Page MenuHomeFreeBSD
Files F148770870

No OneTemporary
Actions

View Options

Index: stable/8/sys/amd64/include/xen
===================================================================
--- stable/8/sys/amd64/include/xen (revision 209264)
+++ stable/8/sys/amd64/include/xen (revision 209265)
Property changes on: stable/8/sys/amd64/include/xen
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/amd64/include/xen:r209260-209261
Index: stable/8/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c
===================================================================
--- stable/8/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c (revision 209264)
+++ stable/8/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c (revision 209265)
@@ -1,2280 +1,2276 @@
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright 2008 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
#include <sys/zfs_context.h>
#include <sys/fm/fs/zfs.h>
#include <sys/spa.h>
#include <sys/txg.h>
#include <sys/spa_impl.h>
#include <sys/vdev_impl.h>
#include <sys/zio_impl.h>
#include <sys/zio_compress.h>
#include <sys/zio_checksum.h>
-#if defined(__amd64__)
-static int zio_use_uma = 1;
-#else
-static int zio_use_uma = 0;
-#endif
SYSCTL_DECL(_vfs_zfs);
SYSCTL_NODE(_vfs_zfs, OID_AUTO, zio, CTLFLAG_RW, 0, "ZFS ZIO");
+static int zio_use_uma = 0;
TUNABLE_INT("vfs.zfs.zio.use_uma", &zio_use_uma);
SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, use_uma, CTLFLAG_RDTUN, &zio_use_uma, 0,
"Use uma(9) for ZIO allocations");
/*
* ==========================================================================
* I/O priority table
* ==========================================================================
*/
uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE] = {
0, /* ZIO_PRIORITY_NOW */
0, /* ZIO_PRIORITY_SYNC_READ */
0, /* ZIO_PRIORITY_SYNC_WRITE */
6, /* ZIO_PRIORITY_ASYNC_READ */
4, /* ZIO_PRIORITY_ASYNC_WRITE */
4, /* ZIO_PRIORITY_FREE */
0, /* ZIO_PRIORITY_CACHE_FILL */
0, /* ZIO_PRIORITY_LOG_WRITE */
10, /* ZIO_PRIORITY_RESILVER */
20, /* ZIO_PRIORITY_SCRUB */
};
/*
* ==========================================================================
* I/O type descriptions
* ==========================================================================
*/
char *zio_type_name[ZIO_TYPES] = {
"null", "read", "write", "free", "claim", "ioctl" };
#define SYNC_PASS_DEFERRED_FREE 1 /* defer frees after this pass */
#define SYNC_PASS_DONT_COMPRESS 4 /* don't compress after this pass */
#define SYNC_PASS_REWRITE 1 /* rewrite new bps after this pass */
/*
* ==========================================================================
* I/O kmem caches
* ==========================================================================
*/
kmem_cache_t *zio_cache;
kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
#ifdef _KERNEL
extern vmem_t *zio_alloc_arena;
#endif
/*
* An allocating zio is one that either currently has the DVA allocate
* stage set or will have it later in its lifetime.
*/
#define IO_IS_ALLOCATING(zio) \
((zio)->io_orig_pipeline & (1U << ZIO_STAGE_DVA_ALLOCATE))
void
zio_init(void)
{
size_t c;
zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0,
NULL, NULL, NULL, NULL, NULL, 0);
/*
* For small buffers, we want a cache for each multiple of
* SPA_MINBLOCKSIZE. For medium-size buffers, we want a cache
* for each quarter-power of 2. For large buffers, we want
* a cache for each multiple of PAGESIZE.
*/
for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) {
size_t size = (c + 1) << SPA_MINBLOCKSHIFT;
size_t p2 = size;
size_t align = 0;
while (p2 & (p2 - 1))
p2 &= p2 - 1;
if (size <= 4 * SPA_MINBLOCKSIZE) {
align = SPA_MINBLOCKSIZE;
} else if (P2PHASE(size, PAGESIZE) == 0) {
align = PAGESIZE;
} else if (P2PHASE(size, p2 >> 2) == 0) {
align = p2 >> 2;
}
if (align != 0) {
char name[36];
(void) sprintf(name, "zio_buf_%lu", (ulong_t)size);
zio_buf_cache[c] = kmem_cache_create(name, size,
align, NULL, NULL, NULL, NULL, NULL, KMC_NODEBUG);
(void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size);
zio_data_buf_cache[c] = kmem_cache_create(name, size,
align, NULL, NULL, NULL, NULL, NULL, KMC_NODEBUG);
}
}
while (--c != 0) {
ASSERT(zio_buf_cache[c] != NULL);
if (zio_buf_cache[c - 1] == NULL)
zio_buf_cache[c - 1] = zio_buf_cache[c];
ASSERT(zio_data_buf_cache[c] != NULL);
if (zio_data_buf_cache[c - 1] == NULL)
zio_data_buf_cache[c - 1] = zio_data_buf_cache[c];
}
zio_inject_init();
}
void
zio_fini(void)
{
size_t c;
kmem_cache_t *last_cache = NULL;
kmem_cache_t *last_data_cache = NULL;
for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) {
if (zio_buf_cache[c] != last_cache) {
last_cache = zio_buf_cache[c];
kmem_cache_destroy(zio_buf_cache[c]);
}
zio_buf_cache[c] = NULL;
if (zio_data_buf_cache[c] != last_data_cache) {
last_data_cache = zio_data_buf_cache[c];
kmem_cache_destroy(zio_data_buf_cache[c]);
}
zio_data_buf_cache[c] = NULL;
}
kmem_cache_destroy(zio_cache);
zio_inject_fini();
}
/*
* ==========================================================================
* Allocate and free I/O buffers
* ==========================================================================
*/
/*
* Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a
* crashdump if the kernel panics, so use it judiciously. Obviously, it's
* useful to inspect ZFS metadata, but if possible, we should avoid keeping
* excess / transient data in-core during a crashdump.
*/
void *
zio_buf_alloc(size_t size)
{
size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
if (zio_use_uma)
return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE));
else
return (kmem_alloc(size, KM_SLEEP));
}
/*
* Use zio_data_buf_alloc to allocate data. The data will not appear in a
* crashdump if the kernel panics. This exists so that we will limit the amount
* of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount
* of kernel heap dumped to disk when the kernel panics)
*/
void *
zio_data_buf_alloc(size_t size)
{
size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
if (zio_use_uma)
return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE));
else
return (kmem_alloc(size, KM_SLEEP));
}
void
zio_buf_free(void *buf, size_t size)
{
size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
if (zio_use_uma)
kmem_cache_free(zio_buf_cache[c], buf);
else
kmem_free(buf, size);
}
void
zio_data_buf_free(void *buf, size_t size)
{
size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
if (zio_use_uma)
kmem_cache_free(zio_data_buf_cache[c], buf);
else
kmem_free(buf, size);
}
/*
* ==========================================================================
* Push and pop I/O transform buffers
* ==========================================================================
*/
static void
zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize,
zio_transform_func_t *transform)
{
zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP);
zt->zt_orig_data = zio->io_data;
zt->zt_orig_size = zio->io_size;
zt->zt_bufsize = bufsize;
zt->zt_transform = transform;
zt->zt_next = zio->io_transform_stack;
zio->io_transform_stack = zt;
zio->io_data = data;
zio->io_size = size;
}
static void
zio_pop_transforms(zio_t *zio)
{
zio_transform_t *zt;
while ((zt = zio->io_transform_stack) != NULL) {
if (zt->zt_transform != NULL)
zt->zt_transform(zio,
zt->zt_orig_data, zt->zt_orig_size);
zio_buf_free(zio->io_data, zt->zt_bufsize);
zio->io_data = zt->zt_orig_data;
zio->io_size = zt->zt_orig_size;
zio->io_transform_stack = zt->zt_next;
kmem_free(zt, sizeof (zio_transform_t));
}
}
/*
* ==========================================================================
* I/O transform callbacks for subblocks and decompression
* ==========================================================================
*/
static void
zio_subblock(zio_t *zio, void *data, uint64_t size)
{
ASSERT(zio->io_size > size);
if (zio->io_type == ZIO_TYPE_READ)
bcopy(zio->io_data, data, size);
}
static void
zio_decompress(zio_t *zio, void *data, uint64_t size)
{
if (zio->io_error == 0 &&
zio_decompress_data(BP_GET_COMPRESS(zio->io_bp),
zio->io_data, zio->io_size, data, size) != 0)
zio->io_error = EIO;
}
/*
* ==========================================================================
* I/O parent/child relationships and pipeline interlocks
* ==========================================================================
*/
static void
zio_add_child(zio_t *pio, zio_t *zio)
{
mutex_enter(&pio->io_lock);
if (zio->io_stage < ZIO_STAGE_READY)
pio->io_children[zio->io_child_type][ZIO_WAIT_READY]++;
if (zio->io_stage < ZIO_STAGE_DONE)
pio->io_children[zio->io_child_type][ZIO_WAIT_DONE]++;
zio->io_sibling_prev = NULL;
zio->io_sibling_next = pio->io_child;
if (pio->io_child != NULL)
pio->io_child->io_sibling_prev = zio;
pio->io_child = zio;
zio->io_parent = pio;
mutex_exit(&pio->io_lock);
}
static void
zio_remove_child(zio_t *pio, zio_t *zio)
{
zio_t *next, *prev;
ASSERT(zio->io_parent == pio);
mutex_enter(&pio->io_lock);
next = zio->io_sibling_next;
prev = zio->io_sibling_prev;
if (next != NULL)
next->io_sibling_prev = prev;
if (prev != NULL)
prev->io_sibling_next = next;
if (pio->io_child == zio)
pio->io_child = next;
mutex_exit(&pio->io_lock);
}
static boolean_t
zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait)
{
uint64_t *countp = &zio->io_children[child][wait];
boolean_t waiting = B_FALSE;
mutex_enter(&zio->io_lock);
ASSERT(zio->io_stall == NULL);
if (*countp != 0) {
zio->io_stage--;
zio->io_stall = countp;
waiting = B_TRUE;
}
mutex_exit(&zio->io_lock);
return (waiting);
}
static void
zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait)
{
uint64_t *countp = &pio->io_children[zio->io_child_type][wait];
int *errorp = &pio->io_child_error[zio->io_child_type];
mutex_enter(&pio->io_lock);
if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
*errorp = zio_worst_error(*errorp, zio->io_error);
pio->io_reexecute |= zio->io_reexecute;
ASSERT3U(*countp, >, 0);
if (--*countp == 0 && pio->io_stall == countp) {
pio->io_stall = NULL;
mutex_exit(&pio->io_lock);
zio_execute(pio);
} else {
mutex_exit(&pio->io_lock);
}
}
static void
zio_inherit_child_errors(zio_t *zio, enum zio_child c)
{
if (zio->io_child_error[c] != 0 && zio->io_error == 0)
zio->io_error = zio->io_child_error[c];
}
/*
* ==========================================================================
* Create the various types of I/O (read, write, free, etc)
* ==========================================================================
*/
static zio_t *
zio_create(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
void *data, uint64_t size, zio_done_func_t *done, void *private,
zio_type_t type, int priority, int flags, vdev_t *vd, uint64_t offset,
const zbookmark_t *zb, uint8_t stage, uint32_t pipeline)
{
zio_t *zio;
ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0);
ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0);
ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER));
ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER));
ASSERT(vd || stage == ZIO_STAGE_OPEN);
zio = kmem_cache_alloc(zio_cache, KM_SLEEP);
bzero(zio, sizeof (zio_t));
mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL);
if (vd != NULL)
zio->io_child_type = ZIO_CHILD_VDEV;
else if (flags & ZIO_FLAG_GANG_CHILD)
zio->io_child_type = ZIO_CHILD_GANG;
else
zio->io_child_type = ZIO_CHILD_LOGICAL;
if (bp != NULL) {
zio->io_bp = bp;
zio->io_bp_copy = *bp;
zio->io_bp_orig = *bp;
if (type != ZIO_TYPE_WRITE)
zio->io_bp = &zio->io_bp_copy; /* so caller can free */
if (zio->io_child_type == ZIO_CHILD_LOGICAL) {
if (BP_IS_GANG(bp))
pipeline |= ZIO_GANG_STAGES;
zio->io_logical = zio;
}
}
zio->io_spa = spa;
zio->io_txg = txg;
zio->io_data = data;
zio->io_size = size;
zio->io_done = done;
zio->io_private = private;
zio->io_type = type;
zio->io_priority = priority;
zio->io_vd = vd;
zio->io_offset = offset;
zio->io_orig_flags = zio->io_flags = flags;
zio->io_orig_stage = zio->io_stage = stage;
zio->io_orig_pipeline = zio->io_pipeline = pipeline;
if (zb != NULL)
zio->io_bookmark = *zb;
if (pio != NULL) {
/*
* Logical I/Os can have logical, gang, or vdev children.
* Gang I/Os can have gang or vdev children.
* Vdev I/Os can only have vdev children.
* The following ASSERT captures all of these constraints.
*/
ASSERT(zio->io_child_type <= pio->io_child_type);
if (zio->io_logical == NULL)
zio->io_logical = pio->io_logical;
zio_add_child(pio, zio);
}
return (zio);
}
static void
zio_destroy(zio_t *zio)
{
spa_t *spa = zio->io_spa;
uint8_t async_root = zio->io_async_root;
mutex_destroy(&zio->io_lock);
cv_destroy(&zio->io_cv);
kmem_cache_free(zio_cache, zio);
if (async_root) {
mutex_enter(&spa->spa_async_root_lock);
if (--spa->spa_async_root_count == 0)
cv_broadcast(&spa->spa_async_root_cv);
mutex_exit(&spa->spa_async_root_lock);
}
}
zio_t *
zio_null(zio_t *pio, spa_t *spa, zio_done_func_t *done, void *private,
int flags)
{
zio_t *zio;
zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private,
ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, NULL, 0, NULL,
ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE);
return (zio);
}
zio_t *
zio_root(spa_t *spa, zio_done_func_t *done, void *private, int flags)
{
return (zio_null(NULL, spa, done, private, flags));
}
zio_t *
zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
void *data, uint64_t size, zio_done_func_t *done, void *private,
int priority, int flags, const zbookmark_t *zb)
{
zio_t *zio;
zio = zio_create(pio, spa, bp->blk_birth, (blkptr_t *)bp,
data, size, done, private,
ZIO_TYPE_READ, priority, flags, NULL, 0, zb,
ZIO_STAGE_OPEN, ZIO_READ_PIPELINE);
return (zio);
}
zio_t *
zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
void *data, uint64_t size, zio_prop_t *zp,
zio_done_func_t *ready, zio_done_func_t *done, void *private,
int priority, int flags, const zbookmark_t *zb)
{
zio_t *zio;
ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF &&
zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS &&
zp->zp_compress >= ZIO_COMPRESS_OFF &&
zp->zp_compress < ZIO_COMPRESS_FUNCTIONS &&
zp->zp_type < DMU_OT_NUMTYPES &&
zp->zp_level < 32 &&
zp->zp_ndvas > 0 &&
zp->zp_ndvas <= spa_max_replication(spa));
ASSERT(ready != NULL);
zio = zio_create(pio, spa, txg, bp, data, size, done, private,
ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb,
ZIO_STAGE_OPEN, ZIO_WRITE_PIPELINE);
zio->io_ready = ready;
zio->io_prop = *zp;
return (zio);
}
zio_t *
zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data,
uint64_t size, zio_done_func_t *done, void *private, int priority,
int flags, zbookmark_t *zb)
{
zio_t *zio;
zio = zio_create(pio, spa, txg, bp, data, size, done, private,
ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb,
ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE);
return (zio);
}
zio_t *
zio_free(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
zio_done_func_t *done, void *private, int flags)
{
zio_t *zio;
ASSERT(!BP_IS_HOLE(bp));
if (bp->blk_fill == BLK_FILL_ALREADY_FREED)
return (zio_null(pio, spa, NULL, NULL, flags));
if (txg == spa->spa_syncing_txg &&
spa_sync_pass(spa) > SYNC_PASS_DEFERRED_FREE) {
bplist_enqueue_deferred(&spa->spa_sync_bplist, bp);
return (zio_null(pio, spa, NULL, NULL, flags));
}
zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp),
done, private, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags,
NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PIPELINE);
return (zio);
}
zio_t *
zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
zio_done_func_t *done, void *private, int flags)
{
zio_t *zio;
/*
* A claim is an allocation of a specific block. Claims are needed
* to support immediate writes in the intent log. The issue is that
* immediate writes contain committed data, but in a txg that was
* *not* committed. Upon opening the pool after an unclean shutdown,
* the intent log claims all blocks that contain immediate write data
* so that the SPA knows they're in use.
*
* All claims *must* be resolved in the first txg -- before the SPA
* starts allocating blocks -- so that nothing is allocated twice.
*/
ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa));
ASSERT3U(spa_first_txg(spa), <=, txg);
zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp),
done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags,
NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE);
return (zio);
}
zio_t *
zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
zio_done_func_t *done, void *private, int priority, int flags)
{
zio_t *zio;
int c;
if (vd->vdev_children == 0) {
zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private,
ZIO_TYPE_IOCTL, priority, flags, vd, 0, NULL,
ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE);
zio->io_cmd = cmd;
} else {
zio = zio_null(pio, spa, NULL, NULL, flags);
for (c = 0; c < vd->vdev_children; c++)
zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd,
done, private, priority, flags));
}
return (zio);
}
zio_t *
zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
void *data, int checksum, zio_done_func_t *done, void *private,
int priority, int flags, boolean_t labels)
{
zio_t *zio;
ASSERT(vd->vdev_children == 0);
ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE ||
offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE);
ASSERT3U(offset + size, <=, vd->vdev_psize);
zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
ZIO_TYPE_READ, priority, flags, vd, offset, NULL,
ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE);
zio->io_prop.zp_checksum = checksum;
return (zio);
}
zio_t *
zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
void *data, int checksum, zio_done_func_t *done, void *private,
int priority, int flags, boolean_t labels)
{
zio_t *zio;
ASSERT(vd->vdev_children == 0);
ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE ||
offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE);
ASSERT3U(offset + size, <=, vd->vdev_psize);
zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL,
ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE);
zio->io_prop.zp_checksum = checksum;
if (zio_checksum_table[checksum].ci_zbt) {
/*
* zbt checksums are necessarily destructive -- they modify
* the end of the write buffer to hold the verifier/checksum.
* Therefore, we must make a local copy in case the data is
* being written to multiple places in parallel.
*/
void *wbuf = zio_buf_alloc(size);
bcopy(data, wbuf, size);
zio_push_transform(zio, wbuf, size, size, NULL);
}
return (zio);
}
/*
* Create a child I/O to do some work for us.
*/
zio_t *
zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset,
void *data, uint64_t size, int type, int priority, int flags,
zio_done_func_t *done, void *private)
{
uint32_t pipeline = ZIO_VDEV_CHILD_PIPELINE;
zio_t *zio;
ASSERT(vd->vdev_parent ==
(pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev));
if (type == ZIO_TYPE_READ && bp != NULL) {
/*
* If we have the bp, then the child should perform the
* checksum and the parent need not. This pushes error
* detection as close to the leaves as possible and
* eliminates redundant checksums in the interior nodes.
*/
pipeline |= 1U << ZIO_STAGE_CHECKSUM_VERIFY;
pio->io_pipeline &= ~(1U << ZIO_STAGE_CHECKSUM_VERIFY);
}
if (vd->vdev_children == 0)
offset += VDEV_LABEL_START_SIZE;
zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size,
done, private, type, priority,
(pio->io_flags & ZIO_FLAG_VDEV_INHERIT) |
ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | flags,
vd, offset, &pio->io_bookmark,
ZIO_STAGE_VDEV_IO_START - 1, pipeline);
return (zio);
}
zio_t *
zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size,
int type, int priority, int flags, zio_done_func_t *done, void *private)
{
zio_t *zio;
ASSERT(vd->vdev_ops->vdev_op_leaf);
zio = zio_create(NULL, vd->vdev_spa, 0, NULL,
data, size, done, private, type, priority,
flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY,
vd, offset, NULL,
ZIO_STAGE_VDEV_IO_START - 1, ZIO_VDEV_CHILD_PIPELINE);
return (zio);
}
void
zio_flush(zio_t *zio, vdev_t *vd)
{
zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE,
NULL, NULL, ZIO_PRIORITY_NOW,
ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY));
}
/*
* ==========================================================================
* Prepare to read and write logical blocks
* ==========================================================================
*/
static int
zio_read_bp_init(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && zio->io_logical == zio) {
uint64_t csize = BP_GET_PSIZE(bp);
void *cbuf = zio_buf_alloc(csize);
zio_push_transform(zio, cbuf, csize, csize, zio_decompress);
}
if (!dmu_ot[BP_GET_TYPE(bp)].ot_metadata && BP_GET_LEVEL(bp) == 0)
zio->io_flags |= ZIO_FLAG_DONT_CACHE;
return (ZIO_PIPELINE_CONTINUE);
}
static int
zio_write_bp_init(zio_t *zio)
{
zio_prop_t *zp = &zio->io_prop;
int compress = zp->zp_compress;
blkptr_t *bp = zio->io_bp;
void *cbuf;
uint64_t lsize = zio->io_size;
uint64_t csize = lsize;
uint64_t cbufsize = 0;
int pass = 1;
/*
* If our children haven't all reached the ready stage,
* wait for them and then repeat this pipeline stage.
