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/slab.h>
22 #include <linux/pagemap.h>
23 #include <linux/init.h>
24 #include <linux/vmalloc.h>
25 #include <linux/bio.h>
26 #include <linux/sysctl.h>
27 #include <linux/proc_fs.h>
28 #include <linux/workqueue.h>
29 #include <linux/percpu.h>
30 #include <linux/blkdev.h>
31 #include <linux/hash.h>
32 #include <linux/kthread.h>
33 #include <linux/migrate.h>
34 #include <linux/backing-dev.h>
35 #include <linux/freezer.h>
40 #include "xfs_dmapi.h"
41 #include "xfs_mount.h"
43 static kmem_zone_t
*xfs_buf_zone
;
44 STATIC
int xfsbufd(void *);
45 STATIC
int xfsbufd_wakeup(int, gfp_t
);
46 STATIC
void xfs_buf_delwri_queue(xfs_buf_t
*, int);
47 static struct shrinker xfs_buf_shake
= {
48 .shrink
= xfsbufd_wakeup
,
49 .seeks
= DEFAULT_SEEKS
,
52 static struct workqueue_struct
*xfslogd_workqueue
;
53 struct workqueue_struct
*xfsdatad_workqueue
;
63 ktrace_enter(xfs_buf_trace_buf
,
65 (void *)(unsigned long)bp
->b_flags
,
66 (void *)(unsigned long)bp
->b_hold
.counter
,
67 (void *)(unsigned long)bp
->b_sema
.count
,
70 (void *)(unsigned long)((bp
->b_file_offset
>>32) & 0xffffffff),
71 (void *)(unsigned long)(bp
->b_file_offset
& 0xffffffff),
72 (void *)(unsigned long)bp
->b_buffer_length
,
73 NULL
, NULL
, NULL
, NULL
, NULL
);
75 ktrace_t
*xfs_buf_trace_buf
;
76 #define XFS_BUF_TRACE_SIZE 4096
77 #define XB_TRACE(bp, id, data) \
78 xfs_buf_trace(bp, id, (void *)data, (void *)__builtin_return_address(0))
80 #define XB_TRACE(bp, id, data) do { } while (0)
83 #ifdef XFS_BUF_LOCK_TRACKING
84 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
85 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
86 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
88 # define XB_SET_OWNER(bp) do { } while (0)
89 # define XB_CLEAR_OWNER(bp) do { } while (0)
90 # define XB_GET_OWNER(bp) do { } while (0)
93 #define xb_to_gfp(flags) \
94 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \
95 ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)
97 #define xb_to_km(flags) \
98 (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)
100 #define xfs_buf_allocate(flags) \
101 kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags))
102 #define xfs_buf_deallocate(bp) \
103 kmem_zone_free(xfs_buf_zone, (bp));
106 * Page Region interfaces.
108 * For pages in filesystems where the blocksize is smaller than the
109 * pagesize, we use the page->private field (long) to hold a bitmap
110 * of uptodate regions within the page.
112 * Each such region is "bytes per page / bits per long" bytes long.
114 * NBPPR == number-of-bytes-per-page-region
115 * BTOPR == bytes-to-page-region (rounded up)
116 * BTOPRT == bytes-to-page-region-truncated (rounded down)
118 #if (BITS_PER_LONG == 32)
119 #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */
120 #elif (BITS_PER_LONG == 64)
121 #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */
123 #error BITS_PER_LONG must be 32 or 64
125 #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG)
126 #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT)
127 #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT))
137 first
= BTOPR(offset
);
138 final
= BTOPRT(offset
+ length
- 1);
139 first
= min(first
, final
);
142 mask
<<= BITS_PER_LONG
- (final
- first
);
143 mask
>>= BITS_PER_LONG
- (final
);
145 ASSERT(offset
+ length
<= PAGE_CACHE_SIZE
);
146 ASSERT((final
- first
) < BITS_PER_LONG
&& (final
- first
) >= 0);
157 set_page_private(page
,
158 page_private(page
) | page_region_mask(offset
, length
));
159 if (page_private(page
) == ~0UL)
160 SetPageUptodate(page
);
169 unsigned long mask
= page_region_mask(offset
, length
);
171 return (mask
&& (page_private(page
) & mask
) == mask
);
175 * Mapping of multi-page buffers into contiguous virtual space
178 typedef struct a_list
{
183 static a_list_t
*as_free_head
;
184 static int as_list_len
;
185 static DEFINE_SPINLOCK(as_lock
);
188 * Try to batch vunmaps because they are costly.
198 * Xen needs to be able to make sure it can get an exclusive
199 * RO mapping of pages it wants to turn into a pagetable. If
200 * a newly allocated page is also still being vmap()ed by xfs,
201 * it will cause pagetable construction to fail. This is a
202 * quick workaround to always eagerly unmap pages so that Xen
209 aentry
= kmalloc(sizeof(a_list_t
), GFP_NOWAIT
);
210 if (likely(aentry
)) {
212 aentry
->next
= as_free_head
;
213 aentry
->vm_addr
= addr
;
214 as_free_head
= aentry
;
216 spin_unlock(&as_lock
);
223 purge_addresses(void)
225 a_list_t
*aentry
, *old
;
227 if (as_free_head
== NULL
)
231 aentry
= as_free_head
;
234 spin_unlock(&as_lock
);
236 while ((old
= aentry
) != NULL
) {
237 vunmap(aentry
->vm_addr
);
238 aentry
= aentry
->next
;
244 * Internal xfs_buf_t object manipulation
250 xfs_buftarg_t
*target
,
251 xfs_off_t range_base
,
253 xfs_buf_flags_t flags
)
256 * We don't want certain flags to appear in b_flags.
258 flags
&= ~(XBF_LOCK
|XBF_MAPPED
|XBF_DONT_BLOCK
|XBF_READ_AHEAD
);
260 memset(bp
, 0, sizeof(xfs_buf_t
));
261 atomic_set(&bp
->b_hold
, 1);
262 init_completion(&bp
->b_iowait
);
263 INIT_LIST_HEAD(&bp
->b_list
);
264 INIT_LIST_HEAD(&bp
->b_hash_list
);
265 init_MUTEX_LOCKED(&bp
->b_sema
); /* held, no waiters */
267 bp
->b_target
= target
;
268 bp
->b_file_offset
= range_base
;
270 * Set buffer_length and count_desired to the same value initially.
271 * I/O routines should use count_desired, which will be the same in
272 * most cases but may be reset (e.g. XFS recovery).
274 bp
->b_buffer_length
= bp
->b_count_desired
= range_length
;
276 bp
->b_bn
= XFS_BUF_DADDR_NULL
;
277 atomic_set(&bp
->b_pin_count
, 0);
278 init_waitqueue_head(&bp
->b_waiters
);
280 XFS_STATS_INC(xb_create
);
281 XB_TRACE(bp
, "initialize", target
);
285 * Allocate a page array capable of holding a specified number
286 * of pages, and point the page buf at it.
292 xfs_buf_flags_t flags
)
294 /* Make sure that we have a page list */
295 if (bp
->b_pages
== NULL
) {
296 bp
->b_offset
= xfs_buf_poff(bp
->b_file_offset
);
297 bp
->b_page_count
= page_count
;
298 if (page_count
<= XB_PAGES
) {
299 bp
->b_pages
= bp
->b_page_array
;
301 bp
->b_pages
= kmem_alloc(sizeof(struct page
*) *
302 page_count
, xb_to_km(flags
));
303 if (bp
->b_pages
== NULL
)
306 memset(bp
->b_pages
, 0, sizeof(struct page
*) * page_count
);
312 * Frees b_pages if it was allocated.
318 if (bp
->b_pages
!= bp
->b_page_array
) {
319 kmem_free(bp
->b_pages
);
324 * Releases the specified buffer.
326 * The modification state of any associated pages is left unchanged.
327 * The buffer most not be on any hash - use xfs_buf_rele instead for
328 * hashed and refcounted buffers
334 XB_TRACE(bp
, "free", 0);
336 ASSERT(list_empty(&bp
->b_hash_list
));
338 if (bp
->b_flags
& (_XBF_PAGE_CACHE
|_XBF_PAGES
)) {
341 if ((bp
->b_flags
& XBF_MAPPED
) && (bp
->b_page_count
> 1))
342 free_address(bp
->b_addr
- bp
->b_offset
);
344 for (i
= 0; i
< bp
->b_page_count
; i
++) {
345 struct page
*page
= bp
->b_pages
[i
];
347 if (bp
->b_flags
& _XBF_PAGE_CACHE
)
348 ASSERT(!PagePrivate(page
));
349 page_cache_release(page
);
351 _xfs_buf_free_pages(bp
);
354 xfs_buf_deallocate(bp
);
358 * Finds all pages for buffer in question and builds it's page list.
361 _xfs_buf_lookup_pages(
365 struct address_space
*mapping
= bp
->b_target
->bt_mapping
;
366 size_t blocksize
= bp
->b_target
->bt_bsize
;
367 size_t size
= bp
->b_count_desired
;
368 size_t nbytes
, offset
;
369 gfp_t gfp_mask
= xb_to_gfp(flags
);
370 unsigned short page_count
, i
;
375 end
= bp
->b_file_offset
+ bp
->b_buffer_length
;
376 page_count
= xfs_buf_btoc(end
) - xfs_buf_btoct(bp
->b_file_offset
);
378 error
= _xfs_buf_get_pages(bp
, page_count
, flags
);
381 bp
->b_flags
|= _XBF_PAGE_CACHE
;
383 offset
= bp
->b_offset
;
384 first
= bp
->b_file_offset
>> PAGE_CACHE_SHIFT
;
386 for (i
= 0; i
< bp
->b_page_count
; i
++) {
391 page
= find_or_create_page(mapping
, first
+ i
, gfp_mask
);
392 if (unlikely(page
== NULL
)) {
393 if (flags
& XBF_READ_AHEAD
) {
394 bp
->b_page_count
= i
;
395 for (i
= 0; i
< bp
->b_page_count
; i
++)
396 unlock_page(bp
->b_pages
[i
]);
401 * This could deadlock.
403 * But until all the XFS lowlevel code is revamped to
404 * handle buffer allocation failures we can't do much.
406 if (!(++retries
% 100))
408 "XFS: possible memory allocation "
409 "deadlock in %s (mode:0x%x)\n",
412 XFS_STATS_INC(xb_page_retries
);
413 xfsbufd_wakeup(0, gfp_mask
);
414 congestion_wait(WRITE
, HZ
/50);
418 XFS_STATS_INC(xb_page_found
);
420 nbytes
= min_t(size_t, size
, PAGE_CACHE_SIZE
- offset
);
423 ASSERT(!PagePrivate(page
));
424 if (!PageUptodate(page
)) {
426 if (blocksize
>= PAGE_CACHE_SIZE
) {
427 if (flags
& XBF_READ
)
428 bp
->b_flags
|= _XBF_PAGE_LOCKED
;
429 } else if (!PagePrivate(page
)) {
430 if (test_page_region(page
, offset
, nbytes
))
435 bp
->b_pages
[i
] = page
;
439 if (!(bp
->b_flags
& _XBF_PAGE_LOCKED
)) {
440 for (i
= 0; i
< bp
->b_page_count
; i
++)
441 unlock_page(bp
->b_pages
[i
]);
444 if (page_count
== bp
->b_page_count
)
445 bp
->b_flags
|= XBF_DONE
;
447 XB_TRACE(bp
, "lookup_pages", (long)page_count
);
452 * Map buffer into kernel address-space if nessecary.
459 /* A single page buffer is always mappable */
460 if (bp
->b_page_count
== 1) {
461 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
462 bp
->b_flags
|= XBF_MAPPED
;
463 } else if (flags
& XBF_MAPPED
) {
464 if (as_list_len
> 64)
466 bp
->b_addr
= vmap(bp
->b_pages
, bp
->b_page_count
,
467 VM_MAP
, PAGE_KERNEL
);
468 if (unlikely(bp
->b_addr
== NULL
))
470 bp
->b_addr
+= bp
->b_offset
;
471 bp
->b_flags
|= XBF_MAPPED
;
478 * Finding and Reading Buffers
482 * Look up, and creates if absent, a lockable buffer for
483 * a given range of an inode. The buffer is returned
484 * locked. If other overlapping buffers exist, they are
485 * released before the new buffer is created and locked,
486 * which may imply that this call will block until those buffers
487 * are unlocked. No I/O is implied by this call.
491 xfs_buftarg_t
*btp
, /* block device target */
492 xfs_off_t ioff
, /* starting offset of range */
493 size_t isize
, /* length of range */
494 xfs_buf_flags_t flags
,
497 xfs_off_t range_base
;
502 range_base
= (ioff
<< BBSHIFT
);
503 range_length
= (isize
<< BBSHIFT
);
505 /* Check for IOs smaller than the sector size / not sector aligned */
506 ASSERT(!(range_length
< (1 << btp
->bt_sshift
)));
507 ASSERT(!(range_base
& (xfs_off_t
)btp
->bt_smask
));
509 hash
= &btp
->bt_hash
[hash_long((unsigned long)ioff
, btp
->bt_hashshift
)];
511 spin_lock(&hash
->bh_lock
);
513 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
514 ASSERT(btp
== bp
->b_target
);
515 if (bp
->b_file_offset
== range_base
&&
516 bp
->b_buffer_length
== range_length
) {
518 * If we look at something, bring it to the
519 * front of the list for next time.
521 atomic_inc(&bp
->b_hold
);
522 list_move(&bp
->b_hash_list
, &hash
->bh_list
);
529 _xfs_buf_initialize(new_bp
, btp
, range_base
,
530 range_length
, flags
);
531 new_bp
->b_hash
= hash
;
532 list_add(&new_bp
->b_hash_list
, &hash
->bh_list
);
534 XFS_STATS_INC(xb_miss_locked
);
537 spin_unlock(&hash
->bh_lock
);
541 spin_unlock(&hash
->bh_lock
);
543 /* Attempt to get the semaphore without sleeping,
544 * if this does not work then we need to drop the
545 * spinlock and do a hard attempt on the semaphore.
547 if (down_trylock(&bp
->b_sema
)) {
548 if (!(flags
& XBF_TRYLOCK
)) {
549 /* wait for buffer ownership */
550 XB_TRACE(bp
, "get_lock", 0);
552 XFS_STATS_INC(xb_get_locked_waited
);
554 /* We asked for a trylock and failed, no need
555 * to look at file offset and length here, we
556 * know that this buffer at least overlaps our
557 * buffer and is locked, therefore our buffer
558 * either does not exist, or is this buffer.
561 XFS_STATS_INC(xb_busy_locked
);
569 if (bp
->b_flags
& XBF_STALE
) {
570 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
571 bp
->b_flags
&= XBF_MAPPED
;
573 XB_TRACE(bp
, "got_lock", 0);
574 XFS_STATS_INC(xb_get_locked
);
579 * Assembles a buffer covering the specified range.
580 * Storage in memory for all portions of the buffer will be allocated,
581 * although backing storage may not be.
585 xfs_buftarg_t
*target
,/* target for buffer */
586 xfs_off_t ioff
, /* starting offset of range */
587 size_t isize
, /* length of range */
588 xfs_buf_flags_t flags
)
590 xfs_buf_t
*bp
, *new_bp
;
593 new_bp
= xfs_buf_allocate(flags
);
594 if (unlikely(!new_bp
))
597 bp
= _xfs_buf_find(target
, ioff
, isize
, flags
, new_bp
);
599 error
= _xfs_buf_lookup_pages(bp
, flags
);
603 xfs_buf_deallocate(new_bp
);
604 if (unlikely(bp
== NULL
))
608 for (i
= 0; i
< bp
->b_page_count
; i
++)
609 mark_page_accessed(bp
->b_pages
[i
]);
611 if (!(bp
->b_flags
& XBF_MAPPED
)) {
612 error
= _xfs_buf_map_pages(bp
, flags
);
613 if (unlikely(error
)) {
614 printk(KERN_WARNING
"%s: failed to map pages\n",
620 XFS_STATS_INC(xb_get
);
623 * Always fill in the block number now, the mapped cases can do
624 * their own overlay of this later.
