]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
afeacc8c | 23 | #include <linux/export.h> |
af7cf057 TM |
24 | #include <linux/freezer.h> |
25 | #include <linux/wait.h> | |
3fcfab16 | 26 | |
7c0f6ba6 | 27 | #include <linux/uaccess.h> |
1da177e4 LT |
28 | |
29 | #include "delegation.h" | |
49a70f27 | 30 | #include "internal.h" |
91d5b470 | 31 | #include "iostat.h" |
def6ed7e | 32 | #include "nfs4_fs.h" |
074cc1de | 33 | #include "fscache.h" |
94ad1c80 | 34 | #include "pnfs.h" |
1da177e4 | 35 | |
f4ce1299 TM |
36 | #include "nfstrace.h" |
37 | ||
1da177e4 LT |
38 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
39 | ||
40 | #define MIN_POOL_WRITE (32) | |
41 | #define MIN_POOL_COMMIT (4) | |
42 | ||
919e3bd9 TM |
43 | struct nfs_io_completion { |
44 | void (*complete)(void *data); | |
45 | void *data; | |
46 | struct kref refcount; | |
47 | }; | |
48 | ||
1da177e4 LT |
49 | /* |
50 | * Local function declarations | |
51 | */ | |
f8512ad0 | 52 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 | 53 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 54 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 55 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
4a0de55c | 56 | static const struct nfs_rw_ops nfs_rw_write_ops; |
d4581383 | 57 | static void nfs_clear_request_commit(struct nfs_page *req); |
02d1426c WAA |
58 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
59 | struct inode *inode); | |
3a3908c8 TM |
60 | static struct nfs_page * |
61 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
62 | struct page *page); | |
1da177e4 | 63 | |
e18b890b | 64 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 65 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 66 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
67 | static mempool_t *nfs_commit_mempool; |
68 | ||
518662e0 | 69 | struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) |
1da177e4 | 70 | { |
518662e0 | 71 | struct nfs_commit_data *p; |
40859d7e | 72 | |
518662e0 N |
73 | if (never_fail) |
74 | p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); | |
75 | else { | |
76 | /* It is OK to do some reclaim, not no safe to wait | |
77 | * for anything to be returned to the pool. | |
78 | * mempool_alloc() cannot handle that particular combination, | |
79 | * so we need two separate attempts. | |
80 | */ | |
81 | p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); | |
82 | if (!p) | |
83 | p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | | |
84 | __GFP_NOWARN | __GFP_NORETRY); | |
85 | if (!p) | |
86 | return NULL; | |
1da177e4 | 87 | } |
518662e0 N |
88 | |
89 | memset(p, 0, sizeof(*p)); | |
90 | INIT_LIST_HEAD(&p->pages); | |
1da177e4 LT |
91 | return p; |
92 | } | |
e0c2b380 | 93 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 94 | |
0b7c0153 | 95 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
96 | { |
97 | mempool_free(p, nfs_commit_mempool); | |
98 | } | |
e0c2b380 | 99 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 100 | |
1e7f3a48 | 101 | static struct nfs_pgio_header *nfs_writehdr_alloc(void) |
3feb2d49 | 102 | { |
1e7f3a48 | 103 | struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
cd841605 | 104 | |
237f8306 N |
105 | memset(p, 0, sizeof(*p)); |
106 | p->rw_mode = FMODE_WRITE; | |
3feb2d49 TM |
107 | return p; |
108 | } | |
6c75dc0d | 109 | |
1e7f3a48 | 110 | static void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 111 | { |
1e7f3a48 | 112 | mempool_free(hdr, nfs_wdata_mempool); |
3feb2d49 | 113 | } |
1da177e4 | 114 | |
919e3bd9 TM |
115 | static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) |
116 | { | |
117 | return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); | |
118 | } | |
119 | ||
120 | static void nfs_io_completion_init(struct nfs_io_completion *ioc, | |
121 | void (*complete)(void *), void *data) | |
122 | { | |
123 | ioc->complete = complete; | |
124 | ioc->data = data; | |
125 | kref_init(&ioc->refcount); | |
126 | } | |
127 | ||
128 | static void nfs_io_completion_release(struct kref *kref) | |
129 | { | |
130 | struct nfs_io_completion *ioc = container_of(kref, | |
131 | struct nfs_io_completion, refcount); | |
132 | ioc->complete(ioc->data); | |
133 | kfree(ioc); | |
134 | } | |
135 | ||
136 | static void nfs_io_completion_get(struct nfs_io_completion *ioc) | |
137 | { | |
138 | if (ioc != NULL) | |
139 | kref_get(&ioc->refcount); | |
140 | } | |
141 | ||
142 | static void nfs_io_completion_put(struct nfs_io_completion *ioc) | |
143 | { | |
144 | if (ioc != NULL) | |
145 | kref_put(&ioc->refcount, nfs_io_completion_release); | |
146 | } | |
147 | ||
bd37d6fc TM |
148 | static struct nfs_page * |
149 | nfs_page_private_request(struct page *page) | |
150 | { | |
151 | if (!PagePrivate(page)) | |
152 | return NULL; | |
153 | return (struct nfs_page *)page_private(page); | |
154 | } | |
155 | ||
84d3a9a9 WAA |
156 | /* |
157 | * nfs_page_find_head_request_locked - find head request associated with @page | |
158 | * | |
159 | * must be called while holding the inode lock. | |
160 | * | |
161 | * returns matching head request with reference held, or NULL if not found. | |
162 | */ | |
29418aa4 | 163 | static struct nfs_page * |
b30d2f04 | 164 | nfs_page_find_private_request(struct page *page) |
277459d2 | 165 | { |
4b9bb25b | 166 | struct address_space *mapping = page_file_mapping(page); |
bd37d6fc | 167 | struct nfs_page *req; |
277459d2 | 168 | |
b30d2f04 TM |
169 | if (!PagePrivate(page)) |
170 | return NULL; | |
4b9bb25b | 171 | spin_lock(&mapping->private_lock); |
bd37d6fc | 172 | req = nfs_page_private_request(page); |
84d3a9a9 WAA |
173 | if (req) { |
174 | WARN_ON_ONCE(req->wb_head != req); | |
29418aa4 | 175 | kref_get(&req->wb_kref); |
84d3a9a9 | 176 | } |
4b9bb25b | 177 | spin_unlock(&mapping->private_lock); |
b30d2f04 TM |
178 | return req; |
179 | } | |
29418aa4 | 180 | |
b30d2f04 TM |
181 | static struct nfs_page * |
182 | nfs_page_find_swap_request(struct page *page) | |
183 | { | |
184 | struct inode *inode = page_file_mapping(page)->host; | |
185 | struct nfs_inode *nfsi = NFS_I(inode); | |
186 | struct nfs_page *req = NULL; | |
187 | if (!PageSwapCache(page)) | |
188 | return NULL; | |
e824f99a | 189 | mutex_lock(&nfsi->commit_mutex); |
b30d2f04 TM |
190 | if (PageSwapCache(page)) { |
191 | req = nfs_page_search_commits_for_head_request_locked(nfsi, | |
192 | page); | |
193 | if (req) { | |
194 | WARN_ON_ONCE(req->wb_head != req); | |
195 | kref_get(&req->wb_kref); | |
196 | } | |
197 | } | |
e824f99a | 198 | mutex_unlock(&nfsi->commit_mutex); |
277459d2 TM |
199 | return req; |
200 | } | |
201 | ||
84d3a9a9 WAA |
202 | /* |
203 | * nfs_page_find_head_request - find head request associated with @page | |
204 | * | |
205 | * returns matching head request with reference held, or NULL if not found. | |
206 | */ | |
207 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 208 | { |
b30d2f04 | 209 | struct nfs_page *req; |
277459d2 | 210 | |
b30d2f04 TM |
211 | req = nfs_page_find_private_request(page); |
212 | if (!req) | |
213 | req = nfs_page_find_swap_request(page); | |
277459d2 TM |
214 | return req; |
215 | } | |
216 | ||
1da177e4 LT |
217 | /* Adjust the file length if we're writing beyond the end */ |
218 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
219 | { | |
d56b4ddf | 220 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
221 | loff_t end, i_size; |
222 | pgoff_t end_index; | |
1da177e4 | 223 | |
a3d01454 TM |
224 | spin_lock(&inode->i_lock); |
225 | i_size = i_size_read(inode); | |
09cbfeaf | 226 | end_index = (i_size - 1) >> PAGE_SHIFT; |
8cd79788 | 227 | if (i_size > 0 && page_index(page) < end_index) |
a3d01454 | 228 | goto out; |
d56b4ddf | 229 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 230 | if (i_size >= end) |
a3d01454 | 231 | goto out; |
1da177e4 | 232 | i_size_write(inode, end); |
a3d01454 TM |
233 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
234 | out: | |
235 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
236 | } |
237 | ||
a301b777 TM |
238 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
239 | static void nfs_set_pageerror(struct page *page) | |
240 | { | |
d56b4ddf | 241 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
242 | } |
243 | ||
d72ddcba WAA |
244 | /* |
245 | * nfs_page_group_search_locked | |
246 | * @head - head request of page group | |
247 | * @page_offset - offset into page | |
248 | * | |
249 | * Search page group with head @head to find a request that contains the | |
250 | * page offset @page_offset. | |
251 | * | |
252 | * Returns a pointer to the first matching nfs request, or NULL if no | |
253 | * match is found. | |
254 | * | |
255 | * Must be called with the page group lock held | |
256 | */ | |
257 | static struct nfs_page * | |
258 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
259 | { | |
260 | struct nfs_page *req; | |
261 | ||
d72ddcba WAA |
262 | req = head; |
263 | do { | |
264 | if (page_offset >= req->wb_pgbase && | |
265 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
266 | return req; | |
267 | ||
268 | req = req->wb_this_page; | |
269 | } while (req != head); | |
270 | ||
271 | return NULL; | |
272 | } | |
273 | ||
274 | /* | |
275 | * nfs_page_group_covers_page | |
276 | * @head - head request of page group | |
277 | * | |
278 | * Return true if the page group with head @head covers the whole page, | |
279 | * returns false otherwise | |
280 | */ | |
281 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
282 | { | |
283 | struct nfs_page *tmp; | |
284 | unsigned int pos = 0; | |
285 | unsigned int len = nfs_page_length(req->wb_page); | |
286 | ||
1344b7ea | 287 | nfs_page_group_lock(req); |
d72ddcba | 288 | |
7e8a30f8 | 289 | for (;;) { |
d72ddcba | 290 | tmp = nfs_page_group_search_locked(req->wb_head, pos); |
7e8a30f8 TM |
291 | if (!tmp) |
292 | break; | |
293 | pos = tmp->wb_pgbase + tmp->wb_bytes; | |
294 | } | |
d72ddcba WAA |
295 | |
296 | nfs_page_group_unlock(req); | |
7e8a30f8 | 297 | return pos >= len; |
d72ddcba WAA |
298 | } |
299 | ||
1da177e4 LT |
300 | /* We can set the PG_uptodate flag if we see that a write request |
301 | * covers the full page. | |
302 | */ | |
d72ddcba | 303 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 304 | { |
d72ddcba | 305 | if (PageUptodate(req->wb_page)) |
1da177e4 | 306 | return; |
d72ddcba | 307 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 308 | return; |
d72ddcba | 309 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
310 | } |
