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/slab.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>
36 #include <linux/list_sort.h>
41 #include "xfs_dmapi.h"
42 #include "xfs_mount.h"
43 #include "xfs_trace.h"
45 static kmem_zone_t *xfs_buf_zone;
46 STATIC int xfsbufd(void *);
47 STATIC int xfsbufd_wakeup(int, gfp_t);
48 STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int);
49 static struct shrinker xfs_buf_shake = {
50 .shrink = xfsbufd_wakeup,
51 .seeks = DEFAULT_SEEKS,
54 static struct workqueue_struct *xfslogd_workqueue;
55 struct workqueue_struct *xfsdatad_workqueue;
56 struct workqueue_struct *xfsconvertd_workqueue;
58 #ifdef XFS_BUF_LOCK_TRACKING
59 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
60 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
61 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
63 # define XB_SET_OWNER(bp) do { } while (0)
64 # define XB_CLEAR_OWNER(bp) do { } while (0)
65 # define XB_GET_OWNER(bp) do { } while (0)
68 #define xb_to_gfp(flags) \
69 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \
70 ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)
72 #define xb_to_km(flags) \
73 (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)
75 #define xfs_buf_allocate(flags) \
76 kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags))
77 #define xfs_buf_deallocate(bp) \
78 kmem_zone_free(xfs_buf_zone, (bp));
85 * Return true if the buffer is vmapped.
87 * The XBF_MAPPED flag is set if the buffer should be mapped, but the
88 * code is clever enough to know it doesn't have to map a single page,
89 * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1.
91 return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1;
98 return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
102 * Page Region interfaces.
104 * For pages in filesystems where the blocksize is smaller than the
105 * pagesize, we use the page->private field (long) to hold a bitmap
106 * of uptodate regions within the page.
108 * Each such region is "bytes per page / bits per long" bytes long.
110 * NBPPR == number-of-bytes-per-page-region
111 * BTOPR == bytes-to-page-region (rounded up)
112 * BTOPRT == bytes-to-page-region-truncated (rounded down)
114 #if (BITS_PER_LONG == 32)
115 #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */
116 #elif (BITS_PER_LONG == 64)
117 #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */
119 #error BITS_PER_LONG must be 32 or 64
121 #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG)
122 #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT)
123 #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT))
133 first = BTOPR(offset);
134 final = BTOPRT(offset + length - 1);
135 first = min(first, final);
138 mask <<= BITS_PER_LONG - (final - first);
139 mask >>= BITS_PER_LONG - (final);
141 ASSERT(offset + length <= PAGE_CACHE_SIZE);
142 ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0);
153 set_page_private(page,
154 page_private(page) | page_region_mask(offset, length));
155 if (page_private(page) == ~0UL)
156 SetPageUptodate(page);
165 unsigned long mask = page_region_mask(offset, length);
167 return (mask && (page_private(page) & mask) == mask);
171 * Mapping of multi-page buffers into contiguous virtual space
174 typedef struct a_list {
179 static a_list_t *as_free_head;
180 static int as_list_len;
181 static DEFINE_SPINLOCK(as_lock);
184 * Try to batch vunmaps because they are costly.
194 * Xen needs to be able to make sure it can get an exclusive
195 * RO mapping of pages it wants to turn into a pagetable. If
196 * a newly allocated page is also still being vmap()ed by xfs,
197 * it will cause pagetable construction to fail. This is a
198 * quick workaround to always eagerly unmap pages so that Xen
205 aentry = kmalloc(sizeof(a_list_t), GFP_NOWAIT);
206 if (likely(aentry)) {
208 aentry->next = as_free_head;
209 aentry->vm_addr = addr;
210 as_free_head = aentry;
212 spin_unlock(&as_lock);
219 purge_addresses(void)
221 a_list_t *aentry, *old;
223 if (as_free_head == NULL)
227 aentry = as_free_head;
230 spin_unlock(&as_lock);
232 while ((old = aentry) != NULL) {
233 vunmap(aentry->vm_addr);
234 aentry = aentry->next;
240 * Internal xfs_buf_t object manipulation
246 xfs_buftarg_t *target,
247 xfs_off_t range_base,
249 xfs_buf_flags_t flags)
252 * We don't want certain flags to appear in b_flags.
254 flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD);
256 memset(bp, 0, sizeof(xfs_buf_t));
257 atomic_set(&bp->b_hold, 1);
258 init_completion(&bp->b_iowait);
259 INIT_LIST_HEAD(&bp->b_list);
260 INIT_LIST_HEAD(&bp->b_hash_list);
261 init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */
263 bp->b_target = target;
264 bp->b_file_offset = range_base;
266 * Set buffer_length and count_desired to the same value initially.
267 * I/O routines should use count_desired, which will be the same in
268 * most cases but may be reset (e.g. XFS recovery).
270 bp->b_buffer_length = bp->b_count_desired = range_length;
272 bp->b_bn = XFS_BUF_DADDR_NULL;
273 atomic_set(&bp->b_pin_count, 0);
274 init_waitqueue_head(&bp->b_waiters);
276 XFS_STATS_INC(xb_create);
278 trace_xfs_buf_init(bp, _RET_IP_);
282 * Allocate a page array capable of holding a specified number
283 * of pages, and point the page buf at it.
289 xfs_buf_flags_t flags)
291 /* Make sure that we have a page list */
292 if (bp->b_pages == NULL) {
293 bp->b_offset = xfs_buf_poff(bp->b_file_offset);
294 bp->b_page_count = page_count;
295 if (page_count <= XB_PAGES) {
296 bp->b_pages = bp->b_page_array;
298 bp->b_pages = kmem_alloc(sizeof(struct page *) *
299 page_count, xb_to_km(flags));
300 if (bp->b_pages == NULL)
303 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
309 * Frees b_pages if it was allocated.
