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 INIT_LIST_HEAD(&bp
->b_hash_list
);
192 init_MUTEX_LOCKED(&bp
->b_sema
); /* 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 ASSERT(list_empty(&bp
->b_hash_list
));
267 if (bp
->b_flags
& (_XBF_PAGE_CACHE
|_XBF_PAGES
)) {
270 if (xfs_buf_is_vmapped(bp
))
271 vm_unmap_ram(bp
->b_addr
- bp
->b_offset
,
274 for (i
= 0; i
< bp
->b_page_count
; i
++) {
275 struct page
*page
= bp
->b_pages
[i
];
277 if (bp
->b_flags
& _XBF_PAGE_CACHE
)
278 ASSERT(!PagePrivate(page
));
279 page_cache_release(page
);
282 _xfs_buf_free_pages(bp
);
283 xfs_buf_deallocate(bp
);
287 * Finds all pages for buffer in question and builds it's page list.
290 _xfs_buf_lookup_pages(
294 struct address_space
*mapping
= bp
->b_target
->bt_mapping
;
295 size_t blocksize
= bp
->b_target
->bt_bsize
;
296 size_t size
= bp
->b_count_desired
;
297 size_t nbytes
, offset
;
298 gfp_t gfp_mask
= xb_to_gfp(flags
);
299 unsigned short page_count
, i
;
304 end
= bp
->b_file_offset
+ bp
->b_buffer_length
;
305 page_count
= xfs_buf_btoc(end
) - xfs_buf_btoct(bp
->b_file_offset
);
307 error
= _xfs_buf_get_pages(bp
, page_count
, flags
);
310 bp
->b_flags
|= _XBF_PAGE_CACHE
;
312 offset
= bp
->b_offset
;
313 first
= bp
->b_file_offset
>> PAGE_CACHE_SHIFT
;
315 for (i
= 0; i
< bp
->b_page_count
; i
++) {
320 page
= find_or_create_page(mapping
, first
+ i
, gfp_mask
);
321 if (unlikely(page
== NULL
)) {
322 if (flags
& XBF_READ_AHEAD
) {
323 bp
->b_page_count
= i
;
324 for (i
= 0; i
< bp
->b_page_count
; i
++)
325 unlock_page(bp
->b_pages
[i
]);
330 * This could deadlock.
332 * But until all the XFS lowlevel code is revamped to
333 * handle buffer allocation failures we can't do much.
335 if (!(++retries
% 100))
337 "XFS: possible memory allocation "
338 "deadlock in %s (mode:0x%x)\n",
341 XFS_STATS_INC(xb_page_retries
);
342 xfsbufd_wakeup(NULL
, 0, gfp_mask
);
343 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
347 XFS_STATS_INC(xb_page_found
);
349 nbytes
= min_t(size_t, size
, PAGE_CACHE_SIZE
- offset
);
352 ASSERT(!PagePrivate(page
));
353 if (!PageUptodate(page
)) {
355 if (blocksize
>= PAGE_CACHE_SIZE
) {
356 if (flags
& XBF_READ
)
357 bp
->b_flags
|= _XBF_PAGE_LOCKED
;
358 } else if (!PagePrivate(page
)) {
359 if (test_page_region(page
, offset
, nbytes
))
364 bp
->b_pages
[i
] = page
;
368 if (!(bp
->b_flags
& _XBF_PAGE_LOCKED
)) {
369 for (i
= 0; i
< bp
->b_page_count
; i
++)
370 unlock_page(bp
->b_pages
[i
]);
373 if (page_count
== bp
->b_page_count
)
374 bp
->b_flags
|= XBF_DONE
;
380 * Map buffer into kernel address-space if nessecary.
387 /* A single page buffer is always mappable */
388 if (bp
->b_page_count
== 1) {
389 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
390 bp
->b_flags
|= XBF_MAPPED
;
391 } else if (flags
& XBF_MAPPED
) {
392 bp
->b_addr
= vm_map_ram(bp
->b_pages
, bp
->b_page_count
,
394 if (unlikely(bp
->b_addr
== NULL
))
396 bp
->b_addr
+= bp
->b_offset
;
397 bp
->b_flags
|= XBF_MAPPED
;
404 * Finding and Reading Buffers
408 * Look up, and creates if absent, a lockable buffer for
409 * a given range of an inode. The buffer is returned
410 * locked. If other overlapping buffers exist, they are
411 * released before the new buffer is created and locked,
412 * which may imply that this call will block until those buffers
413 * are unlocked. No I/O is implied by this call.
417 xfs_buftarg_t
*btp
, /* block device target */
418 xfs_off_t ioff
, /* starting offset of range */
419 size_t isize
, /* length of range */
420 xfs_buf_flags_t flags
,
423 xfs_off_t range_base
;
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
));
435 hash
= &btp
->bt_hash
[hash_long((unsigned long)ioff
, btp
->bt_hashshift
)];
437 spin_lock(&hash
->bh_lock
);
439 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
440 ASSERT(btp
== bp
->b_target
);
441 if (bp
->b_file_offset
== range_base
&&
442 bp
->b_buffer_length
== range_length
) {
443 atomic_inc(&bp
->b_hold
);
450 _xfs_buf_initialize(new_bp
, btp
, range_base
,
451 range_length
, flags
);
452 new_bp
->b_hash
= hash
;
453 list_add(&new_bp
->b_hash_list
, &hash
->bh_list
);
455 XFS_STATS_INC(xb_miss_locked
);
458 spin_unlock(&hash
->bh_lock
);
462 spin_unlock(&hash
->bh_lock
);
464 /* Attempt to get the semaphore without sleeping,
465 * if this does not work then we need to drop the
466 * spinlock and do a hard attempt on the semaphore.
