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>
36 #include <linux/list_sort.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(struct shrinker *, 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 * Internal xfs_buf_t object manipulation
177 xfs_buftarg_t *target,
178 xfs_off_t range_base,
180 xfs_buf_flags_t flags)
183 * We don't want certain flags to appear in b_flags.
185 flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD);
187 memset(bp, 0, sizeof(xfs_buf_t));
188 atomic_set(&bp->b_hold, 1);
189 init_completion(&bp->b_iowait);
190 INIT_LIST_HEAD(&bp->b_list);
191 INIT_LIST_HEAD(&bp->b_hash_list);
192 init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */
194 bp->b_target = target;
195 bp->b_file_offset = range_base;
197 * Set buffer_length and count_desired to the same value initially.
198 * I/O routines should use count_desired, which will be the same in
199 * most cases but may be reset (e.g. XFS recovery).
201 bp->b_buffer_length = bp->b_count_desired = range_length;
203 bp->b_bn = XFS_BUF_DADDR_NULL;
204 atomic_set(&bp->b_pin_count, 0);
205 init_waitqueue_head(&bp->b_waiters);
207 XFS_STATS_INC(xb_create);
209 trace_xfs_buf_init(bp, _RET_IP_);
213 * Allocate a page array capable of holding a specified number
214 * of pages, and point the page buf at it.
220 xfs_buf_flags_t flags)
222 /* Make sure that we have a page list */
223 if (bp->b_pages == NULL) {
224 bp->b_offset = xfs_buf_poff(bp->b_file_offset);
225 bp->b_page_count = page_count;
226 if (page_count <= XB_PAGES) {
227 bp->b_pages = bp->b_page_array;
229 bp->b_pages = kmem_alloc(sizeof(struct page *) *
230 page_count, xb_to_km(flags));
231 if (bp->b_pages == NULL)
234 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
240 * Frees b_pages if it was allocated.
246 if (bp->b_pages != bp->b_page_array) {
247 kmem_free(bp->b_pages);
253 * Releases the specified buffer.
255 * The modification state of any associated pages is left unchanged.
256 * The buffer most not be on any hash - use xfs_buf_rele instead for
257 * hashed and refcounted buffers
263 trace_xfs_buf_free(bp, _RET_IP_);
265 ASSERT(list_empty(&bp->b_hash_list));
267 if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) {
270 if (xfs_buf_is_vmapped(bp))
271 vm_unmap_ram(bp->b_addr - bp->b_offset,
274 for (i = 0; i < bp->b_page_count; i++) {
275 struct page *page = bp->b_pages[i];
277 if (bp->b_flags & _XBF_PAGE_CACHE)
278 ASSERT(!PagePrivate(page));
279 page_cache_release(page);
282 _xfs_buf_free_pages(bp);
283 xfs_buf_deallocate(bp);
287 * Finds all pages for buffer in question and builds it's page list.
290 _xfs_buf_lookup_pages(
294 struct address_space *mapping = bp->b_target->bt_mapping;
295 size_t blocksize = bp->b_target->bt_bsize;
296 size_t size = bp->b_count_desired;
297 size_t nbytes, offset;
298 gfp_t gfp_mask = xb_to_gfp(flags);
299 unsigned short page_count, i;
304 end = bp->b_file_offset + bp->b_buffer_length;
305 page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset);
307 error = _xfs_buf_get_pages(bp, page_count, flags);
310 bp->b_flags |= _XBF_PAGE_CACHE;
312 offset = bp->b_offset;
313 first = bp->b_file_offset >> PAGE_CACHE_SHIFT;
315 for (i = 0; i < bp->b_page_count; i++) {
320 page = find_or_create_page(mapping, first + i, gfp_mask);
321 if (unlikely(page == NULL)) {
322 if (flags & XBF_READ_AHEAD) {
323 bp->b_page_count = i;
324 for (i = 0; i < bp->b_page_count; i++)
325 unlock_page(bp->b_pages[i]);
330 * This could deadlock.
332 * But until all the XFS lowlevel code is revamped to
333 * handle buffer allocation failures we can't do much.
335 if (!(++retries % 100))
337 "XFS: possible memory allocation "
338 "deadlock in %s (mode:0x%x)\n",
341 XFS_STATS_INC(xb_page_retries);
342 xfsbufd_wakeup(NULL, 0, gfp_mask);
343 congestion_wait(BLK_RW_ASYNC, HZ/50);
347 XFS_STATS_INC(xb_page_found);
349 nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset);
352 ASSERT(!PagePrivate(page));
353 if (!PageUptodate(page)) {
355 if (blocksize >= PAGE_CACHE_SIZE) {
356 if (flags & XBF_READ)
357 bp->b_flags |= _XBF_PAGE_LOCKED;
358 } else if (!PagePrivate(page)) {
359 if (test_page_region(page, offset, nbytes))
364 bp->b_pages[i] = page;
368 if (!(bp->b_flags & _XBF_PAGE_LOCKED)) {
369 for (i = 0; i < bp->b_page_count; i++)
370 unlock_page(bp->b_pages[i]);
373 if (page_count == bp->b_page_count)
374 bp->b_flags |= XBF_DONE;
380 * Map buffer into kernel address-space if nessecary.
