| Message ID | 20171130224456.23100-2-bart.vanassche@wdc.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
On 12/01/2017 01:44 AM, Bart Van Assche wrote: > If scsi_eh_scmd_add() is called concurrently with > scsi_host_queue_ready() while shost->host_blocked > 0 then it can > happen that neither function wakes up the SCSI error handler. Fix > this by making every function that decreases the host_busy counter > wake up the error handler if necessary and by protecting the > host_failed checks with the SCSI host lock. > > Reported-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com> > Fixes: commit 746650160866 ("scsi: convert host_busy to atomic_t") > Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> > Cc: Konstantin Khorenko <khorenko@virtuozzo.com> > Cc: Stuart Hayes <stuart.w.hayes@gmail.com> > Cc: Pavel Tikhomirov <ptikhomirov@virtuozzo.com> > Cc: Christoph Hellwig <hch@lst.de> > Cc: Hannes Reinecke <hare@suse.com> > Cc: Johannes Thumshirn <jthumshirn@suse.de> > Cc: <stable@vger.kernel.org> > --- > drivers/scsi/scsi_error.c | 8 +++++++- > drivers/scsi/scsi_lib.c | 39 ++++++++++++++++++++++++++++----------- > 2 files changed, 35 insertions(+), 12 deletions(-) > > diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c > index 5e89049e9b4e..b22a9a23c74c 100644 > --- a/drivers/scsi/scsi_error.c > +++ b/drivers/scsi/scsi_error.c > @@ -233,19 +233,25 @@ static void scsi_eh_reset(struct scsi_cmnd *scmd) > void scsi_eh_scmd_add(struct scsi_cmnd *scmd) > { > struct Scsi_Host *shost = scmd->device->host; > + enum scsi_host_state shost_state; > unsigned long flags; > int ret; > > WARN_ON_ONCE(!shost->ehandler); > > spin_lock_irqsave(shost->host_lock, flags); > + shost_state = shost->shost_state; > if (scsi_host_set_state(shost, SHOST_RECOVERY)) { > ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); > WARN_ON_ONCE(ret); > } > if (shost->eh_deadline != -1 && !shost->last_reset) > shost->last_reset = jiffies; > - > + if (shost_state != shost->shost_state) { > + spin_unlock_irqrestore(shost->host_lock, flags); > + synchronize_rcu(); We can come here from interrupt context, so may be we should use call_rcu() here instead, possible backtrace: => scsi_eh_scmd_add => scsi_times_out => blk_rq_timed_out => blk_abort_request => ata_qc_schedule_eh => ata_qc_complete => ata_do_link_abort => ata_port_abort => ahci_handle_port_interrupt => ahci_single_irq_intr => __handle_irq_event_percpu => handle_irq_event_percpu => handle_irq_event => handle_edge_irq => handle_irq => do_IRQ > + spin_lock_irqsave(shost->host_lock, flags); > + } > scsi_eh_reset(scmd); > list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); > shost->host_failed++; > diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c > index b6d3842b6809..7d18fb245d7d 100644 > --- a/drivers/scsi/scsi_lib.c > +++ b/drivers/scsi/scsi_lib.c > @@ -318,22 +318,39 @@ static void scsi_init_cmd_errh(struct scsi_cmnd *cmd) > cmd->cmd_len = scsi_command_size(cmd->cmnd); > } > > -void scsi_device_unbusy(struct scsi_device *sdev) > +/* > + * Decrement the host_busy counter and wake up the error handler if necessary. > + * Avoid as follows that the error handler is not woken up if shost->host_busy > + * == shost->host_failed: use synchronize_rcu() in scsi_eh_scmd_add() in > + * combination