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.
190 aentry
= kmalloc(sizeof(a_list_t
), GFP_NOWAIT
);
191 if (likely(aentry
)) {
193 aentry
->next
= as_free_head
;
194 aentry
->vm_addr
= addr
;
195 as_free_head
= aentry
;
197 spin_unlock(&as_lock
);
204 purge_addresses(void)
206 a_list_t
*aentry
, *old
;
208 if (as_free_head
== NULL
)
212 aentry
= as_free_head
;
215 spin_unlock(&as_lock
);
217 while ((old
= aentry
) != NULL
) {
218 vunmap(aentry
->vm_addr
);
219 aentry
= aentry
->next
;
225 * Internal xfs_buf_t object manipulation
231 xfs_buftarg_t
*target
,
232 xfs_off_t range_base
,
234 xfs_buf_flags_t flags
)
237 * We don't want certain flags to appear in b_flags.
239 flags
&= ~(XBF_LOCK
|XBF_MAPPED
|XBF_DONT_BLOCK
|XBF_READ_AHEAD
);
241 memset(bp
, 0, sizeof(xfs_buf_t
));
242 atomic_set(&bp
->b_hold
, 1);
243 init_MUTEX_LOCKED(&bp
->b_iodonesema
);
244 INIT_LIST_HEAD(&bp
->b_list
);
245 INIT_LIST_HEAD(&bp
->b_hash_list
);
246 init_MUTEX_LOCKED(&bp
->b_sema
); /* held, no waiters */
248 bp
->b_target
= target
;
249 bp
->b_file_offset
= range_base
;
251 * Set buffer_length and count_desired to the same value initially.
252 * I/O routines should use count_desired, which will be the same in
253 * most cases but may be reset (e.g. XFS recovery).
255 bp
->b_buffer_length
= bp
->b_count_desired
= range_length
;
257 bp
->b_bn
= XFS_BUF_DADDR_NULL
;
258 atomic_set(&bp
->b_pin_count
, 0);
259 init_waitqueue_head(&bp
->b_waiters
);
261 XFS_STATS_INC(xb_create
);
262 XB_TRACE(bp
, "initialize", target
);
266 * Allocate a page array capable of holding a specified number
267 * of pages, and point the page buf at it.
273 xfs_buf_flags_t flags
)
275 /* Make sure that we have a page list */
276 if (bp
->b_pages
== NULL
) {
277 bp
->b_offset
= xfs_buf_poff(bp
->b_file_offset
);
278 bp
->b_page_count
= page_count
;
279 if (page_count
<= XB_PAGES
) {
280 bp
->b_pages
= bp
->b_page_array
;
282 bp
->b_pages
= kmem_alloc(sizeof(struct page
*) *
283 page_count
, xb_to_km(flags
));
284 if (bp
->b_pages
== NULL
)
287 memset(bp
->b_pages
, 0, sizeof(struct page
*) * page_count
);
293 * Frees b_pages if it was allocated.
299 if (bp
->b_pages
!= bp
->b_page_array
) {
300 kmem_free(bp
->b_pages
,
301 bp
->b_page_count
* sizeof(struct page
*));
306 * Releases the specified buffer.
308 * The modification state of any associated pages is left unchanged.
309 * The buffer most not be on any hash - use xfs_buf_rele instead for
310 * hashed and refcounted buffers
316 XB_TRACE(bp
, "free", 0);
318 ASSERT(list_empty(&bp
->b_hash_list
));
320 if (bp
->b_flags
& (_XBF_PAGE_CACHE
|_XBF_PAGES
)) {
323 if ((bp
->b_flags
& XBF_MAPPED
) && (bp
->b_page_count
> 1))
324 free_address(bp
->b_addr
- bp
->b_offset
);
326 for (i
= 0; i
< bp
->b_page_count
; i
++) {
327 struct page
*page
= bp
->b_pages
[i
];
329 if (bp
->b_flags
& _XBF_PAGE_CACHE
)
330 ASSERT(!PagePrivate(page
));
331 page_cache_release(page
);
333 _xfs_buf_free_pages(bp
);
336 xfs_buf_deallocate(bp
);
340 * Finds all pages for buffer in question and builds it's page list.
343 _xfs_buf_lookup_pages(
347 struct address_space
*mapping
= bp
->b_target
->bt_mapping
;
348 size_t blocksize
= bp
->b_target
->bt_bsize
;
349 size_t size
= bp
->b_count_desired
;
350 size_t nbytes
, offset
;
351 gfp_t gfp_mask
= xb_to_gfp(flags
);
352 unsigned short page_count
, i
;
357 end
= bp
->b_file_offset
+ bp
->b_buffer_length
;
358 page_count
= xfs_buf_btoc(end
) - xfs_buf_btoct(bp
->b_file_offset
);
360 error
= _xfs_buf_get_pages(bp
, page_count
, flags
);
363 bp
->b_flags
|= _XBF_PAGE_CACHE
;
365 offset
= bp
->b_offset
;
366 first
= bp
->b_file_offset
>> PAGE_CACHE_SHIFT
;
368 for (i
= 0; i
< bp
->b_page_count
; i
++) {
373 page
= find_or_create_page(mapping
, first
+ i
, gfp_mask
);
374 if (unlikely(page
== NULL
)) {
375 if (flags
& XBF_READ_AHEAD
) {
376 bp
->b_page_count
= i
;
377 for (i
= 0; i
< bp
->b_page_count
; i
++)
378 unlock_page(bp
->b_pages
[i
]);
383 * This could deadlock.