315 if (bp->b_pages != bp->b_page_array) {
316 kmem_free(bp->b_pages);
322 * Releases the specified buffer.
324 * The modification state of any associated pages is left unchanged.
325 * The buffer most not be on any hash - use xfs_buf_rele instead for
326 * hashed and refcounted buffers
332 trace_xfs_buf_free(bp, _RET_IP_);
334 ASSERT(list_empty(&bp->b_hash_list));
336 if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) {
339 if (xfs_buf_is_vmapped(bp))
340 free_address(bp->b_addr - bp->b_offset);
342 for (i = 0; i < bp->b_page_count; i++) {
343 struct page *page = bp->b_pages[i];
345 if (bp->b_flags & _XBF_PAGE_CACHE)
346 ASSERT(!PagePrivate(page));
347 page_cache_release(page);
350 _xfs_buf_free_pages(bp);
351 xfs_buf_deallocate(bp);
355 * Finds all pages for buffer in question and builds it's page list.
358 _xfs_buf_lookup_pages(
362 struct address_space *mapping = bp->b_target->bt_mapping;
363 size_t blocksize = bp->b_target->bt_bsize;
364 size_t size = bp->b_count_desired;
365 size_t nbytes, offset;
366 gfp_t gfp_mask = xb_to_gfp(flags);
367 unsigned short page_count, i;
372 end = bp->b_file_offset + bp->b_buffer_length;
373 page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset);
375 error = _xfs_buf_get_pages(bp, page_count, flags);
378 bp->b_flags |= _XBF_PAGE_CACHE;
380 offset = bp->b_offset;
381 first = bp->b_file_offset >> PAGE_CACHE_SHIFT;
383 for (i = 0; i < bp->b_page_count; i++) {
388 page = find_or_create_page(mapping, first + i, gfp_mask);
389 if (unlikely(page == NULL)) {
390 if (flags & XBF_READ_AHEAD) {
391 bp->b_page_count = i;
392 for (i = 0; i < bp->b_page_count; i++)
393 unlock_page(bp->b_pages[i]);
398 * This could deadlock.
400 * But until all the XFS lowlevel code is revamped to
401 * handle buffer allocation failures we can't do much.
403 if (!(++retries % 100))
405 "XFS: possible memory allocation "
406 "deadlock in %s (mode:0x%x)\n",
409 XFS_STATS_INC(xb_page_retries);
410 xfsbufd_wakeup(0, gfp_mask);
411 congestion_wait(BLK_RW_ASYNC, HZ/50);
415 XFS_STATS_INC(xb_page_found);
417 nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset);
420 ASSERT(!PagePrivate(page));
421 if (!PageUptodate(page)) {
423 if (blocksize >= PAGE_CACHE_SIZE) {
424 if (flags & XBF_READ)
425 bp->b_flags |= _XBF_PAGE_LOCKED;
426 } else if (!PagePrivate(page)) {
427 if (test_page_region(page, offset, nbytes))
432 bp->b_pages[i] = page;
436 if (!(bp->b_flags & _XBF_PAGE_LOCKED)) {
437 for (i = 0; i < bp->b_page_count; i++)
438 unlock_page(bp->b_pages[i]);
441 if (page_count == bp->b_page_count)
442 bp->b_flags |= XBF_DONE;
448 * Map buffer into kernel address-space if nessecary.
455 /* A single page buffer is always mappable */
456 if (bp->b_page_count == 1) {
457 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
458 bp->b_flags |= XBF_MAPPED;
459 } else if (flags & XBF_MAPPED) {
460 if (as_list_len > 64)
462 bp->b_addr = vmap(bp->b_pages, bp->b_page_count,
463 VM_MAP, PAGE_KERNEL);
464 if (unlikely(bp->b_addr == NULL))
466 bp->b_addr += bp->b_offset;
467 bp->b_flags |= XBF_MAPPED;
474 * Finding and Reading Buffers
478 * Look up, and creates if absent, a lockable buffer for
479 * a given range of an inode. The buffer is returned
480 * locked. If other overlapping buffers exist, they are
481 * released before the new buffer is created and locked,
482 * which may imply that this call will block until those buffers
483 * are unlocked. No I/O is implied by this call.
487 xfs_buftarg_t *btp, /* block device target */
488 xfs_off_t ioff, /* starting offset of range */
489 size_t isize, /* length of range */
490 xfs_buf_flags_t flags,
493 xfs_off_t range_base;
498 range_base = (ioff << BBSHIFT);
499 range_length = (isize << BBSHIFT);
501 /* Check for IOs smaller than the sector size / not sector aligned */
502 ASSERT(!(range_length < (1 << btp->bt_sshift)));
503 ASSERT(!(range_base & (xfs_off_t)btp->bt_smask));
505 hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)];
507 spin_lock(&hash->bh_lock);
509 list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) {
510 ASSERT(btp == bp->b_target);
511 if (bp->b_file_offset == range_base &&
512 bp->b_buffer_length == range_length) {
514 * If we look at something, bring it to the
515 * front of the list for next time.
517 atomic_inc(&bp->b_hold);
518 list_move(&bp->b_hash_list, &hash->bh_list);
525 _xfs_buf_initialize(new_bp, btp, range_base,
526 range_length, flags);
527 new_bp->b_hash = hash;
528 list_add(&new_bp->b_hash_list, &hash->bh_list);
530 XFS_STATS_INC(xb_miss_locked);
533 spin_unlock(&hash->bh_lock);
537 spin_unlock(&hash->bh_lock);
539 /* Attempt to get the semaphore without sleeping,
540 * if this does not work then we need to drop the
541 * spinlock and do a hard attempt on the semaphore.
543 if (down_trylock(&bp->b_sema)) {
544 if (!(flags & XBF_TRYLOCK)) {
545 /* wait for buffer ownership */
547 XFS_STATS_INC(xb_get_locked_waited);
549 /* We asked for a trylock and failed, no need
550 * to look at file offset and length here, we
551 * know that this buffer at least overlaps our
552 * buffer and is locked, therefore our buffer
553 * either does not exist, or is this buffer.
556 XFS_STATS_INC(xb_busy_locked);
564 if (bp->b_flags & XBF_STALE) {
565 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
566 bp->b_flags &= XBF_MAPPED;
569 trace_xfs_buf_find(bp, flags, _RET_IP_);
570 XFS_STATS_INC(xb_get_locked);
575 * Assembles a buffer covering the specified range.
576 * Storage in memory for all portions of the buffer will be allocated,
577 * although backing storage may not be.
581 xfs_buftarg_t *target,/* target for buffer */
582 xfs_off_t ioff, /* starting offset of range */
583 size_t isize, /* length of range */
584 xfs_buf_flags_t flags)
586 xfs_buf_t *bp, *new_bp;
589 new_bp = xfs_buf_allocate(flags);
590 if (unlikely(!new_bp))
593 bp = _xfs_buf_find(target, ioff, isize, flags, new_bp);
595 error = _xfs_buf_lookup_pages(bp, flags);
599 xfs_buf_deallocate(new_bp);
600 if (unlikely(bp == NULL))
604 for (i = 0; i < bp->b_page_count; i++)
605 mark_page_accessed(bp->b_pages[i]);
607 if (!(bp->b_flags & XBF_MAPPED)) {
608 error = _xfs_buf_map_pages(bp, flags);
609 if (unlikely(error)) {
610 printk(KERN_WARNING "%s: failed to map pages\n",
616 XFS_STATS_INC(xb_get);
619 * Always fill in the block number now, the mapped cases can do
620 * their own overlay of this later.
623 bp->b_count_desired = bp->b_buffer_length;
625 trace_xfs_buf_get(bp, flags, _RET_IP_);
629 if (flags & (XBF_LOCK | XBF_TRYLOCK))
638 xfs_buf_flags_t flags)
642 ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE)));
643 ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL);
645 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \
646 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
647 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \
648 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
650 status = xfs_buf_iorequest(bp);
651 if (!status && !(flags & XBF_ASYNC))
652 status = xfs_buf_iowait(bp);
658 xfs_buftarg_t *target,
661 xfs_buf_flags_t flags)
667 bp = xfs_buf_get(target, ioff, isize, flags);
669 trace_xfs_buf_read(bp, flags, _RET_IP_);
671 if (!XFS_BUF_ISDONE(bp)) {
672 XFS_STATS_INC(xb_get_read);
673 _xfs_buf_read(bp, flags);
674 } else if (flags & XBF_ASYNC) {
676 * Read ahead call which is already satisfied,
681 /* We do not want read in the flags */
682 bp->b_flags &= ~XBF_READ;
689 if (flags & (XBF_LOCK | XBF_TRYLOCK))
696 * If we are not low on memory then do the readahead in a deadlock
701 xfs_buftarg_t *target,
704 xfs_buf_flags_t flags)
706 struct backing_dev_info *bdi;
708 bdi = target->bt_mapping->backing_dev_info;
709 if (bdi_read_congested(bdi))
712 flags |= (XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD);
713 xfs_buf_read(target, ioff, isize, flags);
719 xfs_buftarg_t *target)
723 bp = xfs_buf_allocate(0);
725 _xfs_buf_initialize(bp, target, 0, len, 0);
729 static inline struct page *
733 if ((!is_vmalloc_addr(addr))) {
734 return virt_to_page(addr);
736 return vmalloc_to_page(addr);
741 xfs_buf_associate_memory(
748 unsigned long pageaddr;
749 unsigned long offset;
753 pageaddr = (unsigned long)mem & PAGE_CACHE_MASK;
754 offset = (unsigned long)mem - pageaddr;
755 buflen = PAGE_CACHE_ALIGN(len + offset);
756 page_count = buflen >> PAGE_CACHE_SHIFT;
758 /* Free any previous set of page pointers */
760 _xfs_buf_free_pages(bp);
765 rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK);
769 bp->b_offset = offset;
771 for (i = 0; i < bp->b_page_count; i++) {
772 bp->b_pages[i] = mem_to_page((void *)pageaddr);
773 pageaddr += PAGE_CACHE_SIZE;
776 bp->b_count_desired = len;
777 bp->b_buffer_length = buflen;
778 bp->b_flags |= XBF_MAPPED;
779 bp->b_flags &= ~_XBF_PAGE_LOCKED;
787 xfs_buftarg_t *target)
789 unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT;
793 bp = xfs_buf_allocate(0);
794 if (unlikely(bp == NULL))
796 _xfs_buf_initialize(bp, target, 0, len, 0);
798 error = _xfs_buf_get_pages(bp, page_count, 0);
802 for (i = 0; i < page_count; i++) {
803 bp->b_pages[i] = alloc_page(GFP_KERNEL);
807 bp->b_flags |= _XBF_PAGES;
809 error = _xfs_buf_map_pages(bp, XBF_MAPPED);
810 if (unlikely(error)) {
811 printk(KERN_WARNING "%s: failed to map pages\n",
818 trace_xfs_buf_get_noaddr(bp, _RET_IP_);
823 __free_page(bp->b_pages[i]);
824 _xfs_buf_free_pages(bp);
826 xfs_buf_deallocate(bp);
832 * Increment reference count on buffer, to hold the buffer concurrently
833 * with another thread which may release (free) the buffer asynchronously.
834 * Must hold the buffer already to call this function.
840 trace_xfs_buf_hold(bp, _RET_IP_);
841 atomic_inc(&bp->b_hold);
845 * Releases a hold on the specified buffer. If the
846 * the hold count is 1, calls xfs_buf_free.
852 xfs_bufhash_t *hash = bp->b_hash;
854 trace_xfs_buf_rele(bp, _RET_IP_);
856 if (unlikely(!hash)) {
857 ASSERT(!bp->b_relse);
858 if (atomic_dec_and_test(&bp->b_hold))
863 ASSERT(atomic_read(&bp->b_hold) > 0);
864 if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) {
866 atomic_inc(&bp->b_hold);
867 spin_unlock(&hash->bh_lock);
868 (*(bp->b_relse)) (bp);
869 } else if (bp->b_flags & XBF_FS_MANAGED) {
870 spin_unlock(&hash->bh_lock);
872 ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)));
873 list_del_init(&bp->b_hash_list);
874 spin_unlock(&hash->bh_lock);
882 * Mutual exclusion on buffers. Locking model:
884 * Buffers associated with inodes for which buffer locking
885 * is not enabled are not protected by semaphores, and are
886 * assumed to be exclusively owned by the caller. There is a
887 * spinlock in the buffer, used by the caller when concurrent
888 * access is possible.
892 * Locks a buffer object, if it is not already locked.
893 * Note that this in no way locks the underlying pages, so it is only
894 * useful for synchronizing concurrent use of buffer objects, not for
895 * synchronizing independent access to the underlying pages.
903 locked = down_trylock(&bp->b_sema) == 0;
907 trace_xfs_buf_cond_lock(bp, _RET_IP_);
908 return locked ? 0 : -EBUSY;
915 return bp->b_sema.count;
919 * Locks a buffer object.
920 * Note that this in no way locks the underlying pages, so it is only
921 * useful for synchronizing concurrent use of buffer objects, not for
922 * synchronizing independent access to the underlying pages.
928 trace_xfs_buf_lock(bp, _RET_IP_);
930 if (atomic_read(&bp->b_io_remaining))
931 blk_run_address_space(bp->b_target->bt_mapping);
935 trace_xfs_buf_lock_done(bp, _RET_IP_);
939 * Releases the lock on the buffer object.
940 * If the buffer is marked delwri but is not queued, do so before we
941 * unlock the buffer as we need to set flags correctly. We also need to
942 * take a reference for the delwri queue because the unlocker is going to
943 * drop their's and they don't know we just queued it.
949 if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) {
950 atomic_inc(&bp->b_hold);
951 bp->b_flags |= XBF_ASYNC;
952 xfs_buf_delwri_queue(bp, 0);
958 trace_xfs_buf_unlock(bp, _RET_IP_);
963 * Pinning Buffer Storage in Memory
964 * Ensure that no attempt to force a buffer to disk will succeed.
970 trace_xfs_buf_pin(bp, _RET_IP_);
971 atomic_inc(&bp->b_pin_count);
978 trace_xfs_buf_unpin(bp, _RET_IP_);
980 if (atomic_dec_and_test(&bp->b_pin_count))
981 wake_up_all(&bp->b_waiters);
988 return atomic_read(&bp->b_pin_count);
995 DECLARE_WAITQUEUE (wait, current);
997 if (atomic_read(&bp->b_pin_count) == 0)
1000 add_wait_queue(&bp->b_waiters, &wait);
1002 set_current_state(TASK_UNINTERRUPTIBLE);
1003 if (atomic_read(&bp->b_pin_count) == 0)
1005 if (atomic_read(&bp->b_io_remaining))
1006 blk_run_address_space(bp->b_target->bt_mapping);
1009 remove_wait_queue(&bp->b_waiters, &wait);
1010 set_current_state(TASK_RUNNING);
1014 * Buffer Utility Routines
1018 xfs_buf_iodone_work(
1019 struct work_struct *work)
1022 container_of(work, xfs_buf_t, b_iodone_work);
1025 * We can get an EOPNOTSUPP to ordered writes. Here we clear the
1026 * ordered flag and reissue them. Because we can't tell the higher
1027 * layers directly that they should not issue ordered I/O anymore, they
1028 * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion.
1030 if ((bp->b_error == EOPNOTSUPP) &&
1031 (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) {
1032 trace_xfs_buf_ordered_retry(bp, _RET_IP_);
1033 bp->b_flags &= ~XBF_ORDERED;
1034 bp->b_flags |= _XFS_BARRIER_FAILED;
1035 xfs_buf_iorequest(bp);
1036 } else if (bp->b_iodone)
1037 (*(bp->b_iodone))(bp);
1038 else if (bp->b_flags & XBF_ASYNC)
1047 trace_xfs_buf_iodone(bp, _RET_IP_);
1049 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1050 if (bp->b_error == 0)
1051 bp->b_flags |= XBF_DONE;
1053 if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) {
1055 INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work);
1056 queue_work(xfslogd_workqueue, &bp->b_iodone_work);
1058 xfs_buf_iodone_work(&bp->b_iodone_work);
1061 complete(&bp->b_iowait);
1070 ASSERT(error >= 0 && error <= 0xffff);
1071 bp->b_error = (unsigned short)error;
1072 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
1077 struct xfs_mount *mp,
1080 int iowait = (bp->b_flags & XBF_ASYNC) == 0;
1083 bp->b_strat = xfs_bdstrat_cb;
1085 bp->b_flags |= XBF_WRITE;
1087 bp->b_flags |= _XBF_RUN_QUEUES;
1089 xfs_buf_delwri_dequeue(bp);
1090 xfs_buf_iostrategy(bp);
1093 error = xfs_buf_iowait(bp);
1095 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1107 trace_xfs_buf_bdwrite(bp, _RET_IP_);
1109 bp->b_strat = xfs_bdstrat_cb;
1112 bp->b_flags &= ~XBF_READ;
1113 bp->b_flags |= (XBF_DELWRI | XBF_ASYNC);
1115 xfs_buf_delwri_queue(bp, 1);
1119 * Called when we want to stop a buffer from getting written or read.