with an RCU read lock in this function to ensure that this > + * function in its entirety either finishes before scsi_eh_scmd_add() > + * increases the host_failed counter or that it notices the shost state change > + * made by scsi_eh_scmd_add(). > + */ > +static void scsi_dec_host_busy(struct Scsi_Host *shost) > { > - struct Scsi_Host *shost = sdev->host; > - struct scsi_target *starget = scsi_target(sdev); > unsigned long flags; > > + rcu_read_lock(); > atomic_dec(&shost->host_busy); > - if (starget->can_queue > 0) > - atomic_dec(&starget->target_busy); > - > - if (unlikely(scsi_host_in_recovery(shost) && > - (shost->host_failed || shost->host_eh_scheduled))) { > + if (unlikely(scsi_host_in_recovery(shost))) { > spin_lock_irqsave(shost->host_lock, flags); > - scsi_eh_wakeup(shost); > + if (shost->host_failed || shost->host_eh_scheduled) > + scsi_eh_wakeup(shost); > spin_unlock_irqrestore(shost->host_lock, flags); > } > + rcu_read_unlock(); > +} > + > +void scsi_device_unbusy(struct scsi_device *sdev) > +{ > + struct Scsi_Host *shost = sdev->host; > + struct scsi_target *starget = scsi_target(sdev); > + > + scsi_dec_host_busy(shost); > + > + if (starget->can_queue > 0) > + atomic_dec(&starget->target_busy); > > atomic_dec(&sdev->device_busy); > } > @@ -1531,7 +1548,7 @@ static inline int scsi_host_queue_ready(struct request_queue *q, > list_add_tail(&sdev->starved_entry, &shost->starved_list); > spin_unlock_irq(shost->host_lock); > out_dec: > - atomic_dec(&shost->host_busy); > + scsi_dec_host_busy(shost); > return 0; > } > > @@ -2017,7 +2034,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, > return BLK_STS_OK; > > out_dec_host_busy: > - atomic_dec(&shost->host_busy); > + scsi_dec_host_busy(shost); > out_dec_target_busy: > if (scsi_target(sdev)->can_queue > 0) > atomic_dec(&scsi_target(sdev)->target_busy); >
Hi Bart, Bart Van Assche <bart.vanassche@wdc.com> writes: [...] > + if (shost_state != shost->shost_state) { > + spin_unlock_irqrestore(shost->host_lock, flags); > + synchronize_rcu(); > + spin_lock_irqsave(shost->host_lock, flags); > + } Plese correct me if I'm wrong, but once you drop the host lock all assumptions about states it protects are void, aren't they?
On Fri, 2017-12-01 at 09:45 +0100, Johannes Thumshirn wrote: > Bart Van Assche <bart.vanassche@wdc.com> writes: > [...] > > > + if (shost_state != shost->shost_state) { > > + spin_unlock_irqrestore(shost->host_lock, flags); > > + synchronize_rcu(); > > + spin_lock_irqsave(shost->host_lock, flags); > > + } > > Plese correct me if I'm wrong, but once you drop the host lock all > assumptions about states it protects are void, aren't they? Hello Johannes, That's a good question. I think it is safe to drop the host lock at that point because waking up the error handler thread will only happen after host_failed has been incremented. Bart.
On Fri, 2017-12-01 at 11:42 +0300, Pavel Tikhomirov wrote: > On 12/01/2017 01:44 AM, Bart Van Assche wrote: > > + if (shost_state != shost->shost_state) { > > + spin_unlock_irqrestore(shost->host_lock, flags); > > + synchronize_rcu(); > > We can come here from interrupt context, so may be we should use > call_rcu() here instead. Hello Pavel, I will rework this patch such that it uses call_rcu() instead of synchronize_rcu(). Bart.
