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 static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
34 p->npages = pagecount;
35 if (pagecount <= ARRAY_SIZE(p->page_array))
36 p->pagevec = p->page_array;
38 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
42 return p->pagevec != NULL;
45 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
46 struct nfs_pgio_header *hdr,
47 void (*release)(struct nfs_pgio_header *hdr))
49 hdr->req = nfs_list_entry(desc->pg_list.next);
50 hdr->inode = desc->pg_inode;
51 hdr->cred = hdr->req->wb_context->cred;
52 hdr->io_start = req_offset(hdr->req);
53 hdr->good_bytes = desc->pg_count;
54 hdr->dreq = desc->pg_dreq;
55 hdr->layout_private = desc->pg_layout_private;
56 hdr->release = release;
57 hdr->completion_ops = desc->pg_completion_ops;
58 if (hdr->completion_ops->init_hdr)
59 hdr->completion_ops->init_hdr(hdr);
61 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
63 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
65 spin_lock(&hdr->lock);
66 if (pos < hdr->io_start + hdr->good_bytes) {
67 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
68 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
69 hdr->good_bytes = pos - hdr->io_start;
72 spin_unlock(&hdr->lock);
75 static inline struct nfs_page *
78 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
80 INIT_LIST_HEAD(&p->wb_list);
85 nfs_page_free(struct nfs_page *p)
87 kmem_cache_free(nfs_page_cachep, p);
91 nfs_iocounter_inc(struct nfs_io_counter *c)
93 atomic_inc(&c->io_count);
97 nfs_iocounter_dec(struct nfs_io_counter *c)
99 if (atomic_dec_and_test(&c->io_count)) {
100 clear_bit(NFS_IO_INPROGRESS, &c->flags);
101 smp_mb__after_atomic();
102 wake_up_bit(&c->flags, NFS_IO_INPROGRESS);
107 __nfs_iocounter_wait(struct nfs_io_counter *c)
109 wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS);
110 DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS);
114 prepare_to_wait(wq, &q.wait, TASK_KILLABLE);
115 set_bit(NFS_IO_INPROGRESS, &c->flags);
116 if (atomic_read(&c->io_count) == 0)
118 ret = nfs_wait_bit_killable(&q.key);
119 } while (atomic_read(&c->io_count) != 0 && !ret);
120 finish_wait(wq, &q.wait);
125 * nfs_iocounter_wait - wait for i/o to complete
126 * @c: nfs_io_counter to use
128 * returns -ERESTARTSYS if interrupted by a fatal signal.
129 * Otherwise returns 0 once the io_count hits 0.
132 nfs_iocounter_wait(struct nfs_io_counter *c)
134 if (atomic_read(&c->io_count) == 0)
136 return __nfs_iocounter_wait(c);
140 * nfs_page_group_lock - lock the head of the page group
141 * @req - request in group that is to be locked
142 * @nonblock - if true don't block waiting for lock
144 * this lock must be held if modifying the page group list
146 * return 0 on success, < 0 on error: -EDELAY if nonblocking or the
147 * result from wait_on_bit_lock
149 * NOTE: calling with nonblock=false should always have set the
150 * lock bit (see fs/buffer.c and other uses of wait_on_bit_lock
151 * with TASK_UNINTERRUPTIBLE), so there is no need to check the result.
154 nfs_page_group_lock(struct nfs_page *req, bool nonblock)
156 struct nfs_page *head = req->wb_head;
158 WARN_ON_ONCE(head != head->wb_head);
160 if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags))
164 return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
165 TASK_UNINTERRUPTIBLE);
171 * nfs_page_group_lock_wait - wait for the lock to clear, but don't grab it
172 * @req - a request in the group
174 * This is a blocking call to wait for the group lock to be cleared.
