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 "XFS: possible memory allocation "
406 "deadlock in %s (mode:0x%x)\n",
409 XFS_STATS_INC(xb_page_retries);
410 congestion_wait(BLK_RW_ASYNC, HZ/50);
414 XFS_STATS_INC(xb_page_found);
416 nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset);
419 ASSERT(!PagePrivate(page));
420 if (!PageUptodate(page)) {
422 if (blocksize >= PAGE_CACHE_SIZE) {
423 if (flags & XBF_READ)
424 bp->b_flags |= _XBF_PAGE_LOCKED;
425 } else if (!PagePrivate(page)) {
426 if (test_page_region(page, offset, nbytes))
431 bp->b_pages[i] = page;
435 if (!(bp->b_flags & _XBF_PAGE_LOCKED)) {
436 for (i = 0; i < bp->b_page_count; i++)
437 unlock_page(bp->b_pages[i]);
440 if (page_count == bp->b_page_count)
441 bp->b_flags |= XBF_DONE;
447 * Map buffer into kernel address-space if nessecary.
454 /* A single page buffer is always mappable */
455 if (bp->b_page_count == 1) {
456 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
457 bp->b_flags |= XBF_MAPPED;
458 } else if (flags & XBF_MAPPED) {
459 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
461 if (unlikely(bp->b_addr == NULL))
463 bp->b_addr += bp->b_offset;
464 bp->b_flags |= XBF_MAPPED;
471 * Finding and Reading Buffers
475 * Look up, and creates if absent, a lockable buffer for
476 * a given range of an inode. The buffer is returned
477 * locked. If other overlapping buffers exist, they are
478 * released before the new buffer is created and locked,
479 * which may imply that this call will block until those buffers
480 * are unlocked. No I/O is implied by this call.
484 xfs_buftarg_t *btp, /* block device target */
485 xfs_off_t ioff, /* starting offset of range */
486 size_t isize, /* length of range */
487 xfs_buf_flags_t flags,
490 xfs_off_t range_base;
492 struct xfs_perag *pag;
493 struct rb_node **rbp;
494 struct rb_node *parent;
497 range_base = (ioff << BBSHIFT);
498 range_length = (isize << BBSHIFT);
500 /* Check for IOs smaller than the sector size / not sector aligned */
501 ASSERT(!(range_length < (1 << btp->bt_sshift)));
502 ASSERT(!(range_base & (xfs_off_t)btp->bt_smask));
505 pag = xfs_perag_get(btp->bt_mount,
506 xfs_daddr_to_agno(btp->bt_mount, ioff));
509 spin_lock(&pag->pag_buf_lock);
510 rbp = &pag->pag_buf_tree.rb_node;
515 bp = rb_entry(parent, struct xfs_buf, b_rbnode);
517 if (range_base < bp->b_file_offset)
518 rbp = &(*rbp)->rb_left;
519 else if (range_base > bp->b_file_offset)
520 rbp = &(*rbp)->rb_right;
523 * found a block offset match. If the range doesn't
524 * match, the only way this is allowed is if the buffer
525 * in the cache is stale and the transaction that made
526 * it stale has not yet committed. i.e. we are
527 * reallocating a busy extent. Skip this buffer and
528 * continue searching to the right for an exact match.
530 if (bp->b_buffer_length != range_length) {
531 ASSERT(bp->b_flags & XBF_STALE);
532 rbp = &(*rbp)->rb_right;
535 atomic_inc(&bp->b_hold);
542 _xfs_buf_initialize(new_bp, btp, range_base,
543 range_length, flags);
544 rb_link_node(&new_bp->b_rbnode, parent, rbp);
545 rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree);
546 /* the buffer keeps the perag reference until it is freed */
548 spin_unlock(&pag->pag_buf_lock);
550 XFS_STATS_INC(xb_miss_locked);
551 spin_unlock(&pag->pag_buf_lock);
557 spin_unlock(&pag->pag_buf_lock);
560 if (xfs_buf_cond_lock(bp)) {
561 /* failed, so wait for the lock if requested. */
562 if (!(flags & XBF_TRYLOCK)) {
564 XFS_STATS_INC(xb_get_locked_waited);
567 XFS_STATS_INC(xb_busy_locked);
572 if (bp->b_flags & XBF_STALE) {
573 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
574 bp->b_flags &= XBF_MAPPED;
577 trace_xfs_buf_find(bp, flags, _RET_IP_);
578 XFS_STATS_INC(xb_get_locked);
583 * Assembles a buffer covering the specified range.
584 * Storage in memory for all portions of the buffer will be allocated,
585 * although backing storage may not be.
