Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 24 10月, 2011 32 次提交
  3. 11 10月, 2011 1 次提交
  5. 16 9月, 2011 1 次提交
    • R
      target: Fix race between multiple invocations of target_qf_do_work() · bcac364a
      由 Roland Dreier 提交于 
      When work is scheduled with schedule_work(), the work can end up
running on multiple CPUs at the same time -- this happens if
the work is already running on one CPU and schedule_work() is called
on another CPU.  This leads to list corruption with target_qf_do_work(),
which is roughly doing:

	spin_lock(...);
	list_for_each_entry_safe(...) {
		list_del(...);
		spin_unlock(...);

		// do stuff

		spin_lock(...);
	}

With multiple CPUs running this code, one CPU can end up deleting the
list entry that the other CPU is about to work on.

Fix this by splicing the list entries onto a local list and then
operating on that in the work function.  This way, each invocation of
target_qf_do_work() operates on its own local list and so multiple
invocations don't corrupt each other's list.  This also avoids dropping
and reacquiring the lock for each list entry.
Signed-off-by: NRoland Dreier <roland@purestorage.com>
Signed-off-by: NNicholas Bellinger <nab@linux-iscsi.org>
      bcac364a
  7. 23 8月, 2011 6 次提交
    • R
      target: Make locking in transport_deregister_session() IRQ safe · e63a8e19
      由 Roland Dreier 提交于 
      At least the tcm_qla2xxx fabric driver calls into transport_deregister_session()
while holding an IRQ-disabled spinlock, so the inner locking needs to
use spin_lock_irqsave() instead of spin_lock_bh().

This fixes warnings seen with tcm_qla2xxx like:

    WARNING: at kernel/softirq.c:159 local_bh_enable_ip+0x98/0xb0()
    Call Trace:
     [<ffffffff8104e65f>] warn_slowpath_common+0x7f/0xc0
     [<ffffffff8104e6ba>] warn_slowpath_null+0x1a/0x20
     [<ffffffff81055368>] local_bh_enable_ip+0x98/0xb0
     [<ffffffff814d5284>] _raw_spin_unlock_bh+0x14/0x20
     [<ffffffffa027b7f6>] transport_deregister_session+0x96/0x180 [target_core_mod]
     [<ffffffffa00f7731>] tcm_qla2xxx_free_session+0xd1/0x170 [tcm_qla2xxx]
     [<ffffffffa01b9173>] qla_tgt_sess_put+0xc3/0x140 [qla2xxx]
     [<ffffffffa01bf40f>] qla_tgt_stop_phase1+0x8f/0x2c0 [qla2xxx]
     [<ffffffffa00f735e>] tcm_qla2xxx_tpg_store_enable+0x6e/0xd0 [tcm_qla2xxx]
     [<ffffffffa026ca29>] target_fabric_tpg_attr_store+0x39/0x40 [target_core_mod]
     [<ffffffffa00a575d>] configfs_write_file+0xbd/0x120 [configfs]
     [<ffffffff811464a6>] vfs_write+0xc6/0x180
     [<ffffffff811467c1>] sys_write+0x51/0x90
     [<ffffffff814dd382>] system_call_fastpath+0x16/0x1b
Signed-off-by: NRoland Dreier <roland@purestorage.com>
Signed-off-by: NNicholas Bellinger <nab@linux-iscsi.org>
      e63a8e19
    • N
      target: Fix task SGL chaining breakage with transport_allocate_data_tasks · c3c74c7a
      由 Nicholas Bellinger 提交于 
      This patch fixes two bugs associated with transport_do_task_sg_chain()
operation where transport_allocate_data_tasks() was incorrectly setting
task_padded_sg for all tasks, and causing bogus task->task_sg_nents
assignments + OOPsen with fabrics depending upon this code.  The first bit
here adds a task_sg_nents_padded check in transport_allocate_data_tasks()
to include an extra SGL vector when necessary for tasks that expect to
be linked using sg_chain().

The second change involves making transport_do_task_sg_chain() properly
account for the extra SGL vector when task->task_padded_sg is set for
the non trailing ->task_sg or single ->task_sg allocations.  Note this
patch also removes the BUG_ON(!task->task_padded_sg) check within
transport_do_task_sg_chain() as we expect this to happen normally
with the updated logic in transport_allocate_data_tasks(), along with
being bogus for CONTROL_SG_IO_CDB type payloads.

So far this bugfix has been tested with tcm_qla2xxx and iblock backends
in (task_count > 1)( and (task_count == 1) operation.
Reported-by: NKiran Patil <kiran.patil@intel.com>
Cc: Kiran Patil <kiran.patil@intel.com>
Cc: Andy Grover <agrover@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: NNicholas Bellinger <nab@linux-iscsi.org>
      c3c74c7a
    • N
      target: Fix task count > 1 handling breakage and use max_sector page alignment · 525a48a2
      由 Nicholas Bellinger 提交于 
      This patch addresses recent breakage with multiple se_task (task_count > 1)
operation following backend dev->se_sub_dev->se_dev_attrib.max_sectors in new
transport_allocate_data_tasks() code.  The initial bug here was a bogus
task->task_sg_nents assignment in transport_allocate_data_tasks() based on
the passed parameter, which now uses DIV_ROUND_UP(task_size, PAGE_SIZE) to
determine the proper number of per task SGL entries for the (task_count > 1)
case.

This also means we now need to enforce a PAGE_SIZE aligned max_sector count
value for this to work as expected without bringing back the pre v3.1
transport_map_mem_to_sg() logic to handle SGL offsets across multiple tasks.
So this patch adds se_dev_align_max_sectors() to round down max_sectors as
necessary to ensure this alignment via se_dev_set_default_attribs() and
se_dev_align_max_sectors() and keeps it simple for (task_count > 1)
operation.

So far this bugfix has been tested with (task_count > 1) operation
using iscsi-target and iblock backends.
Reported-by: NChris Boot <bootc@bootc.net>
Cc: Kiran Patil <kiran.patil@intel.com>
Cc: Andy Grover <agrover@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: NNicholas Bellinger <nab@linux-iscsi.org>
      525a48a2
    • N
      target: Add missing DATA_SG_IO transport_cmd_get_valid_sectors check · 01cde4d5
      由 Nicholas Bellinger 提交于 
      This patch adds the missing transport_cmd_get_valid_sectors() check for
SCF_SCSI_DATA_SG_IO_CDB type payloads to ensure that a received LBA + range
does not exeed past the end of associated backend struct se_device.

This patch also fixes a bug in the failure path of transport_new_cmd_obj()
where this check can fail, so change to use a signed 'rc' and return '-EINVAL'
to signal proper transport_generic_request_failure() handling.
Signed-off-by: NNicholas Bellinger <nab@linux-iscsi.org>
      01cde4d5
    • N
      target: Fix SYNCHRONIZE_CACHE zero LBA + range breakage · 7abbe7f3
      由 Nicholas Bellinger 提交于 
      This patch fixes a SYNCHRONIZE_CACHE CDB handling bug with IBLOCK/FILEIO
backends where transport_cmd_get_valid_sectors() was incorrectly rejecting
a zero LBA + range CDB from being processed, and returning CHECK_CONDITION.

This includes changing transport_cmd_get_valid_sectors() to return '0' on
success and '-EINVAL' on failure (this makes more sense than sectors),
and to only check transport_cmd_get_valid_sectors() when a non zero LBA +
range SYNCHRONIZE_CACHE operation has been receieved for the non passthrough
case.
Signed-off-by: NNicholas Bellinger <nab@linux-iscsi.org>
      7abbe7f3
    • N
      target: Fix WRITE_SAME usage with transport_get_size · 12850626
      由 Nicholas Bellinger 提交于 
      For all flavours of WRITE_SAME, we only expect to handle a single block
of data-out buffer payload, regardless of the number of logical blocks
presented in the CDB.  This patch changes all flavours of WRITE_SAME in
transport_generic_cmd_sequencer() to pass '1' into transport_get_size()
instead of the extracted 'sectors' to properly handle the default usage
of sg_write_same without the --xferlen parameter.
Reported-by: NEric Seppanen <eric@purestorage.com>
Signed-off-by: NNicholas Bellinger <nab@risingtidesystems.com>
      12850626