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>
36 STATIC kmem_zone_t
*xfs_buf_zone
;
37 STATIC kmem_shaker_t xfs_buf_shake
;
38 STATIC
int xfsbufd(void *);
39 STATIC
int xfsbufd_wakeup(int, gfp_t
);
40 STATIC
void xfs_buf_delwri_queue(xfs_buf_t
*, int);
42 STATIC
struct workqueue_struct
*xfslogd_workqueue
;
43 struct workqueue_struct
*xfsdatad_workqueue
;
53 ktrace_enter(xfs_buf_trace_buf
,
55 (void *)(unsigned long)bp
->b_flags
,
56 (void *)(unsigned long)bp
->b_hold
.counter
,
57 (void *)(unsigned long)bp
->b_sema
.count
.counter
,
60 (void *)(unsigned long)((bp
->b_file_offset
>>32) & 0xffffffff),
61 (void *)(unsigned long)(bp
->b_file_offset
& 0xffffffff),
62 (void *)(unsigned long)bp
->b_buffer_length
,
63 NULL
, NULL
, NULL
, NULL
, NULL
);
65 ktrace_t
*xfs_buf_trace_buf
;
66 #define XFS_BUF_TRACE_SIZE 4096
67 #define XB_TRACE(bp, id, data) \
68 xfs_buf_trace(bp, id, (void *)data, (void *)__builtin_return_address(0))
70 #define XB_TRACE(bp, id, data) do { } while (0)
73 #ifdef XFS_BUF_LOCK_TRACKING
74 # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
75 # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
76 # define XB_GET_OWNER(bp) ((bp)->b_last_holder)
78 # define XB_SET_OWNER(bp) do { } while (0)
79 # define XB_CLEAR_OWNER(bp) do { } while (0)
80 # define XB_GET_OWNER(bp) do { } while (0)
83 #define xb_to_gfp(flags) \
84 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \
85 ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)
87 #define xb_to_km(flags) \
88 (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)
90 #define xfs_buf_allocate(flags) \
91 kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags))
92 #define xfs_buf_deallocate(bp) \
93 kmem_zone_free(xfs_buf_zone, (bp));
96 * Page Region interfaces.
98 * For pages in filesystems where the blocksize is smaller than the
99 * pagesize, we use the page->private field (long) to hold a bitmap
100 * of uptodate regions within the page.
102 * Each such region is "bytes per page / bits per long" bytes long.
104 * NBPPR == number-of-bytes-per-page-region
105 * BTOPR == bytes-to-page-region (rounded up)
106 * BTOPRT == bytes-to-page-region-truncated (rounded down)
108 #if (BITS_PER_LONG == 32)
109 #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */
110 #elif (BITS_PER_LONG == 64)
111 #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */
113 #error BITS_PER_LONG must be 32 or 64
115 #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG)
116 #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT)
117 #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT))
127 first
= BTOPR(offset
);
128 final
= BTOPRT(offset
+ length
- 1);
129 first
= min(first
, final
);
132 mask
<<= BITS_PER_LONG
- (final
- first
);
133 mask
>>= BITS_PER_LONG
- (final
);
135 ASSERT(offset
+ length
<= PAGE_CACHE_SIZE
);
136 ASSERT((final
- first
) < BITS_PER_LONG
&& (final
- first
) >= 0);
147 set_page_private(page
,
148 page_private(page
) | page_region_mask(offset
, length
));
149 if (page_private(page
) == ~0UL)
150 SetPageUptodate(page
);
159 unsigned long mask
= page_region_mask(offset
, length
);
161 return (mask
&& (page_private(page
) & mask
) == mask
);
165 * Mapping of multi-page buffers into contiguous virtual space
168 typedef struct a_list
{
173 STATIC a_list_t
*as_free_head
;
174 STATIC
int as_list_len
;
175 STATIC
DEFINE_SPINLOCK(as_lock
);
178 * Try to batch vunmaps because they are costly.
186 aentry
= kmalloc(sizeof(a_list_t
), GFP_NOWAIT
);
187 if (likely(aentry
)) {
189 aentry
->next
= as_free_head
;
190 aentry
->vm_addr
= addr
;
191 as_free_head
= aentry
;
193 spin_unlock(&as_lock
);
200 purge_addresses(void)
202 a_list_t
*aentry
, *old
;
204 if (as_free_head
== NULL
)
208 aentry
= as_free_head
;
211 spin_unlock(&as_lock
);
213 while ((old
= aentry
) != NULL
) {
214 vunmap(aentry
->vm_addr
);
215 aentry
= aentry
->next
;
221 * Internal xfs_buf_t object manipulation
227 xfs_buftarg_t
*target
,
228 xfs_off_t range_base
,
230 xfs_buf_flags_t flags
)
233 * We don't want certain flags to appear in b_flags.
235 flags
&= ~(XBF_LOCK
|XBF_MAPPED
|XBF_DONT_BLOCK
|XBF_READ_AHEAD
);
237 memset(bp
, 0, sizeof(xfs_buf_t
));
238 atomic_set(&bp
->b_hold
, 1);
239 init_MUTEX_LOCKED(&bp
->b_iodonesema
);
240 INIT_LIST_HEAD(&bp
->b_list
);
241 INIT_LIST_HEAD(&bp
->b_hash_list
);
242 init_MUTEX_LOCKED(&bp
->b_sema
); /* held, no waiters */
244 bp
->b_target
= target
;
245 bp
->b_file_offset
= range_base
;
247 * Set buffer_length and count_desired to the same value initially.
248 * I/O routines should use count_desired, which will be the same in
249 * most cases but may be reset (e.g. XFS recovery).
251 bp
->b_buffer_length
= bp
->b_count_desired
= range_length
;
253 bp
->b_bn
= XFS_BUF_DADDR_NULL
;
254 atomic_set(&bp
->b_pin_count
, 0);
255 init_waitqueue_head(&bp
->b_waiters
);
257 XFS_STATS_INC(xb_create
);
258 XB_TRACE(bp
, "initialize", target
);
262 * Allocate a page array capable of holding a specified number
263 * of pages, and point the page buf at it.
