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 struct nfs_pgio_mirror *
46 nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
48 return nfs_pgio_has_mirroring(desc) ?
49 &desc->pg_mirrors[desc->pg_mirror_idx] :
52 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
54 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
55 struct nfs_pgio_header *hdr,
56 void (*release)(struct nfs_pgio_header *hdr))
58 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
61 hdr->req = nfs_list_entry(mirror->pg_list.next);
62 hdr->inode = desc->pg_inode;
63 hdr->cred = hdr->req->wb_context->cred;
64 hdr->io_start = req_offset(hdr->req);
65 hdr->good_bytes = mirror->pg_count;
66 hdr->dreq = desc->pg_dreq;
67 hdr->layout_private = desc->pg_layout_private;
68 hdr->release = release;
69 hdr->completion_ops = desc->pg_completion_ops;
70 if (hdr->completion_ops->init_hdr)
71 hdr->completion_ops->init_hdr(hdr);
73 hdr->pgio_mirror_idx = desc->pg_mirror_idx;
75 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
77 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
79 spin_lock(&hdr->lock);
80 if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags)
81 || pos < hdr->io_start + hdr->good_bytes) {
82 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
83 hdr->good_bytes = pos - hdr->io_start;
86 spin_unlock(&hdr->lock);
89 static inline struct nfs_page *
92 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
94 INIT_LIST_HEAD(&p->wb_list);
99 nfs_page_free(struct nfs_page *p)
101 kmem_cache_free(nfs_page_cachep, p);
105 nfs_iocounter_inc(struct nfs_io_counter *c)
107 atomic_inc(&c->io_count);
111 nfs_iocounter_dec(struct nfs_io_counter *c)
113 if (atomic_dec_and_test(&c->io_count)) {
114 clear_bit(NFS_IO_INPROGRESS, &c->flags);
115 smp_mb__after_atomic();
116 wake_up_bit(&c->flags, NFS_IO_INPROGRESS);
121 __nfs_iocounter_wait(struct nfs_io_counter *c)
123 wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS);
124 DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS);
128 prepare_to_wait(wq, &q.wait, TASK_KILLABLE);
129 set_bit(NFS_IO_INPROGRESS, &c->flags);
130 if (atomic_read(&c->io_count) == 0)
132 ret = nfs_wait_bit_killable(&q.key, TASK_KILLABLE);
133 } while (atomic_read(&c->io_count) != 0 && !ret);
134 finish_wait(wq, &q.wait);
139 * nfs_iocounter_wait - wait for i/o to complete
140 * @c: nfs_io_counter to use
142 * returns -ERESTARTSYS if interrupted by a fatal signal.
143 * Otherwise returns 0 once the io_count hits 0.
146 nfs_iocounter_wait(struct nfs_io_counter *c)
148 if (atomic_read(&c->io_count) == 0)
150 return __nfs_iocounter_wait(c);
154 * nfs_page_group_lock - lock the head of the page group
155 * @req - request in group that is to be locked
156 * @nonblock - if true don't block waiting for lock
158 * this lock must be held if modifying the page group list
160 * return 0 on success, < 0 on error: -EDELAY if nonblocking or the
161 * result from wait_on_bit_lock
163 * NOTE: calling with nonblock=false should always have set the
164 * lock bit (see fs/buffer.c and other uses of wait_on_bit_lock
165 * with TASK_UNINTERRUPTIBLE), so there is no need to check the result.
168 nfs_page_group_lock(struct nfs_page *req, bool nonblock)
170 struct nfs_page *head = req->wb_head;
172 WARN_ON_ONCE(head != head->wb_head);
174 if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags))
178 return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
179 TASK_UNINTERRUPTIBLE);
185 * nfs_page_group_lock_wait - wait for the lock to clear, but don't grab it
186 * @req - a request in the group
188 * This is a blocking call to wait for the group lock to be cleared.
