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 static struct kmem_cache *nfs_page_cachep;
29 bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
31 p->npages = pagecount;
32 if (pagecount <= ARRAY_SIZE(p->page_array))
33 p->pagevec = p->page_array;
35 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
39 return p->pagevec != NULL;
42 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
43 struct nfs_pgio_header *hdr,
44 void (*release)(struct nfs_pgio_header *hdr))
46 hdr->req = nfs_list_entry(desc->pg_list.next);
47 hdr->inode = desc->pg_inode;
48 hdr->cred = hdr->req->wb_context->cred;
49 hdr->io_start = req_offset(hdr->req);
50 hdr->good_bytes = desc->pg_count;
51 hdr->dreq = desc->pg_dreq;
52 hdr->release = release;
53 hdr->completion_ops = desc->pg_completion_ops;
54 if (hdr->completion_ops->init_hdr)
55 hdr->completion_ops->init_hdr(hdr);
57 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
59 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
61 spin_lock(&hdr->lock);
62 if (pos < hdr->io_start + hdr->good_bytes) {
63 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
64 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
65 hdr->good_bytes = pos - hdr->io_start;
68 spin_unlock(&hdr->lock);
71 static inline struct nfs_page *
74 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
76 INIT_LIST_HEAD(&p->wb_list);
81 nfs_page_free(struct nfs_page *p)
83 kmem_cache_free(nfs_page_cachep, p);
87 * nfs_create_request - Create an NFS read/write request.
88 * @ctx: open context to use
89 * @inode: inode to which the request is attached
90 * @page: page to write
91 * @offset: starting offset within the page for the write
92 * @count: number of bytes to read/write
94 * The page must be locked by the caller. This makes sure we never
95 * create two different requests for the same page.
96 * User should ensure it is safe to sleep in this function.
99 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
101 unsigned int offset, unsigned int count)
103 struct nfs_page *req;
105 /* try to allocate the request struct */
106 req = nfs_page_alloc();
108 return ERR_PTR(-ENOMEM);
110 /* get lock context early so we can deal with alloc failures */
111 req->wb_lock_context = nfs_get_lock_context(ctx);
112 if (req->wb_lock_context == NULL) {
114 return ERR_PTR(-ENOMEM);
117 /* Initialize the request struct. Initially, we assume a
118 * long write-back delay. This will be adjusted in
119 * update_nfs_request below if the region is not locked. */
121 req->wb_index = page_file_index(page);
122 page_cache_get(page);
123 req->wb_offset = offset;
124 req->wb_pgbase = offset;
125 req->wb_bytes = count;
126 req->wb_context = get_nfs_open_context(ctx);
127 kref_init(&req->wb_kref);
132 * nfs_unlock_request - Unlock request and wake up sleepers.
135 void nfs_unlock_request(struct nfs_page *req)
137 if (!NFS_WBACK_BUSY(req)) {
138 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
141 smp_mb__before_clear_bit();
142 clear_bit(PG_BUSY, &req->wb_flags);
143 smp_mb__after_clear_bit();
144 wake_up_bit(&req->wb_flags, PG_BUSY);
148 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
151 void nfs_unlock_and_release_request(struct nfs_page *req)
153 nfs_unlock_request(req);
154 nfs_release_request(req);
158 * nfs_clear_request - Free up all resources allocated to the request
161 * Release page and open context resources associated with a read/write
162 * request after it has completed.
164 static void nfs_clear_request(struct nfs_page *req)
166 struct page *page = req->wb_page;
167 struct nfs_open_context *ctx = req->wb_context;
168 struct nfs_lock_context *l_ctx = req->wb_lock_context;
171 page_cache_release(page);
175 nfs_put_lock_context(l_ctx);
176 req->wb_lock_context = NULL;
179 put_nfs_open_context(ctx);
180 req->wb_context = NULL;
186 * nfs_release_request - Release the count on an NFS read/write request
187 * @req: request to release
189 * Note: Should never be called with the spinlock held!
191 static void nfs_free_request(struct kref *kref)
193 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
195 /* Release struct file and open context */
196 nfs_clear_request(req);
200 void nfs_release_request(struct nfs_page *req)
202 kref_put(&req->wb_kref, nfs_free_request);
205 static int nfs_wait_bit_uninterruptible(void *word)
212 * nfs_wait_on_request - Wait for a request to complete.
213 * @req: request to wait upon.
215 * Interruptible by fatal signals only.
216 * The user is responsible for holding a count on the request.
219 nfs_wait_on_request(struct nfs_page *req)
221 return wait_on_bit(&req->wb_flags, PG_BUSY,
222 nfs_wait_bit_uninterruptible,
223 TASK_UNINTERRUPTIBLE);
226 bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
229 * FIXME: ideally we should be able to coalesce all requests
230 * that are not block boundary aligned, but currently this
231 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
232 * since nfs_flush_multi and nfs_pagein_multi assume you
233 * can have only one struct nfs_page.
