[PATCH] fix: scsi: srp_reconnect_rport: unbalanced scsi_block_targets/scsi_target_unblock

[PATCH] fix: scsi: srp_reconnect_rport: unbalanced scsi_block_targets/scsi_target_unblock

[WenTao Liang]

In srp_reconnect_rport(), scsi_block_targets() is called only when
rport->state is not FAIL_FAST and not LOST. However,
scsi_target_unblock() is called unconditionally on the success path and
on some error paths, causing an extra kref_put on sdev_gendev when block
was never called.

Introduce a 'blocked' flag to track whether scsi_block_targets() was
called, and only call scsi_target_unblock() when blocked is true.

Cc: stable@vger.kernel.org
Fixes: 09345f65058b ("[SCSI] add srp transport class")
Signed-off-by: WenTao Liang <vulab@iscas.ac.cn>
---
 drivers/scsi/scsi_transport_srp.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/scsi/scsi_transport_srp.c b/drivers/scsi/scsi_transport_srp.c
index d71ab5fdb758..b9f67e143b52 100644
--- a/drivers/scsi/scsi_transport_srp.c
+++ b/drivers/scsi/scsi_transport_srp.c
@@ -535,6 +535,7 @@ int srp_reconnect_rport(struct srp_rport *rport)
 	struct srp_internal *i = to_srp_internal(shost->transportt);
 	struct scsi_device *sdev;
 	int res;
+	bool blocked = false;
 
 	pr_debug("SCSI host %s\n", dev_name(&shost->shost_gendev));
 
@@ -549,6 +550,7 @@ int srp_reconnect_rport(struct srp_rport *rport)
 		 * treats SDEV_TRANSPORT_OFFLINE like SDEV_BLOCK.
 		 */
 		scsi_block_targets(shost, &shost->shost_gendev);
+		blocked = true;
 	res = rport->state != SRP_RPORT_LOST ? i->f->reconnect(rport) : -ENODEV;
 	pr_debug("%s (state %d): transport.reconnect() returned %d\n",
 		 dev_name(&shost->shost_gendev), rport->state, res);
@@ -558,7 +560,8 @@ int srp_reconnect_rport(struct srp_rport *rport)
 
 		rport->failed_reconnects = 0;
 		srp_rport_set_state(rport, SRP_RPORT_RUNNING);
-		scsi_target_unblock(&shost->shost_gendev, SDEV_RUNNING);
+		if (blocked)
+			scsi_target_unblock(&shost->shost_gendev, SDEV_RUNNING);
 		/*
 		 * If the SCSI error handler has offlined one or more devices,
 		 * invoking scsi_target_unblock() won't change the state of
@@ -579,7 +582,8 @@ int srp_reconnect_rport(struct srp_rport *rport)
 		__rport_fail_io_fast(rport);
 		__srp_start_tl_fail_timers(rport);
 	} else if (rport->state != SRP_RPORT_BLOCKED) {
-		scsi_target_unblock(&shost->shost_gendev,
+		if (blocked)
+			scsi_target_unblock(&shost->shost_gendev,
 				    SDEV_TRANSPORT_OFFLINE);
 	}
 	mutex_unlock(&rport->mutex);
-- 
2.39.5 (Apple Git-154)

Re: [PATCH] fix: scsi: srp_reconnect_rport: unbalanced scsi_block_targets/scsi_target_unblock

[James Bottomley]

On Sat, 2026-06-27 at 00:14 +0800, WenTao Liang wrote:
> In srp_reconnect_rport(), scsi_block_targets() is called only when
> rport->state is not FAIL_FAST and not LOST. However,
> scsi_target_unblock() is called unconditionally on the success path
> and on some error paths, causing an extra kref_put on sdev_gendev
> when block was never called.

I don't believe this is true: block and unblock calls don't have an
unbalanced kref in them (by design, since you're supposed to be able to
have unblock called on a running device and be a no-op).

