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>
24 #include <linux/filelock.h>
30 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
32 static struct kmem_cache *nfs_page_cachep;
33 static const struct rpc_call_ops nfs_pgio_common_ops;
35 static struct nfs_pgio_mirror *
36 nfs_pgio_get_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
38 if (desc->pg_ops->pg_get_mirror)
39 return desc->pg_ops->pg_get_mirror(desc, idx);
40 return &desc->pg_mirrors[0];
43 struct nfs_pgio_mirror *
44 nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
46 return nfs_pgio_get_mirror(desc, desc->pg_mirror_idx);
48 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
51 nfs_pgio_set_current_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
53 if (desc->pg_ops->pg_set_mirror)
54 return desc->pg_ops->pg_set_mirror(desc, idx);
55 return desc->pg_mirror_idx;
58 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
59 struct nfs_pgio_header *hdr,
60 void (*release)(struct nfs_pgio_header *hdr))
62 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
65 hdr->req = nfs_list_entry(mirror->pg_list.next);
66 hdr->inode = desc->pg_inode;
67 hdr->cred = nfs_req_openctx(hdr->req)->cred;
68 hdr->io_start = req_offset(hdr->req);
69 hdr->good_bytes = mirror->pg_count;
70 hdr->io_completion = desc->pg_io_completion;
71 hdr->dreq = desc->pg_dreq;
72 hdr->release = release;
73 hdr->completion_ops = desc->pg_completion_ops;
74 if (hdr->completion_ops->init_hdr)
75 hdr->completion_ops->init_hdr(hdr);
77 hdr->pgio_mirror_idx = desc->pg_mirror_idx;
79 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
81 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
83 unsigned int new = pos - hdr->io_start;
85 trace_nfs_pgio_error(hdr, error, pos);
86 if (hdr->good_bytes > new) {
87 hdr->good_bytes = new;
88 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
89 if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
94 static inline struct nfs_page *nfs_page_alloc(void)
97 kmem_cache_zalloc(nfs_page_cachep, nfs_io_gfp_mask());
99 INIT_LIST_HEAD(&p->wb_list);
104 nfs_page_free(struct nfs_page *p)
106 kmem_cache_free(nfs_page_cachep, p);
110 * nfs_iocounter_wait - wait for i/o to complete
111 * @l_ctx: nfs_lock_context with io_counter to use
113 * returns -ERESTARTSYS if interrupted by a fatal signal.
114 * Otherwise returns 0 once the io_count hits 0.
117 nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
119 return wait_var_event_killable(&l_ctx->io_count,
120 !atomic_read(&l_ctx->io_count));
124 * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
126 * @task: the rpc_task that should wait
127 * @l_ctx: nfs_lock_context with io_counter to check
129 * Returns true if there is outstanding I/O to wait on and the
130 * task has been put to sleep.
133 nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
135 struct inode *inode = d_inode(l_ctx->open_context->dentry);
138 if (atomic_read(&l_ctx->io_count) > 0) {
139 rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
143 if (atomic_read(&l_ctx->io_count) == 0) {
144 rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
150 EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
153 * nfs_page_lock_head_request - page lock the head of the page group
154 * @req: any member of the page group
157 nfs_page_group_lock_head(struct nfs_page *req)
159 struct nfs_page *head = req->wb_head;
161 while (!nfs_lock_request(head)) {
162 int ret = nfs_wait_on_request(head);
167 kref_get(&head->wb_kref);
172 * nfs_unroll_locks - unlock all newly locked reqs and wait on @req
173 * @head: head request of page group, must be holding head lock
174 * @req: request that couldn't lock and needs to wait on the req bit lock
176 * This is a helper function for nfs_lock_and_join_requests
177 * returns 0 on success, < 0 on error.
180 nfs_unroll_locks(struct nfs_page *head, struct nfs_page *req)
182 struct nfs_page *tmp;
184 /* relinquish all the locks successfully grabbed this run */
185 for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) {
186 if (!kref_read(&tmp->wb_kref))
188 nfs_unlock_and_release_request(tmp);
193 * nfs_page_group_lock_subreq - try to lock a subrequest
194 * @head: head request of page group
195 * @subreq: request to lock
197 * This is a helper function for nfs_lock_and_join_requests which
198 * must be called with the head request and page group both locked.