468 if (down_trylock(&bp
->b_sema
)) {
469 if (!(flags
& XBF_TRYLOCK
)) {
470 /* wait for buffer ownership */
472 XFS_STATS_INC(xb_get_locked_waited
);
474 /* We asked for a trylock and failed, no need
475 * to look at file offset and length here, we
476 * know that this buffer at least overlaps our
477 * buffer and is locked, therefore our buffer
478 * either does not exist, or is this buffer.
481 XFS_STATS_INC(xb_busy_locked
);
489 if (bp
->b_flags
& XBF_STALE
) {
490 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
491 bp
->b_flags
&= XBF_MAPPED
;
494 trace_xfs_buf_find(bp
, flags
, _RET_IP_
);
495 XFS_STATS_INC(xb_get_locked
);
500 * Assembles a buffer covering the specified range.
501 * Storage in memory for all portions of the buffer will be allocated,
502 * although backing storage may not be.
506 xfs_buftarg_t
*target
,/* target for buffer */
507 xfs_off_t ioff
, /* starting offset of range */
508 size_t isize
, /* length of range */
509 xfs_buf_flags_t flags
)
511 xfs_buf_t
*bp
, *new_bp
;
514 new_bp
= xfs_buf_allocate(flags
);
515 if (unlikely(!new_bp
))
518 bp
= _xfs_buf_find(target
, ioff
, isize
, flags
, new_bp
);
520 error
= _xfs_buf_lookup_pages(bp
, flags
);
524 xfs_buf_deallocate(new_bp
);
525 if (unlikely(bp
== NULL
))
529 for (i
= 0; i
< bp
->b_page_count
; i
++)
530 mark_page_accessed(bp
->b_pages
[i
]);
532 if (!(bp
->b_flags
& XBF_MAPPED
)) {
533 error
= _xfs_buf_map_pages(bp
, flags
);
534 if (unlikely(error
)) {
535 printk(KERN_WARNING
"%s: failed to map pages\n",
541 XFS_STATS_INC(xb_get
);
544 * Always fill in the block number now, the mapped cases can do
545 * their own overlay of this later.
548 bp
->b_count_desired
= bp
->b_buffer_length
;
550 trace_xfs_buf_get(bp
, flags
, _RET_IP_
);
554 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
563 xfs_buf_flags_t flags
)
567 ASSERT(!(flags
& (XBF_DELWRI
|XBF_WRITE
)));
568 ASSERT(bp
->b_bn
!= XFS_BUF_DADDR_NULL
);
570 bp
->b_flags
&= ~(XBF_WRITE
| XBF_ASYNC
| XBF_DELWRI
| \
571 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
572 bp
->b_flags
|= flags
& (XBF_READ
| XBF_ASYNC
| \
573 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
575 status
= xfs_buf_iorequest(bp
);
576 if (status
|| XFS_BUF_ISERROR(bp
) || (flags
& XBF_ASYNC
))
578 return xfs_buf_iowait(bp
);
583 xfs_buftarg_t
*target
,
586 xfs_buf_flags_t flags
)
592 bp
= xfs_buf_get(target
, ioff
, isize
, flags
);
594 trace_xfs_buf_read(bp
, flags
, _RET_IP_
);
596 if (!XFS_BUF_ISDONE(bp
)) {
597 XFS_STATS_INC(xb_get_read
);
598 _xfs_buf_read(bp
, flags
);
599 } else if (flags
& XBF_ASYNC
) {
601 * Read ahead call which is already satisfied,
606 /* We do not want read in the flags */
607 bp
->b_flags
&= ~XBF_READ
;
614 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
621 * If we are not low on memory then do the readahead in a deadlock
626 xfs_buftarg_t
*target
,
629 xfs_buf_flags_t flags
)
631 struct backing_dev_info
*bdi
;
633 bdi
= target
->bt_mapping
->backing_dev_info
;
634 if (bdi_read_congested(bdi
))
637 flags
|= (XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
);
638 xfs_buf_read(target
, ioff
, isize
, flags
);
642 * Read an uncached buffer from disk. Allocates and returns a locked
643 * buffer containing the disk contents or nothing.
