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 if (current->reclaim_state)
331 current->reclaim_state->reclaimed_slab +=
333 } else if (bp->b_flags & _XBF_KMEM)
334 kmem_free(bp->b_addr);
335 _xfs_buf_free_pages(bp);
336 xfs_buf_free_maps(bp);
337 kmem_cache_free(xfs_buf_zone, bp);
341 * Allocates all the pages for buffer in question and builds it's page list.
344 xfs_buf_allocate_memory(
349 size_t nbytes, offset;
350 gfp_t gfp_mask = xb_to_gfp(flags);
351 unsigned short page_count, i;
352 xfs_off_t start, end;
354 xfs_km_flags_t kmflag_mask = 0;
357 * assure zeroed buffer for non-read cases.
359 if (!(flags & XBF_READ)) {
360 kmflag_mask |= KM_ZERO;
361 gfp_mask |= __GFP_ZERO;
365 * for buffers that are contained within a single page, just allocate
366 * the memory from the heap - there's no need for the complexity of
367 * page arrays to keep allocation down to order 0.
369 size = BBTOB(bp->b_length);
370 if (size < PAGE_SIZE) {
371 int align_mask = xfs_buftarg_dma_alignment(bp->b_target);
372 bp->b_addr = kmem_alloc_io(size, align_mask,
373 KM_NOFS | kmflag_mask);
375 /* low memory - use alloc_page loop instead */
379 if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) !=
380 ((unsigned long)bp->b_addr & PAGE_MASK)) {
381 /* b_addr spans two pages - use alloc_page instead */
382 kmem_free(bp->b_addr);
386 bp->b_offset = offset_in_page(bp->b_addr);
387 bp->b_pages = bp->b_page_array;
388 bp->b_pages[0] = kmem_to_page(bp->b_addr);
389 bp->b_page_count = 1;
390 bp->b_flags |= _XBF_KMEM;
395 start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT;
396 end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1)
398 page_count = end - start;
399 error = _xfs_buf_get_pages(bp, page_count);
403 offset = bp->b_offset;
404 bp->b_flags |= _XBF_PAGES;
406 for (i = 0; i < bp->b_page_count; i++) {
410 page = alloc_page(gfp_mask);
411 if (unlikely(page == NULL)) {
412 if (flags & XBF_READ_AHEAD) {
413 bp->b_page_count = i;
419 * This could deadlock.
421 * But until all the XFS lowlevel code is revamped to
422 * handle buffer allocation failures we can't do much.
424 if (!(++retries % 100))
426 "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)",
427 current->comm, current->pid,
430 XFS_STATS_INC(bp->b_mount, xb_page_retries);
431 congestion_wait(BLK_RW_ASYNC, HZ/50);
435 XFS_STATS_INC(bp->b_mount, xb_page_found);
437 nbytes = min_t(size_t, size, PAGE_SIZE - offset);
439 bp->b_pages[i] = page;
445 for (i = 0; i < bp->b_page_count; i++)
446 __free_page(bp->b_pages[i]);
447 bp->b_flags &= ~_XBF_PAGES;
452 * Map buffer into kernel address-space if necessary.
459 ASSERT(bp->b_flags & _XBF_PAGES);
460 if (bp->b_page_count == 1) {
461 /* A single page buffer is always mappable */
462 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
463 } else if (flags & XBF_UNMAPPED) {
470 * vm_map_ram() will allocate auxiliary structures (e.g.
471 * pagetables) with GFP_KERNEL, yet we are likely to be under
472 * GFP_NOFS context here. Hence we need to tell memory reclaim
473 * that we are in such a context via PF_MEMALLOC_NOFS to prevent
474 * memory reclaim re-entering the filesystem here and
475 * potentially deadlocking.
477 nofs_flag = memalloc_nofs_save();
479 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
484 } while (retried++ <= 1);
485 memalloc_nofs_restore(nofs_flag);
489 bp->b_addr += bp->b_offset;
496 * Finding and Reading Buffers
500 struct rhashtable_compare_arg *arg,
503 const struct xfs_buf_map *map = arg->key;
504 const struct xfs_buf *bp = obj;
507 * The key hashing in the lookup path depends on the key being the
508 * first element of the compare_arg, make sure to assert this.
510 BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0);
512 if (bp->b_bn != map->bm_bn)
515 if (unlikely(bp->b_length != map->bm_len)) {
517 * found a block number match. If the range doesn't
518 * match, the only way this is allowed is if the buffer
519 * in the cache is stale and the transaction that made
520 * it stale has not yet committed. i.e. we are
521 * reallocating a busy extent. Skip this buffer and
522 * continue searching for an exact match.
524 ASSERT(bp->b_flags & XBF_STALE);
530 static const struct rhashtable_params xfs_buf_hash_params = {
531 .min_size = 32, /* empty AGs have minimal footprint */
533 .key_len = sizeof(xfs_daddr_t),
534 .key_offset = offsetof(struct xfs_buf, b_bn),
535 .head_offset = offsetof(struct xfs_buf, b_rhash_head),
536 .automatic_shrinking = true,
537 .obj_cmpfn = _xfs_buf_obj_cmp,
542 struct xfs_perag *pag)
544 spin_lock_init(&pag->pag_buf_lock);
545 return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params);
549 xfs_buf_hash_destroy(
550 struct xfs_perag *pag)
552 rhashtable_destroy(&pag->pag_buf_hash);
556 * Look up a buffer in the buffer cache and return it referenced and locked
559 * If @new_bp is supplied and we have a lookup miss, insert @new_bp into the
562 * If XBF_TRYLOCK is set in @flags, only try to lock the buffer and return
563 * -EAGAIN if we fail to lock it.
