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>
37 static kmem_zone_t
*xfs_buf_zone
;
38 STATIC
int xfsbufd(void *);
39 STATIC
int xfsbufd_wakeup(int, gfp_t
);
40 STATIC
void xfs_buf_delwri_queue(xfs_buf_t
*, int);
41 static struct shrinker xfs_buf_shake
= {
42 .shrink
= xfsbufd_wakeup
,
43 .seeks
= DEFAULT_SEEKS
,
46 static struct workqueue_struct
*xfslogd_workqueue
;
47 struct workqueue_struct
*xfsdatad_workqueue
;
57 ktrace_enter(xfs_buf_trace_buf
,
59 (void *)(unsigned long)bp
->b_flags
,
60 (void *)(unsigned long)bp
->b_hold
.counter
,
61 (void *)(unsigned long)bp
->b_sema
.count
,
64 (void *)(unsigned long)((bp
->b_file_offset
>>32) & 0xffffffff),
65 (void *)(unsigned long)(bp
->b_file_offset
& 0xffffffff),
66 (void *)(unsigned long)bp
->b_buffer_length
,
67 NULL
, NULL
, NULL
, NULL
, NULL
);
69 ktrace_t
*xfs_buf_trace_buf
;
70 #define XFS_BUF_TRACE_SIZE 4096
71 #define XB_TRACE(bp, id, data) \
72 xfs_buf_trace(bp, id, (void *)data, (void *)__builtin_return_address(0))
74 #define XB_TRACE(bp, id, data) do { } while (0)
77 #ifdef XFS_BUF_LOCK_TRACKING
78 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
79 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
80 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
82 # define XB_SET_OWNER(bp) do { } while (0)
83 # define XB_CLEAR_OWNER(bp) do { } while (0)
84 # define XB_GET_OWNER(bp) do { } while (0)
87 #define xb_to_gfp(flags) \
88 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \
89 ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)
91 #define xb_to_km(flags) \
92 (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)
94 #define xfs_buf_allocate(flags) \
95 kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags))
96 #define xfs_buf_deallocate(bp) \
97 kmem_zone_free(xfs_buf_zone, (bp));
100 * Page Region interfaces.
102 * For pages in filesystems where the blocksize is smaller than the
103 * pagesize, we use the page->private field (long) to hold a bitmap
104 * of uptodate regions within the page.
106 * Each such region is "bytes per page / bits per long" bytes long.
108 * NBPPR == number-of-bytes-per-page-region
109 * BTOPR == bytes-to-page-region (rounded up)
110 * BTOPRT == bytes-to-page-region-truncated (rounded down)
112 #if (BITS_PER_LONG == 32)
113 #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */
114 #elif (BITS_PER_LONG == 64)
115 #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */
117 #error BITS_PER_LONG must be 32 or 64
119 #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG)
120 #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT)
121 #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT))
131 first
= BTOPR(offset
);
132 final
= BTOPRT(offset
+ length
- 1);
133 first
= min(first
, final
);
136 mask
<<= BITS_PER_LONG
- (final
- first
);
137 mask
>>= BITS_PER_LONG
- (final
);
139 ASSERT(offset
+ length
<= PAGE_CACHE_SIZE
);
140 ASSERT((final
- first
) < BITS_PER_LONG
&& (final
- first
) >= 0);
151 set_page_private(page
,
152 page_private(page
) | page_region_mask(offset
, length
));
153 if (page_private(page
) == ~0UL)
154 SetPageUptodate(page
);
163 unsigned long mask
= page_region_mask(offset
, length
);
165 return (mask
&& (page_private(page
) & mask
) == mask
);
169 * Mapping of multi-page buffers into contiguous virtual space
172 typedef struct a_list
{
177 static a_list_t
*as_free_head
;
178 static int as_list_len
;
179 static DEFINE_SPINLOCK(as_lock
);
182 * Try to batch vunmaps because they are costly.
192 * Xen needs to be able to make sure it can get an exclusive
193 * RO mapping of pages it wants to turn into a pagetable. If
194 * a newly allocated page is also still being vmap()ed by xfs,
195 * it will cause pagetable construction to fail. This is a
196 * quick workaround to always eagerly unmap pages so that Xen
203 aentry
= kmalloc(sizeof(a_list_t
), GFP_NOWAIT
);
204 if (likely(aentry
)) {
206 aentry
->next
= as_free_head
;
207 aentry
->vm_addr
= addr
;
208 as_free_head
= aentry
;
210 spin_unlock(&as_lock
);
217 purge_addresses(void)
219 a_list_t
*aentry
, *old
;
221 if (as_free_head
== NULL
)
225 aentry
= as_free_head
;
228 spin_unlock(&as_lock
);
230 while ((old
= aentry
) != NULL
) {
231 vunmap(aentry
->vm_addr
);
232 aentry
= aentry
->next
;
238 * Internal xfs_buf_t object manipulation
244 xfs_buftarg_t
*target
,
245 xfs_off_t range_base
,
247 xfs_buf_flags_t flags
)
250 * We don't want certain flags to appear in b_flags.
252 flags
&= ~(XBF_LOCK
|XBF_MAPPED
|XBF_DONT_BLOCK
|XBF_READ_AHEAD
);
254 memset(bp
, 0, sizeof(xfs_buf_t
));
255 atomic_set(&bp
->b_hold
, 1);
256 init_completion(&bp
->b_iowait
);
257 INIT_LIST_HEAD(&bp
->b_list
);
258 INIT_LIST_HEAD(&bp
->b_hash_list
);
259 init_MUTEX_LOCKED(&bp
->b_sema
); /* held, no waiters */
261 bp
->b_target
= target
;
262 bp
->b_file_offset
= range_base
;
264 * Set buffer_length and count_desired to the same value initially.
