*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
* Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
*/
#include <sys/ddt.h>
#include <sys/blkptr.h>
#include <sys/zfeature.h>
+#include <sys/dsl_scan.h>
#include <sys/metaslab_impl.h>
#include <sys/time.h>
#include <sys/trace_zio.h>
#include <sys/abd.h>
+#include <sys/dsl_crypt.h>
/*
* ==========================================================================
"z_null", "z_rd", "z_wr", "z_fr", "z_cl", "z_ioctl"
};
-int zio_dva_throttle_enabled = B_FALSE;
+int zio_dva_throttle_enabled = B_TRUE;
/*
* ==========================================================================
*/
align = 8 * SPA_MINBLOCKSIZE;
#else
- if (size <= 4 * SPA_MINBLOCKSIZE) {
+ if (size < PAGESIZE) {
align = SPA_MINBLOCKSIZE;
} else if (IS_P2ALIGNED(size, p2 >> 2)) {
- align = MIN(p2 >> 2, PAGESIZE);
+ align = PAGESIZE;
}
#endif
kmem_cache_free(zio_data_buf_cache[c], buf);
}
+static void
+zio_abd_free(void *abd, size_t size)
+{
+ abd_free((abd_t *)abd);
+}
+
/*
* ==========================================================================
* Push and pop I/O transform buffers
*/
void
zio_push_transform(zio_t *zio, abd_t *data, uint64_t size, uint64_t bufsize,
- zio_transform_func_t *transform)
+ zio_transform_func_t *transform)
{
zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP);
/*
* ==========================================================================
- * I/O transform callbacks for subblocks and decompression
+ * I/O transform callbacks for subblocks, decompression, and decryption
* ==========================================================================
*/
static void
}
}
+static void
+zio_decrypt(zio_t *zio, abd_t *data, uint64_t size)
+{
+ int ret;
+ void *tmp;
+ blkptr_t *bp = zio->io_bp;
+ spa_t *spa = zio->io_spa;
+ uint64_t dsobj = zio->io_bookmark.zb_objset;
+ uint64_t lsize = BP_GET_LSIZE(bp);
+ dmu_object_type_t ot = BP_GET_TYPE(bp);
+ uint8_t salt[ZIO_DATA_SALT_LEN];
+ uint8_t iv[ZIO_DATA_IV_LEN];
+ uint8_t mac[ZIO_DATA_MAC_LEN];
+ boolean_t no_crypt = B_FALSE;
+
+ ASSERT(BP_USES_CRYPT(bp));
+ ASSERT3U(size, !=, 0);
+
+ if (zio->io_error != 0)
+ return;
+
+ /*
+ * Verify the cksum of MACs stored in an indirect bp. It will always
+ * be possible to verify this since it does not require an encryption
+ * key.
+ */
+ if (BP_HAS_INDIRECT_MAC_CKSUM(bp)) {
+ zio_crypt_decode_mac_bp(bp, mac);
+
+ if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF) {
+ /*
+ * We haven't decompressed the data yet, but
+ * zio_crypt_do_indirect_mac_checksum() requires
+ * decompressed data to be able to parse out the MACs
+ * from the indirect block. We decompress it now and
+ * throw away the result after we are finished.
+ */
+ tmp = zio_buf_alloc(lsize);
+ ret = zio_decompress_data(BP_GET_COMPRESS(bp),
+ zio->io_abd, tmp, zio->io_size, lsize);
+ if (ret != 0) {
+ ret = SET_ERROR(EIO);
+ goto error;
+ }
+ ret = zio_crypt_do_indirect_mac_checksum(B_FALSE,
+ tmp, lsize, BP_SHOULD_BYTESWAP(bp), mac);
+ zio_buf_free(tmp, lsize);
+ } else {
+ ret = zio_crypt_do_indirect_mac_checksum_abd(B_FALSE,
+ zio->io_abd, size, BP_SHOULD_BYTESWAP(bp), mac);
+ }
+ abd_copy(data, zio->io_abd, size);
+
+ if (ret != 0)
+ goto error;
+
+ return;
+ }
+
+ /*
+ * If this is an authenticated block, just check the MAC. It would be
+ * nice to separate this out into its own flag, but for the moment
+ * enum zio_flag is out of bits.
+ */
+ if (BP_IS_AUTHENTICATED(bp)) {
+ if (ot == DMU_OT_OBJSET) {
+ ret = spa_do_crypt_objset_mac_abd(B_FALSE, spa,
+ dsobj, zio->io_abd, size, BP_SHOULD_BYTESWAP(bp));
+ } else {
+ zio_crypt_decode_mac_bp(bp, mac);
+ ret = spa_do_crypt_mac_abd(B_FALSE, spa, dsobj,
+ zio->io_abd, size, mac);
+ }
+ abd_copy(data, zio->io_abd, size);
+
+ if (ret != 0)
+ goto error;
+
+ return;
+ }
+
+ zio_crypt_decode_params_bp(bp, salt, iv);
+
+ if (ot == DMU_OT_INTENT_LOG) {
+ tmp = abd_borrow_buf_copy(zio->io_abd, sizeof (zil_chain_t));
+ zio_crypt_decode_mac_zil(tmp, mac);
+ abd_return_buf(zio->io_abd, tmp, sizeof (zil_chain_t));
+ } else {
+ zio_crypt_decode_mac_bp(bp, mac);
+ }
+
+ ret = spa_do_crypt_abd(B_FALSE, spa, dsobj, bp, bp->blk_birth,
+ size, data, zio->io_abd, iv, mac, salt, &no_crypt);
+ if (no_crypt)
+ abd_copy(data, zio->io_abd, size);
+
+ if (ret != 0)
+ goto error;
+
+ return;
+
+error:
+ /* assert that the key was found unless this was speculative */
+ ASSERT(ret != ENOENT || (zio->io_flags & ZIO_FLAG_SPECULATIVE));
+
+ /*
+ * If there was a decryption / authentication error return EIO as
+ * the io_error. If this was not a speculative zio, create an ereport.
+ */
+ if (ret == ECKSUM) {
+ zio->io_error = SET_ERROR(EIO);
+ if ((zio->io_flags & ZIO_FLAG_SPECULATIVE) == 0) {
+ zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION,
+ spa, NULL, &zio->io_bookmark, zio, 0, 0);
+ }
+ } else {
+ zio->io_error = ret;
+ }
+}
+
/*
* ==========================================================================
* I/O parent/child relationships and pipeline interlocks
{
list_t *cl = &pio->io_child_list;
+ ASSERT(MUTEX_HELD(&pio->io_lock));
+
*zl = (*zl == NULL) ? list_head(cl) : list_next(cl, *zl);
if (*zl == NULL)
return (NULL);
zio_add_child(zio_t *pio, zio_t *cio)
{
zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP);
- int w;
/*
* Logical I/Os can have logical, gang, or vdev children.
* Vdev I/Os can only have vdev children.
* The following ASSERT captures all of these constraints.
