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> | |
1eb5d98f | 26 | #include <linux/iversion.h> |
3fcfab16 | 27 | |
7c0f6ba6 | 28 | #include <linux/uaccess.h> |
875bc3fb | 29 | #include <linux/sched/mm.h> |
1da177e4 LT |
30 | |
31 | #include "delegation.h" | |
49a70f27 | 32 | #include "internal.h" |
91d5b470 | 33 | #include "iostat.h" |
def6ed7e | 34 | #include "nfs4_fs.h" |
074cc1de | 35 | #include "fscache.h" |
94ad1c80 | 36 | #include "pnfs.h" |
1da177e4 | 37 | |
f4ce1299 TM |
38 | #include "nfstrace.h" |
39 | ||
1da177e4 LT |
40 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
41 | ||
42 | #define MIN_POOL_WRITE (32) | |
43 | #define MIN_POOL_COMMIT (4) | |
44 | ||
919e3bd9 TM |
45 | struct nfs_io_completion { |
46 | void (*complete)(void *data); | |
47 | void *data; | |
48 | struct kref refcount; | |
49 | }; | |
50 | ||
1da177e4 LT |
51 | /* |
52 | * Local function declarations | |
53 | */ | |
f8512ad0 | 54 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 | 55 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 56 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 57 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
4a0de55c | 58 | static const struct nfs_rw_ops nfs_rw_write_ops; |
d4581383 | 59 | static void nfs_clear_request_commit(struct nfs_page *req); |
02d1426c WAA |
60 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
61 | struct inode *inode); | |
3a3908c8 TM |
62 | static struct nfs_page * |
63 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
64 | struct page *page); | |
1da177e4 | 65 | |
e18b890b | 66 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 67 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 68 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
69 | static mempool_t *nfs_commit_mempool; |
70 | ||
518662e0 | 71 | struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) |
1da177e4 | 72 | { |
518662e0 | 73 | struct nfs_commit_data *p; |
40859d7e | 74 | |
518662e0 N |
75 | if (never_fail) |
76 | p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); | |
77 | else { | |
78 | /* It is OK to do some reclaim, not no safe to wait | |
79 | * for anything to be returned to the pool. | |
80 | * mempool_alloc() cannot handle that particular combination, | |
81 | * so we need two separate attempts. | |
82 | */ | |
83 | p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); | |
84 | if (!p) | |
85 | p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | | |
86 | __GFP_NOWARN | __GFP_NORETRY); | |
87 | if (!p) | |
88 | return NULL; | |
1da177e4 | 89 | } |
518662e0 N |
90 | |
91 | memset(p, 0, sizeof(*p)); | |
92 | INIT_LIST_HEAD(&p->pages); | |
1da177e4 LT |
93 | return p; |
94 | } | |
e0c2b380 | 95 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 96 | |
0b7c0153 | 97 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
98 | { |
99 | mempool_free(p, nfs_commit_mempool); | |
100 | } | |
e0c2b380 | 101 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 102 | |
1e7f3a48 | 103 | static struct nfs_pgio_header *nfs_writehdr_alloc(void) |
3feb2d49 | 104 | { |
1e7f3a48 | 105 | struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
cd841605 | 106 | |
237f8306 N |
107 | memset(p, 0, sizeof(*p)); |
108 | p->rw_mode = FMODE_WRITE; | |
3feb2d49 TM |
109 | return p; |
110 | } | |
6c75dc0d | 111 | |
1e7f3a48 | 112 | static void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 113 | { |
1e7f3a48 | 114 | mempool_free(hdr, nfs_wdata_mempool); |
3feb2d49 | 115 | } |
1da177e4 | 116 | |
919e3bd9 TM |
117 | static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) |
118 | { | |
119 | return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); | |
120 | } | |
121 | ||
122 | static void nfs_io_completion_init(struct nfs_io_completion *ioc, | |
123 | void (*complete)(void *), void *data) | |
124 | { | |
125 | ioc->complete = complete; | |
126 | ioc->data = data; | |
127 | kref_init(&ioc->refcount); | |
128 | } | |
129 | ||
130 | static void nfs_io_completion_release(struct kref *kref) | |
131 | { | |
132 | struct nfs_io_completion *ioc = container_of(kref, | |
133 | struct nfs_io_completion, refcount); | |
134 | ioc->complete(ioc->data); | |
135 | kfree(ioc); | |
136 | } | |
137 | ||
138 | static void nfs_io_completion_get(struct nfs_io_completion *ioc) | |
139 | { | |
140 | if (ioc != NULL) | |
141 | kref_get(&ioc->refcount); | |
142 | } | |
143 | ||
144 | static void nfs_io_completion_put(struct nfs_io_completion *ioc) | |
145 | { | |
146 | if (ioc != NULL) | |
147 | kref_put(&ioc->refcount, nfs_io_completion_release); | |
148 | } | |
149 | ||
bd37d6fc TM |
150 | static struct nfs_page * |
151 | nfs_page_private_request(struct page *page) | |
152 | { | |
153 | if (!PagePrivate(page)) | |
154 | return NULL; | |
155 | return (struct nfs_page *)page_private(page); | |
156 | } | |
157 | ||
84d3a9a9 WAA |
158 | /* |
159 | * nfs_page_find_head_request_locked - find head request associated with @page | |
160 | * | |
161 | * must be called while holding the inode lock. | |
162 | * | |
163 | * returns matching head request with reference held, or NULL if not found. | |
164 | */ | |
29418aa4 | 165 | static struct nfs_page * |
b30d2f04 | 166 | nfs_page_find_private_request(struct page *page) |
277459d2 | 167 | { |
4b9bb25b | 168 | struct address_space *mapping = page_file_mapping(page); |
bd37d6fc | 169 | struct nfs_page *req; |
277459d2 | 170 | |
b30d2f04 TM |
171 | if (!PagePrivate(page)) |
172 | return NULL; | |
4b9bb25b | 173 | spin_lock(&mapping->private_lock); |
bd37d6fc | 174 | req = nfs_page_private_request(page); |
84d3a9a9 WAA |
175 | if (req) { |
176 | WARN_ON_ONCE(req->wb_head != req); | |
29418aa4 | 177 | kref_get(&req->wb_kref); |
84d3a9a9 | 178 | } |
4b9bb25b | 179 | spin_unlock(&mapping->private_lock); |
b30d2f04 TM |
180 | return req; |
181 | } | |
29418aa4 | 182 | |
b30d2f04 TM |
183 | static struct nfs_page * |
184 | nfs_page_find_swap_request(struct page *page) | |
185 | { | |
186 | struct inode *inode = page_file_mapping(page)->host; | |
187 | struct nfs_inode *nfsi = NFS_I(inode); | |
188 | struct nfs_page *req = NULL; | |
189 | if (!PageSwapCache(page)) | |
190 | return NULL; | |
e824f99a | 191 | mutex_lock(&nfsi->commit_mutex); |
b30d2f04 TM |
192 | if (PageSwapCache(page)) { |
193 | req = nfs_page_search_commits_for_head_request_locked(nfsi, | |
194 | page); | |
195 | if (req) { | |
196 | WARN_ON_ONCE(req->wb_head != req); | |
197 | kref_get(&req->wb_kref); | |
198 | } | |
199 | } | |
e824f99a | 200 | mutex_unlock(&nfsi->commit_mutex); |
277459d2 TM |
201 | return req; |
202 | } | |
203 | ||
84d3a9a9 WAA |
204 | /* |
205 | * nfs_page_find_head_request - find head request associated with @page | |
206 | * | |
207 | * returns matching head request with reference held, or NULL if not found. | |
208 | */ | |
209 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 210 | { |
b30d2f04 | 211 | struct nfs_page *req; |
277459d2 | 212 | |
b30d2f04 TM |
213 | req = nfs_page_find_private_request(page); |
214 | if (!req) | |
215 | req = nfs_page_find_swap_request(page); | |
277459d2 TM |
216 | return req; |
217 | } | |
218 | ||
1da177e4 LT |
219 | /* Adjust the file length if we're writing beyond the end */ |
220 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
221 | { | |
d56b4ddf | 222 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
223 | loff_t end, i_size; |
224 | pgoff_t end_index; | |
1da177e4 | 225 | |
a3d01454 TM |
226 | spin_lock(&inode->i_lock); |
227 | i_size = i_size_read(inode); | |
09cbfeaf | 228 | end_index = (i_size - 1) >> PAGE_SHIFT; |
8cd79788 | 229 | if (i_size > 0 && page_index(page) < end_index) |
a3d01454 | 230 | goto out; |
d56b4ddf | 231 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 232 | if (i_size >= end) |
a3d01454 | 233 | goto out; |
1da177e4 | 234 | i_size_write(inode, end); |
f6cdfa6d | 235 | NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; |
a3d01454 TM |
236 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
237 | out: | |
238 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
239 | } |
240 | ||
a301b777 | 241 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
d2ceb7e5 | 242 | static void nfs_set_pageerror(struct address_space *mapping) |
a301b777 | 243 | { |
d2ceb7e5 | 244 | nfs_zap_mapping(mapping->host, mapping); |
a301b777 TM |
245 | } |
246 | ||
6fbda89b TM |
247 | static void nfs_mapping_set_error(struct page *page, int error) |
248 | { | |
249 | SetPageError(page); | |
250 | mapping_set_error(page_file_mapping(page), error); | |
251 | } | |
252 | ||
d72ddcba WAA |
253 | /* |
254 | * nfs_page_group_search_locked | |
255 | * @head - head request of page group | |
256 | * @page_offset - offset into page | |
257 | * | |
258 | * Search page group with head @head to find a request that contains the | |
259 | * page offset @page_offset. | |
260 | * | |
261 | * Returns a pointer to the first matching nfs request, or NULL if no | |
262 | * match is found. | |
263 | * | |
264 | * Must be called with the page group lock held | |
265 | */ | |
266 | static struct nfs_page * | |
267 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
268 | { | |
269 | struct nfs_page *req; | |
270 | ||
d72ddcba WAA |
271 | req = head; |
272 | do { | |
273 | if (page_offset >= req->wb_pgbase && | |
274 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
275 | return req; | |
276 | ||
277 | req = req->wb_this_page; | |
278 | } while (req != head); | |
279 | ||
280 | return NULL; | |
281 | } | |
282 | ||
283 | /* | |
284 | * nfs_page_group_covers_page | |
285 | * @head - head request of page group | |
286 | * | |
287 | * Return true if the page group with head @head covers the whole page, | |
288 | * returns false otherwise | |
289 | */ | |
290 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
291 | { | |
292 | struct nfs_page *tmp; | |
293 | unsigned int pos = 0; | |
294 | unsigned int len = nfs_page_length(req->wb_page); | |
295 | ||
1344b7ea | 296 | nfs_page_group_lock(req); |
d72ddcba | 297 | |
7e8a30f8 | 298 | for (;;) { |
d72ddcba | 299 | tmp = nfs_page_group_search_locked(req->wb_head, pos); |
7e8a30f8 TM |
300 | if (!tmp) |
301 | break; | |
302 | pos = tmp->wb_pgbase + tmp->wb_bytes; | |
303 | } | |
d72ddcba WAA |
304 | |
305 | nfs_page_group_unlock(req); | |
7e8a30f8 | 306 | return pos >= len; |
d72ddcba WAA |
307 | } |
308 | ||
1da177e4 LT |
309 | /* We can set the PG_uptodate flag if we see that a write request |
310 | * covers the full page. | |
311 | */ | |
d72ddcba | 312 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 313 | { |
d72ddcba | 314 | if (PageUptodate(req->wb_page)) |
1da177e4 | 315 | return; |
d72ddcba | 316 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 317 | return; |