265 * I/O routines should use count_desired, which will be the same in
266 * most cases but may be reset (e.g. XFS recovery).
268 bp
->b_buffer_length
= bp
->b_count_desired
= range_length
;
270 bp
->b_bn
= XFS_BUF_DADDR_NULL
;
271 atomic_set(&bp
->b_pin_count
, 0);
272 init_waitqueue_head(&bp
->b_waiters
);
274 XFS_STATS_INC(xb_create
);
275 XB_TRACE(bp
, "initialize", target
);
279 * Allocate a page array capable of holding a specified number
280 * of pages, and point the page buf at it.
286 xfs_buf_flags_t flags
)
288 /* Make sure that we have a page list */
289 if (bp
->b_pages
== NULL
) {
290 bp
->b_offset
= xfs_buf_poff(bp
->b_file_offset
);
291 bp
->b_page_count
= page_count
;
292 if (page_count
<= XB_PAGES
) {
293 bp
->b_pages
= bp
->b_page_array
;
295 bp
->b_pages
= kmem_alloc(sizeof(struct page
*) *
296 page_count
, xb_to_km(flags
));
297 if (bp
->b_pages
== NULL
)
300 memset(bp
->b_pages
, 0, sizeof(struct page
*) * page_count
);
306 * Frees b_pages if it was allocated.
312 if (bp
->b_pages
!= bp
->b_page_array
) {
313 kmem_free(bp
->b_pages
);
318 * Releases the specified buffer.
320 * The modification state of any associated pages is left unchanged.
321 * The buffer most not be on any hash - use xfs_buf_rele instead for
322 * hashed and refcounted buffers
328 XB_TRACE(bp
, "free", 0);
330 ASSERT(list_empty(&bp
->b_hash_list
));
332 if (bp
->b_flags
& (_XBF_PAGE_CACHE
|_XBF_PAGES
)) {
335 if ((bp
->b_flags
& XBF_MAPPED
) && (bp
->b_page_count
> 1))
336 free_address(bp
->b_addr
- bp
->b_offset
);
338 for (i
= 0; i
< bp
->b_page_count
; i
++) {
339 struct page
*page
= bp
->b_pages
[i
];
341 if (bp
->b_flags
& _XBF_PAGE_CACHE
)
342 ASSERT(!PagePrivate(page
));
343 page_cache_release(page
);
345 _xfs_buf_free_pages(bp
);
348 xfs_buf_deallocate(bp
);
352 * Finds all pages for buffer in question and builds it's page list.
355 _xfs_buf_lookup_pages(
359 struct address_space
*mapping
= bp
->b_target
->bt_mapping
;
360 size_t blocksize
= bp
->b_target
->bt_bsize
;
361 size_t size
= bp
->b_count_desired
;
362 size_t nbytes
, offset
;
363 gfp_t gfp_mask
= xb_to_gfp(flags
);
364 unsigned short page_count
, i
;
369 end
= bp
->b_file_offset
+ bp
->b_buffer_length
;
370 page_count
= xfs_buf_btoc(end
) - xfs_buf_btoct(bp
->b_file_offset
);
372 error
= _xfs_buf_get_pages(bp
, page_count
, flags
);
375 bp
->b_flags
|= _XBF_PAGE_CACHE
;
377 offset
= bp
->b_offset
;
378 first
= bp
->b_file_offset
>> PAGE_CACHE_SHIFT
;
380 for (i
= 0; i
< bp
->b_page_count
; i
++) {
385 page
= find_or_create_page(mapping
, first
+ i
, gfp_mask
);
386 if (unlikely(page
== NULL
)) {
387 if (flags
& XBF_READ_AHEAD
) {
388 bp
->b_page_count
= i
;
389 for (i
= 0; i
< bp
->b_page_count
; i
++)
390 unlock_page(bp
->b_pages
[i
]);
395 * This could deadlock.
397 * But until all the XFS lowlevel code is revamped to
398 * handle buffer allocation failures we can't do much.
400 if (!(++retries
% 100))
402 "XFS: possible memory allocation "
403 "deadlock in %s (mode:0x%x)\n",
406 XFS_STATS_INC(xb_page_retries
);
407 xfsbufd_wakeup(0, gfp_mask
);
408 congestion_wait(WRITE
, HZ
/50);
412 XFS_STATS_INC(xb_page_found
);
414 nbytes
= min_t(size_t, size
, PAGE_CACHE_SIZE
- offset
);
417 ASSERT(!PagePrivate(page
));
418 if (!PageUptodate(page
)) {
420 if (blocksize
>= PAGE_CACHE_SIZE
) {
421 if (flags
& XBF_READ
)
422 bp
->b_flags
|= _XBF_PAGE_LOCKED
;
423 } else if (!PagePrivate(page
)) {
424 if (test_page_region(page
, offset
, nbytes
))
429 bp
->b_pages
[i
] = page
;
433 if (!(bp
->b_flags
& _XBF_PAGE_LOCKED
)) {
434 for (i
= 0; i
< bp
->b_page_count
; i
++)
435 unlock_page(bp
->b_pages
[i
]);
438 if (page_count
== bp
->b_page_count
)
439 bp
->b_flags
|= XBF_DONE
;
441 XB_TRACE(bp
, "lookup_pages", (long)page_count
);
446 * Map buffer into kernel address-space if nessecary.