646 xfs_buf_read_uncached(
647 struct xfs_mount
*mp
,
648 struct xfs_buftarg
*target
,
656 bp
= xfs_buf_get_uncached(target
, length
, flags
);
660 /* set up the buffer for a read IO */
662 XFS_BUF_SET_ADDR(bp
, daddr
);
667 error
= xfs_iowait(bp
);
668 if (error
|| bp
->b_error
) {
678 xfs_buftarg_t
*target
)
682 bp
= xfs_buf_allocate(0);
684 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
688 static inline struct page
*
692 if ((!is_vmalloc_addr(addr
))) {
693 return virt_to_page(addr
);
695 return vmalloc_to_page(addr
);
700 xfs_buf_associate_memory(
707 unsigned long pageaddr
;
708 unsigned long offset
;
712 pageaddr
= (unsigned long)mem
& PAGE_CACHE_MASK
;
713 offset
= (unsigned long)mem
- pageaddr
;
714 buflen
= PAGE_CACHE_ALIGN(len
+ offset
);
715 page_count
= buflen
>> PAGE_CACHE_SHIFT
;
717 /* Free any previous set of page pointers */
719 _xfs_buf_free_pages(bp
);
724 rval
= _xfs_buf_get_pages(bp
, page_count
, XBF_DONT_BLOCK
);
728 bp
->b_offset
= offset
;
730 for (i
= 0; i
< bp
->b_page_count
; i
++) {
731 bp
->b_pages
[i
] = mem_to_page((void *)pageaddr
);
732 pageaddr
+= PAGE_CACHE_SIZE
;
735 bp
->b_count_desired
= len
;
736 bp
->b_buffer_length
= buflen
;
737 bp
->b_flags
|= XBF_MAPPED
;
738 bp
->b_flags
&= ~_XBF_PAGE_LOCKED
;
744 xfs_buf_get_uncached(
745 struct xfs_buftarg
*target
,
749 unsigned long page_count
= PAGE_ALIGN(len
) >> PAGE_SHIFT
;
753 bp
= xfs_buf_allocate(0);
754 if (unlikely(bp
== NULL
))
756 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
758 error
= _xfs_buf_get_pages(bp
, page_count
, 0);
762 for (i
= 0; i
< page_count
; i
++) {
763 bp
->b_pages
[i
] = alloc_page(xb_to_gfp(flags
));
767 bp
->b_flags
|= _XBF_PAGES
;
769 error
= _xfs_buf_map_pages(bp
, XBF_MAPPED
);
770 if (unlikely(error
)) {
771 printk(KERN_WARNING
"%s: failed to map pages\n",
778 trace_xfs_buf_get_uncached(bp
, _RET_IP_
);
783 __free_page(bp
->b_pages
[i
]);
784 _xfs_buf_free_pages(bp
);
786 xfs_buf_deallocate(bp
);
792 * Increment reference count on buffer, to hold the buffer concurrently
793 * with another thread which may release (free) the buffer asynchronously.
794 * Must hold the buffer already to call this function.
800 trace_xfs_buf_hold(bp
, _RET_IP_
);
801 atomic_inc(&bp
->b_hold
);
805 * Releases a hold on the specified buffer. If the
806 * the hold count is 1, calls xfs_buf_free.
812 xfs_bufhash_t
*hash
= bp
->b_hash
;
814 trace_xfs_buf_rele(bp
, _RET_IP_
);
816 if (unlikely(!hash
)) {
817 ASSERT(!bp
->b_relse
);
818 if (atomic_dec_and_test(&bp
->b_hold
))
823 ASSERT(atomic_read(&bp
->b_hold
) > 0);
824 if (atomic_dec_and_lock(&bp
->b_hold
, &hash
->bh_lock
)) {
826 atomic_inc(&bp
->b_hold
);
827 spin_unlock(&hash
->bh_lock
);
828 (*(bp
->b_relse
)) (bp
);
829 } else if (bp
->b_flags
& XBF_FS_MANAGED
) {
830 spin_unlock(&hash
->bh_lock
);
832 ASSERT(!(bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)));
833 list_del_init(&bp
->b_hash_list
);
834 spin_unlock(&hash
->bh_lock
);
842 * Mutual exclusion on buffers. Locking model:
844 * Buffers associated with inodes for which buffer locking
845 * is not enabled are not protected by semaphores, and are
846 * assumed to be exclusively owned by the caller. There is a
847 * spinlock in the buffer, used by the caller when concurrent
848 * access is possible.
852 * Locks a buffer object, if it is not already locked.
853 * Note that this in no way locks the underlying pages, so it is only
854 * useful for synchronizing concurrent use of buffer objects, not for
855 * synchronizing independent access to the underlying pages.
863 locked
= down_trylock(&bp
->b_sema
) == 0;
867 trace_xfs_buf_cond_lock(bp
, _RET_IP_
);
868 return locked
? 0 : -EBUSY
;
875 return bp
->b_sema
.count
;
879 * Locks a buffer object.
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.
884 * If we come across a stale, pinned, locked buffer, we know that we
885 * are being asked to lock a buffer that has been reallocated. Because
886 * it is pinned, we know that the log has not been pushed to disk and
887 * hence it will still be locked. Rather than sleeping until someone
888 * else pushes the log, push it ourselves before trying to get the lock.
894 trace_xfs_buf_lock(bp
, _RET_IP_
);
896 if (atomic_read(&bp
->b_pin_count
) && (bp
->b_flags
& XBF_STALE
))
897 xfs_log_force(bp
->b_target
->bt_mount
, 0);
898 if (atomic_read(&bp
->b_io_remaining
))
899 blk_run_address_space(bp
->b_target
->bt_mapping
);
903 trace_xfs_buf_lock_done(bp
, _RET_IP_
);
907 * Releases the lock on the buffer object.
908 * If the buffer is marked delwri but is not queued, do so before we
909 * unlock the buffer as we need to set flags correctly. We also need to
910 * take a reference for the delwri queue because the unlocker is going to
911 * drop their's and they don't know we just queued it.
917 if ((bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)) == XBF_DELWRI
) {
918 atomic_inc(&bp
->b_hold
);
919 bp
->b_flags
|= XBF_ASYNC
;
920 xfs_buf_delwri_queue(bp
, 0);
926 trace_xfs_buf_unlock(bp
, _RET_IP_
);
933 DECLARE_WAITQUEUE (wait
, current
);
935 if (atomic_read(&bp
->b_pin_count
) == 0)
938 add_wait_queue(&bp
->b_waiters
, &wait
);
940 set_current_state(TASK_UNINTERRUPTIBLE
);
941 if (atomic_read(&bp
->b_pin_count
) == 0)
943 if (atomic_read(&bp
->b_io_remaining
))
944 blk_run_address_space(bp
->b_target
->bt_mapping
);
947 remove_wait_queue(&bp
->b_waiters
, &wait
);
948 set_current_state(TASK_RUNNING
);
952 * Buffer Utility Routines
957 struct work_struct
*work
)
960 container_of(work
, xfs_buf_t
, b_iodone_work
);
963 * We can get an EOPNOTSUPP to ordered writes. Here we clear the
964 * ordered flag and reissue them. Because we can't tell the higher
965 * layers directly that they should not issue ordered I/O anymore, they
966 * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion.
