d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

MFV v242732:

Browse Source 
Merge the ZFS I/O deadman thread from vendor (illumos).
This feature panics the system on hanging ZFS I/O, helps debugging
and resumes failed service.

The panic behavior can be controlled with the loader-only tunables:
vfs.zfs.deadman_enabled (enable or disable panic on stalled ZFS I/O)
vfs.zfs.deadman_synctime (expiration time for stalled ZFS I/O)

By default, ZFS I/O deadman is enabled by default on amd64 and i386
excluding virtual guest machines.

Illumos ZFS issues:
  3246 ZFS I/O deadman thread

References:
  https://www.illumos.org/issues/3246

MFC after:	2 weeks
This commit is contained in:
Martin Matuska 2013-02-25 12:33:31 +00:00
commit e70664bafc
18 changed files with 380 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
cddl/contrib/opensolaris/cmd/zinject/translate.c
View File

@ -20,6 +20,7 @@
*/
/*
* Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012 by Delphix. All rights reserved.
*/
#include <libzfs.h>
@ -455,6 +456,20 @@ translate_device(const char *pool, const char *device, err_type_t label_type,
&record->zi_guid) == 0);
}
/*
* Device faults can take on three different forms:
* 1). delayed or hanging I/O
* 2). zfs label faults
* 3). generic disk faults
*/
if (record->zi_timer != 0) {
record->zi_cmd = ZINJECT_DELAY_IO;
} else if (label_type != TYPE_INVAL) {
record->zi_cmd = ZINJECT_LABEL_FAULT;
} else {
record->zi_cmd = ZINJECT_DEVICE_FAULT;
}
switch (label_type) {
case TYPE_LABEL_UBERBLOCK:
record->zi_start = offsetof(vdev_label_t, vl_uberblock[0]);

32
cddl/contrib/opensolaris/cmd/zinject/zinject.c
View File

@ -20,6 +20,7 @@
*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012 by Delphix. All rights reserved.
*/
/*
@ -603,7 +604,7 @@ main(int argc, char **argv)
}
while ((c = getopt(argc, argv,
":aA:b:d:f:Fg:qhIc:t:T:l:mr:s:e:uL:p:")) != -1) {
":aA:b:d:D:f:Fg:qhIc:t:T:l:mr:s:e:uL:p:")) != -1) {
switch (c) {
case 'a':
flags |= ZINJECT_FLUSH_ARC;
@ -629,6 +630,15 @@ main(int argc, char **argv)
case 'd':
device = optarg;
break;
case 'D':
record.zi_timer = strtoull(optarg, &end, 10);
if (errno != 0 || *end != '\0') {
(void) fprintf(stderr, "invalid i/o delay "
"value: '%s'\n", optarg);
usage();
return (1);
}
break;
case 'e':
if (strcasecmp(optarg, "io") == 0) {
error = EIO;
@ -693,6 +703,7 @@ main(int argc, char **argv)
case 'p':
(void) strlcpy(record.zi_func, optarg,
sizeof (record.zi_func));
record.zi_cmd = ZINJECT_PANIC;
break;
case 'q':
quiet = 1;
@ -766,13 +777,15 @@ main(int argc, char **argv)
argc -= optind;
argv += optind;
if (record.zi_duration != 0)
record.zi_cmd = ZINJECT_IGNORED_WRITES;
if (cancel != NULL) {
/*
* '-c' is invalid with any other options.
*/
if (raw != NULL || range != NULL || type != TYPE_INVAL ||
level != 0 || record.zi_func[0] != '\0' ||
record.zi_duration != 0) {
level != 0 || record.zi_cmd != ZINJECT_UNINITIALIZED) {
(void) fprintf(stderr, "cancel (-c) incompatible with "
"any other options\n");
usage();
@ -804,8 +817,7 @@ main(int argc, char **argv)
* for doing injection, so handle it separately here.
*/
if (raw != NULL || range != NULL || type != TYPE_INVAL ||
level != 0 || record.zi_func[0] != '\0' ||
record.zi_duration != 0) {
level != 0 || record.zi_cmd != ZINJECT_UNINITIALIZED) {
(void) fprintf(stderr, "device (-d) incompatible with "
"data error injection\n");
usage();
@ -839,7 +851,7 @@ main(int argc, char **argv)
} else if (raw != NULL) {
if (range != NULL || type != TYPE_INVAL || level != 0 ||
record.zi_func[0] != '\0' || record.zi_duration != 0) {
record.zi_cmd != ZINJECT_UNINITIALIZED) {
(void) fprintf(stderr, "raw (-b) format with "
"any other options\n");
usage();
@ -862,13 +874,14 @@ main(int argc, char **argv)
return (1);
}
record.zi_cmd = ZINJECT_DATA_FAULT;
if (translate_raw(raw, &record) != 0)
return (1);
if (!error)
error = EIO;
} else if (record.zi_func[0] != '\0') {
} else if (record.zi_cmd == ZINJECT_PANIC) {
if (raw != NULL || range != NULL || type != TYPE_INVAL ||
level != 0 || device != NULL || record.zi_duration != 0) {
level != 0 || device != NULL) {
(void) fprintf(stderr, "panic (-p) incompatible with "
"other options\n");
usage();
@ -886,7 +899,7 @@ main(int argc, char **argv)
if (argv[1] != NULL)
record.zi_type = atoi(argv[1]);
dataset[0] = '\0';
} else if (record.zi_duration != 0) {
} else if (record.zi_cmd == ZINJECT_IGNORED_WRITES) {
if (nowrites == 0) {
(void) fprintf(stderr, "-s or -g meaningless "
"without -I (ignore writes)\n");
@ -940,6 +953,7 @@ main(int argc, char **argv)
return (1);
}
record.zi_cmd = ZINJECT_DATA_FAULT;
if (translate_record(type, argv[0], range, level, &record, pool,
dataset) != 0)
return (1);

