blob: ca4b1d4ff42b0f7d337ad01fa7c1675f959c4134 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/pagelist.c
3 *
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.
7 *
8 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
9 *
10 */
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#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>
Trond Myklebust3f442542006-09-17 14:46:44 -040020#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22#define NFS_PARANOIA 1
23
Christoph Lametere18b8902006-12-06 20:33:20 -080024static struct kmem_cache *nfs_page_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26static inline struct nfs_page *
27nfs_page_alloc(void)
28{
29 struct nfs_page *p;
Christoph Lametere94b1762006-12-06 20:33:17 -080030 p = kmem_cache_alloc(nfs_page_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 if (p) {
32 memset(p, 0, sizeof(*p));
33 INIT_LIST_HEAD(&p->wb_list);
34 }
35 return p;
36}
37
38static inline void
39nfs_page_free(struct nfs_page *p)
40{
41 kmem_cache_free(nfs_page_cachep, p);
42}
43
44/**
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
51 *
52 * The page must be locked by the caller. This makes sure we never
53 * create two different requests for the same page, and avoids
54 * a possible deadlock when we reach the hard limit on the number
55 * of dirty pages.
56 * User should ensure it is safe to sleep in this function.
57 */
58struct nfs_page *
59nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
60 struct page *page,
61 unsigned int offset, unsigned int count)
62{
63 struct nfs_server *server = NFS_SERVER(inode);
64 struct nfs_page *req;
65
66 /* Deal with hard limits. */
67 for (;;) {
68 /* try to allocate the request struct */
69 req = nfs_page_alloc();
70 if (req != NULL)
71 break;
72
73 /* Try to free up at least one request in order to stay
74 * below the hard limit
75 */
76 if (signalled() && (server->flags & NFS_MOUNT_INTR))
77 return ERR_PTR(-ERESTARTSYS);
78 yield();
79 }
80
81 /* Initialize the request struct. Initially, we assume a
82 * long write-back delay. This will be adjusted in
83 * update_nfs_request below if the region is not locked. */
84 req->wb_page = page;
85 atomic_set(&req->wb_complete, 0);
86 req->wb_index = page->index;
87 page_cache_get(page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -050088 BUG_ON(PagePrivate(page));
89 BUG_ON(!PageLocked(page));
90 BUG_ON(page->mapping->host != inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 req->wb_offset = offset;
92 req->wb_pgbase = offset;
93 req->wb_bytes = count;
94 atomic_set(&req->wb_count, 1);
95 req->wb_context = get_nfs_open_context(ctx);
96
97 return req;
98}
99
100/**
101 * nfs_unlock_request - Unlock request and wake up sleepers.
102 * @req:
103 */
104void nfs_unlock_request(struct nfs_page *req)
105{
106 if (!NFS_WBACK_BUSY(req)) {
107 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
108 BUG();
109 }
110 smp_mb__before_clear_bit();
111 clear_bit(PG_BUSY, &req->wb_flags);
112 smp_mb__after_clear_bit();
Trond Myklebust464a98b2005-06-22 17:16:21 +0000113 wake_up_bit(&req->wb_flags, PG_BUSY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 nfs_release_request(req);
115}
116
117/**
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000118 * nfs_set_page_writeback_locked - Lock a request for writeback
119 * @req:
120 */
121int nfs_set_page_writeback_locked(struct nfs_page *req)
122{
123 struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
124
125 if (!nfs_lock_request(req))
126 return 0;
127 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
128 return 1;
129}
130
131/**
132 * nfs_clear_page_writeback - Unlock request and wake up sleepers
133 */
134void nfs_clear_page_writeback(struct nfs_page *req)
135{
136 struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
137
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500138 if (req->wb_page != NULL) {
139 spin_lock(&nfsi->req_lock);
140 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
141 spin_unlock(&nfsi->req_lock);
142 }
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000143 nfs_unlock_request(req);
144}
145
146/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 * nfs_clear_request - Free up all resources allocated to the request
148 * @req:
149 *
150 * Release page resources associated with a write request after it
151 * has completed.
152 */
153void nfs_clear_request(struct nfs_page *req)
154{
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500155 struct page *page = req->wb_page;
156 if (page != NULL) {
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500157 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 req->wb_page = NULL;
159 }
160}
161
162
163/**
164 * nfs_release_request - Release the count on an NFS read/write request
165 * @req: request to release
166 *
167 * Note: Should never be called with the spinlock held!