311 | ||
1da177e4 LT |
312 | static int wb_priority(struct writeback_control *wbc) |
313 | { | |
e87b4c7a | 314 | int ret = 0; |
cca588d6 | 315 | |
e87b4c7a N |
316 | if (wbc->sync_mode == WB_SYNC_ALL) |
317 | ret = FLUSH_COND_STABLE; | |
e87b4c7a | 318 | return ret; |
1da177e4 LT |
319 | } |
320 | ||
89a09141 PZ |
321 | /* |
322 | * NFS congestion control | |
323 | */ | |
324 | ||
325 | int nfs_congestion_kb; | |
326 | ||
327 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
328 | #define NFS_CONGESTION_OFF_THRESH \ | |
329 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
330 | ||
deed85e7 | 331 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 332 | { |
0db10944 JK |
333 | struct inode *inode = page_file_mapping(page)->host; |
334 | struct nfs_server *nfss = NFS_SERVER(inode); | |
5a6d41b3 TM |
335 | int ret = test_set_page_writeback(page); |
336 | ||
deed85e7 | 337 | WARN_ON_ONCE(ret != 0); |
89a09141 | 338 | |
deed85e7 | 339 | if (atomic_long_inc_return(&nfss->writeback) > |
0db10944 JK |
340 | NFS_CONGESTION_ON_THRESH) |
341 | set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); | |
89a09141 PZ |
342 | } |
343 | ||
20633f04 | 344 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 345 | { |
20633f04 | 346 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 | 347 | struct nfs_server *nfss = NFS_SERVER(inode); |
31a01f09 | 348 | bool is_done; |
89a09141 | 349 | |
31a01f09 TM |
350 | is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); |
351 | nfs_unlock_request(req); | |
352 | if (!is_done) | |
20633f04 WAA |
353 | return; |
354 | ||
355 | end_page_writeback(req->wb_page); | |
c4dc4bee | 356 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
0db10944 | 357 | clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); |
89a09141 PZ |
358 | } |
359 | ||
d4581383 WAA |
360 | /* |
361 | * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req | |
362 | * | |
363 | * this is a helper function for nfs_lock_and_join_requests | |
364 | * | |
365 | * @inode - inode associated with request page group, must be holding inode lock | |
366 | * @head - head request of page group, must be holding head lock | |
367 | * @req - request that couldn't lock and needs to wait on the req bit lock | |
d4581383 | 368 | * |
b5bab9bf TM |
369 | * NOTE: this must be called holding page_group bit lock |
370 | * which will be released before returning. | |
d4581383 WAA |
371 | * |
372 | * returns 0 on success, < 0 on error. | |
373 | */ | |
74a6d4b5 TM |
374 | static void |
375 | nfs_unroll_locks(struct inode *inode, struct nfs_page *head, | |
6d17d653 | 376 | struct nfs_page *req) |
d4581383 WAA |
377 | { |
378 | struct nfs_page *tmp; | |
d4581383 WAA |
379 | |
380 | /* relinquish all the locks successfully grabbed this run */ | |
5b2b5187 TM |
381 | for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { |
382 | if (!kref_read(&tmp->wb_kref)) | |
383 | continue; | |
384 | nfs_unlock_and_release_request(tmp); | |
385 | } | |
d4581383 WAA |
386 | } |
387 | ||
388 | /* | |
389 | * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests | |
390 | * | |
391 | * @destroy_list - request list (using wb_this_page) terminated by @old_head | |
392 | * @old_head - the old head of the list | |
393 | * | |
394 | * All subrequests must be locked and removed from all lists, so at this point | |
395 | * they are only "active" in this function, and possibly in nfs_wait_on_request | |
396 | * with a reference held by some other context. | |
397 | */ | |
398 | static void | |
399 | nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, | |
b66aaa8d TM |
400 | struct nfs_page *old_head, |
401 | struct inode *inode) | |
d4581383 WAA |
402 | { |
403 | while (destroy_list) { | |
404 | struct nfs_page *subreq = destroy_list; | |
405 | ||
406 | destroy_list = (subreq->wb_this_page == old_head) ? | |
407 | NULL : subreq->wb_this_page; | |
408 | ||
409 | WARN_ON_ONCE(old_head != subreq->wb_head); | |
410 | ||
411 | /* make sure old group is not used */ | |
d4581383 WAA |
412 | subreq->wb_this_page = subreq; |
413 | ||
902a4c00 TM |
414 | clear_bit(PG_REMOVE, &subreq->wb_flags); |
415 | ||
5b2b5187 TM |
416 | /* Note: races with nfs_page_group_destroy() */ |
417 | if (!kref_read(&subreq->wb_kref)) { | |
5b2b5187 | 418 | /* Check if we raced with nfs_page_group_destroy() */ |
902a4c00 | 419 | if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) |
5b2b5187 TM |
420 | nfs_free_request(subreq); |
421 | continue; | |
422 | } | |
d4581383 | 423 | |
5b2b5187 TM |
424 | subreq->wb_head = subreq; |
425 | ||
426 | if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { | |
427 | nfs_release_request(subreq); | |
a6b6d5b8 | 428 | atomic_long_dec(&NFS_I(inode)->nrequests); |
d4581383 | 429 | } |
5b2b5187 | 430 | |
5b2b5187 TM |
431 | /* subreq is now totally disconnected from page group or any |
432 | * write / commit lists. last chance to wake any waiters */ | |
433 | nfs_unlock_and_release_request(subreq); | |
d4581383 WAA |
434 | } |
435 | } | |
436 | ||
437 | /* | |
438 | * nfs_lock_and_join_requests - join all subreqs to the head req and return | |
439 | * a locked reference, cancelling any pending | |
440 | * operations for this page. | |
441 | * | |
442 | * @page - the page used to lookup the "page group" of nfs_page structures | |
d4581383 WAA |
443 | * |
444 | * This function joins all sub requests to the head request by first | |
445 | * locking all requests in the group, cancelling any pending operations | |
446 | * and finally updating the head request to cover the whole range covered by | |
447 | * the (former) group. All subrequests are removed from any write or commit | |
448 | * lists, unlinked from the group and destroyed. | |
449 | * | |
450 | * Returns a locked, referenced pointer to the head request - which after | |
451 | * this call is guaranteed to be the only request associated with the page. | |
452 | * Returns NULL if no requests are found for @page, or a ERR_PTR if an | |
453 | * error was encountered. | |
454 | */ | |
455 | static struct nfs_page * | |
6d17d653 | 456 | nfs_lock_and_join_requests(struct page *page) |
e261f51f | 457 | { |
d56b4ddf | 458 | struct inode *inode = page_file_mapping(page)->host; |
d4581383 WAA |
459 | struct nfs_page *head, *subreq; |
460 | struct nfs_page *destroy_list = NULL; | |
461 | unsigned int total_bytes; | |
e261f51f TM |
462 | int ret; |
463 | ||
d4581383 | 464 | try_again: |
d4581383 WAA |
465 | /* |
466 | * A reference is taken only on the head request which acts as a | |
467 | * reference to the whole page group - the group will not be destroyed | |
468 | * until the head reference is released. | |
469 | */ | |
bd37d6fc TM |
470 | head = nfs_page_find_head_request(page); |
471 | if (!head) | |
d4581383 | 472 | return NULL; |
d4581383 | 473 | |
a0e265bc TM |
474 | /* lock the page head first in order to avoid an ABBA inefficiency */ |
475 | if (!nfs_lock_request(head)) { | |
a0e265bc TM |
476 | ret = nfs_wait_on_request(head); |
477 | nfs_release_request(head); | |
478 | if (ret < 0) | |
479 | return ERR_PTR(ret); | |
480 | goto try_again; | |
481 | } | |
bd37d6fc TM |
482 | |
483 | /* Ensure that nobody removed the request before we locked it */ | |
484 | if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { | |
485 | nfs_unlock_and_release_request(head); | |
486 | goto try_again; | |
487 | } | |
a0e265bc | 488 | |
1344b7ea | 489 | ret = nfs_page_group_lock(head); |
0671d8f1 ME |
490 | if (ret < 0) |
491 | goto release_request; | |
7c3af975 WAA |
492 | |
493 | /* lock each request in the page group */ | |
a0e265bc TM |
494 | total_bytes = head->wb_bytes; |
495 | for (subreq = head->wb_this_page; subreq != head; | |
496 | subreq = subreq->wb_this_page) { | |
74a6d4b5 | 497 | |
1bd5d6d0 TM |
498 | if (!kref_get_unless_zero(&subreq->wb_kref)) { |
499 | if (subreq->wb_offset == head->wb_offset + total_bytes) | |
500 | total_bytes += subreq->wb_bytes; | |
5b2b5187 | 501 | continue; |
1bd5d6d0 TM |
502 | } |
503 | ||
74a6d4b5 | 504 | while (!nfs_lock_request(subreq)) { |
b5bab9bf | 505 | /* |
74a6d4b5 TM |
506 | * Unlock page to allow nfs_page_group_sync_on_bit() |
507 | * to succeed | |
b5bab9bf | 508 | */ |
74a6d4b5 TM |
509 | nfs_page_group_unlock(head); |
510 | ret = nfs_wait_on_request(subreq); | |
511 | if (!ret) | |
1344b7ea | 512 | ret = nfs_page_group_lock(head); |
74a6d4b5 TM |
513 | if (ret < 0) { |
514 | nfs_unroll_locks(inode, head, subreq); | |
5b2b5187 | 515 | nfs_release_request(subreq); |
0671d8f1 | 516 | goto release_request; |
74a6d4b5 | 517 | } |
e14bebf6 | 518 | } |
d4581383 WAA |
519 | /* |
520 | * Subrequests are always contiguous, non overlapping | |
309a1d65 | 521 | * and in order - but may be repeated (mirrored writes). |
d4581383 | 522 | */ |
309a1d65 WAA |
523 | if (subreq->wb_offset == (head->wb_offset + total_bytes)) { |
524 | /* keep track of how many bytes this group covers */ | |
525 | total_bytes += subreq->wb_bytes; | |
526 | } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || | |
527 | ((subreq->wb_offset + subreq->wb_bytes) > | |
528 | (head->wb_offset + total_bytes)))) { | |
8b77484f | 529 | nfs_page_group_unlock(head); |
74a6d4b5 | 530 | nfs_unroll_locks(inode, head, subreq); |
5b2b5187 | 531 | nfs_unlock_and_release_request(subreq); |
0671d8f1 ME |
532 | ret = -EIO; |
533 | goto release_request; | |
309a1d65 | 534 | } |
a0e265bc | 535 | } |
d4581383 WAA |
536 | |
537 | /* Now that all requests are locked, make sure they aren't on any list. | |
538 | * Commit list removal accounting is done after locks are dropped */ | |
539 | subreq = head; | |
540 | do { | |
411a99ad | 541 | nfs_clear_request_commit(subreq); |
d4581383 WAA |
542 | subreq = subreq->wb_this_page; |
543 | } while (subreq != head); | |
544 | ||
545 | /* unlink subrequests from head, destroy them later */ | |
546 | if (head->wb_this_page != head) { | |
547 | /* destroy list will be terminated by head */ | |
548 | destroy_list = head->wb_this_page; | |
549 | head->wb_this_page = head; | |
550 | ||
551 | /* change head request to cover whole range that | |
552 | * the former page group covered */ | |
553 | head->wb_bytes = total_bytes; | |
e261f51f | 554 | } |
d4581383 | 555 | |
b66aaa8d TM |
556 | /* Postpone destruction of this request */ |
557 | if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { | |
558 | set_bit(PG_INODE_REF, &head->wb_flags); | |
559 | kref_get(&head->wb_kref); | |
a6b6d5b8 | 560 | atomic_long_inc(&NFS_I(inode)->nrequests); |
b66aaa8d TM |