Regards,

James

Re: [PATCH] fix: scsi: srp_reconnect_rport: unbalanced scsi_block_targets/scsi_target_unblock

[kernel test robot]

Hi WenTao,

kernel test robot noticed the following build warnings:

[auto build test WARNING on jejb-scsi/for-next]
[also build test WARNING on mkp-scsi/for-next linus/master v7.2-rc6 next-20260805]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/WenTao-Liang/fix-scsi-srp_reconnect_rport-unbalanced-scsi_block_targets-scsi_target_unblock/20260805-214818
base:   https://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi.git for-next
patch link:    https://lore.kernel.org/r/20260626161402.55116-1-vulab%40iscas.ac.cn
patch subject: [PATCH] fix: scsi: srp_reconnect_rport: unbalanced   scsi_block_targets/scsi_target_unblock
config: alpha-allmodconfig (https://download.01.org/0day-ci/archive/20260806/202608060507.BrNCKCeH-lkp@intel.com/config)
compiler: alpha-linux-gcc (GCC) 16.1.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260806/202608060507.BrNCKCeH-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202608060507.BrNCKCeH-lkp@intel.com/

All warnings (new ones prefixed by >>):

   drivers/scsi/scsi_transport_srp.c: In function 'srp_reconnect_rport':
>> drivers/scsi/scsi_transport_srp.c:545:9: warning: this 'if' clause does not guard... [-Wmisleading-indentation]
     545 |         if (rport->state != SRP_RPORT_FAIL_FAST && rport->state != SRP_RPORT_LOST)
         |         ^~
   drivers/scsi/scsi_transport_srp.c:553:17: note: ...this statement, but the latter is misleadingly indented as if it were guarded by the 'if'
     553 |                 blocked = true;
         |                 ^~~~~~~