968 if ((bp
->b_error
== EOPNOTSUPP
) &&
969 (bp
->b_flags
& (XBF_ORDERED
|XBF_ASYNC
)) == (XBF_ORDERED
|XBF_ASYNC
)) {
970 trace_xfs_buf_ordered_retry(bp
, _RET_IP_
);
971 bp
->b_flags
&= ~XBF_ORDERED
;
972 bp
->b_flags
|= _XFS_BARRIER_FAILED
;
973 xfs_buf_iorequest(bp
);
974 } else if (bp
->b_iodone
)
975 (*(bp
->b_iodone
))(bp
);
976 else if (bp
->b_flags
& XBF_ASYNC
)
985 trace_xfs_buf_iodone(bp
, _RET_IP_
);
987 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_READ_AHEAD
);
988 if (bp
->b_error
== 0)
989 bp
->b_flags
|= XBF_DONE
;
991 if ((bp
->b_iodone
) || (bp
->b_flags
& XBF_ASYNC
)) {
993 INIT_WORK(&bp
->b_iodone_work
, xfs_buf_iodone_work
);
994 queue_work(xfslogd_workqueue
, &bp
->b_iodone_work
);
996 xfs_buf_iodone_work(&bp
->b_iodone_work
);
999 complete(&bp
->b_iowait
);
1008 ASSERT(error
>= 0 && error
<= 0xffff);
1009 bp
->b_error
= (unsigned short)error
;
1010 trace_xfs_buf_ioerror(bp
, error
, _RET_IP_
);
1015 struct xfs_mount
*mp
,
1020 bp
->b_flags
|= XBF_WRITE
;
1021 bp
->b_flags
&= ~(XBF_ASYNC
| XBF_READ
);
1023 xfs_buf_delwri_dequeue(bp
);
1026 error
= xfs_buf_iowait(bp
);
1028 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1038 trace_xfs_buf_bdwrite(bp
, _RET_IP_
);
1040 bp
->b_flags
&= ~XBF_READ
;
1041 bp
->b_flags
|= (XBF_DELWRI
| XBF_ASYNC
);
1043 xfs_buf_delwri_queue(bp
, 1);
1047 * Called when we want to stop a buffer from getting written or read.
1048 * We attach the EIO error, muck with its flags, and call biodone
1049 * so that the proper iodone callbacks get called.
1055 #ifdef XFSERRORDEBUG
1056 ASSERT(XFS_BUF_ISREAD(bp
) || bp
->b_iodone
);
1060 * No need to wait until the buffer is unpinned, we aren't flushing it.
1062 XFS_BUF_ERROR(bp
, EIO
);
1065 * We're calling biodone, so delete XBF_DONE flag.
1068 XFS_BUF_UNDELAYWRITE(bp
);
1078 * Same as xfs_bioerror, except that we are releasing the buffer
1079 * here ourselves, and avoiding the biodone call.
1080 * This is meant for userdata errors; metadata bufs come with
1081 * iodone functions attached, so that we can track down errors.
1087 int64_t fl
= XFS_BUF_BFLAGS(bp
);
1089 * No need to wait until the buffer is unpinned.
1090 * We aren't flushing it.
1092 * chunkhold expects B_DONE to be set, whether
1093 * we actually finish the I/O or not. We don't want to
1094 * change that interface.
1097 XFS_BUF_UNDELAYWRITE(bp
);
1100 XFS_BUF_CLR_IODONE_FUNC(bp
);
1101 if (!(fl
& XBF_ASYNC
)) {
1103 * Mark b_error and B_ERROR _both_.
1104 * Lot's of chunkcache code assumes that.
1105 * There's no reason to mark error for
1108 XFS_BUF_ERROR(bp
, EIO
);
1109 XFS_BUF_FINISH_IOWAIT(bp
);
1119 * All xfs metadata buffers except log state machine buffers
1120 * get this attached as their b_bdstrat callback function.
1121 * This is so that we can catch a buffer
1122 * after prematurely unpinning it to forcibly shutdown the filesystem.
1128 if (XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
1129 trace_xfs_bdstrat_shut(bp
, _RET_IP_
);
1131 * Metadata write that didn't get logged but
1132 * written delayed anyway. These aren't associated
1133 * with a transaction, and can be ignored.
1135 if (!bp
->b_iodone
&& !XFS_BUF_ISREAD(bp
))
1136 return xfs_bioerror_relse(bp
);
1138 return xfs_bioerror(bp
);
1141 xfs_buf_iorequest(bp
);
1146 * Wrapper around bdstrat so that we can stop data from going to disk in case
1147 * we are shutting down the filesystem. Typically user data goes thru this
1148 * path; one of the exceptions is the superblock.
