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
.counter
,
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_MUTEX_LOCKED(&bp
->b_iodonesema
);
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
,
314 bp
->b_page_count
* sizeof(struct page
*));
319 * Releases the specified buffer.
321 * The modification state of any associated pages is left unchanged.
322 * The buffer most not be on any hash - use xfs_buf_rele instead for
323 * hashed and refcounted buffers
329 XB_TRACE(bp
, "free", 0);
331 ASSERT(list_empty(&bp
->b_hash_list
));
333 if (bp
->b_flags
& (_XBF_PAGE_CACHE
|_XBF_PAGES
)) {
336 if ((bp
->b_flags
& XBF_MAPPED
) && (bp
->b_page_count
> 1))
337 free_address(bp
->b_addr
- bp
->b_offset
);
339 for (i
= 0; i
< bp
->b_page_count
; i
++) {
340 struct page
*page
= bp
->b_pages
[i
];
342 if (bp
->b_flags
& _XBF_PAGE_CACHE
)
343 ASSERT(!PagePrivate(page
));
344 page_cache_release(page
);
346 _xfs_buf_free_pages(bp
);
349 xfs_buf_deallocate(bp
);
353 * Finds all pages for buffer in question and builds it's page list.
356 _xfs_buf_lookup_pages(
360 struct address_space
*mapping
= bp
->b_target
->bt_mapping
;
361 size_t blocksize
= bp
->b_target
->bt_bsize
;
362 size_t size
= bp
->b_count_desired
;
363 size_t nbytes
, offset
;
364 gfp_t gfp_mask
= xb_to_gfp(flags
);
365 unsigned short page_count
, i
;
370 end
= bp
->b_file_offset
+ bp
->b_buffer_length
;
371 page_count
= xfs_buf_btoc(end
) - xfs_buf_btoct(bp
->b_file_offset
);
373 error
= _xfs_buf_get_pages(bp
, page_count
, flags
);
376 bp
->b_flags
|= _XBF_PAGE_CACHE
;
378 offset
= bp
->b_offset
;
379 first
= bp
->b_file_offset
>> PAGE_CACHE_SHIFT
;
381 for (i
= 0; i
< bp
->b_page_count
; i
++) {
386 page
= find_or_create_page(mapping
, first
+ i
, gfp_mask
);
387 if (unlikely(page
== NULL
)) {
388 if (flags
& XBF_READ_AHEAD
) {
389 bp
->b_page_count
= i
;
390 for (i
= 0; i
< bp
->b_page_count
; i
++)
391 unlock_page(bp
->b_pages
[i
]);
396 * This could deadlock.
398 * But until all the XFS lowlevel code is revamped to
399 * handle buffer allocation failures we can't do much.
401 if (!(++retries
% 100))
403 "XFS: possible memory allocation "
404 "deadlock in %s (mode:0x%x)\n",
405 __FUNCTION__
, gfp_mask
);
407 XFS_STATS_INC(xb_page_retries
);
408 xfsbufd_wakeup(0, gfp_mask
);
409 congestion_wait(WRITE
, HZ
/50);
413 XFS_STATS_INC(xb_page_found
);
415 nbytes
= min_t(size_t, size
, PAGE_CACHE_SIZE
- offset
);
418 ASSERT(!PagePrivate(page
));
419 if (!PageUptodate(page
)) {
421 if (blocksize
>= PAGE_CACHE_SIZE
) {
422 if (flags
& XBF_READ
)
424 } else if (!PagePrivate(page
)) {
425 if (test_page_region(page
, offset
, nbytes
))
430 bp
->b_pages
[i
] = page
;
435 for (i
= 0; i
< bp
->b_page_count
; i
++)
436 unlock_page(bp
->b_pages
[i
]);
439 if (page_count
== bp
->b_page_count
)
440 bp
->b_flags
|= XBF_DONE
;
442 XB_TRACE(bp
, "lookup_pages", (long)page_count
);
447 * Map buffer into kernel address-space if nessecary.
454 /* A single page buffer is always mappable */
455 if (bp
->b_page_count
== 1) {
456 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
457 bp
->b_flags
|= XBF_MAPPED
;
458 } else if (flags
& XBF_MAPPED
) {
459 if (as_list_len
> 64)
461 bp
->b_addr
= vmap(bp
->b_pages
, bp
->b_page_count
,
462 VM_MAP
, PAGE_KERNEL
);
463 if (unlikely(bp
->b_addr
== NULL
))
465 bp
->b_addr
+= bp
->b_offset
;
466 bp
->b_flags
|= XBF_MAPPED
;
473 * Finding and Reading Buffers
477 * Look up, and creates if absent, a lockable buffer for
478 * a given range of an inode. The buffer is returned
479 * locked. If other overlapping buffers exist, they are
480 * released before the new buffer is created and locked,
481 * which may imply that this call will block until those buffers
482 * are unlocked. No I/O is implied by this call.
