]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
afeacc8c | 23 | #include <linux/export.h> |
af7cf057 TM |
24 | #include <linux/freezer.h> |
25 | #include <linux/wait.h> | |
3fcfab16 | 26 | |
1da177e4 | 27 | #include <asm/uaccess.h> |
1da177e4 LT |
28 | |
29 | #include "delegation.h" | |
49a70f27 | 30 | #include "internal.h" |
91d5b470 | 31 | #include "iostat.h" |
def6ed7e | 32 | #include "nfs4_fs.h" |
074cc1de | 33 | #include "fscache.h" |
94ad1c80 | 34 | #include "pnfs.h" |
1da177e4 | 35 | |
f4ce1299 TM |
36 | #include "nfstrace.h" |
37 | ||
1da177e4 LT |
38 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
39 | ||
40 | #define MIN_POOL_WRITE (32) | |
41 | #define MIN_POOL_COMMIT (4) | |
42 | ||
43 | /* | |
44 | * Local function declarations | |
45 | */ | |
f8512ad0 | 46 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 | 47 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 48 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 49 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
4a0de55c | 50 | static const struct nfs_rw_ops nfs_rw_write_ops; |
d4581383 | 51 | static void nfs_clear_request_commit(struct nfs_page *req); |
02d1426c WAA |
52 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
53 | struct inode *inode); | |
3a3908c8 TM |
54 | static struct nfs_page * |
55 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
56 | struct page *page); | |
1da177e4 | 57 | |
e18b890b | 58 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 59 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 60 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
61 | static mempool_t *nfs_commit_mempool; |
62 | ||
0b7c0153 | 63 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 64 | { |
192e501b | 65 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); |
40859d7e | 66 | |
1da177e4 LT |
67 | if (p) { |
68 | memset(p, 0, sizeof(*p)); | |
69 | INIT_LIST_HEAD(&p->pages); | |
70 | } | |
71 | return p; | |
72 | } | |
e0c2b380 | 73 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 74 | |
0b7c0153 | 75 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
76 | { |
77 | mempool_free(p, nfs_commit_mempool); | |
78 | } | |
e0c2b380 | 79 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 80 | |
1e7f3a48 | 81 | static struct nfs_pgio_header *nfs_writehdr_alloc(void) |
3feb2d49 | 82 | { |
1e7f3a48 | 83 | struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
cd841605 | 84 | |
4a0de55c | 85 | if (p) |
3feb2d49 | 86 | memset(p, 0, sizeof(*p)); |
3feb2d49 TM |
87 | return p; |
88 | } | |
6c75dc0d | 89 | |
1e7f3a48 | 90 | static void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 91 | { |
1e7f3a48 | 92 | mempool_free(hdr, nfs_wdata_mempool); |
3feb2d49 | 93 | } |
1da177e4 | 94 | |
7b159fc1 TM |
95 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
96 | { | |
97 | ctx->error = error; | |
98 | smp_wmb(); | |
99 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
100 | } | |
101 | ||
84d3a9a9 WAA |
102 | /* |
103 | * nfs_page_find_head_request_locked - find head request associated with @page | |
104 | * | |
105 | * must be called while holding the inode lock. | |
106 | * | |
107 | * returns matching head request with reference held, or NULL if not found. | |
108 | */ | |
29418aa4 | 109 | static struct nfs_page * |
84d3a9a9 | 110 | nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) |
277459d2 TM |
111 | { |
112 | struct nfs_page *req = NULL; | |
113 | ||
29418aa4 | 114 | if (PagePrivate(page)) |
277459d2 | 115 | req = (struct nfs_page *)page_private(page); |
02d1426c WAA |
116 | else if (unlikely(PageSwapCache(page))) |
117 | req = nfs_page_search_commits_for_head_request_locked(nfsi, | |
118 | page); | |
29418aa4 | 119 | |
84d3a9a9 WAA |
120 | if (req) { |
121 | WARN_ON_ONCE(req->wb_head != req); | |
29418aa4 | 122 | kref_get(&req->wb_kref); |
84d3a9a9 | 123 | } |
29418aa4 | 124 | |
277459d2 TM |
125 | return req; |
126 | } | |
127 | ||
84d3a9a9 WAA |
128 | /* |
129 | * nfs_page_find_head_request - find head request associated with @page | |
130 | * | |
131 | * returns matching head request with reference held, or NULL if not found. | |
132 | */ | |
133 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 134 | { |