191 nfs_page_group_lock_wait(struct nfs_page *req)
193 struct nfs_page *head = req->wb_head;
195 WARN_ON_ONCE(head != head->wb_head);
197 wait_on_bit(&head->wb_flags, PG_HEADLOCK,
198 TASK_UNINTERRUPTIBLE);
202 * nfs_page_group_unlock - unlock the head of the page group
203 * @req - request in group that is to be unlocked
206 nfs_page_group_unlock(struct nfs_page *req)
208 struct nfs_page *head = req->wb_head;
210 WARN_ON_ONCE(head != head->wb_head);
212 smp_mb__before_atomic();
213 clear_bit(PG_HEADLOCK, &head->wb_flags);
214 smp_mb__after_atomic();
215 wake_up_bit(&head->wb_flags, PG_HEADLOCK);
219 * nfs_page_group_sync_on_bit_locked
221 * must be called with page group lock held
224 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
226 struct nfs_page *head = req->wb_head;
227 struct nfs_page *tmp;
229 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
230 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
232 tmp = req->wb_this_page;
234 if (!test_bit(bit, &tmp->wb_flags))
236 tmp = tmp->wb_this_page;
239 /* true! reset all bits */
242 clear_bit(bit, &tmp->wb_flags);
243 tmp = tmp->wb_this_page;
244 } while (tmp != req);
250 * nfs_page_group_sync_on_bit - set bit on current request, but only
251 * return true if the bit is set for all requests in page group
252 * @req - request in page group
253 * @bit - PG_* bit that is used to sync page group
255 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
259 nfs_page_group_lock(req, false);
260 ret = nfs_page_group_sync_on_bit_locked(req, bit);
261 nfs_page_group_unlock(req);
267 * nfs_page_group_init - Initialize the page group linkage for @req
268 * @req - a new nfs request
269 * @prev - the previous request in page group, or NULL if @req is the first
270 * or only request in the group (the head).
273 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
276 WARN_ON_ONCE(prev == req);
281 req->wb_this_page = req;
284 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
285 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
286 req->wb_head = prev->wb_head;
287 req->wb_this_page = prev->wb_this_page;
288 prev->wb_this_page = req;
290 /* All subrequests take a ref on the head request until
291 * nfs_page_group_destroy is called */
292 kref_get(&req->wb_head->wb_kref);
294 /* grab extra ref and bump the request count if head request
295 * has extra ref from the write/commit path to handle handoff
296 * between write and commit lists. */
297 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
298 inode = page_file_mapping(req->wb_page)->host;
299 set_bit(PG_INODE_REF, &req->wb_flags);
300 kref_get(&req->wb_kref);
301 spin_lock(&inode->i_lock);
302 NFS_I(inode)->nrequests++;
303 spin_unlock(&inode->i_lock);
309 * nfs_page_group_destroy - sync the destruction of page groups
310 * @req - request that no longer needs the page group
312 * releases the page group reference from each member once all
313 * members have called this function.
316 nfs_page_group_destroy(struct kref *kref)
318 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
319 struct nfs_page *tmp, *next;
321 /* subrequests must release the ref on the head request */
322 if (req->wb_head != req)
323 nfs_release_request(req->wb_head);
325 if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
330 next = tmp->wb_this_page;
331 /* unlink and free */
332 tmp->wb_this_page = tmp;
334 nfs_free_request(tmp);
336 } while (tmp != req);
340 * nfs_create_request - Create an NFS read/write request.
341 * @ctx: open context to use
342 * @page: page to write
343 * @last: last nfs request created for this page group or NULL if head
344 * @offset: starting offset within the page for the write
345 * @count: number of bytes to read/write
347 * The page must be locked by the caller. This makes sure we never
348 * create two different requests for the same page.
349 * User should ensure it is safe to sleep in this function.
352 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
353 struct nfs_page *last, unsigned int offset,
356 struct nfs_page *req;
357 struct nfs_lock_context *l_ctx;
359 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
360 return ERR_PTR(-EBADF);
361 /* try to allocate the request struct */
362 req = nfs_page_alloc();
364 return ERR_PTR(-ENOMEM);
366 /* get lock context early so we can deal with alloc failures */
367 l_ctx = nfs_get_lock_context(ctx);
370 return ERR_CAST(l_ctx);
372 req->wb_lock_context = l_ctx;
373 nfs_iocounter_inc(&l_ctx->io_count);
375 /* Initialize the request struct. Initially, we assume a
376 * long write-back delay. This will be adjusted in
377 * update_nfs_request below if the region is not locked. */
379 req->wb_index = page_file_index(page);
380 page_cache_get(page);
381 req->wb_offset = offset;
382 req->wb_pgbase = offset;
383 req->wb_bytes = count;
384 req->wb_context = get_nfs_open_context(ctx);
385 kref_init(&req->wb_kref);
386 nfs_page_group_init(req, last);
391 * nfs_unlock_request - Unlock request and wake up sleepers.