486 xfs_buftarg_t
*btp
, /* block device target */
487 xfs_off_t ioff
, /* starting offset of range */
488 size_t isize
, /* length of range */
489 xfs_buf_flags_t flags
,
492 xfs_off_t range_base
;
497 range_base
= (ioff
<< BBSHIFT
);
498 range_length
= (isize
<< BBSHIFT
);
500 /* Check for IOs smaller than the sector size / not sector aligned */
501 ASSERT(!(range_length
< (1 << btp
->bt_sshift
)));
502 ASSERT(!(range_base
& (xfs_off_t
)btp
->bt_smask
));
504 hash
= &btp
->bt_hash
[hash_long((unsigned long)ioff
, btp
->bt_hashshift
)];
506 spin_lock(&hash
->bh_lock
);
508 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
509 ASSERT(btp
== bp
->b_target
);
510 if (bp
->b_file_offset
== range_base
&&
511 bp
->b_buffer_length
== range_length
) {
513 * If we look at something, bring it to the
514 * front of the list for next time.
516 atomic_inc(&bp
->b_hold
);
517 list_move(&bp
->b_hash_list
, &hash
->bh_list
);
524 _xfs_buf_initialize(new_bp
, btp
, range_base
,
525 range_length
, flags
);
526 new_bp
->b_hash
= hash
;
527 list_add(&new_bp
->b_hash_list
, &hash
->bh_list
);
529 XFS_STATS_INC(xb_miss_locked
);
532 spin_unlock(&hash
->bh_lock
);
536 spin_unlock(&hash
->bh_lock
);
538 /* Attempt to get the semaphore without sleeping,
539 * if this does not work then we need to drop the
540 * spinlock and do a hard attempt on the semaphore.
542 if (down_trylock(&bp
->b_sema
)) {
543 if (!(flags
& XBF_TRYLOCK
)) {
544 /* wait for buffer ownership */
545 XB_TRACE(bp
, "get_lock", 0);
547 XFS_STATS_INC(xb_get_locked_waited
);
549 /* We asked for a trylock and failed, no need
550 * to look at file offset and length here, we
551 * know that this buffer at least overlaps our
552 * buffer and is locked, therefore our buffer
553 * either does not exist, or is this buffer.
556 XFS_STATS_INC(xb_busy_locked
);
564 if (bp
->b_flags
& XBF_STALE
) {
565 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
566 bp
->b_flags
&= XBF_MAPPED
;
568 XB_TRACE(bp
, "got_lock", 0);
569 XFS_STATS_INC(xb_get_locked
);
574 * Assembles a buffer covering the specified range.
575 * Storage in memory for all portions of the buffer will be allocated,
576 * although backing storage may not be.
580 xfs_buftarg_t
*target
,/* target for buffer */
581 xfs_off_t ioff
, /* starting offset of range */
582 size_t isize
, /* length of range */
583 xfs_buf_flags_t flags
)
585 xfs_buf_t
*bp
, *new_bp
;
588 new_bp
= xfs_buf_allocate(flags
);
589 if (unlikely(!new_bp
))
592 bp
= _xfs_buf_find(target
, ioff
, isize
, flags
, new_bp
);
594 error
= _xfs_buf_lookup_pages(bp
, flags
);
598 xfs_buf_deallocate(new_bp
);
599 if (unlikely(bp
== NULL
))
603 for (i
= 0; i
< bp
->b_page_count
; i
++)
604 mark_page_accessed(bp
->b_pages
[i
]);
606 if (!(bp
->b_flags
& XBF_MAPPED
)) {
607 error
= _xfs_buf_map_pages(bp
, flags
);
608 if (unlikely(error
)) {
609 printk(KERN_WARNING
"%s: failed to map pages\n",
615 XFS_STATS_INC(xb_get
);
618 * Always fill in the block number now, the mapped cases can do
619 * their own overlay of this later.