453 /* A single page buffer is always mappable */
454 if (bp
->b_page_count
== 1) {
455 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
456 bp
->b_flags
|= XBF_MAPPED
;
457 } else if (flags
& XBF_MAPPED
) {
458 if (as_list_len
> 64)
460 bp
->b_addr
= vmap(bp
->b_pages
, bp
->b_page_count
,
461 VM_MAP
, PAGE_KERNEL
);
462 if (unlikely(bp
->b_addr
== NULL
))
464 bp
->b_addr
+= bp
->b_offset
;
465 bp
->b_flags
|= XBF_MAPPED
;
472 * Finding and Reading Buffers
476 * Look up, and creates if absent, a lockable buffer for
477 * a given range of an inode. The buffer is returned
478 * locked. If other overlapping buffers exist, they are
479 * released before the new buffer is created and locked,
480 * which may imply that this call will block until those buffers
481 * are unlocked. No I/O is implied by this call.
485 xfs_buftarg_t
*btp
, /* block device target */
486 xfs_off_t ioff
, /* starting offset of range */
487 size_t isize
, /* length of range */
488 xfs_buf_flags_t flags
,
491 xfs_off_t range_base
;
496 range_base
= (ioff
<< BBSHIFT
);
497 range_length
= (isize
<< BBSHIFT
);
499 /* Check for IOs smaller than the sector size / not sector aligned */
500 ASSERT(!(range_length
< (1 << btp
->bt_sshift
)));
501 ASSERT(!(range_base
& (xfs_off_t
)btp
->bt_smask
));
503 hash
= &btp
->bt_hash
[hash_long((unsigned long)ioff
, btp
->bt_hashshift
)];
505 spin_lock(&hash
->bh_lock
);
507 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
508 ASSERT(btp
== bp
->b_target
);
509 if (bp
->b_file_offset
== range_base
&&
510 bp
->b_buffer_length
== range_length
) {
512 * If we look at something, bring it to the
513 * front of the list for next time.
515 atomic_inc(&bp
->b_hold
);
516 list_move(&bp
->b_hash_list
, &hash
->bh_list
);
523 _xfs_buf_initialize(new_bp
, btp
, range_base
,
524 range_length
, flags
);
525 new_bp
->b_hash
= hash
;
526 list_add(&new_bp
->b_hash_list
, &hash
->bh_list
);
528 XFS_STATS_INC(xb_miss_locked
);
531 spin_unlock(&hash
->bh_lock
);
535 spin_unlock(&hash
->bh_lock
);
537 /* Attempt to get the semaphore without sleeping,
538 * if this does not work then we need to drop the
539 * spinlock and do a hard attempt on the semaphore.
541 if (down_trylock(&bp
->b_sema
)) {
542 if (!(flags
& XBF_TRYLOCK
)) {
543 /* wait for buffer ownership */
544 XB_TRACE(bp
, "get_lock", 0);
546 XFS_STATS_INC(xb_get_locked_waited
);
548 /* We asked for a trylock and failed, no need
549 * to look at file offset and length here, we
550 * know that this buffer at least overlaps our
551 * buffer and is locked, therefore our buffer
552 * either does not exist, or is this buffer.
555 XFS_STATS_INC(xb_busy_locked
);
563 if (bp
->b_flags
& XBF_STALE
) {
564 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
565 bp
->b_flags
&= XBF_MAPPED
;
567 XB_TRACE(bp
, "got_lock", 0);
568 XFS_STATS_INC(xb_get_locked
);
573 * Assembles a buffer covering the specified range.
574 * Storage in memory for all portions of the buffer will be allocated,
575 * although backing storage may not be.
579 xfs_buftarg_t
*target
,/* target for buffer */
580 xfs_off_t ioff
, /* starting offset of range */
581 size_t isize
, /* length of range */
582 xfs_buf_flags_t flags
)
584 xfs_buf_t
*bp
, *new_bp
;
587 new_bp
= xfs_buf_allocate(flags
);
588 if (unlikely(!new_bp
))
591 bp
= _xfs_buf_find(target
, ioff
, isize
, flags
, new_bp
);
593 error
= _xfs_buf_lookup_pages(bp
, flags
);
597 xfs_buf_deallocate(new_bp
);
598 if (unlikely(bp
== NULL
))
602 for (i
= 0; i
< bp
->b_page_count
; i
++)
603 mark_page_accessed(bp
->b_pages
[i
]);
605 if (!(bp
->b_flags
& XBF_MAPPED
)) {
606 error
= _xfs_buf_map_pages(bp
, flags
);
607 if (unlikely(error
)) {
608 printk(KERN_WARNING
"%s: failed to map pages\n",
614 XFS_STATS_INC(xb_get
);
617 * Always fill in the block number now, the mapped cases can do
618 * their own overlay of this later.
