2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include <linux/stddef.h>
20 #include <linux/errno.h>
21 #include <linux/gfp.h>
22 #include <linux/pagemap.h>
23 #include <linux/init.h>
24 #include <linux/vmalloc.h>
25 #include <linux/bio.h>
26 #include <linux/sysctl.h>
27 #include <linux/proc_fs.h>
28 #include <linux/workqueue.h>
29 #include <linux/percpu.h>
30 #include <linux/blkdev.h>
31 #include <linux/hash.h>
32 #include <linux/kthread.h>
33 #include <linux/migrate.h>
34 #include <linux/backing-dev.h>
35 #include <linux/freezer.h>
36 #include <linux/list_sort.h>
42 #include "xfs_mount.h"
43 #include "xfs_trace.h"
45 static kmem_zone_t
*xfs_buf_zone
;
46 STATIC
int xfsbufd(void *);
47 STATIC
int xfsbufd_wakeup(struct shrinker
*, int, gfp_t
);
48 STATIC
void xfs_buf_delwri_queue(xfs_buf_t
*, int);
49 static struct shrinker xfs_buf_shake
= {
50 .shrink
= xfsbufd_wakeup
,
51 .seeks
= DEFAULT_SEEKS
,
54 static struct workqueue_struct
*xfslogd_workqueue
;
55 struct workqueue_struct
*xfsdatad_workqueue
;
56 struct workqueue_struct
*xfsconvertd_workqueue
;
58 #ifdef XFS_BUF_LOCK_TRACKING
59 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
60 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
61 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
63 # define XB_SET_OWNER(bp) do { } while (0)
64 # define XB_CLEAR_OWNER(bp) do { } while (0)
65 # define XB_GET_OWNER(bp) do { } while (0)
68 #define xb_to_gfp(flags) \
69 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \
70 ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)
72 #define xb_to_km(flags) \
73 (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)
75 #define xfs_buf_allocate(flags) \
76 kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags))
77 #define xfs_buf_deallocate(bp) \
78 kmem_zone_free(xfs_buf_zone, (bp));
85 * Return true if the buffer is vmapped.
87 * The XBF_MAPPED flag is set if the buffer should be mapped, but the
88 * code is clever enough to know it doesn't have to map a single page,
89 * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1.
91 return (bp
->b_flags
& XBF_MAPPED
) && bp
->b_page_count
> 1;
98 return (bp
->b_page_count
* PAGE_SIZE
) - bp
->b_offset
;
102 * Page Region interfaces.
104 * For pages in filesystems where the blocksize is smaller than the
105 * pagesize, we use the page->private field (long) to hold a bitmap
106 * of uptodate regions within the page.
108 * Each such region is "bytes per page / bits per long" bytes long.
110 * NBPPR == number-of-bytes-per-page-region
111 * BTOPR == bytes-to-page-region (rounded up)
112 * BTOPRT == bytes-to-page-region-truncated (rounded down)
114 #if (BITS_PER_LONG == 32)
115 #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */
116 #elif (BITS_PER_LONG == 64)
117 #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */
119 #error BITS_PER_LONG must be 32 or 64
121 #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG)
122 #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT)
123 #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT))
133 first
= BTOPR(offset
);
134 final
= BTOPRT(offset
+ length
- 1);
135 first
= min(first
, final
);
138 mask
<<= BITS_PER_LONG
- (final
- first
);
139 mask
>>= BITS_PER_LONG
- (final
);
141 ASSERT(offset
+ length
<= PAGE_CACHE_SIZE
);
142 ASSERT((final
- first
) < BITS_PER_LONG
&& (final
- first
) >= 0);
153 set_page_private(page
,
154 page_private(page
) | page_region_mask(offset
, length
));
155 if (page_private(page
) == ~0UL)
156 SetPageUptodate(page
);
165 unsigned long mask
= page_region_mask(offset
, length
);
167 return (mask
&& (page_private(page
) & mask
) == mask
);
171 * Internal xfs_buf_t object manipulation
177 xfs_buftarg_t
*target
,
178 xfs_off_t range_base
,
180 xfs_buf_flags_t flags
)
183 * We don't want certain flags to appear in b_flags.
185 flags
&= ~(XBF_LOCK
|XBF_MAPPED
|XBF_DONT_BLOCK
|XBF_READ_AHEAD
);
187 memset(bp
, 0, sizeof(xfs_buf_t
));
188 atomic_set(&bp
->b_hold
, 1);
189 init_completion(&bp
->b_iowait
);
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
;
195 bp
->b_file_offset
= range_base
;
197 * Set buffer_length and count_desired to the same value initially.
198 * I/O routines should use count_desired, which will be the same in
199 * most cases but may be reset (e.g. XFS recovery).
201 bp
->b_buffer_length
= bp
->b_count_desired
= range_length
;
203 bp
->b_bn
= XFS_BUF_DADDR_NULL
;
204 atomic_set(&bp
->b_pin_count
, 0);
205 init_waitqueue_head(&bp
->b_waiters
);
207 XFS_STATS_INC(xb_create
);
209 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_offset
= xfs_buf_poff(bp
->b_file_offset
);
225 bp
->b_page_count
= page_count
;
226 if (page_count
<= XB_PAGES
) {
227 bp
->b_pages
= bp
->b_page_array
;
229 bp
->b_pages
= kmem_alloc(sizeof(struct page
*) *
230 page_count
, xb_to_km(flags
));
231 if (bp
->b_pages
== NULL
)
234 memset(bp
->b_pages
, 0, sizeof(struct page
*) * page_count
);
240 * Frees b_pages if it was allocated.
246 if (bp
->b_pages
!= bp
->b_page_array
) {
247 kmem_free(bp
->b_pages
);
253 * Releases the specified buffer.
255 * The modification state of any associated pages is left unchanged.
256 * The buffer most not be on any hash - use xfs_buf_rele instead for
257 * hashed and refcounted buffers
263 trace_xfs_buf_free(bp
, _RET_IP_
);
265 if (bp
->b_flags
& (_XBF_PAGE_CACHE
|_XBF_PAGES
)) {
268 if (xfs_buf_is_vmapped(bp
))
269 vm_unmap_ram(bp
->b_addr
- bp
->b_offset
,
272 for (i
= 0; i
< bp
->b_page_count
; i
++) {
273 struct page
*page
= bp
->b_pages
[i
];
275 if (bp
->b_flags
& _XBF_PAGE_CACHE
)
276 ASSERT(!PagePrivate(page
));
277 page_cache_release(page
);
280 _xfs_buf_free_pages(bp
);
281 xfs_buf_deallocate(bp
);
285 * Finds all pages for buffer in question and builds it's page list.