589 xfs_buftarg_t *target,/* target for buffer */
590 xfs_off_t ioff, /* starting offset of range */
591 size_t isize, /* length of range */
592 xfs_buf_flags_t flags)
594 xfs_buf_t *bp, *new_bp;
597 new_bp = xfs_buf_allocate(flags);
598 if (unlikely(!new_bp))
601 bp = _xfs_buf_find(target, ioff, isize, flags, new_bp);
603 error = _xfs_buf_lookup_pages(bp, flags);
607 xfs_buf_deallocate(new_bp);
608 if (unlikely(bp == NULL))
612 for (i = 0; i < bp->b_page_count; i++)
613 mark_page_accessed(bp->b_pages[i]);
615 if (!(bp->b_flags & XBF_MAPPED)) {
616 error = _xfs_buf_map_pages(bp, flags);
617 if (unlikely(error)) {
618 printk(KERN_WARNING "%s: failed to map pages\n",
624 XFS_STATS_INC(xb_get);
627 * Always fill in the block number now, the mapped cases can do
628 * their own overlay of this later.
631 bp->b_count_desired = bp->b_buffer_length;
633 trace_xfs_buf_get(bp, flags, _RET_IP_);
637 if (flags & (XBF_LOCK | XBF_TRYLOCK))
646 xfs_buf_flags_t flags)
650 ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE)));
651 ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL);
653 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \
654 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
655 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \
656 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
658 status = xfs_buf_iorequest(bp);
659 if (status || XFS_BUF_ISERROR(bp) || (flags & XBF_ASYNC))
661 return xfs_buf_iowait(bp);
666 xfs_buftarg_t *target,
669 xfs_buf_flags_t flags)
675 bp = xfs_buf_get(target, ioff, isize, flags);
677 trace_xfs_buf_read(bp, flags, _RET_IP_);
679 if (!XFS_BUF_ISDONE(bp)) {
680 XFS_STATS_INC(xb_get_read);
681 _xfs_buf_read(bp, flags);
682 } else if (flags & XBF_ASYNC) {
684 * Read ahead call which is already satisfied,
689 /* We do not want read in the flags */
690 bp->b_flags &= ~XBF_READ;
697 if (flags & (XBF_LOCK | XBF_TRYLOCK))
704 * If we are not low on memory then do the readahead in a deadlock
709 xfs_buftarg_t *target,
713 struct backing_dev_info *bdi;
715 bdi = target->bt_mapping->backing_dev_info;
716 if (bdi_read_congested(bdi))
719 xfs_buf_read(target, ioff, isize,
720 XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK);
724 * Read an uncached buffer from disk. Allocates and returns a locked
725 * buffer containing the disk contents or nothing.
728 xfs_buf_read_uncached(
729 struct xfs_mount *mp,
730 struct xfs_buftarg *target,
738 bp = xfs_buf_get_uncached(target, length, flags);
742 /* set up the buffer for a read IO */
744 XFS_BUF_SET_ADDR(bp, daddr);
749 error = xfs_buf_iowait(bp);
750 if (error || bp->b_error) {
760 xfs_buftarg_t *target)
764 bp = xfs_buf_allocate(0);
766 _xfs_buf_initialize(bp, target, 0, len, 0);
770 static inline struct page *
774 if ((!is_vmalloc_addr(addr))) {
775 return virt_to_page(addr);
777 return vmalloc_to_page(addr);
782 xfs_buf_associate_memory(
789 unsigned long pageaddr;
790 unsigned long offset;
794 pageaddr = (unsigned long)mem & PAGE_CACHE_MASK;
795 offset = (unsigned long)mem - pageaddr;
796 buflen = PAGE_CACHE_ALIGN(len + offset);
797 page_count = buflen >> PAGE_CACHE_SHIFT;
799 /* Free any previous set of page pointers */
801 _xfs_buf_free_pages(bp);
806 rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK);
810 bp->b_offset = offset;
812 for (i = 0; i < bp->b_page_count; i++) {
813 bp->b_pages[i] = mem_to_page((void *)pageaddr);
814 pageaddr += PAGE_CACHE_SIZE;
817 bp->b_count_desired = len;
818 bp->b_buffer_length = buflen;
819 bp->b_flags |= XBF_MAPPED;
820 bp->b_flags &= ~_XBF_PAGE_LOCKED;
826 xfs_buf_get_uncached(
827 struct xfs_buftarg *target,
831 unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT;
835 bp = xfs_buf_allocate(0);
836 if (unlikely(bp == NULL))
838 _xfs_buf_initialize(bp, target, 0, len, 0);
840 error = _xfs_buf_get_pages(bp, page_count, 0);
844 for (i = 0; i < page_count; i++) {
845 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
849 bp->b_flags |= _XBF_PAGES;
851 error = _xfs_buf_map_pages(bp, XBF_MAPPED);
852 if (unlikely(error)) {
853 printk(KERN_WARNING "%s: failed to map pages\n",
860 trace_xfs_buf_get_uncached(bp, _RET_IP_);
865 __free_page(bp->b_pages[i]);
866 _xfs_buf_free_pages(bp);
868 xfs_buf_deallocate(bp);
874 * Increment reference count on buffer, to hold the buffer concurrently
875 * with another thread which may release (free) the buffer asynchronously.
