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>
40 #include "xfs_mount.h"
41 #include "xfs_trace.h"
43 static kmem_zone_t *xfs_buf_zone;
45 static struct workqueue_struct *xfslogd_workqueue;
47 #ifdef XFS_BUF_LOCK_TRACKING
48 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
49 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
50 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
52 # define XB_SET_OWNER(bp) do { } while (0)
53 # define XB_CLEAR_OWNER(bp) do { } while (0)
54 # define XB_GET_OWNER(bp) do { } while (0)
57 #define xb_to_gfp(flags) \
58 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN)
66 * Return true if the buffer is vmapped.
68 * b_addr is null if the buffer is not mapped, but the code is clever
69 * enough to know it doesn't have to map a single page, so the check has
70 * to be both for b_addr and bp->b_page_count > 1.
72 return bp->b_addr && bp->b_page_count > 1;
79 return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
83 * xfs_buf_lru_add - add a buffer to the LRU.
85 * The LRU takes a new reference to the buffer so that it will only be freed
86 * once the shrinker takes the buffer off the LRU.
92 struct xfs_buftarg *btp = bp->b_target;
94 spin_lock(&btp->bt_lru_lock);
95 if (list_empty(&bp->b_lru)) {
96 atomic_inc(&bp->b_hold);
97 list_add_tail(&bp->b_lru, &btp->bt_lru);
100 spin_unlock(&btp->bt_lru_lock);
104 * xfs_buf_lru_del - remove a buffer from the LRU
106 * The unlocked check is safe here because it only occurs when there are not
107 * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there
108 * to optimise the shrinker removing the buffer from the LRU and calling
109 * xfs_buf_free(). i.e. it removes an unnecessary round trip on the
116 struct xfs_buftarg *btp = bp->b_target;
118 if (list_empty(&bp->b_lru))
121 spin_lock(&btp->bt_lru_lock);
122 if (!list_empty(&bp->b_lru)) {
123 list_del_init(&bp->b_lru);
126 spin_unlock(&btp->bt_lru_lock);
130 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
131 * b_lru_ref count so that the buffer is freed immediately when the buffer
132 * reference count falls to zero. If the buffer is already on the LRU, we need
133 * to remove the reference that LRU holds on the buffer.
135 * This prevents build-up of stale buffers on the LRU.
141 ASSERT(xfs_buf_islocked(bp));
143 bp->b_flags |= XBF_STALE;
146 * Clear the delwri status so that a delwri queue walker will not
147 * flush this buffer to disk now that it is stale. The delwri queue has
148 * a reference to the buffer, so this is safe to do.
150 bp->b_flags &= ~_XBF_DELWRI_Q;
152 atomic_set(&(bp)->b_lru_ref, 0);
153 if (!list_empty(&bp->b_lru)) {
154 struct xfs_buftarg *btp = bp->b_target;
156 spin_lock(&btp->bt_lru_lock);
157 if (!list_empty(&bp->b_lru)) {
158 list_del_init(&bp->b_lru);
160 atomic_dec(&bp->b_hold);
162 spin_unlock(&btp->bt_lru_lock);
164 ASSERT(atomic_read(&bp->b_hold) >= 1);
169 struct xfs_buftarg *target,
172 xfs_buf_flags_t flags)
176 bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS);
181 * We don't want certain flags to appear in b_flags unless they are
182 * specifically set by later operations on the buffer.
184 flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD);
186 atomic_set(&bp->b_hold, 1);
187 atomic_set(&bp->b_lru_ref, 1);
188 init_completion(&bp->b_iowait);
189 INIT_LIST_HEAD(&bp->b_lru);
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;
197 * Set length and io_length to the same value initially.
198 * I/O routines should use io_length, which will be the same in
199 * most cases but may be reset (e.g. XFS recovery).
201 bp->b_length = numblks;
202 bp->b_io_length = numblks;
205 atomic_set(&bp->b_pin_count, 0);
206 init_waitqueue_head(&bp->b_waiters);
208 XFS_STATS_INC(xb_create);
209 trace_xfs_buf_init(bp, _RET_IP_);
215 * Allocate a page array capable of holding a specified number
216 * of pages, and point the page buf at it.
