Linux SCSI subsystem development
 help / color / mirror / Atom feed
* [PATCH] scsi: lpfc: Look up the VMID entry under vmid_lock
@ 2026-11-16  2:39 Changyul Lee
  2026-11-16  3:01 ` sashiko-bot
  0 siblings, 1 reply; 2+ messages in thread
From: Changyul Lee @ 2026-11-16  2:39 UTC (permalink / raw)
  To: Justin Tee, Paul Ely
  Cc: Martin K . Petersen, James E . J . Bottomley, Hannes Reinecke,
	Nigel Kirkland, linux-scsi, linux-kernel, Changyul Lee

lpfc_cmpl_ct_cmd_vmid() looks up the VMID entry for an RAPP_IDENT
response without vport->vmid_lock and takes the lock only after the
lookup. lpfc_vmid_get_appid() can meanwhile hash_del() another entry
from the same bucket under the lock. hash_del() clears the next pointer
of the removed node, so the walk stops early and misses the entry that
waits for the response. That entry keeps LPFC_VMID_REQ_REGISTER, and
the I/O of that VM is sent without a VMID tag from then on.

Take vmid_lock before the lookup, as lpfc_vmid_get_appid() does.

Fixes: 33c79741deaf ("scsi: lpfc: vmid: Introduce VMID in I/O path")
Signed-off-by: Changyul Lee <lcy8047@gmail.com>
---
Found with a static analysis tool we are developing.

Compile-tested only, with x86_64_defconfig plus SCSI_LPFC and W=1.

 drivers/scsi/lpfc/lpfc_ct.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/scsi/lpfc/lpfc_ct.c b/drivers/scsi/lpfc/lpfc_ct.c
index f709e5087577..af720ccdf1b6 100644
--- a/drivers/scsi/lpfc/lpfc_ct.c
+++ b/drivers/scsi/lpfc/lpfc_ct.c
@@ -3613,14 +3613,13 @@ lpfc_cmpl_ct_cmd_vmid(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
 
 		hash = lpfc_vmid_hash_fn(app->obj.entity_id,
 					 app->obj.entity_id_len);
+		write_lock(&vport->vmid_lock);
 		vmp = lpfc_get_vmid_from_hashtable(vport, hash,
 						  app->obj.entity_id);
 		if (vmp) {
-			write_lock(&vport->vmid_lock);
 			vmp->un.app_id = be32_to_cpu(app->app_id);
 			vmp->flag |= LPFC_VMID_REGISTERED;
 			vmp->flag &= ~LPFC_VMID_REQ_REGISTER;
-			write_unlock(&vport->vmid_lock);
 			/* Set IN USE flag */
 			vport->vmid_flag |= LPFC_VMID_IN_USE;
 		} else {
@@ -3628,6 +3627,7 @@ lpfc_cmpl_ct_cmd_vmid(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
 					 "6901 No entry found %s hash %d\n",
 					 app->obj.entity_id, hash);
 		}
+		write_unlock(&vport->vmid_lock);
 		break;
 	case SLI_CTAS_DAPP_IDENT:
 		app = (struct app_id_object *)(DAPP_IDENT_OFFSET + data);

base-commit: 7475ea8b58616b33c04afb976cf0fbaad9dd63fb
-- 
2.43.0


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] scsi: lpfc: Look up the VMID entry under vmid_lock
  2026-11-16  2:39 [PATCH] scsi: lpfc: Look up the VMID entry under vmid_lock Changyul Lee
@ 2026-11-16  3:01 ` sashiko-bot
  0 siblings, 0 replies; 2+ messages in thread
From: sashiko-bot @ 2026-11-16  3:01 UTC (permalink / raw)
  To: Changyul Lee; +Cc: linux-scsi

Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Low] scsi: lpfc: inaccurate Fixes trailer for lpfc_cmpl_ct_cmd_vmid()
  lock fix
It also noticed 1 related pre-existing issue this patch does not
introduce:
- [High] scsi: lpfc: data race and list corruption in
  lpfc_cmpl_ct_cmd_vmid() [1]
--

commit ace8e21fd4e560f69d845916087ba204329824c3
Author: Changyul Lee <lcy8047@gmail.com>