561 | } |
562 | ||
d4581383 WAA |
563 | nfs_page_group_unlock(head); |
564 | ||
b66aaa8d | 565 | nfs_destroy_unlinked_subrequests(destroy_list, head, inode); |
d4581383 | 566 | |
b5bab9bf TM |
567 | /* Did we lose a race with nfs_inode_remove_request()? */ |
568 | if (!(PagePrivate(page) || PageSwapCache(page))) { | |
569 | nfs_unlock_and_release_request(head); | |
570 | return NULL; | |
571 | } | |
572 | ||
bd37d6fc | 573 | /* still holds ref on head from nfs_page_find_head_request |
d4581383 WAA |
574 | * and still has lock on head from lock loop */ |
575 | return head; | |
0671d8f1 ME |
576 | |
577 | release_request: | |
578 | nfs_unlock_and_release_request(head); | |
579 | return ERR_PTR(ret); | |
074cc1de TM |
580 | } |
581 | ||
0bcbf039 PT |
582 | static void nfs_write_error_remove_page(struct nfs_page *req) |
583 | { | |
0bcbf039 | 584 | nfs_end_page_writeback(req); |
0bcbf039 PT |
585 | generic_error_remove_page(page_file_mapping(req->wb_page), |
586 | req->wb_page); | |
1f84ccdf | 587 | nfs_release_request(req); |
0bcbf039 PT |
588 | } |
589 | ||
a6598813 TM |
590 | static bool |
591 | nfs_error_is_fatal_on_server(int err) | |
592 | { | |
593 | switch (err) { | |
594 | case 0: | |
595 | case -ERESTARTSYS: | |
596 | case -EINTR: | |
597 | return false; | |
598 | } | |
599 | return nfs_error_is_fatal(err); | |
0bcbf039 PT |
600 | } |
601 | ||
074cc1de TM |
602 | /* |
603 | * Find an associated nfs write request, and prepare to flush it out | |
604 | * May return an error if the user signalled nfs_wait_on_request(). | |
605 | */ | |
606 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
6d17d653 | 607 | struct page *page) |
074cc1de TM |
608 | { |
609 | struct nfs_page *req; | |
610 | int ret = 0; | |
611 | ||
6d17d653 | 612 | req = nfs_lock_and_join_requests(page); |
074cc1de TM |
613 | if (!req) |
614 | goto out; | |
615 | ret = PTR_ERR(req); | |
616 | if (IS_ERR(req)) | |
617 | goto out; | |
618 | ||
deed85e7 TM |
619 | nfs_set_page_writeback(page); |
620 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 621 | |
deed85e7 | 622 | ret = 0; |
a6598813 TM |
623 | /* If there is a fatal error that covers this write, just exit */ |
624 | if (nfs_error_is_fatal_on_server(req->wb_context->error)) | |
625 | goto out_launder; | |
626 | ||
f8512ad0 | 627 | if (!nfs_pageio_add_request(pgio, req)) { |
074cc1de | 628 | ret = pgio->pg_error; |
0bcbf039 | 629 | /* |
c373fff7 | 630 | * Remove the problematic req upon fatal errors on the server |
0bcbf039 PT |
631 | */ |
632 | if (nfs_error_is_fatal(ret)) { | |
633 | nfs_context_set_write_error(req->wb_context, ret); | |
c373fff7 | 634 | if (nfs_error_is_fatal_on_server(ret)) |
a6598813 | 635 | goto out_launder; |
0bcbf039 | 636 | } |
d6c843b9 PT |
637 | nfs_redirty_request(req); |
638 | ret = -EAGAIN; | |
40f90271 TM |
639 | } else |
640 | nfs_add_stats(page_file_mapping(page)->host, | |
641 | NFSIOS_WRITEPAGES, 1); | |
074cc1de TM |
642 | out: |
643 | return ret; | |
a6598813 TM |
644 | out_launder: |
645 | nfs_write_error_remove_page(req); | |
646 | return ret; | |
e261f51f TM |
647 | } |
648 | ||
d6c843b9 | 649 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, |
c373fff7 | 650 | struct nfs_pageio_descriptor *pgio) |
1da177e4 | 651 | { |
cfb506e1 | 652 | int ret; |
1da177e4 | 653 | |
8cd79788 | 654 | nfs_pageio_cond_complete(pgio, page_index(page)); |
6d17d653 | 655 | ret = nfs_page_async_flush(pgio, page); |
cfb506e1 TM |
656 | if (ret == -EAGAIN) { |
657 | redirty_page_for_writepage(wbc, page); | |
658 | ret = 0; | |
659 | } | |
660 | return ret; | |
f758c885 | 661 | } |
7fe7f848 | 662 | |
f758c885 TM |
663 | /* |
664 | * Write an mmapped page to the server. | |
665 | */ | |
d6c843b9 | 666 | static int nfs_writepage_locked(struct page *page, |
c373fff7 | 667 | struct writeback_control *wbc) |
f758c885 TM |
668 | { |
669 | struct nfs_pageio_descriptor pgio; | |
40f90271 | 670 | struct inode *inode = page_file_mapping(page)->host; |
f758c885 | 671 | int err; |
49a70f27 | 672 | |
40f90271 | 673 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
811ed92e | 674 | nfs_pageio_init_write(&pgio, inode, 0, |
a20c93e3 | 675 | false, &nfs_async_write_completion_ops); |
c373fff7 | 676 | err = nfs_do_writepage(page, wbc, &pgio); |
f758c885 TM |
677 | nfs_pageio_complete(&pgio); |
678 | if (err < 0) | |
679 | return err; | |
680 | if (pgio.pg_error < 0) | |
681 | return pgio.pg_error; | |
682 | return 0; | |
4d770ccf TM |
683 | } |
684 | ||
685 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
686 | { | |
f758c885 | 687 | int ret; |
4d770ccf | 688 | |
c373fff7 | 689 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 690 | unlock_page(page); |
f758c885 TM |
691 | return ret; |
692 | } | |
693 | ||
694 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
695 | { | |
696 | int ret; | |
697 | ||
c373fff7 | 698 | ret = nfs_do_writepage(page, wbc, data); |
f758c885 TM |
699 | unlock_page(page); |
700 | return ret; | |
1da177e4 LT |
701 | } |
702 | ||
919e3bd9 TM |
703 | static void nfs_io_completion_commit(void *inode) |
704 | { | |
705 | nfs_commit_inode(inode, 0); | |
706 | } | |
707 | ||
1da177e4 LT |
708 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
709 | { | |
1da177e4 | 710 | struct inode *inode = mapping->host; |
c63c7b05 | 711 | struct nfs_pageio_descriptor pgio; |
919e3bd9 | 712 | struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); |
1da177e4 LT |
713 | int err; |
714 | ||
91d5b470 CL |
715 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
716 | ||
919e3bd9 TM |
717 | if (ioc) |
718 | nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); | |
719 | ||
a20c93e3 CH |
720 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
721 | &nfs_async_write_completion_ops); | |
919e3bd9 | 722 | pgio.pg_io_completion = ioc; |
f758c885 | 723 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 724 | nfs_pageio_complete(&pgio); |
919e3bd9 | 725 | nfs_io_completion_put(ioc); |
72cb77f4 | 726 | |
f758c885 | 727 | if (err < 0) |
72cb77f4 TM |
728 | goto out_err; |
729 | err = pgio.pg_error; | |
730 | if (err < 0) | |
731 | goto out_err; | |
c63c7b05 | 732 | return 0; |
72cb77f4 TM |
733 | out_err: |
734 | return err; | |
1da177e4 LT |
735 | } |
736 | ||
737 | /* | |
738 | * Insert a write request into an inode | |
739 | */ | |
d6d6dc7c | 740 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 | 741 | { |
4b9bb25b | 742 | struct address_space *mapping = page_file_mapping(req->wb_page); |
1da177e4 | 743 | struct nfs_inode *nfsi = NFS_I(inode); |
e7d39069 | 744 | |
2bfc6e56 WAA |
745 | WARN_ON_ONCE(req->wb_this_page != req); |
746 | ||
e7d39069 | 747 | /* Lock the request! */ |
7ad84aa9 | 748 | nfs_lock_request(req); |
e7d39069 | 749 | |
29418aa4 MG |
750 | /* |
751 | * Swap-space should not get truncated. Hence no need to plug the race | |
752 | * with invalidate/truncate. | |
753 | */ | |
4b9bb25b TM |
754 | spin_lock(&mapping->private_lock); |
755 | if (!nfs_have_writebacks(inode) && | |
756 | NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) { | |
757 | spin_lock(&inode->i_lock); | |
758 | inode->i_version++; | |
759 | spin_unlock(&inode->i_lock); | |
760 | } | |
29418aa4 MG |
761 | if (likely(!PageSwapCache(req->wb_page))) { |
762 | set_bit(PG_MAPPED, &req->wb_flags); | |
763 | SetPagePrivate(req->wb_page); | |
764 | set_page_private(req->wb_page, (unsigned long)req); | |
765 | } | |
4b9bb25b | 766 | spin_unlock(&mapping->private_lock); |
a6b6d5b8 | 767 | atomic_long_inc(&nfsi->nrequests); |
17089a29 | 768 | /* this a head request for a page group - mark it as having an |
cb1410c7 WAA |
769 | * extra reference so sub groups can follow suit. |
770 | * This flag also informs pgio layer when to bump nrequests when | |
771 | * adding subrequests. */ | |
17089a29 | 772 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); |
c03b4024 | 773 | kref_get(&req->wb_kref); |
1da177e4 LT |
774 | } |
775 | ||
776 | /* | |
89a09141 | 777 | * Remove a write request from an inode |
1da177e4 LT |
778 | */ |
779 | static void nfs_inode_remove_request(struct nfs_page *req) | |
780 | { | |
4b9bb25b TM |
781 | struct address_space *mapping = page_file_mapping(req->wb_page); |
782 | struct inode *inode = mapping->host; | |
1da177e4 | 783 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 784 | struct nfs_page *head; |
1da177e4 | 785 | |
a6b6d5b8 | 786 | atomic_long_dec(&nfsi->nrequests); |
20633f04 WAA |
787 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
788 | head = req->wb_head; | |
789 | ||
4b9bb25b | 790 | spin_lock(&mapping->private_lock); |
67911c8f | 791 | if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { |
20633f04 WAA |
792 | set_page_private(head->wb_page, 0); |
793 | ClearPagePrivate(head->wb_page); | |
794 | clear_bit(PG_MAPPED, &head->wb_flags); | |
795 | } | |
4b9bb25b | 796 | spin_unlock(&mapping->private_lock); |
29418aa4 | 797 | } |
17089a29 WAA |
798 | |
799 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) | |
800 | nfs_release_request(req); | |
1da177e4 LT |
801 | } |
802 | ||
61822ab5 | 803 | static void |
6d884e8f | 804 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 805 | { |
67911c8f AS |
806 | if (req->wb_page) |
807 | __set_page_dirty_nobuffers(req->wb_page); | |
61822ab5 TM |
808 | } |
809 | ||
3a3908c8 TM |
810 | /* |
811 | * nfs_page_search_commits_for_head_request_locked | |
812 | * | |
813 | * Search through commit lists on @inode for the head request for @page. | |
814 | * Must be called while holding the inode (which is cinfo) lock. | |
815 | * | |
816 | * Returns the head request if found, or NULL if not found. | |
817 | */ | |
818 | static struct nfs_page * | |
819 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
820 | struct page *page) | |
821 | { | |
822 | struct nfs_page *freq, *t; | |
823 | struct nfs_commit_info cinfo; | |
824 | struct inode *inode = &nfsi->vfs_inode; | |
825 | ||
826 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
827 | ||
828 | /* search through pnfs commit lists */ | |
829 | freq = pnfs_search_commit_reqs(inode, &cinfo, page); | |
830 | if (freq) | |
831 | return freq->wb_head; | |
832 | ||
833 | /* Linearly search the commit list for the correct request */ | |
834 | list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { | |
835 | if (freq->wb_page == page) | |
836 | return freq->wb_head; | |
837 | } | |
838 | ||
839 | return NULL; | |
840 | } | |
841 | ||
86d80f97 TM |
842 | /** |
843 | * nfs_request_add_commit_list_locked - add request to a commit list | |