385 * But until all the XFS lowlevel code is revamped to
386 * handle buffer allocation failures we can't do much.
388 if (!(++retries
% 100))
390 "XFS: possible memory allocation "
391 "deadlock in %s (mode:0x%x)\n",
392 __FUNCTION__
, gfp_mask
);
394 XFS_STATS_INC(xb_page_retries
);
395 xfsbufd_wakeup(0, gfp_mask
);
396 congestion_wait(WRITE
, HZ
/50);
400 XFS_STATS_INC(xb_page_found
);
402 nbytes
= min_t(size_t, size
, PAGE_CACHE_SIZE
- offset
);
405 ASSERT(!PagePrivate(page
));
406 if (!PageUptodate(page
)) {
408 if (blocksize
>= PAGE_CACHE_SIZE
) {
409 if (flags
& XBF_READ
)
411 } else if (!PagePrivate(page
)) {
412 if (test_page_region(page
, offset
, nbytes
))
417 bp
->b_pages
[i
] = page
;
422 for (i
= 0; i
< bp
->b_page_count
; i
++)
423 unlock_page(bp
->b_pages
[i
]);
426 if (page_count
== bp
->b_page_count
)
427 bp
->b_flags
|= XBF_DONE
;
429 XB_TRACE(bp
, "lookup_pages", (long)page_count
);
434 * Map buffer into kernel address-space if nessecary.
441 /* A single page buffer is always mappable */
442 if (bp
->b_page_count
== 1) {
443 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
444 bp
->b_flags
|= XBF_MAPPED
;
445 } else if (flags
& XBF_MAPPED
) {
446 if (as_list_len
> 64)
448 bp
->b_addr
= vmap(bp
->b_pages
, bp
->b_page_count
,
449 VM_MAP
, PAGE_KERNEL
);
450 if (unlikely(bp
->b_addr
== NULL
))
452 bp
->b_addr
+= bp
->b_offset
;
453 bp
->b_flags
|= XBF_MAPPED
;
460 * Finding and Reading Buffers
464 * Look up, and creates if absent, a lockable buffer for
465 * a given range of an inode. The buffer is returned
466 * locked. If other overlapping buffers exist, they are
467 * released before the new buffer is created and locked,
468 * which may imply that this call will block until those buffers
469 * are unlocked. No I/O is implied by this call.
473 xfs_buftarg_t
*btp
, /* block device target */
474 xfs_off_t ioff
, /* starting offset of range */
475 size_t isize
, /* length of range */
476 xfs_buf_flags_t flags
,
479 xfs_off_t range_base
;
484 range_base
= (ioff
<< BBSHIFT
);
485 range_length
= (isize
<< BBSHIFT
);
487 /* Check for IOs smaller than the sector size / not sector aligned */
488 ASSERT(!(range_length
< (1 << btp
->bt_sshift
)));
489 ASSERT(!(range_base
& (xfs_off_t
)btp
->bt_smask
));
491 hash
= &btp
->bt_hash
[hash_long((unsigned long)ioff
, btp
->bt_hashshift
)];
493 spin_lock(&hash
->bh_lock
);
495 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
496 ASSERT(btp
== bp
->b_target
);
497 if (bp
->b_file_offset
== range_base
&&
498 bp
->b_buffer_length
== range_length
) {
500 * If we look at something, bring it to the
501 * front of the list for next time.
503 atomic_inc(&bp
->b_hold
);
504 list_move(&bp
->b_hash_list
, &hash
->bh_list
);
511 _xfs_buf_initialize(new_bp
, btp
, range_base
,
512 range_length
, flags
);
513 new_bp
->b_hash
= hash
;
514 list_add(&new_bp
->b_hash_list
, &hash
->bh_list
);
516 XFS_STATS_INC(xb_miss_locked
);
519 spin_unlock(&hash
->bh_lock
);
523 spin_unlock(&hash
->bh_lock
);
525 /* Attempt to get the semaphore without sleeping,
526 * if this does not work then we need to drop the
527 * spinlock and do a hard attempt on the semaphore.
529 if (down_trylock(&bp
->b_sema
)) {
530 if (!(flags
& XBF_TRYLOCK
)) {
531 /* wait for buffer ownership */
532 XB_TRACE(bp
, "get_lock", 0);
534 XFS_STATS_INC(xb_get_locked_waited
);
536 /* We asked for a trylock and failed, no need
537 * to look at file offset and length here, we
538 * know that this buffer at least overlaps our
539 * buffer and is locked, therefore our buffer
540 * either does not exist, or is this buffer.