vim +/if +545 drivers/scsi/scsi_transport_srp.c

29c17324803c8a Bart Van Assche 2013-10-26  509  
29c17324803c8a Bart Van Assche 2013-10-26  510  /**
29c17324803c8a Bart Van Assche 2013-10-26  511   * srp_reconnect_rport() - reconnect to an SRP target port
0c7f82189d2249 Bart Van Assche 2014-01-13  512   * @rport: SRP target port.
29c17324803c8a Bart Van Assche 2013-10-26  513   *
29c17324803c8a Bart Van Assche 2013-10-26  514   * Blocks SCSI command queueing before invoking reconnect() such that
29c17324803c8a Bart Van Assche 2013-10-26  515   * queuecommand() won't be invoked concurrently with reconnect() from outside
29c17324803c8a Bart Van Assche 2013-10-26  516   * the SCSI EH. This is important since a reconnect() implementation may
29c17324803c8a Bart Van Assche 2013-10-26  517   * reallocate resources needed by queuecommand().
29c17324803c8a Bart Van Assche 2013-10-26  518   *
29c17324803c8a Bart Van Assche 2013-10-26  519   * Notes:
29c17324803c8a Bart Van Assche 2013-10-26  520   * - This function neither waits until outstanding requests have finished nor
29c17324803c8a Bart Van Assche 2013-10-26  521   *   tries to abort these. It is the responsibility of the reconnect()
29c17324803c8a Bart Van Assche 2013-10-26  522   *   function to finish outstanding commands before reconnecting to the target
29c17324803c8a Bart Van Assche 2013-10-26  523   *   port.
29c17324803c8a Bart Van Assche 2013-10-26  524   * - It is the responsibility of the caller to ensure that the resources
29c17324803c8a Bart Van Assche 2013-10-26  525   *   reallocated by the reconnect() function won't be used while this function
29c17324803c8a Bart Van Assche 2013-10-26  526   *   is in progress. One possible strategy is to invoke this function from
29c17324803c8a Bart Van Assche 2013-10-26  527   *   the context of the SCSI EH thread only. Another possible strategy is to
29c17324803c8a Bart Van Assche 2013-10-26  528   *   lock the rport mutex inside each SCSI LLD callback that can be invoked by
29c17324803c8a Bart Van Assche 2013-10-26  529   *   the SCSI EH (the scsi_host_template.eh_*() functions and also the
29c17324803c8a Bart Van Assche 2013-10-26  530   *   scsi_host_template.queuecommand() function).
29c17324803c8a Bart Van Assche 2013-10-26  531   */
29c17324803c8a Bart Van Assche 2013-10-26  532  int srp_reconnect_rport(struct srp_rport *rport)
29c17324803c8a Bart Van Assche 2013-10-26  533  {
29c17324803c8a Bart Van Assche 2013-10-26  534  	struct Scsi_Host *shost = rport_to_shost(rport);
29c17324803c8a Bart Van Assche 2013-10-26  535  	struct srp_internal *i = to_srp_internal(shost->transportt);
29c17324803c8a Bart Van Assche 2013-10-26  536  	struct scsi_device *sdev;
29c17324803c8a Bart Van Assche 2013-10-26  537  	int res;
cc89c54580eff2 WenTao Liang    2026-06-27  538  	bool blocked = false;
29c17324803c8a Bart Van Assche 2013-10-26  539  
29c17324803c8a Bart Van Assche 2013-10-26  540  	pr_debug("SCSI host %s\n", dev_name(&shost->shost_gendev));
29c17324803c8a Bart Van Assche 2013-10-26  541  
29c17324803c8a Bart Van Assche 2013-10-26  542  	res = mutex_lock_interruptible(&rport->mutex);
29c17324803c8a Bart Van Assche 2013-10-26  543  	if (res)
29c17324803c8a Bart Van Assche 2013-10-26  544  		goto out;
5cd0f6f57639c5 Martin Wilck    2021-04-01 @545  	if (rport->state != SRP_RPORT_FAIL_FAST && rport->state != SRP_RPORT_LOST)
72eeb7c7151302 Martin Wilck    2021-01-11  546  		/*
72eeb7c7151302 Martin Wilck    2021-01-11  547  		 * sdev state must be SDEV_TRANSPORT_OFFLINE, transition
72eeb7c7151302 Martin Wilck    2021-01-11  548  		 * to SDEV_BLOCK is illegal. Calling scsi_target_unblock()
72eeb7c7151302 Martin Wilck    2021-01-11  549  		 * later is ok though, scsi_internal_device_unblock_nowait()
