1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/fs/nfs/pagelist.c
5 * A set of helper functions for managing NFS read and write requests.
6 * The main purpose of these routines is to provide support for the
7 * coalescing of several requests into a single RPC call.
9 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
13 #include <linux/slab.h>
14 #include <linux/file.h>
15 #include <linux/sched.h>
16 #include <linux/sunrpc/clnt.h>
17 #include <linux/nfs.h>
18 #include <linux/nfs3.h>
19 #include <linux/nfs4.h>
20 #include <linux/nfs_fs.h>
21 #include <linux/nfs_page.h>
22 #include <linux/nfs_mount.h>
23 #include <linux/export.h>
29 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
31 static struct kmem_cache *nfs_page_cachep;
32 static const struct rpc_call_ops nfs_pgio_common_ops;
34 static struct nfs_pgio_mirror *
35 nfs_pgio_get_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
37 if (desc->pg_ops->pg_get_mirror)
38 return desc->pg_ops->pg_get_mirror(desc, idx);
39 return &desc->pg_mirrors[0];
42 struct nfs_pgio_mirror *
43 nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
45 return nfs_pgio_get_mirror(desc, desc->pg_mirror_idx);
47 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
50 nfs_pgio_set_current_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
52 if (desc->pg_ops->pg_set_mirror)
53 return desc->pg_ops->pg_set_mirror(desc, idx);
54 return desc->pg_mirror_idx;
57 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
58 struct nfs_pgio_header *hdr,
59 void (*release)(struct nfs_pgio_header *hdr))
61 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
64 hdr->req = nfs_list_entry(mirror->pg_list.next);
65 hdr->inode = desc->pg_inode;
66 hdr->cred = nfs_req_openctx(hdr->req)->cred;
67 hdr->io_start = req_offset(hdr->req);
68 hdr->good_bytes = mirror->pg_count;
69 hdr->io_completion = desc->pg_io_completion;
70 hdr->dreq = desc->pg_dreq;
71 hdr->release = release;
72 hdr->completion_ops = desc->pg_completion_ops;
73 if (hdr->completion_ops->init_hdr)
74 hdr->completion_ops->init_hdr(hdr);
76 hdr->pgio_mirror_idx = desc->pg_mirror_idx;
78 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
80 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
82 unsigned int new = pos - hdr->io_start;
84 trace_nfs_pgio_error(hdr, error, pos);
85 if (hdr->good_bytes > new) {
86 hdr->good_bytes = new;
87 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
88 if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
93 static inline struct nfs_page *nfs_page_alloc(void)
96 kmem_cache_zalloc(nfs_page_cachep, nfs_io_gfp_mask());
98 INIT_LIST_HEAD(&p->wb_list);
103 nfs_page_free(struct nfs_page *p)
105 kmem_cache_free(nfs_page_cachep, p);
109 * nfs_iocounter_wait - wait for i/o to complete
110 * @l_ctx: nfs_lock_context with io_counter to use
112 * returns -ERESTARTSYS if interrupted by a fatal signal.
113 * Otherwise returns 0 once the io_count hits 0.
116 nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
118 return wait_var_event_killable(&l_ctx->io_count,
119 !atomic_read(&l_ctx->io_count));
123 * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
125 * @task: the rpc_task that should wait
126 * @l_ctx: nfs_lock_context with io_counter to check
128 * Returns true if there is outstanding I/O to wait on and the
129 * task has been put to sleep.
132 nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
134 struct inode *inode = d_inode(l_ctx->open_context->dentry);
137 if (atomic_read(&l_ctx->io_count) > 0) {
138 rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
142 if (atomic_read(&l_ctx->io_count) == 0) {
143 rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
149 EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
152 * nfs_page_lock_head_request - page lock the head of the page group
153 * @req: any member of the page group
156 nfs_page_group_lock_head(struct nfs_page *req)
158 struct nfs_page *head = req->wb_head;
160 while (!nfs_lock_request(head)) {
161 int ret = nfs_wait_on_request(head);
166 kref_get(&head->wb_kref);
171 * nfs_unroll_locks - unlock all newly locked reqs and wait on @req
172 * @head: head request of page group, must be holding head lock
173 * @req: request that couldn't lock and needs to wait on the req bit lock
175 * This is a helper function for nfs_lock_and_join_requests
176 * returns 0 on success, < 0 on error.
