlock_irqrestore(xafp, iflags); From patchwork Thu Oct 15 02:06:26 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269799 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id BAF85C433E7 for ; Thu, 15 Oct 2020 02:07:37 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 632D222257 for ; Thu, 15 Oct 2020 02:07:37 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729585AbgJOCHh (ORCPT ); Wed, 14 Oct 2020 22:07:37 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40208 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732546AbgJOCH0 (ORCPT ); Wed, 14 Oct 2020 22:07:26 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 83B9720424C; Thu, 15 Oct 2020 04:07:23 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id WzXBF-54mZrI; Thu, 15 Oct 2020 04:07:21 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 414C0204247; Thu, 15 Oct 2020 04:07:15 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 27/44] sg: add sg v4 interface support Date: Wed, 14 Oct 2020 22:06:26 -0400 Message-Id: <20201015020643.432908-28-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Add support for the sg v4 interface based on struct sg_io_v4 found in include/uapi/linux/bsg.h and only previously supported by the bsg driver. Add ioctl(SG_IOSUBMIT) and ioctl(SG_IORECEIVE) for async (non-blocking) usage of the sg v4 interface. Do not accept the v3 interface with these ioctls. Do not accept the v4 interface with this driver's existing write() and read() system calls. For sync (blocking) usage expand the existing ioctl(SG_IO) to additionally accept the sg v4 interface object. Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 442 +++++++++++++++++++++++++++++++++-------- include/uapi/scsi/sg.h | 37 +++- 2 files changed, 397 insertions(+), 82 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index e30fac7e1d86..0e284e38d97d 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -40,11 +40,12 @@ static char *sg_version_date = "20190606"; #include #include #include -#include /* for sg_check_file_access() */ +#include /* for sg_check_file_access() */ #include #include -#include "scsi.h" +#include +#include #include #include #include @@ -76,6 +77,9 @@ static struct kmem_cache *sg_sense_cache; #define SG_MEMPOOL_MIN_NR 4 static mempool_t *sg_sense_pool; +#define uptr64(usp_val) ((void __user *)(uintptr_t)(usp_val)) +#define cuptr64(usp_val) ((const void __user *)(uintptr_t)(usp_val)) + /* Following enum contains the states of sg_request::rq_st */ enum sg_rq_state { /* N.B. sg_rq_state_arr assumes SG_RS_AWAIT_RCV==2 */ SG_RS_INACTIVE = 0, /* request not in use (e.g. on fl) */ @@ -100,6 +104,7 @@ enum sg_rq_state { /* N.B. sg_rq_state_arr assumes SG_RS_AWAIT_RCV==2 */ #define SG_ADD_RQ_MAX_RETRIES 40 /* to stop infinite _trylock(s) */ /* Bit positions (flags) for sg_request::frq_bm bitmask follow */ +#define SG_FRQ_IS_V4I 0 /* true (set) when is v4 interface */ #define SG_FRQ_IS_ORPHAN 1 /* owner of request gone */ #define SG_FRQ_SYNC_INVOC 2 /* synchronous (blocking) invocation */ #define SG_FRQ_NO_US_XFER 3 /* no user space transfer of data */ @@ -164,6 +169,15 @@ struct sg_slice_hdr3 { void __user *usr_ptr; }; +struct sg_slice_hdr4 { /* parts of sg_io_v4 object needed in async usage */ + void __user *sbp; /* derived from sg_io_v4::response */ + u64 usr_ptr; /* hold sg_io_v4::usr_ptr as given (u64) */ + int out_resid; + s16 dir; /* data xfer direction; SG_DXFER_* */ + u16 cmd_len; /* truncated of sg_io_v4::request_len */ + u16 max_sb_len; /* truncated of sg_io_v4::max_response_len */ +}; + struct sg_scatter_hold { /* holding area for scsi scatter gather info */ struct page **pages; /* num_sgat element array of struct page* */ int buflen; /* capacity in bytes (dlen<=buflen) */ @@ -177,7 +191,10 @@ struct sg_fd; struct sg_request { /* active SCSI command or inactive request */ struct sg_scatter_hold sgat_h; /* hold buffer, perhaps scatter list */ - struct sg_slice_hdr3 s_hdr3; /* subset of sg_io_hdr */ + union { + struct sg_slice_hdr3 s_hdr3; /* subset of sg_io_hdr */ + struct sg_slice_hdr4 s_hdr4; /* reduced size struct sg_io_v4 */ + }; u32 duration; /* cmd duration in milliseconds */ u32 rq_flags; /* hold user supplied flags */ u32 rq_idx; /* my index within parent's srp_arr */ @@ -237,7 +254,10 @@ struct sg_device { /* holds the state of each scsi generic device */ struct sg_comm_wr_t { /* arguments to sg_common_write() */ int timeout; unsigned long frq_bm[1]; /* see SG_FRQ_* defines above */ - struct sg_io_hdr *h3p; + union { /* selector is frq_bm.SG_FRQ_IS_V4I */ + struct sg_io_hdr *h3p; + struct sg_io_v4 *h4p; + }; u8 *cmnd; }; @@ -246,12 +266,12 @@ static void sg_rq_end_io(struct request *rq, blk_status_t status); /* Declarations of other static functions used before they are defined */ static int sg_proc_init(void); static int sg_start_req(struct sg_request *srp, u8 *cmd, int cmd_len, - int dxfer_dir); + struct sg_io_v4 *h4p, int dxfer_dir); static void sg_finish_scsi_blk_rq(struct sg_request *srp); static int sg_mk_sgat(struct sg_request *srp, struct sg_fd *sfp, int minlen); -static int sg_submit(struct file *filp, struct sg_fd *sfp, - struct sg_io_hdr *hp, bool sync, - struct sg_request **o_srp); +static int sg_v3_submit(struct file *filp, struct sg_fd *sfp, + struct sg_io_hdr *hp, bool sync, + struct sg_request **o_srp); static struct sg_request *sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp); static int sg_read_append(struct sg_request *srp, void __user *outp, @@ -259,11 +279,11 @@ static int sg_read_append(struct sg_request *srp, void __user *outp, static void sg_remove_sgat(struct sg_request *srp); static struct sg_fd *sg_add_sfp(struct sg_device *sdp); static void sg_remove_sfp(struct kref *); -static struct sg_request *sg_find_srp_by_id(struct sg_fd *sfp, int pack_id); +static struct sg_request *sg_find_srp_by_id(struct sg_fd *sfp, int id); static struct sg_request *sg_setup_req(struct sg_fd *sfp, int dxfr_len, struct sg_comm_wr_t *cwrp); static void sg_deact_request(struct sg_fd *sfp, struct sg_request *srp); -static struct sg_device *sg_get_dev(int dev); +static struct sg_device *sg_get_dev(int min_dev); static void sg_device_destroy(struct kref *kref); static struct sg_request *sg_mk_srp_sgat(struct sg_fd *sfp, bool first, int db_len); @@ -273,8 +293,11 @@ static const char *sg_rq_st_str(enum sg_rq_state rq_st, bool long_str); #define SZ_SG_HEADER ((int)sizeof(struct sg_header)) /* v1 and v2 header */ #define SZ_SG_IO_HDR ((int)sizeof(struct sg_io_hdr)) /* v3 header */ +#define SZ_SG_IO_V4 ((int)sizeof(struct sg_io_v4)) /* v4 header (in bsg.h) */ #define SZ_SG_REQ_INFO ((int)sizeof(struct sg_req_info)) +/* There is a assert that SZ_SG_IO_V4 >= SZ_SG_IO_HDR in first function */ + #define SG_IS_DETACHING(sdp) test_bit(SG_FDEV_DETACHING, (sdp)->fdev_bm) #define SG_HAVE_EXCLUDE(sdp) test_bit(SG_FDEV_EXCLUDE, (sdp)->fdev_bm) #define SG_RS_ACTIVE(srp) (atomic_read(&(srp)->rq_st) != SG_RS_INACTIVE) @@ -331,6 +354,10 @@ static const char *sg_rq_st_str(enum sg_rq_state rq_st, bool long_str); static int sg_check_file_access(struct file *filp, const char *caller) { + /* can't put following in declarations where it belongs */ + compiletime_assert(SZ_SG_IO_V4 >= SZ_SG_IO_HDR, + "struct sg_io_v4 should be larger than sg_io_hdr"); + if (filp->f_cred != current_real_cred()) { pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n", caller, task_tgid_vnr(current), current->comm); @@ -425,21 +452,18 @@ sg_open(struct inode *inode, struct file *filp) o_excl = !!(op_flags & O_EXCL); non_block = !!(op_flags & O_NONBLOCK); if (o_excl && ((op_flags & O_ACCMODE) == O_RDONLY)) - return -EPERM; /* Can't lock it with read only access */ + return -EPERM;/* not permitted, need write access for O_EXCL */ sdp = sg_get_dev(min_dev); /* increments sdp->d_ref */ if (IS_ERR(sdp)) return PTR_ERR(sdp); - /* This driver's module count bumped by fops_get in */ /* Prevent the device driver from vanishing while we sleep */ res = scsi_device_get(sdp->device); if (res) goto sg_put; - res = scsi_autopm_get_device(sdp->device); if (res) goto sdp_put; - res = sg_allow_if_err_recovery(sdp, non_block); if (res) goto error_out; @@ -476,9 +500,10 @@ sg_open(struct inode *inode, struct file *filp) } filp->private_data = sfp; + sfp->tid = (current ? current->pid : -1); mutex_unlock(&sdp->open_rel_lock); - SG_LOG(3, sfp, "%s: minor=%d, op_flags=0x%x; %s count after=%d%s\n", - __func__, min_dev, op_flags, "device open", o_count, + SG_LOG(3, sfp, "%s: o_count after=%d on minor=%d, op_flags=0x%x%s\n", + __func__, o_count, min_dev, op_flags, ((op_flags & O_NONBLOCK) ? " O_NONBLOCK" : "")); res = 0; @@ -501,8 +526,13 @@ sg_open(struct inode *inode, struct file *filp) goto sg_put; } -/* Release resources associated with a successful sg_open() - * Returns 0 on success, else a negated errno value */ +/* + * Release resources associated with a prior, successful sg_open(). It can be + * seen as the (final) close() call on a sg device file descriptor in the user + * space. The real work releasing all resources associated with this file + * descriptor is done by sg_remove_sfp_usercontext() which is scheduled by + * sg_remove_sfp(). + */ static int sg_release(struct inode *inode, struct file *filp) { @@ -525,8 +555,6 @@ sg_release(struct inode *inode, struct file *filp) * Possibly many open()s waiting on exlude clearing, start many; * only open(O_EXCL)'s wait when open_cnt<2 and only start one. */ - /* possibly many open()s waiting on exlude clearing, start many; - * only open(O_EXCL)s wait on 0==open_cnt so only start one */ if (test_and_clear_bit(SG_FDEV_EXCLUDE, sdp->fdev_bm)) wake_up_interruptible_all(&sdp->open_wait); else if (o_count < 2) @@ -606,7 +634,7 @@ sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) __func__); return -EPERM; } - res = sg_submit(filp, sfp, h3p, false, NULL); + res = sg_v3_submit(filp, sfp, h3p, false, NULL); return res < 0 ? res : (int)count; } to_v2: @@ -683,7 +711,7 @@ sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) static inline int sg_chk_mmap(struct sg_fd *sfp, int rq_flags, int len) { - if (!xa_empty(&sfp->srp_arr)) + if (atomic_read(&sfp->submitted) > 0) return -EBUSY; /* already active requests on fd */ if (len > sfp->rsv_srp->sgat_h.buflen) return -ENOMEM; /* MMAP_IO size must fit in reserve */ @@ -714,8 +742,8 @@ sg_fetch_cmnd(struct file *filp, struct sg_fd *sfp, const u8 __user *u_cdbp, } static int -sg_submit(struct file *filp, struct sg_fd *sfp, struct sg_io_hdr *hp, - bool sync, struct