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_log_format.h"
38 #include "xfs_trans_resv.h"
41 #include "xfs_mount.h"
42 #include "xfs_trace.h"
45 static kmem_zone_t
*xfs_buf_zone
;
47 static struct workqueue_struct
*xfslogd_workqueue
;
49 #ifdef XFS_BUF_LOCK_TRACKING
50 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
51 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
52 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
54 # define XB_SET_OWNER(bp) do { } while (0)
55 # define XB_CLEAR_OWNER(bp) do { } while (0)
56 # define XB_GET_OWNER(bp) do { } while (0)
59 #define xb_to_gfp(flags) \
60 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN)
68 * Return true if the buffer is vmapped.
70 * b_addr is null if the buffer is not mapped, but the code is clever
71 * enough to know it doesn't have to map a single page, so the check has
72 * to be both for b_addr and bp->b_page_count > 1.
74 return bp
->b_addr
&& bp
->b_page_count
> 1;
81 return (bp
->b_page_count
* PAGE_SIZE
) - bp
->b_offset
;
85 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
86 * b_lru_ref count so that the buffer is freed immediately when the buffer
87 * reference count falls to zero. If the buffer is already on the LRU, we need
88 * to remove the reference that LRU holds on the buffer.
90 * This prevents build-up of stale buffers on the LRU.
96 ASSERT(xfs_buf_islocked(bp
));
98 bp
->b_flags
|= XBF_STALE
;
101 * Clear the delwri status so that a delwri queue walker will not
102 * flush this buffer to disk now that it is stale. The delwri queue has
103 * a reference to the buffer, so this is safe to do.
105 bp
->b_flags
&= ~_XBF_DELWRI_Q
;
107 spin_lock(&bp
->b_lock
);
108 atomic_set(&bp
->b_lru_ref
, 0);
109 if (!(bp
->b_state
& XFS_BSTATE_DISPOSE
) &&
110 (list_lru_del(&bp
->b_target
->bt_lru
, &bp
->b_lru
)))
111 atomic_dec(&bp
->b_hold
);
113 ASSERT(atomic_read(&bp
->b_hold
) >= 1);
114 spin_unlock(&bp
->b_lock
);
122 ASSERT(bp
->b_maps
== NULL
);
123 bp
->b_map_count
= map_count
;
125 if (map_count
== 1) {
126 bp
->b_maps
= &bp
->__b_map
;
130 bp
->b_maps
= kmem_zalloc(map_count
* sizeof(struct xfs_buf_map
),
138 * Frees b_pages if it was allocated.
144 if (bp
->b_maps
!= &bp
->__b_map
) {
145 kmem_free(bp
->b_maps
);
152 struct xfs_buftarg
*target
,
153 struct xfs_buf_map
*map
,
155 xfs_buf_flags_t flags
)
161 bp
= kmem_zone_zalloc(xfs_buf_zone
, KM_NOFS
);
166 * We don't want certain flags to appear in b_flags unless they are
167 * specifically set by later operations on the buffer.
169 flags
&= ~(XBF_UNMAPPED
| XBF_TRYLOCK
| XBF_ASYNC
| XBF_READ_AHEAD
);
171 atomic_set(&bp
->b_hold
, 1);
172 atomic_set(&bp
->b_lru_ref
, 1);
173 init_completion(&bp
->b_iowait
);
174 INIT_LIST_HEAD(&bp
->b_lru
);
175 INIT_LIST_HEAD(&bp
->b_list
);
176 RB_CLEAR_NODE(&bp
->b_rbnode
);
177 sema_init(&bp
->b_sema
, 0); /* held, no waiters */
178 spin_lock_init(&bp
->b_lock
);
180 bp
->b_target
= target
;
184 * Set length and io_length to the same value initially.
185 * I/O routines should use io_length, which will be the same in
186 * most cases but may be reset (e.g. XFS recovery).
188 error
= xfs_buf_get_maps(bp
, nmaps
);
190 kmem_zone_free(xfs_buf_zone
, bp
);
194 bp
->b_bn
= map
[0].bm_bn
;
196 for (i
= 0; i
< nmaps
; i
++) {
197 bp
->b_maps
[i
].bm_bn
= map
[i
].bm_bn
;
198 bp
->b_maps
[i
].bm_len
= map
[i
].bm_len
;
199 bp
->b_length
+= map
[i
].bm_len
;
201 bp
->b_io_length
= bp
->b_length
;
203 atomic_set(&bp
->b_pin_count
, 0);
204 init_waitqueue_head(&bp
->b_waiters
);
206 XFS_STATS_INC(xb_create
);
207 trace_xfs_buf_init(bp
, _RET_IP_
);
213 * Allocate a page array capable of holding a specified number
214 * of pages, and point the page buf at it.
220 xfs_buf_flags_t flags
)
222 /* Make sure that we have a page list */
223 if (bp
->b_pages
== NULL
) {
224 bp
->b_page_count
= page_count
;
225 if (page_count
<= XB_PAGES
) {
226 bp
->b_pages
= bp
->b_page_array
;
228 bp
->b_pages
= kmem_alloc(sizeof(struct page
*) *
229 page_count
, KM_NOFS
);
230 if (bp
->b_pages
== NULL
)
233 memset(bp
->b_pages
, 0, sizeof(struct page
*) * page_count
);
239 * Frees b_pages if it was allocated.
245 if (bp
->b_pages
!= bp
->b_page_array
) {
246 kmem_free(bp
->b_pages
);
252 * Releases the specified buffer.
254 * The modification state of any associated pages is left unchanged.