269 xfs_buf_flags_t flags
)
271 /* Make sure that we have a page list */
272 if (bp
->b_pages
== NULL
) {
273 bp
->b_offset
= xfs_buf_poff(bp
->b_file_offset
);
274 bp
->b_page_count
= page_count
;
275 if (page_count
<= XB_PAGES
) {
276 bp
->b_pages
= bp
->b_page_array
;
278 bp
->b_pages
= kmem_alloc(sizeof(struct page
*) *
279 page_count
, xb_to_km(flags
));
280 if (bp
->b_pages
== NULL
)
283 memset(bp
->b_pages
, 0, sizeof(struct page
*) * page_count
);
289 * Frees b_pages if it was allocated.
295 if (bp
->b_pages
!= bp
->b_page_array
) {
296 kmem_free(bp
->b_pages
,
297 bp
->b_page_count
* sizeof(struct page
*));
302 * Releases the specified buffer.
304 * The modification state of any associated pages is left unchanged.
305 * The buffer most not be on any hash - use xfs_buf_rele instead for
306 * hashed and refcounted buffers
312 XB_TRACE(bp
, "free", 0);
314 ASSERT(list_empty(&bp
->b_hash_list
));
316 if (bp
->b_flags
& _XBF_PAGE_CACHE
) {
319 if ((bp
->b_flags
& XBF_MAPPED
) && (bp
->b_page_count
> 1))
320 free_address(bp
->b_addr
- bp
->b_offset
);
322 for (i
= 0; i
< bp
->b_page_count
; i
++) {
323 struct page
*page
= bp
->b_pages
[i
];
325 ASSERT(!PagePrivate(page
));
326 page_cache_release(page
);
328 _xfs_buf_free_pages(bp
);
329 } else if (bp
->b_flags
& _XBF_KMEM_ALLOC
) {
331 * XXX(hch): bp->b_count_desired might be incorrect (see
332 * xfs_buf_associate_memory for details), but fortunately
333 * the Linux version of kmem_free ignores the len argument..
335 kmem_free(bp
->b_addr
, bp
->b_count_desired
);
336 _xfs_buf_free_pages(bp
);
339 xfs_buf_deallocate(bp
);
343 * Finds all pages for buffer in question and builds it's page list.
346 _xfs_buf_lookup_pages(
350 struct address_space
*mapping
= bp
->b_target
->bt_mapping
;
351 size_t blocksize
= bp
->b_target
->bt_bsize
;
352 size_t size
= bp
->b_count_desired
;
353 size_t nbytes
, offset
;
354 gfp_t gfp_mask
= xb_to_gfp(flags
);
355 unsigned short page_count
, i
;
360 end
= bp
->b_file_offset
+ bp
->b_buffer_length
;
361 page_count
= xfs_buf_btoc(end
) - xfs_buf_btoct(bp
->b_file_offset
);
363 error
= _xfs_buf_get_pages(bp
, page_count
, flags
);
366 bp
->b_flags
|= _XBF_PAGE_CACHE
;
368 offset
= bp
->b_offset
;
369 first
= bp
->b_file_offset
>> PAGE_CACHE_SHIFT
;
371 for (i
= 0; i
< bp
->b_page_count
; i
++) {
376 page
= find_or_create_page(mapping
, first
+ i
, gfp_mask
);
377 if (unlikely(page
== NULL
)) {
378 if (flags
& XBF_READ_AHEAD
) {
379 bp
->b_page_count
= i
;
380 for (i
= 0; i
< bp
->b_page_count
; i
++)
381 unlock_page(bp
->b_pages
[i
]);
386 * This could deadlock.
388 * But until all the XFS lowlevel code is revamped to
389 * handle buffer allocation failures we can't do much.
391 if (!(++retries
% 100))
393 "XFS: possible memory allocation "
394 "deadlock in %s (mode:0x%x)\n",
395 __FUNCTION__
, gfp_mask
);
397 XFS_STATS_INC(xb_page_retries
);
398 xfsbufd_wakeup(0, gfp_mask
);
399 congestion_wait(WRITE
, HZ
/50);
403 XFS_STATS_INC(xb_page_found
);
405 nbytes
= min_t(size_t, size
, PAGE_CACHE_SIZE
- offset
);
408 ASSERT(!PagePrivate(page
));
409 if (!PageUptodate(page
)) {
411 if (blocksize
>= PAGE_CACHE_SIZE
) {
412 if (flags
& XBF_READ
)
414 } else if (!PagePrivate(page
)) {
415 if (test_page_region(page
, offset
, nbytes
))
420 bp
->b_pages
[i
] = page
;
425 for (i
= 0; i
< bp
->b_page_count
; i
++)
426 unlock_page(bp
->b_pages
[i
]);
429 if (page_count
== bp
->b_page_count
)
430 bp
->b_flags
|= XBF_DONE
;
432 XB_TRACE(bp
, "lookup_pages", (long)page_count
);
437 * Map buffer into kernel address-space if nessecary.
444 /* A single page buffer is always mappable */
445 if (bp
->b_page_count
== 1) {
446 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
447 bp
->b_flags
|= XBF_MAPPED
;
448 } else if (flags
& XBF_MAPPED
) {
449 if (as_list_len
> 64)
451 bp
->b_addr
= vmap(bp
->b_pages
, bp
->b_page_count
,
452 VM_MAP
, PAGE_KERNEL
);
453 if (unlikely(bp
->b_addr
== NULL
))
455 bp
->b_addr
+= bp
->b_offset
;
456 bp
->b_flags
|= XBF_MAPPED
;
463 * Finding and Reading Buffers
467 * Look up, and creates if absent, a lockable buffer for
468 * a given range of an inode. The buffer is returned
469 * locked. If other overlapping buffers exist, they are
470 * released before the new buffer is created and locked,
471 * which may imply that this call will block until those buffers
472 * are unlocked. No I/O is implied by this call.