844 | * @req: pointer to a struct nfs_page | |
845 | * @dst: commit list head | |
846 | * @cinfo: holds list lock and accounting info | |
847 | * | |
848 | * This sets the PG_CLEAN bit, updates the cinfo count of | |
849 | * number of outstanding requests requiring a commit as well as | |
850 | * the MM page stats. | |
851 | * | |
e824f99a TM |
852 | * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the |
853 | * nfs_page lock. | |
86d80f97 TM |
854 | */ |
855 | void | |
856 | nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, | |
857 | struct nfs_commit_info *cinfo) | |
858 | { | |
859 | set_bit(PG_CLEAN, &req->wb_flags); | |
860 | nfs_list_add_request(req, dst); | |
5cb953d4 | 861 | atomic_long_inc(&cinfo->mds->ncommit); |
86d80f97 TM |
862 | } |
863 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); | |
864 | ||
8dd37758 TM |
865 | /** |
866 | * nfs_request_add_commit_list - add request to a commit list | |
867 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
868 | * @dst: commit list head |
869 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 870 | * |
ea2cf228 | 871 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
872 | * number of outstanding requests requiring a commit as well as |
873 | * the MM page stats. | |
874 | * | |
ea2cf228 | 875 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 876 | * holding the nfs_page lock. |
1da177e4 | 877 | */ |
8dd37758 | 878 | void |
6272dcc6 | 879 | nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) |
1da177e4 | 880 | { |
e824f99a | 881 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
6272dcc6 | 882 | nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); |
e824f99a | 883 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
67911c8f AS |
884 | if (req->wb_page) |
885 | nfs_mark_page_unstable(req->wb_page, cinfo); | |
1da177e4 | 886 | } |
8dd37758 TM |
887 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
888 | ||
889 | /** | |
890 | * nfs_request_remove_commit_list - Remove request from a commit list | |
891 | * @req: pointer to a nfs_page | |
ea2cf228 | 892 | * @cinfo: holds list lock and accounting info |
8dd37758 | 893 | * |
ea2cf228 | 894 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
895 | * number of outstanding requests requiring a commit |
896 | * It does not update the MM page stats. | |
897 | * | |
ea2cf228 | 898 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
899 | */ |
900 | void | |
ea2cf228 FI |
901 | nfs_request_remove_commit_list(struct nfs_page *req, |
902 | struct nfs_commit_info *cinfo) | |
8dd37758 | 903 | { |
8dd37758 TM |
904 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
905 | return; | |
906 | nfs_list_remove_request(req); | |
5cb953d4 | 907 | atomic_long_dec(&cinfo->mds->ncommit); |
8dd37758 TM |
908 | } |
909 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
910 | ||
ea2cf228 FI |
911 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
912 | struct inode *inode) | |
913 | { | |
fe238e60 | 914 | cinfo->inode = inode; |
ea2cf228 FI |
915 | cinfo->mds = &NFS_I(inode)->commit_info; |
916 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 917 | cinfo->dreq = NULL; |
f453a54a | 918 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
919 | } |
920 | ||
921 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
922 | struct inode *inode, | |
923 | struct nfs_direct_req *dreq) | |
924 | { | |
1763da12 FI |
925 | if (dreq) |
926 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
927 | else | |
928 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
929 | } |
930 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
931 | |
932 | /* | |
933 | * Add a request to the inode's commit list. | |
934 | */ | |
1763da12 | 935 | void |
ea2cf228 | 936 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
b57ff130 | 937 | struct nfs_commit_info *cinfo, u32 ds_commit_idx) |
8dd37758 | 938 | { |
b57ff130 | 939 | if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) |
8dd37758 | 940 | return; |
6272dcc6 | 941 | nfs_request_add_commit_list(req, cinfo); |
8dd37758 | 942 | } |
8e821cad | 943 | |
d6d6dc7c FI |
944 | static void |
945 | nfs_clear_page_commit(struct page *page) | |
946 | { | |
11fb9989 | 947 | dec_node_page_state(page, NR_UNSTABLE_NFS); |
93f78d88 TH |
948 | dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, |
949 | WB_RECLAIMABLE); | |
d6d6dc7c FI |
950 | } |
951 | ||
b5bab9bf | 952 | /* Called holding the request lock on @req */ |
8dd37758 | 953 | static void |
e468bae9 TM |
954 | nfs_clear_request_commit(struct nfs_page *req) |
955 | { | |
8dd37758 | 956 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
2b0143b5 | 957 | struct inode *inode = d_inode(req->wb_context->dentry); |
ea2cf228 | 958 | struct nfs_commit_info cinfo; |
e468bae9 | 959 | |
ea2cf228 | 960 | nfs_init_cinfo_from_inode(&cinfo, inode); |
e824f99a | 961 | mutex_lock(&NFS_I(inode)->commit_mutex); |
ea2cf228 | 962 | if (!pnfs_clear_request_commit(req, &cinfo)) { |
ea2cf228 | 963 | nfs_request_remove_commit_list(req, &cinfo); |
8dd37758 | 964 | } |
e824f99a | 965 | mutex_unlock(&NFS_I(inode)->commit_mutex); |
d6d6dc7c | 966 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 967 | } |
e468bae9 TM |
968 | } |
969 | ||
d45f60c6 | 970 | int nfs_write_need_commit(struct nfs_pgio_header *hdr) |
8e821cad | 971 | { |
c65e6254 | 972 | if (hdr->verf.committed == NFS_DATA_SYNC) |
d45f60c6 | 973 | return hdr->lseg == NULL; |
c65e6254 | 974 | return hdr->verf.committed != NFS_FILE_SYNC; |
8e821cad TM |
975 | } |
976 | ||
919e3bd9 TM |
977 | static void nfs_async_write_init(struct nfs_pgio_header *hdr) |
978 | { | |
979 | nfs_io_completion_get(hdr->io_completion); | |
980 | } | |
981 | ||
061ae2ed | 982 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 983 | { |
ea2cf228 | 984 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
985 | unsigned long bytes = 0; |
986 | ||
987 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
988 | goto out; | |
ea2cf228 | 989 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
990 | while (!list_empty(&hdr->pages)) { |
991 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
992 | |
993 | bytes += req->wb_bytes; | |
994 | nfs_list_remove_request(req); | |
995 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
996 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 997 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
998 | nfs_context_set_write_error(req->wb_context, hdr->error); |
999 | goto remove_req; | |
1000 | } | |
c65e6254 | 1001 | if (nfs_write_need_commit(hdr)) { |
f79d06f5 | 1002 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
b57ff130 | 1003 | nfs_mark_request_commit(req, hdr->lseg, &cinfo, |
a7d42ddb | 1004 | hdr->pgio_mirror_idx); |
6c75dc0d FI |
1005 | goto next; |
1006 | } | |
1007 | remove_req: | |
1008 | nfs_inode_remove_request(req); | |
1009 | next: | |
20633f04 | 1010 | nfs_end_page_writeback(req); |
3aff4ebb | 1011 | nfs_release_request(req); |
6c75dc0d FI |
1012 | } |
1013 | out: | |
919e3bd9 | 1014 | nfs_io_completion_put(hdr->io_completion); |
6c75dc0d | 1015 | hdr->release(hdr); |
8e821cad | 1016 | } |
1da177e4 | 1017 | |
ce59515c | 1018 | unsigned long |
ea2cf228 | 1019 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 1020 | { |
5cb953d4 | 1021 | return atomic_long_read(&cinfo->mds->ncommit); |
d6d6dc7c FI |
1022 | } |
1023 | ||
e824f99a | 1024 | /* NFS_I(cinfo->inode)->commit_mutex held by caller */ |
1763da12 | 1025 | int |
ea2cf228 FI |
1026 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
1027 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c | 1028 | { |
137da553 | 1029 | struct nfs_page *req, *tmp; |
d6d6dc7c FI |
1030 | int ret = 0; |
1031 | ||
137da553 TM |
1032 | restart: |
1033 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
7ad84aa9 | 1034 | kref_get(&req->wb_kref); |
2ce209c4 TM |
1035 | if (!nfs_lock_request(req)) { |
1036 | int status; | |
137da553 TM |
1037 | |
1038 | /* Prevent deadlock with nfs_lock_and_join_requests */ | |
1039 | if (!list_empty(dst)) { | |
1040 | nfs_release_request(req); | |
1041 | continue; | |
1042 | } | |
1043 | /* Ensure we make progress to prevent livelock */ | |
2ce209c4 TM |
1044 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
1045 | status = nfs_wait_on_request(req); | |
1046 | nfs_release_request(req); | |
1047 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); | |
1048 | if (status < 0) | |
1049 | break; | |
137da553 | 1050 | goto restart; |
2ce209c4 | 1051 | } |
ea2cf228 | 1052 | nfs_request_remove_commit_list(req, cinfo); |
5d2a9d9d | 1053 | clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); |
8dd37758 TM |
1054 | nfs_list_add_request(req, dst); |
1055 | ret++; | |
1763da12 | 1056 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 1057 | break; |
e824f99a | 1058 | cond_resched(); |
d6d6dc7c FI |
1059 | } |
1060 | return ret; | |
fb8a1f11 | 1061 | } |
5d2a9d9d | 1062 | EXPORT_SYMBOL_GPL(nfs_scan_commit_list); |
fb8a1f11 | 1063 | |
1da177e4 LT |
1064 | /* |
1065 | * nfs_scan_commit - Scan an inode for commit requests | |
1066 | * @inode: NFS inode to scan | |
ea2cf228 FI |
1067 | * @dst: mds destination list |
1068 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
1069 | * |
1070 | * Moves requests from the inode's 'commit' request list. | |
1071 | * The requests are *not* checked to ensure that they form a contiguous set. | |
1072 | */ | |
1763da12 | 1073 | int |
ea2cf228 FI |
1074 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
1075 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1076 | { |
d6d6dc7c | 1077 | int ret = 0; |
fb8a1f11 | 1078 | |
5cb953d4 TM |
1079 | if (!atomic_long_read(&cinfo->mds->ncommit)) |
1080 | return 0; | |
e824f99a | 1081 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
5cb953d4 | 1082 | if (atomic_long_read(&cinfo->mds->ncommit) > 0) { |
8dd37758 | 1083 | const int max = INT_MAX; |
d6d6dc7c | 1084 | |
ea2cf228 FI |
1085 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
1086 | cinfo, max); | |
1087 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 1088 | } |
e824f99a | 1089 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
ff778d02 | 1090 | return ret; |
1da177e4 | 1091 | } |
d6d6dc7c | 1092 | |
1da177e4 | 1093 | /* |
e7d39069 TM |
1094 | * Search for an existing write request, and attempt to update |
1095 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 1096 | * |
e7d39069 TM |
1097 | * If the attempt fails, then the existing request is flushed out |
1098 | * to disk. | |