566 * -EFSCORRUPTED if have been supplied with an invalid address
567 * -EAGAIN on trylock failure
568 * -ENOENT if we fail to find a match and @new_bp was NULL
570 * - @new_bp if we inserted it into the cache
571 * - the buffer we found and locked.
575 struct xfs_buftarg *btp,
576 struct xfs_buf_map *map,
578 xfs_buf_flags_t flags,
579 struct xfs_buf *new_bp,
580 struct xfs_buf **found_bp)
582 struct xfs_perag *pag;
584 struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn };
590 for (i = 0; i < nmaps; i++)
591 cmap.bm_len += map[i].bm_len;
593 /* Check for IOs smaller than the sector size / not sector aligned */
594 ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize));
595 ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask));
598 * Corrupted block numbers can get through to here, unfortunately, so we
599 * have to check that the buffer falls within the filesystem bounds.
601 eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks);
602 if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) {
603 xfs_alert(btp->bt_mount,
604 "%s: daddr 0x%llx out of range, EOFS 0x%llx",
605 __func__, cmap.bm_bn, eofs);
607 return -EFSCORRUPTED;
610 pag = xfs_perag_get(btp->bt_mount,
611 xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn));
613 spin_lock(&pag->pag_buf_lock);
614 bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap,
615 xfs_buf_hash_params);
617 atomic_inc(&bp->b_hold);
623 XFS_STATS_INC(btp->bt_mount, xb_miss_locked);
624 spin_unlock(&pag->pag_buf_lock);
629 /* the buffer keeps the perag reference until it is freed */
631 rhashtable_insert_fast(&pag->pag_buf_hash, &new_bp->b_rhash_head,
632 xfs_buf_hash_params);
633 spin_unlock(&pag->pag_buf_lock);
638 spin_unlock(&pag->pag_buf_lock);
641 if (!xfs_buf_trylock(bp)) {
642 if (flags & XBF_TRYLOCK) {
644 XFS_STATS_INC(btp->bt_mount, xb_busy_locked);
648 XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited);
652 * if the buffer is stale, clear all the external state associated with
653 * it. We need to keep flags such as how we allocated the buffer memory
656 if (bp->b_flags & XBF_STALE) {
657 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
658 ASSERT(bp->b_iodone == NULL);
659 bp->b_flags &= _XBF_KMEM | _XBF_PAGES;
663 trace_xfs_buf_find(bp, flags, _RET_IP_);
664 XFS_STATS_INC(btp->bt_mount, xb_get_locked);
671 struct xfs_buftarg *target,
674 xfs_buf_flags_t flags)
678 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
680 error = xfs_buf_find(target, &map, 1, flags, NULL, &bp);
687 * Assembles a buffer covering the specified range. The code is optimised for
688 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
689 * more hits than misses.
693 struct xfs_buftarg *target,
694 struct xfs_buf_map *map,
696 xfs_buf_flags_t flags,
697 struct xfs_buf **bpp)
700 struct xfs_buf *new_bp;
704 error = xfs_buf_find(target, map, nmaps, flags, NULL, &bp);
707 if (error != -ENOENT)
710 error = _xfs_buf_alloc(target, map, nmaps, flags, &new_bp);
714 error = xfs_buf_allocate_memory(new_bp, flags);
716 xfs_buf_free(new_bp);
720 error = xfs_buf_find(target, map, nmaps, flags, new_bp, &bp);
722 xfs_buf_free(new_bp);
727 xfs_buf_free(new_bp);
731 error = _xfs_buf_map_pages(bp, flags);
732 if (unlikely(error)) {
733 xfs_warn_ratelimited(target->bt_mount,
734 "%s: failed to map %u pages", __func__,
742 * Clear b_error if this is a lookup from a caller that doesn't expect
743 * valid data to be found in the buffer.
745 if (!(flags & XBF_READ))
746 xfs_buf_ioerror(bp, 0);
748 XFS_STATS_INC(target->bt_mount, xb_get);
749 trace_xfs_buf_get(bp, flags, _RET_IP_);
757 xfs_buf_flags_t flags)
759 ASSERT(!(flags & XBF_WRITE));
760 ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL);
762 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD);
763 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
765 return xfs_buf_submit(bp);
769 * Reverify a buffer found in cache without an attached ->b_ops.
771 * If the caller passed an ops structure and the buffer doesn't have ops
772 * assigned, set the ops and use it to verify the contents. If verification
773 * fails, clear XBF_DONE. We assume the buffer has no recorded errors and is
774 * already in XBF_DONE state on entry.
776 * Under normal operations, every in-core buffer is verified on read I/O
777 * completion. There are two scenarios that can lead to in-core buffers without
778 * an assigned ->b_ops. The first is during log recovery of buffers on a V4
779 * filesystem, though these buffers are purged at the end of recovery. The
780 * other is online repair, which intentionally reads with a NULL buffer ops to
781 * run several verifiers across an in-core buffer in order to establish buffer
782 * type. If repair can't establish that, the buffer will be left in memory
783 * with NULL buffer ops.