476 xfs_buftarg_t
*btp
, /* block device target */
477 xfs_off_t ioff
, /* starting offset of range */
478 size_t isize
, /* length of range */
479 xfs_buf_flags_t flags
,
482 xfs_off_t range_base
;
487 range_base
= (ioff
<< BBSHIFT
);
488 range_length
= (isize
<< BBSHIFT
);
490 /* Check for IOs smaller than the sector size / not sector aligned */
491 ASSERT(!(range_length
< (1 << btp
->bt_sshift
)));
492 ASSERT(!(range_base
& (xfs_off_t
)btp
->bt_smask
));
494 hash
= &btp
->bt_hash
[hash_long((unsigned long)ioff
, btp
->bt_hashshift
)];
496 spin_lock(&hash
->bh_lock
);
498 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
499 ASSERT(btp
== bp
->b_target
);
500 if (bp
->b_file_offset
== range_base
&&
501 bp
->b_buffer_length
== range_length
) {
503 * If we look at something, bring it to the
504 * front of the list for next time.
506 atomic_inc(&bp
->b_hold
);
507 list_move(&bp
->b_hash_list
, &hash
->bh_list
);
514 _xfs_buf_initialize(new_bp
, btp
, range_base
,
515 range_length
, flags
);
516 new_bp
->b_hash
= hash
;
517 list_add(&new_bp
->b_hash_list
, &hash
->bh_list
);
519 XFS_STATS_INC(xb_miss_locked
);
522 spin_unlock(&hash
->bh_lock
);
526 spin_unlock(&hash
->bh_lock
);
528 /* Attempt to get the semaphore without sleeping,
529 * if this does not work then we need to drop the
530 * spinlock and do a hard attempt on the semaphore.
532 if (down_trylock(&bp
->b_sema
)) {
533 if (!(flags
& XBF_TRYLOCK
)) {
534 /* wait for buffer ownership */
535 XB_TRACE(bp
, "get_lock", 0);
537 XFS_STATS_INC(xb_get_locked_waited
);
539 /* We asked for a trylock and failed, no need
540 * to look at file offset and length here, we
541 * know that this buffer at least overlaps our
542 * buffer and is locked, therefore our buffer
543 * either does not exist, or is this buffer.
546 XFS_STATS_INC(xb_busy_locked
);
554 if (bp
->b_flags
& XBF_STALE
) {
555 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
556 bp
->b_flags
&= XBF_MAPPED
;
558 XB_TRACE(bp
, "got_lock", 0);
559 XFS_STATS_INC(xb_get_locked
);
564 * Assembles a buffer covering the specified range.
565 * Storage in memory for all portions of the buffer will be allocated,
566 * although backing storage may not be.
570 xfs_buftarg_t
*target
,/* target for buffer */
571 xfs_off_t ioff
, /* starting offset of range */
572 size_t isize
, /* length of range */
573 xfs_buf_flags_t flags
)
575 xfs_buf_t
*bp
, *new_bp
;
578 new_bp
= xfs_buf_allocate(flags
);
579 if (unlikely(!new_bp
))
582 bp
= _xfs_buf_find(target
, ioff
, isize
, flags
, new_bp
);
584 error
= _xfs_buf_lookup_pages(bp
, flags
);
588 xfs_buf_deallocate(new_bp
);
589 if (unlikely(bp
== NULL
))
593 for (i
= 0; i
< bp
->b_page_count
; i
++)
594 mark_page_accessed(bp
->b_pages
[i
]);
596 if (!(bp
->b_flags
& XBF_MAPPED
)) {
597 error
= _xfs_buf_map_pages(bp
, flags
);
598 if (unlikely(error
)) {
599 printk(KERN_WARNING
"%s: failed to map pages\n",
605 XFS_STATS_INC(xb_get
);
608 * Always fill in the block number now, the mapped cases can do
609 * their own overlay of this later.
612 bp
->b_count_desired
= bp
->b_buffer_length
;
614 XB_TRACE(bp
, "get", (unsigned long)flags
);
618 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
626 xfs_buftarg_t
*target
,
629 xfs_buf_flags_t flags
)
635 bp
= xfs_buf_get_flags(target
, ioff
, isize
, flags
);
637 if (!XFS_BUF_ISDONE(bp
)) {
638 XB_TRACE(bp
, "read", (unsigned long)flags
);
639 XFS_STATS_INC(xb_get_read
);
640 xfs_buf_iostart(bp
, flags
);
641 } else if (flags
& XBF_ASYNC
) {
642 XB_TRACE(bp
, "read_async", (unsigned long)flags
);
644 * Read ahead call which is already satisfied,
649 XB_TRACE(bp
, "read_done", (unsigned long)flags
);
650 /* We do not want read in the flags */
651 bp
->b_flags
&= ~XBF_READ
;
658 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
665 * If we are not low on memory then do the readahead in a deadlock
670 xfs_buftarg_t
*target
,
673 xfs_buf_flags_t flags
)
675 struct backing_dev_info
*bdi
;
677 bdi
= target
->bt_mapping
->backing_dev_info
;
678 if (bdi_read_congested(bdi
))
681 flags
|= (XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
);
682 xfs_buf_read_flags(target
, ioff
, isize
, flags
);
688 xfs_buftarg_t
*target
)
692 bp
= xfs_buf_allocate(0);
694 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
698 static inline struct page
*
702 if (((unsigned long)addr
< VMALLOC_START
) ||
703 ((unsigned long)addr
>= VMALLOC_END