543 XFS_STATS_INC(xb_busy_locked
);
551 if (bp
->b_flags
& XBF_STALE
) {
552 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
553 bp
->b_flags
&= XBF_MAPPED
;
555 XB_TRACE(bp
, "got_lock", 0);
556 XFS_STATS_INC(xb_get_locked
);
561 * Assembles a buffer covering the specified range.
562 * Storage in memory for all portions of the buffer will be allocated,
563 * although backing storage may not be.
567 xfs_buftarg_t
*target
,/* target for buffer */
568 xfs_off_t ioff
, /* starting offset of range */
569 size_t isize
, /* length of range */
570 xfs_buf_flags_t flags
)
572 xfs_buf_t
*bp
, *new_bp
;
575 new_bp
= xfs_buf_allocate(flags
);
576 if (unlikely(!new_bp
))
579 bp
= _xfs_buf_find(target
, ioff
, isize
, flags
, new_bp
);
581 error
= _xfs_buf_lookup_pages(bp
, flags
);
585 xfs_buf_deallocate(new_bp
);
586 if (unlikely(bp
== NULL
))
590 for (i
= 0; i
< bp
->b_page_count
; i
++)
591 mark_page_accessed(bp
->b_pages
[i
]);
593 if (!(bp
->b_flags
& XBF_MAPPED
)) {
594 error
= _xfs_buf_map_pages(bp
, flags
);
595 if (unlikely(error
)) {
596 printk(KERN_WARNING
"%s: failed to map pages\n",
602 XFS_STATS_INC(xb_get
);
605 * Always fill in the block number now, the mapped cases can do
606 * their own overlay of this later.
609 bp
->b_count_desired
= bp
->b_buffer_length
;
611 XB_TRACE(bp
, "get", (unsigned long)flags
);
615 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
623 xfs_buftarg_t
*target
,
626 xfs_buf_flags_t flags
)
632 bp
= xfs_buf_get_flags(target
, ioff
, isize
, flags
);
634 if (!XFS_BUF_ISDONE(bp
)) {
635 XB_TRACE(bp
, "read", (unsigned long)flags
);
636 XFS_STATS_INC(xb_get_read
);
637 xfs_buf_iostart(bp
, flags
);
638 } else if (flags
& XBF_ASYNC
) {
639 XB_TRACE(bp
, "read_async", (unsigned long)flags
);
641 * Read ahead call which is already satisfied,
646 XB_TRACE(bp
, "read_done", (unsigned long)flags
);
647 /* We do not want read in the flags */
648 bp
->b_flags
&= ~XBF_READ
;
655 if (flags
& (XBF_LOCK
| XBF_TRYLOCK
))
662 * If we are not low on memory then do the readahead in a deadlock
667 xfs_buftarg_t
*target
,
670 xfs_buf_flags_t flags
)
672 struct backing_dev_info
*bdi
;
674 bdi
= target
->bt_mapping
->backing_dev_info
;
675 if (bdi_read_congested(bdi
))
678 flags
|= (XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
);
679 xfs_buf_read_flags(target
, ioff
, isize
, flags
);
685 xfs_buftarg_t
*target
)
689 bp
= xfs_buf_allocate(0);
691 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
695 static inline struct page
*
699 if (((unsigned long)addr
< VMALLOC_START
) ||
700 ((unsigned long)addr
>= VMALLOC_END
)) {
701 return virt_to_page(addr
);
703 return vmalloc_to_page(addr
);
708 xfs_buf_associate_memory(
720 page_count
= PAGE_CACHE_ALIGN(len
) >> PAGE_CACHE_SHIFT
;
721 offset
= (off_t
) mem
- ((off_t
)mem
& PAGE_CACHE_MASK
);
722 if (offset
&& (len
> PAGE_CACHE_SIZE
))
725 /* Free any previous set of page pointers */
727 _xfs_buf_free_pages(bp
);
732 rval
= _xfs_buf_get_pages(bp
, page_count
, 0);
736 bp
->b_offset
= offset
;
737 ptr
= (size_t) mem
& PAGE_CACHE_MASK
;
738 end
= PAGE_CACHE_ALIGN((size_t) mem
+ len
);
740 /* set up first page */
741 bp
->b_pages
[0] = mem_to_page(mem
);
743 ptr
+= PAGE_CACHE_SIZE
;
744 bp
->b_page_count
= ++i
;
746 bp
->b_pages
[i
] = mem_to_page((void *)ptr
);
747 bp
->b_page_count
= ++i
;
748 ptr
+= PAGE_CACHE_SIZE
;
752 bp
->b_count_desired
= bp
->b_buffer_length
= len
;
753 bp
->b_flags
|= XBF_MAPPED
;
761 xfs_buftarg_t
*target
)
763 unsigned long page_count
= PAGE_ALIGN(len
) >> PAGE_SHIFT
;
767 bp
= xfs_buf_allocate(0);
768 if (unlikely(bp
== NULL
))
770 _xfs_buf_initialize(bp
, target
, 0, len
, 0);
772 error
= _xfs_buf_get_pages(bp
, page_count
, 0);
776 for (i
= 0; i
< page_count
; i
++) {
777 bp
->b_pages
[i
] = alloc_page(GFP_KERNEL
);
781 bp
->b_flags
|= _XBF_PAGES
;
783 error
= _xfs_buf_map_pages(bp
, XBF_MAPPED
);
784 if (unlikely(error
)) {
785 printk(KERN_WARNING
"%s: failed to map pages\n",
792 XB_TRACE(bp
, "no_daddr", len
);
797 __free_page(bp
->b_pages
[i
]);
798 _xfs_buf_free_pages(bp
);
800 xfs_buf_deallocate(bp
);
806 * Increment reference count on buffer, to hold the buffer concurrently
807 * with another thread which may release (free) the buffer asynchronously.