168 */
169void
170nfs_release_request(struct nfs_page *req)
171{
172 if (!atomic_dec_and_test(&req->wb_count))
173 return;
174
175#ifdef NFS_PARANOIA
176 BUG_ON (!list_empty(&req->wb_list));
177 BUG_ON (NFS_WBACK_BUSY(req));
178#endif
179
180 /* Release struct file or cached credential */
181 nfs_clear_request(req);
182 put_nfs_open_context(req->wb_context);
183 nfs_page_free(req);
184}
185
Trond Myklebust464a98b2005-06-22 17:16:21 +0000186static int nfs_wait_bit_interruptible(void *word)
187{
188 int ret = 0;
189
190 if (signal_pending(current))
191 ret = -ERESTARTSYS;
192 else
193 schedule();
194 return ret;
195}
196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197/**
198 * nfs_wait_on_request - Wait for a request to complete.
199 * @req: request to wait upon.
200 *
201 * Interruptible by signals only if mounted with intr flag.
202 * The user is responsible for holding a count on the request.
203 */
204int
205nfs_wait_on_request(struct nfs_page *req)
206{
Trond Myklebust464a98b2005-06-22 17:16:21 +0000207 struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
208 sigset_t oldmask;
209 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Trond Myklebust464a98b2005-06-22 17:16:21 +0000211 if (!test_bit(PG_BUSY, &req->wb_flags))
212 goto out;
213 /*
214 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
215 * are not interrupted if intr flag is not set
216 */
217 rpc_clnt_sigmask(clnt, &oldmask);
218 ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
219 nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
220 rpc_clnt_sigunmask(clnt, &oldmask);
221out:
222 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225/**
226 * nfs_coalesce_requests - Split coalesced requests out from a list.
227 * @head: source list
228 * @dst: destination list
229 * @nmax: maximum number of requests to coalesce
230 *
231 * Moves a maximum of 'nmax' elements from one list to another.
232 * The elements are checked to ensure that they form a contiguous set
233 * of pages, and that the RPC credentials are the same.
234 */
235int
236nfs_coalesce_requests(struct list_head *head, struct list_head *dst,
237 unsigned int nmax)
238{
239 struct nfs_page *req = NULL;
240 unsigned int npages = 0;
241
242 while (!list_empty(head)) {
243 struct nfs_page *prev = req;
244
245 req = nfs_list_entry(head->next);
246 if (prev) {
247 if (req->wb_context->cred != prev->wb_context->cred)
248 break;
249 if (req->wb_context->lockowner != prev->wb_context->lockowner)
250 break;
251 if (req->wb_context->state != prev->wb_context->state)
252 break;
253 if (req->wb_index != (prev->wb_index + 1))
254 break;
255
256 if (req->wb_pgbase != 0)
257 break;
258 }
259 nfs_list_remove_request(req);
260 nfs_list_add_request(req, dst);
261 npages++;
262 if (req->wb_pgbase + req->wb_bytes != PAGE_CACHE_SIZE)
263 break;
264 if (npages >= nmax)
265 break;
266 }
267 return npages;
268}
269
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000270#define NFS_SCAN_MAXENTRIES 16
271/**
Trond Myklebust3f442542006-09-17 14:46:44 -0400272 * nfs_scan_dirty - Scan the radix tree for dirty requests
273 * @mapping: pointer to address space
274 * @wbc: writeback_control structure
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000275 * @dst: Destination list
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000276 *
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
282 */
Trond Myklebust3f442542006-09-17 14:46:44 -0400283long nfs_scan_dirty(struct address_space *mapping,
284 struct writeback_control *wbc,
285 struct list_head *dst)
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000286{
Trond Myklebust3f442542006-09-17 14:46:44 -0400287 struct nfs_inode *nfsi = NFS_I(mapping->host);
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000288 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
289 struct nfs_page *req;
Trond Myklebust3f442542006-09-17 14:46:44 -0400290 pgoff_t idx_start, idx_end;
Trond Myklebust3f442542006-09-17 14:46:44 -0400291 long res = 0;
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000292 int found, i;
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000293
Trond Myklebust9cf85e02006-12-05 00:36:56 -0500294 if (nfsi->ndirty == 0)
Trond Myklebust3f442542006-09-17 14:46:44 -0400295 return 0;
296 if (wbc->range_cyclic) {
297 idx_start = 0;
298 idx_end = ULONG_MAX;
299 } else if (wbc->range_end == 0) {
300 idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
301 idx_end = ULONG_MAX;
302 } else {
