Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | ||
1da177e4 LT |
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> | |
3f442547 | 20 | #include <linux/writeback.h> |
1da177e4 LT |
21 | |
22 | #define NFS_PARANOIA 1 | |
23 | ||
e18b890b | 24 | static struct kmem_cache *nfs_page_cachep; |
1da177e4 LT |
25 | |
26 | static inline struct nfs_page * | |
27 | nfs_page_alloc(void) | |
28 | { | |
29 | struct nfs_page *p; | |
e94b1766 | 30 | p = kmem_cache_alloc(nfs_page_cachep, GFP_KERNEL); |
1da177e4 LT |
31 | if (p) { |
32 | memset(p, 0, sizeof(*p)); | |
33 | INIT_LIST_HEAD(&p->wb_list); | |
34 | } | |
35 | return p; | |
36 | } | |
37 | ||
38 | static inline void | |
39 | nfs_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 | */ | |
58 | struct nfs_page * | |
59 | nfs_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); | |
cd52ed35 TM |
88 | BUG_ON(PagePrivate(page)); |
89 | BUG_ON(!PageLocked(page)); | |
90 | BUG_ON(page->mapping->host != inode); | |
1da177e4 LT |
91 | 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 | */ | |
104 | void 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(); | |
464a98bd | 113 | wake_up_bit(&req->wb_flags, PG_BUSY); |
1da177e4 LT |
114 | nfs_release_request(req); |
115 | } | |
116 | ||
c6a556b8 TM |
117 | /** |
118 | * nfs_set_page_writeback_locked - Lock a request for writeback | |
119 | * @req: | |
120 | */ | |
121 | int 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 | */ | |
134 | void nfs_clear_page_writeback(struct nfs_page *req) | |
135 | { | |
136 | struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode); | |
137 | ||
deb7d638 TM |
138 | 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 | } | |
c6a556b8 TM |
143 | nfs_unlock_request(req); |
144 | } | |
145 | ||
1da177e4 LT |
146 | /** |
147 | * 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 | */ | |
153 | void nfs_clear_request(struct nfs_page *req) | |
154 | { | |
cd52ed35 TM |
155 | struct page *page = req->wb_page; |
156 | if (page != NULL) { | |
cd52ed35 | 157 | page_cache_release(page); |
1da177e4 LT |
158 | 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 | */ | |
169 | void | |
170 | nfs_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 | ||
464a98bd TM |
186 | static 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 | ||
1da177e4 LT |
197 | /** |
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 | */ | |
204 | int | |
205 | nfs_wait_on_request(struct nfs_page *req) | |
206 | { | |
464a98bd TM |
207 | struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode); |
208 | sigset_t oldmask; | |
209 | int ret = 0; | |
210 | ||
211 | 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); | |
221 | out: | |
222 | return ret; | |
1da177e4 LT |
223 | } |
224 | ||
225 | /** | |
d8a5ad75 TM |
226 | * nfs_pageio_init - initialise a page io descriptor |
227 | * @desc: pointer to descriptor | |
bcb71bba TM |
228 | * @inode: pointer to inode |
229 | * @doio: pointer to io function | |
230 | * @bsize: io block size | |
231 | * @io_flags: extra parameters for the io function | |
d8a5ad75 | 232 | */ |
bcb71bba TM |
233 | void nfs_pageio_init(struct nfs_pageio_descriptor *desc, |
234 | struct inode *inode, | |
235 | int (*doio)(struct inode *, struct list_head *, size_t, int), | |
236 | unsigned int bsize, | |
237 | int io_flags) | |
d8a5ad75 TM |
238 | { |
239 | INIT_LIST_HEAD(&desc->pg_list); | |
bcb71bba | 240 | desc->pg_bytes_written = 0; |
d8a5ad75 TM |
241 | desc->pg_count = 0; |
242 | desc->pg_bsize = bsize; | |
