2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include <linux/stddef.h>
20 #include <linux/errno.h>
21 #include <linux/gfp.h>
22 #include <linux/pagemap.h>
23 #include <linux/init.h>
24 #include <linux/vmalloc.h>
25 #include <linux/bio.h>
26 #include <linux/sysctl.h>
27 #include <linux/proc_fs.h>
28 #include <linux/workqueue.h>
29 #include <linux/percpu.h>
30 #include <linux/blkdev.h>
31 #include <linux/hash.h>
32 #include <linux/kthread.h>
33 #include <linux/migrate.h>
34 #include <linux/backing-dev.h>
35 #include <linux/freezer.h>
37 #include "xfs_format.h"
38 #include "xfs_log_format.h"
39 #include "xfs_trans_resv.h"
41 #include "xfs_mount.h"
42 #include "xfs_trace.h"
45 static kmem_zone_t *xfs_buf_zone;
47 #ifdef XFS_BUF_LOCK_TRACKING
48 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
49 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
50 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
52 # define XB_SET_OWNER(bp) do { } while (0)
53 # define XB_CLEAR_OWNER(bp) do { } while (0)
54 # define XB_GET_OWNER(bp) do { } while (0)
57 #define xb_to_gfp(flags) \
58 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN)
66 * Return true if the buffer is vmapped.
68 * b_addr is null if the buffer is not mapped, but the code is clever
69 * enough to know it doesn't have to map a single page, so the check has
70 * to be both for b_addr and bp->b_page_count > 1.
72 return bp->b_addr && bp->b_page_count > 1;
79 return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
83 * Bump the I/O in flight count on the buftarg if we haven't yet done so for
84 * this buffer. The count is incremented once per buffer (per hold cycle)
85 * because the corresponding decrement is deferred to buffer release. Buffers
86 * can undergo I/O multiple times in a hold-release cycle and per buffer I/O
87 * tracking adds unnecessary overhead. This is used for sychronization purposes
88 * with unmount (see xfs_wait_buftarg()), so all we really need is a count of
91 * Buffers that are never released (e.g., superblock, iclog buffers) must set
92 * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count
93 * never reaches zero and unmount hangs indefinitely.
99 if (bp->b_flags & (XBF_NO_IOACCT|_XBF_IN_FLIGHT))
102 ASSERT(bp->b_flags & XBF_ASYNC);
103 bp->b_flags |= _XBF_IN_FLIGHT;
104 percpu_counter_inc(&bp->b_target->bt_io_count);
108 * Clear the in-flight state on a buffer about to be released to the LRU or
109 * freed and unaccount from the buftarg.
115 if (!(bp->b_flags & _XBF_IN_FLIGHT))
118 bp->b_flags &= ~_XBF_IN_FLIGHT;
119 percpu_counter_dec(&bp->b_target->bt_io_count);
123 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
124 * b_lru_ref count so that the buffer is freed immediately when the buffer
125 * reference count falls to zero. If the buffer is already on the LRU, we need
126 * to remove the reference that LRU holds on the buffer.
128 * This prevents build-up of stale buffers on the LRU.
134 ASSERT(xfs_buf_islocked(bp));
136 bp->b_flags |= XBF_STALE;
139 * Clear the delwri status so that a delwri queue walker will not
140 * flush this buffer to disk now that it is stale. The delwri queue has
141 * a reference to the buffer, so this is safe to do.
143 bp->b_flags &= ~_XBF_DELWRI_Q;
146 * Once the buffer is marked stale and unlocked, a subsequent lookup
147 * could reset b_flags. There is no guarantee that the buffer is
148 * unaccounted (released to LRU) before that occurs. Drop in-flight
149 * status now to preserve accounting consistency.
151 xfs_buf_ioacct_dec(bp);
153 spin_lock(&bp->b_lock);
154 atomic_set(&bp->b_lru_ref, 0);
155 if (!(bp->b_state & XFS_BSTATE_DISPOSE) &&
156 (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru)))
157 atomic_dec(&bp->b_hold);
159 ASSERT(atomic_read(&bp->b_hold) >= 1);
160 spin_unlock(&bp->b_lock);
168 ASSERT(bp->b_maps == NULL);
169 bp->b_map_count = map_count;
171 if (map_count == 1) {
172 bp->b_maps = &bp->__b_map;
176 bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map),
184 * Frees b_pages if it was allocated.
190 if (bp->b_maps != &bp->__b_map) {
191 kmem_free(bp->b_maps);
198 struct xfs_buftarg *target,
199 struct xfs_buf_map *map,
201 xfs_buf_flags_t flags)
207 bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS);
212 * We don't want certain flags to appear in b_flags unless they are
213 * specifically set by later operations on the buffer.
215 flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD);
217 atomic_set(&bp->b_hold, 1);
218 atomic_set(&bp->b_lru_ref, 1);
219 init_completion(&bp->b_iowait);
220 INIT_LIST_HEAD(&bp->b_lru);
221 INIT_LIST_HEAD(&bp->b_list);
222 sema_init(&bp->b_sema, 0); /* held, no waiters */
223 spin_lock_init(&bp->b_lock);
225 bp->b_target = target;
229 * Set length and io_length to the same value initially.
230 * I/O routines should use io_length, which will be the same in
231 * most cases but may be reset (e.g. XFS recovery).