394 void nfs_unlock_request(struct nfs_page *req)
396 if (!NFS_WBACK_BUSY(req)) {
397 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
400 smp_mb__before_atomic();
401 clear_bit(PG_BUSY, &req->wb_flags);
402 smp_mb__after_atomic();
403 wake_up_bit(&req->wb_flags, PG_BUSY);
407 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
410 void nfs_unlock_and_release_request(struct nfs_page *req)
412 nfs_unlock_request(req);
413 nfs_release_request(req);
417 * nfs_clear_request - Free up all resources allocated to the request
420 * Release page and open context resources associated with a read/write
421 * request after it has completed.
423 static void nfs_clear_request(struct nfs_page *req)
425 struct page *page = req->wb_page;
426 struct nfs_open_context *ctx = req->wb_context;
427 struct nfs_lock_context *l_ctx = req->wb_lock_context;
430 page_cache_release(page);
434 nfs_iocounter_dec(&l_ctx->io_count);
435 nfs_put_lock_context(l_ctx);
436 req->wb_lock_context = NULL;
439 put_nfs_open_context(ctx);
440 req->wb_context = NULL;
445 * nfs_release_request - Release the count on an NFS read/write request
446 * @req: request to release
448 * Note: Should never be called with the spinlock held!
450 void nfs_free_request(struct nfs_page *req)
452 WARN_ON_ONCE(req->wb_this_page != req);
454 /* extra debug: make sure no sync bits are still set */
455 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
456 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
457 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
458 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
459 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
461 /* Release struct file and open context */
462 nfs_clear_request(req);
466 void nfs_release_request(struct nfs_page *req)
468 kref_put(&req->wb_kref, nfs_page_group_destroy);
472 * nfs_wait_on_request - Wait for a request to complete.
473 * @req: request to wait upon.
475 * Interruptible by fatal signals only.
476 * The user is responsible for holding a count on the request.
479 nfs_wait_on_request(struct nfs_page *req)
481 return wait_on_bit_io(&req->wb_flags, PG_BUSY,
482 TASK_UNINTERRUPTIBLE);
486 * nfs_generic_pg_test - determine if requests can be coalesced
487 * @desc: pointer to descriptor
488 * @prev: previous request in desc, or NULL
491 * Returns zero if @req can be coalesced into @desc, otherwise it returns
492 * the size of the request.
494 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
495 struct nfs_page *prev, struct nfs_page *req)
497 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
500 if (mirror->pg_count > mirror->pg_bsize) {
501 /* should never happen */
507 * Limit the request size so that we can still allocate a page array
508 * for it without upsetting the slab allocator.
510 if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
511 sizeof(struct page *) > PAGE_SIZE)
514 return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
516 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
518 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
520 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
523 INIT_LIST_HEAD(&hdr->pages);
524 spin_lock_init(&hdr->lock);
529 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
532 * nfs_pgio_data_destroy - make @hdr suitable for reuse
534 * Frees memory and releases refs from nfs_generic_pgio, so that it may
537 * @hdr: A header that has had nfs_generic_pgio called
539 static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
541 if (hdr->args.context)
542 put_nfs_open_context(hdr->args.context);
543 if (hdr->page_array.pagevec != hdr->page_array.page_array)
544 kfree(hdr->page_array.pagevec);
548 * nfs_pgio_header_free - Free a read or write header
549 * @hdr: The header to free
551 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
553 nfs_pgio_data_destroy(hdr);
554 hdr->rw_ops->rw_free_header(hdr);
556 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
559 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
560 * @hdr: The pageio hdr
561 * @count: Number of bytes to read
562 * @offset: Initial offset
563 * @how: How to commit data (writes only)
564 * @cinfo: Commit information for the call (writes only)
566 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
567 unsigned int count, unsigned int offset,
568 int how, struct nfs_commit_info *cinfo)
570 struct nfs_page *req = hdr->req;
572 /* Set up the RPC argument and reply structs
573 * NB: take care not to mess about with hdr->commit et al. */
575 hdr->args.fh = NFS_FH(hdr->inode);
576 hdr->args.offset = req_offset(req) + offset;
577 /* pnfs_set_layoutcommit needs this */
578 hdr->mds_offset = hdr->args.offset;
579 hdr->args.pgbase = req->wb_pgbase + offset;