876 * Must hold the buffer already to call this function.
882 trace_xfs_buf_hold(bp, _RET_IP_);
883 atomic_inc(&bp->b_hold);
887 * Releases a hold on the specified buffer. If the
888 * the hold count is 1, calls xfs_buf_free.
894 struct xfs_perag *pag = bp->b_pag;
896 trace_xfs_buf_rele(bp, _RET_IP_);
899 ASSERT(!bp->b_relse);
900 ASSERT(list_empty(&bp->b_lru));
901 ASSERT(RB_EMPTY_NODE(&bp->b_rbnode));
902 if (atomic_dec_and_test(&bp->b_hold))
907 ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode));
909 ASSERT(atomic_read(&bp->b_hold) > 0);
910 if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) {
912 atomic_inc(&bp->b_hold);
913 spin_unlock(&pag->pag_buf_lock);
915 } else if (!(bp->b_flags & XBF_STALE) &&
916 atomic_read(&bp->b_lru_ref)) {
918 spin_unlock(&pag->pag_buf_lock);
921 ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)));
922 rb_erase(&bp->b_rbnode, &pag->pag_buf_tree);
923 spin_unlock(&pag->pag_buf_lock);
932 * Mutual exclusion on buffers. Locking model:
934 * Buffers associated with inodes for which buffer locking
935 * is not enabled are not protected by semaphores, and are
936 * assumed to be exclusively owned by the caller. There is a
937 * spinlock in the buffer, used by the caller when concurrent
938 * access is possible.
942 * Locks a buffer object, if it is not already locked. Note that this in
943 * no way locks the underlying pages, so it is only useful for
944 * synchronizing concurrent use of buffer objects, not for synchronizing
945 * independent access to the underlying pages.
947 * If we come across a stale, pinned, locked buffer, we know that we are
948 * being asked to lock a buffer that has been reallocated. Because it is
949 * pinned, we know that the log has not been pushed to disk and hence it
950 * will still be locked. Rather than continuing to have trylock attempts
951 * fail until someone else pushes the log, push it ourselves before
952 * returning. This means that the xfsaild will not get stuck trying
953 * to push on stale inode buffers.
961 locked = down_trylock(&bp->b_sema) == 0;
964 else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
965 xfs_log_force(bp->b_target->bt_mount, 0);
967 trace_xfs_buf_cond_lock(bp, _RET_IP_);
968 return locked ? 0 : -EBUSY;
975 return bp->b_sema.count;
979 * Locks a buffer object.
980 * Note that this in no way locks the underlying pages, so it is only
981 * useful for synchronizing concurrent use of buffer objects, not for
982 * synchronizing independent access to the underlying pages.
984 * If we come across a stale, pinned, locked buffer, we know that we
985 * are being asked to lock a buffer that has been reallocated. Because
986 * it is pinned, we know that the log has not been pushed to disk and
987 * hence it will still be locked. Rather than sleeping until someone
988 * else pushes the log, push it ourselves before trying to get the lock.
994 trace_xfs_buf_lock(bp, _RET_IP_);
996 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
997 xfs_log_force(bp->b_target->bt_mount, 0);
998 if (atomic_read(&bp->b_io_remaining))
999 blk_run_address_space(bp->b_target->bt_mapping);
1003 trace_xfs_buf_lock_done(bp, _RET_IP_);
1007 * Releases the lock on the buffer object.
1008 * If the buffer is marked delwri but is not queued, do so before we
1009 * unlock the buffer as we need to set flags correctly. We also need to
1010 * take a reference for the delwri queue because the unlocker is going to
1011 * drop their's and they don't know we just queued it.