233 error = xfs_buf_get_maps(bp, nmaps);
235 kmem_zone_free(xfs_buf_zone, bp);
239 bp->b_bn = map[0].bm_bn;
241 for (i = 0; i < nmaps; i++) {
242 bp->b_maps[i].bm_bn = map[i].bm_bn;
243 bp->b_maps[i].bm_len = map[i].bm_len;
244 bp->b_length += map[i].bm_len;
246 bp->b_io_length = bp->b_length;
248 atomic_set(&bp->b_pin_count, 0);
249 init_waitqueue_head(&bp->b_waiters);
251 XFS_STATS_INC(target->bt_mount, xb_create);
252 trace_xfs_buf_init(bp, _RET_IP_);
258 * Allocate a page array capable of holding a specified number
259 * of pages, and point the page buf at it.
266 /* Make sure that we have a page list */
267 if (bp->b_pages == NULL) {
268 bp->b_page_count = page_count;
269 if (page_count <= XB_PAGES) {
270 bp->b_pages = bp->b_page_array;
272 bp->b_pages = kmem_alloc(sizeof(struct page *) *
273 page_count, KM_NOFS);
274 if (bp->b_pages == NULL)
277 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
283 * Frees b_pages if it was allocated.
289 if (bp->b_pages != bp->b_page_array) {
290 kmem_free(bp->b_pages);
296 * Releases the specified buffer.
298 * The modification state of any associated pages is left unchanged.
299 * The buffer must not be on any hash - use xfs_buf_rele instead for
300 * hashed and refcounted buffers
306 trace_xfs_buf_free(bp, _RET_IP_);
308 ASSERT(list_empty(&bp->b_lru));
310 if (bp->b_flags & _XBF_PAGES) {
313 if (xfs_buf_is_vmapped(bp))
314 vm_unmap_ram(bp->b_addr - bp->b_offset,
317 for (i = 0; i < bp->b_page_count; i++) {
318 struct page *page = bp->b_pages[i];
322 } else if (bp->b_flags & _XBF_KMEM)
323 kmem_free(bp->b_addr);
324 _xfs_buf_free_pages(bp);
325 xfs_buf_free_maps(bp);
326 kmem_zone_free(xfs_buf_zone, bp);
330 * Allocates all the pages for buffer in question and builds it's page list.
333 xfs_buf_allocate_memory(
338 size_t nbytes, offset;
339 gfp_t gfp_mask = xb_to_gfp(flags);
340 unsigned short page_count, i;
341 xfs_off_t start, end;
345 * for buffers that are contained within a single page, just allocate
346 * the memory from the heap - there's no need for the complexity of
347 * page arrays to keep allocation down to order 0.
349 size = BBTOB(bp->b_length);
350 if (size < PAGE_SIZE) {
351 bp->b_addr = kmem_alloc(size, KM_NOFS);
353 /* low memory - use alloc_page loop instead */
357 if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) !=
358 ((unsigned long)bp->b_addr & PAGE_MASK)) {
359 /* b_addr spans two pages - use alloc_page instead */
360 kmem_free(bp->b_addr);
364 bp->b_offset = offset_in_page(bp->b_addr);
365 bp->b_pages = bp->b_page_array;
366 bp->b_pages[0] = virt_to_page(bp->b_addr);
367 bp->b_page_count = 1;
368 bp->b_flags |= _XBF_KMEM;
373 start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT;
374 end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1)
376 page_count = end - start;
377 error = _xfs_buf_get_pages(bp, page_count);
381 offset = bp->b_offset;
382 bp->b_flags |= _XBF_PAGES;
384 for (i = 0; i < bp->b_page_count; i++) {
388 page = alloc_page(gfp_mask);
389 if (unlikely(page == NULL)) {
390 if (flags & XBF_READ_AHEAD) {
391 bp->b_page_count = i;
397 * This could deadlock.
399 * But until all the XFS lowlevel code is revamped to
400 * handle buffer allocation failures we can't do much.
402 if (!(++retries % 100))
404 "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)",
405 current->comm, current->pid,
408 XFS_STATS_INC(bp->b_target->bt_mount, xb_page_retries);
409 congestion_wait(BLK_RW_ASYNC, HZ/50);
413 XFS_STATS_INC(bp->b_target->bt_mount, xb_page_found);
415 nbytes = min_t(size_t, size, PAGE_SIZE - offset);
417 bp->b_pages[i] = page;
423 for (i = 0; i < bp->b_page_count; i++)
424 __free_page(bp->b_pages[i]);
429 * Map buffer into kernel address-space if necessary.
436 ASSERT(bp->b_flags & _XBF_PAGES);
437 if (bp->b_page_count == 1) {
438 /* A single page buffer is always mappable */
439 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
440 } else if (flags & XBF_UNMAPPED) {
447 * vm_map_ram() will allocate auxillary structures (e.g.
448 * pagetables) with GFP_KERNEL, yet we are likely to be under
449 * GFP_NOFS context here. Hence we need to tell memory reclaim
450 * that we are in such a context via PF_MEMALLOC_NOIO to prevent
451 * memory reclaim re-entering the filesystem here and
452 * potentially deadlocking.
454 noio_flag = memalloc_noio_save();
456 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
461 } while (retried++ <= 1);
462 memalloc_noio_restore(noio_flag);
466 bp->b_addr += bp->b_offset;
473 * Finding and Reading Buffers
477 struct rhashtable_compare_arg *arg,
480 const struct xfs_buf_map *map = arg->key;
481 const struct xfs_buf *bp = obj;
484 * The key hashing in the lookup path depends on the key being the
485 * first element of the compare_arg, make sure to assert this.
487 BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0);
489 if (bp->b_bn != map->bm_bn)
492 if (unlikely(bp->b_length != map->bm_len)) {
494 * found a block number match. If the range doesn't
495 * match, the only way this is allowed is if the buffer
496 * in the cache is stale and the transaction that made
497 * it stale has not yet committed. i.e. we are
498 * reallocating a busy extent. Skip this buffer and
499 * continue searching for an exact match.
