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 RB_CLEAR_NODE(&bp->b_rbnode);
192 sema_init(&bp->b_sema, 0); /* 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 if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) {
268 if (xfs_buf_is_vmapped(bp))
269 vm_unmap_ram(bp->b_addr - bp->b_offset,
272 for (i = 0; i < bp->b_page_count; i++) {
273 struct page *page = bp->b_pages[i];
275 if (bp->b_flags & _XBF_PAGE_CACHE)
276 ASSERT(!PagePrivate(page));
277 page_cache_release(page);
280 _xfs_buf_free_pages(bp);
281 xfs_buf_deallocate(bp);
285 * Finds all pages for buffer in question and builds it's page list.
288 _xfs_buf_lookup_pages(
292 struct address_space *mapping = bp->b_target->bt_mapping;
293 size_t blocksize = bp->b_target->bt_bsize;
294 size_t size = bp->b_count_desired;
295 size_t nbytes, offset;
296 gfp_t gfp_mask = xb_to_gfp(flags);
297 unsigned short page_count, i;
302 end = bp->b_file_offset + bp->b_buffer_length;
303 page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset);
305 error = _xfs_buf_get_pages(bp, page_count, flags);
308 bp->b_flags |= _XBF_PAGE_CACHE;
310 offset = bp->b_offset;
311 first = bp->b_file_offset >> PAGE_CACHE_SHIFT;
313 for (i = 0; i < bp->b_page_count; i++) {
318 page = find_or_create_page(mapping, first + i, gfp_mask);
319 if (unlikely(page == NULL)) {
320 if (flags & XBF_READ_AHEAD) {
321 bp->b_page_count = i;
322 for (i = 0; i < bp->b_page_count; i++)
323 unlock_page(bp->b_pages[i]);
328 * This could deadlock.
330 * But until all the XFS lowlevel code is revamped to
331 * handle buffer allocation failures we can't do much.
333 if (!(++retries % 100))
335 "XFS: possible memory allocation "
336 "deadlock in %s (mode:0x%x)\n",
339 XFS_STATS_INC(xb_page_retries);
340 xfsbufd_wakeup(NULL, 0, gfp_mask);
341 congestion_wait(BLK_RW_ASYNC, HZ/50);
345 XFS_STATS_INC(xb_page_found);
347 nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset);
350 ASSERT(!PagePrivate(page));
351 if (!PageUptodate(page)) {
353 if (blocksize >= PAGE_CACHE_SIZE) {
354 if (flags & XBF_READ)
355 bp->b_flags |= _XBF_PAGE_LOCKED;
356 } else if (!PagePrivate(page)) {
357 if (test_page_region(page, offset, nbytes))
362 bp->b_pages[i] = page;
366 if (!(bp->b_flags & _XBF_PAGE_LOCKED)) {
367 for (i = 0; i < bp->b_page_count; i++)
368 unlock_page(bp->b_pages[i]);
371 if (page_count == bp->b_page_count)
372 bp->b_flags |= XBF_DONE;
378 * Map buffer into kernel address-space if nessecary.
385 /* A single page buffer is always mappable */
386 if (bp->b_page_count == 1) {
387 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
388 bp->b_flags |= XBF_MAPPED;
389 } else if (flags & XBF_MAPPED) {
390 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
392 if (unlikely(bp->b_addr == NULL))
394 bp->b_addr += bp->b_offset;
395 bp->b_flags |= XBF_MAPPED;
402 * Finding and Reading Buffers
406 * Look up, and creates if absent, a lockable buffer for
407 * a given range of an inode. The buffer is returned
408 * locked. If other overlapping buffers exist, they are
409 * released before the new buffer is created and locked,
410 * which may imply that this call will block until those buffers
411 * are unlocked. No I/O is implied by this call.
415 xfs_buftarg_t *btp, /* block device target */
416 xfs_off_t ioff, /* starting offset of range */
417 size_t isize, /* length of range */
418 xfs_buf_flags_t flags,
421 xfs_off_t range_base;
423 struct xfs_perag *pag;
424 struct rb_node **rbp;
425 struct rb_node *parent;
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));
436 pag = xfs_perag_get(btp->bt_mount,
437 xfs_daddr_to_agno(btp->bt_mount, ioff));
440 spin_lock(&pag->pag_buf_lock);
441 rbp = &pag->pag_buf_tree.rb_node;
446 bp = rb_entry(parent, struct xfs_buf, b_rbnode);
448 if (range_base < bp->b_file_offset)
449 rbp = &(*rbp)->rb_left;
450 else if (range_base > bp->b_file_offset)
451 rbp = &(*rbp)->rb_right;
454 * found a block offset match. If the range doesn't
455 * match, the only way this is allowed is if the buffer
456 * in the cache is stale and the transaction that made
457 * it stale has not yet committed. i.e. we are
458 * reallocating a busy extent. Skip this buffer and
459 * continue searching to the right for an exact match.
461 if (bp->b_buffer_length != range_length) {
462 ASSERT(bp->b_flags & XBF_STALE);
463 rbp = &(*rbp)->rb_right;
466 atomic_inc(&bp->b_hold);
473 _xfs_buf_initialize(new_bp, btp, range_base,
474 range_length, flags);
475 rb_link_node(&new_bp->b_rbnode, parent, rbp);
476 rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree);
477 /* the buffer keeps the perag reference until it is freed */
479 spin_unlock(&pag->pag_buf_lock);
481 XFS_STATS_INC(xb_miss_locked);
482 spin_unlock(&pag->pag_buf_lock);
488 spin_unlock(&pag->pag_buf_lock);
491 if (xfs_buf_cond_lock(bp)) {
492 /* failed, so wait for the lock if requested. */
493 if (!(flags & XBF_TRYLOCK)) {
495 XFS_STATS_INC(xb_get_locked_waited);
498 XFS_STATS_INC(xb_busy_locked);
503 if (bp->b_flags & XBF_STALE) {
504 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
505 bp->b_flags &= XBF_MAPPED;
508 trace_xfs_buf_find(bp, flags, _RET_IP_);
509 XFS_STATS_INC(xb_get_locked);
514 * Assembles a buffer covering the specified range.
515 * Storage in memory for all portions of the buffer will be allocated,
516 * although backing storage may not be.
520 xfs_buftarg_t *target,/* target for buffer */
521 xfs_off_t ioff, /* starting offset of range */
522 size_t isize, /* length of range */
523 xfs_buf_flags_t flags)
525 xfs_buf_t *bp, *new_bp;
528 new_bp = xfs_buf_allocate(flags);
529 if (unlikely(!new_bp))
532 bp = _xfs_buf_find(target, ioff, isize, flags, new_bp);
534 error = _xfs_buf_lookup_pages(bp, flags);
538 xfs_buf_deallocate(new_bp);
539 if (unlikely(bp == NULL))
543 for (i = 0; i < bp->b_page_count; i++)
544 mark_page_accessed(bp->b_pages[i]);
546 if (!(bp->b_flags & XBF_MAPPED)) {
547 error = _xfs_buf_map_pages(bp, flags);
548 if (unlikely(error)) {
549 printk(KERN_WARNING "%s: failed to map pages\n",
555 XFS_STATS_INC(xb_get);
558 * Always fill in the block number now, the mapped cases can do
559 * their own overlay of this later.
562 bp->b_count_desired = bp->b_buffer_length;
564 trace_xfs_buf_get(bp, flags, _RET_IP_);
568 if (flags & (XBF_LOCK | XBF_TRYLOCK))
577 xfs_buf_flags_t flags)
581 ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE)));
582 ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL);
584 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \
585 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
586 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \
587 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
589 status = xfs_buf_iorequest(bp);
590 if (status || XFS_BUF_ISERROR(bp) || (flags & XBF_ASYNC))
592 return xfs_buf_iowait(bp);
597 xfs_buftarg_t *target,
600 xfs_buf_flags_t flags)
606 bp = xfs_buf_get(target, ioff, isize, flags);
608 trace_xfs_buf_read(bp, flags, _RET_IP_);
610 if (!XFS_BUF_ISDONE(bp)) {
611 XFS_STATS_INC(xb_get_read);
612 _xfs_buf_read(bp, flags);
613 } else if (flags & XBF_ASYNC) {
615 * Read ahead call which is already satisfied,
620 /* We do not want read in the flags */
621 bp->b_flags &= ~XBF_READ;
628 if (flags & (XBF_LOCK | XBF_TRYLOCK))
635 * If we are not low on memory then do the readahead in a deadlock
640 xfs_buftarg_t *target,
644 struct backing_dev_info *bdi;
646 bdi = target->bt_mapping->backing_dev_info;
647 if (bdi_read_congested(bdi))
650 xfs_buf_read(target, ioff, isize,
651 XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK);
655 * Read an uncached buffer from disk. Allocates and returns a locked
656 * buffer containing the disk contents or nothing.
659 xfs_buf_read_uncached(
660 struct xfs_mount *mp,
661 struct xfs_buftarg *target,
669 bp = xfs_buf_get_uncached(target, length, flags);
673 /* set up the buffer for a read IO */
675 XFS_BUF_SET_ADDR(bp, daddr);
680 error = xfs_buf_iowait(bp);
681 if (error || bp->b_error) {
691 xfs_buftarg_t *target)
695 bp = xfs_buf_allocate(0);
697 _xfs_buf_initialize(bp, target, 0, len, 0);
701 static inline struct page *
705 if ((!is_vmalloc_addr(addr))) {
706 return virt_to_page(addr);
708 return vmalloc_to_page(addr);
713 xfs_buf_associate_memory(
720 unsigned long pageaddr;
721 unsigned long offset;
725 pageaddr = (unsigned long)mem & PAGE_CACHE_MASK;
726 offset = (unsigned long)mem - pageaddr;
727 buflen = PAGE_CACHE_ALIGN(len + offset);
728 page_count = buflen >> PAGE_CACHE_SHIFT;
730 /* Free any previous set of page pointers */
732 _xfs_buf_free_pages(bp);
737 rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK);
741 bp->b_offset = offset;
743 for (i = 0; i < bp->b_page_count; i++) {
744 bp->b_pages[i] = mem_to_page((void *)pageaddr);
745 pageaddr += PAGE_CACHE_SIZE;
748 bp->b_count_desired = len;
749 bp->b_buffer_length = buflen;
750 bp->b_flags |= XBF_MAPPED;
751 bp->b_flags &= ~_XBF_PAGE_LOCKED;
757 xfs_buf_get_uncached(
758 struct xfs_buftarg *target,
762 unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT;
766 bp = xfs_buf_allocate(0);
767 if (unlikely(bp == NULL))
769 _xfs_buf_initialize(bp, target, 0, len, 0);
771 error = _xfs_buf_get_pages(bp, page_count, 0);
775 for (i = 0; i < page_count; i++) {
776 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
780 bp->b_flags |= _XBF_PAGES;
782 error = _xfs_buf_map_pages(bp, XBF_MAPPED);
783 if (unlikely(error)) {
784 printk(KERN_WARNING "%s: failed to map pages\n",
791 trace_xfs_buf_get_uncached(bp, _RET_IP_);
796 __free_page(bp->b_pages[i]);
797 _xfs_buf_free_pages(bp);
799 xfs_buf_deallocate(bp);
805 * Increment reference count on buffer, to hold the buffer concurrently
806 * with another thread which may release (free) the buffer asynchronously.
807 * Must hold the buffer already to call this function.
813 trace_xfs_buf_hold(bp, _RET_IP_);
814 atomic_inc(&bp->b_hold);
818 * Releases a hold on the specified buffer. If the
819 * the hold count is 1, calls xfs_buf_free.
825 struct xfs_perag *pag = bp->b_pag;
827 trace_xfs_buf_rele(bp, _RET_IP_);
830 ASSERT(!bp->b_relse);
831 ASSERT(RB_EMPTY_NODE(&bp->b_rbnode));
832 if (atomic_dec_and_test(&bp->b_hold))
837 ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode));
838 ASSERT(atomic_read(&bp->b_hold) > 0);
839 if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) {
841 atomic_inc(&bp->b_hold);
842 spin_unlock(&pag->pag_buf_lock);
845 ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)));
846 rb_erase(&bp->b_rbnode, &pag->pag_buf_tree);
847 spin_unlock(&pag->pag_buf_lock);
856 * Mutual exclusion on buffers. Locking model:
858 * Buffers associated with inodes for which buffer locking
859 * is not enabled are not protected by semaphores, and are
860 * assumed to be exclusively owned by the caller. There is a
861 * spinlock in the buffer, used by the caller when concurrent
862 * access is possible.
866 * Locks a buffer object, if it is not already locked. Note that this in
867 * no way locks the underlying pages, so it is only useful for
868 * synchronizing concurrent use of buffer objects, not for synchronizing
869 * independent access to the underlying pages.
871 * If we come across a stale, pinned, locked buffer, we know that we are
872 * being asked to lock a buffer that has been reallocated. Because it is
873 * pinned, we know that the log has not been pushed to disk and hence it
874 * will still be locked. Rather than continuing to have trylock attempts
875 * fail until someone else pushes the log, push it ourselves before
876 * returning. This means that the xfsaild will not get stuck trying
877 * to push on stale inode buffers.
885 locked = down_trylock(&bp->b_sema) == 0;
888 else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
889 xfs_log_force(bp->b_target->bt_mount, 0);
891 trace_xfs_buf_cond_lock(bp, _RET_IP_);
892 return locked ? 0 : -EBUSY;
899 return bp->b_sema.count;
903 * Locks a buffer object.
904 * Note that this in no way locks the underlying pages, so it is only
905 * useful for synchronizing concurrent use of buffer objects, not for
906 * synchronizing independent access to the underlying pages.
908 * If we come across a stale, pinned, locked buffer, we know that we
909 * are being asked to lock a buffer that has been reallocated. Because
910 * it is pinned, we know that the log has not been pushed to disk and
911 * hence it will still be locked. Rather than sleeping until someone
912 * else pushes the log, push it ourselves before trying to get the lock.
918 trace_xfs_buf_lock(bp, _RET_IP_);
920 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
921 xfs_log_force(bp->b_target->bt_mount, 0);
922 if (atomic_read(&bp->b_io_remaining))
923 blk_run_address_space(bp->b_target->bt_mapping);
927 trace_xfs_buf_lock_done(bp, _RET_IP_);
931 * Releases the lock on the buffer object.
932 * If the buffer is marked delwri but is not queued, do so before we
933 * unlock the buffer as we need to set flags correctly. We also need to
934 * take a reference for the delwri queue because the unlocker is going to
935 * drop their's and they don't know we just queued it.
941 if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) {
942 atomic_inc(&bp->b_hold);
943 bp->b_flags |= XBF_ASYNC;
944 xfs_buf_delwri_queue(bp, 0);
950 trace_xfs_buf_unlock(bp, _RET_IP_);
957 DECLARE_WAITQUEUE (wait, current);
959 if (atomic_read(&bp->b_pin_count) == 0)
962 add_wait_queue(&bp->b_waiters, &wait);
964 set_current_state(TASK_UNINTERRUPTIBLE);
965 if (atomic_read(&bp->b_pin_count) == 0)
967 if (atomic_read(&bp->b_io_remaining))
968 blk_run_address_space(bp->b_target->bt_mapping);
971 remove_wait_queue(&bp->b_waiters, &wait);
972 set_current_state(TASK_RUNNING);
976 * Buffer Utility Routines
981 struct work_struct *work)
984 container_of(work, xfs_buf_t, b_iodone_work);
987 (*(bp->b_iodone))(bp);
988 else if (bp->b_flags & XBF_ASYNC)
997 trace_xfs_buf_iodone(bp, _RET_IP_);
999 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1000 if (bp->b_error == 0)
1001 bp->b_flags |= XBF_DONE;
1003 if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) {
1005 INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work);
1006 queue_work(xfslogd_workqueue, &bp->b_iodone_work);
1008 xfs_buf_iodone_work(&bp->b_iodone_work);
1011 complete(&bp->b_iowait);
1020 ASSERT(error >= 0 && error <= 0xffff);
1021 bp->b_error = (unsigned short)error;
1022 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
1027 struct xfs_mount *mp,
1032 bp->b_flags |= XBF_WRITE;
1033 bp->b_flags &= ~(XBF_ASYNC | XBF_READ);
1035 xfs_buf_delwri_dequeue(bp);
1038 error = xfs_buf_iowait(bp);
1040 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1050 trace_xfs_buf_bdwrite(bp, _RET_IP_);
1052 bp->b_flags &= ~XBF_READ;
1053 bp->b_flags |= (XBF_DELWRI | XBF_ASYNC);
1055 xfs_buf_delwri_queue(bp, 1);
1059 * Called when we want to stop a buffer from getting written or read.
1060 * We attach the EIO error, muck with its flags, and call xfs_buf_ioend
1061 * so that the proper iodone callbacks get called.
1067 #ifdef XFSERRORDEBUG
1068 ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
1072 * No need to wait until the buffer is unpinned, we aren't flushing it.
1074 XFS_BUF_ERROR(bp, EIO);
1077 * We're calling xfs_buf_ioend, so delete XBF_DONE flag.
1080 XFS_BUF_UNDELAYWRITE(bp);
1084 xfs_buf_ioend(bp, 0);
1090 * Same as xfs_bioerror, except that we are releasing the buffer
1091 * here ourselves, and avoiding the xfs_buf_ioend call.
1092 * This is meant for userdata errors; metadata bufs come with
1093 * iodone functions attached, so that we can track down errors.
1099 int64_t fl = XFS_BUF_BFLAGS(bp);
1101 * No need to wait until the buffer is unpinned.
1102 * We aren't flushing it.
1104 * chunkhold expects B_DONE to be set, whether
1105 * we actually finish the I/O or not. We don't want to
1106 * change that interface.
1109 XFS_BUF_UNDELAYWRITE(bp);
1112 XFS_BUF_CLR_IODONE_FUNC(bp);
1113 if (!(fl & XBF_ASYNC)) {
1115 * Mark b_error and B_ERROR _both_.
1116 * Lot's of chunkcache code assumes that.
1117 * There's no reason to mark error for
1120 XFS_BUF_ERROR(bp, EIO);
1121 XFS_BUF_FINISH_IOWAIT(bp);
1131 * All xfs metadata buffers except log state machine buffers
1132 * get this attached as their b_bdstrat callback function.
1133 * This is so that we can catch a buffer
1134 * after prematurely unpinning it to forcibly shutdown the filesystem.
1140 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1141 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1143 * Metadata write that didn't get logged but
1144 * written delayed anyway. These aren't associated
1145 * with a transaction, and can be ignored.
1147 if (!bp->b_iodone && !XFS_BUF_ISREAD(bp))
1148 return xfs_bioerror_relse(bp);
1150 return xfs_bioerror(bp);
1153 xfs_buf_iorequest(bp);
1158 * Wrapper around bdstrat so that we can stop data from going to disk in case
1159 * we are shutting down the filesystem. Typically user data goes thru this
1160 * path; one of the exceptions is the superblock.
1164 struct xfs_mount *mp,
1167 if (XFS_FORCED_SHUTDOWN(mp)) {
1168 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1169 xfs_bioerror_relse(bp);
1173 xfs_buf_iorequest(bp);
1181 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1182 bp->b_flags &= ~_XBF_PAGE_LOCKED;
1183 xfs_buf_ioend(bp, schedule);
1192 xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private;
1193 unsigned int blocksize = bp->b_target->bt_bsize;
1194 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1196 xfs_buf_ioerror(bp, -error);
1198 if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1199 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1202 struct page *page = bvec->bv_page;
1204 ASSERT(!PagePrivate(page));
1205 if (unlikely(bp->b_error)) {
1206 if (bp->b_flags & XBF_READ)
1207 ClearPageUptodate(page);
1208 } else if (blocksize >= PAGE_CACHE_SIZE) {
1209 SetPageUptodate(page);
1210 } else if (!PagePrivate(page) &&
1211 (bp->b_flags & _XBF_PAGE_CACHE)) {
1212 set_page_region(page, bvec->bv_offset, bvec->bv_len);
1215 if (--bvec >= bio->bi_io_vec)
1216 prefetchw(&bvec->bv_page->flags);
1218 if (bp->b_flags & _XBF_PAGE_LOCKED)
1220 } while (bvec >= bio->bi_io_vec);
1222 _xfs_buf_ioend(bp, 1);
1230 int rw, map_i, total_nr_pages, nr_pages;
1232 int offset = bp->b_offset;
1233 int size = bp->b_count_desired;
1234 sector_t sector = bp->b_bn;
1235 unsigned int blocksize = bp->b_target->bt_bsize;
1237 total_nr_pages = bp->b_page_count;
1240 if (bp->b_flags & XBF_ORDERED) {
1241 ASSERT(!(bp->b_flags & XBF_READ));
1242 rw = WRITE_FLUSH_FUA;
1243 } else if (bp->b_flags & XBF_LOG_BUFFER) {
1244 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1245 bp->b_flags &= ~_XBF_RUN_QUEUES;
1246 rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
1247 } else if (bp->b_flags & _XBF_RUN_QUEUES) {
1248 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1249 bp->b_flags &= ~_XBF_RUN_QUEUES;
1250 rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
1252 rw = (bp->b_flags & XBF_WRITE) ? WRITE :
1253 (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;
1256 /* Special code path for reading a sub page size buffer in --
1257 * we populate up the whole page, and hence the other metadata
1258 * in the same page. This optimization is only valid when the
1259 * filesystem block size is not smaller than the page size.
1261 if ((bp->b_buffer_length < PAGE_CACHE_SIZE) &&
1262 ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) ==
1263 (XBF_READ|_XBF_PAGE_LOCKED)) &&
1264 (blocksize >= PAGE_CACHE_SIZE)) {
1265 bio = bio_alloc(GFP_NOIO, 1);
1267 bio->bi_bdev = bp->b_target->bt_bdev;
1268 bio->bi_sector = sector - (offset >> BBSHIFT);
1269 bio->bi_end_io = xfs_buf_bio_end_io;
1270 bio->bi_private = bp;
1272 bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0);
1275 atomic_inc(&bp->b_io_remaining);
1281 atomic_inc(&bp->b_io_remaining);
1282 nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1283 if (nr_pages > total_nr_pages)
1284 nr_pages = total_nr_pages;
1286 bio = bio_alloc(GFP_NOIO, nr_pages);
1287 bio->bi_bdev = bp->b_target->bt_bdev;
1288 bio->bi_sector = sector;
1289 bio->bi_end_io = xfs_buf_bio_end_io;
1290 bio->bi_private = bp;
1292 for (; size && nr_pages; nr_pages--, map_i++) {
1293 int rbytes, nbytes = PAGE_CACHE_SIZE - offset;
1298 rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset);
1299 if (rbytes < nbytes)
1303 sector += nbytes >> BBSHIFT;
1309 if (likely(bio->bi_size)) {
1310 if (xfs_buf_is_vmapped(bp)) {
1311 flush_kernel_vmap_range(bp->b_addr,
1312 xfs_buf_vmap_len(bp));
1314 submit_bio(rw, bio);
1319 * if we get here, no pages were added to the bio. However,
1320 * we can't just error out here - if the pages are locked then
1321 * we have to unlock them otherwise we can hang on a later
1322 * access to the page.
1324 xfs_buf_ioerror(bp, EIO);
1325 if (bp->b_flags & _XBF_PAGE_LOCKED) {
1327 for (i = 0; i < bp->b_page_count; i++)
1328 unlock_page(bp->b_pages[i]);
1338 trace_xfs_buf_iorequest(bp, _RET_IP_);
1340 if (bp->b_flags & XBF_DELWRI) {
1341 xfs_buf_delwri_queue(bp, 1);
1345 if (bp->b_flags & XBF_WRITE) {
1346 xfs_buf_wait_unpin(bp);
1351 /* Set the count to 1 initially, this will stop an I/O
1352 * completion callout which happens before we have started
1353 * all the I/O from calling xfs_buf_ioend too early.
1355 atomic_set(&bp->b_io_remaining, 1);
1356 _xfs_buf_ioapply(bp);
1357 _xfs_buf_ioend(bp, 0);
1364 * Waits for I/O to complete on the buffer supplied.
1365 * It returns immediately if no I/O is pending.
1366 * It returns the I/O error code, if any, or 0 if there was no error.
1372 trace_xfs_buf_iowait(bp, _RET_IP_);
1374 if (atomic_read(&bp->b_io_remaining))
1375 blk_run_address_space(bp->b_target->bt_mapping);
1376 wait_for_completion(&bp->b_iowait);
1378 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1389 if (bp->b_flags & XBF_MAPPED)
1390 return XFS_BUF_PTR(bp) + offset;
1392 offset += bp->b_offset;
1393 page = bp->b_pages[offset >> PAGE_CACHE_SHIFT];
1394 return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1));
1398 * Move data into or out of a buffer.
1402 xfs_buf_t *bp, /* buffer to process */
1403 size_t boff, /* starting buffer offset */
1404 size_t bsize, /* length to copy */
1405 void *data, /* data address */
1406 xfs_buf_rw_t mode) /* read/write/zero flag */
1408 size_t bend, cpoff, csize;
1411 bend = boff + bsize;
1412 while (boff < bend) {
1413 page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)];
1414 cpoff = xfs_buf_poff(boff + bp->b_offset);
1415 csize = min_t(size_t,
1416 PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff);
1418 ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE));
1422 memset(page_address(page) + cpoff, 0, csize);
1425 memcpy(data, page_address(page) + cpoff, csize);
1428 memcpy(page_address(page) + cpoff, data, csize);
1437 * Handling of buffer targets (buftargs).
1441 * Wait for any bufs with callbacks that have been submitted but
1442 * have not yet returned... walk the hash list for the target.
1446 struct xfs_buftarg *btp)
1448 struct xfs_perag *pag;
1451 for (i = 0; i < btp->bt_mount->m_sb.sb_agcount; i++) {
1452 pag = xfs_perag_get(btp->bt_mount, i);
1453 spin_lock(&pag->pag_buf_lock);
1454 while (rb_first(&pag->pag_buf_tree)) {
1455 spin_unlock(&pag->pag_buf_lock);
1457 spin_lock(&pag->pag_buf_lock);
1459 spin_unlock(&pag->pag_buf_lock);
1465 * buftarg list for delwrite queue processing
1467 static LIST_HEAD(xfs_buftarg_list);
1468 static DEFINE_SPINLOCK(xfs_buftarg_lock);
1471 xfs_register_buftarg(
1474 spin_lock(&xfs_buftarg_lock);
1475 list_add(&btp->bt_list, &xfs_buftarg_list);
1476 spin_unlock(&xfs_buftarg_lock);
1480 xfs_unregister_buftarg(
1483 spin_lock(&xfs_buftarg_lock);
1484 list_del(&btp->bt_list);
1485 spin_unlock(&xfs_buftarg_lock);
1490 struct xfs_mount *mp,
1491 struct xfs_buftarg *btp)
1493 xfs_flush_buftarg(btp, 1);
1494 if (mp->m_flags & XFS_MOUNT_BARRIER)
1495 xfs_blkdev_issue_flush(btp);
1496 iput(btp->bt_mapping->host);
1498 /* Unregister the buftarg first so that we don't get a
1499 * wakeup finding a non-existent task
1501 xfs_unregister_buftarg(btp);
1502 kthread_stop(btp->bt_task);
1508 xfs_setsize_buftarg_flags(
1510 unsigned int blocksize,
1511 unsigned int sectorsize,
1514 btp->bt_bsize = blocksize;
1515 btp->bt_sshift = ffs(sectorsize) - 1;
1516 btp->bt_smask = sectorsize - 1;
1518 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1520 "XFS: Cannot set_blocksize to %u on device %s\n",
1521 sectorsize, XFS_BUFTARG_NAME(btp));
1526 (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) {
1528 "XFS: %u byte sectors in use on device %s. "
1529 "This is suboptimal; %u or greater is ideal.\n",
1530 sectorsize, XFS_BUFTARG_NAME(btp),
1531 (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG);
1538 * When allocating the initial buffer target we have not yet
1539 * read in the superblock, so don't know what sized sectors
1540 * are being used is at this early stage. Play safe.
1543 xfs_setsize_buftarg_early(
1545 struct block_device *bdev)
1547 return xfs_setsize_buftarg_flags(btp,
1548 PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0);
1552 xfs_setsize_buftarg(
1554 unsigned int blocksize,
1555 unsigned int sectorsize)
1557 return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1);
1561 xfs_mapping_buftarg(
1563 struct block_device *bdev)
1565 struct backing_dev_info *bdi;
1566 struct inode *inode;
1567 struct address_space *mapping;
1568 static const struct address_space_operations mapping_aops = {
1569 .sync_page = block_sync_page,
1570 .migratepage = fail_migrate_page,
1573 inode = new_inode(bdev->bd_inode->i_sb);
1576 "XFS: Cannot allocate mapping inode for device %s\n",
1577 XFS_BUFTARG_NAME(btp));
1580 inode->i_ino = get_next_ino();
1581 inode->i_mode = S_IFBLK;
1582 inode->i_bdev = bdev;
1583 inode->i_rdev = bdev->bd_dev;
1584 bdi = blk_get_backing_dev_info(bdev);
1586 bdi = &default_backing_dev_info;
1587 mapping = &inode->i_data;
1588 mapping->a_ops = &mapping_aops;
1589 mapping->backing_dev_info = bdi;
1590 mapping_set_gfp_mask(mapping, GFP_NOFS);
1591 btp->bt_mapping = mapping;
1596 xfs_alloc_delwrite_queue(
1602 INIT_LIST_HEAD(&btp->bt_list);
1603 INIT_LIST_HEAD(&btp->bt_delwrite_queue);
1604 spin_lock_init(&btp->bt_delwrite_lock);
1606 btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname);
1607 if (IS_ERR(btp->bt_task)) {
1608 error = PTR_ERR(btp->bt_task);
1611 xfs_register_buftarg(btp);
1618 struct xfs_mount *mp,
1619 struct block_device *bdev,
1625 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP);
1628 btp->bt_dev = bdev->bd_dev;
1629 btp->bt_bdev = bdev;
1630 if (xfs_setsize_buftarg_early(btp, bdev))
1632 if (xfs_mapping_buftarg(btp, bdev))
1634 if (xfs_alloc_delwrite_queue(btp, fsname))
1645 * Delayed write buffer handling
1648 xfs_buf_delwri_queue(
1652 struct list_head *dwq = &bp->b_target->bt_delwrite_queue;
1653 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1655 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1657 ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC));
1660 /* If already in the queue, dequeue and place at tail */
1661 if (!list_empty(&bp->b_list)) {
1662 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1664 atomic_dec(&bp->b_hold);
1665 list_del(&bp->b_list);
1668 if (list_empty(dwq)) {
1669 /* start xfsbufd as it is about to have something to do */
1670 wake_up_process(bp->b_target->bt_task);
1673 bp->b_flags |= _XBF_DELWRI_Q;
1674 list_add_tail(&bp->b_list, dwq);
1675 bp->b_queuetime = jiffies;
1683 xfs_buf_delwri_dequeue(
1686 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1690 if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) {
1691 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1692 list_del_init(&bp->b_list);
1695 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q);
1701 trace_xfs_buf_delwri_dequeue(bp, _RET_IP_);
1705 * If a delwri buffer needs to be pushed before it has aged out, then promote
1706 * it to the head of the delwri queue so that it will be flushed on the next
1707 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1708 * than the age currently needed to flush the buffer. Hence the next time the
1709 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1712 xfs_buf_delwri_promote(
1715 struct xfs_buftarg *btp = bp->b_target;
1716 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1;
1718 ASSERT(bp->b_flags & XBF_DELWRI);
1719 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1722 * Check the buffer age before locking the delayed write queue as we
1723 * don't need to promote buffers that are already past the flush age.
1725 if (bp->b_queuetime < jiffies - age)
1727 bp->b_queuetime = jiffies - age;
1728 spin_lock(&btp->bt_delwrite_lock);
1729 list_move(&bp->b_list, &btp->bt_delwrite_queue);
1730 spin_unlock(&btp->bt_delwrite_lock);
1734 xfs_buf_runall_queues(
1735 struct workqueue_struct *queue)
1737 flush_workqueue(queue);
1742 struct shrinker *shrink,
1748 spin_lock(&xfs_buftarg_lock);
1749 list_for_each_entry(btp, &xfs_buftarg_list, bt_list) {
1750 if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags))
1752 if (list_empty(&btp->bt_delwrite_queue))
1754 set_bit(XBT_FORCE_FLUSH, &btp->bt_flags);
1755 wake_up_process(btp->bt_task);
1757 spin_unlock(&xfs_buftarg_lock);
1762 * Move as many buffers as specified to the supplied list
1763 * idicating if we skipped any buffers to prevent deadlocks.
1766 xfs_buf_delwri_split(
1767 xfs_buftarg_t *target,
1768 struct list_head *list,
1772 struct list_head *dwq = &target->bt_delwrite_queue;
1773 spinlock_t *dwlk = &target->bt_delwrite_lock;
1777 force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1778 INIT_LIST_HEAD(list);
1780 list_for_each_entry_safe(bp, n, dwq, b_list) {
1781 ASSERT(bp->b_flags & XBF_DELWRI);
1783 if (!XFS_BUF_ISPINNED(bp) && !xfs_buf_cond_lock(bp)) {
1785 time_before(jiffies, bp->b_queuetime + age)) {
1790 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q|
1792 bp->b_flags |= XBF_WRITE;
1793 list_move_tail(&bp->b_list, list);
1794 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1805 * Compare function is more complex than it needs to be because
1806 * the return value is only 32 bits and we are doing comparisons
1812 struct list_head *a,
1813 struct list_head *b)
1815 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1816 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1819 diff = ap->b_bn - bp->b_bn;
1828 xfs_buf_delwri_sort(
1829 xfs_buftarg_t *target,
1830 struct list_head *list)
1832 list_sort(NULL, list, xfs_buf_cmp);
1839 xfs_buftarg_t *target = (xfs_buftarg_t *)data;
1841 current->flags |= PF_MEMALLOC;
1846 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10);
1847 long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10);
1849 struct list_head tmp;
1851 if (unlikely(freezing(current))) {
1852 set_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1855 clear_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1858 /* sleep for a long time if there is nothing to do. */
1859 if (list_empty(&target->bt_delwrite_queue))
1860 tout = MAX_SCHEDULE_TIMEOUT;
1861 schedule_timeout_interruptible(tout);
1863 xfs_buf_delwri_split(target, &tmp, age);
1864 list_sort(NULL, &tmp, xfs_buf_cmp);
1865 while (!list_empty(&tmp)) {
1867 bp = list_first_entry(&tmp, struct xfs_buf, b_list);
1868 list_del_init(&bp->b_list);
1873 blk_run_address_space(target->bt_mapping);
1875 } while (!kthread_should_stop());
1881 * Go through all incore buffers, and release buffers if they belong to
1882 * the given device. This is used in filesystem error handling to
1883 * preserve the consistency of its metadata.
1887 xfs_buftarg_t *target,
1892 LIST_HEAD(tmp_list);
1893 LIST_HEAD(wait_list);
1895 xfs_buf_runall_queues(xfsconvertd_workqueue);
1896 xfs_buf_runall_queues(xfsdatad_workqueue);
1897 xfs_buf_runall_queues(xfslogd_workqueue);
1899 set_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1900 pincount = xfs_buf_delwri_split(target, &tmp_list, 0);
1903 * Dropped the delayed write list lock, now walk the temporary list.
1904 * All I/O is issued async and then if we need to wait for completion
1905 * we do that after issuing all the IO.
1907 list_sort(NULL, &tmp_list, xfs_buf_cmp);
1908 while (!list_empty(&tmp_list)) {
1909 bp = list_first_entry(&tmp_list, struct xfs_buf, b_list);
1910 ASSERT(target == bp->b_target);
1911 list_del_init(&bp->b_list);
1913 bp->b_flags &= ~XBF_ASYNC;
1914 list_add(&bp->b_list, &wait_list);
1920 /* Expedite and wait for IO to complete. */
1921 blk_run_address_space(target->bt_mapping);
1922 while (!list_empty(&wait_list)) {
1923 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
1925 list_del_init(&bp->b_list);
1937 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
1938 KM_ZONE_HWALIGN, NULL);
1942 xfslogd_workqueue = alloc_workqueue("xfslogd",
1943 WQ_MEM_RECLAIM | WQ_HIGHPRI, 1);
1944 if (!xfslogd_workqueue)
1945 goto out_free_buf_zone;
1947 xfsdatad_workqueue = create_workqueue("xfsdatad");
1948 if (!xfsdatad_workqueue)
1949 goto out_destroy_xfslogd_workqueue;
1951 xfsconvertd_workqueue = create_workqueue("xfsconvertd");
1952 if (!xfsconvertd_workqueue)
1953 goto out_destroy_xfsdatad_workqueue;
1955 register_shrinker(&xfs_buf_shake);
1958 out_destroy_xfsdatad_workqueue:
1959 destroy_workqueue(xfsdatad_workqueue);
1960 out_destroy_xfslogd_workqueue:
1961 destroy_workqueue(xfslogd_workqueue);
1963 kmem_zone_destroy(xfs_buf_zone);
1969 xfs_buf_terminate(void)
1971 unregister_shrinker(&xfs_buf_shake);
1972 destroy_workqueue(xfsconvertd_workqueue);
1973 destroy_workqueue(xfsdatad_workqueue);
1974 destroy_workqueue(xfslogd_workqueue);
1975 kmem_zone_destroy(xfs_buf_zone);
1978 #ifdef CONFIG_KDB_MODULES
1980 xfs_get_buftarg_list(void)
1982 return &xfs_buftarg_list;