222 xfs_buf_flags_t flags)
224 /* Make sure that we have a page list */
225 if (bp->b_pages == NULL) {
226 bp->b_page_count = page_count;
227 if (page_count <= XB_PAGES) {
228 bp->b_pages = bp->b_page_array;
230 bp->b_pages = kmem_alloc(sizeof(struct page *) *
231 page_count, KM_NOFS);
232 if (bp->b_pages == NULL)
235 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
241 * Frees b_pages if it was allocated.
247 if (bp->b_pages != bp->b_page_array) {
248 kmem_free(bp->b_pages);
254 * Releases the specified buffer.
256 * The modification state of any associated pages is left unchanged.
257 * The buffer most not be on any hash - use xfs_buf_rele instead for
258 * hashed and refcounted buffers
264 trace_xfs_buf_free(bp, _RET_IP_);
266 ASSERT(list_empty(&bp->b_lru));
268 if (bp->b_flags & _XBF_PAGES) {
271 if (xfs_buf_is_vmapped(bp))
272 vm_unmap_ram(bp->b_addr - bp->b_offset,
275 for (i = 0; i < bp->b_page_count; i++) {
276 struct page *page = bp->b_pages[i];
280 } else if (bp->b_flags & _XBF_KMEM)
281 kmem_free(bp->b_addr);
282 _xfs_buf_free_pages(bp);
283 kmem_zone_free(xfs_buf_zone, bp);
287 * Allocates all the pages for buffer in question and builds it's page list.
290 xfs_buf_allocate_memory(
295 size_t nbytes, offset;
296 gfp_t gfp_mask = xb_to_gfp(flags);
297 unsigned short page_count, i;
298 xfs_off_t start, end;
302 * for buffers that are contained within a single page, just allocate
303 * the memory from the heap - there's no need for the complexity of
304 * page arrays to keep allocation down to order 0.
306 size = BBTOB(bp->b_length);
307 if (size < PAGE_SIZE) {
308 bp->b_addr = kmem_alloc(size, KM_NOFS);
310 /* low memory - use alloc_page loop instead */
314 if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) !=
315 ((unsigned long)bp->b_addr & PAGE_MASK)) {
316 /* b_addr spans two pages - use alloc_page instead */
317 kmem_free(bp->b_addr);
321 bp->b_offset = offset_in_page(bp->b_addr);
322 bp->b_pages = bp->b_page_array;
323 bp->b_pages[0] = virt_to_page(bp->b_addr);
324 bp->b_page_count = 1;
325 bp->b_flags |= _XBF_KMEM;
330 start = BBTOB(bp->b_bn) >> PAGE_SHIFT;
331 end = (BBTOB(bp->b_bn + bp->b_length) + PAGE_SIZE - 1) >> PAGE_SHIFT;
332 page_count = end - start;
333 error = _xfs_buf_get_pages(bp, page_count, flags);
337 offset = bp->b_offset;
338 bp->b_flags |= _XBF_PAGES;
340 for (i = 0; i < bp->b_page_count; i++) {
344 page = alloc_page(gfp_mask);
345 if (unlikely(page == NULL)) {
346 if (flags & XBF_READ_AHEAD) {
347 bp->b_page_count = i;
353 * This could deadlock.
355 * But until all the XFS lowlevel code is revamped to
356 * handle buffer allocation failures we can't do much.
358 if (!(++retries % 100))
360 "possible memory allocation deadlock in %s (mode:0x%x)",
363 XFS_STATS_INC(xb_page_retries);
364 congestion_wait(BLK_RW_ASYNC, HZ/50);
368 XFS_STATS_INC(xb_page_found);
370 nbytes = min_t(size_t, size, PAGE_SIZE - offset);
372 bp->b_pages[i] = page;
378 for (i = 0; i < bp->b_page_count; i++)
379 __free_page(bp->b_pages[i]);
384 * Map buffer into kernel address-space if necessary.
391 ASSERT(bp->b_flags & _XBF_PAGES);
392 if (bp->b_page_count == 1) {
393 /* A single page buffer is always mappable */
394 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
395 } else if (flags & XBF_UNMAPPED) {
401 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
406 } while (retried++ <= 1);
410 bp->b_addr += bp->b_offset;
417 * Finding and Reading Buffers
421 * Look up, and creates if absent, a lockable buffer for
422 * a given range of an inode. The buffer is returned
423 * locked. No I/O is implied by this call.
