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);
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_unlock - unlock the head of the page group
172 * @req - request in group that is to be unlocked
175 nfs_page_group_unlock(struct nfs_page *req)
177 struct nfs_page *head = req->wb_head;
179 WARN_ON_ONCE(head != head->wb_head);
181 smp_mb__before_atomic();
182 clear_bit(PG_HEADLOCK, &head->wb_flags);
183 smp_mb__after_atomic();
184 wake_up_bit(&head->wb_flags, PG_HEADLOCK);
188 * nfs_page_group_sync_on_bit_locked
190 * must be called with page group lock held
193 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
195 struct nfs_page *head = req->wb_head;
196 struct nfs_page *tmp;
198 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
199 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
201 tmp = req->wb_this_page;
203 if (!test_bit(bit, &tmp->wb_flags))
205 tmp = tmp->wb_this_page;
208 /* true! reset all bits */
211 clear_bit(bit, &tmp->wb_flags);
212 tmp = tmp->wb_this_page;
213 } while (tmp != req);
219 * nfs_page_group_sync_on_bit - set bit on current request, but only
220 * return true if the bit is set for all requests in page group
221 * @req - request in page group
222 * @bit - PG_* bit that is used to sync page group
224 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
228 nfs_page_group_lock(req, false);
229 ret = nfs_page_group_sync_on_bit_locked(req, bit);
230 nfs_page_group_unlock(req);
236 * nfs_page_group_init - Initialize the page group linkage for @req
237 * @req - a new nfs request
238 * @prev - the previous request in page group, or NULL if @req is the first
239 * or only request in the group (the head).
242 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
244 WARN_ON_ONCE(prev == req);
249 req->wb_this_page = req;
252 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
253 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
254 req->wb_head = prev->wb_head;
255 req->wb_this_page = prev->wb_this_page;
256 prev->wb_this_page = req;
258 /* All subrequests take a ref on the head request until
259 * nfs_page_group_destroy is called */
260 kref_get(&req->wb_head->wb_kref);
262 /* grab extra ref if head request has extra ref from
263 * the write/commit path to handle handoff between write
264 * and commit lists */
265 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
266 set_bit(PG_INODE_REF, &req->wb_flags);
267 kref_get(&req->wb_kref);
273 * nfs_page_group_destroy - sync the destruction of page groups
274 * @req - request that no longer needs the page group
276 * releases the page group reference from each member once all
277 * members have called this function.
280 nfs_page_group_destroy(struct kref *kref)
282 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
283 struct nfs_page *tmp, *next;
285 /* subrequests must release the ref on the head request */
286 if (req->wb_head != req)
287 nfs_release_request(req->wb_head);
289 if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
294 next = tmp->wb_this_page;
295 /* unlink and free */
296 tmp->wb_this_page = tmp;
298 nfs_free_request(tmp);
300 } while (tmp != req);
304 * nfs_create_request - Create an NFS read/write request.
305 * @ctx: open context to use
306 * @page: page to write
307 * @last: last nfs request created for this page group or NULL if head
308 * @offset: starting offset within the page for the write
309 * @count: number of bytes to read/write
311 * The page must be locked by the caller. This makes sure we never
312 * create two different requests for the same page.
313 * User should ensure it is safe to sleep in this function.
316 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
317 struct nfs_page *last, unsigned int offset,
320 struct nfs_page *req;
321 struct nfs_lock_context *l_ctx;
323 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
324 return ERR_PTR(-EBADF);
325 /* try to allocate the request struct */
326 req = nfs_page_alloc();
328 return ERR_PTR(-ENOMEM);
330 /* get lock context early so we can deal with alloc failures */
331 l_ctx = nfs_get_lock_context(ctx);
334 return ERR_CAST(l_ctx);
336 req->wb_lock_context = l_ctx;
337 nfs_iocounter_inc(&l_ctx->io_count);
339 /* Initialize the request struct. Initially, we assume a
340 * long write-back delay. This will be adjusted in
341 * update_nfs_request below if the region is not locked. */
343 req->wb_index = page_file_index(page);
344 page_cache_get(page);
345 req->wb_offset = offset;
346 req->wb_pgbase = offset;
347 req->wb_bytes = count;
348 req->wb_context = get_nfs_open_context(ctx);
349 kref_init(&req->wb_kref);
350 nfs_page_group_init(req, last);
355 * nfs_unlock_request - Unlock request and wake up sleepers.