*/
- ASSERT(cio->io_child_type <= pio->io_child_type);
+ ASSERT3S(cio->io_child_type, <=, pio->io_child_type);
zl->zl_parent = pio;
zl->zl_child = cio;
- mutex_enter(&cio->io_lock);
mutex_enter(&pio->io_lock);
+ mutex_enter(&cio->io_lock);
ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0);
- for (w = 0; w < ZIO_WAIT_TYPES; w++)
+ for (int w = 0; w < ZIO_WAIT_TYPES; w++)
pio->io_children[cio->io_child_type][w] += !cio->io_state[w];
list_insert_head(&pio->io_child_list, zl);
pio->io_child_count++;
cio->io_parent_count++;
- mutex_exit(&pio->io_lock);
mutex_exit(&cio->io_lock);
+ mutex_exit(&pio->io_lock);
}
static void
ASSERT(zl->zl_parent == pio);
ASSERT(zl->zl_child == cio);
- mutex_enter(&cio->io_lock);
mutex_enter(&pio->io_lock);
+ mutex_enter(&cio->io_lock);
list_remove(&pio->io_child_list, zl);
list_remove(&cio->io_parent_list, zl);
pio->io_child_count--;
cio->io_parent_count--;
- mutex_exit(&pio->io_lock);
mutex_exit(&cio->io_lock);
+ mutex_exit(&pio->io_lock);
kmem_cache_free(zio_link_cache, zl);
}
static boolean_t
-zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait)
+zio_wait_for_children(zio_t *zio, uint8_t childbits, 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 >>= 1;
- ASSERT3U(zio->io_stage, !=, ZIO_STAGE_OPEN);
- zio->io_stall = countp;
- waiting = B_TRUE;
+ for (int c = 0; c < ZIO_CHILD_TYPES; c++) {
+ if (!(ZIO_CHILD_BIT_IS_SET(childbits, c)))
+ continue;
+
+ uint64_t *countp = &zio->io_children[c][wait];
+ if (*countp != 0) {
+ zio->io_stage >>= 1;
+ ASSERT3U(zio->io_stage, !=, ZIO_STAGE_OPEN);
+ zio->io_stall = countp;
+ waiting = B_TRUE;
+ break;
+ }
}
mutex_exit(&zio->io_lock);
-
return (waiting);
}
}
int
-zio_timestamp_compare(const void *x1, const void *x2)
+zio_bookmark_compare(const void *x1, const void *x2)
{
const zio_t *z1 = x1;
const zio_t *z2 = x2;
- int cmp;
- cmp = AVL_CMP(z1->io_queued_timestamp, z2->io_queued_timestamp);
- if (likely(cmp))
- return (cmp);
+ if (z1->io_bookmark.zb_objset < z2->io_bookmark.zb_objset)
+ return (-1);
+ if (z1->io_bookmark.zb_objset > z2->io_bookmark.zb_objset)
+ return (1);
+
+ if (z1->io_bookmark.zb_object < z2->io_bookmark.zb_object)
+ return (-1);
+ if (z1->io_bookmark.zb_object > z2->io_bookmark.zb_object)
+ return (1);
+
+ if (z1->io_bookmark.zb_level < z2->io_bookmark.zb_level)
+ return (-1);
+ if (z1->io_bookmark.zb_level > z2->io_bookmark.zb_level)
+ return (1);
- cmp = AVL_CMP(z1->io_offset, z2->io_offset);
- if (likely(cmp))
- return (cmp);
+ if (z1->io_bookmark.zb_blkid < z2->io_bookmark.zb_blkid)
+ return (-1);
+ if (z1->io_bookmark.zb_blkid > z2->io_bookmark.zb_blkid)
+ return (1);
- return (AVL_PCMP(z1, z2));
+ if (z1 < z2)
+ return (-1);
+ if (z1 > z2)
+ return (1);
+
+ return (0);
}
/*
ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER));
ASSERT(vd || stage == ZIO_STAGE_OPEN);
- IMPLY(lsize != psize, (flags & ZIO_FLAG_RAW) != 0);
+ IMPLY(lsize != psize, (flags & ZIO_FLAG_RAW_COMPRESS) != 0);
zio = kmem_cache_alloc(zio_cache, KM_SLEEP);
bzero(zio, sizeof (zio_t));
offsetof(zio_link_t, zl_parent_node));
list_create(&zio->io_child_list, sizeof (zio_link_t),
offsetof(zio_link_t, zl_child_node));
+ metaslab_trace_init(&zio->io_alloc_list);
if (vd != NULL)
zio->io_child_type = ZIO_CHILD_VDEV;
static void
zio_destroy(zio_t *zio)
{
+ metaslab_trace_fini(&zio->io_alloc_list);
list_destroy(&zio->io_parent_list);
list_destroy(&zio->io_child_list);
mutex_destroy(&zio->io_lock);
void
zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp)
{
- int i;
-
if (!DMU_OT_IS_VALID(BP_GET_TYPE(bp))) {
zfs_panic_recover("blkptr at %p has invalid TYPE %llu",
bp, (longlong_t)BP_GET_TYPE(bp));
* allows the birth time of log blocks (and dmu_sync()-ed blocks
* that are in the log) to be arbitrarily large.
*/
- for (i = 0; i < BP_GET_NDVAS(bp); i++) {
+ for (int i = 0; i < BP_GET_NDVAS(bp); i++) {
uint64_t vdevid = DVA_GET_VDEV(&bp->blk_dva[i]);
- vdev_t *vd;
- uint64_t offset, asize;
+
if (vdevid >= spa->spa_root_vdev->vdev_children) {
zfs_panic_recover("blkptr at %p DVA %u has invalid "
"VDEV %llu",
bp, i, (longlong_t)vdevid);
continue;
}
- vd = spa->spa_root_vdev->vdev_child[vdevid];
+ vdev_t *vd = spa->spa_root_vdev->vdev_child[vdevid];
if (vd == NULL) {
zfs_panic_recover("blkptr at %p DVA %u has invalid "
"VDEV %llu",
*/
continue;
}
- offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
- asize = DVA_GET_ASIZE(&bp->blk_dva[i]);
+ uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
+ uint64_t asize = DVA_GET_ASIZE(&bp->blk_dva[i]);
if (BP_IS_GANG(bp))
asize = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE);
if (offset + asize > vd->vdev_asize) {
* Data can be NULL if we are going to call zio_write_override() to
* provide the already-allocated BP. But we may need the data to
* verify a dedup hit (if requested). In this case, don't try to
- * dedup (just take the already-allocated BP verbatim).
+ * dedup (just take the already-allocated BP verbatim). Encrypted
+ * dedup blocks need data as well so we also disable dedup in this
+ * case.
*/
- if (data == NULL && zio->io_prop.zp_dedup_verify) {
+ if (data == NULL &&
+ (zio->io_prop.zp_dedup_verify || zio->io_prop.zp_encrypt)) {
zio->io_prop.zp_dedup = zio->io_prop.zp_dedup_verify = B_FALSE;
}
zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp)
{
+ zfs_blkptr_verify(spa, bp);
+
/*
* The check for EMBEDDED is a performance optimization. We
* process the free here (by ignoring it) rather than
metaslab_check_free(spa, bp);
arc_freed(spa, bp);
+ dsl_scan_freed(spa, bp);
/*
* GANG and DEDUP blocks can induce a read (for the gang block header,
{
zio_t *zio;
- dprintf_bp(bp, "claiming in txg %llu", txg);
+ zfs_blkptr_verify(spa, bp);
if (BP_IS_EMBEDDED(bp))
return (zio_null(pio, spa, NULL, NULL, NULL, 0));
*/
zio_t *
zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset,
- abd_t *data, uint64_t size, int type, zio_priority_t priority,
- enum zio_flag flags, zio_done_func_t *done, void *private)
+ abd_t *data, uint64_t size, int type, zio_priority_t priority,
+ enum zio_flag flags, zio_done_func_t *done, void *private)
{
enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE;
zio_t *zio;
- ASSERT(vd->vdev_parent ==
- (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev));
+ /*
+ * vdev child I/Os do not propagate their error to the parent.