427 struct xfs_buftarg *btp,
430 xfs_buf_flags_t flags,
434 struct xfs_perag *pag;
435 struct rb_node **rbp;
436 struct rb_node *parent;
439 numbytes = BBTOB(numblks);
441 /* Check for IOs smaller than the sector size / not sector aligned */
442 ASSERT(!(numbytes < (1 << btp->bt_sshift)));
443 ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_smask));
446 pag = xfs_perag_get(btp->bt_mount,
447 xfs_daddr_to_agno(btp->bt_mount, blkno));
450 spin_lock(&pag->pag_buf_lock);
451 rbp = &pag->pag_buf_tree.rb_node;
456 bp = rb_entry(parent, struct xfs_buf, b_rbnode);
458 if (blkno < bp->b_bn)
459 rbp = &(*rbp)->rb_left;
460 else if (blkno > bp->b_bn)
461 rbp = &(*rbp)->rb_right;
464 * found a block number match. If the range doesn't
465 * match, the only way this is allowed is if the buffer
466 * in the cache is stale and the transaction that made
467 * it stale has not yet committed. i.e. we are
468 * reallocating a busy extent. Skip this buffer and
469 * continue searching to the right for an exact match.
471 if (bp->b_length != numblks) {
472 ASSERT(bp->b_flags & XBF_STALE);
473 rbp = &(*rbp)->rb_right;
476 atomic_inc(&bp->b_hold);
483 rb_link_node(&new_bp->b_rbnode, parent, rbp);
484 rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree);
485 /* the buffer keeps the perag reference until it is freed */
487 spin_unlock(&pag->pag_buf_lock);
489 XFS_STATS_INC(xb_miss_locked);
490 spin_unlock(&pag->pag_buf_lock);
496 spin_unlock(&pag->pag_buf_lock);
499 if (!xfs_buf_trylock(bp)) {
500 if (flags & XBF_TRYLOCK) {
502 XFS_STATS_INC(xb_busy_locked);
506 XFS_STATS_INC(xb_get_locked_waited);
510 * if the buffer is stale, clear all the external state associated with
511 * it. We need to keep flags such as how we allocated the buffer memory
514 if (bp->b_flags & XBF_STALE) {
515 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
516 bp->b_flags &= _XBF_KMEM | _XBF_PAGES;
519 trace_xfs_buf_find(bp, flags, _RET_IP_);
520 XFS_STATS_INC(xb_get_locked);
525 * Assembles a buffer covering the specified range. The code is optimised for
526 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
527 * more hits than misses.
531 xfs_buftarg_t *target,
534 xfs_buf_flags_t flags)
537 struct xfs_buf *new_bp;
540 bp = _xfs_buf_find(target, blkno, numblks, flags, NULL);
544 new_bp = xfs_buf_alloc(target, blkno, numblks, flags);
545 if (unlikely(!new_bp))
548 error = xfs_buf_allocate_memory(new_bp, flags);
550 kmem_zone_free(xfs_buf_zone, new_bp);
554 bp = _xfs_buf_find(target, blkno, numblks, flags, new_bp);
556 xfs_buf_free(new_bp);
561 xfs_buf_free(new_bp);
563 bp->b_io_length = bp->b_length;
567 error = _xfs_buf_map_pages(bp, flags);
568 if (unlikely(error)) {
569 xfs_warn(target->bt_mount,
570 "%s: failed to map pages\n", __func__);
576 XFS_STATS_INC(xb_get);
577 trace_xfs_buf_get(bp, flags, _RET_IP_);
584 xfs_buf_flags_t flags)
586 ASSERT(!(flags & XBF_WRITE));
587 ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL);
589 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD);
590 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
592 xfs_buf_iorequest(bp);
593 if (flags & XBF_ASYNC)
595 return xfs_buf_iowait(bp);
600 xfs_buftarg_t *target,
603 xfs_buf_flags_t flags)
609 bp = xfs_buf_get(target, blkno, numblks, flags);
611 trace_xfs_buf_read(bp, flags, _RET_IP_);
613 if (!XFS_BUF_ISDONE(bp)) {
614 XFS_STATS_INC(xb_get_read);
615 _xfs_buf_read(bp, flags);
616 } else if (flags & XBF_ASYNC) {
618 * Read ahead call which is already satisfied,
624 /* We do not want read in the flags */
625 bp->b_flags &= ~XBF_READ;
633 * If we are not low on memory then do the readahead in a deadlock
638 xfs_buftarg_t *target,
642 if (bdi_read_congested(target->bt_bdi))
645 xfs_buf_read(target, blkno, numblks,
646 XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD);
650 * Read an uncached buffer from disk. Allocates and returns a locked
651 * buffer containing the disk contents or nothing.