1da177e4 | 1099 | */ |
e7d39069 TM |
1100 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
1101 | struct page *page, | |
1102 | unsigned int offset, | |
1103 | unsigned int bytes) | |
1da177e4 | 1104 | { |
e7d39069 TM |
1105 | struct nfs_page *req; |
1106 | unsigned int rqend; | |
1107 | unsigned int end; | |
1108 | int error; | |
1109 | ||
1da177e4 LT |
1110 | end = offset + bytes; |
1111 | ||
f6032f21 TM |
1112 | req = nfs_lock_and_join_requests(page); |
1113 | if (IS_ERR_OR_NULL(req)) | |
1114 | return req; | |
1da177e4 | 1115 | |
f6032f21 TM |
1116 | rqend = req->wb_offset + req->wb_bytes; |
1117 | /* | |
1118 | * Tell the caller to flush out the request if | |
1119 | * the offsets are non-contiguous. | |
1120 | * Note: nfs_flush_incompatible() will already | |
1121 | * have flushed out requests having wrong owners. | |
1122 | */ | |
1123 | if (offset > rqend || end < req->wb_offset) | |
1124 | goto out_flushme; | |
1da177e4 LT |
1125 | |
1126 | /* Okay, the request matches. Update the region */ | |
1127 | if (offset < req->wb_offset) { | |
1128 | req->wb_offset = offset; | |
1129 | req->wb_pgbase = offset; | |
1da177e4 | 1130 | } |
1da177e4 LT |
1131 | if (end > rqend) |
1132 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1133 | else |
1134 | req->wb_bytes = rqend - req->wb_offset; | |
e7d39069 TM |
1135 | return req; |
1136 | out_flushme: | |
f6032f21 TM |
1137 | /* |
1138 | * Note: we mark the request dirty here because | |
1139 | * nfs_lock_and_join_requests() cannot preserve | |
1140 | * commit flags, so we have to replay the write. | |
1141 | */ | |
1142 | nfs_mark_request_dirty(req); | |
1143 | nfs_unlock_and_release_request(req); | |
e7d39069 | 1144 | error = nfs_wb_page(inode, page); |
f6032f21 | 1145 | return (error < 0) ? ERR_PTR(error) : NULL; |
e7d39069 TM |
1146 | } |
1147 | ||
1148 | /* | |
1149 | * Try to update an existing write request, or create one if there is none. | |
1150 | * | |
1151 | * Note: Should always be called with the Page Lock held to prevent races | |
1152 | * if we have to add a new request. Also assumes that the caller has | |
1153 | * already called nfs_flush_incompatible() if necessary. | |
1154 | */ | |
1155 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1156 | struct page *page, unsigned int offset, unsigned int bytes) | |
1157 | { | |
d56b4ddf | 1158 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1159 | struct nfs_page *req; |
1da177e4 | 1160 | |
e7d39069 TM |
1161 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1162 | if (req != NULL) | |
1163 | goto out; | |
2bfc6e56 | 1164 | req = nfs_create_request(ctx, page, NULL, offset, bytes); |
e7d39069 TM |
1165 | if (IS_ERR(req)) |
1166 | goto out; | |
d6d6dc7c | 1167 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1168 | out: |
61e930a9 | 1169 | return req; |
1da177e4 LT |
1170 | } |
1171 | ||
e7d39069 TM |
1172 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1173 | unsigned int offset, unsigned int count) | |
1174 | { | |
1175 | struct nfs_page *req; | |
1176 | ||
1177 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1178 | if (IS_ERR(req)) | |
1179 | return PTR_ERR(req); | |
1180 | /* Update file length */ | |
1181 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1182 | nfs_mark_uptodate(req); |
a6305ddb | 1183 | nfs_mark_request_dirty(req); |
1d1afcbc | 1184 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1185 | return 0; |
1186 | } | |
1187 | ||
1da177e4 LT |
1188 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1189 | { | |
cd3758e3 | 1190 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1191 | struct nfs_lock_context *l_ctx; |
bd61e0a9 | 1192 | struct file_lock_context *flctx = file_inode(file)->i_flctx; |
1da177e4 | 1193 | struct nfs_page *req; |
1a54533e | 1194 | int do_flush, status; |
1da177e4 LT |
1195 | /* |
1196 | * Look for a request corresponding to this page. If there | |
1197 | * is one, and it belongs to another file, we flush it out | |
1198 | * before we try to copy anything into the page. Do this | |
1199 | * due to the lack of an ACCESS-type call in NFSv2. | |
1200 | * Also do the same if we find a request from an existing | |
1201 | * dropped page. | |
1202 | */ | |
1a54533e | 1203 | do { |
84d3a9a9 | 1204 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1205 | if (req == NULL) |
1206 | return 0; | |
2a369153 | 1207 | l_ctx = req->wb_lock_context; |
138a2935 TM |
1208 | do_flush = req->wb_page != page || |
1209 | !nfs_match_open_context(req->wb_context, ctx); | |
bd61e0a9 JL |
1210 | if (l_ctx && flctx && |
1211 | !(list_empty_careful(&flctx->flc_posix) && | |
1212 | list_empty_careful(&flctx->flc_flock))) { | |
d51fdb87 | 1213 | do_flush |= l_ctx->lockowner != current->files; |
5263e31e | 1214 | } |
1da177e4 | 1215 | nfs_release_request(req); |
1a54533e TM |
1216 | if (!do_flush) |
1217 | return 0; | |
d56b4ddf | 1218 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1219 | } while (status == 0); |
1220 | return status; | |
1da177e4 LT |
1221 | } |
1222 | ||
dc24826b AA |
1223 | /* |
1224 | * Avoid buffered writes when a open context credential's key would | |
1225 | * expire soon. | |
1226 | * | |
1227 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1228 | * | |
1229 | * Return 0 and set a credential flag which triggers the inode to flush | |
1230 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1231 | * RPC_KEY_EXPIRE_TIMEO. | |
1232 | */ | |
1233 | int | |
1234 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1235 | { | |
1236 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
1237 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
1238 | ||
1239 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
1240 | } | |
1241 | ||
1242 | /* | |
1243 | * Test if the open context credential key is marked to expire soon. | |
1244 | */ | |
ce52914e | 1245 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) |
dc24826b | 1246 | { |
ce52914e SM |
1247 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; |
1248 | ||
1249 | return rpcauth_cred_key_to_expire(auth, ctx->cred); | |
dc24826b AA |
1250 | } |
1251 | ||
5d47a356 TM |
1252 | /* |
1253 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1254 | * the PageUptodate() flag. In this case, we will need to turn off | |
1255 | * write optimisations that depend on the page contents being correct. | |
1256 | */ | |
8d197a56 | 1257 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 1258 | { |
d529ef83 JL |
1259 | struct nfs_inode *nfsi = NFS_I(inode); |
1260 | ||
8d197a56 TM |
1261 | if (nfs_have_delegated_attributes(inode)) |
1262 | goto out; | |
18dd78c4 | 1263 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
d529ef83 | 1264 | return false; |
4db72b40 | 1265 | smp_rmb(); |
d529ef83 | 1266 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
1267 | return false; |
1268 | out: | |
18dd78c4 SM |
1269 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
1270 | return false; | |
8d197a56 | 1271 | return PageUptodate(page) != 0; |
5d47a356 TM |
1272 | } |
1273 | ||
5263e31e JL |
1274 | static bool |
1275 | is_whole_file_wrlock(struct file_lock *fl) | |
1276 | { | |
1277 | return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && | |
1278 | fl->fl_type == F_WRLCK; | |
1279 | } | |
1280 | ||
c7559663 SM |
1281 | /* If we know the page is up to date, and we're not using byte range locks (or |
1282 | * if we have the whole file locked for writing), it may be more efficient to | |
1283 | * extend the write to cover the entire page in order to avoid fragmentation | |
1284 | * inefficiencies. | |
1285 | * | |
263b4509 SM |
1286 | * If the file is opened for synchronous writes then we can just skip the rest |
1287 | * of the checks. | |
c7559663 SM |
1288 | */ |
1289 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
1290 | { | |
5263e31e JL |
1291 | int ret; |
1292 | struct file_lock_context *flctx = inode->i_flctx; | |
1293 | struct file_lock *fl; | |
1294 | ||
c7559663 SM |
1295 | if (file->f_flags & O_DSYNC) |
1296 | return 0; | |
263b4509 SM |
1297 | if (!nfs_write_pageuptodate(page, inode)) |
1298 | return 0; | |
c7559663 SM |
1299 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1300 | return 1; | |
bd61e0a9 JL |
1301 | if (!flctx || (list_empty_careful(&flctx->flc_flock) && |
1302 | list_empty_careful(&flctx->flc_posix))) | |
8fa4592a | 1303 | return 1; |
5263e31e JL |
1304 | |
1305 | /* Check to see if there are whole file write locks */ | |
5263e31e | 1306 | ret = 0; |
6109c850 | 1307 | spin_lock(&flctx->flc_lock); |
bd61e0a9 JL |
1308 | if (!list_empty(&flctx->flc_posix)) { |
1309 | fl = list_first_entry(&flctx->flc_posix, struct file_lock, | |
1310 | fl_list); | |
1311 | if (is_whole_file_wrlock(fl)) | |
1312 | ret = 1; | |
1313 | } else if (!list_empty(&flctx->flc_flock)) { | |
5263e31e JL |
1314 | fl = list_first_entry(&flctx->flc_flock, struct file_lock, |
1315 | fl_list); | |
1316 | if (fl->fl_type == F_WRLCK) | |
1317 | ret = 1; | |
1318 | } | |
6109c850 | 1319 | spin_unlock(&flctx->flc_lock); |
5263e31e | 1320 | return ret; |
c7559663 SM |
1321 | } |
1322 | ||
1da177e4 LT |
1323 | /* |
1324 | * Update and possibly write a cached page of an NFS file. | |
1325 | * | |
1326 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1327 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1328 | */ | |
1329 | int nfs_updatepage(struct file *file, struct page *page, | |
1330 | unsigned int offset, unsigned int count) | |
1331 | { | |
cd3758e3 | 1332 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 1333 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
1334 | int status = 0; |
1335 | ||
91d5b470 CL |
1336 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1337 | ||
6de1472f AV |
1338 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1339 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1340 | |
149a4fdd BC |
1341 | if (!count) |
1342 | goto out; | |
1343 | ||
c7559663 | 1344 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 1345 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 1346 | offset = 0; |
1da177e4 LT |
1347 | } |
1348 | ||
e21195a7 | 1349 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
1350 | if (status < 0) |
1351 | nfs_set_pageerror(page); | |
59b7c05f TM |
1352 | else |
1353 | __set_page_dirty_nobuffers(page); | |
149a4fdd | 1354 | out: |
48186c7d | 1355 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1356 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1357 | return status; |
1358 | } | |
1359 | ||