d72ddcba | 318 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
319 | } |
320 | ||
1da177e4 LT |
321 | static int wb_priority(struct writeback_control *wbc) |
322 | { | |
e87b4c7a | 323 | int ret = 0; |
cca588d6 | 324 | |
e87b4c7a N |
325 | if (wbc->sync_mode == WB_SYNC_ALL) |
326 | ret = FLUSH_COND_STABLE; | |
e87b4c7a | 327 | return ret; |
1da177e4 LT |
328 | } |
329 | ||
89a09141 PZ |
330 | /* |
331 | * NFS congestion control | |
332 | */ | |
333 | ||
334 | int nfs_congestion_kb; | |
335 | ||
336 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
337 | #define NFS_CONGESTION_OFF_THRESH \ | |
338 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
339 | ||
deed85e7 | 340 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 341 | { |
0db10944 JK |
342 | struct inode *inode = page_file_mapping(page)->host; |
343 | struct nfs_server *nfss = NFS_SERVER(inode); | |
5a6d41b3 TM |
344 | int ret = test_set_page_writeback(page); |
345 | ||
deed85e7 | 346 | WARN_ON_ONCE(ret != 0); |
89a09141 | 347 | |
deed85e7 | 348 | if (atomic_long_inc_return(&nfss->writeback) > |
0db10944 JK |
349 | NFS_CONGESTION_ON_THRESH) |
350 | set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); | |
89a09141 PZ |
351 | } |
352 | ||
20633f04 | 353 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 354 | { |
20633f04 | 355 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 | 356 | struct nfs_server *nfss = NFS_SERVER(inode); |
31a01f09 | 357 | bool is_done; |
89a09141 | 358 | |
31a01f09 TM |
359 | is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); |
360 | nfs_unlock_request(req); | |
361 | if (!is_done) | |
20633f04 WAA |
362 | return; |
363 | ||
364 | end_page_writeback(req->wb_page); | |
c4dc4bee | 365 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
0db10944 | 366 | clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); |
89a09141 PZ |
367 | } |
368 | ||
d4581383 WAA |
369 | /* |
370 | * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req | |
371 | * | |
372 | * this is a helper function for nfs_lock_and_join_requests | |
373 | * | |
374 | * @inode - inode associated with request page group, must be holding inode lock | |
375 | * @head - head request of page group, must be holding head lock | |
376 | * @req - request that couldn't lock and needs to wait on the req bit lock | |
d4581383 | 377 | * |
b5bab9bf TM |
378 | * NOTE: this must be called holding page_group bit lock |
379 | * which will be released before returning. | |
d4581383 WAA |
380 | * |
381 | * returns 0 on success, < 0 on error. | |
382 | */ | |
74a6d4b5 TM |
383 | static void |
384 | nfs_unroll_locks(struct inode *inode, struct nfs_page *head, | |
6d17d653 | 385 | struct nfs_page *req) |
d4581383 WAA |
386 | { |
387 | struct nfs_page *tmp; | |
d4581383 WAA |
388 | |
389 | /* relinquish all the locks successfully grabbed this run */ | |
5b2b5187 TM |
390 | for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { |
391 | if (!kref_read(&tmp->wb_kref)) | |
392 | continue; | |
393 | nfs_unlock_and_release_request(tmp); | |
394 | } | |
d4581383 WAA |
395 | } |
396 | ||
397 | /* | |
398 | * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests | |
399 | * | |
400 | * @destroy_list - request list (using wb_this_page) terminated by @old_head | |
401 | * @old_head - the old head of the list | |
402 | * | |
403 | * All subrequests must be locked and removed from all lists, so at this point | |
404 | * they are only "active" in this function, and possibly in nfs_wait_on_request | |
405 | * with a reference held by some other context. | |
406 | */ | |
407 | static void | |
408 | nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, | |
b66aaa8d TM |
409 | struct nfs_page *old_head, |
410 | struct inode *inode) | |
d4581383 WAA |
411 | { |
412 | while (destroy_list) { | |
413 | struct nfs_page *subreq = destroy_list; | |
414 | ||
415 | destroy_list = (subreq->wb_this_page == old_head) ? | |
416 | NULL : subreq->wb_this_page; | |
417 | ||
418 | WARN_ON_ONCE(old_head != subreq->wb_head); | |
419 | ||
420 | /* make sure old group is not used */ | |
d4581383 WAA |
421 | subreq->wb_this_page = subreq; |
422 | ||
902a4c00 TM |
423 | clear_bit(PG_REMOVE, &subreq->wb_flags); |
424 | ||
5b2b5187 TM |
425 | /* Note: races with nfs_page_group_destroy() */ |
426 | if (!kref_read(&subreq->wb_kref)) { | |
5b2b5187 | 427 | /* Check if we raced with nfs_page_group_destroy() */ |
902a4c00 | 428 | if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) |
5b2b5187 TM |
429 | nfs_free_request(subreq); |
430 | continue; | |
431 | } | |
d4581383 | 432 | |
5b2b5187 TM |
433 | subreq->wb_head = subreq; |
434 | ||
435 | if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { | |
436 | nfs_release_request(subreq); | |
a6b6d5b8 | 437 | atomic_long_dec(&NFS_I(inode)->nrequests); |
d4581383 | 438 | } |
5b2b5187 | 439 | |
5b2b5187 TM |
440 | /* subreq is now totally disconnected from page group or any |
441 | * write / commit lists. last chance to wake any waiters */ | |
442 | nfs_unlock_and_release_request(subreq); | |
d4581383 WAA |
443 | } |
444 | } | |
445 | ||
446 | /* | |
447 | * nfs_lock_and_join_requests - join all subreqs to the head req and return | |
448 | * a locked reference, cancelling any pending | |
449 | * operations for this page. | |
450 | * | |
451 | * @page - the page used to lookup the "page group" of nfs_page structures | |
d4581383 WAA |
452 | * |
453 | * This function joins all sub requests to the head request by first | |
454 | * locking all requests in the group, cancelling any pending operations | |
455 | * and finally updating the head request to cover the whole range covered by | |
456 | * the (former) group. All subrequests are removed from any write or commit | |
457 | * lists, unlinked from the group and destroyed. | |
458 | * | |
459 | * Returns a locked, referenced pointer to the head request - which after | |
460 | * this call is guaranteed to be the only request associated with the page. | |
461 | * Returns NULL if no requests are found for @page, or a ERR_PTR if an | |
462 | * error was encountered. | |
463 | */ | |
464 | static struct nfs_page * | |
6d17d653 | 465 | nfs_lock_and_join_requests(struct page *page) |
e261f51f | 466 | { |
d56b4ddf | 467 | struct inode *inode = page_file_mapping(page)->host; |
d4581383 WAA |
468 | struct nfs_page *head, *subreq; |
469 | struct nfs_page *destroy_list = NULL; | |
470 | unsigned int total_bytes; | |
e261f51f TM |
471 | int ret; |
472 | ||
d4581383 | 473 | try_again: |
d4581383 WAA |
474 | /* |
475 | * A reference is taken only on the head request which acts as a | |
476 | * reference to the whole page group - the group will not be destroyed | |
477 | * until the head reference is released. | |
478 | */ | |
bd37d6fc TM |
479 | head = nfs_page_find_head_request(page); |
480 | if (!head) | |
d4581383 | 481 | return NULL; |
d4581383 | 482 | |
a0e265bc TM |
483 | /* lock the page head first in order to avoid an ABBA inefficiency */ |
484 | if (!nfs_lock_request(head)) { | |
a0e265bc TM |
485 | ret = nfs_wait_on_request(head); |
486 | nfs_release_request(head); | |
487 | if (ret < 0) | |
488 | return ERR_PTR(ret); | |
489 | goto try_again; | |
490 | } | |
bd37d6fc TM |
491 | |
492 | /* Ensure that nobody removed the request before we locked it */ | |
493 | if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { | |
494 | nfs_unlock_and_release_request(head); | |
495 | goto try_again; | |
496 | } | |
a0e265bc | 497 | |
1344b7ea | 498 | ret = nfs_page_group_lock(head); |
0671d8f1 ME |
499 | if (ret < 0) |
500 | goto release_request; | |
7c3af975 WAA |
501 | |
502 | /* lock each request in the page group */ | |
a0e265bc TM |
503 | total_bytes = head->wb_bytes; |
504 | for (subreq = head->wb_this_page; subreq != head; | |
505 | subreq = subreq->wb_this_page) { | |
74a6d4b5 | 506 | |
1bd5d6d0 TM |
507 | if (!kref_get_unless_zero(&subreq->wb_kref)) { |
508 | if (subreq->wb_offset == head->wb_offset + total_bytes) | |
509 | total_bytes += subreq->wb_bytes; | |
5b2b5187 | 510 | continue; |
1bd5d6d0 TM |
511 | } |
512 | ||
74a6d4b5 | 513 | while (!nfs_lock_request(subreq)) { |
b5bab9bf | 514 | /* |
74a6d4b5 TM |
515 | * Unlock page to allow nfs_page_group_sync_on_bit() |
516 | * to succeed | |
b5bab9bf | 517 | */ |
74a6d4b5 TM |
518 | nfs_page_group_unlock(head); |
519 | ret = nfs_wait_on_request(subreq); | |
520 | if (!ret) | |
1344b7ea | 521 | ret = nfs_page_group_lock(head); |
74a6d4b5 TM |
522 | if (ret < 0) { |
523 | nfs_unroll_locks(inode, head, subreq); | |
5b2b5187 | 524 | nfs_release_request(subreq); |
0671d8f1 | 525 | goto release_request; |
74a6d4b5 | 526 | } |
e14bebf6 | 527 | } |
d4581383 WAA |
528 | /* |
529 | * Subrequests are always contiguous, non overlapping | |
309a1d65 | 530 | * and in order - but may be repeated (mirrored writes). |
d4581383 | 531 | */ |
309a1d65 WAA |
532 | if (subreq->wb_offset == (head->wb_offset + total_bytes)) { |
533 | /* keep track of how many bytes this group covers */ | |
534 | total_bytes += subreq->wb_bytes; | |
535 | } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || | |
536 | ((subreq->wb_offset + subreq->wb_bytes) > | |
537 | (head->wb_offset + total_bytes)))) { | |
8b77484f | 538 | nfs_page_group_unlock(head); |
74a6d4b5 | 539 | nfs_unroll_locks(inode, head, subreq); |
5b2b5187 | 540 | nfs_unlock_and_release_request(subreq); |
0671d8f1 ME |
541 | ret = -EIO; |
542 | goto release_request; | |
309a1d65 | 543 | } |
a0e265bc | 544 | } |
d4581383 WAA |
545 | |
546 | /* Now that all requests are locked, make sure they aren't on any list. | |
547 | * Commit list removal accounting is done after locks are dropped */ | |
548 | subreq = head; | |
549 | do { | |
411a99ad | 550 | nfs_clear_request_commit(subreq); |
d4581383 WAA |
551 | subreq = subreq->wb_this_page; |
552 | } while (subreq != head); | |
553 | ||
554 | /* unlink subrequests from head, destroy them later */ | |
555 | if (head->wb_this_page != head) { | |
556 | /* destroy list will be terminated by head */ | |
557 | destroy_list = head->wb_this_page; | |
558 | head->wb_this_page = head; | |
559 | ||
560 | /* change head request to cover whole range that | |
561 | * the former page group covered */ | |
562 | head->wb_bytes = total_bytes; | |
e261f51f | 563 | } |
d4581383 | 564 | |
b66aaa8d TM |
565 | /* Postpone destruction of this request */ |
566 | if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { | |
567 | set_bit(PG_INODE_REF, &head->wb_flags); | |
568 | kref_get(&head->wb_kref); | |
a6b6d5b8 | 569 | atomic_long_inc(&NFS_I(inode)->nrequests); |
b66aaa8d TM |
570 | } |
571 | ||
d4581383 WAA |
572 | nfs_page_group_unlock(head); |
573 | ||
b66aaa8d | 574 | nfs_destroy_unlinked_subrequests(destroy_list, head, inode); |