255 * The buffer must not be on any hash - use xfs_buf_rele instead for
256 * hashed and refcounted buffers
262 trace_xfs_buf_free(bp
, _RET_IP_
);
264 ASSERT(list_empty(&bp
->b_lru
));
266 if (bp
->b_flags
& _XBF_PAGES
) {
269 if (xfs_buf_is_vmapped(bp
))
270 vm_unmap_ram(bp
->b_addr
- bp
->b_offset
,
273 for (i
= 0; i
< bp
->b_page_count
; i
++) {
274 struct page
*page
= bp
->b_pages
[i
];
278 } else if (bp
->b_flags
& _XBF_KMEM
)
279 kmem_free(bp
->b_addr
);
280 _xfs_buf_free_pages(bp
);
281 xfs_buf_free_maps(bp
);
282 kmem_zone_free(xfs_buf_zone
, bp
);
286 * Allocates all the pages for buffer in question and builds it's page list.
289 xfs_buf_allocate_memory(
294 size_t nbytes
, offset
;
295 gfp_t gfp_mask
= xb_to_gfp(flags
);
296 unsigned short page_count
, i
;
297 xfs_off_t start
, end
;
301 * for buffers that are contained within a single page, just allocate
302 * the memory from the heap - there's no need for the complexity of
303 * page arrays to keep allocation down to order 0.
305 size
= BBTOB(bp
->b_length
);
306 if (size
< PAGE_SIZE
) {
307 bp
->b_addr
= kmem_alloc(size
, KM_NOFS
);
309 /* low memory - use alloc_page loop instead */
313 if (((unsigned long)(bp
->b_addr
+ size
- 1) & PAGE_MASK
) !=
314 ((unsigned long)bp
->b_addr
& PAGE_MASK
)) {
315 /* b_addr spans two pages - use alloc_page instead */
316 kmem_free(bp
->b_addr
);
320 bp
->b_offset
= offset_in_page(bp
->b_addr
);
321 bp
->b_pages
= bp
->b_page_array
;
322 bp
->b_pages
[0] = virt_to_page(bp
->b_addr
);
323 bp
->b_page_count
= 1;
324 bp
->b_flags
|= _XBF_KMEM
;
329 start
= BBTOB(bp
->b_maps
[0].bm_bn
) >> PAGE_SHIFT
;
330 end
= (BBTOB(bp
->b_maps
[0].bm_bn
+ bp
->b_length
) + PAGE_SIZE
- 1)
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
,
428 struct xfs_buf_map
*map
,
430 xfs_buf_flags_t flags
,
434 struct xfs_perag
*pag
;
435 struct rb_node
**rbp
;
436 struct rb_node
*parent
;
438 xfs_daddr_t blkno
= map
[0].bm_bn
;
443 for (i
= 0; i
< nmaps
; i
++)
444 numblks
+= map
[i
].bm_len
;
445 numbytes
= BBTOB(numblks
);
447 /* Check for IOs smaller than the sector size / not sector aligned */
448 ASSERT(!(numbytes
< (1 << btp
->bt_sshift
)));
449 ASSERT(!(BBTOB(blkno
) & (xfs_off_t
)btp
->bt_smask
));
452 * Corrupted block numbers can get through to here, unfortunately, so we
453 * have to check that the buffer falls within the filesystem bounds.
455 eofs
= XFS_FSB_TO_BB(btp
->bt_mount
, btp
->bt_mount
->m_sb
.sb_dblocks
);
458 * XXX (dgc): we should really be returning EFSCORRUPTED here,
459 * but none of the higher level infrastructure supports
460 * returning a specific error on buffer lookup failures.
462 xfs_alert(btp
->bt_mount
,
463 "%s: Block out of range: block 0x%llx, EOFS 0x%llx ",
464 __func__
, blkno
, eofs
);
470 pag
= xfs_perag_get(btp
->bt_mount
,
471 xfs_daddr_to_agno(btp
->bt_mount
, blkno
));
474 spin_lock(&pag
->pag_buf_lock
);
475 rbp
= &pag
->pag_buf_tree
.rb_node
;
480 bp
= rb_entry(parent
, struct xfs_buf
, b_rbnode
);
482 if (blkno
< bp
->b_bn
)
483 rbp
= &(*rbp
)->rb_left
;
484 else if (blkno
> bp
->b_bn
)
485 rbp
= &(*rbp
)->rb_right
;
488 * found a block number match. If the range doesn't
489 * match, the only way this is allowed is if the buffer
490 * in the cache is stale and the transaction that made
491 * it stale has not yet committed. i.e. we are
492 * reallocating a busy extent. Skip this buffer and
493 * continue searching to the right for an exact match.
495 if (bp
->b_length
!= numblks
) {
496 ASSERT(bp
->b_flags
& XBF_STALE
);
497 rbp
= &(*rbp
)->rb_right
;
500 atomic_inc(&bp
->b_hold
);
507 rb_link_node(&new_bp
->b_rbnode
, parent
, rbp
);
508 rb_insert_color(&new_bp
->b_rbnode
, &pag
->pag_buf_tree
);
509 /* the buffer keeps the perag reference until it is freed */
511 spin_unlock(&pag
->pag_buf_lock
);
513 XFS_STATS_INC(xb_miss_locked
);
514 spin_unlock(&pag
->pag_buf_lock
);
520 spin_unlock(&pag
->pag_buf_lock
);
523 if (!xfs_buf_trylock(bp
)) {
524 if (flags
& XBF_TRYLOCK
) {
526 XFS_STATS_INC(xb_busy_locked
);
530 XFS_STATS_INC(xb_get_locked_waited
);
534 * if the buffer is stale, clear all the external state associated with
535 * it. We need to keep flags such as how we allocated the buffer memory
538 if (bp
->b_flags
& XBF_STALE
) {
539 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
540 ASSERT(bp
->b_iodone
== NULL
);
541 bp
->b_flags
&= _XBF_KMEM
| _XBF_PAGES
;
545 trace_xfs_buf_find(bp
, flags
, _RET_IP_
);
546 XFS_STATS_INC(xb_get_locked
);
551 * Assembles a buffer covering the specified range. The code is optimised for
552 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
553 * more hits than misses.