1017 if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) {
1018 atomic_inc(&bp->b_hold);
1019 bp->b_flags |= XBF_ASYNC;
1020 xfs_buf_delwri_queue(bp, 0);
1026 trace_xfs_buf_unlock(bp, _RET_IP_);
1033 DECLARE_WAITQUEUE (wait, current);
1035 if (atomic_read(&bp->b_pin_count) == 0)
1038 add_wait_queue(&bp->b_waiters, &wait);
1040 set_current_state(TASK_UNINTERRUPTIBLE);
1041 if (atomic_read(&bp->b_pin_count) == 0)
1043 if (atomic_read(&bp->b_io_remaining))
1044 blk_run_address_space(bp->b_target->bt_mapping);
1047 remove_wait_queue(&bp->b_waiters, &wait);
1048 set_current_state(TASK_RUNNING);
1052 * Buffer Utility Routines
1056 xfs_buf_iodone_work(
1057 struct work_struct *work)
1060 container_of(work, xfs_buf_t, b_iodone_work);
1063 (*(bp->b_iodone))(bp);
1064 else if (bp->b_flags & XBF_ASYNC)
1073 trace_xfs_buf_iodone(bp, _RET_IP_);
1075 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1076 if (bp->b_error == 0)
1077 bp->b_flags |= XBF_DONE;
1079 if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) {
1081 INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work);
1082 queue_work(xfslogd_workqueue, &bp->b_iodone_work);
1084 xfs_buf_iodone_work(&bp->b_iodone_work);
1087 complete(&bp->b_iowait);
1096 ASSERT(error >= 0 && error <= 0xffff);
1097 bp->b_error = (unsigned short)error;
1098 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
1103 struct xfs_mount *mp,
1108 bp->b_flags |= XBF_WRITE;
1109 bp->b_flags &= ~(XBF_ASYNC | XBF_READ);
1111 xfs_buf_delwri_dequeue(bp);
1114 error = xfs_buf_iowait(bp);
1116 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1126 trace_xfs_buf_bdwrite(bp, _RET_IP_);
1128 bp->b_flags &= ~XBF_READ;
1129 bp->b_flags |= (XBF_DELWRI | XBF_ASYNC);
1131 xfs_buf_delwri_queue(bp, 1);
1135 * Called when we want to stop a buffer from getting written or read.
1136 * We attach the EIO error, muck with its flags, and call xfs_buf_ioend
1137 * so that the proper iodone callbacks get called.
1143 #ifdef XFSERRORDEBUG
1144 ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
1148 * No need to wait until the buffer is unpinned, we aren't flushing it.
1150 XFS_BUF_ERROR(bp, EIO);
1153 * We're calling xfs_buf_ioend, so delete XBF_DONE flag.
1156 XFS_BUF_UNDELAYWRITE(bp);
1160 xfs_buf_ioend(bp, 0);
1166 * Same as xfs_bioerror, except that we are releasing the buffer
1167 * here ourselves, and avoiding the xfs_buf_ioend call.
1168 * This is meant for userdata errors; metadata bufs come with
1169 * iodone functions attached, so that we can track down errors.
1175 int64_t fl = XFS_BUF_BFLAGS(bp);
1177 * No need to wait until the buffer is unpinned.
1178 * We aren't flushing it.
1180 * chunkhold expects B_DONE to be set, whether
1181 * we actually finish the I/O or not. We don't want to
1182 * change that interface.
1185 XFS_BUF_UNDELAYWRITE(bp);
1188 XFS_BUF_CLR_IODONE_FUNC(bp);
1189 if (!(fl & XBF_ASYNC)) {
1191 * Mark b_error and B_ERROR _both_.
1192 * Lot's of chunkcache code assumes that.
1193 * There's no reason to mark error for
1196 XFS_BUF_ERROR(bp, EIO);
1197 XFS_BUF_FINISH_IOWAIT(bp);
1207 * All xfs metadata buffers except log state machine buffers
1208 * get this attached as their b_bdstrat callback function.
1209 * This is so that we can catch a buffer
1210 * after prematurely unpinning it to forcibly shutdown the filesystem.
1216 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1217 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1219 * Metadata write that didn't get logged but
1220 * written delayed anyway. These aren't associated
1221 * with a transaction, and can be ignored.
1223 if (!bp->b_iodone && !XFS_BUF_ISREAD(bp))
1224 return xfs_bioerror_relse(bp);
1226 return xfs_bioerror(bp);
1229 xfs_buf_iorequest(bp);
1234 * Wrapper around bdstrat so that we can stop data from going to disk in case
1235 * we are shutting down the filesystem. Typically user data goes thru this
1236 * path; one of the exceptions is the superblock.