288 _xfs_buf_lookup_pages(
292 struct address_space
*mapping
= bp
->b_target
->bt_mapping
;
293 size_t blocksize
= bp
->b_target
->bt_bsize
;
294 size_t size
= bp
->b_count_desired
;
295 size_t nbytes
, offset
;
296 gfp_t gfp_mask
= xb_to_gfp(flags
);
297 unsigned short page_count
, i
;
302 end
= bp
->b_file_offset
+ bp
->b_buffer_length
;
303 page_count
= xfs_buf_btoc(end
) - xfs_buf_btoct(bp
->b_file_offset
);
305 error
= _xfs_buf_get_pages(bp
, page_count
, flags
);
308 bp
->b_flags
|= _XBF_PAGE_CACHE
;
310 offset
= bp
->b_offset
;
311 first
= bp
->b_file_offset
>> PAGE_CACHE_SHIFT
;
313 for (i
= 0; i
< bp
->b_page_count
; i
++) {
318 page
= find_or_create_page(mapping
, first
+ i
, gfp_mask
);
319 if (unlikely(page
== NULL
)) {
320 if (flags
& XBF_READ_AHEAD
) {
321 bp
->b_page_count
= i
;
322 for (i
= 0; i
< bp
->b_page_count
; i
++)
323 unlock_page(bp
->b_pages
[i
]);
328 * This could deadlock.
330 * But until all the XFS lowlevel code is revamped to
331 * handle buffer allocation failures we can't do much.
333 if (!(++retries
% 100))
335 "XFS: possible memory allocation "
336 "deadlock in %s (mode:0x%x)\n",
339 XFS_STATS_INC(xb_page_retries
);
340 xfsbufd_wakeup(NULL
, 0, gfp_mask
);
341 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
345 XFS_STATS_INC(xb_page_found
);
347 nbytes
= min_t(size_t, size
, PAGE_CACHE_SIZE
- offset
);
350 ASSERT(!PagePrivate(page
));
351 if (!PageUptodate(page
)) {
353 if (blocksize
>= PAGE_CACHE_SIZE
) {
354 if (flags
& XBF_READ
)
355 bp
->b_flags
|= _XBF_PAGE_LOCKED
;
356 } else if (!PagePrivate(page
)) {
357 if (test_page_region(page
, offset
, nbytes
))
362 bp
->b_pages
[i
] = page
;
366 if (!(bp
->b_flags
& _XBF_PAGE_LOCKED
)) {
367 for (i
= 0; i
< bp
->b_page_count
; i
++)
368 unlock_page(bp
->b_pages
[i
]);
371 if (page_count
== bp
->b_page_count
)
372 bp
->b_flags
|= XBF_DONE
;
378 * Map buffer into kernel address-space if nessecary.
385 /* A single page buffer is always mappable */
386 if (bp
->b_page_count
== 1) {
387 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
388 bp
->b_flags
|= XBF_MAPPED
;
389 } else if (flags
& XBF_MAPPED
) {
390 bp
->b_addr
= vm_map_ram(bp
->b_pages
, bp
->b_page_count
,
392 if (unlikely(bp
->b_addr
== NULL
))
394 bp
->b_addr
+= bp
->b_offset
;
395 bp
->b_flags
|= XBF_MAPPED
;
402 * Finding and Reading Buffers
406 * Look up, and creates if absent, a lockable buffer for
407 * a given range of an inode. The buffer is returned
408 * locked. If other overlapping buffers exist, they are
409 * released before the new buffer is created and locked,
410 * which may imply that this call will block until those buffers
411 * are unlocked. No I/O is implied by this call.
415 xfs_buftarg_t
*btp
, /* block device target */
416 xfs_off_t ioff
, /* starting offset of range */
417 size_t isize
, /* length of range */
418 xfs_buf_flags_t flags
,
421 xfs_off_t range_base
;
423 struct xfs_perag
*pag
;
424 struct rb_node
**rbp
;
425 struct rb_node
*parent
;
428 range_base
= (ioff
<< BBSHIFT
);
429 range_length
= (isize
<< BBSHIFT
);
431 /* Check for IOs smaller than the sector size / not sector aligned */
432 ASSERT(!(range_length
< (1 << btp
->bt_sshift
)));
433 ASSERT(!(range_base
& (xfs_off_t
)btp
->bt_smask
));
436 pag
= xfs_perag_get(btp
->bt_mount
,
437 xfs_daddr_to_agno(btp
->bt_mount
, ioff
));
440 spin_lock(&pag
->pag_buf_lock
);
441 rbp
= &pag
->pag_buf_tree
.rb_node
;
446 bp
= rb_entry(parent
, struct xfs_buf
, b_rbnode
);
448 if (range_base
< bp
->b_file_offset
)
449 rbp
= &(*rbp
)->rb_left
;
450 else if (range_base
> bp
->b_file_offset
)
451 rbp
= &(*rbp
)->rb_right
;
454 * found a block offset match. If the range doesn't
455 * match, the only way this is allowed is if the buffer
456 * in the cache is stale and the transaction that made
457 * it stale has not yet committed. i.e. we are
458 * reallocating a busy extent. Skip this buffer and
459 * continue searching to the right for an exact match.
461 if (bp
->b_buffer_length
!= range_length
) {
462 ASSERT(bp
->b_flags
& XBF_STALE
);
463 rbp
= &(*rbp
)->rb_right
;
466 atomic_inc(&bp
->b_hold
);
473 _xfs_buf_initialize(new_bp
, btp
, range_base
,
474 range_length
, flags
);
475 rb_link_node(&new_bp
->b_rbnode
, parent
, rbp
);
476 rb_insert_color(&new_bp
->b_rbnode
, &pag
->pag_buf_tree
);
477 /* the buffer keeps the perag reference until it is freed */
479 spin_unlock(&pag
->pag_buf_lock
);
481 XFS_STATS_INC(xb_miss_locked
);
482 spin_unlock(&pag
->pag_buf_lock
);
488 spin_unlock(&pag
->pag_buf_lock
);
491 /* Attempt to get the semaphore without sleeping,
492 * if this does not work then we need to drop the
493 * spinlock and do a hard attempt on the semaphore.