621 bp
->b_count_desired
= bp
->b_buffer_length
;
623 XB_TRACE(bp
, "get", (unsigned long)flags
);
627 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
635 xfs_buftarg_t
*target
,
638 xfs_buf_flags_t flags
)
644 bp
= xfs_buf_get_flags(target
, ioff
, isize
, flags
);
646 if (!XFS_BUF_ISDONE(bp
)) {
647 XB_TRACE(bp
, "read", (unsigned long)flags
);
648 XFS_STATS_INC(xb_get_read
);
649 xfs_buf_iostart(bp
, flags
);
650 } else if (flags
& XBF_ASYNC
) {
651 XB_TRACE(bp
, "read_async", (unsigned long)flags
);
653 * Read ahead call which is already satisfied,
658 XB_TRACE(bp
, "read_done", (unsigned long)flags
);
659 /* We do not want read in the flags */
660 bp
->b_flags
&= ~XBF_READ
;
667 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
674 * If we are not low on memory then do the readahead in a deadlock
679 xfs_buftarg_t
*target
,
682 xfs_buf_flags_t flags
)
684 struct backing_dev_info
*bdi
;
686 bdi
= target
->bt_mapping
->backing_dev_info
;
687 if (bdi_read_congested(bdi
))
690 flags
|= (XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
);
691 xfs_buf_read_flags(target
, ioff
, isize
, flags
);
697 xfs_buftarg_t
*target
)
701 bp
= xfs_buf_allocate(0);
703 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
707 static inline struct page
*
711 if ((!is_vmalloc_addr(addr
))) {
712 return virt_to_page(addr
);
714 return vmalloc_to_page(addr
);
719 xfs_buf_associate_memory(
726 unsigned long pageaddr
;
727 unsigned long offset
;
731 pageaddr
= (unsigned long)mem
& PAGE_CACHE_MASK
;
732 offset
= (unsigned long)mem
- pageaddr
;
733 buflen
= PAGE_CACHE_ALIGN(len
+ offset
);
734 page_count
= buflen
>> PAGE_CACHE_SHIFT
;
736 /* Free any previous set of page pointers */
738 _xfs_buf_free_pages(bp
);
743 rval
= _xfs_buf_get_pages(bp
, page_count
, 0);
747 bp
->b_offset
= offset
;
749 for (i
= 0; i
< bp
->b_page_count
; i
++) {
750 bp
->b_pages
[i
] = mem_to_page((void *)pageaddr
);
751 pageaddr
+= PAGE_CACHE_SIZE
;
754 bp
->b_count_desired
= len
;
755 bp
->b_buffer_length
= buflen
;
756 bp
->b_flags
|= XBF_MAPPED
;
757 bp
->b_flags
&= ~_XBF_PAGE_LOCKED
;
765 xfs_buftarg_t
*target
)
767 unsigned long page_count
= PAGE_ALIGN(len
) >> PAGE_SHIFT
;
771 bp
= xfs_buf_allocate(0);
772 if (unlikely(bp
== NULL
))
774 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
776 error
= _xfs_buf_get_pages(bp
, page_count
, 0);
780 for (i
= 0; i
< page_count
; i
++) {
781 bp
->b_pages
[i
] = alloc_page(GFP_KERNEL
);
785 bp
->b_flags
|= _XBF_PAGES
;
787 error
= _xfs_buf_map_pages(bp
, XBF_MAPPED
);
788 if (unlikely(error
)) {
789 printk(KERN_WARNING
"%s: failed to map pages\n",
796 XB_TRACE(bp
, "no_daddr", len
);
801 __free_page(bp
->b_pages
[i
]);
802 _xfs_buf_free_pages(bp
);
804 xfs_buf_deallocate(bp
);
810 * Increment reference count on buffer, to hold the buffer concurrently
811 * with another thread which may release (free) the buffer asynchronously.
812 * Must hold the buffer already to call this function.
818 atomic_inc(&bp
->b_hold
);
819 XB_TRACE(bp
, "hold", 0);
823 * Releases a hold on the specified buffer. If the
824 * the hold count is 1, calls xfs_buf_free.
830 xfs_bufhash_t
*hash
= bp
->b_hash
;
832 XB_TRACE(bp
, "rele", bp
->b_relse
);
834 if (unlikely(!hash
)) {
835 ASSERT(!bp
->b_relse
);
836 if (atomic_dec_and_test(&bp
->b_hold
))
841 ASSERT(atomic_read(&bp
->b_hold
) > 0);
842 if (atomic_dec_and_lock(&bp
->b_hold
, &hash
->bh_lock
)) {
844 atomic_inc(&bp
->b_hold
);
845 spin_unlock(&hash
->bh_lock
);
846 (*(bp
->b_relse
)) (bp
);
847 } else if (bp
->b_flags
& XBF_FS_MANAGED
) {
848 spin_unlock(&hash
->bh_lock
);
850 ASSERT(!(bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)));
851 list_del_init(&bp
->b_hash_list
);
852 spin_unlock(&hash
->bh_lock
);
860 * Mutual exclusion on buffers. Locking model:
862 * Buffers associated with inodes for which buffer locking
863 * is not enabled are not protected by semaphores, and are
864 * assumed to be exclusively owned by the caller. There is a
865 * spinlock in the buffer, used by the caller when concurrent
866 * access is possible.
870 * Locks a buffer object, if it is not already locked.
871 * Note that this in no way locks the underlying pages, so it is only
872 * useful for synchronizing concurrent use of buffer objects, not for
873 * synchronizing independent access to the underlying pages.
881 locked
= down_trylock(&bp
->b_sema
) == 0;
885 XB_TRACE(bp
, "cond_lock", (long)locked
);
886 return locked
? 0 : -EBUSY
;
889 #if defined(DEBUG) || defined(XFS_BLI_TRACE)
894 return bp
->b_sema
.count
;
899 * Locks a buffer object.
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.
908 XB_TRACE(bp
, "lock", 0);
909 if (atomic_read(&bp
->b_io_remaining
))
910 blk_run_address_space(bp
->b_target
->bt_mapping
);
913 XB_TRACE(bp
, "locked", 0);
917 * Releases the lock on the buffer object.
918 * If the buffer is marked delwri but is not queued, do so before we
919 * unlock the buffer as we need to set flags correctly. We also need to
920 * take a reference for the delwri queue because the unlocker is going to
921 * drop their's and they don't know we just queued it.
927 if ((bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)) == XBF_DELWRI
) {
928 atomic_inc(&bp
->b_hold
);
929 bp
->b_flags
|= XBF_ASYNC
;
930 xfs_buf_delwri_queue(bp
, 0);
935 XB_TRACE(bp
, "unlock", 0);
940 * Pinning Buffer Storage in Memory
941 * Ensure that no attempt to force a buffer to disk will succeed.