622 bp
->b_count_desired
= bp
->b_buffer_length
;
624 XB_TRACE(bp
, "get", (unsigned long)flags
);
628 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
636 xfs_buftarg_t
*target
,
639 xfs_buf_flags_t flags
)
645 bp
= xfs_buf_get_flags(target
, ioff
, isize
, flags
);
647 if (!XFS_BUF_ISDONE(bp
)) {
648 XB_TRACE(bp
, "read", (unsigned long)flags
);
649 XFS_STATS_INC(xb_get_read
);
650 xfs_buf_iostart(bp
, flags
);
651 } else if (flags
& XBF_ASYNC
) {
652 XB_TRACE(bp
, "read_async", (unsigned long)flags
);
654 * Read ahead call which is already satisfied,
659 XB_TRACE(bp
, "read_done", (unsigned long)flags
);
660 /* We do not want read in the flags */
661 bp
->b_flags
&= ~XBF_READ
;
668 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
675 * If we are not low on memory then do the readahead in a deadlock
680 xfs_buftarg_t
*target
,
683 xfs_buf_flags_t flags
)
685 struct backing_dev_info
*bdi
;
687 bdi
= target
->bt_mapping
->backing_dev_info
;
688 if (bdi_read_congested(bdi
))
691 flags
|= (XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
);
692 xfs_buf_read_flags(target
, ioff
, isize
, flags
);
698 xfs_buftarg_t
*target
)
702 bp
= xfs_buf_allocate(0);
704 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
708 static inline struct page
*
712 if (((unsigned long)addr
< VMALLOC_START
) ||
713 ((unsigned long)addr
>= VMALLOC_END
)) {
714 return virt_to_page(addr
);
716 return vmalloc_to_page(addr
);
721 xfs_buf_associate_memory(
733 page_count
= PAGE_CACHE_ALIGN(len
) >> PAGE_CACHE_SHIFT
;
734 offset
= (off_t
) mem
- ((off_t
)mem
& PAGE_CACHE_MASK
);
735 if (offset
&& (len
> PAGE_CACHE_SIZE
))
738 /* Free any previous set of page pointers */
740 _xfs_buf_free_pages(bp
);
745 rval
= _xfs_buf_get_pages(bp
, page_count
, 0);
749 bp
->b_offset
= offset
;
750 ptr
= (size_t) mem
& PAGE_CACHE_MASK
;
751 end
= PAGE_CACHE_ALIGN((size_t) mem
+ len
);
753 /* set up first page */
754 bp
->b_pages
[0] = mem_to_page(mem
);
756 ptr
+= PAGE_CACHE_SIZE
;
757 bp
->b_page_count
= ++i
;
759 bp
->b_pages
[i
] = mem_to_page((void *)ptr
);
760 bp
->b_page_count
= ++i
;
761 ptr
+= PAGE_CACHE_SIZE
;
765 bp
->b_count_desired
= bp
->b_buffer_length
= len
;
766 bp
->b_flags
|= XBF_MAPPED
;
774 xfs_buftarg_t
*target
)
776 unsigned long page_count
= PAGE_ALIGN(len
) >> PAGE_SHIFT
;
780 bp
= xfs_buf_allocate(0);
781 if (unlikely(bp
== NULL
))
783 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
785 error
= _xfs_buf_get_pages(bp
, page_count
, 0);
789 for (i
= 0; i
< page_count
; i
++) {
790 bp
->b_pages
[i
] = alloc_page(GFP_KERNEL
);
794 bp
->b_flags
|= _XBF_PAGES
;
796 error
= _xfs_buf_map_pages(bp
, XBF_MAPPED
);
797 if (unlikely(error
)) {
798 printk(KERN_WARNING
"%s: failed to map pages\n",
805 XB_TRACE(bp
, "no_daddr", len
);
810 __free_page(bp
->b_pages
[i
]);
811 _xfs_buf_free_pages(bp
);
813 xfs_buf_deallocate(bp
);
819 * Increment reference count on buffer, to hold the buffer concurrently
820 * with another thread which may release (free) the buffer asynchronously.
821 * Must hold the buffer already to call this function.
827 atomic_inc(&bp
->b_hold
);
828 XB_TRACE(bp
, "hold", 0);
832 * Releases a hold on the specified buffer. If the
833 * the hold count is 1, calls xfs_buf_free.
839 xfs_bufhash_t
*hash
= bp
->b_hash
;
841 XB_TRACE(bp
, "rele", bp
->b_relse
);
843 if (unlikely(!hash
)) {
844 ASSERT(!bp
->b_relse
);
845 if (atomic_dec_and_test(&bp
->b_hold
))
850 if (atomic_dec_and_lock(&bp
->b_hold
, &hash
->bh_lock
)) {
852 atomic_inc(&bp
->b_hold
);
853 spin_unlock(&hash
->bh_lock
);
854 (*(bp
->b_relse
)) (bp
);
855 } else if (bp
->b_flags
& XBF_FS_MANAGED
) {
856 spin_unlock(&hash
->bh_lock
);
858 ASSERT(!(bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)));
859 list_del_init(&bp
->b_hash_list
);
860 spin_unlock(&hash
->bh_lock
);
865 * Catch reference count leaks
867 ASSERT(atomic_read(&bp
->b_hold
) >= 0);
873 * Mutual exclusion on buffers. Locking model:
875 * Buffers associated with inodes for which buffer locking
876 * is not enabled are not protected by semaphores, and are
877 * assumed to be exclusively owned by the caller. There is a
878 * spinlock in the buffer, used by the caller when concurrent
879 * access is possible.
883 * Locks a buffer object, if it is not already locked.
884 * Note that this in no way locks the underlying pages, so it is only
885 * useful for synchronizing concurrent use of buffer objects, not for
886 * synchronizing independent access to the underlying pages.
894 locked
= down_trylock(&bp
->b_sema
) == 0;
898 XB_TRACE(bp
, "cond_lock", (long)locked
);
899 return locked
? 0 : -EBUSY
;
902 #if defined(DEBUG) || defined(XFS_BLI_TRACE)
907 return atomic_read(&bp
->b_sema
.count
);
912 * Locks a buffer object.
913 * Note that this in no way locks the underlying pages, so it is only
914 * useful for synchronizing concurrent use of buffer objects, not for
915 * synchronizing independent access to the underlying pages.
