The session lock is taken in threads, timers, and bottom halves
like softirqs and tasklets. All the code but
iscsi_conn/session_failure take the session lock with the spin_lock_bh
call. This was done because I thought some offload drivers
would be calling these functions from a irq. They never did,
so this patch has iscsi_conn/session_failure use the bh
locking.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
iscsi_tcp, ib_iser, cxgb*, be2iscsi and bnx2i do not use
the host lock and do not take the session lock against
a irq, so this patch drops the DEF_SCSI_QCMD use. Instead
we just take the session lock and disable bhs.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
This adds a more informative error code and message
for the iscsi scsi eh session drop paths. This allows
you to distinguish if the session was dropped due to
a connection failure vs the iscsi layer dropping
the session due to scsi eh failure processing.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
The patch fixes the following situations where NOP-Out pkt is called for:
- local unsolicited NOP-Out requests (requesting no NOP-In response)
- local NOP-Out responses to unsolicited NOP-In requests
kernel panic is observed due to double session spin_lock requests; one in the
bnx2i_process_nopin_local_cmpl routine in bnx2i_hwi.c and the other in the
iscsi_put_task routine in libiscsi.c
The proposed fix is to export the currently static __iscsi_put_task() routine
and have bnx2i call it directly instead of the iscsi_put_task() routine which
holds the session spin lock.
Signed-off-by: Eddie Wai <eddie.wai@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Reviewed-by: Anil Veerabhadrappa <anilgv@broadcom.com>
Acked-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
Move the mid-layer's ->queuecommand() invocation from being locked
with the host lock to being unlocked to facilitate speeding up the
critical path for drivers who don't need this lock taken anyway.
The patch below presents a simple SCSI host lock push-down as an
equivalent transformation. No locking or other behavior should change
with this patch. All existing bugs and locking orders are preserved.
Additionally, add one parameter to queuecommand,
struct Scsi_Host *
and remove one parameter from queuecommand,
void (*done)(struct scsi_cmnd *)
Scsi_Host* is a convenient pointer that most host drivers need anyway,
and 'done' is redundant to struct scsi_cmnd->scsi_done.
Minimal code disturbance was attempted with this change. Most drivers
needed only two one-line modifications for their host lock push-down.
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Acked-by: James Bottomley <James.Bottomley@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This fixes a regression introduced with this commit:
commit d3305f3407
Author: Mike Christie <michaelc@cs.wisc.edu>
Date: Thu Aug 20 15:10:58 2009 -0500
[SCSI] libiscsi: don't increment cmdsn if cmd is not sent
in 2.6.32.
When I moved the hdr->cmdsn after init_task, I added
a bug when header digests are used. The problem is
that the LLD may calculate the header digest in init_task,
so if we then set the cmdsn after the init_task call we
change what the digest will be calculated by the target.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Cc: Stable Tree <stable@kernel.org>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
* git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6:
[SCSI] qla1280: retain firmware for error recovery
[SCSI] attirbute_container: Initialize sysfs attributes with sysfs_attr_init
[SCSI] advansys: fix regression with request_firmware change
[SCSI] qla2xxx: Updated version number to 8.03.02-k2.
[SCSI] qla2xxx: Prevent sending mbx commands from sysfs during isp reset.
[SCSI] qla2xxx: Disable MSI on qla24xx chips other than QLA2432.
[SCSI] qla2xxx: Check to make sure multique and CPU affinity support is not enabled at the same time.
[SCSI] qla2xxx: Correct vp_idx checking during PORT_UPDATE processing.
[SCSI] qla2xxx: Honour "Extended BB credits" bit for CNAs.
[SCSI] scsi_transport_fc: Make sure commands are completed when rport is offline
[SCSI] libiscsi: Fix recovery slowdown regression
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
We could be failing/stopping a connection due to libiscsi starting
recovery/cleanup, but the xmit path or scsi eh thread path
could be dropping the connection at the same time.
As a result the session->state gets set to failed instead of in
recovery. We end up not blocking the session
and so the replacement timeout never gets started and we only end up
failing the IO when scsi_softirq_done sees that the
cmd has been running for (cmd->allowed + 1) * rq->timeout secs.