580 hdr->args.pages = hdr->page_array.pagevec;
581 hdr->args.count = count;
582 hdr->args.context = get_nfs_open_context(req->wb_context);
583 hdr->args.lock_context = req->wb_lock_context;
584 hdr->args.stable = NFS_UNSTABLE;
585 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
588 case FLUSH_COND_STABLE:
589 if (nfs_reqs_to_commit(cinfo))
592 hdr->args.stable = NFS_FILE_SYNC;
595 hdr->res.fattr = &hdr->fattr;
598 hdr->res.verf = &hdr->verf;
599 nfs_fattr_init(&hdr->fattr);
603 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
604 * @task: The current task
605 * @calldata: pageio header to prepare
607 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
609 struct nfs_pgio_header *hdr = calldata;
611 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
616 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
617 struct rpc_cred *cred, const struct nfs_rpc_ops *rpc_ops,
618 const struct rpc_call_ops *call_ops, int how, int flags)
620 struct rpc_task *task;
621 struct rpc_message msg = {
622 .rpc_argp = &hdr->args,
623 .rpc_resp = &hdr->res,
626 struct rpc_task_setup task_setup_data = {
630 .callback_ops = call_ops,
631 .callback_data = hdr,
632 .workqueue = nfsiod_workqueue,
633 .flags = RPC_TASK_ASYNC | flags,
637 hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
639 dprintk("NFS: initiated pgio call "
640 "(req %s/%llu, %u bytes @ offset %llu)\n",
641 hdr->inode->i_sb->s_id,
642 (unsigned long long)NFS_FILEID(hdr->inode),
644 (unsigned long long)hdr->args.offset);
646 task = rpc_run_task(&task_setup_data);
651 if (how & FLUSH_SYNC) {
652 ret = rpc_wait_for_completion_task(task);
654 ret = task->tk_status;
660 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
663 * nfs_pgio_error - Clean up from a pageio error
664 * @desc: IO descriptor
665 * @hdr: pageio header
667 static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
668 struct nfs_pgio_header *hdr)
670 struct nfs_pgio_mirror *mirror;
673 set_bit(NFS_IOHDR_REDO, &hdr->flags);
674 hdr->completion_ops->completion(hdr);
675 /* TODO: Make sure it's right to clean up all mirrors here
676 * and not just hdr->pgio_mirror_idx */
677 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
678 mirror = &desc->pg_mirrors[midx];
679 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
685 * nfs_pgio_release - Release pageio data
686 * @calldata: The pageio header to release
688 static void nfs_pgio_release(void *calldata)
690 struct nfs_pgio_header *hdr = calldata;
691 hdr->completion_ops->completion(hdr);
694 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
697 INIT_LIST_HEAD(&mirror->pg_list);
698 mirror->pg_bytes_written = 0;
699 mirror->pg_count = 0;
700 mirror->pg_bsize = bsize;
702 mirror->pg_recoalesce = 0;
706 * nfs_pageio_init - initialise a page io descriptor
707 * @desc: pointer to descriptor
708 * @inode: pointer to inode
709 * @pg_ops: pointer to pageio operations
710 * @compl_ops: pointer to pageio completion operations
711 * @rw_ops: pointer to nfs read/write operations
712 * @bsize: io block size
713 * @io_flags: extra parameters for the io function
715 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
717 const struct nfs_pageio_ops *pg_ops,
718 const struct nfs_pgio_completion_ops *compl_ops,
719 const struct nfs_rw_ops *rw_ops,
723 struct nfs_pgio_mirror *new;
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;
736 desc->pg_bsize = bsize;
738 desc->pg_mirror_count = 1;
739 desc->pg_mirror_idx = 0;
741 if (pg_ops->pg_get_mirror_count) {
742 /* until we have a request, we don't have an lseg and no
743 * idea how many mirrors there will be */
744 new = kcalloc(NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX,
745 sizeof(struct nfs_pgio_mirror), GFP_KERNEL);
746 desc->pg_mirrors_dynamic = new;
747 desc->pg_mirrors = new;
749 for (i = 0; i < NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX; i++)
750 nfs_pageio_mirror_init(&desc->pg_mirrors[i], bsize);
752 desc->pg_mirrors_dynamic = NULL;
753 desc->pg_mirrors = desc->pg_mirrors_static;
754 nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
757 EXPORT_SYMBOL_GPL(nfs_pageio_init);
760 * nfs_pgio_result - Basic pageio error handling
761 * @task: The task that ran
762 * @calldata: Pageio header to check
764 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
766 struct nfs_pgio_header *hdr = calldata;
767 struct inode *inode = hdr->inode;
769 dprintk("NFS: %s: %5u, (status %d)\n", __func__,
770 task->tk_pid, task->tk_status);
772 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
774 if (task->tk_status < 0)
775 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
777 hdr->rw_ops->rw_result(task, hdr);
781 * Create an RPC task for the given read or write request and kick it.