1152 struct xfs_mount
*mp
,
1155 if (XFS_FORCED_SHUTDOWN(mp
)) {
1156 trace_xfs_bdstrat_shut(bp
, _RET_IP_
);
1157 xfs_bioerror_relse(bp
);
1161 xfs_buf_iorequest(bp
);
1169 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1170 bp
->b_flags
&= ~_XBF_PAGE_LOCKED
;
1171 xfs_buf_ioend(bp
, schedule
);
1180 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1181 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1182 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1184 xfs_buf_ioerror(bp
, -error
);
1186 if (!error
&& xfs_buf_is_vmapped(bp
) && (bp
->b_flags
& XBF_READ
))
1187 invalidate_kernel_vmap_range(bp
->b_addr
, xfs_buf_vmap_len(bp
));
1190 struct page
*page
= bvec
->bv_page
;
1192 ASSERT(!PagePrivate(page
));
1193 if (unlikely(bp
->b_error
)) {
1194 if (bp
->b_flags
& XBF_READ
)
1195 ClearPageUptodate(page
);
1196 } else if (blocksize
>= PAGE_CACHE_SIZE
) {
1197 SetPageUptodate(page
);
1198 } else if (!PagePrivate(page
) &&
1199 (bp
->b_flags
& _XBF_PAGE_CACHE
)) {
1200 set_page_region(page
, bvec
->bv_offset
, bvec
->bv_len
);
1203 if (--bvec
>= bio
->bi_io_vec
)
1204 prefetchw(&bvec
->bv_page
->flags
);
1206 if (bp
->b_flags
& _XBF_PAGE_LOCKED
)
1208 } while (bvec
>= bio
->bi_io_vec
);
1210 _xfs_buf_ioend(bp
, 1);
1218 int rw
, map_i
, total_nr_pages
, nr_pages
;
1220 int offset
= bp
->b_offset
;
1221 int size
= bp
->b_count_desired
;
1222 sector_t sector
= bp
->b_bn
;
1223 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1225 total_nr_pages
= bp
->b_page_count
;
1228 if (bp
->b_flags
& XBF_ORDERED
) {
1229 ASSERT(!(bp
->b_flags
& XBF_READ
));
1231 } else if (bp
->b_flags
& XBF_LOG_BUFFER
) {
1232 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1233 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1234 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_SYNC
: READ_SYNC
;
1235 } else if (bp
->b_flags
& _XBF_RUN_QUEUES
) {
1236 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1237 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1238 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_META
: READ_META
;
1240 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE
:
1241 (bp
->b_flags
& XBF_READ_AHEAD
) ? READA
: READ
;
1244 /* Special code path for reading a sub page size buffer in --
1245 * we populate up the whole page, and hence the other metadata
1246 * in the same page. This optimization is only valid when the
1247 * filesystem block size is not smaller than the page size.
1249 if ((bp
->b_buffer_length
< PAGE_CACHE_SIZE
) &&
1250 ((bp
->b_flags
& (XBF_READ
|_XBF_PAGE_LOCKED
)) ==
1251 (XBF_READ
|_XBF_PAGE_LOCKED
)) &&
1252 (blocksize
>= PAGE_CACHE_SIZE
)) {
1253 bio
= bio_alloc(GFP_NOIO
, 1);
1255 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1256 bio
->bi_sector
= sector
- (offset
>> BBSHIFT
);
1257 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1258 bio
->bi_private
= bp
;
1260 bio_add_page(bio
, bp
->b_pages
[0], PAGE_CACHE_SIZE
, 0);
1263 atomic_inc(&bp
->b_io_remaining
);
1269 atomic_inc(&bp
->b_io_remaining
);
1270 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1271 if (nr_pages
> total_nr_pages
)
1272 nr_pages
= total_nr_pages
;
1274 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1275 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1276 bio
->bi_sector
= sector
;
1277 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1278 bio
->bi_private
= bp
;
1280 for (; size
&& nr_pages
; nr_pages
--, map_i
++) {
1281 int rbytes
, nbytes
= PAGE_CACHE_SIZE
- offset
;
1286 rbytes
= bio_add_page(bio
, bp
->b_pages
[map_i
], nbytes
, offset
);
1287 if (rbytes
< nbytes
)
1291 sector
+= nbytes
>> BBSHIFT
;
1297 if (likely(bio
->bi_size
)) {
1298 if (xfs_buf_is_vmapped(bp
)) {
1299 flush_kernel_vmap_range(bp
->b_addr
,
1300 xfs_buf_vmap_len(bp
));
1302 submit_bio(rw
, bio
);
1307 * if we get here, no pages were added to the bio. However,
1308 * we can't just error out here - if the pages are locked then
1309 * we have to unlock them otherwise we can hang on a later
1310 * access to the page.
1312 xfs_buf_ioerror(bp
, EIO
);
1313 if (bp
->b_flags
& _XBF_PAGE_LOCKED
) {
1315 for (i
= 0; i
< bp
->b_page_count
; i
++)
1316 unlock_page(bp
->b_pages
[i
]);
1326 trace_xfs_buf_iorequest(bp
, _RET_IP_
);
1328 if (bp
->b_flags
& XBF_DELWRI
) {
1329 xfs_buf_delwri_queue(bp
, 1);
1333 if (bp
->b_flags
& XBF_WRITE
) {
1334 xfs_buf_wait_unpin(bp
);
1339 /* Set the count to 1 initially, this will stop an I/O
1340 * completion callout which happens before we have started
1341 * all the I/O from calling xfs_buf_ioend too early.
1343 atomic_set(&bp
->b_io_remaining
, 1);
1344 _xfs_buf_ioapply(bp
);
1345 _xfs_buf_ioend(bp
, 0);
1352 * Waits for I/O to complete on the buffer supplied.
1353 * It returns immediately if no I/O is pending.
1354 * It returns the I/O error code, if any, or 0 if there was no error.
