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 void nfs_free_request(struct nfs_page *);
34 static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
36 p->npages = pagecount;
37 if (pagecount <= ARRAY_SIZE(p->page_array))
38 p->pagevec = p->page_array;
40 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
44 return p->pagevec != NULL;
47 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
48 struct nfs_pgio_header *hdr,
49 void (*release)(struct nfs_pgio_header *hdr))
51 hdr->req = nfs_list_entry(desc->pg_list.next);
52 hdr->inode = desc->pg_inode;
53 hdr->cred = hdr->req->wb_context->cred;
54 hdr->io_start = req_offset(hdr->req);
55 hdr->good_bytes = desc->pg_count;
56 hdr->dreq = desc->pg_dreq;
57 hdr->layout_private = desc->pg_layout_private;
58 hdr->release = release;
59 hdr->completion_ops = desc->pg_completion_ops;
60 if (hdr->completion_ops->init_hdr)
61 hdr->completion_ops->init_hdr(hdr);
63 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
65 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
67 spin_lock(&hdr->lock);
68 if (pos < hdr->io_start + hdr->good_bytes) {
69 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
70 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
71 hdr->good_bytes = pos - hdr->io_start;
74 spin_unlock(&hdr->lock);
77 static inline struct nfs_page *
80 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
82 INIT_LIST_HEAD(&p->wb_list);
87 nfs_page_free(struct nfs_page *p)
89 kmem_cache_free(nfs_page_cachep, p);
93 nfs_iocounter_inc(struct nfs_io_counter *c)
95 atomic_inc(&c->io_count);
99 nfs_iocounter_dec(struct nfs_io_counter *c)
101 if (atomic_dec_and_test(&c->io_count)) {
102 clear_bit(NFS_IO_INPROGRESS, &c->flags);
103 smp_mb__after_clear_bit();
104 wake_up_bit(&c->flags, NFS_IO_INPROGRESS);
109 __nfs_iocounter_wait(struct nfs_io_counter *c)
111 wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS);
112 DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS);
116 prepare_to_wait(wq, &q.wait, TASK_KILLABLE);
117 set_bit(NFS_IO_INPROGRESS, &c->flags);
118 if (atomic_read(&c->io_count) == 0)
120 ret = nfs_wait_bit_killable(&c->flags);
121 } while (atomic_read(&c->io_count) != 0);
122 finish_wait(wq, &q.wait);
127 * nfs_iocounter_wait - wait for i/o to complete
128 * @c: nfs_io_counter to use
130 * returns -ERESTARTSYS if interrupted by a fatal signal.
131 * Otherwise returns 0 once the io_count hits 0.
134 nfs_iocounter_wait(struct nfs_io_counter *c)
136 if (atomic_read(&c->io_count) == 0)
138 return __nfs_iocounter_wait(c);
142 * nfs_page_group_lock - lock the head of the page group
143 * @req - request in group that is to be locked
145 * this lock must be held if modifying the page group list
148 nfs_page_group_lock(struct nfs_page *req)
150 struct nfs_page *head = req->wb_head;
153 WARN_ON_ONCE(head != head->wb_head);
156 err = wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
157 nfs_wait_bit_killable, TASK_KILLABLE);
161 * nfs_page_group_unlock - unlock the head of the page group
162 * @req - request in group that is to be unlocked
165 nfs_page_group_unlock(struct nfs_page *req)
167 struct nfs_page *head = req->wb_head;
169 WARN_ON_ONCE(head != head->wb_head);
171 smp_mb__before_clear_bit();
172 clear_bit(PG_HEADLOCK, &head->wb_flags);
173 smp_mb__after_clear_bit();
174 wake_up_bit(&head->wb_flags, PG_HEADLOCK);
178 * nfs_page_group_sync_on_bit_locked
180 * must be called with page group lock held
183 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
185 struct nfs_page *head = req->wb_head;
186 struct nfs_page *tmp;
188 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
189 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
191 tmp = req->wb_this_page;
193 if (!test_bit(bit, &tmp->wb_flags))
195 tmp = tmp->wb_this_page;
198 /* true! reset all bits */
201 clear_bit(bit, &tmp->wb_flags);
202 tmp = tmp->wb_this_page;
203 } while (tmp != req);
209 * nfs_page_group_sync_on_bit - set bit on current request, but only
210 * return true if the bit is set for all requests in page group
211 * @req - request in page group
212 * @bit - PG_* bit that is used to sync page group
214 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
218 nfs_page_group_lock(req);
219 ret = nfs_page_group_sync_on_bit_locked(req, bit);
220 nfs_page_group_unlock(req);
226 * nfs_page_group_init - Initialize the page group linkage for @req
227 * @req - a new nfs request
228 * @prev - the previous request in page group, or NULL if @req is the first
229 * or only request in the group (the head).
