1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
7 #include <linux/backing-dev.h>
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_trace.h"
17 #include "xfs_errortag.h"
18 #include "xfs_error.h"
20 static kmem_zone_t *xfs_buf_zone;
22 #define xb_to_gfp(flags) \
23 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN)
30 * b_sema (caller holds)
34 * b_sema (caller holds)
43 * xfs_buftarg_wait_rele
45 * b_lock (trylock due to inversion)
49 * b_lock (trylock due to inversion)
57 * Return true if the buffer is vmapped.
59 * b_addr is null if the buffer is not mapped, but the code is clever
60 * enough to know it doesn't have to map a single page, so the check has
61 * to be both for b_addr and bp->b_page_count > 1.
63 return bp->b_addr && bp->b_page_count > 1;
70 return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
74 * Bump the I/O in flight count on the buftarg if we haven't yet done so for
75 * this buffer. The count is incremented once per buffer (per hold cycle)
76 * because the corresponding decrement is deferred to buffer release. Buffers
77 * can undergo I/O multiple times in a hold-release cycle and per buffer I/O
78 * tracking adds unnecessary overhead. This is used for sychronization purposes
79 * with unmount (see xfs_wait_buftarg()), so all we really need is a count of
82 * Buffers that are never released (e.g., superblock, iclog buffers) must set
83 * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count
84 * never reaches zero and unmount hangs indefinitely.
90 if (bp->b_flags & XBF_NO_IOACCT)
93 ASSERT(bp->b_flags & XBF_ASYNC);
94 spin_lock(&bp->b_lock);
95 if (!(bp->b_state & XFS_BSTATE_IN_FLIGHT)) {
96 bp->b_state |= XFS_BSTATE_IN_FLIGHT;
97 percpu_counter_inc(&bp->b_target->bt_io_count);
99 spin_unlock(&bp->b_lock);
103 * Clear the in-flight state on a buffer about to be released to the LRU or
104 * freed and unaccount from the buftarg.
107 __xfs_buf_ioacct_dec(
110 lockdep_assert_held(&bp->b_lock);
112 if (bp->b_state & XFS_BSTATE_IN_FLIGHT) {
113 bp->b_state &= ~XFS_BSTATE_IN_FLIGHT;
114 percpu_counter_dec(&bp->b_target->bt_io_count);
122 spin_lock(&bp->b_lock);
123 __xfs_buf_ioacct_dec(bp);
124 spin_unlock(&bp->b_lock);
128 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
129 * b_lru_ref count so that the buffer is freed immediately when the buffer
130 * reference count falls to zero. If the buffer is already on the LRU, we need
131 * to remove the reference that LRU holds on the buffer.
133 * This prevents build-up of stale buffers on the LRU.
139 ASSERT(xfs_buf_islocked(bp));
141 bp->b_flags |= XBF_STALE;
144 * Clear the delwri status so that a delwri queue walker will not
145 * flush this buffer to disk now that it is stale. The delwri queue has
146 * a reference to the buffer, so this is safe to do.
148 bp->b_flags &= ~_XBF_DELWRI_Q;
151 * Once the buffer is marked stale and unlocked, a subsequent lookup
152 * could reset b_flags. There is no guarantee that the buffer is
153 * unaccounted (released to LRU) before that occurs. Drop in-flight
154 * status now to preserve accounting consistency.
156 spin_lock(&bp->b_lock);
157 __xfs_buf_ioacct_dec(bp);
159 atomic_set(&bp->b_lru_ref, 0);
160 if (!(bp->b_state & XFS_BSTATE_DISPOSE) &&
161 (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru)))
162 atomic_dec(&bp->b_hold);
164 ASSERT(atomic_read(&bp->b_hold) >= 1);
165 spin_unlock(&bp->b_lock);
173 ASSERT(bp->b_maps == NULL);
174 bp->b_map_count = map_count;
176 if (map_count == 1) {
177 bp->b_maps = &bp->__b_map;
181 bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map),
189 * Frees b_pages if it was allocated.
195 if (bp->b_maps != &bp->__b_map) {
196 kmem_free(bp->b_maps);
201 static struct xfs_buf *
203 struct xfs_buftarg *target,
204 struct xfs_buf_map *map,
206 xfs_buf_flags_t flags)
212 bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS);
217 * We don't want certain flags to appear in b_flags unless they are
218 * specifically set by later operations on the buffer.
220 flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD);
222 atomic_set(&bp->b_hold, 1);
223 atomic_set(&bp->b_lru_ref, 1);
224 init_completion(&bp->b_iowait);
225 INIT_LIST_HEAD(&bp->b_lru);
226 INIT_LIST_HEAD(&bp->b_list);
227 INIT_LIST_HEAD(&bp->b_li_list);
228 sema_init(&bp->b_sema, 0); /* held, no waiters */
229 spin_lock_init(&bp->b_lock);
230 bp->b_target = target;
231 bp->b_mount = target->bt_mount;
235 * Set length and io_length to the same value initially.
236 * I/O routines should use io_length, which will be the same in
237 * most cases but may be reset (e.g. XFS recovery).
239 error = xfs_buf_get_maps(bp, nmaps);
241 kmem_zone_free(xfs_buf_zone, bp);
245 bp->b_bn = map[0].bm_bn;
247 for (i = 0; i < nmaps; i++) {
248 bp->b_maps[i].bm_bn = map[i].bm_bn;
249 bp->b_maps[i].bm_len = map[i].bm_len;
250 bp->b_length += map[i].bm_len;
253 atomic_set(&bp->b_pin_count, 0);
254 init_waitqueue_head(&bp->b_waiters);
256 XFS_STATS_INC(bp->b_mount, xb_create);
257 trace_xfs_buf_init(bp, _RET_IP_);
263 * Allocate a page array capable of holding a specified number
264 * of pages, and point the page buf at it.
271 /* Make sure that we have a page list */
272 if (bp->b_pages == NULL) {
273 bp->b_page_count = page_count;
274 if (page_count <= XB_PAGES) {
275 bp->b_pages = bp->b_page_array;
277 bp->b_pages = kmem_alloc(sizeof(struct page *) *
278 page_count, KM_NOFS);
279 if (bp->b_pages == NULL)
282 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
288 * Frees b_pages if it was allocated.