782 * The page must have been locked by the caller.
784 * It may happen that the page we're passed is not marked dirty.
785 * This is the case if nfs_updatepage detects a conflicting request
786 * that has been written but not committed.
788 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
789 struct nfs_pgio_header *hdr)
791 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
793 struct nfs_page *req;
796 struct list_head *head = &mirror->pg_list;
797 struct nfs_commit_info cinfo;
798 unsigned int pagecount, pageused;
800 pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
801 if (!nfs_pgarray_set(&hdr->page_array, pagecount))
802 return nfs_pgio_error(desc, hdr);
804 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
805 pages = hdr->page_array.pagevec;
808 while (!list_empty(head)) {
809 req = nfs_list_entry(head->next);
810 nfs_list_remove_request(req);
811 nfs_list_add_request(req, &hdr->pages);
813 if (!last_page || last_page != req->wb_page) {
815 if (pageused > pagecount)
817 *pages++ = last_page = req->wb_page;
820 if (WARN_ON_ONCE(pageused != pagecount))
821 return nfs_pgio_error(desc, hdr);
823 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
824 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
825 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
827 /* Set up the argument struct */
828 nfs_pgio_rpcsetup(hdr, mirror->pg_count, 0, desc->pg_ioflags, &cinfo);
829 desc->pg_rpc_callops = &nfs_pgio_common_ops;
832 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
834 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
836 struct nfs_pgio_mirror *mirror;
837 struct nfs_pgio_header *hdr;
840 mirror = nfs_pgio_current_mirror(desc);
842 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
844 /* TODO: make sure this is right with mirroring - or
845 * should it back out all mirrors? */
846 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
849 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
850 ret = nfs_generic_pgio(desc, hdr);
852 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
855 NFS_PROTO(hdr->inode),
856 desc->pg_rpc_callops,
857 desc->pg_ioflags, 0);
862 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
863 * by calling the pg_get_mirror_count op
865 static int nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
866 struct nfs_page *req)
868 int mirror_count = 1;
870 if (!pgio->pg_ops->pg_get_mirror_count)
873 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
875 if (pgio->pg_error < 0)
876 return pgio->pg_error;
878 if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX)
881 if (WARN_ON_ONCE(!pgio->pg_mirrors_dynamic))
884 pgio->pg_mirror_count = mirror_count;
889 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
891 pgio->pg_mirror_count = 1;
892 pgio->pg_mirror_idx = 0;
893 pgio->pg_mirrors = pgio->pg_mirrors_static;
894 kfree(pgio->pg_mirrors_dynamic);
895 pgio->pg_mirrors_dynamic = NULL;
898 static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
899 const struct nfs_open_context *ctx2)
901 return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
904 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
905 const struct nfs_lock_context *l2)
907 return l1->lockowner.l_owner == l2->lockowner.l_owner
908 && l1->lockowner.l_pid == l2->lockowner.l_pid;
912 * nfs_can_coalesce_requests - test two requests for compatibility
913 * @prev: pointer to nfs_page
914 * @req: pointer to nfs_page
916 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
917 * page data area they describe is contiguous, and that their RPC
918 * credentials, NFSv4 open state, and lockowners are the same.
920 * Return 'true' if this is the case, else return 'false'.
922 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
923 struct nfs_page *req,
924 struct nfs_pageio_descriptor *pgio)
927 struct file_lock_context *flctx;
930 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
932 flctx = d_inode(req->wb_context->dentry)->i_flctx;
934 !(list_empty_careful(&flctx->flc_posix) &&
935 list_empty_careful(&flctx->flc_flock)) &&
936 !nfs_match_lock_context(req->wb_lock_context,
937 prev->wb_lock_context))
939 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
941 if (req->wb_page == prev->wb_page) {
942 if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
945 if (req->wb_pgbase != 0 ||
946 prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
950 size = pgio->pg_ops->pg_test(pgio, prev, req);
951 WARN_ON_ONCE(size > req->wb_bytes);
952 if (size && size < req->wb_bytes)
953 req->wb_bytes = size;
958 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
959 * @desc: destination io descriptor
962 * Returns true if the request 'req' was successfully coalesced into the
963 * existing list of pages 'desc'.