921 XB_TRACE(bp
, "lock", 0);
922 if (atomic_read(&bp
->b_io_remaining
))
923 blk_run_address_space(bp
->b_target
->bt_mapping
);
926 XB_TRACE(bp
, "locked", 0);
930 * Releases the lock on the buffer object.
931 * If the buffer is marked delwri but is not queued, do so before we
932 * unlock the buffer as we need to set flags correctly. We also need to
933 * take a reference for the delwri queue because the unlocker is going to
934 * drop their's and they don't know we just queued it.
940 if ((bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)) == XBF_DELWRI
) {
941 atomic_inc(&bp
->b_hold
);
942 bp
->b_flags
|= XBF_ASYNC
;
943 xfs_buf_delwri_queue(bp
, 0);
948 XB_TRACE(bp
, "unlock", 0);
953 * Pinning Buffer Storage in Memory
954 * Ensure that no attempt to force a buffer to disk will succeed.
960 atomic_inc(&bp
->b_pin_count
);
961 XB_TRACE(bp
, "pin", (long)bp
->b_pin_count
.counter
);
968 if (atomic_dec_and_test(&bp
->b_pin_count
))
969 wake_up_all(&bp
->b_waiters
);
970 XB_TRACE(bp
, "unpin", (long)bp
->b_pin_count
.counter
);
977 return atomic_read(&bp
->b_pin_count
);
984 DECLARE_WAITQUEUE (wait
, current
);
986 if (atomic_read(&bp
->b_pin_count
) == 0)
989 add_wait_queue(&bp
->b_waiters
, &wait
);
991 set_current_state(TASK_UNINTERRUPTIBLE
);
992 if (atomic_read(&bp
->b_pin_count
) == 0)
994 if (atomic_read(&bp
->b_io_remaining
))
995 blk_run_address_space(bp
->b_target
->bt_mapping
);
998 remove_wait_queue(&bp
->b_waiters
, &wait
);
999 set_current_state(TASK_RUNNING
);
1003 * Buffer Utility Routines
1007 xfs_buf_iodone_work(
1008 struct work_struct
*work
)
1011 container_of(work
, xfs_buf_t
, b_iodone_work
);
1014 * We can get an EOPNOTSUPP to ordered writes. Here we clear the
1015 * ordered flag and reissue them. Because we can't tell the higher
1016 * layers directly that they should not issue ordered I/O anymore, they
1017 * need to check if the ordered flag was cleared during I/O completion.
1019 if ((bp
->b_error
== EOPNOTSUPP
) &&
1020 (bp
->b_flags
& (XBF_ORDERED
|XBF_ASYNC
)) == (XBF_ORDERED
|XBF_ASYNC
)) {
1021 XB_TRACE(bp
, "ordered_retry", bp
->b_iodone
);
1022 bp
->b_flags
&= ~XBF_ORDERED
;
1023 xfs_buf_iorequest(bp
);
1024 } else if (bp
->b_iodone
)
1025 (*(bp
->b_iodone
))(bp
);
1026 else if (bp
->b_flags
& XBF_ASYNC
)
1035 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
);
1036 if (bp
->b_error
== 0)
1037 bp
->b_flags
|= XBF_DONE
;
1039 XB_TRACE(bp
, "iodone", bp
->b_iodone
);
1041 if ((bp
->b_iodone
) || (bp
->b_flags
& XBF_ASYNC
)) {
1043 INIT_WORK(&bp
->b_iodone_work
, xfs_buf_iodone_work
);
1044 queue_work(xfslogd_workqueue
, &bp
->b_iodone_work
);
1046 xfs_buf_iodone_work(&bp
->b_iodone_work
);
1049 up(&bp
->b_iodonesema
);
1058 ASSERT(error
>= 0 && error
<= 0xffff);
1059 bp
->b_error
= (unsigned short)error
;
1060 XB_TRACE(bp
, "ioerror", (unsigned long)error
);
1064 * Initiate I/O on a buffer, based on the flags supplied.
1065 * The b_iodone routine in the buffer supplied will only be called
1066 * when all of the subsidiary I/O requests, if any, have been completed.
1071 xfs_buf_flags_t flags
)
1075 XB_TRACE(bp
, "iostart", (unsigned long)flags
);
1077 if (flags
& XBF_DELWRI
) {
1078 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_ASYNC
);
1079 bp
->b_flags
|= flags
& (XBF_DELWRI
| XBF_ASYNC
);
1080 xfs_buf_delwri_queue(bp
, 1);
1084 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_ASYNC
| XBF_DELWRI
| \
1085 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
1086 bp
->b_flags
|= flags
& (XBF_READ
| XBF_WRITE
| XBF_ASYNC
| \
1087 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
1089 BUG_ON(bp
->b_bn
== XFS_BUF_DADDR_NULL
);
1091 /* For writes allow an alternate strategy routine to precede
1092 * the actual I/O request (which may not be issued at all in
1093 * a shutdown situation, for example).
1095 status
= (flags
& XBF_WRITE
) ?
1096 xfs_buf_iostrategy(bp
) : xfs_buf_iorequest(bp
);
1098 /* Wait for I/O if we are not an async request.