557 struct xfs_buftarg
*target
,
558 struct xfs_buf_map
*map
,
560 xfs_buf_flags_t flags
)
563 struct xfs_buf
*new_bp
;
566 bp
= _xfs_buf_find(target
, map
, nmaps
, flags
, NULL
);
570 new_bp
= _xfs_buf_alloc(target
, map
, nmaps
, flags
);
571 if (unlikely(!new_bp
))
574 error
= xfs_buf_allocate_memory(new_bp
, flags
);
576 xfs_buf_free(new_bp
);
580 bp
= _xfs_buf_find(target
, map
, nmaps
, flags
, new_bp
);
582 xfs_buf_free(new_bp
);
587 xfs_buf_free(new_bp
);
591 error
= _xfs_buf_map_pages(bp
, flags
);
592 if (unlikely(error
)) {
593 xfs_warn(target
->bt_mount
,
594 "%s: failed to map pagesn", __func__
);
600 XFS_STATS_INC(xb_get
);
601 trace_xfs_buf_get(bp
, flags
, _RET_IP_
);
608 xfs_buf_flags_t flags
)
610 ASSERT(!(flags
& XBF_WRITE
));
611 ASSERT(bp
->b_maps
[0].bm_bn
!= XFS_BUF_DADDR_NULL
);
613 bp
->b_flags
&= ~(XBF_WRITE
| XBF_ASYNC
| XBF_READ_AHEAD
);
614 bp
->b_flags
|= flags
& (XBF_READ
| XBF_ASYNC
| XBF_READ_AHEAD
);
616 xfs_buf_iorequest(bp
);
617 if (flags
& XBF_ASYNC
)
619 return xfs_buf_iowait(bp
);
624 struct xfs_buftarg
*target
,
625 struct xfs_buf_map
*map
,
627 xfs_buf_flags_t flags
,
628 const struct xfs_buf_ops
*ops
)
634 bp
= xfs_buf_get_map(target
, map
, nmaps
, flags
);
636 trace_xfs_buf_read(bp
, flags
, _RET_IP_
);
638 if (!XFS_BUF_ISDONE(bp
)) {
639 XFS_STATS_INC(xb_get_read
);
641 _xfs_buf_read(bp
, flags
);
642 } else if (flags
& XBF_ASYNC
) {
644 * Read ahead call which is already satisfied,
650 /* We do not want read in the flags */
651 bp
->b_flags
&= ~XBF_READ
;
659 * If we are not low on memory then do the readahead in a deadlock
663 xfs_buf_readahead_map(
664 struct xfs_buftarg
*target
,
665 struct xfs_buf_map
*map
,
667 const struct xfs_buf_ops
*ops
)
669 if (bdi_read_congested(target
->bt_bdi
))
672 xfs_buf_read_map(target
, map
, nmaps
,
673 XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
, ops
);
677 * Read an uncached buffer from disk. Allocates and returns a locked
678 * buffer containing the disk contents or nothing.
681 xfs_buf_read_uncached(
682 struct xfs_buftarg
*target
,
686 const struct xfs_buf_ops
*ops
)
690 bp
= xfs_buf_get_uncached(target
, numblks
, flags
);
694 /* set up the buffer for a read IO */
695 ASSERT(bp
->b_map_count
== 1);
697 bp
->b_maps
[0].bm_bn
= daddr
;
698 bp
->b_flags
|= XBF_READ
;
701 if (XFS_FORCED_SHUTDOWN(target
->bt_mount
)) {
705 xfs_buf_iorequest(bp
);
711 * Return a buffer allocated as an empty buffer and associated to external
712 * memory via xfs_buf_associate_memory() back to it's empty state.
720 _xfs_buf_free_pages(bp
);
723 bp
->b_page_count
= 0;
725 bp
->b_length
= numblks
;
726 bp
->b_io_length
= numblks
;
728 ASSERT(bp
->b_map_count
== 1);
729 bp
->b_bn
= XFS_BUF_DADDR_NULL
;
730 bp
->b_maps
[0].bm_bn
= XFS_BUF_DADDR_NULL
;
731 bp
->b_maps
[0].bm_len
= bp
->b_length
;
734 static inline struct page
*
738 if ((!is_vmalloc_addr(addr
))) {
739 return virt_to_page(addr
);
741 return vmalloc_to_page(addr
);
746 xfs_buf_associate_memory(
753 unsigned long pageaddr
;
754 unsigned long offset
;
758 pageaddr
= (unsigned long)mem
& PAGE_MASK
;
759 offset
= (unsigned long)mem
- pageaddr
;
760 buflen
= PAGE_ALIGN(len
+ offset
);
761 page_count
= buflen
>> PAGE_SHIFT
;
763 /* Free any previous set of page pointers */
765 _xfs_buf_free_pages(bp
);
770 rval
= _xfs_buf_get_pages(bp
, page_count
, 0);
774 bp
->b_offset
= offset
;
776 for (i
= 0; i
< bp
->b_page_count
; i
++) {
777 bp
->b_pages
[i
] = mem_to_page((void *)pageaddr
);
778 pageaddr
+= PAGE_SIZE
;
781 bp
->b_io_length
= BTOBB(len
);
782 bp
->b_length
= BTOBB(buflen
);
788 xfs_buf_get_uncached(
789 struct xfs_buftarg
*target
,
793 unsigned long page_count
;
796 DEFINE_SINGLE_BUF_MAP(map
, XFS_BUF_DADDR_NULL
, numblks
);
798 bp
= _xfs_buf_alloc(target
, &map
, 1, 0);
799 if (unlikely(bp
== NULL
))
802 page_count
= PAGE_ALIGN(numblks
<< BBSHIFT
) >> PAGE_SHIFT
;
803 error
= _xfs_buf_get_pages(bp
, page_count
, 0);
807 for (i
= 0; i
< page_count
; i
++) {
808 bp
->b_pages
[i
] = alloc_page(xb_to_gfp(flags
));
812 bp
->b_flags
|= _XBF_PAGES
;
814 error
= _xfs_buf_map_pages(bp
, 0);
815 if (unlikely(error
)) {
816 xfs_warn(target
->bt_mount
,
817 "%s: failed to map pages", __func__
);
821 trace_xfs_buf_get_uncached(bp
, _RET_IP_
);
826 __free_page(bp
->b_pages
[i
]);
827 _xfs_buf_free_pages(bp
);
829 xfs_buf_free_maps(bp
);
830 kmem_zone_free(xfs_buf_zone
, bp
);
836 * Increment reference count on buffer, to hold the buffer concurrently
837 * with another thread which may release (free) the buffer asynchronously.