495 if (down_trylock(&bp
->b_sema
)) {
496 if (!(flags
& XBF_TRYLOCK
)) {
497 /* wait for buffer ownership */
499 XFS_STATS_INC(xb_get_locked_waited
);
501 /* We asked for a trylock and failed, no need
502 * to look at file offset and length here, we
503 * know that this buffer at least overlaps our
504 * buffer and is locked, therefore our buffer
505 * either does not exist, or is this buffer.
508 XFS_STATS_INC(xb_busy_locked
);
516 if (bp
->b_flags
& XBF_STALE
) {
517 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
518 bp
->b_flags
&= XBF_MAPPED
;
521 trace_xfs_buf_find(bp
, flags
, _RET_IP_
);
522 XFS_STATS_INC(xb_get_locked
);
527 * Assembles a buffer covering the specified range.
528 * Storage in memory for all portions of the buffer will be allocated,
529 * although backing storage may not be.
533 xfs_buftarg_t
*target
,/* target for buffer */
534 xfs_off_t ioff
, /* starting offset of range */
535 size_t isize
, /* length of range */
536 xfs_buf_flags_t flags
)
538 xfs_buf_t
*bp
, *new_bp
;
541 new_bp
= xfs_buf_allocate(flags
);
542 if (unlikely(!new_bp
))
545 bp
= _xfs_buf_find(target
, ioff
, isize
, flags
, new_bp
);
547 error
= _xfs_buf_lookup_pages(bp
, flags
);
551 xfs_buf_deallocate(new_bp
);
552 if (unlikely(bp
== NULL
))
556 for (i
= 0; i
< bp
->b_page_count
; i
++)
557 mark_page_accessed(bp
->b_pages
[i
]);
559 if (!(bp
->b_flags
& XBF_MAPPED
)) {
560 error
= _xfs_buf_map_pages(bp
, flags
);
561 if (unlikely(error
)) {
562 printk(KERN_WARNING
"%s: failed to map pages\n",
568 XFS_STATS_INC(xb_get
);
571 * Always fill in the block number now, the mapped cases can do
572 * their own overlay of this later.
575 bp
->b_count_desired
= bp
->b_buffer_length
;
577 trace_xfs_buf_get(bp
, flags
, _RET_IP_
);
581 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
590 xfs_buf_flags_t flags
)
594 ASSERT(!(flags
& (XBF_DELWRI
|XBF_WRITE
)));
595 ASSERT(bp
->b_bn
!= XFS_BUF_DADDR_NULL
);
597 bp
->b_flags
&= ~(XBF_WRITE
| XBF_ASYNC
| XBF_DELWRI
| \
598 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
599 bp
->b_flags
|= flags
& (XBF_READ
| XBF_ASYNC
| \
600 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
602 status
= xfs_buf_iorequest(bp
);
603 if (status
|| XFS_BUF_ISERROR(bp
) || (flags
& XBF_ASYNC
))
605 return xfs_buf_iowait(bp
);
610 xfs_buftarg_t
*target
,
613 xfs_buf_flags_t flags
)
619 bp
= xfs_buf_get(target
, ioff
, isize
, flags
);
621 trace_xfs_buf_read(bp
, flags
, _RET_IP_
);
623 if (!XFS_BUF_ISDONE(bp
)) {
624 XFS_STATS_INC(xb_get_read
);
625 _xfs_buf_read(bp
, flags
);
626 } else if (flags
& XBF_ASYNC
) {
628 * Read ahead call which is already satisfied,
633 /* We do not want read in the flags */
634 bp
->b_flags
&= ~XBF_READ
;
641 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
648 * If we are not low on memory then do the readahead in a deadlock
653 xfs_buftarg_t
*target
,
657 struct backing_dev_info
*bdi
;
659 bdi
= target
->bt_mapping
->backing_dev_info
;
660 if (bdi_read_congested(bdi
))
663 xfs_buf_read(target
, ioff
, isize
,
664 XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
|XBF_DONT_BLOCK
);
668 * Read an uncached buffer from disk. Allocates and returns a locked
669 * buffer containing the disk contents or nothing.
672 xfs_buf_read_uncached(
673 struct xfs_mount
*mp
,
674 struct xfs_buftarg
*target
,
682 bp
= xfs_buf_get_uncached(target
, length
, flags
);
686 /* set up the buffer for a read IO */
688 XFS_BUF_SET_ADDR(bp
, daddr
);
693 error
= xfs_buf_iowait(bp
);
694 if (error
|| bp
->b_error
) {
704 xfs_buftarg_t
*target
)
708 bp
= xfs_buf_allocate(0);
710 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
714 static inline struct page
*
718 if ((!is_vmalloc_addr(addr
))) {
719 return virt_to_page(addr
);
721 return vmalloc_to_page(addr
);
726 xfs_buf_associate_memory(
733 unsigned long pageaddr
;
734 unsigned long offset
;
738 pageaddr
= (unsigned long)mem
& PAGE_CACHE_MASK
;
739 offset
= (unsigned long)mem
- pageaddr
;
740 buflen
= PAGE_CACHE_ALIGN(len
+ offset
);
741 page_count
= buflen
>> PAGE_CACHE_SHIFT
;
743 /* Free any previous set of page pointers */
745 _xfs_buf_free_pages(bp
);
750 rval
= _xfs_buf_get_pages(bp
, page_count
, XBF_DONT_BLOCK
);
754 bp
->b_offset
= offset
;
756 for (i
= 0; i
< bp
->b_page_count
; i
++) {
757 bp
->b_pages
[i
] = mem_to_page((void *)pageaddr
);
758 pageaddr
+= PAGE_CACHE_SIZE
;
761 bp
->b_count_desired
= len
;
762 bp
->b_buffer_length
= buflen
;
763 bp
->b_flags
|= XBF_MAPPED
;
764 bp
->b_flags
&= ~_XBF_PAGE_LOCKED
;
770 xfs_buf_get_uncached(
771 struct xfs_buftarg
*target
,
775 unsigned long page_count
= PAGE_ALIGN(len
) >> PAGE_SHIFT
;
779 bp
= xfs_buf_allocate(0);
780 if (unlikely(bp
== NULL
))
782 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
784 error
= _xfs_buf_get_pages(bp
, page_count
, 0);
788 for (i
= 0; i
< page_count
; i
++) {
789 bp
->b_pages
[i
] = alloc_page(xb_to_gfp(flags
));
793 bp
->b_flags
|= _XBF_PAGES
;
795 error
= _xfs_buf_map_pages(bp
, XBF_MAPPED
);
796 if (unlikely(error
)) {
797 printk(KERN_WARNING
"%s: failed to map pages\n",
804 trace_xfs_buf_get_uncached(bp
, _RET_IP_
);
809 __free_page(bp
->b_pages
[i
]);
810 _xfs_buf_free_pages(bp
);
812 xfs_buf_deallocate(bp
);
818 * Increment reference count on buffer, to hold the buffer concurrently
819 * with another thread which may release (free) the buffer asynchronously.
