linux_dsm_epyc7002/drivers/target/iscsi/iscsi_target_erl2.c
Nicholas Bellinger aafc9d158b iscsi-target: Fix iscsit_free_cmd() se_cmd->cmd_kref shutdown handling
With the introduction of target_get_sess_cmd() referencing counting for
ISCSI_OP_SCSI_CMD processing with iser-target, iscsit_free_cmd() usage
in traditional iscsi-target driver code now needs to be aware of the
active I/O shutdown case when a remaining se_cmd->cmd_kref reference may
exist after transport_generic_free_cmd() completes, requiring a final
target_put_sess_cmd() to release iscsi_cmd descriptor memory.

This patch changes iscsit_free_cmd() to invoke __iscsit_free_cmd() before
transport_generic_free_cmd() -> target_put_sess_cmd(), and also avoids
aquiring the per-connection queue locks for typical fast-path calls
during normal ISTATE_REMOVE operation.

Also update iscsit_free_cmd() usage throughout iscsi-target to
use the new 'bool shutdown' parameter.

This patch fixes a regression bug introduced during v3.10-rc1 in
commit 3e1c81a95, that was causing the following WARNING to appear:

[  257.235153] ------------[ cut here]------------
[  257.240314] WARNING: at kernel/softirq.c:160 local_bh_enable_ip+0x3c/0x86()
[  257.248089] Modules linked in: vhost_scsi ib_srpt ib_cm ib_sa ib_mad ib_core tcm_qla2xxx tcm_loop
	tcm_fc libfc iscsi_target_mod target_core_pscsi target_core_file
	target_core_iblock target_core_mod configfs ipv6 iscsi_tcp libiscsi_tcp
	libiscsi scsi_transport_iscsi loop acpi_cpufreq freq_table mperf
	kvm_intel kvm crc32c_intel button ehci_pci pcspkr joydev i2c_i801
	microcode ext3 jbd raid10 raid456 async_pq async_xor xor async_memcpy
	async_raid6_recov raid6_pq async_tx raid1 raid0 linear igb hwmon
	i2c_algo_bit i2c_core ptp ata_piix libata qla2xxx uhci_hcd ehci_hcd
	mlx4_core scsi_transport_fc scsi_tgt pps_core
[  257.308748] CPU: 1 PID: 3295 Comm: iscsi_ttx Not tainted 3.10.0-rc2+ #103
[  257.316329] Hardware name: Intel Corporation S5520HC/S5520HC, BIOS S5500.86B.01.00.0057.031020111721 03/10/2011
[  257.327597]  ffffffff814c24b7 ffff880458331b58 ffffffff8138eef2 ffff880458331b98
[  257.335892]  ffffffff8102c052 ffff880400000008 0000000000000000 ffff88085bdf0000
[  257.344191]  ffff88085bdf00d8 ffff88085bdf00e0 ffff88085bdf00f8 ffff880458331ba8
[  257.352488] Call Trace:
[  257.355223]  [<ffffffff8138eef2>] dump_stack+0x19/0x1f
[  257.360963]  [<ffffffff8102c052>] warn_slowpath_common+0x62/0x7b
[  257.367669]  [<ffffffff8102c080>] warn_slowpath_null+0x15/0x17
[  257.374181]  [<ffffffff81032345>] local_bh_enable_ip+0x3c/0x86
[  257.380697]  [<ffffffff813917fd>] _raw_spin_unlock_bh+0x10/0x12
[  257.387311]  [<ffffffffa029069c>] iscsit_free_r2ts_from_list+0x5e/0x67 [iscsi_target_mod]
[  257.396438]  [<ffffffffa02906c5>] iscsit_release_cmd+0x20/0x223 [iscsi_target_mod]
[  257.404893]  [<ffffffffa02977a4>] lio_release_cmd+0x3a/0x3e [iscsi_target_mod]
[  257.412964]  [<ffffffffa01d59a1>] target_release_cmd_kref+0x7a/0x7c [target_core_mod]
[  257.421712]  [<ffffffffa01d69bc>] target_put_sess_cmd+0x5f/0x7f [target_core_mod]