199 * On error, it returns with the page group unlocked.
202 nfs_page_group_lock_subreq(struct nfs_page *head, struct nfs_page *subreq)
206 if (!kref_get_unless_zero(&subreq->wb_kref))
208 while (!nfs_lock_request(subreq)) {
209 nfs_page_group_unlock(head);
210 ret = nfs_wait_on_request(subreq);
212 ret = nfs_page_group_lock(head);
214 nfs_unroll_locks(head, subreq);
215 nfs_release_request(subreq);
223 * nfs_page_group_lock_subrequests - try to lock the subrequests
224 * @head: head request of page group
226 * This is a helper function for nfs_lock_and_join_requests which
227 * must be called with the head request locked.
229 int nfs_page_group_lock_subrequests(struct nfs_page *head)
231 struct nfs_page *subreq;
234 ret = nfs_page_group_lock(head);
237 /* lock each request in the page group */
238 for (subreq = head->wb_this_page; subreq != head;
239 subreq = subreq->wb_this_page) {
240 ret = nfs_page_group_lock_subreq(head, subreq);
244 nfs_page_group_unlock(head);
249 * nfs_page_set_headlock - set the request PG_HEADLOCK
250 * @req: request that is to be locked
252 * this lock must be held when modifying req->wb_head
254 * return 0 on success, < 0 on error
257 nfs_page_set_headlock(struct nfs_page *req)
259 if (!test_and_set_bit(PG_HEADLOCK, &req->wb_flags))
262 set_bit(PG_CONTENDED1, &req->wb_flags);
263 smp_mb__after_atomic();
264 return wait_on_bit_lock(&req->wb_flags, PG_HEADLOCK,
265 TASK_UNINTERRUPTIBLE);
269 * nfs_page_clear_headlock - clear the request PG_HEADLOCK
270 * @req: request that is to be locked
273 nfs_page_clear_headlock(struct nfs_page *req)
275 clear_bit_unlock(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 clear_bit_unlock(PG_BUSY, &req->wb_flags);
529 smp_mb__after_atomic();
530 if (!test_bit(PG_CONTENDED2, &req->wb_flags))
532 wake_up_bit(&req->wb_flags, PG_BUSY);
536 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
537 * @req: pointer to request
539 void nfs_unlock_and_release_request(struct nfs_page *req)
541 nfs_unlock_request(req);
542 nfs_release_request(req);
546 * nfs_clear_request - Free up all resources allocated to the request
549 * Release page and open context resources associated with a read/write
550 * request after it has completed.
552 static void nfs_clear_request(struct nfs_page *req)
554 struct page *page = req->wb_page;
555 struct nfs_lock_context *l_ctx = req->wb_lock_context;
556 struct nfs_open_context *ctx;
563 if (atomic_dec_and_test(&l_ctx->io_count)) {
564 wake_up_var(&l_ctx->io_count);
565 ctx = l_ctx->open_context;
566 if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
567 rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
569 nfs_put_lock_context(l_ctx);
570 req->wb_lock_context = NULL;
575 * nfs_free_request - Release the count on an NFS read/write request
576 * @req: request to release
578 * Note: Should never be called with the spinlock held!
580 void nfs_free_request(struct nfs_page *req)
582 WARN_ON_ONCE(req->wb_this_page != req);
584 /* extra debug: make sure no sync bits are still set */
585 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
586 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
587 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
588 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
589 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
591 /* Release struct file and open context */
592 nfs_clear_request(req);
596 void nfs_release_request(struct nfs_page *req)
598 kref_put(&req->wb_kref, nfs_page_group_destroy);
600 EXPORT_SYMBOL_GPL(nfs_release_request);
603 * nfs_wait_on_request - Wait for a request to complete.
604 * @req: request to wait upon.
606 * Interruptible by fatal signals only.
607 * The user is responsible for holding a count on the request.
610 nfs_wait_on_request(struct nfs_page *req)
612 if (!test_bit(PG_BUSY, &req->wb_flags))
614 set_bit(PG_CONTENDED2, &req->wb_flags);
615 smp_mb__after_atomic();
616 return wait_on_bit_io(&req->wb_flags, PG_BUSY,
617 TASK_UNINTERRUPTIBLE);
619 EXPORT_SYMBOL_GPL(nfs_wait_on_request);
622 * nfs_generic_pg_test - determine if requests can be coalesced
623 * @desc: pointer to descriptor
624 * @prev: previous request in desc, or NULL
627 * Returns zero if @req cannot be coalesced into @desc, otherwise it returns
628 * the size of the request.
630 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
631 struct nfs_page *prev, struct nfs_page *req)
633 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
636 if (mirror->pg_count > mirror->pg_bsize) {
637 /* should never happen */
643 * Limit the request size so that we can still allocate a page array
644 * for it without upsetting the slab allocator.
646 if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
647 sizeof(struct page *) > PAGE_SIZE)
650 return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
652 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
654 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
656 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
659 INIT_LIST_HEAD(&hdr->pages);
664 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
667 * nfs_pgio_data_destroy - make @hdr suitable for reuse
669 * Frees memory and releases refs from nfs_generic_pgio, so that it may
672 * @hdr: A header that has had nfs_generic_pgio called
674 static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
676 if (hdr->args.context)
677 put_nfs_open_context(hdr->args.context);
678 if (hdr->page_array.pagevec != hdr->page_array.page_array)
679 kfree(hdr->page_array.pagevec);
683 * nfs_pgio_header_free - Free a read or write header
684 * @hdr: The header to free
686 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
688 nfs_pgio_data_destroy(hdr);
689 hdr->rw_ops->rw_free_header(hdr);
691 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
694 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
695 * @hdr: The pageio hdr
696 * @count: Number of bytes to read
697 * @how: How to commit data (writes only)
698 * @cinfo: Commit information for the call (writes only)
700 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
702 int how, struct nfs_commit_info *cinfo)
704 struct nfs_page *req = hdr->req;
706 /* Set up the RPC argument and reply structs
707 * NB: take care not to mess about with hdr->commit et al. */
709 hdr->args.fh = NFS_FH(hdr->inode);
710 hdr->args.offset = req_offset(req);
711 /* pnfs_set_layoutcommit needs this */
712 hdr->mds_offset = hdr->args.offset;
713 hdr->args.pgbase = req->wb_pgbase;
714 hdr->args.pages = hdr->page_array.pagevec;
715 hdr->args.count = count;
716 hdr->args.context = get_nfs_open_context(nfs_req_openctx(req));
717 hdr->args.lock_context = req->wb_lock_context;
718 hdr->args.stable = NFS_UNSTABLE;
719 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
722 case FLUSH_COND_STABLE:
723 if (nfs_reqs_to_commit(cinfo))
727 hdr->args.stable = NFS_FILE_SYNC;
730 hdr->res.fattr = &hdr->fattr;
733 hdr->res.verf = &hdr->verf;
734 nfs_fattr_init(&hdr->fattr);
738 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
739 * @task: The current task
740 * @calldata: pageio header to prepare
742 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
744 struct nfs_pgio_header *hdr = calldata;
746 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
751 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
752 const struct cred *cred, const struct nfs_rpc_ops *rpc_ops,
753 const struct rpc_call_ops *call_ops, int how, int flags)
755 struct rpc_task *task;
756 struct rpc_message msg = {
757 .rpc_argp = &hdr->args,
758 .rpc_resp = &hdr->res,
761 struct rpc_task_setup task_setup_data = {
765 .callback_ops = call_ops,
766 .callback_data = hdr,
767 .workqueue = nfsiod_workqueue,
768 .flags = RPC_TASK_ASYNC | flags,
771 if (nfs_server_capable(hdr->inode, NFS_CAP_MOVEABLE))
772 task_setup_data.flags |= RPC_TASK_MOVEABLE;