808 * Must hold the buffer already to call this function.
814 atomic_inc(&bp
->b_hold
);
815 XB_TRACE(bp
, "hold", 0);
819 * Releases a hold on the specified buffer. If the
820 * the hold count is 1, calls xfs_buf_free.
826 xfs_bufhash_t
*hash
= bp
->b_hash
;
828 XB_TRACE(bp
, "rele", bp
->b_relse
);
830 if (unlikely(!hash
)) {
831 ASSERT(!bp
->b_relse
);
832 if (atomic_dec_and_test(&bp
->b_hold
))
837 if (atomic_dec_and_lock(&bp
->b_hold
, &hash
->bh_lock
)) {
839 atomic_inc(&bp
->b_hold
);
840 spin_unlock(&hash
->bh_lock
);
841 (*(bp
->b_relse
)) (bp
);
842 } else if (bp
->b_flags
& XBF_FS_MANAGED
) {
843 spin_unlock(&hash
->bh_lock
);
845 ASSERT(!(bp
->b_flags
& (XBF_DELWRI
|_XBF_DELWRI_Q
)));
846 list_del_init(&bp
->b_hash_list
);
847 spin_unlock(&hash
->bh_lock
);
852 * Catch reference count leaks
854 ASSERT(atomic_read(&bp
->b_hold
) >= 0);
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 atomic_read(&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 (*(bp
->b_iodone
))(bp
);
1002 else if (bp
->b_flags
& XBF_ASYNC
)
1011 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
);
1012 if (bp
->b_error
== 0)
1013 bp
->b_flags
|= XBF_DONE
;
1015 XB_TRACE(bp
, "iodone", bp
->b_iodone
);
1017 if ((bp
->b_iodone
) || (bp
->b_flags
& XBF_ASYNC
)) {
1019 INIT_WORK(&bp
->b_iodone_work
, xfs_buf_iodone_work
);
1020 queue_work(xfslogd_workqueue
, &bp
->b_iodone_work
);
1022 xfs_buf_iodone_work(&bp
->b_iodone_work
);
1025 up(&bp
->b_iodonesema
);
1034 ASSERT(error
>= 0 && error
<= 0xffff);
1035 bp
->b_error
= (unsigned short)error
;
1036 XB_TRACE(bp
, "ioerror", (unsigned long)error
);
1040 * Initiate I/O on a buffer, based on the flags supplied.
1041 * The b_iodone routine in the buffer supplied will only be called
1042 * when all of the subsidiary I/O requests, if any, have been completed.
1047 xfs_buf_flags_t flags
)
1051 XB_TRACE(bp
, "iostart", (unsigned long)flags
);
1053 if (flags
& XBF_DELWRI
) {
1054 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_ASYNC
);
1055 bp
->b_flags
|= flags
& (XBF_DELWRI
| XBF_ASYNC
);
1056 xfs_buf_delwri_queue(bp
, 1);
1060 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_ASYNC
| XBF_DELWRI
| \
1061 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
1062 bp
->b_flags
|= flags
& (XBF_READ
| XBF_WRITE
| XBF_ASYNC
| \
1063 XBF_READ_AHEAD
| _XBF_RUN_QUEUES
);
1065 BUG_ON(bp
->b_bn
== XFS_BUF_DADDR_NULL
);
1067 /* For writes allow an alternate strategy routine to precede
1068 * the actual I/O request (which may not be issued at all in
1069 * a shutdown situation, for example).
1071 status
= (flags
& XBF_WRITE
) ?
1072 xfs_buf_iostrategy(bp
) : xfs_buf_iorequest(bp
);
1074 /* Wait for I/O if we are not an async request.
1075 * Note: async I/O request completion will release the buffer,
1076 * and that can already be done by this point. So using the
1077 * buffer pointer from here on, after async I/O, is invalid.
