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>
37 #include "xfs_format.h"
38 #include "xfs_log_format.h"
39 #include "xfs_trans_resv.h"
41 #include "xfs_mount.h"
42 #include "xfs_trace.h"
45 static kmem_zone_t
*xfs_buf_zone
;
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 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
84 * b_lru_ref count so that the buffer is freed immediately when the buffer
85 * reference count falls to zero. If the buffer is already on the LRU, we need
86 * to remove the reference that LRU holds on the buffer.
88 * This prevents build-up of stale buffers on the LRU.
94 ASSERT(xfs_buf_islocked(bp
));
96 bp
->b_flags
|= XBF_STALE
;
99 * Clear the delwri status so that a delwri queue walker will not
100 * flush this buffer to disk now that it is stale. The delwri queue has
101 * a reference to the buffer, so this is safe to do.
103 bp
->b_flags
&= ~_XBF_DELWRI_Q
;
105 spin_lock(&bp
->b_lock
);
106 atomic_set(&bp
->b_lru_ref
, 0);
107 if (!(bp
->b_state
& XFS_BSTATE_DISPOSE
) &&
108 (list_lru_del(&bp
->b_target
->bt_lru
, &bp
->b_lru
)))
109 atomic_dec(&bp
->b_hold
);
111 ASSERT(atomic_read(&bp
->b_hold
) >= 1);
112 spin_unlock(&bp
->b_lock
);
120 ASSERT(bp
->b_maps
== NULL
);
121 bp
->b_map_count
= map_count
;
123 if (map_count
== 1) {
124 bp
->b_maps
= &bp
->__b_map
;
128 bp
->b_maps
= kmem_zalloc(map_count
* sizeof(struct xfs_buf_map
),
136 * Frees b_pages if it was allocated.
142 if (bp
->b_maps
!= &bp
->__b_map
) {
143 kmem_free(bp
->b_maps
);
150 struct xfs_buftarg
*target
,
151 struct xfs_buf_map
*map
,
153 xfs_buf_flags_t flags
)
159 bp
= kmem_zone_zalloc(xfs_buf_zone
, KM_NOFS
);
164 * We don't want certain flags to appear in b_flags unless they are
165 * specifically set by later operations on the buffer.
167 flags
&= ~(XBF_UNMAPPED
| XBF_TRYLOCK
| XBF_ASYNC
| XBF_READ_AHEAD
);
169 atomic_set(&bp
->b_hold
, 1);
170 atomic_set(&bp
->b_lru_ref
, 1);
171 init_completion(&bp
->b_iowait
);
172 INIT_LIST_HEAD(&bp
->b_lru
);
173 INIT_LIST_HEAD(&bp
->b_list
);
174 RB_CLEAR_NODE(&bp
->b_rbnode
);
175 sema_init(&bp
->b_sema
, 0); /* held, no waiters */
176 spin_lock_init(&bp
->b_lock
);
178 bp
->b_target
= target
;
182 * Set length and io_length to the same value initially.
183 * I/O routines should use io_length, which will be the same in
184 * most cases but may be reset (e.g. XFS recovery).
186 error
= xfs_buf_get_maps(bp
, nmaps
);
188 kmem_zone_free(xfs_buf_zone
, bp
);
192 bp
->b_bn
= map
[0].bm_bn
;
194 for (i
= 0; i
< nmaps
; i
++) {
195 bp
->b_maps
[i
].bm_bn
= map
[i
].bm_bn
;
196 bp
->b_maps
[i
].bm_len
= map
[i
].bm_len
;
197 bp
->b_length
+= map
[i
].bm_len
;
199 bp
->b_io_length
= bp
->b_length
;
201 atomic_set(&bp
->b_pin_count
, 0);
202 init_waitqueue_head(&bp
->b_waiters
);
204 XFS_STATS_INC(xb_create
);
205 trace_xfs_buf_init(bp
, _RET_IP_
);
211 * Allocate a page array capable of holding a specified number
212 * of pages, and point the page buf at it.
219 /* Make sure that we have a page list */
220 if (bp
->b_pages
== NULL
) {
221 bp
->b_page_count
= page_count
;
222 if (page_count
<= XB_PAGES
) {
223 bp
->b_pages
= bp
->b_page_array
;
225 bp
->b_pages
= kmem_alloc(sizeof(struct page
*) *
226 page_count
, KM_NOFS
);
227 if (bp
->b_pages
== NULL
)
230 memset(bp
->b_pages
, 0, sizeof(struct page
*) * page_count
);
236 * Frees b_pages if it was allocated.
242 if (bp
->b_pages
!= bp
->b_page_array
) {
243 kmem_free(bp
->b_pages
);
249 * Releases the specified buffer.
251 * The modification state of any associated pages is left unchanged.
252 * The buffer must not be on any hash - use xfs_buf_rele instead for
253 * hashed and refcounted buffers
259 trace_xfs_buf_free(bp
, _RET_IP_
);
261 ASSERT(list_empty(&bp
->b_lru
));
263 if (bp
->b_flags
& _XBF_PAGES
) {
266 if (xfs_buf_is_vmapped(bp
))
267 vm_unmap_ram(bp
->b_addr
- bp
->b_offset
,
270 for (i
= 0; i
< bp
->b_page_count
; i
++) {
271 struct page
*page
= bp
->b_pages
[i
];
275 } else if (bp
->b_flags
& _XBF_KMEM
)
276 kmem_free(bp
->b_addr
);
277 _xfs_buf_free_pages(bp
);
278 xfs_buf_free_maps(bp
);
279 kmem_zone_free(xfs_buf_zone
, bp
);
283 * Allocates all the pages for buffer in question and builds it's page list.
286 xfs_buf_allocate_memory(
291 size_t nbytes
, offset
;
292 gfp_t gfp_mask
= xb_to_gfp(flags
);
293 unsigned short page_count
, i
;
294 xfs_off_t start
, end
;
298 * for buffers that are contained within a single page, just allocate
299 * the memory from the heap - there's no need for the complexity of
300 * page arrays to keep allocation down to order 0.