d4581383 | 575 | |
b5bab9bf TM |
576 | /* Did we lose a race with nfs_inode_remove_request()? */ |
577 | if (!(PagePrivate(page) || PageSwapCache(page))) { | |
578 | nfs_unlock_and_release_request(head); | |
579 | return NULL; | |
580 | } | |
581 | ||
bd37d6fc | 582 | /* still holds ref on head from nfs_page_find_head_request |
d4581383 WAA |
583 | * and still has lock on head from lock loop */ |
584 | return head; | |
0671d8f1 ME |
585 | |
586 | release_request: | |
587 | nfs_unlock_and_release_request(head); | |
588 | return ERR_PTR(ret); | |
074cc1de TM |
589 | } |
590 | ||
6fbda89b | 591 | static void nfs_write_error(struct nfs_page *req, int error) |
0bcbf039 | 592 | { |
6fbda89b | 593 | nfs_mapping_set_error(req->wb_page, error); |
0bcbf039 | 594 | nfs_end_page_writeback(req); |
1f84ccdf | 595 | nfs_release_request(req); |
0bcbf039 PT |
596 | } |
597 | ||
a6598813 TM |
598 | static bool |
599 | nfs_error_is_fatal_on_server(int err) | |
600 | { | |
601 | switch (err) { | |
602 | case 0: | |
603 | case -ERESTARTSYS: | |
604 | case -EINTR: | |
605 | return false; | |
606 | } | |
607 | return nfs_error_is_fatal(err); | |
0bcbf039 PT |
608 | } |
609 | ||
074cc1de TM |
610 | /* |
611 | * Find an associated nfs write request, and prepare to flush it out | |
612 | * May return an error if the user signalled nfs_wait_on_request(). | |
613 | */ | |
614 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
6d17d653 | 615 | struct page *page) |
074cc1de | 616 | { |
6fbda89b | 617 | struct address_space *mapping; |
074cc1de TM |
618 | struct nfs_page *req; |
619 | int ret = 0; | |
620 | ||
6d17d653 | 621 | req = nfs_lock_and_join_requests(page); |
074cc1de TM |
622 | if (!req) |
623 | goto out; | |
624 | ret = PTR_ERR(req); | |
625 | if (IS_ERR(req)) | |
626 | goto out; | |
627 | ||
deed85e7 TM |
628 | nfs_set_page_writeback(page); |
629 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 630 | |
a6598813 | 631 | /* If there is a fatal error that covers this write, just exit */ |
6fbda89b TM |
632 | ret = 0; |
633 | mapping = page_file_mapping(page); | |
634 | if (test_bit(AS_ENOSPC, &mapping->flags) || | |
635 | test_bit(AS_EIO, &mapping->flags)) | |
a6598813 TM |
636 | goto out_launder; |
637 | ||
f8512ad0 | 638 | if (!nfs_pageio_add_request(pgio, req)) { |
074cc1de | 639 | ret = pgio->pg_error; |
0bcbf039 | 640 | /* |
c373fff7 | 641 | * Remove the problematic req upon fatal errors on the server |
0bcbf039 PT |
642 | */ |
643 | if (nfs_error_is_fatal(ret)) { | |
c373fff7 | 644 | if (nfs_error_is_fatal_on_server(ret)) |
a6598813 | 645 | goto out_launder; |
8fc75bed TM |
646 | } else |
647 | ret = -EAGAIN; | |
d6c843b9 | 648 | nfs_redirty_request(req); |
40f90271 TM |
649 | } else |
650 | nfs_add_stats(page_file_mapping(page)->host, | |
651 | NFSIOS_WRITEPAGES, 1); | |
074cc1de TM |
652 | out: |
653 | return ret; | |
a6598813 | 654 | out_launder: |
6fbda89b | 655 | nfs_write_error(req, ret); |
14bebe3c | 656 | return 0; |
e261f51f TM |
657 | } |
658 | ||
d6c843b9 | 659 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, |
c373fff7 | 660 | struct nfs_pageio_descriptor *pgio) |
1da177e4 | 661 | { |
cfb506e1 | 662 | int ret; |
1da177e4 | 663 | |
8cd79788 | 664 | nfs_pageio_cond_complete(pgio, page_index(page)); |
6d17d653 | 665 | ret = nfs_page_async_flush(pgio, page); |
cfb506e1 TM |
666 | if (ret == -EAGAIN) { |
667 | redirty_page_for_writepage(wbc, page); | |
668 | ret = 0; | |
669 | } | |
670 | return ret; | |
f758c885 | 671 | } |
7fe7f848 | 672 | |
f758c885 TM |
673 | /* |
674 | * Write an mmapped page to the server. | |
675 | */ | |
d6c843b9 | 676 | static int nfs_writepage_locked(struct page *page, |
c373fff7 | 677 | struct writeback_control *wbc) |
f758c885 TM |
678 | { |
679 | struct nfs_pageio_descriptor pgio; | |
40f90271 | 680 | struct inode *inode = page_file_mapping(page)->host; |
f758c885 | 681 | int err; |
49a70f27 | 682 | |
40f90271 | 683 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
811ed92e | 684 | nfs_pageio_init_write(&pgio, inode, 0, |
a20c93e3 | 685 | false, &nfs_async_write_completion_ops); |
c373fff7 | 686 | err = nfs_do_writepage(page, wbc, &pgio); |
f758c885 TM |
687 | nfs_pageio_complete(&pgio); |
688 | if (err < 0) | |
689 | return err; | |
690 | if (pgio.pg_error < 0) | |
691 | return pgio.pg_error; | |
692 | return 0; | |
4d770ccf TM |
693 | } |
694 | ||
695 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
696 | { | |
f758c885 | 697 | int ret; |
4d770ccf | 698 | |
c373fff7 | 699 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 700 | unlock_page(page); |
f758c885 TM |
701 | return ret; |
702 | } | |
703 | ||
704 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
705 | { | |
706 | int ret; | |
707 | ||
c373fff7 | 708 | ret = nfs_do_writepage(page, wbc, data); |
f758c885 TM |
709 | unlock_page(page); |
710 | return ret; | |
1da177e4 LT |
711 | } |
712 | ||
919e3bd9 TM |
713 | static void nfs_io_completion_commit(void *inode) |
714 | { | |
715 | nfs_commit_inode(inode, 0); | |
716 | } | |
717 | ||
1da177e4 LT |
718 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
719 | { | |
1da177e4 | 720 | struct inode *inode = mapping->host; |
c63c7b05 | 721 | struct nfs_pageio_descriptor pgio; |
875bc3fb TM |
722 | struct nfs_io_completion *ioc; |
723 | unsigned int pflags = memalloc_nofs_save(); | |
1da177e4 LT |
724 | int err; |
725 | ||
91d5b470 CL |
726 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
727 | ||
875bc3fb | 728 | ioc = nfs_io_completion_alloc(GFP_NOFS); |
919e3bd9 TM |
729 | if (ioc) |
730 | nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); | |
731 | ||
a20c93e3 CH |
732 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
733 | &nfs_async_write_completion_ops); | |
919e3bd9 | 734 | pgio.pg_io_completion = ioc; |
f758c885 | 735 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 736 | nfs_pageio_complete(&pgio); |
919e3bd9 | 737 | nfs_io_completion_put(ioc); |
72cb77f4 | 738 | |
875bc3fb TM |
739 | memalloc_nofs_restore(pflags); |
740 | ||
f758c885 | 741 | if (err < 0) |
72cb77f4 TM |
742 | goto out_err; |
743 | err = pgio.pg_error; | |
744 | if (err < 0) | |
745 | goto out_err; | |
c63c7b05 | 746 | return 0; |
72cb77f4 TM |
747 | out_err: |
748 | return err; | |
1da177e4 LT |
749 | } |
750 | ||
751 | /* | |
752 | * Insert a write request into an inode | |
753 | */ | |
d6d6dc7c | 754 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 | 755 | { |
4b9bb25b | 756 | struct address_space *mapping = page_file_mapping(req->wb_page); |
1da177e4 | 757 | struct nfs_inode *nfsi = NFS_I(inode); |
e7d39069 | 758 | |
2bfc6e56 WAA |
759 | WARN_ON_ONCE(req->wb_this_page != req); |
760 | ||
e7d39069 | 761 | /* Lock the request! */ |
7ad84aa9 | 762 | nfs_lock_request(req); |
e7d39069 | 763 | |
29418aa4 MG |
764 | /* |
765 | * Swap-space should not get truncated. Hence no need to plug the race | |
766 | * with invalidate/truncate. | |
767 | */ | |
4b9bb25b TM |
768 | spin_lock(&mapping->private_lock); |
769 | if (!nfs_have_writebacks(inode) && | |
1eb5d98f JL |
770 | NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
771 | inode_inc_iversion_raw(inode); | |
29418aa4 MG |
772 | if (likely(!PageSwapCache(req->wb_page))) { |
773 | set_bit(PG_MAPPED, &req->wb_flags); | |
774 | SetPagePrivate(req->wb_page); | |
775 | set_page_private(req->wb_page, (unsigned long)req); | |
776 | } | |
4b9bb25b | 777 | spin_unlock(&mapping->private_lock); |
a6b6d5b8 | 778 | atomic_long_inc(&nfsi->nrequests); |
17089a29 | 779 | /* this a head request for a page group - mark it as having an |
cb1410c7 WAA |
780 | * extra reference so sub groups can follow suit. |
781 | * This flag also informs pgio layer when to bump nrequests when | |
782 | * adding subrequests. */ | |
17089a29 | 783 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); |
c03b4024 | 784 | kref_get(&req->wb_kref); |
1da177e4 LT |
785 | } |
786 | ||
787 | /* | |
89a09141 | 788 | * Remove a write request from an inode |
1da177e4 LT |
789 | */ |
790 | static void nfs_inode_remove_request(struct nfs_page *req) | |
791 | { | |
4b9bb25b TM |
792 | struct address_space *mapping = page_file_mapping(req->wb_page); |
793 | struct inode *inode = mapping->host; | |
1da177e4 | 794 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 795 | struct nfs_page *head; |
1da177e4 | 796 | |
a6b6d5b8 | 797 | atomic_long_dec(&nfsi->nrequests); |
20633f04 WAA |
798 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
799 | head = req->wb_head; | |
800 | ||
4b9bb25b | 801 | spin_lock(&mapping->private_lock); |
67911c8f | 802 | if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { |
20633f04 WAA |
803 | set_page_private(head->wb_page, 0); |
804 | ClearPagePrivate(head->wb_page); | |
805 | clear_bit(PG_MAPPED, &head->wb_flags); | |
806 | } | |
4b9bb25b | 807 | spin_unlock(&mapping->private_lock); |
29418aa4 | 808 | } |
17089a29 WAA |
809 | |
810 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) | |
811 | nfs_release_request(req); | |
1da177e4 LT |
812 | } |
813 | ||
61822ab5 | 814 | static void |
6d884e8f | 815 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 816 | { |
67911c8f AS |
817 | if (req->wb_page) |
818 | __set_page_dirty_nobuffers(req->wb_page); | |
61822ab5 TM |
819 | } |
820 | ||
3a3908c8 TM |
821 | /* |
822 | * nfs_page_search_commits_for_head_request_locked | |
823 | * | |
824 | * Search through commit lists on @inode for the head request for @page. | |
825 | * Must be called while holding the inode (which is cinfo) lock. | |
826 | * | |
827 | * Returns the head request if found, or NULL if not found. | |
828 | */ | |
829 | static struct nfs_page * | |
830 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
831 | struct page *page) | |
832 | { | |
833 | struct nfs_page *freq, *t; | |
834 | struct nfs_commit_info cinfo; | |
835 | struct inode *inode = &nfsi->vfs_inode; | |
836 | ||
837 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
838 | ||
839 | /* search through pnfs commit lists */ | |
840 | freq = pnfs_search_commit_reqs(inode, &cinfo, page); | |
841 | if (freq) | |
842 | return freq->wb_head; | |
843 | ||
844 | /* Linearly search the commit list for the correct request */ | |
845 | list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { | |
846 | if (freq->wb_page == page) | |
847 | return freq->wb_head; | |
848 | } | |
849 | ||
850 | return NULL; | |
851 | } | |
852 | ||
86d80f97 TM |
853 | /** |
854 | * nfs_request_add_commit_list_locked - add request to a commit list | |
855 | * @req: pointer to a struct nfs_page | |
856 | * @dst: commit list head | |
857 | * @cinfo: holds list lock and accounting info | |
858 | * | |
859 | * This sets the PG_CLEAN bit, updates the cinfo count of | |