1099 * Note: async I/O request completion will release the buffer,
1100 * and that can already be done by this point. So using the
1101 * buffer pointer from here on, after async I/O, is invalid.
1103 if (!status
&& !(flags
& XBF_ASYNC
))
1104 status
= xfs_buf_iowait(bp
);
1113 ASSERT(bp
->b_flags
& (XBF_READ
| XBF_WRITE
));
1114 if (bp
->b_flags
& XBF_READ
)
1115 return bp
->b_locked
;
1124 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1126 xfs_buf_ioend(bp
, schedule
);
1133 unsigned int bytes_done
,
1136 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1137 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1138 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1143 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1147 struct page
*page
= bvec
->bv_page
;
1149 ASSERT(!PagePrivate(page
));
1150 if (unlikely(bp
->b_error
)) {
1151 if (bp
->b_flags
& XBF_READ
)
1152 ClearPageUptodate(page
);
1153 } else if (blocksize
>= PAGE_CACHE_SIZE
) {
1154 SetPageUptodate(page
);
1155 } else if (!PagePrivate(page
) &&
1156 (bp
->b_flags
& _XBF_PAGE_CACHE
)) {
1157 set_page_region(page
, bvec
->bv_offset
, bvec
->bv_len
);
1160 if (--bvec
>= bio
->bi_io_vec
)
1161 prefetchw(&bvec
->bv_page
->flags
);
1163 if (_xfs_buf_iolocked(bp
)) {
1166 } while (bvec
>= bio
->bi_io_vec
);
1168 _xfs_buf_ioend(bp
, 1);
1177 int i
, rw
, map_i
, total_nr_pages
, nr_pages
;
1179 int offset
= bp
->b_offset
;
1180 int size
= bp
->b_count_desired
;
1181 sector_t sector
= bp
->b_bn
;
1182 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1183 int locking
= _xfs_buf_iolocked(bp
);
1185 total_nr_pages
= bp
->b_page_count
;
1188 if (bp
->b_flags
& XBF_ORDERED
) {
1189 ASSERT(!(bp
->b_flags
& XBF_READ
));
1191 } else if (bp
->b_flags
& _XBF_RUN_QUEUES
) {
1192 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1193 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1194 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_SYNC
: READ_SYNC
;
1196 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE
:
1197 (bp
->b_flags
& XBF_READ_AHEAD
) ? READA
: READ
;
1200 /* Special code path for reading a sub page size buffer in --
1201 * we populate up the whole page, and hence the other metadata
1202 * in the same page. This optimization is only valid when the
1203 * filesystem block size is not smaller than the page size.
1205 if ((bp
->b_buffer_length
< PAGE_CACHE_SIZE
) &&
1206 (bp
->b_flags
& XBF_READ
) && locking
&&
1207 (blocksize
>= PAGE_CACHE_SIZE
)) {
1208 bio
= bio_alloc(GFP_NOIO
, 1);
1210 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1211 bio
->bi_sector
= sector
- (offset
>> BBSHIFT
);
1212 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1213 bio
->bi_private
= bp
;
1215 bio_add_page(bio
, bp
->b_pages
[0], PAGE_CACHE_SIZE
, 0);
1218 atomic_inc(&bp
->b_io_remaining
);
1223 /* Lock down the pages which we need to for the request */
1224 if (locking
&& (bp
->b_flags
& XBF_WRITE
) && (bp
->b_locked
== 0)) {
1225 for (i
= 0; size
; i
++) {
1226 int nbytes
= PAGE_CACHE_SIZE
- offset
;
1227 struct page
*page
= bp
->b_pages
[i
];
1237 offset
= bp
->b_offset
;
1238 size
= bp
->b_count_desired
;
1242 atomic_inc(&bp
->b_io_remaining
);
1243 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1244 if (nr_pages
> total_nr_pages
)
1245 nr_pages
= total_nr_pages
;
1247 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1248 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1249 bio
->bi_sector
= sector
;
1250 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1251 bio
->bi_private
= bp
;
1253 for (; size
&& nr_pages
; nr_pages
--, map_i
++) {
1254 int rbytes
, nbytes
= PAGE_CACHE_SIZE
- offset
;
1259 rbytes
= bio_add_page(bio
, bp
->b_pages
[map_i
], nbytes
, offset
);
1260 if (rbytes
< nbytes
)
1264 sector
+= nbytes
>> BBSHIFT
;
1270 if (likely(bio
->bi_size
)) {
1271 submit_bio(rw
, bio
);
1276 xfs_buf_ioerror(bp
, EIO
);
1284 XB_TRACE(bp
, "iorequest", 0);
1286 if (bp
->b_flags
& XBF_DELWRI
) {
1287 xfs_buf_delwri_queue(bp
, 1);
1291 if (bp
->b_flags
& XBF_WRITE
) {
1292 xfs_buf_wait_unpin(bp
);
1297 /* Set the count to 1 initially, this will stop an I/O
1298 * completion callout which happens before we have started
1299 * all the I/O from calling xfs_buf_ioend too early.