1079 if (!status
&& !(flags
& XBF_ASYNC
))
1080 status
= xfs_buf_iowait(bp
);
1089 ASSERT(bp
->b_flags
& (XBF_READ
| XBF_WRITE
));
1090 if (bp
->b_flags
& XBF_READ
)
1091 return bp
->b_locked
;
1100 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1102 xfs_buf_ioend(bp
, schedule
);
1111 xfs_buf_t
*bp
= (xfs_buf_t
*)bio
->bi_private
;
1112 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1113 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
1115 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1119 struct page
*page
= bvec
->bv_page
;
1121 ASSERT(!PagePrivate(page
));
1122 if (unlikely(bp
->b_error
)) {
1123 if (bp
->b_flags
& XBF_READ
)
1124 ClearPageUptodate(page
);
1125 } else if (blocksize
>= PAGE_CACHE_SIZE
) {
1126 SetPageUptodate(page
);
1127 } else if (!PagePrivate(page
) &&
1128 (bp
->b_flags
& _XBF_PAGE_CACHE
)) {
1129 set_page_region(page
, bvec
->bv_offset
, bvec
->bv_len
);
1132 if (--bvec
>= bio
->bi_io_vec
)
1133 prefetchw(&bvec
->bv_page
->flags
);
1135 if (_xfs_buf_iolocked(bp
)) {
1138 } while (bvec
>= bio
->bi_io_vec
);
1140 _xfs_buf_ioend(bp
, 1);
1149 int i
, rw
, map_i
, total_nr_pages
, nr_pages
;
1151 int offset
= bp
->b_offset
;
1152 int size
= bp
->b_count_desired
;
1153 sector_t sector
= bp
->b_bn
;
1154 unsigned int blocksize
= bp
->b_target
->bt_bsize
;
1155 int locking
= _xfs_buf_iolocked(bp
);
1157 total_nr_pages
= bp
->b_page_count
;
1160 if (bp
->b_flags
& XBF_ORDERED
) {
1161 ASSERT(!(bp
->b_flags
& XBF_READ
));
1163 } else if (bp
->b_flags
& _XBF_RUN_QUEUES
) {
1164 ASSERT(!(bp
->b_flags
& XBF_READ_AHEAD
));
1165 bp
->b_flags
&= ~_XBF_RUN_QUEUES
;
1166 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE_SYNC
: READ_SYNC
;
1168 rw
= (bp
->b_flags
& XBF_WRITE
) ? WRITE
:
1169 (bp
->b_flags
& XBF_READ_AHEAD
) ? READA
: READ
;
1172 /* Special code path for reading a sub page size buffer in --
1173 * we populate up the whole page, and hence the other metadata
1174 * in the same page. This optimization is only valid when the
1175 * filesystem block size is not smaller than the page size.
1177 if ((bp
->b_buffer_length
< PAGE_CACHE_SIZE
) &&
1178 (bp
->b_flags
& XBF_READ
) && locking
&&
1179 (blocksize
>= PAGE_CACHE_SIZE
)) {
1180 bio
= bio_alloc(GFP_NOIO
, 1);
1182 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1183 bio
->bi_sector
= sector
- (offset
>> BBSHIFT
);
1184 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1185 bio
->bi_private
= bp
;
1187 bio_add_page(bio
, bp
->b_pages
[0], PAGE_CACHE_SIZE
, 0);
1190 atomic_inc(&bp
->b_io_remaining
);
1195 /* Lock down the pages which we need to for the request */
1196 if (locking
&& (bp
->b_flags
& XBF_WRITE
) && (bp
->b_locked
== 0)) {
1197 for (i
= 0; size
; i
++) {
1198 int nbytes
= PAGE_CACHE_SIZE
- offset
;
1199 struct page
*page
= bp
->b_pages
[i
];
1209 offset
= bp
->b_offset
;
1210 size
= bp
->b_count_desired
;
1214 atomic_inc(&bp
->b_io_remaining
);
1215 nr_pages
= BIO_MAX_SECTORS
>> (PAGE_SHIFT
- BBSHIFT
);
1216 if (nr_pages
> total_nr_pages
)
1217 nr_pages
= total_nr_pages
;
1219 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1220 bio
->bi_bdev
= bp
->b_target
->bt_bdev
;
1221 bio
->bi_sector
= sector
;
1222 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1223 bio
->bi_private
= bp
;
1225 for (; size
&& nr_pages
; nr_pages
--, map_i
++) {
1226 int rbytes
, nbytes
= PAGE_CACHE_SIZE
- offset
;
1231 rbytes
= bio_add_page(bio
, bp
->b_pages
[map_i
], nbytes
, offset
);
1232 if (rbytes
< nbytes
)
1236 sector
+= nbytes
>> BBSHIFT
;
1242 if (likely(bio
->bi_size
)) {
1243 submit_bio(rw
, bio
);
1248 xfs_buf_ioerror(bp
, EIO
);
1256 XB_TRACE(bp
, "iorequest", 0);
1258 if (bp
->b_flags
& XBF_DELWRI
) {
1259 xfs_buf_delwri_queue(bp
, 1);
1263 if (bp
->b_flags
& XBF_WRITE
) {
1264 xfs_buf_wait_unpin(bp
);
1269 /* Set the count to 1 initially, this will stop an I/O
1270 * completion callout which happens before we have started
1271 * all the I/O from calling xfs_buf_ioend too early.
1273 atomic_set(&bp
->b_io_remaining
, 1);
1274 _xfs_buf_ioapply(bp
);
1275 _xfs_buf_ioend(bp
, 0);
1282 * Waits for I/O to complete on the buffer supplied.
1283 * It returns immediately if no I/O is pending.
1284 * It returns the I/O error code, if any, or 0 if there was no error.