820 * Must hold the buffer already to call this function.
826 trace_xfs_buf_hold(bp
, _RET_IP_
);
827 atomic_inc(&bp
->b_hold
);
831 * Releases a hold on the specified buffer. If the
832 * the hold count is 1, calls xfs_buf_free.
838 struct xfs_perag
*pag
= bp
->b_pag
;
840 trace_xfs_buf_rele(bp
, _RET_IP_
);
843 ASSERT(!bp
->b_relse
);
844 ASSERT(RB_EMPTY_NODE(&bp
->b_rbnode
));
845 if (atomic_dec_and_test(&bp
->b_hold
))
850 ASSERT(!RB_EMPTY_NODE(&bp
->b_rbnode
));
851 ASSERT(atomic_read(&bp
->b_hold
) > 0);
852 if (atomic_dec_and_lock(&bp
->b_hold
, &pag
->pag_buf_lock
)) {
854 atomic_inc(&bp
->b_hold
);
855 spin_unlock(&pag
->pag_buf_lock
);
858 ASSERT(!(bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)));
859 rb_erase(&bp
->b_rbnode
, &pag
->pag_buf_tree
);
860 spin_unlock(&pag
->pag_buf_lock
);
869 * Mutual exclusion on buffers. Locking model:
871 * Buffers associated with inodes for which buffer locking
872 * is not enabled are not protected by semaphores, and are
873 * assumed to be exclusively owned by the caller. There is a
874 * spinlock in the buffer, used by the caller when concurrent
875 * access is possible.
879 * Locks a buffer object, if it is not already locked.
880 * Note that this in no way locks the underlying pages, so it is only
881 * useful for synchronizing concurrent use of buffer objects, not for
882 * synchronizing independent access to the underlying pages.
890 locked
= down_trylock(&bp
->b_sema
) == 0;
894 trace_xfs_buf_cond_lock(bp
, _RET_IP_
);
895 return locked
? 0 : -EBUSY
;
902 return bp
->b_sema
.count
;
906 * Locks a buffer object.
907 * Note that this in no way locks the underlying pages, so it is only
908 * useful for synchronizing concurrent use of buffer objects, not for
909 * synchronizing independent access to the underlying pages.
911 * If we come across a stale, pinned, locked buffer, we know that we
912 * are being asked to lock a buffer that has been reallocated. Because
913 * it is pinned, we know that the log has not been pushed to disk and
914 * hence it will still be locked. Rather than sleeping until someone
915 * else pushes the log, push it ourselves before trying to get the lock.
921 trace_xfs_buf_lock(bp
, _RET_IP_
);
923 if (atomic_read(&bp
->b_pin_count
) && (bp
->b_flags
& XBF_STALE
))
924 xfs_log_force(bp
->b_target
->bt_mount
, 0);
925 if (atomic_read(&bp
->b_io_remaining
))
926 blk_run_address_space(bp
->b_target
->bt_mapping
);
930 trace_xfs_buf_lock_done(bp
, _RET_IP_
);
934 * Releases the lock on the buffer object.
935 * If the buffer is marked delwri but is not queued, do so before we
936 * unlock the buffer as we need to set flags correctly. We also need to
937 * take a reference for the delwri queue because the unlocker is going to
938 * drop their's and they don't know we just queued it.
944 if ((bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)) == XBF_DELWRI
) {
945 atomic_inc(&bp
->b_hold
);
946 bp
->b_flags
|= XBF_ASYNC
;
947 xfs_buf_delwri_queue(bp
, 0);
953 trace_xfs_buf_unlock(bp
, _RET_IP_
);
960 DECLARE_WAITQUEUE (wait
, current
);
962 if (atomic_read(&bp
->b_pin_count
) == 0)
965 add_wait_queue(&bp
->b_waiters
, &wait
);
967 set_current_state(TASK_UNINTERRUPTIBLE
);
968 if (atomic_read(&bp
->b_pin_count
) == 0)
970 if (atomic_read(&bp
->b_io_remaining
))
971 blk_run_address_space(bp
->b_target
->bt_mapping
);
974 remove_wait_queue(&bp
->b_waiters
, &wait
);
975 set_current_state(TASK_RUNNING
);
979 * Buffer Utility Routines
984 struct work_struct
*work
)
987 container_of(work
, xfs_buf_t
, b_iodone_work
);
990 (*(bp
->b_iodone
))(bp
);
991 else if (bp
->b_flags
& XBF_ASYNC
)
1000 trace_xfs_buf_iodone(bp
, _RET_IP_
);
1002 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_READ_AHEAD
);
1003 if (bp
->b_error
== 0)
1004 bp
->b_flags
|= XBF_DONE
;
1006 if ((bp
->b_iodone
) || (bp
->b_flags
& XBF_ASYNC
)) {
1008 INIT_WORK(&bp
->b_iodone_work
, xfs_buf_iodone_work
);
1009 queue_work(xfslogd_workqueue
, &bp
->b_iodone_work
);
1011 xfs_buf_iodone_work(&bp
->b_iodone_work
);
1014 complete(&bp
->b_iowait
);
1023 ASSERT(error
>= 0 && error
<= 0xffff);
1024 bp
->b_error
= (unsigned short)error
;
1025 trace_xfs_buf_ioerror(bp
, error
, _RET_IP_
);
1030 struct xfs_mount
*mp
,
1035 bp
->b_flags
|= XBF_WRITE
;
1036 bp
->b_flags
&= ~(XBF_ASYNC
| XBF_READ
);
1038 xfs_buf_delwri_dequeue(bp
);
1041 error
= xfs_buf_iowait(bp
);
1043 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1053 trace_xfs_buf_bdwrite(bp
, _RET_IP_
);
1055 bp
->b_flags
&= ~XBF_READ
;
1056 bp
->b_flags
|= (XBF_DELWRI
| XBF_ASYNC
);
1058 xfs_buf_delwri_queue(bp
, 1);
1062 * Called when we want to stop a buffer from getting written or read.