501 ASSERT(bp->b_flags & XBF_STALE);
507 static const struct rhashtable_params xfs_buf_hash_params = {
508 .min_size = 32, /* empty AGs have minimal footprint */
510 .key_len = sizeof(xfs_daddr_t),
511 .key_offset = offsetof(struct xfs_buf, b_bn),
512 .head_offset = offsetof(struct xfs_buf, b_rhash_head),
513 .automatic_shrinking = true,
514 .obj_cmpfn = _xfs_buf_obj_cmp,
519 struct xfs_perag *pag)
521 spin_lock_init(&pag->pag_buf_lock);
522 return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params);
526 xfs_buf_hash_destroy(
527 struct xfs_perag *pag)
529 rhashtable_destroy(&pag->pag_buf_hash);
533 * Look up, and creates if absent, a lockable buffer for
534 * a given range of an inode. The buffer is returned
535 * locked. No I/O is implied by this call.
539 struct xfs_buftarg *btp,
540 struct xfs_buf_map *map,
542 xfs_buf_flags_t flags,
545 struct xfs_perag *pag;
547 struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn };
551 for (i = 0; i < nmaps; i++)
552 cmap.bm_len += map[i].bm_len;
554 /* Check for IOs smaller than the sector size / not sector aligned */
555 ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize));
556 ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask));
559 * Corrupted block numbers can get through to here, unfortunately, so we
560 * have to check that the buffer falls within the filesystem bounds.
562 eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks);
563 if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) {
565 * XXX (dgc): we should really be returning -EFSCORRUPTED here,
566 * but none of the higher level infrastructure supports
567 * returning a specific error on buffer lookup failures.
569 xfs_alert(btp->bt_mount,
570 "%s: Block out of range: block 0x%llx, EOFS 0x%llx ",
571 __func__, cmap.bm_bn, eofs);
576 pag = xfs_perag_get(btp->bt_mount,
577 xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn));
579 spin_lock(&pag->pag_buf_lock);
580 bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap,
581 xfs_buf_hash_params);
583 atomic_inc(&bp->b_hold);
589 /* the buffer keeps the perag reference until it is freed */
591 rhashtable_insert_fast(&pag->pag_buf_hash,
592 &new_bp->b_rhash_head,
593 xfs_buf_hash_params);
594 spin_unlock(&pag->pag_buf_lock);
596 XFS_STATS_INC(btp->bt_mount, xb_miss_locked);
597 spin_unlock(&pag->pag_buf_lock);
603 spin_unlock(&pag->pag_buf_lock);
606 if (!xfs_buf_trylock(bp)) {
607 if (flags & XBF_TRYLOCK) {
609 XFS_STATS_INC(btp->bt_mount, xb_busy_locked);
613 XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited);
617 * if the buffer is stale, clear all the external state associated with
618 * it. We need to keep flags such as how we allocated the buffer memory
621 if (bp->b_flags & XBF_STALE) {
622 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
623 ASSERT(bp->b_iodone == NULL);
624 bp->b_flags &= _XBF_KMEM | _XBF_PAGES;
628 trace_xfs_buf_find(bp, flags, _RET_IP_);
629 XFS_STATS_INC(btp->bt_mount, xb_get_locked);
634 * Assembles a buffer covering the specified range. The code is optimised for
635 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
636 * more hits than misses.
640 struct xfs_buftarg *target,
641 struct xfs_buf_map *map,
643 xfs_buf_flags_t flags)
646 struct xfs_buf *new_bp;
649 bp = _xfs_buf_find(target, map, nmaps, flags, NULL);
653 new_bp = _xfs_buf_alloc(target, map, nmaps, flags);
654 if (unlikely(!new_bp))
657 error = xfs_buf_allocate_memory(new_bp, flags);
659 xfs_buf_free(new_bp);
663 bp = _xfs_buf_find(target, map, nmaps, flags, new_bp);
665 xfs_buf_free(new_bp);
670 xfs_buf_free(new_bp);
674 error = _xfs_buf_map_pages(bp, flags);
675 if (unlikely(error)) {
676 xfs_warn(target->bt_mount,
677 "%s: failed to map pagesn", __func__);
684 * Clear b_error if this is a lookup from a caller that doesn't expect
685 * valid data to be found in the buffer.
687 if (!(flags & XBF_READ))
688 xfs_buf_ioerror(bp, 0);
690 XFS_STATS_INC(target->bt_mount, xb_get);
691 trace_xfs_buf_get(bp, flags, _RET_IP_);
698 xfs_buf_flags_t flags)
700 ASSERT(!(flags & XBF_WRITE));
701 ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL);
703 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD);
704 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
706 if (flags & XBF_ASYNC) {
710 return xfs_buf_submit_wait(bp);
715 struct xfs_buftarg *target,
716 struct xfs_buf_map *map,
718 xfs_buf_flags_t flags,
719 const struct xfs_buf_ops *ops)
725 bp = xfs_buf_get_map(target, map, nmaps, flags);
727 trace_xfs_buf_read(bp, flags, _RET_IP_);
729 if (!(bp->b_flags & XBF_DONE)) {
730 XFS_STATS_INC(target->bt_mount, xb_get_read);
732 _xfs_buf_read(bp, flags);
733 } else if (flags & XBF_ASYNC) {
735 * Read ahead call which is already satisfied,
741 /* We do not want read in the flags */
742 bp->b_flags &= ~XBF_READ;
750 * If we are not low on memory then do the readahead in a deadlock
754 xfs_buf_readahead_map(
755 struct xfs_buftarg *target,
756 struct xfs_buf_map *map,
758 const struct xfs_buf_ops *ops)
760 if (bdi_read_congested(target->bt_bdi))
763 xfs_buf_read_map(target, map, nmaps,
764 XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops);
768 * Read an uncached buffer from disk. Allocates and returns a locked
769 * buffer containing the disk contents or nothing.