235 if (desc->pg_bsize < PAGE_SIZE)
238 return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
240 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
243 * nfs_pageio_init - initialise a page io descriptor
244 * @desc: pointer to descriptor
245 * @inode: pointer to inode
246 * @doio: pointer to io function
247 * @bsize: io block size
248 * @io_flags: extra parameters for the io function
250 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
252 const struct nfs_pageio_ops *pg_ops,
253 const struct nfs_pgio_completion_ops *compl_ops,
257 INIT_LIST_HEAD(&desc->pg_list);
258 desc->pg_bytes_written = 0;
260 desc->pg_bsize = bsize;
263 desc->pg_recoalesce = 0;
264 desc->pg_inode = inode;
265 desc->pg_ops = pg_ops;
266 desc->pg_completion_ops = compl_ops;
267 desc->pg_ioflags = io_flags;
269 desc->pg_lseg = NULL;
270 desc->pg_dreq = NULL;
272 EXPORT_SYMBOL_GPL(nfs_pageio_init);
275 * nfs_can_coalesce_requests - test two requests for compatibility
276 * @prev: pointer to nfs_page
277 * @req: pointer to nfs_page
279 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
280 * page data area they describe is contiguous, and that their RPC
281 * credentials, NFSv4 open state, and lockowners are the same.
283 * Return 'true' if this is the case, else return 'false'.
285 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
286 struct nfs_page *req,
287 struct nfs_pageio_descriptor *pgio)
289 if (req->wb_context->cred != prev->wb_context->cred)
291 if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner)
293 if (req->wb_context->state != prev->wb_context->state)
295 if (req->wb_pgbase != 0)
297 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
299 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
301 return pgio->pg_ops->pg_test(pgio, prev, req);
305 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
306 * @desc: destination io descriptor
309 * Returns true if the request 'req' was successfully coalesced into the
310 * existing list of pages 'desc'.
312 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
313 struct nfs_page *req)
315 if (desc->pg_count != 0) {
316 struct nfs_page *prev;
318 prev = nfs_list_entry(desc->pg_list.prev);
319 if (!nfs_can_coalesce_requests(prev, req, desc))
322 if (desc->pg_ops->pg_init)
323 desc->pg_ops->pg_init(desc, req);
324 desc->pg_base = req->wb_pgbase;
326 nfs_list_remove_request(req);
327 nfs_list_add_request(req, &desc->pg_list);
328 desc->pg_count += req->wb_bytes;
333 * Helper for nfs_pageio_add_request and nfs_pageio_complete
335 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
337 if (!list_empty(&desc->pg_list)) {
338 int error = desc->pg_ops->pg_doio(desc);
340 desc->pg_error = error;
342 desc->pg_bytes_written += desc->pg_count;
344 if (list_empty(&desc->pg_list)) {
351 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
352 * @desc: destination io descriptor
355 * Returns true if the request 'req' was successfully coalesced into the
356 * existing list of pages 'desc'.
358 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
359 struct nfs_page *req)
361 while (!nfs_pageio_do_add_request(desc, req)) {
363 nfs_pageio_doio(desc);
364 if (desc->pg_error < 0)
367 if (desc->pg_recoalesce)
373 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
378 list_splice_init(&desc->pg_list, &head);
379 desc->pg_bytes_written -= desc->pg_count;
382 desc->pg_recoalesce = 0;
384 while (!list_empty(&head)) {
385 struct nfs_page *req;
387 req = list_first_entry(&head, struct nfs_page, wb_list);
388 nfs_list_remove_request(req);
389 if (__nfs_pageio_add_request(desc, req))
391 if (desc->pg_error < 0)
395 } while (desc->pg_recoalesce);
399 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
400 struct nfs_page *req)
405 ret = __nfs_pageio_add_request(desc, req);
408 if (desc->pg_error < 0)
410 ret = nfs_do_recoalesce(desc);
414 EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
417 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
418 * @desc: pointer to io descriptor
420 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
423 nfs_pageio_doio(desc);
424 if (!desc->pg_recoalesce)
426 if (!nfs_do_recoalesce(desc))
430 EXPORT_SYMBOL_GPL(nfs_pageio_complete);
433 * nfs_pageio_cond_complete - Conditional I/O completion
434 * @desc: pointer to io descriptor
437 * It is important to ensure that processes don't try to take locks
438 * on non-contiguous ranges of pages as that might deadlock. This
439 * function should be called before attempting to wait on a locked
440 * nfs_page. It will complete the I/O if the page index 'index'
441 * is not contiguous with the existing list of pages in 'desc'.
443 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
445 if (!list_empty(&desc->pg_list)) {
446 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
447 if (index != prev->wb_index + 1)
448 nfs_pageio_complete(desc);
452 int __init nfs_init_nfspagecache(void)
454 nfs_page_cachep = kmem_cache_create("nfs_page",
455 sizeof(struct nfs_page),
456 0, SLAB_HWCACHE_ALIGN,
458 if (nfs_page_cachep == NULL)
464 void nfs_destroy_nfspagecache(void)
466 kmem_cache_destroy(nfs_page_cachep);