1301 atomic_set(&bp
->b_io_remaining
, 1);
1302 _xfs_buf_ioapply(bp
);
1303 _xfs_buf_ioend(bp
, 0);
1310 * Waits for I/O to complete on the buffer supplied.
1311 * It returns immediately if no I/O is pending.
1312 * It returns the I/O error code, if any, or 0 if there was no error.
1318 XB_TRACE(bp
, "iowait", 0);
1319 if (atomic_read(&bp
->b_io_remaining
))
1320 blk_run_address_space(bp
->b_target
->bt_mapping
);
1321 down(&bp
->b_iodonesema
);
1322 XB_TRACE(bp
, "iowaited", (long)bp
->b_error
);
1333 if (bp
->b_flags
& XBF_MAPPED
)
1334 return XFS_BUF_PTR(bp
) + offset
;
1336 offset
+= bp
->b_offset
;
1337 page
= bp
->b_pages
[offset
>> PAGE_CACHE_SHIFT
];
1338 return (xfs_caddr_t
)page_address(page
) + (offset
& (PAGE_CACHE_SIZE
-1));
1342 * Move data into or out of a buffer.
1346 xfs_buf_t
*bp
, /* buffer to process */
1347 size_t boff
, /* starting buffer offset */
1348 size_t bsize
, /* length to copy */
1349 caddr_t data
, /* data address */
1350 xfs_buf_rw_t mode
) /* read/write/zero flag */
1352 size_t bend
, cpoff
, csize
;
1355 bend
= boff
+ bsize
;
1356 while (boff
< bend
) {
1357 page
= bp
->b_pages
[xfs_buf_btoct(boff
+ bp
->b_offset
)];
1358 cpoff
= xfs_buf_poff(boff
+ bp
->b_offset
);
1359 csize
= min_t(size_t,
1360 PAGE_CACHE_SIZE
-cpoff
, bp
->b_count_desired
-boff
);
1362 ASSERT(((csize
+ cpoff
) <= PAGE_CACHE_SIZE
));
1366 memset(page_address(page
) + cpoff
, 0, csize
);
1369 memcpy(data
, page_address(page
) + cpoff
, csize
);
1372 memcpy(page_address(page
) + cpoff
, data
, csize
);
1381 * Handling of buffer targets (buftargs).
1385 * Wait for any bufs with callbacks that have been submitted but
1386 * have not yet returned... walk the hash list for the target.
1393 xfs_bufhash_t
*hash
;
1396 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1397 hash
= &btp
->bt_hash
[i
];
1399 spin_lock(&hash
->bh_lock
);
1400 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
1401 ASSERT(btp
== bp
->b_target
);
1402 if (!(bp
->b_flags
& XBF_FS_MANAGED
)) {
1403 spin_unlock(&hash
->bh_lock
);
1405 * Catch superblock reference count leaks
1408 BUG_ON(bp
->b_bn
== 0);
1413 spin_unlock(&hash
->bh_lock
);
1418 * Allocate buffer hash table for a given target.
1419 * For devices containing metadata (i.e. not the log/realtime devices)
1420 * we need to allocate a much larger hash table.
1429 btp
->bt_hashshift
= external
? 3 : 8; /* 8 or 256 buckets */
1430 btp
->bt_hashmask
= (1 << btp
->bt_hashshift
) - 1;
1431 btp
->bt_hash
= kmem_zalloc((1 << btp
->bt_hashshift
) *
1432 sizeof(xfs_bufhash_t
), KM_SLEEP
| KM_LARGE
);
1433 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1434 spin_lock_init(&btp
->bt_hash
[i
].bh_lock
);
1435 INIT_LIST_HEAD(&btp
->bt_hash
[i
].bh_list
);
1443 kmem_free(btp
->bt_hash
, (1<<btp
->bt_hashshift
) * sizeof(xfs_bufhash_t
));
1444 btp
->bt_hash
= NULL
;
1448 * buftarg list for delwrite queue processing
1450 static LIST_HEAD(xfs_buftarg_list
);
1451 static DEFINE_SPINLOCK(xfs_buftarg_lock
);
1454 xfs_register_buftarg(
1457 spin_lock(&xfs_buftarg_lock
);
1458 list_add(&btp
->bt_list
, &xfs_buftarg_list
);
1459 spin_unlock(&xfs_buftarg_lock
);
1463 xfs_unregister_buftarg(
1466 spin_lock(&xfs_buftarg_lock
);
1467 list_del(&btp
->bt_list
);
1468 spin_unlock(&xfs_buftarg_lock
);
1476 xfs_flush_buftarg(btp
, 1);
1477 xfs_blkdev_issue_flush(btp
);
1479 xfs_blkdev_put(btp
->bt_bdev
);
1480 xfs_free_bufhash(btp
);
1481 iput(btp
->bt_mapping
->host
);
1483 /* Unregister the buftarg first so that we don't get a
1484 * wakeup finding a non-existent task
1486 xfs_unregister_buftarg(btp
);
1487 kthread_stop(btp
->bt_task
);
1489 kmem_free(btp
, sizeof(*btp
));
1493 xfs_setsize_buftarg_flags(
1495 unsigned int blocksize
,
1496 unsigned int sectorsize
,
1499 btp
->bt_bsize
= blocksize
;
1500 btp
->bt_sshift
= ffs(sectorsize
) - 1;
1501 btp
->bt_smask
= sectorsize
- 1;
1503 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1505 "XFS: Cannot set_blocksize to %u on device %s\n",
1506 sectorsize
, XFS_BUFTARG_NAME(btp
));
1511 (PAGE_CACHE_SIZE
/ BITS_PER_LONG
) > sectorsize
) {
1513 "XFS: %u byte sectors in use on device %s. "
1514 "This is suboptimal; %u or greater is ideal.\n",
1515 sectorsize
, XFS_BUFTARG_NAME(btp
),
1516 (unsigned int)PAGE_CACHE_SIZE
/ BITS_PER_LONG
);
1523 * When allocating the initial buffer target we have not yet
1524 * read in the superblock, so don't know what sized sectors
1525 * are being used is at this early stage. Play safe.