1063 * We attach the EIO error, muck with its flags, and call xfs_buf_ioend
1064 * so that the proper iodone callbacks get called.
1070 #ifdef XFSERRORDEBUG
1071 ASSERT(XFS_BUF_ISREAD(bp
) || bp
->b_iodone
);
1075 * No need to wait until the buffer is unpinned, we aren't flushing it.
1077 XFS_BUF_ERROR(bp
, EIO
);
1080 * We're calling xfs_buf_ioend, so delete XBF_DONE flag.
1083 XFS_BUF_UNDELAYWRITE(bp
);
1087 xfs_buf_ioend(bp
, 0);
1093 * Same as xfs_bioerror, except that we are releasing the buffer
1094 * here ourselves, and avoiding the xfs_buf_ioend call.
1095 * This is meant for userdata errors; metadata bufs come with
1096 * iodone functions attached, so that we can track down errors.
1102 int64_t fl
= XFS_BUF_BFLAGS(bp
);
1104 * No need to wait until the buffer is unpinned.
1105 * We aren't flushing it.
1107 * chunkhold expects B_DONE to be set, whether
1108 * we actually finish the I/O or not. We don't want to
1109 * change that interface.
1112 XFS_BUF_UNDELAYWRITE(bp
);
1115 XFS_BUF_CLR_IODONE_FUNC(bp
);
1116 if (!(fl
& XBF_ASYNC
)) {
1118 * Mark b_error and B_ERROR _both_.
1119 * Lot's of chunkcache code assumes that.
1120 * There's no reason to mark error for
1123 XFS_BUF_ERROR(bp
, EIO
);
1124 XFS_BUF_FINISH_IOWAIT(bp
);
1134 * All xfs metadata buffers except log state machine buffers
1135 * get this attached as their b_bdstrat callback function.
1136 * This is so that we can catch a buffer
1137 * after prematurely unpinning it to forcibly shutdown the filesystem.
1143 if (XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
1144 trace_xfs_bdstrat_shut(bp
, _RET_IP_
);
1146 * Metadata write that didn't get logged but
1147 * written delayed anyway. These aren't associated
1148 * with a transaction, and can be ignored.
1150 if (!bp
->b_iodone
&& !XFS_BUF_ISREAD(bp
))
1151 return xfs_bioerror_relse(bp
);
1153 return xfs_bioerror(bp
);
1156 xfs_buf_iorequest(bp
);
1161 * Wrapper around bdstrat so that we can stop data from going to disk in case
1162 * we are shutting down the filesystem. Typically user data goes thru this
1163 * path; one of the exceptions is the superblock.
1167 struct xfs_mount
*mp
,
1170 if (XFS_FORCED_SHUTDOWN(mp
)) {
1171 trace_xfs_bdstrat_shut(bp
, _RET_IP_
);
1172 xfs_bioerror_relse(bp
);
1176 xfs_buf_iorequest(bp
);
1184 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1185 bp
->b_flags
&= ~_XBF_PAGE_LOCKED
;
1186 xfs_buf_ioend(bp
, schedule
);
1195 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1196 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1197 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1199 xfs_buf_ioerror(bp
, -error
);
1201 if (!error
&& xfs_buf_is_vmapped(bp
) && (bp
->b_flags
& XBF_READ
))
1202 invalidate_kernel_vmap_range(bp
->b_addr
, xfs_buf_vmap_len(bp
));
1205 struct page
*page
= bvec
->bv_page
;
1207 ASSERT(!PagePrivate(page
));
1208 if (unlikely(bp
->b_error
)) {
1209 if (bp
->b_flags
& XBF_READ
)
1210 ClearPageUptodate(page
);
1211 } else if (blocksize
>= PAGE_CACHE_SIZE
) {
1212 SetPageUptodate(page
);
1213 } else if (!PagePrivate(page
) &&
1214 (bp
->b_flags
& _XBF_PAGE_CACHE
)) {
1215 set_page_region(page
, bvec
->bv_offset
, bvec
->bv_len
);
1218 if (--bvec
>= bio
->bi_io_vec
)
1219 prefetchw(&bvec
->bv_page
->flags
);
1221 if (bp
->b_flags
& _XBF_PAGE_LOCKED
)
1223 } while (bvec
>= bio
->bi_io_vec
);
1225 _xfs_buf_ioend(bp
, 1);
1233 int rw
, map_i
, total_nr_pages
, nr_pages
;
1235 int offset
= bp
->b_offset
;
1236 int size
= bp
->b_count_desired
;
1237 sector_t sector
= bp
->b_bn
;
1238 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1240 total_nr_pages
= bp
->b_page_count
;
1243 if (bp
->b_flags
& XBF_ORDERED
) {
1244 ASSERT(!(bp
->b_flags
& XBF_READ
));
1245 rw
= WRITE_FLUSH_FUA
;
1246 } else if (bp
->b_flags
& XBF_LOG_BUFFER
) {
1247 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1248 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1249 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_SYNC
: READ_SYNC
;
1250 } else if (bp
->b_flags
& _XBF_RUN_QUEUES
) {
1251 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1252 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1253 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_META
: READ_META
;
1255 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE
:
1256 (bp
->b_flags
& XBF_READ_AHEAD
) ? READA
: READ
;
1259 /* Special code path for reading a sub page size buffer in --
1260 * we populate up the whole page, and hence the other metadata
1261 * in the same page. This optimization is only valid when the
1262 * filesystem block size is not smaller than the page size.