72eeb7c7151302 Martin Wilck    2021-01-11  550  		 * treats SDEV_TRANSPORT_OFFLINE like SDEV_BLOCK.
72eeb7c7151302 Martin Wilck    2021-01-11  551  		 */
31950192d939a9 Martin Wilck    2023-06-14  552  		scsi_block_targets(shost, &shost->shost_gendev);
cc89c54580eff2 WenTao Liang    2026-06-27  553  		blocked = true;
93079162bf0ed2 Bart Van Assche 2013-12-11  554  	res = rport->state != SRP_RPORT_LOST ? i->f->reconnect(rport) : -ENODEV;
29c17324803c8a Bart Van Assche 2013-10-26  555  	pr_debug("%s (state %d): transport.reconnect() returned %d\n",
29c17324803c8a Bart Van Assche 2013-10-26  556  		 dev_name(&shost->shost_gendev), rport->state, res);
29c17324803c8a Bart Van Assche 2013-10-26  557  	if (res == 0) {
29c17324803c8a Bart Van Assche 2013-10-26  558  		cancel_delayed_work(&rport->fast_io_fail_work);
29c17324803c8a Bart Van Assche 2013-10-26  559  		cancel_delayed_work(&rport->dev_loss_work);
29c17324803c8a Bart Van Assche 2013-10-26  560  
8c64e4531c3c3b Bart Van Assche 2013-10-26  561  		rport->failed_reconnects = 0;
29c17324803c8a Bart Van Assche 2013-10-26  562  		srp_rport_set_state(rport, SRP_RPORT_RUNNING);
cc89c54580eff2 WenTao Liang    2026-06-27  563  		if (blocked)
29c17324803c8a Bart Van Assche 2013-10-26  564  			scsi_target_unblock(&shost->shost_gendev, SDEV_RUNNING);
29c17324803c8a Bart Van Assche 2013-10-26  565  		/*
29c17324803c8a Bart Van Assche 2013-10-26  566  		 * If the SCSI error handler has offlined one or more devices,
29c17324803c8a Bart Van Assche 2013-10-26  567  		 * invoking scsi_target_unblock() won't change the state of
29c17324803c8a Bart Van Assche 2013-10-26  568  		 * these devices into running so do that explicitly.
29c17324803c8a Bart Van Assche 2013-10-26  569  		 */
0db6ca8a5e1ea5 Bart Van Assche 2017-06-02  570  		shost_for_each_device(sdev, shost) {
0db6ca8a5e1ea5 Bart Van Assche 2017-06-02  571  			mutex_lock(&sdev->state_mutex);
a817e73fe693f0 Linus Torvalds  2017-11-07  572  			if (sdev->sdev_state == SDEV_OFFLINE)
29c17324803c8a Bart Van Assche 2013-10-26  573  				sdev->sdev_state = SDEV_RUNNING;
0db6ca8a5e1ea5 Bart Van Assche 2017-06-02  574  			mutex_unlock(&sdev->state_mutex);
0db6ca8a5e1ea5 Bart Van Assche 2017-06-02  575  		}
29c17324803c8a Bart Van Assche 2013-10-26  576  	} else if (rport->state == SRP_RPORT_RUNNING) {
29c17324803c8a Bart Van Assche 2013-10-26  577  		/*
18cc4e02508e3f Bart Van Assche 2013-12-11  578  		 * srp_reconnect_rport() has been invoked with fast_io_fail
18cc4e02508e3f Bart Van Assche 2013-12-11  579  		 * and dev_loss off. Mark the port as failed and start the TL
18cc4e02508e3f Bart Van Assche 2013-12-11  580  		 * failure timers if these had not yet been started.
29c17324803c8a Bart Van Assche 2013-10-26  581  		 */
29c17324803c8a Bart Van Assche 2013-10-26  582  		__rport_fail_io_fast(rport);
29c17324803c8a Bart Van Assche 2013-10-26  583  		__srp_start_tl_fail_timers(rport);
29c17324803c8a Bart Van Assche 2013-10-26  584  	} else if (rport->state != SRP_RPORT_BLOCKED) {
cc89c54580eff2 WenTao Liang    2026-06-27  585  		if (blocked)
29c17324803c8a Bart Van Assche 2013-10-26  586  			scsi_target_unblock(&shost->shost_gendev,
29c17324803c8a Bart Van Assche 2013-10-26  587  				    SDEV_TRANSPORT_OFFLINE);
29c17324803c8a Bart Van Assche 2013-10-26  588  	}
29c17324803c8a Bart Van Assche 2013-10-26  589  	mutex_unlock(&rport->mutex);
29c17324803c8a Bart Van Assche 2013-10-26  590  
29c17324803c8a Bart Van Assche 2013-10-26  591  out:
29c17324803c8a Bart Van Assche 2013-10-26  592  	return res;
29c17324803c8a Bart Van Assche 2013-10-26  593  }
29c17324803c8a Bart Van Assche 2013-10-26  594  EXPORT_SYMBOL(srp_reconnect_rport);
29c17324803c8a Bart Van Assche 2013-10-26  595  