772 xfs_buf_read_uncached(
773 struct xfs_buftarg *target,
777 struct xfs_buf **bpp,
778 const struct xfs_buf_ops *ops)
784 bp = xfs_buf_get_uncached(target, numblks, flags);
788 /* set up the buffer for a read IO */
789 ASSERT(bp->b_map_count == 1);
790 bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */
791 bp->b_maps[0].bm_bn = daddr;
792 bp->b_flags |= XBF_READ;
795 xfs_buf_submit_wait(bp);
797 int error = bp->b_error;
807 * Return a buffer allocated as an empty buffer and associated to external
808 * memory via xfs_buf_associate_memory() back to it's empty state.
816 _xfs_buf_free_pages(bp);
819 bp->b_page_count = 0;
821 bp->b_length = numblks;
822 bp->b_io_length = numblks;
824 ASSERT(bp->b_map_count == 1);
825 bp->b_bn = XFS_BUF_DADDR_NULL;
826 bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL;
827 bp->b_maps[0].bm_len = bp->b_length;
830 static inline struct page *
834 if ((!is_vmalloc_addr(addr))) {
835 return virt_to_page(addr);
837 return vmalloc_to_page(addr);
842 xfs_buf_associate_memory(
849 unsigned long pageaddr;
850 unsigned long offset;
854 pageaddr = (unsigned long)mem & PAGE_MASK;
855 offset = (unsigned long)mem - pageaddr;
856 buflen = PAGE_ALIGN(len + offset);
857 page_count = buflen >> PAGE_SHIFT;
859 /* Free any previous set of page pointers */
861 _xfs_buf_free_pages(bp);
866 rval = _xfs_buf_get_pages(bp, page_count);
870 bp->b_offset = offset;
872 for (i = 0; i < bp->b_page_count; i++) {
873 bp->b_pages[i] = mem_to_page((void *)pageaddr);
874 pageaddr += PAGE_SIZE;
877 bp->b_io_length = BTOBB(len);
878 bp->b_length = BTOBB(buflen);
884 xfs_buf_get_uncached(
885 struct xfs_buftarg *target,
889 unsigned long page_count;
892 DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks);
894 /* flags might contain irrelevant bits, pass only what we care about */
895 bp = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT);
896 if (unlikely(bp == NULL))
899 page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT;
900 error = _xfs_buf_get_pages(bp, page_count);
904 for (i = 0; i < page_count; i++) {
905 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
909 bp->b_flags |= _XBF_PAGES;
911 error = _xfs_buf_map_pages(bp, 0);
912 if (unlikely(error)) {
913 xfs_warn(target->bt_mount,
914 "%s: failed to map pages", __func__);
918 trace_xfs_buf_get_uncached(bp, _RET_IP_);
923 __free_page(bp->b_pages[i]);
924 _xfs_buf_free_pages(bp);
926 xfs_buf_free_maps(bp);
927 kmem_zone_free(xfs_buf_zone, bp);
933 * Increment reference count on buffer, to hold the buffer concurrently
934 * with another thread which may release (free) the buffer asynchronously.
935 * Must hold the buffer already to call this function.
941 trace_xfs_buf_hold(bp, _RET_IP_);
942 atomic_inc(&bp->b_hold);
946 * Release a hold on the specified buffer. If the hold count is 1, the buffer is
947 * placed on LRU or freed (depending on b_lru_ref).
953 struct xfs_perag *pag = bp->b_pag;
955 bool freebuf = false;
957 trace_xfs_buf_rele(bp, _RET_IP_);
960 ASSERT(list_empty(&bp->b_lru));
961 if (atomic_dec_and_test(&bp->b_hold)) {
962 xfs_buf_ioacct_dec(bp);
968 ASSERT(atomic_read(&bp->b_hold) > 0);
970 release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock);
971 spin_lock(&bp->b_lock);
974 * Drop the in-flight state if the buffer is already on the LRU
975 * and it holds the only reference. This is racy because we
976 * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT
977 * ensures the decrement occurs only once per-buf.
979 if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru))
980 xfs_buf_ioacct_dec(bp);
984 /* the last reference has been dropped ... */
985 xfs_buf_ioacct_dec(bp);
986 if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) {
988 * If the buffer is added to the LRU take a new reference to the
989 * buffer for the LRU and clear the (now stale) dispose list
992 if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) {
993 bp->b_state &= ~XFS_BSTATE_DISPOSE;
994 atomic_inc(&bp->b_hold);
996 spin_unlock(&pag->pag_buf_lock);
999 * most of the time buffers will already be removed from the
1000 * LRU, so optimise that case by checking for the
1001 * XFS_BSTATE_DISPOSE flag indicating the last list the buffer
1002 * was on was the disposal list
1004 if (!(bp->b_state & XFS_BSTATE_DISPOSE)) {
1005 list_lru_del(&bp->b_target->bt_lru, &bp->b_lru);
1007 ASSERT(list_empty(&bp->b_lru));
1010 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1011 rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head,
1012 xfs_buf_hash_params);
1013 spin_unlock(&pag->pag_buf_lock);
1019 spin_unlock(&bp->b_lock);
1027 * Lock a buffer object, if it is not already locked.
1029 * If we come across a stale, pinned, locked buffer, we know that we are
1030 * being asked to lock a buffer that has been reallocated. Because it is
1031 * pinned, we know that the log has not been pushed to disk and hence it
1032 * will still be locked. Rather than continuing to have trylock attempts
1033 * fail until someone else pushes the log, push it ourselves before
1034 * returning. This means that the xfsaild will not get stuck trying
1035 * to push on stale inode buffers.