788 const struct xfs_buf_ops *ops)
790 ASSERT(bp->b_flags & XBF_DONE);
791 ASSERT(bp->b_error == 0);
793 if (!ops || bp->b_ops)
797 bp->b_ops->verify_read(bp);
799 bp->b_flags &= ~XBF_DONE;
805 struct xfs_buftarg *target,
806 struct xfs_buf_map *map,
808 xfs_buf_flags_t flags,
809 struct xfs_buf **bpp,
810 const struct xfs_buf_ops *ops,
819 error = xfs_buf_get_map(target, map, nmaps, flags, &bp);
823 trace_xfs_buf_read(bp, flags, _RET_IP_);
825 if (!(bp->b_flags & XBF_DONE)) {
826 /* Initiate the buffer read and wait. */
827 XFS_STATS_INC(target->bt_mount, xb_get_read);
829 error = _xfs_buf_read(bp, flags);
831 /* Readahead iodone already dropped the buffer, so exit. */
832 if (flags & XBF_ASYNC)
835 /* Buffer already read; all we need to do is check it. */
836 error = xfs_buf_reverify(bp, ops);
838 /* Readahead already finished; drop the buffer and exit. */
839 if (flags & XBF_ASYNC) {
844 /* We do not want read in the flags */
845 bp->b_flags &= ~XBF_READ;
846 ASSERT(bp->b_ops != NULL || ops == NULL);
850 * If we've had a read error, then the contents of the buffer are
851 * invalid and should not be used. To ensure that a followup read tries
852 * to pull the buffer from disk again, we clear the XBF_DONE flag and
853 * mark the buffer stale. This ensures that anyone who has a current
854 * reference to the buffer will interpret it's contents correctly and
855 * future cache lookups will also treat it as an empty, uninitialised
859 if (!XFS_FORCED_SHUTDOWN(target->bt_mount))
860 xfs_buf_ioerror_alert(bp, fa);
862 bp->b_flags &= ~XBF_DONE;
866 /* bad CRC means corrupted metadata */
867 if (error == -EFSBADCRC)
868 error = -EFSCORRUPTED;
877 * If we are not low on memory then do the readahead in a deadlock
881 xfs_buf_readahead_map(
882 struct xfs_buftarg *target,
883 struct xfs_buf_map *map,
885 const struct xfs_buf_ops *ops)
889 if (bdi_read_congested(target->bt_bdev->bd_bdi))
892 xfs_buf_read_map(target, map, nmaps,
893 XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD, &bp, ops,
898 * Read an uncached buffer from disk. Allocates and returns a locked
899 * buffer containing the disk contents or nothing.
902 xfs_buf_read_uncached(
903 struct xfs_buftarg *target,
907 struct xfs_buf **bpp,
908 const struct xfs_buf_ops *ops)
915 error = xfs_buf_get_uncached(target, numblks, flags, &bp);
919 /* set up the buffer for a read IO */
920 ASSERT(bp->b_map_count == 1);
921 bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */
922 bp->b_maps[0].bm_bn = daddr;
923 bp->b_flags |= XBF_READ;
938 xfs_buf_get_uncached(
939 struct xfs_buftarg *target,
942 struct xfs_buf **bpp)
944 unsigned long page_count;
947 DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks);
951 /* flags might contain irrelevant bits, pass only what we care about */
952 error = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT, &bp);
956 page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT;
957 error = _xfs_buf_get_pages(bp, page_count);
961 for (i = 0; i < page_count; i++) {
962 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
963 if (!bp->b_pages[i]) {
968 bp->b_flags |= _XBF_PAGES;
970 error = _xfs_buf_map_pages(bp, 0);
971 if (unlikely(error)) {
972 xfs_warn(target->bt_mount,
973 "%s: failed to map pages", __func__);
977 trace_xfs_buf_get_uncached(bp, _RET_IP_);
983 __free_page(bp->b_pages[i]);
984 _xfs_buf_free_pages(bp);
986 xfs_buf_free_maps(bp);
987 kmem_cache_free(xfs_buf_zone, bp);
993 * Increment reference count on buffer, to hold the buffer concurrently
994 * with another thread which may release (free) the buffer asynchronously.
995 * Must hold the buffer already to call this function.
1001 trace_xfs_buf_hold(bp, _RET_IP_);
1002 atomic_inc(&bp->b_hold);
1006 * Release a hold on the specified buffer. If the hold count is 1, the buffer is
1007 * placed on LRU or freed (depending on b_lru_ref).
1013 struct xfs_perag *pag = bp->b_pag;
1015 bool freebuf = false;
1017 trace_xfs_buf_rele(bp, _RET_IP_);
1020 ASSERT(list_empty(&bp->b_lru));
1021 if (atomic_dec_and_test(&bp->b_hold)) {
1022 xfs_buf_ioacct_dec(bp);
1028 ASSERT(atomic_read(&bp->b_hold) > 0);
1031 * We grab the b_lock here first to serialise racing xfs_buf_rele()
1032 * calls. The pag_buf_lock being taken on the last reference only
1033 * serialises against racing lookups in xfs_buf_find(). IOWs, the second
1034 * to last reference we drop here is not serialised against the last
1035 * reference until we take bp->b_lock. Hence if we don't grab b_lock
1036 * first, the last "release" reference can win the race to the lock and
1037 * free the buffer before the second-to-last reference is processed,
1038 * leading to a use-after-free scenario.
1040 spin_lock(&bp->b_lock);
1041 release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock);
1044 * Drop the in-flight state if the buffer is already on the LRU
1045 * and it holds the only reference. This is racy because we
1046 * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT
1047 * ensures the decrement occurs only once per-buf.