1120 * We attach the EIO error, muck with its flags, and call biodone
1121 * so that the proper iodone callbacks get called.
1127 #ifdef XFSERRORDEBUG
1128 ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
1132 * No need to wait until the buffer is unpinned, we aren't flushing it.
1134 XFS_BUF_ERROR(bp, EIO);
1137 * We're calling biodone, so delete XBF_DONE flag.
1140 XFS_BUF_UNDELAYWRITE(bp);
1144 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
1151 * Same as xfs_bioerror, except that we are releasing the buffer
1152 * here ourselves, and avoiding the biodone call.
1153 * This is meant for userdata errors; metadata bufs come with
1154 * iodone functions attached, so that we can track down errors.
1160 int64_t fl = XFS_BUF_BFLAGS(bp);
1162 * No need to wait until the buffer is unpinned.
1163 * We aren't flushing it.
1165 * chunkhold expects B_DONE to be set, whether
1166 * we actually finish the I/O or not. We don't want to
1167 * change that interface.
1170 XFS_BUF_UNDELAYWRITE(bp);
1173 XFS_BUF_CLR_IODONE_FUNC(bp);
1174 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
1175 if (!(fl & XBF_ASYNC)) {
1177 * Mark b_error and B_ERROR _both_.
1178 * Lot's of chunkcache code assumes that.
1179 * There's no reason to mark error for
1182 XFS_BUF_ERROR(bp, EIO);
1183 XFS_BUF_FINISH_IOWAIT(bp);
1193 * All xfs metadata buffers except log state machine buffers
1194 * get this attached as their b_bdstrat callback function.
1195 * This is so that we can catch a buffer
1196 * after prematurely unpinning it to forcibly shutdown the filesystem.
1202 if (XFS_FORCED_SHUTDOWN(bp->b_mount)) {
1203 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1205 * Metadata write that didn't get logged but
1206 * written delayed anyway. These aren't associated
1207 * with a transaction, and can be ignored.
1209 if (!bp->b_iodone && !XFS_BUF_ISREAD(bp))
1210 return xfs_bioerror_relse(bp);
1212 return xfs_bioerror(bp);
1215 xfs_buf_iorequest(bp);
1220 * Wrapper around bdstrat so that we can stop data from going to disk in case
1221 * we are shutting down the filesystem. Typically user data goes thru this
1222 * path; one of the exceptions is the superblock.
1226 struct xfs_mount *mp,
1229 if (XFS_FORCED_SHUTDOWN(mp)) {
1230 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1231 xfs_bioerror_relse(bp);
1235 xfs_buf_iorequest(bp);
1243 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1244 bp->b_flags &= ~_XBF_PAGE_LOCKED;
1245 xfs_buf_ioend(bp, schedule);
1254 xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private;
1255 unsigned int blocksize = bp->b_target->bt_bsize;
1256 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1258 xfs_buf_ioerror(bp, -error);
1260 if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1261 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1264 struct page *page = bvec->bv_page;
1266 ASSERT(!PagePrivate(page));
1267 if (unlikely(bp->b_error)) {
1268 if (bp->b_flags & XBF_READ)
1269 ClearPageUptodate(page);
1270 } else if (blocksize >= PAGE_CACHE_SIZE) {
1271 SetPageUptodate(page);
1272 } else if (!PagePrivate(page) &&
1273 (bp->b_flags & _XBF_PAGE_CACHE)) {
1274 set_page_region(page, bvec->bv_offset, bvec->bv_len);
1277 if (--bvec >= bio->bi_io_vec)
1278 prefetchw(&bvec->bv_page->flags);
1280 if (bp->b_flags & _XBF_PAGE_LOCKED)
1282 } while (bvec >= bio->bi_io_vec);
1284 _xfs_buf_ioend(bp, 1);
1292 int rw, map_i, total_nr_pages, nr_pages;
1294 int offset = bp->b_offset;
1295 int size = bp->b_count_desired;
1296 sector_t sector = bp->b_bn;
1297 unsigned int blocksize = bp->b_target->bt_bsize;
1299 total_nr_pages = bp->b_page_count;
1302 if (bp->b_flags & XBF_ORDERED) {
1303 ASSERT(!(bp->b_flags & XBF_READ));
1305 } else if (bp->b_flags & XBF_LOG_BUFFER) {
1306 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1307 bp->b_flags &= ~_XBF_RUN_QUEUES;
1308 rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
1309 } else if (bp->b_flags & _XBF_RUN_QUEUES) {
1310 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1311 bp->b_flags &= ~_XBF_RUN_QUEUES;
1312 rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
1314 rw = (bp->b_flags & XBF_WRITE) ? WRITE :
1315 (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;
1318 /* Special code path for reading a sub page size buffer in --
1319 * we populate up the whole page, and hence the other metadata
1320 * in the same page. This optimization is only valid when the
1321 * filesystem block size is not smaller than the page size.
