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]);
425 bp->b_flags &= ~_XBF_PAGES;
430 * Map buffer into kernel address-space if necessary.
437 ASSERT(bp->b_flags & _XBF_PAGES);
438 if (bp->b_page_count == 1) {
439 /* A single page buffer is always mappable */
440 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
441 } else if (flags & XBF_UNMAPPED) {
448 * vm_map_ram() will allocate auxillary structures (e.g.
449 * pagetables) with GFP_KERNEL, yet we are likely to be under
450 * GFP_NOFS context here. Hence we need to tell memory reclaim
451 * that we are in such a context via PF_MEMALLOC_NOIO to prevent
452 * memory reclaim re-entering the filesystem here and
453 * potentially deadlocking.
455 noio_flag = memalloc_noio_save();
457 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
462 } while (retried++ <= 1);
463 memalloc_noio_restore(noio_flag);
467 bp->b_addr += bp->b_offset;
474 * Finding and Reading Buffers
478 struct rhashtable_compare_arg *arg,
481 const struct xfs_buf_map *map = arg->key;
482 const struct xfs_buf *bp = obj;
485 * The key hashing in the lookup path depends on the key being the
486 * first element of the compare_arg, make sure to assert this.
488 BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0);
490 if (bp->b_bn != map->bm_bn)
493 if (unlikely(bp->b_length != map->bm_len)) {
495 * found a block number match. If the range doesn't
496 * match, the only way this is allowed is if the buffer
497 * in the cache is stale and the transaction that made
498 * it stale has not yet committed. i.e. we are
499 * reallocating a busy extent. Skip this buffer and
500 * continue searching for an exact match.
502 ASSERT(bp->b_flags & XBF_STALE);
508 static const struct rhashtable_params xfs_buf_hash_params = {
509 .min_size = 32, /* empty AGs have minimal footprint */
511 .key_len = sizeof(xfs_daddr_t),
512 .key_offset = offsetof(struct xfs_buf, b_bn),
513 .head_offset = offsetof(struct xfs_buf, b_rhash_head),
514 .automatic_shrinking = true,
515 .obj_cmpfn = _xfs_buf_obj_cmp,
520 struct xfs_perag *pag)
522 spin_lock_init(&pag->pag_buf_lock);
523 return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params);
527 xfs_buf_hash_destroy(
528 struct xfs_perag *pag)
530 rhashtable_destroy(&pag->pag_buf_hash);
534 * Look up, and creates if absent, a lockable buffer for
535 * a given range of an inode. The buffer is returned
536 * locked. No I/O is implied by this call.
540 struct xfs_buftarg *btp,
541 struct xfs_buf_map *map,
543 xfs_buf_flags_t flags,
546 struct xfs_perag *pag;
548 struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn };
552 for (i = 0; i < nmaps; i++)
553 cmap.bm_len += map[i].bm_len;
555 /* Check for IOs smaller than the sector size / not sector aligned */
556 ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize));
557 ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask));
560 * Corrupted block numbers can get through to here, unfortunately, so we
561 * have to check that the buffer falls within the filesystem bounds.
563 eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks);
564 if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) {
566 * XXX (dgc): we should really be returning -EFSCORRUPTED here,
567 * but none of the higher level infrastructure supports
568 * returning a specific error on buffer lookup failures.
570 xfs_alert(btp->bt_mount,
571 "%s: Block out of range: block 0x%llx, EOFS 0x%llx ",
572 __func__, cmap.bm_bn, eofs);
577 pag = xfs_perag_get(btp->bt_mount,
578 xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn));
580 spin_lock(&pag->pag_buf_lock);
581 bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap,
582 xfs_buf_hash_params);
584 atomic_inc(&bp->b_hold);
590 /* the buffer keeps the perag reference until it is freed */
592 rhashtable_insert_fast(&pag->pag_buf_hash,
593 &new_bp->b_rhash_head,
594 xfs_buf_hash_params);
595 spin_unlock(&pag->pag_buf_lock);
597 XFS_STATS_INC(btp->bt_mount, xb_miss_locked);
598 spin_unlock(&pag->pag_buf_lock);
604 spin_unlock(&pag->pag_buf_lock);
607 if (!xfs_buf_trylock(bp)) {
608 if (flags & XBF_TRYLOCK) {
610 XFS_STATS_INC(btp->bt_mount, xb_busy_locked);
614 XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited);
618 * if the buffer is stale, clear all the external state associated with
619 * it. We need to keep flags such as how we allocated the buffer memory
622 if (bp->b_flags & XBF_STALE) {
623 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
624 ASSERT(bp->b_iodone == NULL);
625 bp->b_flags &= _XBF_KMEM | _XBF_PAGES;
629 trace_xfs_buf_find(bp, flags, _RET_IP_);
630 XFS_STATS_INC(btp->bt_mount, xb_get_locked);
635 * Assembles a buffer covering the specified range. The code is optimised for
636 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
637 * more hits than misses.
