1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
7 #include <linux/stddef.h>
8 #include <linux/errno.h>
10 #include <linux/pagemap.h>
11 #include <linux/init.h>
12 #include <linux/vmalloc.h>
13 #include <linux/bio.h>
14 #include <linux/sysctl.h>
15 #include <linux/proc_fs.h>
16 #include <linux/workqueue.h>
17 #include <linux/percpu.h>
18 #include <linux/blkdev.h>
19 #include <linux/hash.h>
20 #include <linux/kthread.h>
21 #include <linux/migrate.h>
22 #include <linux/backing-dev.h>
23 #include <linux/freezer.h>
25 #include "xfs_format.h"
26 #include "xfs_log_format.h"
27 #include "xfs_trans_resv.h"
29 #include "xfs_mount.h"
30 #include "xfs_trace.h"
32 #include "xfs_errortag.h"
33 #include "xfs_error.h"
35 static kmem_zone_t
*xfs_buf_zone
;
37 #define xb_to_gfp(flags) \
38 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN)
45 * b_sema (caller holds)
49 * b_sema (caller holds)
58 * xfs_buftarg_wait_rele
60 * b_lock (trylock due to inversion)
64 * b_lock (trylock due to inversion)
72 * Return true if the buffer is vmapped.
74 * b_addr is null if the buffer is not mapped, but the code is clever
75 * enough to know it doesn't have to map a single page, so the check has
76 * to be both for b_addr and bp->b_page_count > 1.
78 return bp
->b_addr
&& bp
->b_page_count
> 1;
85 return (bp
->b_page_count
* PAGE_SIZE
) - bp
->b_offset
;
89 * Bump the I/O in flight count on the buftarg if we haven't yet done so for
90 * this buffer. The count is incremented once per buffer (per hold cycle)
91 * because the corresponding decrement is deferred to buffer release. Buffers
92 * can undergo I/O multiple times in a hold-release cycle and per buffer I/O
93 * tracking adds unnecessary overhead. This is used for sychronization purposes
94 * with unmount (see xfs_wait_buftarg()), so all we really need is a count of
97 * Buffers that are never released (e.g., superblock, iclog buffers) must set
98 * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count
99 * never reaches zero and unmount hangs indefinitely.
105 if (bp
->b_flags
& XBF_NO_IOACCT
)
108 ASSERT(bp
->b_flags
& XBF_ASYNC
);
109 spin_lock(&bp
->b_lock
);
110 if (!(bp
->b_state
& XFS_BSTATE_IN_FLIGHT
)) {
111 bp
->b_state
|= XFS_BSTATE_IN_FLIGHT
;
112 percpu_counter_inc(&bp
->b_target
->bt_io_count
);
114 spin_unlock(&bp
->b_lock
);
118 * Clear the in-flight state on a buffer about to be released to the LRU or
119 * freed and unaccount from the buftarg.
122 __xfs_buf_ioacct_dec(
125 lockdep_assert_held(&bp
->b_lock
);
127 if (bp
->b_state
& XFS_BSTATE_IN_FLIGHT
) {
128 bp
->b_state
&= ~XFS_BSTATE_IN_FLIGHT
;
129 percpu_counter_dec(&bp
->b_target
->bt_io_count
);
137 spin_lock(&bp
->b_lock
);
138 __xfs_buf_ioacct_dec(bp
);
139 spin_unlock(&bp
->b_lock
);
143 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
144 * b_lru_ref count so that the buffer is freed immediately when the buffer
145 * reference count falls to zero. If the buffer is already on the LRU, we need
146 * to remove the reference that LRU holds on the buffer.
148 * This prevents build-up of stale buffers on the LRU.
154 ASSERT(xfs_buf_islocked(bp
));
156 bp
->b_flags
|= XBF_STALE
;
159 * Clear the delwri status so that a delwri queue walker will not
160 * flush this buffer to disk now that it is stale. The delwri queue has
161 * a reference to the buffer, so this is safe to do.
163 bp
->b_flags
&= ~_XBF_DELWRI_Q
;
166 * Once the buffer is marked stale and unlocked, a subsequent lookup
167 * could reset b_flags. There is no guarantee that the buffer is
168 * unaccounted (released to LRU) before that occurs. Drop in-flight
169 * status now to preserve accounting consistency.
171 spin_lock(&bp
->b_lock
);
172 __xfs_buf_ioacct_dec(bp
);
174 atomic_set(&bp
->b_lru_ref
, 0);
175 if (!(bp
->b_state
& XFS_BSTATE_DISPOSE
) &&
176 (list_lru_del(&bp
->b_target
->bt_lru
, &bp
->b_lru
)))
177 atomic_dec(&bp
->b_hold
);
179 ASSERT(atomic_read(&bp
->b_hold
) >= 1);
180 spin_unlock(&bp
->b_lock
);
188 ASSERT(bp
->b_maps
== NULL
);
189 bp
->b_map_count
= map_count
;
191 if (map_count
== 1) {
192 bp
->b_maps
= &bp
->__b_map
;
196 bp
->b_maps
= kmem_zalloc(map_count
* sizeof(struct xfs_buf_map
),
204 * Frees b_pages if it was allocated.
210 if (bp
->b_maps
!= &bp
->__b_map
) {
211 kmem_free(bp
->b_maps
);
218 struct xfs_buftarg
*target
,
219 struct xfs_buf_map
*map
,
221 xfs_buf_flags_t flags
)
227 bp
= kmem_zone_zalloc(xfs_buf_zone
, KM_NOFS
);
232 * We don't want certain flags to appear in b_flags unless they are
233 * specifically set by later operations on the buffer.
235 flags
&= ~(XBF_UNMAPPED
| XBF_TRYLOCK
| XBF_ASYNC
| XBF_READ_AHEAD
);
237 atomic_set(&bp
->b_hold
, 1);
238 atomic_set(&bp
->b_lru_ref
, 1);
239 init_completion(&bp
->b_iowait
);
240 INIT_LIST_HEAD(&bp
->b_lru
);
241 INIT_LIST_HEAD(&bp
->b_list
);
242 INIT_LIST_HEAD(&bp
->b_li_list
);
243 sema_init(&bp
->b_sema
, 0); /* held, no waiters */
244 spin_lock_init(&bp
->b_lock
);
245 bp
->b_target
= target
;
249 * Set length and io_length to the same value initially.
250 * I/O routines should use io_length, which will be the same in
251 * most cases but may be reset (e.g. XFS recovery).
253 error
= xfs_buf_get_maps(bp
, nmaps
);
255 kmem_zone_free(xfs_buf_zone
, bp
);
259 bp
->b_bn
= map
[0].bm_bn
;
261 for (i
= 0; i
< nmaps
; i
++) {
262 bp
->b_maps
[i
].bm_bn
= map
[i
].bm_bn
;
263 bp
->b_maps
[i
].bm_len
= map
[i
].bm_len
;
264 bp
->b_length
+= map
[i
].bm_len
;
266 bp
->b_io_length
= bp
->b_length
;
268 atomic_set(&bp
->b_pin_count
, 0);
269 init_waitqueue_head(&bp
->b_waiters
);
271 XFS_STATS_INC(target
->bt_mount
, xb_create
);
272 trace_xfs_buf_init(bp
, _RET_IP_
);
278 * Allocate a page array capable of holding a specified number
279 * of pages, and point the page buf at it.