387 /* A single page buffer is always mappable */
388 if (bp->b_page_count == 1) {
389 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
390 bp->b_flags |= XBF_MAPPED;
391 } else if (flags & XBF_MAPPED) {
392 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
394 if (unlikely(bp->b_addr == NULL))
396 bp->b_addr += bp->b_offset;
397 bp->b_flags |= XBF_MAPPED;
404 * Finding and Reading Buffers
408 * Look up, and creates if absent, a lockable buffer for
409 * a given range of an inode. The buffer is returned
410 * locked. If other overlapping buffers exist, they are
411 * released before the new buffer is created and locked,
412 * which may imply that this call will block until those buffers
413 * are unlocked. No I/O is implied by this call.
417 xfs_buftarg_t *btp, /* block device target */
418 xfs_off_t ioff, /* starting offset of range */
419 size_t isize, /* length of range */
420 xfs_buf_flags_t flags,
423 xfs_off_t range_base;
428 range_base = (ioff << BBSHIFT);
429 range_length = (isize << BBSHIFT);
431 /* Check for IOs smaller than the sector size / not sector aligned */
432 ASSERT(!(range_length < (1 << btp->bt_sshift)));
433 ASSERT(!(range_base & (xfs_off_t)btp->bt_smask));
435 hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)];
437 spin_lock(&hash->bh_lock);
439 list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) {
440 ASSERT(btp == bp->b_target);
441 if (bp->b_file_offset == range_base &&
442 bp->b_buffer_length == range_length) {
444 * If we look at something, bring it to the
445 * front of the list for next time.
447 atomic_inc(&bp->b_hold);
448 list_move(&bp->b_hash_list, &hash->bh_list);
455 _xfs_buf_initialize(new_bp, btp, range_base,
456 range_length, flags);
457 new_bp->b_hash = hash;
458 list_add(&new_bp->b_hash_list, &hash->bh_list);
460 XFS_STATS_INC(xb_miss_locked);
463 spin_unlock(&hash->bh_lock);
467 spin_unlock(&hash->bh_lock);
469 /* Attempt to get the semaphore without sleeping,
470 * if this does not work then we need to drop the
471 * spinlock and do a hard attempt on the semaphore.
473 if (down_trylock(&bp->b_sema)) {
474 if (!(flags & XBF_TRYLOCK)) {
475 /* wait for buffer ownership */
477 XFS_STATS_INC(xb_get_locked_waited);
479 /* We asked for a trylock and failed, no need
480 * to look at file offset and length here, we
481 * know that this buffer at least overlaps our
482 * buffer and is locked, therefore our buffer
483 * either does not exist, or is this buffer.
486 XFS_STATS_INC(xb_busy_locked);
494 if (bp->b_flags & XBF_STALE) {
495 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
496 bp->b_flags &= XBF_MAPPED;
499 trace_xfs_buf_find(bp, flags, _RET_IP_);
500 XFS_STATS_INC(xb_get_locked);
505 * Assembles a buffer covering the specified range.
506 * Storage in memory for all portions of the buffer will be allocated,
507 * although backing storage may not be.
511 xfs_buftarg_t *target,/* target for buffer */
512 xfs_off_t ioff, /* starting offset of range */
513 size_t isize, /* length of range */
514 xfs_buf_flags_t flags)
516 xfs_buf_t *bp, *new_bp;
519 new_bp = xfs_buf_allocate(flags);
520 if (unlikely(!new_bp))
523 bp = _xfs_buf_find(target, ioff, isize, flags, new_bp);
525 error = _xfs_buf_lookup_pages(bp, flags);
529 xfs_buf_deallocate(new_bp);
530 if (unlikely(bp == NULL))
534 for (i = 0; i < bp->b_page_count; i++)
535 mark_page_accessed(bp->b_pages[i]);
537 if (!(bp->b_flags & XBF_MAPPED)) {
538 error = _xfs_buf_map_pages(bp, flags);
539 if (unlikely(error)) {
540 printk(KERN_WARNING "%s: failed to map pages\n",
546 XFS_STATS_INC(xb_get);
549 * Always fill in the block number now, the mapped cases can do
550 * their own overlay of this later.