294 if (bp->b_pages != bp->b_page_array) {
295 kmem_free(bp->b_pages);
301 * Releases the specified buffer.
303 * The modification state of any associated pages is left unchanged.
304 * The buffer must not be on any hash - use xfs_buf_rele instead for
305 * hashed and refcounted buffers
311 trace_xfs_buf_free(bp, _RET_IP_);
313 ASSERT(list_empty(&bp->b_lru));
315 if (bp->b_flags & _XBF_PAGES) {
318 if (xfs_buf_is_vmapped(bp))
319 vm_unmap_ram(bp->b_addr - bp->b_offset,
322 for (i = 0; i < bp->b_page_count; i++) {
323 struct page *page = bp->b_pages[i];
327 } else if (bp->b_flags & _XBF_KMEM)
328 kmem_free(bp->b_addr);
329 _xfs_buf_free_pages(bp);
330 xfs_buf_free_maps(bp);
331 kmem_zone_free(xfs_buf_zone, bp);
335 * Allocates all the pages for buffer in question and builds it's page list.
338 xfs_buf_allocate_memory(
343 size_t nbytes, offset;
344 gfp_t gfp_mask = xb_to_gfp(flags);
345 unsigned short page_count, i;
346 xfs_off_t start, end;
348 xfs_km_flags_t kmflag_mask = 0;
351 * assure zeroed buffer for non-read cases.
353 if (!(flags & XBF_READ)) {
354 kmflag_mask |= KM_ZERO;
355 gfp_mask |= __GFP_ZERO;
359 * for buffers that are contained within a single page, just allocate
360 * the memory from the heap - there's no need for the complexity of
361 * page arrays to keep allocation down to order 0.
363 size = BBTOB(bp->b_length);
364 if (size < PAGE_SIZE) {
365 int align_mask = xfs_buftarg_dma_alignment(bp->b_target);
366 bp->b_addr = kmem_alloc_io(size, align_mask,
367 KM_NOFS | kmflag_mask);
369 /* low memory - use alloc_page loop instead */
373 if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) !=
374 ((unsigned long)bp->b_addr & PAGE_MASK)) {
375 /* b_addr spans two pages - use alloc_page instead */
376 kmem_free(bp->b_addr);
380 bp->b_offset = offset_in_page(bp->b_addr);
381 bp->b_pages = bp->b_page_array;
382 bp->b_pages[0] = kmem_to_page(bp->b_addr);
383 bp->b_page_count = 1;
384 bp->b_flags |= _XBF_KMEM;
389 start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT;
390 end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1)
392 page_count = end - start;
393 error = _xfs_buf_get_pages(bp, page_count);
397 offset = bp->b_offset;
398 bp->b_flags |= _XBF_PAGES;
400 for (i = 0; i < bp->b_page_count; i++) {
404 page = alloc_page(gfp_mask);
405 if (unlikely(page == NULL)) {
406 if (flags & XBF_READ_AHEAD) {
407 bp->b_page_count = i;
413 * This could deadlock.
415 * But until all the XFS lowlevel code is revamped to
416 * handle buffer allocation failures we can't do much.
418 if (!(++retries % 100))
420 "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)",
421 current->comm, current->pid,
424 XFS_STATS_INC(bp->b_mount, xb_page_retries);
425 congestion_wait(BLK_RW_ASYNC, HZ/50);
429 XFS_STATS_INC(bp->b_mount, xb_page_found);
431 nbytes = min_t(size_t, size, PAGE_SIZE - offset);
433 bp->b_pages[i] = page;
439 for (i = 0; i < bp->b_page_count; i++)
440 __free_page(bp->b_pages[i]);
441 bp->b_flags &= ~_XBF_PAGES;
446 * Map buffer into kernel address-space if necessary.
453 ASSERT(bp->b_flags & _XBF_PAGES);
454 if (bp->b_page_count == 1) {
455 /* A single page buffer is always mappable */
456 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
457 } else if (flags & XBF_UNMAPPED) {
464 * vm_map_ram() will allocate auxillary structures (e.g.
465 * pagetables) with GFP_KERNEL, yet we are likely to be under
466 * GFP_NOFS context here. Hence we need to tell memory reclaim
467 * that we are in such a context via PF_MEMALLOC_NOFS to prevent
468 * memory reclaim re-entering the filesystem here and
469 * potentially deadlocking.
471 nofs_flag = memalloc_nofs_save();
473 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
478 } while (retried++ <= 1);
479 memalloc_nofs_restore(nofs_flag);
483 bp->b_addr += bp->b_offset;
490 * Finding and Reading Buffers
494 struct rhashtable_compare_arg *arg,
497 const struct xfs_buf_map *map = arg->key;
498 const struct xfs_buf *bp = obj;
501 * The key hashing in the lookup path depends on the key being the
502 * first element of the compare_arg, make sure to assert this.
504 BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0);
506 if (bp->b_bn != map->bm_bn)
509 if (unlikely(bp->b_length != map->bm_len)) {
511 * found a block number match. If the range doesn't
512 * match, the only way this is allowed is if the buffer
513 * in the cache is stale and the transaction that made
514 * it stale has not yet committed. i.e. we are
515 * reallocating a busy extent. Skip this buffer and
516 * continue searching for an exact match.
518 ASSERT(bp->b_flags & XBF_STALE);
524 static const struct rhashtable_params xfs_buf_hash_params = {
525 .min_size = 32, /* empty AGs have minimal footprint */
527 .key_len = sizeof(xfs_daddr_t),
528 .key_offset = offsetof(struct xfs_buf, b_bn),
529 .head_offset = offsetof(struct xfs_buf, b_rhash_head),
530 .automatic_shrinking = true,
531 .obj_cmpfn = _xfs_buf_obj_cmp,
536 struct xfs_perag *pag)
538 spin_lock_init(&pag->pag_buf_lock);
539 return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params);
543 xfs_buf_hash_destroy(
544 struct xfs_perag *pag)
546 rhashtable_destroy(&pag->pag_buf_hash);
550 * Look up a buffer in the buffer cache and return it referenced and locked
553 * If @new_bp is supplied and we have a lookup miss, insert @new_bp into the
556 * If XBF_TRYLOCK is set in @flags, only try to lock the buffer and return
557 * -EAGAIN if we fail to lock it.