d56b4ddf | 135 | struct inode *inode = page_file_mapping(page)->host; |
277459d2 | 136 | struct nfs_page *req = NULL; |
277459d2 | 137 | |
587142f8 | 138 | spin_lock(&inode->i_lock); |
84d3a9a9 | 139 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
587142f8 | 140 | spin_unlock(&inode->i_lock); |
277459d2 TM |
141 | return req; |
142 | } | |
143 | ||
1da177e4 LT |
144 | /* Adjust the file length if we're writing beyond the end */ |
145 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
146 | { | |
d56b4ddf | 147 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
148 | loff_t end, i_size; |
149 | pgoff_t end_index; | |
1da177e4 | 150 | |
a3d01454 TM |
151 | spin_lock(&inode->i_lock); |
152 | i_size = i_size_read(inode); | |
153 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
d56b4ddf | 154 | if (i_size > 0 && page_file_index(page) < end_index) |
a3d01454 | 155 | goto out; |
d56b4ddf | 156 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 157 | if (i_size >= end) |
a3d01454 | 158 | goto out; |
1da177e4 | 159 | i_size_write(inode, end); |
a3d01454 TM |
160 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
161 | out: | |
162 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
163 | } |
164 | ||
a301b777 TM |
165 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
166 | static void nfs_set_pageerror(struct page *page) | |
167 | { | |
d56b4ddf | 168 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
169 | } |
170 | ||
d72ddcba WAA |
171 | /* |
172 | * nfs_page_group_search_locked | |
173 | * @head - head request of page group | |
174 | * @page_offset - offset into page | |
175 | * | |
176 | * Search page group with head @head to find a request that contains the | |
177 | * page offset @page_offset. | |
178 | * | |
179 | * Returns a pointer to the first matching nfs request, or NULL if no | |
180 | * match is found. | |
181 | * | |
182 | * Must be called with the page group lock held | |
183 | */ | |
184 | static struct nfs_page * | |
185 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
186 | { | |
187 | struct nfs_page *req; | |
188 | ||
189 | WARN_ON_ONCE(head != head->wb_head); | |
190 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); | |
191 | ||
192 | req = head; | |
193 | do { | |
194 | if (page_offset >= req->wb_pgbase && | |
195 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
196 | return req; | |
197 | ||
198 | req = req->wb_this_page; | |
199 | } while (req != head); | |
200 | ||
201 | return NULL; | |
202 | } | |
203 | ||
204 | /* | |
205 | * nfs_page_group_covers_page | |
206 | * @head - head request of page group | |
207 | * | |
208 | * Return true if the page group with head @head covers the whole page, | |
209 | * returns false otherwise | |
210 | */ | |
211 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
212 | { | |
213 | struct nfs_page *tmp; | |
214 | unsigned int pos = 0; | |
215 | unsigned int len = nfs_page_length(req->wb_page); | |
216 | ||
fd2f3a06 | 217 | nfs_page_group_lock(req, false); |
d72ddcba WAA |
218 | |
219 | do { | |
220 | tmp = nfs_page_group_search_locked(req->wb_head, pos); | |
221 | if (tmp) { | |
222 | /* no way this should happen */ | |
223 | WARN_ON_ONCE(tmp->wb_pgbase != pos); | |
224 | pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); | |
225 | } | |
226 | } while (tmp && pos < len); | |
227 | ||
228 | nfs_page_group_unlock(req); | |
229 | WARN_ON_ONCE(pos > len); | |
230 | return pos == len; | |
231 | } | |
232 | ||
1da177e4 LT |
233 | /* We can set the PG_uptodate flag if we see that a write request |
234 | * covers the full page. | |
235 | */ | |
d72ddcba | 236 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 237 | { |
d72ddcba | 238 | if (PageUptodate(req->wb_page)) |
1da177e4 | 239 | return; |
d72ddcba | 240 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 241 | return; |
d72ddcba | 242 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
243 | } |
244 | ||
1da177e4 LT |
245 | static int wb_priority(struct writeback_control *wbc) |
246 | { | |
e87b4c7a | 247 | int ret = 0; |
1da177e4 | 248 | if (wbc->for_reclaim) |
c63c7b05 | 249 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