1264 if ((bp
->b_buffer_length
< PAGE_CACHE_SIZE
) &&
1265 ((bp
->b_flags
& (XBF_READ
|_XBF_PAGE_LOCKED
)) ==
1266 (XBF_READ
|_XBF_PAGE_LOCKED
)) &&
1267 (blocksize
>= PAGE_CACHE_SIZE
)) {
1268 bio
= bio_alloc(GFP_NOIO
, 1);
1270 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1271 bio
->bi_sector
= sector
- (offset
>> BBSHIFT
);
1272 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1273 bio
->bi_private
= bp
;
1275 bio_add_page(bio
, bp
->b_pages
[0], PAGE_CACHE_SIZE
, 0);
1278 atomic_inc(&bp
->b_io_remaining
);
1284 atomic_inc(&bp
->b_io_remaining
);
1285 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1286 if (nr_pages
> total_nr_pages
)
1287 nr_pages
= total_nr_pages
;
1289 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1290 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1291 bio
->bi_sector
= sector
;
1292 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1293 bio
->bi_private
= bp
;
1295 for (; size
&& nr_pages
; nr_pages
--, map_i
++) {
1296 int rbytes
, nbytes
= PAGE_CACHE_SIZE
- offset
;
1301 rbytes
= bio_add_page(bio
, bp
->b_pages
[map_i
], nbytes
, offset
);
1302 if (rbytes
< nbytes
)
1306 sector
+= nbytes
>> BBSHIFT
;
1312 if (likely(bio
->bi_size
)) {
1313 if (xfs_buf_is_vmapped(bp
)) {
1314 flush_kernel_vmap_range(bp
->b_addr
,
1315 xfs_buf_vmap_len(bp
));
1317 submit_bio(rw
, bio
);
1322 * if we get here, no pages were added to the bio. However,
1323 * we can't just error out here - if the pages are locked then
1324 * we have to unlock them otherwise we can hang on a later
1325 * access to the page.
1327 xfs_buf_ioerror(bp
, EIO
);
1328 if (bp
->b_flags
& _XBF_PAGE_LOCKED
) {
1330 for (i
= 0; i
< bp
->b_page_count
; i
++)
1331 unlock_page(bp
->b_pages
[i
]);
1341 trace_xfs_buf_iorequest(bp
, _RET_IP_
);
1343 if (bp
->b_flags
& XBF_DELWRI
) {
1344 xfs_buf_delwri_queue(bp
, 1);
1348 if (bp
->b_flags
& XBF_WRITE
) {
1349 xfs_buf_wait_unpin(bp
);
1354 /* Set the count to 1 initially, this will stop an I/O
1355 * completion callout which happens before we have started
1356 * all the I/O from calling xfs_buf_ioend too early.
1358 atomic_set(&bp
->b_io_remaining
, 1);
1359 _xfs_buf_ioapply(bp
);
1360 _xfs_buf_ioend(bp
, 0);
1367 * Waits for I/O to complete on the buffer supplied.
1368 * It returns immediately if no I/O is pending.
1369 * It returns the I/O error code, if any, or 0 if there was no error.
1375 trace_xfs_buf_iowait(bp
, _RET_IP_
);
1377 if (atomic_read(&bp
->b_io_remaining
))
1378 blk_run_address_space(bp
->b_target
->bt_mapping
);
1379 wait_for_completion(&bp
->b_iowait
);
1381 trace_xfs_buf_iowait_done(bp
, _RET_IP_
);
1392 if (bp
->b_flags
& XBF_MAPPED
)
1393 return XFS_BUF_PTR(bp
) + offset
;
1395 offset
+= bp
->b_offset
;
1396 page
= bp
->b_pages
[offset
>> PAGE_CACHE_SHIFT
];
1397 return (xfs_caddr_t
)page_address(page
) + (offset
& (PAGE_CACHE_SIZE
-1));
1401 * Move data into or out of a buffer.
1405 xfs_buf_t
*bp
, /* buffer to process */
1406 size_t boff
, /* starting buffer offset */
1407 size_t bsize
, /* length to copy */
1408 void *data
, /* data address */
1409 xfs_buf_rw_t mode
) /* read/write/zero flag */
1411 size_t bend
, cpoff
, csize
;
1414 bend
= boff
+ bsize
;
1415 while (boff
< bend
) {
1416 page
= bp
->b_pages
[xfs_buf_btoct(boff
+ bp
->b_offset
)];
1417 cpoff
= xfs_buf_poff(boff
+ bp
->b_offset
);
1418 csize
= min_t(size_t,
1419 PAGE_CACHE_SIZE
-cpoff
, bp
->b_count_desired
-boff
);
1421 ASSERT(((csize
+ cpoff
) <= PAGE_CACHE_SIZE
));
1425 memset(page_address(page
) + cpoff
, 0, csize
);
1428 memcpy(data
, page_address(page
) + cpoff
, csize
);
1431 memcpy(page_address(page
) + cpoff
, data
, csize
);
1440 * Handling of buffer targets (buftargs).
1444 * Wait for any bufs with callbacks that have been submitted but
1445 * have not yet returned... walk the hash list for the target.
1449 struct xfs_buftarg
*btp
)
1451 struct xfs_perag
*pag
;
1454 for (i
= 0; i
< btp
->bt_mount
->m_sb
.sb_agcount
; i
++) {
1455 pag
= xfs_perag_get(btp
->bt_mount
, i
);
1456 spin_lock(&pag
->pag_buf_lock
);
1457 while (rb_first(&pag
->pag_buf_tree
)) {
1458 spin_unlock(&pag
->pag_buf_lock
);
1460 spin_lock(&pag
->pag_buf_lock
);
1462 spin_unlock(&pag
->pag_buf_lock
);
1468 * buftarg list for delwrite queue processing
1470 static LIST_HEAD(xfs_buftarg_list
);
1471 static DEFINE_SPINLOCK(xfs_buftarg_lock
);
1474 xfs_register_buftarg(
1477 spin_lock(&xfs_buftarg_lock
);
1478 list_add(&btp
->bt_list
, &xfs_buftarg_list
);
1479 spin_unlock(&xfs_buftarg_lock
);
1483 xfs_unregister_buftarg(
1486 spin_lock(&xfs_buftarg_lock
);
1487 list_del(&btp
->bt_list
);
1488 spin_unlock(&xfs_buftarg_lock
);
1493 struct xfs_mount
*mp
,
1494 struct xfs_buftarg
*btp
)
1496 xfs_flush_buftarg(btp
, 1);
1497 if (mp
->m_flags
& XFS_MOUNT_BARRIER
)
1498 xfs_blkdev_issue_flush(btp
);
1499 iput(btp
->bt_mapping
->host
);
1501 /* Unregister the buftarg first so that we don't get a
1502 * wakeup finding a non-existent task
1504 xfs_unregister_buftarg(btp
);
1505 kthread_stop(btp
->bt_task
);
1511 xfs_setsize_buftarg_flags(
1513 unsigned int blocksize
,
1514 unsigned int sectorsize
,
1517 btp
->bt_bsize
= blocksize
;
1518 btp
->bt_sshift
= ffs(sectorsize
) - 1;
1519 btp
->bt_smask
= sectorsize
- 1;
1521 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1523 "XFS: Cannot set_blocksize to %u on device %s\n",
1524 sectorsize
, XFS_BUFTARG_NAME(btp
));
1529 (PAGE_CACHE_SIZE
/ BITS_PER_LONG
) > sectorsize
) {
1531 "XFS: %u byte sectors in use on device %s. "
1532 "This is suboptimal; %u or greater is ideal.\n",
1533 sectorsize
, XFS_BUFTARG_NAME(btp
),
1534 (unsigned int)PAGE_CACHE_SIZE
/ BITS_PER_LONG
);
1541 * When allocating the initial buffer target we have not yet
1542 * read in the superblock, so don't know what sized sectors
1543 * are being used is at this early stage. Play safe.