29
cddl/contrib/opensolaris/lib/libzpool/common/kernel.c
View File

@ -45,6 +45,9 @@ int aok;
uint64_t physmem;
vnode_t *rootdir = (vnode_t *)0xabcd1234;
char hw_serial[HW_HOSTID_LEN];
#ifdef illumos
kmutex_t cpu_lock;
#endif
struct utsname utsname = {
"userland", "libzpool", "1", "1", "na"
@ -842,6 +845,28 @@ ddi_strtoull(const char *str, char **nptr, int base, u_longlong_t *result)
return (0);
}
#ifdef illumos
/* ARGSUSED */
cyclic_id_t
cyclic_add(cyc_handler_t *hdlr, cyc_time_t *when)
{
return (1);
}
/* ARGSUSED */
void
cyclic_remove(cyclic_id_t id)
{
}
/* ARGSUSED */
int
cyclic_reprogram(cyclic_id_t id, hrtime_t expiration)
{
return (1);
}
#endif
/*
* =========================================================================
* kernel emulation setup & teardown
@ -875,6 +900,10 @@ kernel_init(int mode)
system_taskq_init();
#ifdef illumos
mutex_init(&cpu_lock, NULL, MUTEX_DEFAULT, NULL);
#endif
spa_init(mode);
}

33
cddl/contrib/opensolaris/lib/libzpool/common/sys/zfs_context.h
View File

@ -457,6 +457,9 @@ extern vnode_t *rootdir;
extern void delay(clock_t ticks);
#define SEC_TO_TICK(sec) ((sec) * hz)
#define NSEC_TO_TICK(usec) ((usec) / (NANOSEC / hz))
#define gethrestime_sec() time(NULL)
#define gethrestime(t) \
do {\
@ -624,6 +627,36 @@ typedef uint32_t idmap_rid_t;
#define ERESTART (-1)
#endif
#ifdef illumos
/*
* Cyclic information
*/
extern kmutex_t cpu_lock;
typedef uintptr_t cyclic_id_t;
typedef uint16_t cyc_level_t;
typedef void (*cyc_func_t)(void *);
#define CY_LOW_LEVEL 0
#define CY_INFINITY INT64_MAX
#define CYCLIC_NONE ((cyclic_id_t)0)
typedef struct cyc_time {
hrtime_t cyt_when;
hrtime_t cyt_interval;
} cyc_time_t;
typedef struct cyc_handler {
cyc_func_t cyh_func;
void *cyh_arg;
cyc_level_t cyh_level;
} cyc_handler_t;
extern cyclic_id_t cyclic_add(cyc_handler_t *, cyc_time_t *);
extern void cyclic_remove(cyclic_id_t);
extern int cyclic_reprogram(cyclic_id_t, hrtime_t);
#endif /* illumos */
#ifdef __cplusplus
}
#endif