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c index 5e89049e9b4e..b22a9a23c74c 100644 --- a/drivers/scsi/scsi_error.c +++ b/drivers/scsi/scsi_error.c @@ -233,19 +233,25 @@ static void scsi_eh_reset(struct scsi_cmnd *scmd) void scsi_eh_scmd_add(struct scsi_cmnd *scmd) { struct Scsi_Host *shost = scmd->device->host; + enum scsi_host_state shost_state; unsigned long flags; int ret; WARN_ON_ONCE(!shost->ehandler); spin_lock_irqsave(shost->host_lock, flags); + shost_state = shost->shost_state; if (scsi_host_set_state(shost, SHOST_RECOVERY)) { ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); WARN_ON_ONCE(ret); } if (shost->eh_deadline != -1 && !shost->last_reset) shost->last_reset = jiffies; - + if (shost_state != shost->shost_state) { + spin_unlock_irqrestore(shost->host_lock, flags); + synchronize_rcu(); + spin_lock_irqsave(shost->host_lock, flags); + } scsi_eh_reset(scmd); list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); shost->host_failed++; diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index b6d3842b6809..7d18fb245d7d 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -318,22 +318,39 @@ static void scsi_init_cmd_errh(struct scsi_cmnd *cmd) cmd->cmd_len = scsi_command_size(cmd->cmnd); } -void scsi_device_unbusy(struct scsi_device *sdev) +/* + * Decrement the host_busy counter and wake up the error handler if necessary. + * Avoid as follows that the error handler is not woken up if shost->host_busy + * == shost->host_failed: use synchronize_rcu() in scsi_eh_scmd_add() in + * combination with an RCU read lock in this function to ensure that this + * function in its entirety either finishes before scsi_eh_scmd_add() + * increases the host_failed counter or that it notices the shost state change + * made by scsi_eh_scmd_add(). + */ +static void scsi_dec_host_busy(struct Scsi_Host *shost) { - struct Scsi_Host *shost = sdev->host; - struct scsi_target *starget = scsi_target(sdev); unsigned long flags; + rcu_read_lock(); atomic_dec(&shost->host_busy); - if (starget->can_queue > 0) - atomic_dec(&starget->target_busy); - - if (unlikely(scsi_host_in_recovery(shost) && - (shost->host_failed || shost->host_eh_scheduled))) { + if (unlikely(scsi_host_in_recovery(shost))) { spin_lock_irqsave(shost->host_lock, flags); - scsi_eh_wakeup(shost); + if (shost->host_failed || shost->host_eh_scheduled) + scsi_eh_wakeup(shost); spin_unlock_irqrestore(shost->host_lock, flags); } + rcu_read_unlock(); +} + +void scsi_device_unbusy(struct scsi_device *sdev) +{ + struct Scsi_Host *shost = sdev->host; + struct scsi_target *starget = scsi_target(sdev); + + scsi_dec_host_busy(shost); + + if (starget->can_queue > 0) + atomic_dec(&starget->target_busy); atomic_dec(&sdev->device_busy); } @@ -1531,7 +1548,7 @@ static inline int scsi_host_queue_ready(struct request_queue *q, list_add_tail(&sdev->starved_entry, &shost->starved_list); spin_unlock_irq(shost->host_lock); out_dec: - atomic_dec(&shost->host_busy); + scsi_dec_host_busy(shost); return 0; } @@ -2017,7 +2034,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, return BLK_STS_OK; out_dec_host_busy: - atomic_dec(&shost->host_busy); + scsi_dec_host_busy(shost); out_dec_target_busy: if (scsi_target(sdev)->can_queue > 0) atomic_dec(&scsi_target(sdev)->target_busy);
If scsi_eh_scmd_add() is called concurrently with scsi_host_queue_ready() while shost->host_blocked > 0 then it can happen that neither function wakes up the SCSI error handler. Fix this by making every function that decreases the host_busy counter wake up the error handler if necessary and by protecting the host_failed checks with the SCSI host lock. Reported-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com> Fixes: commit 746650160866 ("scsi: convert host_busy to atomic_t") Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Cc: Konstantin Khorenko <khorenko@virtuozzo.com> Cc: Stuart Hayes <stuart.w.hayes@gmail.com> Cc: Pavel Tikhomirov <ptikhomirov@virtuozzo.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Hannes Reinecke <hare@suse.com> Cc: Johannes Thumshirn <jthumshirn@suse.de> Cc: <stable@vger.kernel.org> --- drivers/scsi/scsi_error.c | 8 +++++++- drivers/scsi/scsi_lib.c | 39 ++++++++++++++++++++++++++++----------- 2 files changed, 35 insertions(+), 12 deletions(-)