sg_request **o_srp) +sg_v3_submit(struct file *filp, struct sg_fd *sfp, struct sg_io_hdr *hp, + bool sync, struct sg_request **o_srp) { int res, timeout; unsigned long ul_timeout; @@ -749,6 +777,67 @@ sg_submit(struct file *filp, struct sg_fd *sfp, struct sg_io_hdr *hp, return 0; } +static int +sg_submit_v4(struct file *filp, struct sg_fd *sfp, void __user *p, + struct sg_io_v4 *h4p, bool sync, struct sg_request **o_srp) +{ + int timeout, res; + unsigned long ul_timeout; + struct sg_request *srp; + struct sg_comm_wr_t cwr; + u8 cmnd[SG_MAX_CDB_SIZE]; + + if (h4p->flags & SG_FLAG_MMAP_IO) { + int len = 0; + + if (h4p->din_xferp) + len = h4p->din_xfer_len; + else if (h4p->dout_xferp) + len = h4p->dout_xfer_len; + res = sg_chk_mmap(sfp, h4p->flags, len); + if (res) + return res; + } + /* once v4 (or v3) seen, allow cmd_q on this fd (def: no cmd_q) */ + set_bit(SG_FFD_CMD_Q, sfp->ffd_bm); + ul_timeout = msecs_to_jiffies(h4p->timeout); + timeout = min_t(unsigned long, ul_timeout, INT_MAX); + res = sg_fetch_cmnd(filp, sfp, cuptr64(h4p->request), h4p->request_len, + cmnd); + if (res) + return res; + cwr.frq_bm[0] = 0; + assign_bit(SG_FRQ_SYNC_INVOC, cwr.frq_bm, (int)sync); + set_bit(SG_FRQ_IS_V4I, cwr.frq_bm); + cwr.h4p = h4p; + cwr.timeout = timeout; + cwr.cmnd = cmnd; + srp = sg_common_write(sfp, &cwr); + if (IS_ERR(srp)) + return PTR_ERR(srp); + if (o_srp) + *o_srp = srp; + return res; +} + +static int +sg_ctl_iosubmit(struct file *filp, struct sg_fd *sfp, void __user *p) +{ + int res; + u8 hdr_store[SZ_SG_IO_V4]; + struct sg_io_v4 *h4p = (struct sg_io_v4 *)hdr_store; + struct sg_device *sdp = sfp->parentdp; + + res = sg_allow_if_err_recovery(sdp, (filp->f_flags & O_NONBLOCK)); + if (res) + return res; + if (copy_from_user(hdr_store, p, SZ_SG_IO_V4)) + return -EFAULT; + if (h4p->guard == 'Q') + return sg_submit_v4(filp, sfp, p, h4p, false, NULL); + return -EPERM; +} + #if IS_ENABLED(SG_LOG_ACTIVE) static void sg_rq_state_fail_msg(struct sg_fd *sfp, enum sg_rq_state exp_old_st, @@ -858,16 +947,47 @@ sg_rq_state_chg(struct sg_request *srp, enum sg_rq_state old_st, return 0; } +static void +sg_execute_cmd(struct sg_fd *sfp, struct sg_request *srp) +{ + bool at_head, is_v4h, sync; + struct sg_device *sdp = sfp->parentdp; + + is_v4h = test_bit(SG_FRQ_IS_V4I, srp->frq_bm); + sync = test_bit(SG_FRQ_SYNC_INVOC, srp->frq_bm); + SG_LOG(3, sfp, "%s: is_v4h=%d\n", __func__, (int)is_v4h); + srp->start_ns = ktime_get_boottime_ns(); + srp->duration = 0; + + if (!is_v4h && srp->s_hdr3.interface_id == '\0') + at_head = true; /* backward compatibility: v1+v2 interfaces */ + else if (test_bit(SG_FFD_Q_AT_TAIL, sfp->ffd_bm)) + /* cmd flags can override sfd setting */ + at_head = !!(srp->rq_flags & SG_FLAG_Q_AT_HEAD); + else /* this sfd is defaulting to head */ + at_head = !(srp->rq_flags & SG_FLAG_Q_AT_TAIL); + + kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */ + sg_rq_state_chg(srp, SG_RS_BUSY /* ignored */, SG_RS_INFLIGHT, + true, __func__); + + /* >>>>>>> send cmd/req off to other levels <<<<<<<< */ + if (!sync) + atomic_inc(&sfp->submitted); + blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk, + srp->rq, (int)at_head, sg_rq_end_io); +} + /* * All writes and submits converge on this function to launch the SCSI * command/request (via blk_execute_rq_nowait). Returns a pointer to a * sg_request object holding the request just issued or a negated errno * value twisted by ERR_PTR. + * N.B. pack_id placed in sg_io_v4::request_extra field. */ static struct sg_request * sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) { - bool at_head; int res = 0; int dxfr_len, dir, cmd_len; int pack_id = SG_PACK_ID_WILDCARD; @@ -875,12 +995,32 @@ sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) struct sg_device *sdp = sfp->parentdp; struct sg_request *srp; struct sg_io_hdr *hi_p; - - hi_p = cwrp->h3p; - dir = hi_p->dxfer_direction; - dxfr_len = hi_p->dxfer_len; - rq_flags = hi_p->flags; - pack_id = hi_p->pack_id; + struct sg_io_v4 *h4p; + + if (test_bit(SG_FRQ_IS_V4I, cwrp->frq_bm)) { + h4p = cwrp->h4p; + hi_p = NULL; + dxfr_len = 0; + dir = SG_DXFER_NONE; + rq_flags = h4p->flags; + pack_id = h4p->request_extra; + if (h4p->din_xfer_len && h4p->dout_xfer_len) { + return ERR_PTR(-EOPNOTSUPP); + } else if (h4p->din_xfer_len) { + dxfr_len = h4p->din_xfer_len; + dir = SG_DXFER_FROM_DEV; + } else if (h4p->dout_xfer_len) { + dxfr_len = h4p->dout_xfer_len; + dir = SG_DXFER_TO_DEV; + } + } else { /* sg v3 interface so hi_p valid */ + h4p = NULL; + hi_p = cwrp->h3p; + dir = hi_p->dxfer_direction; + dxfr_len = hi_p->dxfer_len; + rq_flags = hi_p->flags; + pack_id = hi_p->pack_id; + } if (dxfr_len >= SZ_256M) return ERR_PTR(-EINVAL); @@ -890,13 +1030,23 @@ sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) srp->rq_flags = rq_flags; srp->pack_id = pack_id; - cmd_len = hi_p->cmd_len; - memcpy(&srp->s_hdr3, hi_p, sizeof(srp->s_hdr3)); + if (h4p) { + memset(&srp->s_hdr4, 0, sizeof(srp->s_hdr4)); + srp->s_hdr4.usr_ptr = h4p->usr_ptr; + srp->s_hdr4.sbp = uptr64(h4p->response); + srp->s_hdr4.max_sb_len = h4p->max_response_len; + srp->s_hdr4.cmd_len = h4p->request_len; + srp->s_hdr4.dir = dir; + cmd_len = h4p->request_len; + } else { /* v3 interface active */ + cmd_len = hi_p->cmd_len; + memcpy(&srp->s_hdr3, hi_p, sizeof(srp->s_hdr3)); + } srp->cmd_opcode = cwrp->cmnd[0];/* hold opcode of command for debug */ SG_LOG(4, sfp, "%s: opcode=0x%02x, cdb_sz=%d, pack_id=%d\n", __func__, (int)cwrp->cmnd[0], cmd_len, pack_id); - res = sg_start_req(srp, cwrp->cmnd, cmd_len, dir); + res = sg_start_req(srp, cwrp->cmnd, cmd_len, h4p, dir); if (res < 0) /* probably out of space --> -ENOMEM */ goto err_out; if (unlikely(SG_IS_DETACHING(sdp))) { @@ -913,25 +1063,7 @@ sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) goto err_out; } srp->rq->timeout = cwrp->timeout; - kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */ - res = sg_rq_state_chg(srp, SG_RS_BUSY, SG_RS_INFLIGHT, false, - __func__); - if (res) - goto err_out; - srp->start_ns = ktime_get_boottime_ns(); - srp->duration = 0; - - if (srp->s_hdr3.interface_id == '\0') - at_head = true; /* backward compatibility: v1+v2 interfaces */ - else if (test_bit(SG_FFD_Q_AT_TAIL, sfp->ffd_bm)) - /* cmd flags can override sfd setting */ - at_head = !!(srp->rq_flags & SG_FLAG_Q_AT_HEAD); - else /* this sfd is defaulting to head */ - at_head = !(srp->rq_flags & SG_FLAG_Q_AT_TAIL); - if (!test_bit(SG_FRQ_SYNC_INVOC, srp->frq_bm)) - atomic_inc(&sfp->submitted); - blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk, - srp->rq, at_head, sg_rq_end_io); + sg_execute_cmd(sfp, srp); return srp; err_out: sg_finish_scsi_blk_rq(srp); @@ -944,7 +1076,6 @@ sg_common_write(struct sg_fd *sfp, struct sg_comm_wr_t *cwrp) * sg_ctl_ioreceive(). wait_event_interruptible will return if this one * returns true (or an event like a signal (e.g. control-C) occurs). */ - static inline bool sg_get_ready_srp(struct sg_fd *sfp, struct sg_request **srpp, int pack_id) { @@ -964,7 +1095,7 @@ sg_get_ready_srp(struct sg_fd *sfp, struct sg_request **srpp, int pack_id) * negated errno value. */ static int -sg_copy_sense(struct sg_request *srp) +sg_copy_sense(struct sg_request *srp, bool v4_active) { int sb_len_ret = 0; int scsi_stat; @@ -974,11 +1105,18 @@ sg_copy_sense(struct sg_request *srp) if ((scsi_stat & SAM_STAT_CHECK_CONDITION) || (driver_byte(srp->rq_result) & DRIVER_SENSE)) { int sb_len = min_t(int, SCSI_SENSE_BUFFERSIZE, srp->sense_len); - int mx_sb_len = srp->s_hdr3.mx_sb_len; + int mx_sb_len; u8 *sbp = srp->sense_bp; - void __user *up = srp->s_hdr3.sbp; + void __user *up; srp->sense_bp = NULL; + if (v4_active) { + up = uptr64(srp->s_hdr4.sbp); + mx_sb_len = srp->s_hdr4.max_sb_len; + } else { + up = (void __user *)srp->s_hdr3.sbp; + mx_sb_len = srp->s_hdr3.mx_sb_len; + } if (up && mx_sb_len > 0 && sbp) { sb_len = min_t(int, mx_sb_len, sb_len); /* Additional sense length field */ @@ -995,14 +1133,16 @@ sg_copy_sense(struct sg_request *srp) } static int -sg_rec_state_v3(struct sg_fd *sfp, struct sg_request *srp) +sg_rec_state_v3v4(struct sg_fd *sfp, struct sg_request *srp, bool v4_active) { - int sb_len_wr; u32 rq_res = srp->rq_result; - sb_len_wr = sg_copy_sense(srp); - if (sb_len_wr < 0) - return sb_len_wr; + if (unlikely(srp->rq_result & 0xff)) { + int sb_len_wr = sg_copy_sense(srp, v4_active); + + if (sb_len_wr < 0) + return sb_len_wr; + } if (rq_res & SG_ML_RESULT_MSK) srp->rq_info |= SG_INFO_CHECK; if (unlikely(SG_IS_DETACHING(sfp->parentdp))) @@ -1029,7 +1169,7 @@ sg_receive_v3(struct sg_fd *sfp, struct sg_request *srp, size_t count, goto err_out; } SG_LOG(3, sfp, "%s: srp=0x%pK\n", __func__, srp); - err = sg_rec_state_v3(sfp, srp); + err = sg_rec_state_v3v4(sfp, srp, false); memset(hp, 0, sizeof(*hp)); memcpy(hp, &srp->s_hdr3, sizeof(srp->s_hdr3)); hp->sb_len_wr = srp->sense_len; @@ -1053,11 +1193,103 @@ sg_receive_v3(struct sg_fd *sfp, struct sg_request *srp, size_t count, return err; } +static int +sg_receive_v4(struct sg_fd *sfp, struct sg_request *srp, void __user *p, + struct sg_io_v4 *h4p) +{ + int err, err2; + u32 rq_result = srp->rq_result; + + SG_LOG(3, sfp, "%s: p=%s, h4p=%s\n", __func__, + (p ? "given" : "NULL"), (h4p ? "given" : "NULL")); + err = sg_rec_state_v3v4(sfp, srp, true); + h4p->guard = 'Q'; + h4p->protocol = 0; + h4p->subprotocol = 0; + h4p->device_status = rq_result & 0xff; + h4p->driver_status = driver_byte(rq_result); + h4p->transport_status = host_byte(rq_result); + h4p->response_len = srp->sense_len; + h4p->info = srp->rq_info; + h4p->flags = srp->rq_flags; + h4p->duration = srp->duration; + switch (srp->s_hdr4.dir) { + case SG_DXFER_FROM_DEV: + h4p->din_xfer_len = srp->sgat_h.dlen; + break; + case SG_DXFER_TO_DEV: + h4p->dout_xfer_len = srp->sgat_h.dlen; + break; + default: + break; + } + h4p->din_resid = srp->in_resid; + h4p->dout_resid = srp->s_hdr4.out_resid; + h4p->usr_ptr = srp->s_hdr4.usr_ptr; + h4p->response = (u64)srp->s_hdr4.sbp; + h4p->request_extra = srp->pack_id; + if (p) { + if (copy_to_user(p, h4p, SZ_SG_IO_V4)) + err = err ? err : -EFAULT; + } + err2 = sg_rq_state_chg(srp, atomic_read(&srp->rq_st), SG_RS_RCV_DONE, + false, __func__); + if (err2) + err = err ? err : err2; + sg_finish_scsi_blk_rq(srp); + sg_deact_request(sfp, srp); + return err < 0 ? err : 0; +} + /* - * Completes a v3 request/command. Called from sg_read {v2 or v3}, - * ioctl(SG_IO) {for v3}, or from ioctl(SG_IORECEIVE) when its - * completing a v3 request/command. + * Called when ioctl(SG_IORECEIVE) received. Expects a v4 interface object. + * Checks if O_NONBLOCK file flag given, if not checks given 'flags' field + * to see if SGV4_FLAG_IMMED is set. Either of these implies non blocking. + * When non-blocking and there is no request waiting, yields EAGAIN; + * otherwise it waits (i.e. it "blocks"). */ +static int +sg_ctl_ioreceive(struct file *filp, struct sg_fd *sfp, void __user *p) +{ + bool non_block = !!