)) {
704 return virt_to_page(addr
);
706 return vmalloc_to_page(addr
);
711 xfs_buf_associate_memory(
723 page_count
= PAGE_CACHE_ALIGN(len
) >> PAGE_CACHE_SHIFT
;
724 offset
= (off_t
) mem
- ((off_t
)mem
& PAGE_CACHE_MASK
);
725 if (offset
&& (len
> PAGE_CACHE_SIZE
))
728 /* Free any previous set of page pointers */
730 _xfs_buf_free_pages(bp
);
735 rval
= _xfs_buf_get_pages(bp
, page_count
, 0);
739 bp
->b_offset
= offset
;
740 ptr
= (size_t) mem
& PAGE_CACHE_MASK
;
741 end
= PAGE_CACHE_ALIGN((size_t) mem
+ len
);
743 /* set up first page */
744 bp
->b_pages
[0] = mem_to_page(mem
);
746 ptr
+= PAGE_CACHE_SIZE
;
747 bp
->b_page_count
= ++i
;
749 bp
->b_pages
[i
] = mem_to_page((void *)ptr
);
750 bp
->b_page_count
= ++i
;
751 ptr
+= PAGE_CACHE_SIZE
;
755 bp
->b_count_desired
= bp
->b_buffer_length
= len
;
756 bp
->b_flags
|= XBF_MAPPED
;
764 xfs_buftarg_t
*target
)
766 size_t malloc_len
= len
;
771 bp
= xfs_buf_allocate(0);
772 if (unlikely(bp
== NULL
))
774 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
777 data
= kmem_alloc(malloc_len
, KM_SLEEP
| KM_MAYFAIL
| KM_LARGE
);
778 if (unlikely(data
== NULL
))
781 /* check whether alignment matches.. */
782 if ((__psunsigned_t
)data
!=
783 ((__psunsigned_t
)data
& ~target
->bt_smask
)) {
784 /* .. else double the size and try again */
785 kmem_free(data
, malloc_len
);
790 error
= xfs_buf_associate_memory(bp
, data
, len
);
793 bp
->b_flags
|= _XBF_KMEM_ALLOC
;
797 XB_TRACE(bp
, "no_daddr", data
);
800 kmem_free(data
, malloc_len
);
808 * Increment reference count on buffer, to hold the buffer concurrently
809 * with another thread which may release (free) the buffer asynchronously.
810 * Must hold the buffer already to call this function.
816 atomic_inc(&bp
->b_hold
);
817 XB_TRACE(bp
, "hold", 0);
821 * Releases a hold on the specified buffer. If the
822 * the hold count is 1, calls xfs_buf_free.
828 xfs_bufhash_t
*hash
= bp
->b_hash
;
830 XB_TRACE(bp
, "rele", bp
->b_relse
);
832 if (unlikely(!hash
)) {
833 ASSERT(!bp
->b_relse
);
834 if (atomic_dec_and_test(&bp
->b_hold
))
839 if (atomic_dec_and_lock(&bp
->b_hold
, &hash
->bh_lock
)) {
841 atomic_inc(&bp
->b_hold
);
842 spin_unlock(&hash
->bh_lock
);
843 (*(bp
->b_relse
)) (bp
);
844 } else if (bp
->b_flags
& XBF_FS_MANAGED
) {
845 spin_unlock(&hash
->bh_lock
);
847 ASSERT(!(bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)));
848 list_del_init(&bp
->b_hash_list
);
849 spin_unlock(&hash
->bh_lock
);
854 * Catch reference count leaks
856 ASSERT(atomic_read(&bp
->b_hold
) >= 0);
862 * Mutual exclusion on buffers. Locking model:
864 * Buffers associated with inodes for which buffer locking
865 * is not enabled are not protected by semaphores, and are
866 * assumed to be exclusively owned by the caller. There is a
867 * spinlock in the buffer, used by the caller when concurrent
868 * access is possible.
872 * Locks a buffer object, if it is not already locked.
873 * Note that this in no way locks the underlying pages, so it is only
874 * useful for synchronizing concurrent use of buffer objects, not for
875 * synchronizing independent access to the underlying pages.
883 locked
= down_trylock(&bp
->b_sema
) == 0;
887 XB_TRACE(bp
, "cond_lock", (long)locked
);
888 return locked
? 0 : -EBUSY
;
891 #if defined(DEBUG) || defined(XFS_BLI_TRACE)
896 return atomic_read(&bp
->b_sema
.count
);
901 * Locks a buffer object.
902 * Note that this in no way locks the underlying pages, so it is only
903 * useful for synchronizing concurrent use of buffer objects, not for
904 * synchronizing independent access to the underlying pages.
910 XB_TRACE(bp
, "lock", 0);
911 if (atomic_read(&bp
->b_io_remaining
))
912 blk_run_address_space(bp
->b_target
->bt_mapping
);
915 XB_TRACE(bp
, "locked", 0);
919 * Releases the lock on the buffer object.
920 * If the buffer is marked delwri but is not queued, do so before we
921 * unlock the buffer as we need to set flags correctly. We also need to
922 * take a reference for the delwri queue because the unlocker is going to
923 * drop their's and they don't know we just queued it.
929 if ((bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)) == XBF_DELWRI
) {
930 atomic_inc(&bp
->b_hold
);
931 bp
->b_flags
|= XBF_ASYNC
;
932 xfs_buf_delwri_queue(bp
, 0);
937 XB_TRACE(bp
, "unlock", 0);
942 * Pinning Buffer Storage in Memory
943 * Ensure that no attempt to force a buffer to disk will succeed.