947 atomic_inc(&bp
->b_pin_count
);
948 XB_TRACE(bp
, "pin", (long)bp
->b_pin_count
.counter
);
955 if (atomic_dec_and_test(&bp
->b_pin_count
))
956 wake_up_all(&bp
->b_waiters
);
957 XB_TRACE(bp
, "unpin", (long)bp
->b_pin_count
.counter
);
964 return atomic_read(&bp
->b_pin_count
);
971 DECLARE_WAITQUEUE (wait
, current
);
973 if (atomic_read(&bp
->b_pin_count
) == 0)
976 add_wait_queue(&bp
->b_waiters
, &wait
);
978 set_current_state(TASK_UNINTERRUPTIBLE
);
979 if (atomic_read(&bp
->b_pin_count
) == 0)
981 if (atomic_read(&bp
->b_io_remaining
))
982 blk_run_address_space(bp
->b_target
->bt_mapping
);
985 remove_wait_queue(&bp
->b_waiters
, &wait
);
986 set_current_state(TASK_RUNNING
);
990 * Buffer Utility Routines
995 struct work_struct
*work
)
998 container_of(work
, xfs_buf_t
, b_iodone_work
);
1001 * We can get an EOPNOTSUPP to ordered writes. Here we clear the
1002 * ordered flag and reissue them. Because we can't tell the higher
1003 * layers directly that they should not issue ordered I/O anymore, they
1004 * need to check if the ordered flag was cleared during I/O completion.
1006 if ((bp
->b_error
== EOPNOTSUPP
) &&
1007 (bp
->b_flags
& (XBF_ORDERED
|XBF_ASYNC
)) == (XBF_ORDERED
|XBF_ASYNC
)) {
1008 XB_TRACE(bp
, "ordered_retry", bp
->b_iodone
);
1009 bp
->b_flags
&= ~XBF_ORDERED
;
1010 xfs_buf_iorequest(bp
);
1011 } else if (bp
->b_iodone
)
1012 (*(bp
->b_iodone
))(bp
);
1013 else if (bp
->b_flags
& XBF_ASYNC
)
1022 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_READ_AHEAD
);
1023 if (bp
->b_error
== 0)
1024 bp
->b_flags
|= XBF_DONE
;
1026 XB_TRACE(bp
, "iodone", bp
->b_iodone
);
1028 if ((bp
->b_iodone
) || (bp
->b_flags
& XBF_ASYNC
)) {
1030 INIT_WORK(&bp
->b_iodone_work
, xfs_buf_iodone_work
);
1031 queue_work(xfslogd_workqueue
, &bp
->b_iodone_work
);
1033 xfs_buf_iodone_work(&bp
->b_iodone_work
);
1036 complete(&bp
->b_iowait
);
1045 ASSERT(error
>= 0 && error
<= 0xffff);
1046 bp
->b_error
= (unsigned short)error
;
1047 XB_TRACE(bp
, "ioerror", (unsigned long)error
);
1051 * Initiate I/O on a buffer, based on the flags supplied.
1052 * The b_iodone routine in the buffer supplied will only be called
1053 * when all of the subsidiary I/O requests, if any, have been completed.
1058 xfs_buf_flags_t flags
)
1062 XB_TRACE(bp
, "iostart", (unsigned long)flags
);
1064 if (flags
& XBF_DELWRI
) {
1065 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_ASYNC
);
1066 bp
->b_flags
|= flags
& (XBF_DELWRI
| XBF_ASYNC
);
1067 xfs_buf_delwri_queue(bp
, 1);
1071 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_ASYNC
| XBF_DELWRI
| \
1072 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
1073 bp
->b_flags
|= flags
& (XBF_READ
| XBF_WRITE
| XBF_ASYNC
| \
1074 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
1076 BUG_ON(bp
->b_bn
== XFS_BUF_DADDR_NULL
);
1078 /* For writes allow an alternate strategy routine to precede
1079 * the actual I/O request (which may not be issued at all in
1080 * a shutdown situation, for example).
1082 status
= (flags
& XBF_WRITE
) ?
1083 xfs_buf_iostrategy(bp
) : xfs_buf_iorequest(bp
);
1085 /* Wait for I/O if we are not an async request.
1086 * Note: async I/O request completion will release the buffer,
1087 * and that can already be done by this point. So using the
1088 * buffer pointer from here on, after async I/O, is invalid.
1090 if (!status
&& !(flags
& XBF_ASYNC
))
1091 status
= xfs_buf_iowait(bp
);
1101 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1102 bp
->b_flags
&= ~_XBF_PAGE_LOCKED
;
1103 xfs_buf_ioend(bp
, schedule
);
1112 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1113 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1114 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1116 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1120 struct page
*page
= bvec
->bv_page
;
1122 ASSERT(!PagePrivate(page
));
1123 if (unlikely(bp
->b_error
)) {
1124 if (bp
->b_flags
& XBF_READ
)
1125 ClearPageUptodate(page
);
1126 } else if (blocksize
>= PAGE_CACHE_SIZE
) {
1127 SetPageUptodate(page
);
1128 } else if (!PagePrivate(page
) &&
1129 (bp
->b_flags
& _XBF_PAGE_CACHE
)) {
1130 set_page_region(page
, bvec
->bv_offset
, bvec
->bv_len
);
1133 if (--bvec
>= bio
->bi_io_vec
)
1134 prefetchw(&bvec
->bv_page
->flags
);
1136 if (bp
->b_flags
& _XBF_PAGE_LOCKED
)
1138 } while (bvec
>= bio
->bi_io_vec
);
1140 _xfs_buf_ioend(bp
, 1);
1148 int rw
, map_i
, total_nr_pages
, nr_pages
;
1150 int offset
= bp
->b_offset
;
1151 int size
= bp
->b_count_desired
;
1152 sector_t sector
= bp
->b_bn
;
1153 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1155 total_nr_pages
= bp
->b_page_count
;
1158 if (bp
->b_flags
& XBF_ORDERED
) {
1159 ASSERT(!(bp
->b_flags
& XBF_READ
));
1161 } else if (bp
->b_flags
& _XBF_RUN_QUEUES
) {
1162 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1163 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1164 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_SYNC
: READ_SYNC
;
1166 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE
:
1167 (bp
->b_flags
& XBF_READ_AHEAD
) ? READA
: READ
;
1170 /* Special code path for reading a sub page size buffer in --
1171 * we populate up the whole page, and hence the other metadata
1172 * in the same page. This optimization is only valid when the
1173 * filesystem block size is not smaller than the page size.