1240 struct xfs_mount *mp,
1243 if (XFS_FORCED_SHUTDOWN(mp)) {
1244 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1245 xfs_bioerror_relse(bp);
1249 xfs_buf_iorequest(bp);
1257 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1258 bp->b_flags &= ~_XBF_PAGE_LOCKED;
1259 xfs_buf_ioend(bp, schedule);
1268 xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private;
1269 unsigned int blocksize = bp->b_target->bt_bsize;
1270 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1272 xfs_buf_ioerror(bp, -error);
1274 if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1275 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1278 struct page *page = bvec->bv_page;
1280 ASSERT(!PagePrivate(page));
1281 if (unlikely(bp->b_error)) {
1282 if (bp->b_flags & XBF_READ)
1283 ClearPageUptodate(page);
1284 } else if (blocksize >= PAGE_CACHE_SIZE) {
1285 SetPageUptodate(page);
1286 } else if (!PagePrivate(page) &&
1287 (bp->b_flags & _XBF_PAGE_CACHE)) {
1288 set_page_region(page, bvec->bv_offset, bvec->bv_len);
1291 if (--bvec >= bio->bi_io_vec)
1292 prefetchw(&bvec->bv_page->flags);
1294 if (bp->b_flags & _XBF_PAGE_LOCKED)
1296 } while (bvec >= bio->bi_io_vec);
1298 _xfs_buf_ioend(bp, 1);
1306 int rw, map_i, total_nr_pages, nr_pages;
1308 int offset = bp->b_offset;
1309 int size = bp->b_count_desired;
1310 sector_t sector = bp->b_bn;
1311 unsigned int blocksize = bp->b_target->bt_bsize;
1313 total_nr_pages = bp->b_page_count;
1316 if (bp->b_flags & XBF_ORDERED) {
1317 ASSERT(!(bp->b_flags & XBF_READ));
1318 rw = WRITE_FLUSH_FUA;
1319 } else if (bp->b_flags & XBF_LOG_BUFFER) {
1320 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1321 bp->b_flags &= ~_XBF_RUN_QUEUES;
1322 rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
1323 } else if (bp->b_flags & _XBF_RUN_QUEUES) {
1324 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1325 bp->b_flags &= ~_XBF_RUN_QUEUES;
1326 rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
1328 rw = (bp->b_flags & XBF_WRITE) ? WRITE :
1329 (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;
1332 /* Special code path for reading a sub page size buffer in --
1333 * we populate up the whole page, and hence the other metadata
1334 * in the same page. This optimization is only valid when the
1335 * filesystem block size is not smaller than the page size.
1337 if ((bp->b_buffer_length < PAGE_CACHE_SIZE) &&
1338 ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) ==
1339 (XBF_READ|_XBF_PAGE_LOCKED)) &&
1340 (blocksize >= PAGE_CACHE_SIZE)) {
1341 bio = bio_alloc(GFP_NOIO, 1);
1343 bio->bi_bdev = bp->b_target->bt_bdev;
1344 bio->bi_sector = sector - (offset >> BBSHIFT);
1345 bio->bi_end_io = xfs_buf_bio_end_io;
1346 bio->bi_private = bp;
1348 bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0);
1351 atomic_inc(&bp->b_io_remaining);
1357 atomic_inc(&bp->b_io_remaining);
1358 nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1359 if (nr_pages > total_nr_pages)
1360 nr_pages = total_nr_pages;
1362 bio = bio_alloc(GFP_NOIO, nr_pages);
1363 bio->bi_bdev = bp->b_target->bt_bdev;
1364 bio->bi_sector = sector;
1365 bio->bi_end_io = xfs_buf_bio_end_io;
1366 bio->bi_private = bp;
1368 for (; size && nr_pages; nr_pages--, map_i++) {
1369 int rbytes, nbytes = PAGE_CACHE_SIZE - offset;
1374 rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset);
1375 if (rbytes < nbytes)
1379 sector += nbytes >> BBSHIFT;
1385 if (likely(bio->bi_size)) {
1386 if (xfs_buf_is_vmapped(bp)) {
1387 flush_kernel_vmap_range(bp->b_addr,
1388 xfs_buf_vmap_len(bp));
1390 submit_bio(rw, bio);
1395 * if we get here, no pages were added to the bio. However,
1396 * we can't just error out here - if the pages are locked then
1397 * we have to unlock them otherwise we can hang on a later
1398 * access to the page.
1400 xfs_buf_ioerror(bp, EIO);
1401 if (bp->b_flags & _XBF_PAGE_LOCKED) {
1403 for (i = 0; i < bp->b_page_count; i++)
1404 unlock_page(bp->b_pages[i]);
1414 trace_xfs_buf_iorequest(bp, _RET_IP_);
1416 if (bp->b_flags & XBF_DELWRI) {
1417 xfs_buf_delwri_queue(bp, 1);
1421 if (bp->b_flags & XBF_WRITE) {
1422 xfs_buf_wait_unpin(bp);
1427 /* Set the count to 1 initially, this will stop an I/O
1428 * completion callout which happens before we have started
1429 * all the I/O from calling xfs_buf_ioend too early.
1431 atomic_set(&bp->b_io_remaining, 1);
1432 _xfs_buf_ioapply(bp);
1433 _xfs_buf_ioend(bp, 0);
1440 * Waits for I/O to complete on the buffer supplied.
1441 * It returns immediately if no I/O is pending.
1442 * It returns the I/O error code, if any, or 0 if there was no error.