1528 xfs_setsize_buftarg_early(
1530 struct block_device
*bdev
)
1532 return xfs_setsize_buftarg_flags(btp
,
1533 PAGE_CACHE_SIZE
, bdev_hardsect_size(bdev
), 0);
1537 xfs_setsize_buftarg(
1539 unsigned int blocksize
,
1540 unsigned int sectorsize
)
1542 return xfs_setsize_buftarg_flags(btp
, blocksize
, sectorsize
, 1);
1546 xfs_mapping_buftarg(
1548 struct block_device
*bdev
)
1550 struct backing_dev_info
*bdi
;
1551 struct inode
*inode
;
1552 struct address_space
*mapping
;
1553 static const struct address_space_operations mapping_aops
= {
1554 .sync_page
= block_sync_page
,
1555 .migratepage
= fail_migrate_page
,
1558 inode
= new_inode(bdev
->bd_inode
->i_sb
);
1561 "XFS: Cannot allocate mapping inode for device %s\n",
1562 XFS_BUFTARG_NAME(btp
));
1565 inode
->i_mode
= S_IFBLK
;
1566 inode
->i_bdev
= bdev
;
1567 inode
->i_rdev
= bdev
->bd_dev
;
1568 bdi
= blk_get_backing_dev_info(bdev
);
1570 bdi
= &default_backing_dev_info
;
1571 mapping
= &inode
->i_data
;
1572 mapping
->a_ops
= &mapping_aops
;
1573 mapping
->backing_dev_info
= bdi
;
1574 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
1575 btp
->bt_mapping
= mapping
;
1580 xfs_alloc_delwrite_queue(
1585 INIT_LIST_HEAD(&btp
->bt_list
);
1586 INIT_LIST_HEAD(&btp
->bt_delwrite_queue
);
1587 spinlock_init(&btp
->bt_delwrite_lock
, "delwri_lock");
1589 btp
->bt_task
= kthread_run(xfsbufd
, btp
, "xfsbufd");
1590 if (IS_ERR(btp
->bt_task
)) {
1591 error
= PTR_ERR(btp
->bt_task
);
1594 xfs_register_buftarg(btp
);
1601 struct block_device
*bdev
,
1606 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
);
1608 btp
->bt_dev
= bdev
->bd_dev
;
1609 btp
->bt_bdev
= bdev
;
1610 if (xfs_setsize_buftarg_early(btp
, bdev
))
1612 if (xfs_mapping_buftarg(btp
, bdev
))
1614 if (xfs_alloc_delwrite_queue(btp
))
1616 xfs_alloc_bufhash(btp
, external
);
1620 kmem_free(btp
, sizeof(*btp
));
1626 * Delayed write buffer handling
1629 xfs_buf_delwri_queue(
1633 struct list_head
*dwq
= &bp
->b_target
->bt_delwrite_queue
;
1634 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1636 XB_TRACE(bp
, "delwri_q", (long)unlock
);
1637 ASSERT((bp
->b_flags
&(XBF_DELWRI
|XBF_ASYNC
)) == (XBF_DELWRI
|XBF_ASYNC
));
1640 /* If already in the queue, dequeue and place at tail */
1641 if (!list_empty(&bp
->b_list
)) {
1642 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1644 atomic_dec(&bp
->b_hold
);
1645 list_del(&bp
->b_list
);
1648 bp
->b_flags
|= _XBF_DELWRI_Q
;
1649 list_add_tail(&bp
->b_list
, dwq
);
1650 bp
->b_queuetime
= jiffies
;
1658 xfs_buf_delwri_dequeue(
1661 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1665 if ((bp
->b_flags
& XBF_DELWRI
) && !list_empty(&bp
->b_list
)) {
1666 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1667 list_del_init(&bp
->b_list
);
1670 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
);
1676 XB_TRACE(bp
, "delwri_dq", (long)dequeued
);
1680 xfs_buf_runall_queues(
1681 struct workqueue_struct
*queue
)
1683 flush_workqueue(queue
);
1693 spin_lock(&xfs_buftarg_lock
);
1694 list_for_each_entry(btp
, &xfs_buftarg_list
, bt_list
) {
1695 if (test_bit(XBT_FORCE_SLEEP
, &btp
->bt_flags
))
1697 set_bit(XBT_FORCE_FLUSH
, &btp
->bt_flags
);
1698 wake_up_process(btp
->bt_task
);
1700 spin_unlock(&xfs_buftarg_lock
);
1705 * Move as many buffers as specified to the supplied list
1706 * idicating if we skipped any buffers to prevent deadlocks.
