2 * linux/fs/nfs/pagelist.c
4 * A set of helper functions for managing NFS read and write requests.
5 * The main purpose of these routines is to provide support for the
6 * coalescing of several requests into a single RPC call.
8 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/sched.h>
15 #include <linux/sunrpc/clnt.h>
16 #include <linux/nfs.h>
17 #include <linux/nfs3.h>
18 #include <linux/nfs4.h>
19 #include <linux/nfs_page.h>
20 #include <linux/nfs_fs.h>
21 #include <linux/nfs_mount.h>
22 #include <linux/export.h>
27 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
29 static struct kmem_cache
*nfs_page_cachep
;
30 static const struct rpc_call_ops nfs_pgio_common_ops
;
32 struct nfs_pgio_mirror
*
33 nfs_pgio_current_mirror(struct nfs_pageio_descriptor
*desc
)
35 return nfs_pgio_has_mirroring(desc
) ?
36 &desc
->pg_mirrors
[desc
->pg_mirror_idx
] :
39 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror
);
41 void nfs_pgheader_init(struct nfs_pageio_descriptor
*desc
,
42 struct nfs_pgio_header
*hdr
,
43 void (*release
)(struct nfs_pgio_header
*hdr
))
45 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
48 hdr
->req
= nfs_list_entry(mirror
->pg_list
.next
);
49 hdr
->inode
= desc
->pg_inode
;
50 hdr
->cred
= hdr
->req
->wb_context
->cred
;
51 hdr
->io_start
= req_offset(hdr
->req
);
52 hdr
->good_bytes
= mirror
->pg_count
;
53 hdr
->dreq
= desc
->pg_dreq
;
54 hdr
->layout_private
= desc
->pg_layout_private
;
55 hdr
->release
= release
;
56 hdr
->completion_ops
= desc
->pg_completion_ops
;
57 if (hdr
->completion_ops
->init_hdr
)
58 hdr
->completion_ops
->init_hdr(hdr
);
60 hdr
->pgio_mirror_idx
= desc
->pg_mirror_idx
;
62 EXPORT_SYMBOL_GPL(nfs_pgheader_init
);
64 void nfs_set_pgio_error(struct nfs_pgio_header
*hdr
, int error
, loff_t pos
)
66 spin_lock(&hdr
->lock
);
67 if (!test_and_set_bit(NFS_IOHDR_ERROR
, &hdr
->flags
)
68 || pos
< hdr
->io_start
+ hdr
->good_bytes
) {
69 clear_bit(NFS_IOHDR_EOF
, &hdr
->flags
);
70 hdr
->good_bytes
= pos
- hdr
->io_start
;
73 spin_unlock(&hdr
->lock
);
76 static inline struct nfs_page
*
79 struct nfs_page
*p
= kmem_cache_zalloc(nfs_page_cachep
, GFP_NOIO
);
81 INIT_LIST_HEAD(&p
->wb_list
);
86 nfs_page_free(struct nfs_page
*p
)
88 kmem_cache_free(nfs_page_cachep
, p
);
92 * nfs_iocounter_wait - wait for i/o to complete
93 * @l_ctx: nfs_lock_context with io_counter to use
95 * returns -ERESTARTSYS if interrupted by a fatal signal.
96 * Otherwise returns 0 once the io_count hits 0.
99 nfs_iocounter_wait(struct nfs_lock_context
*l_ctx
)
101 return wait_on_atomic_t(&l_ctx
->io_count
, nfs_wait_atomic_killable
,
106 * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
108 * @task: the rpc_task that should wait
109 * @l_ctx: nfs_lock_context with io_counter to check
111 * Returns true if there is outstanding I/O to wait on and the
112 * task has been put to sleep.
115 nfs_async_iocounter_wait(struct rpc_task
*task
, struct nfs_lock_context
*l_ctx
)
117 struct inode
*inode
= d_inode(l_ctx
->open_context
->dentry
);
120 if (atomic_read(&l_ctx
->io_count
) > 0) {
121 rpc_sleep_on(&NFS_SERVER(inode
)->uoc_rpcwaitq
, task
, NULL
);
125 if (atomic_read(&l_ctx
->io_count
) == 0) {
126 rpc_wake_up_queued_task(&NFS_SERVER(inode
)->uoc_rpcwaitq
, task
);
132 EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait
);
135 * nfs_page_group_lock - lock the head of the page group
136 * @req - request in group that is to be locked
137 * @nonblock - if true don't block waiting for lock
139 * this lock must be held if modifying the page group list
141 * return 0 on success, < 0 on error: -EDELAY if nonblocking or the
142 * result from wait_on_bit_lock
144 * NOTE: calling with nonblock=false should always have set the
145 * lock bit (see fs/buffer.c and other uses of wait_on_bit_lock
146 * with TASK_UNINTERRUPTIBLE), so there is no need to check the result.
149 nfs_page_group_lock(struct nfs_page
*req
, bool nonblock
)
151 struct nfs_page
*head
= req
->wb_head
;
153 WARN_ON_ONCE(head
!= head
->wb_head
);
155 if (!test_and_set_bit(PG_HEADLOCK
, &head
->wb_flags
))
159 return wait_on_bit_lock(&head
->wb_flags
, PG_HEADLOCK
,
160 TASK_UNINTERRUPTIBLE
);
166 * nfs_page_group_lock_wait - wait for the lock to clear, but don't grab it
167 * @req - a request in the group
169 * This is a blocking call to wait for the group lock to be cleared.