302 size
= BBTOB(bp
->b_length
);
303 if (size
< PAGE_SIZE
) {
304 bp
->b_addr
= kmem_alloc(size
, KM_NOFS
);
306 /* low memory - use alloc_page loop instead */
310 if (((unsigned long)(bp
->b_addr
+ size
- 1) & PAGE_MASK
) !=
311 ((unsigned long)bp
->b_addr
& PAGE_MASK
)) {
312 /* b_addr spans two pages - use alloc_page instead */
313 kmem_free(bp
->b_addr
);
317 bp
->b_offset
= offset_in_page(bp
->b_addr
);
318 bp
->b_pages
= bp
->b_page_array
;
319 bp
->b_pages
[0] = virt_to_page(bp
->b_addr
);
320 bp
->b_page_count
= 1;
321 bp
->b_flags
|= _XBF_KMEM
;
326 start
= BBTOB(bp
->b_maps
[0].bm_bn
) >> PAGE_SHIFT
;
327 end
= (BBTOB(bp
->b_maps
[0].bm_bn
+ bp
->b_length
) + PAGE_SIZE
- 1)
329 page_count
= end
- start
;
330 error
= _xfs_buf_get_pages(bp
, page_count
);
334 offset
= bp
->b_offset
;
335 bp
->b_flags
|= _XBF_PAGES
;
337 for (i
= 0; i
< bp
->b_page_count
; i
++) {
341 page
= alloc_page(gfp_mask
);
342 if (unlikely(page
== NULL
)) {
343 if (flags
& XBF_READ_AHEAD
) {
344 bp
->b_page_count
= i
;
350 * This could deadlock.
352 * But until all the XFS lowlevel code is revamped to
353 * handle buffer allocation failures we can't do much.
355 if (!(++retries
% 100))
357 "possible memory allocation deadlock in %s (mode:0x%x)",
360 XFS_STATS_INC(xb_page_retries
);
361 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
365 XFS_STATS_INC(xb_page_found
);
367 nbytes
= min_t(size_t, size
, PAGE_SIZE
- offset
);
369 bp
->b_pages
[i
] = page
;
375 for (i
= 0; i
< bp
->b_page_count
; i
++)
376 __free_page(bp
->b_pages
[i
]);
381 * Map buffer into kernel address-space if necessary.
388 ASSERT(bp
->b_flags
& _XBF_PAGES
);
389 if (bp
->b_page_count
== 1) {
390 /* A single page buffer is always mappable */
391 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
392 } else if (flags
& XBF_UNMAPPED
) {
399 * vm_map_ram() will allocate auxillary structures (e.g.
400 * pagetables) with GFP_KERNEL, yet we are likely to be under
401 * GFP_NOFS context here. Hence we need to tell memory reclaim
402 * that we are in such a context via PF_MEMALLOC_NOIO to prevent
403 * memory reclaim re-entering the filesystem here and
404 * potentially deadlocking.
406 noio_flag
= memalloc_noio_save();
408 bp
->b_addr
= vm_map_ram(bp
->b_pages
, bp
->b_page_count
,
413 } while (retried
++ <= 1);
414 memalloc_noio_restore(noio_flag
);
418 bp
->b_addr
+= bp
->b_offset
;
425 * Finding and Reading Buffers
429 * Look up, and creates if absent, a lockable buffer for
430 * a given range of an inode. The buffer is returned
431 * locked. No I/O is implied by this call.
435 struct xfs_buftarg
*btp
,
436 struct xfs_buf_map
*map
,
438 xfs_buf_flags_t flags
,
441 struct xfs_perag
*pag
;
442 struct rb_node
**rbp
;
443 struct rb_node
*parent
;
445 xfs_daddr_t blkno
= map
[0].bm_bn
;
450 for (i
= 0; i
< nmaps
; i
++)
451 numblks
+= map
[i
].bm_len
;
453 /* Check for IOs smaller than the sector size / not sector aligned */
454 ASSERT(!(BBTOB(numblks
) < btp
->bt_meta_sectorsize
));
455 ASSERT(!(BBTOB(blkno
) & (xfs_off_t
)btp
->bt_meta_sectormask
));
458 * Corrupted block numbers can get through to here, unfortunately, so we
459 * have to check that the buffer falls within the filesystem bounds.
461 eofs
= XFS_FSB_TO_BB(btp
->bt_mount
, btp
->bt_mount
->m_sb
.sb_dblocks
);
462 if (blkno
< 0 || blkno
>= eofs
) {
464 * XXX (dgc): we should really be returning -EFSCORRUPTED here,
465 * but none of the higher level infrastructure supports
466 * returning a specific error on buffer lookup failures.
468 xfs_alert(btp
->bt_mount
,
469 "%s: Block out of range: block 0x%llx, EOFS 0x%llx ",
470 __func__
, blkno
, eofs
);
476 pag
= xfs_perag_get(btp
->bt_mount
,
477 xfs_daddr_to_agno(btp
->bt_mount
, blkno
));
480 spin_lock(&pag
->pag_buf_lock
);
481 rbp
= &pag
->pag_buf_tree
.rb_node
;
486 bp
= rb_entry(parent
, struct xfs_buf
, b_rbnode
);
488 if (blkno
< bp
->b_bn
)
489 rbp
= &(*rbp
)->rb_left
;
490 else if (blkno
> bp
->b_bn
)
491 rbp
= &(*rbp
)->rb_right
;
494 * found a block number match. If the range doesn't
495 * match, the only way this is allowed is if the buffer
496 * in the cache is stale and the transaction that made
497 * it stale has not yet committed. i.e. we are
498 * reallocating a busy extent. Skip this buffer and
499 * continue searching to the right for an exact match.
501 if (bp
->b_length
!= numblks
) {
502 ASSERT(bp
->b_flags
& XBF_STALE
);
503 rbp
= &(*rbp
)->rb_right
;
506 atomic_inc(&bp
->b_hold
);
513 rb_link_node(&new_bp
->b_rbnode
, parent
, rbp
);
514 rb_insert_color(&new_bp
->b_rbnode
, &pag
->pag_buf_tree
);
515 /* the buffer keeps the perag reference until it is freed */
517 spin_unlock(&pag
->pag_buf_lock
);
519 XFS_STATS_INC(xb_miss_locked
);
520 spin_unlock(&pag
->pag_buf_lock
);
526 spin_unlock(&pag
->pag_buf_lock
);
529 if (!xfs_buf_trylock(bp
)) {
530 if (flags
& XBF_TRYLOCK
) {
532 XFS_STATS_INC(xb_busy_locked
);
536 XFS_STATS_INC(xb_get_locked_waited
);
540 * if the buffer is stale, clear all the external state associated with
541 * it. We need to keep flags such as how we allocated the buffer memory
544 if (bp
->b_flags
& XBF_STALE
) {
545 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
546 ASSERT(bp
->b_iodone
== NULL
);
547 bp
->b_flags
&= _XBF_KMEM
| _XBF_PAGES
;
551 trace_xfs_buf_find(bp
, flags
, _RET_IP_
);
552 XFS_STATS_INC(xb_get_locked
);
557 * Assembles a buffer covering the specified range. The code is optimised for
558 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
559 * more hits than misses.
563 struct xfs_buftarg
*target
,
564 struct xfs_buf_map
*map
,
566 xfs_buf_flags_t flags
)
569 struct xfs_buf
*new_bp
;
572 bp
= _xfs_buf_find(target
, map
, nmaps
, flags
, NULL
);
576 new_bp
= _xfs_buf_alloc(target
, map
, nmaps
, flags
);
577 if (unlikely(!new_bp
))
580 error
= xfs_buf_allocate_memory(new_bp
, flags
);
582 xfs_buf_free(new_bp
);
586 bp
= _xfs_buf_find(target
, map
, nmaps
, flags
, new_bp
);
588 xfs_buf_free(new_bp
);
593 xfs_buf_free(new_bp
);
597 error
= _xfs_buf_map_pages(bp
, flags
);
598 if (unlikely(error
)) {
599 xfs_warn(target
->bt_mount
,
600 "%s: failed to map pagesn", __func__
);
606 XFS_STATS_INC(xb_get
);
607 trace_xfs_buf_get(bp
, flags
, _RET_IP_
);
614 xfs_buf_flags_t flags
)
616 ASSERT(!(flags
& XBF_WRITE
));
617 ASSERT(bp
->b_maps
[0].bm_bn
!= XFS_BUF_DADDR_NULL
);
619 bp
->b_flags
&= ~(XBF_WRITE
| XBF_ASYNC
| XBF_READ_AHEAD
);
620 bp
->b_flags
|= flags
& (XBF_READ
| XBF_ASYNC
| XBF_READ_AHEAD
);
622 if (flags
& XBF_ASYNC
) {
626 return xfs_buf_submit_wait(bp
);
631 struct xfs_buftarg
*target
,
632 struct xfs_buf_map
*map
,
634 xfs_buf_flags_t flags
,
635 const struct xfs_buf_ops
*ops
)
641 bp
= xfs_buf_get_map(target
, map
, nmaps
, flags
);
643 trace_xfs_buf_read(bp
, flags
, _RET_IP_
);
645 if (!XFS_BUF_ISDONE(bp
)) {
646 XFS_STATS_INC(xb_get_read
);
648 _xfs_buf_read(bp
, flags
);
649 } else if (flags
& XBF_ASYNC
) {
651 * Read ahead call which is already satisfied,
657 /* We do not want read in the flags */
658 bp
->b_flags
&= ~XBF_READ
;
666 * If we are not low on memory then do the readahead in a deadlock
670 xfs_buf_readahead_map(
671 struct xfs_buftarg
*target
,
672 struct xfs_buf_map
*map
,
674 const struct xfs_buf_ops
*ops
)
676 if (bdi_read_congested(target
->bt_bdi
))
679 xfs_buf_read_map(target
, map
, nmaps
,
680 XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
, ops
);
684 * Read an uncached buffer from disk. Allocates and returns a locked
685 * buffer containing the disk contents or nothing.
688 xfs_buf_read_uncached(
689 struct xfs_buftarg
*target
,
693 struct xfs_buf
**bpp
,
694 const struct xfs_buf_ops
*ops
)
700 bp
= xfs_buf_get_uncached(target
, numblks
, flags
);
704 /* set up the buffer for a read IO */
705 ASSERT(bp
->b_map_count
== 1);
706 bp
->b_bn
= XFS_BUF_DADDR_NULL
; /* always null for uncached buffers */
707 bp
->b_maps
[0].bm_bn
= daddr
;
708 bp
->b_flags
|= XBF_READ
;
711 xfs_buf_submit_wait(bp
);
713 int error
= bp
->b_error
;
723 * Return a buffer allocated as an empty buffer and associated to external
724 * memory via xfs_buf_associate_memory() back to it's empty state.
732 _xfs_buf_free_pages(bp
);
735 bp
->b_page_count
= 0;
737 bp
->b_length
= numblks
;
738 bp
->b_io_length
= numblks
;
740 ASSERT(bp
->b_map_count
== 1);
741 bp
->b_bn
= XFS_BUF_DADDR_NULL
;
742 bp
->b_maps
[0].bm_bn
= XFS_BUF_DADDR_NULL
;
743 bp
->b_maps
[0].bm_len
= bp
->b_length
;
746 static inline struct page
*
750 if ((!is_vmalloc_addr(addr
))) {
751 return virt_to_page(addr
);
753 return vmalloc_to_page(addr
);
758 xfs_buf_associate_memory(
765 unsigned long pageaddr
;
766 unsigned long offset
;
770 pageaddr
= (unsigned long)mem
& PAGE_MASK
;
771 offset
= (unsigned long)mem
- pageaddr
;
772 buflen
= PAGE_ALIGN(len
+ offset
);
773 page_count
= buflen
>> PAGE_SHIFT
;
775 /* Free any previous set of page pointers */
777 _xfs_buf_free_pages(bp
);
782 rval
= _xfs_buf_get_pages(bp
, page_count
);
786 bp
->b_offset
= offset
;
788 for (i
= 0; i
< bp
->b_page_count
; i
++) {
789 bp
->b_pages
[i
] = mem_to_page((void *)pageaddr
);
790 pageaddr
+= PAGE_SIZE
;
793 bp
->b_io_length
= BTOBB(len
);
794 bp
->b_length
= BTOBB(buflen
);
800 xfs_buf_get_uncached(
801 struct xfs_buftarg
*target
,
805 unsigned long page_count
;
808 DEFINE_SINGLE_BUF_MAP(map
, XFS_BUF_DADDR_NULL
, numblks
);
810 bp
= _xfs_buf_alloc(target
, &map
, 1, 0);
811 if (unlikely(bp
== NULL
))
814 page_count
= PAGE_ALIGN(numblks
<< BBSHIFT
) >> PAGE_SHIFT
;
815 error
= _xfs_buf_get_pages(bp
, page_count
);
819 for (i
= 0; i
< page_count
; i
++) {
820 bp
->b_pages
[i
] = alloc_page(xb_to_gfp(flags
));
824 bp
->b_flags
|= _XBF_PAGES
;
826 error
= _xfs_buf_map_pages(bp
, 0);
827 if (unlikely(error
)) {
828 xfs_warn(target
->bt_mount
,
829 "%s: failed to map pages", __func__
);
833 trace_xfs_buf_get_uncached(bp
, _RET_IP_
);
838 __free_page(bp
->b_pages
[i
]);
839 _xfs_buf_free_pages(bp
);
841 xfs_buf_free_maps(bp
);
842 kmem_zone_free(xfs_buf_zone
, bp
);
848 * Increment reference count on buffer, to hold the buffer concurrently
849 * with another thread which may release (free) the buffer asynchronously.
850 * Must hold the buffer already to call this function.
856 trace_xfs_buf_hold(bp
, _RET_IP_
);
857 atomic_inc(&bp
->b_hold
);
861 * Releases a hold on the specified buffer. If the
862 * the hold count is 1, calls xfs_buf_free.
868 struct xfs_perag
*pag
= bp
->b_pag
;
870 trace_xfs_buf_rele(bp
, _RET_IP_
);
873 ASSERT(list_empty(&bp
->b_lru
));
874 ASSERT(RB_EMPTY_NODE(&bp
->b_rbnode
));
875 if (atomic_dec_and_test(&bp
->b_hold
))
880 ASSERT(!RB_EMPTY_NODE(&bp
->b_rbnode
));
882 ASSERT(atomic_read(&bp
->b_hold
) > 0);
883 if (atomic_dec_and_lock(&bp
->b_hold
, &pag
->pag_buf_lock
)) {
884 spin_lock(&bp
->b_lock
);
885 if (!(bp
->b_flags
& XBF_STALE
) && atomic_read(&bp
->b_lru_ref
)) {
887 * If the buffer is added to the LRU take a new
888 * reference to the buffer for the LRU and clear the
889 * (now stale) dispose list state flag
891 if (list_lru_add(&bp
->b_target
->bt_lru
, &bp
->b_lru
)) {
892 bp
->b_state
&= ~XFS_BSTATE_DISPOSE
;
893 atomic_inc(&bp
->b_hold
);
895 spin_unlock(&bp
->b_lock
);
896 spin_unlock(&pag
->pag_buf_lock
);
899 * most of the time buffers will already be removed from
900 * the LRU, so optimise that case by checking for the
901 * XFS_BSTATE_DISPOSE flag indicating the last list the
902 * buffer was on was the disposal list
904 if (!(bp
->b_state
& XFS_BSTATE_DISPOSE
)) {
905 list_lru_del(&bp
->b_target
->bt_lru
, &bp
->b_lru
);
907 ASSERT(list_empty(&bp
->b_lru
));
909 spin_unlock(&bp
->b_lock
);
911 ASSERT(!(bp
->b_flags
& _XBF_DELWRI_Q
));
912 rb_erase(&bp
->b_rbnode
, &pag
->pag_buf_tree
);
913 spin_unlock(&pag
->pag_buf_lock
);
922 * Lock a buffer object, if it is not already locked.
924 * If we come across a stale, pinned, locked buffer, we know that we are
925 * being asked to lock a buffer that has been reallocated. Because it is
926 * pinned, we know that the log has not been pushed to disk and hence it
927 * will still be locked. Rather than continuing to have trylock attempts
928 * fail until someone else pushes the log, push it ourselves before
929 * returning. This means that the xfsaild will not get stuck trying
930 * to push on stale inode buffers.
938 locked
= down_trylock(&bp
->b_sema
) == 0;
942 trace_xfs_buf_trylock(bp
, _RET_IP_
);
947 * Lock a buffer object.
949 * If we come across a stale, pinned, locked buffer, we know that we
950 * are being asked to lock a buffer that has been reallocated. Because
951 * it is pinned, we know that the log has not been pushed to disk and
952 * hence it will still be locked. Rather than sleeping until someone
953 * else pushes the log, push it ourselves before trying to get the lock.
959 trace_xfs_buf_lock(bp
, _RET_IP_
);
961 if (atomic_read(&bp
->b_pin_count
) && (bp
->b_flags
& XBF_STALE
))
962 xfs_log_force(bp
->b_target
->bt_mount
, 0);
966 trace_xfs_buf_lock_done(bp
, _RET_IP_
);
976 trace_xfs_buf_unlock(bp
, _RET_IP_
);
983 DECLARE_WAITQUEUE (wait
, current
);
985 if (atomic_read(&bp
->b_pin_count
) == 0)
988 add_wait_queue(&bp
->b_waiters
, &wait
);
990 set_current_state(TASK_UNINTERRUPTIBLE
);
991 if (atomic_read(&bp
->b_pin_count
) == 0)
995 remove_wait_queue(&bp
->b_waiters
, &wait
);
996 set_current_state(TASK_RUNNING
);
1000 * Buffer Utility Routines
1007 bool read
= bp
->b_flags
& XBF_READ
;
1009 trace_xfs_buf_iodone(bp
, _RET_IP_
);
1011 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_READ_AHEAD
);
1014 * Pull in IO completion errors now. We are guaranteed to be running
1015 * single threaded, so we don't need the lock to read b_io_error.
1017 if (!bp
->b_error
&& bp
->b_io_error
)
1018 xfs_buf_ioerror(bp
, bp
->b_io_error
);
1020 /* Only validate buffers that were read without errors */
1021 if (read
&& !bp
->b_error
&& bp
->b_ops
) {
1022 ASSERT(!bp
->b_iodone
);
1023 bp
->b_ops
->verify_read(bp
);
1027 bp
->b_flags
|= XBF_DONE
;
1030 (*(bp
->b_iodone
))(bp
);
1031 else if (bp
->b_flags
& XBF_ASYNC
)
1034 complete(&bp
->b_iowait
);
1039 struct work_struct
*work
)
1041 struct xfs_buf
*bp
=
1042 container_of(work
, xfs_buf_t
, b_ioend_work
);
1048 xfs_buf_ioend_async(
1051 INIT_WORK(&bp
->b_ioend_work
, xfs_buf_ioend_work
);
1052 queue_work(bp
->b_ioend_wq
, &bp
->b_ioend_work
);
1060 ASSERT(error
<= 0 && error
>= -1000);
1061 bp
->b_error
= error
;
1062 trace_xfs_buf_ioerror(bp
, error
, _RET_IP_
);
1066 xfs_buf_ioerror_alert(
1070 xfs_alert(bp
->b_target
->bt_mount
,
1071 "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d",
1072 (__uint64_t
)XFS_BUF_ADDR(bp
), func
, -bp
->b_error
, bp
->b_length
);
1081 ASSERT(xfs_buf_islocked(bp
));
1083 bp
->b_flags
|= XBF_WRITE
;
1084 bp
->b_flags
&= ~(XBF_ASYNC
| XBF_READ
| _XBF_DELWRI_Q
|
1085 XBF_WRITE_FAIL
| XBF_DONE
);
1087 error
= xfs_buf_submit_wait(bp
);
1089 xfs_force_shutdown(bp
->b_target
->bt_mount
,
1090 SHUTDOWN_META_IO_ERROR
);
1099 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1102 * don't overwrite existing errors - otherwise we can lose errors on
1103 * buffers that require multiple bios to complete.
1105 if (bio
->bi_error
) {
1106 spin_lock(&bp
->b_lock
);
1107 if (!bp
->b_io_error
)
1108 bp
->b_io_error
= bio
->bi_error
;
1109 spin_unlock(&bp
->b_lock
);
1112 if (!bp
->b_error
&& xfs_buf_is_vmapped(bp
) && (bp
->b_flags
& XBF_READ
))
1113 invalidate_kernel_vmap_range(bp
->b_addr
, xfs_buf_vmap_len(bp
));
1115 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1)
1116 xfs_buf_ioend_async(bp
);
1121 xfs_buf_ioapply_map(
1129 int total_nr_pages
= bp
->b_page_count
;
1132 sector_t sector
= bp
->b_maps
[map
].bm_bn
;
1136 total_nr_pages
= bp
->b_page_count
;
1138 /* skip the pages in the buffer before the start offset */
1140 offset
= *buf_offset
;
1141 while (offset
>= PAGE_SIZE
) {
1143 offset
-= PAGE_SIZE
;
1147 * Limit the IO size to the length of the current vector, and update the
1148 * remaining IO count for the next time around.
1150 size
= min_t(int, BBTOB(bp
->b_maps
[map
].bm_len
), *count
);
1152 *buf_offset
+= size
;
1155 atomic_inc(&bp
->b_io_remaining
);
1156 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1157 if (nr_pages
> total_nr_pages
)
1158 nr_pages
= total_nr_pages
;
1160 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1161 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1162 bio
->bi_iter
.bi_sector
= sector
;
1163 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1164 bio
->bi_private
= bp
;
1167 for (; size
&& nr_pages
; nr_pages
--, page_index
++) {
1168 int rbytes
, nbytes
= PAGE_SIZE
- offset
;
1173 rbytes
= bio_add_page(bio
, bp
->b_pages
[page_index
], nbytes
,
1175 if (rbytes
< nbytes
)
1179 sector
+= BTOBB(nbytes
);
1184 if (likely(bio
->bi_iter
.bi_size
)) {
1185 if (xfs_buf_is_vmapped(bp
)) {
1186 flush_kernel_vmap_range(bp
->b_addr
,
1187 xfs_buf_vmap_len(bp
));
1189 submit_bio(rw
, bio
);
1194 * This is guaranteed not to be the last io reference count
1195 * because the caller (xfs_buf_submit) holds a count itself.
1197 atomic_dec(&bp
->b_io_remaining
);
1198 xfs_buf_ioerror(bp
, -EIO
);
1208 struct blk_plug plug
;
1215 * Make sure we capture only current IO errors rather than stale errors
1216 * left over from previous use of the buffer (e.g. failed readahead).
1221 * Initialize the I/O completion workqueue if we haven't yet or the
1222 * submitter has not opted to specify a custom one.
1224 if (!bp
->b_ioend_wq
)
1225 bp
->b_ioend_wq
= bp
->b_target
->bt_mount
->m_buf_workqueue
;
1227 if (bp
->b_flags
& XBF_WRITE
) {
1228 if (bp
->b_flags
& XBF_SYNCIO
)
1232 if (bp
->b_flags
& XBF_FUA
)
1234 if (bp
->b_flags
& XBF_FLUSH
)
1238 * Run the write verifier callback function if it exists. If
1239 * this function fails it will mark the buffer with an error and
1240 * the IO should not be dispatched.