358 void nfs_unlock_request(struct nfs_page *req)
360 if (!NFS_WBACK_BUSY(req)) {
361 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
364 smp_mb__before_atomic();
365 clear_bit(PG_BUSY, &req->wb_flags);
366 smp_mb__after_atomic();
367 wake_up_bit(&req->wb_flags, PG_BUSY);
371 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
374 void nfs_unlock_and_release_request(struct nfs_page *req)
376 nfs_unlock_request(req);
377 nfs_release_request(req);
381 * nfs_clear_request - Free up all resources allocated to the request
384 * Release page and open context resources associated with a read/write
385 * request after it has completed.
387 static void nfs_clear_request(struct nfs_page *req)
389 struct page *page = req->wb_page;
390 struct nfs_open_context *ctx = req->wb_context;
391 struct nfs_lock_context *l_ctx = req->wb_lock_context;
394 page_cache_release(page);
398 nfs_iocounter_dec(&l_ctx->io_count);
399 nfs_put_lock_context(l_ctx);
400 req->wb_lock_context = NULL;
403 put_nfs_open_context(ctx);
404 req->wb_context = NULL;
409 * nfs_release_request - Release the count on an NFS read/write request
410 * @req: request to release
412 * Note: Should never be called with the spinlock held!
414 void nfs_free_request(struct nfs_page *req)
416 WARN_ON_ONCE(req->wb_this_page != req);
418 /* extra debug: make sure no sync bits are still set */
419 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
420 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
421 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
422 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
423 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
425 /* Release struct file and open context */
426 nfs_clear_request(req);
430 void nfs_release_request(struct nfs_page *req)
432 kref_put(&req->wb_kref, nfs_page_group_destroy);
436 * nfs_wait_on_request - Wait for a request to complete.
437 * @req: request to wait upon.
439 * Interruptible by fatal signals only.
440 * The user is responsible for holding a count on the request.
443 nfs_wait_on_request(struct nfs_page *req)
445 return wait_on_bit_io(&req->wb_flags, PG_BUSY,
446 TASK_UNINTERRUPTIBLE);
450 * nfs_generic_pg_test - determine if requests can be coalesced
451 * @desc: pointer to descriptor
452 * @prev: previous request in desc, or NULL
455 * Returns zero if @req can be coalesced into @desc, otherwise it returns
456 * the size of the request.
458 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
459 struct nfs_page *prev, struct nfs_page *req)
461 if (desc->pg_count > desc->pg_bsize) {
462 /* should never happen */
467 return min(desc->pg_bsize - desc->pg_count, (size_t)req->wb_bytes);
469 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
471 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
473 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
476 INIT_LIST_HEAD(&hdr->pages);
477 spin_lock_init(&hdr->lock);
482 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
485 * nfs_pgio_header_free - Free a read or write header
486 * @hdr: The header to free
488 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
490 hdr->rw_ops->rw_free_header(hdr);
492 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
495 * nfs_pgio_data_destroy - make @hdr suitable for reuse
497 * Frees memory and releases refs from nfs_generic_pgio, so that it may
500 * @hdr: A header that has had nfs_generic_pgio called
502 void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
504 put_nfs_open_context(hdr->args.context);
505 if (hdr->page_array.pagevec != hdr->page_array.page_array)
506 kfree(hdr->page_array.pagevec);
508 EXPORT_SYMBOL_GPL(nfs_pgio_data_destroy);
511 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
512 * @hdr: The pageio hdr
513 * @count: Number of bytes to read
514 * @offset: Initial offset
515 * @how: How to commit data (writes only)
516 * @cinfo: Commit information for the call (writes only)
518 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
519 unsigned int count, unsigned int offset,
520 int how, struct nfs_commit_info *cinfo)
522 struct nfs_page *req = hdr->req;
524 /* Set up the RPC argument and reply structs