641 struct xfs_buftarg *target,
642 struct xfs_buf_map *map,
644 xfs_buf_flags_t flags)
647 struct xfs_buf *new_bp;
650 bp = _xfs_buf_find(target, map, nmaps, flags, NULL);
654 new_bp = _xfs_buf_alloc(target, map, nmaps, flags);
655 if (unlikely(!new_bp))
658 error = xfs_buf_allocate_memory(new_bp, flags);
660 xfs_buf_free(new_bp);
664 bp = _xfs_buf_find(target, map, nmaps, flags, new_bp);
666 xfs_buf_free(new_bp);
671 xfs_buf_free(new_bp);
675 error = _xfs_buf_map_pages(bp, flags);
676 if (unlikely(error)) {
677 xfs_warn(target->bt_mount,
678 "%s: failed to map pagesn", __func__);
685 * Clear b_error if this is a lookup from a caller that doesn't expect
686 * valid data to be found in the buffer.
688 if (!(flags & XBF_READ))
689 xfs_buf_ioerror(bp, 0);
691 XFS_STATS_INC(target->bt_mount, xb_get);
692 trace_xfs_buf_get(bp, flags, _RET_IP_);
699 xfs_buf_flags_t flags)
701 ASSERT(!(flags & XBF_WRITE));
702 ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL);
704 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD);
705 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
707 if (flags & XBF_ASYNC) {
711 return xfs_buf_submit_wait(bp);
716 struct xfs_buftarg *target,
717 struct xfs_buf_map *map,
719 xfs_buf_flags_t flags,
720 const struct xfs_buf_ops *ops)
726 bp = xfs_buf_get_map(target, map, nmaps, flags);
728 trace_xfs_buf_read(bp, flags, _RET_IP_);
730 if (!(bp->b_flags & XBF_DONE)) {
731 XFS_STATS_INC(target->bt_mount, xb_get_read);
733 _xfs_buf_read(bp, flags);
734 } else if (flags & XBF_ASYNC) {
736 * Read ahead call which is already satisfied,
742 /* We do not want read in the flags */
743 bp->b_flags &= ~XBF_READ;
751 * If we are not low on memory then do the readahead in a deadlock
755 xfs_buf_readahead_map(
756 struct xfs_buftarg *target,
757 struct xfs_buf_map *map,
759 const struct xfs_buf_ops *ops)
761 if (bdi_read_congested(target->bt_bdi))
764 xfs_buf_read_map(target, map, nmaps,
765 XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops);
769 * Read an uncached buffer from disk. Allocates and returns a locked
770 * buffer containing the disk contents or nothing.
773 xfs_buf_read_uncached(
774 struct xfs_buftarg *target,
778 struct xfs_buf **bpp,
779 const struct xfs_buf_ops *ops)
785 bp = xfs_buf_get_uncached(target, numblks, flags);
789 /* set up the buffer for a read IO */
790 ASSERT(bp->b_map_count == 1);
791 bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */
792 bp->b_maps[0].bm_bn = daddr;
793 bp->b_flags |= XBF_READ;
796 xfs_buf_submit_wait(bp);
798 int error = bp->b_error;
808 * Return a buffer allocated as an empty buffer and associated to external
809 * memory via xfs_buf_associate_memory() back to it's empty state.
817 _xfs_buf_free_pages(bp);
820 bp->b_page_count = 0;
822 bp->b_length = numblks;
823 bp->b_io_length = numblks;
825 ASSERT(bp->b_map_count == 1);
826 bp->b_bn = XFS_BUF_DADDR_NULL;
827 bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL;
828 bp->b_maps[0].bm_len = bp->b_length;
831 static inline struct page *
835 if ((!is_vmalloc_addr(addr))) {
836 return virt_to_page(addr);
838 return vmalloc_to_page(addr);
843 xfs_buf_associate_memory(
850 unsigned long pageaddr;
851 unsigned long offset;
855 pageaddr = (unsigned long)mem & PAGE_MASK;
856 offset = (unsigned long)mem - pageaddr;
857 buflen = PAGE_ALIGN(len + offset);
858 page_count = buflen >> PAGE_SHIFT;
860 /* Free any previous set of page pointers */
862 _xfs_buf_free_pages(bp);
867 rval = _xfs_buf_get_pages(bp, page_count);
871 bp->b_offset = offset;
873 for (i = 0; i < bp->b_page_count; i++) {
874 bp->b_pages[i] = mem_to_page((void *)pageaddr);
875 pageaddr += PAGE_SIZE;
878 bp->b_io_length = BTOBB(len);
879 bp->b_length = BTOBB(buflen);
885 xfs_buf_get_uncached(
886 struct xfs_buftarg *target,
890 unsigned long page_count;
893 DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks);
895 /* flags might contain irrelevant bits, pass only what we care about */
896 bp = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT);
897 if (unlikely(bp == NULL))
900 page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT;
901 error = _xfs_buf_get_pages(bp, page_count);
905 for (i = 0; i < page_count; i++) {
906 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
910 bp->b_flags |= _XBF_PAGES;
912 error = _xfs_buf_map_pages(bp, 0);
913 if (unlikely(error)) {
914 xfs_warn(target->bt_mount,
915 "%s: failed to map pages", __func__);
919 trace_xfs_buf_get_uncached(bp, _RET_IP_);
924 __free_page(bp->b_pages[i]);
925 _xfs_buf_free_pages(bp);
927 xfs_buf_free_maps(bp);
928 kmem_zone_free(xfs_buf_zone, bp);
934 * Increment reference count on buffer, to hold the buffer concurrently
935 * with another thread which may release (free) the buffer asynchronously.
936 * Must hold the buffer already to call this function.
942 trace_xfs_buf_hold(bp, _RET_IP_);
943 atomic_inc(&bp->b_hold);
947 * Release a hold on the specified buffer. If the hold count is 1, the buffer is
948 * placed on LRU or freed (depending on b_lru_ref).
954 struct xfs_perag *pag = bp->b_pag;
956 bool freebuf = false;
958 trace_xfs_buf_rele(bp, _RET_IP_);
961 ASSERT(list_empty(&bp->b_lru));
962 if (atomic_dec_and_test(&bp->b_hold)) {
963 xfs_buf_ioacct_dec(bp);
969 ASSERT(atomic_read(&bp->b_hold) > 0);
971 release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock);
972 spin_lock(&bp->b_lock);
975 * Drop the in-flight state if the buffer is already on the LRU
976 * and it holds the only reference. This is racy because we
977 * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT
978 * ensures the decrement occurs only once per-buf.
980 if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru))
981 xfs_buf_ioacct_dec(bp);
985 /* the last reference has been dropped ... */
986 xfs_buf_ioacct_dec(bp);
987 if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) {
989 * If the buffer is added to the LRU take a new reference to the
990 * buffer for the LRU and clear the (now stale) dispose list
993 if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) {
994 bp->b_state &= ~XFS_BSTATE_DISPOSE;
995 atomic_inc(&bp->b_hold);
997 spin_unlock(&pag->pag_buf_lock);
1000 * most of the time buffers will already be removed from the
1001 * LRU, so optimise that case by checking for the
1002 * XFS_BSTATE_DISPOSE flag indicating the last list the buffer
1003 * was on was the disposal list
1005 if (!(bp->b_state & XFS_BSTATE_DISPOSE)) {
1006 list_lru_del(&bp->b_target->bt_lru, &bp->b_lru);
1008 ASSERT(list_empty(&bp->b_lru));
1011 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1012 rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head,
1013 xfs_buf_hash_params);
1014 spin_unlock(&pag->pag_buf_lock);
1020 spin_unlock(&bp->b_lock);
1028 * Lock a buffer object, if it is not already locked.
1030 * If we come across a stale, pinned, locked buffer, we know that we are
1031 * being asked to lock a buffer that has been reallocated. Because it is
1032 * pinned, we know that the log has not been pushed to disk and hence it
1033 * will still be locked. Rather than continuing to have trylock attempts
1034 * fail until someone else pushes the log, push it ourselves before
1035 * returning. This means that the xfsaild will not get stuck trying
1036 * to push on stale inode buffers.
1044 locked = down_trylock(&bp->b_sema) == 0;
1047 trace_xfs_buf_trylock(bp, _RET_IP_);
1049 trace_xfs_buf_trylock_fail(bp, _RET_IP_);
1055 * Lock a buffer object.