(filp->f_flags & O_NONBLOCK); + int res, id; + int pack_id = SG_PACK_ID_WILDCARD; + u8 v4_holder[SZ_SG_IO_V4]; + struct sg_io_v4 *h4p = (struct sg_io_v4 *)v4_holder; + struct sg_device *sdp = sfp->parentdp; + struct sg_request *srp; + + res = sg_allow_if_err_recovery(sdp, non_block); + if (res) + return res; + /* Get first three 32 bit integers: guard, proto+subproto */ + if (copy_from_user(h4p, p, SZ_SG_IO_V4)) + return -EFAULT; + /* for v4: protocol=0 --> SCSI; subprotocol=0 --> SPC++ */ + if (h4p->guard != 'Q' || h4p->protocol != 0 || h4p->subprotocol != 0) + return -EPERM; + if (h4p->flags & SGV4_FLAG_IMMED) + non_block = true; /* set by either this or O_NONBLOCK */ + SG_LOG(3, sfp, "%s: non_block(+IMMED)=%d\n", __func__, non_block); + /* read in part of v3 or v4 header for pack_id or tag based find */ + id = pack_id; + srp = sg_find_srp_by_id(sfp, id); + if (!srp) { /* nothing available so wait on packet or */ + if (unlikely(SG_IS_DETACHING(sdp))) + return -ENODEV; + if (non_block) + return -EAGAIN; + res = wait_event_interruptible(sfp->read_wait, + sg_get_ready_srp(sfp, &srp, + id)); + if (unlikely(SG_IS_DETACHING(sdp))) + return -ENODEV; + if (res) /* -ERESTARTSYS as signal hit process */ + return res; + } /* now srp should be valid */ + return sg_receive_v4(sfp, srp, p, h4p); +} + static int sg_read_v1v2(void __user *buf, int count, struct sg_fd *sfp, struct sg_request *srp) @@ -1334,6 +1566,8 @@ sg_fill_request_element(struct sg_fd *sfp, struct sg_request *srp, rip->sg_io_owned = test_bit(SG_FRQ_SYNC_INVOC, srp->frq_bm); rip->problem = !!(srp->rq_result & SG_ML_RESULT_MSK); rip->pack_id = srp->pack_id; + rip->usr_ptr = test_bit(SG_FRQ_IS_V4I, srp->frq_bm) ? + uptr64(srp->s_hdr4.usr_ptr) : srp->s_hdr3.usr_ptr; rip->usr_ptr = srp->s_hdr3.usr_ptr; xa_unlock_irqrestore(&sfp->srp_arr, iflags); } @@ -1351,7 +1585,7 @@ sg_rq_landed(struct sg_device *sdp, struct sg_request *srp) */ static int sg_wait_event_srp(struct file *filp, struct sg_fd *sfp, void __user *p, - struct sg_request *srp) + struct sg_io_v4 *h4p, struct sg_request *srp) { int res; enum sg_rq_state sr_st; @@ -1379,7 +1613,10 @@ sg_wait_event_srp(struct file *filp, struct sg_fd *sfp, void __user *p, res = sg_rq_state_chg(srp, sr_st, SG_RS_BUSY, false, __func__); if (unlikely(res)) return res; - res = sg_receive_v3(sfp, srp, SZ_SG_IO_HDR, p); + if (test_bit(SG_FRQ_IS_V4I, srp->frq_bm)) + res = sg_receive_v4(sfp, srp, p, h4p); + else + res = sg_receive_v3(sfp, srp, SZ_SG_IO_HDR, p); return (res < 0) ? res : 0; } @@ -1393,8 +1630,9 @@ sg_ctl_sg_io(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, { int res; struct sg_request *srp = NULL; - u8 hu8arr[SZ_SG_IO_HDR]; + u8 hu8arr[SZ_SG_IO_V4]; struct sg_io_hdr *h3p = (struct sg_io_hdr *)hu8arr; + struct sg_io_v4 *h4p = (struct sg_io_v4 *)hu8arr; SG_LOG(3, sfp, "%s: SG_IO%s\n", __func__, ((filp->f_flags & O_NONBLOCK) ? " O_NONBLOCK ignored" : "")); @@ -1403,15 +1641,25 @@ sg_ctl_sg_io(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, return res; if (get_sg_io_hdr(h3p, p)) return -EFAULT; - if (h3p->interface_id == 'S') - res = sg_submit(filp, sfp, h3p, true, &srp); - else + if (h3p->interface_id == 'Q') { + /* copy in rest of sg_io_v4 object */ + if (copy_from_user(hu8arr + SZ_SG_IO_HDR, + ((u8 __user *)p) + SZ_SG_IO_HDR, + SZ_SG_IO_V4 - SZ_SG_IO_HDR)) + return -EFAULT; + res = sg_submit_v4(filp, sfp, p, h4p, true, &srp); + } else if (h3p->interface_id == 'S') { + res = sg_v3_submit(filp, sfp, h3p, true, &srp); + } else { + pr_info_once("sg: %s: v3 or v4 interface only here\n", + __func__); return -EPERM; + } if (unlikely(res < 0)) return res; if (!srp) /* mrq case: already processed all responses */ return res; - res = sg_wait_event_srp(filp, sfp, p, srp); + res = sg_wait_event_srp(filp, sfp, p, h4p, srp); if (res) SG_LOG(1, sfp, "%s: %s=0x%pK state: %s\n", __func__, "unexpected srp", srp, @@ -1629,6 +1877,12 @@ sg_ioctl_common(struct file *filp, struct sg_device *sdp, struct sg_fd *sfp, switch (cmd_in) { case SG_IO: return sg_ctl_sg_io(filp, sdp, sfp, p); + case SG_IOSUBMIT: + SG_LOG(3, sfp, "%s: SG_IOSUBMIT\n", __func__); + return sg_ctl_iosubmit(filp, sfp, p); + case SG_IORECEIVE: + SG_LOG(3, sfp, "%s: SG_IORECEIVE\n", __func__); + return sg_ctl_ioreceive(filp, sfp, p); case SG_GET_SCSI_ID: return sg_ctl_scsi_id(sdev, sfp, p); case SG_SET_FORCE_PACK_ID: @@ -2116,8 +2370,16 @@ sg_rq_end_io(struct request *rq, blk_status_t status) slen = min_t(int, scsi_rp->sense_len, SCSI_SENSE_BUFFERSIZE); a_resid = scsi_rp->resid_len; - if (a_resid) - srp->in_resid = a_resid; + if (a_resid) { + if (test_bit(SG_FRQ_IS_V4I, srp->frq_bm)) { + if (rq_data_dir(rq) == READ) + srp->in_resid = a_resid; + else + srp->s_hdr4.out_resid = a_resid; + } else { + srp->in_resid = a_resid; + } + } SG_LOG(6, sfp, "%s: pack_id=%d, res=0x%x\n", __func__, srp->pack_id, srp->rq_result); @@ -2522,7 +2784,8 @@ sg_set_map_data(const struct sg_scatter_hold *schp, bool up_valid, } static int -sg_start_req(struct sg_request *srp, u8 *cmd, int cmd_len, int dxfer_dir) +sg_start_req(struct sg_request *srp, u8 *cmd, int cmd_len, + struct sg_io_v4 *h4p, int dxfer_dir) { bool reserved, us_xfer; int res = 0; @@ -2539,7 +2802,6 @@ sg_start_req(struct sg_request *srp, u8 *cmd, int cmd_len, int dxfer_dir) struct rq_map_data *md = (void *)srp; /* want any non-NULL value */ u8 *long_cmdp = NULL; __maybe_unused const char *cp = ""; - struct sg_slice_hdr3 *sh3p = &srp->s_hdr3; struct rq_map_data map_data; sdp = sfp->parentdp; @@ -2549,10 +2811,28 @@ sg_start_req(struct sg_request *srp, u8 *cmd, int cmd_len, int dxfer_dir) return -ENOMEM; SG_LOG(5, sfp, "%s: long_cmdp=0x%pK ++\n", __func__, long_cmdp); } - up = sh3p->dxferp; - dxfer_len = (int)sh3p->dxfer_len; - iov_count = sh3p->iovec_count; - r0w = dxfer_dir == SG_DXFER_TO_DEV ? WRITE : READ; + if (h4p) { + if (dxfer_dir == SG_DXFER_TO_DEV) { + r0w = WRITE; + up = uptr64(h4p->dout_xferp); + dxfer_len = (int)h4p->dout_xfer_len; + iov_count = h4p->dout_iovec_count; + } else if (dxfer_dir == SG_DXFER_FROM_DEV) { + r0w = READ; + up = uptr64(h4p->din_xferp); + dxfer_len = (int)h4p->din_xfer_len; + iov_count = h4p->din_iovec_count; + } else { + up = NULL; + } + } else { + struct sg_slice_hdr3 *sh3p = &srp->s_hdr3; + + up = sh3p->dxferp; + dxfer_len = (int)sh3p->dxfer_len; + iov_count = sh3p->iovec_count; + r0w = dxfer_dir == SG_DXFER_TO_DEV ? WRITE : READ; + } SG_LOG(4, sfp, "%s: dxfer_len=%d, data-%s\n", __func__, dxfer_len, (r0w ? "OUT" : "IN")); q = sdp->device->request_queue; diff --git a/include/uapi/scsi/sg.h b/include/uapi/scsi/sg.h index 9bd37c6e4f44..51b86d20a7a5 100644 --- a/include/uapi/scsi/sg.h +++ b/include/uapi/scsi/sg.h @@ -99,6 +99,18 @@ typedef struct sg_io_hdr { #define SG_FLAG_Q_AT_TAIL 0x10 #define SG_FLAG_Q_AT_HEAD 0x20 +/* + * Flags used by ioctl(SG_IOSUBMIT) [abbrev: SG_IOS] and ioctl(SG_IORECEIVE) + * [abbrev: SG_IOR] OR-ed into sg_io_v4::flags. The sync v4 interface uses + * ioctl(SG_IO) and can take these new flags, as can the v3 interface. + * These flags apply for SG_IOS unless otherwise noted. May be OR-ed together. + */ +#define SGV4_FLAG_DIRECT_IO SG_FLAG_DIRECT_IO +#define SGV4_FLAG_MMAP_IO SG_FLAG_MMAP_IO +#define SGV4_FLAG_Q_AT_TAIL SG_FLAG_Q_AT_TAIL +#define SGV4_FLAG_Q_AT_HEAD SG_FLAG_Q_AT_HEAD +#define SGV4_FLAG_IMMED 0x400 /* for polling with SG_IOR, ignored in SG_IOS */ + /* Output (potentially OR-ed together) in v3::info or v4::info field */ #define SG_INFO_OK_MASK 0x1 #define SG_INFO_OK 0x0 /* no sense, host nor driver "noise" */ @@ -134,7 +146,6 @@ typedef struct sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */ /* sg_io_owned set imples synchronous, clear implies asynchronous */ char sg_io_owned;/* 0 -> complete with read(), 1 -> owned by SG_IO */ char problem; /* 0 -> no problem detected, 1 -> error to report */ - /* If SG_CTL_FLAGM_TAG_FOR_PACK_ID set on fd then next field is tag */ int pack_id; /* pack_id, in v4 driver may be tag instead */ void __user *usr_ptr; /* user provided pointer in v3+v4 interface */ unsigned int duration; @@ -163,6 +174,13 @@ typedef struct sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */ #define SG_SET_RESERVED_SIZE 0x2275 /* request new reserved buffer size */ #define SG_GET_RESERVED_SIZE 0x2272 /* actual size of reserved buffer */ +/* + * Historically the scsi/sg driver has used 0x22 as it ioctl base number. + * Add a define for that value and use it for several new ioctls added in + * version 4.0.01 sg driver and later. + */ +#define SG_IOCTL_MAGIC_NUM 0x22 + /* The following ioctl has a 'sg_scsi_id_t *' object as its 3rd argument. */ #define SG_GET_SCSI_ID 0x2276 /* Yields fd's bus, chan, dev, lun + type */ /* SCSI id information can also be obtained from SCSI_IOCTL_GET_IDLUN */ @@ -319,6 +337,23 @@ struct sg_header { */ #define SG_NEXT_CMD_LEN 0x2283 +/* + * New ioctls to replace async (non-blocking) write()/read() interface. + * Present in version 4 and later of the sg driver [>20190427]. The + * SG_IOSUBMIT and SG_IORECEIVE ioctls accept the sg_v4 interface based on + * struct sg_io_v4 found in . These objects are + * passed by a pointer in the third argument of the ioctl. + * + * Data may be transferred both from the user space to the driver by these + * ioctls. Hence the _IOWR macro is used here to generate the ioctl number + * rather than _IOW or _IOR. + */ +/* Submits a v4 interface object to driver, optionally receive tag back */ +#define SG_IOSUBMIT _IOWR(SG_IOCTL_MAGIC_NUM, 0x41, struct sg_io_v4) + +/* Gives some v4 identifying info to driver, receives associated response */ +#define SG_IORECEIVE _IOWR(SG_IOCTL_MAGIC_NUM, 0x42, struct sg_io_v4) + /* command queuing is always on when the v3 or v4 interface is used */ #define SG_DEF_COMMAND_Q 0 From patchwork Thu Oct 15 02:06:27 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269798 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id E1254C4363A for ; Thu, 15 Oct 2020 02:07:38 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id A646E22257 for ; Thu, 15 Oct 2020 02:07:38 