627 bp
->b_count_desired
= bp
->b_buffer_length
;
629 XB_TRACE(bp
, "get", (unsigned long)flags
);
633 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
642 xfs_buf_flags_t flags
)
646 XB_TRACE(bp
, "_xfs_buf_read", (unsigned long)flags
);
648 ASSERT(!(flags
& (XBF_DELWRI
|XBF_WRITE
)));
649 ASSERT(bp
->b_bn
!= XFS_BUF_DADDR_NULL
);
651 bp
->b_flags
&= ~(XBF_WRITE
| XBF_ASYNC
| XBF_DELWRI
| \
652 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
653 bp
->b_flags
|= flags
& (XBF_READ
| XBF_ASYNC
| \
654 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
656 status
= xfs_buf_iorequest(bp
);
657 if (!status
&& !(flags
& XBF_ASYNC
))
658 status
= xfs_buf_iowait(bp
);
664 xfs_buftarg_t
*target
,
667 xfs_buf_flags_t flags
)
673 bp
= xfs_buf_get_flags(target
, ioff
, isize
, flags
);
675 if (!XFS_BUF_ISDONE(bp
)) {
676 XB_TRACE(bp
, "read", (unsigned long)flags
);
677 XFS_STATS_INC(xb_get_read
);
678 _xfs_buf_read(bp
, flags
);
679 } else if (flags
& XBF_ASYNC
) {
680 XB_TRACE(bp
, "read_async", (unsigned long)flags
);
682 * Read ahead call which is already satisfied,
687 XB_TRACE(bp
, "read_done", (unsigned long)flags
);
688 /* We do not want read in the flags */
689 bp
->b_flags
&= ~XBF_READ
;
696 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
703 * If we are not low on memory then do the readahead in a deadlock
708 xfs_buftarg_t
*target
,
711 xfs_buf_flags_t flags
)
713 struct backing_dev_info
*bdi
;
715 bdi
= target
->bt_mapping
->backing_dev_info
;
716 if (bdi_read_congested(bdi
))
719 flags
|= (XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
);
720 xfs_buf_read_flags(target
, ioff
, isize
, flags
);
726 xfs_buftarg_t
*target
)
730 bp
= xfs_buf_allocate(0);
732 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
736 static inline struct page
*
740 if ((!is_vmalloc_addr(addr
))) {
741 return virt_to_page(addr
);
743 return vmalloc_to_page(addr
);
748 xfs_buf_associate_memory(
755 unsigned long pageaddr
;
756 unsigned long offset
;
760 pageaddr
= (unsigned long)mem
& PAGE_CACHE_MASK
;
761 offset
= (unsigned long)mem
- pageaddr
;
762 buflen
= PAGE_CACHE_ALIGN(len
+ offset
);
763 page_count
= buflen
>> PAGE_CACHE_SHIFT
;
765 /* Free any previous set of page pointers */
767 _xfs_buf_free_pages(bp
);
772 rval
= _xfs_buf_get_pages(bp
, page_count
, 0);
776 bp
->b_offset
= offset
;
778 for (i
= 0; i
< bp
->b_page_count
; i
++) {
779 bp
->b_pages
[i
] = mem_to_page((void *)pageaddr
);
780 pageaddr
+= PAGE_CACHE_SIZE
;
783 bp
->b_count_desired
= len
;
784 bp
->b_buffer_length
= buflen
;
785 bp
->b_flags
|= XBF_MAPPED
;
786 bp
->b_flags
&= ~_XBF_PAGE_LOCKED
;
794 xfs_buftarg_t
*target
)
796 unsigned long page_count
= PAGE_ALIGN(len
) >> PAGE_SHIFT
;
800 bp
= xfs_buf_allocate(0);
801 if (unlikely(bp
== NULL
))
803 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
805 error
= _xfs_buf_get_pages(bp
, page_count
, 0);
809 for (i
= 0; i
< page_count
; i
++) {
810 bp
->b_pages
[i
] = alloc_page(GFP_KERNEL
);
814 bp
->b_flags
|= _XBF_PAGES
;
816 error
= _xfs_buf_map_pages(bp
, XBF_MAPPED
);
817 if (unlikely(error
)) {
818 printk(KERN_WARNING
"%s: failed to map pages\n",
825 XB_TRACE(bp
, "no_daddr", len
);
830 __free_page(bp
->b_pages
[i
]);
831 _xfs_buf_free_pages(bp
);
833 xfs_buf_deallocate(bp
);
839 * Increment reference count on buffer, to hold the buffer concurrently
840 * with another thread which may release (free) the buffer asynchronously.
841 * Must hold the buffer already to call this function.
847 atomic_inc(&bp
->b_hold
);
848 XB_TRACE(bp
, "hold", 0);
852 * Releases a hold on the specified buffer. If the
853 * the hold count is 1, calls xfs_buf_free.
859 xfs_bufhash_t
*hash
= bp
->b_hash
;
861 XB_TRACE(bp
, "rele", bp
->b_relse
);
863 if (unlikely(!hash
)) {
864 ASSERT(!bp
->b_relse
);
865 if (atomic_dec_and_test(&bp
->b_hold
))
870 ASSERT(atomic_read(&bp
->b_hold
) > 0);
871 if (atomic_dec_and_lock(&bp
->b_hold
, &hash
->bh_lock
)) {
873 atomic_inc(&bp
->b_hold
);
874 spin_unlock(&hash
->bh_lock
);
875 (*(bp
->b_relse
)) (bp
);
876 } else if (bp
->b_flags
& XBF_FS_MANAGED
) {
877 spin_unlock(&hash
->bh_lock
);
879 ASSERT(!(bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)));
880 list_del_init(&bp
->b_hash_list
);
881 spin_unlock(&hash
->bh_lock
);
889 * Mutual exclusion on buffers. Locking model:
891 * Buffers associated with inodes for which buffer locking
892 * is not enabled are not protected by semaphores, and are
893 * assumed to be exclusively owned by the caller. There is a
894 * spinlock in the buffer, used by the caller when concurrent
895 * access is possible.
899 * Locks a buffer object, if it is not already locked.
900 * Note that this in no way locks the underlying pages, so it is only
901 * useful for synchronizing concurrent use of buffer objects, not for
902 * synchronizing independent access to the underlying pages.
910 locked
= down_trylock(&bp
->b_sema
) == 0;
914 XB_TRACE(bp
, "cond_lock", (long)locked
);
915 return locked
? 0 : -EBUSY
;
918 #if defined(DEBUG) || defined(XFS_BLI_TRACE)
923 return bp
->b_sema
.count
;
928 * Locks a buffer object.
929 * Note that this in no way locks the underlying pages, so it is only
930 * useful for synchronizing concurrent use of buffer objects, not for
931 * synchronizing independent access to the underlying pages.
937 XB_TRACE(bp
, "lock", 0);
938 if (atomic_read(&bp
->b_io_remaining
))
939 blk_run_address_space(bp
->b_target
->bt_mapping
);
942 XB_TRACE(bp
, "locked", 0);
946 * Releases the lock on the buffer object.
947 * If the buffer is marked delwri but is not queued, do so before we
948 * unlock the buffer as we need to set flags correctly. We also need to
949 * take a reference for the delwri queue because the unlocker is going to
950 * drop their's and they don't know we just queued it.