1057 * If we come across a stale, pinned, locked buffer, we know that we
1058 * are being asked to lock a buffer that has been reallocated. Because
1059 * it is pinned, we know that the log has not been pushed to disk and
1060 * hence it will still be locked. Rather than sleeping until someone
1061 * else pushes the log, push it ourselves before trying to get the lock.
1067 trace_xfs_buf_lock(bp, _RET_IP_);
1069 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
1070 xfs_log_force(bp->b_target->bt_mount, 0);
1074 trace_xfs_buf_lock_done(bp, _RET_IP_);
1084 trace_xfs_buf_unlock(bp, _RET_IP_);
1091 DECLARE_WAITQUEUE (wait, current);
1093 if (atomic_read(&bp->b_pin_count) == 0)
1096 add_wait_queue(&bp->b_waiters, &wait);
1098 set_current_state(TASK_UNINTERRUPTIBLE);
1099 if (atomic_read(&bp->b_pin_count) == 0)
1103 remove_wait_queue(&bp->b_waiters, &wait);
1104 set_current_state(TASK_RUNNING);
1108 * Buffer Utility Routines
1115 bool read = bp->b_flags & XBF_READ;
1117 trace_xfs_buf_iodone(bp, _RET_IP_);
1119 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1122 * Pull in IO completion errors now. We are guaranteed to be running
1123 * single threaded, so we don't need the lock to read b_io_error.
1125 if (!bp->b_error && bp->b_io_error)
1126 xfs_buf_ioerror(bp, bp->b_io_error);
1128 /* Only validate buffers that were read without errors */
1129 if (read && !bp->b_error && bp->b_ops) {
1130 ASSERT(!bp->b_iodone);
1131 bp->b_ops->verify_read(bp);
1135 bp->b_flags |= XBF_DONE;
1138 (*(bp->b_iodone))(bp);
1139 else if (bp->b_flags & XBF_ASYNC)
1142 complete(&bp->b_iowait);
1147 struct work_struct *work)
1149 struct xfs_buf *bp =
1150 container_of(work, xfs_buf_t, b_ioend_work);
1156 xfs_buf_ioend_async(
1159 INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work);
1160 queue_work(bp->b_ioend_wq, &bp->b_ioend_work);
1168 ASSERT(error <= 0 && error >= -1000);
1169 bp->b_error = error;
1170 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
1174 xfs_buf_ioerror_alert(
1178 xfs_alert(bp->b_target->bt_mount,
1179 "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d",
1180 (__uint64_t)XFS_BUF_ADDR(bp), func, -bp->b_error, bp->b_length);
1189 ASSERT(xfs_buf_islocked(bp));
1191 bp->b_flags |= XBF_WRITE;
1192 bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q |
1193 XBF_WRITE_FAIL | XBF_DONE);
1195 error = xfs_buf_submit_wait(bp);
1197 xfs_force_shutdown(bp->b_target->bt_mount,
1198 SHUTDOWN_META_IO_ERROR);
1207 struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private;
1210 * don't overwrite existing errors - otherwise we can lose errors on
1211 * buffers that require multiple bios to complete.
1214 cmpxchg(&bp->b_io_error, 0, bio->bi_error);
1216 if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1217 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1219 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1220 xfs_buf_ioend_async(bp);
1225 xfs_buf_ioapply_map(
1234 int total_nr_pages = bp->b_page_count;
1237 sector_t sector = bp->b_maps[map].bm_bn;
1241 total_nr_pages = bp->b_page_count;
1243 /* skip the pages in the buffer before the start offset */
1245 offset = *buf_offset;
1246 while (offset >= PAGE_SIZE) {
1248 offset -= PAGE_SIZE;
1252 * Limit the IO size to the length of the current vector, and update the
1253 * remaining IO count for the next time around.
1255 size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count);
1257 *buf_offset += size;
1260 atomic_inc(&bp->b_io_remaining);
1261 nr_pages = min(total_nr_pages, BIO_MAX_PAGES);
1263 bio = bio_alloc(GFP_NOIO, nr_pages);
1264 bio->bi_bdev = bp->b_target->bt_bdev;
1265 bio->bi_iter.bi_sector = sector;
1266 bio->bi_end_io = xfs_buf_bio_end_io;
1267 bio->bi_private = bp;
1268 bio_set_op_attrs(bio, op, op_flags);
1270 for (; size && nr_pages; nr_pages--, page_index++) {
1271 int rbytes, nbytes = PAGE_SIZE - offset;
1276 rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes,
1278 if (rbytes < nbytes)
1282 sector += BTOBB(nbytes);
1287 if (likely(bio->bi_iter.bi_size)) {
1288 if (xfs_buf_is_vmapped(bp)) {
1289 flush_kernel_vmap_range(bp->b_addr,
1290 xfs_buf_vmap_len(bp));
1297 * This is guaranteed not to be the last io reference count
1298 * because the caller (xfs_buf_submit) holds a count itself.