1323 if ((bp->b_buffer_length < PAGE_CACHE_SIZE) &&
1324 ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) ==
1325 (XBF_READ|_XBF_PAGE_LOCKED)) &&
1326 (blocksize >= PAGE_CACHE_SIZE)) {
1327 bio = bio_alloc(GFP_NOIO, 1);
1329 bio->bi_bdev = bp->b_target->bt_bdev;
1330 bio->bi_sector = sector - (offset >> BBSHIFT);
1331 bio->bi_end_io = xfs_buf_bio_end_io;
1332 bio->bi_private = bp;
1334 bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0);
1337 atomic_inc(&bp->b_io_remaining);
1343 atomic_inc(&bp->b_io_remaining);
1344 nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1345 if (nr_pages > total_nr_pages)
1346 nr_pages = total_nr_pages;
1348 bio = bio_alloc(GFP_NOIO, nr_pages);
1349 bio->bi_bdev = bp->b_target->bt_bdev;
1350 bio->bi_sector = sector;
1351 bio->bi_end_io = xfs_buf_bio_end_io;
1352 bio->bi_private = bp;
1354 for (; size && nr_pages; nr_pages--, map_i++) {
1355 int rbytes, nbytes = PAGE_CACHE_SIZE - offset;
1360 rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset);
1361 if (rbytes < nbytes)
1365 sector += nbytes >> BBSHIFT;
1371 if (likely(bio->bi_size)) {
1372 if (xfs_buf_is_vmapped(bp)) {
1373 flush_kernel_vmap_range(bp->b_addr,
1374 xfs_buf_vmap_len(bp));
1376 submit_bio(rw, bio);
1381 xfs_buf_ioerror(bp, EIO);
1389 trace_xfs_buf_iorequest(bp, _RET_IP_);
1391 if (bp->b_flags & XBF_DELWRI) {
1392 xfs_buf_delwri_queue(bp, 1);
1396 if (bp->b_flags & XBF_WRITE) {
1397 xfs_buf_wait_unpin(bp);
1402 /* Set the count to 1 initially, this will stop an I/O
1403 * completion callout which happens before we have started
1404 * all the I/O from calling xfs_buf_ioend too early.
1406 atomic_set(&bp->b_io_remaining, 1);
1407 _xfs_buf_ioapply(bp);
1408 _xfs_buf_ioend(bp, 0);
1415 * Waits for I/O to complete on the buffer supplied.
1416 * It returns immediately if no I/O is pending.
1417 * It returns the I/O error code, if any, or 0 if there was no error.
1423 trace_xfs_buf_iowait(bp, _RET_IP_);
1425 if (atomic_read(&bp->b_io_remaining))
1426 blk_run_address_space(bp->b_target->bt_mapping);
1427 wait_for_completion(&bp->b_iowait);
1429 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1440 if (bp->b_flags & XBF_MAPPED)
1441 return XFS_BUF_PTR(bp) + offset;
1443 offset += bp->b_offset;
1444 page = bp->b_pages[offset >> PAGE_CACHE_SHIFT];
1445 return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1));
1449 * Move data into or out of a buffer.
1453 xfs_buf_t *bp, /* buffer to process */
1454 size_t boff, /* starting buffer offset */
1455 size_t bsize, /* length to copy */
1456 void *data, /* data address */
1457 xfs_buf_rw_t mode) /* read/write/zero flag */
1459 size_t bend, cpoff, csize;
1462 bend = boff + bsize;
1463 while (boff < bend) {
1464 page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)];
1465 cpoff = xfs_buf_poff(boff + bp->b_offset);
1466 csize = min_t(size_t,
1467 PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff);
1469 ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE));
1473 memset(page_address(page) + cpoff, 0, csize);
1476 memcpy(data, page_address(page) + cpoff, csize);
1479 memcpy(page_address(page) + cpoff, data, csize);
1488 * Handling of buffer targets (buftargs).
1492 * Wait for any bufs with callbacks that have been submitted but
1493 * have not yet returned... walk the hash list for the target.
1500 xfs_bufhash_t *hash;
1503 for (i = 0; i < (1 << btp->bt_hashshift); i++) {
1504 hash = &btp->bt_hash[i];
1506 spin_lock(&hash->bh_lock);
1507 list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) {
1508 ASSERT(btp == bp->b_target);
1509 if (!(bp->b_flags & XBF_FS_MANAGED)) {
1510 spin_unlock(&hash->bh_lock);
1512 * Catch superblock reference count leaks
1515 BUG_ON(bp->b_bn == 0);
1520 spin_unlock(&hash->bh_lock);
1525 * Allocate buffer hash table for a given target.
1526 * For devices containing metadata (i.e. not the log/realtime devices)
1527 * we need to allocate a much larger hash table.
