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/sunrpc/clnt.h>
15 #include <linux/nfs3.h>
16 #include <linux/nfs4.h>
17 #include <linux/nfs_page.h>
18 #include <linux/nfs_fs.h>
19 #include <linux/nfs_mount.h>
21 #define NFS_PARANOIA 1
23 static kmem_cache_t *nfs_page_cachep;
25 static inline struct nfs_page *
29 p = kmem_cache_alloc(nfs_page_cachep, SLAB_KERNEL);
31 memset(p, 0, sizeof(*p));
32 INIT_LIST_HEAD(&p->wb_list);
38 nfs_page_free(struct nfs_page *p)
40 kmem_cache_free(nfs_page_cachep, p);
44 * nfs_create_request - Create an NFS read/write request.
45 * @file: file descriptor to use
46 * @inode: inode to which the request is attached
47 * @page: page to write
48 * @offset: starting offset within the page for the write
49 * @count: number of bytes to read/write
51 * The page must be locked by the caller. This makes sure we never
52 * create two different requests for the same page, and avoids
53 * a possible deadlock when we reach the hard limit on the number
55 * User should ensure it is safe to sleep in this function.
58 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
60 unsigned int offset, unsigned int count)
62 struct nfs_server *server = NFS_SERVER(inode);
65 /* Deal with hard limits. */
67 /* try to allocate the request struct */
68 req = nfs_page_alloc();
72 /* Try to free up at least one request in order to stay
73 * below the hard limit
75 if (signalled() && (server->flags & NFS_MOUNT_INTR))
76 return ERR_PTR(-ERESTARTSYS);
80 /* Initialize the request struct. Initially, we assume a
81 * long write-back delay. This will be adjusted in
82 * update_nfs_request below if the region is not locked. */
84 atomic_set(&req->wb_complete, 0);
85 req->wb_index = page->index;
87 BUG_ON(PagePrivate(page));
88 BUG_ON(!PageLocked(page));
89 BUG_ON(page->mapping->host != inode);
90 req->wb_offset = offset;
91 req->wb_pgbase = offset;
92 req->wb_bytes = count;
93 atomic_set(&req->wb_count, 1);
94 req->wb_context = get_nfs_open_context(ctx);
100 * nfs_unlock_request - Unlock request and wake up sleepers.
103 void nfs_unlock_request(struct nfs_page *req)
105 if (!NFS_WBACK_BUSY(req)) {
106 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
109 smp_mb__before_clear_bit();
110 clear_bit(PG_BUSY, &req->wb_flags);
111 smp_mb__after_clear_bit();
112 wake_up_bit(&req->wb_flags, PG_BUSY);
113 nfs_release_request(req);
117 * nfs_set_page_writeback_locked - Lock a request for writeback
120 int nfs_set_page_writeback_locked(struct nfs_page *req)
122 struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
124 if (!nfs_lock_request(req))
126 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
131 * nfs_clear_page_writeback - Unlock request and wake up sleepers
133 void nfs_clear_page_writeback(struct nfs_page *req)
135 struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
137 if (req->wb_page != NULL) {
138 spin_lock(&nfsi->req_lock);
139 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
140 spin_unlock(&nfsi->req_lock);
142 nfs_unlock_request(req);
146 * nfs_clear_request - Free up all resources allocated to the request
149 * Release page resources associated with a write request after it
152 void nfs_clear_request(struct nfs_page *req)
154 struct page *page = req->wb_page;
156 page_cache_release(page);
163 * nfs_release_request - Release the count on an NFS read/write request
164 * @req: request to release
166 * Note: Should never be called with the spinlock held!
169 nfs_release_request(struct nfs_page *req)
171 if (!atomic_dec_and_test(&req->wb_count))
175 BUG_ON (!list_empty(&req->wb_list));
176 BUG_ON (NFS_WBACK_BUSY(req));
179 /* Release struct file or cached credential */
180 nfs_clear_request(req);
181 put_nfs_open_context(req->wb_context);
185 static int nfs_wait_bit_interruptible(void *word)
189 if (signal_pending(current))
197 * nfs_wait_on_request - Wait for a request to complete.
198 * @req: request to wait upon.
200 * Interruptible by signals only if mounted with intr flag.
201 * The user is responsible for holding a count on the request.