e87b4c7a N |
250 | if (wbc->sync_mode == WB_SYNC_ALL) |
251 | ret = FLUSH_COND_STABLE; | |
b17621fe | 252 | if (wbc->for_kupdate || wbc->for_background) |
e87b4c7a N |
253 | ret |= FLUSH_LOWPRI; |
254 | return ret; | |
1da177e4 LT |
255 | } |
256 | ||
89a09141 PZ |
257 | /* |
258 | * NFS congestion control | |
259 | */ | |
260 | ||
261 | int nfs_congestion_kb; | |
262 | ||
263 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
264 | #define NFS_CONGESTION_OFF_THRESH \ | |
265 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
266 | ||
deed85e7 | 267 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 268 | { |
deed85e7 | 269 | struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); |
5a6d41b3 TM |
270 | int ret = test_set_page_writeback(page); |
271 | ||
deed85e7 | 272 | WARN_ON_ONCE(ret != 0); |
89a09141 | 273 | |
deed85e7 TM |
274 | if (atomic_long_inc_return(&nfss->writeback) > |
275 | NFS_CONGESTION_ON_THRESH) { | |
276 | set_bdi_congested(&nfss->backing_dev_info, | |
277 | BLK_RW_ASYNC); | |
89a09141 PZ |
278 | } |
279 | } | |
280 | ||
20633f04 | 281 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 282 | { |
20633f04 | 283 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 PZ |
284 | struct nfs_server *nfss = NFS_SERVER(inode); |
285 | ||
20633f04 WAA |
286 | if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) |
287 | return; | |
288 | ||
289 | end_page_writeback(req->wb_page); | |
c4dc4bee | 290 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 291 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
292 | } |
293 | ||
d4581383 WAA |
294 | |
295 | /* nfs_page_group_clear_bits | |
296 | * @req - an nfs request | |
297 | * clears all page group related bits from @req | |
298 | */ | |
299 | static void | |
300 | nfs_page_group_clear_bits(struct nfs_page *req) | |
301 | { | |
302 | clear_bit(PG_TEARDOWN, &req->wb_flags); | |
303 | clear_bit(PG_UNLOCKPAGE, &req->wb_flags); | |
304 | clear_bit(PG_UPTODATE, &req->wb_flags); | |
305 | clear_bit(PG_WB_END, &req->wb_flags); | |
306 | clear_bit(PG_REMOVE, &req->wb_flags); | |
307 | } | |
308 | ||
309 | ||
310 | /* | |
311 | * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req | |
312 | * | |
313 | * this is a helper function for nfs_lock_and_join_requests | |
314 | * | |
315 | * @inode - inode associated with request page group, must be holding inode lock | |
316 | * @head - head request of page group, must be holding head lock | |
317 | * @req - request that couldn't lock and needs to wait on the req bit lock | |
318 | * @nonblock - if true, don't actually wait | |
319 | * | |
320 | * NOTE: this must be called holding page_group bit lock and inode spin lock | |
321 | * and BOTH will be released before returning. | |
322 | * | |
323 | * returns 0 on success, < 0 on error. | |
324 | */ | |
325 | static int | |
326 | nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, | |
327 | struct nfs_page *req, bool nonblock) | |
328 | __releases(&inode->i_lock) | |
329 | { | |
330 | struct nfs_page *tmp; | |
331 | int ret; | |
332 | ||
333 | /* relinquish all the locks successfully grabbed this run */ | |
334 | for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) | |
335 | nfs_unlock_request(tmp); | |
336 | ||
337 | WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); | |
338 | ||
339 | /* grab a ref on the request that will be waited on */ | |
340 | kref_get(&req->wb_kref); | |
341 | ||
342 | nfs_page_group_unlock(head); | |
343 | spin_unlock(&inode->i_lock); | |
344 | ||
345 | /* release ref from nfs_page_find_head_request_locked */ | |
346 | nfs_release_request(head); | |
347 | ||
348 | if (!nonblock) | |
349 | ret = nfs_wait_on_request(req); | |
350 | else | |
351 | ret = -EAGAIN; | |
352 | nfs_release_request(req); | |
353 | ||
354 | return ret; | |
355 | } | |
356 | ||
357 | /* | |
358 | * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests | |
359 | * | |
360 | * @destroy_list - request list (using wb_this_page) terminated by @old_head | |
361 | * @old_head - the old head of the list | |
362 | * | |
363 | * All subrequests must be locked and removed from all lists, so at this point | |