774 hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
776 dprintk("NFS: initiated pgio call "
777 "(req %s/%llu, %u bytes @ offset %llu)\n",
778 hdr->inode->i_sb->s_id,
779 (unsigned long long)NFS_FILEID(hdr->inode),
781 (unsigned long long)hdr->args.offset);
783 task = rpc_run_task(&task_setup_data);
785 return PTR_ERR(task);
789 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
792 * nfs_pgio_error - Clean up from a pageio error
793 * @hdr: pageio header
795 static void nfs_pgio_error(struct nfs_pgio_header *hdr)
797 set_bit(NFS_IOHDR_REDO, &hdr->flags);
798 hdr->completion_ops->completion(hdr);
802 * nfs_pgio_release - Release pageio data
803 * @calldata: The pageio header to release
805 static void nfs_pgio_release(void *calldata)
807 struct nfs_pgio_header *hdr = calldata;
808 hdr->completion_ops->completion(hdr);
811 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
814 INIT_LIST_HEAD(&mirror->pg_list);
815 mirror->pg_bytes_written = 0;
816 mirror->pg_count = 0;
817 mirror->pg_bsize = bsize;
819 mirror->pg_recoalesce = 0;
823 * nfs_pageio_init - initialise a page io descriptor
824 * @desc: pointer to descriptor
825 * @inode: pointer to inode
826 * @pg_ops: pointer to pageio operations
827 * @compl_ops: pointer to pageio completion operations
828 * @rw_ops: pointer to nfs read/write operations
829 * @bsize: io block size
830 * @io_flags: extra parameters for the io function
832 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
834 const struct nfs_pageio_ops *pg_ops,
835 const struct nfs_pgio_completion_ops *compl_ops,
836 const struct nfs_rw_ops *rw_ops,
841 desc->pg_inode = inode;
842 desc->pg_ops = pg_ops;
843 desc->pg_completion_ops = compl_ops;
844 desc->pg_rw_ops = rw_ops;
845 desc->pg_ioflags = io_flags;
847 desc->pg_lseg = NULL;
848 desc->pg_io_completion = NULL;
849 desc->pg_dreq = NULL;
850 desc->pg_bsize = bsize;
852 desc->pg_mirror_count = 1;
853 desc->pg_mirror_idx = 0;
855 desc->pg_mirrors_dynamic = NULL;
856 desc->pg_mirrors = desc->pg_mirrors_static;
857 nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
858 desc->pg_maxretrans = 0;
862 * nfs_pgio_result - Basic pageio error handling
863 * @task: The task that ran
864 * @calldata: Pageio header to check
866 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
868 struct nfs_pgio_header *hdr = calldata;
869 struct inode *inode = hdr->inode;
871 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
873 if (task->tk_status < 0)
874 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
876 hdr->rw_ops->rw_result(task, hdr);
880 * Create an RPC task for the given read or write request and kick it.
881 * The page must have been locked by the caller.
883 * It may happen that the page we're passed is not marked dirty.
884 * This is the case if nfs_updatepage detects a conflicting request
885 * that has been written but not committed.
887 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
888 struct nfs_pgio_header *hdr)
890 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
892 struct nfs_page *req;
895 struct list_head *head = &mirror->pg_list;
896 struct nfs_commit_info cinfo;
897 struct nfs_page_array *pg_array = &hdr->page_array;
898 unsigned int pagecount, pageused;
899 gfp_t gfp_flags = nfs_io_gfp_mask();
901 pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
902 pg_array->npages = pagecount;
904 if (pagecount <= ARRAY_SIZE(pg_array->page_array))
905 pg_array->pagevec = pg_array->page_array;
907 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
908 if (!pg_array->pagevec) {
909 pg_array->npages = 0;
911 desc->pg_error = -ENOMEM;
912 return desc->pg_error;
916 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
917 pages = hdr->page_array.pagevec;
920 while (!list_empty(head)) {
921 req = nfs_list_entry(head->next);
922 nfs_list_move_request(req, &hdr->pages);
924 if (!last_page || last_page != req->wb_page) {
926 if (pageused > pagecount)
928 *pages++ = last_page = req->wb_page;
931 if (WARN_ON_ONCE(pageused != pagecount)) {
933 desc->pg_error = -EINVAL;