860 | * number of outstanding requests requiring a commit as well as | |
861 | * the MM page stats. | |
862 | * | |
e824f99a TM |
863 | * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the |
864 | * nfs_page lock. | |
86d80f97 TM |
865 | */ |
866 | void | |
867 | nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, | |
868 | struct nfs_commit_info *cinfo) | |
869 | { | |
870 | set_bit(PG_CLEAN, &req->wb_flags); | |
871 | nfs_list_add_request(req, dst); | |
5cb953d4 | 872 | atomic_long_inc(&cinfo->mds->ncommit); |
86d80f97 TM |
873 | } |
874 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); | |
875 | ||
8dd37758 TM |
876 | /** |
877 | * nfs_request_add_commit_list - add request to a commit list | |
878 | * @req: pointer to a struct nfs_page | |
ea2cf228 | 879 | * @cinfo: holds list lock and accounting info |
8dd37758 | 880 | * |
ea2cf228 | 881 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
882 | * number of outstanding requests requiring a commit as well as |
883 | * the MM page stats. | |
884 | * | |
ea2cf228 | 885 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 886 | * holding the nfs_page lock. |
1da177e4 | 887 | */ |
8dd37758 | 888 | void |
6272dcc6 | 889 | nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) |
1da177e4 | 890 | { |
e824f99a | 891 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
6272dcc6 | 892 | nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); |
e824f99a | 893 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
67911c8f AS |
894 | if (req->wb_page) |
895 | nfs_mark_page_unstable(req->wb_page, cinfo); | |
1da177e4 | 896 | } |
8dd37758 TM |
897 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
898 | ||
899 | /** | |
900 | * nfs_request_remove_commit_list - Remove request from a commit list | |
901 | * @req: pointer to a nfs_page | |
ea2cf228 | 902 | * @cinfo: holds list lock and accounting info |
8dd37758 | 903 | * |
ea2cf228 | 904 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
905 | * number of outstanding requests requiring a commit |
906 | * It does not update the MM page stats. | |
907 | * | |
ea2cf228 | 908 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
909 | */ |
910 | void | |
ea2cf228 FI |
911 | nfs_request_remove_commit_list(struct nfs_page *req, |
912 | struct nfs_commit_info *cinfo) | |
8dd37758 | 913 | { |
8dd37758 TM |
914 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
915 | return; | |
916 | nfs_list_remove_request(req); | |
5cb953d4 | 917 | atomic_long_dec(&cinfo->mds->ncommit); |
8dd37758 TM |
918 | } |
919 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
920 | ||
ea2cf228 FI |
921 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
922 | struct inode *inode) | |
923 | { | |
fe238e60 | 924 | cinfo->inode = inode; |
ea2cf228 FI |
925 | cinfo->mds = &NFS_I(inode)->commit_info; |
926 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 927 | cinfo->dreq = NULL; |
f453a54a | 928 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
929 | } |
930 | ||
931 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
932 | struct inode *inode, | |
933 | struct nfs_direct_req *dreq) | |
934 | { | |
1763da12 FI |
935 | if (dreq) |
936 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
937 | else | |
938 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
939 | } |
940 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
941 | |
942 | /* | |
943 | * Add a request to the inode's commit list. | |
944 | */ | |
1763da12 | 945 | void |
ea2cf228 | 946 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
b57ff130 | 947 | struct nfs_commit_info *cinfo, u32 ds_commit_idx) |
8dd37758 | 948 | { |
b57ff130 | 949 | if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) |
8dd37758 | 950 | return; |
6272dcc6 | 951 | nfs_request_add_commit_list(req, cinfo); |
8dd37758 | 952 | } |
8e821cad | 953 | |
d6d6dc7c FI |
954 | static void |
955 | nfs_clear_page_commit(struct page *page) | |
956 | { | |
11fb9989 | 957 | dec_node_page_state(page, NR_UNSTABLE_NFS); |
93f78d88 TH |
958 | dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, |
959 | WB_RECLAIMABLE); | |
d6d6dc7c FI |
960 | } |
961 | ||
b5bab9bf | 962 | /* Called holding the request lock on @req */ |
8dd37758 | 963 | static void |
e468bae9 TM |
964 | nfs_clear_request_commit(struct nfs_page *req) |
965 | { | |
8dd37758 | 966 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
9fcd5960 TM |
967 | struct nfs_open_context *ctx = nfs_req_openctx(req); |
968 | struct inode *inode = d_inode(ctx->dentry); | |
ea2cf228 | 969 | struct nfs_commit_info cinfo; |
e468bae9 | 970 | |
ea2cf228 | 971 | nfs_init_cinfo_from_inode(&cinfo, inode); |
e824f99a | 972 | mutex_lock(&NFS_I(inode)->commit_mutex); |
ea2cf228 | 973 | if (!pnfs_clear_request_commit(req, &cinfo)) { |
ea2cf228 | 974 | nfs_request_remove_commit_list(req, &cinfo); |
8dd37758 | 975 | } |
e824f99a | 976 | mutex_unlock(&NFS_I(inode)->commit_mutex); |
d6d6dc7c | 977 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 978 | } |
e468bae9 TM |
979 | } |
980 | ||
d45f60c6 | 981 | int nfs_write_need_commit(struct nfs_pgio_header *hdr) |
8e821cad | 982 | { |
c65e6254 | 983 | if (hdr->verf.committed == NFS_DATA_SYNC) |
d45f60c6 | 984 | return hdr->lseg == NULL; |
c65e6254 | 985 | return hdr->verf.committed != NFS_FILE_SYNC; |
8e821cad TM |
986 | } |
987 | ||
919e3bd9 TM |
988 | static void nfs_async_write_init(struct nfs_pgio_header *hdr) |
989 | { | |
990 | nfs_io_completion_get(hdr->io_completion); | |
991 | } | |
992 | ||
061ae2ed | 993 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 994 | { |
ea2cf228 | 995 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
996 | unsigned long bytes = 0; |
997 | ||
998 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
999 | goto out; | |
ea2cf228 | 1000 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
1001 | while (!list_empty(&hdr->pages)) { |
1002 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
1003 | |
1004 | bytes += req->wb_bytes; | |
1005 | nfs_list_remove_request(req); | |
1006 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
1007 | (hdr->good_bytes < bytes)) { | |
d2ceb7e5 | 1008 | nfs_set_pageerror(page_file_mapping(req->wb_page)); |
6fbda89b | 1009 | nfs_mapping_set_error(req->wb_page, hdr->error); |
6c75dc0d FI |
1010 | goto remove_req; |
1011 | } | |
c65e6254 | 1012 | if (nfs_write_need_commit(hdr)) { |
33344e0f TM |
1013 | /* Reset wb_nio, since the write was successful. */ |
1014 | req->wb_nio = 0; | |
f79d06f5 | 1015 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
b57ff130 | 1016 | nfs_mark_request_commit(req, hdr->lseg, &cinfo, |
a7d42ddb | 1017 | hdr->pgio_mirror_idx); |
6c75dc0d FI |
1018 | goto next; |
1019 | } | |
1020 | remove_req: | |
1021 | nfs_inode_remove_request(req); | |
1022 | next: | |
20633f04 | 1023 | nfs_end_page_writeback(req); |
3aff4ebb | 1024 | nfs_release_request(req); |
6c75dc0d FI |
1025 | } |
1026 | out: | |
919e3bd9 | 1027 | nfs_io_completion_put(hdr->io_completion); |
6c75dc0d | 1028 | hdr->release(hdr); |
8e821cad | 1029 | } |
1da177e4 | 1030 | |
ce59515c | 1031 | unsigned long |
ea2cf228 | 1032 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 1033 | { |
5cb953d4 | 1034 | return atomic_long_read(&cinfo->mds->ncommit); |
d6d6dc7c FI |
1035 | } |
1036 | ||
e824f99a | 1037 | /* NFS_I(cinfo->inode)->commit_mutex held by caller */ |
1763da12 | 1038 | int |
ea2cf228 FI |
1039 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
1040 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c | 1041 | { |
137da553 | 1042 | struct nfs_page *req, *tmp; |
d6d6dc7c FI |
1043 | int ret = 0; |
1044 | ||
137da553 TM |
1045 | restart: |
1046 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
7ad84aa9 | 1047 | kref_get(&req->wb_kref); |
2ce209c4 TM |
1048 | if (!nfs_lock_request(req)) { |
1049 | int status; | |
137da553 TM |
1050 | |
1051 | /* Prevent deadlock with nfs_lock_and_join_requests */ | |
1052 | if (!list_empty(dst)) { | |
1053 | nfs_release_request(req); | |
1054 | continue; | |
1055 | } | |
1056 | /* Ensure we make progress to prevent livelock */ | |
2ce209c4 TM |
1057 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
1058 | status = nfs_wait_on_request(req); | |
1059 | nfs_release_request(req); | |
1060 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); | |
1061 | if (status < 0) | |
1062 | break; | |
137da553 | 1063 | goto restart; |
2ce209c4 | 1064 | } |
ea2cf228 | 1065 | nfs_request_remove_commit_list(req, cinfo); |
5d2a9d9d | 1066 | clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); |
8dd37758 TM |
1067 | nfs_list_add_request(req, dst); |
1068 | ret++; | |
1763da12 | 1069 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 1070 | break; |
e824f99a | 1071 | cond_resched(); |
d6d6dc7c FI |
1072 | } |
1073 | return ret; | |
fb8a1f11 | 1074 | } |
5d2a9d9d | 1075 | EXPORT_SYMBOL_GPL(nfs_scan_commit_list); |
fb8a1f11 | 1076 | |
1da177e4 LT |
1077 | /* |
1078 | * nfs_scan_commit - Scan an inode for commit requests | |
1079 | * @inode: NFS inode to scan | |
ea2cf228 FI |
1080 | * @dst: mds destination list |
1081 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
1082 | * |
1083 | * Moves requests from the inode's 'commit' request list. | |
1084 | * The requests are *not* checked to ensure that they form a contiguous set. | |
1085 | */ | |
1763da12 | 1086 | int |
ea2cf228 FI |
1087 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
1088 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1089 | { |
d6d6dc7c | 1090 | int ret = 0; |
fb8a1f11 | 1091 | |
5cb953d4 TM |
1092 | if (!atomic_long_read(&cinfo->mds->ncommit)) |
1093 | return 0; | |
e824f99a | 1094 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
5cb953d4 | 1095 | if (atomic_long_read(&cinfo->mds->ncommit) > 0) { |
8dd37758 | 1096 | const int max = INT_MAX; |
d6d6dc7c | 1097 | |
ea2cf228 FI |
1098 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
1099 | cinfo, max); | |
1100 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 1101 | } |
e824f99a | 1102 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
ff778d02 | 1103 | return ret; |
1da177e4 | 1104 | } |
d6d6dc7c | 1105 | |
1da177e4 | 1106 | /* |
e7d39069 TM |
1107 | * Search for an existing write request, and attempt to update |
1108 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 1109 | * |
e7d39069 TM |
1110 | * If the attempt fails, then the existing request is flushed out |
1111 | * to disk. | |
1da177e4 | 1112 | */ |
e7d39069 TM |
1113 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
1114 | struct page *page, | |
1115 | unsigned int offset, | |
1116 | unsigned int bytes) | |
1da177e4 | 1117 | { |
e7d39069 TM |
1118 | struct nfs_page *req; |
1119 | unsigned int rqend; | |
1120 | unsigned int end; | |