1360 trace_xfs_buf_iowait(bp
, _RET_IP_
);
1362 if (atomic_read(&bp
->b_io_remaining
))
1363 blk_run_address_space(bp
->b_target
->bt_mapping
);
1364 wait_for_completion(&bp
->b_iowait
);
1366 trace_xfs_buf_iowait_done(bp
, _RET_IP_
);
1377 if (bp
->b_flags
& XBF_MAPPED
)
1378 return XFS_BUF_PTR(bp
) + offset
;
1380 offset
+= bp
->b_offset
;
1381 page
= bp
->b_pages
[offset
>> PAGE_CACHE_SHIFT
];
1382 return (xfs_caddr_t
)page_address(page
) + (offset
& (PAGE_CACHE_SIZE
-1));
1386 * Move data into or out of a buffer.
1390 xfs_buf_t
*bp
, /* buffer to process */
1391 size_t boff
, /* starting buffer offset */
1392 size_t bsize
, /* length to copy */
1393 void *data
, /* data address */
1394 xfs_buf_rw_t mode
) /* read/write/zero flag */
1396 size_t bend
, cpoff
, csize
;
1399 bend
= boff
+ bsize
;
1400 while (boff
< bend
) {
1401 page
= bp
->b_pages
[xfs_buf_btoct(boff
+ bp
->b_offset
)];
1402 cpoff
= xfs_buf_poff(boff
+ bp
->b_offset
);
1403 csize
= min_t(size_t,
1404 PAGE_CACHE_SIZE
-cpoff
, bp
->b_count_desired
-boff
);
1406 ASSERT(((csize
+ cpoff
) <= PAGE_CACHE_SIZE
));
1410 memset(page_address(page
) + cpoff
, 0, csize
);
1413 memcpy(data
, page_address(page
) + cpoff
, csize
);
1416 memcpy(page_address(page
) + cpoff
, data
, csize
);
1425 * Handling of buffer targets (buftargs).
1429 * Wait for any bufs with callbacks that have been submitted but
1430 * have not yet returned... walk the hash list for the target.
1437 xfs_bufhash_t
*hash
;
1440 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1441 hash
= &btp
->bt_hash
[i
];
1443 spin_lock(&hash
->bh_lock
);
1444 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
1445 ASSERT(btp
== bp
->b_target
);
1446 if (!(bp
->b_flags
& XBF_FS_MANAGED
)) {
1447 spin_unlock(&hash
->bh_lock
);
1449 * Catch superblock reference count leaks
1452 BUG_ON(bp
->b_bn
== 0);
1457 spin_unlock(&hash
->bh_lock
);
1462 * Allocate buffer hash table for a given target.
1463 * For devices containing metadata (i.e. not the log/realtime devices)
1464 * we need to allocate a much larger hash table.
1473 btp
->bt_hashshift
= external
? 3 : 12; /* 8 or 4096 buckets */
1474 btp
->bt_hash
= kmem_zalloc_large((1 << btp
->bt_hashshift
) *
1475 sizeof(xfs_bufhash_t
));
1476 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1477 spin_lock_init(&btp
->bt_hash
[i
].bh_lock
);
1478 INIT_LIST_HEAD(&btp
->bt_hash
[i
].bh_list
);
1486 kmem_free_large(btp
->bt_hash
);
1487 btp
->bt_hash
= NULL
;
1491 * buftarg list for delwrite queue processing
1493 static LIST_HEAD(xfs_buftarg_list
);
1494 static DEFINE_SPINLOCK(xfs_buftarg_lock
);
1497 xfs_register_buftarg(
1500 spin_lock(&xfs_buftarg_lock
);
1501 list_add(&btp
->bt_list
, &xfs_buftarg_list
);
1502 spin_unlock(&xfs_buftarg_lock
);
1506 xfs_unregister_buftarg(
1509 spin_lock(&xfs_buftarg_lock
);
1510 list_del(&btp
->bt_list
);
1511 spin_unlock(&xfs_buftarg_lock
);
1516 struct xfs_mount
*mp
,
1517 struct xfs_buftarg
*btp
)
1519 xfs_flush_buftarg(btp
, 1);
1520 if (mp
->m_flags
& XFS_MOUNT_BARRIER
)
1521 xfs_blkdev_issue_flush(btp
);
1522 xfs_free_bufhash(btp
);
1523 iput(btp
->bt_mapping
->host
);
1525 /* Unregister the buftarg first so that we don't get a
1526 * wakeup finding a non-existent task
1528 xfs_unregister_buftarg(btp
);
1529 kthread_stop(btp
->bt_task
);
1535 xfs_setsize_buftarg_flags(
1537 unsigned int blocksize
,
1538 unsigned int sectorsize
,
1541 btp
->bt_bsize
= blocksize
;
1542 btp
->bt_sshift
= ffs(sectorsize
) - 1;
1543 btp
->bt_smask
= sectorsize
- 1;
1545 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1547 "XFS: Cannot set_blocksize to %u on device %s\n",
1548 sectorsize
, XFS_BUFTARG_NAME(btp
));
1553 (PAGE_CACHE_SIZE
/ BITS_PER_LONG
) > sectorsize
) {
1555 "XFS: %u byte sectors in use on device %s. "
1556 "This is suboptimal; %u or greater is ideal.\n",
1557 sectorsize
, XFS_BUFTARG_NAME(btp
),
1558 (unsigned int)PAGE_CACHE_SIZE
/ BITS_PER_LONG
);
1565 * When allocating the initial buffer target we have not yet
1566 * read in the superblock, so don't know what sized sectors
1567 * are being used is at this early stage. Play safe.