956 if ((bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)) == XBF_DELWRI
) {
957 atomic_inc(&bp
->b_hold
);
958 bp
->b_flags
|= XBF_ASYNC
;
959 xfs_buf_delwri_queue(bp
, 0);
964 XB_TRACE(bp
, "unlock", 0);
969 * Pinning Buffer Storage in Memory
970 * Ensure that no attempt to force a buffer to disk will succeed.
976 atomic_inc(&bp
->b_pin_count
);
977 XB_TRACE(bp
, "pin", (long)bp
->b_pin_count
.counter
);
984 if (atomic_dec_and_test(&bp
->b_pin_count
))
985 wake_up_all(&bp
->b_waiters
);
986 XB_TRACE(bp
, "unpin", (long)bp
->b_pin_count
.counter
);
993 return atomic_read(&bp
->b_pin_count
);
1000 DECLARE_WAITQUEUE (wait
, current
);
1002 if (atomic_read(&bp
->b_pin_count
) == 0)
1005 add_wait_queue(&bp
->b_waiters
, &wait
);
1007 set_current_state(TASK_UNINTERRUPTIBLE
);
1008 if (atomic_read(&bp
->b_pin_count
) == 0)
1010 if (atomic_read(&bp
->b_io_remaining
))
1011 blk_run_address_space(bp
->b_target
->bt_mapping
);
1014 remove_wait_queue(&bp
->b_waiters
, &wait
);
1015 set_current_state(TASK_RUNNING
);
1019 * Buffer Utility Routines
1023 xfs_buf_iodone_work(
1024 struct work_struct
*work
)
1027 container_of(work
, xfs_buf_t
, b_iodone_work
);
1030 * We can get an EOPNOTSUPP to ordered writes. Here we clear the
1031 * ordered flag and reissue them. Because we can't tell the higher
1032 * layers directly that they should not issue ordered I/O anymore, they
1033 * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion.
1035 if ((bp
->b_error
== EOPNOTSUPP
) &&
1036 (bp
->b_flags
& (XBF_ORDERED
|XBF_ASYNC
)) == (XBF_ORDERED
|XBF_ASYNC
)) {
1037 XB_TRACE(bp
, "ordered_retry", bp
->b_iodone
);
1038 bp
->b_flags
&= ~XBF_ORDERED
;
1039 bp
->b_flags
|= _XFS_BARRIER_FAILED
;
1040 xfs_buf_iorequest(bp
);
1041 } else if (bp
->b_iodone
)
1042 (*(bp
->b_iodone
))(bp
);
1043 else if (bp
->b_flags
& XBF_ASYNC
)
1052 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_READ_AHEAD
);
1053 if (bp
->b_error
== 0)
1054 bp
->b_flags
|= XBF_DONE
;
1056 XB_TRACE(bp
, "iodone", bp
->b_iodone
);
1058 if ((bp
->b_iodone
) || (bp
->b_flags
& XBF_ASYNC
)) {
1060 INIT_WORK(&bp
->b_iodone_work
, xfs_buf_iodone_work
);
1061 queue_work(xfslogd_workqueue
, &bp
->b_iodone_work
);
1063 xfs_buf_iodone_work(&bp
->b_iodone_work
);
1066 complete(&bp
->b_iowait
);
1075 ASSERT(error
>= 0 && error
<= 0xffff);
1076 bp
->b_error
= (unsigned short)error
;
1077 XB_TRACE(bp
, "ioerror", (unsigned long)error
);
1085 XB_TRACE(bp
, "bawrite", 0);
1087 ASSERT(bp
->b_bn
!= XFS_BUF_DADDR_NULL
);
1089 xfs_buf_delwri_dequeue(bp
);
1091 bp
->b_flags
&= ~(XBF_READ
| XBF_DELWRI
| XBF_READ_AHEAD
);
1092 bp
->b_flags
|= (XBF_WRITE
| XBF_ASYNC
| _XBF_RUN_QUEUES
);
1095 bp
->b_strat
= xfs_bdstrat_cb
;
1096 return xfs_bdstrat_cb(bp
);
1104 XB_TRACE(bp
, "bdwrite", 0);
1106 bp
->b_strat
= xfs_bdstrat_cb
;
1109 bp
->b_flags
&= ~XBF_READ
;
1110 bp
->b_flags
|= (XBF_DELWRI
| XBF_ASYNC
);
1112 xfs_buf_delwri_queue(bp
, 1);
1120 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1121 bp
->b_flags
&= ~_XBF_PAGE_LOCKED
;
1122 xfs_buf_ioend(bp
, schedule
);
1131 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1132 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1133 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1135 xfs_buf_ioerror(bp
, -error
);
1138 struct page
*page
= bvec
->bv_page
;
1140 ASSERT(!PagePrivate(page
));
1141 if (unlikely(bp
->b_error
)) {
1142 if (bp
->b_flags
& XBF_READ
)
1143 ClearPageUptodate(page
);
1144 } else if (blocksize
>= PAGE_CACHE_SIZE
) {
1145 SetPageUptodate(page
);
1146 } else if (!PagePrivate(page
) &&
1147 (bp
->b_flags
& _XBF_PAGE_CACHE
)) {
1148 set_page_region(page
, bvec
->bv_offset
, bvec
->bv_len
);
1151 if (--bvec
>= bio
->bi_io_vec
)
1152 prefetchw(&bvec
->bv_page
->flags
);
1154 if (bp
->b_flags
& _XBF_PAGE_LOCKED
)
1156 } while (bvec
>= bio
->bi_io_vec
);
1158 _xfs_buf_ioend(bp
, 1);
1166 int rw
, map_i
, total_nr_pages
, nr_pages
;
1168 int offset
= bp
->b_offset
;
1169 int size
= bp
->b_count_desired
;
1170 sector_t sector
= bp
->b_bn
;
1171 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1173 total_nr_pages
= bp
->b_page_count
;
1176 if (bp
->b_flags
& XBF_ORDERED
) {
1177 ASSERT(!(bp
->b_flags
& XBF_READ
));
1179 } else if (bp
->b_flags
& _XBF_RUN_QUEUES
) {
1180 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1181 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1182 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_SYNC
: READ_SYNC
;
1184 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE
:
1185 (bp
->b_flags
& XBF_READ_AHEAD
) ? READA
: READ
;
1188 /* Special code path for reading a sub page size buffer in --
1189 * we populate up the whole page, and hence the other metadata
1190 * in the same page. This optimization is only valid when the
1191 * filesystem block size is not smaller than the page size.