*/
if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) ||
zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY))
return (ZIO_PIPELINE_STOP);
if (!IO_IS_ALLOCATING(zio))
return (ZIO_PIPELINE_CONTINUE);
ASSERT(compress != ZIO_COMPRESS_INHERIT);
if (bp->blk_birth == zio->io_txg) {
/*
* We're rewriting an existing block, which means we're
* working on behalf of spa_sync(). For spa_sync() to
* converge, it must eventually be the case that we don't
* have to allocate new blocks. But compression changes
* the blocksize, which forces a reallocate, and makes
* convergence take longer. Therefore, after the first
* few passes, stop compressing to ensure convergence.
*/
pass = spa_sync_pass(zio->io_spa);
ASSERT(pass > 1);
if (pass > SYNC_PASS_DONT_COMPRESS)
compress = ZIO_COMPRESS_OFF;
/*
* Only MOS (objset 0) data should need to be rewritten.
*/
ASSERT(zio->io_logical->io_bookmark.zb_objset == 0);
/* Make sure someone doesn't change their mind on overwrites */
ASSERT(MIN(zp->zp_ndvas + BP_IS_GANG(bp),
spa_max_replication(zio->io_spa)) == BP_GET_NDVAS(bp));
}
if (compress != ZIO_COMPRESS_OFF) {
if (!zio_compress_data(compress, zio->io_data, zio->io_size,
&cbuf, &csize, &cbufsize)) {
compress = ZIO_COMPRESS_OFF;
} else if (csize != 0) {
zio_push_transform(zio, cbuf, csize, cbufsize, NULL);
}
}
/*
* The final pass of spa_sync() must be all rewrites, but the first
* few passes offer a trade-off: allocating blocks defers convergence,
* but newly allocated blocks are sequential, so they can be written
* to disk faster. Therefore, we allow the first few passes of
* spa_sync() to allocate new blocks, but force rewrites after that.
* There should only be a handful of blocks after pass 1 in any case.
*/
if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == csize &&
pass > SYNC_PASS_REWRITE) {
ASSERT(csize != 0);
uint32_t gang_stages = zio->io_pipeline & ZIO_GANG_STAGES;
zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages;
zio->io_flags |= ZIO_FLAG_IO_REWRITE;
} else {
BP_ZERO(bp);
zio->io_pipeline = ZIO_WRITE_PIPELINE;
}
if (csize == 0) {
zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
} else {
ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER);
BP_SET_LSIZE(bp, lsize);
BP_SET_PSIZE(bp, csize);
BP_SET_COMPRESS(bp, compress);
BP_SET_CHECKSUM(bp, zp->zp_checksum);
BP_SET_TYPE(bp, zp->zp_type);
BP_SET_LEVEL(bp, zp->zp_level);
BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
}
return (ZIO_PIPELINE_CONTINUE);
}
/*
* ==========================================================================
* Execute the I/O pipeline
* ==========================================================================
*/
static void
zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q)
{
zio_type_t t = zio->io_type;
/*
* If we're a config writer, the normal issue and interrupt threads
* may all be blocked waiting for the config lock. In this case,
* select the otherwise-unused taskq for ZIO_TYPE_NULL.
*/
if (zio->io_flags & ZIO_FLAG_CONFIG_WRITER)
t = ZIO_TYPE_NULL;
/*
* A similar issue exists for the L2ARC write thread until L2ARC 2.0.
*/
if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux)
t = ZIO_TYPE_NULL;
(void) taskq_dispatch_safe(zio->io_spa->spa_zio_taskq[t][q],
(task_func_t *)zio_execute, zio, &zio->io_task);
}
static boolean_t
zio_taskq_member(zio_t *zio, enum zio_taskq_type q)
{
kthread_t *executor = zio->io_executor;
spa_t *spa = zio->io_spa;
for (zio_type_t t = 0; t < ZIO_TYPES; t++)
if (taskq_member(spa->spa_zio_taskq[t][q], executor))
return (B_TRUE);
return (B_FALSE);
}
static int
zio_issue_async(zio_t *zio)
{
zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE);
return (ZIO_PIPELINE_STOP);
}
void
zio_interrupt(zio_t *zio)
{
zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT);
}
/*
* Execute the I/O pipeline until one of the following occurs:
* (1) the I/O completes; (2) the pipeline stalls waiting for
* dependent child I/Os; (3) the I/O issues, so we're waiting
* for an I/O completion interrupt; (4) the I/O is delegated by
* vdev-level caching or aggregation; (5) the I/O is deferred
* due to vdev-level queueing; (6) the I/O is handed off to
* another thread. In all cases, the pipeline stops whenever
* there's no CPU work; it never burns a thread in cv_wait().
*
* There's no locking on io_stage because there's no legitimate way
* for multiple threads to be attempting to process the same I/O.
*/
static zio_pipe_stage_t *zio_pipeline[ZIO_STAGES];
void
zio_execute(zio_t *zio)
{
zio->io_executor = curthread;
while (zio->io_stage < ZIO_STAGE_DONE) {
uint32_t pipeline = zio->io_pipeline;
zio_stage_t stage = zio->io_stage;
int rv;
ASSERT(!MUTEX_HELD(&zio->io_lock));
while (((1U << ++stage) & pipeline) == 0)
continue;
ASSERT(stage <= ZIO_STAGE_DONE);
ASSERT(zio->io_stall == NULL);
/*
* If we are in interrupt context and this pipeline stage
* will grab a config lock that is held across I/O,
* issue async to avoid deadlock.
*/
if (((1U << stage) & ZIO_CONFIG_LOCK_BLOCKING_STAGES) &&
zio->io_vd == NULL &&
zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) {
zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE);
return;
}
zio->io_stage = stage;
rv = zio_pipeline[stage](zio);
if (rv == ZIO_PIPELINE_STOP)
return;
ASSERT(rv == ZIO_PIPELINE_CONTINUE);
}
}
/*
* ==========================================================================
* Initiate I/O, either sync or async
* ==========================================================================
*/
int
zio_wait(zio_t *zio)
{
int error;
ASSERT(zio->io_stage == ZIO_STAGE_OPEN);
ASSERT(zio->io_executor == NULL);
zio->io_waiter = curthread;
zio_execute(zio);
mutex_enter(&zio->io_lock);
while (zio->io_executor != NULL)
cv_wait(&zio->io_cv, &zio->io_lock);
mutex_exit(&zio->io_lock);
error = zio->io_error;
zio_destroy(zio);
return (error);
}
void
zio_nowait(zio_t *zio)
{
ASSERT(zio->io_executor == NULL);
if (zio->io_parent == NULL && zio->io_child_type == ZIO_CHILD_LOGICAL) {
/*
* This is a logical async I/O with no parent to wait for it.
* Attach it to the pool's global async root zio so that
* spa_unload() has a way of waiting for async I/O to finish.
*/
spa_t *spa = zio->io_spa;
zio->io_async_root = B_TRUE;
mutex_enter(&spa->spa_async_root_lock);
spa->spa_async_root_count++;
mutex_exit(&spa->spa_async_root_lock);
}
zio_execute(zio);
}
/*
* ==========================================================================
* Reexecute or suspend/resume failed I/O
* ==========================================================================
*/
static void
zio_reexecute(zio_t *pio)
{
zio_t *zio, *zio_next;
pio->io_flags = pio->io_orig_flags;
pio->io_stage = pio->io_orig_stage;
pio->io_pipeline = pio->io_orig_pipeline;
pio->io_reexecute = 0;
pio->io_error = 0;
for (int c = 0; c < ZIO_CHILD_TYPES; c++)
pio->io_child_error[c] = 0;
if (IO_IS_ALLOCATING(pio)) {
/*
* Remember the failed bp so that the io_ready() callback
* can update its accounting upon reexecution. The block
* was already freed in zio_done(); we indicate this with
* a fill count of -1 so that zio_free() knows to skip it.
*/
blkptr_t *bp = pio->io_bp;
ASSERT(bp->blk_birth == 0 || bp->blk_birth == pio->io_txg);
bp->blk_fill = BLK_FILL_ALREADY_FREED;
pio->io_bp_orig = *bp;
BP_ZERO(bp);
}
/*
* As we reexecute pio's children, new children could be created.
* New children go to the head of the io_child list, however,
* so we will (correctly) not reexecute them. The key is that
* the remainder of the io_child list, from 'zio_next' onward,
* cannot be affected by any side effects of reexecuting 'zio'.
*/
for (zio = pio->io_child; zio != NULL; zio = zio_next) {
zio_next = zio->io_sibling_next;
mutex_enter(&pio->io_lock);
pio->io_children[zio->io_child_type][ZIO_WAIT_READY]++;
pio->io_children[zio->io_child_type][ZIO_WAIT_DONE]++;
mutex_exit(&pio->io_lock);
zio_reexecute(zio);
}
/*
* Now that all children have been reexecuted, execute the parent.
*/
zio_execute(pio);
}
void
zio_suspend(spa_t *spa, zio_t *zio)
{
if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC)
fm_panic("Pool '%s' has encountered an uncorrectable I/O "
"failure and the failure mode property for this pool "
"is set to panic.", spa_name(spa));
zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0);
mutex_enter(&spa->spa_suspend_lock);
if (spa->spa_suspend_zio_root == NULL)
spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, 0);
spa->spa_suspended = B_TRUE;
if (zio != NULL) {
ASSERT(zio != spa->spa_suspend_zio_root);
ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
ASSERT(zio->io_parent == NULL);
ASSERT(zio->io_stage == ZIO_STAGE_DONE);
zio_add_child(spa->spa_suspend_zio_root, zio);
}
mutex_exit(&spa->spa_suspend_lock);
}
void
zio_resume(spa_t *spa)
{
zio_t *pio, *zio;
/*
* Reexecute all previously suspended i/o.
*/
mutex_enter(&spa->spa_suspend_lock);
spa->spa_suspended = B_FALSE;
cv_broadcast(&spa->spa_suspend_cv);
pio = spa->spa_suspend_zio_root;
spa->spa_suspend_zio_root = NULL;
mutex_exit(&spa->spa_suspend_lock);
if (pio == NULL)
return;
while ((zio = pio->io_child) != NULL) {
zio_remove_child(pio, zio);
zio->io_parent = NULL;
zio_reexecute(zio);
}
ASSERT(pio->io_children[ZIO_CHILD_LOGICAL][ZIO_WAIT_DONE] == 0);
(void) zio_wait(pio);
}
void
zio_resume_wait(spa_t *spa)
{
mutex_enter(&spa->spa_suspend_lock);
while (spa_suspended(spa))
cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock);
mutex_exit(&spa->spa_suspend_lock);
}
/*
* ==========================================================================
* Gang blocks.
*
* A gang block is a collection of small blocks that looks to the DMU
* like one large block. When zio_dva_allocate() cannot find a block
* of the requested size, due to either severe fragmentation or the pool
* being nearly full, it calls zio_write_gang_block() to construct the
* block from smaller fragments.
*
* A gang block consists of a gang header (zio_gbh_phys_t) and up to
* three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like
* an indirect block: it's an array of block pointers. It consumes
* only one sector and hence is allocatable regardless of fragmentation.
* The gang header's bps point to its gang members, which hold the data.
*
* Gang blocks are self-checksumming, using the bp's <vdev, offset, txg>
* as the verifier to ensure uniqueness of the SHA256 checksum.
* Critically, the gang block bp's blk_cksum is the checksum of the data,
* not the gang header. This ensures that data block signatures (needed for
* deduplication) are independent of how the block is physically stored.
*
* Gang blocks can be nested: a gang member may itself be a gang block.
* Thus every gang block is a tree in which root and all interior nodes are
* gang headers, and the leaves are normal blocks that contain user data.
* The root of the gang tree is called the gang leader.
*
* To perform any operation (read, rewrite, free, claim) on a gang block,
* zio_gang_assemble() first assembles the gang tree (minus data leaves)
* in the io_gang_tree field of the original logical i/o by recursively
* reading the gang leader and all gang headers below it. This yields
* an in-core tree containing the contents of every gang header and the
* bps for every constituent of the gang block.
*
* With the gang tree now assembled, zio_gang_issue() just walks the gang tree
* and invokes a callback on each bp. To free a gang block, zio_gang_issue()
* calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp.
* zio_claim_gang() provides a similarly trivial wrapper for zio_claim().
* zio_read_gang() is a wrapper around zio_read() that omits reading gang
* headers, since we already have those in io_gang_tree. zio_rewrite_gang()
* performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite()
* of the gang header plus zio_checksum_compute() of the data to update the
* gang header's blk_cksum as described above.
*
* The two-phase assemble/issue model solves the problem of partial failure --
* what if you'd freed part of a gang block but then couldn't read the
* gang header for another part? Assembling the entire gang tree first
* ensures that all the necessary gang header I/O has succeeded before
* starting the actual work of free, claim, or write. Once the gang tree
* is assembled, free and claim are in-memory operations that cannot fail.
*
* In the event that a gang write fails, zio_dva_unallocate() walks the
* gang tree to immediately free (i.e. insert back into the space map)
* everything we've allocated. This ensures that we don't get ENOSPC
* errors during repeated suspend/resume cycles due to a flaky device.
*
* Gang rewrites only happen during sync-to-convergence. If we can't assemble
* the gang tree, we won't modify the block, so we can safely defer the free
* (knowing that the block is still intact). If we *can* assemble the gang
* tree, then even if some of the rewrites fail, zio_dva_unallocate() will free
* each constituent bp and we can allocate a new block on the next sync pass.
*
* In all cases, the gang tree allows complete recovery from partial failure.
* ==========================================================================
*/
static zio_t *
zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
{
if (gn != NULL)
return (pio);
return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp),
NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio),
&pio->io_bookmark));
}
zio_t *
zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
{
zio_t *zio;
if (gn != NULL) {
zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp,
gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority,
ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
/*
* As we rewrite each gang header, the pipeline will compute
* a new gang block header checksum for it; but no one will
* compute a new data checksum, so we do that here. The one
* exception is the gang leader: the pipeline already computed
* its data checksum because that stage precedes gang assembly.
* (Presently, nothing actually uses interior data checksums;
* this is just good hygiene.)
*/
if (gn != pio->io_logical->io_gang_tree) {
zio_checksum_compute(zio, BP_GET_CHECKSUM(bp),
data, BP_GET_PSIZE(bp));
}
} else {
zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp,
data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority,
ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
}
return (zio);
}
/* ARGSUSED */
zio_t *
zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
{
return (zio_free(pio, pio->io_spa, pio->io_txg, bp,
NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio)));
}
/* ARGSUSED */
zio_t *
zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
{
return (zio_claim(pio, pio->io_spa, pio->io_txg, bp,
NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio)));
}
static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = {
NULL,
zio_read_gang,
zio_rewrite_gang,
zio_free_gang,
zio_claim_gang,
NULL
};
static void zio_gang_tree_assemble_done(zio_t *zio);
static zio_gang_node_t *
zio_gang_node_alloc(zio_gang_node_t **gnpp)
{
zio_gang_node_t *gn;
ASSERT(*gnpp == NULL);
gn = kmem_zalloc(sizeof (*gn), KM_SLEEP);
gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE);
*gnpp = gn;
return (gn);
}
static void
zio_gang_node_free(zio_gang_node_t **gnpp)
{
zio_gang_node_t *gn = *gnpp;
for (int g = 0; g < SPA_GBH_NBLKPTRS; g++)
ASSERT(gn->gn_child[g] == NULL);
zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE);
kmem_free(gn, sizeof (*gn));
*gnpp = NULL;
}
static void
zio_gang_tree_free(zio_gang_node_t **gnpp)
{
zio_gang_node_t *gn = *gnpp;
if (gn == NULL)
return;
for (int g = 0; g < SPA_GBH_NBLKPTRS; g++)
zio_gang_tree_free(&gn->gn_child[g]);
zio_gang_node_free(gnpp);
}
static void
zio_gang_tree_assemble(zio_t *lio, blkptr_t *bp, zio_gang_node_t **gnpp)
{
zio_gang_node_t *gn = zio_gang_node_alloc(gnpp);
ASSERT(lio->io_logical == lio);
ASSERT(BP_IS_GANG(bp));
zio_nowait(zio_read(lio, lio->io_spa, bp, gn->gn_gbh,
SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn,
lio->io_priority, ZIO_GANG_CHILD_FLAGS(lio), &lio->io_bookmark));
}
static void
zio_gang_tree_assemble_done(zio_t *zio)
{
zio_t *lio = zio->io_logical;
zio_gang_node_t *gn = zio->io_private;
blkptr_t *bp = zio->io_bp;
ASSERT(zio->io_parent == lio);
ASSERT(zio->io_child == NULL);
if (zio->io_error)
return;
if (BP_SHOULD_BYTESWAP(bp))
byteswap_uint64_array(zio->io_data, zio->io_size);
ASSERT(zio->io_data == gn->gn_gbh);
ASSERT(zio->io_size == SPA_GANGBLOCKSIZE);
ASSERT(gn->gn_gbh->zg_tail.zbt_magic == ZBT_MAGIC);
for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
if (!BP_IS_GANG(gbp))
continue;
zio_gang_tree_assemble(lio, gbp, &gn->gn_child[g]);
}
}
static void
zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data)
{
zio_t *lio = pio->io_logical;
zio_t *zio;
ASSERT(BP_IS_GANG(bp) == !!gn);
ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(lio->io_bp));
ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == lio->io_gang_tree);
/*
* If you're a gang header, your data is in gn->gn_gbh.
* If you're a gang member, your data is in 'data' and gn == NULL.
*/
zio = zio_gang_issue_func[lio->io_type](pio, bp, gn, data);
if (gn != NULL) {
ASSERT(gn->gn_gbh->zg_tail.zbt_magic == ZBT_MAGIC);
for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
if (BP_IS_HOLE(gbp))
continue;
zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data);
data = (char *)data + BP_GET_PSIZE(gbp);
}
}
if (gn == lio->io_gang_tree)
ASSERT3P((char *)lio->io_data + lio->io_size, ==, data);
if (zio != pio)
zio_nowait(zio);
}
static int
zio_gang_assemble(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
ASSERT(BP_IS_GANG(bp) && zio == zio->io_logical);
zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree);
return (ZIO_PIPELINE_CONTINUE);
}
static int
zio_gang_issue(zio_t *zio)
{
zio_t *lio = zio->io_logical;
blkptr_t *bp = zio->io_bp;
if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE))
return (ZIO_PIPELINE_STOP);
ASSERT(BP_IS_GANG(bp) && zio == lio);
if (zio->io_child_error[ZIO_CHILD_GANG] == 0)
zio_gang_tree_issue(lio, lio->io_gang_tree, bp, lio->io_data);
else
zio_gang_tree_free(&lio->io_gang_tree);
zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
return (ZIO_PIPELINE_CONTINUE);
}
static void
zio_write_gang_member_ready(zio_t *zio)
{
zio_t *pio = zio->io_parent;
zio_t *lio = zio->io_logical;
dva_t *cdva = zio->io_bp->blk_dva;
dva_t *pdva = pio->io_bp->blk_dva;
uint64_t asize;
if (BP_IS_HOLE(zio->io_bp))
return;
ASSERT(BP_IS_HOLE(&zio->io_bp_orig));
ASSERT(zio->io_child_type == ZIO_CHILD_GANG);
ASSERT3U(zio->io_prop.zp_ndvas, ==, lio->io_prop.zp_ndvas);
ASSERT3U(zio->io_prop.zp_ndvas, <=, BP_GET_NDVAS(zio->io_bp));
ASSERT3U(pio->io_prop.zp_ndvas, <=, BP_GET_NDVAS(pio->io_bp));
ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp));
mutex_enter(&pio->io_lock);
for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) {
ASSERT(DVA_GET_GANG(&pdva[d]));
asize = DVA_GET_ASIZE(&pdva[d]);
asize += DVA_GET_ASIZE(&cdva[d]);
DVA_SET_ASIZE(&pdva[d], asize);
}
mutex_exit(&pio->io_lock);
}
static int
zio_write_gang_block(zio_t *pio)
{
spa_t *spa = pio->io_spa;
blkptr_t *bp = pio->io_bp;
zio_t *lio = pio->io_logical;
zio_t *zio;
zio_gang_node_t *gn, **gnpp;
zio_gbh_phys_t *gbh;
uint64_t txg = pio->io_txg;
uint64_t resid = pio->io_size;
uint64_t lsize;
int ndvas = lio->io_prop.zp_ndvas;
int gbh_ndvas = MIN(ndvas + 1, spa_max_replication(spa));
zio_prop_t zp;
int error;
error = metaslab_alloc(spa, spa->spa_normal_class, SPA_GANGBLOCKSIZE,
bp, gbh_ndvas, txg, pio == lio ? NULL : lio->io_bp,
METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER);
if (error) {
pio->io_error = error;
return (ZIO_PIPELINE_CONTINUE);
}
if (pio == lio) {
gnpp = &lio->io_gang_tree;
} else {
gnpp = pio->io_private;
ASSERT(pio->io_ready == zio_write_gang_member_ready);
}
gn = zio_gang_node_alloc(gnpp);
gbh = gn->gn_gbh;
bzero(gbh, SPA_GANGBLOCKSIZE);
/*
* Create the gang header.