1175 if ((bp
->b_buffer_length
< PAGE_CACHE_SIZE
) &&
1176 ((bp
->b_flags
& (XBF_READ
|_XBF_PAGE_LOCKED
)) ==
1177 (XBF_READ
|_XBF_PAGE_LOCKED
)) &&
1178 (blocksize
>= PAGE_CACHE_SIZE
)) {
1179 bio
= bio_alloc(GFP_NOIO
, 1);
1181 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1182 bio
->bi_sector
= sector
- (offset
>> BBSHIFT
);
1183 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1184 bio
->bi_private
= bp
;
1186 bio_add_page(bio
, bp
->b_pages
[0], PAGE_CACHE_SIZE
, 0);
1189 atomic_inc(&bp
->b_io_remaining
);
1195 atomic_inc(&bp
->b_io_remaining
);
1196 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1197 if (nr_pages
> total_nr_pages
)
1198 nr_pages
= total_nr_pages
;
1200 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1201 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1202 bio
->bi_sector
= sector
;
1203 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1204 bio
->bi_private
= bp
;
1206 for (; size
&& nr_pages
; nr_pages
--, map_i
++) {
1207 int rbytes
, nbytes
= PAGE_CACHE_SIZE
- offset
;
1212 rbytes
= bio_add_page(bio
, bp
->b_pages
[map_i
], nbytes
, offset
);
1213 if (rbytes
< nbytes
)
1217 sector
+= nbytes
>> BBSHIFT
;
1223 if (likely(bio
->bi_size
)) {
1224 submit_bio(rw
, bio
);
1229 xfs_buf_ioerror(bp
, EIO
);
1237 XB_TRACE(bp
, "iorequest", 0);
1239 if (bp
->b_flags
& XBF_DELWRI
) {
1240 xfs_buf_delwri_queue(bp
, 1);
1244 if (bp
->b_flags
& XBF_WRITE
) {
1245 xfs_buf_wait_unpin(bp
);
1250 /* Set the count to 1 initially, this will stop an I/O
1251 * completion callout which happens before we have started
1252 * all the I/O from calling xfs_buf_ioend too early.
1254 atomic_set(&bp
->b_io_remaining
, 1);
1255 _xfs_buf_ioapply(bp
);
1256 _xfs_buf_ioend(bp
, 0);
1263 * Waits for I/O to complete on the buffer supplied.
1264 * It returns immediately if no I/O is pending.
1265 * It returns the I/O error code, if any, or 0 if there was no error.
1271 XB_TRACE(bp
, "iowait", 0);
1272 if (atomic_read(&bp
->b_io_remaining
))
1273 blk_run_address_space(bp
->b_target
->bt_mapping
);
1274 wait_for_completion(&bp
->b_iowait
);
1275 XB_TRACE(bp
, "iowaited", (long)bp
->b_error
);
1286 if (bp
->b_flags
& XBF_MAPPED
)
1287 return XFS_BUF_PTR(bp
) + offset
;
1289 offset
+= bp
->b_offset
;
1290 page
= bp
->b_pages
[offset
>> PAGE_CACHE_SHIFT
];
1291 return (xfs_caddr_t
)page_address(page
) + (offset
& (PAGE_CACHE_SIZE
-1));
1295 * Move data into or out of a buffer.
1299 xfs_buf_t
*bp
, /* buffer to process */
1300 size_t boff
, /* starting buffer offset */
1301 size_t bsize
, /* length to copy */
1302 caddr_t data
, /* data address */
1303 xfs_buf_rw_t mode
) /* read/write/zero flag */
1305 size_t bend
, cpoff
, csize
;
1308 bend
= boff
+ bsize
;
1309 while (boff
< bend
) {
1310 page
= bp
->b_pages
[xfs_buf_btoct(boff
+ bp
->b_offset
)];
1311 cpoff
= xfs_buf_poff(boff
+ bp
->b_offset
);
1312 csize
= min_t(size_t,
1313 PAGE_CACHE_SIZE
-cpoff
, bp
->b_count_desired
-boff
);
1315 ASSERT(((csize
+ cpoff
) <= PAGE_CACHE_SIZE
));
1319 memset(page_address(page
) + cpoff
, 0, csize
);
1322 memcpy(data
, page_address(page
) + cpoff
, csize
);
1325 memcpy(page_address(page
) + cpoff
, data
, csize
);
1334 * Handling of buffer targets (buftargs).
1338 * Wait for any bufs with callbacks that have been submitted but
1339 * have not yet returned... walk the hash list for the target.
1346 xfs_bufhash_t
*hash
;
1349 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1350 hash
= &btp
->bt_hash
[i
];
1352 spin_lock(&hash
->bh_lock
);
1353 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
1354 ASSERT(btp
== bp
->b_target
);
1355 if (!(bp
->b_flags
& XBF_FS_MANAGED
)) {
1356 spin_unlock(&hash
->bh_lock
);
1358 * Catch superblock reference count leaks
1361 BUG_ON(bp
->b_bn
== 0);
1366 spin_unlock(&hash
->bh_lock
);
1371 * Allocate buffer hash table for a given target.
1372 * For devices containing metadata (i.e. not the log/realtime devices)
1373 * we need to allocate a much larger hash table.
