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/nfs3.h>
17 #include <linux/nfs4.h>
18 #include <linux/nfs_page.h>
19 #include <linux/nfs_fs.h>
20 #include <linux/nfs_mount.h>
24 static struct kmem_cache *nfs_page_cachep;
26 static inline struct nfs_page *
30 p = kmem_cache_alloc(nfs_page_cachep, GFP_KERNEL);
32 memset(p, 0, sizeof(*p));
33 INIT_LIST_HEAD(&p->wb_list);
39 nfs_page_free(struct nfs_page *p)
41 kmem_cache_free(nfs_page_cachep, p);
45 * nfs_create_request - Create an NFS read/write request.
46 * @file: file descriptor to use
47 * @inode: inode to which the request is attached
48 * @page: page to write
49 * @offset: starting offset within the page for the write
50 * @count: number of bytes to read/write
52 * The page must be locked by the caller. This makes sure we never
53 * create two different requests for the same page.
54 * User should ensure it is safe to sleep in this function.
57 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
59 unsigned int offset, unsigned int count)
61 struct nfs_server *server = NFS_SERVER(inode);
65 /* try to allocate the request struct */
66 req = nfs_page_alloc();
70 if (signalled() && (server->flags & NFS_MOUNT_INTR))
71 return ERR_PTR(-ERESTARTSYS);
75 /* Initialize the request struct. Initially, we assume a
76 * long write-back delay. This will be adjusted in
77 * update_nfs_request below if the region is not locked. */
79 atomic_set(&req->wb_complete, 0);
80 req->wb_index = page->index;
82 BUG_ON(PagePrivate(page));
83 BUG_ON(!PageLocked(page));
84 BUG_ON(page->mapping->host != inode);
85 req->wb_offset = offset;
86 req->wb_pgbase = offset;
87 req->wb_bytes = count;
88 atomic_set(&req->wb_count, 1);
89 req->wb_context = get_nfs_open_context(ctx);
95 * nfs_unlock_request - Unlock request and wake up sleepers.
98 void nfs_unlock_request(struct nfs_page *req)
100 if (!NFS_WBACK_BUSY(req)) {
101 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
104 smp_mb__before_clear_bit();
105 clear_bit(PG_BUSY, &req->wb_flags);
106 smp_mb__after_clear_bit();
107 wake_up_bit(&req->wb_flags, PG_BUSY);
108 nfs_release_request(req);
112 * nfs_set_page_writeback_locked - Lock a request for writeback
115 int nfs_set_page_writeback_locked(struct nfs_page *req)
117 struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
119 if (!nfs_lock_request(req))
121 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
126 * nfs_clear_page_writeback - Unlock request and wake up sleepers
128 void nfs_clear_page_writeback(struct nfs_page *req)
130 struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
132 if (req->wb_page != NULL) {
133 spin_lock(&nfsi->req_lock);
134 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
135 spin_unlock(&nfsi->req_lock);
137 nfs_unlock_request(req);
141 * nfs_clear_request - Free up all resources allocated to the request
144 * Release page resources associated with a write request after it
147 void nfs_clear_request(struct nfs_page *req)
149 struct page *page = req->wb_page;
151 page_cache_release(page);
158 * nfs_release_request - Release the count on an NFS read/write request
159 * @req: request to release
161 * Note: Should never be called with the spinlock held!
164 nfs_release_request(struct nfs_page *req)
166 if (!atomic_dec_and_test(&req->wb_count))
169 /* Release struct file or cached credential */
170 nfs_clear_request(req);
171 put_nfs_open_context(req->wb_context);
175 static int nfs_wait_bit_interruptible(void *word)
179 if (signal_pending(current))
187 * nfs_wait_on_request - Wait for a request to complete.
188 * @req: request to wait upon.
190 * Interruptible by signals only if mounted with intr flag.
191 * The user is responsible for holding a count on the request.
194 nfs_wait_on_request(struct nfs_page *req)
196 struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
200 if (!test_bit(PG_BUSY, &req->wb_flags))
203 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
204 * are not interrupted if intr flag is not set
206 rpc_clnt_sigmask(clnt, &oldmask);
207 ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
208 nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
209 rpc_clnt_sigunmask(clnt, &oldmask);
215 * nfs_pageio_init - initialise a page io descriptor
216 * @desc: pointer to descriptor
217 * @inode: pointer to inode
218 * @doio: pointer to io function
219 * @bsize: io block size
220 * @io_flags: extra parameters for the io function
222 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
224 int (*doio)(struct inode *, struct list_head *, unsigned int, size_t, int),
228 INIT_LIST_HEAD(&desc->pg_list);
229 desc->pg_bytes_written = 0;
231 desc->pg_bsize = bsize;
233 desc->pg_inode = inode;
234 desc->pg_doio = doio;
235 desc->pg_ioflags = io_flags;
240 * nfs_can_coalesce_requests - test two requests for compatibility
241 * @prev: pointer to nfs_page
242 * @req: pointer to nfs_page
244 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
245 * page data area they describe is contiguous, and that their RPC
246 * credentials, NFSv4 open state, and lockowners are the same.