1193 if ((bp
->b_buffer_length
< PAGE_CACHE_SIZE
) &&
1194 ((bp
->b_flags
& (XBF_READ
|_XBF_PAGE_LOCKED
)) ==
1195 (XBF_READ
|_XBF_PAGE_LOCKED
)) &&
1196 (blocksize
>= PAGE_CACHE_SIZE
)) {
1197 bio
= bio_alloc(GFP_NOIO
, 1);
1199 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1200 bio
->bi_sector
= sector
- (offset
>> BBSHIFT
);
1201 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1202 bio
->bi_private
= bp
;
1204 bio_add_page(bio
, bp
->b_pages
[0], PAGE_CACHE_SIZE
, 0);
1207 atomic_inc(&bp
->b_io_remaining
);
1213 atomic_inc(&bp
->b_io_remaining
);
1214 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1215 if (nr_pages
> total_nr_pages
)
1216 nr_pages
= total_nr_pages
;
1218 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1219 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1220 bio
->bi_sector
= sector
;
1221 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1222 bio
->bi_private
= bp
;
1224 for (; size
&& nr_pages
; nr_pages
--, map_i
++) {
1225 int rbytes
, nbytes
= PAGE_CACHE_SIZE
- offset
;
1230 rbytes
= bio_add_page(bio
, bp
->b_pages
[map_i
], nbytes
, offset
);
1231 if (rbytes
< nbytes
)
1235 sector
+= nbytes
>> BBSHIFT
;
1241 if (likely(bio
->bi_size
)) {
1242 submit_bio(rw
, bio
);
1247 xfs_buf_ioerror(bp
, EIO
);
1255 XB_TRACE(bp
, "iorequest", 0);
1257 if (bp
->b_flags
& XBF_DELWRI
) {
1258 xfs_buf_delwri_queue(bp
, 1);
1262 if (bp
->b_flags
& XBF_WRITE
) {
1263 xfs_buf_wait_unpin(bp
);
1268 /* Set the count to 1 initially, this will stop an I/O
1269 * completion callout which happens before we have started
1270 * all the I/O from calling xfs_buf_ioend too early.
1272 atomic_set(&bp
->b_io_remaining
, 1);
1273 _xfs_buf_ioapply(bp
);
1274 _xfs_buf_ioend(bp
, 0);
1281 * Waits for I/O to complete on the buffer supplied.
1282 * It returns immediately if no I/O is pending.
1283 * It returns the I/O error code, if any, or 0 if there was no error.
1289 XB_TRACE(bp
, "iowait", 0);
1290 if (atomic_read(&bp
->b_io_remaining
))
1291 blk_run_address_space(bp
->b_target
->bt_mapping
);
1292 wait_for_completion(&bp
->b_iowait
);
1293 XB_TRACE(bp
, "iowaited", (long)bp
->b_error
);
1304 if (bp
->b_flags
& XBF_MAPPED
)
1305 return XFS_BUF_PTR(bp
) + offset
;
1307 offset
+= bp
->b_offset
;
1308 page
= bp
->b_pages
[offset
>> PAGE_CACHE_SHIFT
];
1309 return (xfs_caddr_t
)page_address(page
) + (offset
& (PAGE_CACHE_SIZE
-1));
1313 * Move data into or out of a buffer.
1317 xfs_buf_t
*bp
, /* buffer to process */
1318 size_t boff
, /* starting buffer offset */
1319 size_t bsize
, /* length to copy */
1320 caddr_t data
, /* data address */
1321 xfs_buf_rw_t mode
) /* read/write/zero flag */
1323 size_t bend
, cpoff
, csize
;
1326 bend
= boff
+ bsize
;
1327 while (boff
< bend
) {
1328 page
= bp
->b_pages
[xfs_buf_btoct(boff
+ bp
->b_offset
)];
1329 cpoff
= xfs_buf_poff(boff
+ bp
->b_offset
);
1330 csize
= min_t(size_t,
1331 PAGE_CACHE_SIZE
-cpoff
, bp
->b_count_desired
-boff
);
1333 ASSERT(((csize
+ cpoff
) <= PAGE_CACHE_SIZE
));
1337 memset(page_address(page
) + cpoff
, 0, csize
);
1340 memcpy(data
, page_address(page
) + cpoff
, csize
);
1343 memcpy(page_address(page
) + cpoff
, data
, csize
);
1352 * Handling of buffer targets (buftargs).
1356 * Wait for any bufs with callbacks that have been submitted but
1357 * have not yet returned... walk the hash list for the target.
1364 xfs_bufhash_t
*hash
;
1367 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1368 hash
= &btp
->bt_hash
[i
];
1370 spin_lock(&hash
->bh_lock
);
1371 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
1372 ASSERT(btp
== bp
->b_target
);
1373 if (!(bp
->b_flags
& XBF_FS_MANAGED
)) {
1374 spin_unlock(&hash
->bh_lock
);
1376 * Catch superblock reference count leaks
1379 BUG_ON(bp
->b_bn
== 0);
1384 spin_unlock(&hash
->bh_lock
);
1389 * Allocate buffer hash table for a given target.
1390 * For devices containing metadata (i.e. not the log/realtime devices)
1391 * we need to allocate a much larger hash table.