1382 btp
->bt_hashshift
= external
? 3 : 8; /* 8 or 256 buckets */
1383 btp
->bt_hashmask
= (1 << btp
->bt_hashshift
) - 1;
1384 btp
->bt_hash
= kmem_zalloc((1 << btp
->bt_hashshift
) *
1385 sizeof(xfs_bufhash_t
), KM_SLEEP
| KM_LARGE
);
1386 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1387 spin_lock_init(&btp
->bt_hash
[i
].bh_lock
);
1388 INIT_LIST_HEAD(&btp
->bt_hash
[i
].bh_list
);
1396 kmem_free(btp
->bt_hash
);
1397 btp
->bt_hash
= NULL
;
1401 * buftarg list for delwrite queue processing
1403 static LIST_HEAD(xfs_buftarg_list
);
1404 static DEFINE_SPINLOCK(xfs_buftarg_lock
);
1407 xfs_register_buftarg(
1410 spin_lock(&xfs_buftarg_lock
);
1411 list_add(&btp
->bt_list
, &xfs_buftarg_list
);
1412 spin_unlock(&xfs_buftarg_lock
);
1416 xfs_unregister_buftarg(
1419 spin_lock(&xfs_buftarg_lock
);
1420 list_del(&btp
->bt_list
);
1421 spin_unlock(&xfs_buftarg_lock
);
1428 xfs_flush_buftarg(btp
, 1);
1429 xfs_blkdev_issue_flush(btp
);
1430 xfs_free_bufhash(btp
);
1431 iput(btp
->bt_mapping
->host
);
1433 /* Unregister the buftarg first so that we don't get a
1434 * wakeup finding a non-existent task
1436 xfs_unregister_buftarg(btp
);
1437 kthread_stop(btp
->bt_task
);
1443 xfs_setsize_buftarg_flags(
1445 unsigned int blocksize
,
1446 unsigned int sectorsize
,
1449 btp
->bt_bsize
= blocksize
;
1450 btp
->bt_sshift
= ffs(sectorsize
) - 1;
1451 btp
->bt_smask
= sectorsize
- 1;
1453 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1455 "XFS: Cannot set_blocksize to %u on device %s\n",
1456 sectorsize
, XFS_BUFTARG_NAME(btp
));
1461 (PAGE_CACHE_SIZE
/ BITS_PER_LONG
) > sectorsize
) {
1463 "XFS: %u byte sectors in use on device %s. "
1464 "This is suboptimal; %u or greater is ideal.\n",
1465 sectorsize
, XFS_BUFTARG_NAME(btp
),
1466 (unsigned int)PAGE_CACHE_SIZE
/ BITS_PER_LONG
);
1473 * When allocating the initial buffer target we have not yet
1474 * read in the superblock, so don't know what sized sectors
1475 * are being used is at this early stage. Play safe.
1478 xfs_setsize_buftarg_early(
1480 struct block_device
*bdev
)
1482 return xfs_setsize_buftarg_flags(btp
,
1483 PAGE_CACHE_SIZE
, bdev_hardsect_size(bdev
), 0);
1487 xfs_setsize_buftarg(
1489 unsigned int blocksize
,
1490 unsigned int sectorsize
)
1492 return xfs_setsize_buftarg_flags(btp
, blocksize
, sectorsize
, 1);
1496 xfs_mapping_buftarg(
1498 struct block_device
*bdev
)
1500 struct backing_dev_info
*bdi
;
1501 struct inode
*inode
;
1502 struct address_space
*mapping
;
1503 static const struct address_space_operations mapping_aops
= {
1504 .sync_page
= block_sync_page
,
1505 .migratepage
= fail_migrate_page
,
1508 inode
= new_inode(bdev
->bd_inode
->i_sb
);
1511 "XFS: Cannot allocate mapping inode for device %s\n",
1512 XFS_BUFTARG_NAME(btp
));
1515 inode
->i_mode
= S_IFBLK
;
1516 inode
->i_bdev
= bdev
;
1517 inode
->i_rdev
= bdev
->bd_dev
;
1518 bdi
= blk_get_backing_dev_info(bdev
);
1520 bdi
= &default_backing_dev_info
;
1521 mapping
= &inode
->i_data
;
1522 mapping
->a_ops
= &mapping_aops
;
1523 mapping
->backing_dev_info
= bdi
;
1524 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
1525 btp
->bt_mapping
= mapping
;
1530 xfs_alloc_delwrite_queue(
1535 INIT_LIST_HEAD(&btp
->bt_list
);
1536 INIT_LIST_HEAD(&btp
->bt_delwrite_queue
);
1537 spin_lock_init(&btp
->bt_delwrite_lock
);
1539 btp
->bt_task
= kthread_run(xfsbufd
, btp
, "xfsbufd");
1540 if (IS_ERR(btp
->bt_task
)) {
1541 error
= PTR_ERR(btp
->bt_task
);
1544 xfs_register_buftarg(btp
);
1551 struct block_device
*bdev
,
1556 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
);
1558 btp
->bt_dev
= bdev
->bd_dev
;
1559 btp
->bt_bdev
= bdev
;
1560 if (xfs_setsize_buftarg_early(btp
, bdev
))
1562 if (xfs_mapping_buftarg(btp
, bdev
))
1564 if (xfs_alloc_delwrite_queue(btp
))
1566 xfs_alloc_bufhash(btp
, external
);
1576 * Delayed write buffer handling
1579 xfs_buf_delwri_queue(
1583 struct list_head
*dwq
= &bp
->b_target
->bt_delwrite_queue
;
1584 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1586 XB_TRACE(bp
, "delwri_q", (long)unlock
);
1587 ASSERT((bp
->b_flags
&(XBF_DELWRI
|XBF_ASYNC
)) == (XBF_DELWRI
|XBF_ASYNC
));
1590 /* If already in the queue, dequeue and place at tail */
1591 if (!list_empty(&bp
->b_list
)) {
1592 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1594 atomic_dec(&bp
->b_hold
);
1595 list_del(&bp
->b_list
);
1598 bp
->b_flags
|= _XBF_DELWRI_Q
;
1599 list_add_tail(&bp
->b_list
, dwq
);
1600 bp
->b_queuetime
= jiffies
;
1608 xfs_buf_delwri_dequeue(
1611 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1615 if ((bp
->b_flags
& XBF_DELWRI
) && !list_empty(&bp
->b_list
)) {
1616 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1617 list_del_init(&bp
->b_list
);
1620 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
);
1626 XB_TRACE(bp
, "delwri_dq", (long)dequeued
);
1630 xfs_buf_runall_queues(
1631 struct workqueue_struct
*queue
)
1633 flush_workqueue(queue
);
1643 spin_lock(&xfs_buftarg_lock
);
1644 list_for_each_entry(btp
, &xfs_buftarg_list
, bt_list
) {
1645 if (test_bit(XBT_FORCE_SLEEP
, &btp
->bt_flags
))
1647 set_bit(XBT_FORCE_FLUSH
, &btp
->bt_flags
);
1648 wake_up_process(btp
->bt_task
);
1650 spin_unlock(&xfs_buftarg_lock
);
1655 * Move as many buffers as specified to the supplied list
1656 * idicating if we skipped any buffers to prevent deadlocks.