1300 atomic_dec(&bp->b_io_remaining);
1301 xfs_buf_ioerror(bp, -EIO);
1311 struct blk_plug plug;
1319 * Make sure we capture only current IO errors rather than stale errors
1320 * left over from previous use of the buffer (e.g. failed readahead).
1325 * Initialize the I/O completion workqueue if we haven't yet or the
1326 * submitter has not opted to specify a custom one.
1328 if (!bp->b_ioend_wq)
1329 bp->b_ioend_wq = bp->b_target->bt_mount->m_buf_workqueue;
1331 if (bp->b_flags & XBF_WRITE) {
1333 if (bp->b_flags & XBF_SYNCIO)
1334 op_flags = REQ_SYNC;
1335 if (bp->b_flags & XBF_FUA)
1336 op_flags |= REQ_FUA;
1337 if (bp->b_flags & XBF_FLUSH)
1338 op_flags |= REQ_PREFLUSH;
1341 * Run the write verifier callback function if it exists. If
1342 * this function fails it will mark the buffer with an error and
1343 * the IO should not be dispatched.
1346 bp->b_ops->verify_write(bp);
1348 xfs_force_shutdown(bp->b_target->bt_mount,
1349 SHUTDOWN_CORRUPT_INCORE);
1352 } else if (bp->b_bn != XFS_BUF_DADDR_NULL) {
1353 struct xfs_mount *mp = bp->b_target->bt_mount;
1356 * non-crc filesystems don't attach verifiers during
1357 * log recovery, so don't warn for such filesystems.
1359 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1361 "%s: no ops on block 0x%llx/0x%x",
1362 __func__, bp->b_bn, bp->b_length);
1363 xfs_hex_dump(bp->b_addr, 64);
1367 } else if (bp->b_flags & XBF_READ_AHEAD) {
1369 op_flags = REQ_RAHEAD;
1374 /* we only use the buffer cache for meta-data */
1375 op_flags |= REQ_META;
1378 * Walk all the vectors issuing IO on them. Set up the initial offset
1379 * into the buffer and the desired IO size before we start -
1380 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1383 offset = bp->b_offset;
1384 size = BBTOB(bp->b_io_length);
1385 blk_start_plug(&plug);
1386 for (i = 0; i < bp->b_map_count; i++) {
1387 xfs_buf_ioapply_map(bp, i, &offset, &size, op, op_flags);
1391 break; /* all done */
1393 blk_finish_plug(&plug);
1397 * Asynchronous IO submission path. This transfers the buffer lock ownership and
1398 * the current reference to the IO. It is not safe to reference the buffer after
1399 * a call to this function unless the caller holds an additional reference
1406 trace_xfs_buf_submit(bp, _RET_IP_);
1408 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1409 ASSERT(bp->b_flags & XBF_ASYNC);
1411 /* on shutdown we stale and complete the buffer immediately */
1412 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1413 xfs_buf_ioerror(bp, -EIO);
1414 bp->b_flags &= ~XBF_DONE;
1420 if (bp->b_flags & XBF_WRITE)
1421 xfs_buf_wait_unpin(bp);
1423 /* clear the internal error state to avoid spurious errors */
1427 * The caller's reference is released during I/O completion.
1428 * This occurs some time after the last b_io_remaining reference is
1429 * released, so after we drop our Io reference we have to have some
1430 * other reference to ensure the buffer doesn't go away from underneath
1431 * us. Take a direct reference to ensure we have safe access to the
1432 * buffer until we are finished with it.
1437 * Set the count to 1 initially, this will stop an I/O completion
1438 * callout which happens before we have started all the I/O from calling
1439 * xfs_buf_ioend too early.
1441 atomic_set(&bp->b_io_remaining, 1);
1442 xfs_buf_ioacct_inc(bp);
1443 _xfs_buf_ioapply(bp);
1446 * If _xfs_buf_ioapply failed, we can get back here with only the IO
1447 * reference we took above. If we drop it to zero, run completion so
1448 * that we don't return to the caller with completion still pending.