172 nfs_page_group_lock_wait(struct nfs_page
*req
)
174 struct nfs_page
*head
= req
->wb_head
;
176 WARN_ON_ONCE(head
!= head
->wb_head
);
178 wait_on_bit(&head
->wb_flags
, PG_HEADLOCK
,
179 TASK_UNINTERRUPTIBLE
);
183 * nfs_page_group_unlock - unlock the head of the page group
184 * @req - request in group that is to be unlocked
187 nfs_page_group_unlock(struct nfs_page
*req
)
189 struct nfs_page
*head
= req
->wb_head
;
191 WARN_ON_ONCE(head
!= head
->wb_head
);
193 smp_mb__before_atomic();
194 clear_bit(PG_HEADLOCK
, &head
->wb_flags
);
195 smp_mb__after_atomic();
196 wake_up_bit(&head
->wb_flags
, PG_HEADLOCK
);
200 * nfs_page_group_sync_on_bit_locked
202 * must be called with page group lock held
205 nfs_page_group_sync_on_bit_locked(struct nfs_page
*req
, unsigned int bit
)
207 struct nfs_page
*head
= req
->wb_head
;
208 struct nfs_page
*tmp
;
210 WARN_ON_ONCE(!test_bit(PG_HEADLOCK
, &head
->wb_flags
));
211 WARN_ON_ONCE(test_and_set_bit(bit
, &req
->wb_flags
));
213 tmp
= req
->wb_this_page
;
215 if (!test_bit(bit
, &tmp
->wb_flags
))
217 tmp
= tmp
->wb_this_page
;
220 /* true! reset all bits */
223 clear_bit(bit
, &tmp
->wb_flags
);
224 tmp
= tmp
->wb_this_page
;
225 } while (tmp
!= req
);
231 * nfs_page_group_sync_on_bit - set bit on current request, but only
232 * return true if the bit is set for all requests in page group
233 * @req - request in page group
234 * @bit - PG_* bit that is used to sync page group
236 bool nfs_page_group_sync_on_bit(struct nfs_page
*req
, unsigned int bit
)
240 nfs_page_group_lock(req
, false);
241 ret
= nfs_page_group_sync_on_bit_locked(req
, bit
);
242 nfs_page_group_unlock(req
);
248 * nfs_page_group_init - Initialize the page group linkage for @req
249 * @req - a new nfs request
250 * @prev - the previous request in page group, or NULL if @req is the first
251 * or only request in the group (the head).
254 nfs_page_group_init(struct nfs_page
*req
, struct nfs_page
*prev
)
257 WARN_ON_ONCE(prev
== req
);
262 req
->wb_this_page
= req
;
265 WARN_ON_ONCE(prev
->wb_this_page
!= prev
->wb_head
);
266 WARN_ON_ONCE(!test_bit(PG_HEADLOCK
, &prev
->wb_head
->wb_flags
));
267 req
->wb_head
= prev
->wb_head
;
268 req
->wb_this_page
= prev
->wb_this_page
;
269 prev
->wb_this_page
= req
;
271 /* All subrequests take a ref on the head request until
272 * nfs_page_group_destroy is called */
273 kref_get(&req
->wb_head
->wb_kref
);
275 /* grab extra ref and bump the request count if head request
276 * has extra ref from the write/commit path to handle handoff
277 * between write and commit lists. */
278 if (test_bit(PG_INODE_REF
, &prev
->wb_head
->wb_flags
)) {
279 inode
= page_file_mapping(req
->wb_page
)->host
;
280 set_bit(PG_INODE_REF
, &req
->wb_flags
);
281 kref_get(&req
->wb_kref
);
282 spin_lock(&inode
->i_lock
);
283 NFS_I(inode
)->nrequests
++;
284 spin_unlock(&inode
->i_lock
);
290 * nfs_page_group_destroy - sync the destruction of page groups
291 * @req - request that no longer needs the page group
293 * releases the page group reference from each member once all
294 * members have called this function.
297 nfs_page_group_destroy(struct kref
*kref
)
299 struct nfs_page
*req
= container_of(kref
, struct nfs_page
, wb_kref
);
300 struct nfs_page
*tmp
, *next
;
302 /* subrequests must release the ref on the head request */
303 if (req
->wb_head
!= req
)
304 nfs_release_request(req
->wb_head
);
306 if (!nfs_page_group_sync_on_bit(req
, PG_TEARDOWN
))
311 next
= tmp
->wb_this_page
;
312 /* unlink and free */
313 tmp
->wb_this_page
= tmp
;
315 nfs_free_request(tmp
);
317 } while (tmp
!= req
);
321 * nfs_create_request - Create an NFS read/write request.
322 * @ctx: open context to use
323 * @page: page to write
324 * @last: last nfs request created for this page group or NULL if head
325 * @offset: starting offset within the page for the write
326 * @count: number of bytes to read/write
328 * The page must be locked by the caller. This makes sure we never
329 * create two different requests for the same page.
330 * User should ensure it is safe to sleep in this function.
