We need to hold the queue-lock when checking whether we still have a valid port
handle for the ELS command, i.e whether we can issue this request for this
port. If the error recovery is about to close this port, then it competes for
the queue-lock. If the close request issued by the error recovery wins, then it
is guaranteed that this port has been blocked for other requests.
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: Martin Peschke <mp3@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
zfcp_erp_strategy_check_fsfreq() checks if it is safe to access the
fsf_req associated with the erp_action that gets passed. To test if
it is safe it accesses the fsf_req in order to get its index into
the hash list. This is broken since the fsf_req might be freed already
and the read index has no meaning. It could lead to memory corruption.
Fix this by introducing a new zfcp_reqlist_find_safe() method which
just checks if addresses are equal. This is slower, but only gets
called in case of error recovery.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: Martin Peschke <mp3@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
When adding an invalid LUN, there is a deadlock between the add
via scsi_scan_target and the slave_destroy handler: The handler
waits for the scan to complete, but for an invalid unit,
scsi_scan_target directly calls the slave_destroy handler.
Fix the deadlock by removing the wait in the slave_destroy
handler, it was not necessary anyway.
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
It is not necessary to use jiffies or milliseconds to specify
waiting times that last a couple of seconds.
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Calling zfcp_erp_strategy_check_action() after zfcp_erp_action_to_running()
in zfcp_erp_strategy() might cause an unbalanced up() for erp_ready_sem,
which makes the zfcp recovery fail somewhere along the way:
erp thread processing erp_action:
|
| someone waking up erp thread for erp_action
| |
| | someone else dismissing erp_action:
| | |
V V V
write_lock_irqsave(&adapter->erp_lock, flags);
...
if (zfcp_erp_action_exists(erp_action) == ZFCP_ERP_ACTION_RUNNING) {
zfcp_erp_action_to_ready(erp_action);
up(&adapter->erp_ready_sem); /* first up() for erp_action */
}
write_unlock_irqrestore(&adapter->erp_lock, flags);
write_lock_irqsave(&adapter->erp_lock, flags);
...
zfcp_erp_action_to_running(erp_action);
write_unlock_restore(&adapter->erp_lock, flags);
/* processing erp_action */
write_lock_irqsave(&adapter->erp_lock, flags);
...
erp_action->status |= ZFCP_STATUS_ERP_DISMISSED;
if (zfcp_erp_action_exists(erp_action) ==
ZFCP_ERP_ACTION_RUNNING) {
zfcp_erp_action_to_ready(erp_action);
up(&adapter->erp_ready_sem);
/* second, unbalanced up() for erp_action */
}
...
write_unlock_restore(&adapter->erp_lock, flags);
write_lock_irqsave(&adapter->erp_lock, flags);
if (erp_action->status & ZFCP_STATUS_ERP_DISMISSED) {
zfcp_erp_action_dequeue(erp_action);
retval = ZFCP_ERP_DISMISSED;
}
...
write_unlock_restore(&adapter->erp_lock, flags);
down(&adapter->erp_ready_sem);
/* this down() is meant to balance the first up() */
The erp thread must not dismiss an erp_action after moving that action to
erp_running_head. Instead it should just go through the down() operation,
which balances the first up(), and run through zfcp_erp_strategy one more
time for the second up(), which eventually cleans up erp_action. Which
is similar to the normal processing of an event for erp_action doing
something asynchronously (e.g. waiting for the completion of an fsf_req).
This only works if we make sure that a dismissed erp_action is passed to
zfcp_erp_strategy() prior to the other action, which caused actions to be
dismissed. Therefore the patch implements this rule: running actions go to
the head of the ready list; new actions go to the tail of the ready list;
the erp thread picks actions to be processed from the ready list's head.
Signed-off-by: Martin Peschke <mp3@de.ibm.com>
Acked-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
zfcp_erp_action_dismiss() used to ignore any actions in the ready list. This
is a bug. Any action superseded by a stronger action needs to be dismissed.
This patch changes zfcp_erp_action_dismiss() so that it dismisses actions
regardless of their list affiliation. The ERP thread is able to handle this.
It is important to kick the erp thread only for actions in the running list,
though, as an imbalance of wakeup signals would confuse the erp thread
otherwise.
Signed-off-by: Martin Peschke <mp3@de.ibm.com>
Acked-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Most drivers need to set length and offset as well, so may as well fold
those three lines into one.
Add sg_assign_page() for those two locations that only needed to set
the page, where the offset/length is set outside of the function context.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Cleanup the whitepace from the entire zfcp driver to prevent
to have those changes in future feature or function patches.
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
cleanup, using ERP request mempool for all ERP versions of
the exchange functions (exchange_config (ECD), exchange_port (EPD) )
providing individual versions of the ECD, EPD functions for ERP
and other purposes (_sync).
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Avoid "if (whatever) ;" constructs since they seem to confuse people,
even if there is a comment.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Also removes a bunch of ^L in drivers/s390/cio/cmf.c
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Memory allocated with kmalloc is always initialzed to 0 with memset.
Replace the two calls with kzalloc, that already does both steps.
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
In some cases we did not reset some adapter status flags properly.
This patch clears these flags during FCP adapter shutdown.