1536 btp->bt_hashshift = external ? 3 : 8; /* 8 or 256 buckets */
1537 btp->bt_hashmask = (1 << btp->bt_hashshift) - 1;
1538 btp->bt_hash = kmem_zalloc_large((1 << btp->bt_hashshift) *
1539 sizeof(xfs_bufhash_t));
1540 for (i = 0; i < (1 << btp->bt_hashshift); i++) {
1541 spin_lock_init(&btp->bt_hash[i].bh_lock);
1542 INIT_LIST_HEAD(&btp->bt_hash[i].bh_list);
1550 kmem_free_large(btp->bt_hash);
1551 btp->bt_hash = NULL;
1555 * buftarg list for delwrite queue processing
1557 static LIST_HEAD(xfs_buftarg_list);
1558 static DEFINE_SPINLOCK(xfs_buftarg_lock);
1561 xfs_register_buftarg(
1564 spin_lock(&xfs_buftarg_lock);
1565 list_add(&btp->bt_list, &xfs_buftarg_list);
1566 spin_unlock(&xfs_buftarg_lock);
1570 xfs_unregister_buftarg(
1573 spin_lock(&xfs_buftarg_lock);
1574 list_del(&btp->bt_list);
1575 spin_unlock(&xfs_buftarg_lock);
1580 struct xfs_mount *mp,
1581 struct xfs_buftarg *btp)
1583 xfs_flush_buftarg(btp, 1);
1584 if (mp->m_flags & XFS_MOUNT_BARRIER)
1585 xfs_blkdev_issue_flush(btp);
1586 xfs_free_bufhash(btp);
1587 iput(btp->bt_mapping->host);
1589 /* Unregister the buftarg first so that we don't get a
1590 * wakeup finding a non-existent task
1592 xfs_unregister_buftarg(btp);
1593 kthread_stop(btp->bt_task);
1599 xfs_setsize_buftarg_flags(
1601 unsigned int blocksize,
1602 unsigned int sectorsize,
1605 btp->bt_bsize = blocksize;
1606 btp->bt_sshift = ffs(sectorsize) - 1;
1607 btp->bt_smask = sectorsize - 1;
1609 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1611 "XFS: Cannot set_blocksize to %u on device %s\n",
1612 sectorsize, XFS_BUFTARG_NAME(btp));
1617 (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) {
1619 "XFS: %u byte sectors in use on device %s. "
1620 "This is suboptimal; %u or greater is ideal.\n",
1621 sectorsize, XFS_BUFTARG_NAME(btp),
1622 (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG);
1629 * When allocating the initial buffer target we have not yet
1630 * read in the superblock, so don't know what sized sectors
1631 * are being used is at this early stage. Play safe.
1634 xfs_setsize_buftarg_early(
1636 struct block_device *bdev)
1638 return xfs_setsize_buftarg_flags(btp,
1639 PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0);
1643 xfs_setsize_buftarg(
1645 unsigned int blocksize,
1646 unsigned int sectorsize)
1648 return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1);
1652 xfs_mapping_buftarg(
1654 struct block_device *bdev)
1656 struct backing_dev_info *bdi;
1657 struct inode *inode;
1658 struct address_space *mapping;
1659 static const struct address_space_operations mapping_aops = {
1660 .sync_page = block_sync_page,
1661 .migratepage = fail_migrate_page,
1664 inode = new_inode(bdev->bd_inode->i_sb);
1667 "XFS: Cannot allocate mapping inode for device %s\n",
1668 XFS_BUFTARG_NAME(btp));
1671 inode->i_mode = S_IFBLK;
1672 inode->i_bdev = bdev;
1673 inode->i_rdev = bdev->bd_dev;
1674 bdi = blk_get_backing_dev_info(bdev);
1676 bdi = &default_backing_dev_info;
1677 mapping = &inode->i_data;
1678 mapping->a_ops = &mapping_aops;
1679 mapping->backing_dev_info = bdi;
1680 mapping_set_gfp_mask(mapping, GFP_NOFS);
1681 btp->bt_mapping = mapping;
1686 xfs_alloc_delwrite_queue(
1691 INIT_LIST_HEAD(&btp->bt_list);
1692 INIT_LIST_HEAD(&btp->bt_delwrite_queue);
1693 spin_lock_init(&btp->bt_delwrite_lock);
1695 btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd");
1696 if (IS_ERR(btp->bt_task)) {
1697 error = PTR_ERR(btp->bt_task);
1700 xfs_register_buftarg(btp);
1707 struct block_device *bdev,
1712 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP);
1714 btp->bt_dev = bdev->bd_dev;
1715 btp->bt_bdev = bdev;
1716 if (xfs_setsize_buftarg_early(btp, bdev))
1718 if (xfs_mapping_buftarg(btp, bdev))
1720 if (xfs_alloc_delwrite_queue(btp))
1722 xfs_alloc_bufhash(btp, external);
1732 * Delayed write buffer handling
1735 xfs_buf_delwri_queue(
1739 struct list_head *dwq = &bp->b_target->bt_delwrite_queue;
1740 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1742 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1744 ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC));
1747 /* If already in the queue, dequeue and place at tail */
1748 if (!list_empty(&bp->b_list)) {
1749 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1751 atomic_dec(&bp->b_hold);
1752 list_del(&bp->b_list);
1755 if (list_empty(dwq)) {
1756 /* start xfsbufd as it is about to have something to do */
1757 wake_up_process(bp->b_target->bt_task);
1760 bp->b_flags |= _XBF_DELWRI_Q;
1761 list_add_tail(&bp->b_list, dwq);
1762 bp->b_queuetime = jiffies;
1770 xfs_buf_delwri_dequeue(
1773 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1777 if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) {
1778 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1779 list_del_init(&bp->b_list);
1782 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q);
1788 trace_xfs_buf_delwri_dequeue(bp, _RET_IP_);
1792 * If a delwri buffer needs to be pushed before it has aged out, then promote
1793 * it to the head of the delwri queue so that it will be flushed on the next
1794 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1795 * than the age currently needed to flush the buffer. Hence the next time the
1796 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1799 xfs_buf_delwri_promote(
1802 struct xfs_buftarg *btp = bp->b_target;
1803 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1;
1805 ASSERT(bp->b_flags & XBF_DELWRI);
1806 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1809 * Check the buffer age before locking the delayed write queue as we
1810 * don't need to promote buffers that are already past the flush age.