838 * Must hold the buffer already to call this function.
844 trace_xfs_buf_hold(bp
, _RET_IP_
);
845 atomic_inc(&bp
->b_hold
);
849 * Releases a hold on the specified buffer. If the
850 * the hold count is 1, calls xfs_buf_free.
856 struct xfs_perag
*pag
= bp
->b_pag
;
858 trace_xfs_buf_rele(bp
, _RET_IP_
);
861 ASSERT(list_empty(&bp
->b_lru
));
862 ASSERT(RB_EMPTY_NODE(&bp
->b_rbnode
));
863 if (atomic_dec_and_test(&bp
->b_hold
))
868 ASSERT(!RB_EMPTY_NODE(&bp
->b_rbnode
));
870 ASSERT(atomic_read(&bp
->b_hold
) > 0);
871 if (atomic_dec_and_lock(&bp
->b_hold
, &pag
->pag_buf_lock
)) {
872 spin_lock(&bp
->b_lock
);
873 if (!(bp
->b_flags
& XBF_STALE
) && atomic_read(&bp
->b_lru_ref
)) {
875 * If the buffer is added to the LRU take a new
876 * reference to the buffer for the LRU and clear the
877 * (now stale) dispose list state flag
879 if (list_lru_add(&bp
->b_target
->bt_lru
, &bp
->b_lru
)) {
880 bp
->b_state
&= ~XFS_BSTATE_DISPOSE
;
881 atomic_inc(&bp
->b_hold
);
883 spin_unlock(&bp
->b_lock
);
884 spin_unlock(&pag
->pag_buf_lock
);
887 * most of the time buffers will already be removed from
888 * the LRU, so optimise that case by checking for the
889 * XFS_BSTATE_DISPOSE flag indicating the last list the
890 * buffer was on was the disposal list
892 if (!(bp
->b_state
& XFS_BSTATE_DISPOSE
)) {
893 list_lru_del(&bp
->b_target
->bt_lru
, &bp
->b_lru
);
895 ASSERT(list_empty(&bp
->b_lru
));
897 spin_unlock(&bp
->b_lock
);
899 ASSERT(!(bp
->b_flags
& _XBF_DELWRI_Q
));
900 rb_erase(&bp
->b_rbnode
, &pag
->pag_buf_tree
);
901 spin_unlock(&pag
->pag_buf_lock
);
910 * Lock a buffer object, if it is not already locked.
912 * If we come across a stale, pinned, locked buffer, we know that we are
913 * being asked to lock a buffer that has been reallocated. Because it is
914 * pinned, we know that the log has not been pushed to disk and hence it
915 * will still be locked. Rather than continuing to have trylock attempts
916 * fail until someone else pushes the log, push it ourselves before
917 * returning. This means that the xfsaild will not get stuck trying
918 * to push on stale inode buffers.
926 locked
= down_trylock(&bp
->b_sema
) == 0;
930 trace_xfs_buf_trylock(bp
, _RET_IP_
);
935 * Lock a buffer object.
937 * If we come across a stale, pinned, locked buffer, we know that we
938 * are being asked to lock a buffer that has been reallocated. Because
939 * it is pinned, we know that the log has not been pushed to disk and
940 * hence it will still be locked. Rather than sleeping until someone
941 * else pushes the log, push it ourselves before trying to get the lock.
