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> |
1da177e4 LT |
29 | |
30 | #include "delegation.h" | |
49a70f27 | 31 | #include "internal.h" |
91d5b470 | 32 | #include "iostat.h" |
def6ed7e | 33 | #include "nfs4_fs.h" |
074cc1de | 34 | #include "fscache.h" |
94ad1c80 | 35 | #include "pnfs.h" |
1da177e4 | 36 | |
f4ce1299 TM |
37 | #include "nfstrace.h" |
38 | ||
1da177e4 LT |
39 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
40 | ||
41 | #define MIN_POOL_WRITE (32) | |
42 | #define MIN_POOL_COMMIT (4) | |
43 | ||
919e3bd9 TM |
44 | struct nfs_io_completion { |
45 | void (*complete)(void *data); | |
46 | void *data; | |
47 | struct kref refcount; | |
48 | }; | |
49 | ||
1da177e4 LT |
50 | /* |
51 | * Local function declarations | |
52 | */ | |
f8512ad0 | 53 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 | 54 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 55 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 56 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
4a0de55c | 57 | static const struct nfs_rw_ops nfs_rw_write_ops; |
d4581383 | 58 | static void nfs_clear_request_commit(struct nfs_page *req); |
02d1426c WAA |
59 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
60 | struct inode *inode); | |
3a3908c8 TM |
61 | static struct nfs_page * |
62 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
63 | struct page *page); | |
1da177e4 | 64 | |
e18b890b | 65 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 66 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 67 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
68 | static mempool_t *nfs_commit_mempool; |
69 | ||
518662e0 | 70 | struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) |
1da177e4 | 71 | { |
518662e0 | 72 | struct nfs_commit_data *p; |
40859d7e | 73 | |
518662e0 N |
74 | if (never_fail) |
75 | p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); | |
76 | else { | |
77 | /* It is OK to do some reclaim, not no safe to wait | |
78 | * for anything to be returned to the pool. | |
79 | * mempool_alloc() cannot handle that particular combination, | |
80 | * so we need two separate attempts. | |
81 | */ | |
82 | p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); | |
83 | if (!p) | |
84 | p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | | |
85 | __GFP_NOWARN | __GFP_NORETRY); | |
86 | if (!p) | |
87 | return NULL; | |
1da177e4 | 88 | } |
518662e0 N |
89 | |
90 | memset(p, 0, sizeof(*p)); | |
91 | INIT_LIST_HEAD(&p->pages); | |
1da177e4 LT |
92 | return p; |
93 | } | |
e0c2b380 | 94 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 95 | |
0b7c0153 | 96 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
97 | { |
98 | mempool_free(p, nfs_commit_mempool); | |
99 | } | |
e0c2b380 | 100 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 101 | |
1e7f3a48 | 102 | static struct nfs_pgio_header *nfs_writehdr_alloc(void) |
3feb2d49 | 103 | { |
1e7f3a48 | 104 | struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
cd841605 | 105 | |
237f8306 N |
106 | memset(p, 0, sizeof(*p)); |
107 | p->rw_mode = FMODE_WRITE; | |
3feb2d49 TM |
108 | return p; |
109 | } | |
6c75dc0d | 110 | |
1e7f3a48 | 111 | static void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 112 | { |
1e7f3a48 | 113 | mempool_free(hdr, nfs_wdata_mempool); |
3feb2d49 | 114 | } |
1da177e4 | 115 | |
919e3bd9 TM |
116 | static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) |
117 | { | |
118 | return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); | |
119 | } | |
120 | ||
121 | static void nfs_io_completion_init(struct nfs_io_completion *ioc, | |
122 | void (*complete)(void *), void *data) | |
123 | { | |
124 | ioc->complete = complete; | |
125 | ioc->data = data; | |
126 | kref_init(&ioc->refcount); | |
127 | } | |
128 | ||
129 | static void nfs_io_completion_release(struct kref *kref) | |
130 | { | |
131 | struct nfs_io_completion *ioc = container_of(kref, | |
132 | struct nfs_io_completion, refcount); | |
133 | ioc->complete(ioc->data); | |
134 | kfree(ioc); | |
135 | } | |
136 | ||
137 | static void nfs_io_completion_get(struct nfs_io_completion *ioc) | |
138 | { | |
139 | if (ioc != NULL) | |
140 | kref_get(&ioc->refcount); | |
141 | } | |
142 | ||
143 | static void nfs_io_completion_put(struct nfs_io_completion *ioc) | |
144 | { | |
145 | if (ioc != NULL) | |
146 | kref_put(&ioc->refcount, nfs_io_completion_release); | |
147 | } | |
148 | ||
bd37d6fc TM |
149 | static struct nfs_page * |
150 | nfs_page_private_request(struct page *page) | |
151 | { | |
152 | if (!PagePrivate(page)) | |
153 | return NULL; | |
154 | return (struct nfs_page *)page_private(page); | |
155 | } | |
156 | ||
84d3a9a9 WAA |
157 | /* |
158 | * nfs_page_find_head_request_locked - find head request associated with @page | |
159 | * | |
160 | * must be called while holding the inode lock. | |
161 | * | |
162 | * returns matching head request with reference held, or NULL if not found. | |
163 | */ | |
29418aa4 | 164 | static struct nfs_page * |
b30d2f04 | 165 | nfs_page_find_private_request(struct page *page) |
277459d2 | 166 | { |
4b9bb25b | 167 | struct address_space *mapping = page_file_mapping(page); |
bd37d6fc | 168 | struct nfs_page *req; |
277459d2 | 169 | |
b30d2f04 TM |
170 | if (!PagePrivate(page)) |
171 | return NULL; | |
4b9bb25b | 172 | spin_lock(&mapping->private_lock); |
bd37d6fc | 173 | req = nfs_page_private_request(page); |
84d3a9a9 WAA |
174 | if (req) { |
175 | WARN_ON_ONCE(req->wb_head != req); | |
29418aa4 | 176 | kref_get(&req->wb_kref); |
84d3a9a9 | 177 | } |
4b9bb25b | 178 | spin_unlock(&mapping->private_lock); |
b30d2f04 TM |
179 | return req; |
180 | } | |
29418aa4 | 181 | |
b30d2f04 TM |
182 | static struct nfs_page * |
183 | nfs_page_find_swap_request(struct page *page) | |
184 | { | |
185 | struct inode *inode = page_file_mapping(page)->host; | |
186 | struct nfs_inode *nfsi = NFS_I(inode); | |
187 | struct nfs_page *req = NULL; | |
188 | if (!PageSwapCache(page)) | |
189 | return NULL; | |
e824f99a | 190 | mutex_lock(&nfsi->commit_mutex); |
b30d2f04 TM |
191 | if (PageSwapCache(page)) { |
192 | req = nfs_page_search_commits_for_head_request_locked(nfsi, | |
193 | page); | |
194 | if (req) { | |
195 | WARN_ON_ONCE(req->wb_head != req); | |
196 | kref_get(&req->wb_kref); | |
197 | } | |
198 | } | |
e824f99a | 199 | mutex_unlock(&nfsi->commit_mutex); |
277459d2 TM |
200 | return req; |
201 | } | |
202 | ||
84d3a9a9 WAA |
203 | /* |
204 | * nfs_page_find_head_request - find head request associated with @page | |
205 | * | |
206 | * returns matching head request with reference held, or NULL if not found. | |
207 | */ | |
208 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 209 | { |
b30d2f04 | 210 | struct nfs_page *req; |
277459d2 | 211 | |
b30d2f04 TM |
212 | req = nfs_page_find_private_request(page); |
213 | if (!req) | |
214 | req = nfs_page_find_swap_request(page); | |
277459d2 TM |
215 | return req; |
216 | } | |
217 | ||
1da177e4 LT |
218 | /* Adjust the file length if we're writing beyond the end */ |
219 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
220 | { | |
d56b4ddf | 221 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
222 | loff_t end, i_size; |
223 | pgoff_t end_index; | |
1da177e4 | 224 | |
a3d01454 TM |
225 | spin_lock(&inode->i_lock); |
226 | i_size = i_size_read(inode); | |
09cbfeaf | 227 | end_index = (i_size - 1) >> PAGE_SHIFT; |
8cd79788 | 228 | if (i_size > 0 && page_index(page) < end_index) |
a3d01454 | 229 | goto out; |
d56b4ddf | 230 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 231 | if (i_size >= end) |
a3d01454 | 232 | goto out; |
1da177e4 | 233 | i_size_write(inode, end); |
a3d01454 TM |
234 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
235 | out: | |
236 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
237 | } |
238 | ||
a301b777 TM |
239 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
240 | static void nfs_set_pageerror(struct page *page) | |
241 | { | |
d56b4ddf | 242 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
243 | } |
244 | ||
d72ddcba WAA |
245 | /* |
246 | * nfs_page_group_search_locked | |
247 | * @head - head request of page group | |
248 | * @page_offset - offset into page | |
249 | * | |
250 | * Search page group with head @head to find a request that contains the | |
251 | * page offset @page_offset. | |
252 | * | |
253 | * Returns a pointer to the first matching nfs request, or NULL if no | |
254 | * match is found. | |
255 | * | |
256 | * Must be called with the page group lock held | |
257 | */ | |
258 | static struct nfs_page * | |
259 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
260 | { | |
261 | struct nfs_page *req; | |
262 | ||
d72ddcba WAA |
263 | req = head; |
264 | do { | |
265 | if (page_offset >= req->wb_pgbase && | |
266 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
267 | return req; | |
268 | ||
269 | req = req->wb_this_page; | |
270 | } while (req != head); | |
271 | ||
272 | return NULL; | |
273 | } | |
274 | ||
275 | /* | |
276 | * nfs_page_group_covers_page | |
277 | * @head - head request of page group | |
278 | * | |
279 | * Return true if the page group with head @head covers the whole page, | |
280 | * returns false otherwise | |
281 | */ | |
282 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
283 | { | |
284 | struct nfs_page *tmp; | |
285 | unsigned int pos = 0; | |
286 | unsigned int len = nfs_page_length(req->wb_page); | |
287 | ||
1344b7ea | 288 | nfs_page_group_lock(req); |
d72ddcba | 289 | |
7e8a30f8 | 290 | for (;;) { |
d72ddcba | 291 | tmp = nfs_page_group_search_locked(req->wb_head, pos); |
7e8a30f8 TM |
292 | if (!tmp) |
293 | break; | |
294 | pos = tmp->wb_pgbase + tmp->wb_bytes; | |
295 | } | |
d72ddcba WAA |
296 | |
297 | nfs_page_group_unlock(req); | |
7e8a30f8 | 298 | return pos >= len; |
d72ddcba WAA |
299 | } |
300 | ||
1da177e4 LT |
301 | /* We can set the PG_uptodate flag if we see that a write request |
302 | * covers the full page. | |
303 | */ | |
d72ddcba | 304 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 305 | { |
d72ddcba | 306 | if (PageUptodate(req->wb_page)) |
1da177e4 | 307 | return; |
d72ddcba | 308 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 309 | return; |
d72ddcba | 310 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
311 | } |
312 | ||
1da177e4 LT |
313 | static int wb_priority(struct writeback_control *wbc) |
314 | { | |
e87b4c7a | 315 | int ret = 0; |
cca588d6 | 316 | |
e87b4c7a N |