965 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
966 struct nfs_page *req)
968 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
970 struct nfs_page *prev = NULL;
972 if (mirror->pg_count != 0) {
973 prev = nfs_list_entry(mirror->pg_list.prev);
975 if (desc->pg_ops->pg_init)
976 desc->pg_ops->pg_init(desc, req);
977 if (desc->pg_error < 0)
979 mirror->pg_base = req->wb_pgbase;
981 if (!nfs_can_coalesce_requests(prev, req, desc))
983 nfs_list_remove_request(req);
984 nfs_list_add_request(req, &mirror->pg_list);
985 mirror->pg_count += req->wb_bytes;
990 * Helper for nfs_pageio_add_request and nfs_pageio_complete
992 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
994 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
996 if (!list_empty(&mirror->pg_list)) {
997 int error = desc->pg_ops->pg_doio(desc);
999 desc->pg_error = error;
1000 if (list_empty(&mirror->pg_list)) {
1001 mirror->pg_bytes_written += mirror->pg_count;
1002 mirror->pg_count = 0;
1003 mirror->pg_base = 0;
1004 mirror->pg_recoalesce = 0;
1010 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1011 * @desc: destination io descriptor
1014 * This may split a request into subrequests which are all part of the
1017 * Returns true if the request 'req' was successfully coalesced into the
1018 * existing list of pages 'desc'.
1020 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1021 struct nfs_page *req)
1023 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1025 struct nfs_page *subreq;
1026 unsigned int bytes_left = 0;
1027 unsigned int offset, pgbase;
1029 nfs_page_group_lock(req, false);
1032 bytes_left = subreq->wb_bytes;
1033 offset = subreq->wb_offset;
1034 pgbase = subreq->wb_pgbase;
1037 if (!nfs_pageio_do_add_request(desc, subreq)) {
1038 /* make sure pg_test call(s) did nothing */
1039 WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
1040 WARN_ON_ONCE(subreq->wb_offset != offset);
1041 WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
1043 nfs_page_group_unlock(req);
1044 desc->pg_moreio = 1;
1045 nfs_pageio_doio(desc);
1046 if (desc->pg_error < 0)
1048 if (mirror->pg_recoalesce)
1050 /* retry add_request for this subreq */
1051 nfs_page_group_lock(req, false);
1055 /* check for buggy pg_test call(s) */
1056 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
1057 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
1058 WARN_ON_ONCE(subreq->wb_bytes == 0);
1060 bytes_left -= subreq->wb_bytes;
1061 offset += subreq->wb_bytes;
1062 pgbase += subreq->wb_bytes;
1065 subreq = nfs_create_request(req->wb_context,
1067 subreq, pgbase, bytes_left);
1070 nfs_lock_request(subreq);
1071 subreq->wb_offset = offset;
1072 subreq->wb_index = req->wb_index;
1074 } while (bytes_left > 0);
1076 nfs_page_group_unlock(req);
1079 desc->pg_error = PTR_ERR(subreq);
1080 nfs_page_group_unlock(req);
1084 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1086 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1090 list_splice_init(&mirror->pg_list, &head);
1091 mirror->pg_count = 0;
1092 mirror->pg_base = 0;
1093 mirror->pg_recoalesce = 0;
1095 while (!list_empty(&head)) {
1096 struct nfs_page *req;
1098 req = list_first_entry(&head, struct nfs_page, wb_list);
1099 if (__nfs_pageio_add_request(desc, req))
1101 if (desc->pg_error < 0) {
1102 list_splice_tail(&head, &mirror->pg_list);
1103 mirror->pg_recoalesce = 1;
1108 } while (mirror->pg_recoalesce);
1112 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1113 struct nfs_page *req)
1118 ret = __nfs_pageio_add_request(desc, req);
1121 if (desc->pg_error < 0)
1123 ret = nfs_do_recoalesce(desc);
1129 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1130 struct nfs_page *req)
1133 unsigned int pgbase, offset, bytes;
1134 struct nfs_page *dupreq, *lastreq;
1136 pgbase = req->wb_pgbase;
1137 offset = req->wb_offset;
1138 bytes = req->wb_bytes;
1140 nfs_pageio_setup_mirroring(desc, req);
1141 if (desc->pg_error < 0)
1144 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1146 nfs_page_group_lock(req, false);