947 trace_xfs_buf_lock(bp
, _RET_IP_
);
949 if (atomic_read(&bp
->b_pin_count
) && (bp
->b_flags
& XBF_STALE
))
950 xfs_log_force(bp
->b_target
->bt_mount
, 0);
954 trace_xfs_buf_lock_done(bp
, _RET_IP_
);
964 trace_xfs_buf_unlock(bp
, _RET_IP_
);
971 DECLARE_WAITQUEUE (wait
, current
);
973 if (atomic_read(&bp
->b_pin_count
) == 0)
976 add_wait_queue(&bp
->b_waiters
, &wait
);
978 set_current_state(TASK_UNINTERRUPTIBLE
);
979 if (atomic_read(&bp
->b_pin_count
) == 0)
983 remove_wait_queue(&bp
->b_waiters
, &wait
);
984 set_current_state(TASK_RUNNING
);
988 * Buffer Utility Routines
993 struct work_struct
*work
)
996 container_of(work
, xfs_buf_t
, b_iodone_work
);
997 bool read
= !!(bp
->b_flags
& XBF_READ
);
999 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_READ_AHEAD
);
1001 /* only validate buffers that were read without errors */
1002 if (read
&& bp
->b_ops
&& !bp
->b_error
&& (bp
->b_flags
& XBF_DONE
))
1003 bp
->b_ops
->verify_read(bp
);
1006 (*(bp
->b_iodone
))(bp
);
1007 else if (bp
->b_flags
& XBF_ASYNC
)
1010 ASSERT(read
&& bp
->b_ops
);
1011 complete(&bp
->b_iowait
);
1020 bool read
= !!(bp
->b_flags
& XBF_READ
);
1022 trace_xfs_buf_iodone(bp
, _RET_IP_
);
1024 if (bp
->b_error
== 0)
1025 bp
->b_flags
|= XBF_DONE
;
1027 if (bp
->b_iodone
|| (read
&& bp
->b_ops
) || (bp
->b_flags
& XBF_ASYNC
)) {
1029 INIT_WORK(&bp
->b_iodone_work
, xfs_buf_iodone_work
);
1030 queue_work(xfslogd_workqueue
, &bp
->b_iodone_work
);
1032 xfs_buf_iodone_work(&bp
->b_iodone_work
);
1035 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_READ_AHEAD
);
1036 complete(&bp
->b_iowait
);
1045 ASSERT(error
>= 0 && error
<= 0xffff);
1046 bp
->b_error
= (unsigned short)error
;
1047 trace_xfs_buf_ioerror(bp
, error
, _RET_IP_
);
1051 xfs_buf_ioerror_alert(
1055 xfs_alert(bp
->b_target
->bt_mount
,
1056 "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d",
1057 (__uint64_t
)XFS_BUF_ADDR(bp
), func
, bp
->b_error
, bp
->b_length
);
1061 * Called when we want to stop a buffer from getting written or read.
1062 * We attach the EIO error, muck with its flags, and call xfs_buf_ioend
1063 * so that the proper iodone callbacks get called.
1069 #ifdef XFSERRORDEBUG
1070 ASSERT(XFS_BUF_ISREAD(bp
) || bp
->b_iodone
);
1074 * No need to wait until the buffer is unpinned, we aren't flushing it.
1076 xfs_buf_ioerror(bp
, EIO
);
1079 * We're calling xfs_buf_ioend, so delete XBF_DONE flag.
1085 xfs_buf_ioend(bp
, 0);
1091 * Same as xfs_bioerror, except that we are releasing the buffer
1092 * here ourselves, and avoiding the xfs_buf_ioend call.
1093 * This is meant for userdata errors; metadata bufs come with
1094 * iodone functions attached, so that we can track down errors.
1100 int64_t fl
= bp
->b_flags
;
1102 * No need to wait until the buffer is unpinned.
1103 * We aren't flushing it.
1105 * chunkhold expects B_DONE to be set, whether
1106 * we actually finish the I/O or not. We don't want to
1107 * change that interface.
1112 bp
->b_iodone
= NULL
;
1113 if (!(fl
& XBF_ASYNC
)) {
1115 * Mark b_error and B_ERROR _both_.
1116 * Lot's of chunkcache code assumes that.
1117 * There's no reason to mark error for
1120 xfs_buf_ioerror(bp
, EIO
);
1121 complete(&bp
->b_iowait
);
1133 if (XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
1134 trace_xfs_bdstrat_shut(bp
, _RET_IP_
);
1136 * Metadata write that didn't get logged but
1137 * written delayed anyway. These aren't associated
1138 * with a transaction, and can be ignored.
1140 if (!bp
->b_iodone
&& !XFS_BUF_ISREAD(bp
))
1141 return xfs_bioerror_relse(bp
);
1143 return xfs_bioerror(bp
);
1146 xfs_buf_iorequest(bp
);
1156 ASSERT(xfs_buf_islocked(bp
));
1158 bp
->b_flags
|= XBF_WRITE
;
1159 bp
->b_flags
&= ~(XBF_ASYNC
| XBF_READ
| _XBF_DELWRI_Q
| XBF_WRITE_FAIL
);
1163 error
= xfs_buf_iowait(bp
);
1165 xfs_force_shutdown(bp
->b_target
->bt_mount
,
1166 SHUTDOWN_META_IO_ERROR
);
1176 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1)
1177 xfs_buf_ioend(bp
, schedule
);
1185 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1188 * don't overwrite existing errors - otherwise we can lose errors on
1189 * buffers that require multiple bios to complete.
1192 xfs_buf_ioerror(bp
, -error
);
1194 if (!bp
->b_error
&& xfs_buf_is_vmapped(bp
) && (bp
->b_flags
& XBF_READ
))
1195 invalidate_kernel_vmap_range(bp
->b_addr
, xfs_buf_vmap_len(bp
));
1197 _xfs_buf_ioend(bp
, 1);
1202 xfs_buf_ioapply_map(
1210 int total_nr_pages
= bp
->b_page_count
;
1213 sector_t sector
= bp
->b_maps
[map
].bm_bn
;
1217 total_nr_pages
= bp
->b_page_count
;
1219 /* skip the pages in the buffer before the start offset */
1221 offset
= *buf_offset
;
1222 while (offset
>= PAGE_SIZE
) {
1224 offset
-= PAGE_SIZE
;
1228 * Limit the IO size to the length of the current vector, and update the
1229 * remaining IO count for the next time around.