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726115AbgJOCHi (ORCPT ); Wed, 14 Oct 2020 22:07:38 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40226 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732555AbgJOCH1 (ORCPT ); Wed, 14 Oct 2020 22:07:27 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 0AFF7204269; Thu, 15 Oct 2020 04:07:26 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 5vX35tX5+GaE; Thu, 15 Oct 2020 04:07:23 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 5B312204258; Thu, 15 Oct 2020 04:07:16 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 28/44] sg: rework debug info Date: Wed, 14 Oct 2020 22:06:27 -0400 Message-Id: <20201015020643.432908-29-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Since the version 2 driver, the state of the driver can be found with 'cat /proc/scsi/sg/debug'. As the driver becomes more threaded and IO faster (e.g. scsi_debug with a command timer of 5 microseconds), the existing state dump can become misleading as the state can change during the "snapshot". The new approach in this patch is to allocate a buffer of SG_PROC_DEBUG_SZ bytes and use scnprintf() to populate it. Only when the whole state is captured (or the buffer fills) is the output to the caller's terminal performed. The previous approach was line based: assemble a line of information and then output it. Locks are taken as required for short periods and should not interfere with a disk IO intensive program. Operations such as closing a sg file descriptor or removing a sg device may be held up for a short while (microseconds). Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 256 ++++++++++++++++++++++++++++++++-------------- 1 file changed, 177 insertions(+), 79 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 0e284e38d97d..8b5d41a97950 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -2387,7 +2387,7 @@ sg_rq_end_io(struct request *rq, blk_status_t status) if (unlikely((srp->rq_result & SG_ML_RESULT_MSK) && slen > 0)) sg_check_sense(sdp, srp, slen); if (slen > 0) { - if (scsi_rp->sense) { + if (scsi_rp->sense && !srp->sense_bp) { srp->sense_bp = mempool_alloc(sg_sense_pool, GFP_ATOMIC); if (srp->sense_bp) { @@ -2400,6 +2400,9 @@ sg_rq_end_io(struct request *rq, blk_status_t status) pr_warn("%s: sense but can't alloc buffer\n", __func__); } + } else if (srp->sense_bp) { + slen = 0; + pr_warn("%s: non-NULL srp->sense_bp ? ?\n", __func__); } else { slen = 0; pr_warn("%s: sense_len>0 but sense==NULL\n", __func__); @@ -3923,116 +3926,211 @@ sg_proc_seq_show_devstrs(struct seq_file *s, void *v) return 0; } -/* must be called while holding sg_index_lock */ -static void -sg_proc_debug_helper(struct seq_file *s, struct sg_device *sdp) +/* Writes debug info for one sg_request in obp buffer */ +static int +sg_proc_debug_sreq(struct sg_request *srp, int to, char *obp, int len) { - int k; - unsigned long idx, idx2; + bool is_v3v4, v4, is_dur; + int n = 0; + u32 dur; + enum sg_rq_state rq_st; + const char *cp; + + if (len < 1) + return 0; + v4 = test_bit(SG_FRQ_IS_V4I, srp->frq_bm); + is_v3v4 = v4 ? true : (srp->s_hdr3.interface_id != '\0'); + if (srp->parentfp->rsv_srp == srp) + cp = (is_v3v4 && (srp->rq_flags & SG_FLAG_MMAP_IO)) ? + " mmap>> " : " rsv>> "; + else + cp = (srp->rq_info & SG_INFO_DIRECT_IO_MASK) ? + " dio>> " : " "; + rq_st = atomic_read(&srp->rq_st); + dur = sg_get_dur(srp, &rq_st, &is_dur); + n += scnprintf(obp + n, len - n, "%s%s: dlen=%d/%d id=%d", cp, + sg_rq_st_str(rq_st, false), srp->sgat_h.dlen, + srp->sgat_h.buflen, (int)srp->pack_id); + if (is_dur) /* cmd/req has completed, waiting for ... */ + n += scnprintf(obp + n, len - n, " dur=%ums", dur); + else if (dur < U32_MAX) /* in-flight or busy (so ongoing) */ + n += scnprintf(obp + n, len - n, " t_o/elap=%us/%ums", + to / 1000, dur); + n += scnprintf(obp + n, len - n, " sgat=%d op=0x%02x\n", + srp->sgat_h.num_sgat, srp->cmd_opcode); + return n; +} + +/* Writes debug info for one sg fd (including its sg requests) in obp buffer */ +static int +sg_proc_debug_fd(struct sg_fd *fp, char *obp, int len, unsigned long idx) +{ + int n = 0; + int to, k; + unsigned long iflags; struct sg_request *srp; - struct sg_fd *fp; - const char * cp; - unsigned int ms; + /* sgat=-1 means unavailable */ + to = (fp->timeout >= 0) ? jiffies_to_msecs(fp->timeout) : -999; + if (to < 0) + n += scnprintf(obp + n, len - n, "BAD timeout=%d", + fp->timeout); + else if (to % 1000) + n += scnprintf(obp + n, len - n, "timeout=%dms rs", to); + else + n += scnprintf(obp + n, len - n, "timeout=%ds rs", to / 1000); + n += scnprintf(obp + n, len - n, "v_buflen=%d idx=%lu\n cmd_q=%d ", + fp->rsv_srp->sgat_h.buflen, idx, + (int)test_bit(SG_FFD_CMD_Q, fp->ffd_bm)); + n += scnprintf(obp + n, len - n, + "f_packid=%d k_orphan=%d ffd_bm=0x%lx\n", + (int)test_bit(SG_FFD_FORCE_PACKID, fp->ffd_bm), + (int)test_bit(SG_FFD_KEEP_ORPHAN, fp->ffd_bm), + fp->ffd_bm[0]); + n += scnprintf(obp + n, len - n, " mmap_called=%d\n", + test_bit(SG_FFD_MMAP_CALLED, fp->ffd_bm)); + n += scnprintf(obp + n, len - n, + " submitted=%d waiting=%d open thr_id=%d\n", + atomic_read(&fp->submitted), + atomic_read(&fp->waiting), fp->tid); + k = 0; + xa_lock_irqsave(&fp->srp_arr, iflags); + xa_for_each(&fp->srp_arr, idx, srp) { + if (!srp) + continue; + if (xa_get_mark(&fp->srp_arr, idx, SG_XA_RQ_INACTIVE)) + continue; + n += sg_proc_debug_sreq(srp, fp->timeout, obp + n, len - n); + ++k; + if ((k % 8) == 0) { /* don't hold up isr_s too long */ + xa_unlock_irqrestore(&fp->srp_arr, iflags); + cpu_relax(); + xa_lock_irqsave(&fp->srp_arr, iflags); + } + } + if (k == 0) + n += scnprintf(obp + n, len - n, " No requests active\n"); k = 0; + xa_for_each_marked(&fp->srp_arr, idx, srp, SG_XA_RQ_INACTIVE) { + if (!srp) + continue; + if (k == 0) + n += scnprintf(obp + n, len - n, " Inactives:\n"); + n += sg_proc_debug_sreq(srp, fp->timeout, obp + n, len - n); + ++k; + if ((k % 8) == 0) { /* don't hold up isr_s too long */ + xa_unlock_irqrestore(&fp->srp_arr, iflags); + cpu_relax(); + xa_lock_irqsave(&fp->srp_arr, iflags); + } + } + xa_unlock_irqrestore(&fp->srp_arr, iflags); + return n; +} + +/* Writes debug info for one sg device (including its sg fds) in obp buffer */ +static int +sg_proc_debug_sdev(struct sg_device *sdp, char *obp, int len, int *fd_counterp) +{ + int n = 0; + int my_count = 0; + unsigned long idx; + struct scsi_device *ssdp = sdp->device; + struct sg_fd *fp; + char *disk_name; + int *countp; + + countp = fd_counterp ? fd_counterp : &my_count; + disk_name = (sdp->disk ? sdp->disk->disk_name : "?_?"); + n += scnprintf(obp + n, len - n, " >>> device=%s ", disk_name); + n += scnprintf(obp + n, len - n, "%d:%d:%d:%llu ", ssdp->host->host_no, + ssdp->channel, ssdp->id, ssdp->lun); + n += scnprintf(obp + n, len - n, + " max_sgat_sz,elems=2^%d,%d excl=%d open_cnt=%d\n", + ilog2(sdp->max_sgat_sz), sdp->max_sgat_elems, + SG_HAVE_EXCLUDE(sdp), atomic_read(&sdp->open_cnt)); xa_for_each(&sdp->sfp_arr, idx, fp) { if (!fp) continue; - k++; - seq_printf(s, " FD(%d): timeout=%dms buflen=%d (res)sgat=%d low_dma=%d idx=%lu\n", - k, jiffies_to_msecs(fp->timeout), - fp->rsv_srp->sgat_h.buflen, - (int)fp->rsv_srp->sgat_h.num_sgat, - (int)sdp->device->host->unchecked_isa_dma, idx); - seq_printf(s, " cmd_q=%d f_packid=%d k_orphan=%d closed=0\n", - (int)test_bit(SG_FFD_CMD_Q, fp->ffd_bm), - (int)test_bit(SG_FFD_FORCE_PACKID, fp->ffd_bm), - (int)test_bit(SG_FFD_KEEP_ORPHAN, fp->ffd_bm)); - seq_printf(s, " submitted=%d waiting=%d\n", - atomic_read(&fp->submitted), - atomic_read(&fp->waiting)); - xa_for_each(&fp->srp_arr, idx2, srp) { - const struct sg_slice_hdr3 *sh3p = &srp->s_hdr3; - bool is_v3 = (sh3p->interface_id != '\0'); - enum sg_rq_state rq_st = atomic_read(&srp->rq_st); - - if (!srp) - continue; - if (srp->parentfp->rsv_srp == srp) { - if (is_v3 && (SG_FLAG_MMAP_IO & sh3p->flags)) - cp = " mmap>> "; - else - cp = " rb>> "; - } else { - if (SG_INFO_DIRECT_IO_MASK & srp->rq_info) - cp = " dio>> "; - else - cp = " "; - } - seq_puts(s, cp); - seq_puts(s, sg_rq_st_str(rq_st, false)); - seq_printf(s, ": id=%d len/blen=%d/%d", - srp->pack_id, srp->sgat_h.dlen, - srp->sgat_h.buflen); - if (rq_st == SG_RS_AWAIT_RCV || - rq_st == SG_RS_RCV_DONE) { - seq_printf(s, " dur=%d", srp->duration); - goto fin_line; - } - ms = jiffies_to_msecs(jiffies); - seq_printf(s, " t_o/elap=%d/%d", - (is_v3 ? sh3p->timeout : - jiffies_to_msecs(fp->timeout)), - (ms > srp->duration ? ms - srp->duration : - 0)); -fin_line: - seq_printf(s, "ms sgat=%d op=0x%02x\n", - srp->sgat_h.num_sgat, (int)srp->cmd_opcode); - } - if (xa_empty(&fp->srp_arr)) - seq_puts(s, " No requests active\n"); + ++*countp; + n += scnprintf(obp + n, len - n, " FD(%d): ", *countp); + n += sg_proc_debug_fd(fp, obp + n, len - n, idx); } + return n; } +/* Called via dbg_seq_ops once for each sg device */ static int sg_proc_seq_show_debug(struct seq_file *s, void *v) { + bool found = false; + bool trunc = false; + const int bp_len = SG_PROC_DEBUG_SZ; + int n = 0; + int k = 0; + unsigned long iflags; struct sg_proc_deviter *it = (struct sg_proc_deviter *)v; struct sg_device *sdp; - unsigned long iflags; + int *fdi_p; + char *bp; + char *disk_name; + char b1[128]; + b1[0] = '\0'; if (it && (0 == it->index)) seq_printf(s, "max_active_device=%d def_reserved_size=%d\n", - (int)it->max, sg_big_buff); - + (int)it->max, def_reserved_size); + fdi_p = it ? &it->fd_index : &k; + bp = kzalloc(bp_len, __GFP_NOWARN | GFP_KERNEL); + if (!bp) { + seq_printf(s, "%s: Unable to allocate %d on heap, finish\n", + __func__, bp_len); + return -1; + } read_lock_irqsave(&sg_index_lock, iflags); sdp = it ? sg_lookup_dev(it->index) : NULL; if (NULL == sdp) goto skip; if (!xa_empty(&sdp->sfp_arr)) { - seq_printf(s, " >>> device=%s ", sdp->disk->disk_name); + found = true; + disk_name = (sdp->disk ? sdp->disk->disk_name : "?_?"); if (SG_IS_DETACHING(sdp)) - seq_puts(s, "detaching pending close "); + snprintf(b1, sizeof(b1), " >>> device=%s %s\n", + disk_name, "detaching pending close\n"); else if (sdp->device) { - struct scsi_device *scsidp = sdp->device; - - seq_printf(s, "%d:%d:%d:%llu em=%d", - scsidp->host->host_no, - scsidp->channel, scsidp->id, - scsidp->lun, - scsidp->host->hostt->emulated); + n = sg_proc_debug_sdev(sdp, bp, bp_len, fdi_p); + if (n >= bp_len - 1) { + trunc = true; + if (bp[bp_len - 2] != '\n') + bp[bp_len - 2] = '\n'; + } + } else { + snprintf(b1, sizeof(b1), " >>> device=%s %s\n", + disk_name, "sdp->device==NULL, skip"); } - seq_printf(s, " max_sgat_sz=%d excl=%d open_cnt=%d\n", - sdp->max_sgat_sz, SG_HAVE_EXCLUDE(sdp), - atomic_read(&sdp->open_cnt)); - sg_proc_debug_helper(s, sdp); } skip: read_unlock_irqrestore(&sg_index_lock, iflags); + if (found) { + if (n > 0) { + seq_puts(s, bp); + if (seq_has_overflowed(s)) + goto s_ovfl; + if (trunc) + seq_printf(s, " >> Output truncated %s\n", + "due to buffer size"); + } else if (b1[0]) { + seq_puts(s, b1); + if (seq_has_overflowed(s)) + goto s_ovfl; + } + } +s_ovfl: + kfree(bp); return 0; } -#endif /* CONFIG_SCSI_PROC_FS (~300 lines back) */ +#endif /* CONFIG_SCSI_PROC_FS (~400 lines back) */ module_init(init_sg); module_exit(exit_sg); From patchwork Thu Oct 15 02:06:33 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269794 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id AB192C433DF for ; Thu, 15 Oct 2020 02:07:47 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 6C01722257 for ; Thu, 15 Oct 2020 02:07:47 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1732551AbgJOCHo (ORCPT ); Wed, 14 Oct 2020 22:07:44 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40278 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732546AbgJOCHi (ORCPT ); Wed, 14 Oct 2020 22:07:38 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id CA6AD204274; Thu, 15 Oct 2020 04:07:31 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id nXFGmlocCA2j; Thu, 15 Oct 2020 04:07:29 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 27F2920423F; Thu, 15 Oct 2020 04:07:23 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 34/44] sg: protect multiple receivers Date: Wed, 14 Oct 2020 22:06:33 -0400 Message-Id: <20201015020643.432908-35-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org If two threads call ioctl(SG_IORECEIVE) [or read()] on the same file descriptor there is a potential race on the same request response. Use atomic bit operations to make sure only one thread gets each request response. [The other thread will either get another request response or nothing.] Also make sfp cleanup a bit more robust and report if the number of submitted requests (which are decremented when completed) is other than the expected value of zero. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 54 +++++++++++++++++++++++++++++++++-------------- 1 file changed, 38 insertions(+), 16 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index fa7fca3e7196..af9cb2a9702d 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -109,6 +109,7 @@ enum sg_rq_state { /* N.B. sg_rq_state_arr assumes SG_RS_AWAIT_RCV==2 */ #define SG_FRQ_SYNC_INVOC 2 /* synchronous (blocking) invocation */ #define SG_FRQ_NO_US_XFER 3 /* no user space transfer of data */ #define SG_FRQ_DEACT_ORPHAN 6 /* not keeping orphan so de-activate */ +#define SG_FRQ_RECEIVING 7 /* guard against multiple receivers */ #define SG_FRQ_BLK_PUT_REQ 8 /* set when blk_put_request() called */ /* Bit positions (flags) for sg_fd::ffd_bm bitmask follow */ @@ -1286,6 +1287,7 @@ sg_ctl_ioreceive(struct file *filp, struct sg_fd *sfp, void __user *p) SG_LOG(3, sfp, "%s: non_block(+IMMED)=%d\n", __func__, non_block); /* read in part of v3 or v4 header for pack_id or tag based find */ id = pack_id; +try_again: srp = sg_find_srp_by_id(sfp, id); if (!srp) { /* nothing available so wait on packet or */ if (unlikely(SG_IS_DETACHING(sdp))) @@ -1300,6 +1302,10 @@ sg_ctl_ioreceive(struct file *filp, struct sg_fd *sfp, void __user *p) if (res) return res; /* signal --> -ERESTARTSYS */ } /* now srp should be valid */ + if (test_and_set_bit(SG_FRQ_RECEIVING, srp->frq_bm)) { + cpu_relax(); + goto try_again; + } return sg_receive_v4(sfp, srp, p, h4p); } @@ -1336,7 +1342,7 @@ sg_ctl_ioreceive_v3(struct file *filp, struct sg_fd *sfp, void __user *p) if (test_bit(SG_FFD_FORCE_PACKID, sfp->ffd_bm)) pack_id = h3p->pack_id; - +try_again: srp = sg_find_srp_by_id(sfp, pack_id); if (!srp) { /* nothing available so wait on packet or */ if (unlikely(SG_IS_DETACHING(sdp))) @@ -1351,6 +1357,10 @@ sg_ctl_ioreceive_v3(struct file *filp, struct sg_fd *sfp, void __user *p) if (unlikely(res)) return res; /* signal --> -ERESTARTSYS */ } /* now srp should be valid */ + if (test_and_set_bit(SG_FRQ_RECEIVING, srp->frq_bm)) { + cpu_relax(); + goto try_again; + } return sg_receive_v3(sfp, srp, SZ_SG_IO_HDR, p); } @@ -1503,6 +1513,7 @@ sg_read(struct file *filp, char __user *p, size_t count, loff_t *ppos) want_id = h2p->pack_id; } } +try_again: srp = sg_find_srp_by_id(sfp, want_id); if (!srp) { /* nothing available so wait on packet to arrive or */ if (unlikely(SG_IS_DETACHING(sdp))) @@ -1518,6 +1529,10 @@ sg_read(struct file *filp, char __user *p, size_t count, loff_t *ppos) return ret; /* otherwise srp should be valid */ } + if (test_and_set_bit(SG_FRQ_RECEIVING, srp->frq_bm)) { + cpu_relax(); + goto try_again; + } if (srp->s_hdr3.interface_id == '\0') ret = sg_read_v1v2(p, (int)count, sfp, srp); else @@ -3052,32 +3067,33 @@ sg_finish_scsi_blk_rq(struct sg_request *srp) atomic_dec(&sfp->submitted); atomic_dec(&sfp->waiting); } - if (srp->bio) { - bool us_xfer = !test_bit(SG_FRQ_NO_US_XFER, srp->frq_bm); - - if (us_xfer) { - ret = blk_rq_unmap_user(srp->bio); - if (ret) { /* -EINTR (-4) can be ignored */ - SG_LOG(6, sfp, - "%s: blk_rq_unmap_user() --> %d\n", - __func__, ret); - } - } - srp->bio = NULL; - } - /* In worst case READ data returned to user space by this point */ /* Expect blk_put_request(rq) already called in sg_rq_end_io() */ if (unlikely(!test_and_set_bit(SG_FRQ_BLK_PUT_REQ, srp->frq_bm))) { struct request *rq = srp->rq; + srp->rq = NULL; if (rq) { /* blk_get_request() may have failed */ if (scsi_req(rq)) scsi_req_free_cmd(scsi_req(rq)); - srp->rq = NULL; blk_put_request(rq); } } + if (srp->bio) { + bool us_xfer = !test_bit(SG_FRQ_NO_US_XFER, srp->frq_bm); + struct bio *bio = srp->bio; + + srp->bio = NULL; + if (us_xfer && bio) { + ret = blk_rq_unmap_user(bio); + if (ret) { /* -EINTR (-4) can be ignored */ + SG_LOG(6, sfp, + "%s: blk_rq_unmap_user() --> %d\n", + __func__, ret); + } + } + } + /* In worst case, READ data returned to user space by this point */ } static int @@ -3507,6 +3523,7 @@ sg_deact_request(struct sg_fd *sfp, struct sg_request *srp) return; sbp = srp->sense_bp; srp->sense_bp = NULL; + srp->frq_bm[0] = 0; sg_rq_state_chg(srp, 0, SG_RS_INACTIVE, true /* force */, __func__); /* maybe orphaned req, thus never read */ if (sbp) @@ -3639,6 +3656,7 @@ static void sg_remove_sfp_usercontext(struct work_struct *work) { __maybe_unused int o_count; + int subm; unsigned long idx, iflags; struct sg_device *sdp; struct sg_fd *sfp = container_of(work, struct sg_fd, ew_fd.work); @@ -3676,6 +3694,10 @@ sg_remove_sfp_usercontext(struct work_struct *work) SG_LOG(6, sfp, "%s: kfree: srp=%pK --\n", __func__, srp); kfree(srp); } + subm = atomic_read(&sfp->submitted); + if (subm != 0) + SG_LOG(1, sfp, "%s: expected submitted=0 got %d\n", + __func__, subm); xa_destroy(xafp); xa_lock_irqsave(xadp, iflags); e_sfp = __xa_erase(xadp, sfp->idx); From patchwork Thu Oct 15 02:06:34 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269793 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 909A4C2D0A3 for ; Thu, 15 Oct 2020 02:07:48 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 4138622257 for ; Thu, 15 Oct 2020 02:07:48 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1732572AbgJOCHr (ORCPT ); Wed, 14 Oct 2020 22:07:47 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40279 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732562AbgJOCHj (ORCPT ); Wed, 14 Oct 2020 22:07:39 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 0A827204275; Thu, 15 Oct 2020 04:07:33 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 6Y2TC9GmgqTn; Thu, 15 Oct 2020 04:07:30 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 45672204248; Thu, 15 Oct 2020 04:07:24 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 35/44] sg: first debugfs support Date: Wed, 14 Oct 2020 22:06:34 -0400 Message-Id: <20201015020643.432908-36-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Duplicate the semantics of 'cat /proc/scsi/sg/debug' on 'cat /sys/kernel/debug/scsi_generic/snapshot'. Make code that generates the snapshot conditional on either CONFIG_SCSI_PROC_FS or CONFIG_DEBUG_FS being defined. Also add snapshot_devs which can be written with a list of comma separated integers corresponding to sg (minor) device numbers. That file can also be read showing that list. Minus one (or any negative number) means accept all when in the first position (the default) or means the end of the list in a later position. When a subsequent cat /sys/kernel/debug/scsi_generic/snapshot is performed, only sg device numbers matching an element in that list are output. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 414 +++++++++++++++++++++++++++++++++++++--------- 1 file changed, 333 insertions(+), 81 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index af9cb2a9702d..249d56cb08c7 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -43,6 +43,7 @@ static char *sg_version_date = "20190606"; #include /* for sg_check_file_access() */ #include #include +#include #include #include @@ -67,6 +68,10 @@ static char *sg_version_date = "20190606"; #endif #endif +#if IS_ENABLED(CONFIG_SCSI_PROC_FS) || IS_ENABLED(CONFIG_DEBUG_FS) +#define SG_PROC_OR_DEBUG_FS 1 +#endif + /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater * than 16 bytes are "variable length" whose length is a multiple of 4, so: @@ -269,6 +274,8 @@ struct sg_comm_wr_t { /* arguments to sg_common_write() */ static void sg_rq_end_io(struct request *rq, blk_status_t status); /* Declarations of other static functions used before they are defined */ static int sg_proc_init(void); +static void sg_dfs_init(void); +static void sg_dfs_exit(void); static int sg_start_req(struct sg_request *srp, struct sg_comm_wr_t *cwrp, int dxfer_dir); static void sg_finish_scsi_blk_rq(struct sg_request *srp); @@ -338,7 +345,7 @@ static const char *sg_rq_st_str(enum sg_rq_state rq_st, bool long_str); pr_info("sg: sdp or sfp NULL, " fmt, ##a); \ } while (0) #else -#define SG_LOG(depth, sfp, fmt, a...) { } +#define SG_LOG(depth, sfp, fmt, a...) do { } while (0) #endif /* end of CONFIG_SCSI_LOGGING && SG_DEBUG conditional */ @@ -2748,22 +2755,6 @@ sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf) kref_put(&sdp->d_ref, sg_device_destroy); } -module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR); -module_param_named(def_reserved_size, def_reserved_size, int, - S_IRUGO | S_IWUSR); -module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR); - -MODULE_AUTHOR("Douglas Gilbert"); -MODULE_DESCRIPTION("SCSI generic (sg) driver"); -MODULE_LICENSE("GPL"); -MODULE_VERSION(SG_VERSION_STR); -MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR); - -MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element " - "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))"); -MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd"); -MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))"); - static int __init init_sg(void) { @@ -2813,6 +2804,7 @@ init_sg(void) rc = scsi_register_interface(&sg_interface); if (rc == 0) { sg_proc_init(); + sg_dfs_init(); return 0; } class_destroy(sg_sysfs_class); @@ -2826,17 +2818,10 @@ init_sg(void) return rc; } -#if !IS_ENABLED(CONFIG_SCSI_PROC_FS) -static int -sg_proc_init(void) -{ - return 0; -} -#endif - static void __exit exit_sg(void) { + sg_dfs_exit(); if (IS_ENABLED(CONFIG_SCSI_PROC_FS)) remove_proc_subtree("scsi/sg", NULL); scsi_unregister_interface(&sg_interface); @@ -3287,7 +3272,7 @@ sg_find_srp_by_id(struct sg_fd *sfp, int pack_id) /* here if one of above loops does _not_ find a match */ if (IS_ENABLED(CONFIG_SCSI_PROC_FS)) { if (search_for_1) { - const char *cptp = "pack_id="; + __maybe_unused const char *cptp = "pack_id="; if (is_bad_st) SG_LOG(1, sfp, "%s: %s%d wrong state: %s\n", @@ -3726,17 +3711,6 @@ sg_remove_sfp(struct kref *kref) schedule_work(&sfp->ew_fd.work); } -static int -sg_idr_max_id(int id, void *p, void *data) - __must_hold(sg_index_lock) -{ - int *k = data; - - if (*k < id) - *k = id; - return 0; -} - /* must be called with sg_index_lock held */ static struct sg_device * sg_lookup_dev(int dev) @@ -3766,7 +3740,7 @@ sg_get_dev(int dev) return sdp; } -#if IS_ENABLED(CONFIG_SCSI_PROC_FS) +#if IS_ENABLED(SG_PROC_OR_DEBUG_FS) static const char * sg_rq_st_str(enum sg_rq_state rq_st, bool long_str) { @@ -3795,7 +3769,35 @@ sg_rq_st_str(enum sg_rq_state rq_st, bool long_str) } #endif -#if IS_ENABLED(CONFIG_SCSI_PROC_FS) /* long, almost to end of file */ +#if IS_ENABLED(SG_PROC_OR_DEBUG_FS) + +#define SG_SNAPSHOT_DEV_MAX 4 + +/* + * For snapshot_devs array, -1 or two adjacent the same is terminator. + * -1 in first element of first two elements the same implies all. + */ +static struct sg_dfs_context_t { + struct dentry *dfs_rootdir; + int snapshot_devs[SG_SNAPSHOT_DEV_MAX]; +} sg_dfs_cxt; + +struct sg_proc_deviter { + loff_t index; + size_t max; + int fd_index; +}; + +static int +sg_idr_max_id(int id, void *p, void *data) + __must_hold(sg_index_lock) +{ + int *k = data; + + if (*k < id) + *k = id; + return 0; +} static int sg_last_dev(void) @@ -3809,6 +3811,41 @@ sg_last_dev(void) return k + 1; /* origin 1 */ } +static void * +dev_seq_start(struct seq_file *s, loff_t *pos) +{ + struct sg_proc_deviter *it = kzalloc(sizeof(*it), GFP_KERNEL); + + s->private = it; + if (!it) + return NULL; + + it->index = *pos; + it->max = sg_last_dev(); + if (it->index >= it->max) + return NULL; + return it; +} + +static void * +dev_seq_next(struct seq_file *s, void *v, loff_t *pos) +{ + struct sg_proc_deviter *it = s->private; + + *pos = ++it->index; + return (it->index < it->max) ? it : NULL; +} + +static void +dev_seq_stop(struct seq_file *s, void *v) +{ + kfree(s->private); +} + +#endif /* SG_PROC_OR_DEBUG_FS */ + +#if IS_ENABLED(CONFIG_SCSI_PROC_FS) /* around 100 lines */ + static int sg_proc_seq_show_int(struct seq_file *s, void *v) { @@ -3822,7 +3859,7 @@ sg_proc_single_open_adio(struct inode *inode, struct file *filp) return single_open(filp, sg_proc_seq_show_int, &sg_allow_dio); } -static ssize_t +static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer, size_t count, loff_t *off) { @@ -3844,7 +3881,7 @@ sg_proc_single_open_dressz(struct inode *inode, struct file *filp) return single_open(filp, sg_proc_seq_show_int, &sg_big_buff); } -static ssize_t +static ssize_t sg_proc_write_dressz(struct file *filp, const char __user *buffer, size_t count, loff_t *off) { @@ -3879,43 +3916,6 @@ sg_proc_seq_show_devhdr(struct seq_file *s, void *v) return 0; } -struct sg_proc_deviter { - loff_t index; - size_t max; - int fd_index; -}; - -static void * -dev_seq_start(struct seq_file *s, loff_t *pos) -{ - struct sg_proc_deviter *it = kzalloc(sizeof(*it), GFP_KERNEL); - - s->private = it; - if (! it) - return NULL; - - it->index = *pos; - it->max = sg_last_dev(); - if (it->index >= it->max) - return NULL; - return it; -} - -static void * -dev_seq_next(struct seq_file *s, void *v, loff_t *pos) -{ - struct sg_proc_deviter *it = s->private; - - *pos = ++it->index; - return (it->index < it->max) ? it : NULL; -} - -static void -dev_seq_stop(struct seq_file *s, void *v) -{ - kfree(s->private); -} - static int sg_proc_seq_show_dev(struct seq_file *s, void *v) { @@ -3962,6 +3962,10 @@ sg_proc_seq_show_devstrs(struct seq_file *s, void *v) return 0; } +#endif /* CONFIG_SCSI_PROC_FS (~100 lines back) */ + +#if IS_ENABLED(SG_PROC_OR_DEBUG_FS) + /* Writes debug info for one sg_request in obp buffer */ static int sg_proc_debug_sreq(struct sg_request *srp, int to, char *obp, int len) @@ -4104,18 +4108,20 @@ sg_proc_seq_show_debug(struct seq_file *s, void *v) bool found = false; bool trunc = false; const int bp_len = SG_PROC_DEBUG_SZ; + int j, sd_n; int n = 0; int k = 0; unsigned long iflags; struct sg_proc_deviter *it = (struct sg_proc_deviter *)v; struct sg_device *sdp; int *fdi_p; + const int *dev_arr = sg_dfs_cxt.snapshot_devs; char *bp; char *disk_name; char b1[128]; b1[0] = '\0'; - if (it && (0 == it->index)) + if (it && it->index == 0) seq_printf(s, "max_active_device=%d def_reserved_size=%d\n", (int)it->max, def_reserved_size); fdi_p = it ? &it->fd_index : &k; @@ -4127,8 +4133,31 @@ sg_proc_seq_show_debug(struct seq_file *s, void *v) } read_lock_irqsave(&sg_index_lock, iflags); sdp = it ? sg_lookup_dev(it->index) : NULL; - if (NULL == sdp) + if (!sdp) goto skip; + sd_n = dev_arr[0]; + if (sd_n != -1 && sd_n != sdp->index && sd_n != dev_arr[1]) { + for (j = 1; j < SG_SNAPSHOT_DEV_MAX; ) { + sd_n = dev_arr[j]; + if (sd_n < 0) + goto skip; + ++j; + if (j >= SG_SNAPSHOT_DEV_MAX) { + if (sd_n == sdp->index) { + found = true; + break; + } + } else if (sd_n == dev_arr[j]) { + goto skip; + } else if (sd_n == sdp->index) { + found = true; + break; + } + } + if (!found) + goto skip; + found = false; + } if (!xa_empty(&sdp->sfp_arr)) { found = true; disk_name = (sdp->disk ? sdp->disk->disk_name : "?_?"); @@ -4168,6 +4197,10 @@ sg_proc_seq_show_debug(struct seq_file *s, void *v) return 0; } +#endif /* SG_PROC_OR_DEBUG_FS */ + +#if IS_ENABLED(CONFIG_SCSI_PROC_FS) + static const struct proc_ops adio_proc_ops = { .proc_open = sg_proc_single_open_adio, .proc_read = seq_read, @@ -4226,7 +4259,226 @@ sg_proc_init(void) /* remove_proc_subtree("scsi/sg", NULL) in exit_sg() does cleanup */ -#endif /* CONFIG_SCSI_PROC_FS (~400 lines back) */ +#else + +static int +sg_proc_init(void) +{ + return 0; +} + +#endif /* CONFIG_SCSI_PROC_FS */ + +#if IS_ENABLED(CONFIG_DEBUG_FS) + +struct sg_dfs_attr { + const char *name; + umode_t mode; + int (*show)(void *d, struct seq_file *m); + ssize_t (*write)(void *d, const char __user *b, size_t s, loff_t *o); + /* Set either .show or .seq_ops. */ + const struct seq_operations *seq_ops; +}; +static int +sg_dfs_snapshot_devs_show(void *data, struct seq_file *m) +{ + bool last; + int k, d; + struct sg_dfs_context_t *ctxp = data; + + for (k = 0; k < SG_SNAPSHOT_DEV_MAX; ++k) { + d = ctxp->snapshot_devs[k]; + last = (k + 1 == SG_SNAPSHOT_DEV_MAX); + if (d < 0) { + if (k == 0) + seq_puts(m, "-1"); + break; + } + if (!last && d == ctxp->snapshot_devs[k + 1]) { + if (k == 0) + seq_puts(m, "-1"); + break; + } + if (k != 0) + seq_puts(m, ","); + seq_printf(m, "%d", d); + } + seq_puts(m, "\n"); + return 0; +} + +static ssize_t +sg_dfs_snapshot_devs_write(void *data, const char __user *buf, size_t count, + loff_t *ppos) +{ + bool trailing_comma; + int k, n; + struct sg_dfs_context_t *cxtp = data; + char lbuf[64] = { }, *cp, *c2p; + + if (count >= sizeof(lbuf)) { + pr_err("%s: operation too long\n", __func__); + return -EINVAL; + } + if (copy_from_user(lbuf, buf, count)) + return -EFAULT; + for (k = 0, n = 0, cp = lbuf; k < SG_SNAPSHOT_DEV_MAX; + ++k, cp = c2p + 1) { + c2p = strchr(cp, ','); + if (c2p) + *c2p = '\0'; + trailing_comma = !!c2p; + /* sscanf is easier to use that this ... */ + if (kstrtoint(cp, 10, cxtp->snapshot_devs + k)) + break; + ++n; + if (!trailing_comma) + break; + } + if (n == 0) { + return -EINVAL; + } else if (k >= SG_SNAPSHOT_DEV_MAX && trailing_comma) { + pr_err("%s: only %d elements in snapshot array\n", __func__, + SG_SNAPSHOT_DEV_MAX); + return -EINVAL; + } + if (n < SG_SNAPSHOT_DEV_MAX) + cxtp->snapshot_devs[n] = -1; + return count; +} + +static int +sg_dfs_show(struct seq_file *m, void *v) +{ + const struct sg_dfs_attr *attr = m->private; + void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private; + + return attr->show(data, m); +} + +static ssize_t +sg_dfs_write(struct file *file, const char __user *buf, size_t count, + loff_t *ppos) +{ + struct seq_file *m = file->private_data; + const struct sg_dfs_attr *attr = m->private; + void *data = d_inode(file->f_path.dentry->d_parent)->i_private; + + /* + * Attributes that only implement .seq_ops are read-only and 'attr' is + * the same with 'data' in this case. + */ + if (attr == data || !attr->write) + return -EPERM; + return attr->write(data, buf, count, ppos); +} + +static int +sg_dfs_open(struct inode *inode, struct file *file) +{ + const struct sg_dfs_attr *attr = inode->i_private; + void *data = d_inode(file->f_path.dentry->d_parent)->i_private; + struct seq_file *m; + int ret; + + if (attr->seq_ops) { + ret = seq_open(file, attr->seq_ops); + if (!ret) { + m = file->private_data; + m->private = data; + } + return ret; + } + if (WARN_ON_ONCE(!attr->show)) + return -EPERM; + return single_open(file, sg_dfs_show, inode->i_private); +} + +static int +sg_dfs_release(struct inode *inode, struct file *file) +{ + const struct sg_dfs_attr *attr = inode->i_private; + + if (attr->show) + return single_release(inode, file); + return seq_release(inode, file); +} + +static const struct file_operations sg_dfs_fops = { + .owner = THIS_MODULE, + .open = sg_dfs_open, + .read = seq_read, + .write = sg_dfs_write, + .llseek = seq_lseek, + .release = sg_dfs_release, +}; + +static void sg_dfs_mk_files(struct dentry *parent, void *data, + const struct