177 nfs_page_group_lock_wait(struct nfs_page *req)
179 struct nfs_page *head = req->wb_head;
181 WARN_ON_ONCE(head != head->wb_head);
183 wait_on_bit(&head->wb_flags, PG_HEADLOCK,
184 TASK_UNINTERRUPTIBLE);
188 * nfs_page_group_unlock - unlock the head of the page group
189 * @req - request in group that is to be unlocked
192 nfs_page_group_unlock(struct nfs_page *req)
194 struct nfs_page *head = req->wb_head;
196 WARN_ON_ONCE(head != head->wb_head);
198 smp_mb__before_atomic();
199 clear_bit(PG_HEADLOCK, &head->wb_flags);
200 smp_mb__after_atomic();
201 wake_up_bit(&head->wb_flags, PG_HEADLOCK);
205 * nfs_page_group_sync_on_bit_locked
207 * must be called with page group lock held
210 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
212 struct nfs_page *head = req->wb_head;
213 struct nfs_page *tmp;
215 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
216 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
218 tmp = req->wb_this_page;
220 if (!test_bit(bit, &tmp->wb_flags))
222 tmp = tmp->wb_this_page;
225 /* true! reset all bits */
228 clear_bit(bit, &tmp->wb_flags);
229 tmp = tmp->wb_this_page;
230 } while (tmp != req);
236 * nfs_page_group_sync_on_bit - set bit on current request, but only
237 * return true if the bit is set for all requests in page group
238 * @req - request in page group
239 * @bit - PG_* bit that is used to sync page group
241 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
245 nfs_page_group_lock(req, false);
246 ret = nfs_page_group_sync_on_bit_locked(req, bit);
247 nfs_page_group_unlock(req);
253 * nfs_page_group_init - Initialize the page group linkage for @req
254 * @req - a new nfs request
255 * @prev - the previous request in page group, or NULL if @req is the first
256 * or only request in the group (the head).
259 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
261 WARN_ON_ONCE(prev == req);
266 req->wb_this_page = req;
269 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
270 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
271 req->wb_head = prev->wb_head;
272 req->wb_this_page = prev->wb_this_page;
273 prev->wb_this_page = req;
275 /* All subrequests take a ref on the head request until
276 * nfs_page_group_destroy is called */
277 kref_get(&req->wb_head->wb_kref);
279 /* grab extra ref if head request has extra ref from
280 * the write/commit path to handle handoff between write
281 * and commit lists */
282 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
283 set_bit(PG_INODE_REF, &req->wb_flags);
284 kref_get(&req->wb_kref);
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 nfs_iocounter_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. */
360 req->wb_index = page_file_index(page);
361 page_cache_get(page);
362 req->wb_offset = offset;
363 req->wb_pgbase = offset;
364 req->wb_bytes = count;
365 req->wb_context = get_nfs_open_context(ctx);
366 kref_init(&req->wb_kref);
367 nfs_page_group_init(req, last);
372 * nfs_unlock_request - Unlock request and wake up sleepers.
375 void nfs_unlock_request(struct nfs_page *req)
377 if (!NFS_WBACK_BUSY(req)) {
378 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
381 smp_mb__before_atomic();
382 clear_bit(PG_BUSY, &req->wb_flags);
383 smp_mb__after_atomic();
384 wake_up_bit(&req->wb_flags, PG_BUSY);
388 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
391 void nfs_unlock_and_release_request(struct nfs_page *req)
393 nfs_unlock_request(req);
394 nfs_release_request(req);
398 * nfs_clear_request - Free up all resources allocated to the request
401 * Release page and open context resources associated with a read/write
402 * request after it has completed.
404 static void nfs_clear_request(struct nfs_page *req)
406 struct page *page = req->wb_page;
407 struct nfs_open_context *ctx = req->wb_context;
408 struct nfs_lock_context *l_ctx = req->wb_lock_context;
411 page_cache_release(page);
415 nfs_iocounter_dec(&l_ctx->io_count);
416 nfs_put_lock_context(l_ctx);
417 req->wb_lock_context = NULL;
420 put_nfs_open_context(ctx);
421 req->wb_context = NULL;
426 * nfs_release_request - Release the count on an NFS read/write request
427 * @req: request to release
429 * Note: Should never be called with the spinlock held!