364 | * they are only "active" in this function, and possibly in nfs_wait_on_request | |
365 | * with a reference held by some other context. | |
366 | */ | |
367 | static void | |
368 | nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, | |
369 | struct nfs_page *old_head) | |
370 | { | |
371 | while (destroy_list) { | |
372 | struct nfs_page *subreq = destroy_list; | |
373 | ||
374 | destroy_list = (subreq->wb_this_page == old_head) ? | |
375 | NULL : subreq->wb_this_page; | |
376 | ||
377 | WARN_ON_ONCE(old_head != subreq->wb_head); | |
378 | ||
379 | /* make sure old group is not used */ | |
380 | subreq->wb_head = subreq; | |
381 | subreq->wb_this_page = subreq; | |
382 | ||
d4581383 WAA |
383 | /* subreq is now totally disconnected from page group or any |
384 | * write / commit lists. last chance to wake any waiters */ | |
385 | nfs_unlock_request(subreq); | |
386 | ||
387 | if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { | |
388 | /* release ref on old head request */ | |
389 | nfs_release_request(old_head); | |
390 | ||
391 | nfs_page_group_clear_bits(subreq); | |
392 | ||
393 | /* release the PG_INODE_REF reference */ | |
394 | if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) | |
395 | nfs_release_request(subreq); | |
396 | else | |
397 | WARN_ON_ONCE(1); | |
398 | } else { | |
399 | WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); | |
400 | /* zombie requests have already released the last | |
401 | * reference and were waiting on the rest of the | |
402 | * group to complete. Since it's no longer part of a | |
403 | * group, simply free the request */ | |
404 | nfs_page_group_clear_bits(subreq); | |
405 | nfs_free_request(subreq); | |
406 | } | |
407 | } | |
408 | } | |
409 | ||
410 | /* | |
411 | * nfs_lock_and_join_requests - join all subreqs to the head req and return | |
412 | * a locked reference, cancelling any pending | |
413 | * operations for this page. | |
414 | * | |
415 | * @page - the page used to lookup the "page group" of nfs_page structures | |
416 | * @nonblock - if true, don't block waiting for request locks | |
417 | * | |
418 | * This function joins all sub requests to the head request by first | |
419 | * locking all requests in the group, cancelling any pending operations | |
420 | * and finally updating the head request to cover the whole range covered by | |
421 | * the (former) group. All subrequests are removed from any write or commit | |
422 | * lists, unlinked from the group and destroyed. | |
423 | * | |
424 | * Returns a locked, referenced pointer to the head request - which after | |
425 | * this call is guaranteed to be the only request associated with the page. | |
426 | * Returns NULL if no requests are found for @page, or a ERR_PTR if an | |
427 | * error was encountered. | |
428 | */ | |
429 | static struct nfs_page * | |
430 | nfs_lock_and_join_requests(struct page *page, bool nonblock) | |
e261f51f | 431 | { |
d56b4ddf | 432 | struct inode *inode = page_file_mapping(page)->host; |
d4581383 WAA |
433 | struct nfs_page *head, *subreq; |
434 | struct nfs_page *destroy_list = NULL; | |
435 | unsigned int total_bytes; | |
e261f51f TM |
436 | int ret; |
437 | ||
d4581383 WAA |
438 | try_again: |
439 | total_bytes = 0; | |
440 | ||
441 | WARN_ON_ONCE(destroy_list); | |
442 | ||
587142f8 | 443 | spin_lock(&inode->i_lock); |
d4581383 WAA |
444 | |
445 | /* | |
446 | * A reference is taken only on the head request which acts as a | |
447 | * reference to the whole page group - the group will not be destroyed | |
448 | * until the head reference is released. | |
449 | */ | |
450 | head = nfs_page_find_head_request_locked(NFS_I(inode), page); | |
451 | ||
452 | if (!head) { | |
587142f8 | 453 | spin_unlock(&inode->i_lock); |
d4581383 WAA |
454 | return NULL; |
455 | } | |
456 | ||
7c3af975 WAA |
457 | /* holding inode lock, so always make a non-blocking call to try the |
458 | * page group lock */ | |
fd2f3a06 | 459 | ret = nfs_page_group_lock(head, true); |
94970014 WAA |
460 | if (ret < 0) { |
461 | spin_unlock(&inode->i_lock); | |
7c3af975 WAA |
462 | |
463 | if (!nonblock && ret == -EAGAIN) { | |
464 | nfs_page_group_lock_wait(head); | |
465 | nfs_release_request(head); | |