317 | if (wbc->sync_mode == WB_SYNC_ALL) |
318 | ret = FLUSH_COND_STABLE; | |
e87b4c7a | 319 | return ret; |
1da177e4 LT |
320 | } |
321 | ||
89a09141 PZ |
322 | /* |
323 | * NFS congestion control | |
324 | */ | |
325 | ||
326 | int nfs_congestion_kb; | |
327 | ||
328 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
329 | #define NFS_CONGESTION_OFF_THRESH \ | |
330 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
331 | ||
deed85e7 | 332 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 333 | { |
0db10944 JK |
334 | struct inode *inode = page_file_mapping(page)->host; |
335 | struct nfs_server *nfss = NFS_SERVER(inode); | |
5a6d41b3 TM |
336 | int ret = test_set_page_writeback(page); |
337 | ||
deed85e7 | 338 | WARN_ON_ONCE(ret != 0); |
89a09141 | 339 | |
deed85e7 | 340 | if (atomic_long_inc_return(&nfss->writeback) > |
0db10944 JK |
341 | NFS_CONGESTION_ON_THRESH) |
342 | set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); | |
89a09141 PZ |
343 | } |
344 | ||
20633f04 | 345 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 346 | { |
20633f04 | 347 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 | 348 | struct nfs_server *nfss = NFS_SERVER(inode); |
31a01f09 | 349 | bool is_done; |
89a09141 | 350 | |
31a01f09 TM |
351 | is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); |
352 | nfs_unlock_request(req); | |
353 | if (!is_done) | |
20633f04 WAA |
354 | return; |
355 | ||
356 | end_page_writeback(req->wb_page); | |
c4dc4bee | 357 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
0db10944 | 358 | clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); |
89a09141 PZ |
359 | } |
360 | ||
d4581383 WAA |
361 | /* |
362 | * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req | |
363 | * | |
364 | * this is a helper function for nfs_lock_and_join_requests | |
365 | * | |
366 | * @inode - inode associated with request page group, must be holding inode lock | |
367 | * @head - head request of page group, must be holding head lock | |
368 | * @req - request that couldn't lock and needs to wait on the req bit lock | |
d4581383 | 369 | * |
b5bab9bf TM |
370 | * NOTE: this must be called holding page_group bit lock |
371 | * which will be released before returning. | |
d4581383 WAA |
372 | * |
373 | * returns 0 on success, < 0 on error. | |
374 | */ | |
74a6d4b5 TM |
375 | static void |
376 | nfs_unroll_locks(struct inode *inode, struct nfs_page *head, | |
6d17d653 | 377 | struct nfs_page *req) |
d4581383 WAA |
378 | { |
379 | struct nfs_page *tmp; | |
d4581383 WAA |
380 | |
381 | /* relinquish all the locks successfully grabbed this run */ | |
5b2b5187 TM |
382 | for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { |
383 | if (!kref_read(&tmp->wb_kref)) | |
384 | continue; | |
385 | nfs_unlock_and_release_request(tmp); | |
386 | } | |
d4581383 WAA |
387 | } |
388 | ||
389 | /* | |
390 | * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests | |
391 | * | |
392 | * @destroy_list - request list (using wb_this_page) terminated by @old_head | |
393 | * @old_head - the old head of the list | |
394 | * | |
395 | * All subrequests must be locked and removed from all lists, so at this point | |
396 | * they are only "active" in this function, and possibly in nfs_wait_on_request | |
397 | * with a reference held by some other context. | |
398 | */ | |
399 | static void | |
400 | nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, | |
b66aaa8d TM |
401 | struct nfs_page *old_head, |
402 | struct inode *inode) | |
d4581383 WAA |
403 | { |
404 | while (destroy_list) { | |
405 | struct nfs_page *subreq = destroy_list; | |
406 | ||
407 | destroy_list = (subreq->wb_this_page == old_head) ? | |
408 | NULL : subreq->wb_this_page; | |
409 | ||
410 | WARN_ON_ONCE(old_head != subreq->wb_head); | |
411 | ||
412 | /* make sure old group is not used */ | |
d4581383 WAA |
413 | subreq->wb_this_page = subreq; |
414 | ||
902a4c00 TM |
415 | clear_bit(PG_REMOVE, &subreq->wb_flags); |
416 | ||
5b2b5187 TM |
417 | /* Note: races with nfs_page_group_destroy() */ |
418 | if (!kref_read(&subreq->wb_kref)) { | |
5b2b5187 | 419 | /* Check if we raced with nfs_page_group_destroy() */ |
902a4c00 | 420 | if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) |
5b2b5187 TM |
421 | nfs_free_request(subreq); |
422 | continue; | |
423 | } | |
d4581383 | 424 | |
5b2b5187 TM |
425 | subreq->wb_head = subreq; |
426 | ||
427 | if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { | |
428 | nfs_release_request(subreq); | |
a6b6d5b8 | 429 | atomic_long_dec(&NFS_I(inode)->nrequests); |
d4581383 | 430 | } |
5b2b5187 | 431 | |
5b2b5187 TM |
432 | /* subreq is now totally disconnected from page group or any |
433 | * write / commit lists. last chance to wake any waiters */ | |
434 | nfs_unlock_and_release_request(subreq); | |
d4581383 WAA |
435 | } |
436 | } | |
437 | ||
438 | /* | |
439 | * nfs_lock_and_join_requests - join all subreqs to the head req and return | |
440 | * a locked reference, cancelling any pending | |
441 | * operations for this page. | |
442 | * | |
443 | * @page - the page used to lookup the "page group" of nfs_page structures | |
d4581383 WAA |
444 | * |
445 | * This function joins all sub requests to the head request by first | |
446 | * locking all requests in the group, cancelling any pending operations | |
447 | * and finally updating the head request to cover the whole range covered by | |
448 | * the (former) group. All subrequests are removed from any write or commit | |
449 | * lists, unlinked from the group and destroyed. | |
450 | * | |
451 | * Returns a locked, referenced pointer to the head request - which after | |
452 | * this call is guaranteed to be the only request associated with the page. | |
453 | * Returns NULL if no requests are found for @page, or a ERR_PTR if an | |
454 | * error was encountered. | |
455 | */ | |
456 | static struct nfs_page * | |
6d17d653 | 457 | nfs_lock_and_join_requests(struct page *page) |
e261f51f | 458 | { |
d56b4ddf | 459 | struct inode *inode = page_file_mapping(page)->host; |
d4581383 WAA |
460 | struct nfs_page *head, *subreq; |
461 | struct nfs_page *destroy_list = NULL; | |
462 | unsigned int total_bytes; | |
e261f51f TM |
463 | int ret; |
464 | ||
d4581383 | 465 | try_again: |
d4581383 WAA |
466 | /* |
467 | * A reference is taken only on the head request which acts as a | |
468 | * reference to the whole page group - the group will not be destroyed | |
469 | * until the head reference is released. | |
470 | */ | |
bd37d6fc TM |
471 | head = nfs_page_find_head_request(page); |
472 | if (!head) | |
d4581383 | 473 | return NULL; |
d4581383 | 474 | |
a0e265bc TM |
475 | /* lock the page head first in order to avoid an ABBA inefficiency */ |
476 | if (!nfs_lock_request(head)) { | |
a0e265bc TM |
477 | ret = nfs_wait_on_request(head); |
478 | nfs_release_request(head); | |
479 | if (ret < 0) | |
480 | return ERR_PTR(ret); | |
481 | goto try_again; | |
482 | } | |
bd37d6fc TM |
483 | |
484 | /* Ensure that nobody removed the request before we locked it */ | |
485 | if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { | |
486 | nfs_unlock_and_release_request(head); | |
487 | goto try_again; | |
488 | } | |
a0e265bc | 489 | |
1344b7ea | 490 | ret = nfs_page_group_lock(head); |
0671d8f1 ME |
491 | if (ret < 0) |
492 | goto release_request; | |
7c3af975 WAA |
493 | |
494 | /* lock each request in the page group */ | |
a0e265bc TM |
495 | total_bytes = head->wb_bytes; |
496 | for (subreq = head->wb_this_page; subreq != head; | |
497 | subreq = subreq->wb_this_page) { | |
74a6d4b5 | 498 | |
1bd5d6d0 TM |
499 | if (!kref_get_unless_zero(&subreq->wb_kref)) { |
500 | if (subreq->wb_offset == head->wb_offset + total_bytes) | |
501 | total_bytes += subreq->wb_bytes; | |
5b2b5187 | 502 | continue; |
1bd5d6d0 TM |
503 | } |
504 | ||
74a6d4b5 | 505 | while (!nfs_lock_request(subreq)) { |
b5bab9bf | 506 | /* |
74a6d4b5 TM |
507 | * Unlock page to allow nfs_page_group_sync_on_bit() |
508 | * to succeed | |
b5bab9bf | 509 | */ |
74a6d4b5 TM |
510 | nfs_page_group_unlock(head); |
511 | ret = nfs_wait_on_request(subreq); | |
512 | if (!ret) | |
1344b7ea | 513 | ret = nfs_page_group_lock(head); |
74a6d4b5 TM |
514 | if (ret < 0) { |
515 | nfs_unroll_locks(inode, head, subreq); | |
5b2b5187 | 516 | nfs_release_request(subreq); |
0671d8f1 | 517 | goto release_request; |
74a6d4b5 | 518 | } |
e14bebf6 | 519 | } |
d4581383 WAA |
520 | /* |
521 | * Subrequests are always contiguous, non overlapping | |
309a1d65 | 522 | * and in order - but may be repeated (mirrored writes). |
d4581383 | 523 | */ |
309a1d65 WAA |
524 | if (subreq->wb_offset == (head->wb_offset + total_bytes)) { |
525 | /* keep track of how many bytes this group covers */ | |
526 | total_bytes += subreq->wb_bytes; | |
527 | } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || | |
528 | ((subreq->wb_offset + subreq->wb_bytes) > | |
529 | (head->wb_offset + total_bytes)))) { | |
8b77484f | 530 | nfs_page_group_unlock(head); |
74a6d4b5 | 531 | nfs_unroll_locks(inode, head, subreq); |
5b2b5187 | 532 | nfs_unlock_and_release_request(subreq); |
0671d8f1 ME |
533 | ret = -EIO; |
534 | goto release_request; | |
309a1d65 | 535 | } |
a0e265bc | 536 | } |
d4581383 WAA |
537 | |
538 | /* Now that all requests are locked, make sure they aren't on any list. | |
539 | * Commit list removal accounting is done after locks are dropped */ | |
540 | subreq = head; | |
541 | do { | |
411a99ad | 542 | nfs_clear_request_commit(subreq); |
d4581383 WAA |
543 | subreq = subreq->wb_this_page; |
544 | } while (subreq != head); | |
545 | ||
546 | /* unlink subrequests from head, destroy them later */ | |
547 | if (head->wb_this_page != head) { | |
548 | /* destroy list will be terminated by head */ | |
549 | destroy_list = head->wb_this_page; | |
550 | head->wb_this_page = head; | |
551 | ||
552 | /* change head request to cover whole range that | |
553 | * the former page group covered */ | |
554 | head->wb_bytes = total_bytes; | |
e261f51f | 555 | } |
d4581383 | 556 | |
b66aaa8d TM |
557 | /* Postpone destruction of this request */ |
558 | if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { | |
559 | set_bit(PG_INODE_REF, &head->wb_flags); | |
560 | kref_get(&head->wb_kref); | |
a6b6d5b8 | 561 | atomic_long_inc(&NFS_I(inode)->nrequests); |
b66aaa8d TM |
562 | } |
563 | ||
d4581383 WAA |
564 | nfs_page_group_unlock(head); |
565 | ||
b66aaa8d | 566 | nfs_destroy_unlinked_subrequests(destroy_list, head, inode); |