1243 bp
->b_ops
->verify_write(bp
);
1245 xfs_force_shutdown(bp
->b_target
->bt_mount
,
1246 SHUTDOWN_CORRUPT_INCORE
);
1249 } else if (bp
->b_bn
!= XFS_BUF_DADDR_NULL
) {
1250 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
1253 * non-crc filesystems don't attach verifiers during
1254 * log recovery, so don't warn for such filesystems.
1256 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
1258 "%s: no ops on block 0x%llx/0x%x",
1259 __func__
, bp
->b_bn
, bp
->b_length
);
1260 xfs_hex_dump(bp
->b_addr
, 64);
1264 } else if (bp
->b_flags
& XBF_READ_AHEAD
) {
1270 /* we only use the buffer cache for meta-data */
1274 * Walk all the vectors issuing IO on them. Set up the initial offset
1275 * into the buffer and the desired IO size before we start -
1276 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1279 offset
= bp
->b_offset
;
1280 size
= BBTOB(bp
->b_io_length
);
1281 blk_start_plug(&plug
);
1282 for (i
= 0; i
< bp
->b_map_count
; i
++) {
1283 xfs_buf_ioapply_map(bp
, i
, &offset
, &size
, rw
);
1287 break; /* all done */
1289 blk_finish_plug(&plug
);
1293 * Asynchronous IO submission path. This transfers the buffer lock ownership and
1294 * the current reference to the IO. It is not safe to reference the buffer after
1295 * a call to this function unless the caller holds an additional reference
1302 trace_xfs_buf_submit(bp
, _RET_IP_
);
1304 ASSERT(!(bp
->b_flags
& _XBF_DELWRI_Q
));
1305 ASSERT(bp
->b_flags
& XBF_ASYNC
);
1307 /* on shutdown we stale and complete the buffer immediately */
1308 if (XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
1309 xfs_buf_ioerror(bp
, -EIO
);
1310 bp
->b_flags
&= ~XBF_DONE
;
1316 if (bp
->b_flags
& XBF_WRITE
)
1317 xfs_buf_wait_unpin(bp
);
1319 /* clear the internal error state to avoid spurious errors */
1323 * The caller's reference is released during I/O completion.
1324 * This occurs some time after the last b_io_remaining reference is
1325 * released, so after we drop our Io reference we have to have some
1326 * other reference to ensure the buffer doesn't go away from underneath
1327 * us. Take a direct reference to ensure we have safe access to the
1328 * buffer until we are finished with it.
1333 * Set the count to 1 initially, this will stop an I/O completion
1334 * callout which happens before we have started all the I/O from calling
1335 * xfs_buf_ioend too early.
1337 atomic_set(&bp
->b_io_remaining
, 1);
1338 _xfs_buf_ioapply(bp
);
1341 * If _xfs_buf_ioapply failed, we can get back here with only the IO
1342 * reference we took above. If we drop it to zero, run completion so
1343 * that we don't return to the caller with completion still pending.
1345 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1349 xfs_buf_ioend_async(bp
);
1353 /* Note: it is not safe to reference bp now we've dropped our ref */
1357 * Synchronous buffer IO submission path, read or write.
1360 xfs_buf_submit_wait(
1365 trace_xfs_buf_submit_wait(bp
, _RET_IP_
);
1367 ASSERT(!(bp
->b_flags
& (_XBF_DELWRI_Q
| XBF_ASYNC
)));
1369 if (XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
1370 xfs_buf_ioerror(bp
, -EIO
);
1372 bp
->b_flags
&= ~XBF_DONE
;
1376 if (bp
->b_flags
& XBF_WRITE
)
1377 xfs_buf_wait_unpin(bp
);
1379 /* clear the internal error state to avoid spurious errors */
1383 * For synchronous IO, the IO does not inherit the submitters reference
1384 * count, nor the buffer lock. Hence we cannot release the reference we
1385 * are about to take until we've waited for all IO completion to occur,
1386 * including any xfs_buf_ioend_async() work that may be pending.
1391 * Set the count to 1 initially, this will stop an I/O completion
1392 * callout which happens before we have started all the I/O from calling
1393 * xfs_buf_ioend too early.
1395 atomic_set(&bp
->b_io_remaining
, 1);
1396 _xfs_buf_ioapply(bp
);
1399 * make sure we run completion synchronously if it raced with us and is
1402 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1)
1405 /* wait for completion before gathering the error from the buffer */
1406 trace_xfs_buf_iowait(bp
, _RET_IP_
);
1407 wait_for_completion(&bp
->b_iowait
);
1408 trace_xfs_buf_iowait_done(bp
, _RET_IP_
);
1409 error
= bp
->b_error
;
1412 * all done now, we can release the hold that keeps the buffer
1413 * referenced for the entire IO.
1427 return bp
->b_addr
+ offset
;
1429 offset
+= bp
->b_offset
;
1430 page
= bp
->b_pages
[offset
>> PAGE_SHIFT
];
1431 return page_address(page
) + (offset
& (PAGE_SIZE
-1));
1435 * Move data into or out of a buffer.
1439 xfs_buf_t
*bp
, /* buffer to process */
1440 size_t boff
, /* starting buffer offset */
1441 size_t bsize
, /* length to copy */
1442 void *data
, /* data address */
1443 xfs_buf_rw_t mode
) /* read/write/zero flag */
1447 bend
= boff
+ bsize
;
1448 while (boff
< bend
) {
1450 int page_index
, page_offset
, csize
;
1452 page_index
= (boff
+ bp
->b_offset
) >> PAGE_SHIFT
;
1453 page_offset
= (boff
+ bp
->b_offset
) & ~PAGE_MASK
;
1454 page
= bp
->b_pages
[page_index
];
1455 csize
= min_t(size_t, PAGE_SIZE
- page_offset
,
1456 BBTOB(bp
->b_io_length
) - boff
);
1458 ASSERT((csize
+ page_offset
) <= PAGE_SIZE
);
1462 memset(page_address(page
) + page_offset
, 0, csize
);
1465 memcpy(data
, page_address(page
) + page_offset
, csize
);
1468 memcpy(page_address(page
) + page_offset
, data
, csize
);
1477 * Handling of buffer targets (buftargs).