204 nfs_wait_on_request(struct nfs_page *req)
206 struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
210 if (!test_bit(PG_BUSY, &req->wb_flags))
213 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
214 * are not interrupted if intr flag is not set
216 rpc_clnt_sigmask(clnt, &oldmask);
217 ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
218 nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
219 rpc_clnt_sigunmask(clnt, &oldmask);
225 * nfs_coalesce_requests - Split coalesced requests out from a list.
227 * @dst: destination list
228 * @nmax: maximum number of requests to coalesce
230 * Moves a maximum of 'nmax' elements from one list to another.
231 * The elements are checked to ensure that they form a contiguous set
232 * of pages, and that the RPC credentials are the same.
235 nfs_coalesce_requests(struct list_head *head, struct list_head *dst,
238 struct nfs_page *req = NULL;
239 unsigned int npages = 0;
241 while (!list_empty(head)) {
242 struct nfs_page *prev = req;
244 req = nfs_list_entry(head->next);
246 if (req->wb_context->cred != prev->wb_context->cred)
248 if (req->wb_context->lockowner != prev->wb_context->lockowner)
250 if (req->wb_context->state != prev->wb_context->state)
252 if (req->wb_index != (prev->wb_index + 1))
255 if (req->wb_pgbase != 0)
258 nfs_list_remove_request(req);
259 nfs_list_add_request(req, dst);
261 if (req->wb_pgbase + req->wb_bytes != PAGE_CACHE_SIZE)
269 #define NFS_SCAN_MAXENTRIES 16
271 * nfs_scan_lock_dirty - Scan the radix tree for dirty requests
273 * @dst: Destination list
274 * @idx_start: lower bound of page->index to scan
275 * @npages: idx_start + npages sets the upper bound to scan.
277 * Moves elements from one of the inode request lists.
278 * If the number of requests is set to 0, the entire address_space
279 * starting at index idx_start, is scanned.
280 * The requests are *not* checked to ensure that they form a contiguous set.
281 * You must be holding the inode's req_lock when calling this function
284 nfs_scan_lock_dirty(struct nfs_inode *nfsi, struct list_head *dst,
285 unsigned long idx_start, unsigned int npages)
287 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
288 struct nfs_page *req;
289 unsigned long idx_end;
297 idx_end = idx_start + npages - 1;
300 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
301 (void **)&pgvec[0], idx_start, NFS_SCAN_MAXENTRIES,
305 for (i = 0; i < found; i++) {
307 if (req->wb_index > idx_end)
310 idx_start = req->wb_index + 1;
312 if (nfs_set_page_writeback_locked(req)) {
313 radix_tree_tag_clear(&nfsi->nfs_page_tree,
314 req->wb_index, NFS_PAGE_TAG_DIRTY);
315 nfs_list_remove_request(req);
316 nfs_list_add_request(req, dst);
317 dec_zone_page_state(req->wb_page, NR_FILE_DIRTY);
327 * nfs_scan_list - Scan a list for matching requests
329 * @head: One of the NFS inode request lists
330 * @dst: Destination list
331 * @idx_start: lower bound of page->index to scan
332 * @npages: idx_start + npages sets the upper bound to scan.
334 * Moves elements from one of the inode request lists.
335 * If the number of requests is set to 0, the entire address_space
336 * starting at index idx_start, is scanned.
337 * The requests are *not* checked to ensure that they form a contiguous set.
338 * You must be holding the inode's req_lock when calling this function
340 int nfs_scan_list(struct nfs_inode *nfsi, struct list_head *head,
341 struct list_head *dst, unsigned long idx_start,
344 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
345 struct nfs_page *req;
346 unsigned long idx_end;
354 idx_end = idx_start + npages - 1;
357 found = radix_tree_gang_lookup(&nfsi->nfs_page_tree,
358 (void **)&pgvec[0], idx_start,
359 NFS_SCAN_MAXENTRIES);
362 for (i = 0; i < found; i++) {
364 if (req->wb_index > idx_end)
366 idx_start = req->wb_index + 1;
367 if (req->wb_list_head != head)
369 if (nfs_set_page_writeback_locked(req)) {
370 nfs_list_remove_request(req);
371 nfs_list_add_request(req, dst);
381 int __init nfs_init_nfspagecache(void)
383 nfs_page_cachep = kmem_cache_create("nfs_page",
384 sizeof(struct nfs_page),
385 0, SLAB_HWCACHE_ALIGN,
387 if (nfs_page_cachep == NULL)
393 void nfs_destroy_nfspagecache(void)
395 kmem_cache_destroy(nfs_page_cachep);