1148 /* find the last request */
1149 for (lastreq = req->wb_head;
1150 lastreq->wb_this_page != req->wb_head;
1151 lastreq = lastreq->wb_this_page)
1154 dupreq = nfs_create_request(req->wb_context,
1155 req->wb_page, lastreq, pgbase, bytes);
1157 if (IS_ERR(dupreq)) {
1158 nfs_page_group_unlock(req);
1162 nfs_lock_request(dupreq);
1163 nfs_page_group_unlock(req);
1164 dupreq->wb_offset = offset;
1165 dupreq->wb_index = req->wb_index;
1169 if (nfs_pgio_has_mirroring(desc))
1170 desc->pg_mirror_idx = midx;
1171 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1179 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1180 * nfs_pageio_descriptor
1181 * @desc: pointer to io descriptor
1182 * @mirror_idx: pointer to mirror index
1184 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1187 struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx];
1188 u32 restore_idx = desc->pg_mirror_idx;
1190 if (nfs_pgio_has_mirroring(desc))
1191 desc->pg_mirror_idx = mirror_idx;
1193 nfs_pageio_doio(desc);
1194 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1196 if (!nfs_do_recoalesce(desc))
1199 desc->pg_mirror_idx = restore_idx;
1203 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1204 * @hdr - the pgio header to move request from
1205 * @desc - the pageio descriptor to add requests to
1207 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1210 * Returns 0 on success and < 0 on error.
1212 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1213 struct nfs_pgio_header *hdr)
1217 desc->pg_dreq = hdr->dreq;
1218 while (!list_empty(&hdr->pages)) {
1219 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
1221 nfs_list_remove_request(req);
1222 if (!nfs_pageio_add_request(desc, req))
1223 nfs_list_add_request(req, &failed);
1225 nfs_pageio_complete(desc);
1226 if (!list_empty(&failed)) {
1227 list_move(&failed, &hdr->pages);
1228 return desc->pg_error < 0 ? desc->pg_error : -EIO;
1232 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1235 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1236 * @desc: pointer to io descriptor
1238 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1242 for (midx = 0; midx < desc->pg_mirror_count; midx++)
1243 nfs_pageio_complete_mirror(desc, midx);
1245 if (desc->pg_ops->pg_cleanup)
1246 desc->pg_ops->pg_cleanup(desc);
1247 nfs_pageio_cleanup_mirroring(desc);
1251 * nfs_pageio_cond_complete - Conditional I/O completion
1252 * @desc: pointer to io descriptor
1253 * @index: page index
1255 * It is important to ensure that processes don't try to take locks
1256 * on non-contiguous ranges of pages as that might deadlock. This
1257 * function should be called before attempting to wait on a locked
1258 * nfs_page. It will complete the I/O if the page index 'index'
1259 * is not contiguous with the existing list of pages in 'desc'.
1261 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1263 struct nfs_pgio_mirror *mirror;
1264 struct nfs_page *prev;
1267 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1268 mirror = &desc->pg_mirrors[midx];
1269 if (!list_empty(&mirror->pg_list)) {
1270 prev = nfs_list_entry(mirror->pg_list.prev);
1271 if (index != prev->wb_index + 1) {
1272 nfs_pageio_complete(desc);
1280 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1282 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1284 nfs_pageio_complete(pgio);
1287 int __init nfs_init_nfspagecache(void)
1289 nfs_page_cachep = kmem_cache_create("nfs_page",
1290 sizeof(struct nfs_page),
1291 0, SLAB_HWCACHE_ALIGN,
1293 if (nfs_page_cachep == NULL)
1299 void nfs_destroy_nfspagecache(void)
1301 kmem_cache_destroy(nfs_page_cachep);
1304 static const struct rpc_call_ops nfs_pgio_common_ops = {
1305 .rpc_call_prepare = nfs_pgio_prepare,
1306 .rpc_call_done = nfs_pgio_result,
1307 .rpc_release = nfs_pgio_release,
1310 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1311 .pg_test = nfs_generic_pg_test,
1312 .pg_doio = nfs_generic_pg_pgios,