+ * Therefore, for correct operation the caller *must* check for
+ * and handle the error in the child i/o's done callback.
+ * The only exceptions are i/os that we don't care about
+ * (OPTIONAL or REPAIR).
+ */
+ ASSERT((flags & ZIO_FLAG_OPTIONAL) || (flags & ZIO_FLAG_IO_REPAIR) ||
+ done != NULL);
+
+ /*
+ * In the common case, where the parent zio was to a normal vdev,
+ * the child zio must be to a child vdev of that vdev. Otherwise,
+ * the child zio must be to a top-level vdev.
+ */
+ if (pio->io_vd != NULL && pio->io_vd->vdev_ops != &vdev_indirect_ops) {
+ ASSERT3P(vd->vdev_parent, ==, pio->io_vd);
+ } else {
+ ASSERT3P(vd, ==, vd->vdev_top);
+ }
if (type == ZIO_TYPE_READ && bp != NULL) {
/*
pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY;
}
- if (vd->vdev_children == 0)
+ if (vd->vdev_ops->vdev_op_leaf) {
+ ASSERT0(vd->vdev_children);
offset += VDEV_LABEL_START_SIZE;
+ }
- flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE;
+ flags |= ZIO_VDEV_CHILD_FLAGS(pio);
/*
* If we've decided to do a repair, the write is not speculative --
zio_t *
zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, abd_t *data, uint64_t size,
- int type, zio_priority_t priority, enum zio_flag flags,
- zio_done_func_t *done, void *private)
+ int type, zio_priority_t priority, enum zio_flag flags,
+ zio_done_func_t *done, void *private)
{
zio_t *zio;
void
zio_shrink(zio_t *zio, uint64_t size)
{
- ASSERT(zio->io_executor == NULL);
- ASSERT(zio->io_orig_size == zio->io_size);
- ASSERT(size <= zio->io_size);
+ ASSERT3P(zio->io_executor, ==, NULL);
+ ASSERT3U(zio->io_orig_size, ==, zio->io_size);
+ ASSERT3U(size, <=, zio->io_size);
/*
* We don't shrink for raidz because of problems with the
zio_read_bp_init(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
+ uint64_t psize =
+ BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp);
+
+ ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy);
if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF &&
zio->io_child_type == ZIO_CHILD_LOGICAL &&
- !(zio->io_flags & ZIO_FLAG_RAW)) {
- uint64_t psize =
- BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp);
+ !(zio->io_flags & ZIO_FLAG_RAW_COMPRESS)) {
zio_push_transform(zio, abd_alloc_sametype(zio->io_abd, psize),
psize, psize, zio_decompress);
}
+ if (((BP_IS_PROTECTED(bp) && !(zio->io_flags & ZIO_FLAG_RAW_ENCRYPT)) ||
+ BP_HAS_INDIRECT_MAC_CKSUM(bp)) &&
+ zio->io_child_type == ZIO_CHILD_LOGICAL) {
+ zio_push_transform(zio, abd_alloc_sametype(zio->io_abd, psize),
+ psize, psize, zio_decrypt);
+ }
+
if (BP_IS_EMBEDDED(bp) && BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA) {
int psize = BPE_GET_PSIZE(bp);
void *data = abd_borrow_buf(zio->io_abd, psize);
abd_return_buf_copy(zio->io_abd, data, psize);
} else {
ASSERT(!BP_IS_EMBEDDED(bp));
+ ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy);
}
if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0)
static int
zio_write_bp_init(zio_t *zio)
{
-
if (!IO_IS_ALLOCATING(zio))
return (ZIO_PIPELINE_CONTINUE);
ASSERT((zio_checksum_table[zp->zp_checksum].ci_flags &
ZCHECKSUM_FLAG_DEDUP) || zp->zp_dedup_verify);
- if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) {
+ if (BP_GET_CHECKSUM(bp) == zp->zp_checksum &&
+ !zp->zp_encrypt) {
BP_SET_DEDUP(bp, 1);
zio->io_pipeline |= ZIO_STAGE_DDT_WRITE;
return (ZIO_PIPELINE_CONTINUE);
uint64_t psize = zio->io_size;
int pass = 1;
- EQUIV(lsize != psize, (zio->io_flags & ZIO_FLAG_RAW) != 0);
-
/*
* 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))
+ if (zio_wait_for_children(zio, ZIO_CHILD_LOGICAL_BIT |
+ ZIO_CHILD_GANG_BIT, ZIO_WAIT_READY)) {
return (ZIO_PIPELINE_STOP);
+ }
if (!IO_IS_ALLOCATING(zio))
return (ZIO_PIPELINE_CONTINUE);
}
/* If it's a compressed write that is not raw, compress the buffer. */
- if (compress != ZIO_COMPRESS_OFF && psize == lsize) {
+ if (compress != ZIO_COMPRESS_OFF &&
+ !(zio->io_flags & ZIO_FLAG_RAW_COMPRESS)) {
void *cbuf = zio_buf_alloc(lsize);
psize = zio_compress_data(compress, zio->io_abd, cbuf, lsize);
if (psize == 0 || psize == lsize) {
compress = ZIO_COMPRESS_OFF;
zio_buf_free(cbuf, lsize);
- } else if (!zp->zp_dedup && psize <= BPE_PAYLOAD_SIZE &&
+ } else if (!zp->zp_dedup && !zp->zp_encrypt &&
+ psize <= BPE_PAYLOAD_SIZE &&
zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) &&
spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) {
encode_embedded_bp_compressed(bp,
* in that we charge for the padding used to fill out
* the last sector.
*/
- size_t rounded;
-
ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
-
- rounded = (size_t)P2ROUNDUP(psize,
+ size_t rounded = (size_t)P2ROUNDUP(psize,
1ULL << spa->spa_min_ashift);
if (rounded >= lsize) {
compress = ZIO_COMPRESS_OFF;
*bp = zio->io_bp_orig;
zio->io_pipeline = zio->io_orig_pipeline;
+ } else if ((zio->io_flags & ZIO_FLAG_RAW_ENCRYPT) != 0 &&
+ zp->zp_type == DMU_OT_DNODE) {
+ /*
+ * The DMU actually relies on the zio layer's compression
+ * to free metadnode blocks that have had all contained
+ * dnodes freed. As a result, even when doing a raw
+ * receive, we must check whether the block can be compressed
+ * to a hole.
+ */
+ psize = zio_compress_data(ZIO_COMPRESS_EMPTY,
+ zio->io_abd, NULL, lsize);
+ if (psize == 0)
+ compress = ZIO_COMPRESS_OFF;
} else {
ASSERT3U(psize, !=, 0);
-
}
/*
if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg &&
BP_GET_PSIZE(bp) == psize &&
pass >= zfs_sync_pass_rewrite) {
- enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES;
ASSERT(psize != 0);
+ enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES;
zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages;
zio->io_flags |= ZIO_FLAG_IO_REWRITE;
} else {
if (zp->zp_dedup) {
ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE));
+ ASSERT(!zp->zp_encrypt ||
+ DMU_OT_IS_ENCRYPTED(zp->zp_type));
zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE;
}
if (zp->zp_nopwrite) {
zio->io_pipeline = ZIO_DDT_FREE_PIPELINE;
}
+ ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy);
+
return (ZIO_PIPELINE_CONTINUE);
}
{
kthread_t *executor = zio->io_executor;
spa_t *spa = zio->io_spa;
- zio_type_t t;
- for (t = 0; t < ZIO_TYPES; t++) {
+ for (zio_type_t t = 0; t < ZIO_TYPES; t++) {
spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
uint_t i;
for (i = 0; i < tqs->stqs_count; i++) {
* OpenZFS's timeout_generic().