179 nfs_unroll_locks(struct nfs_page *head, struct nfs_page *req)
181 struct nfs_page *tmp;
183 /* relinquish all the locks successfully grabbed this run */
184 for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) {
185 if (!kref_read(&tmp->wb_kref))
187 nfs_unlock_and_release_request(tmp);
192 * nfs_page_group_lock_subreq - try to lock a subrequest
193 * @head: head request of page group
194 * @subreq: request to lock
196 * This is a helper function for nfs_lock_and_join_requests which
197 * must be called with the head request and page group both locked.
198 * On error, it returns with the page group unlocked.
201 nfs_page_group_lock_subreq(struct nfs_page *head, struct nfs_page *subreq)
205 if (!kref_get_unless_zero(&subreq->wb_kref))
207 while (!nfs_lock_request(subreq)) {
208 nfs_page_group_unlock(head);
209 ret = nfs_wait_on_request(subreq);
211 ret = nfs_page_group_lock(head);
213 nfs_unroll_locks(head, subreq);
214 nfs_release_request(subreq);
222 * nfs_page_group_lock_subrequests - try to lock the subrequests
223 * @head: head request of page group
225 * This is a helper function for nfs_lock_and_join_requests which
226 * must be called with the head request locked.
228 int nfs_page_group_lock_subrequests(struct nfs_page *head)
230 struct nfs_page *subreq;
233 ret = nfs_page_group_lock(head);
236 /* lock each request in the page group */
237 for (subreq = head->wb_this_page; subreq != head;
238 subreq = subreq->wb_this_page) {
239 ret = nfs_page_group_lock_subreq(head, subreq);
243 nfs_page_group_unlock(head);
248 * nfs_page_set_headlock - set the request PG_HEADLOCK
249 * @req: request that is to be locked
251 * this lock must be held when modifying req->wb_head
253 * return 0 on success, < 0 on error
256 nfs_page_set_headlock(struct nfs_page *req)
258 if (!test_and_set_bit(PG_HEADLOCK, &req->wb_flags))
261 set_bit(PG_CONTENDED1, &req->wb_flags);
262 smp_mb__after_atomic();
263 return wait_on_bit_lock(&req->wb_flags, PG_HEADLOCK,
264 TASK_UNINTERRUPTIBLE);
268 * nfs_page_clear_headlock - clear the request PG_HEADLOCK
269 * @req: request that is to be locked
272 nfs_page_clear_headlock(struct nfs_page *req)
274 smp_mb__before_atomic();
275 clear_bit(PG_HEADLOCK, &req->wb_flags);
276 smp_mb__after_atomic();
277 if (!test_bit(PG_CONTENDED1, &req->wb_flags))
279 wake_up_bit(&req->wb_flags, PG_HEADLOCK);
283 * nfs_page_group_lock - lock the head of the page group
284 * @req: request in group that is to be locked
286 * this lock must be held when traversing or modifying the page
289 * return 0 on success, < 0 on error
292 nfs_page_group_lock(struct nfs_page *req)
296 ret = nfs_page_set_headlock(req);
297 if (ret || req->wb_head == req)
299 return nfs_page_set_headlock(req->wb_head);
303 * nfs_page_group_unlock - unlock the head of the page group
304 * @req: request in group that is to be unlocked
307 nfs_page_group_unlock(struct nfs_page *req)
309 if (req != req->wb_head)
310 nfs_page_clear_headlock(req->wb_head);
311 nfs_page_clear_headlock(req);
315 * nfs_page_group_sync_on_bit_locked
317 * must be called with page group lock held
320 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
322 struct nfs_page *head = req->wb_head;
323 struct nfs_page *tmp;
325 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
326 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
328 tmp = req->wb_this_page;
330 if (!test_bit(bit, &tmp->wb_flags))
332 tmp = tmp->wb_this_page;
335 /* true! reset all bits */
338 clear_bit(bit, &tmp->wb_flags);
339 tmp = tmp->wb_this_page;
340 } while (tmp != req);
346 * nfs_page_group_sync_on_bit - set bit on current request, but only
347 * return true if the bit is set for all requests in page group
348 * @req - request in page group
349 * @bit - PG_* bit that is used to sync page group
351 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
355 nfs_page_group_lock(req);
356 ret = nfs_page_group_sync_on_bit_locked(req, bit);
357 nfs_page_group_unlock(req);
363 * nfs_page_group_init - Initialize the page group linkage for @req
364 * @req - a new nfs request
365 * @prev - the previous request in page group, or NULL if @req is the first
366 * or only request in the group (the head).
369 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
372 WARN_ON_ONCE(prev == req);
377 req->wb_this_page = req;
380 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
381 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
382 req->wb_head = prev->wb_head;
383 req->wb_this_page = prev->wb_this_page;
384 prev->wb_this_page = req;
386 /* All subrequests take a ref on the head request until
387 * nfs_page_group_destroy is called */
388 kref_get(&req->wb_head->wb_kref);
390 /* grab extra ref and bump the request count if head request
391 * has extra ref from the write/commit path to handle handoff
392 * between write and commit lists. */
393 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
394 inode = page_file_mapping(req->wb_page)->host;
395 set_bit(PG_INODE_REF, &req->wb_flags);
396 kref_get(&req->wb_kref);
397 atomic_long_inc(&NFS_I(inode)->nrequests);
403 * nfs_page_group_destroy - sync the destruction of page groups
404 * @req - request that no longer needs the page group
406 * releases the page group reference from each member once all
407 * members have called this function.
410 nfs_page_group_destroy(struct kref *kref)
412 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
413 struct nfs_page *head = req->wb_head;
414 struct nfs_page *tmp, *next;
416 if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
421 next = tmp->wb_this_page;
422 /* unlink and free */
423 tmp->wb_this_page = tmp;
425 nfs_free_request(tmp);
427 } while (tmp != req);
429 /* subrequests must release the ref on the head request */
431 nfs_release_request(head);
434 static struct nfs_page *
435 __nfs_create_request(struct nfs_lock_context *l_ctx, struct page *page,
436 unsigned int pgbase, unsigned int offset,
439 struct nfs_page *req;
440 struct nfs_open_context *ctx = l_ctx->open_context;
442 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
443 return ERR_PTR(-EBADF);
444 /* try to allocate the request struct */
445 req = nfs_page_alloc();
447 return ERR_PTR(-ENOMEM);
449 req->wb_lock_context = l_ctx;
450 refcount_inc(&l_ctx->count);
451 atomic_inc(&l_ctx->io_count);
453 /* Initialize the request struct. Initially, we assume a
454 * long write-back delay. This will be adjusted in
455 * update_nfs_request below if the region is not locked. */
458 req->wb_index = page_index(page);
461 req->wb_offset = offset;
462 req->wb_pgbase = pgbase;
463 req->wb_bytes = count;
464 kref_init(&req->wb_kref);
470 * nfs_create_request - Create an NFS read/write request.
471 * @ctx: open context to use
472 * @page: page to write
473 * @offset: starting offset within the page for the write
474 * @count: number of bytes to read/write
476 * The page must be locked by the caller. This makes sure we never
477 * create two different requests for the same page.
478 * User should ensure it is safe to sleep in this function.
481 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
482 unsigned int offset, unsigned int count)
484 struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
485 struct nfs_page *ret;
488 return ERR_CAST(l_ctx);
489 ret = __nfs_create_request(l_ctx, page, offset, offset, count);
491 nfs_page_group_init(ret, NULL);
492 nfs_put_lock_context(l_ctx);
496 static struct nfs_page *
497 nfs_create_subreq(struct nfs_page *req,
502 struct nfs_page *last;
503 struct nfs_page *ret;
505 ret = __nfs_create_request(req->wb_lock_context, req->wb_page,
506 pgbase, offset, count);
508 /* find the last request */
509 for (last = req->wb_head;
510 last->wb_this_page != req->wb_head;
511 last = last->wb_this_page)
514 nfs_lock_request(ret);
515 ret->wb_index = req->wb_index;
516 nfs_page_group_init(ret, last);
517 ret->wb_nio = req->wb_nio;
523 * nfs_unlock_request - Unlock request and wake up sleepers.
524 * @req: pointer to request
526 void nfs_unlock_request(struct nfs_page *req)
528 if (!NFS_WBACK_BUSY(req)) {
529 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
532 smp_mb__before_atomic();
533 clear_bit(PG_BUSY, &req->wb_flags);
534 smp_mb__after_atomic();
535 if (!test_bit(PG_CONTENDED2, &req->wb_flags))
537 wake_up_bit(&req->wb_flags, PG_BUSY);
541 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
542 * @req: pointer to request
544 void nfs_unlock_and_release_request(struct nfs_page *req)
546 nfs_unlock_request(req);
547 nfs_release_request(req);
551 * nfs_clear_request - Free up all resources allocated to the request
554 * Release page and open context resources associated with a read/write
555 * request after it has completed.
557 static void nfs_clear_request(struct nfs_page *req)
559 struct page *page = req->wb_page;
560 struct nfs_lock_context *l_ctx = req->wb_lock_context;
561 struct nfs_open_context *ctx;
568 if (atomic_dec_and_test(&l_ctx->io_count)) {
569 wake_up_var(&l_ctx->io_count);
570 ctx = l_ctx->open_context;
571 if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
572 rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
574 nfs_put_lock_context(l_ctx);
575 req->wb_lock_context = NULL;
580 * nfs_free_request - Release the count on an NFS read/write request
581 * @req: request to release
583 * Note: Should never be called with the spinlock held!
585 void nfs_free_request(struct nfs_page *req)
587 WARN_ON_ONCE(req->wb_this_page != req);
589 /* extra debug: make sure no sync bits are still set */
590 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
591 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
592 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
593 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
594 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
596 /* Release struct file and open context */
597 nfs_clear_request(req);
601 void nfs_release_request(struct nfs_page *req)
603 kref_put(&req->wb_kref, nfs_page_group_destroy);
605 EXPORT_SYMBOL_GPL(nfs_release_request);
608 * nfs_wait_on_request - Wait for a request to complete.
609 * @req: request to wait upon.
611 * Interruptible by fatal signals only.
612 * The user is responsible for holding a count on the request.
615 nfs_wait_on_request(struct nfs_page *req)
617 if (!test_bit(PG_BUSY, &req->wb_flags))
619 set_bit(PG_CONTENDED2, &req->wb_flags);
620 smp_mb__after_atomic();
621 return wait_on_bit_io(&req->wb_flags, PG_BUSY,
622 TASK_UNINTERRUPTIBLE);
624 EXPORT_SYMBOL_GPL(nfs_wait_on_request);
627 * nfs_generic_pg_test - determine if requests can be coalesced
628 * @desc: pointer to descriptor
629 * @prev: previous request in desc, or NULL
632 * Returns zero if @req cannot be coalesced into @desc, otherwise it returns
633 * the size of the request.
635 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
636 struct nfs_page *prev, struct nfs_page *req)
638 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
641 if (mirror->pg_count > mirror->pg_bsize) {
642 /* should never happen */
648 * Limit the request size so that we can still allocate a page array
649 * for it without upsetting the slab allocator.
651 if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
652 sizeof(struct page *) > PAGE_SIZE)
655 return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
657 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
659 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
661 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
664 INIT_LIST_HEAD(&hdr->pages);
669 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
672 * nfs_pgio_data_destroy - make @hdr suitable for reuse
674 * Frees memory and releases refs from nfs_generic_pgio, so that it may
677 * @hdr: A header that has had nfs_generic_pgio called
679 static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
681 if (hdr->args.context)
682 put_nfs_open_context(hdr->args.context);
683 if (hdr->page_array.pagevec != hdr->page_array.page_array)
684 kfree(hdr->page_array.pagevec);
688 * nfs_pgio_header_free - Free a read or write header
689 * @hdr: The header to free
691 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
693 nfs_pgio_data_destroy(hdr);
694 hdr->rw_ops->rw_free_header(hdr);
696 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
699 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
700 * @hdr: The pageio hdr
701 * @count: Number of bytes to read
702 * @how: How to commit data (writes only)
703 * @cinfo: Commit information for the call (writes only)
705 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
707 int how, struct nfs_commit_info *cinfo)
709 struct nfs_page *req = hdr->req;
711 /* Set up the RPC argument and reply structs
712 * NB: take care not to mess about with hdr->commit et al. */
714 hdr->args.fh = NFS_FH(hdr->inode);
715 hdr->args.offset = req_offset(req);
716 /* pnfs_set_layoutcommit needs this */
717 hdr->mds_offset = hdr->args.offset;
718 hdr->args.pgbase = req->wb_pgbase;
719 hdr->args.pages = hdr->page_array.pagevec;
720 hdr->args.count = count;
721 hdr->args.context = get_nfs_open_context(nfs_req_openctx(req));
722 hdr->args.lock_context = req->wb_lock_context;
723 hdr->args.stable = NFS_UNSTABLE;
724 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
727 case FLUSH_COND_STABLE:
728 if (nfs_reqs_to_commit(cinfo))
732 hdr->args.stable = NFS_FILE_SYNC;
735 hdr->res.fattr = &hdr->fattr;
738 hdr->res.verf = &hdr->verf;
739 nfs_fattr_init(&hdr->fattr);
743 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
744 * @task: The current task
745 * @calldata: pageio header to prepare
747 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
749 struct nfs_pgio_header *hdr = calldata;
751 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
756 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
757 const struct cred *cred, const struct nfs_rpc_ops *rpc_ops,
758 const struct rpc_call_ops *call_ops, int how, int flags)
760 struct rpc_task *task;
761 struct rpc_message msg = {
762 .rpc_argp = &hdr->args,
763 .rpc_resp = &hdr->res,
766 struct rpc_task_setup task_setup_data = {
770 .callback_ops = call_ops,
771 .callback_data = hdr,
772 .workqueue = nfsiod_workqueue,
773 .flags = RPC_TASK_ASYNC | flags,
776 if (nfs_server_capable(hdr->inode, NFS_CAP_MOVEABLE))
777 task_setup_data.flags |= RPC_TASK_MOVEABLE;
779 hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
781 dprintk("NFS: initiated pgio call "
782 "(req %s/%llu, %u bytes @ offset %llu)\n",
783 hdr->inode->i_sb->s_id,
784 (unsigned long long)NFS_FILEID(hdr->inode),
786 (unsigned long long)hdr->args.offset);
788 task = rpc_run_task(&task_setup_data);
790 return PTR_ERR(task);
794 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
797 * nfs_pgio_error - Clean up from a pageio error
798 * @hdr: pageio header
800 static void nfs_pgio_error(struct nfs_pgio_header *hdr)
802 set_bit(NFS_IOHDR_REDO, &hdr->flags);
803 hdr->completion_ops->completion(hdr);
807 * nfs_pgio_release - Release pageio data
808 * @calldata: The pageio header to release
810 static void nfs_pgio_release(void *calldata)
812 struct nfs_pgio_header *hdr = calldata;
813 hdr->completion_ops->completion(hdr);
816 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
819 INIT_LIST_HEAD(&mirror->pg_list);
820 mirror->pg_bytes_written = 0;
821 mirror->pg_count = 0;
822 mirror->pg_bsize = bsize;
824 mirror->pg_recoalesce = 0;
828 * nfs_pageio_init - initialise a page io descriptor
829 * @desc: pointer to descriptor
830 * @inode: pointer to inode
831 * @pg_ops: pointer to pageio operations
832 * @compl_ops: pointer to pageio completion operations
833 * @rw_ops: pointer to nfs read/write operations
834 * @bsize: io block size
835 * @io_flags: extra parameters for the io function
837 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
839 const struct nfs_pageio_ops *pg_ops,
840 const struct nfs_pgio_completion_ops *compl_ops,
841 const struct nfs_rw_ops *rw_ops,
846 desc->pg_inode = inode;
847 desc->pg_ops = pg_ops;
848 desc->pg_completion_ops = compl_ops;
849 desc->pg_rw_ops = rw_ops;
850 desc->pg_ioflags = io_flags;
852 desc->pg_lseg = NULL;
853 desc->pg_io_completion = NULL;
854 desc->pg_dreq = NULL;
855 desc->pg_bsize = bsize;
857 desc->pg_mirror_count = 1;
858 desc->pg_mirror_idx = 0;
860 desc->pg_mirrors_dynamic = NULL;
861 desc->pg_mirrors = desc->pg_mirrors_static;
862 nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
863 desc->pg_maxretrans = 0;
867 * nfs_pgio_result - Basic pageio error handling
868 * @task: The task that ran
869 * @calldata: Pageio header to check
871 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
873 struct nfs_pgio_header *hdr = calldata;
874 struct inode *inode = hdr->inode;
876 dprintk("NFS: %s: %5u, (status %d)\n", __func__,
877 task->tk_pid, task->tk_status);
879 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
881 if (task->tk_status < 0)
882 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
884 hdr->rw_ops->rw_result(task, hdr);
888 * Create an RPC task for the given read or write request and kick it.
889 * The page must have been locked by the caller.
891 * It may happen that the page we're passed is not marked dirty.
892 * This is the case if nfs_updatepage detects a conflicting request
893 * that has been written but not committed.
895 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
896 struct nfs_pgio_header *hdr)
898 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
900 struct nfs_page *req;
903 struct list_head *head = &mirror->pg_list;
904 struct nfs_commit_info cinfo;
905 struct nfs_page_array *pg_array = &hdr->page_array;
906 unsigned int pagecount, pageused;
907 gfp_t gfp_flags = nfs_io_gfp_mask();
909 pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
910 pg_array->npages = pagecount;
912 if (pagecount <= ARRAY_SIZE(pg_array->page_array))
913 pg_array->pagevec = pg_array->page_array;
915 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
916 if (!pg_array->pagevec) {
917 pg_array->npages = 0;
919 desc->pg_error = -ENOMEM;
920 return desc->pg_error;
924 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
925 pages = hdr->page_array.pagevec;
928 while (!list_empty(head)) {
929 req = nfs_list_entry(head->next);
930 nfs_list_move_request(req, &hdr->pages);
932 if (!last_page || last_page != req->wb_page) {
934 if (pageused > pagecount)
936 *pages++ = last_page = req->wb_page;
939 if (WARN_ON_ONCE(pageused != pagecount)) {
941 desc->pg_error = -EINVAL;
942 return desc->pg_error;
945 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
946 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
947 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
949 /* Set up the argument struct */
950 nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo);
951 desc->pg_rpc_callops = &nfs_pgio_common_ops;
954 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
956 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
958 struct nfs_pgio_header *hdr;
960 unsigned short task_flags = 0;
962 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
964 desc->pg_error = -ENOMEM;
965 return desc->pg_error;
967 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
968 ret = nfs_generic_pgio(desc, hdr);
970 if (NFS_SERVER(hdr->inode)->nfs_client->cl_minorversion)
971 task_flags = RPC_TASK_MOVEABLE;
972 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
975 NFS_PROTO(hdr->inode),
976 desc->pg_rpc_callops,
978 RPC_TASK_CRED_NOREF | task_flags);
983 static struct nfs_pgio_mirror *
984 nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
985 unsigned int mirror_count)
987 struct nfs_pgio_mirror *ret;
990 kfree(desc->pg_mirrors_dynamic);
991 desc->pg_mirrors_dynamic = NULL;
992 if (mirror_count == 1)
993 return desc->pg_mirrors_static;
994 ret = kmalloc_array(mirror_count, sizeof(*ret), nfs_io_gfp_mask());
996 for (i = 0; i < mirror_count; i++)
997 nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
998 desc->pg_mirrors_dynamic = ret;
1004 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
1005 * by calling the pg_get_mirror_count op
1007 static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
1008 struct nfs_page *req)
1010 unsigned int mirror_count = 1;
1012 if (pgio->pg_ops->pg_get_mirror_count)
1013 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
1014 if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
1017 if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
1018 pgio->pg_error = -EINVAL;
1022 pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
1023 if (pgio->pg_mirrors == NULL) {
1024 pgio->pg_error = -ENOMEM;
1025 pgio->pg_mirrors = pgio->pg_mirrors_static;
1028 pgio->pg_mirror_count = mirror_count;
1031 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
1033 pgio->pg_mirror_count = 1;
1034 pgio->pg_mirror_idx = 0;
1035 pgio->pg_mirrors = pgio->pg_mirrors_static;
1036 kfree(pgio->pg_mirrors_dynamic);
1037 pgio->pg_mirrors_dynamic = NULL;
1040 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
1041 const struct nfs_lock_context *l2)
1043 return l1->lockowner == l2->lockowner;
1047 * nfs_coalesce_size - test two requests for compatibility
1048 * @prev: pointer to nfs_page
1049 * @req: pointer to nfs_page
1050 * @pgio: pointer to nfs_pagio_descriptor
1052 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
1053 * page data area they describe is contiguous, and that their RPC
1054 * credentials, NFSv4 open state, and lockowners are the same.
1056 * Returns size of the request that can be coalesced
1058 static unsigned int nfs_coalesce_size(struct nfs_page *prev,
1059 struct nfs_page *req,
1060 struct nfs_pageio_descriptor *pgio)
1062 struct file_lock_context *flctx;
1065 if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
1067 flctx = d_inode(nfs_req_openctx(req)->dentry)->i_flctx;
1068 if (flctx != NULL &&
1069 !(list_empty_careful(&flctx->flc_posix) &&
1070 list_empty_careful(&flctx->flc_flock)) &&
1071 !nfs_match_lock_context(req->wb_lock_context,
1072 prev->wb_lock_context))
1074 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
1076 if (req->wb_page == prev->wb_page) {
1077 if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
1080 if (req->wb_pgbase != 0 ||
1081 prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE)
1085 return pgio->pg_ops->pg_test(pgio, prev, req);
1089 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
1090 * @desc: destination io descriptor
1093 * If the request 'req' was successfully coalesced into the existing list
1094 * of pages 'desc', it returns the size of req.
1097 nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
1098 struct nfs_page *req)
1100 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1101 struct nfs_page *prev = NULL;
1104 if (list_empty(&mirror->pg_list)) {
1105 if (desc->pg_ops->pg_init)
1106 desc->pg_ops->pg_init(desc, req);
1107 if (desc->pg_error < 0)
1109 mirror->pg_base = req->wb_pgbase;
1110 mirror->pg_count = 0;
1111 mirror->pg_recoalesce = 0;
1113 prev = nfs_list_entry(mirror->pg_list.prev);
1115 if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
1116 if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
1117 desc->pg_error = -ETIMEDOUT;
1119 desc->pg_error = -EIO;
1123 size = nfs_coalesce_size(prev, req, desc);
1124 if (size < req->wb_bytes)
1126 nfs_list_move_request(req, &mirror->pg_list);
1127 mirror->pg_count += req->wb_bytes;
1128 return req->wb_bytes;
1132 * Helper for nfs_pageio_add_request and nfs_pageio_complete
1134 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1136 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1138 if (!list_empty(&mirror->pg_list)) {
1139 int error = desc->pg_ops->pg_doio(desc);
1141 desc->pg_error = error;
1142 if (list_empty(&mirror->pg_list))
1143 mirror->pg_bytes_written += mirror->pg_count;
1148 nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1149 struct nfs_page *req)
1153 nfs_list_move_request(req, &head);
1154 desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1158 * __nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1159 * @desc: destination io descriptor
1162 * This may split a request into subrequests which are all part of the
1163 * same page group. If so, it will submit @req as the last one, to ensure
1164 * the pointer to @req is still valid in case of failure.
1166 * Returns true if the request 'req' was successfully coalesced into the
1167 * existing list of pages 'desc'.
1169 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1170 struct nfs_page *req)
1172 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1173 struct nfs_page *subreq;
1174 unsigned int size, subreq_size;
1176 nfs_page_group_lock(req);
1179 subreq_size = subreq->wb_bytes;
1181 size = nfs_pageio_do_add_request(desc, subreq);
1182 if (size == subreq_size) {
1183 /* We successfully submitted a request */
1186 req->wb_pgbase += size;
1187 req->wb_bytes -= size;
1188 req->wb_offset += size;
1189 subreq_size = req->wb_bytes;
1193 if (WARN_ON_ONCE(subreq != req)) {
1194 nfs_page_group_unlock(req);
1195 nfs_pageio_cleanup_request(desc, subreq);
1197 subreq_size = req->wb_bytes;
1198 nfs_page_group_lock(req);
1201 /* Can't coalesce any more, so do I/O */
1202 nfs_page_group_unlock(req);
1203 desc->pg_moreio = 1;
1204 nfs_pageio_doio(desc);
1205 if (desc->pg_error < 0 || mirror->pg_recoalesce)
1207 /* retry add_request for this subreq */
1208 nfs_page_group_lock(req);
1211 subreq = nfs_create_subreq(req, req->wb_pgbase,
1212 req->wb_offset, size);
1218 nfs_page_group_unlock(req);
1221 desc->pg_error = PTR_ERR(subreq);
1222 nfs_page_group_unlock(req);
1226 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1228 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1232 list_splice_init(&mirror->pg_list, &head);
1233 mirror->pg_recoalesce = 0;
1235 while (!list_empty(&head)) {
1236 struct nfs_page *req;
1238 req = list_first_entry(&head, struct nfs_page, wb_list);
1239 if (__nfs_pageio_add_request(desc, req))
1241 if (desc->pg_error < 0) {
1242 list_splice_tail(&head, &mirror->pg_list);
1243 mirror->pg_recoalesce = 1;
1248 } while (mirror->pg_recoalesce);
1252 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1253 struct nfs_page *req)
1258 ret = __nfs_pageio_add_request(desc, req);
1261 if (desc->pg_error < 0)
1263 ret = nfs_do_recoalesce(desc);
1269 static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1272 struct nfs_pgio_mirror *mirror;
1274 if (!desc->pg_error)
1277 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1278 mirror = nfs_pgio_get_mirror(desc, midx);
1279 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1284 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1285 struct nfs_page *req)
1288 unsigned int pgbase, offset, bytes;
1289 struct nfs_page *dupreq;
1291 pgbase = req->wb_pgbase;
1292 offset = req->wb_offset;
1293 bytes = req->wb_bytes;
1295 nfs_pageio_setup_mirroring(desc, req);
1296 if (desc->pg_error < 0)
1299 /* Create the mirror instances first, and fire them off */
1300 for (midx = 1; midx < desc->pg_mirror_count; midx++) {
1301 nfs_page_group_lock(req);
1303 dupreq = nfs_create_subreq(req,
1304 pgbase, offset, bytes);
1306 nfs_page_group_unlock(req);
1307 if (IS_ERR(dupreq)) {
1308 desc->pg_error = PTR_ERR(dupreq);
1312 nfs_pgio_set_current_mirror(desc, midx);
1313 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1314 goto out_cleanup_subreq;
1317 nfs_pgio_set_current_mirror(desc, 0);
1318 if (!nfs_pageio_add_request_mirror(desc, req))
1324 nfs_pageio_cleanup_request(desc, dupreq);
1326 nfs_pageio_error_cleanup(desc);
1331 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1332 * nfs_pageio_descriptor
1333 * @desc: pointer to io descriptor
1334 * @mirror_idx: pointer to mirror index
1336 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1339 struct nfs_pgio_mirror *mirror;
1342 restore_idx = nfs_pgio_set_current_mirror(desc, mirror_idx);
1343 mirror = nfs_pgio_current_mirror(desc);
1346 nfs_pageio_doio(desc);
1347 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1349 if (!nfs_do_recoalesce(desc))
1352 nfs_pgio_set_current_mirror(desc, restore_idx);
1356 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1357 * @hdr - the pgio header to move request from
1358 * @desc - the pageio descriptor to add requests to
1360 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1363 * Returns 0 on success and < 0 on error.
1365 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1366 struct nfs_pgio_header *hdr)
1370 desc->pg_io_completion = hdr->io_completion;
1371 desc->pg_dreq = hdr->dreq;
1372 list_splice_init(&hdr->pages, &pages);
1373 while (!list_empty(&pages)) {
1374 struct nfs_page *req = nfs_list_entry(pages.next);
1376 if (!nfs_pageio_add_request(desc, req))
1379 nfs_pageio_complete(desc);
1380 if (!list_empty(&pages)) {
1381 int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1382 hdr->completion_ops->error_cleanup(&pages, err);
1383 nfs_set_pgio_error(hdr, err, hdr->io_start);
1388 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1391 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1392 * @desc: pointer to io descriptor
1394 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1398 for (midx = 0; midx < desc->pg_mirror_count; midx++)
1399 nfs_pageio_complete_mirror(desc, midx);
1401 if (desc->pg_error < 0)
1402 nfs_pageio_error_cleanup(desc);
1403 if (desc->pg_ops->pg_cleanup)
1404 desc->pg_ops->pg_cleanup(desc);
1405 nfs_pageio_cleanup_mirroring(desc);
1409 * nfs_pageio_cond_complete - Conditional I/O completion
1410 * @desc: pointer to io descriptor
1411 * @index: page index
1413 * It is important to ensure that processes don't try to take locks
1414 * on non-contiguous ranges of pages as that might deadlock. This
1415 * function should be called before attempting to wait on a locked
1416 * nfs_page. It will complete the I/O if the page index 'index'
1417 * is not contiguous with the existing list of pages in 'desc'.
1419 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1421 struct nfs_pgio_mirror *mirror;
1422 struct nfs_page *prev;
1425 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1426 mirror = nfs_pgio_get_mirror(desc, midx);
1427 if (!list_empty(&mirror->pg_list)) {
1428 prev = nfs_list_entry(mirror->pg_list.prev);
1429 if (index != prev->wb_index + 1) {
1430 nfs_pageio_complete(desc);
1438 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1440 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1442 nfs_pageio_complete(pgio);
1445 int __init nfs_init_nfspagecache(void)
1447 nfs_page_cachep = kmem_cache_create("nfs_page",
1448 sizeof(struct nfs_page),
1449 0, SLAB_HWCACHE_ALIGN,
1451 if (nfs_page_cachep == NULL)
1457 void nfs_destroy_nfspagecache(void)
1459 kmem_cache_destroy(nfs_page_cachep);
1462 static const struct rpc_call_ops nfs_pgio_common_ops = {
1463 .rpc_call_prepare = nfs_pgio_prepare,
1464 .rpc_call_done = nfs_pgio_result,
1465 .rpc_release = nfs_pgio_release,
1468 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1469 .pg_test = nfs_generic_pg_test,
1470 .pg_doio = nfs_generic_pg_pgios,