[  257.430071]  [<ffffffffa01d6d6d>] transport_release_cmd+0x59/0x6f [target_core_mod]
[  257.438625]  [<ffffffffa01d6eb4>] transport_put_cmd+0x131/0x140 [target_core_mod]
[  257.446985]  [<ffffffffa01d6192>] ? transport_wait_for_tasks+0xfa/0x1d5 [target_core_mod]
[  257.456121]  [<ffffffffa01d6f11>] transport_generic_free_cmd+0x4e/0x52 [target_core_mod]
[  257.465159]  [<ffffffff81050537>] ? __migrate_task+0x110/0x110
[  257.471674]  [<ffffffffa02904ba>] iscsit_free_cmd+0x46/0x55 [iscsi_target_mod]
[  257.479741]  [<ffffffffa0291edb>] iscsit_immediate_queue+0x301/0x353 [iscsi_target_mod]
[  257.488683]  [<ffffffffa0292f7e>] iscsi_target_tx_thread+0x1c6/0x2a8 [iscsi_target_mod]
[  257.497623]  [<ffffffff81047486>] ? wake_up_bit+0x25/0x25
[  257.503652]  [<ffffffffa0292db8>] ? iscsit_ack_from_expstatsn+0xd5/0xd5 [iscsi_target_mod]
[  257.512882]  [<ffffffff81046f89>] kthread+0xb0/0xb8
[  257.518329]  [<ffffffff81046ed9>] ? kthread_freezable_should_stop+0x60/0x60
[  257.526105]  [<ffffffff81396fec>] ret_from_fork+0x7c/0xb0
[  257.532133]  [<ffffffff81046ed9>] ? kthread_freezable_should_stop+0x60/0x60
[  257.539906] ---[ end trace 5520397d0f2e0800 ]---

Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
2013-05-31 01:21:28 -07:00

439 lines
12 KiB
C

/*******************************************************************************
* This file contains error recovery level two functions used by
* the iSCSI Target driver.
*
* \u00a9 Copyright 2007-2011 RisingTide Systems LLC.
*
* Licensed to the Linux Foundation under the General Public License (GPL) version 2.
*
* Author: Nicholas A. Bellinger <nab@linux-iscsi.org>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
******************************************************************************/
#include <scsi/iscsi_proto.h>
#include <target/target_core_base.h>
#include <target/target_core_fabric.h>
#include "iscsi_target_core.h"
#include "iscsi_target_datain_values.h"
#include "iscsi_target_util.h"
#include "iscsi_target_erl0.h"
#include "iscsi_target_erl1.h"
#include "iscsi_target_erl2.h"
#include "iscsi_target.h"
/*
* FIXME: Does RData SNACK apply here as well?
*/
void iscsit_create_conn_recovery_datain_values(
struct iscsi_cmd *cmd,
__be32 exp_data_sn)
{
u32 data_sn = 0;
struct iscsi_conn *conn = cmd->conn;
cmd->next_burst_len = 0;
cmd->read_data_done = 0;
while (be32_to_cpu(exp_data_sn) > data_sn) {
if ((cmd->next_burst_len +
conn->conn_ops->MaxRecvDataSegmentLength) <
conn->sess->sess_ops->MaxBurstLength) {
cmd->read_data_done +=
conn->conn_ops->MaxRecvDataSegmentLength;
cmd->next_burst_len +=
conn->conn_ops->MaxRecvDataSegmentLength;
} else {
cmd->read_data_done +=
(conn->sess->sess_ops->MaxBurstLength -
cmd->next_burst_len);
cmd->next_burst_len = 0;
}
data_sn++;
}
}
void iscsit_create_conn_recovery_dataout_values(
struct iscsi_cmd *cmd)
{
u32 write_data_done = 0;
struct iscsi_conn *conn = cmd->conn;
cmd->data_sn = 0;
cmd->next_burst_len = 0;
while (cmd->write_data_done > write_data_done) {