3ff7576d | 1360 | static int flush_task_priority(int how) |
1da177e4 LT |
1361 | { |
1362 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1363 | case FLUSH_HIGHPRI: | |
1364 | return RPC_PRIORITY_HIGH; | |
1365 | case FLUSH_LOWPRI: | |
1366 | return RPC_PRIORITY_LOW; | |
1367 | } | |
1368 | return RPC_PRIORITY_NORMAL; | |
1369 | } | |
1370 | ||
d45f60c6 WAA |
1371 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1372 | struct rpc_message *msg, | |
abde71f4 | 1373 | const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 | 1374 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1375 | { |
3ff7576d | 1376 | int priority = flush_task_priority(how); |
d138d5d1 | 1377 | |
1ed26f33 | 1378 | task_setup_data->priority = priority; |
abde71f4 | 1379 | rpc_ops->write_setup(hdr, msg); |
8224b273 CL |
1380 | trace_nfs_initiate_write(hdr->inode, hdr->io_start, hdr->good_bytes, |
1381 | hdr->args.stable); | |
d138d5d1 | 1382 | |
abde71f4 | 1383 | nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, |
d45f60c6 | 1384 | &task_setup_data->rpc_client, msg, hdr); |
275acaaf TM |
1385 | } |
1386 | ||
6d884e8f F |
1387 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1388 | * call this on each, which will prepare them to be retried on next | |
1389 | * writeback using standard nfs. | |
1390 | */ | |
1391 | static void nfs_redirty_request(struct nfs_page *req) | |
1392 | { | |
1393 | nfs_mark_request_dirty(req); | |
c7070113 | 1394 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
20633f04 | 1395 | nfs_end_page_writeback(req); |
3aff4ebb | 1396 | nfs_release_request(req); |
6d884e8f F |
1397 | } |
1398 | ||
061ae2ed | 1399 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1400 | { |
1401 | struct nfs_page *req; | |
1402 | ||
1403 | while (!list_empty(head)) { | |
1404 | req = nfs_list_entry(head->next); | |
1405 | nfs_list_remove_request(req); | |
1406 | nfs_redirty_request(req); | |
1407 | } | |
1408 | } | |
1409 | ||
dc602dd7 TM |
1410 | static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) |
1411 | { | |
1412 | nfs_async_write_error(&hdr->pages); | |
1413 | } | |
1414 | ||
061ae2ed | 1415 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
919e3bd9 | 1416 | .init_hdr = nfs_async_write_init, |
061ae2ed FI |
1417 | .error_cleanup = nfs_async_write_error, |
1418 | .completion = nfs_write_completion, | |
dc602dd7 | 1419 | .reschedule_io = nfs_async_write_reschedule_io, |
061ae2ed FI |
1420 | }; |
1421 | ||
57208fa7 | 1422 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1423 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1424 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1425 | { |
a20c93e3 | 1426 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1427 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1428 | |
1429 | #ifdef CONFIG_NFS_V4_1 | |
1430 | if (server->pnfs_curr_ld && !force_mds) | |
1431 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1432 | #endif | |
4a0de55c | 1433 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
3bde7afd | 1434 | server->wsize, ioflags); |
1751c363 | 1435 | } |
ddda8e0a | 1436 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1437 | |
dce81290 TM |
1438 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1439 | { | |
a7d42ddb WAA |
1440 | struct nfs_pgio_mirror *mirror; |
1441 | ||
6f29b9bb KM |
1442 | if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) |
1443 | pgio->pg_ops->pg_cleanup(pgio); | |
1444 | ||
41d8d5b7 | 1445 | pgio->pg_ops = &nfs_pgio_rw_ops; |
a7d42ddb WAA |
1446 | |
1447 | nfs_pageio_stop_mirroring(pgio); | |
1448 | ||
1449 | mirror = &pgio->pg_mirrors[0]; | |
1450 | mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
dce81290 | 1451 | } |
1f945357 | 1452 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1453 | |
1da177e4 | 1454 | |
0b7c0153 FI |
1455 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1456 | { | |
1457 | struct nfs_commit_data *data = calldata; | |
1458 | ||
1459 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1460 | } | |
1461 | ||
1f2edbe3 TM |
1462 | /* |
1463 | * Special version of should_remove_suid() that ignores capabilities. | |
1464 | */ | |
1465 | static int nfs_should_remove_suid(const struct inode *inode) | |
1466 | { | |
1467 | umode_t mode = inode->i_mode; | |
1468 | int kill = 0; | |
1469 | ||
1470 | /* suid always must be killed */ | |
1471 | if (unlikely(mode & S_ISUID)) | |
1472 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1473 | |
1f2edbe3 TM |
1474 | /* |
1475 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1476 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1477 | */ | |
1478 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1479 | kill |= ATTR_KILL_SGID; | |
1480 | ||
1481 | if (unlikely(kill && S_ISREG(mode))) | |
1482 | return kill; | |
1483 | ||
1484 | return 0; | |
1485 | } | |
788e7a89 | 1486 | |
a08a8cd3 TM |
1487 | static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, |
1488 | struct nfs_fattr *fattr) | |
1489 | { | |
1490 | struct nfs_pgio_args *argp = &hdr->args; | |
1491 | struct nfs_pgio_res *resp = &hdr->res; | |
2b83d3de | 1492 | u64 size = argp->offset + resp->count; |
a08a8cd3 TM |
1493 | |
1494 | if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) | |
2b83d3de TM |
1495 | fattr->size = size; |
1496 | if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { | |
1497 | fattr->valid &= ~NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 | 1498 | return; |
2b83d3de TM |
1499 | } |
1500 | if (size != fattr->size) | |
a08a8cd3 TM |
1501 | return; |
1502 | /* Set attribute barrier */ | |
1503 | nfs_fattr_set_barrier(fattr); | |
2b83d3de TM |
1504 | /* ...and update size */ |
1505 | fattr->valid |= NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 TM |
1506 | } |
1507 | ||
1508 | void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) | |
1509 | { | |
2b83d3de | 1510 | struct nfs_fattr *fattr = &hdr->fattr; |
a08a8cd3 TM |
1511 | struct inode *inode = hdr->inode; |
1512 | ||
a08a8cd3 TM |
1513 | spin_lock(&inode->i_lock); |
1514 | nfs_writeback_check_extend(hdr, fattr); | |
1515 | nfs_post_op_update_inode_force_wcc_locked(inode, fattr); | |
1516 | spin_unlock(&inode->i_lock); | |
1517 | } | |
1518 | EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); | |
1519 | ||
1da177e4 LT |
1520 | /* |
1521 | * This function is called when the WRITE call is complete. | |
1522 | */ | |
d45f60c6 WAA |
1523 | static int nfs_writeback_done(struct rpc_task *task, |
1524 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1525 | struct inode *inode) |
1da177e4 | 1526 | { |
788e7a89 | 1527 | int status; |
1da177e4 | 1528 | |
f551e44f CL |
1529 | /* |
1530 | * ->write_done will attempt to use post-op attributes to detect | |
1531 | * conflicting writes by other clients. A strict interpretation | |
1532 | * of close-to-open would allow us to continue caching even if | |
1533 | * another writer had changed the file, but some applications | |
1534 | * depend on tighter cache coherency when writing. | |
1535 | */ | |
d45f60c6 | 1536 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1537 | if (status != 0) |
0eecb214 | 1538 | return status; |
8224b273 | 1539 | |
d45f60c6 | 1540 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
8224b273 CL |
1541 | trace_nfs_writeback_done(inode, task->tk_status, |
1542 | hdr->args.offset, hdr->res.verf); | |
91d5b470 | 1543 | |
d45f60c6 WAA |
1544 | if (hdr->res.verf->committed < hdr->args.stable && |
1545 | task->tk_status >= 0) { | |
1da177e4 LT |
1546 | /* We tried a write call, but the server did not |
1547 | * commit data to stable storage even though we | |
1548 | * requested it. | |
1549 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1550 | * the server reports NFS_DATA_SYNC, but performs | |
1551 | * NFS_FILE_SYNC. We therefore implement this checking | |
1552 | * as a dprintk() in order to avoid filling syslog. | |
1553 | */ | |
1554 | static unsigned long complain; | |
1555 | ||
a69aef14 | 1556 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1557 | if (time_before(complain, jiffies)) { |
48186c7d | 1558 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1559 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1560 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
d45f60c6 | 1561 | hdr->res.verf->committed, hdr->args.stable); |
1da177e4 LT |
1562 | complain = jiffies + 300 * HZ; |
1563 | } | |
1564 | } | |
1f2edbe3 TM |
1565 | |
1566 | /* Deal with the suid/sgid bit corner case */ | |
1567 | if (nfs_should_remove_suid(inode)) | |
1568 | nfs_mark_for_revalidate(inode); | |
0eecb214 AS |
1569 | return 0; |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * This function is called when the WRITE call is complete. | |
1574 | */ | |
d45f60c6 WAA |
1575 | static void nfs_writeback_result(struct rpc_task *task, |
1576 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1577 | { |
d45f60c6 WAA |
1578 | struct nfs_pgio_args *argp = &hdr->args; |
1579 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1580 | |
1581 | if (resp->count < argp->count) { | |
1da177e4 LT |
1582 | static unsigned long complain; |
1583 | ||
6c75dc0d | 1584 | /* This a short write! */ |
d45f60c6 | 1585 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1586 | |
1da177e4 | 1587 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1588 | if (resp->count == 0) { |
1589 | if (time_before(complain, jiffies)) { | |
1590 | printk(KERN_WARNING | |
1591 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1592 | argp->count); | |
1593 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1594 | } |
d45f60c6 | 1595 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1596 | task->tk_status = -EIO; |
13602896 | 1597 | return; |
1da177e4 | 1598 | } |
f8417b48 KM |
1599 | |
1600 | /* For non rpc-based layout drivers, retry-through-MDS */ | |
1601 | if (!task->tk_ops) { | |
1602 | hdr->pnfs_error = -EAGAIN; | |
1603 | return; | |
1604 | } | |
1605 | ||
6c75dc0d FI |
1606 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1607 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1608 | /* Resend from where the server left off */ | |
d45f60c6 | 1609 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1610 | argp->offset += resp->count; |
1611 | argp->pgbase += resp->count; | |
1612 | argp->count -= resp->count; | |
1613 | } else { | |
1614 | /* Resend as a stable write in order to avoid | |
1615 | * headaches in the case of a server crash. | |
1616 | */ | |
1617 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1618 | } |