654 xfs_buf_read_uncached(
655 struct xfs_buftarg *target,
663 bp = xfs_buf_get_uncached(target, numblks, flags);
667 /* set up the buffer for a read IO */
668 XFS_BUF_SET_ADDR(bp, daddr);
671 xfsbdstrat(target->bt_mount, bp);
672 error = xfs_buf_iowait(bp);
681 * Return a buffer allocated as an empty buffer and associated to external
682 * memory via xfs_buf_associate_memory() back to it's empty state.
690 _xfs_buf_free_pages(bp);
693 bp->b_page_count = 0;
695 bp->b_length = numblks;
696 bp->b_io_length = numblks;
697 bp->b_bn = XFS_BUF_DADDR_NULL;
700 static inline struct page *
704 if ((!is_vmalloc_addr(addr))) {
705 return virt_to_page(addr);
707 return vmalloc_to_page(addr);
712 xfs_buf_associate_memory(
719 unsigned long pageaddr;
720 unsigned long offset;
724 pageaddr = (unsigned long)mem & PAGE_MASK;
725 offset = (unsigned long)mem - pageaddr;
726 buflen = PAGE_ALIGN(len + offset);
727 page_count = buflen >> PAGE_SHIFT;
729 /* Free any previous set of page pointers */
731 _xfs_buf_free_pages(bp);
736 rval = _xfs_buf_get_pages(bp, page_count, 0);
740 bp->b_offset = offset;
742 for (i = 0; i < bp->b_page_count; i++) {
743 bp->b_pages[i] = mem_to_page((void *)pageaddr);
744 pageaddr += PAGE_SIZE;
747 bp->b_io_length = BTOBB(len);
748 bp->b_length = BTOBB(buflen);
754 xfs_buf_get_uncached(
755 struct xfs_buftarg *target,
759 unsigned long page_count;
763 bp = xfs_buf_alloc(target, XFS_BUF_DADDR_NULL, numblks, 0);
764 if (unlikely(bp == NULL))
767 page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT;
768 error = _xfs_buf_get_pages(bp, page_count, 0);
772 for (i = 0; i < page_count; i++) {
773 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
777 bp->b_flags |= _XBF_PAGES;
779 error = _xfs_buf_map_pages(bp, 0);
780 if (unlikely(error)) {
781 xfs_warn(target->bt_mount,
782 "%s: failed to map pages\n", __func__);
786 trace_xfs_buf_get_uncached(bp, _RET_IP_);
791 __free_page(bp->b_pages[i]);
792 _xfs_buf_free_pages(bp);
794 kmem_zone_free(xfs_buf_zone, bp);
800 * Increment reference count on buffer, to hold the buffer concurrently
801 * with another thread which may release (free) the buffer asynchronously.
802 * Must hold the buffer already to call this function.
808 trace_xfs_buf_hold(bp, _RET_IP_);
809 atomic_inc(&bp->b_hold);
813 * Releases a hold on the specified buffer. If the
814 * the hold count is 1, calls xfs_buf_free.
820 struct xfs_perag *pag = bp->b_pag;
822 trace_xfs_buf_rele(bp, _RET_IP_);
825 ASSERT(list_empty(&bp->b_lru));
826 ASSERT(RB_EMPTY_NODE(&bp->b_rbnode));
827 if (atomic_dec_and_test(&bp->b_hold))
832 ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode));
834 ASSERT(atomic_read(&bp->b_hold) > 0);
835 if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) {
836 if (!(bp->b_flags & XBF_STALE) &&
837 atomic_read(&bp->b_lru_ref)) {
839 spin_unlock(&pag->pag_buf_lock);
842 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
843 rb_erase(&bp->b_rbnode, &pag->pag_buf_tree);
844 spin_unlock(&pag->pag_buf_lock);
853 * Lock a buffer object, if it is not already locked.
855 * If we come across a stale, pinned, locked buffer, we know that we are
856 * being asked to lock a buffer that has been reallocated. Because it is
857 * pinned, we know that the log has not been pushed to disk and hence it
858 * will still be locked. Rather than continuing to have trylock attempts
859 * fail until someone else pushes the log, push it ourselves before
860 * returning. This means that the xfsaild will not get stuck trying
861 * to push on stale inode buffers.