1448 trace_xfs_buf_iowait(bp, _RET_IP_);
1450 if (atomic_read(&bp->b_io_remaining))
1451 blk_run_address_space(bp->b_target->bt_mapping);
1452 wait_for_completion(&bp->b_iowait);
1454 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1465 if (bp->b_flags & XBF_MAPPED)
1466 return XFS_BUF_PTR(bp) + offset;
1468 offset += bp->b_offset;
1469 page = bp->b_pages[offset >> PAGE_CACHE_SHIFT];
1470 return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1));
1474 * Move data into or out of a buffer.
1478 xfs_buf_t *bp, /* buffer to process */
1479 size_t boff, /* starting buffer offset */
1480 size_t bsize, /* length to copy */
1481 void *data, /* data address */
1482 xfs_buf_rw_t mode) /* read/write/zero flag */
1484 size_t bend, cpoff, csize;
1487 bend = boff + bsize;
1488 while (boff < bend) {
1489 page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)];
1490 cpoff = xfs_buf_poff(boff + bp->b_offset);
1491 csize = min_t(size_t,
1492 PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff);
1494 ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE));
1498 memset(page_address(page) + cpoff, 0, csize);
1501 memcpy(data, page_address(page) + cpoff, csize);
1504 memcpy(page_address(page) + cpoff, data, csize);
1513 * Handling of buffer targets (buftargs).
1517 * Wait for any bufs with callbacks that have been submitted but have not yet
1518 * returned. These buffers will have an elevated hold count, so wait on those
1519 * while freeing all the buffers only held by the LRU.
1523 struct xfs_buftarg *btp)
1528 spin_lock(&btp->bt_lru_lock);
1529 while (!list_empty(&btp->bt_lru)) {
1530 bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru);
1531 if (atomic_read(&bp->b_hold) > 1) {
1532 spin_unlock(&btp->bt_lru_lock);
1537 * clear the LRU reference count so the bufer doesn't get
1538 * ignored in xfs_buf_rele().
1540 atomic_set(&bp->b_lru_ref, 0);
1541 spin_unlock(&btp->bt_lru_lock);
1543 spin_lock(&btp->bt_lru_lock);
1545 spin_unlock(&btp->bt_lru_lock);
1550 struct shrinker *shrink,
1554 struct xfs_buftarg *btp = container_of(shrink,
1555 struct xfs_buftarg, bt_shrinker);
1560 return btp->bt_lru_nr;
1562 spin_lock(&btp->bt_lru_lock);
1563 while (!list_empty(&btp->bt_lru)) {
1564 if (nr_to_scan-- <= 0)
1567 bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru);
1570 * Decrement the b_lru_ref count unless the value is already
1571 * zero. If the value is already zero, we need to reclaim the
1572 * buffer, otherwise it gets another trip through the LRU.
1574 if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1575 list_move_tail(&bp->b_lru, &btp->bt_lru);
1580 * remove the buffer from the LRU now to avoid needing another
1581 * lock round trip inside xfs_buf_rele().
1583 list_move(&bp->b_lru, &dispose);
1586 spin_unlock(&btp->bt_lru_lock);
1588 while (!list_empty(&dispose)) {
1589 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1590 list_del_init(&bp->b_lru);
1594 return btp->bt_lru_nr;
1599 struct xfs_mount *mp,
1600 struct xfs_buftarg *btp)
1602 unregister_shrinker(&btp->bt_shrinker);
1604 xfs_flush_buftarg(btp, 1);
1605 if (mp->m_flags & XFS_MOUNT_BARRIER)
1606 xfs_blkdev_issue_flush(btp);
1607 iput(btp->bt_mapping->host);
1609 kthread_stop(btp->bt_task);
1614 xfs_setsize_buftarg_flags(
1616 unsigned int blocksize,
1617 unsigned int sectorsize,
1620 btp->bt_bsize = blocksize;
1621 btp->bt_sshift = ffs(sectorsize) - 1;
1622 btp->bt_smask = sectorsize - 1;
1624 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1626 "XFS: Cannot set_blocksize to %u on device %s\n",
1627 sectorsize, XFS_BUFTARG_NAME(btp));
1632 (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) {
1634 "XFS: %u byte sectors in use on device %s. "
1635 "This is suboptimal; %u or greater is ideal.\n",
1636 sectorsize, XFS_BUFTARG_NAME(btp),
1637 (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG);
1644 * When allocating the initial buffer target we have not yet
1645 * read in the superblock, so don't know what sized sectors
1646 * are being used is at this early stage. Play safe.
1649 xfs_setsize_buftarg_early(
1651 struct block_device *bdev)
1653 return xfs_setsize_buftarg_flags(btp,
1654 PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0);
1658 xfs_setsize_buftarg(
1660 unsigned int blocksize,
1661 unsigned int sectorsize)
1663 return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1);
1667 xfs_mapping_buftarg(
1669 struct block_device *bdev)
1671 struct backing_dev_info *bdi;
1672 struct inode *inode;
1673 struct address_space *mapping;
1674 static const struct address_space_operations mapping_aops = {
1675 .sync_page = block_sync_page,
1676 .migratepage = fail_migrate_page,
1679 inode = new_inode(bdev->bd_inode->i_sb);
1682 "XFS: Cannot allocate mapping inode for device %s\n",
1683 XFS_BUFTARG_NAME(btp));
1686 inode->i_ino = get_next_ino();
1687 inode->i_mode = S_IFBLK;
1688 inode->i_bdev = bdev;
1689 inode->i_rdev = bdev->bd_dev;
1690 bdi = blk_get_backing_dev_info(bdev);
1692 bdi = &default_backing_dev_info;
1693 mapping = &inode->i_data;
1694 mapping->a_ops = &mapping_aops;
1695 mapping->backing_dev_info = bdi;
1696 mapping_set_gfp_mask(mapping, GFP_NOFS);
1697 btp->bt_mapping = mapping;
1702 xfs_alloc_delwrite_queue(
1706 INIT_LIST_HEAD(&btp->bt_delwrite_queue);
1707 spin_lock_init(&btp->bt_delwrite_lock);
1709 btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname);
1710 if (IS_ERR(btp->bt_task))
1711 return PTR_ERR(btp->bt_task);
1717 struct xfs_mount *mp,
1718 struct block_device *bdev,
1724 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP);
1727 btp->bt_dev = bdev->bd_dev;
1728 btp->bt_bdev = bdev;
1729 INIT_LIST_HEAD(&btp->bt_lru);
1730 spin_lock_init(&btp->bt_lru_lock);
1731 if (xfs_setsize_buftarg_early(btp, bdev))
1733 if (xfs_mapping_buftarg(btp, bdev))
1735 if (xfs_alloc_delwrite_queue(btp, fsname))
1737 btp->bt_shrinker.shrink = xfs_buftarg_shrink;
1738 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1739 register_shrinker(&btp->bt_shrinker);
1749 * Delayed write buffer handling
1752 xfs_buf_delwri_queue(
1756 struct list_head *dwq = &bp->b_target->bt_delwrite_queue;
1757 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1759 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1761 ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC));
1764 /* If already in the queue, dequeue and place at tail */
1765 if (!list_empty(&bp->b_list)) {
1766 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1768 atomic_dec(&bp->b_hold);
1769 list_del(&bp->b_list);
1772 if (list_empty(dwq)) {
1773 /* start xfsbufd as it is about to have something to do */
1774 wake_up_process(bp->b_target->bt_task);
1777 bp->b_flags |= _XBF_DELWRI_Q;
1778 list_add_tail(&bp->b_list, dwq);
1779 bp->b_queuetime = jiffies;
1787 xfs_buf_delwri_dequeue(
1790 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
1794 if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) {
1795 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1796 list_del_init(&bp->b_list);
1799 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q);
1805 trace_xfs_buf_delwri_dequeue(bp, _RET_IP_);
1809 * If a delwri buffer needs to be pushed before it has aged out, then promote
1810 * it to the head of the delwri queue so that it will be flushed on the next
1811 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1812 * than the age currently needed to flush the buffer. Hence the next time the
1813 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1816 xfs_buf_delwri_promote(
1819 struct xfs_buftarg *btp = bp->b_target;
1820 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1;
1822 ASSERT(bp->b_flags & XBF_DELWRI);
1823 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1826 * Check the buffer age before locking the delayed write queue as we
1827 * don't need to promote buffers that are already past the flush age.
1829 if (bp->b_queuetime < jiffies - age)
1831 bp->b_queuetime = jiffies - age;
1832 spin_lock(&btp->bt_delwrite_lock);
1833 list_move(&bp->b_list, &btp->bt_delwrite_queue);
1834 spin_unlock(&btp->bt_delwrite_lock);
1838 xfs_buf_runall_queues(
1839 struct workqueue_struct *queue)
1841 flush_workqueue(queue);
1845 * Move as many buffers as specified to the supplied list
1846 * idicating if we skipped any buffers to prevent deadlocks.