1043 locked = down_trylock(&bp->b_sema) == 0;
1046 trace_xfs_buf_trylock(bp, _RET_IP_);
1048 trace_xfs_buf_trylock_fail(bp, _RET_IP_);
1054 * Lock a buffer object.
1056 * If we come across a stale, pinned, locked buffer, we know that we
1057 * are being asked to lock a buffer that has been reallocated. Because
1058 * it is pinned, we know that the log has not been pushed to disk and
1059 * hence it will still be locked. Rather than sleeping until someone
1060 * else pushes the log, push it ourselves before trying to get the lock.
1066 trace_xfs_buf_lock(bp, _RET_IP_);
1068 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
1069 xfs_log_force(bp->b_target->bt_mount, 0);
1073 trace_xfs_buf_lock_done(bp, _RET_IP_);
1083 trace_xfs_buf_unlock(bp, _RET_IP_);
1090 DECLARE_WAITQUEUE (wait, current);
1092 if (atomic_read(&bp->b_pin_count) == 0)
1095 add_wait_queue(&bp->b_waiters, &wait);
1097 set_current_state(TASK_UNINTERRUPTIBLE);
1098 if (atomic_read(&bp->b_pin_count) == 0)
1102 remove_wait_queue(&bp->b_waiters, &wait);
1103 set_current_state(TASK_RUNNING);
1107 * Buffer Utility Routines
1114 bool read = bp->b_flags & XBF_READ;
1116 trace_xfs_buf_iodone(bp, _RET_IP_);
1118 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1121 * Pull in IO completion errors now. We are guaranteed to be running
1122 * single threaded, so we don't need the lock to read b_io_error.
1124 if (!bp->b_error && bp->b_io_error)
1125 xfs_buf_ioerror(bp, bp->b_io_error);
1127 /* Only validate buffers that were read without errors */
1128 if (read && !bp->b_error && bp->b_ops) {
1129 ASSERT(!bp->b_iodone);
1130 bp->b_ops->verify_read(bp);
1134 bp->b_flags |= XBF_DONE;
1137 (*(bp->b_iodone))(bp);
1138 else if (bp->b_flags & XBF_ASYNC)
1141 complete(&bp->b_iowait);
1146 struct work_struct *work)
1148 struct xfs_buf *bp =
1149 container_of(work, xfs_buf_t, b_ioend_work);
1155 xfs_buf_ioend_async(
1158 INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work);
1159 queue_work(bp->b_ioend_wq, &bp->b_ioend_work);
1167 ASSERT(error <= 0 && error >= -1000);
1168 bp->b_error = error;
1169 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
1173 xfs_buf_ioerror_alert(
1177 xfs_alert(bp->b_target->bt_mount,
1178 "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d",
1179 (__uint64_t)XFS_BUF_ADDR(bp), func, -bp->b_error, bp->b_length);
1188 ASSERT(xfs_buf_islocked(bp));
1190 bp->b_flags |= XBF_WRITE;
1191 bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q |
1192 XBF_WRITE_FAIL | XBF_DONE);
1194 error = xfs_buf_submit_wait(bp);
1196 xfs_force_shutdown(bp->b_target->bt_mount,
1197 SHUTDOWN_META_IO_ERROR);
1206 struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private;
1209 * don't overwrite existing errors - otherwise we can lose errors on
1210 * buffers that require multiple bios to complete.
1213 cmpxchg(&bp->b_io_error, 0, bio->bi_error);
1215 if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1216 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1218 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1219 xfs_buf_ioend_async(bp);
1224 xfs_buf_ioapply_map(
1233 int total_nr_pages = bp->b_page_count;
1236 sector_t sector = bp->b_maps[map].bm_bn;
1240 total_nr_pages = bp->b_page_count;
1242 /* skip the pages in the buffer before the start offset */
1244 offset = *buf_offset;
1245 while (offset >= PAGE_SIZE) {
1247 offset -= PAGE_SIZE;
1251 * Limit the IO size to the length of the current vector, and update the
1252 * remaining IO count for the next time around.
1254 size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count);
1256 *buf_offset += size;
1259 atomic_inc(&bp->b_io_remaining);
1260 nr_pages = min(total_nr_pages, BIO_MAX_PAGES);
1262 bio = bio_alloc(GFP_NOIO, nr_pages);
1263 bio->bi_bdev = bp->b_target->bt_bdev;
1264 bio->bi_iter.bi_sector = sector;
1265 bio->bi_end_io = xfs_buf_bio_end_io;
1266 bio->bi_private = bp;
1267 bio_set_op_attrs(bio, op, op_flags);
1269 for (; size && nr_pages; nr_pages--, page_index++) {
1270 int rbytes, nbytes = PAGE_SIZE - offset;
1275 rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes,
1277 if (rbytes < nbytes)
1281 sector += BTOBB(nbytes);
1286 if (likely(bio->bi_iter.bi_size)) {
1287 if (xfs_buf_is_vmapped(bp)) {
1288 flush_kernel_vmap_range(bp->b_addr,
1289 xfs_buf_vmap_len(bp));
1296 * This is guaranteed not to be the last io reference count
1297 * because the caller (xfs_buf_submit) holds a count itself.
1299 atomic_dec(&bp->b_io_remaining);
1300 xfs_buf_ioerror(bp, -EIO);
1310 struct blk_plug plug;
1318 * Make sure we capture only current IO errors rather than stale errors
1319 * left over from previous use of the buffer (e.g. failed readahead).
1324 * Initialize the I/O completion workqueue if we haven't yet or the
1325 * submitter has not opted to specify a custom one.