243 | desc->pg_base = 0; | |
bcb71bba TM |
244 | desc->pg_inode = inode; |
245 | desc->pg_doio = doio; | |
246 | desc->pg_ioflags = io_flags; | |
247 | desc->pg_error = 0; | |
d8a5ad75 TM |
248 | } |
249 | ||
250 | /** | |
251 | * nfs_can_coalesce_requests - test two requests for compatibility | |
252 | * @prev: pointer to nfs_page | |
253 | * @req: pointer to nfs_page | |
254 | * | |
255 | * The nfs_page structures 'prev' and 'req' are compared to ensure that the | |
256 | * page data area they describe is contiguous, and that their RPC | |
257 | * credentials, NFSv4 open state, and lockowners are the same. | |
258 | * | |
259 | * Return 'true' if this is the case, else return 'false'. | |
260 | */ | |
261 | static int nfs_can_coalesce_requests(struct nfs_page *prev, | |
262 | struct nfs_page *req) | |
263 | { | |
264 | if (req->wb_context->cred != prev->wb_context->cred) | |
265 | return 0; | |
266 | if (req->wb_context->lockowner != prev->wb_context->lockowner) | |
267 | return 0; | |
268 | if (req->wb_context->state != prev->wb_context->state) | |
269 | return 0; | |
270 | if (req->wb_index != (prev->wb_index + 1)) | |
271 | return 0; | |
272 | if (req->wb_pgbase != 0) | |
273 | return 0; | |
274 | if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE) | |
275 | return 0; | |
276 | return 1; | |
277 | } | |
278 | ||
279 | /** | |
bcb71bba | 280 | * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list. |
d8a5ad75 TM |
281 | * @desc: destination io descriptor |
282 | * @req: request | |
283 | * | |
284 | * Returns true if the request 'req' was successfully coalesced into the | |
285 | * existing list of pages 'desc'. | |
286 | */ | |
bcb71bba TM |
287 | static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc, |
288 | struct nfs_page *req) | |
d8a5ad75 TM |
289 | { |
290 | size_t newlen = req->wb_bytes; | |
291 | ||
292 | if (desc->pg_count != 0) { | |
293 | struct nfs_page *prev; | |
294 | ||
295 | /* | |
296 | * FIXME: ideally we should be able to coalesce all requests | |
297 | * that are not block boundary aligned, but currently this | |
298 | * is problematic for the case of bsize < PAGE_CACHE_SIZE, | |
299 | * since nfs_flush_multi and nfs_pagein_multi assume you | |
300 | * can have only one struct nfs_page. | |
301 | */ | |
302 | newlen += desc->pg_count; | |
303 | if (desc->pg_base + newlen > desc->pg_bsize) | |
304 | return 0; | |
305 | prev = nfs_list_entry(desc->pg_list.prev); | |
306 | if (!nfs_can_coalesce_requests(prev, req)) | |
307 | return 0; | |
308 | } else | |
309 | desc->pg_base = req->wb_pgbase; | |
310 | nfs_list_remove_request(req); | |
311 | nfs_list_add_request(req, &desc->pg_list); | |
312 | desc->pg_count = newlen; | |
313 | return 1; | |
314 | } | |
315 | ||
bcb71bba TM |
316 | /* |
317 | * Helper for nfs_pageio_add_request and nfs_pageio_complete | |
318 | */ | |
319 | static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc) | |
320 | { | |
321 | if (!list_empty(&desc->pg_list)) { | |
322 | int error = desc->pg_doio(desc->pg_inode, | |
323 | &desc->pg_list, | |
324 | desc->pg_count, | |
325 | desc->pg_ioflags); | |
326 | if (error < 0) | |
327 | desc->pg_error = error; | |
328 | else | |
329 | desc->pg_bytes_written += desc->pg_count; | |
330 | } | |
331 | if (list_empty(&desc->pg_list)) { | |
332 | desc->pg_count = 0; | |
333 | desc->pg_base = 0; | |
334 | } | |
335 | } | |
336 | ||
337 | /** | |
338 | * nfs_pageio_add_request - Attempt to coalesce a request into a page list. | |
339 | * @desc: destination io descriptor | |