1849 xfs_buf_delwri_split(
1850 xfs_buftarg_t *target,
1851 struct list_head *list,
1855 struct list_head *dwq = &target->bt_delwrite_queue;
1856 spinlock_t *dwlk = &target->bt_delwrite_lock;
1860 force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1861 INIT_LIST_HEAD(list);
1863 list_for_each_entry_safe(bp, n, dwq, b_list) {
1864 ASSERT(bp->b_flags & XBF_DELWRI);
1866 if (!XFS_BUF_ISPINNED(bp) && !xfs_buf_cond_lock(bp)) {
1868 time_before(jiffies, bp->b_queuetime + age)) {
1873 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q|
1875 bp->b_flags |= XBF_WRITE;
1876 list_move_tail(&bp->b_list, list);
1877 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1888 * Compare function is more complex than it needs to be because
1889 * the return value is only 32 bits and we are doing comparisons
1895 struct list_head *a,
1896 struct list_head *b)
1898 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1899 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1902 diff = ap->b_bn - bp->b_bn;
1911 xfs_buf_delwri_sort(
1912 xfs_buftarg_t *target,
1913 struct list_head *list)
1915 list_sort(NULL, list, xfs_buf_cmp);
1922 xfs_buftarg_t *target = (xfs_buftarg_t *)data;
1924 current->flags |= PF_MEMALLOC;
1929 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10);
1930 long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10);
1932 struct list_head tmp;
1934 if (unlikely(freezing(current))) {
1935 set_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1938 clear_bit(XBT_FORCE_SLEEP, &target->bt_flags);
1941 /* sleep for a long time if there is nothing to do. */
1942 if (list_empty(&target->bt_delwrite_queue))
1943 tout = MAX_SCHEDULE_TIMEOUT;
1944 schedule_timeout_interruptible(tout);
1946 xfs_buf_delwri_split(target, &tmp, age);
1947 list_sort(NULL, &tmp, xfs_buf_cmp);
1948 while (!list_empty(&tmp)) {
1950 bp = list_first_entry(&tmp, struct xfs_buf, b_list);
1951 list_del_init(&bp->b_list);
1956 blk_run_address_space(target->bt_mapping);
1958 } while (!kthread_should_stop());
1964 * Go through all incore buffers, and release buffers if they belong to
1965 * the given device. This is used in filesystem error handling to
1966 * preserve the consistency of its metadata.
1970 xfs_buftarg_t *target,
1975 LIST_HEAD(tmp_list);
1976 LIST_HEAD(wait_list);
1978 xfs_buf_runall_queues(xfsconvertd_workqueue);
1979 xfs_buf_runall_queues(xfsdatad_workqueue);
1980 xfs_buf_runall_queues(xfslogd_workqueue);
1982 set_bit(XBT_FORCE_FLUSH, &target->bt_flags);
1983 pincount = xfs_buf_delwri_split(target, &tmp_list, 0);
1986 * Dropped the delayed write list lock, now walk the temporary list.
1987 * All I/O is issued async and then if we need to wait for completion
1988 * we do that after issuing all the IO.
1990 list_sort(NULL, &tmp_list, xfs_buf_cmp);
1991 while (!list_empty(&tmp_list)) {
1992 bp = list_first_entry(&tmp_list, struct xfs_buf, b_list);
1993 ASSERT(target == bp->b_target);
1994 list_del_init(&bp->b_list);
1996 bp->b_flags &= ~XBF_ASYNC;
1997 list_add(&bp->b_list, &wait_list);
2003 /* Expedite and wait for IO to complete. */
2004 blk_run_address_space(target->bt_mapping);
2005 while (!list_empty(&wait_list)) {
2006 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
2008 list_del_init(&bp->b_list);
2020 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
2021 KM_ZONE_HWALIGN, NULL);
2025 xfslogd_workqueue = alloc_workqueue("xfslogd",
2026 WQ_MEM_RECLAIM | WQ_HIGHPRI, 1);
2027 if (!xfslogd_workqueue)
2028 goto out_free_buf_zone;
2030 xfsdatad_workqueue = create_workqueue("xfsdatad");
2031 if (!xfsdatad_workqueue)
2032 goto out_destroy_xfslogd_workqueue;
2034 xfsconvertd_workqueue = create_workqueue("xfsconvertd");
2035 if (!xfsconvertd_workqueue)
2036 goto out_destroy_xfsdatad_workqueue;
2040 out_destroy_xfsdatad_workqueue:
2041 destroy_workqueue(xfsdatad_workqueue);
2042 out_destroy_xfslogd_workqueue:
2043 destroy_workqueue(xfslogd_workqueue);
2045 kmem_zone_destroy(xfs_buf_zone);
2051 xfs_buf_terminate(void)
2053 destroy_workqueue(xfsconvertd_workqueue);
2054 destroy_workqueue(xfsdatad_workqueue);
2055 destroy_workqueue(xfslogd_workqueue);
2056 kmem_zone_destroy(xfs_buf_zone);
2059 #ifdef CONFIG_KDB_MODULES
2061 xfs_get_buftarg_list(void)
2063 return &xfs_buftarg_list;