1121 | int error; | |
1122 | ||
1da177e4 LT |
1123 | end = offset + bytes; |
1124 | ||
f6032f21 TM |
1125 | req = nfs_lock_and_join_requests(page); |
1126 | if (IS_ERR_OR_NULL(req)) | |
1127 | return req; | |
1da177e4 | 1128 | |
f6032f21 TM |
1129 | rqend = req->wb_offset + req->wb_bytes; |
1130 | /* | |
1131 | * Tell the caller to flush out the request if | |
1132 | * the offsets are non-contiguous. | |
1133 | * Note: nfs_flush_incompatible() will already | |
1134 | * have flushed out requests having wrong owners. | |
1135 | */ | |
1136 | if (offset > rqend || end < req->wb_offset) | |
1137 | goto out_flushme; | |
1da177e4 LT |
1138 | |
1139 | /* Okay, the request matches. Update the region */ | |
1140 | if (offset < req->wb_offset) { | |
1141 | req->wb_offset = offset; | |
1142 | req->wb_pgbase = offset; | |
1da177e4 | 1143 | } |
1da177e4 LT |
1144 | if (end > rqend) |
1145 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1146 | else |
1147 | req->wb_bytes = rqend - req->wb_offset; | |
33344e0f | 1148 | req->wb_nio = 0; |
e7d39069 TM |
1149 | return req; |
1150 | out_flushme: | |
f6032f21 TM |
1151 | /* |
1152 | * Note: we mark the request dirty here because | |
1153 | * nfs_lock_and_join_requests() cannot preserve | |
1154 | * commit flags, so we have to replay the write. | |
1155 | */ | |
1156 | nfs_mark_request_dirty(req); | |
1157 | nfs_unlock_and_release_request(req); | |
e7d39069 | 1158 | error = nfs_wb_page(inode, page); |
f6032f21 | 1159 | return (error < 0) ? ERR_PTR(error) : NULL; |
e7d39069 TM |
1160 | } |
1161 | ||
1162 | /* | |
1163 | * Try to update an existing write request, or create one if there is none. | |
1164 | * | |
1165 | * Note: Should always be called with the Page Lock held to prevent races | |
1166 | * if we have to add a new request. Also assumes that the caller has | |
1167 | * already called nfs_flush_incompatible() if necessary. | |
1168 | */ | |
1169 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1170 | struct page *page, unsigned int offset, unsigned int bytes) | |
1171 | { | |
d56b4ddf | 1172 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1173 | struct nfs_page *req; |
1da177e4 | 1174 | |
e7d39069 TM |
1175 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1176 | if (req != NULL) | |
1177 | goto out; | |
28b1d3f5 | 1178 | req = nfs_create_request(ctx, page, offset, bytes); |
e7d39069 TM |
1179 | if (IS_ERR(req)) |
1180 | goto out; | |
d6d6dc7c | 1181 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1182 | out: |
61e930a9 | 1183 | return req; |
1da177e4 LT |
1184 | } |
1185 | ||
e7d39069 TM |
1186 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1187 | unsigned int offset, unsigned int count) | |
1188 | { | |
1189 | struct nfs_page *req; | |
1190 | ||
1191 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1192 | if (IS_ERR(req)) | |
1193 | return PTR_ERR(req); | |
1194 | /* Update file length */ | |
1195 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1196 | nfs_mark_uptodate(req); |
a6305ddb | 1197 | nfs_mark_request_dirty(req); |
1d1afcbc | 1198 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1199 | return 0; |
1200 | } | |
1201 | ||
1da177e4 LT |
1202 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1203 | { | |
cd3758e3 | 1204 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1205 | struct nfs_lock_context *l_ctx; |
bd61e0a9 | 1206 | struct file_lock_context *flctx = file_inode(file)->i_flctx; |
1da177e4 | 1207 | struct nfs_page *req; |
1a54533e | 1208 | int do_flush, status; |
1da177e4 LT |
1209 | /* |
1210 | * Look for a request corresponding to this page. If there | |
1211 | * is one, and it belongs to another file, we flush it out | |
1212 | * before we try to copy anything into the page. Do this | |
1213 | * due to the lack of an ACCESS-type call in NFSv2. | |
1214 | * Also do the same if we find a request from an existing | |
1215 | * dropped page. | |
1216 | */ | |
1a54533e | 1217 | do { |
84d3a9a9 | 1218 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1219 | if (req == NULL) |
1220 | return 0; | |
2a369153 | 1221 | l_ctx = req->wb_lock_context; |
138a2935 | 1222 | do_flush = req->wb_page != page || |
9fcd5960 | 1223 | !nfs_match_open_context(nfs_req_openctx(req), ctx); |
bd61e0a9 JL |
1224 | if (l_ctx && flctx && |
1225 | !(list_empty_careful(&flctx->flc_posix) && | |
1226 | list_empty_careful(&flctx->flc_flock))) { | |
d51fdb87 | 1227 | do_flush |= l_ctx->lockowner != current->files; |
5263e31e | 1228 | } |
1da177e4 | 1229 | nfs_release_request(req); |
1a54533e TM |
1230 | if (!do_flush) |
1231 | return 0; | |
d56b4ddf | 1232 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1233 | } while (status == 0); |
1234 | return status; | |
1da177e4 LT |
1235 | } |
1236 | ||
dc24826b AA |
1237 | /* |
1238 | * Avoid buffered writes when a open context credential's key would | |
1239 | * expire soon. | |
1240 | * | |
1241 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1242 | * | |
1243 | * Return 0 and set a credential flag which triggers the inode to flush | |
1244 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1245 | * RPC_KEY_EXPIRE_TIMEO. | |
1246 | */ | |
1247 | int | |
1248 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1249 | { | |
1250 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
dc24826b | 1251 | |
ddf529ee N |
1252 | if (nfs_ctx_key_to_expire(ctx, inode) && |
1253 | !ctx->ll_cred) | |
1254 | /* Already expired! */ | |
1255 | return -EACCES; | |
1256 | return 0; | |
dc24826b AA |
1257 | } |
1258 | ||
1259 | /* | |
1260 | * Test if the open context credential key is marked to expire soon. | |
1261 | */ | |
ce52914e | 1262 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) |
dc24826b | 1263 | { |
ce52914e | 1264 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; |
ddf529ee N |
1265 | struct rpc_cred *cred = ctx->ll_cred; |
1266 | struct auth_cred acred = { | |
a52458b4 | 1267 | .cred = ctx->cred, |
ddf529ee N |
1268 | }; |
1269 | ||
1270 | if (cred && !cred->cr_ops->crmatch(&acred, cred, 0)) { | |
1271 | put_rpccred(cred); | |
1272 | ctx->ll_cred = NULL; | |
1273 | cred = NULL; | |
1274 | } | |
1275 | if (!cred) | |
1276 | cred = auth->au_ops->lookup_cred(auth, &acred, 0); | |
1277 | if (!cred || IS_ERR(cred)) | |
1278 | return true; | |
1279 | ctx->ll_cred = cred; | |
1280 | return !!(cred->cr_ops->crkey_timeout && | |
1281 | cred->cr_ops->crkey_timeout(cred)); | |
dc24826b AA |
1282 | } |
1283 | ||
5d47a356 TM |
1284 | /* |
1285 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1286 | * the PageUptodate() flag. In this case, we will need to turn off | |
1287 | * write optimisations that depend on the page contents being correct. | |
1288 | */ | |
8d197a56 | 1289 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 1290 | { |
d529ef83 JL |
1291 | struct nfs_inode *nfsi = NFS_I(inode); |
1292 | ||
8d197a56 TM |
1293 | if (nfs_have_delegated_attributes(inode)) |
1294 | goto out; | |
18dd78c4 | 1295 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
d529ef83 | 1296 | return false; |
4db72b40 | 1297 | smp_rmb(); |
d529ef83 | 1298 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
1299 | return false; |
1300 | out: | |
18dd78c4 SM |
1301 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
1302 | return false; | |
8d197a56 | 1303 | return PageUptodate(page) != 0; |
5d47a356 TM |
1304 | } |
1305 | ||
5263e31e JL |
1306 | static bool |
1307 | is_whole_file_wrlock(struct file_lock *fl) | |
1308 | { | |
1309 | return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && | |
1310 | fl->fl_type == F_WRLCK; | |
1311 | } | |
1312 | ||
c7559663 SM |
1313 | /* If we know the page is up to date, and we're not using byte range locks (or |
1314 | * if we have the whole file locked for writing), it may be more efficient to | |
1315 | * extend the write to cover the entire page in order to avoid fragmentation | |
1316 | * inefficiencies. | |
1317 | * | |
263b4509 SM |
1318 | * If the file is opened for synchronous writes then we can just skip the rest |
1319 | * of the checks. | |
c7559663 SM |
1320 | */ |
1321 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
1322 | { | |
5263e31e JL |
1323 | int ret; |
1324 | struct file_lock_context *flctx = inode->i_flctx; | |
1325 | struct file_lock *fl; | |
1326 | ||
c7559663 SM |
1327 | if (file->f_flags & O_DSYNC) |
1328 | return 0; | |
263b4509 SM |
1329 | if (!nfs_write_pageuptodate(page, inode)) |
1330 | return 0; | |
c7559663 SM |
1331 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1332 | return 1; | |
bd61e0a9 JL |
1333 | if (!flctx || (list_empty_careful(&flctx->flc_flock) && |
1334 | list_empty_careful(&flctx->flc_posix))) | |
8fa4592a | 1335 | return 1; |
5263e31e JL |
1336 | |
1337 | /* Check to see if there are whole file write locks */ | |
5263e31e | 1338 | ret = 0; |
6109c850 | 1339 | spin_lock(&flctx->flc_lock); |
bd61e0a9 JL |
1340 | if (!list_empty(&flctx->flc_posix)) { |
1341 | fl = list_first_entry(&flctx->flc_posix, struct file_lock, | |
1342 | fl_list); | |
1343 | if (is_whole_file_wrlock(fl)) | |
1344 | ret = 1; | |
1345 | } else if (!list_empty(&flctx->flc_flock)) { | |
5263e31e JL |
1346 | fl = list_first_entry(&flctx->flc_flock, struct file_lock, |
1347 | fl_list); | |
1348 | if (fl->fl_type == F_WRLCK) | |
1349 | ret = 1; | |
1350 | } | |
6109c850 | 1351 | spin_unlock(&flctx->flc_lock); |
5263e31e | 1352 | return ret; |
c7559663 SM |
1353 | } |
1354 | ||
1da177e4 LT |
1355 | /* |
1356 | * Update and possibly write a cached page of an NFS file. | |
1357 | * | |
1358 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1359 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1360 | */ | |
1361 | int nfs_updatepage(struct file *file, struct page *page, | |
1362 | unsigned int offset, unsigned int count) | |
1363 | { | |
cd3758e3 | 1364 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d2ceb7e5 BC |
1365 | struct address_space *mapping = page_file_mapping(page); |
1366 | struct inode *inode = mapping->host; | |
1da177e4 LT |
1367 | int status = 0; |
1368 | ||
91d5b470 CL |
1369 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1370 | ||
6de1472f AV |
1371 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1372 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1373 | |
149a4fdd BC |
1374 | if (!count) |
1375 | goto out; | |
1376 | ||
c7559663 | 1377 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 1378 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 1379 | offset = 0; |
1da177e4 LT |
1380 | } |
1381 | ||
e21195a7 | 1382 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 | 1383 | if (status < 0) |
d2ceb7e5 | 1384 | nfs_set_pageerror(mapping); |
59b7c05f TM |
1385 | else |