949 atomic_inc(&bp
->b_pin_count
);
950 XB_TRACE(bp
, "pin", (long)bp
->b_pin_count
.counter
);
957 if (atomic_dec_and_test(&bp
->b_pin_count
))
958 wake_up_all(&bp
->b_waiters
);
959 XB_TRACE(bp
, "unpin", (long)bp
->b_pin_count
.counter
);
966 return atomic_read(&bp
->b_pin_count
);
973 DECLARE_WAITQUEUE (wait
, current
);
975 if (atomic_read(&bp
->b_pin_count
) == 0)
978 add_wait_queue(&bp
->b_waiters
, &wait
);
980 set_current_state(TASK_UNINTERRUPTIBLE
);
981 if (atomic_read(&bp
->b_pin_count
) == 0)
983 if (atomic_read(&bp
->b_io_remaining
))
984 blk_run_address_space(bp
->b_target
->bt_mapping
);
987 remove_wait_queue(&bp
->b_waiters
, &wait
);
988 set_current_state(TASK_RUNNING
);
992 * Buffer Utility Routines
997 struct work_struct
*work
)
1000 container_of(work
, xfs_buf_t
, b_iodone_work
);
1003 (*(bp
->b_iodone
))(bp
);
1004 else if (bp
->b_flags
& XBF_ASYNC
)
1013 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
);
1014 if (bp
->b_error
== 0)
1015 bp
->b_flags
|= XBF_DONE
;
1017 XB_TRACE(bp
, "iodone", bp
->b_iodone
);
1019 if ((bp
->b_iodone
) || (bp
->b_flags
& XBF_ASYNC
)) {
1021 INIT_WORK(&bp
->b_iodone_work
, xfs_buf_iodone_work
);
1022 queue_work(xfslogd_workqueue
, &bp
->b_iodone_work
);
1024 xfs_buf_iodone_work(&bp
->b_iodone_work
);
1027 up(&bp
->b_iodonesema
);
1036 ASSERT(error
>= 0 && error
<= 0xffff);
1037 bp
->b_error
= (unsigned short)error
;
1038 XB_TRACE(bp
, "ioerror", (unsigned long)error
);
1042 * Initiate I/O on a buffer, based on the flags supplied.
1043 * The b_iodone routine in the buffer supplied will only be called
1044 * when all of the subsidiary I/O requests, if any, have been completed.
1049 xfs_buf_flags_t flags
)
1053 XB_TRACE(bp
, "iostart", (unsigned long)flags
);
1055 if (flags
& XBF_DELWRI
) {
1056 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_ASYNC
);
1057 bp
->b_flags
|= flags
& (XBF_DELWRI
| XBF_ASYNC
);
1058 xfs_buf_delwri_queue(bp
, 1);
1062 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_ASYNC
| XBF_DELWRI
| \
1063 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
1064 bp
->b_flags
|= flags
& (XBF_READ
| XBF_WRITE
| XBF_ASYNC
| \
1065 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
1067 BUG_ON(bp
->b_bn
== XFS_BUF_DADDR_NULL
);
1069 /* For writes allow an alternate strategy routine to precede
1070 * the actual I/O request (which may not be issued at all in
1071 * a shutdown situation, for example).
1073 status
= (flags
& XBF_WRITE
) ?
1074 xfs_buf_iostrategy(bp
) : xfs_buf_iorequest(bp
);
1076 /* Wait for I/O if we are not an async request.
1077 * Note: async I/O request completion will release the buffer,
1078 * and that can already be done by this point. So using the
1079 * buffer pointer from here on, after async I/O, is invalid.
1081 if (!status
&& !(flags
& XBF_ASYNC
))
1082 status
= xfs_buf_iowait(bp
);
1087 STATIC __inline__
int
1091 ASSERT(bp
->b_flags
& (XBF_READ
| XBF_WRITE
));
1092 if (bp
->b_flags
& XBF_READ
)
1093 return bp
->b_locked
;
1097 STATIC __inline__
void
1102 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1104 xfs_buf_ioend(bp
, schedule
);
1111 unsigned int bytes_done
,
1114 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1115 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1116 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1121 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1125 struct page
*page
= bvec
->bv_page
;
1127 ASSERT(!PagePrivate(page
));
1128 if (unlikely(bp
->b_error
)) {
1129 if (bp
->b_flags
& XBF_READ
)
1130 ClearPageUptodate(page
);
1131 } else if (blocksize
>= PAGE_CACHE_SIZE
) {
1132 SetPageUptodate(page
);
1133 } else if (!PagePrivate(page
) &&
1134 (bp
->b_flags
& _XBF_PAGE_CACHE
)) {
1135 set_page_region(page
, bvec
->bv_offset
, bvec
->bv_len
);
1138 if (--bvec
>= bio
->bi_io_vec
)
1139 prefetchw(&bvec
->bv_page
->flags
);
1141 if (_xfs_buf_iolocked(bp
)) {
1144 } while (bvec
>= bio
->bi_io_vec
);
1146 _xfs_buf_ioend(bp
, 1);
1155 int i
, rw
, map_i
, total_nr_pages
, nr_pages
;
1157 int offset
= bp
->b_offset
;
1158 int size
= bp
->b_count_desired
;
1159 sector_t sector
= bp
->b_bn
;
1160 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1161 int locking
= _xfs_buf_iolocked(bp
);
1163 total_nr_pages
= bp
->b_page_count
;
1166 if (bp
->b_flags
& XBF_ORDERED
) {
1167 ASSERT(!(bp
->b_flags
& XBF_READ
));
1169 } else if (bp
->b_flags
& _XBF_RUN_QUEUES
) {
1170 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1171 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1172 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_SYNC
: READ_SYNC
;
1174 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE
:
1175 (bp
->b_flags
& XBF_READ_AHEAD
) ? READA
: READ
;
1178 /* Special code path for reading a sub page size buffer in --
1179 * we populate up the whole page, and hence the other metadata
1180 * in the same page. This optimization is only valid when the
1181 * filesystem block size is not smaller than the page size.