d4581383 | 567 | |
b5bab9bf TM |
568 | /* Did we lose a race with nfs_inode_remove_request()? */ |
569 | if (!(PagePrivate(page) || PageSwapCache(page))) { | |
570 | nfs_unlock_and_release_request(head); | |
571 | return NULL; | |
572 | } | |
573 | ||
bd37d6fc | 574 | /* still holds ref on head from nfs_page_find_head_request |
d4581383 WAA |
575 | * and still has lock on head from lock loop */ |
576 | return head; | |
0671d8f1 ME |
577 | |
578 | release_request: | |
579 | nfs_unlock_and_release_request(head); | |
580 | return ERR_PTR(ret); | |
074cc1de TM |
581 | } |
582 | ||
0bcbf039 PT |
583 | static void nfs_write_error_remove_page(struct nfs_page *req) |
584 | { | |
0bcbf039 | 585 | nfs_end_page_writeback(req); |
0bcbf039 PT |
586 | generic_error_remove_page(page_file_mapping(req->wb_page), |
587 | req->wb_page); | |
1f84ccdf | 588 | nfs_release_request(req); |
0bcbf039 PT |
589 | } |
590 | ||
a6598813 TM |
591 | static bool |
592 | nfs_error_is_fatal_on_server(int err) | |
593 | { | |
594 | switch (err) { | |
595 | case 0: | |
596 | case -ERESTARTSYS: | |
597 | case -EINTR: | |
598 | return false; | |
599 | } | |
600 | return nfs_error_is_fatal(err); | |
0bcbf039 PT |
601 | } |
602 | ||
074cc1de TM |
603 | /* |
604 | * Find an associated nfs write request, and prepare to flush it out | |
605 | * May return an error if the user signalled nfs_wait_on_request(). | |
606 | */ | |
607 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
6d17d653 | 608 | struct page *page) |
074cc1de TM |
609 | { |
610 | struct nfs_page *req; | |
611 | int ret = 0; | |
612 | ||
6d17d653 | 613 | req = nfs_lock_and_join_requests(page); |
074cc1de TM |
614 | if (!req) |
615 | goto out; | |
616 | ret = PTR_ERR(req); | |
617 | if (IS_ERR(req)) | |
618 | goto out; | |
619 | ||
deed85e7 TM |
620 | nfs_set_page_writeback(page); |
621 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 622 | |
deed85e7 | 623 | ret = 0; |
a6598813 TM |
624 | /* If there is a fatal error that covers this write, just exit */ |
625 | if (nfs_error_is_fatal_on_server(req->wb_context->error)) | |
626 | goto out_launder; | |
627 | ||
f8512ad0 | 628 | if (!nfs_pageio_add_request(pgio, req)) { |
074cc1de | 629 | ret = pgio->pg_error; |
0bcbf039 | 630 | /* |
c373fff7 | 631 | * Remove the problematic req upon fatal errors on the server |
0bcbf039 PT |
632 | */ |
633 | if (nfs_error_is_fatal(ret)) { | |
634 | nfs_context_set_write_error(req->wb_context, ret); | |
c373fff7 | 635 | if (nfs_error_is_fatal_on_server(ret)) |
a6598813 | 636 | goto out_launder; |
0bcbf039 | 637 | } |
d6c843b9 PT |
638 | nfs_redirty_request(req); |
639 | ret = -EAGAIN; | |
40f90271 TM |
640 | } else |
641 | nfs_add_stats(page_file_mapping(page)->host, | |
642 | NFSIOS_WRITEPAGES, 1); | |
074cc1de TM |
643 | out: |
644 | return ret; | |
a6598813 TM |
645 | out_launder: |
646 | nfs_write_error_remove_page(req); | |
647 | return ret; | |
e261f51f TM |
648 | } |
649 | ||
d6c843b9 | 650 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, |
c373fff7 | 651 | struct nfs_pageio_descriptor *pgio) |
1da177e4 | 652 | { |
cfb506e1 | 653 | int ret; |
1da177e4 | 654 | |
8cd79788 | 655 | nfs_pageio_cond_complete(pgio, page_index(page)); |
6d17d653 | 656 | ret = nfs_page_async_flush(pgio, page); |
cfb506e1 TM |
657 | if (ret == -EAGAIN) { |
658 | redirty_page_for_writepage(wbc, page); | |
659 | ret = 0; | |
660 | } | |
661 | return ret; | |
f758c885 | 662 | } |
7fe7f848 | 663 | |
f758c885 TM |
664 | /* |
665 | * Write an mmapped page to the server. | |
666 | */ | |
d6c843b9 | 667 | static int nfs_writepage_locked(struct page *page, |
c373fff7 | 668 | struct writeback_control *wbc) |
f758c885 TM |
669 | { |
670 | struct nfs_pageio_descriptor pgio; | |
40f90271 | 671 | struct inode *inode = page_file_mapping(page)->host; |
f758c885 | 672 | int err; |
49a70f27 | 673 | |
40f90271 | 674 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
811ed92e | 675 | nfs_pageio_init_write(&pgio, inode, 0, |
a20c93e3 | 676 | false, &nfs_async_write_completion_ops); |
c373fff7 | 677 | err = nfs_do_writepage(page, wbc, &pgio); |
f758c885 TM |
678 | nfs_pageio_complete(&pgio); |
679 | if (err < 0) | |
680 | return err; | |
681 | if (pgio.pg_error < 0) | |
682 | return pgio.pg_error; | |
683 | return 0; | |
4d770ccf TM |
684 | } |
685 | ||
686 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
687 | { | |
f758c885 | 688 | int ret; |
4d770ccf | 689 | |
c373fff7 | 690 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 691 | unlock_page(page); |
f758c885 TM |
692 | return ret; |
693 | } | |
694 | ||
695 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
696 | { | |
697 | int ret; | |
698 | ||
c373fff7 | 699 | ret = nfs_do_writepage(page, wbc, data); |
f758c885 TM |
700 | unlock_page(page); |
701 | return ret; | |
1da177e4 LT |
702 | } |
703 | ||
919e3bd9 TM |
704 | static void nfs_io_completion_commit(void *inode) |
705 | { | |
706 | nfs_commit_inode(inode, 0); | |
707 | } | |
708 | ||
1da177e4 LT |
709 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
710 | { | |
1da177e4 | 711 | struct inode *inode = mapping->host; |
c63c7b05 | 712 | struct nfs_pageio_descriptor pgio; |
919e3bd9 | 713 | struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); |
1da177e4 LT |
714 | int err; |
715 | ||
91d5b470 CL |
716 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
717 | ||
919e3bd9 TM |
718 | if (ioc) |
719 | nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); | |
720 | ||
a20c93e3 CH |
721 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
722 | &nfs_async_write_completion_ops); | |
919e3bd9 | 723 | pgio.pg_io_completion = ioc; |
f758c885 | 724 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 725 | nfs_pageio_complete(&pgio); |
919e3bd9 | 726 | nfs_io_completion_put(ioc); |
72cb77f4 | 727 | |
f758c885 | 728 | if (err < 0) |
72cb77f4 TM |
729 | goto out_err; |
730 | err = pgio.pg_error; | |
731 | if (err < 0) | |
732 | goto out_err; | |
c63c7b05 | 733 | return 0; |
72cb77f4 TM |
734 | out_err: |
735 | return err; | |
1da177e4 LT |
736 | } |
737 | ||
738 | /* | |
739 | * Insert a write request into an inode | |
740 | */ | |
d6d6dc7c | 741 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 | 742 | { |
4b9bb25b | 743 | struct address_space *mapping = page_file_mapping(req->wb_page); |
1da177e4 | 744 | struct nfs_inode *nfsi = NFS_I(inode); |
e7d39069 | 745 | |
2bfc6e56 WAA |
746 | WARN_ON_ONCE(req->wb_this_page != req); |
747 | ||
e7d39069 | 748 | /* Lock the request! */ |
7ad84aa9 | 749 | nfs_lock_request(req); |
e7d39069 | 750 | |
29418aa4 MG |
751 | /* |
752 | * Swap-space should not get truncated. Hence no need to plug the race | |
753 | * with invalidate/truncate. | |
754 | */ | |
4b9bb25b TM |
755 | spin_lock(&mapping->private_lock); |
756 | if (!nfs_have_writebacks(inode) && | |
1eb5d98f JL |
757 | NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
758 | inode_inc_iversion_raw(inode); | |
29418aa4 MG |
759 | if (likely(!PageSwapCache(req->wb_page))) { |
760 | set_bit(PG_MAPPED, &req->wb_flags); | |
761 | SetPagePrivate(req->wb_page); | |
762 | set_page_private(req->wb_page, (unsigned long)req); | |
763 | } | |
4b9bb25b | 764 | spin_unlock(&mapping->private_lock); |
a6b6d5b8 | 765 | atomic_long_inc(&nfsi->nrequests); |
17089a29 | 766 | /* this a head request for a page group - mark it as having an |
cb1410c7 WAA |
767 | * extra reference so sub groups can follow suit. |
768 | * This flag also informs pgio layer when to bump nrequests when | |
769 | * adding subrequests. */ | |
17089a29 | 770 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); |
c03b4024 | 771 | kref_get(&req->wb_kref); |
1da177e4 LT |
772 | } |
773 | ||
774 | /* | |
89a09141 | 775 | * Remove a write request from an inode |
1da177e4 LT |
776 | */ |
777 | static void nfs_inode_remove_request(struct nfs_page *req) | |
778 | { | |
4b9bb25b TM |
779 | struct address_space *mapping = page_file_mapping(req->wb_page); |
780 | struct inode *inode = mapping->host; | |
1da177e4 | 781 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 782 | struct nfs_page *head; |
1da177e4 | 783 | |
a6b6d5b8 | 784 | atomic_long_dec(&nfsi->nrequests); |
20633f04 WAA |
785 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
786 | head = req->wb_head; | |
787 | ||
4b9bb25b | 788 | spin_lock(&mapping->private_lock); |
67911c8f | 789 | if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { |
20633f04 WAA |
790 | set_page_private(head->wb_page, 0); |
791 | ClearPagePrivate(head->wb_page); | |
792 | clear_bit(PG_MAPPED, &head->wb_flags); | |
793 | } | |
4b9bb25b | 794 | spin_unlock(&mapping->private_lock); |
29418aa4 | 795 | } |
17089a29 WAA |
796 | |
797 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) | |
798 | nfs_release_request(req); | |
1da177e4 LT |
799 | } |
800 | ||
61822ab5 | 801 | static void |
6d884e8f | 802 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 803 | { |
67911c8f AS |
804 | if (req->wb_page) |
805 | __set_page_dirty_nobuffers(req->wb_page); | |
61822ab5 TM |
806 | } |
807 | ||
3a3908c8 TM |
808 | /* |
809 | * nfs_page_search_commits_for_head_request_locked | |
810 | * | |
811 | * Search through commit lists on @inode for the head request for @page. | |
812 | * Must be called while holding the inode (which is cinfo) lock. | |
813 | * | |
814 | * Returns the head request if found, or NULL if not found. | |
815 | */ | |
816 | static struct nfs_page * | |
817 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
818 | struct page *page) | |
819 | { | |
820 | struct nfs_page *freq, *t; | |
821 | struct nfs_commit_info cinfo; | |
822 | struct inode *inode = &nfsi->vfs_inode; | |
823 | ||
824 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
825 | ||
826 | /* search through pnfs commit lists */ | |
827 | freq = pnfs_search_commit_reqs(inode, &cinfo, page); | |
828 | if (freq) | |
829 | return freq->wb_head; | |
830 | ||
831 | /* Linearly search the commit list for the correct request */ | |
832 | list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { | |
833 | if (freq->wb_page == page) | |
834 | return freq->wb_head; | |
835 | } | |
836 | ||
837 | return NULL; | |
838 | } | |
839 | ||
86d80f97 TM |
840 | /** |
841 | * nfs_request_add_commit_list_locked - add request to a commit list | |
842 | * @req: pointer to a struct nfs_page | |
843 | * @dst: commit list head | |
844 | * @cinfo: holds list lock and accounting info | |
845 | * | |
846 | * This sets the PG_CLEAN bit, updates the cinfo count of | |
847 | * number of outstanding requests requiring a commit as well as | |
848 | * the MM page stats. | |
849 | * | |
e824f99a TM |
850 | * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the |
851 | * nfs_page lock. | |
86d80f97 TM |
852 | */ |
853 | void | |
854 | nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, | |
855 | struct nfs_commit_info *cinfo) | |
856 | { | |
857 | set_bit(PG_CLEAN, &req->wb_flags); | |
858 | nfs_list_add_request(req, dst); | |
5cb953d4 | 859 | atomic_long_inc(&cinfo->mds->ncommit); |
86d80f97 TM |
860 | } |
861 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); | |
862 | ||
8dd37758 TM |
863 | /** |
864 | * nfs_request_add_commit_list - add request to a commit list | |
865 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
866 | * @dst: commit list head |
867 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 868 | * |
ea2cf228 | 869 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
870 | * number of outstanding requests requiring a commit as well as |
871 | * the MM page stats. | |
872 | * | |
ea2cf228 | 873 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 874 | * holding the nfs_page lock. |
1da177e4 | 875 | */ |
8dd37758 | 876 | void |
6272dcc6 | 877 | nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) |
1da177e4 | 878 | { |
e824f99a | 879 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
6272dcc6 | 880 | nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); |
e824f99a | 881 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
67911c8f AS |
882 | if (req->wb_page) |
883 | nfs_mark_page_unstable(req->wb_page, cinfo); | |
1da177e4 | 884 | } |
8dd37758 TM |
885 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
886 | ||
887 | /** | |
888 | * nfs_request_remove_commit_list - Remove request from a commit list | |
889 | * @req: pointer to a nfs_page | |
ea2cf228 | 890 | * @cinfo: holds list lock and accounting info |
8dd37758 | 891 | * |
ea2cf228 | 892 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
893 | * number of outstanding requests requiring a commit |
894 | * It does not update the MM page stats. | |
895 | * | |
ea2cf228 | 896 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
897 | */ |
898 | void | |
ea2cf228 FI |
899 | nfs_request_remove_commit_list(struct nfs_page *req, |
900 | struct nfs_commit_info *cinfo) | |
8dd37758 | 901 | { |
8dd37758 TM |
902 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
903 | return; | |
904 | nfs_list_remove_request(req); | |
5cb953d4 | 905 | atomic_long_dec(&cinfo->mds->ncommit); |
8dd37758 TM |
906 | } |
907 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
908 | ||
ea2cf228 FI |
909 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
910 | struct inode *inode) | |
911 | { | |
fe238e60 | 912 | cinfo->inode = inode; |
ea2cf228 FI |
913 | cinfo->mds = &NFS_I(inode)->commit_info; |
914 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 915 | cinfo->dreq = NULL; |
f453a54a | 916 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
917 | } |
918 | ||
919 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
920 | struct inode *inode, | |
921 | struct nfs_direct_req *dreq) | |
922 | { | |
1763da12 FI |
923 | if (dreq) |
924 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
925 | else | |
926 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
927 | } |
928 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
929 | |
930 | /* | |
931 | * Add a request to the inode's commit list. | |
932 | */ | |
1763da12 | 933 | void |
ea2cf228 | 934 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
b57ff130 | 935 | struct nfs_commit_info *cinfo, u32 ds_commit_idx) |
8dd37758 | 936 | { |
b57ff130 | 937 | if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) |
8dd37758 | 938 | return; |
6272dcc6 | 939 | nfs_request_add_commit_list(req, cinfo); |
8dd37758 | 940 | } |
8e821cad | 941 | |
d6d6dc7c FI |
942 | static void |
943 | nfs_clear_page_commit(struct page *page) | |
944 | { | |
11fb9989 | 945 | dec_node_page_state(page, NR_UNSTABLE_NFS); |
93f78d88 TH |
946 | dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, |
947 | WB_RECLAIMABLE); | |
d6d6dc7c FI |
948 | } |
949 | ||
b5bab9bf | 950 | /* Called holding the request lock on @req */ |
8dd37758 | 951 | static void |
e468bae9 TM |
952 | nfs_clear_request_commit(struct nfs_page *req) |
953 | { | |
8dd37758 | 954 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
2b0143b5 | 955 | struct inode *inode = d_inode(req->wb_context->dentry); |
ea2cf228 | 956 | struct nfs_commit_info cinfo; |
e468bae9 | 957 | |
ea2cf228 | 958 | nfs_init_cinfo_from_inode(&cinfo, inode); |
e824f99a | 959 | mutex_lock(&NFS_I(inode)->commit_mutex); |
ea2cf228 | 960 | if (!pnfs_clear_request_commit(req, &cinfo)) { |
ea2cf228 | 961 | nfs_request_remove_commit_list(req, &cinfo); |
8dd37758 | 962 | } |
e824f99a | 963 | mutex_unlock(&NFS_I(inode)->commit_mutex); |
d6d6dc7c | 964 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 965 | } |
e468bae9 TM |
966 | } |
967 | ||
d45f60c6 | 968 | int nfs_write_need_commit(struct nfs_pgio_header *hdr) |
8e821cad | 969 | { |
c65e6254 | 970 | if (hdr->verf.committed == NFS_DATA_SYNC) |
d45f60c6 | 971 | return hdr->lseg == NULL; |
c65e6254 | 972 | return hdr->verf.committed != NFS_FILE_SYNC; |
8e821cad TM |
973 | } |
974 | ||
919e3bd9 TM |
975 | static void nfs_async_write_init(struct nfs_pgio_header *hdr) |
976 | { | |
977 | nfs_io_completion_get(hdr->io_completion); | |
978 | } | |
979 | ||
061ae2ed | 980 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 981 | { |
ea2cf228 | 982 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
983 | unsigned long bytes = 0; |
984 | ||
985 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
986 | goto out; | |
ea2cf228 | 987 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
988 | while (!list_empty(&hdr->pages)) { |
989 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
990 | |
991 | bytes += req->wb_bytes; | |
992 | nfs_list_remove_request(req); | |
993 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
994 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 995 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
996 | nfs_context_set_write_error(req->wb_context, hdr->error); |
997 | goto remove_req; | |
998 | } | |
c65e6254 | 999 | if (nfs_write_need_commit(hdr)) { |
f79d06f5 | 1000 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
b57ff130 | 1001 | nfs_mark_request_commit(req, hdr->lseg, &cinfo, |
a7d42ddb | 1002 | hdr->pgio_mirror_idx); |
6c75dc0d FI |
1003 | goto next; |
1004 | } | |
1005 | remove_req: | |
1006 | nfs_inode_remove_request(req); | |
1007 | next: | |
20633f04 | 1008 | nfs_end_page_writeback(req); |
3aff4ebb | 1009 | nfs_release_request(req); |
6c75dc0d FI |
1010 | } |
1011 | out: | |
919e3bd9 | 1012 | nfs_io_completion_put(hdr->io_completion); |
6c75dc0d | 1013 | hdr->release(hdr); |
8e821cad | 1014 | } |
1da177e4 | 1015 | |
ce59515c | 1016 | unsigned long |
ea2cf228 | 1017 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 1018 | { |
5cb953d4 | 1019 | return atomic_long_read(&cinfo->mds->ncommit); |
d6d6dc7c FI |
1020 | } |
1021 | ||
e824f99a | 1022 | /* NFS_I(cinfo->inode)->commit_mutex held by caller */ |
1763da12 | 1023 | int |
ea2cf228 FI |
1024 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
1025 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c | 1026 | { |
137da553 | 1027 | struct nfs_page *req, *tmp; |
d6d6dc7c FI |
1028 | int ret = 0; |
1029 | ||
137da553 TM |
1030 | restart: |
1031 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
7ad84aa9 | 1032 | kref_get(&req->wb_kref); |
2ce209c4 TM |
1033 | if (!nfs_lock_request(req)) { |
1034 | int status; | |
137da553 TM |
1035 | |
1036 | /* Prevent deadlock with nfs_lock_and_join_requests */ | |
1037 | if (!list_empty(dst)) { | |
1038 | nfs_release_request(req); | |
1039 | continue; | |
1040 | } | |
1041 | /* Ensure we make progress to prevent livelock */ | |
2ce209c4 TM |
1042 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
1043 | status = nfs_wait_on_request(req); | |
1044 | nfs_release_request(req); | |
1045 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); | |
1046 | if (status < 0) | |
1047 | break; | |
137da553 | 1048 | goto restart; |
2ce209c4 | 1049 | } |
ea2cf228 | 1050 | nfs_request_remove_commit_list(req, cinfo); |
5d2a9d9d | 1051 | clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); |
8dd37758 TM |
1052 | nfs_list_add_request(req, dst); |
1053 | ret++; | |
1763da12 | 1054 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 1055 | break; |
e824f99a | 1056 | cond_resched(); |
d6d6dc7c FI |
1057 | } |
1058 | return ret; | |
fb8a1f11 | 1059 | } |
5d2a9d9d | 1060 | EXPORT_SYMBOL_GPL(nfs_scan_commit_list); |
fb8a1f11 | 1061 | |
1da177e4 LT |
1062 | /* |
1063 | * nfs_scan_commit - Scan an inode for commit requests | |
1064 | * @inode: NFS inode to scan | |
ea2cf228 FI |
1065 | * @dst: mds destination list |
1066 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
1067 | * |
1068 | * Moves requests from the inode's 'commit' request list. | |
1069 | * The requests are *not* checked to ensure that they form a contiguous set. | |
1070 | */ | |
1763da12 | 1071 | int |
ea2cf228 FI |
1072 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
1073 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1074 | { |
d6d6dc7c | 1075 | int ret = 0; |
fb8a1f11 | 1076 | |
5cb953d4 TM |
1077 | if (!atomic_long_read(&cinfo->mds->ncommit)) |
1078 | return 0; | |
e824f99a | 1079 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
5cb953d4 | 1080 | if (atomic_long_read(&cinfo->mds->ncommit) > 0) { |
8dd37758 | 1081 | const int max = INT_MAX; |
d6d6dc7c | 1082 | |
ea2cf228 FI |
1083 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
1084 | cinfo, max); | |
1085 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 1086 | } |
e824f99a | 1087 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
ff778d02 | 1088 | return ret; |
1da177e4 | 1089 | } |
d6d6dc7c | 1090 | |
1da177e4 | 1091 | /* |
e7d39069 TM |
1092 | * Search for an existing write request, and attempt to update |
1093 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 1094 | * |
e7d39069 TM |
1095 | * If the attempt fails, then the existing request is flushed out |
1096 | * to disk. | |
1da177e4 | 1097 | */ |
e7d39069 TM |
1098 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
1099 | struct page *page, | |
1100 | unsigned int offset, | |
1101 | unsigned int bytes) | |