560 * -EFSCORRUPTED if have been supplied with an invalid address
561 * -EAGAIN on trylock failure
562 * -ENOENT if we fail to find a match and @new_bp was NULL
564 * - @new_bp if we inserted it into the cache
565 * - the buffer we found and locked.
569 struct xfs_buftarg *btp,
570 struct xfs_buf_map *map,
572 xfs_buf_flags_t flags,
573 struct xfs_buf *new_bp,
574 struct xfs_buf **found_bp)
576 struct xfs_perag *pag;
578 struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn };
584 for (i = 0; i < nmaps; i++)
585 cmap.bm_len += map[i].bm_len;
587 /* Check for IOs smaller than the sector size / not sector aligned */
588 ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize));
589 ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask));
592 * Corrupted block numbers can get through to here, unfortunately, so we
593 * have to check that the buffer falls within the filesystem bounds.
595 eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks);
596 if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) {
597 xfs_alert(btp->bt_mount,
598 "%s: daddr 0x%llx out of range, EOFS 0x%llx",
599 __func__, cmap.bm_bn, eofs);
601 return -EFSCORRUPTED;
604 pag = xfs_perag_get(btp->bt_mount,
605 xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn));
607 spin_lock(&pag->pag_buf_lock);
608 bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap,
609 xfs_buf_hash_params);
611 atomic_inc(&bp->b_hold);
617 XFS_STATS_INC(btp->bt_mount, xb_miss_locked);
618 spin_unlock(&pag->pag_buf_lock);
623 /* the buffer keeps the perag reference until it is freed */
625 rhashtable_insert_fast(&pag->pag_buf_hash, &new_bp->b_rhash_head,
626 xfs_buf_hash_params);
627 spin_unlock(&pag->pag_buf_lock);
632 spin_unlock(&pag->pag_buf_lock);
635 if (!xfs_buf_trylock(bp)) {
636 if (flags & XBF_TRYLOCK) {
638 XFS_STATS_INC(btp->bt_mount, xb_busy_locked);
642 XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited);
646 * if the buffer is stale, clear all the external state associated with
647 * it. We need to keep flags such as how we allocated the buffer memory
650 if (bp->b_flags & XBF_STALE) {
651 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
652 ASSERT(bp->b_iodone == NULL);
653 bp->b_flags &= _XBF_KMEM | _XBF_PAGES;
657 trace_xfs_buf_find(bp, flags, _RET_IP_);
658 XFS_STATS_INC(btp->bt_mount, xb_get_locked);
665 struct xfs_buftarg *target,
668 xfs_buf_flags_t flags)
672 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
674 error = xfs_buf_find(target, &map, 1, flags, NULL, &bp);
681 * Assembles a buffer covering the specified range. The code is optimised for
682 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
683 * more hits than misses.
687 struct xfs_buftarg *target,
688 struct xfs_buf_map *map,
690 xfs_buf_flags_t flags)
693 struct xfs_buf *new_bp;
696 error = xfs_buf_find(target, map, nmaps, flags, NULL, &bp);
703 /* cache hit, trylock failure, caller handles failure */
704 ASSERT(flags & XBF_TRYLOCK);
707 /* cache miss, go for insert */
712 * None of the higher layers understand failure types
713 * yet, so return NULL to signal a fatal lookup error.
718 new_bp = _xfs_buf_alloc(target, map, nmaps, flags);
719 if (unlikely(!new_bp))
722 error = xfs_buf_allocate_memory(new_bp, flags);
724 xfs_buf_free(new_bp);
728 error = xfs_buf_find(target, map, nmaps, flags, new_bp, &bp);
730 xfs_buf_free(new_bp);
735 xfs_buf_free(new_bp);
739 error = _xfs_buf_map_pages(bp, flags);
740 if (unlikely(error)) {
741 xfs_warn(target->bt_mount,
742 "%s: failed to map pagesn", __func__);
749 * Clear b_error if this is a lookup from a caller that doesn't expect
750 * valid data to be found in the buffer.
752 if (!(flags & XBF_READ))
753 xfs_buf_ioerror(bp, 0);
755 XFS_STATS_INC(target->bt_mount, xb_get);
756 trace_xfs_buf_get(bp, flags, _RET_IP_);
763 xfs_buf_flags_t flags)
765 ASSERT(!(flags & XBF_WRITE));
766 ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL);
768 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD);
769 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
771 return xfs_buf_submit(bp);
775 * Reverify a buffer found in cache without an attached ->b_ops.
777 * If the caller passed an ops structure and the buffer doesn't have ops
778 * assigned, set the ops and use it to verify the contents. If verification
779 * fails, clear XBF_DONE. We assume the buffer has no recorded errors and is
780 * already in XBF_DONE state on entry.
782 * Under normal operations, every in-core buffer is verified on read I/O
783 * completion. There are two scenarios that can lead to in-core buffers without
784 * an assigned ->b_ops. The first is during log recovery of buffers on a V4
785 * filesystem, though these buffers are purged at the end of recovery. The
786 * other is online repair, which intentionally reads with a NULL buffer ops to
787 * run several verifiers across an in-core buffer in order to establish buffer
788 * type. If repair can't establish that, the buffer will be left in memory
789 * with NULL buffer ops.