3
sys/cddl/compat/opensolaris/sys/time.h
View File

@ -46,6 +46,9 @@ typedef longlong_t hrtime_t;
((ts)->tv_sec < INT64_MIN || (ts)->tv_sec > INT64_MAX)
#endif
#define SEC_TO_TICK(sec) ((sec) * hz)
#define NSEC_TO_TICK(usec) ((usec) / (NANOSEC / hz))
#ifdef _KERNEL
static __inline hrtime_t
gethrtime(void) {

24
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c
View File

@ -22,6 +22,7 @@
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012 by Delphix. All rights reserved.
* Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
*/
/*
@ -141,6 +142,10 @@ uint_t zio_taskq_basedc = 80; /* base duty cycle */
boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
extern int zfs_sync_pass_deferred_free;
#ifndef illumos
extern void spa_deadman(void *arg);
#endif
/*
* This (illegal) pool name is used when temporarily importing a spa_t in order
* to get the vdev stats associated with the imported devices.
@ -6258,6 +6263,17 @@ spa_sync(spa_t *spa, uint64_t txg)
tx = dmu_tx_create_assigned(dp, txg);
spa->spa_sync_starttime = gethrtime();
#ifdef illumos
VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
spa->spa_sync_starttime + spa->spa_deadman_synctime));
#else /* FreeBSD */
#ifdef _KERNEL
callout_reset(&spa->spa_deadman_cycid,
hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
#endif
#endif
/*
* If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
* set spa_deflate if we have no raid-z vdevs.
@ -6386,6 +6402,14 @@ spa_sync(spa_t *spa, uint64_t txg)
}
dmu_tx_commit(tx);
#ifdef illumos
VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
#else /* FreeBSD */
#ifdef _KERNEL
callout_drain(&spa->spa_deadman_cycid);
#endif
#endif
/*
* Clear the dirty config list.
*/