*/
zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL,
pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
/*
* Create and nowait the gang children.
*/
for (int g = 0; resid != 0; resid -= lsize, g++) {
lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g),
SPA_MINBLOCKSIZE);
ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid);
zp.zp_checksum = lio->io_prop.zp_checksum;
zp.zp_compress = ZIO_COMPRESS_OFF;
zp.zp_type = DMU_OT_NONE;
zp.zp_level = 0;
zp.zp_ndvas = lio->io_prop.zp_ndvas;
zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g],
(char *)pio->io_data + (pio->io_size - resid), lsize, &zp,
zio_write_gang_member_ready, NULL, &gn->gn_child[g],
pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio),
&pio->io_bookmark));
}
/*
* Set pio's pipeline to just wait for zio to finish.
*/
pio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
zio_nowait(zio);
return (ZIO_PIPELINE_CONTINUE);
}
/*
* ==========================================================================
* Allocate and free blocks
* ==========================================================================
*/
static int
zio_dva_allocate(zio_t *zio)
{
spa_t *spa = zio->io_spa;
metaslab_class_t *mc = spa->spa_normal_class;
blkptr_t *bp = zio->io_bp;
int error;
ASSERT(BP_IS_HOLE(bp));
ASSERT3U(BP_GET_NDVAS(bp), ==, 0);
ASSERT3U(zio->io_prop.zp_ndvas, >, 0);
ASSERT3U(zio->io_prop.zp_ndvas, <=, spa_max_replication(spa));
ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp));
error = metaslab_alloc(spa, mc, zio->io_size, bp,
zio->io_prop.zp_ndvas, zio->io_txg, NULL, 0);
if (error) {
if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE)
return (zio_write_gang_block(zio));
zio->io_error = error;
}
return (ZIO_PIPELINE_CONTINUE);
}
static int
zio_dva_free(zio_t *zio)
{
metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE);
return (ZIO_PIPELINE_CONTINUE);
}
static int
zio_dva_claim(zio_t *zio)
{
int error;
error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg);
if (error)
zio->io_error = error;
return (ZIO_PIPELINE_CONTINUE);
}
/*
* Undo an allocation. This is used by zio_done() when an I/O fails
* and we want to give back the block we just allocated.
* This handles both normal blocks and gang blocks.
*/
static void
zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp)
{
spa_t *spa = zio->io_spa;
boolean_t now = !(zio->io_flags & ZIO_FLAG_IO_REWRITE);
ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp));
if (zio->io_bp == bp && !now) {
/*
* This is a rewrite for sync-to-convergence.
* We can't do a metaslab_free(NOW) because bp wasn't allocated
* during this sync pass, which means that metaslab_sync()
* already committed the allocation.
*/
ASSERT(DVA_EQUAL(BP_IDENTITY(bp),
BP_IDENTITY(&zio->io_bp_orig)));
ASSERT(spa_sync_pass(spa) > 1);
if (BP_IS_GANG(bp) && gn == NULL) {
/*
* This is a gang leader whose gang header(s) we
* couldn't read now, so defer the free until later.
* The block should still be intact because without
* the headers, we'd never even start the rewrite.
*/
bplist_enqueue_deferred(&spa->spa_sync_bplist, bp);
return;
}
}
if (!BP_IS_HOLE(bp))
metaslab_free(spa, bp, bp->blk_birth, now);
if (gn != NULL) {
for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
zio_dva_unallocate(zio, gn->gn_child[g],
&gn->gn_gbh->zg_blkptr[g]);
}
}
}
/*
* Try to allocate an intent log block. Return 0 on success, errno on failure.
*/
int
zio_alloc_blk(spa_t *spa, uint64_t size, blkptr_t *new_bp, blkptr_t *old_bp,
uint64_t txg)
{
int error;
error = metaslab_alloc(spa, spa->spa_log_class, size,
new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID);
if (error)
error = metaslab_alloc(spa, spa->spa_normal_class, size,
new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID);
if (error == 0) {
BP_SET_LSIZE(new_bp, size);
BP_SET_PSIZE(new_bp, size);
BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF);
BP_SET_CHECKSUM(new_bp, ZIO_CHECKSUM_ZILOG);
BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG);
BP_SET_LEVEL(new_bp, 0);
BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER);
}
return (error);
}
/*
* Free an intent log block. We know it can't be a gang block, so there's
* nothing to do except metaslab_free() it.
*/
void
zio_free_blk(spa_t *spa, blkptr_t *bp, uint64_t txg)
{
ASSERT(!BP_IS_GANG(bp));
metaslab_free(spa, bp, txg, B_FALSE);
}
/*
* ==========================================================================
* Read and write to physical devices
* ==========================================================================
*/
static void
zio_vdev_io_probe_done(zio_t *zio)
{
zio_t *dio;
vdev_t *vd = zio->io_private;
mutex_enter(&vd->vdev_probe_lock);
ASSERT(vd->vdev_probe_zio == zio);
vd->vdev_probe_zio = NULL;
mutex_exit(&vd->vdev_probe_lock);
while ((dio = zio->io_delegate_list) != NULL) {
zio->io_delegate_list = dio->io_delegate_next;
dio->io_delegate_next = NULL;
if (!vdev_accessible(vd, dio))
dio->io_error = ENXIO;
zio_execute(dio);
}
}
/*
* Probe the device to determine whether I/O failure is specific to this
* zio (e.g. a bad sector) or affects the entire vdev (e.g. unplugged).
*/
static int
zio_vdev_io_probe(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
zio_t *pio = NULL;
boolean_t created_pio = B_FALSE;
/*
* Don't probe the probe.
*/
if (zio->io_flags & ZIO_FLAG_PROBE)
return (ZIO_PIPELINE_CONTINUE);
/*
* To prevent 'probe storms' when a device fails, we create
* just one probe i/o at a time. All zios that want to probe
* this vdev will join the probe zio's io_delegate_list.
*/
mutex_enter(&vd->vdev_probe_lock);
if ((pio = vd->vdev_probe_zio) == NULL) {
vd->vdev_probe_zio = pio = zio_root(zio->io_spa,
zio_vdev_io_probe_done, vd, ZIO_FLAG_CANFAIL);
created_pio = B_TRUE;
vd->vdev_probe_wanted = B_TRUE;
spa_async_request(zio->io_spa, SPA_ASYNC_PROBE);
}
zio->io_delegate_next = pio->io_delegate_list;
pio->io_delegate_list = zio;
mutex_exit(&vd->vdev_probe_lock);
if (created_pio) {
zio_nowait(vdev_probe(vd, pio));
zio_nowait(pio);
}
return (ZIO_PIPELINE_STOP);
}
static int
zio_vdev_io_start(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
uint64_t align;
spa_t *spa = zio->io_spa;
ASSERT(zio->io_error == 0);
ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0);
if (vd == NULL) {
if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER))
spa_config_enter(spa, SCL_ZIO, zio, RW_READER);
/*
* The mirror_ops handle multiple DVAs in a single BP.
*/
return (vdev_mirror_ops.vdev_op_io_start(zio));
}
align = 1ULL << vd->vdev_top->vdev_ashift;
if (P2PHASE(zio->io_size, align) != 0) {
uint64_t asize = P2ROUNDUP(zio->io_size, align);
char *abuf = zio_buf_alloc(asize);
ASSERT(vd == vd->vdev_top);
if (zio->io_type == ZIO_TYPE_WRITE) {
bcopy(zio->io_data, abuf, zio->io_size);
bzero(abuf + zio->io_size, asize - zio->io_size);
}
zio_push_transform(zio, abuf, asize, asize, zio_subblock);
}
ASSERT(P2PHASE(zio->io_offset, align) == 0);
ASSERT(P2PHASE(zio->io_size, align) == 0);
ASSERT(zio->io_type != ZIO_TYPE_WRITE || (spa_mode & FWRITE));
if (vd->vdev_ops->vdev_op_leaf &&
(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) {
if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
return (ZIO_PIPELINE_STOP);
if ((zio = vdev_queue_io(zio)) == NULL)
return (ZIO_PIPELINE_STOP);
if (!vdev_accessible(vd, zio)) {
zio->io_error = ENXIO;
zio_interrupt(zio);
return (ZIO_PIPELINE_STOP);
}
}
return (vd->vdev_ops->vdev_op_io_start(zio));
}
static int
zio_vdev_io_done(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops;
boolean_t unexpected_error = B_FALSE;
if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE))
return (ZIO_PIPELINE_STOP);
ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
if (vd != NULL && vd->vdev_ops->vdev_op_leaf) {
vdev_queue_io_done(zio);
if (zio->io_type == ZIO_TYPE_WRITE)
vdev_cache_write(zio);
if (zio_injection_enabled && zio->io_error == 0)
zio->io_error = zio_handle_device_injection(vd, EIO);
if (zio_injection_enabled && zio->io_error == 0)
zio->io_error = zio_handle_label_injection(zio, EIO);
if (zio->io_error) {
if (!vdev_accessible(vd, zio)) {
zio->io_error = ENXIO;
} else {
unexpected_error = B_TRUE;
}
}
}
ops->vdev_op_io_done(zio);
if (unexpected_error)
return (zio_vdev_io_probe(zio));
return (ZIO_PIPELINE_CONTINUE);
}
static int
zio_vdev_io_assess(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE))
return (ZIO_PIPELINE_STOP);
if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER))
spa_config_exit(zio->io_spa, SCL_ZIO, zio);
if (zio->io_vsd != NULL) {
zio->io_vsd_free(zio);
zio->io_vsd = NULL;
}
if (zio_injection_enabled && zio->io_error == 0)
zio->io_error = zio_handle_fault_injection(zio, EIO);
/*
* If the I/O failed, determine whether we should attempt to retry it.
*/
if (zio->io_error && vd == NULL &&
!(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) {
ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */
ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */
zio->io_error = 0;
zio->io_flags |= ZIO_FLAG_IO_RETRY |
ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE;
zio->io_stage = ZIO_STAGE_VDEV_IO_START - 1;
zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE);
return (ZIO_PIPELINE_STOP);
}
/*
* If we got an error on a leaf device, convert it to ENXIO
* if the device is not accessible at all.
*/
if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf &&
!vdev_accessible(vd, zio))
zio->io_error = ENXIO;
/*
* If we can't write to an interior vdev (mirror or RAID-Z),
* set vdev_cant_write so that we stop trying to allocate from it.
*/
if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE &&
vd != NULL && !vd->vdev_ops->vdev_op_leaf)
vd->vdev_cant_write = B_TRUE;
if (zio->io_error)
zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
return (ZIO_PIPELINE_CONTINUE);
}
void
zio_vdev_io_reissue(zio_t *zio)
{
ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START);
ASSERT(zio->io_error == 0);
zio->io_stage--;
}
void
zio_vdev_io_redone(zio_t *zio)
{
ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE);
zio->io_stage--;
}
void
zio_vdev_io_bypass(zio_t *zio)
{
ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START);
ASSERT(zio->io_error == 0);
zio->io_flags |= ZIO_FLAG_IO_BYPASS;
zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS - 1;
}
/*
* ==========================================================================
* Generate and verify checksums
* ==========================================================================
*/
static int
zio_checksum_generate(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
enum zio_checksum checksum;
if (bp == NULL) {
/*
* This is zio_write_phys().
* We're either generating a label checksum, or none at all.
*/
checksum = zio->io_prop.zp_checksum;
if (checksum == ZIO_CHECKSUM_OFF)
return (ZIO_PIPELINE_CONTINUE);
ASSERT(checksum == ZIO_CHECKSUM_LABEL);
} else {
if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) {
ASSERT(!IO_IS_ALLOCATING(zio));
checksum = ZIO_CHECKSUM_GANG_HEADER;
} else {
checksum = BP_GET_CHECKSUM(bp);
}
}
zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size);
return (ZIO_PIPELINE_CONTINUE);
}
static int
zio_checksum_verify(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
int error;
if (bp == NULL) {
/*
* This is zio_read_phys().
* We're either verifying a label checksum, or nothing at all.
*/
if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF)
return (ZIO_PIPELINE_CONTINUE);
ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL);
}
if ((error = zio_checksum_error(zio)) != 0) {
zio->io_error = error;
if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
zfs_ereport_post(FM_EREPORT_ZFS_CHECKSUM,
zio->io_spa, zio->io_vd, zio, 0, 0);
}
}
return (ZIO_PIPELINE_CONTINUE);
}
/*
* Called by RAID-Z to ensure we don't compute the checksum twice.
*/
void
zio_checksum_verified(zio_t *zio)
{
zio->io_pipeline &= ~(1U << ZIO_STAGE_CHECKSUM_VERIFY);
}
/*
* ==========================================================================
* Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other.
* An error of 0 indictes success. ENXIO indicates whole-device failure,
* which may be transient (e.g. unplugged) or permament. ECKSUM and EIO
* indicate errors that are specific to one I/O, and most likely permanent.
* Any other error is presumed to be worse because we weren't expecting it.
* ==========================================================================
*/
int
zio_worst_error(int e1, int e2)
{
static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO };
int r1, r2;
for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++)
if (e1 == zio_error_rank[r1])
break;
for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++)
if (e2 == zio_error_rank[r2])
break;
return (r1 > r2 ? e1 : e2);
}
/*
* ==========================================================================
* I/O completion
* ==========================================================================
*/
static int
zio_ready(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
zio_t *pio = zio->io_parent;
if (zio->io_ready) {
if (BP_IS_GANG(bp) &&
zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY))
return (ZIO_PIPELINE_STOP);
ASSERT(IO_IS_ALLOCATING(zio));
ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp));
ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0);
zio->io_ready(zio);
}
if (bp != NULL && bp != &zio->io_bp_copy)
zio->io_bp_copy = *bp;
if (zio->io_error)
zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
if (pio != NULL)
zio_notify_parent(pio, zio, ZIO_WAIT_READY);
return (ZIO_PIPELINE_CONTINUE);
}
static int
zio_done(zio_t *zio)
{
spa_t *spa = zio->io_spa;
zio_t *pio = zio->io_parent;
zio_t *lio = zio->io_logical;
blkptr_t *bp = zio->io_bp;
vdev_t *vd = zio->io_vd;
uint64_t psize = zio->io_size;
/*
* If our of children haven't all completed,
* wait for them and then repeat this pipeline stage.
*/
if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) ||
zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) ||
zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE))
return (ZIO_PIPELINE_STOP);
for (int c = 0; c < ZIO_CHILD_TYPES; c++)
for (int w = 0; w < ZIO_WAIT_TYPES; w++)
ASSERT(zio->io_children[c][w] == 0);
if (bp != NULL) {
ASSERT(bp->blk_pad[0] == 0);
ASSERT(bp->blk_pad[1] == 0);
ASSERT(bp->blk_pad[2] == 0);
ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 ||
(pio != NULL && bp == pio->io_bp));
if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(bp) &&
!(zio->io_flags & ZIO_FLAG_IO_REPAIR)) {
ASSERT(!BP_SHOULD_BYTESWAP(bp));
ASSERT3U(zio->io_prop.zp_ndvas, <=, BP_GET_NDVAS(bp));
ASSERT(BP_COUNT_GANG(bp) == 0 ||
(BP_COUNT_GANG(bp) == BP_GET_NDVAS(bp)));
}
}
/*
* If there were child vdev or gang errors, they apply to us now.
*/
zio_inherit_child_errors(zio, ZIO_CHILD_VDEV);
zio_inherit_child_errors(zio, ZIO_CHILD_GANG);
zio_pop_transforms(zio); /* note: may set zio->io_error */
vdev_stat_update(zio, psize);
if (zio->io_error) {
/*
* If this I/O is attached to a particular vdev,
* generate an error message describing the I/O failure
* at the block level. We ignore these errors if the
* device is currently unavailable.
*/
if (zio->io_error != ECKSUM && vd != NULL && !vdev_is_dead(vd))
zfs_ereport_post(FM_EREPORT_ZFS_IO, spa, vd, zio, 0, 0);
if ((zio->io_error == EIO ||
!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) && zio == lio) {
/*
* For logical I/O requests, tell the SPA to log the
* error and generate a logical data ereport.
*/
spa_log_error(spa, zio);
zfs_ereport_post(FM_EREPORT_ZFS_DATA, spa, NULL, zio,
0, 0);
}
}
if (zio->io_error && zio == lio) {
/*
* Determine whether zio should be reexecuted. This will
* propagate all the way to the root via zio_notify_parent().
*/
ASSERT(vd == NULL && bp != NULL);
if (IO_IS_ALLOCATING(zio))
if (zio->io_error != ENOSPC)
zio->io_reexecute |= ZIO_REEXECUTE_NOW;
else
zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
if ((zio->io_type == ZIO_TYPE_READ ||
zio->io_type == ZIO_TYPE_FREE) &&
zio->io_error == ENXIO &&
spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE)
zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute)
zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
}
/*
* If there were logical child errors, they apply to us now.
* We defer this until now to avoid conflating logical child
* errors with errors that happened to the zio itself when
* updating vdev stats and reporting FMA events above.
*/
zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL);
if (zio->io_reexecute) {
/*
* This is a logical I/O that wants to reexecute.
*
* Reexecute is top-down. When an i/o fails, if it's not
* the root, it simply notifies its parent and sticks around.
* The parent, seeing that it still has children in zio_done(),
* does the same. This percolates all the way up to the root.
* The root i/o will reexecute or suspend the entire tree.
*
* This approach ensures that zio_reexecute() honors
* all the original i/o dependency relationships, e.g.
* parents not executing until children are ready.
*/
ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
if (IO_IS_ALLOCATING(zio))
zio_dva_unallocate(zio, zio->io_gang_tree, bp);
zio_gang_tree_free(&zio->io_gang_tree);
if (pio != NULL) {
/*
* We're not a root i/o, so there's nothing to do
* but notify our parent. Don't propagate errors
* upward since we haven't permanently failed yet.
*/
zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE;
zio_notify_parent(pio, zio, ZIO_WAIT_DONE);
} else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) {
/*
* We'd fail again if we reexecuted now, so suspend
* until conditions improve (e.g. device comes online).
*/
zio_suspend(spa, zio);
} else {
/*
* Reexecution is potentially a huge amount of work.
* Hand it off to the otherwise-unused claim taskq.
*/
(void) taskq_dispatch_safe(
spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE],
(task_func_t *)zio_reexecute, zio, &zio->io_task);
}
return (ZIO_PIPELINE_STOP);
}
ASSERT(zio->io_child == NULL);
ASSERT(zio->io_reexecute == 0);
ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL));
if (zio->io_done)
zio->io_done(zio);
zio_gang_tree_free(&zio->io_gang_tree);
ASSERT(zio->io_delegate_list == NULL);
ASSERT(zio->io_delegate_next == NULL);
if (pio != NULL) {
zio_remove_child(pio, zio);
zio_notify_parent(pio, zio, ZIO_WAIT_DONE);
}
if (zio->io_waiter != NULL) {
mutex_enter(&zio->io_lock);
zio->io_executor = NULL;
cv_broadcast(&zio->io_cv);
mutex_exit(&zio->io_lock);
} else {
zio_destroy(zio);
}
return (ZIO_PIPELINE_STOP);
}
/*
* ==========================================================================
* I/O pipeline definition
* ==========================================================================
*/
static zio_pipe_stage_t *zio_pipeline[ZIO_STAGES] = {
NULL,
zio_issue_async,
zio_read_bp_init,
zio_write_bp_init,
zio_checksum_generate,
zio_gang_assemble,
zio_gang_issue,
zio_dva_allocate,
zio_dva_free,
zio_dva_claim,
zio_ready,
zio_vdev_io_start,
zio_vdev_io_done,
zio_vdev_io_assess,
zio_checksum_verify,
zio_done
};
Index: stable/8/sys/cddl/contrib/opensolaris
===================================================================
--- stable/8/sys/cddl/contrib/opensolaris (revision 209264)
+++ stable/8/sys/cddl/contrib/opensolaris (revision 209265)
Property changes on: stable/8/sys/cddl/contrib/opensolaris
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/cddl/contrib/opensolaris:r209260-209261
Index: stable/8/sys/contrib/dev/acpica
===================================================================
--- stable/8/sys/contrib/dev/acpica (revision 209264)
+++ stable/8/sys/contrib/dev/acpica (revision 209265)
Property changes on: stable/8/sys/contrib/dev/acpica
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/contrib/dev/acpica:r209260-209261
Index: stable/8/sys/contrib/pf
===================================================================
--- stable/8/sys/contrib/pf (revision 209264)
+++ stable/8/sys/contrib/pf (revision 209265)
Property changes on: stable/8/sys/contrib/pf
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/contrib/pf:r209260-209261
Index: stable/8/sys/dev/xen/xenpci
===================================================================
--- stable/8/sys/dev/xen/xenpci (revision 209264)
+++ stable/8/sys/dev/xen/xenpci (revision 209265)
Property changes on: stable/8/sys/dev/xen/xenpci
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/dev/xen/xenpci:r209260-209261
Index: stable/8/sys/geom/sched
===================================================================
--- stable/8/sys/geom/sched (revision 209264)
+++ stable/8/sys/geom/sched (revision 209265)
Property changes on: stable/8/sys/geom/sched
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys/geom/sched:r209260-209261
Index: stable/8/sys/kern/vfs_subr.c
===================================================================
--- stable/8/sys/kern/vfs_subr.c (revision 209264)
+++ stable/8/sys/kern/vfs_subr.c (revision 209265)
@@ -1,4359 +1,4346 @@
/*-
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
* (c) UNIX System Laboratories, Inc.