1049 if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru))
1050 __xfs_buf_ioacct_dec(bp);
1054 /* the last reference has been dropped ... */
1055 __xfs_buf_ioacct_dec(bp);
1056 if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) {
1058 * If the buffer is added to the LRU take a new reference to the
1059 * buffer for the LRU and clear the (now stale) dispose list
1062 if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) {
1063 bp->b_state &= ~XFS_BSTATE_DISPOSE;
1064 atomic_inc(&bp->b_hold);
1066 spin_unlock(&pag->pag_buf_lock);
1069 * most of the time buffers will already be removed from the
1070 * LRU, so optimise that case by checking for the
1071 * XFS_BSTATE_DISPOSE flag indicating the last list the buffer
1072 * was on was the disposal list
1074 if (!(bp->b_state & XFS_BSTATE_DISPOSE)) {
1075 list_lru_del(&bp->b_target->bt_lru, &bp->b_lru);
1077 ASSERT(list_empty(&bp->b_lru));
1080 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1081 rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head,
1082 xfs_buf_hash_params);
1083 spin_unlock(&pag->pag_buf_lock);
1089 spin_unlock(&bp->b_lock);
1097 * Lock a buffer object, if it is not already locked.
1099 * If we come across a stale, pinned, locked buffer, we know that we are
1100 * being asked to lock a buffer that has been reallocated. Because it is
1101 * pinned, we know that the log has not been pushed to disk and hence it
1102 * will still be locked. Rather than continuing to have trylock attempts
1103 * fail until someone else pushes the log, push it ourselves before
1104 * returning. This means that the xfsaild will not get stuck trying
1105 * to push on stale inode buffers.
1113 locked = down_trylock(&bp->b_sema) == 0;
1115 trace_xfs_buf_trylock(bp, _RET_IP_);
1117 trace_xfs_buf_trylock_fail(bp, _RET_IP_);
1122 * Lock a buffer object.
1124 * If we come across a stale, pinned, locked buffer, we know that we
1125 * are being asked to lock a buffer that has been reallocated. Because
1126 * it is pinned, we know that the log has not been pushed to disk and
1127 * hence it will still be locked. Rather than sleeping until someone
1128 * else pushes the log, push it ourselves before trying to get the lock.
1134 trace_xfs_buf_lock(bp, _RET_IP_);
1136 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
1137 xfs_log_force(bp->b_mount, 0);
1140 trace_xfs_buf_lock_done(bp, _RET_IP_);
1147 ASSERT(xfs_buf_islocked(bp));
1150 trace_xfs_buf_unlock(bp, _RET_IP_);
1157 DECLARE_WAITQUEUE (wait, current);
1159 if (atomic_read(&bp->b_pin_count) == 0)
1162 add_wait_queue(&bp->b_waiters, &wait);
1164 set_current_state(TASK_UNINTERRUPTIBLE);
1165 if (atomic_read(&bp->b_pin_count) == 0)
1169 remove_wait_queue(&bp->b_waiters, &wait);
1170 set_current_state(TASK_RUNNING);
1174 * Buffer Utility Routines
1181 bool read = bp->b_flags & XBF_READ;
1183 trace_xfs_buf_iodone(bp, _RET_IP_);
1185 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1188 * Pull in IO completion errors now. We are guaranteed to be running
1189 * single threaded, so we don't need the lock to read b_io_error.
1191 if (!bp->b_error && bp->b_io_error)
1192 xfs_buf_ioerror(bp, bp->b_io_error);
1194 /* Only validate buffers that were read without errors */
1195 if (read && !bp->b_error && bp->b_ops) {
1196 ASSERT(!bp->b_iodone);
1197 bp->b_ops->verify_read(bp);
1201 bp->b_flags &= ~XBF_WRITE_FAIL;
1202 bp->b_flags |= XBF_DONE;
1206 (*(bp->b_iodone))(bp);
1207 else if (bp->b_flags & XBF_ASYNC)
1210 complete(&bp->b_iowait);
1215 struct work_struct *work)
1217 struct xfs_buf *bp =
1218 container_of(work, xfs_buf_t, b_ioend_work);
1224 xfs_buf_ioend_async(
1227 INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work);
1228 queue_work(bp->b_mount->m_buf_workqueue, &bp->b_ioend_work);
1235 xfs_failaddr_t failaddr)
1237 ASSERT(error <= 0 && error >= -1000);
1238 bp->b_error = error;
1239 trace_xfs_buf_ioerror(bp, error, failaddr);
1243 xfs_buf_ioerror_alert(
1245 xfs_failaddr_t func)
1247 xfs_buf_alert_ratelimited(bp, "XFS: metadata IO error",
1248 "metadata I/O error in \"%pS\" at daddr 0x%llx len %d error %d",
1249 func, (uint64_t)XFS_BUF_ADDR(bp),
1250 bp->b_length, -bp->b_error);
1254 * To simulate an I/O failure, the buffer must be locked and held with at least
1255 * three references. The LRU reference is dropped by the stale call. The buf
1256 * item reference is dropped via ioend processing. The third reference is owned
1257 * by the caller and is dropped on I/O completion if the buffer is XBF_ASYNC.
1263 bp->b_flags &= ~XBF_DONE;
1265 xfs_buf_ioerror(bp, -EIO);
1275 ASSERT(xfs_buf_islocked(bp));
1277 bp->b_flags |= XBF_WRITE;
1278 bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q |
1281 error = xfs_buf_submit(bp);
1283 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
1291 struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private;
1293 if (!bio->bi_status &&
1294 (bp->b_flags & XBF_WRITE) && (bp->b_flags & XBF_ASYNC) &&
1295 XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_IOERROR))
1296 bio->bi_status = BLK_STS_IOERR;
1299 * don't overwrite existing errors - otherwise we can lose errors on
1300 * buffers that require multiple bios to complete.