232 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
234 WARN_ON_ONCE(prev == req);
238 req->wb_this_page = req;
240 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
241 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
242 req->wb_head = prev->wb_head;
243 req->wb_this_page = prev->wb_this_page;
244 prev->wb_this_page = req;
246 /* grab extra ref if head request has extra ref from
247 * the write/commit path to handle handoff between write
248 * and commit lists */
249 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags))
250 kref_get(&req->wb_kref);
255 * nfs_page_group_destroy - sync the destruction of page groups
256 * @req - request that no longer needs the page group
258 * releases the page group reference from each member once all
259 * members have called this function.
262 nfs_page_group_destroy(struct kref *kref)
264 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
265 struct nfs_page *tmp, *next;
267 if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
272 next = tmp->wb_this_page;
273 /* unlink and free */
274 tmp->wb_this_page = tmp;
276 nfs_free_request(tmp);
278 } while (tmp != req);
282 * nfs_create_request - Create an NFS read/write request.
283 * @ctx: open context to use
284 * @page: page to write
285 * @last: last nfs request created for this page group or NULL if head
286 * @offset: starting offset within the page for the write
287 * @count: number of bytes to read/write
289 * The page must be locked by the caller. This makes sure we never
290 * create two different requests for the same page.
291 * User should ensure it is safe to sleep in this function.
294 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
295 struct nfs_page *last, unsigned int offset,
298 struct nfs_page *req;
299 struct nfs_lock_context *l_ctx;
301 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
302 return ERR_PTR(-EBADF);
303 /* try to allocate the request struct */
304 req = nfs_page_alloc();
306 return ERR_PTR(-ENOMEM);
308 /* get lock context early so we can deal with alloc failures */
309 l_ctx = nfs_get_lock_context(ctx);
312 return ERR_CAST(l_ctx);
314 req->wb_lock_context = l_ctx;
315 nfs_iocounter_inc(&l_ctx->io_count);
317 /* Initialize the request struct. Initially, we assume a
318 * long write-back delay. This will be adjusted in
319 * update_nfs_request below if the region is not locked. */
321 req->wb_index = page_file_index(page);
322 page_cache_get(page);
323 req->wb_offset = offset;
324 req->wb_pgbase = offset;
325 req->wb_bytes = count;
326 req->wb_context = get_nfs_open_context(ctx);
327 kref_init(&req->wb_kref);
328 nfs_page_group_init(req, last);
333 * nfs_unlock_request - Unlock request and wake up sleepers.
336 void nfs_unlock_request(struct nfs_page *req)
338 if (!NFS_WBACK_BUSY(req)) {
339 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
342 smp_mb__before_clear_bit();
343 clear_bit(PG_BUSY, &req->wb_flags);
344 smp_mb__after_clear_bit();
345 wake_up_bit(&req->wb_flags, PG_BUSY);
349 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
352 void nfs_unlock_and_release_request(struct nfs_page *req)
354 nfs_unlock_request(req);
355 nfs_release_request(req);
359 * nfs_clear_request - Free up all resources allocated to the request
362 * Release page and open context resources associated with a read/write
363 * request after it has completed.
365 static void nfs_clear_request(struct nfs_page *req)
367 struct page *page = req->wb_page;
368 struct nfs_open_context *ctx = req->wb_context;
369 struct nfs_lock_context *l_ctx = req->wb_lock_context;
372 page_cache_release(page);
376 nfs_iocounter_dec(&l_ctx->io_count);
377 nfs_put_lock_context(l_ctx);
378 req->wb_lock_context = NULL;
381 put_nfs_open_context(ctx);
382 req->wb_context = NULL;
387 * nfs_release_request - Release the count on an NFS read/write request
388 * @req: request to release
390 * Note: Should never be called with the spinlock held!