286 /* Make sure that we have a page list */
287 if (bp
->b_pages
== NULL
) {
288 bp
->b_page_count
= page_count
;
289 if (page_count
<= XB_PAGES
) {
290 bp
->b_pages
= bp
->b_page_array
;
292 bp
->b_pages
= kmem_alloc(sizeof(struct page
*) *
293 page_count
, KM_NOFS
);
294 if (bp
->b_pages
== NULL
)
297 memset(bp
->b_pages
, 0, sizeof(struct page
*) * page_count
);
303 * Frees b_pages if it was allocated.
309 if (bp
->b_pages
!= bp
->b_page_array
) {
310 kmem_free(bp
->b_pages
);
316 * Releases the specified buffer.
318 * The modification state of any associated pages is left unchanged.
319 * The buffer must not be on any hash - use xfs_buf_rele instead for
320 * hashed and refcounted buffers
326 trace_xfs_buf_free(bp
, _RET_IP_
);
328 ASSERT(list_empty(&bp
->b_lru
));
330 if (bp
->b_flags
& _XBF_PAGES
) {
333 if (xfs_buf_is_vmapped(bp
))
334 vm_unmap_ram(bp
->b_addr
- bp
->b_offset
,
337 for (i
= 0; i
< bp
->b_page_count
; i
++) {
338 struct page
*page
= bp
->b_pages
[i
];
342 } else if (bp
->b_flags
& _XBF_KMEM
)
343 kmem_free(bp
->b_addr
);
344 _xfs_buf_free_pages(bp
);
345 xfs_buf_free_maps(bp
);
346 kmem_zone_free(xfs_buf_zone
, bp
);
350 * Allocates all the pages for buffer in question and builds it's page list.
353 xfs_buf_allocate_memory(
358 size_t nbytes
, offset
;
359 gfp_t gfp_mask
= xb_to_gfp(flags
);
360 unsigned short page_count
, i
;
361 xfs_off_t start
, end
;
365 * for buffers that are contained within a single page, just allocate
366 * the memory from the heap - there's no need for the complexity of
367 * page arrays to keep allocation down to order 0.
369 size
= BBTOB(bp
->b_length
);
370 if (size
< PAGE_SIZE
) {
371 bp
->b_addr
= kmem_alloc(size
, KM_NOFS
);
373 /* low memory - use alloc_page loop instead */
377 if (((unsigned long)(bp
->b_addr
+ size
- 1) & PAGE_MASK
) !=
378 ((unsigned long)bp
->b_addr
& PAGE_MASK
)) {
379 /* b_addr spans two pages - use alloc_page instead */
380 kmem_free(bp
->b_addr
);
384 bp
->b_offset
= offset_in_page(bp
->b_addr
);
385 bp
->b_pages
= bp
->b_page_array
;
386 bp
->b_pages
[0] = virt_to_page(bp
->b_addr
);
387 bp
->b_page_count
= 1;
388 bp
->b_flags
|= _XBF_KMEM
;
393 start
= BBTOB(bp
->b_maps
[0].bm_bn
) >> PAGE_SHIFT
;
394 end
= (BBTOB(bp
->b_maps
[0].bm_bn
+ bp
->b_length
) + PAGE_SIZE
- 1)
396 page_count
= end
- start
;
397 error
= _xfs_buf_get_pages(bp
, page_count
);
401 offset
= bp
->b_offset
;
402 bp
->b_flags
|= _XBF_PAGES
;
404 for (i
= 0; i
< bp
->b_page_count
; i
++) {
408 page
= alloc_page(gfp_mask
);
409 if (unlikely(page
== NULL
)) {
410 if (flags
& XBF_READ_AHEAD
) {
411 bp
->b_page_count
= i
;
417 * This could deadlock.
419 * But until all the XFS lowlevel code is revamped to
420 * handle buffer allocation failures we can't do much.
422 if (!(++retries
% 100))
424 "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)",
425 current
->comm
, current
->pid
,
428 XFS_STATS_INC(bp
->b_target
->bt_mount
, xb_page_retries
);
429 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
433 XFS_STATS_INC(bp
->b_target
->bt_mount
, xb_page_found
);
435 nbytes
= min_t(size_t, size
, PAGE_SIZE
- offset
);
437 bp
->b_pages
[i
] = page
;
443 for (i
= 0; i
< bp
->b_page_count
; i
++)
444 __free_page(bp
->b_pages
[i
]);
445 bp
->b_flags
&= ~_XBF_PAGES
;
450 * Map buffer into kernel address-space if necessary.
457 ASSERT(bp
->b_flags
& _XBF_PAGES
);
458 if (bp
->b_page_count
== 1) {
459 /* A single page buffer is always mappable */
460 bp
->b_addr
= page_address(bp
->b_pages
[0]) + bp
->b_offset
;
461 } else if (flags
& XBF_UNMAPPED
) {
468 * vm_map_ram() will allocate auxillary structures (e.g.
469 * pagetables) with GFP_KERNEL, yet we are likely to be under
470 * GFP_NOFS context here. Hence we need to tell memory reclaim
471 * that we are in such a context via PF_MEMALLOC_NOFS to prevent
472 * memory reclaim re-entering the filesystem here and
473 * potentially deadlocking.
475 nofs_flag
= memalloc_nofs_save();
477 bp
->b_addr
= vm_map_ram(bp
->b_pages
, bp
->b_page_count
,
482 } while (retried
++ <= 1);
483 memalloc_nofs_restore(nofs_flag
);
487 bp
->b_addr
+= bp
->b_offset
;
494 * Finding and Reading Buffers
498 struct rhashtable_compare_arg
*arg
,
501 const struct xfs_buf_map
*map
= arg
->key
;
502 const struct xfs_buf
*bp
= obj
;
505 * The key hashing in the lookup path depends on the key being the
506 * first element of the compare_arg, make sure to assert this.
508 BUILD_BUG_ON(offsetof(struct xfs_buf_map
, bm_bn
) != 0);
510 if (bp
->b_bn
!= map
->bm_bn
)
513 if (unlikely(bp
->b_length
!= map
->bm_len
)) {
515 * found a block number match. If the range doesn't
516 * match, the only way this is allowed is if the buffer
517 * in the cache is stale and the transaction that made
518 * it stale has not yet committed. i.e. we are
519 * reallocating a busy extent. Skip this buffer and
520 * continue searching for an exact match.
522 ASSERT(bp
->b_flags
& XBF_STALE
);
528 static const struct rhashtable_params xfs_buf_hash_params
= {
529 .min_size
= 32, /* empty AGs have minimal footprint */
531 .key_len
= sizeof(xfs_daddr_t
),
532 .key_offset
= offsetof(struct xfs_buf
, b_bn
),
533 .head_offset
= offsetof(struct xfs_buf
, b_rhash_head
),
534 .automatic_shrinking
= true,
535 .obj_cmpfn
= _xfs_buf_obj_cmp
,
540 struct xfs_perag
*pag
)
542 spin_lock_init(&pag
->pag_buf_lock
);
543 return rhashtable_init(&pag
->pag_buf_hash
, &xfs_buf_hash_params
);
547 xfs_buf_hash_destroy(
548 struct xfs_perag
*pag
)
550 rhashtable_destroy(&pag
->pag_buf_hash
);
554 * Look up a buffer in the buffer cache and return it referenced and locked
557 * If @new_bp is supplied and we have a lookup miss, insert @new_bp into the
560 * If XBF_TRYLOCK is set in @flags, only try to lock the buffer and return
561 * -EAGAIN if we fail to lock it.
564 * -EFSCORRUPTED if have been supplied with an invalid address
565 * -EAGAIN on trylock failure
566 * -ENOENT if we fail to find a match and @new_bp was NULL
568 * - @new_bp if we inserted it into the cache
569 * - the buffer we found and locked.