1302 if (bio->bi_status) {
1303 int error = blk_status_to_errno(bio->bi_status);
1305 cmpxchg(&bp->b_io_error, 0, error);
1308 if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1309 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1311 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1312 xfs_buf_ioend_async(bp);
1317 xfs_buf_ioapply_map(
1325 int total_nr_pages = bp->b_page_count;
1328 sector_t sector = bp->b_maps[map].bm_bn;
1332 /* skip the pages in the buffer before the start offset */
1334 offset = *buf_offset;
1335 while (offset >= PAGE_SIZE) {
1337 offset -= PAGE_SIZE;
1341 * Limit the IO size to the length of the current vector, and update the
1342 * remaining IO count for the next time around.
1344 size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count);
1346 *buf_offset += size;
1349 atomic_inc(&bp->b_io_remaining);
1350 nr_pages = min(total_nr_pages, BIO_MAX_PAGES);
1352 bio = bio_alloc(GFP_NOIO, nr_pages);
1353 bio_set_dev(bio, bp->b_target->bt_bdev);
1354 bio->bi_iter.bi_sector = sector;
1355 bio->bi_end_io = xfs_buf_bio_end_io;
1356 bio->bi_private = bp;
1359 for (; size && nr_pages; nr_pages--, page_index++) {
1360 int rbytes, nbytes = PAGE_SIZE - offset;
1365 rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes,
1367 if (rbytes < nbytes)
1371 sector += BTOBB(nbytes);
1376 if (likely(bio->bi_iter.bi_size)) {
1377 if (xfs_buf_is_vmapped(bp)) {
1378 flush_kernel_vmap_range(bp->b_addr,
1379 xfs_buf_vmap_len(bp));
1386 * This is guaranteed not to be the last io reference count
1387 * because the caller (xfs_buf_submit) holds a count itself.
1389 atomic_dec(&bp->b_io_remaining);
1390 xfs_buf_ioerror(bp, -EIO);
1400 struct blk_plug plug;
1407 * Make sure we capture only current IO errors rather than stale errors
1408 * left over from previous use of the buffer (e.g. failed readahead).
1412 if (bp->b_flags & XBF_WRITE) {
1416 * Run the write verifier callback function if it exists. If
1417 * this function fails it will mark the buffer with an error and
1418 * the IO should not be dispatched.
1421 bp->b_ops->verify_write(bp);
1423 xfs_force_shutdown(bp->b_mount,
1424 SHUTDOWN_CORRUPT_INCORE);
1427 } else if (bp->b_bn != XFS_BUF_DADDR_NULL) {
1428 struct xfs_mount *mp = bp->b_mount;
1431 * non-crc filesystems don't attach verifiers during
1432 * log recovery, so don't warn for such filesystems.
1434 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1436 "%s: no buf ops on daddr 0x%llx len %d",
1437 __func__, bp->b_bn, bp->b_length);
1438 xfs_hex_dump(bp->b_addr,
1439 XFS_CORRUPTION_DUMP_LEN);
1445 if (bp->b_flags & XBF_READ_AHEAD)
1449 /* we only use the buffer cache for meta-data */
1453 * Walk all the vectors issuing IO on them. Set up the initial offset
1454 * into the buffer and the desired IO size before we start -
1455 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1458 offset = bp->b_offset;
1459 size = BBTOB(bp->b_length);
1460 blk_start_plug(&plug);
1461 for (i = 0; i < bp->b_map_count; i++) {
1462 xfs_buf_ioapply_map(bp, i, &offset, &size, op);
1466 break; /* all done */
1468 blk_finish_plug(&plug);
1472 * Wait for I/O completion of a sync buffer and return the I/O error code.
1478 ASSERT(!(bp->b_flags & XBF_ASYNC));
1480 trace_xfs_buf_iowait(bp, _RET_IP_);
1481 wait_for_completion(&bp->b_iowait);
1482 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1488 * Buffer I/O submission path, read or write. Asynchronous submission transfers
1489 * the buffer lock ownership and the current reference to the IO. It is not
1490 * safe to reference the buffer after a call to this function unless the caller
1491 * holds an additional reference itself.
1500 trace_xfs_buf_submit(bp, _RET_IP_);
1502 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1504 /* on shutdown we stale and complete the buffer immediately */
1505 if (XFS_FORCED_SHUTDOWN(bp->b_mount)) {
1506 xfs_buf_ioend_fail(bp);
1511 * Grab a reference so the buffer does not go away underneath us. For
1512 * async buffers, I/O completion drops the callers reference, which
1513 * could occur before submission returns.
1517 if (bp->b_flags & XBF_WRITE)
1518 xfs_buf_wait_unpin(bp);
1520 /* clear the internal error state to avoid spurious errors */
1524 * Set the count to 1 initially, this will stop an I/O completion
1525 * callout which happens before we have started all the I/O from calling
1526 * xfs_buf_ioend too early.
1528 atomic_set(&bp->b_io_remaining, 1);
1529 if (bp->b_flags & XBF_ASYNC)
1530 xfs_buf_ioacct_inc(bp);
1531 _xfs_buf_ioapply(bp);
1534 * If _xfs_buf_ioapply failed, we can get back here with only the IO
1535 * reference we took above. If we drop it to zero, run completion so
1536 * that we don't return to the caller with completion still pending.
1538 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1539 if (bp->b_error || !(bp->b_flags & XBF_ASYNC))
1542 xfs_buf_ioend_async(bp);
1546 error = xfs_buf_iowait(bp);
1549 * Release the hold that keeps the buffer referenced for the entire
1550 * I/O. Note that if the buffer is async, it is not safe to reference
1551 * after this release.