392 static void nfs_free_request(struct nfs_page *req)
394 WARN_ON_ONCE(req->wb_this_page != req);
396 /* extra debug: make sure no sync bits are still set */
397 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
398 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
399 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
400 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
401 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
403 /* Release struct file and open context */
404 nfs_clear_request(req);
408 void nfs_release_request(struct nfs_page *req)
410 kref_put(&req->wb_kref, nfs_page_group_destroy);
413 static int nfs_wait_bit_uninterruptible(void *word)
420 * nfs_wait_on_request - Wait for a request to complete.
421 * @req: request to wait upon.
423 * Interruptible by fatal signals only.
424 * The user is responsible for holding a count on the request.
427 nfs_wait_on_request(struct nfs_page *req)
429 return wait_on_bit(&req->wb_flags, PG_BUSY,
430 nfs_wait_bit_uninterruptible,
431 TASK_UNINTERRUPTIBLE);
435 * nfs_generic_pg_test - determine if requests can be coalesced
436 * @desc: pointer to descriptor
437 * @prev: previous request in desc, or NULL
440 * Returns zero if @req can be coalesced into @desc, otherwise it returns
441 * the size of the request.
443 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
444 struct nfs_page *prev, struct nfs_page *req)
446 if (desc->pg_count > desc->pg_bsize) {
447 /* should never happen */
452 return min(desc->pg_bsize - desc->pg_count, (size_t)req->wb_bytes);
454 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
456 static inline struct nfs_rw_header *NFS_RW_HEADER(struct nfs_pgio_header *hdr)
458 return container_of(hdr, struct nfs_rw_header, header);
462 * nfs_rw_header_alloc - Allocate a header for a read or write
463 * @ops: Read or write function vector
465 struct nfs_rw_header *nfs_rw_header_alloc(const struct nfs_rw_ops *ops)
467 struct nfs_rw_header *header = ops->rw_alloc_header();
470 struct nfs_pgio_header *hdr = &header->header;
472 INIT_LIST_HEAD(&hdr->pages);
473 INIT_LIST_HEAD(&hdr->rpc_list);
474 spin_lock_init(&hdr->lock);
475 atomic_set(&hdr->refcnt, 0);
480 EXPORT_SYMBOL_GPL(nfs_rw_header_alloc);
483 * nfs_rw_header_free - Free a read or write header
484 * @hdr: The header to free
486 void nfs_rw_header_free(struct nfs_pgio_header *hdr)
488 hdr->rw_ops->rw_free_header(NFS_RW_HEADER(hdr));
490 EXPORT_SYMBOL_GPL(nfs_rw_header_free);
493 * nfs_pgio_data_alloc - Allocate pageio data
494 * @hdr: The header making a request
495 * @pagecount: Number of pages to create
497 static struct nfs_pgio_data *nfs_pgio_data_alloc(struct nfs_pgio_header *hdr,
498 unsigned int pagecount)
500 struct nfs_pgio_data *data, *prealloc;
502 prealloc = &NFS_RW_HEADER(hdr)->rpc_data;
503 if (prealloc->header == NULL)
506 data = kzalloc(sizeof(*data), GFP_KERNEL);
510 if (nfs_pgarray_set(&data->pages, pagecount)) {
512 atomic_inc(&hdr->refcnt);
514 if (data != prealloc)
523 * nfs_pgio_data_release - Properly free pageio data
524 * @data: The data to release
526 void nfs_pgio_data_release(struct nfs_pgio_data *data)
528 struct nfs_pgio_header *hdr = data->header;
529 struct nfs_rw_header *pageio_header = NFS_RW_HEADER(hdr);
531 put_nfs_open_context(data->args.context);
532 if (data->pages.pagevec != data->pages.page_array)
533 kfree(data->pages.pagevec);
534 if (data == &pageio_header->rpc_data) {
538 if (atomic_dec_and_test(&hdr->refcnt))
539 hdr->completion_ops->completion(hdr);
540 /* Note: we only free the rpc_task after callbacks are done.