303 idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
304 idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
305 }
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000306
307 for (;;) {
Trond Myklebust3f442542006-09-17 14:46:44 -0400308 unsigned int toscan = NFS_SCAN_MAXENTRIES;
309
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000310 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
Trond Myklebust3f442542006-09-17 14:46:44 -0400311 (void **)&pgvec[0], idx_start, toscan,
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000312 NFS_PAGE_TAG_DIRTY);
Trond Myklebust3f442542006-09-17 14:46:44 -0400313
314 /* Did we make progress? */
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000315 if (found <= 0)
316 break;
Trond Myklebust3f442542006-09-17 14:46:44 -0400317
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000318 for (i = 0; i < found; i++) {
319 req = pgvec[i];
Trond Myklebust3f442542006-09-17 14:46:44 -0400320 if (!wbc->range_cyclic && req->wb_index > idx_end)
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000321 goto out;
322
Trond Myklebust3f442542006-09-17 14:46:44 -0400323 /* Try to lock request and mark it for writeback */
324 if (!nfs_set_page_writeback_locked(req))
325 goto next;
326 radix_tree_tag_clear(&nfsi->nfs_page_tree,
327 req->wb_index, NFS_PAGE_TAG_DIRTY);
328 nfsi->ndirty--;
329 nfs_list_remove_request(req);
330 nfs_list_add_request(req, dst);
Trond Myklebust3f442542006-09-17 14:46:44 -0400331 res++;
332 if (res == LONG_MAX)
333 goto out;
Trond Myklebust3f442542006-09-17 14:46:44 -0400334next:
335 idx_start = req->wb_index + 1;
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000336 }
337 }
338out:
Trond Myklebust3f442542006-09-17 14:46:44 -0400339 WARN_ON ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty));
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000340 return res;
341}
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343/**
344 * nfs_scan_list - Scan a list for matching requests
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400345 * @nfsi: NFS inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 * @head: One of the NFS inode request lists
347 * @dst: Destination list
348 * @idx_start: lower bound of page->index to scan
349 * @npages: idx_start + npages sets the upper bound to scan.
350 *
351 * Moves elements from one of the inode request lists.
352 * If the number of requests is set to 0, the entire address_space
353 * starting at index idx_start, is scanned.
354 * The requests are *not* checked to ensure that they form a contiguous set.
355 * You must be holding the inode's req_lock when calling this function
356 */
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400357int nfs_scan_list(struct nfs_inode *nfsi, struct list_head *head,
358 struct list_head *dst, unsigned long idx_start,
359 unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360{
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400361 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
362 struct nfs_page *req;
363 unsigned long idx_end;
364 int found, i;
365 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367 res = 0;
368 if (npages == 0)
369 idx_end = ~0;
370 else
371 idx_end = idx_start + npages - 1;
372
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400373 for (;;) {
374 found = radix_tree_gang_lookup(&nfsi->nfs_page_tree,
375 (void **)&pgvec[0], idx_start,
376 NFS_SCAN_MAXENTRIES);
377 if (found <= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400379 for (i = 0; i < found; i++) {
380 req = pgvec[i];
381 if (req->wb_index > idx_end)
382 goto out;
383 idx_start = req->wb_index + 1;
384 if (req->wb_list_head != head)
385 continue;
386 if (nfs_set_page_writeback_locked(req)) {
387 nfs_list_remove_request(req);
388 nfs_list_add_request(req, dst);
389 res++;
390 }
391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400394out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 return res;
396}
397
David Howellsf7b422b2006-06-09 09:34:33 -0400398int __init nfs_init_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
400 nfs_page_cachep = kmem_cache_create("nfs_page",
401 sizeof(struct nfs_page),
402 0, SLAB_HWCACHE_ALIGN,
403 NULL, NULL);
404 if (nfs_page_cachep == NULL)
405 return -ENOMEM;
406
407 return 0;
408}
409
David Brownell266bee82006-06-27 12:59:15 -0700410void nfs_destroy_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700412 kmem_cache_destroy(nfs_page_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413}
414