1183 if ((bp
->b_buffer_length
< PAGE_CACHE_SIZE
) &&
1184 (bp
->b_flags
& XBF_READ
) && locking
&&
1185 (blocksize
>= PAGE_CACHE_SIZE
)) {
1186 bio
= bio_alloc(GFP_NOIO
, 1);
1188 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1189 bio
->bi_sector
= sector
- (offset
>> BBSHIFT
);
1190 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1191 bio
->bi_private
= bp
;
1193 bio_add_page(bio
, bp
->b_pages
[0], PAGE_CACHE_SIZE
, 0);
1196 atomic_inc(&bp
->b_io_remaining
);
1201 /* Lock down the pages which we need to for the request */
1202 if (locking
&& (bp
->b_flags
& XBF_WRITE
) && (bp
->b_locked
== 0)) {
1203 for (i
= 0; size
; i
++) {
1204 int nbytes
= PAGE_CACHE_SIZE
- offset
;
1205 struct page
*page
= bp
->b_pages
[i
];
1215 offset
= bp
->b_offset
;
1216 size
= bp
->b_count_desired
;
1220 atomic_inc(&bp
->b_io_remaining
);
1221 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1222 if (nr_pages
> total_nr_pages
)
1223 nr_pages
= total_nr_pages
;
1225 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1226 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1227 bio
->bi_sector
= sector
;
1228 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1229 bio
->bi_private
= bp
;
1231 for (; size
&& nr_pages
; nr_pages
--, map_i
++) {
1232 int rbytes
, nbytes
= PAGE_CACHE_SIZE
- offset
;
1237 rbytes
= bio_add_page(bio
, bp
->b_pages
[map_i
], nbytes
, offset
);
1238 if (rbytes
< nbytes
)
1242 sector
+= nbytes
>> BBSHIFT
;
1248 if (likely(bio
->bi_size
)) {
1249 submit_bio(rw
, bio
);
1254 xfs_buf_ioerror(bp
, EIO
);
1262 XB_TRACE(bp
, "iorequest", 0);
1264 if (bp
->b_flags
& XBF_DELWRI
) {
1265 xfs_buf_delwri_queue(bp
, 1);
1269 if (bp
->b_flags
& XBF_WRITE
) {
1270 xfs_buf_wait_unpin(bp
);
1275 /* Set the count to 1 initially, this will stop an I/O
1276 * completion callout which happens before we have started
1277 * all the I/O from calling xfs_buf_ioend too early.
1279 atomic_set(&bp
->b_io_remaining
, 1);
1280 _xfs_buf_ioapply(bp
);
1281 _xfs_buf_ioend(bp
, 0);
1288 * Waits for I/O to complete on the buffer supplied.
1289 * It returns immediately if no I/O is pending.
1290 * It returns the I/O error code, if any, or 0 if there was no error.
1296 XB_TRACE(bp
, "iowait", 0);
1297 if (atomic_read(&bp
->b_io_remaining
))
1298 blk_run_address_space(bp
->b_target
->bt_mapping
);
1299 down(&bp
->b_iodonesema
);
1300 XB_TRACE(bp
, "iowaited", (long)bp
->b_error
);
1311 if (bp
->b_flags
& XBF_MAPPED
)
1312 return XFS_BUF_PTR(bp
) + offset
;
1314 offset
+= bp
->b_offset
;
1315 page
= bp
->b_pages
[offset
>> PAGE_CACHE_SHIFT
];
1316 return (xfs_caddr_t
)page_address(page
) + (offset
& (PAGE_CACHE_SIZE
-1));
1320 * Move data into or out of a buffer.
1324 xfs_buf_t
*bp
, /* buffer to process */
1325 size_t boff
, /* starting buffer offset */
1326 size_t bsize
, /* length to copy */
1327 caddr_t data
, /* data address */
1328 xfs_buf_rw_t mode
) /* read/write/zero flag */
1330 size_t bend
, cpoff
, csize
;
1333 bend
= boff
+ bsize
;
1334 while (boff
< bend
) {
1335 page
= bp
->b_pages
[xfs_buf_btoct(boff
+ bp
->b_offset
)];
1336 cpoff
= xfs_buf_poff(boff
+ bp
->b_offset
);
1337 csize
= min_t(size_t,
1338 PAGE_CACHE_SIZE
-cpoff
, bp
->b_count_desired
-boff
);
1340 ASSERT(((csize
+ cpoff
) <= PAGE_CACHE_SIZE
));
1344 memset(page_address(page
) + cpoff
, 0, csize
);
1347 memcpy(data
, page_address(page
) + cpoff
, csize
);
1350 memcpy(page_address(page
) + cpoff
, data
, csize
);
1359 * Handling of buffer targets (buftargs).
1363 * Wait for any bufs with callbacks that have been submitted but
1364 * have not yet returned... walk the hash list for the target.
1371 xfs_bufhash_t
*hash
;
1374 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1375 hash
= &btp
->bt_hash
[i
];
1377 spin_lock(&hash
->bh_lock
);
1378 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
1379 ASSERT(btp
== bp
->b_target
);
1380 if (!(bp
->b_flags
& XBF_FS_MANAGED
)) {
1381 spin_unlock(&hash
->bh_lock
);
1383 * Catch superblock reference count leaks
1386 BUG_ON(bp
->b_bn
== 0);
1391 spin_unlock(&hash
->bh_lock
);
1396 * Allocate buffer hash table for a given target.
1397 * For devices containing metadata (i.e. not the log/realtime devices)
1398 * we need to allocate a much larger hash table.