1481 * Wait for any bufs with callbacks that have been submitted but have not yet
1482 * returned. These buffers will have an elevated hold count, so wait on those
1483 * while freeing all the buffers only held by the LRU.
1485 static enum lru_status
1486 xfs_buftarg_wait_rele(
1487 struct list_head
*item
,
1488 struct list_lru_one
*lru
,
1489 spinlock_t
*lru_lock
,
1493 struct xfs_buf
*bp
= container_of(item
, struct xfs_buf
, b_lru
);
1494 struct list_head
*dispose
= arg
;
1496 if (atomic_read(&bp
->b_hold
) > 1) {
1497 /* need to wait, so skip it this pass */
1498 trace_xfs_buf_wait_buftarg(bp
, _RET_IP_
);
1501 if (!spin_trylock(&bp
->b_lock
))
1505 * clear the LRU reference count so the buffer doesn't get
1506 * ignored in xfs_buf_rele().
1508 atomic_set(&bp
->b_lru_ref
, 0);
1509 bp
->b_state
|= XFS_BSTATE_DISPOSE
;
1510 list_lru_isolate_move(lru
, item
, dispose
);
1511 spin_unlock(&bp
->b_lock
);
1517 struct xfs_buftarg
*btp
)
1522 /* loop until there is nothing left on the lru list. */
1523 while (list_lru_count(&btp
->bt_lru
)) {
1524 list_lru_walk(&btp
->bt_lru
, xfs_buftarg_wait_rele
,
1525 &dispose
, LONG_MAX
);
1527 while (!list_empty(&dispose
)) {
1529 bp
= list_first_entry(&dispose
, struct xfs_buf
, b_lru
);
1530 list_del_init(&bp
->b_lru
);
1531 if (bp
->b_flags
& XBF_WRITE_FAIL
) {
1532 xfs_alert(btp
->bt_mount
,
1533 "Corruption Alert: Buffer at block 0x%llx had permanent write failures!",
1534 (long long)bp
->b_bn
);
1535 xfs_alert(btp
->bt_mount
,
1536 "Please run xfs_repair to determine the extent of the problem.");
1545 static enum lru_status
1546 xfs_buftarg_isolate(
1547 struct list_head
*item
,
1548 struct list_lru_one
*lru
,
1549 spinlock_t
*lru_lock
,
1552 struct xfs_buf
*bp
= container_of(item
, struct xfs_buf
, b_lru
);
1553 struct list_head
*dispose
= arg
;
1556 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1557 * If we fail to get the lock, just skip it.
1559 if (!spin_trylock(&bp
->b_lock
))
1562 * Decrement the b_lru_ref count unless the value is already
1563 * zero. If the value is already zero, we need to reclaim the
1564 * buffer, otherwise it gets another trip through the LRU.
1566 if (!atomic_add_unless(&bp
->b_lru_ref
, -1, 0)) {
1567 spin_unlock(&bp
->b_lock
);
1571 bp
->b_state
|= XFS_BSTATE_DISPOSE
;
1572 list_lru_isolate_move(lru
, item
, dispose
);
1573 spin_unlock(&bp
->b_lock
);
1577 static unsigned long
1578 xfs_buftarg_shrink_scan(
1579 struct shrinker
*shrink
,
1580 struct shrink_control
*sc
)
1582 struct xfs_buftarg
*btp
= container_of(shrink
,
1583 struct xfs_buftarg
, bt_shrinker
);
1585 unsigned long freed
;
1587 freed
= list_lru_shrink_walk(&btp
->bt_lru
, sc
,
1588 xfs_buftarg_isolate
, &dispose
);
1590 while (!list_empty(&dispose
)) {
1592 bp
= list_first_entry(&dispose
, struct xfs_buf
, b_lru
);
1593 list_del_init(&bp
->b_lru
);
1600 static unsigned long
1601 xfs_buftarg_shrink_count(
1602 struct shrinker
*shrink
,
1603 struct shrink_control
*sc
)
1605 struct xfs_buftarg
*btp
= container_of(shrink
,
1606 struct xfs_buftarg
, bt_shrinker
);
1607 return list_lru_shrink_count(&btp
->bt_lru
, sc
);
1612 struct xfs_mount
*mp
,
1613 struct xfs_buftarg
*btp
)
1615 unregister_shrinker(&btp
->bt_shrinker
);
1616 list_lru_destroy(&btp
->bt_lru
);
1618 if (mp
->m_flags
& XFS_MOUNT_BARRIER
)
1619 xfs_blkdev_issue_flush(btp
);
1625 xfs_setsize_buftarg(
1627 unsigned int sectorsize
)
1629 /* Set up metadata sector size info */
1630 btp
->bt_meta_sectorsize
= sectorsize
;
1631 btp
->bt_meta_sectormask
= sectorsize
- 1;
1633 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1634 char name
[BDEVNAME_SIZE
];
1636 bdevname(btp
->bt_bdev
, name
);
1638 xfs_warn(btp
->bt_mount
,
1639 "Cannot set_blocksize to %u on device %s",
1644 /* Set up device logical sector size mask */
1645 btp
->bt_logical_sectorsize
= bdev_logical_block_size(btp
->bt_bdev
);
1646 btp
->bt_logical_sectormask
= bdev_logical_block_size(btp
->bt_bdev
) - 1;
1652 * When allocating the initial buffer target we have not yet
1653 * read in the superblock, so don't know what sized sectors
1654 * are being used at this early stage. Play safe.