1da177e4 | 1102 | { |
e7d39069 TM |
1103 | struct nfs_page *req; |
1104 | unsigned int rqend; | |
1105 | unsigned int end; | |
1106 | int error; | |
1107 | ||
1da177e4 LT |
1108 | end = offset + bytes; |
1109 | ||
f6032f21 TM |
1110 | req = nfs_lock_and_join_requests(page); |
1111 | if (IS_ERR_OR_NULL(req)) | |
1112 | return req; | |
1da177e4 | 1113 | |
f6032f21 TM |
1114 | rqend = req->wb_offset + req->wb_bytes; |
1115 | /* | |
1116 | * Tell the caller to flush out the request if | |
1117 | * the offsets are non-contiguous. | |
1118 | * Note: nfs_flush_incompatible() will already | |
1119 | * have flushed out requests having wrong owners. | |
1120 | */ | |
1121 | if (offset > rqend || end < req->wb_offset) | |
1122 | goto out_flushme; | |
1da177e4 LT |
1123 | |
1124 | /* Okay, the request matches. Update the region */ | |
1125 | if (offset < req->wb_offset) { | |
1126 | req->wb_offset = offset; | |
1127 | req->wb_pgbase = offset; | |
1da177e4 | 1128 | } |
1da177e4 LT |
1129 | if (end > rqend) |
1130 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1131 | else |
1132 | req->wb_bytes = rqend - req->wb_offset; | |
e7d39069 TM |
1133 | return req; |
1134 | out_flushme: | |
f6032f21 TM |
1135 | /* |
1136 | * Note: we mark the request dirty here because | |
1137 | * nfs_lock_and_join_requests() cannot preserve | |
1138 | * commit flags, so we have to replay the write. | |
1139 | */ | |
1140 | nfs_mark_request_dirty(req); | |
1141 | nfs_unlock_and_release_request(req); | |
e7d39069 | 1142 | error = nfs_wb_page(inode, page); |
f6032f21 | 1143 | return (error < 0) ? ERR_PTR(error) : NULL; |
e7d39069 TM |
1144 | } |
1145 | ||
1146 | /* | |
1147 | * Try to update an existing write request, or create one if there is none. | |
1148 | * | |
1149 | * Note: Should always be called with the Page Lock held to prevent races | |
1150 | * if we have to add a new request. Also assumes that the caller has | |
1151 | * already called nfs_flush_incompatible() if necessary. | |
1152 | */ | |
1153 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1154 | struct page *page, unsigned int offset, unsigned int bytes) | |
1155 | { | |
d56b4ddf | 1156 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1157 | struct nfs_page *req; |
1da177e4 | 1158 | |
e7d39069 TM |
1159 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1160 | if (req != NULL) | |
1161 | goto out; | |
2bfc6e56 | 1162 | req = nfs_create_request(ctx, page, NULL, offset, bytes); |
e7d39069 TM |
1163 | if (IS_ERR(req)) |
1164 | goto out; | |
d6d6dc7c | 1165 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1166 | out: |
61e930a9 | 1167 | return req; |
1da177e4 LT |
1168 | } |
1169 | ||
e7d39069 TM |
1170 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1171 | unsigned int offset, unsigned int count) | |
1172 | { | |
1173 | struct nfs_page *req; | |
1174 | ||
1175 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1176 | if (IS_ERR(req)) | |
1177 | return PTR_ERR(req); | |
1178 | /* Update file length */ | |
1179 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1180 | nfs_mark_uptodate(req); |
a6305ddb | 1181 | nfs_mark_request_dirty(req); |
1d1afcbc | 1182 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1183 | return 0; |
1184 | } | |
1185 | ||
1da177e4 LT |
1186 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1187 | { | |
cd3758e3 | 1188 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1189 | struct nfs_lock_context *l_ctx; |
bd61e0a9 | 1190 | struct file_lock_context *flctx = file_inode(file)->i_flctx; |
1da177e4 | 1191 | struct nfs_page *req; |
1a54533e | 1192 | int do_flush, status; |
1da177e4 LT |
1193 | /* |
1194 | * Look for a request corresponding to this page. If there | |
1195 | * is one, and it belongs to another file, we flush it out | |
1196 | * before we try to copy anything into the page. Do this | |
1197 | * due to the lack of an ACCESS-type call in NFSv2. | |
1198 | * Also do the same if we find a request from an existing | |
1199 | * dropped page. | |
1200 | */ | |
1a54533e | 1201 | do { |
84d3a9a9 | 1202 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1203 | if (req == NULL) |
1204 | return 0; | |
2a369153 | 1205 | l_ctx = req->wb_lock_context; |
138a2935 TM |
1206 | do_flush = req->wb_page != page || |
1207 | !nfs_match_open_context(req->wb_context, ctx); | |
bd61e0a9 JL |
1208 | if (l_ctx && flctx && |
1209 | !(list_empty_careful(&flctx->flc_posix) && | |
1210 | list_empty_careful(&flctx->flc_flock))) { | |
d51fdb87 | 1211 | do_flush |= l_ctx->lockowner != current->files; |
5263e31e | 1212 | } |
1da177e4 | 1213 | nfs_release_request(req); |
1a54533e TM |
1214 | if (!do_flush) |
1215 | return 0; | |
d56b4ddf | 1216 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1217 | } while (status == 0); |
1218 | return status; | |
1da177e4 LT |
1219 | } |
1220 | ||
dc24826b AA |
1221 | /* |
1222 | * Avoid buffered writes when a open context credential's key would | |
1223 | * expire soon. | |
1224 | * | |
1225 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1226 | * | |
1227 | * Return 0 and set a credential flag which triggers the inode to flush | |
1228 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1229 | * RPC_KEY_EXPIRE_TIMEO. | |
1230 | */ | |
1231 | int | |
1232 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1233 | { | |
1234 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
1235 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
1236 | ||
1237 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
1238 | } | |
1239 | ||
1240 | /* | |
1241 | * Test if the open context credential key is marked to expire soon. | |
1242 | */ | |
ce52914e | 1243 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) |
dc24826b | 1244 | { |
ce52914e SM |
1245 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; |
1246 | ||
1247 | return rpcauth_cred_key_to_expire(auth, ctx->cred); | |
dc24826b AA |
1248 | } |
1249 | ||
5d47a356 TM |
1250 | /* |
1251 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1252 | * the PageUptodate() flag. In this case, we will need to turn off | |
1253 | * write optimisations that depend on the page contents being correct. | |
1254 | */ | |
8d197a56 | 1255 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 1256 | { |
d529ef83 JL |
1257 | struct nfs_inode *nfsi = NFS_I(inode); |
1258 | ||
8d197a56 TM |
1259 | if (nfs_have_delegated_attributes(inode)) |
1260 | goto out; | |
18dd78c4 | 1261 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
d529ef83 | 1262 | return false; |
4db72b40 | 1263 | smp_rmb(); |
d529ef83 | 1264 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
1265 | return false; |
1266 | out: | |
18dd78c4 SM |
1267 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
1268 | return false; | |
8d197a56 | 1269 | return PageUptodate(page) != 0; |
5d47a356 TM |
1270 | } |
1271 | ||
5263e31e JL |
1272 | static bool |
1273 | is_whole_file_wrlock(struct file_lock *fl) | |
1274 | { | |
1275 | return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && | |
1276 | fl->fl_type == F_WRLCK; | |
1277 | } | |
1278 | ||
c7559663 SM |
1279 | /* If we know the page is up to date, and we're not using byte range locks (or |
1280 | * if we have the whole file locked for writing), it may be more efficient to | |
1281 | * extend the write to cover the entire page in order to avoid fragmentation | |
1282 | * inefficiencies. | |
1283 | * | |
263b4509 SM |
1284 | * If the file is opened for synchronous writes then we can just skip the rest |
1285 | * of the checks. | |
c7559663 SM |
1286 | */ |
1287 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
1288 | { | |
5263e31e JL |
1289 | int ret; |
1290 | struct file_lock_context *flctx = inode->i_flctx; | |
1291 | struct file_lock *fl; | |
1292 | ||
c7559663 SM |
1293 | if (file->f_flags & O_DSYNC) |
1294 | return 0; | |
263b4509 SM |
1295 | if (!nfs_write_pageuptodate(page, inode)) |
1296 | return 0; | |
c7559663 SM |
1297 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1298 | return 1; | |
bd61e0a9 JL |
1299 | if (!flctx || (list_empty_careful(&flctx->flc_flock) && |
1300 | list_empty_careful(&flctx->flc_posix))) | |
8fa4592a | 1301 | return 1; |
5263e31e JL |
1302 | |
1303 | /* Check to see if there are whole file write locks */ | |
5263e31e | 1304 | ret = 0; |
6109c850 | 1305 | spin_lock(&flctx->flc_lock); |
bd61e0a9 JL |
1306 | if (!list_empty(&flctx->flc_posix)) { |
1307 | fl = list_first_entry(&flctx->flc_posix, struct file_lock, | |
1308 | fl_list); | |
1309 | if (is_whole_file_wrlock(fl)) | |
1310 | ret = 1; | |
1311 | } else if (!list_empty(&flctx->flc_flock)) { | |
5263e31e JL |
1312 | fl = list_first_entry(&flctx->flc_flock, struct file_lock, |
1313 | fl_list); | |
1314 | if (fl->fl_type == F_WRLCK) | |
1315 | ret = 1; | |
1316 | } | |
6109c850 | 1317 | spin_unlock(&flctx->flc_lock); |
5263e31e | 1318 | return ret; |
c7559663 SM |
1319 | } |
1320 | ||
1da177e4 LT |
1321 | /* |
1322 | * Update and possibly write a cached page of an NFS file. | |
1323 | * | |
1324 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1325 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1326 | */ | |
1327 | int nfs_updatepage(struct file *file, struct page *page, | |
1328 | unsigned int offset, unsigned int count) | |
1329 | { | |
cd3758e3 | 1330 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 1331 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
1332 | int status = 0; |
1333 | ||
91d5b470 CL |
1334 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1335 | ||
6de1472f AV |
1336 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1337 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1338 | |
149a4fdd BC |
1339 | if (!count) |
1340 | goto out; | |
1341 | ||
c7559663 | 1342 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 1343 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 1344 | offset = 0; |
1da177e4 LT |
1345 | } |
1346 | ||
e21195a7 | 1347 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
1348 | if (status < 0) |
1349 | nfs_set_pageerror(page); | |
59b7c05f TM |
1350 | else |
1351 | __set_page_dirty_nobuffers(page); | |
149a4fdd | 1352 | out: |
48186c7d | 1353 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1354 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1355 | return status; |
1356 | } | |
1357 | ||
3ff7576d | 1358 | static int flush_task_priority(int how) |
1da177e4 LT |
1359 | { |
1360 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1361 | case FLUSH_HIGHPRI: | |