* All or some portions of this file are derived from material licensed
* to the University of California by American Telephone and Telegraph
* Co. or Unix System Laboratories, Inc. and are reproduced herein with
* the permission of UNIX System Laboratories, Inc.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
*/
/*
* External virtual filesystem routines
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_ddb.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bio.h>
#include <sys/buf.h>
#include <sys/condvar.h>
#include <sys/conf.h>
#include <sys/dirent.h>
#include <sys/event.h>
#include <sys/eventhandler.h>
#include <sys/extattr.h>
#include <sys/file.h>
#include <sys/fcntl.h>
#include <sys/jail.h>
#include <sys/kdb.h>
#include <sys/kernel.h>
#include <sys/kthread.h>
#include <sys/lockf.h>
#include <sys/malloc.h>
#include <sys/mount.h>
#include <sys/namei.h>
#include <sys/priv.h>
#include <sys/reboot.h>
#include <sys/sleepqueue.h>
#include <sys/stat.h>
#include <sys/sysctl.h>
#include <sys/syslog.h>
#include <sys/vmmeter.h>
#include <sys/vnode.h>
#include <machine/stdarg.h>
#include <security/mac/mac_framework.h>
#include <vm/vm.h>
#include <vm/vm_object.h>
#include <vm/vm_extern.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_page.h>
#include <vm/vm_kern.h>
#include <vm/uma.h>
#ifdef DDB
#include <ddb/ddb.h>
#endif
#define WI_MPSAFEQ 0
#define WI_GIANTQ 1
static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
static void delmntque(struct vnode *vp);
static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
int slpflag, int slptimeo);
static void syncer_shutdown(void *arg, int howto);
static int vtryrecycle(struct vnode *vp);
static void vbusy(struct vnode *vp);
static void vinactive(struct vnode *, struct thread *);
static void v_incr_usecount(struct vnode *);
static void v_decr_usecount(struct vnode *);
static void v_decr_useonly(struct vnode *);
static void v_upgrade_usecount(struct vnode *);
static void vfree(struct vnode *);
static void vnlru_free(int);
static void vgonel(struct vnode *);
static void vfs_knllock(void *arg);
static void vfs_knlunlock(void *arg);
static void vfs_knl_assert_locked(void *arg);
static void vfs_knl_assert_unlocked(void *arg);
static void destroy_vpollinfo(struct vpollinfo *vi);
/*
* Number of vnodes in existence. Increased whenever getnewvnode()
* allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
* vnode.
*/
static unsigned long numvnodes;
SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
/*
* Conversion tables for conversion from vnode types to inode formats
* and back.
*/
enum vtype iftovt_tab[16] = {
VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
};
int vttoif_tab[10] = {
0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
};
/*
* List of vnodes that are ready for recycling.
*/
static TAILQ_HEAD(freelst, vnode) vnode_free_list;
/*
* Free vnode target. Free vnodes may simply be files which have been stat'd
* but not read. This is somewhat common, and a small cache of such files
* should be kept to avoid recreation costs.
*/
static u_long wantfreevnodes;
SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
/* Number of vnodes in the free list. */
static u_long freevnodes;
SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
static int vlru_allow_cache_src;
SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
&vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
/*
* Various variables used for debugging the new implementation of
* reassignbuf().
* XXX these are probably of (very) limited utility now.
*/
static int reassignbufcalls;
SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
/*
* Cache for the mount type id assigned to NFS. This is used for
* special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
*/
int nfs_mount_type = -1;
/* To keep more than one thread at a time from running vfs_getnewfsid */
static struct mtx mntid_mtx;
/*
* Lock for any access to the following:
* vnode_free_list
* numvnodes
* freevnodes
*/
static struct mtx vnode_free_list_mtx;
/* Publicly exported FS */
struct nfs_public nfs_pub;
/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
static uma_zone_t vnode_zone;
static uma_zone_t vnodepoll_zone;
/* Set to 1 to print out reclaim of active vnodes */
int prtactive;
/*
* The workitem queue.
*
* It is useful to delay writes of file data and filesystem metadata
* for tens of seconds so that quickly created and deleted files need
* not waste disk bandwidth being created and removed. To realize this,
* we append vnodes to a "workitem" queue. When running with a soft
* updates implementation, most pending metadata dependencies should
* not wait for more than a few seconds. Thus, mounted on block devices
* are delayed only about a half the time that file data is delayed.
* Similarly, directory updates are more critical, so are only delayed
* about a third the time that file data is delayed. Thus, there are
* SYNCER_MAXDELAY queues that are processed round-robin at a rate of
* one each second (driven off the filesystem syncer process). The
* syncer_delayno variable indicates the next queue that is to be processed.
* Items that need to be processed soon are placed in this queue:
*
* syncer_workitem_pending[syncer_delayno]
*
* A delay of fifteen seconds is done by placing the request fifteen
* entries later in the queue:
*
* syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
*
*/
static int syncer_delayno;
static long syncer_mask;
LIST_HEAD(synclist, bufobj);
static struct synclist *syncer_workitem_pending[2];
/*
* The sync_mtx protects:
* bo->bo_synclist
* sync_vnode_count
* syncer_delayno
* syncer_state
* syncer_workitem_pending
* syncer_worklist_len
* rushjob
*/
static struct mtx sync_mtx;
static struct cv sync_wakeup;
#define SYNCER_MAXDELAY 32
static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
static int syncdelay = 30; /* max time to delay syncing data */
static int filedelay = 30; /* time to delay syncing files */
SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
static int dirdelay = 29; /* time to delay syncing directories */
SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
static int metadelay = 28; /* time to delay syncing metadata */
SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
static int rushjob; /* number of slots to run ASAP */
static int stat_rush_requests; /* number of times I/O speeded up */
SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
/*
* When shutting down the syncer, run it at four times normal speed.
*/
#define SYNCER_SHUTDOWN_SPEEDUP 4
static int sync_vnode_count;
static int syncer_worklist_len;
static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
syncer_state;
/*
* Number of vnodes we want to exist at any one time. This is mostly used
* to size hash tables in vnode-related code. It is normally not used in
* getnewvnode(), as wantfreevnodes is normally nonzero.)
*
* XXX desiredvnodes is historical cruft and should not exist.
*/
int desiredvnodes;
SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
&desiredvnodes, 0, "Maximum number of vnodes");
SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
&wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
static int vnlru_nowhere;
SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
&vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
/*
* Macros to control when a vnode is freed and recycled. All require
* the vnode interlock.
*/
#define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
#define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
#define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
/*
* Initialize the vnode management data structures.
*/
#ifndef MAXVNODES_MAX
#define MAXVNODES_MAX 100000
#endif
static void
vntblinit(void *dummy __unused)
{
/*
* Desiredvnodes is a function of the physical memory size and
* the kernel's heap size. Specifically, desiredvnodes scales
* in proportion to the physical memory size until two fifths
* of the kernel's heap size is consumed by vnodes and vm
* objects.
*/
desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
(5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
if (desiredvnodes > MAXVNODES_MAX) {
if (bootverbose)
printf("Reducing kern.maxvnodes %d -> %d\n",
desiredvnodes, MAXVNODES_MAX);
desiredvnodes = MAXVNODES_MAX;
}
wantfreevnodes = desiredvnodes / 4;
mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
TAILQ_INIT(&vnode_free_list);
mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
NULL, NULL, UMA_ALIGN_PTR, 0);
vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
/*
* Initialize the filesystem syncer.
*/
syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
&syncer_mask);
syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
&syncer_mask);
syncer_maxdelay = syncer_mask + 1;
mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
cv_init(&sync_wakeup, "syncer");
}
SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
/*
* Mark a mount point as busy. Used to synchronize access and to delay
* unmounting. Eventually, mountlist_mtx is not released on failure.
*/
int
vfs_busy(struct mount *mp, int flags)
{
MPASS((flags & ~MBF_MASK) == 0);
CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
MNT_ILOCK(mp);
MNT_REF(mp);
/*
* If mount point is currenly being unmounted, sleep until the
* mount point fate is decided. If thread doing the unmounting fails,
* it will clear MNTK_UNMOUNT flag before waking us up, indicating
* that this mount point has survived the unmount attempt and vfs_busy
* should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
* flag in addition to MNTK_UNMOUNT, indicating that mount point is
* about to be really destroyed. vfs_busy needs to release its
* reference on the mount point in this case and return with ENOENT,
* telling the caller that mount mount it tried to busy is no longer
* valid.
*/
while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
MNT_REL(mp);
MNT_IUNLOCK(mp);
CTR1(KTR_VFS, "%s: failed busying before sleeping",
__func__);
return (ENOENT);
}
if (flags & MBF_MNTLSTLOCK)
mtx_unlock(&mountlist_mtx);
mp->mnt_kern_flag |= MNTK_MWAIT;
msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
if (flags & MBF_MNTLSTLOCK)
mtx_lock(&mountlist_mtx);
}
if (flags & MBF_MNTLSTLOCK)
mtx_unlock(&mountlist_mtx);
mp->mnt_lockref++;
MNT_IUNLOCK(mp);
return (0);
}
/*
* Free a busy filesystem.
*/
void
vfs_unbusy(struct mount *mp)
{
CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
MNT_ILOCK(mp);
MNT_REL(mp);
KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
mp->mnt_lockref--;
if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
CTR1(KTR_VFS, "%s: waking up waiters", __func__);
mp->mnt_kern_flag &= ~MNTK_DRAINING;
wakeup(&mp->mnt_lockref);
}
MNT_IUNLOCK(mp);
}
/*
* Lookup a mount point by filesystem identifier.
*/
struct mount *
vfs_getvfs(fsid_t *fsid)
{
struct mount *mp;
CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
mtx_lock(&mountlist_mtx);
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
vfs_ref(mp);
mtx_unlock(&mountlist_mtx);
return (mp);
}
}
mtx_unlock(&mountlist_mtx);
CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
return ((struct mount *) 0);
}
/*
* Lookup a mount point by filesystem identifier, busying it before
* returning.
*/
struct mount *
vfs_busyfs(fsid_t *fsid)
{
struct mount *mp;
int error;
CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
mtx_lock(&mountlist_mtx);
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
error = vfs_busy(mp, MBF_MNTLSTLOCK);
if (error) {
mtx_unlock(&mountlist_mtx);
return (NULL);
}
return (mp);
}
}
CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
mtx_unlock(&mountlist_mtx);
return ((struct mount *) 0);
}
/*
* Check if a user can access privileged mount options.
*/
int
vfs_suser(struct mount *mp, struct thread *td)
{
int error;
/*
* If the thread is jailed, but this is not a jail-friendly file
* system, deny immediately.
*/
if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
return (EPERM);
/*
* If the file system was mounted outside the jail of the calling
* thread, deny immediately.
*/
if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
return (EPERM);
/*
* If file system supports delegated administration, we don't check
* for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
* by the file system itself.
* If this is not the user that did original mount, we check for
* the PRIV_VFS_MOUNT_OWNER privilege.
*/
if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
return (error);
}
return (0);
}
/*
* Get a new unique fsid. Try to make its val[0] unique, since this value
* will be used to create fake device numbers for stat(). Also try (but
* not so hard) make its val[0] unique mod 2^16, since some emulators only
* support 16-bit device numbers. We end up with unique val[0]'s for the
* first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
*
* Keep in mind that several mounts may be running in parallel. Starting
* the search one past where the previous search terminated is both a
* micro-optimization and a defense against returning the same fsid to
* different mounts.
*/
void
vfs_getnewfsid(struct mount *mp)
{
static u_int16_t mntid_base;
struct mount *nmp;
fsid_t tfsid;
int mtype;
CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
mtx_lock(&mntid_mtx);
mtype = mp->mnt_vfc->vfc_typenum;
tfsid.val[1] = mtype;
mtype = (mtype & 0xFF) << 24;
for (;;) {
tfsid.val[0] = makedev(255,
mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
mntid_base++;
if ((nmp = vfs_getvfs(&tfsid)) == NULL)
break;
vfs_rel(nmp);
}
mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
mtx_unlock(&mntid_mtx);
}
/*
* Knob to control the precision of file timestamps:
*
* 0 = seconds only; nanoseconds zeroed.
* 1 = seconds and nanoseconds, accurate within 1/HZ.
* 2 = seconds and nanoseconds, truncated to microseconds.
* >=3 = seconds and nanoseconds, maximum precision.
*/
enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
static int timestamp_precision = TSP_SEC;
SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
&timestamp_precision, 0, "");
/*
* Get a current timestamp.
*/
void
vfs_timestamp(struct timespec *tsp)
{
struct timeval tv;
switch (timestamp_precision) {
case TSP_SEC:
tsp->tv_sec = time_second;
tsp->tv_nsec = 0;
break;
case TSP_HZ:
getnanotime(tsp);
break;
case TSP_USEC:
microtime(&tv);
TIMEVAL_TO_TIMESPEC(&tv, tsp);
break;
case TSP_NSEC:
default:
nanotime(tsp);
break;
}
}
/*
* Set vnode attributes to VNOVAL
*/
void
vattr_null(struct vattr *vap)
{
vap->va_type = VNON;
vap->va_size = VNOVAL;
vap->va_bytes = VNOVAL;
vap->va_mode = VNOVAL;
vap->va_nlink = VNOVAL;
vap->va_uid = VNOVAL;
vap->va_gid = VNOVAL;
vap->va_fsid = VNOVAL;
vap->va_fileid = VNOVAL;
vap->va_blocksize = VNOVAL;
vap->va_rdev = VNOVAL;
vap->va_atime.tv_sec = VNOVAL;
vap->va_atime.tv_nsec = VNOVAL;
vap->va_mtime.tv_sec = VNOVAL;
vap->va_mtime.tv_nsec = VNOVAL;
vap->va_ctime.tv_sec = VNOVAL;
vap->va_ctime.tv_nsec = VNOVAL;
vap->va_birthtime.tv_sec = VNOVAL;
vap->va_birthtime.tv_nsec = VNOVAL;
vap->va_flags = VNOVAL;
vap->va_gen = VNOVAL;
vap->va_vaflags = 0;
}
/*
* This routine is called when we have too many vnodes. It attempts
* to free <count> vnodes and will potentially free vnodes that still
* have VM backing store (VM backing store is typically the cause
* of a vnode blowout so we want to do this). Therefore, this operation
* is not considered cheap.
*
* A number of conditions may prevent a vnode from being reclaimed.
* the buffer cache may have references on the vnode, a directory
* vnode may still have references due to the namei cache representing
* underlying files, or the vnode may be in active use. It is not
* desireable to reuse such vnodes. These conditions may cause the
* number of vnodes to reach some minimum value regardless of what
* you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
*/
static int
vlrureclaim(struct mount *mp)
{
struct vnode *vp;
int done;
int trigger;
int usevnodes;
int count;
/*
* Calculate the trigger point, don't allow user
* screwups to blow us up. This prevents us from
* recycling vnodes with lots of resident pages. We
* aren't trying to free memory, we are trying to
* free vnodes.
*/
usevnodes = desiredvnodes;
if (usevnodes <= 0)
usevnodes = 1;
trigger = cnt.v_page_count * 2 / usevnodes;
done = 0;
vn_start_write(NULL, &mp, V_WAIT);
MNT_ILOCK(mp);
count = mp->mnt_nvnodelistsize / 10 + 1;
while (count != 0) {
vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
while (vp != NULL && vp->v_type == VMARKER)
vp = TAILQ_NEXT(vp, v_nmntvnodes);
if (vp == NULL)
break;
TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
--count;
if (!VI_TRYLOCK(vp))
goto next_iter;
/*
* If it's been deconstructed already, it's still
* referenced, or it exceeds the trigger, skip it.
*/
if (vp->v_usecount ||
(!vlru_allow_cache_src &&
!LIST_EMPTY(&(vp)->v_cache_src)) ||
(vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
vp->v_object->resident_page_count > trigger)) {
VI_UNLOCK(vp);
goto next_iter;
}
MNT_IUNLOCK(mp);
vholdl(vp);
if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
vdrop(vp);
goto next_iter_mntunlocked;
}
VI_LOCK(vp);
/*
* v_usecount may have been bumped after VOP_LOCK() dropped
* the vnode interlock and before it was locked again.
*
* It is not necessary to recheck VI_DOOMED because it can
* only be set by another thread that holds both the vnode
* lock and vnode interlock. If another thread has the
* vnode lock before we get to VOP_LOCK() and obtains the
* vnode interlock after VOP_LOCK() drops the vnode
* interlock, the other thread will be unable to drop the
* vnode lock before our VOP_LOCK() call fails.
*/
if (vp->v_usecount ||
(!vlru_allow_cache_src &&
!LIST_EMPTY(&(vp)->v_cache_src)) ||
(vp->v_object != NULL &&
vp->v_object->resident_page_count > trigger)) {
VOP_UNLOCK(vp, LK_INTERLOCK);
goto next_iter_mntunlocked;
}
KASSERT((vp->v_iflag & VI_DOOMED) == 0,
("VI_DOOMED unexpectedly detected in vlrureclaim()"));
vgonel(vp);
VOP_UNLOCK(vp, 0);
vdropl(vp);
done++;
next_iter_mntunlocked:
if ((count % 256) != 0)
goto relock_mnt;
goto yield;
next_iter:
if ((count % 256) != 0)
continue;
MNT_IUNLOCK(mp);
yield:
uio_yield();
relock_mnt:
MNT_ILOCK(mp);
}
MNT_IUNLOCK(mp);
vn_finished_write(mp);
return done;
}
/*
* Attempt to keep the free list at wantfreevnodes length.
*/
static void
vnlru_free(int count)
{
struct vnode *vp;
int vfslocked;
mtx_assert(&vnode_free_list_mtx, MA_OWNED);
for (; count > 0; count--) {
vp = TAILQ_FIRST(&vnode_free_list);
/*
* The list can be modified while the free_list_mtx
* has been dropped and vp could be NULL here.
*/
if (!vp)
break;
VNASSERT(vp->v_op != NULL, vp,
("vnlru_free: vnode already reclaimed."));
TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
/*
* Don't recycle if we can't get the interlock.
*/
if (!VI_TRYLOCK(vp)) {
TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
continue;
}
VNASSERT(VCANRECYCLE(vp), vp,
("vp inconsistent on freelist"));
freevnodes--;
vp->v_iflag &= ~VI_FREE;
vholdl(vp);
mtx_unlock(&vnode_free_list_mtx);
VI_UNLOCK(vp);
vfslocked = VFS_LOCK_GIANT(vp->v_mount);
vtryrecycle(vp);
VFS_UNLOCK_GIANT(vfslocked);
/*
* If the recycled succeeded this vdrop will actually free
* the vnode. If not it will simply place it back on
* the free list.
*/
vdrop(vp);
mtx_lock(&vnode_free_list_mtx);
}
}
/*
* Attempt to recycle vnodes in a context that is always safe to block.
* Calling vlrurecycle() from the bowels of filesystem code has some
* interesting deadlock problems.
*/
static struct proc *vnlruproc;
static int vnlruproc_sig;
static void
vnlru_proc(void)
{
struct mount *mp, *nmp;
int done, vfslocked;
struct proc *p = vnlruproc;
EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
SHUTDOWN_PRI_FIRST);
for (;;) {
kproc_suspend_check(p);
mtx_lock(&vnode_free_list_mtx);
if (freevnodes > wantfreevnodes)
vnlru_free(freevnodes - wantfreevnodes);
if (numvnodes <= desiredvnodes * 9 / 10) {
vnlruproc_sig = 0;
wakeup(&vnlruproc_sig);
msleep(vnlruproc, &vnode_free_list_mtx,
PVFS|PDROP, "vlruwt", hz);
continue;
}
mtx_unlock(&vnode_free_list_mtx);
done = 0;
mtx_lock(&mountlist_mtx);
for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
nmp = TAILQ_NEXT(mp, mnt_list);
continue;
}
vfslocked = VFS_LOCK_GIANT(mp);
done += vlrureclaim(mp);
VFS_UNLOCK_GIANT(vfslocked);
mtx_lock(&mountlist_mtx);
nmp = TAILQ_NEXT(mp, mnt_list);
vfs_unbusy(mp);
}
mtx_unlock(&mountlist_mtx);
if (done == 0) {
#if 0
/* These messages are temporary debugging aids */
if (vnlru_nowhere < 5)
printf("vnlru process getting nowhere..\n");
else if (vnlru_nowhere == 5)
printf("vnlru process messages stopped.\n");
#endif
vnlru_nowhere++;
tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
} else
uio_yield();
}
}
static struct kproc_desc vnlru_kp = {
"vnlru",
vnlru_proc,
&vnlruproc
};
SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
&vnlru_kp);
-static void
-vfs_lowmem(void *arg __unused)
-{
-
- /*
- * On low memory condition free 1/8th of the free vnodes.