122
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c
View File

@ -22,10 +22,12 @@
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012 by Delphix. All rights reserved.
* Copyright 2011 Nexenta Systems, Inc. All rights reserved.
* Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
*/
#include <sys/zfs_context.h>
#include <sys/spa_impl.h>
#include <sys/spa_boot.h>
#include <sys/zio.h>
#include <sys/zio_checksum.h>
#include <sys/zio_compress.h>
@ -253,6 +255,52 @@ TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0,
"Try to recover from otherwise-fatal errors.");
extern int zfs_txg_synctime_ms;
/*
* Expiration time in units of zfs_txg_synctime_ms. This value has two
* meanings. First it is used to determine when the spa_deadman logic
* should fire. By default the spa_deadman will fire if spa_sync has
* not completed in 1000 * zfs_txg_synctime_ms (i.e. 1000 seconds).
* Secondly, the value determines if an I/O is considered "hung".
* Any I/O that has not completed in zfs_deadman_synctime is considered
* "hung" resulting in a system panic.
* 1000 zfs_txg_synctime_ms (i.e. 1000 seconds).
*/
uint64_t zfs_deadman_synctime = 1000ULL;
TUNABLE_QUAD("vfs.zfs.deadman_synctime", &zfs_deadman_synctime);
SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime, CTLFLAG_RDTUN,
&zfs_deadman_synctime, 0,
"Stalled ZFS I/O expiration time in units of vfs.zfs.txg_synctime_ms");
/*
* Default value of -1 for zfs_deadman_enabled is resolved in
* zfs_deadman_init()
*/
int zfs_deadman_enabled = -1;
TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled);
SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
&zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
#ifndef illumos
#ifdef _KERNEL
static void
zfs_deadman_init()
{
/*
* If we are not i386 or amd64 or in a virtual machine,
* disable ZFS deadman thread by default
*/
if (zfs_deadman_enabled == -1) {
#if defined(__amd64__) || defined(__i386__)
zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
#else
zfs_deadman_enabled = 0;
#endif
}
}
#endif /* _KERNEL */
#endif /* !illumos */
/*
* ==========================================================================
@ -421,6 +469,23 @@ spa_lookup(const char *name)
return (spa);
}
/*
* Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
* If the zfs_deadman_enabled flag is set then it inspects all vdev queues
* looking for potentially hung I/Os.
*/
void
spa_deadman(void *arg)
{
spa_t *spa = arg;
zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
(gethrtime() - spa->spa_sync_starttime) / NANOSEC,
++spa->spa_deadman_calls);
if (zfs_deadman_enabled)
vdev_deadman(spa->spa_root_vdev);
}
/*
* Create an uninitialized spa_t with the given name. Requires
* spa_namespace_lock. The caller must ensure that the spa_t doesn't already
@ -431,6 +496,10 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
{
spa_t *spa;
spa_config_dirent_t *dp;
#ifdef illumos
cyc_handler_t hdlr;
cyc_time_t when;
#endif
ASSERT(MUTEX_HELD(&spa_namespace_lock));
@ -462,6 +531,32 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
spa->spa_proc = &p0;
spa->spa_proc_state = SPA_PROC_NONE;
#ifdef illumos
hdlr.cyh_func = spa_deadman;
hdlr.cyh_arg = spa;
hdlr.cyh_level = CY_LOW_LEVEL;
#endif
spa->spa_deadman_synctime = zfs_deadman_synctime *
zfs_txg_synctime_ms * MICROSEC;
#ifdef illumos
/*
* This determines how often we need to check for hung I/Os after
* the cyclic has already fired. Since checking for hung I/Os is
* an expensive operation we don't want to check too frequently.
* Instead wait for 5 synctimes before checking again.
*/
when.cyt_interval = 5ULL * zfs_txg_synctime_ms * MICROSEC;
when.cyt_when = CY_INFINITY;
mutex_enter(&cpu_lock);
spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
mutex_exit(&cpu_lock);
#else /* !illumos */
#ifdef _KERNEL
callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE);
#endif
#endif
refcount_create(&spa->spa_refcount);
spa_config_lock_init(spa);
@ -544,6 +639,18 @@ spa_remove(spa_t *spa)
nvlist_free(spa->spa_load_info);
spa_config_set(spa, NULL);
#ifdef illumos
mutex_enter(&cpu_lock);
if (spa->spa_deadman_cycid != CYCLIC_NONE)
cyclic_remove(spa->spa_deadman_cycid);
mutex_exit(&cpu_lock);
spa->spa_deadman_cycid = CYCLIC_NONE;
#else /* !illumos */
#ifdef _KERNEL
callout_drain(&spa->spa_deadman_cycid);
#endif
#endif
refcount_destroy(&spa->spa_refcount);
spa_config_lock_destroy(spa);
@ -1510,6 +1617,12 @@ spa_prev_software_version(spa_t *spa)
return (spa->spa_prev_software_version);
}
uint64_t
spa_deadman_synctime(spa_t *spa)
{
return (spa->spa_deadman_synctime);
}
uint64_t
dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
{
@ -1605,7 +1718,9 @@ spa_init(int mode)
spa_mode_global = mode;
#ifdef illumos
#ifndef _KERNEL
#ifdef _KERNEL
spa_arch_init();
#else
if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
arc_procfd = open("/proc/self/ctl", O_WRONLY);
if (arc_procfd == -1) {
@ -1629,6 +1744,11 @@ spa_init(int mode)
zpool_feature_init();
spa_config_load();
l2arc_start();
#ifndef illumos
#ifdef _KERNEL
zfs_deadman_init();
#endif
#endif /* !illumos */
}
void