1565 return bp->b_addr + offset;
1567 offset += bp->b_offset;
1568 page = bp->b_pages[offset >> PAGE_SHIFT];
1569 return page_address(page) + (offset & (PAGE_SIZE-1));
1580 bend = boff + bsize;
1581 while (boff < bend) {
1583 int page_index, page_offset, csize;
1585 page_index = (boff + bp->b_offset) >> PAGE_SHIFT;
1586 page_offset = (boff + bp->b_offset) & ~PAGE_MASK;
1587 page = bp->b_pages[page_index];
1588 csize = min_t(size_t, PAGE_SIZE - page_offset,
1589 BBTOB(bp->b_length) - boff);
1591 ASSERT((csize + page_offset) <= PAGE_SIZE);
1593 memset(page_address(page) + page_offset, 0, csize);
1600 * Log a message about and stale a buffer that a caller has decided is corrupt.
1602 * This function should be called for the kinds of metadata corruption that
1603 * cannot be detect from a verifier, such as incorrect inter-block relationship
1604 * data. Do /not/ call this function from a verifier function.
1606 * The buffer must be XBF_DONE prior to the call. Afterwards, the buffer will
1607 * be marked stale, but b_error will not be set. The caller is responsible for
1608 * releasing the buffer or fixing it.
1611 __xfs_buf_mark_corrupt(
1615 ASSERT(bp->b_flags & XBF_DONE);
1617 xfs_buf_corruption_error(bp, fa);
1622 * Handling of buffer targets (buftargs).
1626 * Wait for any bufs with callbacks that have been submitted but have not yet
1627 * returned. These buffers will have an elevated hold count, so wait on those
1628 * while freeing all the buffers only held by the LRU.
1630 static enum lru_status
1631 xfs_buftarg_wait_rele(
1632 struct list_head *item,
1633 struct list_lru_one *lru,
1634 spinlock_t *lru_lock,
1638 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1639 struct list_head *dispose = arg;
1641 if (atomic_read(&bp->b_hold) > 1) {
1642 /* need to wait, so skip it this pass */
1643 trace_xfs_buf_wait_buftarg(bp, _RET_IP_);
1646 if (!spin_trylock(&bp->b_lock))
1650 * clear the LRU reference count so the buffer doesn't get
1651 * ignored in xfs_buf_rele().
1653 atomic_set(&bp->b_lru_ref, 0);
1654 bp->b_state |= XFS_BSTATE_DISPOSE;
1655 list_lru_isolate_move(lru, item, dispose);
1656 spin_unlock(&bp->b_lock);
1662 struct xfs_buftarg *btp)
1666 bool write_fail = false;
1669 * First wait on the buftarg I/O count for all in-flight buffers to be
1670 * released. This is critical as new buffers do not make the LRU until
1671 * they are released.
1673 * Next, flush the buffer workqueue to ensure all completion processing
1674 * has finished. Just waiting on buffer locks is not sufficient for
1675 * async IO as the reference count held over IO is not released until
1676 * after the buffer lock is dropped. Hence we need to ensure here that
1677 * all reference counts have been dropped before we start walking the
1680 while (percpu_counter_sum(&btp->bt_io_count))
1682 flush_workqueue(btp->bt_mount->m_buf_workqueue);
1684 /* loop until there is nothing left on the lru list. */
1685 while (list_lru_count(&btp->bt_lru)) {
1686 list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele,
1687 &dispose, LONG_MAX);
1689 while (!list_empty(&dispose)) {
1691 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1692 list_del_init(&bp->b_lru);
1693 if (bp->b_flags & XBF_WRITE_FAIL) {
1695 xfs_buf_alert_ratelimited(bp,
1696 "XFS: Corruption Alert",
1697 "Corruption Alert: Buffer at daddr 0x%llx had permanent write failures!",
1698 (long long)bp->b_bn);
1707 * If one or more failed buffers were freed, that means dirty metadata
1708 * was thrown away. This should only ever happen after I/O completion
1709 * handling has elevated I/O error(s) to permanent failures and shuts
1713 ASSERT(XFS_FORCED_SHUTDOWN(btp->bt_mount));
1714 xfs_alert(btp->bt_mount,
1715 "Please run xfs_repair to determine the extent of the problem.");
1719 static enum lru_status
1720 xfs_buftarg_isolate(
1721 struct list_head *item,
1722 struct list_lru_one *lru,
1723 spinlock_t *lru_lock,
1726 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1727 struct list_head *dispose = arg;
1730 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1731 * If we fail to get the lock, just skip it.
1733 if (!spin_trylock(&bp->b_lock))
1736 * Decrement the b_lru_ref count unless the value is already
1737 * zero. If the value is already zero, we need to reclaim the
1738 * buffer, otherwise it gets another trip through the LRU.