- */
- mtx_lock(&vnode_free_list_mtx);
- vnlru_free(freevnodes / 8);
- mtx_unlock(&vnode_free_list_mtx);
-}
-EVENTHANDLER_DEFINE(vm_lowmem, vfs_lowmem, NULL, 0);
-
/*
* Routines having to do with the management of the vnode table.
*/
void
vdestroy(struct vnode *vp)
{
struct bufobj *bo;
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
mtx_lock(&vnode_free_list_mtx);
numvnodes--;
mtx_unlock(&vnode_free_list_mtx);
bo = &vp->v_bufobj;
VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
("cleaned vnode still on the free list."));
VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
VI_UNLOCK(vp);
#ifdef MAC
mac_vnode_destroy(vp);
#endif
if (vp->v_pollinfo != NULL)
destroy_vpollinfo(vp->v_pollinfo);
#ifdef INVARIANTS
/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
vp->v_op = NULL;
#endif
lockdestroy(vp->v_vnlock);
mtx_destroy(&vp->v_interlock);
mtx_destroy(BO_MTX(bo));
uma_zfree(vnode_zone, vp);
}
/*
* Try to recycle a freed vnode. We abort if anyone picks up a reference
* before we actually vgone(). This function must be called with the vnode
* held to prevent the vnode from being returned to the free list midway
* through vgone().
*/
static int
vtryrecycle(struct vnode *vp)
{
struct mount *vnmp;
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
VNASSERT(vp->v_holdcnt, vp,
("vtryrecycle: Recycling vp %p without a reference.", vp));
/*
* This vnode may found and locked via some other list, if so we
* can't recycle it yet.
*/
if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
CTR2(KTR_VFS,
"%s: impossible to recycle, vp %p lock is already held",
__func__, vp);
return (EWOULDBLOCK);
}
/*
* Don't recycle if its filesystem is being suspended.
*/
if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
VOP_UNLOCK(vp, 0);
CTR2(KTR_VFS,
"%s: impossible to recycle, cannot start the write for %p",
__func__, vp);
return (EBUSY);
}
/*
* If we got this far, we need to acquire the interlock and see if
* anyone picked up this vnode from another list. If not, we will
* mark it with DOOMED via vgonel() so that anyone who does find it
* will skip over it.
*/
VI_LOCK(vp);
if (vp->v_usecount) {
VOP_UNLOCK(vp, LK_INTERLOCK);
vn_finished_write(vnmp);
CTR2(KTR_VFS,
"%s: impossible to recycle, %p is already referenced",
__func__, vp);
return (EBUSY);
}
if ((vp->v_iflag & VI_DOOMED) == 0)
vgonel(vp);
VOP_UNLOCK(vp, LK_INTERLOCK);
vn_finished_write(vnmp);
return (0);
}
/*
* Return the next vnode from the free list.
*/
int
getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
struct vnode **vpp)
{
struct vnode *vp = NULL;
struct bufobj *bo;
CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
mtx_lock(&vnode_free_list_mtx);
/*
* Lend our context to reclaim vnodes if they've exceeded the max.
*/
if (freevnodes > wantfreevnodes)
vnlru_free(1);
/*
* Wait for available vnodes.
*/
if (numvnodes > desiredvnodes) {
if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
/*
* File system is beeing suspended, we cannot risk a
* deadlock here, so allocate new vnode anyway.
*/
if (freevnodes > wantfreevnodes)
vnlru_free(freevnodes - wantfreevnodes);
goto alloc;
}
if (vnlruproc_sig == 0) {
vnlruproc_sig = 1; /* avoid unnecessary wakeups */
wakeup(vnlruproc);
}
msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
"vlruwk", hz);
#if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
if (numvnodes > desiredvnodes) {
mtx_unlock(&vnode_free_list_mtx);
return (ENFILE);
}
#endif
}
alloc:
numvnodes++;
mtx_unlock(&vnode_free_list_mtx);
vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
/*
* Setup locks.
*/
vp->v_vnlock = &vp->v_lock;
mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
/*
* By default, don't allow shared locks unless filesystems
* opt-in.
*/
lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
/*
* Initialize bufobj.
*/
bo = &vp->v_bufobj;
bo->__bo_vnode = vp;
mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
bo->bo_ops = &buf_ops_bio;
bo->bo_private = vp;
TAILQ_INIT(&bo->bo_clean.bv_hd);
TAILQ_INIT(&bo->bo_dirty.bv_hd);
/*
* Initialize namecache.
*/
LIST_INIT(&vp->v_cache_src);
TAILQ_INIT(&vp->v_cache_dst);
/*
* Finalize various vnode identity bits.
*/
vp->v_type = VNON;
vp->v_tag = tag;
vp->v_op = vops;
v_incr_usecount(vp);
vp->v_data = 0;
#ifdef MAC
mac_vnode_init(vp);
if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
mac_vnode_associate_singlelabel(mp, vp);
else if (mp == NULL && vops != &dead_vnodeops)
printf("NULL mp in getnewvnode()\n");
#endif
if (mp != NULL) {
bo->bo_bsize = mp->mnt_stat.f_iosize;
if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
vp->v_vflag |= VV_NOKNOTE;
}
*vpp = vp;
return (0);
}
/*
* Delete from old mount point vnode list, if on one.
*/
static void
delmntque(struct vnode *vp)
{
struct mount *mp;
mp = vp->v_mount;
if (mp == NULL)
return;
MNT_ILOCK(mp);
vp->v_mount = NULL;
VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
("bad mount point vnode list size"));
TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
mp->mnt_nvnodelistsize--;
MNT_REL(mp);
MNT_IUNLOCK(mp);
}
static void
insmntque_stddtr(struct vnode *vp, void *dtr_arg)
{
vp->v_data = NULL;
vp->v_op = &dead_vnodeops;
/* XXX non mp-safe fs may still call insmntque with vnode
unlocked */
if (!VOP_ISLOCKED(vp))
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vgone(vp);
vput(vp);
}
/*
* Insert into list of vnodes for the new mount point, if available.
*/
int
insmntque1(struct vnode *vp, struct mount *mp,
void (*dtr)(struct vnode *, void *), void *dtr_arg)
{
int locked;
KASSERT(vp->v_mount == NULL,
("insmntque: vnode already on per mount vnode list"));
VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
#ifdef DEBUG_VFS_LOCKS
if (!VFS_NEEDSGIANT(mp))
ASSERT_VOP_ELOCKED(vp,
"insmntque: mp-safe fs and non-locked vp");
#endif
MNT_ILOCK(mp);
if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
mp->mnt_nvnodelistsize == 0)) {
locked = VOP_ISLOCKED(vp);
if (!locked || (locked == LK_EXCLUSIVE &&
(vp->v_vflag & VV_FORCEINSMQ) == 0)) {
MNT_IUNLOCK(mp);
if (dtr != NULL)
dtr(vp, dtr_arg);
return (EBUSY);
}
}
vp->v_mount = mp;
MNT_REF(mp);
TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
("neg mount point vnode list size"));
mp->mnt_nvnodelistsize++;
MNT_IUNLOCK(mp);
return (0);
}
int
insmntque(struct vnode *vp, struct mount *mp)
{
return (insmntque1(vp, mp, insmntque_stddtr, NULL));
}
/*
* Flush out and invalidate all buffers associated with a bufobj
* Called with the underlying object locked.
*/
int
bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
{
int error;
BO_LOCK(bo);
if (flags & V_SAVE) {
error = bufobj_wwait(bo, slpflag, slptimeo);
if (error) {
BO_UNLOCK(bo);
return (error);
}
if (bo->bo_dirty.bv_cnt > 0) {
BO_UNLOCK(bo);
if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
return (error);
/*
* XXX We could save a lock/unlock if this was only
* enabled under INVARIANTS
*/
BO_LOCK(bo);
if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
panic("vinvalbuf: dirty bufs");
}
}
/*
* If you alter this loop please notice that interlock is dropped and
* reacquired in flushbuflist. Special care is needed to ensure that
* no race conditions occur from this.
*/
do {
error = flushbuflist(&bo->bo_clean,
flags, bo, slpflag, slptimeo);
if (error == 0)
error = flushbuflist(&bo->bo_dirty,
flags, bo, slpflag, slptimeo);
if (error != 0 && error != EAGAIN) {
BO_UNLOCK(bo);
return (error);
}
} while (error != 0);
/*
* Wait for I/O to complete. XXX needs cleaning up. The vnode can
* have write I/O in-progress but if there is a VM object then the
* VM object can also have read-I/O in-progress.
*/
do {
bufobj_wwait(bo, 0, 0);
BO_UNLOCK(bo);
if (bo->bo_object != NULL) {
VM_OBJECT_LOCK(bo->bo_object);
vm_object_pip_wait(bo->bo_object, "bovlbx");
VM_OBJECT_UNLOCK(bo->bo_object);
}
BO_LOCK(bo);
} while (bo->bo_numoutput > 0);
BO_UNLOCK(bo);
/*
* Destroy the copy in the VM cache, too.
*/
if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
VM_OBJECT_LOCK(bo->bo_object);
vm_object_page_remove(bo->bo_object, 0, 0,
(flags & V_SAVE) ? TRUE : FALSE);
VM_OBJECT_UNLOCK(bo->bo_object);
}
#ifdef INVARIANTS
BO_LOCK(bo);
if ((flags & (V_ALT | V_NORMAL)) == 0 &&
(bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
panic("vinvalbuf: flush failed");
BO_UNLOCK(bo);
#endif
return (0);
}
/*
* Flush out and invalidate all buffers associated with a vnode.
* Called with the underlying object locked.
*/
int
vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
{
CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
ASSERT_VOP_LOCKED(vp, "vinvalbuf");
return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
}
/*
* Flush out buffers on the specified list.
*
*/
static int
flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
int slptimeo)
{
struct buf *bp, *nbp;
int retval, error;
daddr_t lblkno;
b_xflags_t xflags;
ASSERT_BO_LOCKED(bo);
retval = 0;
TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
continue;
}
lblkno = 0;
xflags = 0;
if (nbp != NULL) {
lblkno = nbp->b_lblkno;
xflags = nbp->b_xflags &
(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
}
retval = EAGAIN;
error = BUF_TIMELOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
"flushbuf", slpflag, slptimeo);
if (error) {
BO_LOCK(bo);
return (error != ENOLCK ? error : EAGAIN);
}
KASSERT(bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p",
bp, bp->b_bufobj, bo));
if (bp->b_bufobj != bo) { /* XXX: necessary ? */
BUF_UNLOCK(bp);
BO_LOCK(bo);
return (EAGAIN);
}
/*
* XXX Since there are no node locks for NFS, I
* believe there is a slight chance that a delayed
* write will occur while sleeping just above, so
* check for it.
*/
if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
(flags & V_SAVE)) {
bremfree(bp);
bp->b_flags |= B_ASYNC;
bwrite(bp);
BO_LOCK(bo);
return (EAGAIN); /* XXX: why not loop ? */
}
bremfree(bp);
bp->b_flags |= (B_INVAL | B_RELBUF);
bp->b_flags &= ~B_ASYNC;
brelse(bp);
BO_LOCK(bo);
if (nbp != NULL &&
(nbp->b_bufobj != bo ||
nbp->b_lblkno != lblkno ||
(nbp->b_xflags &
(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
break; /* nbp invalid */
}
return (retval);
}
/*
* Truncate a file's buffer and pages to a specified length. This
* is in lieu of the old vinvalbuf mechanism, which performed unneeded
* sync activity.
*/
int
vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
off_t length, int blksize)
{
struct buf *bp, *nbp;
int anyfreed;
int trunclbn;
struct bufobj *bo;
CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
vp, cred, blksize, (uintmax_t)length);
/*
* Round up to the *next* lbn.
*/
trunclbn = (length + blksize - 1) / blksize;
ASSERT_VOP_LOCKED(vp, "vtruncbuf");
restart:
bo = &vp->v_bufobj;
BO_LOCK(bo);
anyfreed = 1;
for (;anyfreed;) {
anyfreed = 0;
TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
if (bp->b_lblkno < trunclbn)
continue;
if (BUF_LOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
BO_MTX(bo)) == ENOLCK)
goto restart;
bremfree(bp);
bp->b_flags |= (B_INVAL | B_RELBUF);
bp->b_flags &= ~B_ASYNC;
brelse(bp);
anyfreed = 1;
if (nbp != NULL &&
(((nbp->b_xflags & BX_VNCLEAN) == 0) ||
(nbp->b_vp != vp) ||
(nbp->b_flags & B_DELWRI))) {
goto restart;
}
BO_LOCK(bo);
}
TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
if (bp->b_lblkno < trunclbn)
continue;
if (BUF_LOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
BO_MTX(bo)) == ENOLCK)
goto restart;
bremfree(bp);
bp->b_flags |= (B_INVAL | B_RELBUF);
bp->b_flags &= ~B_ASYNC;
brelse(bp);
anyfreed = 1;
if (nbp != NULL &&
(((nbp->b_xflags & BX_VNDIRTY) == 0) ||
(nbp->b_vp != vp) ||
(nbp->b_flags & B_DELWRI) == 0)) {
goto restart;
}
BO_LOCK(bo);
}
}
if (length > 0) {
restartsync:
TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
if (bp->b_lblkno > 0)
continue;
/*
* Since we hold the vnode lock this should only
* fail if we're racing with the buf daemon.
*/
if (BUF_LOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
BO_MTX(bo)) == ENOLCK) {
goto restart;
}
VNASSERT((bp->b_flags & B_DELWRI), vp,
("buf(%p) on dirty queue without DELWRI", bp));
bremfree(bp);
bawrite(bp);
BO_LOCK(bo);
goto restartsync;
}
}
bufobj_wwait(bo, 0, 0);
BO_UNLOCK(bo);
vnode_pager_setsize(vp, length);
return (0);
}
/*
* buf_splay() - splay tree core for the clean/dirty list of buffers in
* a vnode.
*
* NOTE: We have to deal with the special case of a background bitmap
* buffer, a situation where two buffers will have the same logical
* block offset. We want (1) only the foreground buffer to be accessed
* in a lookup and (2) must differentiate between the foreground and
* background buffer in the splay tree algorithm because the splay
* tree cannot normally handle multiple entities with the same 'index'.
* We accomplish this by adding differentiating flags to the splay tree's
* numerical domain.
*/
static
struct buf *
buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
{
struct buf dummy;
struct buf *lefttreemax, *righttreemin, *y;
if (root == NULL)
return (NULL);
lefttreemax = righttreemin = &dummy;
for (;;) {
if (lblkno < root->b_lblkno ||
(lblkno == root->b_lblkno &&
(xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
if ((y = root->b_left) == NULL)
break;
if (lblkno < y->b_lblkno) {
/* Rotate right. */
root->b_left = y->b_right;
y->b_right = root;
root = y;
if ((y = root->b_left) == NULL)
break;
}
/* Link into the new root's right tree. */
righttreemin->b_left = root;
righttreemin = root;
} else if (lblkno > root->b_lblkno ||
(lblkno == root->b_lblkno &&
(xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
if ((y = root->b_right) == NULL)
break;
if (lblkno > y->b_lblkno) {
/* Rotate left. */
root->b_right = y->b_left;
y->b_left = root;
root = y;
if ((y = root->b_right) == NULL)
break;
}
/* Link into the new root's left tree. */
lefttreemax->b_right = root;
lefttreemax = root;
} else {
break;
}
root = y;
}
/* Assemble the new root. */
lefttreemax->b_right = root->b_left;
righttreemin->b_left = root->b_right;
root->b_left = dummy.b_right;
root->b_right = dummy.b_left;
return (root);
}
static void
buf_vlist_remove(struct buf *bp)
{
struct buf *root;
struct bufv *bv;
KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
ASSERT_BO_LOCKED(bp->b_bufobj);
KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
(BX_VNDIRTY|BX_VNCLEAN),
("buf_vlist_remove: Buf %p is on two lists", bp));
if (bp->b_xflags & BX_VNDIRTY)
bv = &bp->b_bufobj->bo_dirty;
else
bv = &bp->b_bufobj->bo_clean;
if (bp != bv->bv_root) {
root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
KASSERT(root == bp, ("splay lookup failed in remove"));
}
if (bp->b_left == NULL) {
root = bp->b_right;
} else {
root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
root->b_right = bp->b_right;
}
bv->bv_root = root;
TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
bv->bv_cnt--;
bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
}
/*
* Add the buffer to the sorted clean or dirty block list using a
* splay tree algorithm.
*
* NOTE: xflags is passed as a constant, optimizing this inline function!
*/
static void
buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
{
struct buf *root;
struct bufv *bv;
ASSERT_BO_LOCKED(bo);
KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
bp->b_xflags |= xflags;
if (xflags & BX_VNDIRTY)
bv = &bo->bo_dirty;
else
bv = &bo->bo_clean;
root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
if (root == NULL) {
bp->b_left = NULL;
bp->b_right = NULL;
TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
} else if (bp->b_lblkno < root->b_lblkno ||
(bp->b_lblkno == root->b_lblkno &&
(bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
bp->b_left = root->b_left;
bp->b_right = root;
root->b_left = NULL;
TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
} else {
bp->b_right = root->b_right;
bp->b_left = root;
root->b_right = NULL;
TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
}
bv->bv_cnt++;
bv->bv_root = bp;
}
/*
* Lookup a buffer using the splay tree. Note that we specifically avoid
* shadow buffers used in background bitmap writes.
*
* This code isn't quite efficient as it could be because we are maintaining
* two sorted lists and do not know which list the block resides in.
*
* During a "make buildworld" the desired buffer is found at one of
* the roots more than 60% of the time. Thus, checking both roots
* before performing either splay eliminates unnecessary splays on the
* first tree splayed.
*/
struct buf *
gbincore(struct bufobj *bo, daddr_t lblkno)
{
struct buf *bp;
ASSERT_BO_LOCKED(bo);
if ((bp = bo->bo_clean.bv_root) != NULL &&
bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
return (bp);
if ((bp = bo->bo_dirty.bv_root) != NULL &&
bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
return (bp);
if ((bp = bo->bo_clean.bv_root) != NULL) {
bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
return (bp);
}
if ((bp = bo->bo_dirty.bv_root) != NULL) {
bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
return (bp);
}
return (NULL);
}
/*
* Associate a buffer with a vnode.
*/
void
bgetvp(struct vnode *vp, struct buf *bp)
{
struct bufobj *bo;
bo = &vp->v_bufobj;
ASSERT_BO_LOCKED(bo);
VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
("bgetvp: bp already attached! %p", bp));
vhold(vp);
if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
bp->b_flags |= B_NEEDSGIANT;
bp->b_vp = vp;
bp->b_bufobj = bo;
/*
* Insert onto list for new vnode.
*/
buf_vlist_add(bp, bo, BX_VNCLEAN);
}
/*
* Disassociate a buffer from a vnode.
*/
void
brelvp(struct buf *bp)
{
struct bufobj *bo;
struct vnode *vp;
CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
/*
* Delete from old vnode list, if on one.
*/
vp = bp->b_vp; /* XXX */
bo = bp->b_bufobj;
BO_LOCK(bo);
if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
buf_vlist_remove(bp);
else
panic("brelvp: Buffer %p not on queue.", bp);
if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
bo->bo_flag &= ~BO_ONWORKLST;
mtx_lock(&sync_mtx);
LIST_REMOVE(bo, bo_synclist);
syncer_worklist_len--;
mtx_unlock(&sync_mtx);
}
bp->b_flags &= ~B_NEEDSGIANT;
bp->b_vp = NULL;
bp->b_bufobj = NULL;
BO_UNLOCK(bo);
vdrop(vp);
}
/*
* Add an item to the syncer work queue.
*/
static void
vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
{
int queue, slot;
ASSERT_BO_LOCKED(bo);
mtx_lock(&sync_mtx);
if (bo->bo_flag & BO_ONWORKLST)
LIST_REMOVE(bo, bo_synclist);
else {
bo->bo_flag |= BO_ONWORKLST;
syncer_worklist_len++;
}
if (delay > syncer_maxdelay - 2)
delay = syncer_maxdelay - 2;
slot = (syncer_delayno + delay) & syncer_mask;
queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
WI_MPSAFEQ;
LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
bo_synclist);
mtx_unlock(&sync_mtx);
}
static int
sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
{
int error, len;
mtx_lock(&sync_mtx);
len = syncer_worklist_len - sync_vnode_count;
mtx_unlock(&sync_mtx);
error = SYSCTL_OUT(req, &len, sizeof(len));
return (error);
}
SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
static struct proc *updateproc;
static void sched_sync(void);
static struct kproc_desc up_kp = {
"syncer",
sched_sync,
&updateproc
};
SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
static int
sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
{
struct vnode *vp;
struct mount *mp;
*bo = LIST_FIRST(slp);
if (*bo == NULL)
return (0);
vp = (*bo)->__bo_vnode; /* XXX */
if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
return (1);
/*
* We use vhold in case the vnode does not
* successfully sync. vhold prevents the vnode from
* going away when we unlock the sync_mtx so that
* we can acquire the vnode interlock.