340 | * @req: request | |
341 | * | |
342 | * Returns true if the request 'req' was successfully coalesced into the | |
343 | * existing list of pages 'desc'. | |
344 | */ | |
345 | static int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, | |
346 | struct nfs_page *req) | |
347 | { | |
348 | while (!nfs_pageio_do_add_request(desc, req)) { | |
349 | nfs_pageio_doio(desc); | |
350 | if (desc->pg_error < 0) | |
351 | return 0; | |
352 | } | |
353 | return 1; | |
354 | } | |
355 | ||
d8a5ad75 TM |
356 | /** |
357 | * nfs_pageio_add_list - Split coalesced requests out from a list. | |
358 | * @desc: destination io descriptor | |
1da177e4 | 359 | * @head: source list |
1da177e4 LT |
360 | * |
361 | * Moves a maximum of 'nmax' elements from one list to another. | |
362 | * The elements are checked to ensure that they form a contiguous set | |
363 | * of pages, and that the RPC credentials are the same. | |
364 | */ | |
d8a5ad75 TM |
365 | void nfs_pageio_add_list(struct nfs_pageio_descriptor *desc, |
366 | struct list_head *head) | |
1da177e4 | 367 | { |
1da177e4 | 368 | while (!list_empty(head)) { |
d8a5ad75 TM |
369 | struct nfs_page *req = nfs_list_entry(head->next); |
370 | if (!nfs_pageio_add_request(desc, req)) | |
1da177e4 LT |
371 | break; |
372 | } | |
1da177e4 LT |
373 | } |
374 | ||
bcb71bba TM |
375 | /** |
376 | * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor | |
377 | * @desc: pointer to io descriptor | |
378 | */ | |
379 | void nfs_pageio_complete(struct nfs_pageio_descriptor *desc) | |
380 | { | |
381 | nfs_pageio_doio(desc); | |
382 | } | |
383 | ||
3da28eb1 TM |
384 | #define NFS_SCAN_MAXENTRIES 16 |
385 | /** | |
3f442547 TM |
386 | * nfs_scan_dirty - Scan the radix tree for dirty requests |
387 | * @mapping: pointer to address space | |
388 | * @wbc: writeback_control structure | |
3da28eb1 | 389 | * @dst: Destination list |
3da28eb1 TM |
390 | * |
391 | * Moves elements from one of the inode request lists. | |
392 | * If the number of requests is set to 0, the entire address_space | |
393 | * starting at index idx_start, is scanned. | |
394 | * The requests are *not* checked to ensure that they form a contiguous set. | |
395 | * You must be holding the inode's req_lock when calling this function | |
396 | */ | |
3f442547 TM |
397 | long nfs_scan_dirty(struct address_space *mapping, |
398 | struct writeback_control *wbc, | |
399 | struct list_head *dst) | |
3da28eb1 | 400 | { |
3f442547 | 401 | struct nfs_inode *nfsi = NFS_I(mapping->host); |
3da28eb1 TM |
402 | struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES]; |
403 | struct nfs_page *req; | |
3f442547 | 404 | pgoff_t idx_start, idx_end; |
3f442547 | 405 | long res = 0; |
3da28eb1 | 406 | int found, i; |
3da28eb1 | 407 | |
9cf85e0a | 408 | if (nfsi->ndirty == 0) |
3f442547 TM |
409 | return 0; |
410 | if (wbc->range_cyclic) { | |
411 | idx_start = 0; | |
412 | idx_end = ULONG_MAX; | |
413 | } else if (wbc->range_end == 0) { | |
414 | idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
415 | idx_end = ULONG_MAX; | |
416 | } else { | |
417 | idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
418 | idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
419 | } | |
3da28eb1 TM |
420 | |
421 | for (;;) { | |
3f442547 TM |
422 | unsigned int toscan = NFS_SCAN_MAXENTRIES; |
423 | ||
3da28eb1 | 424 | found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, |
3f442547 | 425 | (void **)&pgvec[0], idx_start, toscan, |
3da28eb1 | 426 | NFS_PAGE_TAG_DIRTY); |
3f442547 TM |
427 | |