1740 if (atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1741 spin_unlock(&bp->b_lock);
1745 bp->b_state |= XFS_BSTATE_DISPOSE;
1746 list_lru_isolate_move(lru, item, dispose);
1747 spin_unlock(&bp->b_lock);
1751 static unsigned long
1752 xfs_buftarg_shrink_scan(
1753 struct shrinker *shrink,
1754 struct shrink_control *sc)
1756 struct xfs_buftarg *btp = container_of(shrink,
1757 struct xfs_buftarg, bt_shrinker);
1759 unsigned long freed;
1761 freed = list_lru_shrink_walk(&btp->bt_lru, sc,
1762 xfs_buftarg_isolate, &dispose);
1764 while (!list_empty(&dispose)) {
1766 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1767 list_del_init(&bp->b_lru);
1774 static unsigned long
1775 xfs_buftarg_shrink_count(
1776 struct shrinker *shrink,
1777 struct shrink_control *sc)
1779 struct xfs_buftarg *btp = container_of(shrink,
1780 struct xfs_buftarg, bt_shrinker);
1781 return list_lru_shrink_count(&btp->bt_lru, sc);
1786 struct xfs_buftarg *btp)
1788 unregister_shrinker(&btp->bt_shrinker);
1789 ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0);
1790 percpu_counter_destroy(&btp->bt_io_count);
1791 list_lru_destroy(&btp->bt_lru);
1793 xfs_blkdev_issue_flush(btp);
1799 xfs_setsize_buftarg(
1801 unsigned int sectorsize)
1803 /* Set up metadata sector size info */
1804 btp->bt_meta_sectorsize = sectorsize;
1805 btp->bt_meta_sectormask = sectorsize - 1;
1807 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1808 xfs_warn(btp->bt_mount,
1809 "Cannot set_blocksize to %u on device %pg",
1810 sectorsize, btp->bt_bdev);
1814 /* Set up device logical sector size mask */
1815 btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev);
1816 btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1;
1822 * When allocating the initial buffer target we have not yet
1823 * read in the superblock, so don't know what sized sectors
1824 * are being used at this early stage. Play safe.
1827 xfs_setsize_buftarg_early(
1829 struct block_device *bdev)
1831 return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev));
1836 struct xfs_mount *mp,
1837 struct block_device *bdev,
1838 struct dax_device *dax_dev)
1842 btp = kmem_zalloc(sizeof(*btp), KM_NOFS);
1845 btp->bt_dev = bdev->bd_dev;
1846 btp->bt_bdev = bdev;
1847 btp->bt_daxdev = dax_dev;
1850 * Buffer IO error rate limiting. Limit it to no more than 10 messages
1851 * per 30 seconds so as to not spam logs too much on repeated errors.
1853 ratelimit_state_init(&btp->bt_ioerror_rl, 30 * HZ,
1854 DEFAULT_RATELIMIT_BURST);
1856 if (xfs_setsize_buftarg_early(btp, bdev))
1859 if (list_lru_init(&btp->bt_lru))
1862 if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL))
1865 btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count;
1866 btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan;
1867 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1868 btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE;
1869 if (register_shrinker(&btp->bt_shrinker))
1874 percpu_counter_destroy(&btp->bt_io_count);
1876 list_lru_destroy(&btp->bt_lru);
1883 * Cancel a delayed write list.
1885 * Remove each buffer from the list, clear the delwri queue flag and drop the
1886 * associated buffer reference.
1889 xfs_buf_delwri_cancel(
1890 struct list_head *list)
1894 while (!list_empty(list)) {
1895 bp = list_first_entry(list, struct xfs_buf, b_list);
1898 bp->b_flags &= ~_XBF_DELWRI_Q;
1899 list_del_init(&bp->b_list);
1905 * Add a buffer to the delayed write list.
1907 * This queues a buffer for writeout if it hasn't already been. Note that
1908 * neither this routine nor the buffer list submission functions perform
1909 * any internal synchronization. It is expected that the lists are thread-local
1912 * Returns true if we queued up the buffer, or false if it already had
1913 * been on the buffer list.
1916 xfs_buf_delwri_queue(
1918 struct list_head *list)
1920 ASSERT(xfs_buf_islocked(bp));
1921 ASSERT(!(bp->b_flags & XBF_READ));
1924 * If the buffer is already marked delwri it already is queued up
1925 * by someone else for imediate writeout. Just ignore it in that
1928 if (bp->b_flags & _XBF_DELWRI_Q) {
1929 trace_xfs_buf_delwri_queued(bp, _RET_IP_);
1933 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1936 * If a buffer gets written out synchronously or marked stale while it
1937 * is on a delwri list we lazily remove it. To do this, the other party
1938 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1939 * It remains referenced and on the list. In a rare corner case it
1940 * might get readded to a delwri list after the synchronous writeout, in
1941 * which case we need just need to re-add the flag here.
1943 bp->b_flags |= _XBF_DELWRI_Q;
1944 if (list_empty(&bp->b_list)) {
1945 atomic_inc(&bp->b_hold);
1946 list_add_tail(&bp->b_list, list);
1953 * Compare function is more complex than it needs to be because
1954 * the return value is only 32 bits and we are doing comparisons
1960 struct list_head *a,
1961 struct list_head *b)
1963 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1964 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1967 diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn;
1976 * Submit buffers for write. If wait_list is specified, the buffers are
1977 * submitted using sync I/O and placed on the wait list such that the caller can
1978 * iowait each buffer. Otherwise async I/O is used and the buffers are released
1979 * at I/O completion time. In either case, buffers remain locked until I/O
1980 * completes and the buffer is released from the queue.
1983 xfs_buf_delwri_submit_buffers(
1984 struct list_head *buffer_list,
1985 struct list_head *wait_list)
1987 struct xfs_buf *bp, *n;
1989 struct blk_plug plug;
1991 list_sort(NULL, buffer_list, xfs_buf_cmp);
1993 blk_start_plug(&plug);
1994 list_for_each_entry_safe(bp, n, buffer_list, b_list) {
1996 if (xfs_buf_ispinned(bp)) {
2000 if (!xfs_buf_trylock(bp))
2007 * Someone else might have written the buffer synchronously or
2008 * marked it stale in the meantime. In that case only the
2009 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
2010 * reference and remove it from the list here.