1327 if (!bp->b_ioend_wq)
1328 bp->b_ioend_wq = bp->b_target->bt_mount->m_buf_workqueue;
1330 if (bp->b_flags & XBF_WRITE) {
1332 if (bp->b_flags & XBF_SYNCIO)
1333 op_flags = REQ_SYNC;
1334 if (bp->b_flags & XBF_FUA)
1335 op_flags |= REQ_FUA;
1336 if (bp->b_flags & XBF_FLUSH)
1337 op_flags |= REQ_PREFLUSH;
1340 * Run the write verifier callback function if it exists. If
1341 * this function fails it will mark the buffer with an error and
1342 * the IO should not be dispatched.
1345 bp->b_ops->verify_write(bp);
1347 xfs_force_shutdown(bp->b_target->bt_mount,
1348 SHUTDOWN_CORRUPT_INCORE);
1351 } else if (bp->b_bn != XFS_BUF_DADDR_NULL) {
1352 struct xfs_mount *mp = bp->b_target->bt_mount;
1355 * non-crc filesystems don't attach verifiers during
1356 * log recovery, so don't warn for such filesystems.
1358 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1360 "%s: no ops on block 0x%llx/0x%x",
1361 __func__, bp->b_bn, bp->b_length);
1362 xfs_hex_dump(bp->b_addr, 64);
1366 } else if (bp->b_flags & XBF_READ_AHEAD) {
1368 op_flags = REQ_RAHEAD;
1373 /* we only use the buffer cache for meta-data */
1374 op_flags |= REQ_META;
1377 * Walk all the vectors issuing IO on them. Set up the initial offset
1378 * into the buffer and the desired IO size before we start -
1379 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1382 offset = bp->b_offset;
1383 size = BBTOB(bp->b_io_length);
1384 blk_start_plug(&plug);
1385 for (i = 0; i < bp->b_map_count; i++) {
1386 xfs_buf_ioapply_map(bp, i, &offset, &size, op, op_flags);
1390 break; /* all done */
1392 blk_finish_plug(&plug);
1396 * Asynchronous IO submission path. This transfers the buffer lock ownership and
1397 * the current reference to the IO. It is not safe to reference the buffer after
1398 * a call to this function unless the caller holds an additional reference
1405 trace_xfs_buf_submit(bp, _RET_IP_);
1407 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1408 ASSERT(bp->b_flags & XBF_ASYNC);
1410 /* on shutdown we stale and complete the buffer immediately */
1411 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1412 xfs_buf_ioerror(bp, -EIO);
1413 bp->b_flags &= ~XBF_DONE;
1419 if (bp->b_flags & XBF_WRITE)
1420 xfs_buf_wait_unpin(bp);
1422 /* clear the internal error state to avoid spurious errors */
1426 * The caller's reference is released during I/O completion.
1427 * This occurs some time after the last b_io_remaining reference is
1428 * released, so after we drop our Io reference we have to have some
1429 * other reference to ensure the buffer doesn't go away from underneath
1430 * us. Take a direct reference to ensure we have safe access to the
1431 * buffer until we are finished with it.
1436 * Set the count to 1 initially, this will stop an I/O completion
1437 * callout which happens before we have started all the I/O from calling
1438 * xfs_buf_ioend too early.
1440 atomic_set(&bp->b_io_remaining, 1);
1441 xfs_buf_ioacct_inc(bp);
1442 _xfs_buf_ioapply(bp);
1445 * If _xfs_buf_ioapply failed, we can get back here with only the IO
1446 * reference we took above. If we drop it to zero, run completion so
1447 * that we don't return to the caller with completion still pending.
1449 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1453 xfs_buf_ioend_async(bp);
1457 /* Note: it is not safe to reference bp now we've dropped our ref */
1461 * Synchronous buffer IO submission path, read or write.
1464 xfs_buf_submit_wait(
1469 trace_xfs_buf_submit_wait(bp, _RET_IP_);
1471 ASSERT(!(bp->b_flags & (_XBF_DELWRI_Q | XBF_ASYNC)));
1473 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1474 xfs_buf_ioerror(bp, -EIO);
1476 bp->b_flags &= ~XBF_DONE;
1480 if (bp->b_flags & XBF_WRITE)
1481 xfs_buf_wait_unpin(bp);
1483 /* clear the internal error state to avoid spurious errors */
1487 * For synchronous IO, the IO does not inherit the submitters reference
1488 * count, nor the buffer lock. Hence we cannot release the reference we
1489 * are about to take until we've waited for all IO completion to occur,
1490 * including any xfs_buf_ioend_async() work that may be pending.
1495 * Set the count to 1 initially, this will stop an I/O completion
1496 * callout which happens before we have started all the I/O from calling
1497 * xfs_buf_ioend too early.
1499 atomic_set(&bp->b_io_remaining, 1);
1500 _xfs_buf_ioapply(bp);
1503 * make sure we run completion synchronously if it raced with us and is
1506 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1509 /* wait for completion before gathering the error from the buffer */
1510 trace_xfs_buf_iowait(bp, _RET_IP_);
1511 wait_for_completion(&bp->b_iowait);
1512 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1513 error = bp->b_error;
1516 * all done now, we can release the hold that keeps the buffer
1517 * referenced for the entire IO.
1531 return bp->b_addr + offset;
1533 offset += bp->b_offset;
1534 page = bp->b_pages[offset >> PAGE_SHIFT];
1535 return page_address(page) + (offset & (PAGE_SIZE-1));
1539 * Move data into or out of a buffer.