541 * See the comment in rpc_free_task() for why
545 EXPORT_SYMBOL_GPL(nfs_pgio_data_release);
548 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
549 * @data: The pageio data
550 * @count: Number of bytes to read
551 * @offset: Initial offset
552 * @how: How to commit data (writes only)
553 * @cinfo: Commit information for the call (writes only)
555 static void nfs_pgio_rpcsetup(struct nfs_pgio_data *data,
556 unsigned int count, unsigned int offset,
557 int how, struct nfs_commit_info *cinfo)
559 struct nfs_page *req = data->header->req;
561 /* Set up the RPC argument and reply structs
562 * NB: take care not to mess about with data->commit et al. */
564 data->args.fh = NFS_FH(data->header->inode);
565 data->args.offset = req_offset(req) + offset;
566 /* pnfs_set_layoutcommit needs this */
567 data->mds_offset = data->args.offset;
568 data->args.pgbase = req->wb_pgbase + offset;
569 data->args.pages = data->pages.pagevec;
570 data->args.count = count;
571 data->args.context = get_nfs_open_context(req->wb_context);
572 data->args.lock_context = req->wb_lock_context;
573 data->args.stable = NFS_UNSTABLE;
574 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
577 case FLUSH_COND_STABLE:
578 if (nfs_reqs_to_commit(cinfo))
581 data->args.stable = NFS_FILE_SYNC;
584 data->res.fattr = &data->fattr;
585 data->res.count = count;
587 data->res.verf = &data->verf;
588 nfs_fattr_init(&data->fattr);
592 * nfs_pgio_prepare - Prepare pageio data to go over the wire
593 * @task: The current task
594 * @calldata: pageio data to prepare
596 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
598 struct nfs_pgio_data *data = calldata;
600 err = NFS_PROTO(data->header->inode)->pgio_rpc_prepare(task, data);
605 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_data *data,
606 const struct rpc_call_ops *call_ops, int how, int flags)
608 struct rpc_task *task;
609 struct rpc_message msg = {
610 .rpc_argp = &data->args,
611 .rpc_resp = &data->res,
612 .rpc_cred = data->header->cred,
614 struct rpc_task_setup task_setup_data = {
618 .callback_ops = call_ops,
619 .callback_data = data,
620 .workqueue = nfsiod_workqueue,
621 .flags = RPC_TASK_ASYNC | flags,
625 data->header->rw_ops->rw_initiate(data, &msg, &task_setup_data, how);
627 dprintk("NFS: %5u initiated pgio call "
628 "(req %s/%llu, %u bytes @ offset %llu)\n",
630 data->header->inode->i_sb->s_id,
631 (unsigned long long)NFS_FILEID(data->header->inode),
633 (unsigned long long)data->args.offset);
635 task = rpc_run_task(&task_setup_data);
640 if (how & FLUSH_SYNC) {
641 ret = rpc_wait_for_completion_task(task);
643 ret = task->tk_status;
649 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
651 static int nfs_do_multiple_pgios(struct list_head *head,
652 const struct rpc_call_ops *call_ops,
655 struct nfs_pgio_data *data;
658 while (!list_empty(head)) {
661 data = list_first_entry(head, struct nfs_pgio_data, list);
662 list_del_init(&data->list);
664 ret2 = nfs_initiate_pgio(NFS_CLIENT(data->header->inode),
665 data, call_ops, how, 0);
673 * nfs_pgio_error - Clean up from a pageio error
674 * @desc: IO descriptor
675 * @hdr: pageio header
677 static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
678 struct nfs_pgio_header *hdr)
680 struct nfs_pgio_data *data;
682 set_bit(NFS_IOHDR_REDO, &hdr->flags);
683 while (!list_empty(&hdr->rpc_list)) {
684 data = list_first_entry(&hdr->rpc_list, struct nfs_pgio_data, list);
685 list_del(&data->list);
686 nfs_pgio_data_release(data);
688 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
693 * nfs_pgio_release - Release pageio data
694 * @calldata: The pageio data to release
696 static void nfs_pgio_release(void *calldata)
698 struct nfs_pgio_data *data = calldata;
699 if (data->header->rw_ops->rw_release)
700 data->header->rw_ops->rw_release(data);
701 nfs_pgio_data_release(data);
705 * nfs_pageio_init - initialise a page io descriptor
706 * @desc: pointer to descriptor
707 * @inode: pointer to inode
708 * @doio: pointer to io function
709 * @bsize: io block size
710 * @io_flags: extra parameters for the io function
712 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
714 const struct nfs_pageio_ops *pg_ops,
715 const struct nfs_pgio_completion_ops *compl_ops,
716 const struct nfs_rw_ops *rw_ops,
720 INIT_LIST_HEAD(&desc->pg_list);
721 desc->pg_bytes_written = 0;
723 desc->pg_bsize = bsize;
726 desc->pg_recoalesce = 0;
727 desc->pg_inode = inode;
728 desc->pg_ops = pg_ops;
729 desc->pg_completion_ops = compl_ops;
730 desc->pg_rw_ops = rw_ops;
731 desc->pg_ioflags = io_flags;
733 desc->pg_lseg = NULL;
734 desc->pg_dreq = NULL;
735 desc->pg_layout_private = NULL;
737 EXPORT_SYMBOL_GPL(nfs_pageio_init);
740 * nfs_pgio_result - Basic pageio error handling
741 * @task: The task that ran
742 * @calldata: Pageio data to check
744 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
746 struct nfs_pgio_data *data = calldata;
747 struct inode *inode = data->header->inode;
749 dprintk("NFS: %s: %5u, (status %d)\n", __func__,
750 task->tk_pid, task->tk_status);
752 if (data->header->rw_ops->rw_done(task, data, inode) != 0)
754 if (task->tk_status < 0)
755 nfs_set_pgio_error(data->header, task->tk_status, data->args.offset);
757 data->header->rw_ops->rw_result(task, data);
761 * Create an RPC task for the given read or write request and kick it.