1570 xfs_setsize_buftarg_early(
1572 struct block_device
*bdev
)
1574 return xfs_setsize_buftarg_flags(btp
,
1575 PAGE_CACHE_SIZE
, bdev_logical_block_size(bdev
), 0);
1579 xfs_setsize_buftarg(
1581 unsigned int blocksize
,
1582 unsigned int sectorsize
)
1584 return xfs_setsize_buftarg_flags(btp
, blocksize
, sectorsize
, 1);
1588 xfs_mapping_buftarg(
1590 struct block_device
*bdev
)
1592 struct backing_dev_info
*bdi
;
1593 struct inode
*inode
;
1594 struct address_space
*mapping
;
1595 static const struct address_space_operations mapping_aops
= {
1596 .sync_page
= block_sync_page
,
1597 .migratepage
= fail_migrate_page
,
1600 inode
= new_inode(bdev
->bd_inode
->i_sb
);
1603 "XFS: Cannot allocate mapping inode for device %s\n",
1604 XFS_BUFTARG_NAME(btp
));
1607 inode
->i_mode
= S_IFBLK
;
1608 inode
->i_bdev
= bdev
;
1609 inode
->i_rdev
= bdev
->bd_dev
;
1610 bdi
= blk_get_backing_dev_info(bdev
);
1612 bdi
= &default_backing_dev_info
;
1613 mapping
= &inode
->i_data
;
1614 mapping
->a_ops
= &mapping_aops
;
1615 mapping
->backing_dev_info
= bdi
;
1616 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
1617 btp
->bt_mapping
= mapping
;
1622 xfs_alloc_delwrite_queue(
1628 INIT_LIST_HEAD(&btp
->bt_list
);
1629 INIT_LIST_HEAD(&btp
->bt_delwrite_queue
);
1630 spin_lock_init(&btp
->bt_delwrite_lock
);
1632 btp
->bt_task
= kthread_run(xfsbufd
, btp
, "xfsbufd/%s", fsname
);
1633 if (IS_ERR(btp
->bt_task
)) {
1634 error
= PTR_ERR(btp
->bt_task
);
1637 xfs_register_buftarg(btp
);
1644 struct xfs_mount
*mp
,
1645 struct block_device
*bdev
,
1651 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
);
1654 btp
->bt_dev
= bdev
->bd_dev
;
1655 btp
->bt_bdev
= bdev
;
1656 if (xfs_setsize_buftarg_early(btp
, bdev
))
1658 if (xfs_mapping_buftarg(btp
, bdev
))
1660 if (xfs_alloc_delwrite_queue(btp
, fsname
))
1662 xfs_alloc_bufhash(btp
, external
);
1672 * Delayed write buffer handling
1675 xfs_buf_delwri_queue(
1679 struct list_head
*dwq
= &bp
->b_target
->bt_delwrite_queue
;
1680 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1682 trace_xfs_buf_delwri_queue(bp
, _RET_IP_
);
1684 ASSERT((bp
->b_flags
&(XBF_DELWRI
|XBF_ASYNC
)) == (XBF_DELWRI
|XBF_ASYNC
));
1687 /* If already in the queue, dequeue and place at tail */
1688 if (!list_empty(&bp
->b_list
)) {
1689 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1691 atomic_dec(&bp
->b_hold
);
1692 list_del(&bp
->b_list
);
1695 if (list_empty(dwq
)) {
1696 /* start xfsbufd as it is about to have something to do */
1697 wake_up_process(bp
->b_target
->bt_task
);
1700 bp
->b_flags
|= _XBF_DELWRI_Q
;
1701 list_add_tail(&bp
->b_list
, dwq
);
1702 bp
->b_queuetime
= jiffies
;
1710 xfs_buf_delwri_dequeue(
1713 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1717 if ((bp
->b_flags
& XBF_DELWRI
) && !list_empty(&bp
->b_list
)) {
1718 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1719 list_del_init(&bp
->b_list
);
1722 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
);
1728 trace_xfs_buf_delwri_dequeue(bp
, _RET_IP_
);
1732 * If a delwri buffer needs to be pushed before it has aged out, then promote
1733 * it to the head of the delwri queue so that it will be flushed on the next
1734 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1735 * than the age currently needed to flush the buffer. Hence the next time the
1736 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1739 xfs_buf_delwri_promote(
1742 struct xfs_buftarg
*btp
= bp
->b_target
;
1743 long age
= xfs_buf_age_centisecs
* msecs_to_jiffies(10) + 1;
1745 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1746 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1749 * Check the buffer age before locking the delayed write queue as we
1750 * don't need to promote buffers that are already past the flush age.
1752 if (bp
->b_queuetime
< jiffies
- age
)
1754 bp
->b_queuetime
= jiffies
- age
;
1755 spin_lock(&btp
->bt_delwrite_lock
);
1756 list_move(&bp
->b_list
, &btp
->bt_delwrite_queue
);
1757 spin_unlock(&btp
->bt_delwrite_lock
);
1761 xfs_buf_runall_queues(
1762 struct workqueue_struct
*queue
)
1764 flush_workqueue(queue
);
1769 struct shrinker
*shrink
,
1775 spin_lock(&xfs_buftarg_lock
);
1776 list_for_each_entry(btp
, &xfs_buftarg_list
, bt_list
) {
1777 if (test_bit(XBT_FORCE_SLEEP
, &btp
->bt_flags
))
1779 if (list_empty(&btp
->bt_delwrite_queue
))
1781 set_bit(XBT_FORCE_FLUSH
, &btp
->bt_flags
);
1782 wake_up_process(btp
->bt_task
);
1784 spin_unlock(&xfs_buftarg_lock
);
1789 * Move as many buffers as specified to the supplied list
1790 * idicating if we skipped any buffers to prevent deadlocks.