1543 xfs_buf_t *bp, /* buffer to process */
1544 size_t boff, /* starting buffer offset */
1545 size_t bsize, /* length to copy */
1546 void *data, /* data address */
1547 xfs_buf_rw_t mode) /* read/write/zero flag */
1551 bend = boff + bsize;
1552 while (boff < bend) {
1554 int page_index, page_offset, csize;
1556 page_index = (boff + bp->b_offset) >> PAGE_SHIFT;
1557 page_offset = (boff + bp->b_offset) & ~PAGE_MASK;
1558 page = bp->b_pages[page_index];
1559 csize = min_t(size_t, PAGE_SIZE - page_offset,
1560 BBTOB(bp->b_io_length) - boff);
1562 ASSERT((csize + page_offset) <= PAGE_SIZE);
1566 memset(page_address(page) + page_offset, 0, csize);
1569 memcpy(data, page_address(page) + page_offset, csize);
1572 memcpy(page_address(page) + page_offset, data, csize);
1581 * Handling of buffer targets (buftargs).
1585 * Wait for any bufs with callbacks that have been submitted but have not yet
1586 * returned. These buffers will have an elevated hold count, so wait on those
1587 * while freeing all the buffers only held by the LRU.
1589 static enum lru_status
1590 xfs_buftarg_wait_rele(
1591 struct list_head *item,
1592 struct list_lru_one *lru,
1593 spinlock_t *lru_lock,
1597 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1598 struct list_head *dispose = arg;
1600 if (atomic_read(&bp->b_hold) > 1) {
1601 /* need to wait, so skip it this pass */
1602 trace_xfs_buf_wait_buftarg(bp, _RET_IP_);
1605 if (!spin_trylock(&bp->b_lock))
1609 * clear the LRU reference count so the buffer doesn't get
1610 * ignored in xfs_buf_rele().
1612 atomic_set(&bp->b_lru_ref, 0);
1613 bp->b_state |= XFS_BSTATE_DISPOSE;
1614 list_lru_isolate_move(lru, item, dispose);
1615 spin_unlock(&bp->b_lock);
1621 struct xfs_buftarg *btp)
1627 * First wait on the buftarg I/O count for all in-flight buffers to be
1628 * released. This is critical as new buffers do not make the LRU until
1629 * they are released.
1631 * Next, flush the buffer workqueue to ensure all completion processing
1632 * has finished. Just waiting on buffer locks is not sufficient for
1633 * async IO as the reference count held over IO is not released until
1634 * after the buffer lock is dropped. Hence we need to ensure here that
1635 * all reference counts have been dropped before we start walking the
1638 while (percpu_counter_sum(&btp->bt_io_count))
1640 flush_workqueue(btp->bt_mount->m_buf_workqueue);
1642 /* loop until there is nothing left on the lru list. */
1643 while (list_lru_count(&btp->bt_lru)) {
1644 list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele,
1645 &dispose, LONG_MAX);
1647 while (!list_empty(&dispose)) {
1649 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1650 list_del_init(&bp->b_lru);
1651 if (bp->b_flags & XBF_WRITE_FAIL) {
1652 xfs_alert(btp->bt_mount,
1653 "Corruption Alert: Buffer at block 0x%llx had permanent write failures!",
1654 (long long)bp->b_bn);
1655 xfs_alert(btp->bt_mount,
1656 "Please run xfs_repair to determine the extent of the problem.");
1665 static enum lru_status
1666 xfs_buftarg_isolate(
1667 struct list_head *item,
1668 struct list_lru_one *lru,
1669 spinlock_t *lru_lock,
1672 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1673 struct list_head *dispose = arg;
1676 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1677 * If we fail to get the lock, just skip it.
1679 if (!spin_trylock(&bp->b_lock))
1682 * Decrement the b_lru_ref count unless the value is already
1683 * zero. If the value is already zero, we need to reclaim the
1684 * buffer, otherwise it gets another trip through the LRU.
1686 if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1687 spin_unlock(&bp->b_lock);
1691 bp->b_state |= XFS_BSTATE_DISPOSE;
1692 list_lru_isolate_move(lru, item, dispose);
1693 spin_unlock(&bp->b_lock);
1697 static unsigned long
1698 xfs_buftarg_shrink_scan(
1699 struct shrinker *shrink,
1700 struct shrink_control *sc)
1702 struct xfs_buftarg *btp = container_of(shrink,
1703 struct xfs_buftarg, bt_shrinker);
1705 unsigned long freed;
1707 freed = list_lru_shrink_walk(&btp->bt_lru, sc,
1708 xfs_buftarg_isolate, &dispose);
1710 while (!list_empty(&dispose)) {
1712 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1713 list_del_init(&bp->b_lru);
1720 static unsigned long
1721 xfs_buftarg_shrink_count(
1722 struct shrinker *shrink,
1723 struct shrink_control *sc)
1725 struct xfs_buftarg *btp = container_of(shrink,
1726 struct xfs_buftarg, bt_shrinker);
1727 return list_lru_shrink_count(&btp->bt_lru, sc);
1732 struct xfs_mount *mp,
1733 struct xfs_buftarg *btp)
1735 unregister_shrinker(&btp->bt_shrinker);
1736 ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0);
1737 percpu_counter_destroy(&btp->bt_io_count);
1738 list_lru_destroy(&btp->bt_lru);
1740 xfs_blkdev_issue_flush(btp);
1746 xfs_setsize_buftarg(
1748 unsigned int sectorsize)
1750 /* Set up metadata sector size info */
1751 btp->bt_meta_sectorsize = sectorsize;
1752 btp->bt_meta_sectormask = sectorsize - 1;
1754 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1755 xfs_warn(btp->bt_mount,
1756 "Cannot set_blocksize to %u on device %pg",
1757 sectorsize, btp->bt_bdev);
1761 /* Set up device logical sector size mask */
1762 btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev);
1763 btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1;
1769 * When allocating the initial buffer target we have not yet
1770 * read in the superblock, so don't know what sized sectors
1771 * are being used at this early stage. Play safe.