6c75dc0d | 1619 | rpc_restart_call_prepare(task); |
1da177e4 | 1620 | } |
1da177e4 LT |
1621 | } |
1622 | ||
af7cf057 | 1623 | static int wait_on_commit(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1624 | { |
af7cf057 TM |
1625 | return wait_on_atomic_t(&cinfo->rpcs_out, |
1626 | nfs_wait_atomic_killable, TASK_KILLABLE); | |
1627 | } | |
b8413f98 | 1628 | |
af7cf057 TM |
1629 | static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) |
1630 | { | |
1631 | atomic_inc(&cinfo->rpcs_out); | |
71d0a611 TM |
1632 | } |
1633 | ||
af7cf057 | 1634 | static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1635 | { |
af7cf057 TM |
1636 | if (atomic_dec_and_test(&cinfo->rpcs_out)) |
1637 | wake_up_atomic_t(&cinfo->rpcs_out); | |
71d0a611 TM |
1638 | } |
1639 | ||
0b7c0153 | 1640 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1641 | { |
0b7c0153 FI |
1642 | put_nfs_open_context(data->context); |
1643 | nfs_commit_free(data); | |
1da177e4 | 1644 | } |
e0c2b380 | 1645 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1646 | |
0b7c0153 | 1647 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
c36aae9a | 1648 | const struct nfs_rpc_ops *nfs_ops, |
9ace33cd | 1649 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1650 | int how, int flags) |
1da177e4 | 1651 | { |
07737691 | 1652 | struct rpc_task *task; |
9ace33cd | 1653 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1654 | struct rpc_message msg = { |
1655 | .rpc_argp = &data->args, | |
1656 | .rpc_resp = &data->res, | |
9ace33cd | 1657 | .rpc_cred = data->cred, |
bdc7f021 | 1658 | }; |
84115e1c | 1659 | struct rpc_task_setup task_setup_data = { |
07737691 | 1660 | .task = &data->task, |
9ace33cd | 1661 | .rpc_client = clnt, |
bdc7f021 | 1662 | .rpc_message = &msg, |
9ace33cd | 1663 | .callback_ops = call_ops, |
84115e1c | 1664 | .callback_data = data, |
101070ca | 1665 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1666 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1667 | .priority = priority, |
84115e1c | 1668 | }; |
9ace33cd | 1669 | /* Set up the initial task struct. */ |
c36aae9a | 1670 | nfs_ops->commit_setup(data, &msg); |
8224b273 | 1671 | trace_nfs_initiate_commit(data); |
9ace33cd | 1672 | |
b4839ebe | 1673 | dprintk("NFS: initiated commit call\n"); |
9ace33cd | 1674 | |
8c21c62c WAA |
1675 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1676 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1677 | ||
9ace33cd FI |
1678 | task = rpc_run_task(&task_setup_data); |
1679 | if (IS_ERR(task)) | |
1680 | return PTR_ERR(task); | |
1681 | if (how & FLUSH_SYNC) | |
1682 | rpc_wait_for_completion_task(task); | |
1683 | rpc_put_task(task); | |
1684 | return 0; | |
1685 | } | |
e0c2b380 | 1686 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1687 | |
378520b8 PT |
1688 | static loff_t nfs_get_lwb(struct list_head *head) |
1689 | { | |
1690 | loff_t lwb = 0; | |
1691 | struct nfs_page *req; | |
1692 | ||
1693 | list_for_each_entry(req, head, wb_list) | |
1694 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1695 | lwb = req_offset(req) + req->wb_bytes; | |
1696 | ||
1697 | return lwb; | |
1698 | } | |
1699 | ||
9ace33cd FI |
1700 | /* |
1701 | * Set up the argument/result storage required for the RPC call. | |
1702 | */ | |
0b7c0153 | 1703 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1704 | struct list_head *head, |
1705 | struct pnfs_layout_segment *lseg, | |
1706 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1707 | { |
1708 | struct nfs_page *first = nfs_list_entry(head->next); | |
2b0143b5 | 1709 | struct inode *inode = d_inode(first->wb_context->dentry); |
1da177e4 LT |
1710 | |
1711 | /* Set up the RPC argument and reply structs | |
1712 | * NB: take care not to mess about with data->commit et al. */ | |
1713 | ||
1714 | list_splice_init(head, &data->pages); | |
1da177e4 | 1715 | |
1da177e4 | 1716 | data->inode = inode; |
9ace33cd | 1717 | data->cred = first->wb_context->cred; |
988b6dce | 1718 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1719 | /* only set lwb for pnfs commit */ |
1720 | if (lseg) | |
1721 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1722 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1723 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1724 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1725 | |
1726 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1727 | /* Note: we always request a commit of the entire inode */ |
1728 | data->args.offset = 0; | |
1729 | data->args.count = 0; | |
0b7c0153 | 1730 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1731 | data->res.fattr = &data->fattr; |
1732 | data->res.verf = &data->verf; | |
0e574af1 | 1733 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1734 | } |
e0c2b380 | 1735 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1736 | |
e0c2b380 | 1737 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 | 1738 | struct pnfs_layout_segment *lseg, |
b57ff130 WAA |
1739 | struct nfs_commit_info *cinfo, |
1740 | u32 ds_commit_idx) | |
64bfeb49 FI |
1741 | { |
1742 | struct nfs_page *req; | |
1743 | ||
1744 | while (!list_empty(page_list)) { | |
1745 | req = nfs_list_entry(page_list->next); | |
1746 | nfs_list_remove_request(req); | |
b57ff130 | 1747 | nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); |
487b9b8a TH |
1748 | if (!cinfo->dreq) |
1749 | nfs_clear_page_commit(req->wb_page); | |
1d1afcbc | 1750 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1751 | } |
1752 | } | |
e0c2b380 | 1753 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1754 | |
b20135d0 TM |
1755 | static void |
1756 | nfs_commit_resched_write(struct nfs_commit_info *cinfo, | |
1757 | struct nfs_page *req) | |
1758 | { | |
1759 | __set_page_dirty_nobuffers(req->wb_page); | |
1760 | } | |
1761 | ||
1da177e4 LT |
1762 | /* |
1763 | * Commit dirty pages | |
1764 | */ | |
1765 | static int | |
ea2cf228 FI |
1766 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1767 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1768 | { |
0b7c0153 | 1769 | struct nfs_commit_data *data; |
1da177e4 | 1770 | |
ade8febd WAA |
1771 | /* another commit raced with us */ |
1772 | if (list_empty(head)) | |
1773 | return 0; | |
1774 | ||
518662e0 | 1775 | data = nfs_commitdata_alloc(true); |
1da177e4 LT |
1776 | |
1777 | /* Set up the argument struct */ | |
f453a54a FI |
1778 | nfs_init_commit(data, head, NULL, cinfo); |
1779 | atomic_inc(&cinfo->mds->rpcs_out); | |
c36aae9a PT |
1780 | return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), |
1781 | data->mds_ops, how, 0); | |
67911c8f | 1782 | } |
67911c8f | 1783 | |
1da177e4 LT |
1784 | /* |
1785 | * COMMIT call returned | |
1786 | */ | |
788e7a89 | 1787 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1788 | { |
0b7c0153 | 1789 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1790 | |
a3f565b1 | 1791 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1792 | task->tk_pid, task->tk_status); |
1793 | ||
788e7a89 | 1794 | /* Call the NFS version-specific code */ |
c0d0e96b | 1795 | NFS_PROTO(data->inode)->commit_done(task, data); |
8224b273 | 1796 | trace_nfs_commit_done(data); |
c9d8f89d TM |
1797 | } |
1798 | ||
f453a54a | 1799 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1800 | { |
5917ce84 | 1801 | struct nfs_page *req; |
c9d8f89d | 1802 | int status = data->task.tk_status; |
f453a54a | 1803 | struct nfs_commit_info cinfo; |
353db796 | 1804 | struct nfs_server *nfss; |
788e7a89 | 1805 | |
1da177e4 LT |
1806 | while (!list_empty(&data->pages)) { |
1807 | req = nfs_list_entry(data->pages.next); | |
1808 | nfs_list_remove_request(req); | |
67911c8f AS |
1809 | if (req->wb_page) |
1810 | nfs_clear_page_commit(req->wb_page); | |
1da177e4 | 1811 | |
1e8968c5 | 1812 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1813 | req->wb_context->dentry->d_sb->s_id, |
2b0143b5 | 1814 | (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), |
1da177e4 LT |
1815 | req->wb_bytes, |
1816 | (long long)req_offset(req)); | |
c9d8f89d TM |
1817 | if (status < 0) { |
1818 | nfs_context_set_write_error(req->wb_context, status); | |
38a33101 KM |
1819 | if (req->wb_page) |
1820 | nfs_inode_remove_request(req); | |
ddeaa637 | 1821 | dprintk_cont(", error = %d\n", status); |
1da177e4 LT |
1822 | goto next; |
1823 | } | |
1824 | ||
1825 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1826 | * returned by the server against all stored verfs. */ | |
8fc3c386 | 1827 | if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { |
1da177e4 | 1828 | /* We have a match */ |
38a33101 KM |
1829 | if (req->wb_page) |
1830 | nfs_inode_remove_request(req); | |
ddeaa637 | 1831 | dprintk_cont(" OK\n"); |
1da177e4 LT |
1832 | goto next; |
1833 | } | |
1834 | /* We have a mismatch. Write the page again */ | |
ddeaa637 | 1835 | dprintk_cont(" mismatch\n"); |
6d884e8f | 1836 | nfs_mark_request_dirty(req); |
05990d1b | 1837 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1838 | next: |
1d1afcbc | 1839 | nfs_unlock_and_release_request(req); |
8aea358b TM |
1840 | /* Latency breaker */ |
1841 | cond_resched(); | |
1da177e4 | 1842 | } |
353db796 N |
1843 | nfss = NFS_SERVER(data->inode); |
1844 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
0db10944 | 1845 | clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); |
353db796 | 1846 | |
f453a54a | 1847 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
af7cf057 | 1848 | nfs_commit_end(cinfo.mds); |
5917ce84 FI |
1849 | } |
1850 | ||
1851 | static void nfs_commit_release(void *calldata) | |
1852 | { | |
0b7c0153 | 1853 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1854 | |
f453a54a | 1855 | data->completion_ops->completion(data); |
c9d8f89d | 1856 | nfs_commitdata_release(calldata); |
1da177e4 | 1857 | } |
788e7a89 TM |
1858 | |
1859 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1860 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1861 | .rpc_call_done = nfs_commit_done, |
1862 | .rpc_release = nfs_commit_release, | |
1863 | }; | |
1da177e4 | 1864 | |
f453a54a FI |
1865 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1866 | .completion = nfs_commit_release_pages, | |
b20135d0 | 1867 | .resched_write = nfs_commit_resched_write, |
f453a54a FI |
1868 | }; |
1869 | ||
1763da12 FI |
1870 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1871 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1872 | { |
1873 | int status; | |
1874 | ||