525 * NB: take care not to mess about with hdr->commit et al. */
527 hdr->args.fh = NFS_FH(hdr->inode);
528 hdr->args.offset = req_offset(req) + offset;
529 /* pnfs_set_layoutcommit needs this */
530 hdr->mds_offset = hdr->args.offset;
531 hdr->args.pgbase = req->wb_pgbase + offset;
532 hdr->args.pages = hdr->page_array.pagevec;
533 hdr->args.count = count;
534 hdr->args.context = get_nfs_open_context(req->wb_context);
535 hdr->args.lock_context = req->wb_lock_context;
536 hdr->args.stable = NFS_UNSTABLE;
537 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
540 case FLUSH_COND_STABLE:
541 if (nfs_reqs_to_commit(cinfo))
544 hdr->args.stable = NFS_FILE_SYNC;
547 hdr->res.fattr = &hdr->fattr;
548 hdr->res.count = count;
550 hdr->res.verf = &hdr->verf;
551 nfs_fattr_init(&hdr->fattr);
555 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
556 * @task: The current task
557 * @calldata: pageio header to prepare
559 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
561 struct nfs_pgio_header *hdr = calldata;
563 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
568 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
569 const struct rpc_call_ops *call_ops, int how, int flags)
571 struct rpc_task *task;
572 struct rpc_message msg = {
573 .rpc_argp = &hdr->args,
574 .rpc_resp = &hdr->res,
575 .rpc_cred = hdr->cred,
577 struct rpc_task_setup task_setup_data = {
581 .callback_ops = call_ops,
582 .callback_data = hdr,
583 .workqueue = nfsiod_workqueue,
584 .flags = RPC_TASK_ASYNC | flags,
588 hdr->rw_ops->rw_initiate(hdr, &msg, &task_setup_data, how);
590 dprintk("NFS: %5u initiated pgio call "
591 "(req %s/%llu, %u bytes @ offset %llu)\n",
593 hdr->inode->i_sb->s_id,
594 (unsigned long long)NFS_FILEID(hdr->inode),
596 (unsigned long long)hdr->args.offset);
598 task = rpc_run_task(&task_setup_data);
603 if (how & FLUSH_SYNC) {
604 ret = rpc_wait_for_completion_task(task);
606 ret = task->tk_status;
612 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
615 * nfs_pgio_error - Clean up from a pageio error
616 * @desc: IO descriptor
617 * @hdr: pageio header
619 static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
620 struct nfs_pgio_header *hdr)
622 set_bit(NFS_IOHDR_REDO, &hdr->flags);
623 nfs_pgio_data_destroy(hdr);
624 hdr->completion_ops->completion(hdr);
625 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
630 * nfs_pgio_release - Release pageio data
631 * @calldata: The pageio header to release
633 static void nfs_pgio_release(void *calldata)
635 struct nfs_pgio_header *hdr = calldata;
636 if (hdr->rw_ops->rw_release)
637 hdr->rw_ops->rw_release(hdr);
638 nfs_pgio_data_destroy(hdr);
639 hdr->completion_ops->completion(hdr);
643 * nfs_pageio_init - initialise a page io descriptor
644 * @desc: pointer to descriptor
645 * @inode: pointer to inode
646 * @doio: pointer to io function
647 * @bsize: io block size
648 * @io_flags: extra parameters for the io function
650 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
652 const struct nfs_pageio_ops *pg_ops,
653 const struct nfs_pgio_completion_ops *compl_ops,
654 const struct nfs_rw_ops *rw_ops,
658 INIT_LIST_HEAD(&desc->pg_list);
659 desc->pg_bytes_written = 0;
661 desc->pg_bsize = bsize;
664 desc->pg_recoalesce = 0;
665 desc->pg_inode = inode;
666 desc->pg_ops = pg_ops;
667 desc->pg_completion_ops = compl_ops;
668 desc->pg_rw_ops = rw_ops;
669 desc->pg_ioflags = io_flags;
671 desc->pg_lseg = NULL;
672 desc->pg_dreq = NULL;
673 desc->pg_layout_private = NULL;
675 EXPORT_SYMBOL_GPL(nfs_pageio_init);
678 * nfs_pgio_result - Basic pageio error handling
679 * @task: The task that ran
680 * @calldata: Pageio header to check
682 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
684 struct nfs_pgio_header *hdr = calldata;
685 struct inode *inode = hdr->inode;
687 dprintk("NFS: %s: %5u, (status %d)\n", __func__,
688 task->tk_pid, task->tk_status);
690 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
692 if (task->tk_status < 0)
693 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
695 hdr->rw_ops->rw_result(task, hdr);
699 * Create an RPC task for the given read or write request and kick it.