869 locked = down_trylock(&bp->b_sema) == 0;
872 else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
873 xfs_log_force(bp->b_target->bt_mount, 0);
875 trace_xfs_buf_trylock(bp, _RET_IP_);
880 * Lock a buffer object.
882 * If we come across a stale, pinned, locked buffer, we know that we
883 * are being asked to lock a buffer that has been reallocated. Because
884 * it is pinned, we know that the log has not been pushed to disk and
885 * hence it will still be locked. Rather than sleeping until someone
886 * else pushes the log, push it ourselves before trying to get the lock.
892 trace_xfs_buf_lock(bp, _RET_IP_);
894 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
895 xfs_log_force(bp->b_target->bt_mount, 0);
899 trace_xfs_buf_lock_done(bp, _RET_IP_);
909 trace_xfs_buf_unlock(bp, _RET_IP_);
916 DECLARE_WAITQUEUE (wait, current);
918 if (atomic_read(&bp->b_pin_count) == 0)
921 add_wait_queue(&bp->b_waiters, &wait);
923 set_current_state(TASK_UNINTERRUPTIBLE);
924 if (atomic_read(&bp->b_pin_count) == 0)
928 remove_wait_queue(&bp->b_waiters, &wait);
929 set_current_state(TASK_RUNNING);
933 * Buffer Utility Routines
938 struct work_struct *work)
941 container_of(work, xfs_buf_t, b_iodone_work);
944 (*(bp->b_iodone))(bp);
945 else if (bp->b_flags & XBF_ASYNC)
954 trace_xfs_buf_iodone(bp, _RET_IP_);
956 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
957 if (bp->b_error == 0)
958 bp->b_flags |= XBF_DONE;
960 if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) {
962 INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work);
963 queue_work(xfslogd_workqueue, &bp->b_iodone_work);
965 xfs_buf_iodone_work(&bp->b_iodone_work);
968 complete(&bp->b_iowait);
977 ASSERT(error >= 0 && error <= 0xffff);
978 bp->b_error = (unsigned short)error;
979 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
983 xfs_buf_ioerror_alert(
987 xfs_alert(bp->b_target->bt_mount,
988 "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d",
989 (__uint64_t)XFS_BUF_ADDR(bp), func, bp->b_error, bp->b_length);
993 * Called when we want to stop a buffer from getting written or read.
994 * We attach the EIO error, muck with its flags, and call xfs_buf_ioend
995 * so that the proper iodone callbacks get called.
1001 #ifdef XFSERRORDEBUG
1002 ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
1006 * No need to wait until the buffer is unpinned, we aren't flushing it.
1008 xfs_buf_ioerror(bp, EIO);
1011 * We're calling xfs_buf_ioend, so delete XBF_DONE flag.
1017 xfs_buf_ioend(bp, 0);
1023 * Same as xfs_bioerror, except that we are releasing the buffer
1024 * here ourselves, and avoiding the xfs_buf_ioend call.
1025 * This is meant for userdata errors; metadata bufs come with
1026 * iodone functions attached, so that we can track down errors.
1032 int64_t fl = bp->b_flags;
1034 * No need to wait until the buffer is unpinned.
1035 * We aren't flushing it.
1037 * chunkhold expects B_DONE to be set, whether
1038 * we actually finish the I/O or not. We don't want to
1039 * change that interface.
1044 bp->b_iodone = NULL;
1045 if (!(fl & XBF_ASYNC)) {
1047 * Mark b_error and B_ERROR _both_.
1048 * Lot's of chunkcache code assumes that.
1049 * There's no reason to mark error for
1052 xfs_buf_ioerror(bp, EIO);
1053 complete(&bp->b_iowait);
1065 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1066 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1068 * Metadata write that didn't get logged but
1069 * written delayed anyway. These aren't associated
1070 * with a transaction, and can be ignored.
1072 if (!bp->b_iodone && !XFS_BUF_ISREAD(bp))
1073 return xfs_bioerror_relse(bp);
1075 return xfs_bioerror(bp);
1078 xfs_buf_iorequest(bp);
1088 ASSERT(xfs_buf_islocked(bp));
1090 bp->b_flags |= XBF_WRITE;
1091 bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q);
1095 error = xfs_buf_iowait(bp);
1097 xfs_force_shutdown(bp->b_target->bt_mount,
1098 SHUTDOWN_META_IO_ERROR);
1104 * Wrapper around bdstrat so that we can stop data from going to disk in case
1105 * we are shutting down the filesystem. Typically user data goes thru this
1106 * path; one of the exceptions is the superblock.