1450 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1454 xfs_buf_ioend_async(bp);
1458 /* Note: it is not safe to reference bp now we've dropped our ref */
1462 * Synchronous buffer IO submission path, read or write.
1465 xfs_buf_submit_wait(
1470 trace_xfs_buf_submit_wait(bp, _RET_IP_);
1472 ASSERT(!(bp->b_flags & (_XBF_DELWRI_Q | XBF_ASYNC)));
1474 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1475 xfs_buf_ioerror(bp, -EIO);
1477 bp->b_flags &= ~XBF_DONE;
1481 if (bp->b_flags & XBF_WRITE)
1482 xfs_buf_wait_unpin(bp);
1484 /* clear the internal error state to avoid spurious errors */
1488 * For synchronous IO, the IO does not inherit the submitters reference
1489 * count, nor the buffer lock. Hence we cannot release the reference we
1490 * are about to take until we've waited for all IO completion to occur,
1491 * including any xfs_buf_ioend_async() work that may be pending.
1496 * Set the count to 1 initially, this will stop an I/O completion
1497 * callout which happens before we have started all the I/O from calling
1498 * xfs_buf_ioend too early.
1500 atomic_set(&bp->b_io_remaining, 1);
1501 _xfs_buf_ioapply(bp);
1504 * make sure we run completion synchronously if it raced with us and is
1507 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1510 /* wait for completion before gathering the error from the buffer */
1511 trace_xfs_buf_iowait(bp, _RET_IP_);
1512 wait_for_completion(&bp->b_iowait);
1513 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1514 error = bp->b_error;
1517 * all done now, we can release the hold that keeps the buffer
1518 * referenced for the entire IO.
1532 return bp->b_addr + offset;
1534 offset += bp->b_offset;
1535 page = bp->b_pages[offset >> PAGE_SHIFT];
1536 return page_address(page) + (offset & (PAGE_SIZE-1));
1540 * Move data into or out of a buffer.
1544 xfs_buf_t *bp, /* buffer to process */
1545 size_t boff, /* starting buffer offset */
1546 size_t bsize, /* length to copy */
1547 void *data, /* data address */
1548 xfs_buf_rw_t mode) /* read/write/zero flag */
1552 bend = boff + bsize;
1553 while (boff < bend) {
1555 int page_index, page_offset, csize;
1557 page_index = (boff + bp->b_offset) >> PAGE_SHIFT;
1558 page_offset = (boff + bp->b_offset) & ~PAGE_MASK;
1559 page = bp->b_pages[page_index];
1560 csize = min_t(size_t, PAGE_SIZE - page_offset,
1561 BBTOB(bp->b_io_length) - boff);
1563 ASSERT((csize + page_offset) <= PAGE_SIZE);
1567 memset(page_address(page) + page_offset, 0, csize);
1570 memcpy(data, page_address(page) + page_offset, csize);
1573 memcpy(page_address(page) + page_offset, data, csize);
1582 * Handling of buffer targets (buftargs).
1586 * Wait for any bufs with callbacks that have been submitted but have not yet
1587 * returned. These buffers will have an elevated hold count, so wait on those
1588 * while freeing all the buffers only held by the LRU.
1590 static enum lru_status
1591 xfs_buftarg_wait_rele(
1592 struct list_head *item,
1593 struct list_lru_one *lru,
1594 spinlock_t *lru_lock,
1598 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1599 struct list_head *dispose = arg;
1601 if (atomic_read(&bp->b_hold) > 1) {
1602 /* need to wait, so skip it this pass */
1603 trace_xfs_buf_wait_buftarg(bp, _RET_IP_);
1606 if (!spin_trylock(&bp->b_lock))
1610 * clear the LRU reference count so the buffer doesn't get
1611 * ignored in xfs_buf_rele().
1613 atomic_set(&bp->b_lru_ref, 0);
1614 bp->b_state |= XFS_BSTATE_DISPOSE;
1615 list_lru_isolate_move(lru, item, dispose);
1616 spin_unlock(&bp->b_lock);
1622 struct xfs_buftarg *btp)
1628 * First wait on the buftarg I/O count for all in-flight buffers to be
1629 * released. This is critical as new buffers do not make the LRU until
1630 * they are released.
1632 * Next, flush the buffer workqueue to ensure all completion processing
1633 * has finished. Just waiting on buffer locks is not sufficient for
1634 * async IO as the reference count held over IO is not released until
1635 * after the buffer lock is dropped. Hence we need to ensure here that
1636 * all reference counts have been dropped before we start walking the
1639 while (percpu_counter_sum(&btp->bt_io_count))
1641 flush_workqueue(btp->bt_mount->m_buf_workqueue);
1643 /* loop until there is nothing left on the lru list. */
1644 while (list_lru_count(&btp->bt_lru)) {
1645 list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele,
1646 &dispose, LONG_MAX);
1648 while (!list_empty(&dispose)) {
1650 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1651 list_del_init(&bp->b_lru);
1652 if (bp->b_flags & XBF_WRITE_FAIL) {
1653 xfs_alert(btp->bt_mount,
1654 "Corruption Alert: Buffer at block 0x%llx had permanent write failures!",
1655 (long long)bp->b_bn);
1656 xfs_alert(btp->bt_mount,
1657 "Please run xfs_repair to determine the extent of the problem.");
1666 static enum lru_status
1667 xfs_buftarg_isolate(
1668 struct list_head *item,
1669 struct list_lru_one *lru,
1670 spinlock_t *lru_lock,
1673 struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru);
1674 struct list_head *dispose = arg;
1677 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1678 * If we fail to get the lock, just skip it.
1680 if (!spin_trylock(&bp->b_lock))
1683 * Decrement the b_lru_ref count unless the value is already
1684 * zero. If the value is already zero, we need to reclaim the
1685 * buffer, otherwise it gets another trip through the LRU.
1687 if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1688 spin_unlock(&bp->b_lock);
1692 bp->b_state |= XFS_BSTATE_DISPOSE;
1693 list_lru_isolate_move(lru, item, dispose);
1694 spin_unlock(&bp->b_lock);
1698 static unsigned long
1699 xfs_buftarg_shrink_scan(
1700 struct shrinker *shrink,
1701 struct shrink_control *sc)
1703 struct xfs_buftarg *btp = container_of(shrink,
1704 struct xfs_buftarg, bt_shrinker);
1706 unsigned long freed;
1708 freed = list_lru_shrink_walk(&btp->bt_lru, sc,
1709 xfs_buftarg_isolate, &dispose);
1711 while (!list_empty(&dispose)) {
1713 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1714 list_del_init(&bp->b_lru);
1721 static unsigned long
1722 xfs_buftarg_shrink_count(
1723 struct shrinker *shrink,
1724 struct shrink_control *sc)
1726 struct xfs_buftarg *btp = container_of(shrink,
1727 struct xfs_buftarg, bt_shrinker);
1728 return list_lru_shrink_count(&btp->bt_lru, sc);
1733 struct xfs_mount *mp,
1734 struct xfs_buftarg *btp)
1736 unregister_shrinker(&btp->bt_shrinker);
1737 ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0);
1738 percpu_counter_destroy(&btp->bt_io_count);
1739 list_lru_destroy(&btp->bt_lru);
1741 xfs_blkdev_issue_flush(btp);
1747 xfs_setsize_buftarg(
1749 unsigned int sectorsize)
1751 /* Set up metadata sector size info */
1752 btp->bt_meta_sectorsize = sectorsize;
1753 btp->bt_meta_sectormask = sectorsize - 1;
1755 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1756 xfs_warn(btp->bt_mount,
1757 "Cannot set_blocksize to %u on device %pg",
1758 sectorsize, btp->bt_bdev);
1762 /* Set up device logical sector size mask */
1763 btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev);
1764 btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1;
1770 * When allocating the initial buffer target we have not yet
1771 * read in the superblock, so don't know what sized sectors
1772 * are being used at this early stage. Play safe.
1775 xfs_setsize_buftarg_early(
1777 struct block_device *bdev)
1779 return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev));
1784 struct xfs_mount *mp,
1785 struct block_device *bdev)
1789 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS);
1792 btp->bt_dev = bdev->bd_dev;
1793 btp->bt_bdev = bdev;
1794 btp->bt_bdi = blk_get_backing_dev_info(bdev);
1796 if (xfs_setsize_buftarg_early(btp, bdev))
1799 if (list_lru_init(&btp->bt_lru))
1802 if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL))
1805 btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count;
1806 btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan;
1807 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1808 btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE;
1809 register_shrinker(&btp->bt_shrinker);
1818 * Add a buffer to the delayed write list.
1820 * This queues a buffer for writeout if it hasn't already been. Note that
1821 * neither this routine nor the buffer list submission functions perform
1822 * any internal synchronization. It is expected that the lists are thread-local
1825 * Returns true if we queued up the buffer, or false if it already had
1826 * been on the buffer list.
