| Message ID | 20240208093657.19617-1-pchelkin@ispras.ru |
|---|---|
| State | New |
| Headers | show |
| Series | scsi: elx: efct: adjust error handling inside efct_hw_setup_io | expand |
On Thu, Feb 08, 2024 at 12:36:57PM +0300, Fedor Pchelkin wrote: > IO and WQE buffers are allocated once per HW and can be reused later. If > WQE buffers allocation fails then the whole allocation is marked as failed > but already created IO array internal objects are not freed. hw->io is > freed but not nullified in that specific case - it may become a problem > later as efct_hw_setup_io() is supposed to be reusable for the same HW. > > While at it, use kcalloc instead of kmalloc_array/memset-zero combination > and get rid of some needless NULL assignments: nullifying hw->io[i] > elements just before freeing hw->io is not really useful. > > Found by Linux Verification Center (linuxtesting.org). > > Fixes: 4df84e846624 ("scsi: elx: efct: Driver initialization routines") > Cc: stable@vger.kernel.org > Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru> The patch looks okay. Though I think this funktion leaks all over the place memory as soon we take the error path. Could you also prepare a fix for these path while you are at it? Thanks! Daniel
diff --git a/drivers/scsi/elx/efct/efct_hw.c b/drivers/scsi/elx/efct/efct_hw.c index 5a5525054d71..e5486e6949f9 100644 --- a/drivers/scsi/elx/efct/efct_hw.c +++ b/drivers/scsi/elx/efct/efct_hw.c @@ -487,12 +487,10 @@ efct_hw_setup_io(struct efct_hw *hw) struct efct *efct = hw->os; if (!hw->io) { - hw->io = kmalloc_array(hw->config.n_io, sizeof(io), GFP_KERNEL); + hw->io = kcalloc(hw->config.n_io, sizeof(io), GFP_KERNEL); if (!hw->io) return -ENOMEM; - memset(hw->io, 0, hw->config.n_io * sizeof(io)); - for (i = 0; i < hw->config.n_io; i++) { hw->io[i] = kzalloc(sizeof(*io), GFP_KERNEL); if (!hw->io[i]) @@ -502,10 +500,8 @@ efct_hw_setup_io(struct efct_hw *hw) /* Create WQE buffs for IO */ hw->wqe_buffs = kzalloc((hw->config.n_io * hw->sli.wqe_size), GFP_KERNEL); - if (!hw->wqe_buffs) { - kfree(hw->io); - return -ENOMEM; - } + if (!hw->wqe_buffs) + goto error; } else { /* re-use existing IOs, including SGLs */ @@ -586,10 +582,8 @@ efct_hw_setup_io(struct efct_hw *hw) return 0; error: - for (i = 0; i < hw->config.n_io && hw->io[i]; i++) { + for (i = 0; i < hw->config.n_io && hw->io[i]; i++) kfree(hw->io[i]); - hw->io[i] = NULL; - } kfree(hw->io); hw->io = NULL;
IO and WQE buffers are allocated once per HW and can be reused later. If WQE buffers allocation fails then the whole allocation is marked as failed but already created IO array internal objects are not freed. hw->io is freed but not nullified in that specific case - it may become a problem later as efct_hw_setup_io() is supposed to be reusable for the same HW. While at it, use kcalloc instead of kmalloc_array/memset-zero combination and get rid of some needless NULL assignments: nullifying hw->io[i] elements just before freeing hw->io is not really useful. Found by Linux Verification Center (linuxtesting.org). Fixes: 4df84e846624 ("scsi: elx: efct: Driver initialization routines") Cc: stable@vger.kernel.org Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru> --- drivers/scsi/elx/efct/efct_hw.c | 14 ++++---------- 1 file changed, 4 insertions(+), 10 deletions(-)