1110 struct xfs_mount *mp,
1113 if (XFS_FORCED_SHUTDOWN(mp)) {
1114 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1115 xfs_bioerror_relse(bp);
1119 xfs_buf_iorequest(bp);
1127 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1128 xfs_buf_ioend(bp, schedule);
1136 xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private;
1138 xfs_buf_ioerror(bp, -error);
1140 if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1141 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1143 _xfs_buf_ioend(bp, 1);
1151 int rw, map_i, total_nr_pages, nr_pages;
1153 int offset = bp->b_offset;
1154 int size = BBTOB(bp->b_io_length);
1155 sector_t sector = bp->b_bn;
1157 total_nr_pages = bp->b_page_count;
1160 if (bp->b_flags & XBF_WRITE) {
1161 if (bp->b_flags & XBF_SYNCIO)
1165 if (bp->b_flags & XBF_FUA)
1167 if (bp->b_flags & XBF_FLUSH)
1169 } else if (bp->b_flags & XBF_READ_AHEAD) {
1175 /* we only use the buffer cache for meta-data */
1179 atomic_inc(&bp->b_io_remaining);
1180 nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1181 if (nr_pages > total_nr_pages)
1182 nr_pages = total_nr_pages;
1184 bio = bio_alloc(GFP_NOIO, nr_pages);
1185 bio->bi_bdev = bp->b_target->bt_bdev;
1186 bio->bi_sector = sector;
1187 bio->bi_end_io = xfs_buf_bio_end_io;
1188 bio->bi_private = bp;
1191 for (; size && nr_pages; nr_pages--, map_i++) {
1192 int rbytes, nbytes = PAGE_SIZE - offset;
1197 rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset);
1198 if (rbytes < nbytes)
1202 sector += BTOBB(nbytes);
1207 if (likely(bio->bi_size)) {
1208 if (xfs_buf_is_vmapped(bp)) {
1209 flush_kernel_vmap_range(bp->b_addr,
1210 xfs_buf_vmap_len(bp));
1212 submit_bio(rw, bio);
1216 xfs_buf_ioerror(bp, EIO);
1225 trace_xfs_buf_iorequest(bp, _RET_IP_);
1227 ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1229 if (bp->b_flags & XBF_WRITE)
1230 xfs_buf_wait_unpin(bp);
1233 /* Set the count to 1 initially, this will stop an I/O
1234 * completion callout which happens before we have started
1235 * all the I/O from calling xfs_buf_ioend too early.
1237 atomic_set(&bp->b_io_remaining, 1);
1238 _xfs_buf_ioapply(bp);
1239 _xfs_buf_ioend(bp, 1);
1245 * Waits for I/O to complete on the buffer supplied. It returns immediately if
1246 * no I/O is pending or there is already a pending error on the buffer. It
1247 * returns the I/O error code, if any, or 0 if there was no error.
1253 trace_xfs_buf_iowait(bp, _RET_IP_);
1256 wait_for_completion(&bp->b_iowait);
1258 trace_xfs_buf_iowait_done(bp, _RET_IP_);
1270 return bp->b_addr + offset;
1272 offset += bp->b_offset;
1273 page = bp->b_pages[offset >> PAGE_SHIFT];
1274 return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1));
1278 * Move data into or out of a buffer.
1282 xfs_buf_t *bp, /* buffer to process */
1283 size_t boff, /* starting buffer offset */
1284 size_t bsize, /* length to copy */
1285 void *data, /* data address */
1286 xfs_buf_rw_t mode) /* read/write/zero flag */
1290 bend = boff + bsize;
1291 while (boff < bend) {
1293 int page_index, page_offset, csize;
1295 page_index = (boff + bp->b_offset) >> PAGE_SHIFT;
1296 page_offset = (boff + bp->b_offset) & ~PAGE_MASK;
1297 page = bp->b_pages[page_index];
1298 csize = min_t(size_t, PAGE_SIZE - page_offset,
1299 BBTOB(bp->b_io_length) - boff);
1301 ASSERT((csize + page_offset) <= PAGE_SIZE);
1305 memset(page_address(page) + page_offset, 0, csize);
1308 memcpy(data, page_address(page) + page_offset, csize);
1311 memcpy(page_address(page) + page_offset, data, csize);
1320 * Handling of buffer targets (buftargs).