*/
tid = taskq_dispatch_delay(system_taskq,
- (task_func_t *) zio_interrupt,
+ (task_func_t *)zio_interrupt,
zio, TQ_NOSLEEP, expire_at_tick);
if (tid == TASKQID_INVALID) {
/*
zio_interrupt(zio);
}
+static void
+zio_deadman_impl(zio_t *pio)
+{
+ zio_t *cio, *cio_next;
+ zio_link_t *zl = NULL;
+ vdev_t *vd = pio->io_vd;
+
+ if (vd != NULL && vd->vdev_ops->vdev_op_leaf) {
+ vdev_queue_t *vq = &vd->vdev_queue;
+ zbookmark_phys_t *zb = &pio->io_bookmark;
+ uint64_t delta = gethrtime() - pio->io_timestamp;
+ uint64_t failmode = spa_get_deadman_failmode(pio->io_spa);
+
+ zfs_dbgmsg("slow zio: zio=%p timestamp=%llu "
+ "delta=%llu queued=%llu io=%llu "
+ "path=%s last=%llu "
+ "type=%d priority=%d flags=0x%x "
+ "stage=0x%x pipeline=0x%x pipeline-trace=0x%x "
+ "objset=%llu object=%llu level=%llu blkid=%llu "
+ "offset=%llu size=%llu error=%d",
+ pio, pio->io_timestamp,
+ delta, pio->io_delta, pio->io_delay,
+ vd->vdev_path, vq->vq_io_complete_ts,
+ pio->io_type, pio->io_priority, pio->io_flags,
+ pio->io_state, pio->io_pipeline, pio->io_pipeline_trace,
+ zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
+ pio->io_offset, pio->io_size, pio->io_error);
+ zfs_ereport_post(FM_EREPORT_ZFS_DEADMAN,
+ pio->io_spa, vd, zb, pio, 0, 0);
+
+ if (failmode == ZIO_FAILURE_MODE_CONTINUE &&
+ taskq_empty_ent(&pio->io_tqent)) {
+ zio_interrupt(pio);
+ }
+ }
+
+ mutex_enter(&pio->io_lock);
+ for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) {
+ cio_next = zio_walk_children(pio, &zl);
+ zio_deadman_impl(cio);
+ }
+ mutex_exit(&pio->io_lock);
+}
+
+/*
+ * Log the critical information describing this zio and all of its children
+ * using the zfs_dbgmsg() interface then post deadman event for the ZED.
+ */
+void
+zio_deadman(zio_t *pio, char *tag)
+{
+ spa_t *spa = pio->io_spa;
+ char *name = spa_name(spa);
+
+ if (!zfs_deadman_enabled || spa_suspended(spa))
+ return;
+
+ zio_deadman_impl(pio);
+
+ switch (spa_get_deadman_failmode(spa)) {
+ case ZIO_FAILURE_MODE_WAIT:
+ zfs_dbgmsg("%s waiting for hung I/O to pool '%s'", tag, name);
+ break;
+
+ case ZIO_FAILURE_MODE_CONTINUE:
+ zfs_dbgmsg("%s restarting hung I/O for pool '%s'", tag, name);
+ break;
+
+ case ZIO_FAILURE_MODE_PANIC:
+ fm_panic("%s determined I/O to pool '%s' is hung.", tag, name);
+ break;
+ }
+}
+
/*
* Execute the I/O pipeline until one of the following occurs:
* (1) the I/O completes; (2) the pipeline stalls waiting for
int
zio_wait(zio_t *zio)
{
+ long timeout = MSEC_TO_TICK(zfs_deadman_ziotime_ms);
int error;
- ASSERT(zio->io_stage == ZIO_STAGE_OPEN);
- ASSERT(zio->io_executor == NULL);
+ ASSERT3S(zio->io_stage, ==, ZIO_STAGE_OPEN);
+ ASSERT3P(zio->io_executor, ==, NULL);
zio->io_waiter = curthread;
ASSERT0(zio->io_queued_timestamp);
__zio_execute(zio);
mutex_enter(&zio->io_lock);
- while (zio->io_executor != NULL)
- cv_wait_io(&zio->io_cv, &zio->io_lock);
+ while (zio->io_executor != NULL) {
+ error = cv_timedwait_io(&zio->io_cv, &zio->io_lock,
+ ddi_get_lbolt() + timeout);
+
+ if (zfs_deadman_enabled && error == -1 &&
+ gethrtime() - zio->io_queued_timestamp >
+ spa_deadman_ziotime(zio->io_spa)) {
+ mutex_exit(&zio->io_lock);
+ timeout = MSEC_TO_TICK(zfs_deadman_checktime_ms);
+ zio_deadman(zio, FTAG);
+ mutex_enter(&zio->io_lock);
+ }
+ }
mutex_exit(&zio->io_lock);
error = zio->io_error;
void
zio_nowait(zio_t *zio)
{
- ASSERT(zio->io_executor == NULL);
+ ASSERT3P(zio->io_executor, ==, NULL);
if (zio->io_child_type == ZIO_CHILD_LOGICAL &&
zio_unique_parent(zio) == NULL) {
/*
* ==========================================================================
- * Reexecute or suspend/resume failed I/O
+ * Reexecute, cancel, or suspend/resume failed I/O
* ==========================================================================
*/
zio_reexecute(zio_t *pio)
{
zio_t *cio, *cio_next;
- int c, w;
- zio_link_t *zl = NULL;
ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL);
ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN);
pio->io_flags |= ZIO_FLAG_REEXECUTED;
pio->io_pipeline_trace = 0;
pio->io_error = 0;
- for (w = 0; w < ZIO_WAIT_TYPES; w++)
+ for (int w = 0; w < ZIO_WAIT_TYPES; w++)
pio->io_state[w] = 0;
- for (c = 0; c < ZIO_CHILD_TYPES; c++)
+ for (int c = 0; c < ZIO_CHILD_TYPES; c++)
pio->io_child_error[c] = 0;
if (IO_IS_ALLOCATING(pio))
* the remainder of pio's io_child_list, from 'cio_next' onward,
* cannot be affected by any side effects of reexecuting 'cio'.
*/
+ zio_link_t *zl = NULL;
+ mutex_enter(&pio->io_lock);
for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) {
cio_next = zio_walk_children(pio, &zl);
- mutex_enter(&pio->io_lock);
- for (w = 0; w < ZIO_WAIT_TYPES; w++)
+ for (int w = 0; w < ZIO_WAIT_TYPES; w++)
pio->io_children[cio->io_child_type][w]++;
mutex_exit(&pio->io_lock);
zio_reexecute(cio);
+ mutex_enter(&pio->io_lock);
}
+ mutex_exit(&pio->io_lock);
/*
* Now that all children have been reexecuted, execute the parent.
* We don't reexecute "The Godfather" I/O here as it's the
- * responsibility of the caller to wait on him.
+ * responsibility of the caller to wait on it.