794 const struct xfs_buf_ops *ops)
796 ASSERT(bp->b_flags & XBF_DONE);
797 ASSERT(bp->b_error == 0);
799 if (!ops || bp->b_ops)
803 bp->b_ops->verify_read(bp);
805 bp->b_flags &= ~XBF_DONE;
811 struct xfs_buftarg *target,
812 struct xfs_buf_map *map,
814 xfs_buf_flags_t flags,
815 const struct xfs_buf_ops *ops)
821 bp = xfs_buf_get_map(target, map, nmaps, flags);
825 trace_xfs_buf_read(bp, flags, _RET_IP_);
827 if (!(bp->b_flags & XBF_DONE)) {
828 XFS_STATS_INC(target->bt_mount, xb_get_read);
830 _xfs_buf_read(bp, flags);
834 xfs_buf_reverify(bp, ops);
836 if (flags & XBF_ASYNC) {
838 * Read ahead call which is already satisfied,
845 /* We do not want read in the flags */
846 bp->b_flags &= ~XBF_READ;
847 ASSERT(bp->b_ops != NULL || ops == NULL);
852 * If we are not low on memory then do the readahead in a deadlock
856 xfs_buf_readahead_map(
857 struct xfs_buftarg *target,
858 struct xfs_buf_map *map,
860 const struct xfs_buf_ops *ops)
862 if (bdi_read_congested(target->bt_bdev->bd_bdi))
865 xfs_buf_read_map(target, map, nmaps,
866 XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops);
870 * Read an uncached buffer from disk. Allocates and returns a locked
871 * buffer containing the disk contents or nothing.
874 xfs_buf_read_uncached(
875 struct xfs_buftarg *target,
879 struct xfs_buf **bpp,
880 const struct xfs_buf_ops *ops)
886 bp = xfs_buf_get_uncached(target, numblks, flags);
890 /* set up the buffer for a read IO */
891 ASSERT(bp->b_map_count == 1);
892 bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */
893 bp->b_maps[0].bm_bn = daddr;
894 bp->b_flags |= XBF_READ;
899 int error = bp->b_error;
909 xfs_buf_get_uncached(
910 struct xfs_buftarg *target,
914 unsigned long page_count;
917 DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks);
919 /* flags might contain irrelevant bits, pass only what we care about */
920 bp = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT);
921 if (unlikely(bp == NULL))
924 page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT;
925 error = _xfs_buf_get_pages(bp, page_count);
929 for (i = 0; i < page_count; i++) {
930 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
934 bp->b_flags |= _XBF_PAGES;
936 error = _xfs_buf_map_pages(bp, 0);
937 if (unlikely(error)) {
938 xfs_warn(target->bt_mount,
939 "%s: failed to map pages", __func__);
943 trace_xfs_buf_get_uncached(bp, _RET_IP_);
948 __free_page(bp->b_pages[i]);
949 _xfs_buf_free_pages(bp);
951 xfs_buf_free_maps(bp);
952 kmem_zone_free(xfs_buf_zone, bp);
958 * Increment reference count on buffer, to hold the buffer concurrently
959 * with another thread which may release (free) the buffer asynchronously.
960 * Must hold the buffer already to call this function.
966 trace_xfs_buf_hold(bp, _RET_IP_);
967 atomic_inc(&bp->b_hold);
971 * Release a hold on the specified buffer. If the hold count is 1, the buffer is
972 * placed on LRU or freed (depending on b_lru_ref).
978 struct xfs_perag *pag = bp->b_pag;
980 bool freebuf = false;
982 trace_xfs_buf_rele(bp, _RET_IP_);
985 ASSERT(list_empty(&bp->b_lru));
986 if (atomic_dec_and_test(&bp->b_hold)) {
987 xfs_buf_ioacct_dec(bp);
993 ASSERT(atomic_read(&bp->b_hold) > 0);
996 * We grab the b_lock here first to serialise racing xfs_buf_rele()
997 * calls. The pag_buf_lock being taken on the last reference only
998 * serialises against racing lookups in xfs_buf_find(). IOWs, the second
999 * to last reference we drop here is not serialised against the last
1000 * reference until we take bp->b_lock. Hence if we don't grab b_lock
1001 * first, the last "release" reference can win the race to the lock and
1002 * free the buffer before the second-to-last reference is processed,
1003 * leading to a use-after-free scenario.
1005 spin_lock(&bp->b_lock);
1006 release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock);
1009 * Drop the in-flight state if the buffer is already on the LRU
1010 * and it holds the only reference. This is racy because we
1011 * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT
1012 * ensures the decrement occurs only once per-buf.
1014 if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru))
1015 __xfs_buf_ioacct_dec(bp);
1019 /* the last reference has been dropped ... */
1020 __xfs_buf_ioacct_dec(bp);
1021 if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) {
1023 * If the buffer is added to the LRU take a new reference to the
1024 * buffer for the LRU and clear the (now stale) dispose list
1027 if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) {
1028 bp->b_state &= ~XFS_BSTATE_DISPOSE;
1029 atomic_inc(&bp->b_hold);
1031 spin_unlock(&pag->pag_buf_lock);
1034 * most of the time buffers will already be removed from the
1035 * LRU, so optimise that case by checking for the
1036 * XFS_BSTATE_DISPOSE flag indicating the last list the buffer
1037 * was on was the disposal list
1039 if (!(bp->b_state & XFS_BSTATE_DISPOSE)) {
1040 list_lru_del(&bp->b_target->bt_lru, &bp->b_lru);
1042 ASSERT(list_empty(&bp->b_lru));
1045 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1046 rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head,
1047 xfs_buf_hash_params);
1048 spin_unlock(&pag->pag_buf_lock);
1054 spin_unlock(&bp->b_lock);
1062 * Lock a buffer object, if it is not already locked.
1064 * If we come across a stale, pinned, locked buffer, we know that we are
1065 * being asked to lock a buffer that has been reallocated. Because it is
1066 * pinned, we know that the log has not been pushed to disk and hence it
1067 * will still be locked. Rather than continuing to have trylock attempts
1068 * fail until someone else pushes the log, push it ourselves before
1069 * returning. This means that the xfsaild will not get stuck trying
1070 * to push on stale inode buffers.
1078 locked = down_trylock(&bp->b_sema) == 0;
1080 trace_xfs_buf_trylock(bp, _RET_IP_);
1082 trace_xfs_buf_trylock_fail(bp, _RET_IP_);
1087 * Lock a buffer object.
1089 * If we come across a stale, pinned, locked buffer, we know that we
1090 * are being asked to lock a buffer that has been reallocated. Because
1091 * it is pinned, we know that the log has not been pushed to disk and
1092 * hence it will still be locked. Rather than sleeping until someone
1093 * else pushes the log, push it ourselves before trying to get the lock.