762 * The page must have been locked by the caller.
764 * It may happen that the page we're passed is not marked dirty.
765 * This is the case if nfs_updatepage detects a conflicting request
766 * that has been written but not committed.
768 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
769 struct nfs_pgio_header *hdr)
771 struct nfs_page *req;
773 struct nfs_pgio_data *data;
774 struct list_head *head = &desc->pg_list;
775 struct nfs_commit_info cinfo;
777 data = nfs_pgio_data_alloc(hdr, nfs_page_array_len(desc->pg_base,
780 return nfs_pgio_error(desc, hdr);
782 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
783 pages = data->pages.pagevec;
784 while (!list_empty(head)) {
785 req = nfs_list_entry(head->next);
786 nfs_list_remove_request(req);
787 nfs_list_add_request(req, &hdr->pages);
788 *pages++ = req->wb_page;
791 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
792 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
793 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
795 /* Set up the argument struct */
796 nfs_pgio_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
797 list_add(&data->list, &hdr->rpc_list);
798 desc->pg_rpc_callops = &nfs_pgio_common_ops;
801 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
803 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
805 struct nfs_rw_header *rw_hdr;
806 struct nfs_pgio_header *hdr;
809 rw_hdr = nfs_rw_header_alloc(desc->pg_rw_ops);
811 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
814 hdr = &rw_hdr->header;
815 nfs_pgheader_init(desc, hdr, nfs_rw_header_free);
816 atomic_inc(&hdr->refcnt);
817 ret = nfs_generic_pgio(desc, hdr);
819 ret = nfs_do_multiple_pgios(&hdr->rpc_list,
820 desc->pg_rpc_callops,
822 if (atomic_dec_and_test(&hdr->refcnt))
823 hdr->completion_ops->completion(hdr);
827 static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
828 const struct nfs_open_context *ctx2)
830 return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
833 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
834 const struct nfs_lock_context *l2)
836 return l1->lockowner.l_owner == l2->lockowner.l_owner
837 && l1->lockowner.l_pid == l2->lockowner.l_pid;
841 * nfs_can_coalesce_requests - test two requests for compatibility
842 * @prev: pointer to nfs_page
843 * @req: pointer to nfs_page
845 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
846 * page data area they describe is contiguous, and that their RPC
847 * credentials, NFSv4 open state, and lockowners are the same.
849 * Return 'true' if this is the case, else return 'false'.
851 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
852 struct nfs_page *req,
853 struct nfs_pageio_descriptor *pgio)
858 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
860 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
861 !nfs_match_lock_context(req->wb_lock_context,
862 prev->wb_lock_context))
864 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
867 size = pgio->pg_ops->pg_test(pgio, prev, req);
868 WARN_ON_ONCE(size > req->wb_bytes);
869 if (size && size < req->wb_bytes)
870 req->wb_bytes = size;
875 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
876 * @desc: destination io descriptor
879 * Returns true if the request 'req' was successfully coalesced into the
880 * existing list of pages 'desc'.