1657 xfs_setsize_buftarg_early(
1659 struct block_device
*bdev
)
1661 return xfs_setsize_buftarg(btp
, bdev_logical_block_size(bdev
));
1666 struct xfs_mount
*mp
,
1667 struct block_device
*bdev
)
1671 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
| KM_NOFS
);
1674 btp
->bt_dev
= bdev
->bd_dev
;
1675 btp
->bt_bdev
= bdev
;
1676 btp
->bt_bdi
= blk_get_backing_dev_info(bdev
);
1678 if (xfs_setsize_buftarg_early(btp
, bdev
))
1681 if (list_lru_init(&btp
->bt_lru
))
1684 btp
->bt_shrinker
.count_objects
= xfs_buftarg_shrink_count
;
1685 btp
->bt_shrinker
.scan_objects
= xfs_buftarg_shrink_scan
;
1686 btp
->bt_shrinker
.seeks
= DEFAULT_SEEKS
;
1687 btp
->bt_shrinker
.flags
= SHRINKER_NUMA_AWARE
;
1688 register_shrinker(&btp
->bt_shrinker
);
1697 * Add a buffer to the delayed write list.
1699 * This queues a buffer for writeout if it hasn't already been. Note that
1700 * neither this routine nor the buffer list submission functions perform
1701 * any internal synchronization. It is expected that the lists are thread-local
1704 * Returns true if we queued up the buffer, or false if it already had
1705 * been on the buffer list.
1708 xfs_buf_delwri_queue(
1710 struct list_head
*list
)
1712 ASSERT(xfs_buf_islocked(bp
));
1713 ASSERT(!(bp
->b_flags
& XBF_READ
));
1716 * If the buffer is already marked delwri it already is queued up
1717 * by someone else for imediate writeout. Just ignore it in that
1720 if (bp
->b_flags
& _XBF_DELWRI_Q
) {
1721 trace_xfs_buf_delwri_queued(bp
, _RET_IP_
);
1725 trace_xfs_buf_delwri_queue(bp
, _RET_IP_
);
1728 * If a buffer gets written out synchronously or marked stale while it
1729 * is on a delwri list we lazily remove it. To do this, the other party
1730 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1731 * It remains referenced and on the list. In a rare corner case it
1732 * might get readded to a delwri list after the synchronous writeout, in
1733 * which case we need just need to re-add the flag here.
1735 bp
->b_flags
|= _XBF_DELWRI_Q
;
1736 if (list_empty(&bp
->b_list
)) {
1737 atomic_inc(&bp
->b_hold
);
1738 list_add_tail(&bp
->b_list
, list
);
1745 * Compare function is more complex than it needs to be because
1746 * the return value is only 32 bits and we are doing comparisons
1752 struct list_head
*a
,
1753 struct list_head
*b
)
1755 struct xfs_buf
*ap
= container_of(a
, struct xfs_buf
, b_list
);
1756 struct xfs_buf
*bp
= container_of(b
, struct xfs_buf
, b_list
);
1759 diff
= ap
->b_maps
[0].bm_bn
- bp
->b_maps
[0].bm_bn
;
1768 __xfs_buf_delwri_submit(
1769 struct list_head
*buffer_list
,
1770 struct list_head
*io_list
,
1773 struct blk_plug plug
;
1774 struct xfs_buf
*bp
, *n
;
1777 list_for_each_entry_safe(bp
, n
, buffer_list
, b_list
) {
1779 if (xfs_buf_ispinned(bp
)) {
1783 if (!xfs_buf_trylock(bp
))
1790 * Someone else might have written the buffer synchronously or
1791 * marked it stale in the meantime. In that case only the
1792 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
1793 * reference and remove it from the list here.
1795 if (!(bp
->b_flags
& _XBF_DELWRI_Q
)) {
1796 list_del_init(&bp
->b_list
);
1801 list_move_tail(&bp
->b_list
, io_list
);
1802 trace_xfs_buf_delwri_split(bp
, _RET_IP_
);
1805 list_sort(NULL
, io_list
, xfs_buf_cmp
);
1807 blk_start_plug(&plug
);
1808 list_for_each_entry_safe(bp
, n
, io_list
, b_list
) {
1809 bp
->b_flags
&= ~(_XBF_DELWRI_Q
| XBF_ASYNC
| XBF_WRITE_FAIL
);
1810 bp
->b_flags
|= XBF_WRITE
| XBF_ASYNC
;
1813 * we do all Io submission async. This means if we need to wait
1814 * for IO completion we need to take an extra reference so the
1815 * buffer is still valid on the other side.
1820 list_del_init(&bp
->b_list
);
1824 blk_finish_plug(&plug
);
1830 * Write out a buffer list asynchronously.
1832 * This will take the @buffer_list, write all non-locked and non-pinned buffers
1833 * out and not wait for I/O completion on any of the buffers. This interface
1834 * is only safely useable for callers that can track I/O completion by higher
1835 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
1839 xfs_buf_delwri_submit_nowait(
1840 struct list_head
*buffer_list
)
1842 LIST_HEAD (io_list
);
1843 return __xfs_buf_delwri_submit(buffer_list
, &io_list
, false);
1847 * Write out a buffer list synchronously.
1849 * This will take the @buffer_list, write all buffers out and wait for I/O
1850 * completion on all of the buffers. @buffer_list is consumed by the function,
1851 * so callers must have some other way of tracking buffers if they require such
1855 xfs_buf_delwri_submit(
1856 struct list_head
*buffer_list
)
1858 LIST_HEAD (io_list
);
1859 int error
= 0, error2
;
1862 __xfs_buf_delwri_submit(buffer_list
, &io_list
, true);
1864 /* Wait for IO to complete. */
1865 while (!list_empty(&io_list
)) {
1866 bp
= list_first_entry(&io_list
, struct xfs_buf
, b_list
);
1868 list_del_init(&bp
->b_list
);
1870 /* locking the buffer will wait for async IO completion. */
1872 error2
= bp
->b_error
;
1884 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1885 KM_ZONE_HWALIGN
, NULL
);
1896 xfs_buf_terminate(void)
1898 kmem_zone_destroy(xfs_buf_zone
);
This page took 0.109899 seconds and 6 git commands to generate.