if ((write_data_done + conn->sess->sess_ops->MaxBurstLength) <=
cmd->write_data_done)
write_data_done += conn->sess->sess_ops->MaxBurstLength;
else
break;
}
cmd->write_data_done = write_data_done;
}
static int iscsit_attach_active_connection_recovery_entry(
struct iscsi_session *sess,
struct iscsi_conn_recovery *cr)
{
spin_lock(&sess->cr_a_lock);
list_add_tail(&cr->cr_list, &sess->cr_active_list);
spin_unlock(&sess->cr_a_lock);
return 0;
}
static int iscsit_attach_inactive_connection_recovery_entry(
struct iscsi_session *sess,
struct iscsi_conn_recovery *cr)
{
spin_lock(&sess->cr_i_lock);
list_add_tail(&cr->cr_list, &sess->cr_inactive_list);
sess->conn_recovery_count++;
pr_debug("Incremented connection recovery count to %u for"
" SID: %u\n", sess->conn_recovery_count, sess->sid);
spin_unlock(&sess->cr_i_lock);
return 0;
}
struct iscsi_conn_recovery *iscsit_get_inactive_connection_recovery_entry(
struct iscsi_session *sess,
u16 cid)
{
struct iscsi_conn_recovery *cr;
spin_lock(&sess->cr_i_lock);
list_for_each_entry(cr, &sess->cr_inactive_list, cr_list) {
if (cr->cid == cid) {
spin_unlock(&sess->cr_i_lock);
return cr;
}
}
spin_unlock(&sess->cr_i_lock);
return NULL;
}
void iscsit_free_connection_recovery_entires(struct iscsi_session *sess)
{
struct iscsi_cmd *cmd, *cmd_tmp;
struct iscsi_conn_recovery *cr, *cr_tmp;
spin_lock(&sess->cr_a_lock);
list_for_each_entry_safe(cr, cr_tmp, &sess->cr_active_list, cr_list) {
list_del(&cr->cr_list);
spin_unlock(&sess->cr_a_lock);
spin_lock(&cr->conn_recovery_cmd_lock);
list_for_each_entry_safe(cmd, cmd_tmp,
&cr->conn_recovery_cmd_list, i_conn_node) {
list_del(&cmd->i_conn_node);
cmd->conn = NULL;
spin_unlock(&cr->conn_recovery_cmd_lock);
iscsit_free_cmd(cmd, true);
spin_lock(&cr->conn_recovery_cmd_lock);
}
spin_unlock(&cr->conn_recovery_cmd_lock);
spin_lock(&sess->cr_a_lock);
kfree(cr);
}
spin_unlock(&sess->cr_a_lock);
spin_lock(&sess->cr_i_lock);
list_for_each_entry_safe(cr, cr_tmp, &sess->cr_inactive_list, cr_list) {
list_del(&cr->cr_list);
spin_unlock(&sess->cr_i_lock);
spin_lock(&cr->conn_recovery_cmd_lock);
list_for_each_entry_safe(cmd, cmd_tmp,
&cr->conn_recovery_cmd_list, i_conn_node) {
list_del(&cmd->i_conn_node);
cmd->conn = NULL;
spin_unlock(&cr->conn_recovery_cmd_lock);
iscsit_free_cmd(cmd, true);
spin_lock(&cr->conn_recovery_cmd_lock);
}
spin_unlock(&cr->conn_recovery_cmd_lock);
spin_lock(&sess->cr_i_lock);
kfree(cr);
}
spin_unlock(&sess->cr_i_lock);
}
int iscsit_remove_active_connection_recovery_entry(
struct iscsi_conn_recovery *cr,
struct iscsi_session *sess)
{
spin_lock(&sess->cr_a_lock);
list_del(&cr->cr_list);
sess->conn_recovery_count--;
pr_debug("Decremented connection recovery count to %u for"
" SID: %u\n", sess->conn_recovery_count, sess->sid);
spin_unlock(&sess->cr_a_lock);
kfree(cr);
return 0;
}
static void iscsit_remove_inactive_connection_recovery_entry(
struct iscsi_conn_recovery *cr,
struct iscsi_session *sess)
{
spin_lock(&sess->cr_i_lock);
list_del(&cr->cr_list);
spin_unlock(&sess->cr_i_lock);
}
/*
* Called with cr->conn_recovery_cmd_lock help.
*/
int iscsit_remove_cmd_from_connection_recovery(
struct iscsi_cmd *cmd,
struct iscsi_session *sess)
{
struct iscsi_conn_recovery *cr;
if (!cmd->cr) {
pr_err("struct iscsi_conn_recovery pointer for ITT: 0x%08x"
" is NULL!\n", cmd->init_task_tag);
BUG();
}
cr = cmd->cr;
list_del(&cmd->i_conn_node);
return --cr->cmd_count;
}
void iscsit_discard_cr_cmds_by_expstatsn(
struct iscsi_conn_recovery *cr,
u32 exp_statsn)
{
u32 dropped_count = 0;
struct iscsi_cmd *cmd, *cmd_tmp;
struct iscsi_session *sess = cr->sess;
spin_lock(&cr->conn_recovery_cmd_lock);
list_for_each_entry_safe(cmd, cmd_tmp,
&cr->conn_recovery_cmd_list, i_conn_node) {
if (((cmd->deferred_i_state != ISTATE_SENT_STATUS) &&
(cmd->deferred_i_state != ISTATE_REMOVE)) ||
(cmd->stat_sn >= exp_statsn)) {
continue;
}
dropped_count++;
pr_debug("Dropping Acknowledged ITT: 0x%08x, StatSN:"
" 0x%08x, CID: %hu.\n", cmd->init_task_tag,
cmd->stat_sn, cr->cid);
iscsit_remove_cmd_from_connection_recovery(cmd, sess);
spin_unlock(&cr->conn_recovery_cmd_lock);
iscsit_free_cmd(cmd, true);
spin_lock(&cr->conn_recovery_cmd_lock);
}
spin_unlock(&cr->conn_recovery_cmd_lock);
pr_debug("Dropped %u total acknowledged commands on"
" CID: %hu less than old ExpStatSN: 0x%08x\n",
dropped_count, cr->cid, exp_statsn);
if (!cr->cmd_count) {
pr_debug("No commands to be reassigned for failed"
" connection CID: %hu on SID: %u\n",
cr->cid, sess->sid);
iscsit_remove_inactive_connection_recovery_entry(cr, sess);
iscsit_attach_active_connection_recovery_entry(sess, cr);
pr_debug("iSCSI connection recovery successful for CID:"
" %hu on SID: %u\n", cr->cid, sess->sid);
iscsit_remove_active_connection_recovery_entry(cr, sess);
} else {
iscsit_remove_inactive_connection_recovery_entry(cr, sess);
iscsit_attach_active_connection_recovery_entry(sess, cr);
}
}
int iscsit_discard_unacknowledged_ooo_cmdsns_for_conn(struct iscsi_conn *conn)
{
u32 dropped_count = 0;
struct iscsi_cmd *cmd, *cmd_tmp;
struct iscsi_ooo_cmdsn *ooo_cmdsn, *ooo_cmdsn_tmp;
struct iscsi_session *sess = conn->sess;
mutex_lock(&sess->cmdsn_mutex);
list_for_each_entry_safe(ooo_cmdsn, ooo_cmdsn_tmp,
&sess->sess_ooo_cmdsn_list, ooo_list) {
if (ooo_cmdsn->cid != conn->cid)
continue;
dropped_count++;
pr_debug("Dropping unacknowledged CmdSN:"
" 0x%08x during connection recovery on CID: %hu\n",
ooo_cmdsn->cmdsn, conn->cid);
iscsit_remove_ooo_cmdsn(sess, ooo_cmdsn);
}
mutex_unlock(&sess->cmdsn_mutex);
spin_lock_bh(&conn->cmd_lock);
list_for_each_entry_safe(cmd, cmd_tmp, &conn->conn_cmd_list, i_conn_node) {
if (!(cmd->cmd_flags & ICF_OOO_CMDSN))
continue;
list_del(&cmd->i_conn_node);
spin_unlock_bh(&conn->cmd_lock);
iscsit_free_cmd(cmd, true);
spin_lock_bh(&conn->cmd_lock);
}
spin_unlock_bh(&conn->cmd_lock);
pr_debug("Dropped %u total unacknowledged commands on CID:"
" %hu for ExpCmdSN: 0x%08x.\n", dropped_count, conn->cid,
sess->exp_cmd_sn);
return 0;
}
int iscsit_prepare_cmds_for_realligance(struct iscsi_conn *conn)
{
u32 cmd_count = 0;
struct iscsi_cmd *cmd, *cmd_tmp;
struct iscsi_conn_recovery *cr;
/*
* Allocate an struct iscsi_conn_recovery for this connection.
* Each struct iscsi_cmd contains an struct iscsi_conn_recovery pointer
* (struct iscsi_cmd->cr) so we need to allocate this before preparing the
* connection's command list for connection recovery.