Signed-off-by: Volker Sameske <sameske@de.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Must clear adapter failed flag if an fsf request times out. This is
necessary because on link down situations the failed flags gets set
but the QDIO queues are still up. Since an adapter reopen will be
skipped if the failed flag is set an adapter_reopen that is issued
on fsf request timeout has no effect if the local link is down.
Might lead to locked up system if the SCSI stack is waiting for abort
completion.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Simplify request ID management and make sure that frequently used
functions are inlined. Also fix a memory leak in zfcp_adapter_enqueue()
which only gets hit in error handling.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
The SCSI stack requires low level drivers to register and
unregister devices. For zfcp this leads to the situation where
zfcp calls the SCSI stack, the SCSI tries to scan the new device
and the scan SCSI command fails. This would require the zfcp erp,
but the erp thread is already blocked in the register call.
The fix is to make sure that the calls from the ERP thread to
the SCSI stack do not block the ERP thread. In detail:
1) Use a workqueue to avoid blocking of the scsi_scan_target calls.
2) When removing a unit make sure that no scsi_scan_target call is
pending.
3) Replace scsi_flush_work with scsi_target_unblock. This avoids
blocking and has the same result.
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
S_ID and D_ID are defined in the FCP spec as 3 byte fields.
Change the output in zfcp print statements accordingly to print
them with only 3 bytes.
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Fix the fix ... One of my previous fixes introduced removal of all fsf
requests in zfcp's eh_host_reset_handler. But this must not happen
before qdio queues are shut down. So, I revert the changes of
zfcp_scsi_eh_host_reset_handler.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
This instance will be used whenever a timer is needed for
a request by zfcp.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Bug fixes for zfcp's erp:
- trigger adapter reopen if do_QDIO fails
- avoid erp deadlock if registration of scsi target or remote port hang
- do not treat as error if exchange port data fails
- decrease timeout for target reset and aborts
- mark unit failed if slave_destroy is called
Additionally some code cleanup was done:
- made some functions void when retval is not of interest
- shortened initialization of zfcp's host_template
- corrected some comments
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Improve request handling. Use hash table to manage request IDs.
Signed-off-by: Volker Sameske <sameske@de.ibm.com>
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
If zfcp's port erp fails we now call fc_remote_port_delete. This helps
to avoid offlined scsi devices if scsi commands time out due to path
failures. When an adapter erp fails we call fc_remote_port_delete for
all ports on that adapter.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Signed-off-by: Ralph Wuerthner <rwuerthn@de.ibm.com>
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Removed some macros, struct members and typedefs which were
unused or not necessary.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Replace kmalloc/memset by kzalloc or kcalloc.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
The patch fixes following issues:
(1) Replace scsi_add_device with scsi_scan_target.
(Thus the rport instead of the scsi_host becomes parent of a
scsi_target again.)
(2) Avoid scsi_device allocation during registration of an remote port.
(Would be done during fc_scsi_scan_rport.)
(3) Fix queuecommand behaviour when an zfcp unit is blocked.
(Call scsi_done with DID_NO_CONNECT instead of returning
SCSI_MLQUEUE_DEVICE_BUSY otherwise we might end up waiting
for completion in blk_execute_rq for ever.)
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Flush workqueue of a scsi host after a remote port for that host
is registered at the fc transport class. Otherwise immediate
registration of a scsi device on that host is racy.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Remove endless polling for replug of the local link. Just wait for
link up notification.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
- Remove all CVS generated information like e.g. revision IDs from
drivers/s390 and include/asm-s390 (none present in arch/s390).
- Add newline at end of arch/s390/lib/Makefile to avoid diff message.
Acked-by: Andreas Herrmann <aherrman@de.ibm.com>
Acked-by: Frank Pavlic <pavlic@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Replaced zfcp adapter attributes with fc_host attributes:
fc_topology by port_type, physical_wwpn by permanent_port_name.
Make use of fc_host attribute supported_speeds.
Removed zfcp adapter attribute physical_s_id.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Fixed various problems in opening sequence of adapters which was previously
changed with NPIV support:
o corrected handling when exchange port data function is not supported,
otherwise adapters on z900 cannot be opened anymore
o corrected setup of timer for exchange port data if called from error
recovery
o corrected check of return code of exchange config data
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Move initialization of locks and lists to adapter allocation function.
Otherwise we might end up with some uninitialized locks, like e.g. the
erp locks which only will be inititialized if an error recovery thread
for an adapter will be started.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
this patch adds some fc host attributes and removes its equivalents
from the zfcp_adapter structure and zfcp specific sysfs subtree.
Furthermore it removes superfluous calls to fc_remort_port_delete when
an adapter is set offline because rports will be removed by
fc_remove_host anyway.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
N_Port ID Virtualization (NPIV) allows a single FCP port to appear as
multiple, distinct ports providing separate port identification. NPIV
is supported by FC HBAs on System z9. zfcp was adapted to support this
new feature.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
o always use locking when changing erp_action lists,
o avoid escalation to ERP_ACTION_REOPEN_PORT_FORCED if erp_action is
still in use for ERP_ACTION_REOPEN_PORT
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
This patch fixes a severe problem with 2.6.13-rc7.
Due to recent SCSI changes it is not possible to add any LUNs to the zfcp
device driver anymore. With registration of remote ports this is fixed.
Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Acked-by: James Bottomley <jejb@steeleye.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Joe Perches