1
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa.h
View File

@ -599,6 +599,7 @@ extern boolean_t spa_suspended(spa_t *spa);
extern uint64_t spa_bootfs(spa_t *spa);
extern uint64_t spa_delegation(spa_t *spa);
extern objset_t *spa_meta_objset(spa_t *spa);
extern uint64_t spa_deadman_synctime(spa_t *spa);
/* Miscellaneous support routines */
extern void spa_activate_mos_feature(spa_t *spa, const char *feature);

6
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa_boot.h
View File

@ -23,6 +23,10 @@
* Use is subject to license terms.
*/
/*
* Copyright (c) 2012 by Delphix. All rights reserved.
*/
#ifndef _SYS_SPA_BOOT_H
#define _SYS_SPA_BOOT_H
@ -35,6 +39,8 @@ extern "C" {
extern char *spa_get_bootprop(char *prop);
extern void spa_free_bootprop(char *prop);
extern void spa_arch_init(void);
#ifdef __cplusplus
}
#endif

11
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa_impl.h
View File

@ -22,6 +22,7 @@
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012 by Delphix. All rights reserved.
* Copyright 2011 Nexenta Systems, Inc. All rights reserved.
* Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
*/
#ifndef _SYS_SPA_IMPL_H
@ -230,6 +231,16 @@ struct spa {
uint64_t spa_feat_for_write_obj; /* required to write to pool */
uint64_t spa_feat_for_read_obj; /* required to read from pool */
uint64_t spa_feat_desc_obj; /* Feature descriptions */
#ifdef illumos
cyclic_id_t spa_deadman_cycid; /* cyclic id */
#else /* FreeBSD */
#ifdef _KERNEL
struct callout spa_deadman_cycid; /* callout id */
#endif
#endif /* illumos */
uint64_t spa_deadman_calls; /* number of deadman calls */
uint64_t spa_sync_starttime; /* starting time fo spa_sync */
uint64_t spa_deadman_synctime; /* deadman expiration timer */
/*
* spa_refcnt & spa_config_lock must be the last elements
* because refcount_t changes size based on compilation options.

1
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/vdev.h
View File

@ -80,6 +80,7 @@ extern void vdev_metaslab_fini(vdev_t *vd);
extern void vdev_metaslab_set_size(vdev_t *);
extern void vdev_expand(vdev_t *vd, uint64_t txg);
extern void vdev_split(vdev_t *vd);
extern void vdev_deadman(vdev_t *vd);
extern void vdev_get_stats(vdev_t *vd, vdev_stat_t *vs);

2
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/vdev_impl.h
View File

@ -104,6 +104,8 @@ struct vdev_queue {
avl_tree_t vq_read_tree;
avl_tree_t vq_write_tree;
avl_tree_t vq_pending_tree;
uint64_t vq_io_complete_ts;
uint64_t vq_io_delta_ts;
kmutex_t vq_lock;
};

9
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zfs_context.h
View File

@ -22,6 +22,10 @@
* Copyright 2009 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
/*
* Copyright 2011 Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2012 by Delphix. All rights reserved.
*/
#ifndef _SYS_ZFS_CONTEXT_H
#define _SYS_ZFS_CONTEXT_H
@ -88,6 +92,11 @@ extern "C" {
#include <sys/u8_textprep.h>
#include <sys/fm/util.h>
#include <sys/sunddi.h>
#ifdef illumos
#include <sys/cyclic.h>
#else /* FreeBSD */
#include <sys/callout.h>
#endif
#include <machine/stdarg.h>