1290 XB_TRACE(bp
, "iowait", 0);
1291 if (atomic_read(&bp
->b_io_remaining
))
1292 blk_run_address_space(bp
->b_target
->bt_mapping
);
1293 down(&bp
->b_iodonesema
);
1294 XB_TRACE(bp
, "iowaited", (long)bp
->b_error
);
1305 if (bp
->b_flags
& XBF_MAPPED
)
1306 return XFS_BUF_PTR(bp
) + offset
;
1308 offset
+= bp
->b_offset
;
1309 page
= bp
->b_pages
[offset
>> PAGE_CACHE_SHIFT
];
1310 return (xfs_caddr_t
)page_address(page
) + (offset
& (PAGE_CACHE_SIZE
-1));
1314 * Move data into or out of a buffer.
1318 xfs_buf_t
*bp
, /* buffer to process */
1319 size_t boff
, /* starting buffer offset */
1320 size_t bsize
, /* length to copy */
1321 caddr_t data
, /* data address */
1322 xfs_buf_rw_t mode
) /* read/write/zero flag */
1324 size_t bend
, cpoff
, csize
;
1327 bend
= boff
+ bsize
;
1328 while (boff
< bend
) {
1329 page
= bp
->b_pages
[xfs_buf_btoct(boff
+ bp
->b_offset
)];
1330 cpoff
= xfs_buf_poff(boff
+ bp
->b_offset
);
1331 csize
= min_t(size_t,
1332 PAGE_CACHE_SIZE
-cpoff
, bp
->b_count_desired
-boff
);
1334 ASSERT(((csize
+ cpoff
) <= PAGE_CACHE_SIZE
));
1338 memset(page_address(page
) + cpoff
, 0, csize
);
1341 memcpy(data
, page_address(page
) + cpoff
, csize
);
1344 memcpy(page_address(page
) + cpoff
, data
, csize
);
1353 * Handling of buffer targets (buftargs).
1357 * Wait for any bufs with callbacks that have been submitted but
1358 * have not yet returned... walk the hash list for the target.
1365 xfs_bufhash_t
*hash
;
1368 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1369 hash
= &btp
->bt_hash
[i
];
1371 spin_lock(&hash
->bh_lock
);
1372 list_for_each_entry_safe(bp
, n
, &hash
->bh_list
, b_hash_list
) {
1373 ASSERT(btp
== bp
->b_target
);
1374 if (!(bp
->b_flags
& XBF_FS_MANAGED
)) {
1375 spin_unlock(&hash
->bh_lock
);
1377 * Catch superblock reference count leaks
1380 BUG_ON(bp
->b_bn
== 0);
1385 spin_unlock(&hash
->bh_lock
);
1390 * Allocate buffer hash table for a given target.
1391 * For devices containing metadata (i.e. not the log/realtime devices)
1392 * we need to allocate a much larger hash table.
1401 btp
->bt_hashshift
= external
? 3 : 8; /* 8 or 256 buckets */
1402 btp
->bt_hashmask
= (1 << btp
->bt_hashshift
) - 1;
1403 btp
->bt_hash
= kmem_zalloc((1 << btp
->bt_hashshift
) *
1404 sizeof(xfs_bufhash_t
), KM_SLEEP
| KM_LARGE
);
1405 for (i
= 0; i
< (1 << btp
->bt_hashshift
); i
++) {
1406 spin_lock_init(&btp
->bt_hash
[i
].bh_lock
);
1407 INIT_LIST_HEAD(&btp
->bt_hash
[i
].bh_list
);
1415 kmem_free(btp
->bt_hash
, (1<<btp
->bt_hashshift
) * sizeof(xfs_bufhash_t
));
1416 btp
->bt_hash
= NULL
;
1420 * buftarg list for delwrite queue processing
1422 static LIST_HEAD(xfs_buftarg_list
);
1423 static DEFINE_SPINLOCK(xfs_buftarg_lock
);
1426 xfs_register_buftarg(
1429 spin_lock(&xfs_buftarg_lock
);
1430 list_add(&btp
->bt_list
, &xfs_buftarg_list
);
1431 spin_unlock(&xfs_buftarg_lock
);
1435 xfs_unregister_buftarg(
1438 spin_lock(&xfs_buftarg_lock
);
1439 list_del(&btp
->bt_list
);
1440 spin_unlock(&xfs_buftarg_lock
);
1448 xfs_flush_buftarg(btp
, 1);
1449 xfs_blkdev_issue_flush(btp
);
1451 xfs_blkdev_put(btp
->bt_bdev
);
1452 xfs_free_bufhash(btp
);
1453 iput(btp
->bt_mapping
->host
);
1455 /* Unregister the buftarg first so that we don't get a
1456 * wakeup finding a non-existent task
1458 xfs_unregister_buftarg(btp
);
1459 kthread_stop(btp
->bt_task
);
1461 kmem_free(btp
, sizeof(*btp
));
1465 xfs_setsize_buftarg_flags(
1467 unsigned int blocksize
,
1468 unsigned int sectorsize
,
1471 btp
->bt_bsize
= blocksize
;
1472 btp
->bt_sshift
= ffs(sectorsize
) - 1;
1473 btp
->bt_smask
= sectorsize
- 1;
1475 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1477 "XFS: Cannot set_blocksize to %u on device %s\n",
1478 sectorsize
, XFS_BUFTARG_NAME(btp
));
1483 (PAGE_CACHE_SIZE
/ BITS_PER_LONG
) > sectorsize
) {
1485 "XFS: %u byte sectors in use on device %s. "
1486 "This is suboptimal; %u or greater is ideal.\n",
1487 sectorsize
, XFS_BUFTARG_NAME(btp
),
1488 (unsigned int)PAGE_CACHE_SIZE
/ BITS_PER_LONG
);
1495 * When allocating the initial buffer target we have not yet
1496 * read in the superblock, so don't know what sized sectors
1497 * are being used is at this early stage. Play safe.
1500 xfs_setsize_buftarg_early(
1502 struct block_device
*bdev
)
1504 return xfs_setsize_buftarg_flags(btp
,
1505 PAGE_CACHE_SIZE
, bdev_hardsect_size(bdev
), 0);
1509 xfs_setsize_buftarg(
1511 unsigned int blocksize
,
1512 unsigned int sectorsize
)
1514 return xfs_setsize_buftarg_flags(btp
, blocksize
, sectorsize
, 1);
1518 xfs_mapping_buftarg(
1520 struct block_device
*bdev
)
1522 struct backing_dev_info
*bdi
;
1523 struct inode
*inode
;
1524 struct address_space
*mapping
;
1525 static const struct address_space_operations mapping_aops
= {
1526 .sync_page
= block_sync_page
,
1527 .migratepage
= fail_migrate_page
,
1530 inode
= new_inode(bdev
->bd_inode
->i_sb
);
1533 "XFS: Cannot allocate mapping inode for device %s\n",
1534 XFS_BUFTARG_NAME(btp
));
1537 inode
->i_mode
= S_IFBLK
;
1538 inode
->i_bdev
= bdev
;
1539 inode
->i_rdev
= bdev
->bd_dev
;
1540 bdi
= blk_get_backing_dev_info(bdev
);
1542 bdi
= &default_backing_dev_info
;
1543 mapping
= &inode
->i_data
;
1544 mapping
->a_ops
= &mapping_aops
;
1545 mapping
->backing_dev_info
= bdi
;
1546 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
1547 btp
->bt_mapping
= mapping
;
1552 xfs_alloc_delwrite_queue(
1557 INIT_LIST_HEAD(&btp
->bt_list
);
1558 INIT_LIST_HEAD(&btp
->bt_delwrite_queue
);
1559 spinlock_init(&btp
->bt_delwrite_lock
, "delwri_lock");
1561 btp
->bt_task
= kthread_run(xfsbufd
, btp
, "xfsbufd");
1562 if (IS_ERR(btp
->bt_task
)) {
1563 error
= PTR_ERR(btp
->bt_task
);
1566 xfs_register_buftarg(btp
);
1573 struct block_device
*bdev
,
1578 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
);
1580 btp
->bt_dev
= bdev
->bd_dev
;
1581 btp
->bt_bdev
= bdev
;
1582 if (xfs_setsize_buftarg_early(btp
, bdev
))
1584 if (xfs_mapping_buftarg(btp
, bdev
))
1586 if (xfs_alloc_delwrite_queue(btp
))
1588 xfs_alloc_bufhash(btp
, external
);
1592 kmem_free(btp
, sizeof(*btp
));
1598 * Delayed write buffer handling
1601 xfs_buf_delwri_queue(
1605 struct list_head
*dwq
= &bp
->b_target
->bt_delwrite_queue
;
1606 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1608 XB_TRACE(bp
, "delwri_q", (long)unlock
);
1609 ASSERT((bp
->b_flags
&(XBF_DELWRI
|XBF_ASYNC
)) == (XBF_DELWRI
|XBF_ASYNC
));
1612 /* If already in the queue, dequeue and place at tail */
1613 if (!list_empty(&bp
->b_list
)) {
1614 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1616 atomic_dec(&bp
->b_hold
);
1617 list_del(&bp
->b_list
);
1620 bp
->b_flags
|= _XBF_DELWRI_Q
;
1621 list_add_tail(&bp
->b_list
, dwq
);
1622 bp
->b_queuetime
= jiffies
;
1630 xfs_buf_delwri_dequeue(
1633 spinlock_t
*dwlk
= &bp
->b_target
->bt_delwrite_lock
;
1637 if ((bp
->b_flags
& XBF_DELWRI
) && !list_empty(&bp
->b_list
)) {
1638 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
1639 list_del_init(&bp
->b_list
);
1642 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
);
1648 XB_TRACE(bp
, "delwri_dq", (long)dequeued
);
1652 xfs_buf_runall_queues(
1653 struct workqueue_struct
*queue
)
1655 flush_workqueue(queue
);
1665 spin_lock(&xfs_buftarg_lock
);
1666 list_for_each_entry(btp
, &xfs_buftarg_list
, bt_list
) {
1667 if (test_bit(XBT_FORCE_SLEEP
, &btp
->bt_flags
))
1669 set_bit(XBT_FORCE_FLUSH
, &btp
->bt_flags
);
1670 wake_up_process(btp
->bt_task
);
1672 spin_unlock(&xfs_buftarg_lock
);
1677 * Move as many buffers as specified to the supplied list
1678 * idicating if we skipped any buffers to prevent deadlocks.
1681 xfs_buf_delwri_split(
1682 xfs_buftarg_t
*target
,
1683 struct list_head
*list
,
1687 struct list_head
*dwq
= &target
->bt_delwrite_queue
;
1688 spinlock_t
*dwlk
= &target
->bt_delwrite_lock
;
1692 force
= test_and_clear_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1693 INIT_LIST_HEAD(list
);
1695 list_for_each_entry_safe(bp
, n
, dwq
, b_list
) {
1696 XB_TRACE(bp
, "walkq1", (long)xfs_buf_ispin(bp
));
1697 ASSERT(bp
->b_flags
& XBF_DELWRI
);
1699 if (!xfs_buf_ispin(bp
) && !xfs_buf_cond_lock(bp
)) {
1701 time_before(jiffies
, bp
->b_queuetime
+ age
)) {
1706 bp
->b_flags
&= ~(XBF_DELWRI
|_XBF_DELWRI_Q
|
1708 bp
->b_flags
|= XBF_WRITE
;
1709 list_move_tail(&bp
->b_list
, list
);
1723 struct list_head tmp
;
1724 xfs_buftarg_t
*target
= (xfs_buftarg_t
*)data
;
1728 current
->flags
|= PF_MEMALLOC
;
1731 if (unlikely(freezing(current
))) {
1732 set_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1735 clear_bit(XBT_FORCE_SLEEP
, &target
->bt_flags
);
1738 schedule_timeout_interruptible(
1739 xfs_buf_timer_centisecs
* msecs_to_jiffies(10));
1741 xfs_buf_delwri_split(target
, &tmp
,
1742 xfs_buf_age_centisecs
* msecs_to_jiffies(10));
1745 while (!list_empty(&tmp
)) {
1746 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1747 ASSERT(target
== bp
->b_target
);
1749 list_del_init(&bp
->b_list
);
1750 xfs_buf_iostrategy(bp
);
1754 if (as_list_len
> 0)
1757 blk_run_address_space(target
->bt_mapping
);
1759 } while (!kthread_should_stop());
1765 * Go through all incore buffers, and release buffers if they belong to
1766 * the given device. This is used in filesystem error handling to
1767 * preserve the consistency of its metadata.
1771 xfs_buftarg_t
*target
,
1774 struct list_head tmp
;
1778 xfs_buf_runall_queues(xfsdatad_workqueue
);
1779 xfs_buf_runall_queues(xfslogd_workqueue
);
1781 set_bit(XBT_FORCE_FLUSH
, &target
->bt_flags
);
1782 pincount
= xfs_buf_delwri_split(target
, &tmp
, 0);
1785 * Dropped the delayed write list lock, now walk the temporary list
1787 list_for_each_entry_safe(bp
, n
, &tmp
, b_list
) {
1788 ASSERT(target
== bp
->b_target
);
1790 bp
->b_flags
&= ~XBF_ASYNC
;
1792 list_del_init(&bp
->b_list
);
1794 xfs_buf_iostrategy(bp
);
1798 blk_run_address_space(target
->bt_mapping
);
1801 * Remaining list items must be flushed before returning
1803 while (!list_empty(&tmp
)) {
1804 bp
= list_entry(tmp
.next
, xfs_buf_t
, b_list
);
1806 list_del_init(&bp
->b_list
);
1817 #ifdef XFS_BUF_TRACE
1818 xfs_buf_trace_buf
= ktrace_alloc(XFS_BUF_TRACE_SIZE
, KM_SLEEP
);
1821 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
1822 KM_ZONE_HWALIGN
, NULL
);
1824 goto out_free_trace_buf
;
1826 xfslogd_workqueue
= create_workqueue("xfslogd");
1827 if (!xfslogd_workqueue
)
1828 goto out_free_buf_zone
;
1830 xfsdatad_workqueue
= create_workqueue("xfsdatad");
1831 if (!xfsdatad_workqueue
)
1832 goto out_destroy_xfslogd_workqueue
;
1834 register_shrinker(&xfs_buf_shake
);
1837 out_destroy_xfslogd_workqueue
:
1838 destroy_workqueue(xfslogd_workqueue
);
1840 kmem_zone_destroy(xfs_buf_zone
);
1842 #ifdef XFS_BUF_TRACE
1843 ktrace_free(xfs_buf_trace_buf
);
1849 xfs_buf_terminate(void)
1851 unregister_shrinker(&xfs_buf_shake
);
1852 destroy_workqueue(xfsdatad_workqueue
);
1853 destroy_workqueue(xfslogd_workqueue
);
1854 kmem_zone_destroy(xfs_buf_zone
);
1855 #ifdef XFS_BUF_TRACE
1856 ktrace_free(xfs_buf_trace_buf
);
1860 #ifdef CONFIG_KDB_MODULES
1862 xfs_get_buftarg_list(void)
1864 return &xfs_buftarg_list
;