1231 size
= min_t(int, BBTOB(bp
->b_maps
[map
].bm_len
), *count
);
1233 *buf_offset
+= size
;
1236 atomic_inc(&bp
->b_io_remaining
);
1237 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1238 if (nr_pages
> total_nr_pages
)
1239 nr_pages
= total_nr_pages
;
1241 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1242 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1243 bio
->bi_sector
= sector
;
1244 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1245 bio
->bi_private
= bp
;
1248 for (; size
&& nr_pages
; nr_pages
--, page_index
++) {
1249 int rbytes
, nbytes
= PAGE_SIZE
- offset
;
1254 rbytes
= bio_add_page(bio
, bp
->b_pages
[page_index
], nbytes
,
1256 if (rbytes
< nbytes
)
1260 sector
+= BTOBB(nbytes
);
1265 if (likely(bio
->bi_size
)) {
1266 if (xfs_buf_is_vmapped(bp
)) {
1267 flush_kernel_vmap_range(bp
->b_addr
,
1268 xfs_buf_vmap_len(bp
));
1270 submit_bio(rw
, bio
);
1275 * This is guaranteed not to be the last io reference count
1276 * because the caller (xfs_buf_iorequest) holds a count itself.
1278 atomic_dec(&bp
->b_io_remaining
);
1279 xfs_buf_ioerror(bp
, EIO
);
1289 struct blk_plug plug
;
1296 * Make sure we capture only current IO errors rather than stale errors
1297 * left over from previous use of the buffer (e.g. failed readahead).
1301 if (bp
->b_flags
& XBF_WRITE
) {
1302 if (bp
->b_flags
& XBF_SYNCIO
)
1306 if (bp
->b_flags
& XBF_FUA
)
1308 if (bp
->b_flags
& XBF_FLUSH
)
1312 * Run the write verifier callback function if it exists. If
1313 * this function fails it will mark the buffer with an error and
1314 * the IO should not be dispatched.
1317 bp
->b_ops
->verify_write(bp
);
1319 xfs_force_shutdown(bp
->b_target
->bt_mount
,
1320 SHUTDOWN_CORRUPT_INCORE
);
1324 } else if (bp
->b_flags
& XBF_READ_AHEAD
) {
1330 /* we only use the buffer cache for meta-data */
1334 * Walk all the vectors issuing IO on them. Set up the initial offset
1335 * into the buffer and the desired IO size before we start -
1336 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1339 offset
= bp
->b_offset
;
1340 size
= BBTOB(bp
->b_io_length
);
1341 blk_start_plug(&plug
);
1342 for (i
= 0; i
< bp
->b_map_count
; i
++) {
1343 xfs_buf_ioapply_map(bp
, i
, &offset
, &size
, rw
);
1347 break; /* all done */
1349 blk_finish_plug(&plug
);
1356 trace_xfs_buf_iorequest(bp
, _RET_IP_
);
1358 ASSERT(!(bp
->b_flags
& _XBF_DELWRI_Q
));
1360 if (bp
->b_flags
& XBF_WRITE
)
1361 xfs_buf_wait_unpin(bp
);
1364 /* Set the count to 1 initially, this will stop an I/O
1365 * completion callout which happens before we have started
1366 * all the I/O from calling xfs_buf_ioend too early.
1368 atomic_set(&bp
->b_io_remaining
, 1);
1369 _xfs_buf_ioapply(bp
);
1370 _xfs_buf_ioend(bp
, 1);
1376 * Waits for I/O to complete on the buffer supplied. It returns immediately if
1377 * no I/O is pending or there is already a pending error on the buffer. It
1378 * returns the I/O error code, if any, or 0 if there was no error.
1384 trace_xfs_buf_iowait(bp
, _RET_IP_
);
1387 wait_for_completion(&bp
->b_iowait
);
1389 trace_xfs_buf_iowait_done(bp
, _RET_IP_
);
1401 return bp
->b_addr
+ offset
;
1403 offset
+= bp
->b_offset
;
1404 page
= bp
->b_pages
[offset
>> PAGE_SHIFT
];
1405 return (xfs_caddr_t
)page_address(page
) + (offset
& (PAGE_SIZE
-1));
1409 * Move data into or out of a buffer.
1413 xfs_buf_t
*bp
, /* buffer to process */
1414 size_t boff
, /* starting buffer offset */
1415 size_t bsize
, /* length to copy */
1416 void *data
, /* data address */
1417 xfs_buf_rw_t mode
) /* read/write/zero flag */
1421 bend
= boff
+ bsize
;
1422 while (boff
< bend
) {
1424 int page_index
, page_offset
, csize
;
1426 page_index
= (boff
+ bp
->b_offset
) >> PAGE_SHIFT
;
1427 page_offset
= (boff
+ bp
->b_offset
) & ~PAGE_MASK
;
1428 page
= bp
->b_pages
[page_index
];
1429 csize
= min_t(size_t, PAGE_SIZE
- page_offset
,
1430 BBTOB(bp
->b_io_length
) - boff
);
1432 ASSERT((csize
+ page_offset
) <= PAGE_SIZE
);
1436 memset(page_address(page
) + page_offset
, 0, csize
);
1439 memcpy(data
, page_address(page
) + page_offset
, csize
);
1442 memcpy(page_address(page
) + page_offset
, data
, csize
);
1451 * Handling of buffer targets (buftargs).
1455 * Wait for any bufs with callbacks that have been submitted but have not yet
1456 * returned. These buffers will have an elevated hold count, so wait on those
1457 * while freeing all the buffers only held by the LRU.