1362 | return RPC_PRIORITY_HIGH; | |
1363 | case FLUSH_LOWPRI: | |
1364 | return RPC_PRIORITY_LOW; | |
1365 | } | |
1366 | return RPC_PRIORITY_NORMAL; | |
1367 | } | |
1368 | ||
d45f60c6 WAA |
1369 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1370 | struct rpc_message *msg, | |
abde71f4 | 1371 | const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 | 1372 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1373 | { |
3ff7576d | 1374 | int priority = flush_task_priority(how); |
d138d5d1 | 1375 | |
1ed26f33 | 1376 | task_setup_data->priority = priority; |
abde71f4 | 1377 | rpc_ops->write_setup(hdr, msg); |
8224b273 CL |
1378 | trace_nfs_initiate_write(hdr->inode, hdr->io_start, hdr->good_bytes, |
1379 | hdr->args.stable); | |
d138d5d1 | 1380 | |
abde71f4 | 1381 | nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, |
d45f60c6 | 1382 | &task_setup_data->rpc_client, msg, hdr); |
275acaaf TM |
1383 | } |
1384 | ||
6d884e8f F |
1385 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1386 | * call this on each, which will prepare them to be retried on next | |
1387 | * writeback using standard nfs. | |
1388 | */ | |
1389 | static void nfs_redirty_request(struct nfs_page *req) | |
1390 | { | |
1391 | nfs_mark_request_dirty(req); | |
c7070113 | 1392 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
20633f04 | 1393 | nfs_end_page_writeback(req); |
3aff4ebb | 1394 | nfs_release_request(req); |
6d884e8f F |
1395 | } |
1396 | ||
061ae2ed | 1397 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1398 | { |
1399 | struct nfs_page *req; | |
1400 | ||
1401 | while (!list_empty(head)) { | |
1402 | req = nfs_list_entry(head->next); | |
1403 | nfs_list_remove_request(req); | |
1404 | nfs_redirty_request(req); | |
1405 | } | |
1406 | } | |
1407 | ||
dc602dd7 TM |
1408 | static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) |
1409 | { | |
1410 | nfs_async_write_error(&hdr->pages); | |
1411 | } | |
1412 | ||
061ae2ed | 1413 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
919e3bd9 | 1414 | .init_hdr = nfs_async_write_init, |
061ae2ed FI |
1415 | .error_cleanup = nfs_async_write_error, |
1416 | .completion = nfs_write_completion, | |
dc602dd7 | 1417 | .reschedule_io = nfs_async_write_reschedule_io, |
061ae2ed FI |
1418 | }; |
1419 | ||
57208fa7 | 1420 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1421 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1422 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1423 | { |
a20c93e3 | 1424 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1425 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1426 | |
1427 | #ifdef CONFIG_NFS_V4_1 | |
1428 | if (server->pnfs_curr_ld && !force_mds) | |
1429 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1430 | #endif | |
4a0de55c | 1431 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
3bde7afd | 1432 | server->wsize, ioflags); |
1751c363 | 1433 | } |
ddda8e0a | 1434 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1435 | |
dce81290 TM |
1436 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1437 | { | |
a7d42ddb WAA |
1438 | struct nfs_pgio_mirror *mirror; |
1439 | ||
6f29b9bb KM |
1440 | if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) |
1441 | pgio->pg_ops->pg_cleanup(pgio); | |
1442 | ||
41d8d5b7 | 1443 | pgio->pg_ops = &nfs_pgio_rw_ops; |
a7d42ddb WAA |
1444 | |
1445 | nfs_pageio_stop_mirroring(pgio); | |
1446 | ||
1447 | mirror = &pgio->pg_mirrors[0]; | |
1448 | mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
dce81290 | 1449 | } |
1f945357 | 1450 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1451 | |
1da177e4 | 1452 | |
0b7c0153 FI |
1453 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1454 | { | |
1455 | struct nfs_commit_data *data = calldata; | |
1456 | ||
1457 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1458 | } | |
1459 | ||
1f2edbe3 TM |
1460 | /* |
1461 | * Special version of should_remove_suid() that ignores capabilities. | |
1462 | */ | |
1463 | static int nfs_should_remove_suid(const struct inode *inode) | |
1464 | { | |
1465 | umode_t mode = inode->i_mode; | |
1466 | int kill = 0; | |
1467 | ||
1468 | /* suid always must be killed */ | |
1469 | if (unlikely(mode & S_ISUID)) | |
1470 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1471 | |
1f2edbe3 TM |
1472 | /* |
1473 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1474 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1475 | */ | |
1476 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1477 | kill |= ATTR_KILL_SGID; | |
1478 | ||
1479 | if (unlikely(kill && S_ISREG(mode))) | |
1480 | return kill; | |
1481 | ||
1482 | return 0; | |
1483 | } | |
788e7a89 | 1484 | |
a08a8cd3 TM |
1485 | static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, |
1486 | struct nfs_fattr *fattr) | |
1487 | { | |
1488 | struct nfs_pgio_args *argp = &hdr->args; | |
1489 | struct nfs_pgio_res *resp = &hdr->res; | |
2b83d3de | 1490 | u64 size = argp->offset + resp->count; |
a08a8cd3 TM |
1491 | |
1492 | if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) | |
2b83d3de TM |
1493 | fattr->size = size; |
1494 | if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { | |
1495 | fattr->valid &= ~NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 | 1496 | return; |
2b83d3de TM |
1497 | } |
1498 | if (size != fattr->size) | |
a08a8cd3 TM |
1499 | return; |
1500 | /* Set attribute barrier */ | |
1501 | nfs_fattr_set_barrier(fattr); | |
2b83d3de TM |
1502 | /* ...and update size */ |
1503 | fattr->valid |= NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 TM |
1504 | } |
1505 | ||
1506 | void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) | |
1507 | { | |
2b83d3de | 1508 | struct nfs_fattr *fattr = &hdr->fattr; |
a08a8cd3 TM |
1509 | struct inode *inode = hdr->inode; |
1510 | ||
a08a8cd3 TM |
1511 | spin_lock(&inode->i_lock); |
1512 | nfs_writeback_check_extend(hdr, fattr); | |
1513 | nfs_post_op_update_inode_force_wcc_locked(inode, fattr); | |
1514 | spin_unlock(&inode->i_lock); | |
1515 | } | |
1516 | EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); | |
1517 | ||
1da177e4 LT |
1518 | /* |
1519 | * This function is called when the WRITE call is complete. | |
1520 | */ | |
d45f60c6 WAA |
1521 | static int nfs_writeback_done(struct rpc_task *task, |
1522 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1523 | struct inode *inode) |
1da177e4 | 1524 | { |
788e7a89 | 1525 | int status; |
1da177e4 | 1526 | |
f551e44f CL |
1527 | /* |
1528 | * ->write_done will attempt to use post-op attributes to detect | |
1529 | * conflicting writes by other clients. A strict interpretation | |
1530 | * of close-to-open would allow us to continue caching even if | |
1531 | * another writer had changed the file, but some applications | |
1532 | * depend on tighter cache coherency when writing. | |
1533 | */ | |
d45f60c6 | 1534 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1535 | if (status != 0) |
0eecb214 | 1536 | return status; |
8224b273 | 1537 | |
d45f60c6 | 1538 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
8224b273 CL |
1539 | trace_nfs_writeback_done(inode, task->tk_status, |
1540 | hdr->args.offset, hdr->res.verf); | |
91d5b470 | 1541 | |
d45f60c6 WAA |
1542 | if (hdr->res.verf->committed < hdr->args.stable && |
1543 | task->tk_status >= 0) { | |
1da177e4 LT |
1544 | /* We tried a write call, but the server did not |
1545 | * commit data to stable storage even though we | |
1546 | * requested it. | |
1547 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1548 | * the server reports NFS_DATA_SYNC, but performs | |
1549 | * NFS_FILE_SYNC. We therefore implement this checking | |
1550 | * as a dprintk() in order to avoid filling syslog. | |
1551 | */ | |
1552 | static unsigned long complain; | |
1553 | ||
a69aef14 | 1554 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1555 | if (time_before(complain, jiffies)) { |
48186c7d | 1556 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1557 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1558 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
d45f60c6 | 1559 | hdr->res.verf->committed, hdr->args.stable); |
1da177e4 LT |
1560 | complain = jiffies + 300 * HZ; |
1561 | } | |
1562 | } | |
1f2edbe3 TM |
1563 | |
1564 | /* Deal with the suid/sgid bit corner case */ | |
16e14375 TM |
1565 | if (nfs_should_remove_suid(inode)) { |
1566 | spin_lock(&inode->i_lock); | |
1567 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_OTHER; | |
1568 | spin_unlock(&inode->i_lock); | |
1569 | } | |
0eecb214 AS |
1570 | return 0; |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * This function is called when the WRITE call is complete. | |
1575 | */ | |
d45f60c6 WAA |
1576 | static void nfs_writeback_result(struct rpc_task *task, |
1577 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1578 | { |
d45f60c6 WAA |
1579 | struct nfs_pgio_args *argp = &hdr->args; |
1580 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1581 | |
1582 | if (resp->count < argp->count) { | |
1da177e4 LT |
1583 | static unsigned long complain; |
1584 | ||
6c75dc0d | 1585 | /* This a short write! */ |
d45f60c6 | 1586 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1587 | |
1da177e4 | 1588 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1589 | if (resp->count == 0) { |
1590 | if (time_before(complain, jiffies)) { | |
1591 | printk(KERN_WARNING | |
1592 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1593 | argp->count); | |
1594 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1595 | } |
d45f60c6 | 1596 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1597 | task->tk_status = -EIO; |
13602896 | 1598 | return; |
1da177e4 | 1599 | } |
f8417b48 KM |
1600 | |
1601 | /* For non rpc-based layout drivers, retry-through-MDS */ | |
1602 | if (!task->tk_ops) { | |
1603 | hdr->pnfs_error = -EAGAIN; | |
1604 | return; | |
1605 | } | |
1606 | ||
6c75dc0d FI |
1607 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1608 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1609 | /* Resend from where the server left off */ | |
d45f60c6 | 1610 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1611 | argp->offset += resp->count; |
1612 | argp->pgbase += resp->count; | |
1613 | argp->count -= resp->count; | |
1614 | } else { | |
1615 | /* Resend as a stable write in order to avoid | |
1616 | * headaches in the case of a server crash. | |
1617 | */ | |
1618 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1619 | } |