1407 btp
->bt_hashshift
= external
? 3 : 8; /* 8 or 256 buckets */
1408 btp
->bt_hashmask
= (1 << btp
->bt_hashshift
) - 1;
1409 btp
->bt_hash
= kmem_zalloc((1 << btp
->bt_hashshift
) *
1410 sizeof(xfs_bufhash_t
), KM_SLEEP
| KM_LARGE
);
1411 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1412 spin_lock_init(&btp
->bt_hash
[i
].bh_lock
);
1413 INIT_LIST_HEAD(&btp
->bt_hash
[i
].bh_list
);
1421 kmem_free(btp
->bt_hash
, (1<<btp
->bt_hashshift
) * sizeof(xfs_bufhash_t
));
1422 btp
->bt_hash
= NULL
;
1426 * buftarg list for delwrite queue processing
1428 STATIC
LIST_HEAD(xfs_buftarg_list
);
1429 STATIC
DEFINE_SPINLOCK(xfs_buftarg_lock
);
1432 xfs_register_buftarg(
1435 spin_lock(&xfs_buftarg_lock
);
1436 list_add(&btp
->bt_list
, &xfs_buftarg_list
);
1437 spin_unlock(&xfs_buftarg_lock
);
1441 xfs_unregister_buftarg(
1444 spin_lock(&xfs_buftarg_lock
);
1445 list_del(&btp
->bt_list
);
1446 spin_unlock(&xfs_buftarg_lock
);
1454 xfs_flush_buftarg(btp
, 1);
1456 xfs_blkdev_put(btp
->bt_bdev
);
1457 xfs_free_bufhash(btp
);
1458 iput(btp
->bt_mapping
->host
);
1460 /* Unregister the buftarg first so that we don't get a
1461 * wakeup finding a non-existent task
1463 xfs_unregister_buftarg(btp
);
1464 kthread_stop(btp
->bt_task
);
1466 kmem_free(btp
, sizeof(*btp
));
1470 xfs_setsize_buftarg_flags(
1472 unsigned int blocksize
,
1473 unsigned int sectorsize
,
1476 btp
->bt_bsize
= blocksize
;
1477 btp
->bt_sshift
= ffs(sectorsize
) - 1;
1478 btp
->bt_smask
= sectorsize
- 1;
1480 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1482 "XFS: Cannot set_blocksize to %u on device %s\n",
1483 sectorsize
, XFS_BUFTARG_NAME(btp
));
1488 (PAGE_CACHE_SIZE
/ BITS_PER_LONG
) > sectorsize
) {
1490 "XFS: %u byte sectors in use on device %s. "
1491 "This is suboptimal; %u or greater is ideal.\n",
1492 sectorsize
, XFS_BUFTARG_NAME(btp
),
1493 (unsigned int)PAGE_CACHE_SIZE
/ BITS_PER_LONG
);
1500 * When allocating the initial buffer target we have not yet
1501 * read in the superblock, so don't know what sized sectors
1502 * are being used is at this early stage. Play safe.
1505 xfs_setsize_buftarg_early(
1507 struct block_device
*bdev
)
1509 return xfs_setsize_buftarg_flags(btp
,
1510 PAGE_CACHE_SIZE
, bdev_hardsect_size(bdev
), 0);
1514 xfs_setsize_buftarg(
1516 unsigned int blocksize
,
1517 unsigned int sectorsize
)
1519 return xfs_setsize_buftarg_flags(btp
, blocksize
, sectorsize
, 1);
1523 xfs_mapping_buftarg(
1525 struct block_device
*bdev
)
1527 struct backing_dev_info
*bdi
;
1528 struct inode
*inode
;
1529 struct address_space
*mapping
;
1530 static const struct address_space_operations mapping_aops
= {
1531 .sync_page
= block_sync_page
,
1532 .migratepage
= fail_migrate_page
,
1535 inode
= new_inode(bdev
->bd_inode
->i_sb
);
1538 "XFS: Cannot allocate mapping inode for device %s\n",
1539 XFS_BUFTARG_NAME(btp
));
1542 inode
->i_mode
= S_IFBLK
;
1543 inode
->i_bdev
= bdev
;
1544 inode
->i_rdev
= bdev
->bd_dev
;
1545 bdi
= blk_get_backing_dev_info(bdev
);
1547 bdi
= &default_backing_dev_info
;
1548 mapping
= &inode
->i_data
;
1549 mapping
->a_ops
= &mapping_aops
;
1550 mapping
->backing_dev_info
= bdi
;
1551 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
1552 btp
->bt_mapping
= mapping
;
1557 xfs_alloc_delwrite_queue(
1562 INIT_LIST_HEAD(&btp
->bt_list
);
1563 INIT_LIST_HEAD(&btp
->bt_delwrite_queue
);
1564 spinlock_init(&btp
->bt_delwrite_lock
, "delwri_lock");
1566 btp
->bt_task
= kthread_run(xfsbufd
, btp
, "xfsbufd");
1567 if (IS_ERR(btp
->bt_task
)) {
1568 error
= PTR_ERR(btp
->bt_task
);
1571 xfs_register_buftarg(btp
);
1578 struct block_device
*bdev
,
1583 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
);
1585 btp
->bt_dev
= bdev
->bd_dev
;
1586 btp
->bt_bdev
= bdev
;
1587 if (xfs_setsize_buftarg_early(btp
, bdev
))
1589 if (xfs_mapping_buftarg(btp
, bdev
))
1591 if (xfs_alloc_delwrite_queue(btp
))
1593 xfs_alloc_bufhash(btp
, external
);
1597 kmem_free(btp
, sizeof(*btp
));
1603 * Delayed write buffer handling
1606 xfs_buf_delwri_queue(
1610 struct list_head
*dwq
= &bp
->b_target
->bt_delwrite_queue
;
1611 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1613 XB_TRACE(bp
, "delwri_q", (long)unlock
);
1614 ASSERT((bp
->b_flags
&(XBF_DELWRI
|XBF_ASYNC
)) == (XBF_DELWRI
|XBF_ASYNC
));
1617 /* If already in the queue, dequeue and place at tail */
1618 if (!list_empty(&bp
->b_list
)) {
1619 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1621 atomic_dec(&bp
->b_hold
);
1622 list_del(&bp
->b_list
);
1625 bp
->b_flags
|= _XBF_DELWRI_Q
;
1626 list_add_tail(&bp
->b_list
, dwq
);
1627 bp
->b_queuetime
= jiffies
;