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH] fix: scsi: srp_reconnect_rport: unbalanced scsi_block_targets/scsi_target_unblock

[kernel test robot]

Hi WenTao,

kernel test robot noticed the following build warnings:

[auto build test WARNING on jejb-scsi/for-next]
[also build test WARNING on mkp-scsi/for-next linus/master v7.2-rc6 next-20260805]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/WenTao-Liang/fix-scsi-srp_reconnect_rport-unbalanced-scsi_block_targets-scsi_target_unblock/20260805-214818
base:   https://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi.git for-next
patch link:    https://lore.kernel.org/r/20260626161402.55116-1-vulab%40iscas.ac.cn
patch subject: [PATCH] fix: scsi: srp_reconnect_rport: unbalanced   scsi_block_targets/scsi_target_unblock
config: x86_64-kexec (https://download.01.org/0day-ci/archive/20260806/202608060245.Qn5OJAhT-lkp@intel.com/config)
compiler: clang version 22.1.8 (https://github.com/llvm/llvm-project ca7933e47d3a3451d81e72ac174dcb5aa28b59d1)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260806/202608060245.Qn5OJAhT-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202608060245.Qn5OJAhT-lkp@intel.com/

All warnings (new ones prefixed by >>):

>> drivers/scsi/scsi_transport_srp.c:553:3: warning: misleading indentation; statement is not part of the previous 'if' [-Wmisleading-indentation]
     553 |                 blocked = true;
         |                 ^
   drivers/scsi/scsi_transport_srp.c:545:2: note: previous statement is here
     545 |         if (rport->state != SRP_RPORT_FAIL_FAST && rport->state != SRP_RPORT_LOST)
         |         ^
   1 warning generated.


vim +/if +553 drivers/scsi/scsi_transport_srp.c

   509	
   510	/**
   511	 * srp_reconnect_rport() - reconnect to an SRP target port
   512	 * @rport: SRP target port.
   513	 *
   514	 * Blocks SCSI command queueing before invoking reconnect() such that
   515	 * queuecommand() won't be invoked concurrently with reconnect() from outside
   516	 * the SCSI EH. This is important since a reconnect() implementation may
   517	 * reallocate resources needed by queuecommand().
   518	 *
   519	 * Notes:
   520	 * - This function neither waits until outstanding requests have finished nor
   521	 *   tries to abort these. It is the responsibility of the reconnect()
   522	 *   function to finish outstanding commands before reconnecting to the target
   523	 *   port.
   524	 * - It is the responsibility of the caller to ensure that the resources
   525	 *   reallocated by the reconnect() function won't be used while this function
   526	 *   is in progress. One possible strategy is to invoke this function from
   527	 *   the context of the SCSI EH thread only. Another possible strategy is to
   528	 *   lock the rport mutex inside each SCSI LLD callback that can be invoked by
   529	 *   the SCSI EH (the scsi_host_template.eh_*() functions and also the
   530	 *   scsi_host_template.queuecommand() function).
   531	 */
   532	int srp_reconnect_rport(struct srp_rport *rport)
   533	{
   534		struct Scsi_Host *shost = rport_to_shost(rport);
   535		struct srp_internal *i = to_srp_internal(shost->transportt);
   536		struct scsi_device *sdev;
   537		int res;
   538		bool blocked = false;
   539	
   540		pr_debug("SCSI host %s\n", dev_name(&shost->shost_gendev));
   541	
   542		res = mutex_lock_interruptible(&rport->mutex);
   543		if (res)
   544			goto out;
   545		if (rport->state != SRP_RPORT_FAIL_FAST && rport->state != SRP_RPORT_LOST)
   546			/*
   547			 * sdev state must be SDEV_TRANSPORT_OFFLINE, transition
   548			 * to SDEV_BLOCK is illegal. Calling scsi_target_unblock()
   549			 * later is ok though, scsi_internal_device_unblock_nowait()
   550			 * treats SDEV_TRANSPORT_OFFLINE like SDEV_BLOCK.
   551			 */
   552			scsi_block_targets(shost, &shost->shost_gendev);
 > 553			blocked = true;
   554		res = rport->state != SRP_RPORT_LOST ? i->f->reconnect(rport) : -ENODEV;
   555		pr_debug("%s (state %d): transport.reconnect() returned %d\n",
   556			 dev_name(&shost->shost_gendev), rport->state, res);
   557		if (res == 0) {
   558			cancel_delayed_work(&rport->fast_io_fail_work);
   559			cancel_delayed_work(&rport->dev_loss_work);
   560	
   561			rport->failed_reconnects = 0;
   562			srp_rport_set_state(rport, SRP_RPORT_RUNNING);
   563			if (blocked)
   564				scsi_target_unblock(&shost->shost_gendev, SDEV_RUNNING);
   565			/*
   566			 * If the SCSI error handler has offlined one or more devices,
   567			 * invoking scsi_target_unblock() won't change the state of
   568			 * these devices into running so do that explicitly.
   569			 */
   570			shost_for_each_device(sdev, shost) {
   571				mutex_lock(&sdev->state_mutex);
   572				if (sdev->sdev_state == SDEV_OFFLINE)
   573					sdev->sdev_state = SDEV_RUNNING;
   574				mutex_unlock(&sdev->state_mutex);
   575			}
   576		} else if (rport->state == SRP_RPORT_RUNNING) {
   577			/*
   578			 * srp_reconnect_rport() has been invoked with fast_io_fail
   579			 * and dev_loss off. Mark the port as failed and start the TL
   580			 * failure timers if these had not yet been started.
   581			 */
   582			__rport_fail_io_fast(rport);
   583			__srp_start_tl_fail_timers(rport);
   584		} else if (rport->state != SRP_RPORT_BLOCKED) {
   585			if (blocked)
   586				scsi_target_unblock(&shost->shost_gendev,
   587					    SDEV_TRANSPORT_OFFLINE);
   588		}
   589		mutex_unlock(&rport->mutex);
   590	
   591	out:
   592		return res;
   593	}
   594	EXPORT_SYMBOL(srp_reconnect_rport);
   595	