We used to fail the IO right away so users are seeing a long
delay when using dm-multipath. This problem was added in
2.6.28.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Cc: stable@kernel.org
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
The iscsi_eh_target_reset has been modified to attempt
target reset only. If it fails, then iscsi_eh_session_reset
will be called.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Jayamohan Kallickal <jayamohank@serverengines.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
This patch resets the cmd timer if cmds started before
the timedout command are making progress. The idea is
that the cmd probably timed out because we are trying
to exeucte too many commands. If it turns out that the
device the IO timedout on was bad or the cmd just got
screwed up but other IO/devs were ok then we will
will figure this out when the cmds ahead of the timed
out one complete ok.
This also fixes a bug where we were sort of detecting
this by setting the last_timeout and last_xfer to the
same value when the task was allocated. That caught
the case where we never got to send any IO for it. However,
if the problem had started right before we started the
new task, then we were forced to wait an extra cmd
timeout seconds to start the scsi eh.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
rename kfifo_put... into kfifo_in... to prevent miss use of old non in
kernel-tree drivers
ditto for kfifo_get... -> kfifo_out...
Improve the prototypes of kfifo_in and kfifo_out to make the kerneldoc
annotations more readable.
Add mini "howto porting to the new API" in kfifo.h
Signed-off-by: Stefani Seibold <stefani@seibold.net>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Move the pointer to the spinlock out of struct kfifo. Most users in
tree do not actually use a spinlock, so the few exceptions now have to
call kfifo_{get,put}_locked, which takes an extra argument to a
spinlock.
Signed-off-by: Stefani Seibold <stefani@seibold.net>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is a new generic kernel FIFO implementation.
The current kernel fifo API is not very widely used, because it has to
many constrains. Only 17 files in the current 2.6.31-rc5 used it.
FIFO's are like list's a very basic thing and a kfifo API which handles
the most use case would save a lot of development time and memory
resources.
I think this are the reasons why kfifo is not in use:
- The API is to simple, important functions are missing
- A fifo can be only allocated dynamically
- There is a requirement of a spinlock whether you need it or not
- There is no support for data records inside a fifo
So I decided to extend the kfifo in a more generic way without blowing up
the API to much. The new API has the following benefits:
- Generic usage: For kernel internal use and/or device driver.
- Provide an API for the most use case.
- Slim API: The whole API provides 25 functions.
- Linux style habit.
- DECLARE_KFIFO, DEFINE_KFIFO and INIT_KFIFO Macros
- Direct copy_to_user from the fifo and copy_from_user into the fifo.
- The kfifo itself is an in place member of the using data structure, this save an
indirection access and does not waste the kernel allocator.
- Lockless access: if only one reader and one writer is active on the fifo,
which is the common use case, no additional locking is necessary.
- Remove spinlock - give the user the freedom of choice what kind of locking to use if
one is required.
- Ability to handle records. Three type of records are supported:
- Variable length records between 0-255 bytes, with a record size
field of 1 bytes.
- Variable length records between 0-65535 bytes, with a record size
field of 2 bytes.
- Fixed size records, which no record size field.
- Preserve memory resource.
- Performance!
- Easy to use!
This patch:
Since most users want to have the kfifo as part of another object,
reorganize the code to allow including struct kfifo in another data
structure. This requires changing the kfifo_alloc and kfifo_init
prototypes so that we pass an existing kfifo pointer into them. This
patch changes the implementation and all existing users.
[akpm@linux-foundation.org: fix warning]
Signed-off-by: Stefani Seibold <stefani@seibold.net>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It is rare to get a queue full with iscsi, because targets seem to
just reduce the iscsi cmd window. However, there is at least
one iscsi target that will throw a queue full when overloaded.