1459 static enum lru_status
1460 xfs_buftarg_wait_rele(
1461 struct list_head
*item
,
1462 spinlock_t
*lru_lock
,
1466 struct xfs_buf
*bp
= container_of(item
, struct xfs_buf
, b_lru
);
1467 struct list_head
*dispose
= arg
;
1469 if (atomic_read(&bp
->b_hold
) > 1) {
1470 /* need to wait, so skip it this pass */
1471 trace_xfs_buf_wait_buftarg(bp
, _RET_IP_
);
1474 if (!spin_trylock(&bp
->b_lock
))
1478 * clear the LRU reference count so the buffer doesn't get
1479 * ignored in xfs_buf_rele().
1481 atomic_set(&bp
->b_lru_ref
, 0);
1482 bp
->b_state
|= XFS_BSTATE_DISPOSE
;
1483 list_move(item
, dispose
);
1484 spin_unlock(&bp
->b_lock
);
1490 struct xfs_buftarg
*btp
)
1495 /* loop until there is nothing left on the lru list. */
1496 while (list_lru_count(&btp
->bt_lru
)) {
1497 list_lru_walk(&btp
->bt_lru
, xfs_buftarg_wait_rele
,
1498 &dispose
, LONG_MAX
);
1500 while (!list_empty(&dispose
)) {
1502 bp
= list_first_entry(&dispose
, struct xfs_buf
, b_lru
);
1503 list_del_init(&bp
->b_lru
);
1504 if (bp
->b_flags
& XBF_WRITE_FAIL
) {
1505 xfs_alert(btp
->bt_mount
,
1506 "Corruption Alert: Buffer at block 0x%llx had permanent write failures!\n"
1507 "Please run xfs_repair to determine the extent of the problem.",
1508 (long long)bp
->b_bn
);
1517 static enum lru_status
1518 xfs_buftarg_isolate(
1519 struct list_head
*item
,
1520 spinlock_t
*lru_lock
,
1523 struct xfs_buf
*bp
= container_of(item
, struct xfs_buf
, b_lru
);
1524 struct list_head
*dispose
= arg
;
1527 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1528 * If we fail to get the lock, just skip it.
1530 if (!spin_trylock(&bp
->b_lock
))
1533 * Decrement the b_lru_ref count unless the value is already
1534 * zero. If the value is already zero, we need to reclaim the
1535 * buffer, otherwise it gets another trip through the LRU.
1537 if (!atomic_add_unless(&bp
->b_lru_ref
, -1, 0)) {
1538 spin_unlock(&bp
->b_lock
);
1542 bp
->b_state
|= XFS_BSTATE_DISPOSE
;
1543 list_move(item
, dispose
);
1544 spin_unlock(&bp
->b_lock
);
1548 static unsigned long
1549 xfs_buftarg_shrink_scan(
1550 struct shrinker
*shrink
,
1551 struct shrink_control
*sc
)
1553 struct xfs_buftarg
*btp
= container_of(shrink
,
1554 struct xfs_buftarg
, bt_shrinker
);
1556 unsigned long freed
;
1557 unsigned long nr_to_scan
= sc
->nr_to_scan
;
1559 freed
= list_lru_walk_node(&btp
->bt_lru
, sc
->nid
, xfs_buftarg_isolate
,
1560 &dispose
, &nr_to_scan
);
1562 while (!list_empty(&dispose
)) {
1564 bp
= list_first_entry(&dispose
, struct xfs_buf
, b_lru
);
1565 list_del_init(&bp
->b_lru
);
1572 static unsigned long
1573 xfs_buftarg_shrink_count(
1574 struct shrinker
*shrink
,
1575 struct shrink_control
*sc
)
1577 struct xfs_buftarg
*btp
= container_of(shrink
,
1578 struct xfs_buftarg
, bt_shrinker
);
1579 return list_lru_count_node(&btp
->bt_lru
, sc
->nid
);
1584 struct xfs_mount
*mp
,
1585 struct xfs_buftarg
*btp
)
1587 unregister_shrinker(&btp
->bt_shrinker
);
1588 list_lru_destroy(&btp
->bt_lru
);
1590 if (mp
->m_flags
& XFS_MOUNT_BARRIER
)
1591 xfs_blkdev_issue_flush(btp
);
1597 xfs_setsize_buftarg_flags(
1599 unsigned int blocksize
,
1600 unsigned int sectorsize
,
1603 btp
->bt_bsize
= blocksize
;
1604 btp
->bt_sshift
= ffs(sectorsize
) - 1;
1605 btp
->bt_smask
= sectorsize
- 1;
1607 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1608 char name
[BDEVNAME_SIZE
];
1610 bdevname(btp
->bt_bdev
, name
);
1612 xfs_warn(btp
->bt_mount
,
1613 "Cannot set_blocksize to %u on device %s",
1622 * When allocating the initial buffer target we have not yet
1623 * read in the superblock, so don't know what sized sectors
1624 * are being used at this early stage. Play safe.
1627 xfs_setsize_buftarg_early(
1629 struct block_device
*bdev
)
1631 return xfs_setsize_buftarg_flags(btp
,
1632 PAGE_SIZE
, bdev_logical_block_size(bdev
), 0);
1636 xfs_setsize_buftarg(
1638 unsigned int blocksize
,
1639 unsigned int sectorsize
)
1641 return xfs_setsize_buftarg_flags(btp
, blocksize
, sectorsize
, 1);
1646 struct xfs_mount
*mp
,
1647 struct block_device
*bdev
,
1653 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
| KM_NOFS
);
1656 btp
->bt_dev
= bdev
->bd_dev
;
1657 btp
->bt_bdev
= bdev
;
1658 btp
->bt_bdi
= blk_get_backing_dev_info(bdev
);
1662 if (xfs_setsize_buftarg_early(btp
, bdev
))
1665 if (list_lru_init(&btp
->bt_lru
))
1668 btp
->bt_shrinker
.count_objects
= xfs_buftarg_shrink_count
;
1669 btp
->bt_shrinker
.scan_objects
= xfs_buftarg_shrink_scan
;
1670 btp
->bt_shrinker
.seeks
= DEFAULT_SEEKS
;
1671 btp
->bt_shrinker
.flags
= SHRINKER_NUMA_AWARE
;
1672 register_shrinker(&btp
->bt_shrinker
);
1681 * Add a buffer to the delayed write list.