553 bp->b_count_desired = bp->b_buffer_length;
555 trace_xfs_buf_get(bp, flags, _RET_IP_);
559 if (flags & (XBF_LOCK | XBF_TRYLOCK))
568 xfs_buf_flags_t flags)
572 ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE)));
573 ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL);
575 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \
576 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
577 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \
578 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
580 status = xfs_buf_iorequest(bp);
581 if (status || XFS_BUF_ISERROR(bp) || (flags & XBF_ASYNC))
583 return xfs_buf_iowait(bp);
588 xfs_buftarg_t *target,
591 xfs_buf_flags_t flags)
597 bp = xfs_buf_get(target, ioff, isize, flags);
599 trace_xfs_buf_read(bp, flags, _RET_IP_);
601 if (!XFS_BUF_ISDONE(bp)) {
602 XFS_STATS_INC(xb_get_read);
603 _xfs_buf_read(bp, flags);
604 } else if (flags & XBF_ASYNC) {
606 * Read ahead call which is already satisfied,
611 /* We do not want read in the flags */
612 bp->b_flags &= ~XBF_READ;
619 if (flags & (XBF_LOCK | XBF_TRYLOCK))
626 * If we are not low on memory then do the readahead in a deadlock
631 xfs_buftarg_t *target,
634 xfs_buf_flags_t flags)
636 struct backing_dev_info *bdi;
638 bdi = target->bt_mapping->backing_dev_info;
639 if (bdi_read_congested(bdi))
642 flags |= (XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD);
643 xfs_buf_read(target, ioff, isize, flags);
649 xfs_buftarg_t *target)
653 bp = xfs_buf_allocate(0);
655 _xfs_buf_initialize(bp, target, 0, len, 0);
659 static inline struct page *
663 if ((!is_vmalloc_addr(addr))) {
664 return virt_to_page(addr);
666 return vmalloc_to_page(addr);
671 xfs_buf_associate_memory(
678 unsigned long pageaddr;
679 unsigned long offset;
683 pageaddr = (unsigned long)mem & PAGE_CACHE_MASK;
684 offset = (unsigned long)mem - pageaddr;
685 buflen = PAGE_CACHE_ALIGN(len + offset);
686 page_count = buflen >> PAGE_CACHE_SHIFT;
688 /* Free any previous set of page pointers */
690 _xfs_buf_free_pages(bp);
695 rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK);
699 bp->b_offset = offset;
701 for (i = 0; i < bp->b_page_count; i++) {
702 bp->b_pages[i] = mem_to_page((void *)pageaddr);
703 pageaddr += PAGE_CACHE_SIZE;
706 bp->b_count_desired = len;
707 bp->b_buffer_length = buflen;
708 bp->b_flags |= XBF_MAPPED;
709 bp->b_flags &= ~_XBF_PAGE_LOCKED;
717 xfs_buftarg_t *target)
719 unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT;
723 bp = xfs_buf_allocate(0);
724 if (unlikely(bp == NULL))
726 _xfs_buf_initialize(bp, target, 0, len, 0);
728 error = _xfs_buf_get_pages(bp, page_count, 0);
732 for (i = 0; i < page_count; i++) {
733 bp->b_pages[i] = alloc_page(GFP_KERNEL);
737 bp->b_flags |= _XBF_PAGES;
739 error = _xfs_buf_map_pages(bp, XBF_MAPPED);
740 if (unlikely(error)) {
741 printk(KERN_WARNING "%s: failed to map pages\n",
748 trace_xfs_buf_get_noaddr(bp, _RET_IP_);
753 __free_page(bp->b_pages[i]);
754 _xfs_buf_free_pages(bp);
756 xfs_buf_deallocate(bp);
762 * Increment reference count on buffer, to hold the buffer concurrently
763 * with another thread which may release (free) the buffer asynchronously.
764 * Must hold the buffer already to call this function.
770 trace_xfs_buf_hold(bp, _RET_IP_);
771 atomic_inc(&bp->b_hold);
775 * Releases a hold on the specified buffer. If the
776 * the hold count is 1, calls xfs_buf_free.
782 xfs_bufhash_t *hash = bp->b_hash;
784 trace_xfs_buf_rele(bp, _RET_IP_);
786 if (unlikely(!hash)) {
787 ASSERT(!bp->b_relse);
788 if (atomic_dec_and_test(&bp->b_hold))
793 ASSERT(atomic_read(&bp->b_hold) > 0);
794 if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) {
796 atomic_inc(&bp->b_hold);
797 spin_unlock(&hash->bh_lock);
798 (*(bp->b_relse)) (bp);
799 } else if (bp->b_flags & XBF_FS_MANAGED) {
800 spin_unlock(&hash->bh_lock);
802 ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)));
803 list_del_init(&bp->b_hash_list);
804 spin_unlock(&hash->bh_lock);
812 * Mutual exclusion on buffers. Locking model:
814 * Buffers associated with inodes for which buffer locking
815 * is not enabled are not protected by semaphores, and are
816 * assumed to be exclusively owned by the caller. There is a
817 * spinlock in the buffer, used by the caller when concurrent
818 * access is possible.
822 * Locks a buffer object, if it is not already locked.
823 * Note that this in no way locks the underlying pages, so it is only
824 * useful for synchronizing concurrent use of buffer objects, not for
825 * synchronizing independent access to the underlying pages.
833 locked = down_trylock(&bp->b_sema) == 0;
837 trace_xfs_buf_cond_lock(bp, _RET_IP_);
838 return locked ? 0 : -EBUSY;
845 return bp->b_sema.count;
849 * Locks a buffer object.
850 * Note that this in no way locks the underlying pages, so it is only
851 * useful for synchronizing concurrent use of buffer objects, not for
852 * synchronizing independent access to the underlying pages.
854 * If we come across a stale, pinned, locked buffer, we know that we
855 * are being asked to lock a buffer that has been reallocated. Because
856 * it is pinned, we know that the log has not been pushed to disk and
857 * hence it will still be locked. Rather than sleeping until someone
858 * else pushes the log, push it ourselves before trying to get the lock.
864 trace_xfs_buf_lock(bp, _RET_IP_);
866 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
867 xfs_log_force(bp->b_mount, 0);
868 if (atomic_read(&bp->b_io_remaining))
869 blk_run_address_space(bp->b_target->bt_mapping);
873 trace_xfs_buf_lock_done(bp, _RET_IP_);
877 * Releases the lock on the buffer object.
878 * If the buffer is marked delwri but is not queued, do so before we
879 * unlock the buffer as we need to set flags correctly. We also need to
880 * take a reference for the delwri queue because the unlocker is going to
881 * drop their's and they don't know we just queued it.
887 if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) {
888 atomic_inc(&bp->b_hold);
889 bp->b_flags |= XBF_ASYNC;
890 xfs_buf_delwri_queue(bp, 0);
896 trace_xfs_buf_unlock(bp, _RET_IP_);
903 DECLARE_WAITQUEUE (wait, current);
905 if (atomic_read(&bp->b_pin_count) == 0)
908 add_wait_queue(&bp->b_waiters, &wait);
910 set_current_state(TASK_UNINTERRUPTIBLE);
911 if (atomic_read(&bp->b_pin_count) == 0)
913 if (atomic_read(&bp->b_io_remaining))
914 blk_run_address_space(bp->b_target->bt_mapping);
917 remove_wait_queue(&bp->b_waiters, &wait);
918 set_current_state(TASK_RUNNING);
922 * Buffer Utility Routines
927 struct work_struct *work)
930 container_of(work, xfs_buf_t, b_iodone_work);
933 * We can get an EOPNOTSUPP to ordered writes. Here we clear the
934 * ordered flag and reissue them. Because we can't tell the higher
935 * layers directly that they should not issue ordered I/O anymore, they
936 * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion.
938 if ((bp->b_error == EOPNOTSUPP) &&
939 (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) {
940 trace_xfs_buf_ordered_retry(bp, _RET_IP_);
941 bp->b_flags &= ~XBF_ORDERED;
942 bp->b_flags |= _XFS_BARRIER_FAILED;
943 xfs_buf_iorequest(bp);
944 } else if (bp->b_iodone)
945 (*(bp->b_iodone))(bp);
946 else if (bp->b_flags & XBF_ASYNC)
955 trace_xfs_buf_iodone(bp, _RET_IP_);
957 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
958 if (bp->b_error == 0)
959 bp->b_flags |= XBF_DONE;
961 if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) {
963 INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work);
964 queue_work(xfslogd_workqueue, &bp->b_iodone_work);
966 xfs_buf_iodone_work(&bp->b_iodone_work);
969 complete(&bp->b_iowait);
978 ASSERT(error >= 0 && error <= 0xffff);
979 bp->b_error = (unsigned short)error;
980 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
985 struct xfs_mount *mp,
990 bp->b_strat = xfs_bdstrat_cb;
992 bp->b_flags |= XBF_WRITE;
993 bp->b_flags &= ~(XBF_ASYNC | XBF_READ);
995 xfs_buf_delwri_dequeue(bp);
996 xfs_buf_iostrategy(bp);
998 error = xfs_buf_iowait(bp);
1000 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1010 trace_xfs_buf_bdwrite(bp, _RET_IP_);
1012 bp->b_strat = xfs_bdstrat_cb;
1015 bp->b_flags &= ~XBF_READ;
1016 bp->b_flags |= (XBF_DELWRI | XBF_ASYNC);
1018 xfs_buf_delwri_queue(bp, 1);
1022 * Called when we want to stop a buffer from getting written or read.
1023 * We attach the EIO error, muck with its flags, and call biodone
1024 * so that the proper iodone callbacks get called.
1030 #ifdef XFSERRORDEBUG
1031 ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
1035 * No need to wait until the buffer is unpinned, we aren't flushing it.
1037 XFS_BUF_ERROR(bp, EIO);
1040 * We're calling biodone, so delete XBF_DONE flag.
1043 XFS_BUF_UNDELAYWRITE(bp);
1047 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
1054 * Same as xfs_bioerror, except that we are releasing the buffer
1055 * here ourselves, and avoiding the biodone call.
1056 * This is meant for userdata errors; metadata bufs come with
1057 * iodone functions attached, so that we can track down errors.
1063 int64_t fl = XFS_BUF_BFLAGS(bp);
1065 * No need to wait until the buffer is unpinned.
1066 * We aren't flushing it.
1068 * chunkhold expects B_DONE to be set, whether
1069 * we actually finish the I/O or not. We don't want to
1070 * change that interface.
1073 XFS_BUF_UNDELAYWRITE(bp);
1076 XFS_BUF_CLR_IODONE_FUNC(bp);
1077 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
1078 if (!(fl & XBF_ASYNC)) {
1080 * Mark b_error and B_ERROR _both_.
1081 * Lot's of chunkcache code assumes that.
1082 * There's no reason to mark error for
1085 XFS_BUF_ERROR(bp, EIO);
1086 XFS_BUF_FINISH_IOWAIT(bp);
1096 * All xfs metadata buffers except log state machine buffers
1097 * get this attached as their b_bdstrat callback function.
1098 * This is so that we can catch a buffer
1099 * after prematurely unpinning it to forcibly shutdown the filesystem.
1105 if (XFS_FORCED_SHUTDOWN(bp->b_mount)) {
1106 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1108 * Metadata write that didn't get logged but
1109 * written delayed anyway. These aren't associated
1110 * with a transaction, and can be ignored.
1112 if (!bp->b_iodone && !XFS_BUF_ISREAD(bp))
1113 return xfs_bioerror_relse(bp);
1115 return xfs_bioerror(bp);
1118 xfs_buf_iorequest(bp);
1123 * Wrapper around bdstrat so that we can stop data from going to disk in case
1124 * we are shutting down the filesystem. Typically user data goes thru this
1125 * path; one of the exceptions is the superblock.
1129 struct xfs_mount *mp,
1132 if (XFS_FORCED_SHUTDOWN(mp)) {
1133 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1134 xfs_bioerror_relse(bp);
1138 xfs_buf_iorequest(bp);
1146 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1147 bp->b_flags &= ~_XBF_PAGE_LOCKED;
1148 xfs_buf_ioend(bp, schedule);
1157 xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private;
1158 unsigned int blocksize = bp->b_target->bt_bsize;
1159 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1161 xfs_buf_ioerror(bp, -error);
1163 if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1164 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1167 struct page *page = bvec->bv_page;
1169 ASSERT(!PagePrivate(page));
1170 if (unlikely(bp->b_error)) {
1171 if (bp->b_flags & XBF_READ)
1172 ClearPageUptodate(page);
1173 } else if (blocksize >= PAGE_CACHE_SIZE) {
1174 SetPageUptodate(page);
1175 } else if (!PagePrivate(page) &&
1176 (bp->b_flags & _XBF_PAGE_CACHE)) {
1177 set_page_region(page, bvec->bv_offset, bvec->bv_len);
1180 if (--bvec >= bio->bi_io_vec)
1181 prefetchw(&bvec->bv_page->flags);
1183 if (bp->b_flags & _XBF_PAGE_LOCKED)
1185 } while (bvec >= bio->bi_io_vec);
1187 _xfs_buf_ioend(bp, 1);
1195 int rw, map_i, total_nr_pages, nr_pages;
1197 int offset = bp->b_offset;
1198 int size = bp->b_count_desired;
1199 sector_t sector = bp->b_bn;
1200 unsigned int blocksize = bp->b_target->bt_bsize;
1202 total_nr_pages = bp->b_page_count;
1205 if (bp->b_flags & XBF_ORDERED) {
1206 ASSERT(!(bp->b_flags & XBF_READ));
1208 } else if (bp->b_flags & XBF_LOG_BUFFER) {
1209 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1210 bp->b_flags &= ~_XBF_RUN_QUEUES;
1211 rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
1212 } else if (bp->b_flags & _XBF_RUN_QUEUES) {
1213 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1214 bp->b_flags &= ~_XBF_RUN_QUEUES;
1215 rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
1217 rw = (bp->b_flags & XBF_WRITE) ? WRITE :
1218 (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;
1221 /* Special code path for reading a sub page size buffer in --
1222 * we populate up the whole page, and hence the other metadata
1223 * in the same page. This optimization is only valid when the
1224 * filesystem block size is not smaller than the page size.
1226 if ((bp->b_buffer_length < PAGE_CACHE_SIZE) &&
1227 ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) ==
1228 (XBF_READ|_XBF_PAGE_LOCKED)) &&
1229 (blocksize >= PAGE_CACHE_SIZE)) {
1230 bio = bio_alloc(GFP_NOIO, 1);
1232 bio->bi_bdev = bp->b_target->bt_bdev;
1233 bio->bi_sector = sector - (offset >> BBSHIFT);
1234 bio->bi_end_io = xfs_buf_bio_end_io;
1235 bio->bi_private = bp;
1237 bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0);
1240 atomic_inc(&bp->b_io_remaining);
1246 atomic_inc(&bp->b_io_remaining);
1247 nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1248 if (nr_pages > total_nr_pages)
1249 nr_pages = total_nr_pages;
1251 bio = bio_alloc(GFP_NOIO, nr_pages);
1252 bio->bi_bdev = bp->b_target->bt_bdev;
1253 bio->bi_sector = sector;
1254 bio->bi_end_io = xfs_buf_bio_end_io;
1255 bio->bi_private = bp;
1257 for (; size && nr_pages; nr_pages--, map_i++) {
1258 int rbytes, nbytes = PAGE_CACHE_SIZE - offset;
1263 rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset);
1264 if (rbytes < nbytes)
1268 sector += nbytes >> BBSHIFT;
1274 if (likely(bio->bi_size)) {
1275 if (xfs_buf_is_vmapped(bp)) {
1276 flush_kernel_vmap_range(bp->b_addr,
1277 xfs_buf_vmap_len(bp));
1279 submit_bio(rw, bio);
1284 * if we get here, no pages were added to the bio. However,
1285 * we can't just error out here - if the pages are locked then
1286 * we have to unlock them otherwise we can hang on a later
1287 * access to the page.
1289 xfs_buf_ioerror(bp, EIO);
1290 if (bp->b_flags & _XBF_PAGE_LOCKED) {
1292 for (i = 0; i < bp->b_page_count; i++)
1293 unlock_page(bp->b_pages[i]);
1303 trace_xfs_buf_iorequest(bp, _RET_IP_);
1305 if (bp->b_flags & XBF_DELWRI) {
1306 xfs_buf_delwri_queue(bp, 1);
1310 if (bp->b_flags & XBF_WRITE) {
1311 xfs_buf_wait_unpin(bp);
1316 /* Set the count to 1 initially, this will stop an I/O
1317 * completion callout which happens before we have started
1318 * all the I/O from calling xfs_buf_ioend too early.
1320 atomic_set(&bp->b_io_remaining, 1);
1321 _xfs_buf_ioapply(bp);
1322 _xfs_buf_ioend(bp, 0);
1329 * Waits for I/O to complete on the buffer supplied.
1330 * It returns immediately if no I/O is pending.
1331 * It returns the I/O error code, if any, or 0 if there was no error.
1337 trace_xfs_buf_iowait(bp, _RET_IP_);
1339 if (atomic_read(&bp->b_io_remaining))
1340 blk_run_address_space(bp->b_target->bt_mapping);
1341 wait_for_completion(&bp->b_iowait);
1343 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1354 if (bp->b_flags & XBF_MAPPED)
1355 return XFS_BUF_PTR(bp) + offset;
1357 offset += bp->b_offset;
1358 page = bp->b_pages[offset >> PAGE_CACHE_SHIFT];
1359 return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1));
1363 * Move data into or out of a buffer.
1367 xfs_buf_t *bp, /* buffer to process */
1368 size_t boff, /* starting buffer offset */
1369 size_t bsize, /* length to copy */
1370 void *data, /* data address */
1371 xfs_buf_rw_t mode) /* read/write/zero flag */
1373 size_t bend, cpoff, csize;
1376 bend = boff + bsize;
1377 while (boff < bend) {
1378 page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)];
1379 cpoff = xfs_buf_poff(boff + bp->b_offset);
1380 csize = min_t(size_t,
1381 PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff);
1383 ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE));
1387 memset(page_address(page) + cpoff, 0, csize);
1390 memcpy(data, page_address(page) + cpoff, csize);
1393 memcpy(page_address(page) + cpoff, data, csize);
1402 * Handling of buffer targets (buftargs).
1406 * Wait for any bufs with callbacks that have been submitted but
1407 * have not yet returned... walk the hash list for the target.
1414 xfs_bufhash_t *hash;
1417 for (i = 0; i < (1 << btp->bt_hashshift); i++) {
1418 hash = &btp->bt_hash[i];
1420 spin_lock(&hash->bh_lock);
1421 list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) {
1422 ASSERT(btp == bp->b_target);
1423 if (!(bp->b_flags & XBF_FS_MANAGED)) {
1424 spin_unlock(&hash->bh_lock);
1426 * Catch superblock reference count leaks
1429 BUG_ON(bp->b_bn == 0);
1434 spin_unlock(&hash->bh_lock);
1439 * Allocate buffer hash table for a given target.
1440 * For devices containing metadata (i.e. not the log/realtime devices)
1441 * we need to allocate a much larger hash table.
1450 btp->bt_hashshift = external ? 3 : 8; /* 8 or 256 buckets */
1451 btp->bt_hashmask = (1 << btp->bt_hashshift) - 1;
1452 btp->bt_hash = kmem_zalloc_large((1 << btp->bt_hashshift) *
1453 sizeof(xfs_bufhash_t));
1454 for (i = 0; i < (1 << btp->bt_hashshift); i++) {
1455 spin_lock_init(&btp->bt_hash[i].bh_lock);
1456 INIT_LIST_HEAD(&btp->bt_hash[i].bh_list);
1464 kmem_free_large(btp->bt_hash);
1465 btp->bt_hash = NULL;
1469 * buftarg list for delwrite queue processing
1471 static LIST_HEAD(xfs_buftarg_list);
1472 static DEFINE_SPINLOCK(xfs_buftarg_lock);
1475 xfs_register_buftarg(
1478 spin_lock(&xfs_buftarg_lock);
1479 list_add(&btp->bt_list, &xfs_buftarg_list);
1480 spin_unlock(&xfs_buftarg_lock);
1484 xfs_unregister_buftarg(
1487 spin_lock(&xfs_buftarg_lock);
1488 list_del(&btp->bt_list);
1489 spin_unlock(&xfs_buftarg_lock);
1494 struct xfs_mount *mp,
1495 struct xfs_buftarg *btp)
1497 xfs_flush_buftarg(btp, 1);
1498 if (mp->m_flags & XFS_MOUNT_BARRIER)
1499 xfs_blkdev_issue_flush(btp);
1500 xfs_free_bufhash(btp);
1501 iput(btp->bt_mapping->host);
1503 /* Unregister the buftarg first so that we don't get a
1504 * wakeup finding a non-existent task
1506 xfs_unregister_buftarg(btp);
1507 kthread_stop(btp->bt_task);
1513 xfs_setsize_buftarg_flags(
1515 unsigned int blocksize,
1516 unsigned int sectorsize,
1519 btp->bt_bsize = blocksize;
1520 btp->bt_sshift = ffs(sectorsize) - 1;
1521 btp->bt_smask = sectorsize - 1;
1523 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1525 "XFS: Cannot set_blocksize to %u on device %s\n",
1526 sectorsize, XFS_BUFTARG_NAME(btp));
1531 (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) {
1533 "XFS: %u byte sectors in use on device %s. "
1534 "This is suboptimal; %u or greater is ideal.\n",
1535 sectorsize, XFS_BUFTARG_NAME(btp),
1536 (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG);
1543 * When allocating the initial buffer target we have not yet
1544 * read in the superblock, so don't know what sized sectors
1545 * are being used is at this early stage. Play safe.