882 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
883 struct nfs_page *req)
885 struct nfs_page *prev = NULL;
886 if (desc->pg_count != 0) {
887 prev = nfs_list_entry(desc->pg_list.prev);
889 if (desc->pg_ops->pg_init)
890 desc->pg_ops->pg_init(desc, req);
891 desc->pg_base = req->wb_pgbase;
893 if (!nfs_can_coalesce_requests(prev, req, desc))
895 nfs_list_remove_request(req);
896 nfs_list_add_request(req, &desc->pg_list);
897 desc->pg_count += req->wb_bytes;
902 * Helper for nfs_pageio_add_request and nfs_pageio_complete
904 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
906 if (!list_empty(&desc->pg_list)) {
907 int error = desc->pg_ops->pg_doio(desc);
909 desc->pg_error = error;
911 desc->pg_bytes_written += desc->pg_count;
913 if (list_empty(&desc->pg_list)) {
920 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
921 * @desc: destination io descriptor
924 * This may split a request into subrequests which are all part of the
927 * Returns true if the request 'req' was successfully coalesced into the
928 * existing list of pages 'desc'.
930 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
931 struct nfs_page *req)
933 struct nfs_page *subreq;
934 unsigned int bytes_left = 0;
935 unsigned int offset, pgbase;
937 nfs_page_group_lock(req);
940 bytes_left = subreq->wb_bytes;
941 offset = subreq->wb_offset;
942 pgbase = subreq->wb_pgbase;
945 if (!nfs_pageio_do_add_request(desc, subreq)) {
946 /* make sure pg_test call(s) did nothing */
947 WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
948 WARN_ON_ONCE(subreq->wb_offset != offset);
949 WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
951 nfs_page_group_unlock(req);
953 nfs_pageio_doio(desc);
954 if (desc->pg_error < 0)
957 if (desc->pg_recoalesce)
959 /* retry add_request for this subreq */
960 nfs_page_group_lock(req);
964 /* check for buggy pg_test call(s) */
965 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
966 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
967 WARN_ON_ONCE(subreq->wb_bytes == 0);
969 bytes_left -= subreq->wb_bytes;
970 offset += subreq->wb_bytes;
971 pgbase += subreq->wb_bytes;
974 subreq = nfs_create_request(req->wb_context,
976 subreq, pgbase, bytes_left);
977 nfs_lock_request(subreq);
978 subreq->wb_offset = offset;
979 subreq->wb_index = req->wb_index;
981 } while (bytes_left > 0);
983 nfs_page_group_unlock(req);
987 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
992 list_splice_init(&desc->pg_list, &head);
993 desc->pg_bytes_written -= desc->pg_count;
996 desc->pg_recoalesce = 0;
998 while (!list_empty(&head)) {
999 struct nfs_page *req;
1001 req = list_first_entry(&head, struct nfs_page, wb_list);
1002 nfs_list_remove_request(req);
1003 if (__nfs_pageio_add_request(desc, req))
1005 if (desc->pg_error < 0)
1009 } while (desc->pg_recoalesce);
1013 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1014 struct nfs_page *req)
1019 ret = __nfs_pageio_add_request(desc, req);
1022 if (desc->pg_error < 0)
1024 ret = nfs_do_recoalesce(desc);
1028 EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
1031 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
1032 * @desc: pointer to io descriptor
1034 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1037 nfs_pageio_doio(desc);
1038 if (!desc->pg_recoalesce)
1040 if (!nfs_do_recoalesce(desc))
1044 EXPORT_SYMBOL_GPL(nfs_pageio_complete);
1047 * nfs_pageio_cond_complete - Conditional I/O completion
1048 * @desc: pointer to io descriptor
1049 * @index: page index
1051 * It is important to ensure that processes don't try to take locks
1052 * on non-contiguous ranges of pages as that might deadlock. This
1053 * function should be called before attempting to wait on a locked
1054 * nfs_page. It will complete the I/O if the page index 'index'
1055 * is not contiguous with the existing list of pages in 'desc'.
1057 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1059 if (!list_empty(&desc->pg_list)) {
1060 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
1061 if (index != prev->wb_index + 1)
1062 nfs_pageio_complete(desc);
1066 int __init nfs_init_nfspagecache(void)
1068 nfs_page_cachep = kmem_cache_create("nfs_page",
1069 sizeof(struct nfs_page),
1070 0, SLAB_HWCACHE_ALIGN,
1072 if (nfs_page_cachep == NULL)
1078 void nfs_destroy_nfspagecache(void)
1080 kmem_cache_destroy(nfs_page_cachep);
1083 static const struct rpc_call_ops nfs_pgio_common_ops = {
1084 .rpc_call_prepare = nfs_pgio_prepare,
1085 .rpc_call_done = nfs_pgio_result,
1086 .rpc_release = nfs_pgio_release,
1089 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1090 .pg_test = nfs_generic_pg_test,
1091 .pg_doio = nfs_generic_pg_pgios,