1774 xfs_setsize_buftarg_early(
1776 struct block_device *bdev)
1778 return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev));
1783 struct xfs_mount *mp,
1784 struct block_device *bdev)
1788 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS);
1791 btp->bt_dev = bdev->bd_dev;
1792 btp->bt_bdev = bdev;
1793 btp->bt_bdi = blk_get_backing_dev_info(bdev);
1795 if (xfs_setsize_buftarg_early(btp, bdev))
1798 if (list_lru_init(&btp->bt_lru))
1801 if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL))
1804 btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count;
1805 btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan;
1806 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1807 btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE;
1808 register_shrinker(&btp->bt_shrinker);
1817 * Add a buffer to the delayed write list.
1819 * This queues a buffer for writeout if it hasn't already been. Note that
1820 * neither this routine nor the buffer list submission functions perform
1821 * any internal synchronization. It is expected that the lists are thread-local
1824 * Returns true if we queued up the buffer, or false if it already had
1825 * been on the buffer list.
1828 xfs_buf_delwri_queue(
1830 struct list_head *list)
1832 ASSERT(xfs_buf_islocked(bp));
1833 ASSERT(!(bp->b_flags & XBF_READ));
1836 * If the buffer is already marked delwri it already is queued up
1837 * by someone else for imediate writeout. Just ignore it in that
1840 if (bp->b_flags & _XBF_DELWRI_Q) {
1841 trace_xfs_buf_delwri_queued(bp, _RET_IP_);
1845 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1848 * If a buffer gets written out synchronously or marked stale while it
1849 * is on a delwri list we lazily remove it. To do this, the other party
1850 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1851 * It remains referenced and on the list. In a rare corner case it
1852 * might get readded to a delwri list after the synchronous writeout, in
1853 * which case we need just need to re-add the flag here.
1855 bp->b_flags |= _XBF_DELWRI_Q;
1856 if (list_empty(&bp->b_list)) {
1857 atomic_inc(&bp->b_hold);
1858 list_add_tail(&bp->b_list, list);
1865 * Compare function is more complex than it needs to be because
1866 * the return value is only 32 bits and we are doing comparisons
1872 struct list_head *a,
1873 struct list_head *b)
1875 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1876 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1879 diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn;
1888 * submit buffers for write.
1890 * When we have a large buffer list, we do not want to hold all the buffers
1891 * locked while we block on the request queue waiting for IO dispatch. To avoid
1892 * this problem, we lock and submit buffers in groups of 50, thereby minimising
1893 * the lock hold times for lists which may contain thousands of objects.
1895 * To do this, we sort the buffer list before we walk the list to lock and
1896 * submit buffers, and we plug and unplug around each group of buffers we
1900 xfs_buf_delwri_submit_buffers(
1901 struct list_head *buffer_list,
1902 struct list_head *wait_list)
1904 struct xfs_buf *bp, *n;
1905 LIST_HEAD (submit_list);
1907 struct blk_plug plug;
1909 list_sort(NULL, buffer_list, xfs_buf_cmp);
1911 blk_start_plug(&plug);
1912 list_for_each_entry_safe(bp, n, buffer_list, b_list) {
1914 if (xfs_buf_ispinned(bp)) {
1918 if (!xfs_buf_trylock(bp))
1925 * Someone else might have written the buffer synchronously or
1926 * marked it stale in the meantime. In that case only the
1927 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
1928 * reference and remove it from the list here.
1930 if (!(bp->b_flags & _XBF_DELWRI_Q)) {
1931 list_del_init(&bp->b_list);
1936 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1939 * We do all IO submission async. This means if we need
1940 * to wait for IO completion we need to take an extra
1941 * reference so the buffer is still valid on the other
1942 * side. We need to move the buffer onto the io_list
1943 * at this point so the caller can still access it.
1945 bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_WRITE_FAIL);
1946 bp->b_flags |= XBF_WRITE | XBF_ASYNC;
1949 list_move_tail(&bp->b_list, wait_list);
1951 list_del_init(&bp->b_list);
1955 blk_finish_plug(&plug);
1961 * Write out a buffer list asynchronously.
1963 * This will take the @buffer_list, write all non-locked and non-pinned buffers
1964 * out and not wait for I/O completion on any of the buffers. This interface
1965 * is only safely useable for callers that can track I/O completion by higher
1966 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
1970 xfs_buf_delwri_submit_nowait(
1971 struct list_head *buffer_list)
1973 return xfs_buf_delwri_submit_buffers(buffer_list, NULL);
1977 * Write out a buffer list synchronously.
1979 * This will take the @buffer_list, write all buffers out and wait for I/O
1980 * completion on all of the buffers. @buffer_list is consumed by the function,
1981 * so callers must have some other way of tracking buffers if they require such
1985 xfs_buf_delwri_submit(
1986 struct list_head *buffer_list)
1988 LIST_HEAD (wait_list);
1989 int error = 0, error2;
1992 xfs_buf_delwri_submit_buffers(buffer_list, &wait_list);
1994 /* Wait for IO to complete. */
1995 while (!list_empty(&wait_list)) {
1996 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
1998 list_del_init(&bp->b_list);
2000 /* locking the buffer will wait for async IO completion. */
2002 error2 = bp->b_error;
2014 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
2015 KM_ZONE_HWALIGN, NULL);
2026 xfs_buf_terminate(void)
2028 kmem_zone_destroy(xfs_buf_zone);