700 * The page must have been locked by the caller.
702 * It may happen that the page we're passed is not marked dirty.
703 * This is the case if nfs_updatepage detects a conflicting request
704 * that has been written but not committed.
706 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
707 struct nfs_pgio_header *hdr)
709 struct nfs_page *req;
711 struct list_head *head = &desc->pg_list;
712 struct nfs_commit_info cinfo;
713 unsigned int pagecount;
715 pagecount = nfs_page_array_len(desc->pg_base, desc->pg_count);
716 if (!nfs_pgarray_set(&hdr->page_array, pagecount))
717 return nfs_pgio_error(desc, hdr);
719 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
720 pages = hdr->page_array.pagevec;
721 while (!list_empty(head)) {
722 req = nfs_list_entry(head->next);
723 nfs_list_remove_request(req);
724 nfs_list_add_request(req, &hdr->pages);
725 *pages++ = req->wb_page;
728 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
729 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
730 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
732 /* Set up the argument struct */
733 nfs_pgio_rpcsetup(hdr, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
734 desc->pg_rpc_callops = &nfs_pgio_common_ops;
737 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
739 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
741 struct nfs_pgio_header *hdr;
744 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
746 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
749 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
750 ret = nfs_generic_pgio(desc, hdr);
752 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
753 hdr, desc->pg_rpc_callops,
754 desc->pg_ioflags, 0);
758 static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
759 const struct nfs_open_context *ctx2)
761 return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
764 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
765 const struct nfs_lock_context *l2)
767 return l1->lockowner.l_owner == l2->lockowner.l_owner
768 && l1->lockowner.l_pid == l2->lockowner.l_pid;
772 * nfs_can_coalesce_requests - test two requests for compatibility
773 * @prev: pointer to nfs_page
774 * @req: pointer to nfs_page
776 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
777 * page data area they describe is contiguous, and that their RPC
778 * credentials, NFSv4 open state, and lockowners are the same.
780 * Return 'true' if this is the case, else return 'false'.
782 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
783 struct nfs_page *req,
784 struct nfs_pageio_descriptor *pgio)
789 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
791 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
792 !nfs_match_lock_context(req->wb_lock_context,
793 prev->wb_lock_context))
795 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
798 size = pgio->pg_ops->pg_test(pgio, prev, req);
799 WARN_ON_ONCE(size > req->wb_bytes);
800 if (size && size < req->wb_bytes)
801 req->wb_bytes = size;
806 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
807 * @desc: destination io descriptor
810 * Returns true if the request 'req' was successfully coalesced into the
811 * existing list of pages 'desc'.
813 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
814 struct nfs_page *req)
816 struct nfs_page *prev = NULL;
817 if (desc->pg_count != 0) {
818 prev = nfs_list_entry(desc->pg_list.prev);
820 if (desc->pg_ops->pg_init)
821 desc->pg_ops->pg_init(desc, req);
822 desc->pg_base = req->wb_pgbase;
824 if (!nfs_can_coalesce_requests(prev, req, desc))
826 nfs_list_remove_request(req);
827 nfs_list_add_request(req, &desc->pg_list);
828 desc->pg_count += req->wb_bytes;
833 * Helper for nfs_pageio_add_request and nfs_pageio_complete
835 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
837 if (!list_empty(&desc->pg_list)) {
838 int error = desc->pg_ops->pg_doio(desc);
840 desc->pg_error = error;
842 desc->pg_bytes_written += desc->pg_count;
844 if (list_empty(&desc->pg_list)) {
851 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
852 * @desc: destination io descriptor
855 * This may split a request into subrequests which are all part of the
858 * Returns true if the request 'req' was successfully coalesced into the
859 * existing list of pages 'desc'.