333 nfs_create_request(struct nfs_open_context
*ctx
, struct page
*page
,
334 struct nfs_page
*last
, unsigned int offset
,
337 struct nfs_page
*req
;
338 struct nfs_lock_context
*l_ctx
;
340 if (test_bit(NFS_CONTEXT_BAD
, &ctx
->flags
))
341 return ERR_PTR(-EBADF
);
342 /* try to allocate the request struct */
343 req
= nfs_page_alloc();
345 return ERR_PTR(-ENOMEM
);
347 /* get lock context early so we can deal with alloc failures */
348 l_ctx
= nfs_get_lock_context(ctx
);
351 return ERR_CAST(l_ctx
);
353 req
->wb_lock_context
= l_ctx
;
354 atomic_inc(&l_ctx
->io_count
);
356 /* Initialize the request struct. Initially, we assume a
357 * long write-back delay. This will be adjusted in
358 * update_nfs_request below if the region is not locked. */
361 req
->wb_index
= page_index(page
);
364 req
->wb_offset
= offset
;
365 req
->wb_pgbase
= offset
;
366 req
->wb_bytes
= count
;
367 req
->wb_context
= get_nfs_open_context(ctx
);
368 kref_init(&req
->wb_kref
);
369 nfs_page_group_init(req
, last
);
374 * nfs_unlock_request - Unlock request and wake up sleepers.
377 void nfs_unlock_request(struct nfs_page
*req
)
379 if (!NFS_WBACK_BUSY(req
)) {
380 printk(KERN_ERR
"NFS: Invalid unlock attempted\n");
383 smp_mb__before_atomic();
384 clear_bit(PG_BUSY
, &req
->wb_flags
);
385 smp_mb__after_atomic();
386 wake_up_bit(&req
->wb_flags
, PG_BUSY
);
390 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
393 void nfs_unlock_and_release_request(struct nfs_page
*req
)
395 nfs_unlock_request(req
);
396 nfs_release_request(req
);
400 * nfs_clear_request - Free up all resources allocated to the request
403 * Release page and open context resources associated with a read/write
404 * request after it has completed.
406 static void nfs_clear_request(struct nfs_page
*req
)
408 struct page
*page
= req
->wb_page
;
409 struct nfs_open_context
*ctx
= req
->wb_context
;
410 struct nfs_lock_context
*l_ctx
= req
->wb_lock_context
;
417 if (atomic_dec_and_test(&l_ctx
->io_count
)) {
418 wake_up_atomic_t(&l_ctx
->io_count
);
419 if (test_bit(NFS_CONTEXT_UNLOCK
, &ctx
->flags
))
420 rpc_wake_up(&NFS_SERVER(d_inode(ctx
->dentry
))->uoc_rpcwaitq
);
422 nfs_put_lock_context(l_ctx
);
423 req
->wb_lock_context
= NULL
;
426 put_nfs_open_context(ctx
);
427 req
->wb_context
= NULL
;
432 * nfs_release_request - Release the count on an NFS read/write request
433 * @req: request to release
435 * Note: Should never be called with the spinlock held!
437 void nfs_free_request(struct nfs_page
*req
)
439 WARN_ON_ONCE(req
->wb_this_page
!= req
);
441 /* extra debug: make sure no sync bits are still set */
442 WARN_ON_ONCE(test_bit(PG_TEARDOWN
, &req
->wb_flags
));
443 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE
, &req
->wb_flags
));
444 WARN_ON_ONCE(test_bit(PG_UPTODATE
, &req
->wb_flags
));
445 WARN_ON_ONCE(test_bit(PG_WB_END
, &req
->wb_flags
));
446 WARN_ON_ONCE(test_bit(PG_REMOVE
, &req
->wb_flags
));
448 /* Release struct file and open context */
449 nfs_clear_request(req
);
453 void nfs_release_request(struct nfs_page
*req
)
455 kref_put(&req
->wb_kref
, nfs_page_group_destroy
);
459 * nfs_wait_on_request - Wait for a request to complete.
460 * @req: request to wait upon.
462 * Interruptible by fatal signals only.
463 * The user is responsible for holding a count on the request.
466 nfs_wait_on_request(struct nfs_page
*req
)
468 return wait_on_bit_io(&req
->wb_flags
, PG_BUSY
,
469 TASK_UNINTERRUPTIBLE
);
473 * nfs_generic_pg_test - determine if requests can be coalesced
474 * @desc: pointer to descriptor
475 * @prev: previous request in desc, or NULL
478 * Returns zero if @req can be coalesced into @desc, otherwise it returns
479 * the size of the request.
481 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor
*desc
,
482 struct nfs_page
*prev
, struct nfs_page
*req
)
484 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
487 if (mirror
->pg_count
> mirror
->pg_bsize
) {
488 /* should never happen */
494 * Limit the request size so that we can still allocate a page array
495 * for it without upsetting the slab allocator.