1812 if (bp->b_queuetime < jiffies - age)
1814 bp->b_queuetime = jiffies - age;
1815 spin_lock(&btp->bt_delwrite_lock);
1816 list_move(&bp->b_list, &btp->bt_delwrite_queue);
1817 spin_unlock(&btp->bt_delwrite_lock);
1821 xfs_buf_runall_queues(
1822 struct workqueue_struct *queue)
1824 flush_workqueue(queue);
1834 spin_lock(&xfs_buftarg_lock);
1835 list_for_each_entry(btp, &xfs_buftarg_list, bt_list) {
1836 if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags))
1838 if (list_empty(&btp->bt_delwrite_queue))
1840 set_bit(XBT_FORCE_FLUSH, &btp->bt_flags);
1841 wake_up_process(btp->bt_task);
1843 spin_unlock(&xfs_buftarg_lock);
1848 * Move as many buffers as specified to the supplied list
1849 * idicating if we skipped any buffers to prevent deadlocks.
1852 xfs_buf_delwri_split(
1853 xfs_buftarg_t *target,
1854 struct list_head *list,
1858 struct list_head *dwq = &target->bt_delwrite_queue;
1859 spinlock_t *dwlk = &target->bt_delwrite_lock;
1863 force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1864 INIT_LIST_HEAD(list);
1866 list_for_each_entry_safe(bp, n, dwq, b_list) {
1867 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1868 ASSERT(bp->b_flags & XBF_DELWRI);
1870 if (!xfs_buf_ispin(bp) && !xfs_buf_cond_lock(bp)) {
1872 time_before(jiffies, bp->b_queuetime + age)) {
1877 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q|
1879 bp->b_flags |= XBF_WRITE;
1880 list_move_tail(&bp->b_list, list);
1891 * Compare function is more complex than it needs to be because
1892 * the return value is only 32 bits and we are doing comparisons
1898 struct list_head *a,
1899 struct list_head *b)
1901 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1902 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1905 diff = ap->b_bn - bp->b_bn;
1914 xfs_buf_delwri_sort(
1915 xfs_buftarg_t *target,
1916 struct list_head *list)
1918 list_sort(NULL, list, xfs_buf_cmp);
1925 xfs_buftarg_t *target = (xfs_buftarg_t *)data;
1927 current->flags |= PF_MEMALLOC;
1932 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10);
1933 long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10);
1935 struct list_head tmp;
1937 if (unlikely(freezing(current))) {
1938 set_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1941 clear_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1944 /* sleep for a long time if there is nothing to do. */
1945 if (list_empty(&target->bt_delwrite_queue))
1946 tout = MAX_SCHEDULE_TIMEOUT;
1947 schedule_timeout_interruptible(tout);
1949 xfs_buf_delwri_split(target, &tmp, age);
1950 list_sort(NULL, &tmp, xfs_buf_cmp);
1951 while (!list_empty(&tmp)) {
1953 bp = list_first_entry(&tmp, struct xfs_buf, b_list);
1954 list_del_init(&bp->b_list);
1955 xfs_buf_iostrategy(bp);
1959 if (as_list_len > 0)
1962 blk_run_address_space(target->bt_mapping);
1964 } while (!kthread_should_stop());
1970 * Go through all incore buffers, and release buffers if they belong to
1971 * the given device. This is used in filesystem error handling to
1972 * preserve the consistency of its metadata.
1976 xfs_buftarg_t *target,
1981 LIST_HEAD(tmp_list);
1982 LIST_HEAD(wait_list);
1984 xfs_buf_runall_queues(xfsconvertd_workqueue);
1985 xfs_buf_runall_queues(xfsdatad_workqueue);
1986 xfs_buf_runall_queues(xfslogd_workqueue);
1988 set_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1989 pincount = xfs_buf_delwri_split(target, &tmp_list, 0);
1992 * Dropped the delayed write list lock, now walk the temporary list.
1993 * All I/O is issued async and then if we need to wait for completion
1994 * we do that after issuing all the IO.
1996 list_sort(NULL, &tmp_list, xfs_buf_cmp);
1997 while (!list_empty(&tmp_list)) {
1998 bp = list_first_entry(&tmp_list, struct xfs_buf, b_list);
1999 ASSERT(target == bp->b_target);
2000 list_del_init(&bp->b_list);
2002 bp->b_flags &= ~XBF_ASYNC;
2003 list_add(&bp->b_list, &wait_list);
2005 xfs_buf_iostrategy(bp);
2009 /* Expedite and wait for IO to complete. */
2010 blk_run_address_space(target->bt_mapping);
2011 while (!list_empty(&wait_list)) {
2012 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
2014 list_del_init(&bp->b_list);
2026 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
2027 KM_ZONE_HWALIGN, NULL);
2031 xfslogd_workqueue = create_workqueue("xfslogd");
2032 if (!xfslogd_workqueue)
2033 goto out_free_buf_zone;
2035 xfsdatad_workqueue = create_workqueue("xfsdatad");
2036 if (!xfsdatad_workqueue)
2037 goto out_destroy_xfslogd_workqueue;
2039 xfsconvertd_workqueue = create_workqueue("xfsconvertd");
2040 if (!xfsconvertd_workqueue)
2041 goto out_destroy_xfsdatad_workqueue;
2043 register_shrinker(&xfs_buf_shake);
2046 out_destroy_xfsdatad_workqueue:
2047 destroy_workqueue(xfsdatad_workqueue);
2048 out_destroy_xfslogd_workqueue:
2049 destroy_workqueue(xfslogd_workqueue);
2051 kmem_zone_destroy(xfs_buf_zone);
2057 xfs_buf_terminate(void)
2059 unregister_shrinker(&xfs_buf_shake);
2060 destroy_workqueue(xfsconvertd_workqueue);
2061 destroy_workqueue(xfsdatad_workqueue);
2062 destroy_workqueue(xfslogd_workqueue);
2063 kmem_zone_destroy(xfs_buf_zone);
2066 #ifdef CONFIG_KDB_MODULES
2068 xfs_get_buftarg_list(void)
2070 return &xfs_buftarg_list;