*/
vholdl(vp);
mtx_unlock(&sync_mtx);
VI_UNLOCK(vp);
if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
vdrop(vp);
mtx_lock(&sync_mtx);
return (*bo == LIST_FIRST(slp));
}
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
(void) VOP_FSYNC(vp, MNT_LAZY, td);
VOP_UNLOCK(vp, 0);
vn_finished_write(mp);
BO_LOCK(*bo);
if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
/*
* Put us back on the worklist. The worklist
* routine will remove us from our current
* position and then add us back in at a later
* position.
*/
vn_syncer_add_to_worklist(*bo, syncdelay);
}
BO_UNLOCK(*bo);
vdrop(vp);
mtx_lock(&sync_mtx);
return (0);
}
/*
* System filesystem synchronizer daemon.
*/
static void
sched_sync(void)
{
struct synclist *gnext, *next;
struct synclist *gslp, *slp;
struct bufobj *bo;
long starttime;
struct thread *td = curthread;
int last_work_seen;
int net_worklist_len;
int syncer_final_iter;
int first_printf;
int error;
last_work_seen = 0;
syncer_final_iter = 0;
first_printf = 1;
syncer_state = SYNCER_RUNNING;
starttime = time_uptime;
td->td_pflags |= TDP_NORUNNINGBUF;
EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
SHUTDOWN_PRI_LAST);
mtx_lock(&sync_mtx);
for (;;) {
if (syncer_state == SYNCER_FINAL_DELAY &&
syncer_final_iter == 0) {
mtx_unlock(&sync_mtx);
kproc_suspend_check(td->td_proc);
mtx_lock(&sync_mtx);
}
net_worklist_len = syncer_worklist_len - sync_vnode_count;
if (syncer_state != SYNCER_RUNNING &&
starttime != time_uptime) {
if (first_printf) {
printf("\nSyncing disks, vnodes remaining...");
first_printf = 0;
}
printf("%d ", net_worklist_len);
}
starttime = time_uptime;
/*
* Push files whose dirty time has expired. Be careful
* of interrupt race on slp queue.
*
* Skip over empty worklist slots when shutting down.
*/
do {
slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
syncer_delayno += 1;
if (syncer_delayno == syncer_maxdelay)
syncer_delayno = 0;
next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
/*
* If the worklist has wrapped since the
* it was emptied of all but syncer vnodes,
* switch to the FINAL_DELAY state and run
* for one more second.
*/
if (syncer_state == SYNCER_SHUTTING_DOWN &&
net_worklist_len == 0 &&
last_work_seen == syncer_delayno) {
syncer_state = SYNCER_FINAL_DELAY;
syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
}
} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
LIST_EMPTY(gslp) && syncer_worklist_len > 0);
/*
* Keep track of the last time there was anything
* on the worklist other than syncer vnodes.
* Return to the SHUTTING_DOWN state if any
* new work appears.
*/
if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
last_work_seen = syncer_delayno;
if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
syncer_state = SYNCER_SHUTTING_DOWN;
while (!LIST_EMPTY(slp)) {
error = sync_vnode(slp, &bo, td);
if (error == 1) {
LIST_REMOVE(bo, bo_synclist);
LIST_INSERT_HEAD(next, bo, bo_synclist);
continue;
}
}
if (!LIST_EMPTY(gslp)) {
mtx_unlock(&sync_mtx);
mtx_lock(&Giant);
mtx_lock(&sync_mtx);
while (!LIST_EMPTY(gslp)) {
error = sync_vnode(gslp, &bo, td);
if (error == 1) {
LIST_REMOVE(bo, bo_synclist);
LIST_INSERT_HEAD(gnext, bo,
bo_synclist);
continue;
}
}
mtx_unlock(&Giant);
}
if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
syncer_final_iter--;
/*
* The variable rushjob allows the kernel to speed up the
* processing of the filesystem syncer process. A rushjob
* value of N tells the filesystem syncer to process the next
* N seconds worth of work on its queue ASAP. Currently rushjob
* is used by the soft update code to speed up the filesystem
* syncer process when the incore state is getting so far
* ahead of the disk that the kernel memory pool is being
* threatened with exhaustion.
*/
if (rushjob > 0) {
rushjob -= 1;
continue;
}
/*
* Just sleep for a short period of time between
* iterations when shutting down to allow some I/O
* to happen.
*
* If it has taken us less than a second to process the
* current work, then wait. Otherwise start right over
* again. We can still lose time if any single round
* takes more than two seconds, but it does not really
* matter as we are just trying to generally pace the
* filesystem activity.
*/
if (syncer_state != SYNCER_RUNNING)
cv_timedwait(&sync_wakeup, &sync_mtx,
hz / SYNCER_SHUTDOWN_SPEEDUP);
else if (time_uptime == starttime)
cv_timedwait(&sync_wakeup, &sync_mtx, hz);
}
}
/*
* Request the syncer daemon to speed up its work.
* We never push it to speed up more than half of its
* normal turn time, otherwise it could take over the cpu.
*/
int
speedup_syncer(void)
{
int ret = 0;
mtx_lock(&sync_mtx);
if (rushjob < syncdelay / 2) {
rushjob += 1;
stat_rush_requests += 1;
ret = 1;
}
mtx_unlock(&sync_mtx);
cv_broadcast(&sync_wakeup);
return (ret);
}
/*
* Tell the syncer to speed up its work and run though its work
* list several times, then tell it to shut down.
*/
static void
syncer_shutdown(void *arg, int howto)
{
if (howto & RB_NOSYNC)
return;
mtx_lock(&sync_mtx);
syncer_state = SYNCER_SHUTTING_DOWN;
rushjob = 0;
mtx_unlock(&sync_mtx);
cv_broadcast(&sync_wakeup);
kproc_shutdown(arg, howto);
}
/*
* Reassign a buffer from one vnode to another.
* Used to assign file specific control information
* (indirect blocks) to the vnode to which they belong.
*/
void
reassignbuf(struct buf *bp)
{
struct vnode *vp;
struct bufobj *bo;
int delay;
#ifdef INVARIANTS
struct bufv *bv;
#endif
vp = bp->b_vp;
bo = bp->b_bufobj;
++reassignbufcalls;
CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
bp, bp->b_vp, bp->b_flags);
/*
* B_PAGING flagged buffers cannot be reassigned because their vp
* is not fully linked in.
*/
if (bp->b_flags & B_PAGING)
panic("cannot reassign paging buffer");
/*
* Delete from old vnode list, if on one.
*/
BO_LOCK(bo);
if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
buf_vlist_remove(bp);
else
panic("reassignbuf: Buffer %p not on queue.", bp);
/*
* If dirty, put on list of dirty buffers; otherwise insert onto list
* of clean buffers.
*/
if (bp->b_flags & B_DELWRI) {
if ((bo->bo_flag & BO_ONWORKLST) == 0) {
switch (vp->v_type) {
case VDIR:
delay = dirdelay;
break;
case VCHR:
delay = metadelay;
break;
default:
delay = filedelay;
}
vn_syncer_add_to_worklist(bo, delay);
}
buf_vlist_add(bp, bo, BX_VNDIRTY);
} else {
buf_vlist_add(bp, bo, BX_VNCLEAN);
if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
mtx_lock(&sync_mtx);
LIST_REMOVE(bo, bo_synclist);
syncer_worklist_len--;
mtx_unlock(&sync_mtx);
bo->bo_flag &= ~BO_ONWORKLST;
}
}
#ifdef INVARIANTS
bv = &bo->bo_clean;
bp = TAILQ_FIRST(&bv->bv_hd);
KASSERT(bp == NULL || bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
bp = TAILQ_LAST(&bv->bv_hd, buflists);
KASSERT(bp == NULL || bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
bv = &bo->bo_dirty;
bp = TAILQ_FIRST(&bv->bv_hd);
KASSERT(bp == NULL || bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
bp = TAILQ_LAST(&bv->bv_hd, buflists);
KASSERT(bp == NULL || bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
#endif
BO_UNLOCK(bo);
}
/*
* Increment the use and hold counts on the vnode, taking care to reference
* the driver's usecount if this is a chardev. The vholdl() will remove
* the vnode from the free list if it is presently free. Requires the
* vnode interlock and returns with it held.
*/
static void
v_incr_usecount(struct vnode *vp)
{
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
vp->v_usecount++;
if (vp->v_type == VCHR && vp->v_rdev != NULL) {
dev_lock();
vp->v_rdev->si_usecount++;
dev_unlock();
}
vholdl(vp);
}
/*
* Turn a holdcnt into a use+holdcnt such that only one call to
* v_decr_usecount is needed.
*/
static void
v_upgrade_usecount(struct vnode *vp)
{
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
vp->v_usecount++;
if (vp->v_type == VCHR && vp->v_rdev != NULL) {
dev_lock();
vp->v_rdev->si_usecount++;
dev_unlock();
}
}
/*
* Decrement the vnode use and hold count along with the driver's usecount
* if this is a chardev. The vdropl() below releases the vnode interlock
* as it may free the vnode.
*/
static void
v_decr_usecount(struct vnode *vp)
{
ASSERT_VI_LOCKED(vp, __FUNCTION__);
VNASSERT(vp->v_usecount > 0, vp,
("v_decr_usecount: negative usecount"));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
vp->v_usecount--;
if (vp->v_type == VCHR && vp->v_rdev != NULL) {
dev_lock();
vp->v_rdev->si_usecount--;
dev_unlock();
}
vdropl(vp);
}
/*
* Decrement only the use count and driver use count. This is intended to
* be paired with a follow on vdropl() to release the remaining hold count.
* In this way we may vgone() a vnode with a 0 usecount without risk of
* having it end up on a free list because the hold count is kept above 0.
*/
static void
v_decr_useonly(struct vnode *vp)
{
ASSERT_VI_LOCKED(vp, __FUNCTION__);
VNASSERT(vp->v_usecount > 0, vp,
("v_decr_useonly: negative usecount"));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
vp->v_usecount--;
if (vp->v_type == VCHR && vp->v_rdev != NULL) {
dev_lock();
vp->v_rdev->si_usecount--;
dev_unlock();
}
}
/*
* Grab a particular vnode from the free list, increment its
* reference count and lock it. VI_DOOMED is set if the vnode
* is being destroyed. Only callers who specify LK_RETRY will
* see doomed vnodes. If inactive processing was delayed in
* vput try to do it here.
*/
int
vget(struct vnode *vp, int flags, struct thread *td)
{
int error;
error = 0;
VFS_ASSERT_GIANT(vp->v_mount);
VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
("vget: invalid lock operation"));
CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
if ((flags & LK_INTERLOCK) == 0)
VI_LOCK(vp);
vholdl(vp);
if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
vdrop(vp);
CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
vp);
return (error);
}
if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
panic("vget: vn_lock failed to return ENOENT\n");
VI_LOCK(vp);
/* Upgrade our holdcnt to a usecount. */
v_upgrade_usecount(vp);
/*
* We don't guarantee that any particular close will
* trigger inactive processing so just make a best effort
* here at preventing a reference to a removed file. If
* we don't succeed no harm is done.
*/
if (vp->v_iflag & VI_OWEINACT) {
if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
(flags & LK_NOWAIT) == 0)
vinactive(vp, td);
vp->v_iflag &= ~VI_OWEINACT;
}
VI_UNLOCK(vp);
return (0);
}
/*
* Increase the reference count of a vnode.
*/
void
vref(struct vnode *vp)
{
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
VI_LOCK(vp);
v_incr_usecount(vp);
VI_UNLOCK(vp);
}
/*
* Return reference count of a vnode.
*
* The results of this call are only guaranteed when some mechanism other
* than the VI lock is used to stop other processes from gaining references
* to the vnode. This may be the case if the caller holds the only reference.
* This is also useful when stale data is acceptable as race conditions may
* be accounted for by some other means.
*/
int
vrefcnt(struct vnode *vp)
{
int usecnt;
VI_LOCK(vp);
usecnt = vp->v_usecount;
VI_UNLOCK(vp);
return (usecnt);
}
#define VPUTX_VRELE 1
#define VPUTX_VPUT 2
#define VPUTX_VUNREF 3
static void
vputx(struct vnode *vp, int func)
{
int error;
KASSERT(vp != NULL, ("vputx: null vp"));
if (func == VPUTX_VUNREF)
ASSERT_VOP_ELOCKED(vp, "vunref");
else if (func == VPUTX_VPUT)
ASSERT_VOP_LOCKED(vp, "vput");
else
KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
VFS_ASSERT_GIANT(vp->v_mount);
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
VI_LOCK(vp);
/* Skip this v_writecount check if we're going to panic below. */
VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
("vputx: missed vn_close"));
error = 0;
if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
vp->v_usecount == 1)) {
if (func == VPUTX_VPUT)
VOP_UNLOCK(vp, 0);
v_decr_usecount(vp);
return;
}
if (vp->v_usecount != 1) {
#ifdef DIAGNOSTIC
vprint("vputx: negative ref count", vp);
#endif
panic("vputx: negative ref cnt");
}
CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
/*
* We want to hold the vnode until the inactive finishes to
* prevent vgone() races. We drop the use count here and the
* hold count below when we're done.
*/
v_decr_useonly(vp);
/*
* We must call VOP_INACTIVE with the node locked. Mark
* as VI_DOINGINACT to avoid recursion.
*/
vp->v_iflag |= VI_OWEINACT;
if (func == VPUTX_VRELE) {
error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
VI_LOCK(vp);
} else if (func == VPUTX_VPUT && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT);
VI_LOCK(vp);
}
if (vp->v_usecount > 0)
vp->v_iflag &= ~VI_OWEINACT;
if (error == 0) {
if (vp->v_iflag & VI_OWEINACT)
vinactive(vp, curthread);
if (func != VPUTX_VUNREF)
VOP_UNLOCK(vp, 0);
}
vdropl(vp);
}
/*
* Vnode put/release.
* If count drops to zero, call inactive routine and return to freelist.
*/
void
vrele(struct vnode *vp)
{
vputx(vp, VPUTX_VRELE);
}
/*
* Release an already locked vnode. This give the same effects as
* unlock+vrele(), but takes less time and avoids releasing and
* re-aquiring the lock (as vrele() acquires the lock internally.)
*/
void
vput(struct vnode *vp)
{
vputx(vp, VPUTX_VPUT);
}
/*
* Release an exclusively locked vnode. Do not unlock the vnode lock.
*/
void
vunref(struct vnode *vp)
{
vputx(vp, VPUTX_VUNREF);
}
/*
* Somebody doesn't want the vnode recycled.
*/
void
vhold(struct vnode *vp)
{
VI_LOCK(vp);
vholdl(vp);
VI_UNLOCK(vp);
}
void
vholdl(struct vnode *vp)
{
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
vp->v_holdcnt++;
if (VSHOULDBUSY(vp))
vbusy(vp);
}
/*
* Note that there is one less who cares about this vnode. vdrop() is the
* opposite of vhold().
*/
void
vdrop(struct vnode *vp)
{
VI_LOCK(vp);
vdropl(vp);
}
/*
* Drop the hold count of the vnode. If this is the last reference to
* the vnode we will free it if it has been vgone'd otherwise it is
* placed on the free list.
*/
void
vdropl(struct vnode *vp)
{
ASSERT_VI_LOCKED(vp, "vdropl");
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
if (vp->v_holdcnt <= 0)
panic("vdrop: holdcnt %d", vp->v_holdcnt);
vp->v_holdcnt--;
if (vp->v_holdcnt == 0) {
if (vp->v_iflag & VI_DOOMED) {
CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
vp);
vdestroy(vp);
return;
} else
vfree(vp);
}
VI_UNLOCK(vp);
}
/*
* Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
* flags. DOINGINACT prevents us from recursing in calls to vinactive.
* OWEINACT tracks whether a vnode missed a call to inactive due to a
* failed lock upgrade.
*/
static void
vinactive(struct vnode *vp, struct thread *td)
{
ASSERT_VOP_ELOCKED(vp, "vinactive");
ASSERT_VI_LOCKED(vp, "vinactive");
VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
("vinactive: recursed on VI_DOINGINACT"));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
vp->v_iflag |= VI_DOINGINACT;
vp->v_iflag &= ~VI_OWEINACT;
VI_UNLOCK(vp);
VOP_INACTIVE(vp, td);
VI_LOCK(vp);
VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
("vinactive: lost VI_DOINGINACT"));
vp->v_iflag &= ~VI_DOINGINACT;
}
/*
* Remove any vnodes in the vnode table belonging to mount point mp.
*
* If FORCECLOSE is not specified, there should not be any active ones,
* return error if any are found (nb: this is a user error, not a
* system error). If FORCECLOSE is specified, detach any active vnodes
* that are found.
*
* If WRITECLOSE is set, only flush out regular file vnodes open for
* writing.
*
* SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
*
* `rootrefs' specifies the base reference count for the root vnode
* of this filesystem. The root vnode is considered busy if its
* v_usecount exceeds this value. On a successful return, vflush(, td)
* will call vrele() on the root vnode exactly rootrefs times.
* If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
* be zero.
*/
#ifdef DIAGNOSTIC
static int busyprt = 0; /* print out busy vnodes */
SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
#endif
int
vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
{
struct vnode *vp, *mvp, *rootvp = NULL;
struct vattr vattr;
int busy = 0, error;
CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
rootrefs, flags);
if (rootrefs > 0) {
KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
("vflush: bad args"));
/*
* Get the filesystem root vnode. We can vput() it
* immediately, since with rootrefs > 0, it won't go away.
*/
if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
__func__, error);
return (error);
}
vput(rootvp);
}
MNT_ILOCK(mp);
loop:
MNT_VNODE_FOREACH(vp, mp, mvp) {
VI_LOCK(vp);
vholdl(vp);
MNT_IUNLOCK(mp);
error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
if (error) {
vdrop(vp);
MNT_ILOCK(mp);
MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
goto loop;
}
/*
* Skip over a vnodes marked VV_SYSTEM.
*/
if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
VOP_UNLOCK(vp, 0);
vdrop(vp);
MNT_ILOCK(mp);
continue;
}
/*
* If WRITECLOSE is set, flush out unlinked but still open
* files (even if open only for reading) and regular file
* vnodes open for writing.
*/
if (flags & WRITECLOSE) {
error = VOP_GETATTR(vp, &vattr, td->td_ucred);
VI_LOCK(vp);
if ((vp->v_type == VNON ||
(error == 0 && vattr.va_nlink > 0)) &&
(vp->v_writecount == 0 || vp->v_type != VREG)) {
VOP_UNLOCK(vp, 0);
vdropl(vp);
MNT_ILOCK(mp);
continue;
}
} else
VI_LOCK(vp);
/*
* With v_usecount == 0, all we need to do is clear out the
* vnode data structures and we are done.
*
* If FORCECLOSE is set, forcibly close the vnode.
*/
if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
VNASSERT(vp->v_usecount == 0 ||
(vp->v_type != VCHR && vp->v_type != VBLK), vp,
("device VNODE %p is FORCECLOSED", vp));
vgonel(vp);
} else {
busy++;
#ifdef DIAGNOSTIC
if (busyprt)
vprint("vflush: busy vnode", vp);
#endif
}
VOP_UNLOCK(vp, 0);
vdropl(vp);
MNT_ILOCK(mp);
}
MNT_IUNLOCK(mp);
if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
/*
* If just the root vnode is busy, and if its refcount
* is equal to `rootrefs', then go ahead and kill it.
*/
VI_LOCK(rootvp);
KASSERT(busy > 0, ("vflush: not busy"));
VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
("vflush: usecount %d < rootrefs %d",
rootvp->v_usecount, rootrefs));
if (busy == 1 && rootvp->v_usecount == rootrefs) {
VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
vgone(rootvp);
VOP_UNLOCK(rootvp, 0);
busy = 0;
} else
VI_UNLOCK(rootvp);
}
if (busy) {
CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
busy);
return (EBUSY);
}
for (; rootrefs > 0; rootrefs--)
vrele(rootvp);
return (0);
}
/*
* Recycle an unused vnode to the front of the free list.
*/
int
vrecycle(struct vnode *vp, struct thread *td)
{
int recycled;
ASSERT_VOP_ELOCKED(vp, "vrecycle");
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
recycled = 0;
VI_LOCK(vp);
if (vp->v_usecount == 0) {
recycled = 1;
vgonel(vp);
}
VI_UNLOCK(vp);
return (recycled);
}
/*
* Eliminate all activity associated with a vnode
* in preparation for reuse.
*/
void
vgone(struct vnode *vp)
{
VI_LOCK(vp);
vgonel(vp);
VI_UNLOCK(vp);
}
/*
* vgone, with the vp interlock held.
*/
void
vgonel(struct vnode *vp)
{
struct thread *td;
int oweinact;
int active;
struct mount *mp;
ASSERT_VOP_ELOCKED(vp, "vgonel");
ASSERT_VI_LOCKED(vp, "vgonel");
VNASSERT(vp->v_holdcnt, vp,
("vgonel: vp %p has no reference.", vp));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
td = curthread;
/*
* Don't vgonel if we're already doomed.
*/
if (vp->v_iflag & VI_DOOMED)
return;
vp->v_iflag |= VI_DOOMED;
/*
* Check to see if the vnode is in use. If so, we have to call
* VOP_CLOSE() and VOP_INACTIVE().
*/
active = vp->v_usecount;
oweinact = (vp->v_iflag & VI_OWEINACT);
VI_UNLOCK(vp);
/*
* Clean out any buffers associated with the vnode.