1546 xfs_setsize_buftarg_early(
1548 struct block_device
*bdev
)
1550 return xfs_setsize_buftarg_flags(btp
,
1551 PAGE_CACHE_SIZE
, bdev_logical_block_size(bdev
), 0);
1555 xfs_setsize_buftarg(
1557 unsigned int blocksize
,
1558 unsigned int sectorsize
)
1560 return xfs_setsize_buftarg_flags(btp
, blocksize
, sectorsize
, 1);
1564 xfs_mapping_buftarg(
1566 struct block_device
*bdev
)
1568 struct backing_dev_info
*bdi
;
1569 struct inode
*inode
;
1570 struct address_space
*mapping
;
1571 static const struct address_space_operations mapping_aops
= {
1572 .sync_page
= block_sync_page
,
1573 .migratepage
= fail_migrate_page
,
1576 inode
= new_inode(bdev
->bd_inode
->i_sb
);
1579 "XFS: Cannot allocate mapping inode for device %s\n",
1580 XFS_BUFTARG_NAME(btp
));
1583 inode
->i_ino
= get_next_ino();
1584 inode
->i_mode
= S_IFBLK
;
1585 inode
->i_bdev
= bdev
;
1586 inode
->i_rdev
= bdev
->bd_dev
;
1587 bdi
= blk_get_backing_dev_info(bdev
);
1589 bdi
= &default_backing_dev_info
;
1590 mapping
= &inode
->i_data
;
1591 mapping
->a_ops
= &mapping_aops
;
1592 mapping
->backing_dev_info
= bdi
;
1593 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
1594 btp
->bt_mapping
= mapping
;
1599 xfs_alloc_delwrite_queue(
1605 INIT_LIST_HEAD(&btp
->bt_list
);
1606 INIT_LIST_HEAD(&btp
->bt_delwrite_queue
);
1607 spin_lock_init(&btp
->bt_delwrite_lock
);
1609 btp
->bt_task
= kthread_run(xfsbufd
, btp
, "xfsbufd/%s", fsname
);
1610 if (IS_ERR(btp
->bt_task
)) {
1611 error
= PTR_ERR(btp
->bt_task
);
1614 xfs_register_buftarg(btp
);
1621 struct xfs_mount
*mp
,
1622 struct block_device
*bdev
,
1628 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
);
1631 btp
->bt_dev
= bdev
->bd_dev
;
1632 btp
->bt_bdev
= bdev
;
1633 if (xfs_setsize_buftarg_early(btp
, bdev
))
1635 if (xfs_mapping_buftarg(btp
, bdev
))
1637 if (xfs_alloc_delwrite_queue(btp
, fsname
))
1648 * Delayed write buffer handling
1651 xfs_buf_delwri_queue(
1655 struct list_head
*dwq
= &bp
->b_target
->bt_delwrite_queue
;
1656 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1658 trace_xfs_buf_delwri_queue(bp
, _RET_IP_
);
1660 ASSERT((bp
->b_flags
&(XBF_DELWRI
|XBF_ASYNC
)) == (XBF_DELWRI
|XBF_ASYNC
));
1663 /* If already in the queue, dequeue and place at tail */
1664 if (!list_empty(&bp
->b_list
)) {
1665 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1667 atomic_dec(&bp
->b_hold
);
1668 list_del(&bp
->b_list
);
1671 if (list_empty(dwq
)) {
1672 /* start xfsbufd as it is about to have something to do */
1673 wake_up_process(bp
->b_target
->bt_task
);
1676 bp
->b_flags
|= _XBF_DELWRI_Q
;
1677 list_add_tail(&bp
->b_list
, dwq
);
1678 bp
->b_queuetime
= jiffies
;
1686 xfs_buf_delwri_dequeue(
1689 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1693 if ((bp
->b_flags
& XBF_DELWRI
) && !list_empty(&bp
->b_list
)) {
1694 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1695 list_del_init(&bp
->b_list
);
1698 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
);
1704 trace_xfs_buf_delwri_dequeue(bp
, _RET_IP_
);
1708 * If a delwri buffer needs to be pushed before it has aged out, then promote
1709 * it to the head of the delwri queue so that it will be flushed on the next
1710 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1711 * than the age currently needed to flush the buffer. Hence the next time the
1712 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1715 xfs_buf_delwri_promote(
1718 struct xfs_buftarg
*btp
= bp
->b_target
;
1719 long age
= xfs_buf_age_centisecs
* msecs_to_jiffies(10) + 1;
1721 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1722 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1725 * Check the buffer age before locking the delayed write queue as we
1726 * don't need to promote buffers that are already past the flush age.
1728 if (bp
->b_queuetime
< jiffies
- age
)
1730 bp
->b_queuetime
= jiffies
- age
;
1731 spin_lock(&btp
->bt_delwrite_lock
);
1732 list_move(&bp
->b_list
, &btp
->bt_delwrite_queue
);
1733 spin_unlock(&btp
->bt_delwrite_lock
);
1737 xfs_buf_runall_queues(
1738 struct workqueue_struct
*queue
)
1740 flush_workqueue(queue
);
1745 struct shrinker
*shrink
,
1751 spin_lock(&xfs_buftarg_lock
);
1752 list_for_each_entry(btp
, &xfs_buftarg_list
, bt_list
) {
1753 if (test_bit(XBT_FORCE_SLEEP
, &btp
->bt_flags
))
1755 if (list_empty(&btp
->bt_delwrite_queue
))
1757 set_bit(XBT_FORCE_FLUSH
, &btp
->bt_flags
);
1758 wake_up_process(btp
->bt_task
);
1760 spin_unlock(&xfs_buftarg_lock
);
1765 * Move as many buffers as specified to the supplied list
1766 * idicating if we skipped any buffers to prevent deadlocks.