248 * Return 'true' if this is the case, else return 'false'.
250 static int nfs_can_coalesce_requests(struct nfs_page *prev,
251 struct nfs_page *req)
253 if (req->wb_context->cred != prev->wb_context->cred)
255 if (req->wb_context->lockowner != prev->wb_context->lockowner)
257 if (req->wb_context->state != prev->wb_context->state)
259 if (req->wb_index != (prev->wb_index + 1))
261 if (req->wb_pgbase != 0)
263 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
269 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
270 * @desc: destination io descriptor
273 * Returns true if the request 'req' was successfully coalesced into the
274 * existing list of pages 'desc'.
276 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
277 struct nfs_page *req)
279 size_t newlen = req->wb_bytes;
281 if (desc->pg_count != 0) {
282 struct nfs_page *prev;
285 * FIXME: ideally we should be able to coalesce all requests
286 * that are not block boundary aligned, but currently this
287 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
288 * since nfs_flush_multi and nfs_pagein_multi assume you
289 * can have only one struct nfs_page.
291 if (desc->pg_bsize < PAGE_SIZE)
293 newlen += desc->pg_count;
294 if (newlen > desc->pg_bsize)
296 prev = nfs_list_entry(desc->pg_list.prev);
297 if (!nfs_can_coalesce_requests(prev, req))
300 desc->pg_base = req->wb_pgbase;
301 nfs_list_remove_request(req);
302 nfs_list_add_request(req, &desc->pg_list);
303 desc->pg_count = newlen;
308 * Helper for nfs_pageio_add_request and nfs_pageio_complete
310 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
312 if (!list_empty(&desc->pg_list)) {
313 int error = desc->pg_doio(desc->pg_inode,
315 nfs_page_array_len(desc->pg_base,
320 desc->pg_error = error;
322 desc->pg_bytes_written += desc->pg_count;
324 if (list_empty(&desc->pg_list)) {
331 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
332 * @desc: destination io descriptor
335 * Returns true if the request 'req' was successfully coalesced into the
336 * existing list of pages 'desc'.
338 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
339 struct nfs_page *req)
341 while (!nfs_pageio_do_add_request(desc, req)) {
342 nfs_pageio_doio(desc);
343 if (desc->pg_error < 0)
350 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
351 * @desc: pointer to io descriptor
353 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
355 nfs_pageio_doio(desc);
359 * nfs_pageio_cond_complete - Conditional I/O completion
360 * @desc: pointer to io descriptor
363 * It is important to ensure that processes don't try to take locks
364 * on non-contiguous ranges of pages as that might deadlock. This
365 * function should be called before attempting to wait on a locked
366 * nfs_page. It will complete the I/O if the page index 'index'
367 * is not contiguous with the existing list of pages in 'desc'.
369 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
371 if (!list_empty(&desc->pg_list)) {
372 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
373 if (index != prev->wb_index + 1)
374 nfs_pageio_doio(desc);
378 #define NFS_SCAN_MAXENTRIES 16
380 * nfs_scan_list - Scan a list for matching requests
382 * @head: One of the NFS inode request lists
383 * @dst: Destination list
384 * @idx_start: lower bound of page->index to scan
385 * @npages: idx_start + npages sets the upper bound to scan.
387 * Moves elements from one of the inode request lists.
388 * If the number of requests is set to 0, the entire address_space
389 * starting at index idx_start, is scanned.
390 * The requests are *not* checked to ensure that they form a contiguous set.
391 * You must be holding the inode's req_lock when calling this function
393 int nfs_scan_list(struct nfs_inode *nfsi, struct list_head *head,
394 struct list_head *dst, pgoff_t idx_start,
397 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
398 struct nfs_page *req;
407 idx_end = idx_start + npages - 1;
410 found = radix_tree_gang_lookup(&nfsi->nfs_page_tree,
411 (void **)&pgvec[0], idx_start,
412 NFS_SCAN_MAXENTRIES);
415 for (i = 0; i < found; i++) {
417 if (req->wb_index > idx_end)
419 idx_start = req->wb_index + 1;
420 if (req->wb_list_head != head)
422 if (nfs_set_page_writeback_locked(req)) {
423 nfs_list_remove_request(req);
424 nfs_list_add_request(req, dst);
434 int __init nfs_init_nfspagecache(void)
436 nfs_page_cachep = kmem_cache_create("nfs_page",
437 sizeof(struct nfs_page),
438 0, SLAB_HWCACHE_ALIGN,
440 if (nfs_page_cachep == NULL)
446 void nfs_destroy_nfspagecache(void)
448 kmem_cache_destroy(nfs_page_cachep);