431 void nfs_free_request(struct nfs_page *req)
433 WARN_ON_ONCE(req->wb_this_page != req);
435 /* extra debug: make sure no sync bits are still set */
436 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
437 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
438 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
439 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
440 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
442 /* Release struct file and open context */
443 nfs_clear_request(req);
447 void nfs_release_request(struct nfs_page *req)
449 kref_put(&req->wb_kref, nfs_page_group_destroy);
453 * nfs_wait_on_request - Wait for a request to complete.
454 * @req: request to wait upon.
456 * Interruptible by fatal signals only.
457 * The user is responsible for holding a count on the request.
460 nfs_wait_on_request(struct nfs_page *req)
462 return wait_on_bit_io(&req->wb_flags, PG_BUSY,
463 TASK_UNINTERRUPTIBLE);
467 * nfs_generic_pg_test - determine if requests can be coalesced
468 * @desc: pointer to descriptor
469 * @prev: previous request in desc, or NULL
472 * Returns zero if @req can be coalesced into @desc, otherwise it returns
473 * the size of the request.
475 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
476 struct nfs_page *prev, struct nfs_page *req)
478 if (desc->pg_count > desc->pg_bsize) {
479 /* should never happen */
484 return min(desc->pg_bsize - desc->pg_count, (size_t)req->wb_bytes);
486 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
488 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
490 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
493 INIT_LIST_HEAD(&hdr->pages);
494 spin_lock_init(&hdr->lock);
499 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
502 * nfs_pgio_header_free - Free a read or write header
503 * @hdr: The header to free
505 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
507 hdr->rw_ops->rw_free_header(hdr);
509 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
512 * nfs_pgio_data_destroy - make @hdr suitable for reuse
514 * Frees memory and releases refs from nfs_generic_pgio, so that it may
517 * @hdr: A header that has had nfs_generic_pgio called
519 void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
521 put_nfs_open_context(hdr->args.context);
522 if (hdr->page_array.pagevec != hdr->page_array.page_array)
523 kfree(hdr->page_array.pagevec);
525 EXPORT_SYMBOL_GPL(nfs_pgio_data_destroy);
528 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
529 * @hdr: The pageio hdr
530 * @count: Number of bytes to read
531 * @offset: Initial offset
532 * @how: How to commit data (writes only)
533 * @cinfo: Commit information for the call (writes only)
535 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
536 unsigned int count, unsigned int offset,
537 int how, struct nfs_commit_info *cinfo)
539 struct nfs_page *req = hdr->req;
541 /* Set up the RPC argument and reply structs
542 * NB: take care not to mess about with hdr->commit et al. */
544 hdr->args.fh = NFS_FH(hdr->inode);
545 hdr->args.offset = req_offset(req) + offset;
546 /* pnfs_set_layoutcommit needs this */
547 hdr->mds_offset = hdr->args.offset;
548 hdr->args.pgbase = req->wb_pgbase + offset;
549 hdr->args.pages = hdr->page_array.pagevec;
550 hdr->args.count = count;
551 hdr->args.context = get_nfs_open_context(req->wb_context);
552 hdr->args.lock_context = req->wb_lock_context;
553 hdr->args.stable = NFS_UNSTABLE;
554 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
557 case FLUSH_COND_STABLE:
558 if (nfs_reqs_to_commit(cinfo))
561 hdr->args.stable = NFS_FILE_SYNC;
564 hdr->res.fattr = &hdr->fattr;
565 hdr->res.count = count;
567 hdr->res.verf = &hdr->verf;
568 nfs_fattr_init(&hdr->fattr);
572 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
573 * @task: The current task
574 * @calldata: pageio header to prepare
576 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
578 struct nfs_pgio_header *hdr = calldata;
580 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
585 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
586 const struct rpc_call_ops *call_ops, int how, int flags)
588 struct rpc_task *task;
589 struct rpc_message msg = {
590 .rpc_argp = &hdr->args,
591 .rpc_resp = &hdr->res,
592 .rpc_cred = hdr->cred,
594 struct rpc_task_setup task_setup_data = {
598 .callback_ops = call_ops,
599 .callback_data = hdr,
600 .workqueue = nfsiod_workqueue,
601 .flags = RPC_TASK_ASYNC | flags,
605 hdr->rw_ops->rw_initiate(hdr, &msg, &task_setup_data, how);
607 dprintk("NFS: %5u initiated pgio call "
608 "(req %s/%llu, %u bytes @ offset %llu)\n",
610 hdr->inode->i_sb->s_id,
611 (unsigned long long)NFS_FILEID(hdr->inode),
613 (unsigned long long)hdr->args.offset);
615 task = rpc_run_task(&task_setup_data);
620 if (how & FLUSH_SYNC) {
621 ret = rpc_wait_for_completion_task(task);
623 ret = task->tk_status;
629 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
632 * nfs_pgio_error - Clean up from a pageio error
633 * @desc: IO descriptor
634 * @hdr: pageio header
636 static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
637 struct nfs_pgio_header *hdr)
639 set_bit(NFS_IOHDR_REDO, &hdr->flags);
640 nfs_pgio_data_destroy(hdr);
641 hdr->completion_ops->completion(hdr);
642 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
647 * nfs_pgio_release - Release pageio data
648 * @calldata: The pageio header to release
650 static void nfs_pgio_release(void *calldata)
652 struct nfs_pgio_header *hdr = calldata;
653 if (hdr->rw_ops->rw_release)
654 hdr->rw_ops->rw_release(hdr);
655 nfs_pgio_data_destroy(hdr);
656 hdr->completion_ops->completion(hdr);
660 * nfs_pageio_init - initialise a page io descriptor
661 * @desc: pointer to descriptor
662 * @inode: pointer to inode
663 * @doio: pointer to io function
664 * @bsize: io block size
665 * @io_flags: extra parameters for the io function
667 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
669 const struct nfs_pageio_ops *pg_ops,
670 const struct nfs_pgio_completion_ops *compl_ops,
671 const struct nfs_rw_ops *rw_ops,
675 INIT_LIST_HEAD(&desc->pg_list);
676 desc->pg_bytes_written = 0;
678 desc->pg_bsize = bsize;
681 desc->pg_recoalesce = 0;
682 desc->pg_inode = inode;
683 desc->pg_ops = pg_ops;
684 desc->pg_completion_ops = compl_ops;
685 desc->pg_rw_ops = rw_ops;
686 desc->pg_ioflags = io_flags;
688 desc->pg_lseg = NULL;
689 desc->pg_dreq = NULL;
690 desc->pg_layout_private = NULL;
692 EXPORT_SYMBOL_GPL(nfs_pageio_init);
695 * nfs_pgio_result - Basic pageio error handling
696 * @task: The task that ran
697 * @calldata: Pageio header to check
699 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
701 struct nfs_pgio_header *hdr = calldata;
702 struct inode *inode = hdr->inode;
704 dprintk("NFS: %s: %5u, (status %d)\n", __func__,
705 task->tk_pid, task->tk_status);
707 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
709 if (task->tk_status < 0)
710 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
712 hdr->rw_ops->rw_result(task, hdr);
716 * Create an RPC task for the given read or write request and kick it.
717 * The page must have been locked by the caller.
719 * It may happen that the page we're passed is not marked dirty.
720 * This is the case if nfs_updatepage detects a conflicting request
721 * that has been written but not committed.
723 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
724 struct nfs_pgio_header *hdr)
726 struct nfs_page *req;
729 struct list_head *head = &desc->pg_list;
730 struct nfs_commit_info cinfo;
731 unsigned int pagecount, pageused;
733 pagecount = nfs_page_array_len(desc->pg_base, desc->pg_count);
734 if (!nfs_pgarray_set(&hdr->page_array, pagecount))
735 return nfs_pgio_error(desc, hdr);
737 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
738 pages = hdr->page_array.pagevec;
741 while (!list_empty(head)) {
742 req = nfs_list_entry(head->next);
743 nfs_list_remove_request(req);
744 nfs_list_add_request(req, &hdr->pages);
746 if (WARN_ON_ONCE(pageused >= pagecount))
747 return nfs_pgio_error(desc, hdr);
749 if (!last_page || last_page != req->wb_page) {
750 *pages++ = last_page = req->wb_page;
754 if (WARN_ON_ONCE(pageused != pagecount))
755 return nfs_pgio_error(desc, hdr);
757 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
758 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
759 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
761 /* Set up the argument struct */
762 nfs_pgio_rpcsetup(hdr, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
763 desc->pg_rpc_callops = &nfs_pgio_common_ops;
766 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
768 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
770 struct nfs_pgio_header *hdr;
773 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
775 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
778 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
779 ret = nfs_generic_pgio(desc, hdr);
781 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
782 hdr, desc->pg_rpc_callops,
783 desc->pg_ioflags, 0);
787 static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
788 const struct nfs_open_context *ctx2)
790 return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
793 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
794 const struct nfs_lock_context *l2)
796 return l1->lockowner.l_owner == l2->lockowner.l_owner
797 && l1->lockowner.l_pid == l2->lockowner.l_pid;
801 * nfs_can_coalesce_requests - test two requests for compatibility
802 * @prev: pointer to nfs_page
803 * @req: pointer to nfs_page
805 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
806 * page data area they describe is contiguous, and that their RPC
807 * credentials, NFSv4 open state, and lockowners are the same.
809 * Return 'true' if this is the case, else return 'false'.
811 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
812 struct nfs_page *req,
813 struct nfs_pageio_descriptor *pgio)
818 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
820 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
821 !nfs_match_lock_context(req->wb_lock_context,
822 prev->wb_lock_context))
824 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
826 if (req->wb_page == prev->wb_page) {
827 if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
830 if (req->wb_pgbase != 0 ||
831 prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
835 size = pgio->pg_ops->pg_test(pgio, prev, req);
836 WARN_ON_ONCE(size > req->wb_bytes);
837 if (size && size < req->wb_bytes)
838 req->wb_bytes = size;
843 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
844 * @desc: destination io descriptor
847 * Returns true if the request 'req' was successfully coalesced into the
848 * existing list of pages 'desc'.
850 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
851 struct nfs_page *req)
853 struct nfs_page *prev = NULL;
854 if (desc->pg_count != 0) {
855 prev = nfs_list_entry(desc->pg_list.prev);
857 if (desc->pg_ops->pg_init)
858 desc->pg_ops->pg_init(desc, req);
859 desc->pg_base = req->wb_pgbase;
861 if (!nfs_can_coalesce_requests(prev, req, desc))
863 nfs_list_remove_request(req);
864 nfs_list_add_request(req, &desc->pg_list);
865 desc->pg_count += req->wb_bytes;
870 * Helper for nfs_pageio_add_request and nfs_pageio_complete
872 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
874 if (!list_empty(&desc->pg_list)) {
875 int error = desc->pg_ops->pg_doio(desc);
877 desc->pg_error = error;
879 desc->pg_bytes_written += desc->pg_count;
881 if (list_empty(&desc->pg_list)) {
888 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
889 * @desc: destination io descriptor
892 * This may split a request into subrequests which are all part of the
895 * Returns true if the request 'req' was successfully coalesced into the
896 * existing list of pages 'desc'.
898 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
899 struct nfs_page *req)
901 struct nfs_page *subreq;
902 unsigned int bytes_left = 0;
903 unsigned int offset, pgbase;
905 nfs_page_group_lock(req, false);
908 bytes_left = subreq->wb_bytes;
909 offset = subreq->wb_offset;
910 pgbase = subreq->wb_pgbase;
913 if (!nfs_pageio_do_add_request(desc, subreq)) {
914 /* make sure pg_test call(s) did nothing */
915 WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
916 WARN_ON_ONCE(subreq->wb_offset != offset);
917 WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
919 nfs_page_group_unlock(req);
921 nfs_pageio_doio(desc);
922 if (desc->pg_error < 0)
924 if (desc->pg_recoalesce)
926 /* retry add_request for this subreq */
927 nfs_page_group_lock(req, false);
931 /* check for buggy pg_test call(s) */
932 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
933 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
934 WARN_ON_ONCE(subreq->wb_bytes == 0);
936 bytes_left -= subreq->wb_bytes;
937 offset += subreq->wb_bytes;
938 pgbase += subreq->wb_bytes;
941 subreq = nfs_create_request(req->wb_context,
943 subreq, pgbase, bytes_left);
946 nfs_lock_request(subreq);
947 subreq->wb_offset = offset;
948 subreq->wb_index = req->wb_index;
950 } while (bytes_left > 0);
952 nfs_page_group_unlock(req);
955 desc->pg_error = PTR_ERR(subreq);
956 nfs_page_group_unlock(req);
960 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
965 list_splice_init(&desc->pg_list, &head);
966 desc->pg_bytes_written -= desc->pg_count;
969 desc->pg_recoalesce = 0;
972 while (!list_empty(&head)) {
973 struct nfs_page *req;
975 req = list_first_entry(&head, struct nfs_page, wb_list);
976 nfs_list_remove_request(req);
977 if (__nfs_pageio_add_request(desc, req))
979 if (desc->pg_error < 0)
983 } while (desc->pg_recoalesce);
987 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
988 struct nfs_page *req)
993 ret = __nfs_pageio_add_request(desc, req);
996 if (desc->pg_error < 0)
998 ret = nfs_do_recoalesce(desc);
1004 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1005 * @hdr - the pgio header to move request from
1006 * @desc - the pageio descriptor to add requests to
1008 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1011 * Returns 0 on success and < 0 on error.
1013 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1014 struct nfs_pgio_header *hdr)
1018 desc->pg_dreq = hdr->dreq;
1019 while (!list_empty(&hdr->pages)) {
1020 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
1022 nfs_list_remove_request(req);
1023 if (!nfs_pageio_add_request(desc, req))
1024 nfs_list_add_request(req, &failed);
1026 nfs_pageio_complete(desc);
1027 if (!list_empty(&failed)) {
1028 list_move(&failed, &hdr->pages);
1033 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1036 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
1037 * @desc: pointer to io descriptor
1039 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1042 nfs_pageio_doio(desc);
1043 if (!desc->pg_recoalesce)
1045 if (!nfs_do_recoalesce(desc))
1051 * nfs_pageio_cond_complete - Conditional I/O completion
1052 * @desc: pointer to io descriptor
1053 * @index: page index
1055 * It is important to ensure that processes don't try to take locks
1056 * on non-contiguous ranges of pages as that might deadlock. This
1057 * function should be called before attempting to wait on a locked
1058 * nfs_page. It will complete the I/O if the page index 'index'
1059 * is not contiguous with the existing list of pages in 'desc'.
1061 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1063 if (!list_empty(&desc->pg_list)) {
1064 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
1065 if (index != prev->wb_index + 1)
1066 nfs_pageio_complete(desc);
1070 int __init nfs_init_nfspagecache(void)
1072 nfs_page_cachep = kmem_cache_create("nfs_page",
1073 sizeof(struct nfs_page),
1074 0, SLAB_HWCACHE_ALIGN,
1076 if (nfs_page_cachep == NULL)
1082 void nfs_destroy_nfspagecache(void)
1084 kmem_cache_destroy(nfs_page_cachep);
1087 static const struct rpc_call_ops nfs_pgio_common_ops = {
1088 .rpc_call_prepare = nfs_pgio_prepare,
1089 .rpc_call_done = nfs_pgio_result,
1090 .rpc_release = nfs_pgio_release,
1093 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1094 .pg_test = nfs_generic_pg_test,
1095 .pg_doio = nfs_generic_pg_pgios,