1635 xfs_buf_delwri_dequeue(
1638 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1642 if ((bp
->b_flags
& XBF_DELWRI
) && !list_empty(&bp
->b_list
)) {
1643 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1644 list_del_init(&bp
->b_list
);
1647 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
);
1653 XB_TRACE(bp
, "delwri_dq", (long)dequeued
);
1657 xfs_buf_runall_queues(
1658 struct workqueue_struct
*queue
)
1660 flush_workqueue(queue
);
1670 spin_lock(&xfs_buftarg_lock
);
1671 list_for_each_entry(btp
, &xfs_buftarg_list
, bt_list
) {
1672 if (test_bit(XBT_FORCE_SLEEP
, &btp
->bt_flags
))
1674 set_bit(XBT_FORCE_FLUSH
, &btp
->bt_flags
);
1675 wake_up_process(btp
->bt_task
);
1677 spin_unlock(&xfs_buftarg_lock
);
1685 struct list_head tmp
;
1687 xfs_buftarg_t
*target
= (xfs_buftarg_t
*)data
;
1689 struct list_head
*dwq
= &target
->bt_delwrite_queue
;
1690 spinlock_t
*dwlk
= &target
->bt_delwrite_lock
;
1693 current
->flags
|= PF_MEMALLOC
;
1695 INIT_LIST_HEAD(&tmp
);
1697 if (unlikely(freezing(current
))) {
1698 set_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1701 clear_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1704 schedule_timeout_interruptible(
1705 xfs_buf_timer_centisecs
* msecs_to_jiffies(10));
1708 age
= xfs_buf_age_centisecs
* msecs_to_jiffies(10);
1710 list_for_each_entry_safe(bp
, n
, dwq
, b_list
) {
1711 XB_TRACE(bp
, "walkq1", (long)xfs_buf_ispin(bp
));
1712 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1714 if (!xfs_buf_ispin(bp
) && !xfs_buf_cond_lock(bp
)) {
1715 if (!test_bit(XBT_FORCE_FLUSH
,
1716 &target
->bt_flags
) &&
1717 time_before(jiffies
,
1718 bp
->b_queuetime
+ age
)) {
1723 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
|
1725 bp
->b_flags
|= XBF_WRITE
;
1726 list_move_tail(&bp
->b_list
, &tmp
);
1732 while (!list_empty(&tmp
)) {
1733 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1734 ASSERT(target
== bp
->b_target
);
1736 list_del_init(&bp
->b_list
);
1737 xfs_buf_iostrategy(bp
);
1740 if (as_list_len
> 0)
1743 blk_run_address_space(target
->bt_mapping
);
1745 clear_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1746 } while (!kthread_should_stop());
1752 * Go through all incore buffers, and release buffers if they belong to
1753 * the given device. This is used in filesystem error handling to
1754 * preserve the consistency of its metadata.
1758 xfs_buftarg_t
*target
,
1761 struct list_head tmp
;
1764 struct list_head
*dwq
= &target
->bt_delwrite_queue
;
1765 spinlock_t
*dwlk
= &target
->bt_delwrite_lock
;
1767 xfs_buf_runall_queues(xfsdatad_workqueue
);
1768 xfs_buf_runall_queues(xfslogd_workqueue
);
1770 INIT_LIST_HEAD(&tmp
);
1772 list_for_each_entry_safe(bp
, n
, dwq
, b_list
) {
1773 ASSERT(bp
->b_target
== target
);
1774 ASSERT(bp
->b_flags
& (XBF_DELWRI
| _XBF_DELWRI_Q
));
1775 XB_TRACE(bp
, "walkq2", (long)xfs_buf_ispin(bp
));
1776 if (xfs_buf_ispin(bp
)) {
1781 list_move_tail(&bp
->b_list
, &tmp
);
1786 * Dropped the delayed write list lock, now walk the temporary list
1788 list_for_each_entry_safe(bp
, n
, &tmp
, b_list
) {
1790 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
|_XBF_RUN_QUEUES
);
1791 bp
->b_flags
|= XBF_WRITE
;
1793 bp
->b_flags
&= ~XBF_ASYNC
;
1795 list_del_init(&bp
->b_list
);
1797 xfs_buf_iostrategy(bp
);
1801 blk_run_address_space(target
->bt_mapping
);
1804 * Remaining list items must be flushed before returning
1806 while (!list_empty(&tmp
)) {
1807 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1809 list_del_init(&bp
->b_list
);
1820 #ifdef XFS_BUF_TRACE
1821 xfs_buf_trace_buf
= ktrace_alloc(XFS_BUF_TRACE_SIZE
, KM_SLEEP
);
1824 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1825 KM_ZONE_HWALIGN
, NULL
);
1827 goto out_free_trace_buf
;
1829 xfslogd_workqueue
= create_workqueue("xfslogd");
1830 if (!xfslogd_workqueue
)
1831 goto out_free_buf_zone
;
1833 xfsdatad_workqueue
= create_workqueue("xfsdatad");
1834 if (!xfsdatad_workqueue
)
1835 goto out_destroy_xfslogd_workqueue
;
1837 xfs_buf_shake
= kmem_shake_register(xfsbufd_wakeup
);
1839 goto out_destroy_xfsdatad_workqueue
;
1843 out_destroy_xfsdatad_workqueue
:
1844 destroy_workqueue(xfsdatad_workqueue
);
1845 out_destroy_xfslogd_workqueue
:
1846 destroy_workqueue(xfslogd_workqueue
);
1848 kmem_zone_destroy(xfs_buf_zone
);
1850 #ifdef XFS_BUF_TRACE
1851 ktrace_free(xfs_buf_trace_buf
);
1857 xfs_buf_terminate(void)
1859 kmem_shake_deregister(xfs_buf_shake
);
1860 destroy_workqueue(xfsdatad_workqueue
);
1861 destroy_workqueue(xfslogd_workqueue
);
1862 kmem_zone_destroy(xfs_buf_zone
);
1863 #ifdef XFS_BUF_TRACE
1864 ktrace_free(xfs_buf_trace_buf
);