466 | goto try_again; | |
467 | } | |
468 | ||
94970014 | 469 | nfs_release_request(head); |
e7029206 | 470 | return ERR_PTR(ret); |
94970014 | 471 | } |
7c3af975 WAA |
472 | |
473 | /* lock each request in the page group */ | |
d4581383 WAA |
474 | subreq = head; |
475 | do { | |
476 | /* | |
477 | * Subrequests are always contiguous, non overlapping | |
309a1d65 | 478 | * and in order - but may be repeated (mirrored writes). |
d4581383 | 479 | */ |
309a1d65 WAA |
480 | if (subreq->wb_offset == (head->wb_offset + total_bytes)) { |
481 | /* keep track of how many bytes this group covers */ | |
482 | total_bytes += subreq->wb_bytes; | |
483 | } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || | |
484 | ((subreq->wb_offset + subreq->wb_bytes) > | |
485 | (head->wb_offset + total_bytes)))) { | |
486 | nfs_page_group_unlock(head); | |
487 | spin_unlock(&inode->i_lock); | |
488 | return ERR_PTR(-EIO); | |
489 | } | |
d4581383 WAA |
490 | |
491 | if (!nfs_lock_request(subreq)) { | |
492 | /* releases page group bit lock and | |
493 | * inode spin lock and all references */ | |
494 | ret = nfs_unroll_locks_and_wait(inode, head, | |
495 | subreq, nonblock); | |
496 | ||
497 | if (ret == 0) | |
498 | goto try_again; | |
499 | ||
074cc1de | 500 | return ERR_PTR(ret); |
d4581383 WAA |
501 | } |
502 | ||
503 | subreq = subreq->wb_this_page; | |
504 | } while (subreq != head); | |
505 | ||
506 | /* Now that all requests are locked, make sure they aren't on any list. | |
507 | * Commit list removal accounting is done after locks are dropped */ | |
508 | subreq = head; | |
509 | do { | |
411a99ad | 510 | nfs_clear_request_commit(subreq); |
d4581383 WAA |
511 | subreq = subreq->wb_this_page; |
512 | } while (subreq != head); | |
513 | ||
514 | /* unlink subrequests from head, destroy them later */ | |
515 | if (head->wb_this_page != head) { | |
516 | /* destroy list will be terminated by head */ | |
517 | destroy_list = head->wb_this_page; | |
518 | head->wb_this_page = head; | |
519 | ||
520 | /* change head request to cover whole range that | |
521 | * the former page group covered */ | |
522 | head->wb_bytes = total_bytes; | |
e261f51f | 523 | } |
d4581383 WAA |
524 | |
525 | /* | |
526 | * prepare head request to be added to new pgio descriptor | |
527 | */ | |
528 | nfs_page_group_clear_bits(head); | |
529 | ||
530 | /* | |
531 | * some part of the group was still on the inode list - otherwise | |
532 | * the group wouldn't be involved in async write. | |
533 | * grab a reference for the head request, iff it needs one. | |
534 | */ | |
535 | if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) | |
536 | kref_get(&head->wb_kref); | |
537 | ||
538 | nfs_page_group_unlock(head); | |
539 | ||
411a99ad | 540 | /* drop lock to clean uprequests on destroy list */ |
587142f8 | 541 | spin_unlock(&inode->i_lock); |
d4581383 WAA |
542 | |
543 | nfs_destroy_unlinked_subrequests(destroy_list, head); | |
544 | ||
d4581383 WAA |
545 | /* still holds ref on head from nfs_page_find_head_request_locked |
546 | * and still has lock on head from lock loop */ | |
547 | return head; | |
074cc1de TM |
548 | } |
549 | ||
550 | /* | |
551 | * Find an associated nfs write request, and prepare to flush it out | |
552 | * May return an error if the user signalled nfs_wait_on_request(). | |
553 | */ | |
554 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 555 | struct page *page, bool nonblock) |
074cc1de TM |
556 | { |
557 | struct nfs_page *req; | |
558 | int ret = 0; | |
559 | ||
d4581383 | 560 | req = nfs_lock_and_join_requests(page, nonblock); |
074cc1de TM |
561 | if (!req) |
562 | goto out; | |
563 | ret = PTR_ERR(req); | |
564 | if (IS_ERR(req)) | |
565 | goto out; | |
566 | ||
deed85e7 TM |
567 | nfs_set_page_writeback(page); |
568 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 569 | |
deed85e7 | 570 | ret = 0; |
f8512ad0 FI |
571 | if (!nfs_pageio_add_request(pgio, req)) { |
572 | nfs_redirty_request(req); | |
074cc1de | 573 | ret = pgio->pg_error; |
40f90271 TM |
574 | } else |
575 | nfs_add_stats(page_file_mapping(page)->host, | |
576 | NFSIOS_WRITEPAGES, 1); | |
074cc1de TM |
577 | out: |
578 | return ret; | |