1099 trace_xfs_buf_lock(bp, _RET_IP_);
1101 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
1102 xfs_log_force(bp->b_mount, 0);
1105 trace_xfs_buf_lock_done(bp, _RET_IP_);
1112 ASSERT(xfs_buf_islocked(bp));
1115 trace_xfs_buf_unlock(bp, _RET_IP_);
1122 DECLARE_WAITQUEUE (wait, current);
1124 if (atomic_read(&bp->b_pin_count) == 0)
1127 add_wait_queue(&bp->b_waiters, &wait);
1129 set_current_state(TASK_UNINTERRUPTIBLE);
1130 if (atomic_read(&bp->b_pin_count) == 0)
1134 remove_wait_queue(&bp->b_waiters, &wait);
1135 set_current_state(TASK_RUNNING);
1139 * Buffer Utility Routines
1146 bool read = bp->b_flags & XBF_READ;
1148 trace_xfs_buf_iodone(bp, _RET_IP_);
1150 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1153 * Pull in IO completion errors now. We are guaranteed to be running
1154 * single threaded, so we don't need the lock to read b_io_error.
1156 if (!bp->b_error && bp->b_io_error)
1157 xfs_buf_ioerror(bp, bp->b_io_error);
1159 /* Only validate buffers that were read without errors */
1160 if (read && !bp->b_error && bp->b_ops) {
1161 ASSERT(!bp->b_iodone);
1162 bp->b_ops->verify_read(bp);
1166 bp->b_flags &= ~XBF_WRITE_FAIL;
1167 bp->b_flags |= XBF_DONE;
1171 (*(bp->b_iodone))(bp);
1172 else if (bp->b_flags & XBF_ASYNC)
1175 complete(&bp->b_iowait);
1180 struct work_struct *work)
1182 struct xfs_buf *bp =
1183 container_of(work, xfs_buf_t, b_ioend_work);
1189 xfs_buf_ioend_async(
1192 INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work);
1193 queue_work(bp->b_mount->m_buf_workqueue, &bp->b_ioend_work);
1200 xfs_failaddr_t failaddr)
1202 ASSERT(error <= 0 && error >= -1000);
1203 bp->b_error = error;
1204 trace_xfs_buf_ioerror(bp, error, failaddr);
1208 xfs_buf_ioerror_alert(
1212 xfs_alert(bp->b_mount,
1213 "metadata I/O error in \"%s\" at daddr 0x%llx len %d error %d",
1214 func, (uint64_t)XFS_BUF_ADDR(bp), bp->b_length,
1224 ASSERT(xfs_buf_islocked(bp));
1226 bp->b_flags |= XBF_WRITE;
1227 bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q |
1230 error = xfs_buf_submit(bp);
1232 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
1240 struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private;
1243 * don't overwrite existing errors - otherwise we can lose errors on
1244 * buffers that require multiple bios to complete.
1246 if (bio->bi_status) {
1247 int error = blk_status_to_errno(bio->bi_status);
1249 cmpxchg(&bp->b_io_error, 0, error);
1252 if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1253 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1255 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1256 xfs_buf_ioend_async(bp);
1261 xfs_buf_ioapply_map(
1270 int total_nr_pages = bp->b_page_count;
1273 sector_t sector = bp->b_maps[map].bm_bn;
1277 /* skip the pages in the buffer before the start offset */
1279 offset = *buf_offset;
1280 while (offset >= PAGE_SIZE) {
1282 offset -= PAGE_SIZE;
1286 * Limit the IO size to the length of the current vector, and update the
1287 * remaining IO count for the next time around.
1289 size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count);
1291 *buf_offset += size;
1294 atomic_inc(&bp->b_io_remaining);
1295 nr_pages = min(total_nr_pages, BIO_MAX_PAGES);
1297 bio = bio_alloc(GFP_NOIO, nr_pages);
1298 bio_set_dev(bio, bp->b_target->bt_bdev);
1299 bio->bi_iter.bi_sector = sector;
1300 bio->bi_end_io = xfs_buf_bio_end_io;
1301 bio->bi_private = bp;
1302 bio_set_op_attrs(bio, op, op_flags);
1304 for (; size && nr_pages; nr_pages--, page_index++) {
1305 int rbytes, nbytes = PAGE_SIZE - offset;
1310 rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes,
1312 if (rbytes < nbytes)
1316 sector += BTOBB(nbytes);
1321 if (likely(bio->bi_iter.bi_size)) {
1322 if (xfs_buf_is_vmapped(bp)) {
1323 flush_kernel_vmap_range(bp->b_addr,
1324 xfs_buf_vmap_len(bp));
1331 * This is guaranteed not to be the last io reference count
1332 * because the caller (xfs_buf_submit) holds a count itself.
1334 atomic_dec(&bp->b_io_remaining);
1335 xfs_buf_ioerror(bp, -EIO);
1345 struct blk_plug plug;
1353 * Make sure we capture only current IO errors rather than stale errors
1354 * left over from previous use of the buffer (e.g. failed readahead).
1358 if (bp->b_flags & XBF_WRITE) {
1362 * Run the write verifier callback function if it exists. If
1363 * this function fails it will mark the buffer with an error and
1364 * the IO should not be dispatched.
1367 bp->b_ops->verify_write(bp);
1369 xfs_force_shutdown(bp->b_mount,
1370 SHUTDOWN_CORRUPT_INCORE);
1373 } else if (bp->b_bn != XFS_BUF_DADDR_NULL) {
1374 struct xfs_mount *mp = bp->b_mount;
1377 * non-crc filesystems don't attach verifiers during
1378 * log recovery, so don't warn for such filesystems.
1380 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1382 "%s: no buf ops on daddr 0x%llx len %d",
1383 __func__, bp->b_bn, bp->b_length);
1384 xfs_hex_dump(bp->b_addr,
1385 XFS_CORRUPTION_DUMP_LEN);
1389 } else if (bp->b_flags & XBF_READ_AHEAD) {
1391 op_flags = REQ_RAHEAD;
1396 /* we only use the buffer cache for meta-data */
1397 op_flags |= REQ_META;
1400 * Walk all the vectors issuing IO on them. Set up the initial offset
1401 * into the buffer and the desired IO size before we start -
1402 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1405 offset = bp->b_offset;
1406 size = BBTOB(bp->b_length);
1407 blk_start_plug(&plug);
1408 for (i = 0; i < bp->b_map_count; i++) {
1409 xfs_buf_ioapply_map(bp, i, &offset, &size, op, op_flags);
1413 break; /* all done */
1415 blk_finish_plug(&plug);
1419 * Wait for I/O completion of a sync buffer and return the I/O error code.