1709 xfs_buf_delwri_split(
1710 xfs_buftarg_t
*target
,
1711 struct list_head
*list
,
1715 struct list_head
*dwq
= &target
->bt_delwrite_queue
;
1716 spinlock_t
*dwlk
= &target
->bt_delwrite_lock
;
1720 force
= test_and_clear_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1721 INIT_LIST_HEAD(list
);
1723 list_for_each_entry_safe(bp
, n
, dwq
, b_list
) {
1724 XB_TRACE(bp
, "walkq1", (long)xfs_buf_ispin(bp
));
1725 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1727 if (!xfs_buf_ispin(bp
) && !xfs_buf_cond_lock(bp
)) {
1729 time_before(jiffies
, bp
->b_queuetime
+ age
)) {
1734 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
|
1736 bp
->b_flags
|= XBF_WRITE
;
1737 list_move_tail(&bp
->b_list
, list
);
1751 struct list_head tmp
;
1752 xfs_buftarg_t
*target
= (xfs_buftarg_t
*)data
;
1756 current
->flags
|= PF_MEMALLOC
;
1759 if (unlikely(freezing(current
))) {
1760 set_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1763 clear_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1766 schedule_timeout_interruptible(
1767 xfs_buf_timer_centisecs
* msecs_to_jiffies(10));
1769 xfs_buf_delwri_split(target
, &tmp
,
1770 xfs_buf_age_centisecs
* msecs_to_jiffies(10));
1773 while (!list_empty(&tmp
)) {
1774 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1775 ASSERT(target
== bp
->b_target
);
1777 list_del_init(&bp
->b_list
);
1778 xfs_buf_iostrategy(bp
);
1782 if (as_list_len
> 0)
1785 blk_run_address_space(target
->bt_mapping
);
1787 } while (!kthread_should_stop());
1793 * Go through all incore buffers, and release buffers if they belong to
1794 * the given device. This is used in filesystem error handling to
1795 * preserve the consistency of its metadata.
1799 xfs_buftarg_t
*target
,
1802 struct list_head tmp
;
1806 xfs_buf_runall_queues(xfsdatad_workqueue
);
1807 xfs_buf_runall_queues(xfslogd_workqueue
);
1809 set_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1810 pincount
= xfs_buf_delwri_split(target
, &tmp
, 0);
1813 * Dropped the delayed write list lock, now walk the temporary list
1815 list_for_each_entry_safe(bp
, n
, &tmp
, b_list
) {
1816 ASSERT(target
== bp
->b_target
);
1818 bp
->b_flags
&= ~XBF_ASYNC
;
1820 list_del_init(&bp
->b_list
);
1822 xfs_buf_iostrategy(bp
);
1826 blk_run_address_space(target
->bt_mapping
);
1829 * Remaining list items must be flushed before returning
1831 while (!list_empty(&tmp
)) {
1832 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1834 list_del_init(&bp
->b_list
);
1845 #ifdef XFS_BUF_TRACE
1846 xfs_buf_trace_buf
= ktrace_alloc(XFS_BUF_TRACE_SIZE
, KM_SLEEP
);
1849 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1850 KM_ZONE_HWALIGN
, NULL
);
1852 goto out_free_trace_buf
;
1854 xfslogd_workqueue
= create_workqueue("xfslogd");
1855 if (!xfslogd_workqueue
)
1856 goto out_free_buf_zone
;
1858 xfsdatad_workqueue
= create_workqueue("xfsdatad");
1859 if (!xfsdatad_workqueue
)
1860 goto out_destroy_xfslogd_workqueue
;
1862 register_shrinker(&xfs_buf_shake
);
1865 out_destroy_xfslogd_workqueue
:
1866 destroy_workqueue(xfslogd_workqueue
);
1868 kmem_zone_destroy(xfs_buf_zone
);
1870 #ifdef XFS_BUF_TRACE
1871 ktrace_free(xfs_buf_trace_buf
);
1877 xfs_buf_terminate(void)
1879 unregister_shrinker(&xfs_buf_shake
);
1880 destroy_workqueue(xfsdatad_workqueue
);
1881 destroy_workqueue(xfslogd_workqueue
);
1882 kmem_zone_destroy(xfs_buf_zone
);
1883 #ifdef XFS_BUF_TRACE
1884 ktrace_free(xfs_buf_trace_buf
);
1888 #ifdef CONFIG_KDB_MODULES
1890 xfs_get_buftarg_list(void)
1892 return &xfs_buftarg_list
;