1683 * This queues a buffer for writeout if it hasn't already been. Note that
1684 * neither this routine nor the buffer list submission functions perform
1685 * any internal synchronization. It is expected that the lists are thread-local
1688 * Returns true if we queued up the buffer, or false if it already had
1689 * been on the buffer list.
1692 xfs_buf_delwri_queue(
1694 struct list_head
*list
)
1696 ASSERT(xfs_buf_islocked(bp
));
1697 ASSERT(!(bp
->b_flags
& XBF_READ
));
1700 * If the buffer is already marked delwri it already is queued up
1701 * by someone else for imediate writeout. Just ignore it in that
1704 if (bp
->b_flags
& _XBF_DELWRI_Q
) {
1705 trace_xfs_buf_delwri_queued(bp
, _RET_IP_
);
1709 trace_xfs_buf_delwri_queue(bp
, _RET_IP_
);
1712 * If a buffer gets written out synchronously or marked stale while it
1713 * is on a delwri list we lazily remove it. To do this, the other party
1714 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1715 * It remains referenced and on the list. In a rare corner case it
1716 * might get readded to a delwri list after the synchronous writeout, in
1717 * which case we need just need to re-add the flag here.
1719 bp
->b_flags
|= _XBF_DELWRI_Q
;
1720 if (list_empty(&bp
->b_list
)) {
1721 atomic_inc(&bp
->b_hold
);
1722 list_add_tail(&bp
->b_list
, list
);
1729 * Compare function is more complex than it needs to be because
1730 * the return value is only 32 bits and we are doing comparisons
1736 struct list_head
*a
,
1737 struct list_head
*b
)
1739 struct xfs_buf
*ap
= container_of(a
, struct xfs_buf
, b_list
);
1740 struct xfs_buf
*bp
= container_of(b
, struct xfs_buf
, b_list
);
1743 diff
= ap
->b_maps
[0].bm_bn
- bp
->b_maps
[0].bm_bn
;
1752 __xfs_buf_delwri_submit(
1753 struct list_head
*buffer_list
,
1754 struct list_head
*io_list
,
1757 struct blk_plug plug
;
1758 struct xfs_buf
*bp
, *n
;
1761 list_for_each_entry_safe(bp
, n
, buffer_list
, b_list
) {
1763 if (xfs_buf_ispinned(bp
)) {
1767 if (!xfs_buf_trylock(bp
))
1774 * Someone else might have written the buffer synchronously or
1775 * marked it stale in the meantime. In that case only the
1776 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
1777 * reference and remove it from the list here.
1779 if (!(bp
->b_flags
& _XBF_DELWRI_Q
)) {
1780 list_del_init(&bp
->b_list
);
1785 list_move_tail(&bp
->b_list
, io_list
);
1786 trace_xfs_buf_delwri_split(bp
, _RET_IP_
);
1789 list_sort(NULL
, io_list
, xfs_buf_cmp
);
1791 blk_start_plug(&plug
);
1792 list_for_each_entry_safe(bp
, n
, io_list
, b_list
) {
1793 bp
->b_flags
&= ~(_XBF_DELWRI_Q
| XBF_ASYNC
| XBF_WRITE_FAIL
);
1794 bp
->b_flags
|= XBF_WRITE
;
1797 bp
->b_flags
|= XBF_ASYNC
;
1798 list_del_init(&bp
->b_list
);
1802 blk_finish_plug(&plug
);
1808 * Write out a buffer list asynchronously.
1810 * This will take the @buffer_list, write all non-locked and non-pinned buffers
1811 * out and not wait for I/O completion on any of the buffers. This interface
1812 * is only safely useable for callers that can track I/O completion by higher
1813 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
1817 xfs_buf_delwri_submit_nowait(
1818 struct list_head
*buffer_list
)
1820 LIST_HEAD (io_list
);
1821 return __xfs_buf_delwri_submit(buffer_list
, &io_list
, false);
1825 * Write out a buffer list synchronously.
1827 * This will take the @buffer_list, write all buffers out and wait for I/O
1828 * completion on all of the buffers. @buffer_list is consumed by the function,
1829 * so callers must have some other way of tracking buffers if they require such
1833 xfs_buf_delwri_submit(
1834 struct list_head
*buffer_list
)
1836 LIST_HEAD (io_list
);
1837 int error
= 0, error2
;
1840 __xfs_buf_delwri_submit(buffer_list
, &io_list
, true);
1842 /* Wait for IO to complete. */
1843 while (!list_empty(&io_list
)) {
1844 bp
= list_first_entry(&io_list
, struct xfs_buf
, b_list
);
1846 list_del_init(&bp
->b_list
);
1847 error2
= xfs_buf_iowait(bp
);
1859 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1860 KM_ZONE_HWALIGN
, NULL
);
1864 xfslogd_workqueue
= alloc_workqueue("xfslogd",
1865 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 1);
1866 if (!xfslogd_workqueue
)
1867 goto out_free_buf_zone
;
1872 kmem_zone_destroy(xfs_buf_zone
);
1878 xfs_buf_terminate(void)
1880 destroy_workqueue(xfslogd_workqueue
);
1881 kmem_zone_destroy(xfs_buf_zone
);
This page took 0.097058 seconds and 5 git commands to generate.