1324 * Wait for any bufs with callbacks that have been submitted but have not yet
1325 * returned. These buffers will have an elevated hold count, so wait on those
1326 * while freeing all the buffers only held by the LRU.
1330 struct xfs_buftarg *btp)
1335 spin_lock(&btp->bt_lru_lock);
1336 while (!list_empty(&btp->bt_lru)) {
1337 bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru);
1338 if (atomic_read(&bp->b_hold) > 1) {
1339 spin_unlock(&btp->bt_lru_lock);
1344 * clear the LRU reference count so the buffer doesn't get
1345 * ignored in xfs_buf_rele().
1347 atomic_set(&bp->b_lru_ref, 0);
1348 spin_unlock(&btp->bt_lru_lock);
1350 spin_lock(&btp->bt_lru_lock);
1352 spin_unlock(&btp->bt_lru_lock);
1357 struct shrinker *shrink,
1358 struct shrink_control *sc)
1360 struct xfs_buftarg *btp = container_of(shrink,
1361 struct xfs_buftarg, bt_shrinker);
1363 int nr_to_scan = sc->nr_to_scan;
1367 return btp->bt_lru_nr;
1369 spin_lock(&btp->bt_lru_lock);
1370 while (!list_empty(&btp->bt_lru)) {
1371 if (nr_to_scan-- <= 0)
1374 bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru);
1377 * Decrement the b_lru_ref count unless the value is already
1378 * zero. If the value is already zero, we need to reclaim the
1379 * buffer, otherwise it gets another trip through the LRU.
1381 if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1382 list_move_tail(&bp->b_lru, &btp->bt_lru);
1387 * remove the buffer from the LRU now to avoid needing another
1388 * lock round trip inside xfs_buf_rele().
1390 list_move(&bp->b_lru, &dispose);
1393 spin_unlock(&btp->bt_lru_lock);
1395 while (!list_empty(&dispose)) {
1396 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1397 list_del_init(&bp->b_lru);
1401 return btp->bt_lru_nr;
1406 struct xfs_mount *mp,
1407 struct xfs_buftarg *btp)
1409 unregister_shrinker(&btp->bt_shrinker);
1411 if (mp->m_flags & XFS_MOUNT_BARRIER)
1412 xfs_blkdev_issue_flush(btp);
1418 xfs_setsize_buftarg_flags(
1420 unsigned int blocksize,
1421 unsigned int sectorsize,
1424 btp->bt_bsize = blocksize;
1425 btp->bt_sshift = ffs(sectorsize) - 1;
1426 btp->bt_smask = sectorsize - 1;
1428 if (set_blocksize(btp->bt_bdev, sectorsize)) {
1429 char name[BDEVNAME_SIZE];
1431 bdevname(btp->bt_bdev, name);
1433 xfs_warn(btp->bt_mount,
1434 "Cannot set_blocksize to %u on device %s\n",
1443 * When allocating the initial buffer target we have not yet
1444 * read in the superblock, so don't know what sized sectors
1445 * are being used is at this early stage. Play safe.
1448 xfs_setsize_buftarg_early(
1450 struct block_device *bdev)
1452 return xfs_setsize_buftarg_flags(btp,
1453 PAGE_SIZE, bdev_logical_block_size(bdev), 0);
1457 xfs_setsize_buftarg(
1459 unsigned int blocksize,
1460 unsigned int sectorsize)
1462 return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1);
1467 struct xfs_mount *mp,
1468 struct block_device *bdev,
1474 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP);
1477 btp->bt_dev = bdev->bd_dev;
1478 btp->bt_bdev = bdev;
1479 btp->bt_bdi = blk_get_backing_dev_info(bdev);
1483 INIT_LIST_HEAD(&btp->bt_lru);
1484 spin_lock_init(&btp->bt_lru_lock);
1485 if (xfs_setsize_buftarg_early(btp, bdev))
1487 btp->bt_shrinker.shrink = xfs_buftarg_shrink;
1488 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1489 register_shrinker(&btp->bt_shrinker);
1498 * Add a buffer to the delayed write list.
1500 * This queues a buffer for writeout if it hasn't already been. Note that
1501 * neither this routine nor the buffer list submission functions perform
1502 * any internal synchronization. It is expected that the lists are thread-local
1505 * Returns true if we queued up the buffer, or false if it already had
1506 * been on the buffer list.
