Skip to contentSkip to conversation

Question about scsi_qla_host_t.fcports list protection (qla2xxx)

DISCUSSION Vladislav Bolkhovitin 2 participants 5 messages Aug 15, 2006 – Sep 5, 2006 SCSI

lore
Share & tools
lore
#0 lore raw

Hello

Looking at the qla2xxx driver source code I noticed that scsi_qla_host_t.fcports is used without any protection. For instance, qla2x00_loop_reset() walks over this list. This function is called from qla2xxx_eh_bus_reset(), which could be called at any time. What does protect fcports from being changed, eg, by some async loop event during the walking and eventually oops'ing? Do I miss something? There are many such unprotected fcports list usages in the driver.

Vlad

#1 lore raw
Quoted 7 lines · Vladislav Bolkhovitin

Looking at the qla2xxx driver source code I noticed that scsi_qla_host_t.fcports is used without any protection. For instance, qla2x00_loop_reset() walks over this list. This function is called from qla2xxx_eh_bus_reset(), which could be called at any time. What does protect fcports from being changed, eg, by some async loop event during the walking and eventually oops'ing? Do I miss something? There are many such unprotected fcports list usages in the driver.

Nothing is ever removed from the fcports list during driver execution. fcports are also only added from a single context (post init-time) within the DPC thread. We are safe here.

Regards,
Andrew Vasquez

#2 lore raw
Quoted 9 lines · Andrew Vasquez
Quoted 1 line · Vladislav Bolkhovitin

Looking at the qla2xxx driver source code I noticed that scsi_qla_host_t.fcports is used without any protection. For instance, qla2x00_loop_reset() walks over this list. This function is called from qla2xxx_eh_bus_reset(), which could be called at any time. What does protect fcports from being changed, eg, by some async loop event during the walking and eventually oops'ing? Do I miss something? There are many such unprotected fcports list usages in the driver.

Nothing is ever removed from the fcports list during driver execution. fcports are also only added from a single context (post init-time) within the DPC thread. We are safe here.

Sorry, I don't think so. For lockless iterating over a list concurrently with the addition to it you need to use *_rcu() list functions and do some RCU locking or, at least, disable preemption. Otherwise it is not safe.

Also, I don't think that keeping dead FC ports entries forever is a good idea. What does prevent you from deleting them, for instance, after fc_remote_port_delete() call?

Vlad

#3 lore raw
In reply to Andrew Vasquez
Quoted 9 lines · Andrew Vasquez
Quoted 1 line · Vladislav Bolkhovitin

Looking at the qla2xxx driver source code I noticed that scsi_qla_host_t.fcports is used without any protection. For instance, qla2x00_loop_reset() walks over this list. This function is called from qla2xxx_eh_bus_reset(), which could be called at any time. What does protect fcports from being changed, eg, by some async loop event during the walking and eventually oops'ing? Do I miss something? There are many such unprotected fcports list usages in the driver.

Nothing is ever removed from the fcports list during driver execution. fcports are also only added from a single context (post init-time) within the DPC thread.

How about qla2x00_send_login_iocb_cb(), where an entry is added to fcports list and which is called from IRQ context from qla2x00_process_iodesc()?

Quoted 5 lines

We are safe here.

Regards,
Andrew Vasquez

#4 lore raw
Quoted 9 lines · Vladislav Bolkhovitin
Quoted 7 lines · Andrew Vasquez
Quoted 1 line · Vladislav Bolkhovitin

Looking at the qla2xxx driver source code I noticed that scsi_qla_host_t.fcports is used without any protection. For instance, qla2x00_loop_reset() walks over this list. This function is called from qla2xxx_eh_bus_reset(), which could be called at any time. What does protect fcports from being changed, eg, by some async loop event during the walking and eventually oops'ing? Do I miss something? There are many such unprotected fcports list usages in the driver.

Quoted 6 lines
Quoted 3 lines

Nothing is ever removed from the fcports list during driver execution. fcports are also only added from a single context (post init-time) within the DPC thread.

How about qla2x00_send_login_iocb_cb(), where an entry is added to fcports list and which is called from IRQ context from qla2x00_process_iodesc()?

Sorry for the noise, I just checked in 2.6.18-rc6 and from it all that code was removed. Anyway, the problem with unsafe fcports list iterating remains valid.

Vlad

Quoted 12 lines
Quoted 6 lines

We are safe here.

Regards,
Andrew Vasquez

-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html