573 struct xfs_buftarg
*btp
,
574 struct xfs_buf_map
*map
,
576 xfs_buf_flags_t flags
,
577 struct xfs_buf
*new_bp
,
578 struct xfs_buf
**found_bp
)
580 struct xfs_perag
*pag
;
582 struct xfs_buf_map cmap
= { .bm_bn
= map
[0].bm_bn
};
588 for (i
= 0; i
< nmaps
; i
++)
589 cmap
.bm_len
+= map
[i
].bm_len
;
591 /* Check for IOs smaller than the sector size / not sector aligned */
592 ASSERT(!(BBTOB(cmap
.bm_len
) < btp
->bt_meta_sectorsize
));
593 ASSERT(!(BBTOB(cmap
.bm_bn
) & (xfs_off_t
)btp
->bt_meta_sectormask
));
596 * Corrupted block numbers can get through to here, unfortunately, so we
597 * have to check that the buffer falls within the filesystem bounds.
599 eofs
= XFS_FSB_TO_BB(btp
->bt_mount
, btp
->bt_mount
->m_sb
.sb_dblocks
);
600 if (cmap
.bm_bn
< 0 || cmap
.bm_bn
>= eofs
) {
601 xfs_alert(btp
->bt_mount
,
602 "%s: daddr 0x%llx out of range, EOFS 0x%llx",
603 __func__
, cmap
.bm_bn
, eofs
);
605 return -EFSCORRUPTED
;
608 pag
= xfs_perag_get(btp
->bt_mount
,
609 xfs_daddr_to_agno(btp
->bt_mount
, cmap
.bm_bn
));
611 spin_lock(&pag
->pag_buf_lock
);
612 bp
= rhashtable_lookup_fast(&pag
->pag_buf_hash
, &cmap
,
613 xfs_buf_hash_params
);
615 atomic_inc(&bp
->b_hold
);
621 XFS_STATS_INC(btp
->bt_mount
, xb_miss_locked
);
622 spin_unlock(&pag
->pag_buf_lock
);
627 /* the buffer keeps the perag reference until it is freed */
629 rhashtable_insert_fast(&pag
->pag_buf_hash
, &new_bp
->b_rhash_head
,
630 xfs_buf_hash_params
);
631 spin_unlock(&pag
->pag_buf_lock
);
636 spin_unlock(&pag
->pag_buf_lock
);
639 if (!xfs_buf_trylock(bp
)) {
640 if (flags
& XBF_TRYLOCK
) {
642 XFS_STATS_INC(btp
->bt_mount
, xb_busy_locked
);
646 XFS_STATS_INC(btp
->bt_mount
, xb_get_locked_waited
);
650 * if the buffer is stale, clear all the external state associated with
651 * it. We need to keep flags such as how we allocated the buffer memory
654 if (bp
->b_flags
& XBF_STALE
) {
655 ASSERT((bp
->b_flags
& _XBF_DELWRI_Q
) == 0);
656 ASSERT(bp
->b_iodone
== NULL
);
657 bp
->b_flags
&= _XBF_KMEM
| _XBF_PAGES
;
661 trace_xfs_buf_find(bp
, flags
, _RET_IP_
);
662 XFS_STATS_INC(btp
->bt_mount
, xb_get_locked
);
669 struct xfs_buftarg
*target
,
672 xfs_buf_flags_t flags
)
676 DEFINE_SINGLE_BUF_MAP(map
, blkno
, numblks
);
678 error
= xfs_buf_find(target
, &map
, 1, flags
, NULL
, &bp
);
685 * Assembles a buffer covering the specified range. The code is optimised for
686 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
687 * more hits than misses.
691 struct xfs_buftarg
*target
,
692 struct xfs_buf_map
*map
,
694 xfs_buf_flags_t flags
)
697 struct xfs_buf
*new_bp
;
700 error
= xfs_buf_find(target
, map
, nmaps
, flags
, NULL
, &bp
);
707 /* cache hit, trylock failure, caller handles failure */
708 ASSERT(flags
& XBF_TRYLOCK
);
711 /* cache miss, go for insert */
716 * None of the higher layers understand failure types
717 * yet, so return NULL to signal a fatal lookup error.
722 new_bp
= _xfs_buf_alloc(target
, map
, nmaps
, flags
);
723 if (unlikely(!new_bp
))
726 error
= xfs_buf_allocate_memory(new_bp
, flags
);
728 xfs_buf_free(new_bp
);
732 error
= xfs_buf_find(target
, map
, nmaps
, flags
, new_bp
, &bp
);
734 xfs_buf_free(new_bp
);
739 xfs_buf_free(new_bp
);
743 error
= _xfs_buf_map_pages(bp
, flags
);
744 if (unlikely(error
)) {
745 xfs_warn(target
->bt_mount
,
746 "%s: failed to map pagesn", __func__
);
753 * Clear b_error if this is a lookup from a caller that doesn't expect
754 * valid data to be found in the buffer.
756 if (!(flags
& XBF_READ
))
757 xfs_buf_ioerror(bp
, 0);
759 XFS_STATS_INC(target
->bt_mount
, xb_get
);
760 trace_xfs_buf_get(bp
, flags
, _RET_IP_
);
767 xfs_buf_flags_t flags
)
769 ASSERT(!(flags
& XBF_WRITE
));
770 ASSERT(bp
->b_maps
[0].bm_bn
!= XFS_BUF_DADDR_NULL
);
772 bp
->b_flags
&= ~(XBF_WRITE
| XBF_ASYNC
| XBF_READ_AHEAD
);
773 bp
->b_flags
|= flags
& (XBF_READ
| XBF_ASYNC
| XBF_READ_AHEAD
);
775 return xfs_buf_submit(bp
);
779 * If the caller passed in an ops structure and the buffer doesn't have ops
780 * assigned, set the ops and use them to verify the contents. If the contents
781 * cannot be verified, we'll clear XBF_DONE. We assume the buffer has no
782 * recorded errors and is already in XBF_DONE state.
787 const struct xfs_buf_ops
*ops
)
789 ASSERT(bp
->b_flags
& XBF_DONE
);
790 ASSERT(bp
->b_error
== 0);
792 if (!ops
|| bp
->b_ops
)
796 bp
->b_ops
->verify_read(bp
);
798 bp
->b_flags
&= ~XBF_DONE
;
804 struct xfs_buftarg
*target
,
805 struct xfs_buf_map
*map
,
807 xfs_buf_flags_t flags
,
808 const struct xfs_buf_ops
*ops
)
814 bp
= xfs_buf_get_map(target
, map
, nmaps
, flags
);
818 trace_xfs_buf_read(bp
, flags
, _RET_IP_
);
820 if (!(bp
->b_flags
& XBF_DONE
)) {
821 XFS_STATS_INC(target
->bt_mount
, xb_get_read
);
823 _xfs_buf_read(bp
, flags
);
827 xfs_buf_ensure_ops(bp
, ops
);
829 if (flags
& XBF_ASYNC
) {
831 * Read ahead call which is already satisfied,
838 /* We do not want read in the flags */
839 bp
->b_flags
&= ~XBF_READ
;
840 ASSERT(bp
->b_ops
!= NULL
|| ops
== NULL
);
845 * If we are not low on memory then do the readahead in a deadlock
849 xfs_buf_readahead_map(
850 struct xfs_buftarg
*target
,
851 struct xfs_buf_map
*map
,
853 const struct xfs_buf_ops
*ops
)
855 if (bdi_read_congested(target
->bt_bdev
->bd_bdi
))
858 xfs_buf_read_map(target
, map
, nmaps
,
859 XBF_TRYLOCK
|XBF_ASYNC
|XBF_READ_AHEAD
, ops
);
863 * Read an uncached buffer from disk. Allocates and returns a locked
864 * buffer containing the disk contents or nothing.
867 xfs_buf_read_uncached(
868 struct xfs_buftarg
*target
,
872 struct xfs_buf
**bpp
,
873 const struct xfs_buf_ops
*ops
)
879 bp
= xfs_buf_get_uncached(target
, numblks
, flags
);
883 /* set up the buffer for a read IO */
884 ASSERT(bp
->b_map_count
== 1);
885 bp
->b_bn
= XFS_BUF_DADDR_NULL
; /* always null for uncached buffers */
886 bp
->b_maps
[0].bm_bn
= daddr
;
887 bp
->b_flags
|= XBF_READ
;
892 int error
= bp
->b_error
;
902 * Return a buffer allocated as an empty buffer and associated to external
903 * memory via xfs_buf_associate_memory() back to it's empty state.
911 _xfs_buf_free_pages(bp
);
914 bp
->b_page_count
= 0;
916 bp
->b_length
= numblks
;
917 bp
->b_io_length
= numblks
;
919 ASSERT(bp
->b_map_count
== 1);
920 bp
->b_bn
= XFS_BUF_DADDR_NULL
;
921 bp
->b_maps
[0].bm_bn
= XFS_BUF_DADDR_NULL
;
922 bp
->b_maps
[0].bm_len
= bp
->b_length
;
925 static inline struct page
*
929 if ((!is_vmalloc_addr(addr
))) {
930 return virt_to_page(addr
);
932 return vmalloc_to_page(addr
);
937 xfs_buf_associate_memory(
944 unsigned long pageaddr
;
945 unsigned long offset
;
949 pageaddr
= (unsigned long)mem
& PAGE_MASK
;
950 offset
= (unsigned long)mem
- pageaddr
;
951 buflen
= PAGE_ALIGN(len
+ offset
);
952 page_count
= buflen
>> PAGE_SHIFT
;
954 /* Free any previous set of page pointers */
956 _xfs_buf_free_pages(bp
);
961 rval
= _xfs_buf_get_pages(bp
, page_count
);
965 bp
->b_offset
= offset
;
967 for (i
= 0; i
< bp
->b_page_count
; i
++) {
968 bp
->b_pages
[i
] = mem_to_page((void *)pageaddr
);
969 pageaddr
+= PAGE_SIZE
;
972 bp
->b_io_length
= BTOBB(len
);
973 bp
->b_length
= BTOBB(buflen
);
979 xfs_buf_get_uncached(
980 struct xfs_buftarg
*target
,
984 unsigned long page_count
;
987 DEFINE_SINGLE_BUF_MAP(map
, XFS_BUF_DADDR_NULL
, numblks
);
989 /* flags might contain irrelevant bits, pass only what we care about */
990 bp
= _xfs_buf_alloc(target
, &map
, 1, flags
& XBF_NO_IOACCT
);
991 if (unlikely(bp
== NULL
))
994 page_count
= PAGE_ALIGN(numblks
<< BBSHIFT
) >> PAGE_SHIFT
;
995 error
= _xfs_buf_get_pages(bp
, page_count
);
999 for (i
= 0; i
< page_count
; i
++) {
1000 bp
->b_pages
[i
] = alloc_page(xb_to_gfp(flags
));
1001 if (!bp
->b_pages
[i
])
1004 bp
->b_flags
|= _XBF_PAGES
;
1006 error
= _xfs_buf_map_pages(bp
, 0);
1007 if (unlikely(error
)) {
1008 xfs_warn(target
->bt_mount
,
1009 "%s: failed to map pages", __func__
);
1013 trace_xfs_buf_get_uncached(bp
, _RET_IP_
);
1018 __free_page(bp
->b_pages
[i
]);
1019 _xfs_buf_free_pages(bp
);
1021 xfs_buf_free_maps(bp
);
1022 kmem_zone_free(xfs_buf_zone
, bp
);
1028 * Increment reference count on buffer, to hold the buffer concurrently
1029 * with another thread which may release (free) the buffer asynchronously.
1030 * Must hold the buffer already to call this function.
1036 trace_xfs_buf_hold(bp
, _RET_IP_
);
1037 atomic_inc(&bp
->b_hold
);
1041 * Release a hold on the specified buffer. If the hold count is 1, the buffer is
1042 * placed on LRU or freed (depending on b_lru_ref).
1048 struct xfs_perag
*pag
= bp
->b_pag
;
1050 bool freebuf
= false;
1052 trace_xfs_buf_rele(bp
, _RET_IP_
);
1055 ASSERT(list_empty(&bp
->b_lru
));
1056 if (atomic_dec_and_test(&bp
->b_hold
)) {
1057 xfs_buf_ioacct_dec(bp
);
1063 ASSERT(atomic_read(&bp
->b_hold
) > 0);
1066 * We grab the b_lock here first to serialise racing xfs_buf_rele()
1067 * calls. The pag_buf_lock being taken on the last reference only
1068 * serialises against racing lookups in xfs_buf_find(). IOWs, the second
1069 * to last reference we drop here is not serialised against the last
1070 * reference until we take bp->b_lock. Hence if we don't grab b_lock
1071 * first, the last "release" reference can win the race to the lock and
1072 * free the buffer before the second-to-last reference is processed,
1073 * leading to a use-after-free scenario.
1075 spin_lock(&bp
->b_lock
);
1076 release
= atomic_dec_and_lock(&bp
->b_hold
, &pag
->pag_buf_lock
);
1079 * Drop the in-flight state if the buffer is already on the LRU
1080 * and it holds the only reference. This is racy because we
1081 * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT
1082 * ensures the decrement occurs only once per-buf.
1084 if ((atomic_read(&bp
->b_hold
) == 1) && !list_empty(&bp
->b_lru
))
1085 __xfs_buf_ioacct_dec(bp
);
1089 /* the last reference has been dropped ... */
1090 __xfs_buf_ioacct_dec(bp
);
1091 if (!(bp
->b_flags
& XBF_STALE
) && atomic_read(&bp
->b_lru_ref
)) {
1093 * If the buffer is added to the LRU take a new reference to the
1094 * buffer for the LRU and clear the (now stale) dispose list
1097 if (list_lru_add(&bp
->b_target
->bt_lru
, &bp
->b_lru
)) {
1098 bp
->b_state
&= ~XFS_BSTATE_DISPOSE
;
1099 atomic_inc(&bp
->b_hold
);
1101 spin_unlock(&pag
->pag_buf_lock
);
1104 * most of the time buffers will already be removed from the
1105 * LRU, so optimise that case by checking for the
1106 * XFS_BSTATE_DISPOSE flag indicating the last list the buffer
1107 * was on was the disposal list
1109 if (!(bp
->b_state
& XFS_BSTATE_DISPOSE
)) {
1110 list_lru_del(&bp
->b_target
->bt_lru
, &bp
->b_lru
);
1112 ASSERT(list_empty(&bp
->b_lru
));
1115 ASSERT(!(bp
->b_flags
& _XBF_DELWRI_Q
));
1116 rhashtable_remove_fast(&pag
->pag_buf_hash
, &bp
->b_rhash_head
,
1117 xfs_buf_hash_params
);
1118 spin_unlock(&pag
->pag_buf_lock
);
1124 spin_unlock(&bp
->b_lock
);
1132 * Lock a buffer object, if it is not already locked.
1134 * If we come across a stale, pinned, locked buffer, we know that we are
1135 * being asked to lock a buffer that has been reallocated. Because it is
1136 * pinned, we know that the log has not been pushed to disk and hence it
1137 * will still be locked. Rather than continuing to have trylock attempts
1138 * fail until someone else pushes the log, push it ourselves before
1139 * returning. This means that the xfsaild will not get stuck trying
1140 * to push on stale inode buffers.
1148 locked
= down_trylock(&bp
->b_sema
) == 0;
1150 trace_xfs_buf_trylock(bp
, _RET_IP_
);
1152 trace_xfs_buf_trylock_fail(bp
, _RET_IP_
);
1157 * Lock a buffer object.
1159 * If we come across a stale, pinned, locked buffer, we know that we
1160 * are being asked to lock a buffer that has been reallocated. Because
1161 * it is pinned, we know that the log has not been pushed to disk and
1162 * hence it will still be locked. Rather than sleeping until someone
1163 * else pushes the log, push it ourselves before trying to get the lock.
1169 trace_xfs_buf_lock(bp
, _RET_IP_
);
1171 if (atomic_read(&bp
->b_pin_count
) && (bp
->b_flags
& XBF_STALE
))
1172 xfs_log_force(bp
->b_target
->bt_mount
, 0);
1175 trace_xfs_buf_lock_done(bp
, _RET_IP_
);
1182 ASSERT(xfs_buf_islocked(bp
));
1185 trace_xfs_buf_unlock(bp
, _RET_IP_
);
1192 DECLARE_WAITQUEUE (wait
, current
);
1194 if (atomic_read(&bp
->b_pin_count
) == 0)
1197 add_wait_queue(&bp
->b_waiters
, &wait
);
1199 set_current_state(TASK_UNINTERRUPTIBLE
);
1200 if (atomic_read(&bp
->b_pin_count
) == 0)
1204 remove_wait_queue(&bp
->b_waiters
, &wait
);
1205 set_current_state(TASK_RUNNING
);
1209 * Buffer Utility Routines
1216 bool read
= bp
->b_flags
& XBF_READ
;
1218 trace_xfs_buf_iodone(bp
, _RET_IP_
);
1220 bp
->b_flags
&= ~(XBF_READ
| XBF_WRITE
| XBF_READ_AHEAD
);
1223 * Pull in IO completion errors now. We are guaranteed to be running
1224 * single threaded, so we don't need the lock to read b_io_error.
1226 if (!bp
->b_error
&& bp
->b_io_error
)
1227 xfs_buf_ioerror(bp
, bp
->b_io_error
);
1229 /* Only validate buffers that were read without errors */
1230 if (read
&& !bp
->b_error
&& bp
->b_ops
) {
1231 ASSERT(!bp
->b_iodone
);
1232 bp
->b_ops
->verify_read(bp
);
1236 bp
->b_flags
|= XBF_DONE
;
1239 (*(bp
->b_iodone
))(bp
);
1240 else if (bp
->b_flags
& XBF_ASYNC
)
1243 complete(&bp
->b_iowait
);
1248 struct work_struct
*work
)
1250 struct xfs_buf
*bp
=
1251 container_of(work
, xfs_buf_t
, b_ioend_work
);
1257 xfs_buf_ioend_async(
1260 INIT_WORK(&bp
->b_ioend_work
, xfs_buf_ioend_work
);
1261 queue_work(bp
->b_ioend_wq
, &bp
->b_ioend_work
);
1268 xfs_failaddr_t failaddr
)
1270 ASSERT(error
<= 0 && error
>= -1000);
1271 bp
->b_error
= error
;
1272 trace_xfs_buf_ioerror(bp
, error
, failaddr
);
1276 xfs_buf_ioerror_alert(
1280 xfs_alert(bp
->b_target
->bt_mount
,
1281 "metadata I/O error in \"%s\" at daddr 0x%llx len %d error %d",
1282 func
, (uint64_t)XFS_BUF_ADDR(bp
), bp
->b_length
,
1292 ASSERT(xfs_buf_islocked(bp
));
1294 bp
->b_flags
|= XBF_WRITE
;
1295 bp
->b_flags
&= ~(XBF_ASYNC
| XBF_READ
| _XBF_DELWRI_Q
|
1296 XBF_WRITE_FAIL
| XBF_DONE
);
1298 error
= xfs_buf_submit(bp
);
1300 xfs_force_shutdown(bp
->b_target
->bt_mount
,
1301 SHUTDOWN_META_IO_ERROR
);
1310 struct xfs_buf
*bp
= (struct xfs_buf
*)bio
->bi_private
;
1313 * don't overwrite existing errors - otherwise we can lose errors on
1314 * buffers that require multiple bios to complete.
1316 if (bio
->bi_status
) {
1317 int error
= blk_status_to_errno(bio
->bi_status
);
1319 cmpxchg(&bp
->b_io_error
, 0, error
);
1322 if (!bp
->b_error
&& xfs_buf_is_vmapped(bp
) && (bp
->b_flags
& XBF_READ
))
1323 invalidate_kernel_vmap_range(bp
->b_addr
, xfs_buf_vmap_len(bp
));
1325 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1)
1326 xfs_buf_ioend_async(bp
);
1331 xfs_buf_ioapply_map(
1340 int total_nr_pages
= bp
->b_page_count
;
1343 sector_t sector
= bp
->b_maps
[map
].bm_bn
;
1347 /* skip the pages in the buffer before the start offset */
1349 offset
= *buf_offset
;
1350 while (offset
>= PAGE_SIZE
) {
1352 offset
-= PAGE_SIZE
;
1356 * Limit the IO size to the length of the current vector, and update the
1357 * remaining IO count for the next time around.
1359 size
= min_t(int, BBTOB(bp
->b_maps
[map
].bm_len
), *count
);
1361 *buf_offset
+= size
;
1364 atomic_inc(&bp
->b_io_remaining
);
1365 nr_pages
= min(total_nr_pages
, BIO_MAX_PAGES
);
1367 bio
= bio_alloc(GFP_NOIO
, nr_pages
);
1368 bio_set_dev(bio
, bp
->b_target
->bt_bdev
);
1369 bio
->bi_iter
.bi_sector
= sector
;
1370 bio
->bi_end_io
= xfs_buf_bio_end_io
;
1371 bio
->bi_private
= bp
;
1372 bio_set_op_attrs(bio
, op
, op_flags
);
1374 for (; size
&& nr_pages
; nr_pages
--, page_index
++) {
1375 int rbytes
, nbytes
= PAGE_SIZE
- offset
;
1380 rbytes
= bio_add_page(bio
, bp
->b_pages
[page_index
], nbytes
,
1382 if (rbytes
< nbytes
)
1386 sector
+= BTOBB(nbytes
);
1391 if (likely(bio
->bi_iter
.bi_size
)) {
1392 if (xfs_buf_is_vmapped(bp
)) {
1393 flush_kernel_vmap_range(bp
->b_addr
,
1394 xfs_buf_vmap_len(bp
));
1401 * This is guaranteed not to be the last io reference count
1402 * because the caller (xfs_buf_submit) holds a count itself.
1404 atomic_dec(&bp
->b_io_remaining
);
1405 xfs_buf_ioerror(bp
, -EIO
);
1415 struct blk_plug plug
;
1423 * Make sure we capture only current IO errors rather than stale errors
1424 * left over from previous use of the buffer (e.g. failed readahead).
1429 * Initialize the I/O completion workqueue if we haven't yet or the
1430 * submitter has not opted to specify a custom one.
1432 if (!bp
->b_ioend_wq
)
1433 bp
->b_ioend_wq
= bp
->b_target
->bt_mount
->m_buf_workqueue
;
1435 if (bp
->b_flags
& XBF_WRITE
) {
1437 if (bp
->b_flags
& XBF_SYNCIO
)
1438 op_flags
= REQ_SYNC
;
1439 if (bp
->b_flags
& XBF_FUA
)
1440 op_flags
|= REQ_FUA
;
1441 if (bp
->b_flags
& XBF_FLUSH
)
1442 op_flags
|= REQ_PREFLUSH
;
1445 * Run the write verifier callback function if it exists. If
1446 * this function fails it will mark the buffer with an error and
1447 * the IO should not be dispatched.
1450 bp
->b_ops
->verify_write(bp
);
1452 xfs_force_shutdown(bp
->b_target
->bt_mount
,
1453 SHUTDOWN_CORRUPT_INCORE
);
1456 } else if (bp
->b_bn
!= XFS_BUF_DADDR_NULL
) {
1457 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
1460 * non-crc filesystems don't attach verifiers during
1461 * log recovery, so don't warn for such filesystems.
1463 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
1465 "%s: no buf ops on daddr 0x%llx len %d",
1466 __func__
, bp
->b_bn
, bp
->b_length
);
1467 xfs_hex_dump(bp
->b_addr
,
1468 XFS_CORRUPTION_DUMP_LEN
);
1472 } else if (bp
->b_flags
& XBF_READ_AHEAD
) {
1474 op_flags
= REQ_RAHEAD
;
1479 /* we only use the buffer cache for meta-data */
1480 op_flags
|= REQ_META
;
1483 * Walk all the vectors issuing IO on them. Set up the initial offset
1484 * into the buffer and the desired IO size before we start -
1485 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1488 offset
= bp
->b_offset
;
1489 size
= BBTOB(bp
->b_io_length
);
1490 blk_start_plug(&plug
);
1491 for (i
= 0; i
< bp
->b_map_count
; i
++) {
1492 xfs_buf_ioapply_map(bp
, i
, &offset
, &size
, op
, op_flags
);
1496 break; /* all done */
1498 blk_finish_plug(&plug
);
1502 * Wait for I/O completion of a sync buffer and return the I/O error code.
1508 ASSERT(!(bp
->b_flags
& XBF_ASYNC
));
1510 trace_xfs_buf_iowait(bp
, _RET_IP_
);
1511 wait_for_completion(&bp
->b_iowait
);
1512 trace_xfs_buf_iowait_done(bp
, _RET_IP_
);
1518 * Buffer I/O submission path, read or write. Asynchronous submission transfers
1519 * the buffer lock ownership and the current reference to the IO. It is not
1520 * safe to reference the buffer after a call to this function unless the caller
1521 * holds an additional reference itself.
1530 trace_xfs_buf_submit(bp
, _RET_IP_
);
1532 ASSERT(!(bp
->b_flags
& _XBF_DELWRI_Q
));
1534 /* on shutdown we stale and complete the buffer immediately */
1535 if (XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
1536 xfs_buf_ioerror(bp
, -EIO
);
1537 bp
->b_flags
&= ~XBF_DONE
;
1539 if (bp
->b_flags
& XBF_ASYNC
)
1545 * Grab a reference so the buffer does not go away underneath us. For
1546 * async buffers, I/O completion drops the callers reference, which
1547 * could occur before submission returns.
1551 if (bp
->b_flags
& XBF_WRITE
)
1552 xfs_buf_wait_unpin(bp
);
1554 /* clear the internal error state to avoid spurious errors */
1558 * Set the count to 1 initially, this will stop an I/O completion
1559 * callout which happens before we have started all the I/O from calling
1560 * xfs_buf_ioend too early.
1562 atomic_set(&bp
->b_io_remaining
, 1);
1563 if (bp
->b_flags
& XBF_ASYNC
)
1564 xfs_buf_ioacct_inc(bp
);
1565 _xfs_buf_ioapply(bp
);
1568 * If _xfs_buf_ioapply failed, we can get back here with only the IO
1569 * reference we took above. If we drop it to zero, run completion so
1570 * that we don't return to the caller with completion still pending.
1572 if (atomic_dec_and_test(&bp
->b_io_remaining
) == 1) {
1573 if (bp
->b_error
|| !(bp
->b_flags
& XBF_ASYNC
))
1576 xfs_buf_ioend_async(bp
);
1580 error
= xfs_buf_iowait(bp
);
1583 * Release the hold that keeps the buffer referenced for the entire
1584 * I/O. Note that if the buffer is async, it is not safe to reference
1585 * after this release.
1599 return bp
->b_addr
+ offset
;
1601 offset
+= bp
->b_offset
;
1602 page
= bp
->b_pages
[offset
>> PAGE_SHIFT
];
1603 return page_address(page
) + (offset
& (PAGE_SIZE
-1));
1607 * Move data into or out of a buffer.
1611 xfs_buf_t
*bp
, /* buffer to process */
1612 size_t boff
, /* starting buffer offset */
1613 size_t bsize
, /* length to copy */
1614 void *data
, /* data address */
1615 xfs_buf_rw_t mode
) /* read/write/zero flag */
1619 bend
= boff
+ bsize
;
1620 while (boff
< bend
) {
1622 int page_index
, page_offset
, csize
;
1624 page_index
= (boff
+ bp
->b_offset
) >> PAGE_SHIFT
;
1625 page_offset
= (boff
+ bp
->b_offset
) & ~PAGE_MASK
;
1626 page
= bp
->b_pages
[page_index
];
1627 csize
= min_t(size_t, PAGE_SIZE
- page_offset
,
1628 BBTOB(bp
->b_io_length
) - boff
);
1630 ASSERT((csize
+ page_offset
) <= PAGE_SIZE
);
1634 memset(page_address(page
) + page_offset
, 0, csize
);
1637 memcpy(data
, page_address(page
) + page_offset
, csize
);
1640 memcpy(page_address(page
) + page_offset
, data
, csize
);
1649 * Handling of buffer targets (buftargs).
1653 * Wait for any bufs with callbacks that have been submitted but have not yet
1654 * returned. These buffers will have an elevated hold count, so wait on those
1655 * while freeing all the buffers only held by the LRU.
1657 static enum lru_status
1658 xfs_buftarg_wait_rele(
1659 struct list_head
*item
,
1660 struct list_lru_one
*lru
,
1661 spinlock_t
*lru_lock
,
1665 struct xfs_buf
*bp
= container_of(item
, struct xfs_buf
, b_lru
);
1666 struct list_head
*dispose
= arg
;
1668 if (atomic_read(&bp
->b_hold
) > 1) {
1669 /* need to wait, so skip it this pass */
1670 trace_xfs_buf_wait_buftarg(bp
, _RET_IP_
);
1673 if (!spin_trylock(&bp
->b_lock
))
1677 * clear the LRU reference count so the buffer doesn't get
1678 * ignored in xfs_buf_rele().
1680 atomic_set(&bp
->b_lru_ref
, 0);
1681 bp
->b_state
|= XFS_BSTATE_DISPOSE
;
1682 list_lru_isolate_move(lru
, item
, dispose
);
1683 spin_unlock(&bp
->b_lock
);
1689 struct xfs_buftarg
*btp
)
1695 * First wait on the buftarg I/O count for all in-flight buffers to be
1696 * released. This is critical as new buffers do not make the LRU until
1697 * they are released.
1699 * Next, flush the buffer workqueue to ensure all completion processing
1700 * has finished. Just waiting on buffer locks is not sufficient for
1701 * async IO as the reference count held over IO is not released until
1702 * after the buffer lock is dropped. Hence we need to ensure here that
1703 * all reference counts have been dropped before we start walking the
1706 while (percpu_counter_sum(&btp
->bt_io_count
))
1708 flush_workqueue(btp
->bt_mount
->m_buf_workqueue
);
1710 /* loop until there is nothing left on the lru list. */
1711 while (list_lru_count(&btp
->bt_lru
)) {
1712 list_lru_walk(&btp
->bt_lru
, xfs_buftarg_wait_rele
,
1713 &dispose
, LONG_MAX
);
1715 while (!list_empty(&dispose
)) {
1717 bp
= list_first_entry(&dispose
, struct xfs_buf
, b_lru
);
1718 list_del_init(&bp
->b_lru
);
1719 if (bp
->b_flags
& XBF_WRITE_FAIL
) {
1720 xfs_alert(btp
->bt_mount
,
1721 "Corruption Alert: Buffer at daddr 0x%llx had permanent write failures!",
1722 (long long)bp
->b_bn
);
1723 xfs_alert(btp
->bt_mount
,
1724 "Please run xfs_repair to determine the extent of the problem.");
1733 static enum lru_status
1734 xfs_buftarg_isolate(
1735 struct list_head
*item
,
1736 struct list_lru_one
*lru
,
1737 spinlock_t
*lru_lock
,
1740 struct xfs_buf
*bp
= container_of(item
, struct xfs_buf
, b_lru
);
1741 struct list_head
*dispose
= arg
;
1744 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1745 * If we fail to get the lock, just skip it.
1747 if (!spin_trylock(&bp
->b_lock
))
1750 * Decrement the b_lru_ref count unless the value is already
1751 * zero. If the value is already zero, we need to reclaim the
1752 * buffer, otherwise it gets another trip through the LRU.
1754 if (atomic_add_unless(&bp
->b_lru_ref
, -1, 0)) {
1755 spin_unlock(&bp
->b_lock
);
1759 bp
->b_state
|= XFS_BSTATE_DISPOSE
;
1760 list_lru_isolate_move(lru
, item
, dispose
);
1761 spin_unlock(&bp
->b_lock
);
1765 static unsigned long
1766 xfs_buftarg_shrink_scan(
1767 struct shrinker
*shrink
,
1768 struct shrink_control
*sc
)
1770 struct xfs_buftarg
*btp
= container_of(shrink
,
1771 struct xfs_buftarg
, bt_shrinker
);
1773 unsigned long freed
;
1775 freed
= list_lru_shrink_walk(&btp
->bt_lru
, sc
,
1776 xfs_buftarg_isolate
, &dispose
);
1778 while (!list_empty(&dispose
)) {
1780 bp
= list_first_entry(&dispose
, struct xfs_buf
, b_lru
);
1781 list_del_init(&bp
->b_lru
);
1788 static unsigned long
1789 xfs_buftarg_shrink_count(
1790 struct shrinker
*shrink
,
1791 struct shrink_control
*sc
)
1793 struct xfs_buftarg
*btp
= container_of(shrink
,
1794 struct xfs_buftarg
, bt_shrinker
);
1795 return list_lru_shrink_count(&btp
->bt_lru
, sc
);
1800 struct xfs_buftarg
*btp
)
1802 unregister_shrinker(&btp
->bt_shrinker
);
1803 ASSERT(percpu_counter_sum(&btp
->bt_io_count
) == 0);
1804 percpu_counter_destroy(&btp
->bt_io_count
);
1805 list_lru_destroy(&btp
->bt_lru
);
1807 xfs_blkdev_issue_flush(btp
);
1813 xfs_setsize_buftarg(
1815 unsigned int sectorsize
)
1817 /* Set up metadata sector size info */
1818 btp
->bt_meta_sectorsize
= sectorsize
;
1819 btp
->bt_meta_sectormask
= sectorsize
- 1;
1821 if (set_blocksize(btp
->bt_bdev
, sectorsize
)) {
1822 xfs_warn(btp
->bt_mount
,
1823 "Cannot set_blocksize to %u on device %pg",
1824 sectorsize
, btp
->bt_bdev
);
1828 /* Set up device logical sector size mask */
1829 btp
->bt_logical_sectorsize
= bdev_logical_block_size(btp
->bt_bdev
);
1830 btp
->bt_logical_sectormask
= bdev_logical_block_size(btp
->bt_bdev
) - 1;
1836 * When allocating the initial buffer target we have not yet
1837 * read in the superblock, so don't know what sized sectors
1838 * are being used at this early stage. Play safe.
1841 xfs_setsize_buftarg_early(
1843 struct block_device
*bdev
)
1845 return xfs_setsize_buftarg(btp
, bdev_logical_block_size(bdev
));
1850 struct xfs_mount
*mp
,
1851 struct block_device
*bdev
,
1852 struct dax_device
*dax_dev
)
1856 btp
= kmem_zalloc(sizeof(*btp
), KM_SLEEP
| KM_NOFS
);
1859 btp
->bt_dev
= bdev
->bd_dev
;
1860 btp
->bt_bdev
= bdev
;
1861 btp
->bt_daxdev
= dax_dev
;
1863 if (xfs_setsize_buftarg_early(btp
, bdev
))
1866 if (list_lru_init(&btp
->bt_lru
))
1869 if (percpu_counter_init(&btp
->bt_io_count
, 0, GFP_KERNEL
))
1872 btp
->bt_shrinker
.count_objects
= xfs_buftarg_shrink_count
;
1873 btp
->bt_shrinker
.scan_objects
= xfs_buftarg_shrink_scan
;
1874 btp
->bt_shrinker
.seeks
= DEFAULT_SEEKS
;
1875 btp
->bt_shrinker
.flags
= SHRINKER_NUMA_AWARE
;
1876 if (register_shrinker(&btp
->bt_shrinker
))
1881 percpu_counter_destroy(&btp
->bt_io_count
);
1883 list_lru_destroy(&btp
->bt_lru
);
1890 * Cancel a delayed write list.
1892 * Remove each buffer from the list, clear the delwri queue flag and drop the
1893 * associated buffer reference.
1896 xfs_buf_delwri_cancel(
1897 struct list_head
*list
)
1901 while (!list_empty(list
)) {
1902 bp
= list_first_entry(list
, struct xfs_buf
, b_list
);
1905 bp
->b_flags
&= ~_XBF_DELWRI_Q
;
1906 list_del_init(&bp
->b_list
);
1912 * Add a buffer to the delayed write list.
1914 * This queues a buffer for writeout if it hasn't already been. Note that
1915 * neither this routine nor the buffer list submission functions perform
1916 * any internal synchronization. It is expected that the lists are thread-local
1919 * Returns true if we queued up the buffer, or false if it already had
1920 * been on the buffer list.
1923 xfs_buf_delwri_queue(
1925 struct list_head
*list
)
1927 ASSERT(xfs_buf_islocked(bp
));
1928 ASSERT(!(bp
->b_flags
& XBF_READ
));
1931 * If the buffer is already marked delwri it already is queued up
1932 * by someone else for imediate writeout. Just ignore it in that
1935 if (bp
->b_flags
& _XBF_DELWRI_Q
) {
1936 trace_xfs_buf_delwri_queued(bp
, _RET_IP_
);
1940 trace_xfs_buf_delwri_queue(bp
, _RET_IP_
);
1943 * If a buffer gets written out synchronously or marked stale while it
1944 * is on a delwri list we lazily remove it. To do this, the other party
1945 * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1946 * It remains referenced and on the list. In a rare corner case it
1947 * might get readded to a delwri list after the synchronous writeout, in
1948 * which case we need just need to re-add the flag here.
1950 bp
->b_flags
|= _XBF_DELWRI_Q
;
1951 if (list_empty(&bp
->b_list
)) {
1952 atomic_inc(&bp
->b_hold
);
1953 list_add_tail(&bp
->b_list
, list
);
1960 * Compare function is more complex than it needs to be because
1961 * the return value is only 32 bits and we are doing comparisons
1967 struct list_head
*a
,
1968 struct list_head
*b
)
1970 struct xfs_buf
*ap
= container_of(a
, struct xfs_buf
, b_list
);
1971 struct xfs_buf
*bp
= container_of(b
, struct xfs_buf
, b_list
);
1974 diff
= ap
->b_maps
[0].bm_bn
- bp
->b_maps
[0].bm_bn
;
1983 * Submit buffers for write. If wait_list is specified, the buffers are
1984 * submitted using sync I/O and placed on the wait list such that the caller can
1985 * iowait each buffer. Otherwise async I/O is used and the buffers are released
1986 * at I/O completion time. In either case, buffers remain locked until I/O
1987 * completes and the buffer is released from the queue.
1990 xfs_buf_delwri_submit_buffers(
1991 struct list_head
*buffer_list
,
1992 struct list_head
*wait_list
)
1994 struct xfs_buf
*bp
, *n
;
1996 struct blk_plug plug
;
1998 list_sort(NULL
, buffer_list
, xfs_buf_cmp
);
2000 blk_start_plug(&plug
);
2001 list_for_each_entry_safe(bp
, n
, buffer_list
, b_list
) {
2003 if (xfs_buf_ispinned(bp
)) {
2007 if (!xfs_buf_trylock(bp
))
2014 * Someone else might have written the buffer synchronously or
2015 * marked it stale in the meantime. In that case only the
2016 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
2017 * reference and remove it from the list here.
2019 if (!(bp
->b_flags
& _XBF_DELWRI_Q
)) {
2020 list_del_init(&bp
->b_list
);
2025 trace_xfs_buf_delwri_split(bp
, _RET_IP_
);
2028 * If we have a wait list, each buffer (and associated delwri
2029 * queue reference) transfers to it and is submitted
2030 * synchronously. Otherwise, drop the buffer from the delwri
2031 * queue and submit async.
2033 bp
->b_flags
&= ~(_XBF_DELWRI_Q
| XBF_WRITE_FAIL
);
2034 bp
->b_flags
|= XBF_WRITE
;
2036 bp
->b_flags
&= ~XBF_ASYNC
;
2037 list_move_tail(&bp
->b_list
, wait_list
);
2039 bp
->b_flags
|= XBF_ASYNC
;
2040 list_del_init(&bp
->b_list
);
2042 __xfs_buf_submit(bp
, false);
2044 blk_finish_plug(&plug
);
2050 * Write out a buffer list asynchronously.
2052 * This will take the @buffer_list, write all non-locked and non-pinned buffers
2053 * out and not wait for I/O completion on any of the buffers. This interface
2054 * is only safely useable for callers that can track I/O completion by higher
2055 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
2058 * Note: this function will skip buffers it would block on, and in doing so
2059 * leaves them on @buffer_list so they can be retried on a later pass. As such,
2060 * it is up to the caller to ensure that the buffer list is fully submitted or
2061 * cancelled appropriately when they are finished with the list. Failure to
2062 * cancel or resubmit the list until it is empty will result in leaked buffers
2066 xfs_buf_delwri_submit_nowait(
2067 struct list_head
*buffer_list
)
2069 return xfs_buf_delwri_submit_buffers(buffer_list
, NULL
);
2073 * Write out a buffer list synchronously.
2075 * This will take the @buffer_list, write all buffers out and wait for I/O
2076 * completion on all of the buffers. @buffer_list is consumed by the function,
2077 * so callers must have some other way of tracking buffers if they require such
2081 xfs_buf_delwri_submit(
2082 struct list_head
*buffer_list
)
2084 LIST_HEAD (wait_list
);
2085 int error
= 0, error2
;
2088 xfs_buf_delwri_submit_buffers(buffer_list
, &wait_list
);
2090 /* Wait for IO to complete. */
2091 while (!list_empty(&wait_list
)) {
2092 bp
= list_first_entry(&wait_list
, struct xfs_buf
, b_list
);
2094 list_del_init(&bp
->b_list
);
2097 * Wait on the locked buffer, check for errors and unlock and
2098 * release the delwri queue reference.
2100 error2
= xfs_buf_iowait(bp
);
2110 * Push a single buffer on a delwri queue.
2112 * The purpose of this function is to submit a single buffer of a delwri queue
2113 * and return with the buffer still on the original queue. The waiting delwri
2114 * buffer submission infrastructure guarantees transfer of the delwri queue
2115 * buffer reference to a temporary wait list. We reuse this infrastructure to
2116 * transfer the buffer back to the original queue.
2118 * Note the buffer transitions from the queued state, to the submitted and wait
2119 * listed state and back to the queued state during this call. The buffer
2120 * locking and queue management logic between _delwri_pushbuf() and
2121 * _delwri_queue() guarantee that the buffer cannot be queued to another list
2125 xfs_buf_delwri_pushbuf(
2127 struct list_head
*buffer_list
)
2129 LIST_HEAD (submit_list
);
2132 ASSERT(bp
->b_flags
& _XBF_DELWRI_Q
);
2134 trace_xfs_buf_delwri_pushbuf(bp
, _RET_IP_
);
2137 * Isolate the buffer to a new local list so we can submit it for I/O
2138 * independently from the rest of the original list.
2141 list_move(&bp
->b_list
, &submit_list
);
2145 * Delwri submission clears the DELWRI_Q buffer flag and returns with
2146 * the buffer on the wait list with the original reference. Rather than
2147 * bounce the buffer from a local wait list back to the original list
2148 * after I/O completion, reuse the original list as the wait list.
2150 xfs_buf_delwri_submit_buffers(&submit_list
, buffer_list
);
2153 * The buffer is now locked, under I/O and wait listed on the original
2154 * delwri queue. Wait for I/O completion, restore the DELWRI_Q flag and
2155 * return with the buffer unlocked and on the original queue.
2157 error
= xfs_buf_iowait(bp
);
2158 bp
->b_flags
|= _XBF_DELWRI_Q
;
2167 xfs_buf_zone
= kmem_zone_init_flags(sizeof(xfs_buf_t
), "xfs_buf",
2168 KM_ZONE_HWALIGN
, NULL
);
2179 xfs_buf_terminate(void)
2181 kmem_zone_destroy(xfs_buf_zone
);
2184 void xfs_buf_set_ref(struct xfs_buf
*bp
, int lru_ref
)
2187 * Set the lru reference count to 0 based on the error injection tag.
2188 * This allows userspace to disrupt buffer caching for debug/testing
2191 if (XFS_TEST_ERROR(false, bp
->b_target
->bt_mount
,
2192 XFS_ERRTAG_BUF_LRU_REF
))
2195 atomic_set(&bp
->b_lru_ref
, lru_ref
);