934 return desc->pg_error;
937 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
938 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
939 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
941 /* Set up the argument struct */
942 nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo);
943 desc->pg_rpc_callops = &nfs_pgio_common_ops;
946 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
948 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
950 struct nfs_pgio_header *hdr;
952 unsigned short task_flags = 0;
954 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
956 desc->pg_error = -ENOMEM;
957 return desc->pg_error;
959 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
960 ret = nfs_generic_pgio(desc, hdr);
962 if (NFS_SERVER(hdr->inode)->nfs_client->cl_minorversion)
963 task_flags = RPC_TASK_MOVEABLE;
964 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
967 NFS_PROTO(hdr->inode),
968 desc->pg_rpc_callops,
970 RPC_TASK_CRED_NOREF | task_flags);
975 static struct nfs_pgio_mirror *
976 nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
977 unsigned int mirror_count)
979 struct nfs_pgio_mirror *ret;
982 kfree(desc->pg_mirrors_dynamic);
983 desc->pg_mirrors_dynamic = NULL;
984 if (mirror_count == 1)
985 return desc->pg_mirrors_static;
986 ret = kmalloc_array(mirror_count, sizeof(*ret), nfs_io_gfp_mask());
988 for (i = 0; i < mirror_count; i++)
989 nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
990 desc->pg_mirrors_dynamic = ret;
996 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
997 * by calling the pg_get_mirror_count op
999 static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
1000 struct nfs_page *req)
1002 unsigned int mirror_count = 1;
1004 if (pgio->pg_ops->pg_get_mirror_count)
1005 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
1006 if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
1009 if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
1010 pgio->pg_error = -EINVAL;
1014 pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
1015 if (pgio->pg_mirrors == NULL) {
1016 pgio->pg_error = -ENOMEM;
1017 pgio->pg_mirrors = pgio->pg_mirrors_static;
1020 pgio->pg_mirror_count = mirror_count;
1023 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
1025 pgio->pg_mirror_count = 1;
1026 pgio->pg_mirror_idx = 0;
1027 pgio->pg_mirrors = pgio->pg_mirrors_static;
1028 kfree(pgio->pg_mirrors_dynamic);
1029 pgio->pg_mirrors_dynamic = NULL;
1032 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
1033 const struct nfs_lock_context *l2)
1035 return l1->lockowner == l2->lockowner;
1039 * nfs_coalesce_size - test two requests for compatibility
1040 * @prev: pointer to nfs_page
1041 * @req: pointer to nfs_page
1042 * @pgio: pointer to nfs_pagio_descriptor
1044 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
1045 * page data area they describe is contiguous, and that their RPC
1046 * credentials, NFSv4 open state, and lockowners are the same.
1048 * Returns size of the request that can be coalesced
1050 static unsigned int nfs_coalesce_size(struct nfs_page *prev,
1051 struct nfs_page *req,
1052 struct nfs_pageio_descriptor *pgio)
1054 struct file_lock_context *flctx;
1057 if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
1059 flctx = locks_inode_context(d_inode(nfs_req_openctx(req)->dentry));
1060 if (flctx != NULL &&
1061 !(list_empty_careful(&flctx->flc_posix) &&
1062 list_empty_careful(&flctx->flc_flock)) &&
1063 !nfs_match_lock_context(req->wb_lock_context,
1064 prev->wb_lock_context))
1066 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
1068 if (req->wb_page == prev->wb_page) {
1069 if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
1072 if (req->wb_pgbase != 0 ||
1073 prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE)
1077 return pgio->pg_ops->pg_test(pgio, prev, req);
1081 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
1082 * @desc: destination io descriptor
1085 * If the request 'req' was successfully coalesced into the existing list
1086 * of pages 'desc', it returns the size of req.
1089 nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
1090 struct nfs_page *req)
1092 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1093 struct nfs_page *prev = NULL;
1096 if (list_empty(&mirror->pg_list)) {
1097 if (desc->pg_ops->pg_init)
1098 desc->pg_ops->pg_init(desc, req);
1099 if (desc->pg_error < 0)
1101 mirror->pg_base = req->wb_pgbase;
1102 mirror->pg_count = 0;
1103 mirror->pg_recoalesce = 0;
1105 prev = nfs_list_entry(mirror->pg_list.prev);
1107 if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
1108 if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
1109 desc->pg_error = -ETIMEDOUT;
1111 desc->pg_error = -EIO;
1115 size = nfs_coalesce_size(prev, req, desc);
1116 if (size < req->wb_bytes)
1118 nfs_list_move_request(req, &mirror->pg_list);
1119 mirror->pg_count += req->wb_bytes;
1120 return req->wb_bytes;
1124 * Helper for nfs_pageio_add_request and nfs_pageio_complete
1126 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1128 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1130 if (!list_empty(&mirror->pg_list)) {
1131 int error = desc->pg_ops->pg_doio(desc);
1133 desc->pg_error = error;
1134 if (list_empty(&mirror->pg_list))
1135 mirror->pg_bytes_written += mirror->pg_count;
1140 nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1141 struct nfs_page *req)
1145 nfs_list_move_request(req, &head);
1146 desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1150 * __nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1151 * @desc: destination io descriptor
1154 * This may split a request into subrequests which are all part of the
1155 * same page group. If so, it will submit @req as the last one, to ensure
1156 * the pointer to @req is still valid in case of failure.
1158 * Returns true if the request 'req' was successfully coalesced into the
1159 * existing list of pages 'desc'.
1161 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1162 struct nfs_page *req)
1164 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1165 struct nfs_page *subreq;
1166 unsigned int size, subreq_size;
1168 nfs_page_group_lock(req);
1171 subreq_size = subreq->wb_bytes;
1173 size = nfs_pageio_do_add_request(desc, subreq);
1174 if (size == subreq_size) {
1175 /* We successfully submitted a request */
1178 req->wb_pgbase += size;
1179 req->wb_bytes -= size;
1180 req->wb_offset += size;
1181 subreq_size = req->wb_bytes;
1185 if (WARN_ON_ONCE(subreq != req)) {
1186 nfs_page_group_unlock(req);
1187 nfs_pageio_cleanup_request(desc, subreq);
1189 subreq_size = req->wb_bytes;
1190 nfs_page_group_lock(req);
1193 /* Can't coalesce any more, so do I/O */
1194 nfs_page_group_unlock(req);
1195 desc->pg_moreio = 1;
1196 nfs_pageio_doio(desc);
1197 if (desc->pg_error < 0 || mirror->pg_recoalesce)
1199 /* retry add_request for this subreq */
1200 nfs_page_group_lock(req);
1203 subreq = nfs_create_subreq(req, req->wb_pgbase,
1204 req->wb_offset, size);
1210 nfs_page_group_unlock(req);
1213 desc->pg_error = PTR_ERR(subreq);
1214 nfs_page_group_unlock(req);
1218 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1220 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1224 list_splice_init(&mirror->pg_list, &head);
1225 mirror->pg_recoalesce = 0;
1227 while (!list_empty(&head)) {
1228 struct nfs_page *req;
1230 req = list_first_entry(&head, struct nfs_page, wb_list);
1231 if (__nfs_pageio_add_request(desc, req))
1233 if (desc->pg_error < 0) {
1234 list_splice_tail(&head, &mirror->pg_list);
1235 mirror->pg_recoalesce = 1;
1240 } while (mirror->pg_recoalesce);
1244 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1245 struct nfs_page *req)
1250 ret = __nfs_pageio_add_request(desc, req);
1253 if (desc->pg_error < 0)
1255 ret = nfs_do_recoalesce(desc);
1261 static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1264 struct nfs_pgio_mirror *mirror;
1266 if (!desc->pg_error)
1269 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1270 mirror = nfs_pgio_get_mirror(desc, midx);
1271 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1276 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1277 struct nfs_page *req)
1280 unsigned int pgbase, offset, bytes;
1281 struct nfs_page *dupreq;
1283 pgbase = req->wb_pgbase;
1284 offset = req->wb_offset;
1285 bytes = req->wb_bytes;
1287 nfs_pageio_setup_mirroring(desc, req);
1288 if (desc->pg_error < 0)
1291 /* Create the mirror instances first, and fire them off */
1292 for (midx = 1; midx < desc->pg_mirror_count; midx++) {
1293 nfs_page_group_lock(req);
1295 dupreq = nfs_create_subreq(req,
1296 pgbase, offset, bytes);
1298 nfs_page_group_unlock(req);
1299 if (IS_ERR(dupreq)) {
1300 desc->pg_error = PTR_ERR(dupreq);
1304 nfs_pgio_set_current_mirror(desc, midx);
1305 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1306 goto out_cleanup_subreq;
1309 nfs_pgio_set_current_mirror(desc, 0);
1310 if (!nfs_pageio_add_request_mirror(desc, req))
1316 nfs_pageio_cleanup_request(desc, dupreq);
1318 nfs_pageio_error_cleanup(desc);
1323 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1324 * nfs_pageio_descriptor
1325 * @desc: pointer to io descriptor
1326 * @mirror_idx: pointer to mirror index
1328 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1331 struct nfs_pgio_mirror *mirror;
1334 restore_idx = nfs_pgio_set_current_mirror(desc, mirror_idx);
1335 mirror = nfs_pgio_current_mirror(desc);
1338 nfs_pageio_doio(desc);
1339 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1341 if (!nfs_do_recoalesce(desc))
1344 nfs_pgio_set_current_mirror(desc, restore_idx);
1348 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1349 * @hdr - the pgio header to move request from
1350 * @desc - the pageio descriptor to add requests to
1352 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1355 * Returns 0 on success and < 0 on error.
1357 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1358 struct nfs_pgio_header *hdr)
1362 desc->pg_io_completion = hdr->io_completion;
1363 desc->pg_dreq = hdr->dreq;
1364 list_splice_init(&hdr->pages, &pages);
1365 while (!list_empty(&pages)) {
1366 struct nfs_page *req = nfs_list_entry(pages.next);
1368 if (!nfs_pageio_add_request(desc, req))
1371 nfs_pageio_complete(desc);
1372 if (!list_empty(&pages)) {
1373 int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1374 hdr->completion_ops->error_cleanup(&pages, err);
1375 nfs_set_pgio_error(hdr, err, hdr->io_start);
1380 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1383 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1384 * @desc: pointer to io descriptor
1386 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1390 for (midx = 0; midx < desc->pg_mirror_count; midx++)
1391 nfs_pageio_complete_mirror(desc, midx);
1393 if (desc->pg_error < 0)
1394 nfs_pageio_error_cleanup(desc);
1395 if (desc->pg_ops->pg_cleanup)
1396 desc->pg_ops->pg_cleanup(desc);
1397 nfs_pageio_cleanup_mirroring(desc);
1401 * nfs_pageio_cond_complete - Conditional I/O completion
1402 * @desc: pointer to io descriptor
1403 * @index: page index
1405 * It is important to ensure that processes don't try to take locks
1406 * on non-contiguous ranges of pages as that might deadlock. This
1407 * function should be called before attempting to wait on a locked
1408 * nfs_page. It will complete the I/O if the page index 'index'
1409 * is not contiguous with the existing list of pages in 'desc'.
1411 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1413 struct nfs_pgio_mirror *mirror;
1414 struct nfs_page *prev;
1417 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1418 mirror = nfs_pgio_get_mirror(desc, midx);
1419 if (!list_empty(&mirror->pg_list)) {
1420 prev = nfs_list_entry(mirror->pg_list.prev);
1421 if (index != prev->wb_index + 1) {
1422 nfs_pageio_complete(desc);
1430 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1432 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1434 nfs_pageio_complete(pgio);
1437 int __init nfs_init_nfspagecache(void)
1439 nfs_page_cachep = kmem_cache_create("nfs_page",
1440 sizeof(struct nfs_page),
1441 0, SLAB_HWCACHE_ALIGN,
1443 if (nfs_page_cachep == NULL)
1449 void nfs_destroy_nfspagecache(void)
1451 kmem_cache_destroy(nfs_page_cachep);
1454 static const struct rpc_call_ops nfs_pgio_common_ops = {
1455 .rpc_call_prepare = nfs_pgio_prepare,
1456 .rpc_call_done = nfs_pgio_result,
1457 .rpc_release = nfs_pgio_release,
1460 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1461 .pg_test = nfs_generic_pg_test,
1462 .pg_doio = nfs_generic_pg_pgios,