1793 xfs_buf_delwri_split(
1794 xfs_buftarg_t
*target
,
1795 struct list_head
*list
,
1799 struct list_head
*dwq
= &target
->bt_delwrite_queue
;
1800 spinlock_t
*dwlk
= &target
->bt_delwrite_lock
;
1804 force
= test_and_clear_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1805 INIT_LIST_HEAD(list
);
1807 list_for_each_entry_safe(bp
, n
, dwq
, b_list
) {
1808 trace_xfs_buf_delwri_split(bp
, _RET_IP_
);
1809 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1811 if (!XFS_BUF_ISPINNED(bp
) && !xfs_buf_cond_lock(bp
)) {
1813 time_before(jiffies
, bp
->b_queuetime
+ age
)) {
1818 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
|
1820 bp
->b_flags
|= XBF_WRITE
;
1821 list_move_tail(&bp
->b_list
, list
);
1832 * Compare function is more complex than it needs to be because
1833 * the return value is only 32 bits and we are doing comparisons
1839 struct list_head
*a
,
1840 struct list_head
*b
)
1842 struct xfs_buf
*ap
= container_of(a
, struct xfs_buf
, b_list
);
1843 struct xfs_buf
*bp
= container_of(b
, struct xfs_buf
, b_list
);
1846 diff
= ap
->b_bn
- bp
->b_bn
;
1855 xfs_buf_delwri_sort(
1856 xfs_buftarg_t
*target
,
1857 struct list_head
*list
)
1859 list_sort(NULL
, list
, xfs_buf_cmp
);
1866 xfs_buftarg_t
*target
= (xfs_buftarg_t
*)data
;
1868 current
->flags
|= PF_MEMALLOC
;
1873 long age
= xfs_buf_age_centisecs
* msecs_to_jiffies(10);
1874 long tout
= xfs_buf_timer_centisecs
* msecs_to_jiffies(10);
1876 struct list_head tmp
;
1878 if (unlikely(freezing(current
))) {
1879 set_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1882 clear_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1885 /* sleep for a long time if there is nothing to do. */
1886 if (list_empty(&target
->bt_delwrite_queue
))
1887 tout
= MAX_SCHEDULE_TIMEOUT
;
1888 schedule_timeout_interruptible(tout
);
1890 xfs_buf_delwri_split(target
, &tmp
, age
);
1891 list_sort(NULL
, &tmp
, xfs_buf_cmp
);
1892 while (!list_empty(&tmp
)) {
1894 bp
= list_first_entry(&tmp
, struct xfs_buf
, b_list
);
1895 list_del_init(&bp
->b_list
);
1900 blk_run_address_space(target
->bt_mapping
);
1902 } while (!kthread_should_stop());
1908 * Go through all incore buffers, and release buffers if they belong to
1909 * the given device. This is used in filesystem error handling to
1910 * preserve the consistency of its metadata.
1914 xfs_buftarg_t
*target
,
1919 LIST_HEAD(tmp_list
);
1920 LIST_HEAD(wait_list
);
1922 xfs_buf_runall_queues(xfsconvertd_workqueue
);
1923 xfs_buf_runall_queues(xfsdatad_workqueue
);
1924 xfs_buf_runall_queues(xfslogd_workqueue
);
1926 set_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1927 pincount
= xfs_buf_delwri_split(target
, &tmp_list
, 0);
1930 * Dropped the delayed write list lock, now walk the temporary list.
1931 * All I/O is issued async and then if we need to wait for completion
1932 * we do that after issuing all the IO.
1934 list_sort(NULL
, &tmp_list
, xfs_buf_cmp
);
1935 while (!list_empty(&tmp_list
)) {
1936 bp
= list_first_entry(&tmp_list
, struct xfs_buf
, b_list
);
1937 ASSERT(target
== bp
->b_target
);
1938 list_del_init(&bp
->b_list
);
1940 bp
->b_flags
&= ~XBF_ASYNC
;
1941 list_add(&bp
->b_list
, &wait_list
);
1947 /* Expedite and wait for IO to complete. */
1948 blk_run_address_space(target
->bt_mapping
);
1949 while (!list_empty(&wait_list
)) {
1950 bp
= list_first_entry(&wait_list
, struct xfs_buf
, b_list
);
1952 list_del_init(&bp
->b_list
);
1964 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1965 KM_ZONE_HWALIGN
, NULL
);
1969 xfslogd_workqueue
= alloc_workqueue("xfslogd",
1970 WQ_RESCUER
| WQ_HIGHPRI
, 1);
1971 if (!xfslogd_workqueue
)
1972 goto out_free_buf_zone
;
1974 xfsdatad_workqueue
= create_workqueue("xfsdatad");
1975 if (!xfsdatad_workqueue
)
1976 goto out_destroy_xfslogd_workqueue
;
1978 xfsconvertd_workqueue
= create_workqueue("xfsconvertd");
1979 if (!xfsconvertd_workqueue
)
1980 goto out_destroy_xfsdatad_workqueue
;
1982 register_shrinker(&xfs_buf_shake
);
1985 out_destroy_xfsdatad_workqueue
:
1986 destroy_workqueue(xfsdatad_workqueue
);
1987 out_destroy_xfslogd_workqueue
:
1988 destroy_workqueue(xfslogd_workqueue
);
1990 kmem_zone_destroy(xfs_buf_zone
);
1996 xfs_buf_terminate(void)
1998 unregister_shrinker(&xfs_buf_shake
);
1999 destroy_workqueue(xfsconvertd_workqueue
);
2000 destroy_workqueue(xfsdatad_workqueue
);
2001 destroy_workqueue(xfslogd_workqueue
);
2002 kmem_zone_destroy(xfs_buf_zone
);
2005 #ifdef CONFIG_KDB_MODULES
2007 xfs_get_buftarg_list(void)
2009 return &xfs_buftarg_list
;