> scsi: lpfc: Look up the VMID entry under vmid_lock
>
> lpfc_cmpl_ct_cmd_vmid() looks up the VMID entry for an RAPP_IDENT
> response without vport->vmid_lock and takes the lock only after the
> lookup. lpfc_vmid_get_appid() can meanwhile hash_del() another entry
> from the same bucket under the lock. hash_del() clears the next pointer
> of the removed node, so the walk stops early and misses the entry that
> waits for the response. That entry keeps LPFC_VMID_REQ_REGISTER, and
> the I/O of that VM is sent without a VMID tag from then on.
>
> Take vmid_lock before the lookup, as lpfc_vmid_get_appid() does.
>
> Fixes: 33c79741deaf ("scsi: lpfc: vmid: Introduce VMID in I/O path")
> Signed-off-by: Changyul Lee <lcy8047@gmail.com>

[Severity: Low]
33c79741deaf ("scsi: lpfc: vmid: Introduce VMID in I/O path") only touched
drivers/scsi/lpfc/lpfc_scsi.c.  The unlocked lpfc_get_vmid_from_hashtable()
call in lpfc_cmpl_ct_cmd_vmid() was added by 742b0cf87a8f ("scsi: lpfc:
vmid: Implement CT commands for appid"), so should the Fixes tag be:

  Fixes: 742b0cf87a8f ("scsi: lpfc: vmid: Implement CT commands for appid")

> diff --git a/drivers/scsi/lpfc/lpfc_ct.c b/drivers/scsi/lpfc/lpfc_ct.c
> index f709e50875770..af720ccdf1b6f 100644
> --- a/drivers/scsi/lpfc/lpfc_ct.c
> +++ b/drivers/scsi/lpfc/lpfc_ct.c
> @@ -3613,14 +3613,13 @@ lpfc_cmpl_ct_cmd_vmid(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
>  
>  		hash = lpfc_vmid_hash_fn(app->obj.entity_id,
>  					 app->obj.entity_id_len);
> +		write_lock(&vport->vmid_lock);
>  		vmp = lpfc_get_vmid_from_hashtable(vport, hash,
>  						  app->obj.entity_id);
>  		if (vmp) {
> -			write_lock(&vport->vmid_lock);
>  			vmp->un.app_id = be32_to_cpu(app->app_id);
>  			vmp->flag |= LPFC_VMID_REGISTERED;
>  			vmp->flag &= ~LPFC_VMID_REQ_REGISTER;
> -			write_unlock(&vport->vmid_lock);
>  			/* Set IN USE flag */
>  			vport->vmid_flag |= LPFC_VMID_IN_USE;
>  		} else {
> @@ -3628,6 +3627,7 @@ lpfc_cmpl_ct_cmd_vmid(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
>  					 "6901 No entry found %s hash %d\n",
>  					 app->obj.entity_id, hash);
>  		}
> +		write_unlock(&vport->vmid_lock);
>  		break;
>  	case SLI_CTAS_DAPP_IDENT:
>  		app = (struct app_id_object *)(DAPP_IDENT_OFFSET + data);

[Severity: High]
This problem wasn't introduced by this patch, but the SLI_CTAS_DALLAPP_ID
branch right below in lpfc_cmpl_ct_cmd_vmid() causes a NULL pointer
dereference, list corruption, and a memory leak [1]:

drivers/scsi/lpfc/lpfc_ct.c:lpfc_cmpl_ct_cmd_vmid() {
	...
	case SLI_CTAS_DALLAPP_ID:
		...
		read_lock(&vport->vmid_lock);
		for (i = 0; i < phba->cfg_max_vmid; i++) {
			vmp = &vport->vmid[i];
			if (vmp->flag != LPFC_VMID_SLOT_FREE)
				memset(vmp, 0, sizeof(struct lpfc_vmid));
		}
		read_unlock(&vport->vmid_lock);
		/* for all elements in the hash table */
		if (!hash_empty(vport->hash_table))
			hash_for_each(vport->hash_table, bucket, cur, hnode)
				hash_del(&cur->hnode);
	...
}

Can this crash when deregistering all app IDs?  Zeroing vmp with memset()
while its hnode is still linked in vport->hash_table clears hnode.next and
hnode.pprev, so the subsequent hash_del(&cur->hnode) dereferences a NULL
pprev pointer.  In addition, the loop mutates vport->vmid under read_lock()
instead of write_lock(), overwrites vmp->last_io_time without calling
free_percpu(), and walks vport->hash_table without holding vport->vmid_lock
or using hash_for_each_safe().

[1] https://sashiko.dev/#/bug/linux-b856b4ae-5c8f-4fb3-812c-1d152bd8d42f

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261116023939.588838-1-lcy8047@gmail.com?part=1

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-16  3:01 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16  2:39 [PATCH] scsi: lpfc: Look up the VMID entry under vmid_lock Changyul Lee
2026-11-16  3:01 ` sashiko-bot

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox