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);
203 struct xfs_buftarg *target,
204 struct xfs_buf_map *map,
206 xfs_buf_flags_t flags,
207 struct xfs_buf **bpp)
214 bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS);
219 * We don't want certain flags to appear in b_flags unless they are
220 * specifically set by later operations on the buffer.
222 flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD);
224 atomic_set(&bp->b_hold, 1);
225 atomic_set(&bp->b_lru_ref, 1);
226 init_completion(&bp->b_iowait);
227 INIT_LIST_HEAD(&bp->b_lru);
228 INIT_LIST_HEAD(&bp->b_list);
229 INIT_LIST_HEAD(&bp->b_li_list);
230 sema_init(&bp->b_sema, 0); /* held, no waiters */
231 spin_lock_init(&bp->b_lock);
232 bp->b_target = target;
233 bp->b_mount = target->bt_mount;
237 * Set length and io_length to the same value initially.
238 * I/O routines should use io_length, which will be the same in
239 * most cases but may be reset (e.g. XFS recovery).
241 error = xfs_buf_get_maps(bp, nmaps);
243 kmem_cache_free(xfs_buf_zone, bp);
247 bp->b_bn = map[0].bm_bn;
249 for (i = 0; i < nmaps; i++) {
250 bp->b_maps[i].bm_bn = map[i].bm_bn;
251 bp->b_maps[i].bm_len = map[i].bm_len;
252 bp->b_length += map[i].bm_len;
255 atomic_set(&bp->b_pin_count, 0);
256 init_waitqueue_head(&bp->b_waiters);
258 XFS_STATS_INC(bp->b_mount, xb_create);
259 trace_xfs_buf_init(bp, _RET_IP_);
266 * Allocate a page array capable of holding a specified number
267 * of pages, and point the page buf at it.
274 /* Make sure that we have a page list */
275 if (bp->b_pages == NULL) {
276 bp->b_page_count = page_count;
277 if (page_count <= XB_PAGES) {
278 bp->b_pages = bp->b_page_array;
280 bp->b_pages = kmem_alloc(sizeof(struct page *) *
281 page_count, KM_NOFS);
282 if (bp->b_pages == NULL)
285 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
291 * Frees b_pages if it was allocated.
297 if (bp->b_pages != bp->b_page_array) {
298 kmem_free(bp->b_pages);
304 * Releases the specified buffer.
306 * The modification state of any associated pages is left unchanged.
307 * The buffer must not be on any hash - use xfs_buf_rele instead for
308 * hashed and refcounted buffers
314 trace_xfs_buf_free(bp, _RET_IP_);
316 ASSERT(list_empty(&bp->b_lru));
318 if (bp->b_flags & _XBF_PAGES) {
321 if (xfs_buf_is_vmapped(bp))
322 vm_unmap_ram(bp->b_addr - bp->b_offset,
325 for (i = 0; i < bp->b_page_count; i++) {
326 struct page *page = bp->b_pages[i];
330 } else if (bp->b_flags & _XBF_KMEM)
331 kmem_free(bp->b_addr);
332 _xfs_buf_free_pages(bp);
333 xfs_buf_free_maps(bp);
334 kmem_cache_free(xfs_buf_zone, bp);
338 * Allocates all the pages for buffer in question and builds it's page list.
341 xfs_buf_allocate_memory(
346 size_t nbytes, offset;
347 gfp_t gfp_mask = xb_to_gfp(flags);
348 unsigned short page_count, i;
349 xfs_off_t start, end;
351 xfs_km_flags_t kmflag_mask = 0;
354 * assure zeroed buffer for non-read cases.
356 if (!(flags & XBF_READ)) {
357 kmflag_mask |= KM_ZERO;
358 gfp_mask |= __GFP_ZERO;
362 * for buffers that are contained within a single page, just allocate
363 * the memory from the heap - there's no need for the complexity of
364 * page arrays to keep allocation down to order 0.
366 size = BBTOB(bp->b_length);
367 if (size < PAGE_SIZE) {
368 int align_mask = xfs_buftarg_dma_alignment(bp->b_target);
369 bp->b_addr = kmem_alloc_io(size, align_mask,
370 KM_NOFS | kmflag_mask);
372 /* low memory - use alloc_page loop instead */
376 if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) !=
377 ((unsigned long)bp->b_addr & PAGE_MASK)) {
378 /* b_addr spans two pages - use alloc_page instead */
379 kmem_free(bp->b_addr);
383 bp->b_offset = offset_in_page(bp->b_addr);
384 bp->b_pages = bp->b_page_array;
385 bp->b_pages[0] = kmem_to_page(bp->b_addr);
386 bp->b_page_count = 1;
387 bp->b_flags |= _XBF_KMEM;
392 start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT;
393 end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1)
395 page_count = end - start;
396 error = _xfs_buf_get_pages(bp, page_count);
400 offset = bp->b_offset;
401 bp->b_flags |= _XBF_PAGES;
403 for (i = 0; i < bp->b_page_count; i++) {
407 page = alloc_page(gfp_mask);
408 if (unlikely(page == NULL)) {
409 if (flags & XBF_READ_AHEAD) {
410 bp->b_page_count = i;
416 * This could deadlock.
418 * But until all the XFS lowlevel code is revamped to
419 * handle buffer allocation failures we can't do much.
421 if (!(++retries % 100))
423 "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)",
424 current->comm, current->pid,
427 XFS_STATS_INC(bp->b_mount, xb_page_retries);
428 congestion_wait(BLK_RW_ASYNC, HZ/50);
432 XFS_STATS_INC(bp->b_mount, xb_page_found);
434 nbytes = min_t(size_t, size, PAGE_SIZE - offset);
436 bp->b_pages[i] = page;
442 for (i = 0; i < bp->b_page_count; i++)
443 __free_page(bp->b_pages[i]);
444 bp->b_flags &= ~_XBF_PAGES;
449 * Map buffer into kernel address-space if necessary.
456 ASSERT(bp->b_flags & _XBF_PAGES);
457 if (bp->b_page_count == 1) {
458 /* A single page buffer is always mappable */
459 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
460 } else if (flags & XBF_UNMAPPED) {
467 * vm_map_ram() will allocate auxiliary structures (e.g.
468 * pagetables) with GFP_KERNEL, yet we are likely to be under
469 * GFP_NOFS context here. Hence we need to tell memory reclaim
470 * that we are in such a context via PF_MEMALLOC_NOFS to prevent
471 * memory reclaim re-entering the filesystem here and
472 * potentially deadlocking.
474 nofs_flag = memalloc_nofs_save();
476 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
481 } while (retried++ <= 1);
482 memalloc_nofs_restore(nofs_flag);
486 bp->b_addr += bp->b_offset;
493 * Finding and Reading Buffers
497 struct rhashtable_compare_arg *arg,
500 const struct xfs_buf_map *map = arg->key;
501 const struct xfs_buf *bp = obj;
504 * The key hashing in the lookup path depends on the key being the
505 * first element of the compare_arg, make sure to assert this.
507 BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0);
509 if (bp->b_bn != map->bm_bn)
512 if (unlikely(bp->b_length != map->bm_len)) {
514 * found a block number match. If the range doesn't
515 * match, the only way this is allowed is if the buffer
516 * in the cache is stale and the transaction that made
517 * it stale has not yet committed. i.e. we are
518 * reallocating a busy extent. Skip this buffer and
519 * continue searching for an exact match.
521 ASSERT(bp->b_flags & XBF_STALE);
527 static const struct rhashtable_params xfs_buf_hash_params = {
528 .min_size = 32, /* empty AGs have minimal footprint */
530 .key_len = sizeof(xfs_daddr_t),
531 .key_offset = offsetof(struct xfs_buf, b_bn),
532 .head_offset = offsetof(struct xfs_buf, b_rhash_head),
533 .automatic_shrinking = true,
534 .obj_cmpfn = _xfs_buf_obj_cmp,
539 struct xfs_perag *pag)
541 spin_lock_init(&pag->pag_buf_lock);
542 return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params);
546 xfs_buf_hash_destroy(
547 struct xfs_perag *pag)
549 rhashtable_destroy(&pag->pag_buf_hash);
553 * Look up a buffer in the buffer cache and return it referenced and locked
556 * If @new_bp is supplied and we have a lookup miss, insert @new_bp into the
559 * If XBF_TRYLOCK is set in @flags, only try to lock the buffer and return
560 * -EAGAIN if we fail to lock it.
563 * -EFSCORRUPTED if have been supplied with an invalid address
564 * -EAGAIN on trylock failure
565 * -ENOENT if we fail to find a match and @new_bp was NULL
567 * - @new_bp if we inserted it into the cache
568 * - the buffer we found and locked.
572 struct xfs_buftarg *btp,
573 struct xfs_buf_map *map,
575 xfs_buf_flags_t flags,
576 struct xfs_buf *new_bp,
577 struct xfs_buf **found_bp)
579 struct xfs_perag *pag;
581 struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn };
587 for (i = 0; i < nmaps; i++)
588 cmap.bm_len += map[i].bm_len;
590 /* Check for IOs smaller than the sector size / not sector aligned */
591 ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize));
592 ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask));
595 * Corrupted block numbers can get through to here, unfortunately, so we
596 * have to check that the buffer falls within the filesystem bounds.
598 eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks);
599 if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) {
600 xfs_alert(btp->bt_mount,
601 "%s: daddr 0x%llx out of range, EOFS 0x%llx",
602 __func__, cmap.bm_bn, eofs);
604 return -EFSCORRUPTED;
607 pag = xfs_perag_get(btp->bt_mount,
608 xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn));
610 spin_lock(&pag->pag_buf_lock);
611 bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap,
612 xfs_buf_hash_params);
614 atomic_inc(&bp->b_hold);
620 XFS_STATS_INC(btp->bt_mount, xb_miss_locked);
621 spin_unlock(&pag->pag_buf_lock);
626 /* the buffer keeps the perag reference until it is freed */
628 rhashtable_insert_fast(&pag->pag_buf_hash, &new_bp->b_rhash_head,
629 xfs_buf_hash_params);
630 spin_unlock(&pag->pag_buf_lock);
635 spin_unlock(&pag->pag_buf_lock);
638 if (!xfs_buf_trylock(bp)) {
639 if (flags & XBF_TRYLOCK) {
641 XFS_STATS_INC(btp->bt_mount, xb_busy_locked);
645 XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited);
649 * if the buffer is stale, clear all the external state associated with
650 * it. We need to keep flags such as how we allocated the buffer memory
653 if (bp->b_flags & XBF_STALE) {
654 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
655 ASSERT(bp->b_iodone == NULL);
656 bp->b_flags &= _XBF_KMEM | _XBF_PAGES;
660 trace_xfs_buf_find(bp, flags, _RET_IP_);
661 XFS_STATS_INC(btp->bt_mount, xb_get_locked);
668 struct xfs_buftarg *target,
671 xfs_buf_flags_t flags)
675 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
677 error = xfs_buf_find(target, &map, 1, flags, NULL, &bp);
684 * Assembles a buffer covering the specified range. The code is optimised for
685 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
686 * more hits than misses.
690 struct xfs_buftarg *target,
691 struct xfs_buf_map *map,
693 xfs_buf_flags_t flags,
694 struct xfs_buf **bpp)
697 struct xfs_buf *new_bp;
701 error = xfs_buf_find(target, map, nmaps, flags, NULL, &bp);
704 if (error != -ENOENT)
707 error = _xfs_buf_alloc(target, map, nmaps, flags, &new_bp);
711 error = xfs_buf_allocate_memory(new_bp, flags);
713 xfs_buf_free(new_bp);
717 error = xfs_buf_find(target, map, nmaps, flags, new_bp, &bp);
719 xfs_buf_free(new_bp);
724 xfs_buf_free(new_bp);
728 error = _xfs_buf_map_pages(bp, flags);
729 if (unlikely(error)) {
730 xfs_warn_ratelimited(target->bt_mount,
731 "%s: failed to map %u pages", __func__,
739 * Clear b_error if this is a lookup from a caller that doesn't expect
740 * valid data to be found in the buffer.
742 if (!(flags & XBF_READ))
743 xfs_buf_ioerror(bp, 0);
745 XFS_STATS_INC(target->bt_mount, xb_get);
746 trace_xfs_buf_get(bp, flags, _RET_IP_);
754 xfs_buf_flags_t flags)
756 ASSERT(!(flags & XBF_WRITE));
757 ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL);
759 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD);
760 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
762 return xfs_buf_submit(bp);
766 * Reverify a buffer found in cache without an attached ->b_ops.
768 * If the caller passed an ops structure and the buffer doesn't have ops
769 * assigned, set the ops and use it to verify the contents. If verification
770 * fails, clear XBF_DONE. We assume the buffer has no recorded errors and is
771 * already in XBF_DONE state on entry.
773 * Under normal operations, every in-core buffer is verified on read I/O
774 * completion. There are two scenarios that can lead to in-core buffers without
775 * an assigned ->b_ops. The first is during log recovery of buffers on a V4
776 * filesystem, though these buffers are purged at the end of recovery. The
777 * other is online repair, which intentionally reads with a NULL buffer ops to
778 * run several verifiers across an in-core buffer in order to establish buffer
779 * type. If repair can't establish that, the buffer will be left in memory
780 * with NULL buffer ops.
785 const struct xfs_buf_ops *ops)
787 ASSERT(bp->b_flags & XBF_DONE);
788 ASSERT(bp->b_error == 0);
790 if (!ops || bp->b_ops)
794 bp->b_ops->verify_read(bp);
796 bp->b_flags &= ~XBF_DONE;
802 struct xfs_buftarg *target,
803 struct xfs_buf_map *map,
805 xfs_buf_flags_t flags,
806 struct xfs_buf **bpp,
807 const struct xfs_buf_ops *ops,
816 error = xfs_buf_get_map(target, map, nmaps, flags, &bp);
820 trace_xfs_buf_read(bp, flags, _RET_IP_);
822 if (!(bp->b_flags & XBF_DONE)) {
823 /* Initiate the buffer read and wait. */
824 XFS_STATS_INC(target->bt_mount, xb_get_read);
826 error = _xfs_buf_read(bp, flags);
828 /* Readahead iodone already dropped the buffer, so exit. */
829 if (flags & XBF_ASYNC)
832 /* Buffer already read; all we need to do is check it. */
833 error = xfs_buf_reverify(bp, ops);
835 /* Readahead already finished; drop the buffer and exit. */
836 if (flags & XBF_ASYNC) {
841 /* We do not want read in the flags */
842 bp->b_flags &= ~XBF_READ;
843 ASSERT(bp->b_ops != NULL || ops == NULL);
847 * If we've had a read error, then the contents of the buffer are
848 * invalid and should not be used. To ensure that a followup read tries
849 * to pull the buffer from disk again, we clear the XBF_DONE flag and
850 * mark the buffer stale. This ensures that anyone who has a current
851 * reference to the buffer will interpret it's contents correctly and
852 * future cache lookups will also treat it as an empty, uninitialised
856 if (!XFS_FORCED_SHUTDOWN(target->bt_mount))
857 xfs_buf_ioerror_alert(bp, fa);
859 bp->b_flags &= ~XBF_DONE;
863 /* bad CRC means corrupted metadata */
864 if (error == -EFSBADCRC)
865 error = -EFSCORRUPTED;
874 * If we are not low on memory then do the readahead in a deadlock
878 xfs_buf_readahead_map(
879 struct xfs_buftarg *target,
880 struct xfs_buf_map *map,
882 const struct xfs_buf_ops *ops)
886 if (bdi_read_congested(target->bt_bdev->bd_bdi))
889 xfs_buf_read_map(target, map, nmaps,
890 XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD, &bp, ops,
895 * Read an uncached buffer from disk. Allocates and returns a locked
896 * buffer containing the disk contents or nothing.
899 xfs_buf_read_uncached(
900 struct xfs_buftarg *target,
904 struct xfs_buf **bpp,
905 const struct xfs_buf_ops *ops)
912 error = xfs_buf_get_uncached(target, numblks, flags, &bp);
916 /* set up the buffer for a read IO */
917 ASSERT(bp->b_map_count == 1);
918 bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */
919 bp->b_maps[0].bm_bn = daddr;
920 bp->b_flags |= XBF_READ;
935 xfs_buf_get_uncached(
936 struct xfs_buftarg *target,
939 struct xfs_buf **bpp)
941 unsigned long page_count;
944 DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks);
948 /* flags might contain irrelevant bits, pass only what we care about */
949 error = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT, &bp);
953 page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT;
954 error = _xfs_buf_get_pages(bp, page_count);
958 for (i = 0; i < page_count; i++) {
959 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
960 if (!bp->b_pages[i]) {
965 bp->b_flags |= _XBF_PAGES;
967 error = _xfs_buf_map_pages(bp, 0);
968 if (unlikely(error)) {
969 xfs_warn(target->bt_mount,
970 "%s: failed to map pages", __func__);
974 trace_xfs_buf_get_uncached(bp, _RET_IP_);
980 __free_page(bp->b_pages[i]);
981 _xfs_buf_free_pages(bp);
983 xfs_buf_free_maps(bp);
984 kmem_cache_free(xfs_buf_zone, bp);
990 * Increment reference count on buffer, to hold the buffer concurrently
991 * with another thread which may release (free) the buffer asynchronously.
992 * Must hold the buffer already to call this function.
998 trace_xfs_buf_hold(bp, _RET_IP_);
999 atomic_inc(&bp->b_hold);
1003 * Release a hold on the specified buffer. If the hold count is 1, the buffer is
1004 * placed on LRU or freed (depending on b_lru_ref).
1010 struct xfs_perag *pag = bp->b_pag;
1012 bool freebuf = false;
1014 trace_xfs_buf_rele(bp, _RET_IP_);
1017 ASSERT(list_empty(&bp->b_lru));
1018 if (atomic_dec_and_test(&bp->b_hold)) {
1019 xfs_buf_ioacct_dec(bp);
1025 ASSERT(atomic_read(&bp->b_hold) > 0);
1028 * We grab the b_lock here first to serialise racing xfs_buf_rele()
1029 * calls. The pag_buf_lock being taken on the last reference only
1030 * serialises against racing lookups in xfs_buf_find(). IOWs, the second
1031 * to last reference we drop here is not serialised against the last
1032 * reference until we take bp->b_lock. Hence if we don't grab b_lock
1033 * first, the last "release" reference can win the race to the lock and
1034 * free the buffer before the second-to-last reference is processed,
1035 * leading to a use-after-free scenario.
1037 spin_lock(&bp->b_lock);
1038 release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock);
1041 * Drop the in-flight state if the buffer is already on the LRU
1042 * and it holds the only reference. This is racy because we
1043 * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT
1044 * ensures the decrement occurs only once per-buf.
1046 if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru))
1047 __xfs_buf_ioacct_dec(bp);
1051 /* the last reference has been dropped ... */
1052 __xfs_buf_ioacct_dec(bp);
1053 if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) {
1055 * If the buffer is added to the LRU take a new reference to the
1056 * buffer for the LRU and clear the (now stale) dispose list
1059 if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) {
1060 bp->b_state &= ~XFS_BSTATE_DISPOSE;
1061 atomic_inc(&bp->b_hold);
1063 spin_unlock(&pag->pag_buf_lock);
1066 * most of the time buffers will already be removed from the
1067 * LRU, so optimise that case by checking for the
1068 * XFS_BSTATE_DISPOSE flag indicating the last list the buffer
1069 * was on was the disposal list
1071 if (!(bp->b_state & XFS_BSTATE_DISPOSE)) {
1072 list_lru_del(&bp->b_target->bt_lru, &bp->b_lru);
1074 ASSERT(list_empty(&bp->b_lru));
1077 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1078 rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head,
1079 xfs_buf_hash_params);
1080 spin_unlock(&pag->pag_buf_lock);
1086 spin_unlock(&bp->b_lock);
1094 * Lock a buffer object, if it is not already locked.
1096 * If we come across a stale, pinned, locked buffer, we know that we are
1097 * being asked to lock a buffer that has been reallocated. Because it is
1098 * pinned, we know that the log has not been pushed to disk and hence it
1099 * will still be locked. Rather than continuing to have trylock attempts
1100 * fail until someone else pushes the log, push it ourselves before
1101 * returning. This means that the xfsaild will not get stuck trying
1102 * to push on stale inode buffers.
1110 locked = down_trylock(&bp->b_sema) == 0;
1112 trace_xfs_buf_trylock(bp, _RET_IP_);
1114 trace_xfs_buf_trylock_fail(bp, _RET_IP_);
1119 * Lock a buffer object.
1121 * If we come across a stale, pinned, locked buffer, we know that we
1122 * are being asked to lock a buffer that has been reallocated. Because
1123 * it is pinned, we know that the log has not been pushed to disk and
1124 * hence it will still be locked. Rather than sleeping until someone
1125 * else pushes the log, push it ourselves before trying to get the lock.
1131 trace_xfs_buf_lock(bp, _RET_IP_);
1133 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
1134 xfs_log_force(bp->b_mount, 0);
1137 trace_xfs_buf_lock_done(bp, _RET_IP_);
1144 ASSERT(xfs_buf_islocked(bp));
1147 trace_xfs_buf_unlock(bp, _RET_IP_);
1154 DECLARE_WAITQUEUE (wait, current);
1156 if (atomic_read(&bp->b_pin_count) == 0)
1159 add_wait_queue(&bp->b_waiters, &wait);
1161 set_current_state(TASK_UNINTERRUPTIBLE);
1162 if (atomic_read(&bp->b_pin_count) == 0)
1166 remove_wait_queue(&bp->b_waiters, &wait);
1167 set_current_state(TASK_RUNNING);
1171 * Buffer Utility Routines
1178 bool read = bp->b_flags & XBF_READ;
1180 trace_xfs_buf_iodone(bp, _RET_IP_);
1182 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1185 * Pull in IO completion errors now. We are guaranteed to be running
1186 * single threaded, so we don't need the lock to read b_io_error.
1188 if (!bp->b_error && bp->b_io_error)
1189 xfs_buf_ioerror(bp, bp->b_io_error);
1191 /* Only validate buffers that were read without errors */
1192 if (read && !bp->b_error && bp->b_ops) {
1193 ASSERT(!bp->b_iodone);
1194 bp->b_ops->verify_read(bp);
1198 bp->b_flags |= XBF_DONE;
1201 (*(bp->b_iodone))(bp);
1202 else if (bp->b_flags & XBF_ASYNC)
1205 complete(&bp->b_iowait);
1210 struct work_struct *work)
1212 struct xfs_buf *bp =
1213 container_of(work, xfs_buf_t, b_ioend_work);
1219 xfs_buf_ioend_async(
1222 INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work);
1223 queue_work(bp->b_mount->m_buf_workqueue, &bp->b_ioend_work);
1230 xfs_failaddr_t failaddr)
1232 ASSERT(error <= 0 && error >= -1000);
1233 bp->b_error = error;
1234 trace_xfs_buf_ioerror(bp, error, failaddr);
1238 xfs_buf_ioerror_alert(
1240 xfs_failaddr_t func)
1242 xfs_alert_ratelimited(bp->b_mount,
1243 "metadata I/O error in \"%pS\" at daddr 0x%llx len %d error %d",
1244 func, (uint64_t)XFS_BUF_ADDR(bp), bp->b_length,
1254 ASSERT(xfs_buf_islocked(bp));
1256 bp->b_flags |= XBF_WRITE;
1257 bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q |
1258 XBF_WRITE_FAIL | XBF_DONE);
1260 error = xfs_buf_submit(bp);
1262 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
1270 struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private;
1273 * don't overwrite existing errors - otherwise we can lose errors on
1274 * buffers that require multiple bios to complete.
1276 if (bio->bi_status) {
1277 int error = blk_status_to_errno(bio->bi_status);
1279 cmpxchg(&bp->b_io_error, 0, error);
1282 if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1283 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1285 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1286 xfs_buf_ioend_async(bp);
1291 xfs_buf_ioapply_map(
1299 int total_nr_pages = bp->b_page_count;
1302 sector_t sector = bp->b_maps[map].bm_bn;
1306 /* skip the pages in the buffer before the start offset */
1308 offset = *buf_offset;
1309 while (offset >= PAGE_SIZE) {
1311 offset -= PAGE_SIZE;
1315 * Limit the IO size to the length of the current vector, and update the
1316 * remaining IO count for the next time around.
1318 size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count);
1320 *buf_offset += size;
1323 atomic_inc(&bp->b_io_remaining);
1324 nr_pages = min(total_nr_pages, BIO_MAX_PAGES);
1326 bio = bio_alloc(GFP_NOIO, nr_pages);
1327 bio_set_dev(bio, bp->b_target->bt_bdev);
1328 bio->bi_iter.bi_sector = sector;
1329 bio->bi_end_io = xfs_buf_bio_end_io;
1330 bio->bi_private = bp;
1333 for (; size && nr_pages; nr_pages--, page_index++) {
1334 int rbytes, nbytes = PAGE_SIZE - offset;
1339 rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes,
1341 if (rbytes < nbytes)
1345 sector += BTOBB(nbytes);
1350 if (likely(bio->bi_iter.bi_size)) {
1351 if (xfs_buf_is_vmapped(bp)) {
1352 flush_kernel_vmap_range(bp->b_addr,
1353 xfs_buf_vmap_len(bp));
1360 * This is guaranteed not to be the last io reference count
1361 * because the caller (xfs_buf_submit) holds a count itself.
1363 atomic_dec(&bp->b_io_remaining);
1364 xfs_buf_ioerror(bp, -EIO);
1374 struct blk_plug plug;
1381 * Make sure we capture only current IO errors rather than stale errors
1382 * left over from previous use of the buffer (e.g. failed readahead).
1386 if (bp->b_flags & XBF_WRITE) {
1390 * Run the write verifier callback function if it exists. If
1391 * this function fails it will mark the buffer with an error and
1392 * the IO should not be dispatched.
1395 bp->b_ops->verify_write(bp);
1397 xfs_force_shutdown(bp->b_mount,
1398 SHUTDOWN_CORRUPT_INCORE);
1401 } else if (bp->b_bn != XFS_BUF_DADDR_NULL) {
1402 struct xfs_mount *mp = bp->b_mount;
1405 * non-crc filesystems don't attach verifiers during
1406 * log recovery, so don't warn for such filesystems.
1408 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1410 "%s: no buf ops on daddr 0x%llx len %d",
1411 __func__, bp->b_bn, bp->b_length);
1412 xfs_hex_dump(bp->b_addr,
1413 XFS_CORRUPTION_DUMP_LEN);
1419 if (bp->b_flags & XBF_READ_AHEAD)
1423 /* we only use the buffer cache for meta-data */
1427 * Walk all the vectors issuing IO on them. Set up the initial offset
1428 * into the buffer and the desired IO size before we start -
1429 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1432 offset = bp->b_offset;
1433 size = BBTOB(bp->b_length);
1434 blk_start_plug(&plug);
1435 for (i = 0; i < bp->b_map_count; i++) {
1436 xfs_buf_ioapply_map(bp, i, &offset, &size, op);
1440 break; /* all done */
1442 blk_finish_plug(&plug);
1446 * Wait for I/O completion of a sync buffer and return the I/O error code.
1452 ASSERT(!(bp->b_flags & XBF_ASYNC));
1454 trace_xfs_buf_iowait(bp, _RET_IP_);
1455 wait_for_completion(&bp->b_iowait);
1456 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1462 * Buffer I/O submission path, read or write. Asynchronous submission transfers
1463 * the buffer lock ownership and the current reference to the IO. It is not
1464 * safe to reference the buffer after a call to this function unless the caller
1465 * holds an additional reference itself.
1474 trace_xfs_buf_submit(bp, _RET_IP_);
1476 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1478 /* on shutdown we stale and complete the buffer immediately */
1479 if (XFS_FORCED_SHUTDOWN(bp->b_mount)) {
1480 xfs_buf_ioerror(bp, -EIO);
1481 bp->b_flags &= ~XBF_DONE;
1488 * Grab a reference so the buffer does not go away underneath us. For
1489 * async buffers, I/O completion drops the callers reference, which
1490 * could occur before submission returns.
1494 if (bp->b_flags & XBF_WRITE)
1495 xfs_buf_wait_unpin(bp);
1497 /* clear the internal error state to avoid spurious errors */
1501 * Set the count to 1 initially, this will stop an I/O completion
1502 * callout which happens before we have started all the I/O from calling
1503 * xfs_buf_ioend too early.
1505 atomic_set(&bp->b_io_remaining, 1);
1506 if (bp->b_flags & XBF_ASYNC)
1507 xfs_buf_ioacct_inc(bp);
1508 _xfs_buf_ioapply(bp);
1511 * If _xfs_buf_ioapply failed, we can get back here with only the IO
1512 * reference we took above. If we drop it to zero, run completion so
1513 * that we don't return to the caller with completion still pending.
1515 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1516 if (bp->b_error || !(bp->b_flags & XBF_ASYNC))
1519 xfs_buf_ioend_async(bp);
1523 error = xfs_buf_iowait(bp);
1526 * Release the hold that keeps the buffer referenced for the entire
1527 * I/O. Note that if the buffer is async, it is not safe to reference
1528 * after this release.
1542 return bp->b_addr + offset;
1544 offset += bp->b_offset;
1545 page = bp->b_pages[offset >> PAGE_SHIFT];
1546 return page_address(page) + (offset & (PAGE_SIZE-1));
1557 bend = boff + bsize;
1558 while (boff < bend) {
1560 int page_index, page_offset, csize;
1562 page_index = (boff + bp->b_offset) >> PAGE_SHIFT;
1563 page_offset = (boff + bp->b_offset) & ~PAGE_MASK;
1564 page = bp->b_pages[page_index];
1565 csize = min_t(size_t, PAGE_SIZE - page_offset,
1566 BBTOB(bp->b_length) - boff);
1568 ASSERT((csize + page_offset) <= PAGE_SIZE);
1570 memset(page_address(page) + page_offset, 0, csize);
1577 * Log a message about and stale a buffer that a caller has decided is corrupt.
1579 * This function should be called for the kinds of metadata corruption that
1580 * cannot be detect from a verifier, such as incorrect inter-block relationship
1581 * data. Do /not/ call this function from a verifier function.
1583 * The buffer must be XBF_DONE prior to the call. Afterwards, the buffer will
1584 * be marked stale, but b_error will not be set. The caller is responsible for
1585 * releasing the buffer or fixing it.
1588 __xfs_buf_mark_corrupt(
1592 ASSERT(bp->b_flags & XBF_DONE);
1594 xfs_buf_corruption_error(bp, fa);
1599 * Handling of buffer targets (buftargs).
1603 * Wait for any bufs with callbacks that have been submitted but have not yet
1604 * returned. These buffers will have an elevated hold count, so wait on those
1605 * while freeing all the buffers only held by the LRU.
1607 static enum lru_status
1608 xfs_buftarg_wait_rele(
1609 struct list_head *item,
1610 struct list_lru_one *lru,
1611 spinlock_t *lru_lock,
1615 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1616 struct list_head *dispose = arg;
1618 if (atomic_read(&bp->b_hold) > 1) {
1619 /* need to wait, so skip it this pass */
1620 trace_xfs_buf_wait_buftarg(bp, _RET_IP_);
1623 if (!spin_trylock(&bp->b_lock))
1627 * clear the LRU reference count so the buffer doesn't get
1628 * ignored in xfs_buf_rele().
1630 atomic_set(&bp->b_lru_ref, 0);
1631 bp->b_state |= XFS_BSTATE_DISPOSE;
1632 list_lru_isolate_move(lru, item, dispose);
1633 spin_unlock(&bp->b_lock);
1639 struct xfs_buftarg *btp)
1645 * First wait on the buftarg I/O count for all in-flight buffers to be
1646 * released. This is critical as new buffers do not make the LRU until
1647 * they are released.
1649 * Next, flush the buffer workqueue to ensure all completion processing
1650 * has finished. Just waiting on buffer locks is not sufficient for
1651 * async IO as the reference count held over IO is not released until
1652 * after the buffer lock is dropped. Hence we need to ensure here that
1653 * all reference counts have been dropped before we start walking the
1656 while (percpu_counter_sum(&btp->bt_io_count))
1658 flush_workqueue(btp->bt_mount->m_buf_workqueue);
1660 /* loop until there is nothing left on the lru list. */
1661 while (list_lru_count(&btp->bt_lru)) {
1662 list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele,
1663 &dispose, LONG_MAX);
1665 while (!list_empty(&dispose)) {
1667 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1668 list_del_init(&bp->b_lru);
1669 if (bp->b_flags & XBF_WRITE_FAIL) {
1670 xfs_alert(btp->bt_mount,
1671 "Corruption Alert: Buffer at daddr 0x%llx had permanent write failures!",
1672 (long long)bp->b_bn);
1673 xfs_alert(btp->bt_mount,
1674 "Please run xfs_repair to determine the extent of the problem.");
1683 static enum lru_status
1684 xfs_buftarg_isolate(
1685 struct list_head *item,
1686 struct list_lru_one *lru,
1687 spinlock_t *lru_lock,
1690 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1691 struct list_head *dispose = arg;
1694 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1695 * If we fail to get the lock, just skip it.
1697 if (!spin_trylock(&bp->b_lock))
1700 * Decrement the b_lru_ref count unless the value is already
1701 * zero. If the value is already zero, we need to reclaim the
1702 * buffer, otherwise it gets another trip through the LRU.
1704 if (atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1705 spin_unlock(&bp->b_lock);
1709 bp->b_state |= XFS_BSTATE_DISPOSE;
1710 list_lru_isolate_move(lru, item, dispose);
1711 spin_unlock(&bp->b_lock);
1715 static unsigned long
1716 xfs_buftarg_shrink_scan(
1717 struct shrinker *shrink,
1718 struct shrink_control *sc)
1720 struct xfs_buftarg *btp = container_of(shrink,
1721 struct xfs_buftarg, bt_shrinker);
1723 unsigned long freed;
1725 freed = list_lru_shrink_walk(&btp->bt_lru, sc,
1726 xfs_buftarg_isolate, &dispose);
1728 while (!list_empty(&dispose)) {
1730 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1731 list_del_init(&bp->b_lru);
1738 static unsigned long
1739 xfs_buftarg_shrink_count(
1740 struct shrinker *shrink,
1741 struct shrink_control *sc)
1743 struct xfs_buftarg *btp = container_of(shrink,
1744 struct xfs_buftarg, bt_shrinker);
1745 return list_lru_shrink_count(&btp->bt_lru, sc);
1750 struct xfs_buftarg *btp)
1752 unregister_shrinker(&btp->bt_shrinker);
1753 ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0);
1754 percpu_counter_destroy(&btp->bt_io_count);
1755 list_lru_destroy(&btp->bt_lru);
1757 xfs_blkdev_issue_flush(btp);
1763 xfs_setsize_buftarg(
1765 unsigned int sectorsize)
1767 /* Set up metadata sector size info */
1768 btp->bt_meta_sectorsize = sectorsize;
1769 btp->bt_meta_sectormask = sectorsize - 1;
1771 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1772 xfs_warn(btp->bt_mount,
1773 "Cannot set_blocksize to %u on device %pg",
1774 sectorsize, btp->bt_bdev);
1778 /* Set up device logical sector size mask */
1779 btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev);
1780 btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1;
1786 * When allocating the initial buffer target we have not yet
1787 * read in the superblock, so don't know what sized sectors
1788 * are being used at this early stage. Play safe.
1791 xfs_setsize_buftarg_early(
1793 struct block_device *bdev)
1795 return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev));
1800 struct xfs_mount *mp,
1801 struct block_device *bdev,
1802 struct dax_device *dax_dev)
1806 btp = kmem_zalloc(sizeof(*btp), KM_NOFS);
1809 btp->bt_dev = bdev->bd_dev;
1810 btp->bt_bdev = bdev;
1811 btp->bt_daxdev = dax_dev;
1813 if (xfs_setsize_buftarg_early(btp, bdev))
1816 if (list_lru_init(&btp->bt_lru))
1819 if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL))
1822 btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count;
1823 btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan;
1824 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1825 btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE;
1826 if (register_shrinker(&btp->bt_shrinker))
1831 percpu_counter_destroy(&btp->bt_io_count);
1833 list_lru_destroy(&btp->bt_lru);
1840 * Cancel a delayed write list.
1842 * Remove each buffer from the list, clear the delwri queue flag and drop the
1843 * associated buffer reference.
1846 xfs_buf_delwri_cancel(
1847 struct list_head *list)
1851 while (!list_empty(list)) {
1852 bp = list_first_entry(list, struct xfs_buf, b_list);
1855 bp->b_flags &= ~_XBF_DELWRI_Q;
1856 list_del_init(&bp->b_list);
1862 * Add a buffer to the delayed write list.
1864 * This queues a buffer for writeout if it hasn't already been. Note that
1865 * neither this routine nor the buffer list submission functions perform
1866 * any internal synchronization. It is expected that the lists are thread-local
1869 * Returns true if we queued up the buffer, or false if it already had
1870 * been on the buffer list.
1873 xfs_buf_delwri_queue(
1875 struct list_head *list)
1877 ASSERT(xfs_buf_islocked(bp));
1878 ASSERT(!(bp->b_flags & XBF_READ));
1881 * If the buffer is already marked delwri it already is queued up
1882 * by someone else for imediate writeout. Just ignore it in that
1885 if (bp->b_flags & _XBF_DELWRI_Q) {
1886 trace_xfs_buf_delwri_queued(bp, _RET_IP_);
1890 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1893 * If a buffer gets written out synchronously or marked stale while it
1894 * is on a delwri list we lazily remove it. To do this, the other party
1895 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1896 * It remains referenced and on the list. In a rare corner case it
1897 * might get readded to a delwri list after the synchronous writeout, in
1898 * which case we need just need to re-add the flag here.
1900 bp->b_flags |= _XBF_DELWRI_Q;
1901 if (list_empty(&bp->b_list)) {
1902 atomic_inc(&bp->b_hold);
1903 list_add_tail(&bp->b_list, list);
1910 * Compare function is more complex than it needs to be because
1911 * the return value is only 32 bits and we are doing comparisons
1917 struct list_head *a,
1918 struct list_head *b)
1920 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1921 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1924 diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn;
1933 * Submit buffers for write. If wait_list is specified, the buffers are
1934 * submitted using sync I/O and placed on the wait list such that the caller can
1935 * iowait each buffer. Otherwise async I/O is used and the buffers are released
1936 * at I/O completion time. In either case, buffers remain locked until I/O
1937 * completes and the buffer is released from the queue.
1940 xfs_buf_delwri_submit_buffers(
1941 struct list_head *buffer_list,
1942 struct list_head *wait_list)
1944 struct xfs_buf *bp, *n;
1946 struct blk_plug plug;
1948 list_sort(NULL, buffer_list, xfs_buf_cmp);
1950 blk_start_plug(&plug);
1951 list_for_each_entry_safe(bp, n, buffer_list, b_list) {
1953 if (xfs_buf_ispinned(bp)) {
1957 if (!xfs_buf_trylock(bp))
1964 * Someone else might have written the buffer synchronously or
1965 * marked it stale in the meantime. In that case only the
1966 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
1967 * reference and remove it from the list here.
1969 if (!(bp->b_flags & _XBF_DELWRI_Q)) {
1970 list_del_init(&bp->b_list);
1975 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1978 * If we have a wait list, each buffer (and associated delwri
1979 * queue reference) transfers to it and is submitted
1980 * synchronously. Otherwise, drop the buffer from the delwri
1981 * queue and submit async.
1983 bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_WRITE_FAIL);
1984 bp->b_flags |= XBF_WRITE;
1986 bp->b_flags &= ~XBF_ASYNC;
1987 list_move_tail(&bp->b_list, wait_list);
1989 bp->b_flags |= XBF_ASYNC;
1990 list_del_init(&bp->b_list);
1992 __xfs_buf_submit(bp, false);
1994 blk_finish_plug(&plug);
2000 * Write out a buffer list asynchronously.
2002 * This will take the @buffer_list, write all non-locked and non-pinned buffers
2003 * out and not wait for I/O completion on any of the buffers. This interface
2004 * is only safely useable for callers that can track I/O completion by higher
2005 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
2008 * Note: this function will skip buffers it would block on, and in doing so
2009 * leaves them on @buffer_list so they can be retried on a later pass. As such,
2010 * it is up to the caller to ensure that the buffer list is fully submitted or
2011 * cancelled appropriately when they are finished with the list. Failure to
2012 * cancel or resubmit the list until it is empty will result in leaked buffers
2016 xfs_buf_delwri_submit_nowait(
2017 struct list_head *buffer_list)
2019 return xfs_buf_delwri_submit_buffers(buffer_list, NULL);
2023 * Write out a buffer list synchronously.
2025 * This will take the @buffer_list, write all buffers out and wait for I/O
2026 * completion on all of the buffers. @buffer_list is consumed by the function,
2027 * so callers must have some other way of tracking buffers if they require such
2031 xfs_buf_delwri_submit(
2032 struct list_head *buffer_list)
2034 LIST_HEAD (wait_list);
2035 int error = 0, error2;
2038 xfs_buf_delwri_submit_buffers(buffer_list, &wait_list);
2040 /* Wait for IO to complete. */
2041 while (!list_empty(&wait_list)) {
2042 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
2044 list_del_init(&bp->b_list);
2047 * Wait on the locked buffer, check for errors and unlock and
2048 * release the delwri queue reference.
2050 error2 = xfs_buf_iowait(bp);
2060 * Push a single buffer on a delwri queue.
2062 * The purpose of this function is to submit a single buffer of a delwri queue
2063 * and return with the buffer still on the original queue. The waiting delwri
2064 * buffer submission infrastructure guarantees transfer of the delwri queue
2065 * buffer reference to a temporary wait list. We reuse this infrastructure to
2066 * transfer the buffer back to the original queue.
2068 * Note the buffer transitions from the queued state, to the submitted and wait
2069 * listed state and back to the queued state during this call. The buffer
2070 * locking and queue management logic between _delwri_pushbuf() and
2071 * _delwri_queue() guarantee that the buffer cannot be queued to another list
2075 xfs_buf_delwri_pushbuf(
2077 struct list_head *buffer_list)
2079 LIST_HEAD (submit_list);
2082 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
2084 trace_xfs_buf_delwri_pushbuf(bp, _RET_IP_);
2087 * Isolate the buffer to a new local list so we can submit it for I/O
2088 * independently from the rest of the original list.
2091 list_move(&bp->b_list, &submit_list);
2095 * Delwri submission clears the DELWRI_Q buffer flag and returns with
2096 * the buffer on the wait list with the original reference. Rather than
2097 * bounce the buffer from a local wait list back to the original list
2098 * after I/O completion, reuse the original list as the wait list.
2100 xfs_buf_delwri_submit_buffers(&submit_list, buffer_list);
2103 * The buffer is now locked, under I/O and wait listed on the original
2104 * delwri queue. Wait for I/O completion, restore the DELWRI_Q flag and
2105 * return with the buffer unlocked and on the original queue.
2107 error = xfs_buf_iowait(bp);
2108 bp->b_flags |= _XBF_DELWRI_Q;
2117 xfs_buf_zone = kmem_cache_create("xfs_buf",
2118 sizeof(struct xfs_buf), 0,
2119 SLAB_HWCACHE_ALIGN, NULL);
2130 xfs_buf_terminate(void)
2132 kmem_cache_destroy(xfs_buf_zone);
2135 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref)
2138 * Set the lru reference count to 0 based on the error injection tag.
2139 * This allows userspace to disrupt buffer caching for debug/testing
2142 if (XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_LRU_REF))
2145 atomic_set(&bp->b_lru_ref, lru_ref);
2149 * Verify an on-disk magic value against the magic value specified in the
2150 * verifier structure. The verifier magic is in disk byte order so the caller is
2151 * expected to pass the value directly from disk.
2158 struct xfs_mount *mp = bp->b_mount;
2161 idx = xfs_sb_version_hascrc(&mp->m_sb);
2162 if (WARN_ON(!bp->b_ops || !bp->b_ops->magic[idx]))
2164 return dmagic == bp->b_ops->magic[idx];
2167 * Verify an on-disk magic value against the magic value specified in the
2168 * verifier structure. The verifier magic is in disk byte order so the caller is
2169 * expected to pass the value directly from disk.
2176 struct xfs_mount *mp = bp->b_mount;
2179 idx = xfs_sb_version_hascrc(&mp->m_sb);
2180 if (WARN_ON(!bp->b_ops || !bp->b_ops->magic16[idx]))
2182 return dmagic == bp->b_ops->magic16[idx];