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH] fix: scsi: srp_reconnect_rport: unbalanced scsi_block_targets/scsi_target_unblock

[kernel test robot]

Hi WenTao,

kernel test robot noticed the following build warnings:

[auto build test WARNING on jejb-scsi/for-next]
[also build test WARNING on mkp-scsi/for-next linus/master v7.2-rc6 next-20260805]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/WenTao-Liang/fix-scsi-srp_reconnect_rport-unbalanced-scsi_block_targets-scsi_target_unblock/20260805-214818
base:   https://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi.git for-next
patch link:    https://lore.kernel.org/r/20260626161402.55116-1-vulab%40iscas.ac.cn
patch subject: [PATCH] fix: scsi: srp_reconnect_rport: unbalanced   scsi_block_targets/scsi_target_unblock
config: x86_64-kexec (https://download.01.org/0day-ci/archive/20260806/202608061517.9mRANAFw-lkp@intel.com/config)
compiler: clang version 22.1.3 (https://github.com/llvm/llvm-project e9846648fd6183ee6d8cbdb4502213fcf902a211)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260806/202608061517.9mRANAFw-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202608061517.9mRANAFw-lkp@intel.com/

All warnings (new ones prefixed by >>):

>> drivers/scsi/scsi_transport_srp.c:553:3: warning: misleading indentation; statement is not part of the previous 'if' [-Wmisleading-indentation]
     553 |                 blocked = true;
         |                 ^
   drivers/scsi/scsi_transport_srp.c:545:2: note: previous statement is here
     545 |         if (rport->state != SRP_RPORT_FAIL_FAST && rport->state != SRP_RPORT_LOST)
         |         ^
   1 warning generated.