1509 xfs_buf_delwri_queue(
1511 struct list_head *list)
1513 ASSERT(xfs_buf_islocked(bp));
1514 ASSERT(!(bp->b_flags & XBF_READ));
1517 * If the buffer is already marked delwri it already is queued up
1518 * by someone else for imediate writeout. Just ignore it in that
1521 if (bp->b_flags & _XBF_DELWRI_Q) {
1522 trace_xfs_buf_delwri_queued(bp, _RET_IP_);
1526 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1529 * If a buffer gets written out synchronously or marked stale while it
1530 * is on a delwri list we lazily remove it. To do this, the other party
1531 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1532 * It remains referenced and on the list. In a rare corner case it
1533 * might get readded to a delwri list after the synchronous writeout, in
1534 * which case we need just need to re-add the flag here.
1536 bp->b_flags |= _XBF_DELWRI_Q;
1537 if (list_empty(&bp->b_list)) {
1538 atomic_inc(&bp->b_hold);
1539 list_add_tail(&bp->b_list, list);
1546 * Compare function is more complex than it needs to be because
1547 * the return value is only 32 bits and we are doing comparisons
1553 struct list_head *a,
1554 struct list_head *b)
1556 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1557 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1560 diff = ap->b_bn - bp->b_bn;
1569 __xfs_buf_delwri_submit(
1570 struct list_head *buffer_list,
1571 struct list_head *io_list,
1574 struct blk_plug plug;
1575 struct xfs_buf *bp, *n;
1578 list_for_each_entry_safe(bp, n, buffer_list, b_list) {
1580 if (xfs_buf_ispinned(bp)) {
1584 if (!xfs_buf_trylock(bp))
1591 * Someone else might have written the buffer synchronously or
1592 * marked it stale in the meantime. In that case only the
1593 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
1594 * reference and remove it from the list here.
1596 if (!(bp->b_flags & _XBF_DELWRI_Q)) {
1597 list_del_init(&bp->b_list);
1602 list_move_tail(&bp->b_list, io_list);
1603 trace_xfs_buf_delwri_split(bp, _RET_IP_);
1606 list_sort(NULL, io_list, xfs_buf_cmp);
1608 blk_start_plug(&plug);
1609 list_for_each_entry_safe(bp, n, io_list, b_list) {
1610 bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC);
1611 bp->b_flags |= XBF_WRITE;
1614 bp->b_flags |= XBF_ASYNC;
1615 list_del_init(&bp->b_list);
1619 blk_finish_plug(&plug);
1625 * Write out a buffer list asynchronously.
1627 * This will take the @buffer_list, write all non-locked and non-pinned buffers
1628 * out and not wait for I/O completion on any of the buffers. This interface
1629 * is only safely useable for callers that can track I/O completion by higher
1630 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
1634 xfs_buf_delwri_submit_nowait(
1635 struct list_head *buffer_list)
1637 LIST_HEAD (io_list);
1638 return __xfs_buf_delwri_submit(buffer_list, &io_list, false);
1642 * Write out a buffer list synchronously.
1644 * This will take the @buffer_list, write all buffers out and wait for I/O
1645 * completion on all of the buffers. @buffer_list is consumed by the function,
1646 * so callers must have some other way of tracking buffers if they require such
1650 xfs_buf_delwri_submit(
1651 struct list_head *buffer_list)
1653 LIST_HEAD (io_list);
1654 int error = 0, error2;
1657 __xfs_buf_delwri_submit(buffer_list, &io_list, true);
1659 /* Wait for IO to complete. */
1660 while (!list_empty(&io_list)) {
1661 bp = list_first_entry(&io_list, struct xfs_buf, b_list);
1663 list_del_init(&bp->b_list);
1664 error2 = xfs_buf_iowait(bp);
1676 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
1677 KM_ZONE_HWALIGN, NULL);
1681 xfslogd_workqueue = alloc_workqueue("xfslogd",
1682 WQ_MEM_RECLAIM | WQ_HIGHPRI, 1);
1683 if (!xfslogd_workqueue)
1684 goto out_free_buf_zone;
1689 kmem_zone_destroy(xfs_buf_zone);
1695 xfs_buf_terminate(void)
1697 destroy_workqueue(xfslogd_workqueue);
1698 kmem_zone_destroy(xfs_buf_zone);