1425 ASSERT(!(bp->b_flags & XBF_ASYNC));
1427 trace_xfs_buf_iowait(bp, _RET_IP_);
1428 wait_for_completion(&bp->b_iowait);
1429 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1435 * Buffer I/O submission path, read or write. Asynchronous submission transfers
1436 * the buffer lock ownership and the current reference to the IO. It is not
1437 * safe to reference the buffer after a call to this function unless the caller
1438 * holds an additional reference itself.
1447 trace_xfs_buf_submit(bp, _RET_IP_);
1449 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1451 /* on shutdown we stale and complete the buffer immediately */
1452 if (XFS_FORCED_SHUTDOWN(bp->b_mount)) {
1453 xfs_buf_ioerror(bp, -EIO);
1454 bp->b_flags &= ~XBF_DONE;
1461 * Grab a reference so the buffer does not go away underneath us. For
1462 * async buffers, I/O completion drops the callers reference, which
1463 * could occur before submission returns.
1467 if (bp->b_flags & XBF_WRITE)
1468 xfs_buf_wait_unpin(bp);
1470 /* clear the internal error state to avoid spurious errors */
1474 * Set the count to 1 initially, this will stop an I/O completion
1475 * callout which happens before we have started all the I/O from calling
1476 * xfs_buf_ioend too early.
1478 atomic_set(&bp->b_io_remaining, 1);
1479 if (bp->b_flags & XBF_ASYNC)
1480 xfs_buf_ioacct_inc(bp);
1481 _xfs_buf_ioapply(bp);
1484 * If _xfs_buf_ioapply failed, we can get back here with only the IO
1485 * reference we took above. If we drop it to zero, run completion so
1486 * that we don't return to the caller with completion still pending.
1488 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1489 if (bp->b_error || !(bp->b_flags & XBF_ASYNC))
1492 xfs_buf_ioend_async(bp);
1496 error = xfs_buf_iowait(bp);
1499 * Release the hold that keeps the buffer referenced for the entire
1500 * I/O. Note that if the buffer is async, it is not safe to reference
1501 * after this release.
1515 return bp->b_addr + offset;
1517 offset += bp->b_offset;
1518 page = bp->b_pages[offset >> PAGE_SHIFT];
1519 return page_address(page) + (offset & (PAGE_SIZE-1));
1530 bend = boff + bsize;
1531 while (boff < bend) {
1533 int page_index, page_offset, csize;
1535 page_index = (boff + bp->b_offset) >> PAGE_SHIFT;
1536 page_offset = (boff + bp->b_offset) & ~PAGE_MASK;
1537 page = bp->b_pages[page_index];
1538 csize = min_t(size_t, PAGE_SIZE - page_offset,
1539 BBTOB(bp->b_length) - boff);
1541 ASSERT((csize + page_offset) <= PAGE_SIZE);
1543 memset(page_address(page) + page_offset, 0, csize);
1550 * Log a message about and stale a buffer that a caller has decided is corrupt.
1552 * This function should be called for the kinds of metadata corruption that
1553 * cannot be detect from a verifier, such as incorrect inter-block relationship
1554 * data. Do /not/ call this function from a verifier function.
1556 * The buffer must be XBF_DONE prior to the call. Afterwards, the buffer will
1557 * be marked stale, but b_error will not be set. The caller is responsible for
1558 * releasing the buffer or fixing it.
1561 __xfs_buf_mark_corrupt(
1565 ASSERT(bp->b_flags & XBF_DONE);
1567 xfs_buf_corruption_error(bp, fa);
1572 * Handling of buffer targets (buftargs).
1576 * Wait for any bufs with callbacks that have been submitted but have not yet
1577 * returned. These buffers will have an elevated hold count, so wait on those
1578 * while freeing all the buffers only held by the LRU.
1580 static enum lru_status
1581 xfs_buftarg_wait_rele(
1582 struct list_head *item,
1583 struct list_lru_one *lru,
1584 spinlock_t *lru_lock,
1588 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1589 struct list_head *dispose = arg;
1591 if (atomic_read(&bp->b_hold) > 1) {
1592 /* need to wait, so skip it this pass */
1593 trace_xfs_buf_wait_buftarg(bp, _RET_IP_);
1596 if (!spin_trylock(&bp->b_lock))
1600 * clear the LRU reference count so the buffer doesn't get
1601 * ignored in xfs_buf_rele().
1603 atomic_set(&bp->b_lru_ref, 0);
1604 bp->b_state |= XFS_BSTATE_DISPOSE;
1605 list_lru_isolate_move(lru, item, dispose);
1606 spin_unlock(&bp->b_lock);
1612 struct xfs_buftarg *btp)
1618 * First wait on the buftarg I/O count for all in-flight buffers to be
1619 * released. This is critical as new buffers do not make the LRU until
1620 * they are released.
1622 * Next, flush the buffer workqueue to ensure all completion processing
1623 * has finished. Just waiting on buffer locks is not sufficient for
1624 * async IO as the reference count held over IO is not released until
1625 * after the buffer lock is dropped. Hence we need to ensure here that
1626 * all reference counts have been dropped before we start walking the
1629 while (percpu_counter_sum(&btp->bt_io_count))
1631 flush_workqueue(btp->bt_mount->m_buf_workqueue);
1633 /* loop until there is nothing left on the lru list. */
1634 while (list_lru_count(&btp->bt_lru)) {
1635 list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele,
1636 &dispose, LONG_MAX);
1638 while (!list_empty(&dispose)) {
1640 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1641 list_del_init(&bp->b_lru);
1642 if (bp->b_flags & XBF_WRITE_FAIL) {
1643 xfs_alert(btp->bt_mount,
1644 "Corruption Alert: Buffer at daddr 0x%llx had permanent write failures!",
1645 (long long)bp->b_bn);
1646 xfs_alert(btp->bt_mount,
1647 "Please run xfs_repair to determine the extent of the problem.");
1656 static enum lru_status
1657 xfs_buftarg_isolate(
1658 struct list_head *item,
1659 struct list_lru_one *lru,
1660 spinlock_t *lru_lock,
1663 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1664 struct list_head *dispose = arg;
1667 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1668 * If we fail to get the lock, just skip it.
1670 if (!spin_trylock(&bp->b_lock))
1673 * Decrement the b_lru_ref count unless the value is already
1674 * zero. If the value is already zero, we need to reclaim the
1675 * buffer, otherwise it gets another trip through the LRU.
1677 if (atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1678 spin_unlock(&bp->b_lock);
1682 bp->b_state |= XFS_BSTATE_DISPOSE;
1683 list_lru_isolate_move(lru, item, dispose);
1684 spin_unlock(&bp->b_lock);
1688 static unsigned long
1689 xfs_buftarg_shrink_scan(
1690 struct shrinker *shrink,
1691 struct shrink_control *sc)
1693 struct xfs_buftarg *btp = container_of(shrink,
1694 struct xfs_buftarg, bt_shrinker);
1696 unsigned long freed;
1698 freed = list_lru_shrink_walk(&btp->bt_lru, sc,
1699 xfs_buftarg_isolate, &dispose);
1701 while (!list_empty(&dispose)) {
1703 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1704 list_del_init(&bp->b_lru);
1711 static unsigned long
1712 xfs_buftarg_shrink_count(
1713 struct shrinker *shrink,
1714 struct shrink_control *sc)
1716 struct xfs_buftarg *btp = container_of(shrink,
1717 struct xfs_buftarg, bt_shrinker);
1718 return list_lru_shrink_count(&btp->bt_lru, sc);
1723 struct xfs_buftarg *btp)
1725 unregister_shrinker(&btp->bt_shrinker);
1726 ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0);
1727 percpu_counter_destroy(&btp->bt_io_count);
1728 list_lru_destroy(&btp->bt_lru);
1730 xfs_blkdev_issue_flush(btp);
1736 xfs_setsize_buftarg(
1738 unsigned int sectorsize)
1740 /* Set up metadata sector size info */
1741 btp->bt_meta_sectorsize = sectorsize;
1742 btp->bt_meta_sectormask = sectorsize - 1;
1744 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1745 xfs_warn(btp->bt_mount,
1746 "Cannot set_blocksize to %u on device %pg",
1747 sectorsize, btp->bt_bdev);
1751 /* Set up device logical sector size mask */
1752 btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev);
1753 btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1;
1759 * When allocating the initial buffer target we have not yet
1760 * read in the superblock, so don't know what sized sectors
1761 * are being used at this early stage. Play safe.
1764 xfs_setsize_buftarg_early(
1766 struct block_device *bdev)
1768 return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev));
1773 struct xfs_mount *mp,
1774 struct block_device *bdev,
1775 struct dax_device *dax_dev)
1779 btp = kmem_zalloc(sizeof(*btp), KM_NOFS);
1782 btp->bt_dev = bdev->bd_dev;
1783 btp->bt_bdev = bdev;
1784 btp->bt_daxdev = dax_dev;
1786 if (xfs_setsize_buftarg_early(btp, bdev))
1789 if (list_lru_init(&btp->bt_lru))
1792 if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL))
1795 btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count;
1796 btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan;
1797 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1798 btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE;
1799 if (register_shrinker(&btp->bt_shrinker))
1804 percpu_counter_destroy(&btp->bt_io_count);
1806 list_lru_destroy(&btp->bt_lru);
1813 * Cancel a delayed write list.
1815 * Remove each buffer from the list, clear the delwri queue flag and drop the
1816 * associated buffer reference.
1819 xfs_buf_delwri_cancel(
1820 struct list_head *list)
1824 while (!list_empty(list)) {
1825 bp = list_first_entry(list, struct xfs_buf, b_list);
1828 bp->b_flags &= ~_XBF_DELWRI_Q;
1829 list_del_init(&bp->b_list);
1835 * Add a buffer to the delayed write list.
1837 * This queues a buffer for writeout if it hasn't already been. Note that
1838 * neither this routine nor the buffer list submission functions perform
1839 * any internal synchronization. It is expected that the lists are thread-local
1842 * Returns true if we queued up the buffer, or false if it already had
1843 * been on the buffer list.
1846 xfs_buf_delwri_queue(
1848 struct list_head *list)
1850 ASSERT(xfs_buf_islocked(bp));
1851 ASSERT(!(bp->b_flags & XBF_READ));
1854 * If the buffer is already marked delwri it already is queued up
1855 * by someone else for imediate writeout. Just ignore it in that
1858 if (bp->b_flags & _XBF_DELWRI_Q) {
1859 trace_xfs_buf_delwri_queued(bp, _RET_IP_);
1863 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1866 * If a buffer gets written out synchronously or marked stale while it
1867 * is on a delwri list we lazily remove it. To do this, the other party
1868 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1869 * It remains referenced and on the list. In a rare corner case it
1870 * might get readded to a delwri list after the synchronous writeout, in
1871 * which case we need just need to re-add the flag here.
1873 bp->b_flags |= _XBF_DELWRI_Q;
1874 if (list_empty(&bp->b_list)) {
1875 atomic_inc(&bp->b_hold);
1876 list_add_tail(&bp->b_list, list);
1883 * Compare function is more complex than it needs to be because
1884 * the return value is only 32 bits and we are doing comparisons
1890 struct list_head *a,
1891 struct list_head *b)
1893 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1894 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1897 diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn;
1906 * Submit buffers for write. If wait_list is specified, the buffers are
1907 * submitted using sync I/O and placed on the wait list such that the caller can
1908 * iowait each buffer. Otherwise async I/O is used and the buffers are released
1909 * at I/O completion time. In either case, buffers remain locked until I/O
1910 * completes and the buffer is released from the queue.