1400 btp
->bt_hashshift
= external
? 3 : 8; /* 8 or 256 buckets */
1401 btp
->bt_hashmask
= (1 << btp
->bt_hashshift
) - 1;
1402 btp
->bt_hash
= kmem_zalloc((1 << btp
->bt_hashshift
) *
1403 sizeof(xfs_bufhash_t
), KM_SLEEP
| KM_LARGE
);
1404 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1405 spin_lock_init(&btp
->bt_hash
[i
].bh_lock
);
1406 INIT_LIST_HEAD(&btp
->bt_hash
[i
].bh_list
);
1414 kmem_free(btp
->bt_hash
);
1415 btp
->bt_hash
= NULL
;
1419 * buftarg list for delwrite queue processing
1421 static LIST_HEAD(xfs_buftarg_list
);
1422 static DEFINE_SPINLOCK(xfs_buftarg_lock
);
1425 xfs_register_buftarg(
1428 spin_lock(&xfs_buftarg_lock
);
1429 list_add(&btp
->bt_list
, &xfs_buftarg_list
);
1430 spin_unlock(&xfs_buftarg_lock
);
1434 xfs_unregister_buftarg(
1437 spin_lock(&xfs_buftarg_lock
);
1438 list_del(&btp
->bt_list
);
1439 spin_unlock(&xfs_buftarg_lock
);
1444 struct xfs_mount
*mp
,
1445 struct xfs_buftarg
*btp
)
1447 xfs_flush_buftarg(btp
, 1);
1448 if (mp
->m_flags
& XFS_MOUNT_BARRIER
)
1449 xfs_blkdev_issue_flush(btp
);
1450 xfs_free_bufhash(btp
);
1451 iput(btp
->bt_mapping
->host
);
1453 /* Unregister the buftarg first so that we don't get a
1454 * wakeup finding a non-existent task
1456 xfs_unregister_buftarg(btp
);
1457 kthread_stop(btp
->bt_task
);
1463 xfs_setsize_buftarg_flags(
1465 unsigned int blocksize
,
1466 unsigned int sectorsize
,
1469 btp
->bt_bsize
= blocksize
;
1470 btp
->bt_sshift
= ffs(sectorsize
) - 1;
1471 btp
->bt_smask
= sectorsize
- 1;
1473 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1475 "XFS: Cannot set_blocksize to %u on device %s\n",
1476 sectorsize
, XFS_BUFTARG_NAME(btp
));
1481 (PAGE_CACHE_SIZE
/ BITS_PER_LONG
) > sectorsize
) {
1483 "XFS: %u byte sectors in use on device %s. "
1484 "This is suboptimal; %u or greater is ideal.\n",
1485 sectorsize
, XFS_BUFTARG_NAME(btp
),
1486 (unsigned int)PAGE_CACHE_SIZE
/ BITS_PER_LONG
);
1493 * When allocating the initial buffer target we have not yet
1494 * read in the superblock, so don't know what sized sectors
1495 * are being used is at this early stage. Play safe.
1498 xfs_setsize_buftarg_early(
1500 struct block_device
*bdev
)
1502 return xfs_setsize_buftarg_flags(btp
,
1503 PAGE_CACHE_SIZE
, bdev_hardsect_size(bdev
), 0);
1507 xfs_setsize_buftarg(
1509 unsigned int blocksize
,
1510 unsigned int sectorsize
)
1512 return xfs_setsize_buftarg_flags(btp
, blocksize
, sectorsize
, 1);
1516 xfs_mapping_buftarg(
1518 struct block_device
*bdev
)
1520 struct backing_dev_info
*bdi
;
1521 struct inode
*inode
;
1522 struct address_space
*mapping
;
1523 static const struct address_space_operations mapping_aops
= {
1524 .sync_page
= block_sync_page
,
1525 .migratepage
= fail_migrate_page
,
1528 inode
= new_inode(bdev
->bd_inode
->i_sb
);
1531 "XFS: Cannot allocate mapping inode for device %s\n",
1532 XFS_BUFTARG_NAME(btp
));
1535 inode
->i_mode
= S_IFBLK
;
1536 inode
->i_bdev
= bdev
;
1537 inode
->i_rdev
= bdev
->bd_dev
;
1538 bdi
= blk_get_backing_dev_info(bdev
);
1540 bdi
= &default_backing_dev_info
;
1541 mapping
= &inode
->i_data
;
1542 mapping
->a_ops
= &mapping_aops
;
1543 mapping
->backing_dev_info
= bdi
;
1544 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
1545 btp
->bt_mapping
= mapping
;
1550 xfs_alloc_delwrite_queue(
1555 INIT_LIST_HEAD(&btp
->bt_list
);
1556 INIT_LIST_HEAD(&btp
->bt_delwrite_queue
);
1557 spin_lock_init(&btp
->bt_delwrite_lock
);
1559 btp
->bt_task
= kthread_run(xfsbufd
, btp
, "xfsbufd");
1560 if (IS_ERR(btp
->bt_task
)) {
1561 error
= PTR_ERR(btp
->bt_task
);
1564 xfs_register_buftarg(btp
);
1571 struct block_device
*bdev
,
1576 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
);
1578 btp
->bt_dev
= bdev
->bd_dev
;
1579 btp
->bt_bdev
= bdev
;
1580 if (xfs_setsize_buftarg_early(btp
, bdev
))
1582 if (xfs_mapping_buftarg(btp
, bdev
))
1584 if (xfs_alloc_delwrite_queue(btp
))
1586 xfs_alloc_bufhash(btp
, external
);
1596 * Delayed write buffer handling
1599 xfs_buf_delwri_queue(
1603 struct list_head
*dwq
= &bp
->b_target
->bt_delwrite_queue
;
1604 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1606 XB_TRACE(bp
, "delwri_q", (long)unlock
);
1607 ASSERT((bp
->b_flags
&(XBF_DELWRI
|XBF_ASYNC
)) == (XBF_DELWRI
|XBF_ASYNC
));
1610 /* If already in the queue, dequeue and place at tail */
1611 if (!list_empty(&bp
->b_list
)) {
1612 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1614 atomic_dec(&bp
->b_hold
);
1615 list_del(&bp
->b_list
);
1618 bp
->b_flags
|= _XBF_DELWRI_Q
;
1619 list_add_tail(&bp
->b_list
, dwq
);
1620 bp
->b_queuetime
= jiffies
;
1628 xfs_buf_delwri_dequeue(
1631 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1635 if ((bp
->b_flags
& XBF_DELWRI
) && !list_empty(&bp
->b_list
)) {
1636 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1637 list_del_init(&bp
->b_list
);
1640 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
);
1646 XB_TRACE(bp
, "delwri_dq", (long)dequeued
);
1650 xfs_buf_runall_queues(
1651 struct workqueue_struct
*queue
)
1653 flush_workqueue(queue
);
1663 spin_lock(&xfs_buftarg_lock
);
1664 list_for_each_entry(btp
, &xfs_buftarg_list
, bt_list
) {
1665 if (test_bit(XBT_FORCE_SLEEP
, &btp
->bt_flags
))
1667 set_bit(XBT_FORCE_FLUSH
, &btp
->bt_flags
);
1668 wake_up_process(btp
->bt_task
);
1670 spin_unlock(&xfs_buftarg_lock
);
1675 * Move as many buffers as specified to the supplied list
1676 * idicating if we skipped any buffers to prevent deadlocks.