*/
if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) {
pio->io_queued_timestamp = gethrtime();
}
void
-zio_suspend(spa_t *spa, zio_t *zio)
+zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t reason)
{
if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC)
fm_panic("Pool '%s' has encountered an uncorrectable I/O "
cmn_err(CE_WARN, "Pool '%s' has encountered an uncorrectable I/O "
"failure and has been suspended.\n", spa_name(spa));
- zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0);
+ zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL,
+ NULL, NULL, 0, 0);
mutex_enter(&spa->spa_suspend_lock);
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
ZIO_FLAG_GODFATHER);
- spa->spa_suspended = B_TRUE;
+ spa->spa_suspended = reason;
if (zio != NULL) {
ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER));
* Reexecute all previously suspended i/o.
*/
mutex_enter(&spa->spa_suspend_lock);
- spa->spa_suspended = B_FALSE;
+ spa->spa_suspended = ZIO_SUSPEND_NONE;
cv_broadcast(&spa->spa_suspend_cv);
pio = spa->spa_suspend_zio_root;
spa->spa_suspend_zio_root = NULL;
zio_gang_node_free(zio_gang_node_t **gnpp)
{
zio_gang_node_t *gn = *gnpp;
- int g;
- for (g = 0; g < SPA_GBH_NBLKPTRS; g++)
+ for (int g = 0; g < SPA_GBH_NBLKPTRS; g++)
ASSERT(gn->gn_child[g] == NULL);
zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE);
zio_gang_tree_free(zio_gang_node_t **gnpp)
{
zio_gang_node_t *gn = *gnpp;
- int g;
if (gn == NULL)
return;
- for (g = 0; g < SPA_GBH_NBLKPTRS; g++)
+ for (int g = 0; g < SPA_GBH_NBLKPTRS; g++)
zio_gang_tree_free(&gn->gn_child[g]);
zio_gang_node_free(gnpp);
zio_t *gio = zio->io_gang_leader;
zio_gang_node_t *gn = zio->io_private;
blkptr_t *bp = zio->io_bp;
- int g;
ASSERT(gio == zio_unique_parent(zio));
ASSERT(zio->io_child_count == 0);
abd_put(zio->io_abd);
- for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
+ 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_t *gio = pio->io_gang_leader;
zio_t *zio;
- int g;
ASSERT(BP_IS_GANG(bp) == !!gn);
ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp));
if (gn != NULL) {
ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC);
- for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
+ for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
if (BP_IS_HOLE(gbp))
continue;
{
blkptr_t *bp = zio->io_bp;
- if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE))
+ if (zio_wait_for_children(zio, ZIO_CHILD_GANG_BIT, ZIO_WAIT_DONE)) {
return (ZIO_PIPELINE_STOP);
+ }
ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio);
ASSERT(zio->io_child_type > ZIO_CHILD_GANG);
dva_t *cdva = zio->io_bp->blk_dva;
dva_t *pdva = pio->io_bp->blk_dva;
uint64_t asize;
- int d;
ASSERTV(zio_t *gio = zio->io_gang_leader);
if (BP_IS_HOLE(zio->io_bp))
ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp));
mutex_enter(&pio->io_lock);
- for (d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) {
+ 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]);
uint64_t resid = pio->io_size;
uint64_t lsize;
int copies = gio->io_prop.zp_copies;
- int gbh_copies = MIN(copies + 1, spa_max_replication(spa));
+ int gbh_copies;
zio_prop_t zp;
- int g, error;
+ int error;
+
+ /*
+ * encrypted blocks need DVA[2] free so encrypted gang headers can't
+ * have a third copy.
+ */
+ gbh_copies = MIN(copies + 1, spa_max_replication(spa));
+ if (gio->io_prop.zp_encrypt && gbh_copies >= SPA_DVAS_PER_BP)
+ gbh_copies = SPA_DVAS_PER_BP - 1;
int flags = METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER;
if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) {
}
error = metaslab_alloc(spa, mc, SPA_GANGBLOCKSIZE,
- bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, flags, pio);
+ bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, flags,
+ &pio->io_alloc_list, pio);
if (error) {
if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) {
ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE);
/*
* Create and nowait the gang children.
*/
- for (g = 0; resid != 0; resid -= lsize, g++) {
- zio_t *cio;
-
+ 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_dedup = B_FALSE;
zp.zp_dedup_verify = B_FALSE;
zp.zp_nopwrite = B_FALSE;
+ zp.zp_encrypt = gio->io_prop.zp_encrypt;
+ zp.zp_byteorder = gio->io_prop.zp_byteorder;
+ bzero(zp.zp_salt, ZIO_DATA_SALT_LEN);
+ bzero(zp.zp_iv, ZIO_DATA_IV_LEN);
+ bzero(zp.zp_mac, ZIO_DATA_MAC_LEN);
- cio = zio_write(zio, spa, txg, &gbh->zg_blkptr[g],
+ zio_t *cio = zio_write(zio, spa, txg, &gbh->zg_blkptr[g],
abd_get_offset(pio->io_abd, pio->io_size - resid), lsize,
lsize, &zp, zio_write_gang_member_ready, NULL, NULL,
zio_write_gang_done, &gn->gn_child[g], pio->io_priority,
if (BP_IS_HOLE(bp_orig) ||
!(zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_flags &
ZCHECKSUM_FLAG_NOPWRITE) ||
+ BP_IS_ENCRYPTED(bp) || BP_IS_ENCRYPTED(bp_orig) ||
BP_GET_CHECKSUM(bp) != BP_GET_CHECKSUM(bp_orig) ||
BP_GET_COMPRESS(bp) != BP_GET_COMPRESS(bp_orig) ||
BP_GET_DEDUP(bp) != BP_GET_DEDUP(bp_orig) ||
zio_ddt_read_start(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
- int p;
ASSERT(BP_GET_DEDUP(bp));
ASSERT(BP_GET_PSIZE(bp) == zio->io_size);
if (ddp_self == NULL)
return (ZIO_PIPELINE_CONTINUE);
- for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
+ for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
if (ddp->ddp_phys_birth == 0 || ddp == ddp_self)
continue;
ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp,
{
blkptr_t *bp = zio->io_bp;
- if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE))
+ if (zio_wait_for_children(zio, ZIO_CHILD_DDT_BIT, ZIO_WAIT_DONE)) {
return (ZIO_PIPELINE_STOP);
+ }
ASSERT(BP_GET_DEDUP(bp));
ASSERT(BP_GET_PSIZE(bp) == zio->io_size);
zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde)
{
spa_t *spa = zio->io_spa;
- int p;
boolean_t do_raw = !!(zio->io_flags & ZIO_FLAG_RAW);
ASSERT(!(zio->io_bp_override && do_raw));
* pushed the I/O transforms. That's an important optimization
* because otherwise we'd compress/encrypt all dmu_sync() data twice.
* However, we should never get a raw, override zio so in these
- * cases we can compare the io_data directly. This is useful because
+ * cases we can compare the io_abd directly. This is useful because
* it allows us to do dedup verification even if we don't have access
* to the original data (for instance, if the encryption keys aren't
* loaded).