1386 | __set_page_dirty_nobuffers(page); | |
149a4fdd | 1387 | out: |
48186c7d | 1388 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1389 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1390 | return status; |
1391 | } | |
1392 | ||
3ff7576d | 1393 | static int flush_task_priority(int how) |
1da177e4 LT |
1394 | { |
1395 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1396 | case FLUSH_HIGHPRI: | |
1397 | return RPC_PRIORITY_HIGH; | |
1398 | case FLUSH_LOWPRI: | |
1399 | return RPC_PRIORITY_LOW; | |
1400 | } | |
1401 | return RPC_PRIORITY_NORMAL; | |
1402 | } | |
1403 | ||
d45f60c6 WAA |
1404 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1405 | struct rpc_message *msg, | |
abde71f4 | 1406 | const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 | 1407 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1408 | { |
3ff7576d | 1409 | int priority = flush_task_priority(how); |
d138d5d1 | 1410 | |
1ed26f33 | 1411 | task_setup_data->priority = priority; |
fb91fb0e | 1412 | rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); |
8224b273 CL |
1413 | trace_nfs_initiate_write(hdr->inode, hdr->io_start, hdr->good_bytes, |
1414 | hdr->args.stable); | |
275acaaf TM |
1415 | } |
1416 | ||
6d884e8f F |
1417 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1418 | * call this on each, which will prepare them to be retried on next | |
1419 | * writeback using standard nfs. | |
1420 | */ | |
1421 | static void nfs_redirty_request(struct nfs_page *req) | |
1422 | { | |
33344e0f TM |
1423 | /* Bump the transmission count */ |
1424 | req->wb_nio++; | |
6d884e8f | 1425 | nfs_mark_request_dirty(req); |
9fcd5960 | 1426 | set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); |
20633f04 | 1427 | nfs_end_page_writeback(req); |
3aff4ebb | 1428 | nfs_release_request(req); |
6d884e8f F |
1429 | } |
1430 | ||
df3accb8 | 1431 | static void nfs_async_write_error(struct list_head *head, int error) |
6c75dc0d FI |
1432 | { |
1433 | struct nfs_page *req; | |
1434 | ||
1435 | while (!list_empty(head)) { | |
1436 | req = nfs_list_entry(head->next); | |
1437 | nfs_list_remove_request(req); | |
6fbda89b TM |
1438 | if (nfs_error_is_fatal(error)) |
1439 | nfs_write_error(req, error); | |
1440 | else | |
1441 | nfs_redirty_request(req); | |
6c75dc0d FI |
1442 | } |
1443 | } | |
1444 | ||
dc602dd7 TM |
1445 | static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) |
1446 | { | |
df3accb8 | 1447 | nfs_async_write_error(&hdr->pages, 0); |
7be7b3ca TM |
1448 | filemap_fdatawrite_range(hdr->inode->i_mapping, hdr->args.offset, |
1449 | hdr->args.offset + hdr->args.count - 1); | |
dc602dd7 TM |
1450 | } |
1451 | ||
061ae2ed | 1452 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
919e3bd9 | 1453 | .init_hdr = nfs_async_write_init, |
061ae2ed FI |
1454 | .error_cleanup = nfs_async_write_error, |
1455 | .completion = nfs_write_completion, | |
dc602dd7 | 1456 | .reschedule_io = nfs_async_write_reschedule_io, |
061ae2ed FI |
1457 | }; |
1458 | ||
57208fa7 | 1459 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1460 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1461 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1462 | { |
a20c93e3 | 1463 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1464 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1465 | |
1466 | #ifdef CONFIG_NFS_V4_1 | |
1467 | if (server->pnfs_curr_ld && !force_mds) | |
1468 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1469 | #endif | |
4a0de55c | 1470 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
3bde7afd | 1471 | server->wsize, ioflags); |
1751c363 | 1472 | } |
ddda8e0a | 1473 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1474 | |
dce81290 TM |
1475 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1476 | { | |
a7d42ddb WAA |
1477 | struct nfs_pgio_mirror *mirror; |
1478 | ||
6f29b9bb KM |
1479 | if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) |
1480 | pgio->pg_ops->pg_cleanup(pgio); | |
1481 | ||
41d8d5b7 | 1482 | pgio->pg_ops = &nfs_pgio_rw_ops; |
a7d42ddb WAA |
1483 | |
1484 | nfs_pageio_stop_mirroring(pgio); | |
1485 | ||
1486 | mirror = &pgio->pg_mirrors[0]; | |
1487 | mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
dce81290 | 1488 | } |
1f945357 | 1489 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1490 | |
1da177e4 | 1491 | |
0b7c0153 FI |
1492 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1493 | { | |
1494 | struct nfs_commit_data *data = calldata; | |
1495 | ||
1496 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1497 | } | |
1498 | ||
1f2edbe3 TM |
1499 | /* |
1500 | * Special version of should_remove_suid() that ignores capabilities. | |
1501 | */ | |
1502 | static int nfs_should_remove_suid(const struct inode *inode) | |
1503 | { | |
1504 | umode_t mode = inode->i_mode; | |
1505 | int kill = 0; | |
1506 | ||
1507 | /* suid always must be killed */ | |
1508 | if (unlikely(mode & S_ISUID)) | |
1509 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1510 | |
1f2edbe3 TM |
1511 | /* |
1512 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1513 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1514 | */ | |
1515 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1516 | kill |= ATTR_KILL_SGID; | |
1517 | ||
1518 | if (unlikely(kill && S_ISREG(mode))) | |
1519 | return kill; | |
1520 | ||
1521 | return 0; | |
1522 | } | |
788e7a89 | 1523 | |
a08a8cd3 TM |
1524 | static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, |
1525 | struct nfs_fattr *fattr) | |
1526 | { | |
1527 | struct nfs_pgio_args *argp = &hdr->args; | |
1528 | struct nfs_pgio_res *resp = &hdr->res; | |
2b83d3de | 1529 | u64 size = argp->offset + resp->count; |
a08a8cd3 TM |
1530 | |
1531 | if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) | |
2b83d3de TM |
1532 | fattr->size = size; |
1533 | if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { | |
1534 | fattr->valid &= ~NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 | 1535 | return; |
2b83d3de TM |
1536 | } |
1537 | if (size != fattr->size) | |
a08a8cd3 TM |
1538 | return; |
1539 | /* Set attribute barrier */ | |
1540 | nfs_fattr_set_barrier(fattr); | |
2b83d3de TM |
1541 | /* ...and update size */ |
1542 | fattr->valid |= NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 TM |
1543 | } |
1544 | ||
1545 | void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) | |
1546 | { | |
2b83d3de | 1547 | struct nfs_fattr *fattr = &hdr->fattr; |
a08a8cd3 TM |
1548 | struct inode *inode = hdr->inode; |
1549 | ||
a08a8cd3 TM |
1550 | spin_lock(&inode->i_lock); |
1551 | nfs_writeback_check_extend(hdr, fattr); | |
1552 | nfs_post_op_update_inode_force_wcc_locked(inode, fattr); | |
1553 | spin_unlock(&inode->i_lock); | |
1554 | } | |
1555 | EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); | |
1556 | ||
1da177e4 LT |
1557 | /* |
1558 | * This function is called when the WRITE call is complete. | |
1559 | */ | |
d45f60c6 WAA |
1560 | static int nfs_writeback_done(struct rpc_task *task, |
1561 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1562 | struct inode *inode) |
1da177e4 | 1563 | { |
788e7a89 | 1564 | int status; |
1da177e4 | 1565 | |
f551e44f CL |
1566 | /* |
1567 | * ->write_done will attempt to use post-op attributes to detect | |
1568 | * conflicting writes by other clients. A strict interpretation | |
1569 | * of close-to-open would allow us to continue caching even if | |
1570 | * another writer had changed the file, but some applications | |
1571 | * depend on tighter cache coherency when writing. | |
1572 | */ | |
d45f60c6 | 1573 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1574 | if (status != 0) |
0eecb214 | 1575 | return status; |
8224b273 | 1576 | |
d45f60c6 | 1577 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
8224b273 CL |
1578 | trace_nfs_writeback_done(inode, task->tk_status, |
1579 | hdr->args.offset, hdr->res.verf); | |
91d5b470 | 1580 | |
d45f60c6 WAA |
1581 | if (hdr->res.verf->committed < hdr->args.stable && |
1582 | task->tk_status >= 0) { | |
1da177e4 LT |
1583 | /* We tried a write call, but the server did not |
1584 | * commit data to stable storage even though we | |
1585 | * requested it. | |
1586 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1587 | * the server reports NFS_DATA_SYNC, but performs | |
1588 | * NFS_FILE_SYNC. We therefore implement this checking | |
1589 | * as a dprintk() in order to avoid filling syslog. | |
1590 | */ | |
1591 | static unsigned long complain; | |
1592 | ||
a69aef14 | 1593 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1594 | if (time_before(complain, jiffies)) { |
48186c7d | 1595 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1596 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1597 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
d45f60c6 | 1598 | hdr->res.verf->committed, hdr->args.stable); |
1da177e4 LT |
1599 | complain = jiffies + 300 * HZ; |
1600 | } | |
1601 | } | |
1f2edbe3 TM |
1602 | |
1603 | /* Deal with the suid/sgid bit corner case */ | |
16e14375 TM |
1604 | if (nfs_should_remove_suid(inode)) { |
1605 | spin_lock(&inode->i_lock); | |
1606 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_OTHER; | |
1607 | spin_unlock(&inode->i_lock); | |
1608 | } | |
0eecb214 AS |
1609 | return 0; |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * This function is called when the WRITE call is complete. | |
1614 | */ | |
d45f60c6 WAA |
1615 | static void nfs_writeback_result(struct rpc_task *task, |
1616 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1617 | { |
d45f60c6 WAA |
1618 | struct nfs_pgio_args *argp = &hdr->args; |
1619 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1620 | |
1621 | if (resp->count < argp->count) { | |
1da177e4 LT |
1622 | static unsigned long complain; |
1623 | ||
6c75dc0d | 1624 | /* This a short write! */ |
d45f60c6 | 1625 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1626 | |
1da177e4 | 1627 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1628 | if (resp->count == 0) { |
1629 | if (time_before(complain, jiffies)) { | |
1630 | printk(KERN_WARNING | |
1631 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1632 | argp->count); | |
1633 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1634 | } |
d45f60c6 | 1635 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1636 | task->tk_status = -EIO; |
13602896 | 1637 | return; |
1da177e4 | 1638 | } |
f8417b48 KM |
1639 | |
1640 | /* For non rpc-based layout drivers, retry-through-MDS */ | |
1641 | if (!task->tk_ops) { | |
1642 | hdr->pnfs_error = -EAGAIN; | |
1643 | return; | |
1644 | } | |
1645 | ||
6c75dc0d FI |
1646 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1647 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1648 | /* Resend from where the server left off */ | |
