kernel.git/drivers/target/target_core_device.c, branch linux-4.9.y

scsi: target: Fix ordered tag handling

2021-11-26T10:48:41Z

[ Upstream commit ed1227e080990ffec5bf39006ec8a57358e6689a ] This patch fixes the following bugs: 1. If there are multiple ordered cmds queued and multiple simple cmds completing, target_restart_delayed_cmds() could be called on different CPUs and each instance could start a ordered cmd. They could then run in different orders than they were queued. 2. target_restart_delayed_cmds() and target_handle_task_attr() can race where: 1. target_handle_task_attr() has passed the simple_cmds == 0 check. 2. transport_complete_task_attr() then decrements simple_cmds to 0. 3. transport_complete_task_attr() runs target_restart_delayed_cmds() and it does not see any cmds on the delayed_cmd_list. 4. target_handle_task_attr() adds the cmd to the delayed_cmd_list. The cmd will then end up timing out. 3. If we are sent > 1 ordered cmds and simple_cmds == 0, we can execute them out of order, because target_handle_task_attr() will hit that simple_cmds check first and return false for all ordered cmds sent. 4. We run target_restart_delayed_cmds() after every cmd completion, so if there is more than 1 simple cmd running, we start executing ordered cmds after that first cmd instead of waiting for all of them to complete. 5. Ordered cmds are not supposed to start until HEAD OF QUEUE and all older cmds have completed, and not just simple. 6. It's not a bug but it doesn't make sense to take the delayed_cmd_lock for every cmd completion when ordered cmds are almost never used. Just replacing that lock with an atomic increases IOPs by up to 10% when completions are spread over multiple CPUs and there are multiple sessions/ mqs/thread accessing the same device. This patch moves the queued delayed handling to a per device work to serialze the cmd executions for each device and adds a new counter to track HEAD_OF_QUEUE and SIMPLE cmds. We can then check the new counter to determine when to run the work on the completion path. Link: https://lore.kernel.org/r/20210930020422.92578-3-michael.christie@oracle.com Signed-off-by: Mike Christie Signed-off-by: Martin K. Petersen Signed-off-by: Sasha Levin

scsi: target: core: Do not overwrite CDB byte 1

2019-11-10T10:23:18Z

[ Upstream commit 27e84243cb63601a10e366afe3e2d05bb03c1cb5 ] passthrough_parse_cdb() - used by TCMU and PSCSI - attepts to reset the LUN field of SCSI-2 CDBs (bits 5,6,7 of byte 1). The current code is wrong as for newer commands not having the LUN field it overwrites relevant command bits (e.g. for SECURITY PROTOCOL IN / OUT). We think this code was unnecessary from the beginning or at least it is no longer useful. So we remove it entirely. Link: https://lore.kernel.org/r/12498eab-76fd-eaad-1316-c2827badb76a@ts.fujitsu.com Signed-off-by: Bodo Stroesser Reviewed-by: Bart Van Assche Reviewed-by: Hannes Reinecke Signed-off-by: Martin K. Petersen Signed-off-by: Sasha Levin

target: Fix NULL dereference during LUN lookup + active I/O shutdown

2017-03-15T02:02:46Z

commit bd4e2d2907fa23a11d46217064ecf80470ddae10 upstream. When transport_clear_lun_ref() is shutting down a se_lun via configfs with new I/O in-flight, it's possible to trigger a NULL pointer dereference in transport_lookup_cmd_lun() due to the fact percpu_ref_get() doesn't do any __PERCPU_REF_DEAD checking before incrementing lun->lun_ref.count after lun->lun_ref has switched to atomic_t mode. This results in a NULL pointer dereference as LUN shutdown code in core_tpg_remove_lun() continues running after the existing ->release() -> core_tpg_lun_ref_release() callback completes, and clears the RCU protected se_lun->lun_se_dev pointer. During the OOPs, the state of lun->lun_ref in the process which triggered the NULL pointer dereference looks like the following on v4.1.y stable code: struct se_lun { lun_link_magic = 4294932337, lun_status = TRANSPORT_LUN_STATUS_FREE, ..... lun_se_dev = 0x0, lun_sep = 0x0, ..... lun_ref = { count = { counter = 1 }, percpu_count_ptr = 3, release = 0xffffffffa02fa1e0 , confirm_switch = 0x0, force_atomic = false, rcu = { next = 0xffff88154fa1a5d0, func = 0xffffffff8137c4c0 } } } To address this bug, use percpu_ref_tryget_live() to ensure once __PERCPU_REF_DEAD is visable on all CPUs and ->lun_ref has switched to atomic_t, all new I/Os will fail to obtain a new lun->lun_ref reference. Also use an explicit percpu_ref_kill_and_confirm() callback to block on ->lun_ref_comp to allow the first stage and associated RCU grace period to complete, and then block on ->lun_ref_shutdown waiting for the final percpu_ref_put() to drop the last reference via transport_lun_remove_cmd() before continuing with core_tpg_remove_lun() shutdown. Reported-by: Rob Millner Tested-by: Rob Millner Cc: Rob Millner Tested-by: Vaibhav Tandon Cc: Vaibhav Tandon Tested-by: Bryant G. Ly Signed-off-by: Nicholas Bellinger Signed-off-by: Greg Kroah-Hartman

target: Don't BUG_ON during NodeACL dynamic -> explicit conversion

2017-02-14T23:25:36Z

commit 391e2a6de9781e4906dd7e0b1cc097050bf43e11 upstream. After the v4.2+ RCU conversion to se_node_acl->lun_entry_hlist, a BUG_ON() was added in core_enable_device_list_for_node() to detect when the located orig->se_lun_acl contains an existing se_lun_acl pointer reference. However, this scenario can happen when a dynamically generated NodeACL is being converted to an explicit NodeACL, when the explicit NodeACL contains a different LUN mapping than the default provided by the WWN endpoint. So instead of triggering BUG_ON(), go ahead and fail instead following the original pre RCU conversion logic. Reported-by: Benjamin ESTRABAUD Cc: Benjamin ESTRABAUD Reviewed-by: Christoph Hellwig Signed-off-by: Nicholas Bellinger Signed-off-by: Greg Kroah-Hartman

target: Fix max_unmap_lba_count calc overflow

2016-07-20T07:58:36Z

max_discard_sectors only 32bits, and some non scsi backend devices will set this to the max 0xffffffff, so we can end up overflowing during the max_unmap_lba_count calculation. This fixes a regression caused by my patch: commit 8a9ebe717a133ba7bc90b06047f43cc6b8bcb8b3 Author: Mike Christie Date: Mon Jan 18 14:09:27 2016 -0600 target: Fix WRITE_SAME/DISCARD conversion to linux 512b sectors which can result in extra discards being sent to due the overflow causing max_unmap_lba_count to be smaller than what the backing device can actually support. Signed-off-by: Mike Christie Reviewed-by: Bart Van Assche Cc: stable@vger.kernel.org Signed-off-by: Nicholas Bellinger

target: Remove enum transport_lunflags_table

2016-03-11T05:48:55Z

se_dev_entry.lun_flags and se_lun.lun_access are only used for keeping track of read-write vs. read-only state. Since this is an either/or thing we can represent it as bool, and remove the unneeded enum transport_lunflags_table, which is left over from when there were more flags. Change code that uses this enum to just use true/false, and make it clear through variable and param names that true means read-only, false means read-write. Signed-off-by: Andy Grover Reviewed-by: Christoph Hellwig Signed-off-by: Nicholas Bellinger

target: Fix WRITE_SAME/DISCARD conversion to linux 512b sectors

2016-01-30T03:29:34Z

In a couple places we are not converting to/from the Linux block layer 512 bytes sectors. 1. The request queue values and what we do are a mismatch of things: max_discard_sectors - This is in linux block layer 512 byte sectors. We are just copying this to max_unmap_lba_count. discard_granularity - This is in bytes. We are converting it to Linux block layer 512 byte sectors. discard_alignment - This is in bytes. We are just copying this over. The problem is that the core LIO code exports these values in spc_emulate_evpd_b0 and we use them to test request arguments in sbc_execute_unmap, but we never convert to the block size we export to the initiator. If we are not using 512 byte sectors then we are exporting the wrong values or are checks are off. And, for the discard_alignment/bytes case we are just plain messed up. 2. blkdev_issue_discard's start and number of sector arguments are supposed to be in linux block layer 512 byte sectors. We are currently passing in the values we get from the initiator which might be based on some other sector size. There is a similar problem in iblock_execute_write_same where the bio functions want values in 512 byte sectors but we are passing in what we got from the initiator. Signed-off-by: Mike Christie Cc: stable@vger.kernel.org # 3.10+ Signed-off-by: Nicholas Bellinger

target/sbc: Add LBPRZ attribute + control CDB emulation

2016-01-06T21:39:55Z

This change sets the LBPRZ flag in EVPD page b2h and READ CAPACITY (16) based on a new unmap_zeroes_data device attribute. This flag is set automatically for iblock based on underlying block device queue's discard_zeroes_data flag. Signed-off-by: Jamie Pocas Signed-off-by: Nicholas Bellinger

target: Make TCM_WRITE_PROTECT failure honor D_SENSE bit

2015-09-25T06:17:27Z

This patch changes transport_lookup_cmd_lun() to obtain se_lun->lun_ref + se_cmd->se_device rcu_dereference during TCM_WRITE_PROTECT -> CHECK_CONDITION failure status. Do this to ensure the active control D_SENSE mode page bit is being honored. Reported-by: Sagi Grimberg Reviewed-by: Sagi Grimberg Signed-off-by: Nicholas Bellinger

target: Fix PR registration + APTPL RCU conversion regression

2015-09-25T06:17:07Z

This patch fixes a v4.2+ regression introduced by commit 79dc9c9e86 where lookup of t10_pr_registration->pr_reg_deve and associated ->pr_kref get was missing from __core_scsi3_do_alloc_registration(), which is responsible for setting DEF_PR_REG_ACTIVE. This would result in REGISTER operations completing successfully, but subsequent core_scsi3_pr_seq_non_holder() checking would fail with !DEF_PR_REG_ACTIVE -> RESERVATION CONFLICT status. Update __core_scsi3_add_registration() to drop ->pr_kref reference after registration and any optional ALL_TG_PT=1 processing has completed. Update core_scsi3_decode_spec_i_port() to release the new parent local_pr_reg->pr_kref as well. Also, update __core_scsi3_check_aptpl_registration() to perform the same target_nacl_find_deve() lookup + ->pr_kref get, now that __core_scsi3_add_registration() expects to drop the reference. Finally, since there are cases when se_dev_entry->se_lun_acl can still be dereferenced in core_scsi3_lunacl_undepend_item() while holding ->pr_kref, go ahead and move explicit rcu_assign_pointer() NULL assignments within core_disable_device_list_for_node() until after orig->pr_comp finishes. Reported-by: Scott L. Lykens Tested-by: Scott L. Lykens Cc: Christoph Hellwig Cc: Hannes Reinecke Cc: Lee Duncan Cc: # v4.2+ Signed-off-by: Nicholas Bellinger