This hooks the iscsi code in to the ramp up/down code, so we
can handle it.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
This implements warm target reset tmf support for
the scsi-ml target reset callback. Previously we would
just drop the session in that callback. This patch will
now try a target reset and if that fails drop the session.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
Patch and mail from both MikeC and HannesR:
Before we're trying to send a PDU we have to check whether a TMF
is active. If so and if the PDU will be affected by the TMF
we should allow only Data-out PDUs to be sent.
If fast_abort is set, no Data-out PDUs will be sent while
a LUN reset is being processed for a affected LUN.
fast_abort is now ingored during a ABORT TASK tmf. We will not
send any Data-outs for a task if the task is being aborted.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
For some reason we used to check for the the immediate bit
set and the opcocde in many places instead of just masking
the opcode. In the passthrough code this is a problem
because userspace may or may not have set the immediate bit
and it does not have to. This fixes up the opcode checks
in the passthrough code, so we mask off the opcode then
check against the iscsi proto definition like is done in
other places.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
This patch modifies scsi_host_template->change_queue_depth so that
it takes an argument indicating why it is being called. This will be
used so that if a LLD needs to do some extra processing when
handling queue fulls or later ramp ups, it can do so.
This is a simple port of the drivers setting a change_queue_depth
callback. In the patch I just have these LLDs adjust the queue depth
if the user was requesting it.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
[Vasu.Dev: v2
Also converted pmcraid_change_queue_depth and then verified
all modules compile using "make allmodconfig" for any new build
warnings on X86_64.
Updated original description after combing two original
patches from Mike to make this patch git bisectable.]
Signed-off-by: Vasu Dev <vasu.dev@intel.com>
[jejb: fixed up 53c700]
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
This patch contains changes that allow iscsi_session_setup
to allocate private space for LLD's
Signed-off-by: Jayamohan Kallickal <jayamohank@serverengines.com>
Acked-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
bnx2i currently has a check for if a ep is properly bound, so if
iscsi_queuecommand/xmit_task is called while there is no ep
we will not queue IO.
be2iscsi sends IO from queuecommand/xmit_task like how bnx2i does
and needs a similar test. This patch has us just use the suspend_bit
test for this.
When ep_poll has succeeed iscsid will call conn_bind, the LLD will
then call iscsi_conn_bind which will clear the suspend bit.
When ep_disconnect is called (or if there is a conn error) we set
the suspend bit. For the ep_disconnect case I am adding a helper
in this patch that will take the session lock to make sure
iscsi_queuecommand/xmit_task is not running and it will set
the suspend bit.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Jayamohan Kallickal <jayamohank@serverengines.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
beiscsi does not need the iscsi scsi cmd processing. It does not
even get this info on the completion path. This adds a function
to just update the sequencing numbers and complete a task.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Jayamohan Kallickal <jayamohank@serverengines.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
If we had multiple tasks on the cmd or requeue lists, and iscsi_tcp
returns a error, the write_space function can still run and queue
iscsi_data_xmit. If it was a legetimate problem and iscsi_conn_failure
was run but we raced and iscsi_data_xmit was run first it could miss
the suspend bit checks, and start trying to send data again and hit
another timeout. A similar problem is present when using cxgb3i.
This has libiscsi check the suspend bit before calling the xmit
task callout, so we at least do not try sending multiple tasks
(one could be sent).
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
If we sent multiple pdus as immediate the target could be
rejecting some and we have just been dropping the rejection
notification. This adds code to handle nop-out rejections,
so if a nop-out was sent as a ping and rejected we do not
mark the connection bad. Instead we just clean up the timers
since we have pdu making a rount trip we know the connection
is good.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
We increment session->cmdsn at the top of iscsi_prep_scsi_cmd_pdu, but
if the prep ecb or prep bidi or init_task calls fails then we leave the
session->cmdsn incremented. This moves the cmdsn manipulation to the end
of the function when we know it has succeeded.
It also adds a session->cmdsn--; in queuecommand for if a driver like
bnx2i tries to send a a task from that context but it fails. We do not
have to do this in the xmit thread context because that code will retry
the same task if the initial call fails.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
The session lock can be held in the scsi eh thread or the completion
paths run from the net softirq. This disables bhs in iscsi_eh_abort when
taking the session lock.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Allow the user to control the debug logs in libiscsi. We will now
have a module param for connection, session & error handling.