d45f60c6 | 1649 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1650 | argp->offset += resp->count; |
1651 | argp->pgbase += resp->count; | |
1652 | argp->count -= resp->count; | |
1653 | } else { | |
1654 | /* Resend as a stable write in order to avoid | |
1655 | * headaches in the case of a server crash. | |
1656 | */ | |
1657 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1658 | } |
6c75dc0d | 1659 | rpc_restart_call_prepare(task); |
1da177e4 | 1660 | } |
1da177e4 LT |
1661 | } |
1662 | ||
af7cf057 | 1663 | static int wait_on_commit(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1664 | { |
723c921e PZ |
1665 | return wait_var_event_killable(&cinfo->rpcs_out, |
1666 | !atomic_read(&cinfo->rpcs_out)); | |
af7cf057 | 1667 | } |
b8413f98 | 1668 | |
af7cf057 TM |
1669 | static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) |
1670 | { | |
1671 | atomic_inc(&cinfo->rpcs_out); | |
71d0a611 TM |
1672 | } |
1673 | ||
af7cf057 | 1674 | static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1675 | { |
af7cf057 | 1676 | if (atomic_dec_and_test(&cinfo->rpcs_out)) |
723c921e | 1677 | wake_up_var(&cinfo->rpcs_out); |
71d0a611 TM |
1678 | } |
1679 | ||
0b7c0153 | 1680 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1681 | { |
0b7c0153 FI |
1682 | put_nfs_open_context(data->context); |
1683 | nfs_commit_free(data); | |
1da177e4 | 1684 | } |
e0c2b380 | 1685 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1686 | |
0b7c0153 | 1687 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
c36aae9a | 1688 | const struct nfs_rpc_ops *nfs_ops, |
9ace33cd | 1689 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1690 | int how, int flags) |
1da177e4 | 1691 | { |
07737691 | 1692 | struct rpc_task *task; |
9ace33cd | 1693 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1694 | struct rpc_message msg = { |
1695 | .rpc_argp = &data->args, | |
1696 | .rpc_resp = &data->res, | |
9ace33cd | 1697 | .rpc_cred = data->cred, |
bdc7f021 | 1698 | }; |
84115e1c | 1699 | struct rpc_task_setup task_setup_data = { |
07737691 | 1700 | .task = &data->task, |
9ace33cd | 1701 | .rpc_client = clnt, |
bdc7f021 | 1702 | .rpc_message = &msg, |
9ace33cd | 1703 | .callback_ops = call_ops, |
84115e1c | 1704 | .callback_data = data, |
101070ca | 1705 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1706 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1707 | .priority = priority, |
84115e1c | 1708 | }; |
9ace33cd | 1709 | /* Set up the initial task struct. */ |
e9ae1ee2 | 1710 | nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); |
8224b273 | 1711 | trace_nfs_initiate_commit(data); |
9ace33cd | 1712 | |
b4839ebe | 1713 | dprintk("NFS: initiated commit call\n"); |
9ace33cd FI |
1714 | |
1715 | task = rpc_run_task(&task_setup_data); | |
1716 | if (IS_ERR(task)) | |
1717 | return PTR_ERR(task); | |
1718 | if (how & FLUSH_SYNC) | |
1719 | rpc_wait_for_completion_task(task); | |
1720 | rpc_put_task(task); | |
1721 | return 0; | |
1722 | } | |
e0c2b380 | 1723 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1724 | |
378520b8 PT |
1725 | static loff_t nfs_get_lwb(struct list_head *head) |
1726 | { | |
1727 | loff_t lwb = 0; | |
1728 | struct nfs_page *req; | |
1729 | ||
1730 | list_for_each_entry(req, head, wb_list) | |
1731 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1732 | lwb = req_offset(req) + req->wb_bytes; | |
1733 | ||
1734 | return lwb; | |
1735 | } | |
1736 | ||
9ace33cd FI |
1737 | /* |
1738 | * Set up the argument/result storage required for the RPC call. | |
1739 | */ | |
0b7c0153 | 1740 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1741 | struct list_head *head, |
1742 | struct pnfs_layout_segment *lseg, | |
1743 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1744 | { |
1745 | struct nfs_page *first = nfs_list_entry(head->next); | |
9fcd5960 TM |
1746 | struct nfs_open_context *ctx = nfs_req_openctx(first); |
1747 | struct inode *inode = d_inode(ctx->dentry); | |
1da177e4 LT |
1748 | |
1749 | /* Set up the RPC argument and reply structs | |
1750 | * NB: take care not to mess about with data->commit et al. */ | |
1751 | ||
1752 | list_splice_init(head, &data->pages); | |
1da177e4 | 1753 | |
1da177e4 | 1754 | data->inode = inode; |
9fcd5960 | 1755 | data->cred = ctx->cred; |
988b6dce | 1756 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1757 | /* only set lwb for pnfs commit */ |
1758 | if (lseg) | |
1759 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1760 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1761 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1762 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1763 | |
1764 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1765 | /* Note: we always request a commit of the entire inode */ |
1766 | data->args.offset = 0; | |
1767 | data->args.count = 0; | |
9fcd5960 | 1768 | data->context = get_nfs_open_context(ctx); |
1da177e4 LT |
1769 | data->res.fattr = &data->fattr; |
1770 | data->res.verf = &data->verf; | |
0e574af1 | 1771 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1772 | } |
e0c2b380 | 1773 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1774 | |
e0c2b380 | 1775 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 | 1776 | struct pnfs_layout_segment *lseg, |
b57ff130 WAA |
1777 | struct nfs_commit_info *cinfo, |
1778 | u32 ds_commit_idx) | |
64bfeb49 FI |
1779 | { |
1780 | struct nfs_page *req; | |
1781 | ||
1782 | while (!list_empty(page_list)) { | |
1783 | req = nfs_list_entry(page_list->next); | |
1784 | nfs_list_remove_request(req); | |
b57ff130 | 1785 | nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); |
487b9b8a TH |
1786 | if (!cinfo->dreq) |
1787 | nfs_clear_page_commit(req->wb_page); | |
1d1afcbc | 1788 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1789 | } |
1790 | } | |
e0c2b380 | 1791 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1792 | |
b20135d0 TM |
1793 | static void |
1794 | nfs_commit_resched_write(struct nfs_commit_info *cinfo, | |
1795 | struct nfs_page *req) | |
1796 | { | |
1797 | __set_page_dirty_nobuffers(req->wb_page); | |
1798 | } | |
1799 | ||
1da177e4 LT |
1800 | /* |
1801 | * Commit dirty pages | |
1802 | */ | |
1803 | static int | |
ea2cf228 FI |
1804 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1805 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1806 | { |
0b7c0153 | 1807 | struct nfs_commit_data *data; |
1da177e4 | 1808 | |
ade8febd WAA |
1809 | /* another commit raced with us */ |
1810 | if (list_empty(head)) | |
1811 | return 0; | |
1812 | ||
518662e0 | 1813 | data = nfs_commitdata_alloc(true); |
1da177e4 LT |
1814 | |
1815 | /* Set up the argument struct */ | |
f453a54a FI |
1816 | nfs_init_commit(data, head, NULL, cinfo); |
1817 | atomic_inc(&cinfo->mds->rpcs_out); | |
c36aae9a PT |
1818 | return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), |
1819 | data->mds_ops, how, 0); | |
67911c8f | 1820 | } |
67911c8f | 1821 | |
1da177e4 LT |
1822 | /* |
1823 | * COMMIT call returned | |
1824 | */ | |
788e7a89 | 1825 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1826 | { |
0b7c0153 | 1827 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1828 | |
a3f565b1 | 1829 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1830 | task->tk_pid, task->tk_status); |
1831 | ||
788e7a89 | 1832 | /* Call the NFS version-specific code */ |
c0d0e96b | 1833 | NFS_PROTO(data->inode)->commit_done(task, data); |
8224b273 | 1834 | trace_nfs_commit_done(data); |
c9d8f89d TM |
1835 | } |
1836 | ||
f453a54a | 1837 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1838 | { |
5917ce84 | 1839 | struct nfs_page *req; |
c9d8f89d | 1840 | int status = data->task.tk_status; |
f453a54a | 1841 | struct nfs_commit_info cinfo; |
353db796 | 1842 | struct nfs_server *nfss; |
788e7a89 | 1843 | |
1da177e4 LT |
1844 | while (!list_empty(&data->pages)) { |
1845 | req = nfs_list_entry(data->pages.next); | |
1846 | nfs_list_remove_request(req); | |
67911c8f AS |
1847 | if (req->wb_page) |
1848 | nfs_clear_page_commit(req->wb_page); | |
1da177e4 | 1849 | |
1e8968c5 | 1850 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
9fcd5960 TM |
1851 | nfs_req_openctx(req)->dentry->d_sb->s_id, |
1852 | (unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)), | |
1da177e4 LT |
1853 | req->wb_bytes, |
1854 | (long long)req_offset(req)); | |
c9d8f89d | 1855 | if (status < 0) { |
6fbda89b TM |
1856 | if (req->wb_page) { |
1857 | nfs_mapping_set_error(req->wb_page, status); | |
38a33101 | 1858 | nfs_inode_remove_request(req); |
6fbda89b | 1859 | } |
ddeaa637 | 1860 | dprintk_cont(", error = %d\n", status); |
1da177e4 LT |
1861 | goto next; |
1862 | } | |
1863 | ||
1864 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1865 | * returned by the server against all stored verfs. */ | |
8fc3c386 | 1866 | if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { |
1da177e4 | 1867 | /* We have a match */ |
38a33101 KM |
1868 | if (req->wb_page) |
1869 | nfs_inode_remove_request(req); | |
ddeaa637 | 1870 | dprintk_cont(" OK\n"); |
1da177e4 LT |
1871 | goto next; |
1872 | } | |
1873 | /* We have a mismatch. Write the page again */ | |
ddeaa637 | 1874 | dprintk_cont(" mismatch\n"); |
6d884e8f | 1875 | nfs_mark_request_dirty(req); |
9fcd5960 | 1876 | set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); |
1da177e4 | 1877 | next: |
1d1afcbc | 1878 | nfs_unlock_and_release_request(req); |
7f1bda44 TM |
1879 | /* Latency breaker */ |
1880 | cond_resched(); | |
1da177e4 | 1881 | } |
353db796 N |
1882 | nfss = NFS_SERVER(data->inode); |
1883 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
0db10944 | 1884 | clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); |
353db796 | 1885 | |
f453a54a | 1886 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
af7cf057 | 1887 | nfs_commit_end(cinfo.mds); |
5917ce84 FI |
1888 | } |
1889 | ||
1890 | static void nfs_commit_release(void *calldata) | |
1891 | { | |
0b7c0153 | 1892 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1893 | |
f453a54a | 1894 | data->completion_ops->completion(data); |
c9d8f89d | 1895 | nfs_commitdata_release(calldata); |
1da177e4 | 1896 | } |
788e7a89 TM |
1897 | |
1898 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1899 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1900 | .rpc_call_done = nfs_commit_done, |
1901 | .rpc_release = nfs_commit_release, | |
1902 | }; | |
1da177e4 | 1903 | |
f453a54a FI |
1904 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1905 | .completion = nfs_commit_release_pages, | |
b20135d0 | 1906 | .resched_write = nfs_commit_resched_write, |
f453a54a FI |
1907 | }; |
1908 | ||
1763da12 FI |