sg_dfs_attr *attr) +{ + if (IS_ERR_OR_NULL(parent)) + return; + + d_inode(parent)->i_private = data; + for (; attr->name; ++attr) + debugfs_create_file(attr->name, attr->mode, parent, + (void *)attr, &sg_dfs_fops); +} + +static const struct seq_operations sg_snapshot_seq_ops = { + .start = dev_seq_start, + .next = dev_seq_next, + .stop = dev_seq_stop, + .show = sg_proc_seq_show_debug, +}; + +static const struct sg_dfs_attr sg_dfs_attrs[] = { + {"snapshot", 0400, .seq_ops = &sg_snapshot_seq_ops}, + {"snapshot_devs", 0600, sg_dfs_snapshot_devs_show, + sg_dfs_snapshot_devs_write}, + { }, +}; + +static void +sg_dfs_init(void) +{ + /* create and populate /sys/kernel/debug/scsi_generic directory */ + if (!sg_dfs_cxt.dfs_rootdir) { + sg_dfs_cxt.dfs_rootdir = debugfs_create_dir("scsi_generic", + NULL); + sg_dfs_mk_files(sg_dfs_cxt.dfs_rootdir, &sg_dfs_cxt, + sg_dfs_attrs); + } + sg_dfs_cxt.snapshot_devs[0] = -1; /* show all sg devices */ +} + +static void +sg_dfs_exit(void) +{ + debugfs_remove_recursive(sg_dfs_cxt.dfs_rootdir); + sg_dfs_cxt.dfs_rootdir = NULL; +} + +#else /* not defined: CONFIG_DEBUG_FS */ + +static void sg_dfs_init(void) {} +static void sg_dfs_exit(void) {} + +#endif /* CONFIG_DEBUG_FS */ + +module_param_named(scatter_elem_sz, scatter_elem_sz, int, 0644); +module_param_named(def_reserved_size, def_reserved_size, int, 0644); +module_param_named(allow_dio, sg_allow_dio, int, 0644); + +MODULE_AUTHOR("Douglas Gilbert"); +MODULE_DESCRIPTION("SCSI generic (sg) driver"); +MODULE_LICENSE("GPL"); +MODULE_VERSION(SG_VERSION_STR); +MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR); + +MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element size (default: max(SG_SCATTER_SZ, PAGE_SIZE))"); +MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd"); +MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))"); module_init(init_sg); module_exit(exit_sg); From patchwork Thu Oct 15 02:06:36 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269796 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 70871C433E7 for ; Thu, 15 Oct 2020 02:07:47 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 30BA02225C for ; Thu, 15 Oct 2020 02:07:47 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1732549AbgJOCHq (ORCPT ); Wed, 14 Oct 2020 22:07:46 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40281 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732565AbgJOCHk (ORCPT ); Wed, 14 Oct 2020 22:07:40 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id D23BB204278; Thu, 15 Oct 2020 04:07:34 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id wc0gujEjcB54; Thu, 15 Oct 2020 04:07:33 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 7F19F204258; Thu, 15 Oct 2020 04:07:26 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 37/44] sg: defang allow_dio Date: Wed, 14 Oct 2020 22:06:36 -0400 Message-Id: <20201015020643.432908-38-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Before direct IO was permitted by this driver, the user either had to give 'allow_dio=1' as a module parameter or write to procfs with 'echo 1 > /proc/scsi/sg/allow_dio'. The user also needs to set the SG_FLAG_DIRECT_IO flag in the flags field of either the sg v3 or v3 interface. The reason this "belts and braces" approach was taken is lost in the mists of time (done over 20 years ago). So this patch keeps the allow_dio attribute for backward compatibility but ignores it, relying on the SG_FLAG_DIRECT_IO flag alone. This brings the use of the SG_FLAG_DIRECT_IO flag into line with the SG_FLAG_MMAP_IO flag; the two mechanisms are no more, or less, safe than one another in recent Linux kernels. Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 26 ++++++++------------------ 1 file changed, 8 insertions(+), 18 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index b9fa76b2ea15..da10e6d32f53 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -142,7 +142,7 @@ int sg_big_buff = SG_DEF_RESERVED_SIZE; * not enough memory) will be reserved for use by this file descriptor. */ static int def_reserved_size = -1; /* picks up init parameter */ -static int sg_allow_dio = SG_ALLOW_DIO_DEF; +static int sg_allow_dio = SG_ALLOW_DIO_DEF; /* ignored by code */ static int scatter_elem_sz = SG_SCATTER_SZ; @@ -2862,19 +2862,6 @@ exit_sg(void) idr_destroy(&sg_index_idr); } -static inline bool -sg_chk_dio_allowed(struct sg_device *sdp, struct sg_request *srp, - int iov_count, int dir) -{ - if (sg_allow_dio && (srp->rq_flags & SG_FLAG_DIRECT_IO)) { - if (dir != SG_DXFER_UNKNOWN && !iov_count) { - if (!sdp->device->host->unchecked_isa_dma) - return true; - } - } - return false; -} - static inline void sg_set_map_data(const struct sg_scatter_hold *schp, bool up_valid, struct rq_map_data *mdp) @@ -2893,6 +2880,7 @@ sg_start_req(struct sg_request *srp, struct sg_comm_wr_t *cwrp, int dxfer_dir) int res = 0; int dxfer_len = 0; int r0w = READ; + u32 rq_flags = srp->rq_flags; unsigned int iov_count = 0; void __user *up; struct request *rq; @@ -2973,7 +2961,7 @@ sg_start_req(struct sg_request *srp, struct sg_comm_wr_t *cwrp, int dxfer_dir) } scsi_rp->cmd_len = cwrp->cmd_len; srp->cmd_opcode = scsi_rp->cmd[0]; - us_xfer = !(srp->rq_flags & (SG_FLAG_NO_DXFER | SG_FLAG_MMAP_IO)); + us_xfer = !(rq_flags & (SG_FLAG_NO_DXFER | SG_FLAG_MMAP_IO)); assign_bit(SG_FRQ_NO_US_XFER, srp->frq_bm, !us_xfer); reserved = (sfp->rsv_srp == srp); rq->end_io_data = srp; @@ -2984,7 +2972,8 @@ sg_start_req(struct sg_request *srp, struct sg_comm_wr_t *cwrp, int dxfer_dir) SG_LOG(4, sfp, "%s: no data xfer [0x%pK]\n", __func__, srp); set_bit(SG_FRQ_NO_US_XFER, srp->frq_bm); goto fini; /* path of reqs with no din nor dout */ - } else if (sg_chk_dio_allowed(sdp, srp, iov_count, dxfer_dir) && + } else if ((rq_flags & SG_FLAG_DIRECT_IO) && iov_count == 0 && + !sdp->device->host->unchecked_isa_dma && blk_rq_aligned(q, (unsigned long)up, dxfer_len)) { srp->rq_info |= SG_INFO_DIRECT_IO; md = NULL; @@ -2995,7 +2984,7 @@ sg_start_req(struct sg_request *srp, struct sg_comm_wr_t *cwrp, int dxfer_dir) } if (likely(md)) { /* normal, "indirect" IO */ - if (unlikely(srp->rq_flags & SG_FLAG_MMAP_IO)) { + if (unlikely(rq_flags & SG_FLAG_MMAP_IO)) { /* mmap IO must use and fit in reserve request */ if (!reserved || dxfer_len > req_schp->buflen) res = reserved ? -ENOMEM : -EBUSY; @@ -3894,6 +3883,7 @@ sg_proc_single_open_adio(struct inode *inode, struct file *filp) return single_open(filp, sg_proc_seq_show_int, &sg_allow_dio); } +/* Kept for backward compatibility. sg_allow_dio is now ignored. */ static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer, size_t count, loff_t *off) @@ -4514,6 +4504,6 @@ MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR); MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element size (default: max(SG_SCATTER_SZ, PAGE_SIZE))"); MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd"); -MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))"); +MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow)); now ignored"); module_init(init_sg); module_exit(exit_sg); From patchwork Thu Oct 15 02:06:37 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269795 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 2A2C9C2BD10 for ; Thu, 15 Oct 2020 02:07:48 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id D4B1A22257 for ; Thu, 15 Oct 2020 02:07:47 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1732565AbgJOCHq (ORCPT ); Wed, 14 Oct 2020 22:07:46 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40282 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732567AbgJOCHj (ORCPT ); Wed, 14 Oct 2020 22:07:39 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 18952204258; Thu, 15 Oct 2020 04:07:36 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id c7mzQV-XQWz9; Thu, 15 Oct 2020 04:07:34 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 9B59020426D; Thu, 15 Oct 2020 04:07:27 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 38/44] sg: warn v3 write system call users Date: Wed, 14 Oct 2020 22:06:37 -0400 Message-Id: <20201015020643.432908-39-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Should generate one log message per kernel run when the write() system call is used with the sg interface version 3. Due to security concerns suggest that they use ioctl(SG_SUBMIT_v3) instead. Sg interface version 1 or 2 based code may also be calling write() in this context. There is no easy solution for them (short of upgrading their interface to version 3 or 4), so don't produce a warning suggesting the conversion will be simple. Reviewed-by: Hannes Reinecke Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index da10e6d32f53..b95eb87a3c43 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -645,6 +645,9 @@ sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) __func__); return -EPERM; } + pr_warn_once("Please use %s instead of write(),\n%s\n", + "ioctl(SG_SUBMIT_V3)", + " See: http://sg.danny.cz/sg/sg_v40.html"); res = sg_v3_submit(filp, sfp, h3p, false, NULL); return res < 0 ? res : (int)count; } From patchwork Thu Oct 15 02:06:38 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269797 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id E4BB7C43457 for ; Thu, 15 Oct 2020 02:07:44 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 9874C22257 for ; Thu, 15 Oct 2020 02:07:44 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729760AbgJOCHo (ORCPT ); Wed, 14 Oct 2020 22:07:44 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40283 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732566AbgJOCHj (ORCPT ); Wed, 14 Oct 2020 22:07:39 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 7CC6220426D; Thu, 15 Oct 2020 04:07:36 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 7+QWf238fNqn; Thu, 15 Oct 2020 04:07:34 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id C064A20425B; Thu, 15 Oct 2020 04:07:28 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 39/44] sg: add mmap_sz tracking Date: Wed, 14 Oct 2020 22:06:38 -0400 Message-Id: <20201015020643.432908-40-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org Track mmap_sz from prior mmap(2) call, per sg file descriptor. Also reset this value whenever munmap(2) is called. Fail SG_FLAG_MMAP_IO uses if mmap(2) hasn't been called or the memory associated with it is not large enough for the current request. Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index b95eb87a3c43..cba1db278775 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -232,6 +232,7 @@ struct sg_fd { /* holds the state of a file descriptor */ atomic_t waiting; /* number of requests awaiting receive */ atomic_t req_cnt; /* number of requests */ int sgat_elem_sz; /* initialized to scatter_elem_sz */ + int mmap_sz; /* byte size of previous mmap() call */ unsigned long ffd_bm[1]; /* see SG_FFD_* defines above */ pid_t tid; /* thread id when opened */ u8 next_cmd_len; /* 0: automatic, >0: use on next write() */ @@ -725,10 +726,14 @@ sg_write(struct file *filp, const char __user *p, size_t count, loff_t *ppos) static inline int sg_chk_mmap(struct sg_fd *sfp, int rq_flags, int len) { + if (!test_bit(SG_FFD_MMAP_CALLED, sfp->ffd_bm)) + return -EBADFD; if (atomic_read(&sfp->submitted) > 0) return -EBUSY; /* already active requests on fd */ if (len > sfp->rsv_srp->sgat_h.buflen) return -ENOMEM; /* MMAP_IO size must fit in reserve */ + if (unlikely(len > sfp->mmap_sz)) + return -ENOMEM; /* MMAP_IO size can't exceed mmap() size */ if (rq_flags & SG_FLAG_DIRECT_IO) return -EINVAL; /* not both MMAP_IO and DIRECT_IO */ return 0; @@ -2262,6 +2267,8 @@ sg_vma_close(struct vm_area_struct *vma) pr_warn("%s: sfp null\n", __func__); return; } + sfp->mmap_sz = 0; + clear_bit(SG_FFD_MMAP_CALLED, sfp->ffd_bm); kref_put(&sfp->f_ref, sg_remove_sfp); /* get in: sg_vma_open() */ } @@ -2383,6 +2390,7 @@ sg_mmap(struct file *filp, struct vm_area_struct *vma) goto fini; } } + sfp->mmap_sz = req_sz; vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP; vma->vm_private_data = sfp; vma->vm_ops = &sg_mmap_vm_ops; @@ -4056,8 +4064,7 @@ sg_proc_debug_fd(struct sg_fd *fp, char *obp, int len, unsigned long idx) (int)test_bit(SG_FFD_FORCE_PACKID, fp->ffd_bm), (int)test_bit(SG_FFD_KEEP_ORPHAN, fp->ffd_bm), fp->ffd_bm[0]); - n += scnprintf(obp + n, len - n, " mmap_called=%d\n", - test_bit(SG_FFD_MMAP_CALLED, fp->ffd_bm)); + n += scnprintf(obp + n, len - n, " mmap_sz=%d\n", fp->mmap_sz); n += scnprintf(obp + n, len - n, " submitted=%d waiting=%d open thr_id=%d\n", atomic_read(&fp->submitted), From patchwork Thu Oct 15 02:06:41 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269792 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id AA167C433E7 for ; Thu, 15 Oct 2020 02:07:56 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 6448622258 for ; Thu, 15 Oct 2020 02:07:56 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S2389379AbgJOCH4 (ORCPT ); Wed, 14 Oct 2020 22:07:56 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40283 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732570AbgJOCHp (ORCPT ); Wed, 14 Oct 2020 22:07:45 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 138F120414F; Thu, 15 Oct 2020 04:07:40 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id IjKWiJR7BeUr; Thu, 15 Oct 2020 04:07:38 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 228C520423F; Thu, 15 Oct 2020 04:07:32 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 42/44] sg: remove unit attention check for device changed Date: Wed, 14 Oct 2020 22:06:41 -0400 Message-Id: <20201015020643.432908-43-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org The SCSI mid-layer now checks for SCSI UNIT ATTENTIONs and takes the appropriate actions. This means that the sg driver no longer needs to do this check. Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 49 ++++++++++++----------------------------------- 1 file changed, 12 insertions(+), 37 deletions(-) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index d6acba00a3e4..75c29d0bceef 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -2457,39 +2457,6 @@ sg_rq_end_io_usercontext(struct work_struct *work) kref_put(&sfp->f_ref, sg_remove_sfp); } -static void -sg_check_sense(struct sg_device *sdp, struct sg_request *srp, int sense_len) -{ - int driver_stat; - u32 rq_res = srp->rq_result; - struct scsi_request *scsi_rp = scsi_req(srp->rq); - u8 *sbp = scsi_rp ? scsi_rp->sense : NULL; - - if (!sbp) - return; - driver_stat = driver_byte(rq_res); - if (driver_stat & DRIVER_SENSE) { - struct scsi_sense_hdr ssh; - - if (scsi_normalize_sense(sbp, sense_len, &ssh)) { - if (!scsi_sense_is_deferred(&ssh)) { - if (ssh.sense_key == UNIT_ATTENTION) { - if (sdp->device->removable) - sdp->device->changed = 1; - } - } - } - } - if (test_bit(SG_FDEV_LOG_SENSE, sdp->fdev_bm) > 0) { - int scsi_stat = rq_res & 0xff; - - if (scsi_stat == SAM_STAT_CHECK_CONDITION || - scsi_stat == SAM_STAT_COMMAND_TERMINATED) - __scsi_print_sense(sdp->device, __func__, sbp, - sense_len); - } -} - /* * This "bottom half" (soft interrupt) handler is called by the mid-level * when a request has completed or failed. This callback is registered in a @@ -2501,6 +2468,7 @@ sg_rq_end_io(struct request *rq, blk_status_t status) { enum sg_rq_state rqq_state = SG_RS_AWAIT_RCV; int a_resid, slen; + u32 rq_result; struct sg_request *srp = rq->end_io_data; struct scsi_request *scsi_rp = scsi_req(rq); struct sg_device *sdp; @@ -2514,7 +2482,8 @@ sg_rq_end_io(struct request *rq, blk_status_t status) sfp = srp->parentfp; sdp = sfp->parentdp; - srp->rq_result = scsi_rp->result; + rq_result = scsi_rp->result; + srp->rq_result = rq_result; slen = min_t(int, scsi_rp->sense_len, SCSI_SENSE_BUFFERSIZE); a_resid = scsi_rp->resid_len; @@ -2530,10 +2499,16 @@ sg_rq_end_io(struct request *rq, blk_status_t status) } SG_LOG(6, sfp, "%s: pack_id=%d, res=0x%x\n", __func__, srp->pack_id, - srp->rq_result); + rq_result); srp->duration = sg_calc_rq_dur(srp); - if (unlikely((srp->rq_result & SG_ML_RESULT_MSK) && slen > 0)) - sg_check_sense(sdp, srp, slen); + if (unlikely((rq_result & SG_ML_RESULT_MSK) && slen > 0 && + test_bit(SG_FDEV_LOG_SENSE, sdp->fdev_bm))) { + u32 scsi_stat = rq_result & 0xff; + + if (scsi_stat == SAM_STAT_CHECK_CONDITION || + scsi_stat == SAM_STAT_COMMAND_TERMINATED) + __scsi_print_sense(sdp->device, __func__, scsi_rp->sense, slen); + } if (slen > 0) { if (scsi_rp->sense && !srp->sense_bp) { srp->sense_bp = mempool_alloc(sg_sense_pool, From patchwork Thu Oct 15 02:06:42 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Douglas Gilbert X-Patchwork-Id: 269791 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-12.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH, MAILING_LIST_MULTI, SIGNED_OFF_BY, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id ED266C43457 for ; Thu, 15 Oct 2020 02:07:57 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id B283022257 for ; Thu, 15 Oct 2020 02:07:57 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S2389514AbgJOCH4 (ORCPT ); Wed, 14 Oct 2020 22:07:56 -0400 Received: from smtp.infotech.no ([82.134.31.41]:40278 "EHLO smtp.infotech.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732555AbgJOCHp (ORCPT ); Wed, 14 Oct 2020 22:07:45 -0400 Received: from localhost (localhost [127.0.0.1]) by smtp.infotech.no (Postfix) with ESMTP id 2DDF720423F; Thu, 15 Oct 2020 04:07:41 +0200 (CEST) X-Virus-Scanned: by amavisd-new-2.6.6 (20110518) (Debian) at infotech.no Received: from smtp.infotech.no ([127.0.0.1]) by localhost (smtp.infotech.no [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id Sr-h4XW6eyJk; Thu, 15 Oct 2020 04:07:39 +0200 (CEST) Received: from xtwo70.bingwo.ca (host-104-157-204-209.dyn.295.ca [104.157.204.209]) by smtp.infotech.no (Postfix) with ESMTPA id 3E81B204248; Thu, 15 Oct 2020 04:07:33 +0200 (CEST) From: Douglas Gilbert To: linux-scsi@vger.kernel.org Cc: martin.petersen@oracle.com, jejb@linux.vnet.ibm.com, hare@suse.de Subject: [PATCH v11 43/44] sg: no_dxfer: move to/from kernel buffers Date: Wed, 14 Oct 2020 22:06:42 -0400 Message-Id: <20201015020643.432908-44-dgilbert@interlog.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20201015020643.432908-1-dgilbert@interlog.com> References: <20201015020643.432908-1-dgilbert@interlog.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-scsi@vger.kernel.org When the NO_DXFER flag is use on a command/request, the data-in and data-out buffers (if present) should not be ignored. Add sg_rq_map_kern() function to handle this. Uses a single bio with multiple bvec_s usually each holding multiple pages, if necessary. The driver default element size is 32 KiB so if PAGE_SIZE is 4096 then get_order()==3 . Signed-off-by: Douglas Gilbert --- drivers/scsi/sg.c | 59 +++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 59 insertions(+) diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c index 75c29d0bceef..770e8c1ef53e 100644 --- a/drivers/scsi/sg.c +++ b/drivers/scsi/sg.c @@ -2865,6 +2865,63 @@ exit_sg(void) idr_destroy(&sg_index_idr); } +static struct bio * +sg_mk_kern_bio(int bvec_cnt) +{ + struct bio *biop; + + if (bvec_cnt > BIO_MAX_PAGES) + return NULL; + biop = bio_alloc(GFP_ATOMIC, bvec_cnt); + if (!biop) + return NULL; + biop->bi_end_io = bio_put; + return biop; +} + +/* + * Setup to move data between kernel buffers managed by this driver and a SCSI device. Note that + * there is no corresponding 'unmap' call as is required by blk_rq_map_user() . Uses a single + * bio with an expanded appended bvec if necessary. + */ +static int +sg_rq_map_kern(struct sg_request *srp, struct request_queue *q, struct request *rqq, int rw_ind) +{ + struct sg_scatter_hold *schp = &srp->sgat_h; + struct bio *bio; + int k, ln; + int op_flags = 0; + int num_sgat = schp->num_sgat; + int dlen = schp->dlen; + int pg_sz = 1 << (PAGE_SHIFT + schp->page_order); + int num_segs = (1 << schp->page_order) * num_sgat; + int res = 0; + + SG_LOG(4, srp->parentfp, "%s: dlen=%d, pg_sz=%d\n", __func__, dlen, pg_sz); + if (num_sgat <= 0) + return 0; + if (rw_ind == WRITE) + op_flags = REQ_SYNC | REQ_IDLE; + bio = sg_mk_kern_bio(num_sgat - k); + if (!bio) + return -ENOMEM; + bio->bi_opf = req_op(rqq) | op_flags; + + for (k = 0; k < num_sgat && dlen > 0; ++k, dlen -= ln) { + ln = min_t(int, dlen, pg_sz); + if (bio_add_pc_page(q, bio, schp->pages[k], ln, 0) < ln) { + bio_put(bio); + return -EINVAL; + } + } + res = blk_rq_append_bio(rqq, &bio); + if (unlikely(res)) + bio_put(bio); + else + rqq->nr_phys_segments = num_segs; + return res; +} + static inline void sg_set_map_data(const struct sg_scatter_hold *schp, bool up_valid, struct rq_map_data *mdp) @@ -3036,6 +3093,8 @@ sg_start_req(struct sg_request *srp, struct sg_comm_wr_t *cwrp, int dxfer_dir) if (IS_ENABLED(CONFIG_SCSI_PROC_FS) && res) SG_LOG(1, sfp, "%s: blk_rq_map_user() res=%d\n", __func__, res); + } else { /* transfer data to/from kernel buffers */ + res = sg_rq_map_kern(srp, q, rq, r0w); } fini: if (unlikely(res)) { /* failure, free up resources */