497 if (((mirror
->pg_count
+ req
->wb_bytes
) >> PAGE_SHIFT
) *
498 sizeof(struct page
*) > PAGE_SIZE
)
501 return min(mirror
->pg_bsize
- mirror
->pg_count
, (size_t)req
->wb_bytes
);
503 EXPORT_SYMBOL_GPL(nfs_generic_pg_test
);
505 struct nfs_pgio_header
*nfs_pgio_header_alloc(const struct nfs_rw_ops
*ops
)
507 struct nfs_pgio_header
*hdr
= ops
->rw_alloc_header();
510 INIT_LIST_HEAD(&hdr
->pages
);
511 spin_lock_init(&hdr
->lock
);
516 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc
);
519 * nfs_pgio_header_free - Free a read or write header
520 * @hdr: The header to free
522 void nfs_pgio_header_free(struct nfs_pgio_header
*hdr
)
524 hdr
->rw_ops
->rw_free_header(hdr
);
526 EXPORT_SYMBOL_GPL(nfs_pgio_header_free
);
529 * nfs_pgio_data_destroy - make @hdr suitable for reuse
531 * Frees memory and releases refs from nfs_generic_pgio, so that it may
534 * @hdr: A header that has had nfs_generic_pgio called
536 void nfs_pgio_data_destroy(struct nfs_pgio_header
*hdr
)
538 if (hdr
->args
.context
)
539 put_nfs_open_context(hdr
->args
.context
);
540 if (hdr
->page_array
.pagevec
!= hdr
->page_array
.page_array
)
541 kfree(hdr
->page_array
.pagevec
);
543 EXPORT_SYMBOL_GPL(nfs_pgio_data_destroy
);
546 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
547 * @hdr: The pageio hdr
548 * @count: Number of bytes to read
549 * @offset: Initial offset
550 * @how: How to commit data (writes only)
551 * @cinfo: Commit information for the call (writes only)
553 static void nfs_pgio_rpcsetup(struct nfs_pgio_header
*hdr
,
554 unsigned int count
, unsigned int offset
,
555 int how
, struct nfs_commit_info
*cinfo
)
557 struct nfs_page
*req
= hdr
->req
;
559 /* Set up the RPC argument and reply structs
560 * NB: take care not to mess about with hdr->commit et al. */
562 hdr
->args
.fh
= NFS_FH(hdr
->inode
);
563 hdr
->args
.offset
= req_offset(req
) + offset
;
564 /* pnfs_set_layoutcommit needs this */
565 hdr
->mds_offset
= hdr
->args
.offset
;
566 hdr
->args
.pgbase
= req
->wb_pgbase
+ offset
;
567 hdr
->args
.pages
= hdr
->page_array
.pagevec
;
568 hdr
->args
.count
= count
;
569 hdr
->args
.context
= get_nfs_open_context(req
->wb_context
);
570 hdr
->args
.lock_context
= req
->wb_lock_context
;
571 hdr
->args
.stable
= NFS_UNSTABLE
;
572 switch (how
& (FLUSH_STABLE
| FLUSH_COND_STABLE
)) {
575 case FLUSH_COND_STABLE
:
576 if (nfs_reqs_to_commit(cinfo
))
579 hdr
->args
.stable
= NFS_FILE_SYNC
;
582 hdr
->res
.fattr
= &hdr
->fattr
;
583 hdr
->res
.count
= count
;
585 hdr
->res
.verf
= &hdr
->verf
;
586 nfs_fattr_init(&hdr
->fattr
);
590 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
591 * @task: The current task
592 * @calldata: pageio header to prepare
594 static void nfs_pgio_prepare(struct rpc_task
*task
, void *calldata
)
596 struct nfs_pgio_header
*hdr
= calldata
;
598 err
= NFS_PROTO(hdr
->inode
)->pgio_rpc_prepare(task
, hdr
);
603 int nfs_initiate_pgio(struct rpc_clnt
*clnt
, struct nfs_pgio_header
*hdr
,
604 struct rpc_cred
*cred
, const struct nfs_rpc_ops
*rpc_ops
,
605 const struct rpc_call_ops
*call_ops
, int how
, int flags
)
607 struct rpc_task
*task
;
608 struct rpc_message msg
= {
609 .rpc_argp
= &hdr
->args
,
610 .rpc_resp
= &hdr
->res
,
613 struct rpc_task_setup task_setup_data
= {
617 .callback_ops
= call_ops
,
618 .callback_data
= hdr
,
619 .workqueue
= nfsiod_workqueue
,
620 .flags
= RPC_TASK_ASYNC
| flags
,
624 hdr
->rw_ops
->rw_initiate(hdr
, &msg
, rpc_ops
, &task_setup_data
, how
);
626 dprintk("NFS: initiated pgio call "
627 "(req %s/%llu, %u bytes @ offset %llu)\n",
628 hdr
->inode
->i_sb
->s_id
,
629 (unsigned long long)NFS_FILEID(hdr
->inode
),
631 (unsigned long long)hdr
->args
.offset
);
633 task
= rpc_run_task(&task_setup_data
);
638 if (how
& FLUSH_SYNC
) {
639 ret
= rpc_wait_for_completion_task(task
);
641 ret
= task
->tk_status
;
647 EXPORT_SYMBOL_GPL(nfs_initiate_pgio
);
650 * nfs_pgio_error - Clean up from a pageio error
651 * @desc: IO descriptor
652 * @hdr: pageio header
654 static void nfs_pgio_error(struct nfs_pgio_header
*hdr
)
656 set_bit(NFS_IOHDR_REDO
, &hdr
->flags
);
657 nfs_pgio_data_destroy(hdr
);
658 hdr
->completion_ops
->completion(hdr
);
662 * nfs_pgio_release - Release pageio data
663 * @calldata: The pageio header to release
665 static void nfs_pgio_release(void *calldata
)
667 struct nfs_pgio_header
*hdr
= calldata
;
668 nfs_pgio_data_destroy(hdr
);
669 hdr
->completion_ops
->completion(hdr
);
672 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror
*mirror
,
675 INIT_LIST_HEAD(&mirror
->pg_list
);
676 mirror
->pg_bytes_written
= 0;
677 mirror
->pg_count
= 0;
678 mirror
->pg_bsize
= bsize
;
680 mirror
->pg_recoalesce
= 0;
684 * nfs_pageio_init - initialise a page io descriptor
685 * @desc: pointer to descriptor
686 * @inode: pointer to inode
687 * @pg_ops: pointer to pageio operations
688 * @compl_ops: pointer to pageio completion operations
689 * @rw_ops: pointer to nfs read/write operations
690 * @bsize: io block size
691 * @io_flags: extra parameters for the io function
693 void nfs_pageio_init(struct nfs_pageio_descriptor
*desc
,
695 const struct nfs_pageio_ops
*pg_ops
,
696 const struct nfs_pgio_completion_ops
*compl_ops
,
697 const struct nfs_rw_ops
*rw_ops
,
702 struct nfs_pgio_mirror
*new;
706 desc
->pg_inode
= inode
;
707 desc
->pg_ops
= pg_ops
;
708 desc
->pg_completion_ops
= compl_ops
;
709 desc
->pg_rw_ops
= rw_ops
;
710 desc
->pg_ioflags
= io_flags
;
712 desc
->pg_lseg
= NULL
;
713 desc
->pg_dreq
= NULL
;
714 desc
->pg_layout_private
= NULL
;
715 desc
->pg_bsize
= bsize
;
717 desc
->pg_mirror_count
= 1;
718 desc
->pg_mirror_idx
= 0;
720 if (pg_ops
->pg_get_mirror_count
) {
721 /* until we have a request, we don't have an lseg and no
722 * idea how many mirrors there will be */
723 new = kcalloc(NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX
,
724 sizeof(struct nfs_pgio_mirror
), gfp_flags
);
725 desc
->pg_mirrors_dynamic
= new;
726 desc
->pg_mirrors
= new;
728 for (i
= 0; i
< NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX
; i
++)
729 nfs_pageio_mirror_init(&desc
->pg_mirrors
[i
], bsize
);
731 desc
->pg_mirrors_dynamic
= NULL
;
732 desc
->pg_mirrors
= desc
->pg_mirrors_static
;
733 nfs_pageio_mirror_init(&desc
->pg_mirrors
[0], bsize
);
736 EXPORT_SYMBOL_GPL(nfs_pageio_init
);
739 * nfs_pgio_result - Basic pageio error handling
740 * @task: The task that ran
741 * @calldata: Pageio header to check
743 static void nfs_pgio_result(struct rpc_task
*task
, void *calldata
)
745 struct nfs_pgio_header
*hdr
= calldata
;
746 struct inode
*inode
= hdr
->inode
;
748 dprintk("NFS: %s: %5u, (status %d)\n", __func__
,
749 task
->tk_pid
, task
->tk_status
);
751 if (hdr
->rw_ops
->rw_done(task
, hdr
, inode
) != 0)
753 if (task
->tk_status
< 0)
754 nfs_set_pgio_error(hdr
, task
->tk_status
, hdr
->args
.offset
);
756 hdr
->rw_ops
->rw_result(task
, hdr
);
760 * Create an RPC task for the given read or write request and kick it.
761 * The page must have been locked by the caller.
763 * It may happen that the page we're passed is not marked dirty.
764 * This is the case if nfs_updatepage detects a conflicting request
765 * that has been written but not committed.
767 int nfs_generic_pgio(struct nfs_pageio_descriptor
*desc
,
768 struct nfs_pgio_header
*hdr
)
770 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
772 struct nfs_page
*req
;
775 struct list_head
*head
= &mirror
->pg_list
;
776 struct nfs_commit_info cinfo
;
777 struct nfs_page_array
*pg_array
= &hdr
->page_array
;
778 unsigned int pagecount
, pageused
;
779 gfp_t gfp_flags
= GFP_KERNEL
;
781 pagecount
= nfs_page_array_len(mirror
->pg_base
, mirror
->pg_count
);
783 if (pagecount
<= ARRAY_SIZE(pg_array
->page_array
))
784 pg_array
->pagevec
= pg_array
->page_array
;
786 if (hdr
->rw_mode
== FMODE_WRITE
)
787 gfp_flags
= GFP_NOIO
;
788 pg_array
->pagevec
= kcalloc(pagecount
, sizeof(struct page
*), gfp_flags
);
789 if (!pg_array
->pagevec
) {
790 pg_array
->npages
= 0;
792 desc
->pg_error
= -ENOMEM
;
793 return desc
->pg_error
;
797 nfs_init_cinfo(&cinfo
, desc
->pg_inode
, desc
->pg_dreq
);
798 pages
= hdr
->page_array
.pagevec
;
801 while (!list_empty(head
)) {
802 req
= nfs_list_entry(head
->next
);
803 nfs_list_remove_request(req
);
804 nfs_list_add_request(req
, &hdr
->pages
);
806 if (!last_page
|| last_page
!= req
->wb_page
) {
808 if (pageused
> pagecount
)
810 *pages
++ = last_page
= req
->wb_page
;
813 if (WARN_ON_ONCE(pageused
!= pagecount
)) {
815 desc
->pg_error
= -EINVAL
;
816 return desc
->pg_error
;
819 if ((desc
->pg_ioflags
& FLUSH_COND_STABLE
) &&
820 (desc
->pg_moreio
|| nfs_reqs_to_commit(&cinfo
)))
821 desc
->pg_ioflags
&= ~FLUSH_COND_STABLE
;
823 /* Set up the argument struct */
824 nfs_pgio_rpcsetup(hdr
, mirror
->pg_count
, 0, desc
->pg_ioflags
, &cinfo
);
825 desc
->pg_rpc_callops
= &nfs_pgio_common_ops
;
828 EXPORT_SYMBOL_GPL(nfs_generic_pgio
);
830 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor
*desc
)
832 struct nfs_pgio_header
*hdr
;
835 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
837 desc
->pg_error
= -ENOMEM
;
838 return desc
->pg_error
;
840 nfs_pgheader_init(desc
, hdr
, nfs_pgio_header_free
);
841 ret
= nfs_generic_pgio(desc
, hdr
);
843 ret
= nfs_initiate_pgio(NFS_CLIENT(hdr
->inode
),
846 NFS_PROTO(hdr
->inode
),
847 desc
->pg_rpc_callops
,
848 desc
->pg_ioflags
, 0);
853 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
854 * by calling the pg_get_mirror_count op
856 static int nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor
*pgio
,
857 struct nfs_page
*req
)
859 int mirror_count
= 1;
861 if (!pgio
->pg_ops
->pg_get_mirror_count
)
864 mirror_count
= pgio
->pg_ops
->pg_get_mirror_count(pgio
, req
);
866 if (pgio
->pg_error
< 0)
867 return pgio
->pg_error
;
869 if (!mirror_count
|| mirror_count
> NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX
)
872 if (WARN_ON_ONCE(!pgio
->pg_mirrors_dynamic
))
875 pgio
->pg_mirror_count
= mirror_count
;
881 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
883 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor
*pgio
)
885 pgio
->pg_mirror_count
= 1;
886 pgio
->pg_mirror_idx
= 0;
889 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor
*pgio
)
891 pgio
->pg_mirror_count
= 1;
892 pgio
->pg_mirror_idx
= 0;
893 pgio
->pg_mirrors
= pgio
->pg_mirrors_static
;
894 kfree(pgio
->pg_mirrors_dynamic
);
895 pgio
->pg_mirrors_dynamic
= NULL
;
898 static bool nfs_match_lock_context(const struct nfs_lock_context
*l1
,
899 const struct nfs_lock_context
*l2
)
901 return l1
->lockowner
== l2
->lockowner
;
905 * nfs_can_coalesce_requests - test two requests for compatibility
906 * @prev: pointer to nfs_page
907 * @req: pointer to nfs_page
909 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
910 * page data area they describe is contiguous, and that their RPC
911 * credentials, NFSv4 open state, and lockowners are the same.
913 * Return 'true' if this is the case, else return 'false'.
915 static bool nfs_can_coalesce_requests(struct nfs_page
*prev
,
916 struct nfs_page
*req
,
917 struct nfs_pageio_descriptor
*pgio
)
920 struct file_lock_context
*flctx
;
923 if (!nfs_match_open_context(req
->wb_context
, prev
->wb_context
))
925 flctx
= d_inode(req
->wb_context
->dentry
)->i_flctx
;
927 !(list_empty_careful(&flctx
->flc_posix
) &&
928 list_empty_careful(&flctx
->flc_flock
)) &&
929 !nfs_match_lock_context(req
->wb_lock_context
,
930 prev
->wb_lock_context
))
932 if (req_offset(req
) != req_offset(prev
) + prev
->wb_bytes
)
934 if (req
->wb_page
== prev
->wb_page
) {
935 if (req
->wb_pgbase
!= prev
->wb_pgbase
+ prev
->wb_bytes
)
938 if (req
->wb_pgbase
!= 0 ||
939 prev
->wb_pgbase
+ prev
->wb_bytes
!= PAGE_SIZE
)
943 size
= pgio
->pg_ops
->pg_test(pgio
, prev
, req
);
944 WARN_ON_ONCE(size
> req
->wb_bytes
);
945 if (size
&& size
< req
->wb_bytes
)
946 req
->wb_bytes
= size
;
951 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
952 * @desc: destination io descriptor
955 * Returns true if the request 'req' was successfully coalesced into the
956 * existing list of pages 'desc'.
958 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor
*desc
,
959 struct nfs_page
*req
)
961 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
963 struct nfs_page
*prev
= NULL
;
965 if (mirror
->pg_count
!= 0) {
966 prev
= nfs_list_entry(mirror
->pg_list
.prev
);
968 if (desc
->pg_ops
->pg_init
)
969 desc
->pg_ops
->pg_init(desc
, req
);
970 if (desc
->pg_error
< 0)
972 mirror
->pg_base
= req
->wb_pgbase
;
974 if (!nfs_can_coalesce_requests(prev
, req
, desc
))
976 nfs_list_remove_request(req
);
977 nfs_list_add_request(req
, &mirror
->pg_list
);
978 mirror
->pg_count
+= req
->wb_bytes
;
983 * Helper for nfs_pageio_add_request and nfs_pageio_complete
985 static void nfs_pageio_doio(struct nfs_pageio_descriptor
*desc
)
987 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
990 if (!list_empty(&mirror
->pg_list
)) {
991 int error
= desc
->pg_ops
->pg_doio(desc
);
993 desc
->pg_error
= error
;
995 mirror
->pg_bytes_written
+= mirror
->pg_count
;
997 if (list_empty(&mirror
->pg_list
)) {
998 mirror
->pg_count
= 0;
1004 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1005 * @desc: destination io descriptor
1008 * This may split a request into subrequests which are all part of the
1011 * Returns true if the request 'req' was successfully coalesced into the
1012 * existing list of pages 'desc'.
1014 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor
*desc
,
1015 struct nfs_page
*req
)
1017 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
1019 struct nfs_page
*subreq
;
1020 unsigned int bytes_left
= 0;
1021 unsigned int offset
, pgbase
;
1023 nfs_page_group_lock(req
, false);
1026 bytes_left
= subreq
->wb_bytes
;
1027 offset
= subreq
->wb_offset
;
1028 pgbase
= subreq
->wb_pgbase
;
1031 if (!nfs_pageio_do_add_request(desc
, subreq
)) {
1032 /* make sure pg_test call(s) did nothing */
1033 WARN_ON_ONCE(subreq
->wb_bytes
!= bytes_left
);
1034 WARN_ON_ONCE(subreq
->wb_offset
!= offset
);
1035 WARN_ON_ONCE(subreq
->wb_pgbase
!= pgbase
);
1037 nfs_page_group_unlock(req
);
1038 desc
->pg_moreio
= 1;
1039 nfs_pageio_doio(desc
);
1040 if (desc
->pg_error
< 0)
1042 if (mirror
->pg_recoalesce
)
1044 /* retry add_request for this subreq */
1045 nfs_page_group_lock(req
, false);
1049 /* check for buggy pg_test call(s) */
1050 WARN_ON_ONCE(subreq
->wb_bytes
+ subreq
->wb_pgbase
> PAGE_SIZE
);
1051 WARN_ON_ONCE(subreq
->wb_bytes
> bytes_left
);
1052 WARN_ON_ONCE(subreq
->wb_bytes
== 0);
1054 bytes_left
-= subreq
->wb_bytes
;
1055 offset
+= subreq
->wb_bytes
;
1056 pgbase
+= subreq
->wb_bytes
;
1059 subreq
= nfs_create_request(req
->wb_context
,
1061 subreq
, pgbase
, bytes_left
);
1064 nfs_lock_request(subreq
);
1065 subreq
->wb_offset
= offset
;
1066 subreq
->wb_index
= req
->wb_index
;
1068 } while (bytes_left
> 0);
1070 nfs_page_group_unlock(req
);
1073 desc
->pg_error
= PTR_ERR(subreq
);
1074 nfs_page_group_unlock(req
);
1078 static int nfs_do_recoalesce(struct nfs_pageio_descriptor
*desc
)
1080 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
1084 list_splice_init(&mirror
->pg_list
, &head
);
1085 mirror
->pg_bytes_written
-= mirror
->pg_count
;
1086 mirror
->pg_count
= 0;
1087 mirror
->pg_base
= 0;
1088 mirror
->pg_recoalesce
= 0;
1090 while (!list_empty(&head
)) {
1091 struct nfs_page
*req
;
1093 req
= list_first_entry(&head
, struct nfs_page
, wb_list
);
1094 nfs_list_remove_request(req
);
1095 if (__nfs_pageio_add_request(desc
, req
))
1097 if (desc
->pg_error
< 0) {
1098 list_splice_tail(&head
, &mirror
->pg_list
);
1099 mirror
->pg_recoalesce
= 1;
1104 } while (mirror
->pg_recoalesce
);
1108 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor
*desc
,
1109 struct nfs_page
*req
)
1114 ret
= __nfs_pageio_add_request(desc
, req
);
1117 if (desc
->pg_error
< 0)
1119 ret
= nfs_do_recoalesce(desc
);
1125 int nfs_pageio_add_request(struct nfs_pageio_descriptor
*desc
,
1126 struct nfs_page
*req
)
1129 unsigned int pgbase
, offset
, bytes
;
1130 struct nfs_page
*dupreq
, *lastreq
;
1132 pgbase
= req
->wb_pgbase
;
1133 offset
= req
->wb_offset
;
1134 bytes
= req
->wb_bytes
;
1136 nfs_pageio_setup_mirroring(desc
, req
);
1137 if (desc
->pg_error
< 0)
1140 for (midx
= 0; midx
< desc
->pg_mirror_count
; midx
++) {
1142 nfs_page_group_lock(req
, false);
1144 /* find the last request */
1145 for (lastreq
= req
->wb_head
;
1146 lastreq
->wb_this_page
!= req
->wb_head
;
1147 lastreq
= lastreq
->wb_this_page
)
1150 dupreq
= nfs_create_request(req
->wb_context
,
1151 req
->wb_page
, lastreq
, pgbase
, bytes
);
1153 if (IS_ERR(dupreq
)) {
1154 nfs_page_group_unlock(req
);
1155 desc
->pg_error
= PTR_ERR(dupreq
);
1159 nfs_lock_request(dupreq
);
1160 nfs_page_group_unlock(req
);
1161 dupreq
->wb_offset
= offset
;
1162 dupreq
->wb_index
= req
->wb_index
;
1166 if (nfs_pgio_has_mirroring(desc
))
1167 desc
->pg_mirror_idx
= midx
;
1168 if (!nfs_pageio_add_request_mirror(desc
, dupreq
))
1176 * We might have failed before sending any reqs over wire.
1177 * Clean up rest of the reqs in mirror pg_list.
1179 if (desc
->pg_error
) {
1180 struct nfs_pgio_mirror
*mirror
;
1181 void (*func
)(struct list_head
*);
1183 /* remember fatal errors */
1184 if (nfs_error_is_fatal(desc
->pg_error
))
1185 mapping_set_error(desc
->pg_inode
->i_mapping
,
1188 func
= desc
->pg_completion_ops
->error_cleanup
;
1189 for (midx
= 0; midx
< desc
->pg_mirror_count
; midx
++) {
1190 mirror
= &desc
->pg_mirrors
[midx
];
1191 func(&mirror
->pg_list
);
1198 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1199 * nfs_pageio_descriptor
1200 * @desc: pointer to io descriptor
1201 * @mirror_idx: pointer to mirror index
1203 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor
*desc
,
1206 struct nfs_pgio_mirror
*mirror
= &desc
->pg_mirrors
[mirror_idx
];
1207 u32 restore_idx
= desc
->pg_mirror_idx
;
1209 if (nfs_pgio_has_mirroring(desc
))
1210 desc
->pg_mirror_idx
= mirror_idx
;
1212 nfs_pageio_doio(desc
);
1213 if (!mirror
->pg_recoalesce
)
1215 if (!nfs_do_recoalesce(desc
))
1218 desc
->pg_mirror_idx
= restore_idx
;
1222 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1223 * @hdr - the pgio header to move request from
1224 * @desc - the pageio descriptor to add requests to
1226 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1229 * Returns 0 on success and < 0 on error.
1231 int nfs_pageio_resend(struct nfs_pageio_descriptor
*desc
,
1232 struct nfs_pgio_header
*hdr
)
1236 desc
->pg_dreq
= hdr
->dreq
;
1237 while (!list_empty(&hdr
->pages
)) {
1238 struct nfs_page
*req
= nfs_list_entry(hdr
->pages
.next
);
1240 nfs_list_remove_request(req
);
1241 if (!nfs_pageio_add_request(desc
, req
))
1242 nfs_list_add_request(req
, &failed
);
1244 nfs_pageio_complete(desc
);
1245 if (!list_empty(&failed
)) {
1246 list_move(&failed
, &hdr
->pages
);
1247 return desc
->pg_error
< 0 ? desc
->pg_error
: -EIO
;
1251 EXPORT_SYMBOL_GPL(nfs_pageio_resend
);
1254 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1255 * @desc: pointer to io descriptor
1257 void nfs_pageio_complete(struct nfs_pageio_descriptor
*desc
)
1261 for (midx
= 0; midx
< desc
->pg_mirror_count
; midx
++)
1262 nfs_pageio_complete_mirror(desc
, midx
);
1264 if (desc
->pg_ops
->pg_cleanup
)
1265 desc
->pg_ops
->pg_cleanup(desc
);
1266 nfs_pageio_cleanup_mirroring(desc
);
1270 * nfs_pageio_cond_complete - Conditional I/O completion
1271 * @desc: pointer to io descriptor
1272 * @index: page index
1274 * It is important to ensure that processes don't try to take locks
1275 * on non-contiguous ranges of pages as that might deadlock. This
1276 * function should be called before attempting to wait on a locked
1277 * nfs_page. It will complete the I/O if the page index 'index'
1278 * is not contiguous with the existing list of pages in 'desc'.
1280 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor
*desc
, pgoff_t index
)
1282 struct nfs_pgio_mirror
*mirror
;
1283 struct nfs_page
*prev
;
1286 for (midx
= 0; midx
< desc
->pg_mirror_count
; midx
++) {
1287 mirror
= &desc
->pg_mirrors
[midx
];
1288 if (!list_empty(&mirror
->pg_list
)) {
1289 prev
= nfs_list_entry(mirror
->pg_list
.prev
);
1290 if (index
!= prev
->wb_index
+ 1) {
1291 nfs_pageio_complete(desc
);
1298 int __init
nfs_init_nfspagecache(void)
1300 nfs_page_cachep
= kmem_cache_create("nfs_page",
1301 sizeof(struct nfs_page
),
1302 0, SLAB_HWCACHE_ALIGN
,
1304 if (nfs_page_cachep
== NULL
)
1310 void nfs_destroy_nfspagecache(void)
1312 kmem_cache_destroy(nfs_page_cachep
);
1315 static const struct rpc_call_ops nfs_pgio_common_ops
= {
1316 .rpc_call_prepare
= nfs_pgio_prepare
,
1317 .rpc_call_done
= nfs_pgio_result
,
1318 .rpc_release
= nfs_pgio_release
,
1321 const struct nfs_pageio_ops nfs_pgio_rw_ops
= {
1322 .pg_test
= nfs_generic_pg_test
,
1323 .pg_doio
= nfs_generic_pg_pgios
,