* If the flush fails, just toss the buffers.
*/
mp = NULL;
if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
(void) vn_start_secondary_write(vp, &mp, V_WAIT);
if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
vinvalbuf(vp, 0, 0, 0);
/*
* If purging an active vnode, it must be closed and
* deactivated before being reclaimed.
*/
if (active)
VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
if (oweinact || active) {
VI_LOCK(vp);
if ((vp->v_iflag & VI_DOINGINACT) == 0)
vinactive(vp, td);
VI_UNLOCK(vp);
}
/*
* Reclaim the vnode.
*/
if (VOP_RECLAIM(vp, td))
panic("vgone: cannot reclaim");
if (mp != NULL)
vn_finished_secondary_write(mp);
VNASSERT(vp->v_object == NULL, vp,
("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
/*
* Clear the advisory locks and wake up waiting threads.
*/
lf_purgelocks(vp, &(vp->v_lockf));
/*
* Delete from old mount point vnode list.
*/
delmntque(vp);
cache_purge(vp);
/*
* Done with purge, reset to the standard lock and invalidate
* the vnode.
*/
VI_LOCK(vp);
vp->v_vnlock = &vp->v_lock;
vp->v_op = &dead_vnodeops;
vp->v_tag = "none";
vp->v_type = VBAD;
}
/*
* Calculate the total number of references to a special device.
*/
int
vcount(struct vnode *vp)
{
int count;
dev_lock();
count = vp->v_rdev->si_usecount;
dev_unlock();
return (count);
}
/*
* Same as above, but using the struct cdev *as argument
*/
int
count_dev(struct cdev *dev)
{
int count;
dev_lock();
count = dev->si_usecount;
dev_unlock();
return(count);
}
/*
* Print out a description of a vnode.
*/
static char *typename[] =
{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
"VMARKER"};
void
vn_printf(struct vnode *vp, const char *fmt, ...)
{
va_list ap;
char buf[256], buf2[16];
u_long flags;
va_start(ap, fmt);
vprintf(fmt, ap);
va_end(ap);
printf("%p: ", (void *)vp);
printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
buf[0] = '\0';
buf[1] = '\0';
if (vp->v_vflag & VV_ROOT)
strlcat(buf, "|VV_ROOT", sizeof(buf));
if (vp->v_vflag & VV_ISTTY)
strlcat(buf, "|VV_ISTTY", sizeof(buf));
if (vp->v_vflag & VV_NOSYNC)
strlcat(buf, "|VV_NOSYNC", sizeof(buf));
if (vp->v_vflag & VV_CACHEDLABEL)
strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
if (vp->v_vflag & VV_TEXT)
strlcat(buf, "|VV_TEXT", sizeof(buf));
if (vp->v_vflag & VV_COPYONWRITE)
strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
if (vp->v_vflag & VV_SYSTEM)
strlcat(buf, "|VV_SYSTEM", sizeof(buf));
if (vp->v_vflag & VV_PROCDEP)
strlcat(buf, "|VV_PROCDEP", sizeof(buf));
if (vp->v_vflag & VV_NOKNOTE)
strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
if (vp->v_vflag & VV_DELETED)
strlcat(buf, "|VV_DELETED", sizeof(buf));
if (vp->v_vflag & VV_MD)
strlcat(buf, "|VV_MD", sizeof(buf));
flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
VV_NOKNOTE | VV_DELETED | VV_MD);
if (flags != 0) {
snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
strlcat(buf, buf2, sizeof(buf));
}
if (vp->v_iflag & VI_MOUNT)
strlcat(buf, "|VI_MOUNT", sizeof(buf));
if (vp->v_iflag & VI_AGE)
strlcat(buf, "|VI_AGE", sizeof(buf));
if (vp->v_iflag & VI_DOOMED)
strlcat(buf, "|VI_DOOMED", sizeof(buf));
if (vp->v_iflag & VI_FREE)
strlcat(buf, "|VI_FREE", sizeof(buf));
if (vp->v_iflag & VI_DOINGINACT)
strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
if (vp->v_iflag & VI_OWEINACT)
strlcat(buf, "|VI_OWEINACT", sizeof(buf));
flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
VI_DOINGINACT | VI_OWEINACT);
if (flags != 0) {
snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
strlcat(buf, buf2, sizeof(buf));
}
printf(" flags (%s)\n", buf + 1);
if (mtx_owned(VI_MTX(vp)))
printf(" VI_LOCKed");
if (vp->v_object != NULL)
printf(" v_object %p ref %d pages %d\n",
vp->v_object, vp->v_object->ref_count,
vp->v_object->resident_page_count);
printf(" ");
lockmgr_printinfo(vp->v_vnlock);
if (vp->v_data != NULL)
VOP_PRINT(vp);
}
#ifdef DDB
/*
* List all of the locked vnodes in the system.
* Called when debugging the kernel.
*/
DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
{
struct mount *mp, *nmp;
struct vnode *vp;
/*
* Note: because this is DDB, we can't obey the locking semantics
* for these structures, which means we could catch an inconsistent
* state and dereference a nasty pointer. Not much to be done
* about that.
*/
db_printf("Locked vnodes\n");
for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
nmp = TAILQ_NEXT(mp, mnt_list);
TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
if (vp->v_type != VMARKER &&
VOP_ISLOCKED(vp))
vprint("", vp);
}
nmp = TAILQ_NEXT(mp, mnt_list);
}
}
/*
* Show details about the given vnode.
*/
DB_SHOW_COMMAND(vnode, db_show_vnode)
{
struct vnode *vp;
if (!have_addr)
return;
vp = (struct vnode *)addr;
vn_printf(vp, "vnode ");
}
/*
* Show details about the given mount point.
*/
DB_SHOW_COMMAND(mount, db_show_mount)
{
struct mount *mp;
struct vfsopt *opt;
struct statfs *sp;
struct vnode *vp;
char buf[512];
u_int flags;
if (!have_addr) {
/* No address given, print short info about all mount points. */
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
db_printf("%p %s on %s (%s)\n", mp,
mp->mnt_stat.f_mntfromname,
mp->mnt_stat.f_mntonname,
mp->mnt_stat.f_fstypename);
if (db_pager_quit)
break;
}
db_printf("\nMore info: show mount <addr>\n");
return;
}
mp = (struct mount *)addr;
db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
buf[0] = '\0';
flags = mp->mnt_flag;
#define MNT_FLAG(flag) do { \
if (flags & (flag)) { \
if (buf[0] != '\0') \
strlcat(buf, ", ", sizeof(buf)); \
strlcat(buf, (#flag) + 4, sizeof(buf)); \
flags &= ~(flag); \
} \
} while (0)
MNT_FLAG(MNT_RDONLY);
MNT_FLAG(MNT_SYNCHRONOUS);
MNT_FLAG(MNT_NOEXEC);
MNT_FLAG(MNT_NOSUID);
MNT_FLAG(MNT_UNION);
MNT_FLAG(MNT_ASYNC);
MNT_FLAG(MNT_SUIDDIR);
MNT_FLAG(MNT_SOFTDEP);
MNT_FLAG(MNT_NOSYMFOLLOW);
MNT_FLAG(MNT_GJOURNAL);
MNT_FLAG(MNT_MULTILABEL);
MNT_FLAG(MNT_ACLS);
MNT_FLAG(MNT_NOATIME);
MNT_FLAG(MNT_NOCLUSTERR);
MNT_FLAG(MNT_NOCLUSTERW);
MNT_FLAG(MNT_NFS4ACLS);
MNT_FLAG(MNT_EXRDONLY);
MNT_FLAG(MNT_EXPORTED);
MNT_FLAG(MNT_DEFEXPORTED);
MNT_FLAG(MNT_EXPORTANON);
MNT_FLAG(MNT_EXKERB);
MNT_FLAG(MNT_EXPUBLIC);
MNT_FLAG(MNT_LOCAL);
MNT_FLAG(MNT_QUOTA);
MNT_FLAG(MNT_ROOTFS);
MNT_FLAG(MNT_USER);
MNT_FLAG(MNT_IGNORE);
MNT_FLAG(MNT_UPDATE);
MNT_FLAG(MNT_DELEXPORT);
MNT_FLAG(MNT_RELOAD);
MNT_FLAG(MNT_FORCE);
MNT_FLAG(MNT_SNAPSHOT);
MNT_FLAG(MNT_BYFSID);
#undef MNT_FLAG
if (flags != 0) {
if (buf[0] != '\0')
strlcat(buf, ", ", sizeof(buf));
snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
"0x%08x", flags);
}
db_printf(" mnt_flag = %s\n", buf);
buf[0] = '\0';
flags = mp->mnt_kern_flag;
#define MNT_KERN_FLAG(flag) do { \
if (flags & (flag)) { \
if (buf[0] != '\0') \
strlcat(buf, ", ", sizeof(buf)); \
strlcat(buf, (#flag) + 5, sizeof(buf)); \
flags &= ~(flag); \
} \
} while (0)
MNT_KERN_FLAG(MNTK_UNMOUNTF);
MNT_KERN_FLAG(MNTK_ASYNC);
MNT_KERN_FLAG(MNTK_SOFTDEP);
MNT_KERN_FLAG(MNTK_NOINSMNTQ);
MNT_KERN_FLAG(MNTK_UNMOUNT);
MNT_KERN_FLAG(MNTK_MWAIT);
MNT_KERN_FLAG(MNTK_SUSPEND);
MNT_KERN_FLAG(MNTK_SUSPEND2);
MNT_KERN_FLAG(MNTK_SUSPENDED);
MNT_KERN_FLAG(MNTK_MPSAFE);
MNT_KERN_FLAG(MNTK_NOKNOTE);
MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
#undef MNT_KERN_FLAG
if (flags != 0) {
if (buf[0] != '\0')
strlcat(buf, ", ", sizeof(buf));
snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
"0x%08x", flags);
}
db_printf(" mnt_kern_flag = %s\n", buf);
db_printf(" mnt_opt = ");
opt = TAILQ_FIRST(mp->mnt_opt);
if (opt != NULL) {
db_printf("%s", opt->name);
opt = TAILQ_NEXT(opt, link);
while (opt != NULL) {
db_printf(", %s", opt->name);
opt = TAILQ_NEXT(opt, link);
}
}
db_printf("\n");
sp = &mp->mnt_stat;
db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
"bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
"ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
"asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
(u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
(uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
(uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
(intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
(intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
(uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
(uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
(u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
db_printf(" mnt_cred = { uid=%u ruid=%u",
(u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
if (jailed(mp->mnt_cred))
db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
db_printf(" }\n");
db_printf(" mnt_ref = %d\n", mp->mnt_ref);
db_printf(" mnt_gen = %d\n", mp->mnt_gen);
db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
db_printf(" mnt_secondary_accwrites = %d\n",
mp->mnt_secondary_accwrites);
db_printf(" mnt_gjprovider = %s\n",
mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
db_printf("\n");
TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
if (vp->v_type != VMARKER) {
vn_printf(vp, "vnode ");
if (db_pager_quit)
break;
}
}
}
#endif /* DDB */
/*
* Fill in a struct xvfsconf based on a struct vfsconf.
*/
static void
vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
{
strcpy(xvfsp->vfc_name, vfsp->vfc_name);
xvfsp->vfc_typenum = vfsp->vfc_typenum;
xvfsp->vfc_refcount = vfsp->vfc_refcount;
xvfsp->vfc_flags = vfsp->vfc_flags;
/*
* These are unused in userland, we keep them
* to not break binary compatibility.
*/
xvfsp->vfc_vfsops = NULL;
xvfsp->vfc_next = NULL;
}
/*
* Top level filesystem related information gathering.
*/
static int
sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
{
struct vfsconf *vfsp;
struct xvfsconf xvfsp;
int error;
error = 0;
TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
bzero(&xvfsp, sizeof(xvfsp));
vfsconf2x(vfsp, &xvfsp);
error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
if (error)
break;
}
return (error);
}
SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
"S,xvfsconf", "List of all configured filesystems");
#ifndef BURN_BRIDGES
static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
static int
vfs_sysctl(SYSCTL_HANDLER_ARGS)
{
int *name = (int *)arg1 - 1; /* XXX */
u_int namelen = arg2 + 1; /* XXX */
struct vfsconf *vfsp;
struct xvfsconf xvfsp;
printf("WARNING: userland calling deprecated sysctl, "
"please rebuild world\n");
#if 1 || defined(COMPAT_PRELITE2)
/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
if (namelen == 1)
return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
#endif
switch (name[1]) {
case VFS_MAXTYPENUM:
if (namelen != 2)
return (ENOTDIR);
return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
case VFS_CONF:
if (namelen != 3)
return (ENOTDIR); /* overloaded */
TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
if (vfsp->vfc_typenum == name[2])
break;
if (vfsp == NULL)
return (EOPNOTSUPP);
bzero(&xvfsp, sizeof(xvfsp));
vfsconf2x(vfsp, &xvfsp);
return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
}
return (EOPNOTSUPP);
}
static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
vfs_sysctl, "Generic filesystem");
#if 1 || defined(COMPAT_PRELITE2)
static int
sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
{
int error;
struct vfsconf *vfsp;
struct ovfsconf ovfs;
TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
bzero(&ovfs, sizeof(ovfs));
ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
strcpy(ovfs.vfc_name, vfsp->vfc_name);
ovfs.vfc_index = vfsp->vfc_typenum;
ovfs.vfc_refcount = vfsp->vfc_refcount;
ovfs.vfc_flags = vfsp->vfc_flags;
error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
if (error)
return error;
}
return 0;
}
#endif /* 1 || COMPAT_PRELITE2 */
#endif /* !BURN_BRIDGES */
#define KINFO_VNODESLOP 10
#ifdef notyet
/*
* Dump vnode list (via sysctl).
*/
/* ARGSUSED */
static int
sysctl_vnode(SYSCTL_HANDLER_ARGS)
{
struct xvnode *xvn;
struct mount *mp;
struct vnode *vp;
int error, len, n;
/*
* Stale numvnodes access is not fatal here.
*/
req->lock = 0;
len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
if (!req->oldptr)
/* Make an estimate */
return (SYSCTL_OUT(req, 0, len));
error = sysctl_wire_old_buffer(req, 0);
if (error != 0)
return (error);
xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
n = 0;
mtx_lock(&mountlist_mtx);
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
continue;
MNT_ILOCK(mp);
TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
if (n == len)
break;
vref(vp);
xvn[n].xv_size = sizeof *xvn;
xvn[n].xv_vnode = vp;
xvn[n].xv_id = 0; /* XXX compat */
#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
XV_COPY(usecount);
XV_COPY(writecount);
XV_COPY(holdcnt);
XV_COPY(mount);
XV_COPY(numoutput);
XV_COPY(type);
#undef XV_COPY
xvn[n].xv_flag = vp->v_vflag;
switch (vp->v_type) {
case VREG:
case VDIR:
case VLNK:
break;
case VBLK:
case VCHR:
if (vp->v_rdev == NULL) {
vrele(vp);
continue;
}
xvn[n].xv_dev = dev2udev(vp->v_rdev);
break;
case VSOCK:
xvn[n].xv_socket = vp->v_socket;
break;
case VFIFO:
xvn[n].xv_fifo = vp->v_fifoinfo;
break;
case VNON:
case VBAD:
default:
/* shouldn't happen? */
vrele(vp);
continue;
}
vrele(vp);
++n;
}
MNT_IUNLOCK(mp);
mtx_lock(&mountlist_mtx);
vfs_unbusy(mp);
if (n == len)
break;
}
mtx_unlock(&mountlist_mtx);
error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
free(xvn, M_TEMP);
return (error);
}
SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
0, 0, sysctl_vnode, "S,xvnode", "");
#endif
/*
* Unmount all filesystems. The list is traversed in reverse order
* of mounting to avoid dependencies.
*/
void
vfs_unmountall(void)
{
struct mount *mp;
struct thread *td;
int error;
KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
td = curthread;
/*
* Since this only runs when rebooting, it is not interlocked.
*/
while(!TAILQ_EMPTY(&mountlist)) {
mp = TAILQ_LAST(&mountlist, mntlist);
error = dounmount(mp, MNT_FORCE, td);
if (error) {
TAILQ_REMOVE(&mountlist, mp, mnt_list);
/*
* XXX: Due to the way in which we mount the root
* file system off of devfs, devfs will generate a
* "busy" warning when we try to unmount it before
* the root. Don't print a warning as a result in
* order to avoid false positive errors that may
* cause needless upset.
*/
if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
printf("unmount of %s failed (",
mp->mnt_stat.f_mntonname);
if (error == EBUSY)
printf("BUSY)\n");
else
printf("%d)\n", error);
}
} else {
/* The unmount has removed mp from the mountlist */
}
}
}
/*
* perform msync on all vnodes under a mount point
* the mount point must be locked.
*/
void
vfs_msync(struct mount *mp, int flags)
{
struct vnode *vp, *mvp;
struct vm_object *obj;
CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
MNT_ILOCK(mp);
MNT_VNODE_FOREACH(vp, mp, mvp) {
VI_LOCK(vp);
obj = vp->v_object;
if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
(flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
MNT_IUNLOCK(mp);
if (!vget(vp,
LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
curthread)) {
if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
vput(vp);
MNT_ILOCK(mp);
continue;
}
obj = vp->v_object;
if (obj != NULL) {
VM_OBJECT_LOCK(obj);
vm_object_page_clean(obj, 0, 0,
flags == MNT_WAIT ?
OBJPC_SYNC : OBJPC_NOSYNC);
VM_OBJECT_UNLOCK(obj);
}
vput(vp);
}
MNT_ILOCK(mp);
} else
VI_UNLOCK(vp);
}
MNT_IUNLOCK(mp);
}
/*
* Mark a vnode as free, putting it up for recycling.
*/
static void
vfree(struct vnode *vp)
{
ASSERT_VI_LOCKED(vp, "vfree");
mtx_lock(&vnode_free_list_mtx);
VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
("vfree: Freeing doomed vnode"));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
if (vp->v_iflag & VI_AGE) {
TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
} else {
TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
}
freevnodes++;
vp->v_iflag &= ~VI_AGE;
vp->v_iflag |= VI_FREE;
mtx_unlock(&vnode_free_list_mtx);
}
/*
* Opposite of vfree() - mark a vnode as in use.
*/
static void
vbusy(struct vnode *vp)
{
ASSERT_VI_LOCKED(vp, "vbusy");
VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
mtx_lock(&vnode_free_list_mtx);
TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
freevnodes--;
vp->v_iflag &= ~(VI_FREE|VI_AGE);
mtx_unlock(&vnode_free_list_mtx);
}
static void
destroy_vpollinfo(struct vpollinfo *vi)
{
knlist_destroy(&vi->vpi_selinfo.si_note);
mtx_destroy(&vi->vpi_lock);
uma_zfree(vnodepoll_zone, vi);
}
/*
* Initalize per-vnode helper structure to hold poll-related state.
*/
void
v_addpollinfo(struct vnode *vp)
{
struct vpollinfo *vi;
if (vp->v_pollinfo != NULL)
return;
vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
VI_LOCK(vp);
if (vp->v_pollinfo != NULL) {
VI_UNLOCK(vp);
destroy_vpollinfo(vi);
return;
}
vp->v_pollinfo = vi;
VI_UNLOCK(vp);
}
/*
* Record a process's interest in events which might happen to
* a vnode. Because poll uses the historic select-style interface
* internally, this routine serves as both the ``check for any
* pending events'' and the ``record my interest in future events''
* functions. (These are done together, while the lock is held,
* to avoid race conditions.)
*/
int
vn_pollrecord(struct vnode *vp, struct thread *td, int events)
{
v_addpollinfo(vp);
mtx_lock(&vp->v_pollinfo->vpi_lock);
if (vp->v_pollinfo->vpi_revents & events) {
/*
* This leaves events we are not interested
* in available for the other process which
* which presumably had requested them
* (otherwise they would never have been
* recorded).
*/
events &= vp->v_pollinfo->vpi_revents;
vp->v_pollinfo->vpi_revents &= ~events;
mtx_unlock(&vp->v_pollinfo->vpi_lock);
return (events);
}
vp->v_pollinfo->vpi_events |= events;
selrecord(td, &vp->v_pollinfo->vpi_selinfo);
mtx_unlock(&vp->v_pollinfo->vpi_lock);
return (0);
}
/*
* Routine to create and manage a filesystem syncer vnode.
*/
#define sync_close ((int (*)(struct vop_close_args *))nullop)
static int sync_fsync(struct vop_fsync_args *);
static int sync_inactive(struct vop_inactive_args *);
static int sync_reclaim(struct vop_reclaim_args *);
static struct vop_vector sync_vnodeops = {
.vop_bypass = VOP_EOPNOTSUPP,
.vop_close = sync_close, /* close */
.vop_fsync = sync_fsync, /* fsync */
.vop_inactive = sync_inactive, /* inactive */
.vop_reclaim = sync_reclaim, /* reclaim */
.vop_lock1 = vop_stdlock, /* lock */
.vop_unlock = vop_stdunlock, /* unlock */
.vop_islocked = vop_stdislocked, /* islocked */
};
/*
* Create a new filesystem syncer vnode for the specified mount point.