6c75dc0d | 1620 | rpc_restart_call_prepare(task); |
1da177e4 | 1621 | } |
1da177e4 LT |
1622 | } |
1623 | ||
af7cf057 | 1624 | static int wait_on_commit(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1625 | { |
af7cf057 TM |
1626 | return wait_on_atomic_t(&cinfo->rpcs_out, |
1627 | nfs_wait_atomic_killable, TASK_KILLABLE); | |
1628 | } | |
b8413f98 | 1629 | |
af7cf057 TM |
1630 | static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) |
1631 | { | |
1632 | atomic_inc(&cinfo->rpcs_out); | |
71d0a611 TM |
1633 | } |
1634 | ||
af7cf057 | 1635 | static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1636 | { |
af7cf057 TM |
1637 | if (atomic_dec_and_test(&cinfo->rpcs_out)) |
1638 | wake_up_atomic_t(&cinfo->rpcs_out); | |
71d0a611 TM |
1639 | } |
1640 | ||
0b7c0153 | 1641 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1642 | { |
0b7c0153 FI |
1643 | put_nfs_open_context(data->context); |
1644 | nfs_commit_free(data); | |
1da177e4 | 1645 | } |
e0c2b380 | 1646 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1647 | |
0b7c0153 | 1648 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
c36aae9a | 1649 | const struct nfs_rpc_ops *nfs_ops, |
9ace33cd | 1650 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1651 | int how, int flags) |
1da177e4 | 1652 | { |
07737691 | 1653 | struct rpc_task *task; |
9ace33cd | 1654 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1655 | struct rpc_message msg = { |
1656 | .rpc_argp = &data->args, | |
1657 | .rpc_resp = &data->res, | |
9ace33cd | 1658 | .rpc_cred = data->cred, |
bdc7f021 | 1659 | }; |
84115e1c | 1660 | struct rpc_task_setup task_setup_data = { |
07737691 | 1661 | .task = &data->task, |
9ace33cd | 1662 | .rpc_client = clnt, |
bdc7f021 | 1663 | .rpc_message = &msg, |
9ace33cd | 1664 | .callback_ops = call_ops, |
84115e1c | 1665 | .callback_data = data, |
101070ca | 1666 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1667 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1668 | .priority = priority, |
84115e1c | 1669 | }; |
9ace33cd | 1670 | /* Set up the initial task struct. */ |
c36aae9a | 1671 | nfs_ops->commit_setup(data, &msg); |
8224b273 | 1672 | trace_nfs_initiate_commit(data); |
9ace33cd | 1673 | |
b4839ebe | 1674 | dprintk("NFS: initiated commit call\n"); |
9ace33cd | 1675 | |
8c21c62c WAA |
1676 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1677 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1678 | ||
9ace33cd FI |
1679 | task = rpc_run_task(&task_setup_data); |
1680 | if (IS_ERR(task)) | |
1681 | return PTR_ERR(task); | |
1682 | if (how & FLUSH_SYNC) | |
1683 | rpc_wait_for_completion_task(task); | |
1684 | rpc_put_task(task); | |
1685 | return 0; | |
1686 | } | |
e0c2b380 | 1687 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1688 | |
378520b8 PT |
1689 | static loff_t nfs_get_lwb(struct list_head *head) |
1690 | { | |
1691 | loff_t lwb = 0; | |
1692 | struct nfs_page *req; | |
1693 | ||
1694 | list_for_each_entry(req, head, wb_list) | |
1695 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1696 | lwb = req_offset(req) + req->wb_bytes; | |
1697 | ||
1698 | return lwb; | |
1699 | } | |
1700 | ||
9ace33cd FI |
1701 | /* |
1702 | * Set up the argument/result storage required for the RPC call. | |
1703 | */ | |
0b7c0153 | 1704 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1705 | struct list_head *head, |
1706 | struct pnfs_layout_segment *lseg, | |
1707 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1708 | { |
1709 | struct nfs_page *first = nfs_list_entry(head->next); | |
2b0143b5 | 1710 | struct inode *inode = d_inode(first->wb_context->dentry); |
1da177e4 LT |
1711 | |
1712 | /* Set up the RPC argument and reply structs | |
1713 | * NB: take care not to mess about with data->commit et al. */ | |
1714 | ||
1715 | list_splice_init(head, &data->pages); | |
1da177e4 | 1716 | |
1da177e4 | 1717 | data->inode = inode; |
9ace33cd | 1718 | data->cred = first->wb_context->cred; |
988b6dce | 1719 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1720 | /* only set lwb for pnfs commit */ |
1721 | if (lseg) | |
1722 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1723 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1724 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1725 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1726 | |
1727 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1728 | /* Note: we always request a commit of the entire inode */ |
1729 | data->args.offset = 0; | |
1730 | data->args.count = 0; | |
0b7c0153 | 1731 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1732 | data->res.fattr = &data->fattr; |
1733 | data->res.verf = &data->verf; | |
0e574af1 | 1734 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1735 | } |
e0c2b380 | 1736 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1737 | |
e0c2b380 | 1738 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 | 1739 | struct pnfs_layout_segment *lseg, |
b57ff130 WAA |
1740 | struct nfs_commit_info *cinfo, |
1741 | u32 ds_commit_idx) | |
64bfeb49 FI |
1742 | { |
1743 | struct nfs_page *req; | |
1744 | ||
1745 | while (!list_empty(page_list)) { | |
1746 | req = nfs_list_entry(page_list->next); | |
1747 | nfs_list_remove_request(req); | |
b57ff130 | 1748 | nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); |
487b9b8a TH |
1749 | if (!cinfo->dreq) |
1750 | nfs_clear_page_commit(req->wb_page); | |
1d1afcbc | 1751 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1752 | } |
1753 | } | |
e0c2b380 | 1754 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1755 | |
b20135d0 TM |
1756 | static void |
1757 | nfs_commit_resched_write(struct nfs_commit_info *cinfo, | |
1758 | struct nfs_page *req) | |
1759 | { | |
1760 | __set_page_dirty_nobuffers(req->wb_page); | |
1761 | } | |
1762 | ||
1da177e4 LT |
1763 | /* |
1764 | * Commit dirty pages | |
1765 | */ | |
1766 | static int | |
ea2cf228 FI |
1767 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1768 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1769 | { |
0b7c0153 | 1770 | struct nfs_commit_data *data; |
1da177e4 | 1771 | |
ade8febd WAA |
1772 | /* another commit raced with us */ |
1773 | if (list_empty(head)) | |
1774 | return 0; | |
1775 | ||
518662e0 | 1776 | data = nfs_commitdata_alloc(true); |
1da177e4 LT |
1777 | |
1778 | /* Set up the argument struct */ | |
f453a54a FI |
1779 | nfs_init_commit(data, head, NULL, cinfo); |
1780 | atomic_inc(&cinfo->mds->rpcs_out); | |
c36aae9a PT |
1781 | return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), |
1782 | data->mds_ops, how, 0); | |
67911c8f | 1783 | } |
67911c8f | 1784 | |
1da177e4 LT |
1785 | /* |
1786 | * COMMIT call returned | |
1787 | */ | |
788e7a89 | 1788 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1789 | { |
0b7c0153 | 1790 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1791 | |
a3f565b1 | 1792 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1793 | task->tk_pid, task->tk_status); |
1794 | ||
788e7a89 | 1795 | /* Call the NFS version-specific code */ |
c0d0e96b | 1796 | NFS_PROTO(data->inode)->commit_done(task, data); |
8224b273 | 1797 | trace_nfs_commit_done(data); |
c9d8f89d TM |
1798 | } |
1799 | ||
f453a54a | 1800 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1801 | { |
5917ce84 | 1802 | struct nfs_page *req; |
c9d8f89d | 1803 | int status = data->task.tk_status; |
f453a54a | 1804 | struct nfs_commit_info cinfo; |
353db796 | 1805 | struct nfs_server *nfss; |
788e7a89 | 1806 | |
1da177e4 LT |
1807 | while (!list_empty(&data->pages)) { |
1808 | req = nfs_list_entry(data->pages.next); | |
1809 | nfs_list_remove_request(req); | |
67911c8f AS |
1810 | if (req->wb_page) |
1811 | nfs_clear_page_commit(req->wb_page); | |
1da177e4 | 1812 | |
1e8968c5 | 1813 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1814 | req->wb_context->dentry->d_sb->s_id, |
2b0143b5 | 1815 | (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), |
1da177e4 LT |
1816 | req->wb_bytes, |
1817 | (long long)req_offset(req)); | |
c9d8f89d TM |
1818 | if (status < 0) { |
1819 | nfs_context_set_write_error(req->wb_context, status); | |
38a33101 KM |
1820 | if (req->wb_page) |
1821 | nfs_inode_remove_request(req); | |
ddeaa637 | 1822 | dprintk_cont(", error = %d\n", status); |
1da177e4 LT |
1823 | goto next; |
1824 | } | |
1825 | ||
1826 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1827 | * returned by the server against all stored verfs. */ | |
8fc3c386 | 1828 | if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { |
1da177e4 | 1829 | /* We have a match */ |
38a33101 KM |
1830 | if (req->wb_page) |
1831 | nfs_inode_remove_request(req); | |
ddeaa637 | 1832 | dprintk_cont(" OK\n"); |
1da177e4 LT |
1833 | goto next; |
1834 | } | |
1835 | /* We have a mismatch. Write the page again */ | |
ddeaa637 | 1836 | dprintk_cont(" mismatch\n"); |
6d884e8f | 1837 | nfs_mark_request_dirty(req); |
05990d1b | 1838 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1839 | next: |
1d1afcbc | 1840 | nfs_unlock_and_release_request(req); |
7f1bda44 TM |
1841 | /* Latency breaker */ |
1842 | cond_resched(); | |
1da177e4 | 1843 | } |
353db796 N |
1844 | nfss = NFS_SERVER(data->inode); |
1845 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
0db10944 | 1846 | clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); |
353db796 | 1847 | |
f453a54a | 1848 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
af7cf057 | 1849 | nfs_commit_end(cinfo.mds); |
5917ce84 FI |
1850 | } |
1851 | ||
1852 | static void nfs_commit_release(void *calldata) | |
1853 | { | |
0b7c0153 | 1854 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1855 | |
f453a54a | 1856 | data->completion_ops->completion(data); |
c9d8f89d | 1857 | nfs_commitdata_release(calldata); |
1da177e4 | 1858 | } |
788e7a89 TM |
1859 | |
1860 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1861 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1862 | .rpc_call_done = nfs_commit_done, |
1863 | .rpc_release = nfs_commit_release, | |
1864 | }; | |
1da177e4 | 1865 | |
f453a54a FI |
1866 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1867 | .completion = nfs_commit_release_pages, | |
b20135d0 | 1868 | .resched_write = nfs_commit_resched_write, |
f453a54a FI |
1869 | }; |
1870 | ||
1763da12 FI |
1871 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1872 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1873 | { |
1874 | int status; | |
1875 | ||