[Mike Christie - Fixed up to compile on current code and added
missing ISCSI_DBG_EH conversions]
Signed-off-by: Erez Zilber <erezzi.list@gmail.com>
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If we are sending or receiving data for the task successfully do
not run the scsi eh, because we know the task is making progress.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
This patch just adds some debug statements for the abort
and completion paths.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If a task did not complete normally due to a TMF, libiscsi will
now complete the task with the state ISCSI_TASK_ABRT_TMF. Drivers
like bnx2i that need to free resources if a command did not complete normally
can then check the task state. If a driver does not need to send
a special command if we have dropped the session then they can check
for ISCSI_TASK_ABRT_SESS_RECOV.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Instead of having libiscsi check if the offload bit is set, have
it check if the lld created a work queue. I think this is more
clear.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If the session is failed, but we have not yet fully transitioned
to the recovery stage we were still queueuing IO. The idea is
that for some failures we can recvover at the command level
and still continue to execute other IO. Well, we never have
added the recovery within a command code, so queueing up IO here
just creates the possibility that it might time time out so
this just has us requeue the IO the scsi layer for now.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
bnx2i needs to send a hardware specific cleanup command if
a command has not completed normally (iscsi/scsi response from
target), and the session is still ok (this is the case when we
send a TMF to stop the command).
At this time it will need to drop the session lock. The problem
with the current code is that fail_all_commands assumes we
will hold the lock the entire time, so it uses list_for_each_entry_safe.
If while bnx2i drops the session lock multiple cmds complete then
list_for_each_entry_safe will not handle this correctly.
This patch removes the running lists and just has us loop over
the cmds array (in later patches we will then replace that
array with a block tag map at the session level). It also fixes
up the completion path so that if the TMF code and the normal recv
path were completing the same command then they both do not try
to do release the refcount taken when the task is queued.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If we have not got any pdus for recv_timeout seconds, then we will
send a iscsi ping/nop to make sure the target is still around. The
problem is if this is a slow link, and the ping got queued after
the data for a data_out (read), then the transport code could think
the ping has failed when it is just slowly making its way through
the network. This patch has us check if we are making progress while
the nop is outstanding. If we are still reading in data, then we
do not fail the session at that time.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If we are responding to a nop from the target by sending our nop,
and the session is getting torn down, then iscsi_start_session_recovery
could set the conn stop bits while the recv path is sending the nop
response and we will hit the bug ons in __iscsi_conn_send_pdu.
This has us check the state in __iscsi_conn_send_pdu and fail all
incoming mgmt IO if we are not logged in and if the pdu is not login
related. It also changes the ordering of the setting of conn stop state
bits so they are set after the session state is set (both are set under
the session lock).
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
This has iscsi_data_in_rsp call iscsi_update_cmdsn when a pdu is
completed like is done for other pdu's that are don.
For libiscsi_tcp, this means that it calls iscsi_update_cmdsn when
it is handling the pdu internally to only transfer data, but if there is
status then it does not need to call it since the completion handling
will do it.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
bnx2i needs to be able to look up mgmt task like login and nop, because
it does some processing of them on the completion path. This exports
iscsi_itt_to_task so it can look up the task.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If we could not allocate the initiator name or some other id like
the hwaddress or netdev, then userspace could deal with the failure
by just running in a dregraded mode.
Now we want to be able to switch values for the params and we
want some feedback, so this patch will check if a string like
the initiatorname could not be allocated and return an error.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
bnx2i does not have one. It currently preallocates the bdt
when the session is setup.
We probably want to change that to a dma pool, then allocate from
the pool in the alloc pdu. Until then check if there is a alloc
pdu callout.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Set target can queue limit to the number of preallocated
session tasks we have.
This along with the cxgb3i can_queue patch will fix a throughput
problem where it could only queue one LU worth of data at a time.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Le lundi 30 mars 2009, Chris Wright a écrit :
> q->queue could be ERR_PTR(-ENOMEM) which will break unwinding
> on error. Make iscsi_pool_free more defensive.