1909 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1910 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1911 | { |
1912 | int status; | |
1913 | ||
ea2cf228 | 1914 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1915 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1916 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1917 | return status; |
1918 | } | |
1919 | ||
c4f24df9 TM |
1920 | static int __nfs_commit_inode(struct inode *inode, int how, |
1921 | struct writeback_control *wbc) | |
1da177e4 | 1922 | { |
1da177e4 | 1923 | LIST_HEAD(head); |
ea2cf228 | 1924 | struct nfs_commit_info cinfo; |
71d0a611 | 1925 | int may_wait = how & FLUSH_SYNC; |
c4f24df9 | 1926 | int ret, nscan; |
1da177e4 | 1927 | |
ea2cf228 | 1928 | nfs_init_cinfo_from_inode(&cinfo, inode); |
af7cf057 | 1929 | nfs_commit_begin(cinfo.mds); |
c4f24df9 TM |
1930 | for (;;) { |
1931 | ret = nscan = nfs_scan_commit(inode, &head, &cinfo); | |
1932 | if (ret <= 0) | |
1933 | break; | |
1934 | ret = nfs_generic_commit_list(inode, &head, how, &cinfo); | |
1935 | if (ret < 0) | |
1936 | break; | |
1937 | ret = 0; | |
1938 | if (wbc && wbc->sync_mode == WB_SYNC_NONE) { | |
1939 | if (nscan < wbc->nr_to_write) | |
1940 | wbc->nr_to_write -= nscan; | |
1941 | else | |
1942 | wbc->nr_to_write = 0; | |
1943 | } | |
1944 | if (nscan < INT_MAX) | |
1945 | break; | |
1946 | cond_resched(); | |
1947 | } | |
af7cf057 | 1948 | nfs_commit_end(cinfo.mds); |
c4f24df9 TM |
1949 | if (ret || !may_wait) |
1950 | return ret; | |
1951 | return wait_on_commit(cinfo.mds); | |
1952 | } | |
1953 | ||
1954 | int nfs_commit_inode(struct inode *inode, int how) | |
1955 | { | |
1956 | return __nfs_commit_inode(inode, how, NULL); | |
1da177e4 | 1957 | } |
b20135d0 | 1958 | EXPORT_SYMBOL_GPL(nfs_commit_inode); |
8fc795f7 | 1959 | |
ae09c31f | 1960 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
8fc795f7 | 1961 | { |
420e3646 TM |
1962 | struct nfs_inode *nfsi = NFS_I(inode); |
1963 | int flags = FLUSH_SYNC; | |
1964 | int ret = 0; | |
8fc795f7 | 1965 | |
a00dd6c0 | 1966 | if (wbc->sync_mode == WB_SYNC_NONE) { |
c4f24df9 TM |
1967 | /* no commits means nothing needs to be done */ |
1968 | if (!atomic_long_read(&nfsi->commit_info.ncommit)) | |
1969 | goto check_requests_outstanding; | |
1970 | ||
a00dd6c0 JL |
1971 | /* Don't commit yet if this is a non-blocking flush and there |
1972 | * are a lot of outstanding writes for this mapping. | |
1973 | */ | |
1a4edf0f | 1974 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) |
a00dd6c0 | 1975 | goto out_mark_dirty; |
420e3646 | 1976 | |
a00dd6c0 | 1977 | /* don't wait for the COMMIT response */ |
420e3646 | 1978 | flags = 0; |
a00dd6c0 JL |
1979 | } |
1980 | ||
c4f24df9 TM |
1981 | ret = __nfs_commit_inode(inode, flags, wbc); |
1982 | if (!ret) { | |
1983 | if (flags & FLUSH_SYNC) | |
1984 | return 0; | |
1985 | } else if (atomic_long_read(&nfsi->commit_info.ncommit)) | |
1986 | goto out_mark_dirty; | |
1987 | ||
1988 | check_requests_outstanding: | |
1989 | if (!atomic_read(&nfsi->commit_info.rpcs_out)) | |
1990 | return ret; | |
420e3646 | 1991 | out_mark_dirty: |
8fc795f7 TM |
1992 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1993 | return ret; | |
1994 | } | |
89d77c8f | 1995 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1996 | |
837bb1d7 TM |
1997 | /* |
1998 | * Wrapper for filemap_write_and_wait_range() | |
1999 | * | |
2000 | * Needed for pNFS in order to ensure data becomes visible to the | |
2001 | * client. | |
2002 | */ | |
2003 | int nfs_filemap_write_and_wait_range(struct address_space *mapping, | |
2004 | loff_t lstart, loff_t lend) | |
2005 | { | |
2006 | int ret; | |
2007 | ||
2008 | ret = filemap_write_and_wait_range(mapping, lstart, lend); | |
2009 | if (ret == 0) | |
2010 | ret = pnfs_sync_inode(mapping->host, true); | |
2011 | return ret; | |
2012 | } | |
2013 | EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); | |
2014 | ||
acdc53b2 TM |
2015 | /* |
2016 | * flush the inode to disk. | |
2017 | */ | |
2018 | int nfs_wb_all(struct inode *inode) | |
34901f70 | 2019 | { |
f4ce1299 TM |
2020 | int ret; |
2021 | ||
2022 | trace_nfs_writeback_inode_enter(inode); | |
2023 | ||
5bb89b47 | 2024 | ret = filemap_write_and_wait(inode->i_mapping); |
6b196875 CL |
2025 | if (ret) |
2026 | goto out; | |
2027 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
2028 | if (ret < 0) | |
2029 | goto out; | |
2030 | pnfs_sync_inode(inode, true); | |
2031 | ret = 0; | |
34901f70 | 2032 | |
6b196875 | 2033 | out: |
f4ce1299 TM |
2034 | trace_nfs_writeback_inode_exit(inode, ret); |
2035 | return ret; | |
1c75950b | 2036 | } |
ddda8e0a | 2037 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 2038 | |
1b3b4a1a TM |
2039 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
2040 | { | |
2041 | struct nfs_page *req; | |
1b3b4a1a TM |
2042 | int ret = 0; |
2043 | ||
3e217045 WAA |
2044 | wait_on_page_writeback(page); |
2045 | ||
2046 | /* blocking call to cancel all requests and join to a single (head) | |
2047 | * request */ | |
6d17d653 | 2048 | req = nfs_lock_and_join_requests(page); |
3e217045 WAA |
2049 | |
2050 | if (IS_ERR(req)) { | |
2051 | ret = PTR_ERR(req); | |
2052 | } else if (req) { | |
2053 | /* all requests from this page have been cancelled by | |
2054 | * nfs_lock_and_join_requests, so just remove the head | |
2055 | * request from the inode / page_private pointer and | |
2056 | * release it */ | |
2057 | nfs_inode_remove_request(req); | |
3e217045 | 2058 | nfs_unlock_and_release_request(req); |
1b3b4a1a | 2059 | } |
3e217045 | 2060 | |
1b3b4a1a TM |
2061 | return ret; |
2062 | } | |
2063 | ||
7f2f12d9 TM |
2064 | /* |
2065 | * Write back all requests on one page - we do this before reading it. | |
2066 | */ | |
c373fff7 | 2067 | int nfs_wb_page(struct inode *inode, struct page *page) |
1c75950b | 2068 | { |
29418aa4 | 2069 | loff_t range_start = page_file_offset(page); |
09cbfeaf | 2070 | loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); |
4d770ccf | 2071 | struct writeback_control wbc = { |
4d770ccf | 2072 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 2073 | .nr_to_write = 0, |
4d770ccf TM |
2074 | .range_start = range_start, |
2075 | .range_end = range_end, | |
2076 | }; | |
2077 | int ret; | |
1c75950b | 2078 | |
f4ce1299 TM |
2079 | trace_nfs_writeback_page_enter(inode); |
2080 | ||
0522f6ad | 2081 | for (;;) { |
ba8b06e6 | 2082 | wait_on_page_writeback(page); |
73e3302f | 2083 | if (clear_page_dirty_for_io(page)) { |
c373fff7 | 2084 | ret = nfs_writepage_locked(page, &wbc); |
73e3302f TM |
2085 | if (ret < 0) |
2086 | goto out_error; | |
0522f6ad | 2087 | continue; |
7f2f12d9 | 2088 | } |
f4ce1299 | 2089 | ret = 0; |
0522f6ad TM |
2090 | if (!PagePrivate(page)) |
2091 | break; | |
2092 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 2093 | if (ret < 0) |
73e3302f | 2094 | goto out_error; |
7f2f12d9 | 2095 | } |
73e3302f | 2096 | out_error: |
f4ce1299 | 2097 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 2098 | return ret; |
1c75950b TM |
2099 | } |
2100 | ||
074cc1de TM |
2101 | #ifdef CONFIG_MIGRATION |
2102 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 2103 | struct page *page, enum migrate_mode mode) |
074cc1de | 2104 | { |
2da95652 JL |
2105 | /* |
2106 | * If PagePrivate is set, then the page is currently associated with | |
2107 | * an in-progress read or write request. Don't try to migrate it. | |
2108 | * | |
2109 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
2110 | * that we can safely release the inode reference while holding | |
2111 | * the page lock. | |
2112 | */ | |
2113 | if (PagePrivate(page)) | |
2114 | return -EBUSY; | |
074cc1de | 2115 | |
8c209ce7 DH |
2116 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
2117 | return -EBUSY; | |
074cc1de | 2118 | |
a6bc32b8 | 2119 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
2120 | } |
2121 | #endif | |
2122 | ||
f7b422b1 | 2123 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
2124 | { |
2125 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 2126 | sizeof(struct nfs_pgio_header), |
1da177e4 | 2127 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 2128 | NULL); |
1da177e4 LT |
2129 | if (nfs_wdata_cachep == NULL) |
2130 | return -ENOMEM; | |
2131 | ||
93d2341c MD |
2132 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
2133 | nfs_wdata_cachep); | |
1da177e4 | 2134 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 2135 | goto out_destroy_write_cache; |
1da177e4 | 2136 | |
0b7c0153 FI |
2137 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
2138 | sizeof(struct nfs_commit_data), | |
2139 | 0, SLAB_HWCACHE_ALIGN, | |
2140 | NULL); | |
2141 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 2142 | goto out_destroy_write_mempool; |
0b7c0153 | 2143 | |
93d2341c | 2144 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 2145 | nfs_cdata_cachep); |
1da177e4 | 2146 | if (nfs_commit_mempool == NULL) |
3dd4765f | 2147 | goto out_destroy_commit_cache; |
1da177e4 | 2148 | |
89a09141 PZ |
2149 | /* |
2150 | * NFS congestion size, scale with available memory. | |
2151 | * | |
2152 | * 64MB: 8192k | |
2153 | * 128MB: 11585k | |
2154 | * 256MB: 16384k | |
2155 | * 512MB: 23170k | |
2156 | * 1GB: 32768k | |
2157 | * 2GB: 46340k | |
2158 | * 4GB: 65536k | |
2159 | * 8GB: 92681k | |
2160 | * 16GB: 131072k | |
2161 | * | |
2162 | * This allows larger machines to have larger/more transfers. | |
2163 | * Limit the default to 256M | |
2164 | */ | |
ca79b0c2 | 2165 | nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); |
89a09141 PZ |
2166 | if (nfs_congestion_kb > 256*1024) |
2167 | nfs_congestion_kb = 256*1024; | |
2168 | ||
1da177e4 | 2169 | return 0; |
3dd4765f JL |
2170 | |
2171 | out_destroy_commit_cache: | |
2172 | kmem_cache_destroy(nfs_cdata_cachep); | |
2173 | out_destroy_write_mempool: | |
2174 | mempool_destroy(nfs_wdata_mempool); | |
2175 | out_destroy_write_cache: | |
2176 | kmem_cache_destroy(nfs_wdata_cachep); | |
2177 | return -ENOMEM; | |
1da177e4 LT |
2178 | } |
2179 | ||
266bee88 | 2180 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
2181 | { |
2182 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 2183 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 2184 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 2185 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
2186 | } |
2187 | ||
4a0de55c AS |
2188 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
2189 | .rw_alloc_header = nfs_writehdr_alloc, | |
2190 | .rw_free_header = nfs_writehdr_free, | |
0eecb214 AS |
2191 | .rw_done = nfs_writeback_done, |
2192 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 2193 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 2194 | }; |