2012 if (!(bp->b_flags & _XBF_DELWRI_Q)) {
2013 list_del_init(&bp->b_list);
2018 trace_xfs_buf_delwri_split(bp, _RET_IP_);
2021 * If we have a wait list, each buffer (and associated delwri
2022 * queue reference) transfers to it and is submitted
2023 * synchronously. Otherwise, drop the buffer from the delwri
2024 * queue and submit async.
2026 bp->b_flags &= ~_XBF_DELWRI_Q;
2027 bp->b_flags |= XBF_WRITE;
2029 bp->b_flags &= ~XBF_ASYNC;
2030 list_move_tail(&bp->b_list, wait_list);
2032 bp->b_flags |= XBF_ASYNC;
2033 list_del_init(&bp->b_list);
2035 __xfs_buf_submit(bp, false);
2037 blk_finish_plug(&plug);
2043 * Write out a buffer list asynchronously.
2045 * This will take the @buffer_list, write all non-locked and non-pinned buffers
2046 * out and not wait for I/O completion on any of the buffers. This interface
2047 * is only safely useable for callers that can track I/O completion by higher
2048 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
2051 * Note: this function will skip buffers it would block on, and in doing so
2052 * leaves them on @buffer_list so they can be retried on a later pass. As such,
2053 * it is up to the caller to ensure that the buffer list is fully submitted or
2054 * cancelled appropriately when they are finished with the list. Failure to
2055 * cancel or resubmit the list until it is empty will result in leaked buffers
2059 xfs_buf_delwri_submit_nowait(
2060 struct list_head *buffer_list)
2062 return xfs_buf_delwri_submit_buffers(buffer_list, NULL);
2066 * Write out a buffer list synchronously.
2068 * This will take the @buffer_list, write all buffers out and wait for I/O
2069 * completion on all of the buffers. @buffer_list is consumed by the function,
2070 * so callers must have some other way of tracking buffers if they require such
2074 xfs_buf_delwri_submit(
2075 struct list_head *buffer_list)
2077 LIST_HEAD (wait_list);
2078 int error = 0, error2;
2081 xfs_buf_delwri_submit_buffers(buffer_list, &wait_list);
2083 /* Wait for IO to complete. */
2084 while (!list_empty(&wait_list)) {
2085 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
2087 list_del_init(&bp->b_list);
2090 * Wait on the locked buffer, check for errors and unlock and
2091 * release the delwri queue reference.
2093 error2 = xfs_buf_iowait(bp);
2103 * Push a single buffer on a delwri queue.
2105 * The purpose of this function is to submit a single buffer of a delwri queue
2106 * and return with the buffer still on the original queue. The waiting delwri
2107 * buffer submission infrastructure guarantees transfer of the delwri queue
2108 * buffer reference to a temporary wait list. We reuse this infrastructure to
2109 * transfer the buffer back to the original queue.
2111 * Note the buffer transitions from the queued state, to the submitted and wait
2112 * listed state and back to the queued state during this call. The buffer
2113 * locking and queue management logic between _delwri_pushbuf() and
2114 * _delwri_queue() guarantee that the buffer cannot be queued to another list
2118 xfs_buf_delwri_pushbuf(
2120 struct list_head *buffer_list)
2122 LIST_HEAD (submit_list);
2125 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
2127 trace_xfs_buf_delwri_pushbuf(bp, _RET_IP_);
2130 * Isolate the buffer to a new local list so we can submit it for I/O
2131 * independently from the rest of the original list.
2134 list_move(&bp->b_list, &submit_list);
2138 * Delwri submission clears the DELWRI_Q buffer flag and returns with
2139 * the buffer on the wait list with the original reference. Rather than
2140 * bounce the buffer from a local wait list back to the original list
2141 * after I/O completion, reuse the original list as the wait list.
2143 xfs_buf_delwri_submit_buffers(&submit_list, buffer_list);
2146 * The buffer is now locked, under I/O and wait listed on the original
2147 * delwri queue. Wait for I/O completion, restore the DELWRI_Q flag and
2148 * return with the buffer unlocked and on the original queue.
2150 error = xfs_buf_iowait(bp);
2151 bp->b_flags |= _XBF_DELWRI_Q;
2160 xfs_buf_zone = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
2161 SLAB_HWCACHE_ALIGN |
2162 SLAB_RECLAIM_ACCOUNT |
2175 xfs_buf_terminate(void)
2177 kmem_cache_destroy(xfs_buf_zone);
2180 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref)
2183 * Set the lru reference count to 0 based on the error injection tag.
2184 * This allows userspace to disrupt buffer caching for debug/testing
2187 if (XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_LRU_REF))
2190 atomic_set(&bp->b_lru_ref, lru_ref);
2194 * Verify an on-disk magic value against the magic value specified in the
2195 * verifier structure. The verifier magic is in disk byte order so the caller is
2196 * expected to pass the value directly from disk.
2203 struct xfs_mount *mp = bp->b_mount;
2206 idx = xfs_sb_version_hascrc(&mp->m_sb);
2207 if (WARN_ON(!bp->b_ops || !bp->b_ops->magic[idx]))
2209 return dmagic == bp->b_ops->magic[idx];
2212 * Verify an on-disk magic value against the magic value specified in the
2213 * verifier structure. The verifier magic is in disk byte order so the caller is
2214 * expected to pass the value directly from disk.
2221 struct xfs_mount *mp = bp->b_mount;
2224 idx = xfs_sb_version_hascrc(&mp->m_sb);
2225 if (WARN_ON(!bp->b_ops || !bp->b_ops->magic16[idx]))
2227 return dmagic == bp->b_ops->magic16[idx];