1659 xfs_buf_delwri_split(
1660 xfs_buftarg_t
*target
,
1661 struct list_head
*list
,
1665 struct list_head
*dwq
= &target
->bt_delwrite_queue
;
1666 spinlock_t
*dwlk
= &target
->bt_delwrite_lock
;
1670 force
= test_and_clear_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1671 INIT_LIST_HEAD(list
);
1673 list_for_each_entry_safe(bp
, n
, dwq
, b_list
) {
1674 XB_TRACE(bp
, "walkq1", (long)xfs_buf_ispin(bp
));
1675 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1677 if (!xfs_buf_ispin(bp
) && !xfs_buf_cond_lock(bp
)) {
1679 time_before(jiffies
, bp
->b_queuetime
+ age
)) {
1684 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
|
1686 bp
->b_flags
|= XBF_WRITE
;
1687 list_move_tail(&bp
->b_list
, list
);
1701 struct list_head tmp
;
1702 xfs_buftarg_t
*target
= (xfs_buftarg_t
*)data
;
1706 current
->flags
|= PF_MEMALLOC
;
1711 if (unlikely(freezing(current
))) {
1712 set_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1715 clear_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1718 schedule_timeout_interruptible(
1719 xfs_buf_timer_centisecs
* msecs_to_jiffies(10));
1721 xfs_buf_delwri_split(target
, &tmp
,
1722 xfs_buf_age_centisecs
* msecs_to_jiffies(10));
1725 while (!list_empty(&tmp
)) {
1726 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1727 ASSERT(target
== bp
->b_target
);
1729 list_del_init(&bp
->b_list
);
1730 xfs_buf_iostrategy(bp
);
1734 if (as_list_len
> 0)
1737 blk_run_address_space(target
->bt_mapping
);
1739 } while (!kthread_should_stop());
1745 * Go through all incore buffers, and release buffers if they belong to
1746 * the given device. This is used in filesystem error handling to
1747 * preserve the consistency of its metadata.
1751 xfs_buftarg_t
*target
,
1754 struct list_head tmp
;
1758 xfs_buf_runall_queues(xfsdatad_workqueue
);
1759 xfs_buf_runall_queues(xfslogd_workqueue
);
1761 set_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1762 pincount
= xfs_buf_delwri_split(target
, &tmp
, 0);
1765 * Dropped the delayed write list lock, now walk the temporary list
1767 list_for_each_entry_safe(bp
, n
, &tmp
, b_list
) {
1768 ASSERT(target
== bp
->b_target
);
1770 bp
->b_flags
&= ~XBF_ASYNC
;
1772 list_del_init(&bp
->b_list
);
1774 xfs_buf_iostrategy(bp
);
1778 blk_run_address_space(target
->bt_mapping
);
1781 * Remaining list items must be flushed before returning
1783 while (!list_empty(&tmp
)) {
1784 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1786 list_del_init(&bp
->b_list
);
1797 #ifdef XFS_BUF_TRACE
1798 xfs_buf_trace_buf
= ktrace_alloc(XFS_BUF_TRACE_SIZE
, KM_NOFS
);
1801 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1802 KM_ZONE_HWALIGN
, NULL
);
1804 goto out_free_trace_buf
;
1806 xfslogd_workqueue
= create_workqueue("xfslogd");
1807 if (!xfslogd_workqueue
)
1808 goto out_free_buf_zone
;
1810 xfsdatad_workqueue
= create_workqueue("xfsdatad");
1811 if (!xfsdatad_workqueue
)
1812 goto out_destroy_xfslogd_workqueue
;
1814 register_shrinker(&xfs_buf_shake
);
1817 out_destroy_xfslogd_workqueue
:
1818 destroy_workqueue(xfslogd_workqueue
);
1820 kmem_zone_destroy(xfs_buf_zone
);
1822 #ifdef XFS_BUF_TRACE
1823 ktrace_free(xfs_buf_trace_buf
);
1829 xfs_buf_terminate(void)
1831 unregister_shrinker(&xfs_buf_shake
);
1832 destroy_workqueue(xfsdatad_workqueue
);
1833 destroy_workqueue(xfslogd_workqueue
);
1834 kmem_zone_destroy(xfs_buf_zone
);
1835 #ifdef XFS_BUF_TRACE
1836 ktrace_free(xfs_buf_trace_buf
);
1840 #ifdef CONFIG_KDB_MODULES
1842 xfs_get_buftarg_list(void)
1844 return &xfs_buftarg_list
;