861 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
862 struct nfs_page *req)
864 struct nfs_page *subreq;
865 unsigned int bytes_left = 0;
866 unsigned int offset, pgbase;
868 nfs_page_group_lock(req, false);
871 bytes_left = subreq->wb_bytes;
872 offset = subreq->wb_offset;
873 pgbase = subreq->wb_pgbase;
876 if (!nfs_pageio_do_add_request(desc, subreq)) {
877 /* make sure pg_test call(s) did nothing */
878 WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
879 WARN_ON_ONCE(subreq->wb_offset != offset);
880 WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
882 nfs_page_group_unlock(req);
884 nfs_pageio_doio(desc);
885 if (desc->pg_error < 0)
887 if (desc->pg_recoalesce)
889 /* retry add_request for this subreq */
890 nfs_page_group_lock(req, false);
894 /* check for buggy pg_test call(s) */
895 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
896 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
897 WARN_ON_ONCE(subreq->wb_bytes == 0);
899 bytes_left -= subreq->wb_bytes;
900 offset += subreq->wb_bytes;
901 pgbase += subreq->wb_bytes;
904 subreq = nfs_create_request(req->wb_context,
906 subreq, pgbase, bytes_left);
909 nfs_lock_request(subreq);
910 subreq->wb_offset = offset;
911 subreq->wb_index = req->wb_index;
913 } while (bytes_left > 0);
915 nfs_page_group_unlock(req);
918 desc->pg_error = PTR_ERR(subreq);
919 nfs_page_group_unlock(req);
923 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
928 list_splice_init(&desc->pg_list, &head);
929 desc->pg_bytes_written -= desc->pg_count;
932 desc->pg_recoalesce = 0;
935 while (!list_empty(&head)) {
936 struct nfs_page *req;
938 req = list_first_entry(&head, struct nfs_page, wb_list);
939 nfs_list_remove_request(req);
940 if (__nfs_pageio_add_request(desc, req))
942 if (desc->pg_error < 0)
946 } while (desc->pg_recoalesce);
950 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
951 struct nfs_page *req)
956 ret = __nfs_pageio_add_request(desc, req);
959 if (desc->pg_error < 0)
961 ret = nfs_do_recoalesce(desc);
967 * nfs_pageio_resend - Transfer requests to new descriptor and resend
968 * @hdr - the pgio header to move request from
969 * @desc - the pageio descriptor to add requests to
971 * Try to move each request (nfs_page) from @hdr to @desc then attempt
974 * Returns 0 on success and < 0 on error.
976 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
977 struct nfs_pgio_header *hdr)
981 desc->pg_dreq = hdr->dreq;
982 while (!list_empty(&hdr->pages)) {
983 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
985 nfs_list_remove_request(req);
986 if (!nfs_pageio_add_request(desc, req))
987 nfs_list_add_request(req, &failed);
989 nfs_pageio_complete(desc);
990 if (!list_empty(&failed)) {
991 list_move(&failed, &hdr->pages);
996 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
999 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
1000 * @desc: pointer to io descriptor
1002 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1005 nfs_pageio_doio(desc);
1006 if (!desc->pg_recoalesce)
1008 if (!nfs_do_recoalesce(desc))
1014 * nfs_pageio_cond_complete - Conditional I/O completion
1015 * @desc: pointer to io descriptor
1016 * @index: page index
1018 * It is important to ensure that processes don't try to take locks
1019 * on non-contiguous ranges of pages as that might deadlock. This
1020 * function should be called before attempting to wait on a locked
1021 * nfs_page. It will complete the I/O if the page index 'index'
1022 * is not contiguous with the existing list of pages in 'desc'.
1024 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1026 if (!list_empty(&desc->pg_list)) {
1027 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
1028 if (index != prev->wb_index + 1)
1029 nfs_pageio_complete(desc);
1033 int __init nfs_init_nfspagecache(void)
1035 nfs_page_cachep = kmem_cache_create("nfs_page",
1036 sizeof(struct nfs_page),
1037 0, SLAB_HWCACHE_ALIGN,
1039 if (nfs_page_cachep == NULL)
1045 void nfs_destroy_nfspagecache(void)
1047 kmem_cache_destroy(nfs_page_cachep);
1050 static const struct rpc_call_ops nfs_pgio_common_ops = {
1051 .rpc_call_prepare = nfs_pgio_prepare,
1052 .rpc_call_done = nfs_pgio_result,
1053 .rpc_release = nfs_pgio_release,
1056 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1057 .pg_test = nfs_generic_pg_test,
1058 .pg_doio = nfs_generic_pg_pgios,