vim +/if +553 drivers/scsi/scsi_transport_srp.c

   509	
   510	/**
   511	 * srp_reconnect_rport() - reconnect to an SRP target port
   512	 * @rport: SRP target port.
   513	 *
   514	 * Blocks SCSI command queueing before invoking reconnect() such that
   515	 * queuecommand() won't be invoked concurrently with reconnect() from outside
   516	 * the SCSI EH. This is important since a reconnect() implementation may
   517	 * reallocate resources needed by queuecommand().
   518	 *
   519	 * Notes:
   520	 * - This function neither waits until outstanding requests have finished nor
   521	 *   tries to abort these. It is the responsibility of the reconnect()
   522	 *   function to finish outstanding commands before reconnecting to the target
   523	 *   port.
   524	 * - It is the responsibility of the caller to ensure that the resources
   525	 *   reallocated by the reconnect() function won't be used while this function
   526	 *   is in progress. One possible strategy is to invoke this function from
   527	 *   the context of the SCSI EH thread only. Another possible strategy is to
   528	 *   lock the rport mutex inside each SCSI LLD callback that can be invoked by
   529	 *   the SCSI EH (the scsi_host_template.eh_*() functions and also the
   530	 *   scsi_host_template.queuecommand() function).
   531	 */
   532	int srp_reconnect_rport(struct srp_rport *rport)
   533	{
   534		struct Scsi_Host *shost = rport_to_shost(rport);
   535		struct srp_internal *i = to_srp_internal(shost->transportt);
   536		struct scsi_device *sdev;
   537		int res;
   538		bool blocked = false;
   539	
   540		pr_debug("SCSI host %s\n", dev_name(&shost->shost_gendev));
   541	
   542		res = mutex_lock_interruptible(&rport->mutex);
   543		if (res)
   544			goto out;
   545		if (rport->state != SRP_RPORT_FAIL_FAST && rport->state != SRP_RPORT_LOST)
   546			/*
   547			 * sdev state must be SDEV_TRANSPORT_OFFLINE, transition
   548			 * to SDEV_BLOCK is illegal. Calling scsi_target_unblock()
   549			 * later is ok though, scsi_internal_device_unblock_nowait()
   550			 * treats SDEV_TRANSPORT_OFFLINE like SDEV_BLOCK.
   551			 */
   552			scsi_block_targets(shost, &shost->shost_gendev);
 > 553			blocked = true;
   554		res = rport->state != SRP_RPORT_LOST ? i->f->reconnect(rport) : -ENODEV;
   555		pr_debug("%s (state %d): transport.reconnect() returned %d\n",
   556			 dev_name(&shost->shost_gendev), rport->state, res);
   557		if (res == 0) {
   558			cancel_delayed_work(&rport->fast_io_fail_work);
   559			cancel_delayed_work(&rport->dev_loss_work);
   560	
   561			rport->failed_reconnects = 0;
   562			srp_rport_set_state(rport, SRP_RPORT_RUNNING);
   563			if (blocked)
   564				scsi_target_unblock(&shost->shost_gendev, SDEV_RUNNING);
   565			/*
   566			 * If the SCSI error handler has offlined one or more devices,
   567			 * invoking scsi_target_unblock() won't change the state of
   568			 * these devices into running so do that explicitly.
   569			 */
   570			shost_for_each_device(sdev, shost) {
   571				mutex_lock(&sdev->state_mutex);
   572				if (sdev->sdev_state == SDEV_OFFLINE)
   573					sdev->sdev_state = SDEV_RUNNING;
   574				mutex_unlock(&sdev->state_mutex);
   575			}
   576		} else if (rport->state == SRP_RPORT_RUNNING) {
   577			/*
   578			 * srp_reconnect_rport() has been invoked with fast_io_fail
   579			 * and dev_loss off. Mark the port as failed and start the TL
   580			 * failure timers if these had not yet been started.
   581			 */
   582			__rport_fail_io_fast(rport);
   583			__srp_start_tl_fail_timers(rport);
   584		} else if (rport->state != SRP_RPORT_BLOCKED) {
   585			if (blocked)
   586				scsi_target_unblock(&shost->shost_gendev,
   587					    SDEV_TRANSPORT_OFFLINE);
   588		}
   589		mutex_unlock(&rport->mutex);
   590	
   591	out:
   592		return res;
   593	}
   594	EXPORT_SYMBOL(srp_reconnect_rport);
   595	

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki