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 void xfs_buf_delwri_queue(xfs_buf_t *, int);
49 static struct workqueue_struct *xfslogd_workqueue;
50 struct workqueue_struct *xfsdatad_workqueue;
51 struct workqueue_struct *xfsconvertd_workqueue;
53 #ifdef XFS_BUF_LOCK_TRACKING
54 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
55 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
56 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
58 # define XB_SET_OWNER(bp) do { } while (0)
59 # define XB_CLEAR_OWNER(bp) do { } while (0)
60 # define XB_GET_OWNER(bp) do { } while (0)
63 #define xb_to_gfp(flags) \
64 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \
65 ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)
67 #define xb_to_km(flags) \
68 (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)
70 #define xfs_buf_allocate(flags) \
71 kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags))
72 #define xfs_buf_deallocate(bp) \
73 kmem_zone_free(xfs_buf_zone, (bp));
80 * Return true if the buffer is vmapped.
82 * The XBF_MAPPED flag is set if the buffer should be mapped, but the
83 * code is clever enough to know it doesn't have to map a single page,
84 * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1.
86 return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1;
93 return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
97 * Page Region interfaces.
99 * For pages in filesystems where the blocksize is smaller than the
100 * pagesize, we use the page->private field (long) to hold a bitmap
101 * of uptodate regions within the page.
103 * Each such region is "bytes per page / bits per long" bytes long.
105 * NBPPR == number-of-bytes-per-page-region
106 * BTOPR == bytes-to-page-region (rounded up)
107 * BTOPRT == bytes-to-page-region-truncated (rounded down)
109 #if (BITS_PER_LONG == 32)
110 #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */
111 #elif (BITS_PER_LONG == 64)
112 #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */
114 #error BITS_PER_LONG must be 32 or 64
116 #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG)
117 #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT)
118 #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT))
128 first = BTOPR(offset);
129 final = BTOPRT(offset + length - 1);
130 first = min(first, final);
133 mask <<= BITS_PER_LONG - (final - first);
134 mask >>= BITS_PER_LONG - (final);
136 ASSERT(offset + length <= PAGE_CACHE_SIZE);
137 ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0);
148 set_page_private(page,
149 page_private(page) | page_region_mask(offset, length));
150 if (page_private(page) == ~0UL)
151 SetPageUptodate(page);
160 unsigned long mask = page_region_mask(offset, length);
162 return (mask && (page_private(page) & mask) == mask);
166 * xfs_buf_lru_add - add a buffer to the LRU.
168 * The LRU takes a new reference to the buffer so that it will only be freed
169 * once the shrinker takes the buffer off the LRU.
175 struct xfs_buftarg *btp = bp->b_target;
177 spin_lock(&btp->bt_lru_lock);
178 if (list_empty(&bp->b_lru)) {
179 atomic_inc(&bp->b_hold);
180 list_add_tail(&bp->b_lru, &btp->bt_lru);
183 spin_unlock(&btp->bt_lru_lock);
187 * xfs_buf_lru_del - remove a buffer from the LRU
189 * The unlocked check is safe here because it only occurs when there are not
190 * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there
191 * to optimise the shrinker removing the buffer from the LRU and calling
192 * xfs_buf_free(). i.e. it removes an unneccessary round trip on the
199 struct xfs_buftarg *btp = bp->b_target;
201 if (list_empty(&bp->b_lru))
204 spin_lock(&btp->bt_lru_lock);
205 if (!list_empty(&bp->b_lru)) {
206 list_del_init(&bp->b_lru);
209 spin_unlock(&btp->bt_lru_lock);
213 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
214 * b_lru_ref count so that the buffer is freed immediately when the buffer
215 * reference count falls to zero. If the buffer is already on the LRU, we need
216 * to remove the reference that LRU holds on the buffer.
218 * This prevents build-up of stale buffers on the LRU.
224 bp->b_flags |= XBF_STALE;
225 atomic_set(&(bp)->b_lru_ref, 0);
226 if (!list_empty(&bp->b_lru)) {
227 struct xfs_buftarg *btp = bp->b_target;
229 spin_lock(&btp->bt_lru_lock);
230 if (!list_empty(&bp->b_lru)) {
231 list_del_init(&bp->b_lru);
233 atomic_dec(&bp->b_hold);
235 spin_unlock(&btp->bt_lru_lock);
237 ASSERT(atomic_read(&bp->b_hold) >= 1);
243 xfs_buftarg_t *target,
244 xfs_off_t range_base,
246 xfs_buf_flags_t flags)
249 * We don't want certain flags to appear in b_flags.
251 flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD);
253 memset(bp, 0, sizeof(xfs_buf_t));
254 atomic_set(&bp->b_hold, 1);
255 atomic_set(&bp->b_lru_ref, 1);
256 init_completion(&bp->b_iowait);
257 INIT_LIST_HEAD(&bp->b_lru);
258 INIT_LIST_HEAD(&bp->b_list);
259 RB_CLEAR_NODE(&bp->b_rbnode);
260 sema_init(&bp->b_sema, 0); /* held, no waiters */
262 bp->b_target = target;
263 bp->b_file_offset = range_base;
265 * Set buffer_length and count_desired to the same value initially.
266 * I/O routines should use count_desired, which will be the same in
267 * most cases but may be reset (e.g. XFS recovery).
269 bp->b_buffer_length = bp->b_count_desired = range_length;
271 bp->b_bn = XFS_BUF_DADDR_NULL;
272 atomic_set(&bp->b_pin_count, 0);
273 init_waitqueue_head(&bp->b_waiters);
275 XFS_STATS_INC(xb_create);
277 trace_xfs_buf_init(bp, _RET_IP_);
281 * Allocate a page array capable of holding a specified number
282 * of pages, and point the page buf at it.
288 xfs_buf_flags_t flags)
290 /* Make sure that we have a page list */
291 if (bp->b_pages == NULL) {
292 bp->b_offset = xfs_buf_poff(bp->b_file_offset);
293 bp->b_page_count = page_count;
294 if (page_count <= XB_PAGES) {
295 bp->b_pages = bp->b_page_array;
297 bp->b_pages = kmem_alloc(sizeof(struct page *) *
298 page_count, xb_to_km(flags));
299 if (bp->b_pages == NULL)
302 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
308 * Frees b_pages if it was allocated.
314 if (bp->b_pages != bp->b_page_array) {
315 kmem_free(bp->b_pages);
321 * Releases the specified buffer.
323 * The modification state of any associated pages is left unchanged.
324 * The buffer most not be on any hash - use xfs_buf_rele instead for
325 * hashed and refcounted buffers
331 trace_xfs_buf_free(bp, _RET_IP_);
333 ASSERT(list_empty(&bp->b_lru));
335 if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) {
338 if (xfs_buf_is_vmapped(bp))
339 vm_unmap_ram(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 "possible memory allocation deadlock in %s (mode:0x%x)",
408 XFS_STATS_INC(xb_page_retries);
409 congestion_wait(BLK_RW_ASYNC, HZ/50);
413 XFS_STATS_INC(xb_page_found);
415 nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset);
418 ASSERT(!PagePrivate(page));
419 if (!PageUptodate(page)) {
421 if (blocksize >= PAGE_CACHE_SIZE) {
422 if (flags & XBF_READ)
423 bp->b_flags |= _XBF_PAGE_LOCKED;
424 } else if (!PagePrivate(page)) {
425 if (test_page_region(page, offset, nbytes))
430 bp->b_pages[i] = page;
434 if (!(bp->b_flags & _XBF_PAGE_LOCKED)) {
435 for (i = 0; i < bp->b_page_count; i++)
436 unlock_page(bp->b_pages[i]);
439 if (page_count == bp->b_page_count)
440 bp->b_flags |= XBF_DONE;
446 * Map buffer into kernel address-space if nessecary.
453 /* A single page buffer is always mappable */
454 if (bp->b_page_count == 1) {
455 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
456 bp->b_flags |= XBF_MAPPED;
457 } else if (flags & XBF_MAPPED) {
458 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
460 if (unlikely(bp->b_addr == NULL))
462 bp->b_addr += bp->b_offset;
463 bp->b_flags |= XBF_MAPPED;
470 * Finding and Reading Buffers
474 * Look up, and creates if absent, a lockable buffer for
475 * a given range of an inode. The buffer is returned
476 * locked. If other overlapping buffers exist, they are
477 * released before the new buffer is created and locked,
478 * which may imply that this call will block until those buffers
479 * are unlocked. No I/O is implied by this call.
483 xfs_buftarg_t *btp, /* block device target */
484 xfs_off_t ioff, /* starting offset of range */
485 size_t isize, /* length of range */
486 xfs_buf_flags_t flags,
489 xfs_off_t range_base;
491 struct xfs_perag *pag;
492 struct rb_node **rbp;
493 struct rb_node *parent;
496 range_base = (ioff << BBSHIFT);
497 range_length = (isize << BBSHIFT);
499 /* Check for IOs smaller than the sector size / not sector aligned */
500 ASSERT(!(range_length < (1 << btp->bt_sshift)));
501 ASSERT(!(range_base & (xfs_off_t)btp->bt_smask));
504 pag = xfs_perag_get(btp->bt_mount,
505 xfs_daddr_to_agno(btp->bt_mount, ioff));
508 spin_lock(&pag->pag_buf_lock);
509 rbp = &pag->pag_buf_tree.rb_node;
514 bp = rb_entry(parent, struct xfs_buf, b_rbnode);
516 if (range_base < bp->b_file_offset)
517 rbp = &(*rbp)->rb_left;
518 else if (range_base > bp->b_file_offset)
519 rbp = &(*rbp)->rb_right;
522 * found a block offset match. If the range doesn't
523 * match, the only way this is allowed is if the buffer
524 * in the cache is stale and the transaction that made
525 * it stale has not yet committed. i.e. we are
526 * reallocating a busy extent. Skip this buffer and
527 * continue searching to the right for an exact match.
529 if (bp->b_buffer_length != range_length) {
530 ASSERT(bp->b_flags & XBF_STALE);
531 rbp = &(*rbp)->rb_right;
534 atomic_inc(&bp->b_hold);
541 _xfs_buf_initialize(new_bp, btp, range_base,
542 range_length, flags);
543 rb_link_node(&new_bp->b_rbnode, parent, rbp);
544 rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree);
545 /* the buffer keeps the perag reference until it is freed */
547 spin_unlock(&pag->pag_buf_lock);
549 XFS_STATS_INC(xb_miss_locked);
550 spin_unlock(&pag->pag_buf_lock);
556 spin_unlock(&pag->pag_buf_lock);
559 if (xfs_buf_cond_lock(bp)) {
560 /* failed, so wait for the lock if requested. */
561 if (!(flags & XBF_TRYLOCK)) {
563 XFS_STATS_INC(xb_get_locked_waited);
566 XFS_STATS_INC(xb_busy_locked);
571 if (bp->b_flags & XBF_STALE) {
572 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
573 bp->b_flags &= XBF_MAPPED;
576 trace_xfs_buf_find(bp, flags, _RET_IP_);
577 XFS_STATS_INC(xb_get_locked);
582 * Assembles a buffer covering the specified range.
583 * Storage in memory for all portions of the buffer will be allocated,
584 * although backing storage may not be.
588 xfs_buftarg_t *target,/* target for buffer */
589 xfs_off_t ioff, /* starting offset of range */
590 size_t isize, /* length of range */
591 xfs_buf_flags_t flags)
593 xfs_buf_t *bp, *new_bp;
596 new_bp = xfs_buf_allocate(flags);
597 if (unlikely(!new_bp))
600 bp = _xfs_buf_find(target, ioff, isize, flags, new_bp);
602 error = _xfs_buf_lookup_pages(bp, flags);
606 xfs_buf_deallocate(new_bp);
607 if (unlikely(bp == NULL))
611 for (i = 0; i < bp->b_page_count; i++)
612 mark_page_accessed(bp->b_pages[i]);
614 if (!(bp->b_flags & XBF_MAPPED)) {
615 error = _xfs_buf_map_pages(bp, flags);
616 if (unlikely(error)) {
617 xfs_warn(target->bt_mount,
618 "%s: failed to map pages\n", __func__);
623 XFS_STATS_INC(xb_get);
626 * Always fill in the block number now, the mapped cases can do
627 * their own overlay of this later.
630 bp->b_count_desired = bp->b_buffer_length;
632 trace_xfs_buf_get(bp, flags, _RET_IP_);
636 if (flags & (XBF_LOCK | XBF_TRYLOCK))
645 xfs_buf_flags_t flags)
649 ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE)));
650 ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL);
652 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \
653 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
654 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \
655 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
657 status = xfs_buf_iorequest(bp);
658 if (status || XFS_BUF_ISERROR(bp) || (flags & XBF_ASYNC))
660 return xfs_buf_iowait(bp);
665 xfs_buftarg_t *target,
668 xfs_buf_flags_t flags)
674 bp = xfs_buf_get(target, ioff, isize, flags);
676 trace_xfs_buf_read(bp, flags, _RET_IP_);
678 if (!XFS_BUF_ISDONE(bp)) {
679 XFS_STATS_INC(xb_get_read);
680 _xfs_buf_read(bp, flags);
681 } else if (flags & XBF_ASYNC) {
683 * Read ahead call which is already satisfied,
688 /* We do not want read in the flags */
689 bp->b_flags &= ~XBF_READ;
696 if (flags & (XBF_LOCK | XBF_TRYLOCK))
703 * If we are not low on memory then do the readahead in a deadlock
708 xfs_buftarg_t *target,
712 struct backing_dev_info *bdi;
714 bdi = target->bt_mapping->backing_dev_info;
715 if (bdi_read_congested(bdi))
718 xfs_buf_read(target, ioff, isize,
719 XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK);
723 * Read an uncached buffer from disk. Allocates and returns a locked
724 * buffer containing the disk contents or nothing.
727 xfs_buf_read_uncached(
728 struct xfs_mount *mp,
729 struct xfs_buftarg *target,
737 bp = xfs_buf_get_uncached(target, length, flags);
741 /* set up the buffer for a read IO */
743 XFS_BUF_SET_ADDR(bp, daddr);
748 error = xfs_buf_iowait(bp);
749 if (error || bp->b_error) {
759 xfs_buftarg_t *target)
763 bp = xfs_buf_allocate(0);
765 _xfs_buf_initialize(bp, target, 0, len, 0);
769 static inline struct page *
773 if ((!is_vmalloc_addr(addr))) {
774 return virt_to_page(addr);
776 return vmalloc_to_page(addr);
781 xfs_buf_associate_memory(
788 unsigned long pageaddr;
789 unsigned long offset;
793 pageaddr = (unsigned long)mem & PAGE_CACHE_MASK;
794 offset = (unsigned long)mem - pageaddr;
795 buflen = PAGE_CACHE_ALIGN(len + offset);
796 page_count = buflen >> PAGE_CACHE_SHIFT;
798 /* Free any previous set of page pointers */
800 _xfs_buf_free_pages(bp);
805 rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK);
809 bp->b_offset = offset;
811 for (i = 0; i < bp->b_page_count; i++) {
812 bp->b_pages[i] = mem_to_page((void *)pageaddr);
813 pageaddr += PAGE_CACHE_SIZE;
816 bp->b_count_desired = len;
817 bp->b_buffer_length = buflen;
818 bp->b_flags |= XBF_MAPPED;
819 bp->b_flags &= ~_XBF_PAGE_LOCKED;
825 xfs_buf_get_uncached(
826 struct xfs_buftarg *target,
830 unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT;
834 bp = xfs_buf_allocate(0);
835 if (unlikely(bp == NULL))
837 _xfs_buf_initialize(bp, target, 0, len, 0);
839 error = _xfs_buf_get_pages(bp, page_count, 0);
843 for (i = 0; i < page_count; i++) {
844 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
848 bp->b_flags |= _XBF_PAGES;
850 error = _xfs_buf_map_pages(bp, XBF_MAPPED);
851 if (unlikely(error)) {
852 xfs_warn(target->bt_mount,
853 "%s: failed to map pages\n", __func__);
859 trace_xfs_buf_get_uncached(bp, _RET_IP_);
864 __free_page(bp->b_pages[i]);
865 _xfs_buf_free_pages(bp);
867 xfs_buf_deallocate(bp);
873 * Increment reference count on buffer, to hold the buffer concurrently
874 * with another thread which may release (free) the buffer asynchronously.
875 * Must hold the buffer already to call this function.
881 trace_xfs_buf_hold(bp, _RET_IP_);
882 atomic_inc(&bp->b_hold);
886 * Releases a hold on the specified buffer. If the
887 * the hold count is 1, calls xfs_buf_free.
893 struct xfs_perag *pag = bp->b_pag;
895 trace_xfs_buf_rele(bp, _RET_IP_);
898 ASSERT(list_empty(&bp->b_lru));
899 ASSERT(RB_EMPTY_NODE(&bp->b_rbnode));
900 if (atomic_dec_and_test(&bp->b_hold))
905 ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode));
907 ASSERT(atomic_read(&bp->b_hold) > 0);
908 if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) {
909 if (!(bp->b_flags & XBF_STALE) &&
910 atomic_read(&bp->b_lru_ref)) {
912 spin_unlock(&pag->pag_buf_lock);
915 ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)));
916 rb_erase(&bp->b_rbnode, &pag->pag_buf_tree);
917 spin_unlock(&pag->pag_buf_lock);
926 * Mutual exclusion on buffers. Locking model:
928 * Buffers associated with inodes for which buffer locking
929 * is not enabled are not protected by semaphores, and are
930 * assumed to be exclusively owned by the caller. There is a
931 * spinlock in the buffer, used by the caller when concurrent
932 * access is possible.
936 * Locks a buffer object, if it is not already locked. Note that this in
937 * no way locks the underlying pages, so it is only useful for
938 * synchronizing concurrent use of buffer objects, not for synchronizing
939 * independent access to the underlying pages.
941 * If we come across a stale, pinned, locked buffer, we know that we are
942 * being asked to lock a buffer that has been reallocated. Because it is
943 * pinned, we know that the log has not been pushed to disk and hence it
944 * will still be locked. Rather than continuing to have trylock attempts
945 * fail until someone else pushes the log, push it ourselves before
946 * returning. This means that the xfsaild will not get stuck trying
947 * to push on stale inode buffers.
955 locked = down_trylock(&bp->b_sema) == 0;
958 else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
959 xfs_log_force(bp->b_target->bt_mount, 0);
961 trace_xfs_buf_cond_lock(bp, _RET_IP_);
962 return locked ? 0 : -EBUSY;
969 return bp->b_sema.count;
973 * Locks a buffer object.
974 * Note that this in no way locks the underlying pages, so it is only
975 * useful for synchronizing concurrent use of buffer objects, not for
976 * synchronizing independent access to the underlying pages.
978 * If we come across a stale, pinned, locked buffer, we know that we
979 * are being asked to lock a buffer that has been reallocated. Because
980 * it is pinned, we know that the log has not been pushed to disk and
981 * hence it will still be locked. Rather than sleeping until someone
982 * else pushes the log, push it ourselves before trying to get the lock.
988 trace_xfs_buf_lock(bp, _RET_IP_);
990 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
991 xfs_log_force(bp->b_target->bt_mount, 0);
992 if (atomic_read(&bp->b_io_remaining))
993 blk_run_address_space(bp->b_target->bt_mapping);
997 trace_xfs_buf_lock_done(bp, _RET_IP_);
1001 * Releases the lock on the buffer object.
1002 * If the buffer is marked delwri but is not queued, do so before we
1003 * unlock the buffer as we need to set flags correctly. We also need to
1004 * take a reference for the delwri queue because the unlocker is going to
1005 * drop their's and they don't know we just queued it.
1011 if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) {
1012 atomic_inc(&bp->b_hold);
1013 bp->b_flags |= XBF_ASYNC;
1014 xfs_buf_delwri_queue(bp, 0);
1020 trace_xfs_buf_unlock(bp, _RET_IP_);
1027 DECLARE_WAITQUEUE (wait, current);
1029 if (atomic_read(&bp->b_pin_count) == 0)
1032 add_wait_queue(&bp->b_waiters, &wait);
1034 set_current_state(TASK_UNINTERRUPTIBLE);
1035 if (atomic_read(&bp->b_pin_count) == 0)
1037 if (atomic_read(&bp->b_io_remaining))
1038 blk_run_address_space(bp->b_target->bt_mapping);
1041 remove_wait_queue(&bp->b_waiters, &wait);
1042 set_current_state(TASK_RUNNING);
1046 * Buffer Utility Routines
1050 xfs_buf_iodone_work(
1051 struct work_struct *work)
1054 container_of(work, xfs_buf_t, b_iodone_work);
1057 (*(bp->b_iodone))(bp);
1058 else if (bp->b_flags & XBF_ASYNC)
1067 trace_xfs_buf_iodone(bp, _RET_IP_);
1069 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1070 if (bp->b_error == 0)
1071 bp->b_flags |= XBF_DONE;
1073 if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) {
1075 INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work);
1076 queue_work(xfslogd_workqueue, &bp->b_iodone_work);
1078 xfs_buf_iodone_work(&bp->b_iodone_work);
1081 complete(&bp->b_iowait);
1090 ASSERT(error >= 0 && error <= 0xffff);
1091 bp->b_error = (unsigned short)error;
1092 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
1097 struct xfs_mount *mp,
1102 bp->b_flags |= XBF_WRITE;
1103 bp->b_flags &= ~(XBF_ASYNC | XBF_READ);
1105 xfs_buf_delwri_dequeue(bp);
1108 error = xfs_buf_iowait(bp);
1110 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1120 trace_xfs_buf_bdwrite(bp, _RET_IP_);
1122 bp->b_flags &= ~XBF_READ;
1123 bp->b_flags |= (XBF_DELWRI | XBF_ASYNC);
1125 xfs_buf_delwri_queue(bp, 1);
1129 * Called when we want to stop a buffer from getting written or read.
1130 * We attach the EIO error, muck with its flags, and call xfs_buf_ioend
1131 * so that the proper iodone callbacks get called.
1137 #ifdef XFSERRORDEBUG
1138 ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
1142 * No need to wait until the buffer is unpinned, we aren't flushing it.
1144 XFS_BUF_ERROR(bp, EIO);
1147 * We're calling xfs_buf_ioend, so delete XBF_DONE flag.
1150 XFS_BUF_UNDELAYWRITE(bp);
1154 xfs_buf_ioend(bp, 0);
1160 * Same as xfs_bioerror, except that we are releasing the buffer
1161 * here ourselves, and avoiding the xfs_buf_ioend call.
1162 * This is meant for userdata errors; metadata bufs come with
1163 * iodone functions attached, so that we can track down errors.
1169 int64_t fl = XFS_BUF_BFLAGS(bp);
1171 * No need to wait until the buffer is unpinned.
1172 * We aren't flushing it.
1174 * chunkhold expects B_DONE to be set, whether
1175 * we actually finish the I/O or not. We don't want to
1176 * change that interface.
1179 XFS_BUF_UNDELAYWRITE(bp);
1182 XFS_BUF_CLR_IODONE_FUNC(bp);
1183 if (!(fl & XBF_ASYNC)) {
1185 * Mark b_error and B_ERROR _both_.
1186 * Lot's of chunkcache code assumes that.
1187 * There's no reason to mark error for
1190 XFS_BUF_ERROR(bp, EIO);
1191 XFS_BUF_FINISH_IOWAIT(bp);
1201 * All xfs metadata buffers except log state machine buffers
1202 * get this attached as their b_bdstrat callback function.
1203 * This is so that we can catch a buffer
1204 * after prematurely unpinning it to forcibly shutdown the filesystem.
1210 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1211 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1213 * Metadata write that didn't get logged but
1214 * written delayed anyway. These aren't associated
1215 * with a transaction, and can be ignored.
1217 if (!bp->b_iodone && !XFS_BUF_ISREAD(bp))
1218 return xfs_bioerror_relse(bp);
1220 return xfs_bioerror(bp);
1223 xfs_buf_iorequest(bp);
1228 * Wrapper around bdstrat so that we can stop data from going to disk in case
1229 * we are shutting down the filesystem. Typically user data goes thru this
1230 * path; one of the exceptions is the superblock.
1234 struct xfs_mount *mp,
1237 if (XFS_FORCED_SHUTDOWN(mp)) {
1238 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1239 xfs_bioerror_relse(bp);
1243 xfs_buf_iorequest(bp);
1251 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1252 bp->b_flags &= ~_XBF_PAGE_LOCKED;
1253 xfs_buf_ioend(bp, schedule);
1262 xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private;
1263 unsigned int blocksize = bp->b_target->bt_bsize;
1264 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1266 xfs_buf_ioerror(bp, -error);
1268 if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1269 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1272 struct page *page = bvec->bv_page;
1274 ASSERT(!PagePrivate(page));
1275 if (unlikely(bp->b_error)) {
1276 if (bp->b_flags & XBF_READ)
1277 ClearPageUptodate(page);
1278 } else if (blocksize >= PAGE_CACHE_SIZE) {
1279 SetPageUptodate(page);
1280 } else if (!PagePrivate(page) &&
1281 (bp->b_flags & _XBF_PAGE_CACHE)) {
1282 set_page_region(page, bvec->bv_offset, bvec->bv_len);
1285 if (--bvec >= bio->bi_io_vec)
1286 prefetchw(&bvec->bv_page->flags);
1288 if (bp->b_flags & _XBF_PAGE_LOCKED)
1290 } while (bvec >= bio->bi_io_vec);
1292 _xfs_buf_ioend(bp, 1);
1300 int rw, map_i, total_nr_pages, nr_pages;
1302 int offset = bp->b_offset;
1303 int size = bp->b_count_desired;
1304 sector_t sector = bp->b_bn;
1305 unsigned int blocksize = bp->b_target->bt_bsize;
1307 total_nr_pages = bp->b_page_count;
1310 if (bp->b_flags & XBF_ORDERED) {
1311 ASSERT(!(bp->b_flags & XBF_READ));
1312 rw = WRITE_FLUSH_FUA;
1313 } else if (bp->b_flags & XBF_LOG_BUFFER) {
1314 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1315 bp->b_flags &= ~_XBF_RUN_QUEUES;
1316 rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
1317 } else if (bp->b_flags & _XBF_RUN_QUEUES) {
1318 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1319 bp->b_flags &= ~_XBF_RUN_QUEUES;
1320 rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
1322 rw = (bp->b_flags & XBF_WRITE) ? WRITE :
1323 (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;
1326 /* Special code path for reading a sub page size buffer in --
1327 * we populate up the whole page, and hence the other metadata
1328 * in the same page. This optimization is only valid when the
1329 * filesystem block size is not smaller than the page size.
1331 if ((bp->b_buffer_length < PAGE_CACHE_SIZE) &&
1332 ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) ==
1333 (XBF_READ|_XBF_PAGE_LOCKED)) &&
1334 (blocksize >= PAGE_CACHE_SIZE)) {
1335 bio = bio_alloc(GFP_NOIO, 1);
1337 bio->bi_bdev = bp->b_target->bt_bdev;
1338 bio->bi_sector = sector - (offset >> BBSHIFT);
1339 bio->bi_end_io = xfs_buf_bio_end_io;
1340 bio->bi_private = bp;
1342 bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0);
1345 atomic_inc(&bp->b_io_remaining);
1351 atomic_inc(&bp->b_io_remaining);
1352 nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1353 if (nr_pages > total_nr_pages)
1354 nr_pages = total_nr_pages;
1356 bio = bio_alloc(GFP_NOIO, nr_pages);
1357 bio->bi_bdev = bp->b_target->bt_bdev;
1358 bio->bi_sector = sector;
1359 bio->bi_end_io = xfs_buf_bio_end_io;
1360 bio->bi_private = bp;
1362 for (; size && nr_pages; nr_pages--, map_i++) {
1363 int rbytes, nbytes = PAGE_CACHE_SIZE - offset;
1368 rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset);
1369 if (rbytes < nbytes)
1373 sector += nbytes >> BBSHIFT;
1379 if (likely(bio->bi_size)) {
1380 if (xfs_buf_is_vmapped(bp)) {
1381 flush_kernel_vmap_range(bp->b_addr,
1382 xfs_buf_vmap_len(bp));
1384 submit_bio(rw, bio);
1389 * if we get here, no pages were added to the bio. However,
1390 * we can't just error out here - if the pages are locked then
1391 * we have to unlock them otherwise we can hang on a later
1392 * access to the page.
1394 xfs_buf_ioerror(bp, EIO);
1395 if (bp->b_flags & _XBF_PAGE_LOCKED) {
1397 for (i = 0; i < bp->b_page_count; i++)
1398 unlock_page(bp->b_pages[i]);
1408 trace_xfs_buf_iorequest(bp, _RET_IP_);
1410 if (bp->b_flags & XBF_DELWRI) {
1411 xfs_buf_delwri_queue(bp, 1);
1415 if (bp->b_flags & XBF_WRITE) {
1416 xfs_buf_wait_unpin(bp);
1421 /* Set the count to 1 initially, this will stop an I/O
1422 * completion callout which happens before we have started
1423 * all the I/O from calling xfs_buf_ioend too early.
1425 atomic_set(&bp->b_io_remaining, 1);
1426 _xfs_buf_ioapply(bp);
1427 _xfs_buf_ioend(bp, 0);
1434 * Waits for I/O to complete on the buffer supplied.
1435 * It returns immediately if no I/O is pending.
1436 * It returns the I/O error code, if any, or 0 if there was no error.
1442 trace_xfs_buf_iowait(bp, _RET_IP_);
1444 if (atomic_read(&bp->b_io_remaining))
1445 blk_run_address_space(bp->b_target->bt_mapping);
1446 wait_for_completion(&bp->b_iowait);
1448 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1459 if (bp->b_flags & XBF_MAPPED)
1460 return XFS_BUF_PTR(bp) + offset;
1462 offset += bp->b_offset;
1463 page = bp->b_pages[offset >> PAGE_CACHE_SHIFT];
1464 return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1));
1468 * Move data into or out of a buffer.
1472 xfs_buf_t *bp, /* buffer to process */
1473 size_t boff, /* starting buffer offset */
1474 size_t bsize, /* length to copy */
1475 void *data, /* data address */
1476 xfs_buf_rw_t mode) /* read/write/zero flag */
1478 size_t bend, cpoff, csize;
1481 bend = boff + bsize;
1482 while (boff < bend) {
1483 page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)];
1484 cpoff = xfs_buf_poff(boff + bp->b_offset);
1485 csize = min_t(size_t,
1486 PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff);
1488 ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE));
1492 memset(page_address(page) + cpoff, 0, csize);
1495 memcpy(data, page_address(page) + cpoff, csize);
1498 memcpy(page_address(page) + cpoff, data, csize);
1507 * Handling of buffer targets (buftargs).
1511 * Wait for any bufs with callbacks that have been submitted but have not yet
1512 * returned. These buffers will have an elevated hold count, so wait on those
1513 * while freeing all the buffers only held by the LRU.
1517 struct xfs_buftarg *btp)
1522 spin_lock(&btp->bt_lru_lock);
1523 while (!list_empty(&btp->bt_lru)) {
1524 bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru);
1525 if (atomic_read(&bp->b_hold) > 1) {
1526 spin_unlock(&btp->bt_lru_lock);
1531 * clear the LRU reference count so the bufer doesn't get
1532 * ignored in xfs_buf_rele().
1534 atomic_set(&bp->b_lru_ref, 0);
1535 spin_unlock(&btp->bt_lru_lock);
1537 spin_lock(&btp->bt_lru_lock);
1539 spin_unlock(&btp->bt_lru_lock);
1544 struct shrinker *shrink,
1548 struct xfs_buftarg *btp = container_of(shrink,
1549 struct xfs_buftarg, bt_shrinker);
1554 return btp->bt_lru_nr;
1556 spin_lock(&btp->bt_lru_lock);
1557 while (!list_empty(&btp->bt_lru)) {
1558 if (nr_to_scan-- <= 0)
1561 bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru);
1564 * Decrement the b_lru_ref count unless the value is already
1565 * zero. If the value is already zero, we need to reclaim the
1566 * buffer, otherwise it gets another trip through the LRU.
1568 if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1569 list_move_tail(&bp->b_lru, &btp->bt_lru);
1574 * remove the buffer from the LRU now to avoid needing another
1575 * lock round trip inside xfs_buf_rele().
1577 list_move(&bp->b_lru, &dispose);
1580 spin_unlock(&btp->bt_lru_lock);
1582 while (!list_empty(&dispose)) {
1583 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1584 list_del_init(&bp->b_lru);
1588 return btp->bt_lru_nr;
1593 struct xfs_mount *mp,
1594 struct xfs_buftarg *btp)
1596 unregister_shrinker(&btp->bt_shrinker);
1598 xfs_flush_buftarg(btp, 1);
1599 if (mp->m_flags & XFS_MOUNT_BARRIER)
1600 xfs_blkdev_issue_flush(btp);
1601 iput(btp->bt_mapping->host);
1603 kthread_stop(btp->bt_task);
1608 xfs_setsize_buftarg_flags(
1610 unsigned int blocksize,
1611 unsigned int sectorsize,
1614 btp->bt_bsize = blocksize;
1615 btp->bt_sshift = ffs(sectorsize) - 1;
1616 btp->bt_smask = sectorsize - 1;
1618 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1619 xfs_warn(btp->bt_mount,
1620 "Cannot set_blocksize to %u on device %s\n",
1621 sectorsize, XFS_BUFTARG_NAME(btp));
1626 (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) {
1628 "XFS: %u byte sectors in use on device %s. "
1629 "This is suboptimal; %u or greater is ideal.\n",
1630 sectorsize, XFS_BUFTARG_NAME(btp),
1631 (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG);
1638 * When allocating the initial buffer target we have not yet
1639 * read in the superblock, so don't know what sized sectors
1640 * are being used is at this early stage. Play safe.
1643 xfs_setsize_buftarg_early(
1645 struct block_device *bdev)
1647 return xfs_setsize_buftarg_flags(btp,
1648 PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0);
1652 xfs_setsize_buftarg(
1654 unsigned int blocksize,
1655 unsigned int sectorsize)
1657 return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1);
1661 xfs_mapping_buftarg(
1663 struct block_device *bdev)
1665 struct backing_dev_info *bdi;
1666 struct inode *inode;
1667 struct address_space *mapping;
1668 static const struct address_space_operations mapping_aops = {
1669 .sync_page = block_sync_page,
1670 .migratepage = fail_migrate_page,
1673 inode = new_inode(bdev->bd_inode->i_sb);
1676 "XFS: Cannot allocate mapping inode for device %s\n",
1677 XFS_BUFTARG_NAME(btp));
1680 inode->i_ino = get_next_ino();
1681 inode->i_mode = S_IFBLK;
1682 inode->i_bdev = bdev;
1683 inode->i_rdev = bdev->bd_dev;
1684 bdi = blk_get_backing_dev_info(bdev);
1686 bdi = &default_backing_dev_info;
1687 mapping = &inode->i_data;
1688 mapping->a_ops = &mapping_aops;
1689 mapping->backing_dev_info = bdi;
1690 mapping_set_gfp_mask(mapping, GFP_NOFS);
1691 btp->bt_mapping = mapping;
1696 xfs_alloc_delwrite_queue(
1700 INIT_LIST_HEAD(&btp->bt_delwrite_queue);
1701 spin_lock_init(&btp->bt_delwrite_lock);
1703 btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname);
1704 if (IS_ERR(btp->bt_task))
1705 return PTR_ERR(btp->bt_task);
1711 struct xfs_mount *mp,
1712 struct block_device *bdev,
1718 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP);
1721 btp->bt_dev = bdev->bd_dev;
1722 btp->bt_bdev = bdev;
1723 INIT_LIST_HEAD(&btp->bt_lru);
1724 spin_lock_init(&btp->bt_lru_lock);
1725 if (xfs_setsize_buftarg_early(btp, bdev))
1727 if (xfs_mapping_buftarg(btp, bdev))
1729 if (xfs_alloc_delwrite_queue(btp, fsname))
1731 btp->bt_shrinker.shrink = xfs_buftarg_shrink;
1732 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1733 register_shrinker(&btp->bt_shrinker);
1743 * Delayed write buffer handling
1746 xfs_buf_delwri_queue(
1750 struct list_head *dwq = &bp->b_target->bt_delwrite_queue;
1751 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1753 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1755 ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC));
1758 /* If already in the queue, dequeue and place at tail */
1759 if (!list_empty(&bp->b_list)) {
1760 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1762 atomic_dec(&bp->b_hold);
1763 list_del(&bp->b_list);
1766 if (list_empty(dwq)) {
1767 /* start xfsbufd as it is about to have something to do */
1768 wake_up_process(bp->b_target->bt_task);
1771 bp->b_flags |= _XBF_DELWRI_Q;
1772 list_add_tail(&bp->b_list, dwq);
1773 bp->b_queuetime = jiffies;
1781 xfs_buf_delwri_dequeue(
1784 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1788 if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) {
1789 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1790 list_del_init(&bp->b_list);
1793 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q);
1799 trace_xfs_buf_delwri_dequeue(bp, _RET_IP_);
1803 * If a delwri buffer needs to be pushed before it has aged out, then promote
1804 * it to the head of the delwri queue so that it will be flushed on the next
1805 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1806 * than the age currently needed to flush the buffer. Hence the next time the
1807 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1810 xfs_buf_delwri_promote(
1813 struct xfs_buftarg *btp = bp->b_target;
1814 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1;
1816 ASSERT(bp->b_flags & XBF_DELWRI);
1817 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1820 * Check the buffer age before locking the delayed write queue as we
1821 * don't need to promote buffers that are already past the flush age.
1823 if (bp->b_queuetime < jiffies - age)
1825 bp->b_queuetime = jiffies - age;
1826 spin_lock(&btp->bt_delwrite_lock);
1827 list_move(&bp->b_list, &btp->bt_delwrite_queue);
1828 spin_unlock(&btp->bt_delwrite_lock);
1832 xfs_buf_runall_queues(
1833 struct workqueue_struct *queue)
1835 flush_workqueue(queue);
1839 * Move as many buffers as specified to the supplied list
1840 * idicating if we skipped any buffers to prevent deadlocks.
1843 xfs_buf_delwri_split(
1844 xfs_buftarg_t *target,
1845 struct list_head *list,
1849 struct list_head *dwq = &target->bt_delwrite_queue;
1850 spinlock_t *dwlk = &target->bt_delwrite_lock;
1854 force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1855 INIT_LIST_HEAD(list);
1857 list_for_each_entry_safe(bp, n, dwq, b_list) {
1858 ASSERT(bp->b_flags & XBF_DELWRI);
1860 if (!XFS_BUF_ISPINNED(bp) && !xfs_buf_cond_lock(bp)) {
1862 time_before(jiffies, bp->b_queuetime + age)) {
1867 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q|
1869 bp->b_flags |= XBF_WRITE;
1870 list_move_tail(&bp->b_list, list);
1871 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1882 * Compare function is more complex than it needs to be because
1883 * the return value is only 32 bits and we are doing comparisons
1889 struct list_head *a,
1890 struct list_head *b)
1892 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1893 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1896 diff = ap->b_bn - bp->b_bn;
1905 xfs_buf_delwri_sort(
1906 xfs_buftarg_t *target,
1907 struct list_head *list)
1909 list_sort(NULL, list, xfs_buf_cmp);
1916 xfs_buftarg_t *target = (xfs_buftarg_t *)data;
1918 current->flags |= PF_MEMALLOC;
1923 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10);
1924 long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10);
1926 struct list_head tmp;
1928 if (unlikely(freezing(current))) {
1929 set_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1932 clear_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1935 /* sleep for a long time if there is nothing to do. */
1936 if (list_empty(&target->bt_delwrite_queue))
1937 tout = MAX_SCHEDULE_TIMEOUT;
1938 schedule_timeout_interruptible(tout);
1940 xfs_buf_delwri_split(target, &tmp, age);
1941 list_sort(NULL, &tmp, xfs_buf_cmp);
1942 while (!list_empty(&tmp)) {
1944 bp = list_first_entry(&tmp, struct xfs_buf, b_list);
1945 list_del_init(&bp->b_list);
1950 blk_run_address_space(target->bt_mapping);
1952 } while (!kthread_should_stop());
1958 * Go through all incore buffers, and release buffers if they belong to
1959 * the given device. This is used in filesystem error handling to
1960 * preserve the consistency of its metadata.
1964 xfs_buftarg_t *target,
1969 LIST_HEAD(tmp_list);
1970 LIST_HEAD(wait_list);
1972 xfs_buf_runall_queues(xfsconvertd_workqueue);
1973 xfs_buf_runall_queues(xfsdatad_workqueue);
1974 xfs_buf_runall_queues(xfslogd_workqueue);
1976 set_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1977 pincount = xfs_buf_delwri_split(target, &tmp_list, 0);
1980 * Dropped the delayed write list lock, now walk the temporary list.
1981 * All I/O is issued async and then if we need to wait for completion
1982 * we do that after issuing all the IO.
1984 list_sort(NULL, &tmp_list, xfs_buf_cmp);
1985 while (!list_empty(&tmp_list)) {
1986 bp = list_first_entry(&tmp_list, struct xfs_buf, b_list);
1987 ASSERT(target == bp->b_target);
1988 list_del_init(&bp->b_list);
1990 bp->b_flags &= ~XBF_ASYNC;
1991 list_add(&bp->b_list, &wait_list);
1997 /* Expedite and wait for IO to complete. */
1998 blk_run_address_space(target->bt_mapping);
1999 while (!list_empty(&wait_list)) {
2000 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
2002 list_del_init(&bp->b_list);
2014 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
2015 KM_ZONE_HWALIGN, NULL);
2019 xfslogd_workqueue = alloc_workqueue("xfslogd",
2020 WQ_MEM_RECLAIM | WQ_HIGHPRI, 1);
2021 if (!xfslogd_workqueue)
2022 goto out_free_buf_zone;
2024 xfsdatad_workqueue = alloc_workqueue("xfsdatad", WQ_MEM_RECLAIM, 1);
2025 if (!xfsdatad_workqueue)
2026 goto out_destroy_xfslogd_workqueue;
2028 xfsconvertd_workqueue = alloc_workqueue("xfsconvertd",
2030 if (!xfsconvertd_workqueue)
2031 goto out_destroy_xfsdatad_workqueue;
2035 out_destroy_xfsdatad_workqueue:
2036 destroy_workqueue(xfsdatad_workqueue);
2037 out_destroy_xfslogd_workqueue:
2038 destroy_workqueue(xfslogd_workqueue);
2040 kmem_zone_destroy(xfs_buf_zone);
2046 xfs_buf_terminate(void)
2048 destroy_workqueue(xfsconvertd_workqueue);
2049 destroy_workqueue(xfsdatad_workqueue);
2050 destroy_workqueue(xfslogd_workqueue);
2051 kmem_zone_destroy(xfs_buf_zone);
2054 #ifdef CONFIG_KDB_MODULES
2056 xfs_get_buftarg_list(void)
2058 return &xfs_buftarg_list;