*/
- for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
+ for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
zio_t *lio = dde->dde_lead_zio[p];
if (lio != NULL && do_raw) {
}
}
- for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
+ for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
ddt_phys_t *ddp = &dde->dde_phys[p];
if (ddp->ddp_phys_birth != 0 && do_raw) {
ddt_entry_t *dde = zio->io_private;
ddt_phys_t *ddp = &dde->dde_phys[p];
zio_t *pio;
- zio_link_t *zl;
if (zio->io_error)
return;
ddt_phys_fill(ddp, zio->io_bp);
- zl = NULL;
+ zio_link_t *zl = NULL;
while ((pio = zio_walk_parents(zio, &zl)) != NULL)
ddt_bp_fill(ddp, pio->io_bp, zio->io_txg);
zio_ddt_ditto_write_done(zio_t *zio)
{
int p = DDT_PHYS_DITTO;
+ ASSERTV(zio_prop_t *zp = &zio->io_prop);
blkptr_t *bp = zio->io_bp;
ddt_t *ddt = ddt_select(zio->io_spa, bp);
ddt_entry_t *dde = zio->io_private;
ddt_phys_t *ddp = &dde->dde_phys[p];
ddt_key_t *ddk = &dde->dde_key;
- ASSERTV(zio_prop_t *zp = &zio->io_prop);
ddt_enter(ddt);
return (ZIO_PIPELINE_CONTINUE);
if (nio != NULL) {
- ASSERT3U(nio->io_queued_timestamp, <=,
- zio->io_queued_timestamp);
ASSERT(nio->io_stage == ZIO_STAGE_DVA_THROTTLE);
/*
* We are passing control to a new zio so make sure that
flags |= METASLAB_ASYNC_ALLOC;
error = metaslab_alloc(spa, mc, zio->io_size, bp,
- zio->io_prop.zp_copies, zio->io_txg, NULL, flags, zio);
+ zio->io_prop.zp_copies, zio->io_txg, NULL, flags,
+ &zio->io_alloc_list, zio);
if (error != 0) {
spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, "
static void
zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp)
{
- int g;
-
ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp));
ASSERT(zio->io_bp_override == NULL);
metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE);
if (gn != NULL) {
- for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
+ 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_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, uint64_t size,
- boolean_t use_slog)
+zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, blkptr_t *new_bp,
+ uint64_t size, boolean_t *slog)
{
int error = 1;
+ zio_alloc_list_t io_alloc_list;
ASSERT(txg > spa_syncing_txg(spa));
- if (use_slog) {
- error = metaslab_alloc(spa, spa_log_class(spa), size,
- new_bp, 1, txg, NULL, METASLAB_FASTWRITE, NULL);
- }
-
- if (error) {
+ metaslab_trace_init(&io_alloc_list);
+ error = metaslab_alloc(spa, spa_log_class(spa), size, new_bp, 1,
+ txg, NULL, METASLAB_FASTWRITE, &io_alloc_list, NULL);
+ if (error == 0) {
+ *slog = TRUE;
+ } else {
error = metaslab_alloc(spa, spa_normal_class(spa), size,
- new_bp, 1, txg, NULL, METASLAB_FASTWRITE, NULL);
+ new_bp, 1, txg, NULL, METASLAB_FASTWRITE,
+ &io_alloc_list, NULL);
+ if (error == 0)
+ *slog = FALSE;
}
+ metaslab_trace_fini(&io_alloc_list);
if (error == 0) {
BP_SET_LSIZE(new_bp, size);
BP_SET_LEVEL(new_bp, 0);
BP_SET_DEDUP(new_bp, 0);
BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER);
+
+ /*
+ * encrypted blocks will require an IV and salt. We generate
+ * these now since we will not be rewriting the bp at
+ * rewrite time.
+ */
+ if (os->os_encrypted) {
+ uint8_t iv[ZIO_DATA_IV_LEN];
+ uint8_t salt[ZIO_DATA_SALT_LEN];
+
+ BP_SET_CRYPT(new_bp, B_TRUE);
+ VERIFY0(spa_crypt_get_salt(spa,
+ dmu_objset_id(os), salt));
+ VERIFY0(zio_crypt_generate_iv(iv));
+
+ zio_crypt_encode_params_bp(new_bp, salt, iv);
+ }
+ } else {
+ zfs_dbgmsg("%s: zil block allocation failure: "
+ "size %llu, error %d", spa_name(spa), size, error);
}
return (error);
}
ASSERT3P(zio->io_logical, !=, zio);
-
- /*
- * We keep track of time-sensitive I/Os so that the scan thread
- * can quickly react to certain workloads. In particular, we care
- * about non-scrubbing, top-level reads and writes with the following
- * characteristics:
- * - synchronous writes of user data to non-slog devices
- * - any reads of user data
- * When these conditions are met, adjust the timestamp of spa_last_io
- * which allows the scan thread to adjust its workload accordingly.
- */
- if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL &&
- vd == vd->vdev_top && !vd->vdev_islog &&
- zio->io_bookmark.zb_objset != DMU_META_OBJSET &&
- zio->io_txg != spa_syncing_txg(spa)) {
- uint64_t old = spa->spa_last_io;
- uint64_t new = ddi_get_lbolt64();
- if (old != new)
- (void) atomic_cas_64(&spa->spa_last_io, old, new);
+ if (zio->io_type == ZIO_TYPE_WRITE && zio->io_vd->vdev_removing) {
+ /*
+ * Note: the code can handle other kinds of writes,
+ * but we don't expect them.
+ */
+ ASSERT(zio->io_flags &
+ (ZIO_FLAG_PHYSICAL | ZIO_FLAG_SELF_HEAL |
+ ZIO_FLAG_INDUCE_DAMAGE));
}
align = 1ULL << vd->vdev_top->vdev_ashift;
zio_interrupt(zio);
return (ZIO_PIPELINE_STOP);
}
+ zio->io_delay = gethrtime();
}
- zio->io_delay = gethrtime();
vd->vdev_ops->vdev_op_io_start(zio);
return (ZIO_PIPELINE_STOP);
}
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))
+ if (zio_wait_for_children(zio, ZIO_CHILD_VDEV_BIT, ZIO_WAIT_DONE)) {
return (ZIO_PIPELINE_STOP);
+ }
ASSERT(zio->io_type == ZIO_TYPE_READ || 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,
- zio, EIO);
+ zio->io_error = zio_handle_device_injections(vd, zio,
+ EIO, EILSEQ);
if (zio_injection_enabled && zio->io_error == 0)
zio->io_error = zio_handle_label_injection(zio, EIO);
return (ZIO_PIPELINE_CONTINUE);
}
+/*
+ * This function is used to change the priority of an existing zio that is
+ * currently in-flight. This is used by the arc to upgrade priority in the
+ * event that a demand read is made for a block that is currently queued
+ * as a scrub or async read IO. Otherwise, the high priority read request
+ * would end up having to wait for the lower priority IO.
+ */
+void
+zio_change_priority(zio_t *pio, zio_priority_t priority)
+{
+ zio_t *cio, *cio_next;
+ zio_link_t *zl = NULL;
+
+ ASSERT3U(priority, <, ZIO_PRIORITY_NUM_QUEUEABLE);
+
+ if (pio->io_vd != NULL && pio->io_vd->vdev_ops->vdev_op_leaf) {
+ vdev_queue_change_io_priority(pio, priority);
+ } else {
+ pio->io_priority = priority;
+ }
+
+ mutex_enter(&pio->io_lock);
+ for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) {
+ cio_next = zio_walk_children(pio, &zl);
+ zio_change_priority(cio, priority);
+ }
+ mutex_exit(&pio->io_lock);
+}
+
/*
* For non-raidz ZIOs, we can just copy aside the bad data read from the
* disk, and use that to finish the checksum ereport later.