1548 xfs_setsize_buftarg_early(
1550 struct block_device *bdev)
1552 return xfs_setsize_buftarg_flags(btp,
1553 PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0);
1557 xfs_setsize_buftarg(
1559 unsigned int blocksize,
1560 unsigned int sectorsize)
1562 return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1);
1566 xfs_mapping_buftarg(
1568 struct block_device *bdev)
1570 struct backing_dev_info *bdi;
1571 struct inode *inode;
1572 struct address_space *mapping;
1573 static const struct address_space_operations mapping_aops = {
1574 .sync_page = block_sync_page,
1575 .migratepage = fail_migrate_page,
1578 inode = new_inode(bdev->bd_inode->i_sb);
1581 "XFS: Cannot allocate mapping inode for device %s\n",
1582 XFS_BUFTARG_NAME(btp));
1585 inode->i_mode = S_IFBLK;
1586 inode->i_bdev = bdev;
1587 inode->i_rdev = bdev->bd_dev;
1588 bdi = blk_get_backing_dev_info(bdev);
1590 bdi = &default_backing_dev_info;
1591 mapping = &inode->i_data;
1592 mapping->a_ops = &mapping_aops;
1593 mapping->backing_dev_info = bdi;
1594 mapping_set_gfp_mask(mapping, GFP_NOFS);
1595 btp->bt_mapping = mapping;
1600 xfs_alloc_delwrite_queue(
1606 INIT_LIST_HEAD(&btp->bt_list);
1607 INIT_LIST_HEAD(&btp->bt_delwrite_queue);
1608 spin_lock_init(&btp->bt_delwrite_lock);
1610 btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname);
1611 if (IS_ERR(btp->bt_task)) {
1612 error = PTR_ERR(btp->bt_task);
1615 xfs_register_buftarg(btp);
1622 struct block_device *bdev,
1628 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP);
1630 btp->bt_dev = bdev->bd_dev;
1631 btp->bt_bdev = bdev;
1632 if (xfs_setsize_buftarg_early(btp, bdev))
1634 if (xfs_mapping_buftarg(btp, bdev))
1636 if (xfs_alloc_delwrite_queue(btp, fsname))
1638 xfs_alloc_bufhash(btp, external);
1648 * Delayed write buffer handling
1651 xfs_buf_delwri_queue(
1655 struct list_head *dwq = &bp->b_target->bt_delwrite_queue;
1656 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1658 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1660 ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC));
1663 /* If already in the queue, dequeue and place at tail */
1664 if (!list_empty(&bp->b_list)) {
1665 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1667 atomic_dec(&bp->b_hold);
1668 list_del(&bp->b_list);
1671 if (list_empty(dwq)) {
1672 /* start xfsbufd as it is about to have something to do */
1673 wake_up_process(bp->b_target->bt_task);
1676 bp->b_flags |= _XBF_DELWRI_Q;
1677 list_add_tail(&bp->b_list, dwq);
1678 bp->b_queuetime = jiffies;
1686 xfs_buf_delwri_dequeue(
1689 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1693 if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) {
1694 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1695 list_del_init(&bp->b_list);
1698 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q);
1704 trace_xfs_buf_delwri_dequeue(bp, _RET_IP_);
1708 * If a delwri buffer needs to be pushed before it has aged out, then promote
1709 * it to the head of the delwri queue so that it will be flushed on the next
1710 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1711 * than the age currently needed to flush the buffer. Hence the next time the
1712 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1715 xfs_buf_delwri_promote(
1718 struct xfs_buftarg *btp = bp->b_target;
1719 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1;
1721 ASSERT(bp->b_flags & XBF_DELWRI);
1722 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1725 * Check the buffer age before locking the delayed write queue as we
1726 * don't need to promote buffers that are already past the flush age.
1728 if (bp->b_queuetime < jiffies - age)
1730 bp->b_queuetime = jiffies - age;
1731 spin_lock(&btp->bt_delwrite_lock);
1732 list_move(&bp->b_list, &btp->bt_delwrite_queue);
1733 spin_unlock(&btp->bt_delwrite_lock);
1737 xfs_buf_runall_queues(
1738 struct workqueue_struct *queue)
1740 flush_workqueue(queue);
1745 struct shrinker *shrink,
1751 spin_lock(&xfs_buftarg_lock);
1752 list_for_each_entry(btp, &xfs_buftarg_list, bt_list) {
1753 if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags))
1755 if (list_empty(&btp->bt_delwrite_queue))
1757 set_bit(XBT_FORCE_FLUSH, &btp->bt_flags);
1758 wake_up_process(btp->bt_task);
1760 spin_unlock(&xfs_buftarg_lock);
1765 * Move as many buffers as specified to the supplied list
1766 * idicating if we skipped any buffers to prevent deadlocks.
1769 xfs_buf_delwri_split(
1770 xfs_buftarg_t *target,
1771 struct list_head *list,
1775 struct list_head *dwq = &target->bt_delwrite_queue;
1776 spinlock_t *dwlk = &target->bt_delwrite_lock;
1780 force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1781 INIT_LIST_HEAD(list);
1783 list_for_each_entry_safe(bp, n, dwq, b_list) {
1784 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1785 ASSERT(bp->b_flags & XBF_DELWRI);
1787 if (!XFS_BUF_ISPINNED(bp) && !xfs_buf_cond_lock(bp)) {
1789 time_before(jiffies, bp->b_queuetime + age)) {
1794 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q|
1796 bp->b_flags |= XBF_WRITE;
1797 list_move_tail(&bp->b_list, list);
1808 * Compare function is more complex than it needs to be because
1809 * the return value is only 32 bits and we are doing comparisons
1815 struct list_head *a,
1816 struct list_head *b)
1818 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1819 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1822 diff = ap->b_bn - bp->b_bn;
1831 xfs_buf_delwri_sort(
1832 xfs_buftarg_t *target,
1833 struct list_head *list)
1835 list_sort(NULL, list, xfs_buf_cmp);
1842 xfs_buftarg_t *target = (xfs_buftarg_t *)data;
1844 current->flags |= PF_MEMALLOC;
1849 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10);
1850 long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10);
1852 struct list_head tmp;
1854 if (unlikely(freezing(current))) {
1855 set_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1858 clear_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1861 /* sleep for a long time if there is nothing to do. */
1862 if (list_empty(&target->bt_delwrite_queue))
1863 tout = MAX_SCHEDULE_TIMEOUT;
1864 schedule_timeout_interruptible(tout);
1866 xfs_buf_delwri_split(target, &tmp, age);
1867 list_sort(NULL, &tmp, xfs_buf_cmp);
1868 while (!list_empty(&tmp)) {
1870 bp = list_first_entry(&tmp, struct xfs_buf, b_list);
1871 list_del_init(&bp->b_list);
1872 xfs_buf_iostrategy(bp);
1876 blk_run_address_space(target->bt_mapping);
1878 } while (!kthread_should_stop());
1884 * Go through all incore buffers, and release buffers if they belong to
1885 * the given device. This is used in filesystem error handling to
1886 * preserve the consistency of its metadata.
1890 xfs_buftarg_t *target,
1895 LIST_HEAD(tmp_list);
1896 LIST_HEAD(wait_list);
1898 xfs_buf_runall_queues(xfsconvertd_workqueue);
1899 xfs_buf_runall_queues(xfsdatad_workqueue);
1900 xfs_buf_runall_queues(xfslogd_workqueue);
1902 set_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1903 pincount = xfs_buf_delwri_split(target, &tmp_list, 0);
1906 * Dropped the delayed write list lock, now walk the temporary list.
1907 * All I/O is issued async and then if we need to wait for completion
1908 * we do that after issuing all the IO.
1910 list_sort(NULL, &tmp_list, xfs_buf_cmp);
1911 while (!list_empty(&tmp_list)) {
1912 bp = list_first_entry(&tmp_list, struct xfs_buf, b_list);
1913 ASSERT(target == bp->b_target);
1914 list_del_init(&bp->b_list);
1916 bp->b_flags &= ~XBF_ASYNC;
1917 list_add(&bp->b_list, &wait_list);
1919 xfs_buf_iostrategy(bp);
1923 /* Expedite and wait for IO to complete. */
1924 blk_run_address_space(target->bt_mapping);
1925 while (!list_empty(&wait_list)) {
1926 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
1928 list_del_init(&bp->b_list);
1940 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
1941 KM_ZONE_HWALIGN, NULL);
1945 xfslogd_workqueue = create_workqueue("xfslogd");
1946 if (!xfslogd_workqueue)
1947 goto out_free_buf_zone;
1949 xfsdatad_workqueue = create_workqueue("xfsdatad");
1950 if (!xfsdatad_workqueue)
1951 goto out_destroy_xfslogd_workqueue;
1953 xfsconvertd_workqueue = create_workqueue("xfsconvertd");
1954 if (!xfsconvertd_workqueue)
1955 goto out_destroy_xfsdatad_workqueue;
1957 register_shrinker(&xfs_buf_shake);
1960 out_destroy_xfsdatad_workqueue:
1961 destroy_workqueue(xfsdatad_workqueue);
1962 out_destroy_xfslogd_workqueue:
1963 destroy_workqueue(xfslogd_workqueue);
1965 kmem_zone_destroy(xfs_buf_zone);
1971 xfs_buf_terminate(void)
1973 unregister_shrinker(&xfs_buf_shake);
1974 destroy_workqueue(xfsconvertd_workqueue);
1975 destroy_workqueue(xfsdatad_workqueue);
1976 destroy_workqueue(xfslogd_workqueue);
1977 kmem_zone_destroy(xfs_buf_zone);
1980 #ifdef CONFIG_KDB_MODULES
1982 xfs_get_buftarg_list(void)
1984 return &xfs_buftarg_list;