>
Making the freeing of q->queue dependent on q->pool being set looks
really weird (although it is correct at the moment. But this seems
to be fixable in a much simpler way.
With the benefit that only the error case is slowed down. In both
cases we have a problem if q->queue contains an error value but it's
not -ENOMEM. Apparently this can't happen today, but it doesn't feel
right to assume this will always be true. Maybe it's the right time
to fix this as well.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If the iscsi eh fires when the current task is a nop, then
the task->sc pointer is null. fail_all_commands could
then try to do task->sc->device and oops. We actually do
not need to access the curr task in this path, because
if it is a cmd task the fail_command call will handle
this and if it is mgmt task then the flush of the mgmt
queues will handle that.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The api for conn and session failures is akward because
one takes a conn from the lib and one takes a session
from the class. This syncs up the interfaces to use
structs from the lib.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The qdepth setting was useful when we needed libiscsi to verify
the setting. Now we just need to make sure if older tools
passed in zero then we need to set some default.
So this patch just has us use the sht->cmd_per_lun or if
for LLD does a host per session then we can set it on per
host basis.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
We were using the shost work queue which ended up being
a little akward since all iscsi hosts need a thread for
scanning, but only drivers hooked into libiscsi need
a workqueue for transmitting. So this patch moves the
xmit workqueue to the lib.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
There is no need to cap the queue depth in the modules. We set
this in userspace and can do that there. For performance testing
with ram based targets, this is helpful since we can have very
high queue depths.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
This makes the logging a compile time option and replaces
the scsi_debug macro with session and connection ones
that print out a driver model id prefix.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Memory freeing in iscsi_pool_free() looks wrong to me. Either q->pool
can be NULL and this should be tested before dereferencing it, or it
can't be NULL and it shouldn't be tested at all. As far as I can see,
the only case where q->pool is NULL is on early error in
iscsi_pool_init(). One possible way to fix the bug is thus to not
call iscsi_pool_free() in this case (nothing needs to be freed anyway)
and then we can get rid of the q->pool check.
Signed-off-by: Jean Delvare <jdelvare@suse.de>
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Yanling Qi from LSI found the root cause of the panic, below is his
analysis:
Problem description: the open iscsi driver installs eh_timed_out handler
to the
blank_transport_template of the scsi middle level that causes panic of
timed
out command of other host
Here are the details
Iscsi Session creation
During iscsi session creation time, the iscsi_tcp_session_create() of
iscsi_tpc.c will create a scsi-host for the session. See the statement
marked
with the label A. The statement B replaces the shost->transportt point
with a
local struct variable.
static struct iscsi_cls_session *
iscsi_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
uint16_t qdepth, uint32_t initial_cmdsn,
uint32_t *hostno)
{
struct iscsi_cls_session *cls_session;
struct iscsi_session *session;
struct Scsi_Host *shost;
int cmd_i;
if (ep) {
printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
return NULL;
}
A shost = iscsi_host_alloc(&iscsi_sht, 0, qdepth);
if (!shost)
return NULL;
B shost->transportt = iscsi_tcp_scsi_transport;
shost->max_lun = iscsi_max_lun;
Please note the scsi host is allocated by invoking isccsi_host_alloc()
in
libiscsi.c
Polluting the middle level blank_transport_template in
iscsi_host_alloc() of
libiscsi.c
The iscsi_host_alloc() invokes the middle level function
scsi_host_alloc() in
hosts.c for allocating a scsi_host. Then the statement marked with C
assigns
the iscsi_eh_cmd_timed_out handler to the eh_timed_out callback
function.
struct Scsi_Host *iscsi_host_alloc(struct scsi_host_template *sht,
int dd_data_size, uint16_t qdepth)
{
struct Scsi_Host *shost;
struct iscsi_host *ihost;
shost = scsi_host_alloc(sht, sizeof(struct iscsi_host) +
dd_data_size);
if (!shost)
return NULL;
C shost->transportt->eh_timed_out = iscsi_eh_cmd_timed_out;
Please note the shost->transport is the middle level
blank_transport_template
as shown in the code segment below. We see two problems here. 1.
iscsi_eh_cmd_timed_out is installed to the blank_transport_template that
will
cause some body else problem. 2. iscsi_eh_cmd_timed_out will never be
invoked
when iscsi command gets timeout because the statement B resets the
pointer.
Middle level blank_transport_template
In the middle level function scsi_host_alloc() of hosts.c, the middle
level
assigns a blank_transport_template for those hosts not implementing its
transport layer. All HBAs without supporting a specific scsi_transport
will
share the middle level blank_transport_template. Please see the
statement D
struct Scsi_Host *scsi_host_alloc(struct scsi_host_template *sht, int
privsize)
{
struct Scsi_Host *shost;
gfp_t gfp_mask = GFP_KERNEL;
int rval;
if (sht->unchecked_isa_dma && privsize)
gfp_mask |= __GFP_DMA;
shost = kzalloc(sizeof(struct Scsi_Host) + privsize, gfp_mask);
if (!shost)
return NULL;
shost->host_lock = &shost->default_lock;
spin_lock_init(shost->host_lock);
shost->shost_state = SHOST_CREATED;
INIT_LIST_HEAD(&shost->__devices);
INIT_LIST_HEAD(&shost->__targets);
INIT_LIST_HEAD(&shost->eh_cmd_q);
INIT_LIST_HEAD(&shost->starved_list);
init_waitqueue_head(&shost->host_wait);
mutex_init(&shost->scan_mutex);
shost->host_no = scsi_host_next_hn++; /* XXX(hch): still racy */
shost->dma_channel = 0xff;
/* These three are default values which can be overridden */
shost->max_channel = 0;
shost->max_id = 8;
shost->max_lun = 8;
/* Give each shost a default transportt */
D shost->transportt = &blank_transport_template;
Why we see panic at iscsi_eh_cmd_timed_out()
The mpp virtual HBA doesn’t have a specific scsi_transport. Therefore,
the
blank_transport_template will be assigned to the virtual host of the MPP
virtual HBA by SCSI middle level. Please note that the statement C has
assigned
iscsi-transport eh_timedout handler to the blank_transport_template.
When a mpp
virtual command gets timedout, the iscsi_eh_cmd_timed_out() will be
invoked to
handle mpp virtual command timeout from the middle level
scsi_times_out()
function of the scsi_error.c.
enum blk_eh_timer_return scsi_times_out(struct request *req)
{
struct scsi_cmnd *scmd = req->special;
enum blk_eh_timer_return (*eh_timed_out)(struct scsi_cmnd *);
enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
scsi_log_completion(scmd, TIMEOUT_ERROR);
if (scmd->device->host->transportt->eh_timed_out)
E eh_timed_out =
scmd->device->host->transportt->eh_timed_out;
else if (scmd->device->host->hostt->eh_timed_out)
eh_timed_out = scmd->device->host->hostt->eh_timed_out;
else
eh_timed_out = NULL;
if (eh_timed_out) {
rtn = eh_timed_out(scmd);
It is very easy to understand why we get panic in the
iscsi_eh_cmd_timed_out().
A scsi_cmnd from a no-iscsi device definitely can not resolve out a
session and
session->lock. The panic can be happed anywhere during the differencing.
static enum blk_eh_timer_return iscsi_eh_cmd_timed_out(struct scsi_cmnd
*scmd)
{
struct iscsi_cls_session *cls_session;
struct iscsi_session *session;
struct iscsi_conn *conn;
enum blk_eh_timer_return rc = BLK_EH_NOT_HANDLED;
cls_session = starget_to_session(scsi_target(scmd->device));
session = cls_session->dd_data;
debug_scsi("scsi cmd %p timedout\n", scmd);
spin_lock(&session->lock);
This patch fixes the problem by moving the setting of the
iscsi_eh_cmd_timed_out to iscsi_add_host, which is after the LLDs
have set their transport template to shost->transportt.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Mike Christie