*/
int
vfs_allocate_syncvnode(struct mount *mp)
{
struct vnode *vp;
struct bufobj *bo;
static long start, incr, next;
int error;
/* Allocate a new vnode */
if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
mp->mnt_syncer = NULL;
return (error);
}
vp->v_type = VNON;
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vp->v_vflag |= VV_FORCEINSMQ;
error = insmntque(vp, mp);
if (error != 0)
panic("vfs_allocate_syncvnode: insmntque failed");
vp->v_vflag &= ~VV_FORCEINSMQ;
VOP_UNLOCK(vp, 0);
/*
* Place the vnode onto the syncer worklist. We attempt to
* scatter them about on the list so that they will go off
* at evenly distributed times even if all the filesystems
* are mounted at once.
*/
next += incr;
if (next == 0 || next > syncer_maxdelay) {
start /= 2;
incr /= 2;
if (start == 0) {
start = syncer_maxdelay / 2;
incr = syncer_maxdelay;
}
next = start;
}
bo = &vp->v_bufobj;
BO_LOCK(bo);
vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
mtx_lock(&sync_mtx);
sync_vnode_count++;
mtx_unlock(&sync_mtx);
BO_UNLOCK(bo);
mp->mnt_syncer = vp;
return (0);
}
/*
* Do a lazy sync of the filesystem.
*/
static int
sync_fsync(struct vop_fsync_args *ap)
{
struct vnode *syncvp = ap->a_vp;
struct mount *mp = syncvp->v_mount;
int error;
struct bufobj *bo;
/*
* We only need to do something if this is a lazy evaluation.
*/
if (ap->a_waitfor != MNT_LAZY)
return (0);
/*
* Move ourselves to the back of the sync list.
*/
bo = &syncvp->v_bufobj;
BO_LOCK(bo);
vn_syncer_add_to_worklist(bo, syncdelay);
BO_UNLOCK(bo);
/*
* Walk the list of vnodes pushing all that are dirty and
* not already on the sync list.
*/
mtx_lock(&mountlist_mtx);
if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
mtx_unlock(&mountlist_mtx);
return (0);
}
if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
vfs_unbusy(mp);
return (0);
}
MNT_ILOCK(mp);
mp->mnt_noasync++;
mp->mnt_kern_flag &= ~MNTK_ASYNC;
MNT_IUNLOCK(mp);
vfs_msync(mp, MNT_NOWAIT);
error = VFS_SYNC(mp, MNT_LAZY);
MNT_ILOCK(mp);
mp->mnt_noasync--;
if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
mp->mnt_kern_flag |= MNTK_ASYNC;
MNT_IUNLOCK(mp);
vn_finished_write(mp);
vfs_unbusy(mp);
return (error);
}
/*
* The syncer vnode is no referenced.
*/
static int
sync_inactive(struct vop_inactive_args *ap)
{
vgone(ap->a_vp);
return (0);
}
/*
* The syncer vnode is no longer needed and is being decommissioned.
*
* Modifications to the worklist must be protected by sync_mtx.
*/
static int
sync_reclaim(struct vop_reclaim_args *ap)
{
struct vnode *vp = ap->a_vp;
struct bufobj *bo;
bo = &vp->v_bufobj;
BO_LOCK(bo);
vp->v_mount->mnt_syncer = NULL;
if (bo->bo_flag & BO_ONWORKLST) {
mtx_lock(&sync_mtx);
LIST_REMOVE(bo, bo_synclist);
syncer_worklist_len--;
sync_vnode_count--;
mtx_unlock(&sync_mtx);
bo->bo_flag &= ~BO_ONWORKLST;
}
BO_UNLOCK(bo);
return (0);
}
/*
* Check if vnode represents a disk device
*/
int
vn_isdisk(struct vnode *vp, int *errp)
{
int error;
error = 0;
dev_lock();
if (vp->v_type != VCHR)
error = ENOTBLK;
else if (vp->v_rdev == NULL)
error = ENXIO;
else if (vp->v_rdev->si_devsw == NULL)
error = ENXIO;
else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
error = ENOTBLK;
dev_unlock();
if (errp != NULL)
*errp = error;
return (error == 0);
}
/*
* Common filesystem object access control check routine. Accepts a
* vnode's type, "mode", uid and gid, requested access mode, credentials,
* and optional call-by-reference privused argument allowing vaccess()
* to indicate to the caller whether privilege was used to satisfy the
* request (obsoleted). Returns 0 on success, or an errno on failure.
*
* The ifdef'd CAPABILITIES version is here for reference, but is not
* actually used.
*/
int
vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
accmode_t accmode, struct ucred *cred, int *privused)
{
accmode_t dac_granted;
accmode_t priv_granted;
KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
("invalid bit in accmode"));
/*
* Look for a normal, non-privileged way to access the file/directory
* as requested. If it exists, go with that.
*/
if (privused != NULL)
*privused = 0;
dac_granted = 0;
/* Check the owner. */
if (cred->cr_uid == file_uid) {
dac_granted |= VADMIN;
if (file_mode & S_IXUSR)
dac_granted |= VEXEC;
if (file_mode & S_IRUSR)
dac_granted |= VREAD;
if (file_mode & S_IWUSR)
dac_granted |= (VWRITE | VAPPEND);
if ((accmode & dac_granted) == accmode)
return (0);
goto privcheck;
}
/* Otherwise, check the groups (first match) */
if (groupmember(file_gid, cred)) {
if (file_mode & S_IXGRP)
dac_granted |= VEXEC;
if (file_mode & S_IRGRP)
dac_granted |= VREAD;
if (file_mode & S_IWGRP)
dac_granted |= (VWRITE | VAPPEND);
if ((accmode & dac_granted) == accmode)
return (0);
goto privcheck;
}
/* Otherwise, check everyone else. */
if (file_mode & S_IXOTH)
dac_granted |= VEXEC;
if (file_mode & S_IROTH)
dac_granted |= VREAD;
if (file_mode & S_IWOTH)
dac_granted |= (VWRITE | VAPPEND);
if ((accmode & dac_granted) == accmode)
return (0);
privcheck:
/*
* Build a privilege mask to determine if the set of privileges
* satisfies the requirements when combined with the granted mask
* from above. For each privilege, if the privilege is required,
* bitwise or the request type onto the priv_granted mask.
*/
priv_granted = 0;
if (type == VDIR) {
/*
* For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
* requests, instead of PRIV_VFS_EXEC.
*/
if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
!priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
priv_granted |= VEXEC;
} else {
if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
!priv_check_cred(cred, PRIV_VFS_EXEC, 0))
priv_granted |= VEXEC;
}
if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
!priv_check_cred(cred, PRIV_VFS_READ, 0))
priv_granted |= VREAD;
if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
!priv_check_cred(cred, PRIV_VFS_WRITE, 0))
priv_granted |= (VWRITE | VAPPEND);
if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
!priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
priv_granted |= VADMIN;
if ((accmode & (priv_granted | dac_granted)) == accmode) {
/* XXX audit: privilege used */
if (privused != NULL)
*privused = 1;
return (0);
}
return ((accmode & VADMIN) ? EPERM : EACCES);
}
/*
* Credential check based on process requesting service, and per-attribute
* permissions.
*/
int
extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
struct thread *td, accmode_t accmode)
{
/*
* Kernel-invoked always succeeds.
*/
if (cred == NOCRED)
return (0);
/*
* Do not allow privileged processes in jail to directly manipulate
* system attributes.
*/
switch (attrnamespace) {
case EXTATTR_NAMESPACE_SYSTEM:
/* Potentially should be: return (EPERM); */
return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
case EXTATTR_NAMESPACE_USER:
return (VOP_ACCESS(vp, accmode, cred, td));
default:
return (EPERM);
}
}
#ifdef DEBUG_VFS_LOCKS
/*
* This only exists to supress warnings from unlocked specfs accesses. It is
* no longer ok to have an unlocked VFS.
*/
#define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
(vp)->v_type == VCHR || (vp)->v_type == VBAD)
int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
int vfs_badlock_print = 1; /* Print lock violations. */
SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
#ifdef KDB
int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
#endif
static void
vfs_badlock(const char *msg, const char *str, struct vnode *vp)
{
#ifdef KDB
if (vfs_badlock_backtrace)
kdb_backtrace();
#endif
if (vfs_badlock_print)
printf("%s: %p %s\n", str, (void *)vp, msg);
if (vfs_badlock_ddb)
kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
}
void
assert_vi_locked(struct vnode *vp, const char *str)
{
if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
vfs_badlock("interlock is not locked but should be", str, vp);
}
void
assert_vi_unlocked(struct vnode *vp, const char *str)
{
if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
vfs_badlock("interlock is locked but should not be", str, vp);
}
void
assert_vop_locked(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
vfs_badlock("is not locked but should be", str, vp);
}
void
assert_vop_unlocked(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
vfs_badlock("is locked but should not be", str, vp);
}
void
assert_vop_elocked(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
vfs_badlock("is not exclusive locked but should be", str, vp);
}
#if 0
void
assert_vop_elocked_other(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
vfs_badlock("is not exclusive locked by another thread",
str, vp);
}
void
assert_vop_slocked(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
vfs_badlock("is not locked shared but should be", str, vp);
}
#endif /* 0 */
#endif /* DEBUG_VFS_LOCKS */
void
vop_rename_fail(struct vop_rename_args *ap)
{
if (ap->a_tvp != NULL)
vput(ap->a_tvp);
if (ap->a_tdvp == ap->a_tvp)
vrele(ap->a_tdvp);
else
vput(ap->a_tdvp);
vrele(ap->a_fdvp);
vrele(ap->a_fvp);
}
void
vop_rename_pre(void *ap)
{
struct vop_rename_args *a = ap;
#ifdef DEBUG_VFS_LOCKS
if (a->a_tvp)
ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
/* Check the source (from). */
if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
if (a->a_tvp != a->a_fvp)
ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
/* Check the target. */
if (a->a_tvp)
ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
#endif
if (a->a_tdvp != a->a_fdvp)
vhold(a->a_fdvp);
if (a->a_tvp != a->a_fvp)
vhold(a->a_fvp);
vhold(a->a_tdvp);
if (a->a_tvp)
vhold(a->a_tvp);
}
void
vop_strategy_pre(void *ap)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_strategy_args *a;
struct buf *bp;
a = ap;
bp = a->a_bp;
/*
* Cluster ops lock their component buffers but not the IO container.
*/
if ((bp->b_flags & B_CLUSTER) != 0)
return;
if (!BUF_ISLOCKED(bp)) {
if (vfs_badlock_print)
printf(
"VOP_STRATEGY: bp is not locked but should be\n");
if (vfs_badlock_ddb)
kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
}
#endif
}
void
vop_lookup_pre(void *ap)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_lookup_args *a;
struct vnode *dvp;
a = ap;
dvp = a->a_dvp;
ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
#endif
}
void
vop_lookup_post(void *ap, int rc)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_lookup_args *a;
struct vnode *dvp;
struct vnode *vp;
a = ap;
dvp = a->a_dvp;
vp = *(a->a_vpp);
ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
if (!rc)
ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
#endif
}
void
vop_lock_pre(void *ap)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_lock1_args *a = ap;
if ((a->a_flags & LK_INTERLOCK) == 0)
ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
else
ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
#endif
}
void
vop_lock_post(void *ap, int rc)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_lock1_args *a = ap;
ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
if (rc == 0)
ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
#endif
}
void
vop_unlock_pre(void *ap)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_unlock_args *a = ap;
if (a->a_flags & LK_INTERLOCK)
ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
#endif
}
void
vop_unlock_post(void *ap, int rc)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_unlock_args *a = ap;
if (a->a_flags & LK_INTERLOCK)
ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
#endif
}
void
vop_create_post(void *ap, int rc)
{
struct vop_create_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
}
void
vop_link_post(void *ap, int rc)
{
struct vop_link_args *a = ap;
if (!rc) {
VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
}
}
void
vop_mkdir_post(void *ap, int rc)
{
struct vop_mkdir_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
}
void
vop_mknod_post(void *ap, int rc)
{
struct vop_mknod_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
}
void
vop_remove_post(void *ap, int rc)
{
struct vop_remove_args *a = ap;
if (!rc) {
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
}
}
void
vop_rename_post(void *ap, int rc)
{
struct vop_rename_args *a = ap;
if (!rc) {
VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
if (a->a_tvp)
VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
}
if (a->a_tdvp != a->a_fdvp)
vdrop(a->a_fdvp);
if (a->a_tvp != a->a_fvp)
vdrop(a->a_fvp);
vdrop(a->a_tdvp);
if (a->a_tvp)
vdrop(a->a_tvp);
}
void
vop_rmdir_post(void *ap, int rc)
{
struct vop_rmdir_args *a = ap;
if (!rc) {
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
}
}
void
vop_setattr_post(void *ap, int rc)
{
struct vop_setattr_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
}
void
vop_symlink_post(void *ap, int rc)
{
struct vop_symlink_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
}
static struct knlist fs_knlist;
static void
vfs_event_init(void *arg)
{
knlist_init_mtx(&fs_knlist, NULL);
}
/* XXX - correct order? */
SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
void
vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
{
KNOTE_UNLOCKED(&fs_knlist, event);
}
static int filt_fsattach(struct knote *kn);
static void filt_fsdetach(struct knote *kn);
static int filt_fsevent(struct knote *kn, long hint);
struct filterops fs_filtops =
{ 0, filt_fsattach, filt_fsdetach, filt_fsevent };
static int
filt_fsattach(struct knote *kn)
{
kn->kn_flags |= EV_CLEAR;
knlist_add(&fs_knlist, kn, 0);
return (0);
}
static void
filt_fsdetach(struct knote *kn)
{
knlist_remove(&fs_knlist, kn, 0);
}
static int
filt_fsevent(struct knote *kn, long hint)
{
kn->kn_fflags |= hint;
return (kn->kn_fflags != 0);
}
static int
sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
{
struct vfsidctl vc;
int error;
struct mount *mp;
error = SYSCTL_IN(req, &vc, sizeof(vc));
if (error)
return (error);
if (vc.vc_vers != VFS_CTL_VERS1)
return (EINVAL);
mp = vfs_getvfs(&vc.vc_fsid);
if (mp == NULL)
return (ENOENT);
/* ensure that a specific sysctl goes to the right filesystem. */
if (strcmp(vc.vc_fstypename, "*") != 0 &&
strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
vfs_rel(mp);
return (EINVAL);
}
VCTLTOREQ(&vc, req);
error = VFS_SYSCTL(mp, vc.vc_op, req);
vfs_rel(mp);
return (error);
}
SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
"Sysctl by fsid");
/*
* Function to initialize a va_filerev field sensibly.
* XXX: Wouldn't a random number make a lot more sense ??
*/
u_quad_t
init_va_filerev(void)
{
struct bintime bt;
getbinuptime(&bt);
return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
}
static int filt_vfsread(struct knote *kn, long hint);
static int filt_vfswrite(struct knote *kn, long hint);
static int filt_vfsvnode(struct knote *kn, long hint);
static void filt_vfsdetach(struct knote *kn);
static struct filterops vfsread_filtops =
{ 1, NULL, filt_vfsdetach, filt_vfsread };
static struct filterops vfswrite_filtops =
{ 1, NULL, filt_vfsdetach, filt_vfswrite };
static struct filterops vfsvnode_filtops =
{ 1, NULL, filt_vfsdetach, filt_vfsvnode };
static void
vfs_knllock(void *arg)
{
struct vnode *vp = arg;
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
}
static void
vfs_knlunlock(void *arg)
{
struct vnode *vp = arg;
VOP_UNLOCK(vp, 0);
}
static void
vfs_knl_assert_locked(void *arg)
{
#ifdef DEBUG_VFS_LOCKS
struct vnode *vp = arg;
ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
#endif
}
static void
vfs_knl_assert_unlocked(void *arg)
{
#ifdef DEBUG_VFS_LOCKS
struct vnode *vp = arg;
ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
#endif
}
int
vfs_kqfilter(struct vop_kqfilter_args *ap)
{
struct vnode *vp = ap->a_vp;
struct knote *kn = ap->a_kn;
struct knlist *knl;
switch (kn->kn_filter) {
case EVFILT_READ:
kn->kn_fop = &vfsread_filtops;
break;
case EVFILT_WRITE:
kn->kn_fop = &vfswrite_filtops;
break;
case EVFILT_VNODE:
kn->kn_fop = &vfsvnode_filtops;
break;
default:
return (EINVAL);
}
kn->kn_hook = (caddr_t)vp;
v_addpollinfo(vp);
if (vp->v_pollinfo == NULL)
return (ENOMEM);
knl = &vp->v_pollinfo->vpi_selinfo.si_note;
knlist_add(knl, kn, 0);
return (0);
}
/*
* Detach knote from vnode
*/
static void
filt_vfsdetach(struct knote *kn)
{
struct vnode *vp = (struct vnode *)kn->kn_hook;
KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
}
/*ARGSUSED*/
static int
filt_vfsread(struct knote *kn, long hint)
{
struct vnode *vp = (struct vnode *)kn->kn_hook;
struct vattr va;
int res;
/*
* filesystem is gone, so set the EOF flag and schedule
* the knote for deletion.
*/
if (hint == NOTE_REVOKE) {
VI_LOCK(vp);
kn->kn_flags |= (EV_EOF | EV_ONESHOT);
VI_UNLOCK(vp);
return (1);
}
if (VOP_GETATTR(vp, &va, curthread->td_ucred))
return (0);
VI_LOCK(vp);
kn->kn_data = va.va_size - kn->kn_fp->f_offset;
res = (kn->kn_data != 0);
VI_UNLOCK(vp);
return (res);
}
/*ARGSUSED*/
static int
filt_vfswrite(struct knote *kn, long hint)
{
struct vnode *vp = (struct vnode *)kn->kn_hook;
VI_LOCK(vp);
/*
* filesystem is gone, so set the EOF flag and schedule
* the knote for deletion.
*/
if (hint == NOTE_REVOKE)
kn->kn_flags |= (EV_EOF | EV_ONESHOT);
kn->kn_data = 0;
VI_UNLOCK(vp);
return (1);
}
static int
filt_vfsvnode(struct knote *kn, long hint)
{
struct vnode *vp = (struct vnode *)kn->kn_hook;
int res;
VI_LOCK(vp);
if (kn->kn_sfflags & hint)
kn->kn_fflags |= hint;
if (hint == NOTE_REVOKE) {
kn->kn_flags |= EV_EOF;
VI_UNLOCK(vp);
return (1);
}
res = (kn->kn_fflags != 0);
VI_UNLOCK(vp);
return (res);
}
int
vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
{
int error;
if (dp->d_reclen > ap->a_uio->uio_resid)
return (ENAMETOOLONG);
error = uiomove(dp, dp->d_reclen, ap->a_uio);
if (error) {
if (ap->a_ncookies != NULL) {
if (ap->a_cookies != NULL)
free(ap->a_cookies, M_TEMP);
ap->a_cookies = NULL;
*ap->a_ncookies = 0;
}
return (error);
}
if (ap->a_ncookies == NULL)
return (0);
KASSERT(ap->a_cookies,
("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
*ap->a_cookies = realloc(*ap->a_cookies,
(*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
(*ap->a_cookies)[*ap->a_ncookies] = off;
return (0);
}
/*
* Mark for update the access time of the file if the filesystem
* supports VOP_MARKATIME. This functionality is used by execve and
* mmap, so we want to avoid the I/O implied by directly setting
* va_atime for the sake of efficiency.
*/
void
vfs_mark_atime(struct vnode *vp, struct ucred *cred)
{
struct mount *mp;
mp = vp->v_mount;
VFS_ASSERT_GIANT(mp);
ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
(void)VOP_MARKATIME(vp);
}
/*
* The purpose of this routine is to remove granularity from accmode_t,
* reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
* VADMIN and VAPPEND.
*
* If it returns 0, the caller is supposed to continue with the usual
* access checks using 'accmode' as modified by this routine. If it
* returns nonzero value, the caller is supposed to return that value
* as errno.
*
* Note that after this routine runs, accmode may be zero.
*/
int
vfs_unixify_accmode(accmode_t *accmode)
{
/*
* There is no way to specify explicit "deny" rule using
* file mode or POSIX.1e ACLs.
*/
if (*accmode & VEXPLICIT_DENY) {
*accmode = 0;
return (0);
}
/*
* None of these can be translated into usual access bits.
* Also, the common case for NFSv4 ACLs is to not contain
* either of these bits. Caller should check for VWRITE
* on the containing directory instead.
*/
if (*accmode & (VDELETE_CHILD | VDELETE))
return (EPERM);
if (*accmode & VADMIN_PERMS) {
*accmode &= ~VADMIN_PERMS;
*accmode |= VADMIN;
}
/*
* There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
* or VSYNCHRONIZE using file mode or POSIX.1e ACL.
*/
*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
return (0);
}
Index: stable/8/sys
===================================================================
--- stable/8/sys (revision 209264)
+++ stable/8/sys (revision 209265)
Property changes on: stable/8/sys
___________________________________________________________________
Modified: svn:mergeinfo
## -0,0 +0,1 ##
Merged /head/sys:r209260-209261

File Metadata

Mime Type
text/x-c
Expires
Sun, Mar 22, 3:34 AM (1 d, 16 h)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
30057908
Default Alt Text
(179 KB)

Event Timeline