*/
cr = kzalloc(sizeof(struct iscsi_conn_recovery), GFP_KERNEL);
if (!cr) {
pr_err("Unable to allocate memory for"
" struct iscsi_conn_recovery.\n");
return -1;
}
INIT_LIST_HEAD(&cr->cr_list);
INIT_LIST_HEAD(&cr->conn_recovery_cmd_list);
spin_lock_init(&cr->conn_recovery_cmd_lock);
/*
* Only perform connection recovery on ISCSI_OP_SCSI_CMD or
* ISCSI_OP_NOOP_OUT opcodes. For all other opcodes call
* list_del(&cmd->i_conn_node); to release the command to the
* session pool and remove it from the connection's list.
*
* Also stop the DataOUT timer, which will be restarted after
* sending the TMR response.
*/
spin_lock_bh(&conn->cmd_lock);
list_for_each_entry_safe(cmd, cmd_tmp, &conn->conn_cmd_list, i_conn_node) {
if ((cmd->iscsi_opcode != ISCSI_OP_SCSI_CMD) &&
(cmd->iscsi_opcode != ISCSI_OP_NOOP_OUT)) {
pr_debug("Not performing realligence on"
" Opcode: 0x%02x, ITT: 0x%08x, CmdSN: 0x%08x,"
" CID: %hu\n", cmd->iscsi_opcode,
cmd->init_task_tag, cmd->cmd_sn, conn->cid);
list_del(&cmd->i_conn_node);
spin_unlock_bh(&conn->cmd_lock);
iscsit_free_cmd(cmd, true);
spin_lock_bh(&conn->cmd_lock);
continue;
}
/*
* Special case where commands greater than or equal to
* the session's ExpCmdSN are attached to the connection
* list but not to the out of order CmdSN list. The one
* obvious case is when a command with immediate data
* attached must only check the CmdSN against ExpCmdSN
* after the data is received. The special case below
* is when the connection fails before data is received,
* but also may apply to other PDUs, so it has been
* made generic here.
*/
if (!(cmd->cmd_flags & ICF_OOO_CMDSN) && !cmd->immediate_cmd &&
iscsi_sna_gte(cmd->cmd_sn, conn->sess->exp_cmd_sn)) {
list_del(&cmd->i_conn_node);
spin_unlock_bh(&conn->cmd_lock);
iscsit_free_cmd(cmd, true);
spin_lock_bh(&conn->cmd_lock);
continue;
}
cmd_count++;
pr_debug("Preparing Opcode: 0x%02x, ITT: 0x%08x,"
" CmdSN: 0x%08x, StatSN: 0x%08x, CID: %hu for"
" realligence.\n", cmd->iscsi_opcode,
cmd->init_task_tag, cmd->cmd_sn, cmd->stat_sn,
conn->cid);
cmd->deferred_i_state = cmd->i_state;
cmd->i_state = ISTATE_IN_CONNECTION_RECOVERY;
if (cmd->data_direction == DMA_TO_DEVICE)
iscsit_stop_dataout_timer(cmd);
cmd->sess = conn->sess;
list_del(&cmd->i_conn_node);
spin_unlock_bh(&conn->cmd_lock);
iscsit_free_all_datain_reqs(cmd);
transport_wait_for_tasks(&cmd->se_cmd);
/*
* Add the struct iscsi_cmd to the connection recovery cmd list
*/
spin_lock(&cr->conn_recovery_cmd_lock);
list_add_tail(&cmd->i_conn_node, &cr->conn_recovery_cmd_list);
spin_unlock(&cr->conn_recovery_cmd_lock);
spin_lock_bh(&conn->cmd_lock);
cmd->cr = cr;
cmd->conn = NULL;
}
spin_unlock_bh(&conn->cmd_lock);
/*
* Fill in the various values in the preallocated struct iscsi_conn_recovery.
*/
cr->cid = conn->cid;
cr->cmd_count = cmd_count;
cr->maxrecvdatasegmentlength = conn->conn_ops->MaxRecvDataSegmentLength;
cr->maxxmitdatasegmentlength = conn->conn_ops->MaxXmitDataSegmentLength;
cr->sess = conn->sess;
iscsit_attach_inactive_connection_recovery_entry(conn->sess, cr);
return 0;
}
int iscsit_connection_recovery_transport_reset(struct iscsi_conn *conn)
{
atomic_set(&conn->connection_recovery, 1);
if (iscsit_close_connection(conn) < 0)
return -1;
return 0;
}