1769 xfs_buf_delwri_split(
1770 xfs_buftarg_t
*target
,
1771 struct list_head
*list
,
1775 struct list_head
*dwq
= &target
->bt_delwrite_queue
;
1776 spinlock_t
*dwlk
= &target
->bt_delwrite_lock
;
1780 force
= test_and_clear_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1781 INIT_LIST_HEAD(list
);
1783 list_for_each_entry_safe(bp
, n
, dwq
, b_list
) {
1784 trace_xfs_buf_delwri_split(bp
, _RET_IP_
);
1785 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1787 if (!XFS_BUF_ISPINNED(bp
) && !xfs_buf_cond_lock(bp
)) {
1789 time_before(jiffies
, bp
->b_queuetime
+ age
)) {
1794 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
|
1796 bp
->b_flags
|= XBF_WRITE
;
1797 list_move_tail(&bp
->b_list
, list
);
1808 * Compare function is more complex than it needs to be because
1809 * the return value is only 32 bits and we are doing comparisons
1815 struct list_head
*a
,
1816 struct list_head
*b
)
1818 struct xfs_buf
*ap
= container_of(a
, struct xfs_buf
, b_list
);
1819 struct xfs_buf
*bp
= container_of(b
, struct xfs_buf
, b_list
);
1822 diff
= ap
->b_bn
- bp
->b_bn
;
1831 xfs_buf_delwri_sort(
1832 xfs_buftarg_t
*target
,
1833 struct list_head
*list
)
1835 list_sort(NULL
, list
, xfs_buf_cmp
);
1842 xfs_buftarg_t
*target
= (xfs_buftarg_t
*)data
;
1844 current
->flags
|= PF_MEMALLOC
;
1849 long age
= xfs_buf_age_centisecs
* msecs_to_jiffies(10);
1850 long tout
= xfs_buf_timer_centisecs
* msecs_to_jiffies(10);
1852 struct list_head tmp
;
1854 if (unlikely(freezing(current
))) {
1855 set_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1858 clear_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1861 /* sleep for a long time if there is nothing to do. */
1862 if (list_empty(&target
->bt_delwrite_queue
))
1863 tout
= MAX_SCHEDULE_TIMEOUT
;
1864 schedule_timeout_interruptible(tout
);
1866 xfs_buf_delwri_split(target
, &tmp
, age
);
1867 list_sort(NULL
, &tmp
, xfs_buf_cmp
);
1868 while (!list_empty(&tmp
)) {
1870 bp
= list_first_entry(&tmp
, struct xfs_buf
, b_list
);
1871 list_del_init(&bp
->b_list
);
1876 blk_run_address_space(target
->bt_mapping
);
1878 } while (!kthread_should_stop());
1884 * Go through all incore buffers, and release buffers if they belong to
1885 * the given device. This is used in filesystem error handling to
1886 * preserve the consistency of its metadata.
1890 xfs_buftarg_t
*target
,
1895 LIST_HEAD(tmp_list
);
1896 LIST_HEAD(wait_list
);
1898 xfs_buf_runall_queues(xfsconvertd_workqueue
);
1899 xfs_buf_runall_queues(xfsdatad_workqueue
);
1900 xfs_buf_runall_queues(xfslogd_workqueue
);
1902 set_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1903 pincount
= xfs_buf_delwri_split(target
, &tmp_list
, 0);
1906 * Dropped the delayed write list lock, now walk the temporary list.
1907 * All I/O is issued async and then if we need to wait for completion
1908 * we do that after issuing all the IO.
1910 list_sort(NULL
, &tmp_list
, xfs_buf_cmp
);
1911 while (!list_empty(&tmp_list
)) {
1912 bp
= list_first_entry(&tmp_list
, struct xfs_buf
, b_list
);
1913 ASSERT(target
== bp
->b_target
);
1914 list_del_init(&bp
->b_list
);
1916 bp
->b_flags
&= ~XBF_ASYNC
;
1917 list_add(&bp
->b_list
, &wait_list
);
1923 /* Expedite and wait for IO to complete. */
1924 blk_run_address_space(target
->bt_mapping
);
1925 while (!list_empty(&wait_list
)) {
1926 bp
= list_first_entry(&wait_list
, struct xfs_buf
, b_list
);
1928 list_del_init(&bp
->b_list
);
1940 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1941 KM_ZONE_HWALIGN
, NULL
);
1945 xfslogd_workqueue
= alloc_workqueue("xfslogd",
1946 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 1);
1947 if (!xfslogd_workqueue
)
1948 goto out_free_buf_zone
;
1950 xfsdatad_workqueue
= create_workqueue("xfsdatad");
1951 if (!xfsdatad_workqueue
)
1952 goto out_destroy_xfslogd_workqueue
;
1954 xfsconvertd_workqueue
= create_workqueue("xfsconvertd");
1955 if (!xfsconvertd_workqueue
)
1956 goto out_destroy_xfsdatad_workqueue
;
1958 register_shrinker(&xfs_buf_shake
);
1961 out_destroy_xfsdatad_workqueue
:
1962 destroy_workqueue(xfsdatad_workqueue
);
1963 out_destroy_xfslogd_workqueue
:
1964 destroy_workqueue(xfslogd_workqueue
);
1966 kmem_zone_destroy(xfs_buf_zone
);
1972 xfs_buf_terminate(void)
1974 unregister_shrinker(&xfs_buf_shake
);
1975 destroy_workqueue(xfsconvertd_workqueue
);
1976 destroy_workqueue(xfsdatad_workqueue
);
1977 destroy_workqueue(xfslogd_workqueue
);
1978 kmem_zone_destroy(xfs_buf_zone
);
1981 #ifdef CONFIG_KDB_MODULES
1983 xfs_get_buftarg_list(void)
1985 return &xfs_buftarg_list
;