linux-stable.git/drivers/target, branch linux-2.6.38.y

target: Fix task->task_execute_queue=1 clear bug + LUN_RESET OOPs

2011-06-03T01:34:05+00:00

commit af57c3ac9947990da2608561b71f4799eb7795c6 upstream.

This patch fixes a bug where task->task_execute_queue=1 was not being
cleared once se_task had been removed from se_device->execute_task_list,
resulting in an OOPs in core_tmr_lun_reset() for the task->task_active=0
case where transport_remove_task_from_execute_queue() was incorrectly
being called.

This patch fixes two cases in transport_get_task_from_execute_queue()
and transport_remove_task_from_execute_queue() to properly clear
task->task_execute_queue=0 once list_del(&task->t_execute_list) has
been called.

It also adds an explict check in transport_remove_task_from_execute_queue()
to dump_stack + return if called with task->task_execute_queue=0.

Signed-off-by: Nicholas Bellinger 
Signed-off-by: James Bottomley 
Signed-off-by: Greg Kroah-Hartman

target: Fix bug with task_sg chained transport_free_dev_tasks release

2011-06-03T01:34:05+00:00

commit f436677262a5b524ac87675014c6d4e8ee153029 upstream.

This patch addresses a bug in the target core release path for HW
operation where transport_free_dev_tasks() was incorrectly being called
from transport_lun_remove_cmd() while releasing a se_cmd reference and
calling struct target_core_fabric_ops->queue_data_in().

This would result in a OOPs with HW target mode when the release of
se_task->task_sg[] would happen before pci_unmap_sg() can be called in
HW target mode fabric module code.  This patch addresses the issue by
moving transport_free_dev_tasks() from transport_lun_remove_cmd() into
transport_generic_free_cmd(), and adding TRANSPORT_FREE_CMD_INTR and
transport_generic_free_cmd_intr() to allow se_cmd descriptor release
to happen fromfrom within transport_processing_thread() process context
when release of se_cmd is not possible from HW interrupt context.

Signed-off-by: Nicholas Bellinger 
Signed-off-by: James Bottomley 
Signed-off-by: Greg Kroah-Hartman

target: Fix interrupt context bug with stats_lock and core_tmr_alloc_req

2011-06-03T01:34:04+00:00

commit 53ab6709b4d35b1924240854d794482fd7d33d4a upstream.

This patch fixes two bugs wrt to the interrupt context usage of target
core with HW target mode drivers.  It first converts the usage of struct
se_device->stats_lock in transport_get_lun_for_cmd() and core_tmr_lun_reset()
to properly use spin_lock_irq() to address an BUG with CONFIG_LOCKDEP_SUPPORT=y
enabled.

This patch also adds a 'in_interrupt()' check to allow GFP_ATOMIC usage from
core_tmr_alloc_req() to fix a 'sleeping in interrupt context' BUG with HW
target fabrics that require this logic to function.

Signed-off-by: Nicholas Bellinger 
Signed-off-by: James Bottomley 
Signed-off-by: Greg Kroah-Hartman

target: Fix multi task->task_sg[] chaining logic bug

2011-06-03T01:34:04+00:00

commit 97868c8905a1537153d406c4a3aa39a503a5c299 upstream.

This patch fixes a bug in transport_do_task_sg_chain() used by HW target
mode modules with sg_chain() to provide a single sg_next() walkable memory
layout for use with pci_map_sg() and friends.  This patch addresses an
issue with mapping multiple small block max_sector tasks across multiple
struct se_task->task_sg[] mappings for HW target mode operation.

This was causing OOPs with (cmd->t_task->t_tasks_no > 1) I/O traffic for
HW target drivers using transport_do_task_sg_chain(), and has been tested
so far with tcm_fc(openfcoe), tcm_qla2xxx, and ib_srpt fabrics with
t_tasks_no > 1 IBLOCK backends using a smaller max_sectors to trigger the
original issue.

Signed-off-by: Nicholas Bellinger 
Acked-by: Kiran Patil 
Signed-off-by: James Bottomley 
Signed-off-by: Greg Kroah-Hartman

target: Fix volume size misreporting for volumes > 2TB

2011-03-23T20:04:02+00:00

commit 904f0bc482201fa86e75c330d79dfd11be494cf8 upstream.

the target infrastructure fails to send the correct conventional size
to READ_CAPACITY that force a retry with READ_CAPACITY_16, which reads
the capacity for devices > 2TB.  Fix by adding the correct return to
trigger RC(16).

Reported-by: Ben Jarvis 
Signed-off-by: Signed-off-by: Nicholas A. Bellinger 
Signed-off-by: James Bottomley 
Signed-off-by: Greg Kroah-Hartman

[SCSI] target: Fix t_transport_aborted handling in LUN_RESET + active I/O shutdown

2011-02-28T17:23:32+00:00

This patch addresses two outstanding bugs related to
T_TASK(cmd)->t_transport_aborted handling during TMR LUN_RESET and
active I/O shutdown.

This first involves adding two explict t_transport_aborted=1
assignments in core_tmr_lun_reset() in order to signal the task has
been aborted, and updating transport_generic_wait_for_tasks() to skip
sleeping when t_transport_aborted=1 has been set.  This fixes an issue
where transport_generic_wait_for_tasks() would end up sleeping
indefinately when called from fabric module context while TMR
LUN_RESET was happening with long outstanding backend struct se_task
not yet being completed.

The second adds a missing call to
transport_remove_task_from_execute_queue() when
task->task_execute_queue=1 is set in order to fix an OOPs when
task->t_execute_list has not been dropped.  It also fixes the same
case in transport_processing_shutdown() to prevent the issue from
happening during active I/O struct se_device shutdown.

Signed-off-by: Nicholas A. Bellinger 
Signed-off-by: James Bottomley

[SCSI] target: fix use after free detected by SLUB poison

2011-02-12T18:32:41+00:00

This patch moves a large number of memory release paths inside of the
configfs callback target_core_hba_item_ops->release() called from
within fs/configfs/item.c: config_item_cleanup() context.  This patch
resolves the SLUB 'Poison overwritten' warnings.

Signed-off-by: Nicholas A. Bellinger 
Signed-off-by: James Bottomley

[SCSI] target: Remove procfs based target_core_mib.c code

2011-02-12T18:15:47+00:00

This patch removes the legacy procfs based target_core_mib.c code,
and moves the necessary scsi_index_tables functions and defines into
target_core_transport.c and target_core_base.h code to allow existing
fabric independent statistics to function.

This includes the removal of a handful of 'atomic_t mib_ref_count'
counters used in struct se_node_acl, se_session and se_hba to prevent
removal while using seq_list procfs walking logic.

[jejb: fix up compile failures]
Signed-off-by: Nicholas A. Bellinger 
Signed-off-by: James Bottomley

[SCSI] target: Fix SCF_SCSI_CONTROL_SG_IO_CDB breakage

2011-02-12T18:01:42+00:00

This patch fixes a bug introduced during the v4 control CDB emulation
refactoring that broke SCF_SCSI_CONTROL_SG_IO_CDB operation within
transport_map_control_cmd_to_task().  It moves the BUG_ON() into
transport_do_se_mem_map() after the TRANSPORT(dev)->do_se_mem_map()
RAMDISK_DR special case, and adds the proper struct se_mem assignment
when !list_empty() for normal non RAMDISK_DR backend device cases.

Reported-by: Kai-Thorsten Hambrecht 
Signed-off-by: Nicholas A. Bellinger 
Signed-off-by: James Bottomley

[SCSI] target: Fix top-level configfs_subsystem default_group shutdown breakage

2011-02-12T17:39:14+00:00

This patch fixes two bugs uncovered during testing with
slub_debug=FPUZ during module_exit() -> target_core_exit_configfs()
with release of configfs subsystem consumer default groups, namely how
this should be working with
fs/configfs/dir.c:configfs_unregister_subsystem() release logic for
struct config_group->default_group.

The first issue involves configfs_unregister_subsystem() expecting to
walk+drain the top-level subsys->su_group.default_groups directly in
unlink_group(), and not directly from the configfs subsystem consumer
for the top level struct config_group->default_groups.  This patch
drops the walk+drain of subsys->su_group.default_groups from TCM
configfs subsystem consumer code, and moves the top-level
->default_groups kfree() after configfs_unregister_subsystem() has
been called.

The second issue involves calling
core_alua_free_lu_gp(se_global->default_lu_gp) to release the
default_lu_gp->lu_gp_group before configfs_unregister_subsystem() has
been called.  This patches also moves the core_alua_free_lu_gp() call
to release default_lu_group->lu_gp_group after the subsys has been
unregistered.

Finally, this patch explictly clears the
[lu_gp,alua,hba]_cg->default_groups pointers after kfree() to ensure
that no stale memory is picked up from child struct
config_group->default_group[] while configfs_unregister_subsystem() is
called.

Reported-by: Fubo Chen 
Signed-off-by: Nicholas A. Bellinger 
Signed-off-by: James Bottomley