*/
static void
zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr,
- const void *good_buf)
+ const abd_t *good_buf)
{
/* no processing needed */
zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE);
void
zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored)
{
- void *buf = zio_buf_alloc(zio->io_size);
+ void *abd = abd_alloc_sametype(zio->io_abd, zio->io_size);
- abd_copy_to_buf(buf, zio->io_abd, zio->io_size);
+ abd_copy(abd, zio->io_abd, zio->io_size);
zcr->zcr_cbinfo = zio->io_size;
- zcr->zcr_cbdata = buf;
+ zcr->zcr_cbdata = abd;
zcr->zcr_finish = zio_vsd_default_cksum_finish;
- zcr->zcr_free = zio_buf_free;
+ zcr->zcr_free = zio_abd_free;
}
static int
{
vdev_t *vd = zio->io_vd;
- if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE))
+ if (zio_wait_for_children(zio, ZIO_CHILD_VDEV_BIT, 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);
vd->vdev_cant_write = B_TRUE;
}
+ /*
+ * If a cache flush returns ENOTSUP or ENOTTY, we know that no future
+ * attempts will ever succeed. In this case we set a persistent bit so
+ * that we don't bother with it in the future.
+ */
+ if ((zio->io_error == ENOTSUP || zio->io_error == ENOTTY) &&
+ zio->io_type == ZIO_TYPE_IOCTL &&
+ zio->io_cmd == DKIOCFLUSHWRITECACHE && vd != NULL)
+ vd->vdev_nowritecache = B_TRUE;
+
if (zio->io_error)
zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1;
}
+/*
+ * ==========================================================================
+ * Encrypt and store encryption parameters
+ * ==========================================================================
+ */
+
+
+/*
+ * This function is used for ZIO_STAGE_ENCRYPT. It is responsible for
+ * managing the storage of encryption parameters and passing them to the
+ * lower-level encryption functions.
+ */
+static int
+zio_encrypt(zio_t *zio)
+{
+ zio_prop_t *zp = &zio->io_prop;
+ spa_t *spa = zio->io_spa;
+ blkptr_t *bp = zio->io_bp;
+ uint64_t psize = BP_GET_PSIZE(bp);
+ uint64_t dsobj = zio->io_bookmark.zb_objset;
+ dmu_object_type_t ot = BP_GET_TYPE(bp);
+ void *enc_buf = NULL;
+ abd_t *eabd = NULL;
+ uint8_t salt[ZIO_DATA_SALT_LEN];
+ uint8_t iv[ZIO_DATA_IV_LEN];
+ uint8_t mac[ZIO_DATA_MAC_LEN];
+ boolean_t no_crypt = B_FALSE;
+
+ /* the root zio already encrypted the data */
+ if (zio->io_child_type == ZIO_CHILD_GANG)
+ return (ZIO_PIPELINE_CONTINUE);
+
+ /* only ZIL blocks are re-encrypted on rewrite */
+ if (!IO_IS_ALLOCATING(zio) && ot != DMU_OT_INTENT_LOG)
+ return (ZIO_PIPELINE_CONTINUE);
+
+ if (!(zp->zp_encrypt || BP_IS_ENCRYPTED(bp))) {
+ BP_SET_CRYPT(bp, B_FALSE);
+ return (ZIO_PIPELINE_CONTINUE);
+ }
+
+ /* if we are doing raw encryption set the provided encryption params */
+ if (zio->io_flags & ZIO_FLAG_RAW_ENCRYPT) {
+ ASSERT0(BP_GET_LEVEL(bp));
+ BP_SET_CRYPT(bp, B_TRUE);
+ BP_SET_BYTEORDER(bp, zp->zp_byteorder);
+ if (ot != DMU_OT_OBJSET)
+ zio_crypt_encode_mac_bp(bp, zp->zp_mac);
+
+ /* dnode blocks must be written out in the provided byteorder */
+ if (zp->zp_byteorder != ZFS_HOST_BYTEORDER &&
+ ot == DMU_OT_DNODE) {
+ void *bswap_buf = zio_buf_alloc(psize);
+ abd_t *babd = abd_get_from_buf(bswap_buf, psize);
+
+ ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF);
+ abd_copy_to_buf(bswap_buf, zio->io_abd, psize);
+ dmu_ot_byteswap[DMU_OT_BYTESWAP(ot)].ob_func(bswap_buf,
+ psize);
+
+ abd_take_ownership_of_buf(babd, B_TRUE);
+ zio_push_transform(zio, babd, psize, psize, NULL);
+ }
+
+ if (DMU_OT_IS_ENCRYPTED(ot))
+ zio_crypt_encode_params_bp(bp, zp->zp_salt, zp->zp_iv);
+ return (ZIO_PIPELINE_CONTINUE);
+ }
+
+ /* indirect blocks only maintain a cksum of the lower level MACs */
+ if (BP_GET_LEVEL(bp) > 0) {
+ BP_SET_CRYPT(bp, B_TRUE);
+ VERIFY0(zio_crypt_do_indirect_mac_checksum_abd(B_TRUE,
+ zio->io_orig_abd, BP_GET_LSIZE(bp), BP_SHOULD_BYTESWAP(bp),
+ mac));
+ zio_crypt_encode_mac_bp(bp, mac);
+ return (ZIO_PIPELINE_CONTINUE);
+ }
+
+ /*
+ * Objset blocks are a special case since they have 2 256-bit MACs
+ * embedded within them.
+ */
+ if (ot == DMU_OT_OBJSET) {
+ ASSERT0(DMU_OT_IS_ENCRYPTED(ot));
+ ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF);
+ BP_SET_CRYPT(bp, B_TRUE);
+ VERIFY0(spa_do_crypt_objset_mac_abd(B_TRUE, spa, dsobj,
+ zio->io_abd, psize, BP_SHOULD_BYTESWAP(bp)));
+ return (ZIO_PIPELINE_CONTINUE);
+ }
+
+ /* unencrypted object types are only authenticated with a MAC */
+ if (!DMU_OT_IS_ENCRYPTED(ot)) {
+ BP_SET_CRYPT(bp, B_TRUE);
+ VERIFY0(spa_do_crypt_mac_abd(B_TRUE, spa, dsobj,
+ zio->io_abd, psize, mac));
+ zio_crypt_encode_mac_bp(bp, mac);
+ return (ZIO_PIPELINE_CONTINUE);
+ }
+
+ /*
+ * Later passes of sync-to-convergence may decide to rewrite data
+ * in place to avoid more disk reallocations. This presents a problem
+ * for encryption because this consitutes rewriting the new data with
+ * the same encryption key and IV. However, this only applies to blocks
+ * in the MOS (particularly the spacemaps) and we do not encrypt the
+ * MOS. We assert that the zio is allocating or an intent log write
+ * to enforce this.
+ */
+ ASSERT(IO_IS_ALLOCATING(zio) || ot == DMU_OT_INTENT_LOG);
+ ASSERT(BP_GET_LEVEL(bp) == 0 || ot == DMU_OT_INTENT_LOG);
+ ASSERT(spa_feature_is_active(spa, SPA_FEATURE_ENCRYPTION));
+ ASSERT3U(psize, !=, 0);
+
+ enc_buf = zio_buf_alloc(psize);
+ eabd = abd_get_from_buf(enc_buf, psize);
+ abd_take_ownership_of_buf(eabd, B_TRUE);
+
+ /*
+ * For an explanation of what encryption parameters are stored
+ * where, see the block comment in zio_crypt.c.