1829 xfs_buf_delwri_queue(
1831 struct list_head *list)
1833 ASSERT(xfs_buf_islocked(bp));
1834 ASSERT(!(bp->b_flags & XBF_READ));
1837 * If the buffer is already marked delwri it already is queued up
1838 * by someone else for imediate writeout. Just ignore it in that
1841 if (bp->b_flags & _XBF_DELWRI_Q) {
1842 trace_xfs_buf_delwri_queued(bp, _RET_IP_);
1846 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1849 * If a buffer gets written out synchronously or marked stale while it
1850 * is on a delwri list we lazily remove it. To do this, the other party
1851 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1852 * It remains referenced and on the list. In a rare corner case it
1853 * might get readded to a delwri list after the synchronous writeout, in
1854 * which case we need just need to re-add the flag here.
1856 bp->b_flags |= _XBF_DELWRI_Q;
1857 if (list_empty(&bp->b_list)) {
1858 atomic_inc(&bp->b_hold);
1859 list_add_tail(&bp->b_list, list);
1866 * Compare function is more complex than it needs to be because
1867 * the return value is only 32 bits and we are doing comparisons
1873 struct list_head *a,
1874 struct list_head *b)
1876 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1877 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1880 diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn;
1889 * submit buffers for write.
1891 * When we have a large buffer list, we do not want to hold all the buffers
1892 * locked while we block on the request queue waiting for IO dispatch. To avoid
1893 * this problem, we lock and submit buffers in groups of 50, thereby minimising
1894 * the lock hold times for lists which may contain thousands of objects.
1896 * To do this, we sort the buffer list before we walk the list to lock and
1897 * submit buffers, and we plug and unplug around each group of buffers we
1901 xfs_buf_delwri_submit_buffers(
1902 struct list_head *buffer_list,
1903 struct list_head *wait_list)
1905 struct xfs_buf *bp, *n;
1906 LIST_HEAD (submit_list);
1908 struct blk_plug plug;
1910 list_sort(NULL, buffer_list, xfs_buf_cmp);
1912 blk_start_plug(&plug);
1913 list_for_each_entry_safe(bp, n, buffer_list, b_list) {
1915 if (xfs_buf_ispinned(bp)) {
1919 if (!xfs_buf_trylock(bp))
1926 * Someone else might have written the buffer synchronously or
1927 * marked it stale in the meantime. In that case only the
1928 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
1929 * reference and remove it from the list here.
1931 if (!(bp->b_flags & _XBF_DELWRI_Q)) {
1932 list_del_init(&bp->b_list);
1937 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1940 * We do all IO submission async. This means if we need
1941 * to wait for IO completion we need to take an extra
1942 * reference so the buffer is still valid on the other
1943 * side. We need to move the buffer onto the io_list
1944 * at this point so the caller can still access it.
1946 bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_WRITE_FAIL);
1947 bp->b_flags |= XBF_WRITE | XBF_ASYNC;
1950 list_move_tail(&bp->b_list, wait_list);
1952 list_del_init(&bp->b_list);
1956 blk_finish_plug(&plug);
1962 * Write out a buffer list asynchronously.
1964 * This will take the @buffer_list, write all non-locked and non-pinned buffers
1965 * out and not wait for I/O completion on any of the buffers. This interface
1966 * is only safely useable for callers that can track I/O completion by higher
1967 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
1971 xfs_buf_delwri_submit_nowait(
1972 struct list_head *buffer_list)
1974 return xfs_buf_delwri_submit_buffers(buffer_list, NULL);
1978 * Write out a buffer list synchronously.
1980 * This will take the @buffer_list, write all buffers out and wait for I/O
1981 * completion on all of the buffers. @buffer_list is consumed by the function,
1982 * so callers must have some other way of tracking buffers if they require such
1986 xfs_buf_delwri_submit(
1987 struct list_head *buffer_list)
1989 LIST_HEAD (wait_list);
1990 int error = 0, error2;
1993 xfs_buf_delwri_submit_buffers(buffer_list, &wait_list);
1995 /* Wait for IO to complete. */
1996 while (!list_empty(&wait_list)) {
1997 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
1999 list_del_init(&bp->b_list);
2001 /* locking the buffer will wait for async IO completion. */
2003 error2 = bp->b_error;
2015 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
2016 KM_ZONE_HWALIGN, NULL);
2027 xfs_buf_terminate(void)
2029 kmem_zone_destroy(xfs_buf_zone);