1679 xfs_buf_delwri_split(
1680 xfs_buftarg_t
*target
,
1681 struct list_head
*list
,
1685 struct list_head
*dwq
= &target
->bt_delwrite_queue
;
1686 spinlock_t
*dwlk
= &target
->bt_delwrite_lock
;
1690 force
= test_and_clear_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1691 INIT_LIST_HEAD(list
);
1693 list_for_each_entry_safe(bp
, n
, dwq
, b_list
) {
1694 XB_TRACE(bp
, "walkq1", (long)xfs_buf_ispin(bp
));
1695 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1697 if (!xfs_buf_ispin(bp
) && !xfs_buf_cond_lock(bp
)) {
1699 time_before(jiffies
, bp
->b_queuetime
+ age
)) {
1704 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
|
1706 bp
->b_flags
|= XBF_WRITE
;
1707 list_move_tail(&bp
->b_list
, list
);
1721 struct list_head tmp
;
1722 xfs_buftarg_t
*target
= (xfs_buftarg_t
*)data
;
1726 current
->flags
|= PF_MEMALLOC
;
1731 if (unlikely(freezing(current
))) {
1732 set_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1735 clear_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1738 schedule_timeout_interruptible(
1739 xfs_buf_timer_centisecs
* msecs_to_jiffies(10));
1741 xfs_buf_delwri_split(target
, &tmp
,
1742 xfs_buf_age_centisecs
* msecs_to_jiffies(10));
1745 while (!list_empty(&tmp
)) {
1746 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1747 ASSERT(target
== bp
->b_target
);
1749 list_del_init(&bp
->b_list
);
1750 xfs_buf_iostrategy(bp
);
1754 if (as_list_len
> 0)
1757 blk_run_address_space(target
->bt_mapping
);
1759 } while (!kthread_should_stop());
1765 * Go through all incore buffers, and release buffers if they belong to
1766 * the given device. This is used in filesystem error handling to
1767 * preserve the consistency of its metadata.
1771 xfs_buftarg_t
*target
,
1774 struct list_head tmp
;
1778 xfs_buf_runall_queues(xfsdatad_workqueue
);
1779 xfs_buf_runall_queues(xfslogd_workqueue
);
1781 set_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1782 pincount
= xfs_buf_delwri_split(target
, &tmp
, 0);
1785 * Dropped the delayed write list lock, now walk the temporary list
1787 list_for_each_entry_safe(bp
, n
, &tmp
, b_list
) {
1788 ASSERT(target
== bp
->b_target
);
1790 bp
->b_flags
&= ~XBF_ASYNC
;
1792 list_del_init(&bp
->b_list
);
1794 xfs_buf_iostrategy(bp
);
1798 blk_run_address_space(target
->bt_mapping
);
1801 * Remaining list items must be flushed before returning
1803 while (!list_empty(&tmp
)) {
1804 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1806 list_del_init(&bp
->b_list
);
1817 #ifdef XFS_BUF_TRACE
1818 xfs_buf_trace_buf
= ktrace_alloc(XFS_BUF_TRACE_SIZE
, KM_NOFS
);
1821 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1822 KM_ZONE_HWALIGN
, NULL
);
1824 goto out_free_trace_buf
;
1826 xfslogd_workqueue
= create_workqueue("xfslogd");
1827 if (!xfslogd_workqueue
)
1828 goto out_free_buf_zone
;
1830 xfsdatad_workqueue
= create_workqueue("xfsdatad");
1831 if (!xfsdatad_workqueue
)
1832 goto out_destroy_xfslogd_workqueue
;
1834 register_shrinker(&xfs_buf_shake
);
1837 out_destroy_xfslogd_workqueue
:
1838 destroy_workqueue(xfslogd_workqueue
);
1840 kmem_zone_destroy(xfs_buf_zone
);
1842 #ifdef XFS_BUF_TRACE
1843 ktrace_free(xfs_buf_trace_buf
);
1849 xfs_buf_terminate(void)
1851 unregister_shrinker(&xfs_buf_shake
);
1852 destroy_workqueue(xfsdatad_workqueue
);
1853 destroy_workqueue(xfslogd_workqueue
);
1854 kmem_zone_destroy(xfs_buf_zone
);
1855 #ifdef XFS_BUF_TRACE
1856 ktrace_free(xfs_buf_trace_buf
);
1860 #ifdef CONFIG_KDB_MODULES
1862 xfs_get_buftarg_list(void)
1864 return &xfs_buftarg_list
;