+ */
+ if (ot == DMU_OT_INTENT_LOG) {
+ zio_crypt_decode_params_bp(bp, salt, iv);
+ } else {
+ BP_SET_CRYPT(bp, B_TRUE);
+ }
+
+ /* Perform the encryption. This should not fail */
+ VERIFY0(spa_do_crypt_abd(B_TRUE, spa, dsobj, bp, zio->io_txg,
+ psize, zio->io_abd, eabd, iv, mac, salt, &no_crypt));
+
+ /* encode encryption metadata into the bp */
+ if (ot == DMU_OT_INTENT_LOG) {
+ /*
+ * ZIL blocks store the MAC in the embedded checksum, so the
+ * transform must always be applied.
+ */
+ zio_crypt_encode_mac_zil(enc_buf, mac);
+ zio_push_transform(zio, eabd, psize, psize, NULL);
+ } else {
+ BP_SET_CRYPT(bp, B_TRUE);
+ zio_crypt_encode_params_bp(bp, salt, iv);
+ zio_crypt_encode_mac_bp(bp, mac);
+
+ if (no_crypt) {
+ ASSERT3U(ot, ==, DMU_OT_DNODE);
+ abd_free(eabd);
+ } else {
+ zio_push_transform(zio, eabd, psize, psize, NULL);
+ }
+ }
+
+ return (ZIO_PIPELINE_CONTINUE);
+}
+
/*
* ==========================================================================
* Generate and verify checksums
if (error == ECKSUM &&
!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
zfs_ereport_start_checksum(zio->io_spa,
- zio->io_vd, zio, zio->io_offset,
- zio->io_size, NULL, &info);
+ zio->io_vd, &zio->io_bookmark, zio,
+ zio->io_offset, zio->io_size, NULL, &info);
}
}
zio_t *pio, *pio_next;
zio_link_t *zl = NULL;
- if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) ||
- zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY))
+ if (zio_wait_for_children(zio, ZIO_CHILD_GANG_BIT | ZIO_CHILD_DDT_BIT,
+ ZIO_WAIT_READY)) {
return (ZIO_PIPELINE_STOP);
+ }
if (zio->io_ready) {
ASSERT(IO_IS_ALLOCATING(zio));
static void
zio_dva_throttle_done(zio_t *zio)
{
+ ASSERTV(zio_t *lio = zio->io_logical);
zio_t *pio = zio_unique_parent(zio);
vdev_t *vd = zio->io_vd;
int flags = METASLAB_ASYNC_ALLOC;
- ASSERTV(zio_t *lio = zio->io_logical);
ASSERT3P(zio->io_bp, !=, NULL);
ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE);
ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV);
ASSERT(vd != NULL);
ASSERT3P(vd, ==, vd->vdev_top);
- ASSERT(!(zio->io_flags & (ZIO_FLAG_IO_REPAIR | ZIO_FLAG_IO_RETRY)));
+ ASSERT(zio_injection_enabled || !(zio->io_flags & ZIO_FLAG_IO_RETRY));
+ ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REPAIR));
ASSERT(zio->io_flags & ZIO_FLAG_IO_ALLOCATING);
ASSERT(!(lio->io_flags & ZIO_FLAG_IO_REWRITE));
ASSERT(!(lio->io_orig_flags & ZIO_FLAG_NODATA));
* Always attempt to keep stack usage minimal here since
* we can be called recurisvely up to 19 levels deep.
*/
- uint64_t psize = zio->io_size;
+ const uint64_t psize = zio->io_size;
zio_t *pio, *pio_next;
- int c, w;
zio_link_t *zl = NULL;
/*
* If our 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_DDT, ZIO_WAIT_DONE) ||
- zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE))
+ if (zio_wait_for_children(zio, ZIO_CHILD_ALL_BITS, ZIO_WAIT_DONE)) {
return (ZIO_PIPELINE_STOP);
+ }
/*
* If the allocation throttle is enabled, then update the accounting.
}
- for (c = 0; c < ZIO_CHILD_TYPES; c++)
- for (w = 0; w < ZIO_WAIT_TYPES; w++)
+ 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 (zio->io_bp != NULL && !BP_IS_EMBEDDED(zio->io_bp)) {
if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(zio->io_bp) &&
zio->io_bp_override == NULL &&
!(zio->io_flags & ZIO_FLAG_IO_REPAIR)) {
- ASSERT(!BP_SHOULD_BYTESWAP(zio->io_bp));
ASSERT3U(zio->io_prop.zp_copies, <=,
BP_GET_NDVAS(zio->io_bp));
ASSERT(BP_COUNT_GANG(zio->io_bp) == 0 ||
zio_cksum_report_t *zcr = zio->io_cksum_report;
uint64_t align = zcr->zcr_align;
uint64_t asize = P2ROUNDUP(psize, align);
- char *abuf = NULL;
abd_t *adata = zio->io_abd;
if (asize != psize) {
- adata = abd_alloc_linear(asize, B_TRUE);
+ adata = abd_alloc(asize, B_TRUE);
abd_copy(adata, zio->io_abd, psize);
abd_zero_off(adata, psize, asize - psize);
}
- if (adata != NULL)
- abuf = abd_borrow_buf_copy(adata, asize);
-
zio->io_cksum_report = zcr->zcr_next;
zcr->zcr_next = NULL;
- zcr->zcr_finish(zcr, abuf);
+ zcr->zcr_finish(zcr, adata);
zfs_ereport_free_checksum(zcr);
- if (adata != NULL)
- abd_return_buf(adata, abuf, asize);
-
if (asize != psize)
abd_free(adata);
}
if (zio->io_delay >= MSEC2NSEC(zio_delay_max)) {
if (zio->io_vd != NULL && !vdev_is_dead(zio->io_vd))
zfs_ereport_post(FM_EREPORT_ZFS_DELAY, zio->io_spa,
- zio->io_vd, zio, 0, 0);
+ zio->io_vd, &zio->io_bookmark, zio, 0, 0);
}
if (zio->io_error) {
* device is currently unavailable.
*/
if (zio->io_error != ECKSUM && zio->io_vd != NULL &&
- !vdev_is_dead(zio->io_vd))
+ !vdev_is_dead(zio->io_vd))
zfs_ereport_post(FM_EREPORT_ZFS_IO, zio->io_spa,
- zio->io_vd, zio, 0, 0);
+ zio->io_vd, &zio->io_bookmark, zio, 0, 0);
if ((zio->io_error == EIO || !(zio->io_flags &
(ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) &&
* For logical I/O requests, tell the SPA to log the
* error and generate a logical data ereport.
*/
- spa_log_error(zio->io_spa, zio);
+ spa_log_error(zio->io_spa, &zio->io_bookmark);
zfs_ereport_post(FM_EREPORT_ZFS_DATA, zio->io_spa,
- NULL, zio, 0, 0);
+ NULL, &zio->io_bookmark, zio, 0, 0);
}
}
*/
if ((zio->io_flags & ZIO_FLAG_GODFATHER) &&
(zio->io_reexecute & ZIO_REEXECUTE_SUSPEND))
- zio->io_reexecute = 0;
+ zio->io_reexecute &= ~ZIO_REEXECUTE_SUSPEND;
if (zio->io_reexecute) {
/*
* We'd fail again if we reexecuted now, so suspend
* until conditions improve (e.g. device comes online).
*/
- zio_suspend(zio->io_spa, zio);
+ zio_suspend(zio->io_spa, zio, ZIO_SUSPEND_IOERR);
} else {
/*
* Reexecution is potentially a huge amount of work.
zio_free_bp_init,
zio_issue_async,
zio_write_compress,
+ zio_encrypt,
zio_checksum_generate,
zio_nop_write,
zio_ddt_read_start,
projects / mirror_zfs.git / blobdiff