428 | /* Did we make progress? */ | |
3da28eb1 TM |
429 | if (found <= 0) |
430 | break; | |
3f442547 | 431 | |
3da28eb1 TM |
432 | for (i = 0; i < found; i++) { |
433 | req = pgvec[i]; | |
3f442547 | 434 | if (!wbc->range_cyclic && req->wb_index > idx_end) |
3da28eb1 TM |
435 | goto out; |
436 | ||
3f442547 TM |
437 | /* Try to lock request and mark it for writeback */ |
438 | if (!nfs_set_page_writeback_locked(req)) | |
439 | goto next; | |
440 | radix_tree_tag_clear(&nfsi->nfs_page_tree, | |
441 | req->wb_index, NFS_PAGE_TAG_DIRTY); | |
442 | nfsi->ndirty--; | |
443 | nfs_list_remove_request(req); | |
444 | nfs_list_add_request(req, dst); | |
3f442547 TM |
445 | res++; |
446 | if (res == LONG_MAX) | |
447 | goto out; | |
3f442547 TM |
448 | next: |
449 | idx_start = req->wb_index + 1; | |
3da28eb1 TM |
450 | } |
451 | } | |
452 | out: | |
3f442547 | 453 | WARN_ON ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty)); |
3da28eb1 TM |
454 | return res; |
455 | } | |
456 | ||
1da177e4 LT |
457 | /** |
458 | * nfs_scan_list - Scan a list for matching requests | |
d2ccddf0 | 459 | * @nfsi: NFS inode |
1da177e4 LT |
460 | * @head: One of the NFS inode request lists |
461 | * @dst: Destination list | |
462 | * @idx_start: lower bound of page->index to scan | |
463 | * @npages: idx_start + npages sets the upper bound to scan. | |
464 | * | |
465 | * Moves elements from one of the inode request lists. | |
466 | * If the number of requests is set to 0, the entire address_space | |
467 | * starting at index idx_start, is scanned. | |
468 | * The requests are *not* checked to ensure that they form a contiguous set. | |
469 | * You must be holding the inode's req_lock when calling this function | |
470 | */ | |
d2ccddf0 TM |
471 | int nfs_scan_list(struct nfs_inode *nfsi, struct list_head *head, |
472 | struct list_head *dst, unsigned long idx_start, | |
473 | unsigned int npages) | |
1da177e4 | 474 | { |
d2ccddf0 TM |
475 | struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES]; |
476 | struct nfs_page *req; | |
477 | unsigned long idx_end; | |
478 | int found, i; | |
479 | int res; | |
1da177e4 LT |
480 | |
481 | res = 0; | |
482 | if (npages == 0) | |
483 | idx_end = ~0; | |
484 | else | |
485 | idx_end = idx_start + npages - 1; | |
486 | ||
d2ccddf0 TM |
487 | for (;;) { |
488 | found = radix_tree_gang_lookup(&nfsi->nfs_page_tree, | |
489 | (void **)&pgvec[0], idx_start, | |
490 | NFS_SCAN_MAXENTRIES); | |
491 | if (found <= 0) | |
1da177e4 | 492 | break; |
d2ccddf0 TM |
493 | for (i = 0; i < found; i++) { |
494 | req = pgvec[i]; | |
495 | if (req->wb_index > idx_end) | |
496 | goto out; | |
497 | idx_start = req->wb_index + 1; | |
498 | if (req->wb_list_head != head) | |
499 | continue; | |
500 | if (nfs_set_page_writeback_locked(req)) { | |
501 | nfs_list_remove_request(req); | |
502 | nfs_list_add_request(req, dst); | |
503 | res++; | |
504 | } | |
505 | } | |
1da177e4 | 506 | |
1da177e4 | 507 | } |
d2ccddf0 | 508 | out: |
1da177e4 LT |
509 | return res; |
510 | } | |
511 | ||
f7b422b1 | 512 | int __init nfs_init_nfspagecache(void) |
1da177e4 LT |
513 | { |
514 | nfs_page_cachep = kmem_cache_create("nfs_page", | |
515 | sizeof(struct nfs_page), | |
516 | 0, SLAB_HWCACHE_ALIGN, | |
517 | NULL, NULL); | |
518 | if (nfs_page_cachep == NULL) | |
519 | return -ENOMEM; | |
520 | ||
521 | return 0; | |
522 | } | |
523 | ||
266bee88 | 524 | void nfs_destroy_nfspagecache(void) |
1da177e4 | 525 | { |
1a1d92c1 | 526 | kmem_cache_destroy(nfs_page_cachep); |
1da177e4 LT |
527 | } |
528 |