ea2cf228 | 1875 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1876 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1877 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1878 | return status; |
1879 | } | |
1880 | ||
b608b283 | 1881 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1882 | { |
1da177e4 | 1883 | LIST_HEAD(head); |
ea2cf228 | 1884 | struct nfs_commit_info cinfo; |
71d0a611 | 1885 | int may_wait = how & FLUSH_SYNC; |
af7cf057 | 1886 | int error = 0; |
b8413f98 | 1887 | int res; |
1da177e4 | 1888 | |
ea2cf228 | 1889 | nfs_init_cinfo_from_inode(&cinfo, inode); |
af7cf057 | 1890 | nfs_commit_begin(cinfo.mds); |
ea2cf228 | 1891 | res = nfs_scan_commit(inode, &head, &cinfo); |
af7cf057 | 1892 | if (res) |
ea2cf228 | 1893 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
af7cf057 | 1894 | nfs_commit_end(cinfo.mds); |
dc4fd9ab SM |
1895 | if (res == 0) |
1896 | return res; | |
af7cf057 TM |
1897 | if (error < 0) |
1898 | goto out_error; | |
1899 | if (!may_wait) | |
1900 | goto out_mark_dirty; | |
1901 | error = wait_on_commit(cinfo.mds); | |
1902 | if (error < 0) | |
1903 | return error; | |
c5efa5fc | 1904 | return res; |
af7cf057 TM |
1905 | out_error: |
1906 | res = error; | |
c5efa5fc TM |
1907 | /* Note: If we exit without ensuring that the commit is complete, |
1908 | * we must mark the inode as dirty. Otherwise, future calls to | |
1909 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1910 | * that the data is on the disk. | |
1911 | */ | |
1912 | out_mark_dirty: | |
1913 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1914 | return res; |
1915 | } | |
b20135d0 | 1916 | EXPORT_SYMBOL_GPL(nfs_commit_inode); |
8fc795f7 | 1917 | |
ae09c31f | 1918 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
8fc795f7 | 1919 | { |
420e3646 TM |
1920 | struct nfs_inode *nfsi = NFS_I(inode); |
1921 | int flags = FLUSH_SYNC; | |
1922 | int ret = 0; | |
8fc795f7 | 1923 | |
3236c3e1 | 1924 | /* no commits means nothing needs to be done */ |
5cb953d4 | 1925 | if (!atomic_long_read(&nfsi->commit_info.ncommit)) |
3236c3e1 JL |
1926 | return ret; |
1927 | ||
a00dd6c0 JL |
1928 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1929 | /* Don't commit yet if this is a non-blocking flush and there | |
1930 | * are a lot of outstanding writes for this mapping. | |
1931 | */ | |
1a4edf0f | 1932 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) |
a00dd6c0 | 1933 | goto out_mark_dirty; |
420e3646 | 1934 | |
a00dd6c0 | 1935 | /* don't wait for the COMMIT response */ |
420e3646 | 1936 | flags = 0; |
a00dd6c0 JL |
1937 | } |
1938 | ||
420e3646 TM |
1939 | ret = nfs_commit_inode(inode, flags); |
1940 | if (ret >= 0) { | |
1941 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1942 | if (ret < wbc->nr_to_write) | |
1943 | wbc->nr_to_write -= ret; | |
1944 | else | |
1945 | wbc->nr_to_write = 0; | |
1946 | } | |
8fc795f7 | 1947 | return 0; |
420e3646 TM |
1948 | } |
1949 | out_mark_dirty: | |
8fc795f7 TM |
1950 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1951 | return ret; | |
1952 | } | |
89d77c8f | 1953 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1954 | |
837bb1d7 TM |
1955 | /* |
1956 | * Wrapper for filemap_write_and_wait_range() | |
1957 | * | |
1958 | * Needed for pNFS in order to ensure data becomes visible to the | |
1959 | * client. | |
1960 | */ | |
1961 | int nfs_filemap_write_and_wait_range(struct address_space *mapping, | |
1962 | loff_t lstart, loff_t lend) | |
1963 | { | |
1964 | int ret; | |
1965 | ||
1966 | ret = filemap_write_and_wait_range(mapping, lstart, lend); | |
1967 | if (ret == 0) | |
1968 | ret = pnfs_sync_inode(mapping->host, true); | |
1969 | return ret; | |
1970 | } | |
1971 | EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); | |
1972 | ||
acdc53b2 TM |
1973 | /* |
1974 | * flush the inode to disk. | |
1975 | */ | |
1976 | int nfs_wb_all(struct inode *inode) | |
34901f70 | 1977 | { |
f4ce1299 TM |
1978 | int ret; |
1979 | ||
1980 | trace_nfs_writeback_inode_enter(inode); | |
1981 | ||
5bb89b47 | 1982 | ret = filemap_write_and_wait(inode->i_mapping); |
6b196875 CL |
1983 | if (ret) |
1984 | goto out; | |
1985 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
1986 | if (ret < 0) | |
1987 | goto out; | |
1988 | pnfs_sync_inode(inode, true); | |
1989 | ret = 0; | |
34901f70 | 1990 | |
6b196875 | 1991 | out: |
f4ce1299 TM |
1992 | trace_nfs_writeback_inode_exit(inode, ret); |
1993 | return ret; | |
1c75950b | 1994 | } |
ddda8e0a | 1995 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1996 | |
1b3b4a1a TM |
1997 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1998 | { | |
1999 | struct nfs_page *req; | |
1b3b4a1a TM |
2000 | int ret = 0; |
2001 | ||
3e217045 WAA |
2002 | wait_on_page_writeback(page); |
2003 | ||
2004 | /* blocking call to cancel all requests and join to a single (head) | |
2005 | * request */ | |
6d17d653 | 2006 | req = nfs_lock_and_join_requests(page); |
3e217045 WAA |
2007 | |
2008 | if (IS_ERR(req)) { | |
2009 | ret = PTR_ERR(req); | |
2010 | } else if (req) { | |
2011 | /* all requests from this page have been cancelled by | |
2012 | * nfs_lock_and_join_requests, so just remove the head | |
2013 | * request from the inode / page_private pointer and | |
2014 | * release it */ | |
2015 | nfs_inode_remove_request(req); | |
3e217045 | 2016 | nfs_unlock_and_release_request(req); |
1b3b4a1a | 2017 | } |
3e217045 | 2018 | |
1b3b4a1a TM |
2019 | return ret; |
2020 | } | |
2021 | ||
7f2f12d9 TM |
2022 | /* |
2023 | * Write back all requests on one page - we do this before reading it. | |
2024 | */ | |
c373fff7 | 2025 | int nfs_wb_page(struct inode *inode, struct page *page) |
1c75950b | 2026 | { |
29418aa4 | 2027 | loff_t range_start = page_file_offset(page); |
09cbfeaf | 2028 | loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); |
4d770ccf | 2029 | struct writeback_control wbc = { |
4d770ccf | 2030 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 2031 | .nr_to_write = 0, |
4d770ccf TM |
2032 | .range_start = range_start, |
2033 | .range_end = range_end, | |
2034 | }; | |
2035 | int ret; | |
1c75950b | 2036 | |
f4ce1299 TM |
2037 | trace_nfs_writeback_page_enter(inode); |
2038 | ||
0522f6ad | 2039 | for (;;) { |
ba8b06e6 | 2040 | wait_on_page_writeback(page); |
73e3302f | 2041 | if (clear_page_dirty_for_io(page)) { |
c373fff7 | 2042 | ret = nfs_writepage_locked(page, &wbc); |
73e3302f TM |
2043 | if (ret < 0) |
2044 | goto out_error; | |
0522f6ad | 2045 | continue; |
7f2f12d9 | 2046 | } |
f4ce1299 | 2047 | ret = 0; |
0522f6ad TM |
2048 | if (!PagePrivate(page)) |
2049 | break; | |
2050 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 2051 | if (ret < 0) |
73e3302f | 2052 | goto out_error; |
7f2f12d9 | 2053 | } |
73e3302f | 2054 | out_error: |
f4ce1299 | 2055 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 2056 | return ret; |
1c75950b TM |
2057 | } |
2058 | ||
074cc1de TM |
2059 | #ifdef CONFIG_MIGRATION |
2060 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 2061 | struct page *page, enum migrate_mode mode) |
074cc1de | 2062 | { |
2da95652 JL |
2063 | /* |
2064 | * If PagePrivate is set, then the page is currently associated with | |
2065 | * an in-progress read or write request. Don't try to migrate it. | |
2066 | * | |
2067 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
2068 | * that we can safely release the inode reference while holding | |
2069 | * the page lock. | |
2070 | */ | |
2071 | if (PagePrivate(page)) | |
2072 | return -EBUSY; | |
074cc1de | 2073 | |
8c209ce7 DH |
2074 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
2075 | return -EBUSY; | |
074cc1de | 2076 | |
a6bc32b8 | 2077 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
2078 | } |
2079 | #endif | |
2080 | ||
f7b422b1 | 2081 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
2082 | { |
2083 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 2084 | sizeof(struct nfs_pgio_header), |
1da177e4 | 2085 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 2086 | NULL); |
1da177e4 LT |
2087 | if (nfs_wdata_cachep == NULL) |
2088 | return -ENOMEM; | |
2089 | ||
93d2341c MD |
2090 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
2091 | nfs_wdata_cachep); | |
1da177e4 | 2092 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 2093 | goto out_destroy_write_cache; |
1da177e4 | 2094 | |
0b7c0153 FI |
2095 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
2096 | sizeof(struct nfs_commit_data), | |
2097 | 0, SLAB_HWCACHE_ALIGN, | |
2098 | NULL); | |
2099 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 2100 | goto out_destroy_write_mempool; |
0b7c0153 | 2101 | |
93d2341c | 2102 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 2103 | nfs_cdata_cachep); |
1da177e4 | 2104 | if (nfs_commit_mempool == NULL) |
3dd4765f | 2105 | goto out_destroy_commit_cache; |
1da177e4 | 2106 | |
89a09141 PZ |
2107 | /* |
2108 | * NFS congestion size, scale with available memory. | |
2109 | * | |
2110 | * 64MB: 8192k | |
2111 | * 128MB: 11585k | |
2112 | * 256MB: 16384k | |
2113 | * 512MB: 23170k | |
2114 | * 1GB: 32768k | |
2115 | * 2GB: 46340k | |
2116 | * 4GB: 65536k | |
2117 | * 8GB: 92681k | |
2118 | * 16GB: 131072k | |
2119 | * | |
2120 | * This allows larger machines to have larger/more transfers. | |
2121 | * Limit the default to 256M | |
2122 | */ | |
2123 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
2124 | if (nfs_congestion_kb > 256*1024) | |
2125 | nfs_congestion_kb = 256*1024; | |
2126 | ||
1da177e4 | 2127 | return 0; |
3dd4765f JL |
2128 | |
2129 | out_destroy_commit_cache: | |
2130 | kmem_cache_destroy(nfs_cdata_cachep); | |
2131 | out_destroy_write_mempool: | |
2132 | mempool_destroy(nfs_wdata_mempool); | |
2133 | out_destroy_write_cache: | |
2134 | kmem_cache_destroy(nfs_wdata_cachep); | |
2135 | return -ENOMEM; | |
1da177e4 LT |
2136 | } |
2137 | ||
266bee88 | 2138 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
2139 | { |
2140 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 2141 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 2142 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 2143 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
2144 | } |
2145 | ||
4a0de55c AS |
2146 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
2147 | .rw_alloc_header = nfs_writehdr_alloc, | |
2148 | .rw_free_header = nfs_writehdr_free, | |
0eecb214 AS |
2149 | .rw_done = nfs_writeback_done, |
2150 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 2151 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 2152 | }; |