12
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zfs_ioctl.h
View File

@ -246,12 +246,24 @@ typedef struct zinject_record {
uint32_t zi_iotype;
int32_t zi_duration;
uint64_t zi_timer;
uint32_t zi_cmd;
uint32_t zi_pad;
} zinject_record_t;
#define ZINJECT_NULL 0x1
#define ZINJECT_FLUSH_ARC 0x2
#define ZINJECT_UNLOAD_SPA 0x4
typedef enum zinject_type {
ZINJECT_UNINITIALIZED,
ZINJECT_DATA_FAULT,
ZINJECT_DEVICE_FAULT,
ZINJECT_LABEL_FAULT,
ZINJECT_IGNORED_WRITES,
ZINJECT_PANIC,
ZINJECT_DELAY_IO,
} zinject_type_t;
typedef struct zfs_share {
uint64_t z_exportdata;
uint64_t z_sharedata;

2
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h
View File

@ -443,6 +443,7 @@ struct zio {
uint64_t io_offset;
uint64_t io_deadline;
uint64_t io_timestamp;
avl_node_t io_offset_node;
avl_node_t io_deadline_node;
avl_tree_t *io_vdev_tree;
@ -596,6 +597,7 @@ extern int zio_handle_fault_injection(zio_t *zio, int error);
extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
extern int zio_handle_label_injection(zio_t *zio, int error);
extern void zio_handle_ignored_writes(zio_t *zio);
extern uint64_t zio_handle_io_delay(zio_t *zio);
/*
* Checksum ereport functions

38
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c
View File

@ -3173,3 +3173,41 @@ vdev_split(vdev_t *vd)
}
vdev_propagate_state(cvd);
}
void
vdev_deadman(vdev_t *vd)
{
for (int c = 0; c < vd->vdev_children; c++) {
vdev_t *cvd = vd->vdev_child[c];
vdev_deadman(cvd);
}
if (vd->vdev_ops->vdev_op_leaf) {
vdev_queue_t *vq = &vd->vdev_queue;
mutex_enter(&vq->vq_lock);
if (avl_numnodes(&vq->vq_pending_tree) > 0) {
spa_t *spa = vd->vdev_spa;
zio_t *fio;
uint64_t delta;
/*
* Look at the head of all the pending queues,
* if any I/O has been outstanding for longer than
* the spa_deadman_synctime we panic the system.
*/
fio = avl_first(&vq->vq_pending_tree);
delta = ddi_get_lbolt64() - fio->io_timestamp;
if (delta > NSEC_TO_TICK(spa_deadman_synctime(spa))) {
zfs_dbgmsg("SLOW IO: zio timestamp %llu, "
"delta %llu, last io %llu",
fio->io_timestamp, delta,
vq->vq_io_complete_ts);
fm_panic("I/O to pool '%s' appears to be "
"hung.", spa_name(spa));
}
}
mutex_exit(&vq->vq_lock);
}
}

14
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c
View File

@ -23,6 +23,10 @@
* Use is subject to license terms.
*/
/*
* Copyright (c) 2012 by Delphix. All rights reserved.
*/
#include <sys/zfs_context.h>
#include <sys/vdev_impl.h>
#include <sys/zio.h>
@ -315,6 +319,7 @@ again:
zio_buf_alloc(size), size, fio->io_type, ZIO_PRIORITY_AGG,
flags | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_QUEUE,
vdev_queue_agg_io_done, NULL);
aio->io_timestamp = fio->io_timestamp;
nio = fio;
do {
@ -386,7 +391,8 @@ vdev_queue_io(zio_t *zio)
mutex_enter(&vq->vq_lock);
zio->io_deadline = (ddi_get_lbolt64() >> zfs_vdev_time_shift) +
zio->io_timestamp = ddi_get_lbolt64();
zio->io_deadline = (zio->io_timestamp >> zfs_vdev_time_shift) +
zio->io_priority;
vdev_queue_io_add(vq, zio);
@ -411,10 +417,16 @@ vdev_queue_io_done(zio_t *zio)
{
vdev_queue_t *vq = &zio->io_vd->vdev_queue;
if (zio_injection_enabled)
delay(SEC_TO_TICK(zio_handle_io_delay(zio)));
mutex_enter(&vq->vq_lock);
avl_remove(&vq->vq_pending_tree, zio);
vq->vq_io_complete_ts = ddi_get_lbolt64();
vq->vq_io_delta_ts = vq->vq_io_complete_ts - zio->io_timestamp;
for (int i = 0; i < zfs_vdev_ramp_rate; i++) {
zio_t *nio = vdev_queue_io_to_issue(vq, zfs_vdev_max_pending);
if (nio == NULL)