1913 xfs_buf_delwri_submit_buffers(
1914 struct list_head *buffer_list,
1915 struct list_head *wait_list)
1917 struct xfs_buf *bp, *n;
1919 struct blk_plug plug;
1921 list_sort(NULL, buffer_list, xfs_buf_cmp);
1923 blk_start_plug(&plug);
1924 list_for_each_entry_safe(bp, n, buffer_list, b_list) {
1926 if (xfs_buf_ispinned(bp)) {
1930 if (!xfs_buf_trylock(bp))
1937 * Someone else might have written the buffer synchronously or
1938 * marked it stale in the meantime. In that case only the
1939 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
1940 * reference and remove it from the list here.
1942 if (!(bp->b_flags & _XBF_DELWRI_Q)) {
1943 list_del_init(&bp->b_list);
1948 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1951 * If we have a wait list, each buffer (and associated delwri
1952 * queue reference) transfers to it and is submitted
1953 * synchronously. Otherwise, drop the buffer from the delwri
1954 * queue and submit async.
1956 bp->b_flags &= ~_XBF_DELWRI_Q;
1957 bp->b_flags |= XBF_WRITE;
1959 bp->b_flags &= ~XBF_ASYNC;
1960 list_move_tail(&bp->b_list, wait_list);
1962 bp->b_flags |= XBF_ASYNC;
1963 list_del_init(&bp->b_list);
1965 __xfs_buf_submit(bp, false);
1967 blk_finish_plug(&plug);
1973 * Write out a buffer list asynchronously.
1975 * This will take the @buffer_list, write all non-locked and non-pinned buffers
1976 * out and not wait for I/O completion on any of the buffers. This interface
1977 * is only safely useable for callers that can track I/O completion by higher
1978 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
1981 * Note: this function will skip buffers it would block on, and in doing so
1982 * leaves them on @buffer_list so they can be retried on a later pass. As such,
1983 * it is up to the caller to ensure that the buffer list is fully submitted or
1984 * cancelled appropriately when they are finished with the list. Failure to
1985 * cancel or resubmit the list until it is empty will result in leaked buffers
1989 xfs_buf_delwri_submit_nowait(
1990 struct list_head *buffer_list)
1992 return xfs_buf_delwri_submit_buffers(buffer_list, NULL);
1996 * Write out a buffer list synchronously.
1998 * This will take the @buffer_list, write all buffers out and wait for I/O
1999 * completion on all of the buffers. @buffer_list is consumed by the function,
2000 * so callers must have some other way of tracking buffers if they require such
2004 xfs_buf_delwri_submit(
2005 struct list_head *buffer_list)
2007 LIST_HEAD (wait_list);
2008 int error = 0, error2;
2011 xfs_buf_delwri_submit_buffers(buffer_list, &wait_list);
2013 /* Wait for IO to complete. */
2014 while (!list_empty(&wait_list)) {
2015 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
2017 list_del_init(&bp->b_list);
2020 * Wait on the locked buffer, check for errors and unlock and
2021 * release the delwri queue reference.
2023 error2 = xfs_buf_iowait(bp);
2033 * Push a single buffer on a delwri queue.
2035 * The purpose of this function is to submit a single buffer of a delwri queue
2036 * and return with the buffer still on the original queue. The waiting delwri
2037 * buffer submission infrastructure guarantees transfer of the delwri queue
2038 * buffer reference to a temporary wait list. We reuse this infrastructure to
2039 * transfer the buffer back to the original queue.
2041 * Note the buffer transitions from the queued state, to the submitted and wait
2042 * listed state and back to the queued state during this call. The buffer
2043 * locking and queue management logic between _delwri_pushbuf() and
2044 * _delwri_queue() guarantee that the buffer cannot be queued to another list
2048 xfs_buf_delwri_pushbuf(
2050 struct list_head *buffer_list)
2052 LIST_HEAD (submit_list);
2055 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
2057 trace_xfs_buf_delwri_pushbuf(bp, _RET_IP_);
2060 * Isolate the buffer to a new local list so we can submit it for I/O
2061 * independently from the rest of the original list.
2064 list_move(&bp->b_list, &submit_list);
2068 * Delwri submission clears the DELWRI_Q buffer flag and returns with
2069 * the buffer on the wait list with the original reference. Rather than
2070 * bounce the buffer from a local wait list back to the original list
2071 * after I/O completion, reuse the original list as the wait list.
2073 xfs_buf_delwri_submit_buffers(&submit_list, buffer_list);
2076 * The buffer is now locked, under I/O and wait listed on the original
2077 * delwri queue. Wait for I/O completion, restore the DELWRI_Q flag and
2078 * return with the buffer unlocked and on the original queue.
2080 error = xfs_buf_iowait(bp);
2081 bp->b_flags |= _XBF_DELWRI_Q;
2090 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
2091 KM_ZONE_HWALIGN, NULL);
2102 xfs_buf_terminate(void)
2104 kmem_zone_destroy(xfs_buf_zone);
2107 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref)
2110 * Set the lru reference count to 0 based on the error injection tag.
2111 * This allows userspace to disrupt buffer caching for debug/testing
2114 if (XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_LRU_REF))
2117 atomic_set(&bp->b_lru_ref, lru_ref);
2121 * Verify an on-disk magic value against the magic value specified in the
2122 * verifier structure. The verifier magic is in disk byte order so the caller is
2123 * expected to pass the value directly from disk.
2130 struct xfs_mount *mp = bp->b_mount;
2133 idx = xfs_sb_version_hascrc(&mp->m_sb);
2134 if (WARN_ON(!bp->b_ops || !bp->b_ops->magic[idx]))
2136 return dmagic == bp->b_ops->magic[idx];
2139 * Verify an on-disk magic value against the magic value specified in the
2140 * verifier structure. The verifier magic is in disk byte order so the caller is
2141 * expected to pass the value directly from disk.
2148 struct xfs_mount *mp = bp->b_mount;
2151 idx = xfs_sb_version_hascrc(&mp->m_sb);
2152 if (WARN_ON(!bp->b_ops || !bp->b_ops->magic16[idx]))
2154 return dmagic == bp->b_ops->magic16[idx];