ea2cf228 | 1876 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1877 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1878 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1879 | return status; |
1880 | } | |
1881 | ||
b608b283 | 1882 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1883 | { |
1da177e4 | 1884 | LIST_HEAD(head); |
ea2cf228 | 1885 | struct nfs_commit_info cinfo; |
71d0a611 | 1886 | int may_wait = how & FLUSH_SYNC; |
af7cf057 | 1887 | int error = 0; |
b8413f98 | 1888 | int res; |
1da177e4 | 1889 | |
ea2cf228 | 1890 | nfs_init_cinfo_from_inode(&cinfo, inode); |
af7cf057 | 1891 | nfs_commit_begin(cinfo.mds); |
ea2cf228 | 1892 | res = nfs_scan_commit(inode, &head, &cinfo); |
af7cf057 | 1893 | if (res) |
ea2cf228 | 1894 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
af7cf057 | 1895 | nfs_commit_end(cinfo.mds); |
dc4fd9ab SM |
1896 | if (res == 0) |
1897 | return res; | |
af7cf057 TM |
1898 | if (error < 0) |
1899 | goto out_error; | |
1900 | if (!may_wait) | |
1901 | goto out_mark_dirty; | |
1902 | error = wait_on_commit(cinfo.mds); | |
1903 | if (error < 0) | |
1904 | return error; | |
c5efa5fc | 1905 | return res; |
af7cf057 TM |
1906 | out_error: |
1907 | res = error; | |
c5efa5fc TM |
1908 | /* Note: If we exit without ensuring that the commit is complete, |
1909 | * we must mark the inode as dirty. Otherwise, future calls to | |
1910 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1911 | * that the data is on the disk. | |
1912 | */ | |
1913 | out_mark_dirty: | |
1914 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1915 | return res; |
1916 | } | |
b20135d0 | 1917 | EXPORT_SYMBOL_GPL(nfs_commit_inode); |
8fc795f7 | 1918 | |
ae09c31f | 1919 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
8fc795f7 | 1920 | { |
420e3646 TM |
1921 | struct nfs_inode *nfsi = NFS_I(inode); |
1922 | int flags = FLUSH_SYNC; | |
1923 | int ret = 0; | |
8fc795f7 | 1924 | |
3236c3e1 | 1925 | /* no commits means nothing needs to be done */ |
5cb953d4 | 1926 | if (!atomic_long_read(&nfsi->commit_info.ncommit)) |
3236c3e1 JL |
1927 | return ret; |
1928 | ||
a00dd6c0 JL |
1929 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1930 | /* Don't commit yet if this is a non-blocking flush and there | |
1931 | * are a lot of outstanding writes for this mapping. | |
1932 | */ | |
1a4edf0f | 1933 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) |
a00dd6c0 | 1934 | goto out_mark_dirty; |
420e3646 | 1935 | |
a00dd6c0 | 1936 | /* don't wait for the COMMIT response */ |
420e3646 | 1937 | flags = 0; |
a00dd6c0 JL |
1938 | } |
1939 | ||
420e3646 TM |
1940 | ret = nfs_commit_inode(inode, flags); |
1941 | if (ret >= 0) { | |
1942 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1943 | if (ret < wbc->nr_to_write) | |
1944 | wbc->nr_to_write -= ret; | |
1945 | else | |
1946 | wbc->nr_to_write = 0; | |
1947 | } | |
8fc795f7 | 1948 | return 0; |
420e3646 TM |
1949 | } |
1950 | out_mark_dirty: | |
8fc795f7 TM |
1951 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1952 | return ret; | |
1953 | } | |
89d77c8f | 1954 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1955 | |
837bb1d7 TM |
1956 | /* |
1957 | * Wrapper for filemap_write_and_wait_range() | |
1958 | * | |
1959 | * Needed for pNFS in order to ensure data becomes visible to the | |
1960 | * client. | |
1961 | */ | |
1962 | int nfs_filemap_write_and_wait_range(struct address_space *mapping, | |
1963 | loff_t lstart, loff_t lend) | |
1964 | { | |
1965 | int ret; | |
1966 | ||
1967 | ret = filemap_write_and_wait_range(mapping, lstart, lend); | |
1968 | if (ret == 0) | |
1969 | ret = pnfs_sync_inode(mapping->host, true); | |
1970 | return ret; | |
1971 | } | |
1972 | EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); | |
1973 | ||
acdc53b2 TM |
1974 | /* |
1975 | * flush the inode to disk. | |
1976 | */ | |
1977 | int nfs_wb_all(struct inode *inode) | |
34901f70 | 1978 | { |
f4ce1299 TM |
1979 | int ret; |
1980 | ||
1981 | trace_nfs_writeback_inode_enter(inode); | |
1982 | ||
5bb89b47 | 1983 | ret = filemap_write_and_wait(inode->i_mapping); |
6b196875 CL |
1984 | if (ret) |
1985 | goto out; | |
1986 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
1987 | if (ret < 0) | |
1988 | goto out; | |
1989 | pnfs_sync_inode(inode, true); | |
1990 | ret = 0; | |
34901f70 | 1991 | |
6b196875 | 1992 | out: |
f4ce1299 TM |
1993 | trace_nfs_writeback_inode_exit(inode, ret); |
1994 | return ret; | |
1c75950b | 1995 | } |
ddda8e0a | 1996 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1997 | |
1b3b4a1a TM |
1998 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1999 | { | |
2000 | struct nfs_page *req; | |
1b3b4a1a TM |
2001 | int ret = 0; |
2002 | ||
3e217045 WAA |
2003 | wait_on_page_writeback(page); |
2004 | ||
2005 | /* blocking call to cancel all requests and join to a single (head) | |
2006 | * request */ | |
6d17d653 | 2007 | req = nfs_lock_and_join_requests(page); |
3e217045 WAA |
2008 | |
2009 | if (IS_ERR(req)) { | |
2010 | ret = PTR_ERR(req); | |
2011 | } else if (req) { | |
2012 | /* all requests from this page have been cancelled by | |
2013 | * nfs_lock_and_join_requests, so just remove the head | |
2014 | * request from the inode / page_private pointer and | |
2015 | * release it */ | |
2016 | nfs_inode_remove_request(req); | |
3e217045 | 2017 | nfs_unlock_and_release_request(req); |
1b3b4a1a | 2018 | } |
3e217045 | 2019 | |
1b3b4a1a TM |
2020 | return ret; |
2021 | } | |
2022 | ||
7f2f12d9 TM |
2023 | /* |
2024 | * Write back all requests on one page - we do this before reading it. | |
2025 | */ | |
c373fff7 | 2026 | int nfs_wb_page(struct inode *inode, struct page *page) |
1c75950b | 2027 | { |
29418aa4 | 2028 | loff_t range_start = page_file_offset(page); |
09cbfeaf | 2029 | loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); |
4d770ccf | 2030 | struct writeback_control wbc = { |
4d770ccf | 2031 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 2032 | .nr_to_write = 0, |
4d770ccf TM |
2033 | .range_start = range_start, |
2034 | .range_end = range_end, | |
2035 | }; | |
2036 | int ret; | |
1c75950b | 2037 | |
f4ce1299 TM |
2038 | trace_nfs_writeback_page_enter(inode); |
2039 | ||
0522f6ad | 2040 | for (;;) { |
ba8b06e6 | 2041 | wait_on_page_writeback(page); |
73e3302f | 2042 | if (clear_page_dirty_for_io(page)) { |
c373fff7 | 2043 | ret = nfs_writepage_locked(page, &wbc); |
73e3302f TM |
2044 | if (ret < 0) |
2045 | goto out_error; | |
0522f6ad | 2046 | continue; |
7f2f12d9 | 2047 | } |
f4ce1299 | 2048 | ret = 0; |
0522f6ad TM |
2049 | if (!PagePrivate(page)) |
2050 | break; | |
2051 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 2052 | if (ret < 0) |
73e3302f | 2053 | goto out_error; |
7f2f12d9 | 2054 | } |
73e3302f | 2055 | out_error: |
f4ce1299 | 2056 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 2057 | return ret; |
1c75950b TM |
2058 | } |
2059 | ||
074cc1de TM |
2060 | #ifdef CONFIG_MIGRATION |
2061 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 2062 | struct page *page, enum migrate_mode mode) |
074cc1de | 2063 | { |
2da95652 JL |
2064 | /* |
2065 | * If PagePrivate is set, then the page is currently associated with | |
2066 | * an in-progress read or write request. Don't try to migrate it. | |
2067 | * | |
2068 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
2069 | * that we can safely release the inode reference while holding | |
2070 | * the page lock. | |
2071 | */ | |
2072 | if (PagePrivate(page)) | |
2073 | return -EBUSY; | |
074cc1de | 2074 | |
8c209ce7 DH |
2075 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
2076 | return -EBUSY; | |
074cc1de | 2077 | |
a6bc32b8 | 2078 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
2079 | } |
2080 | #endif | |
2081 | ||
f7b422b1 | 2082 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
2083 | { |
2084 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 2085 | sizeof(struct nfs_pgio_header), |
1da177e4 | 2086 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 2087 | NULL); |
1da177e4 LT |
2088 | if (nfs_wdata_cachep == NULL) |
2089 | return -ENOMEM; | |
2090 | ||
93d2341c MD |
2091 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
2092 | nfs_wdata_cachep); | |
1da177e4 | 2093 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 2094 | goto out_destroy_write_cache; |
1da177e4 | 2095 | |
0b7c0153 FI |
2096 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
2097 | sizeof(struct nfs_commit_data), | |
2098 | 0, SLAB_HWCACHE_ALIGN, | |
2099 | NULL); | |
2100 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 2101 | goto out_destroy_write_mempool; |
0b7c0153 | 2102 | |
93d2341c | 2103 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 2104 | nfs_cdata_cachep); |
1da177e4 | 2105 | if (nfs_commit_mempool == NULL) |
3dd4765f | 2106 | goto out_destroy_commit_cache; |
1da177e4 | 2107 | |
89a09141 PZ |
2108 | /* |
2109 | * NFS congestion size, scale with available memory. | |
2110 | * | |
2111 | * 64MB: 8192k | |
2112 | * 128MB: 11585k | |
2113 | * 256MB: 16384k | |
2114 | * 512MB: 23170k | |
2115 | * 1GB: 32768k | |
2116 | * 2GB: 46340k | |
2117 | * 4GB: 65536k | |
2118 | * 8GB: 92681k | |
2119 | * 16GB: 131072k | |
2120 | * | |
2121 | * This allows larger machines to have larger/more transfers. | |
2122 | * Limit the default to 256M | |
2123 | */ | |
2124 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
2125 | if (nfs_congestion_kb > 256*1024) | |
2126 | nfs_congestion_kb = 256*1024; | |
2127 | ||
1da177e4 | 2128 | return 0; |
3dd4765f JL |
2129 | |
2130 | out_destroy_commit_cache: | |
2131 | kmem_cache_destroy(nfs_cdata_cachep); | |
2132 | out_destroy_write_mempool: | |
2133 | mempool_destroy(nfs_wdata_mempool); | |
2134 | out_destroy_write_cache: | |
2135 | kmem_cache_destroy(nfs_wdata_cachep); | |
2136 | return -ENOMEM; | |
1da177e4 LT |
2137 | } |
2138 | ||
266bee88 | 2139 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
2140 | { |
2141 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 2142 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 2143 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 2144 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
2145 | } |
2146 | ||
4a0de55c AS |
2147 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
2148 | .rw_alloc_header = nfs_writehdr_alloc, | |
2149 | .rw_free_header = nfs_writehdr_free, | |
0eecb214 AS |
2150 | .rw_done = nfs_writeback_done, |
2151 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 2152 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 2153 | }; |