65
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio_inject.c
View File

@ -20,6 +20,7 @@
*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012 by Delphix. All rights reserved.
*/
/*
@ -147,14 +148,8 @@ zio_handle_fault_injection(zio_t *zio, int error)
for (handler = list_head(&inject_handlers); handler != NULL;
handler = list_next(&inject_handlers, handler)) {
/* Ignore errors not destined for this pool */
if (zio->io_spa != handler->zi_spa)
continue;
/* Ignore device errors and panic injection */
if (handler->zi_record.zi_guid != 0 ||
handler->zi_record.zi_func[0] != '\0' ||
handler->zi_record.zi_duration != 0)
if (zio->io_spa != handler->zi_spa ||
handler->zi_record.zi_cmd != ZINJECT_DATA_FAULT)
continue;
/* If this handler matches, return EIO */
@ -197,10 +192,7 @@ zio_handle_label_injection(zio_t *zio, int error)
uint64_t start = handler->zi_record.zi_start;
uint64_t end = handler->zi_record.zi_end;
/* Ignore device only faults or panic injection */
if (handler->zi_record.zi_start == 0 ||
handler->zi_record.zi_func[0] != '\0' ||
handler->zi_record.zi_duration != 0)
if (handler->zi_record.zi_cmd != ZINJECT_LABEL_FAULT)
continue;
/*
@ -246,13 +238,7 @@ zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error)
for (handler = list_head(&inject_handlers); handler != NULL;
handler = list_next(&inject_handlers, handler)) {
/*
* Ignore label specific faults, panic injection
* or fake writes
*/
if (handler->zi_record.zi_start != 0 ||
handler->zi_record.zi_func[0] != '\0' ||
handler->zi_record.zi_duration != 0)
if (handler->zi_record.zi_cmd != ZINJECT_DEVICE_FAULT)
continue;
if (vd->vdev_guid == handler->zi_record.zi_guid) {
@ -316,10 +302,8 @@ zio_handle_ignored_writes(zio_t *zio)
handler = list_next(&inject_handlers, handler)) {
/* Ignore errors not destined for this pool */
if (zio->io_spa != handler->zi_spa)
continue;
if (handler->zi_record.zi_duration == 0)
if (zio->io_spa != handler->zi_spa ||
handler->zi_record.zi_cmd != ZINJECT_IGNORED_WRITES)
continue;
/*
@ -355,11 +339,8 @@ spa_handle_ignored_writes(spa_t *spa)
for (handler = list_head(&inject_handlers); handler != NULL;
handler = list_next(&inject_handlers, handler)) {
/* Ignore errors not destined for this pool */
if (spa != handler->zi_spa)
continue;
if (handler->zi_record.zi_duration == 0)
if (spa != handler->zi_spa ||
handler->zi_record.zi_cmd != ZINJECT_IGNORED_WRITES)
continue;
if (handler->zi_record.zi_duration > 0) {
@ -379,6 +360,34 @@ spa_handle_ignored_writes(spa_t *spa)
rw_exit(&inject_lock);
}
uint64_t
zio_handle_io_delay(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
inject_handler_t *handler;
uint64_t seconds = 0;
if (zio_injection_enabled == 0)
return (0);
rw_enter(&inject_lock, RW_READER);
for (handler = list_head(&inject_handlers); handler != NULL;
handler = list_next(&inject_handlers, handler)) {
if (handler->zi_record.zi_cmd != ZINJECT_DELAY_IO)
continue;
if (vd->vdev_guid == handler->zi_record.zi_guid) {
seconds = handler->zi_record.zi_timer;
break;
}
}
rw_exit(&inject_lock);
return (seconds);
}
/*
* Create a new handler for the given record. We add it to the list, adding
* a reference to the spa_t in the process. We increment zio_injection_enabled,