e261f51f TM |
579 | } |
580 | ||
f758c885 | 581 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 582 | { |
cfb506e1 | 583 | int ret; |
1da177e4 | 584 | |
d56b4ddf | 585 | nfs_pageio_cond_complete(pgio, page_file_index(page)); |
1b430bee | 586 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
587 | if (ret == -EAGAIN) { |
588 | redirty_page_for_writepage(wbc, page); | |
589 | ret = 0; | |
590 | } | |
591 | return ret; | |
f758c885 | 592 | } |
7fe7f848 | 593 | |
f758c885 TM |
594 | /* |
595 | * Write an mmapped page to the server. | |
596 | */ | |
597 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
598 | { | |
599 | struct nfs_pageio_descriptor pgio; | |
40f90271 | 600 | struct inode *inode = page_file_mapping(page)->host; |
f758c885 | 601 | int err; |
49a70f27 | 602 | |
40f90271 TM |
603 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
604 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), | |
a20c93e3 | 605 | false, &nfs_async_write_completion_ops); |
f758c885 TM |
606 | err = nfs_do_writepage(page, wbc, &pgio); |
607 | nfs_pageio_complete(&pgio); | |
608 | if (err < 0) | |
609 | return err; | |
610 | if (pgio.pg_error < 0) | |
611 | return pgio.pg_error; | |
612 | return 0; | |
4d770ccf TM |
613 | } |
614 | ||
615 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
616 | { | |
f758c885 | 617 | int ret; |
4d770ccf | 618 | |
f758c885 | 619 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 620 | unlock_page(page); |
f758c885 TM |
621 | return ret; |
622 | } | |
623 | ||
624 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
625 | { | |
626 | int ret; | |
627 | ||
628 | ret = nfs_do_writepage(page, wbc, data); | |
629 | unlock_page(page); | |
630 | return ret; | |
1da177e4 LT |
631 | } |
632 | ||
1da177e4 LT |
633 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
634 | { | |
1da177e4 | 635 | struct inode *inode = mapping->host; |
72cb77f4 | 636 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 637 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
638 | int err; |
639 | ||
72cb77f4 | 640 | /* Stop dirtying of new pages while we sync */ |
74316201 | 641 | err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING, |
72cb77f4 TM |
642 | nfs_wait_bit_killable, TASK_KILLABLE); |
643 | if (err) | |
644 | goto out_err; | |
645 | ||
91d5b470 CL |
646 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
647 | ||
a20c93e3 CH |
648 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
649 | &nfs_async_write_completion_ops); | |
f758c885 | 650 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 651 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
652 | |
653 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
4e857c58 | 654 | smp_mb__after_atomic(); |
72cb77f4 TM |
655 | wake_up_bit(bitlock, NFS_INO_FLUSHING); |
656 | ||
f758c885 | 657 | if (err < 0) |
72cb77f4 TM |
658 | goto out_err; |
659 | err = pgio.pg_error; | |
660 | if (err < 0) | |
661 | goto out_err; | |
c63c7b05 | 662 | return 0; |
72cb77f4 TM |
663 | out_err: |
664 | return err; | |
1da177e4 LT |
665 | } |
666 | ||
667 | /* | |
668 | * Insert a write request into an inode | |
669 | */ | |
d6d6dc7c | 670 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
671 | { |
672 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 | 673 | |
2bfc6e56 WAA |
674 | WARN_ON_ONCE(req->wb_this_page != req); |
675 | ||
e7d39069 | 676 | /* Lock the request! */ |
7ad84aa9 | 677 | nfs_lock_request(req); |
e7d39069 TM |
678 | |
679 | spin_lock(&inode->i_lock); | |
cb1410c7 WAA |
680 | if (!nfsi->nrequests && |
681 | NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) | |
a9a4a87a | 682 | inode->i_version++; |
29418aa4 MG |
683 | /* |
684 | * Swap-space should not get truncated. Hence no need to plug the race | |
685 | * with invalidate/truncate. | |
686 | */ | |
687 | if (likely(!PageSwapCache(req->wb_page))) { | |
688 | set_bit(PG_MAPPED, &req->wb_flags); | |
689 | SetPagePrivate(req->wb_page); | |
690 | set_page_private(req->wb_page, (unsigned long)req); | |
691 | } | |
cb1410c7 | 692 | nfsi->nrequests++; |
17089a29 | 693 | /* this a head request for a page group - mark it as having an |
cb1410c7 WAA |
694 | * extra reference so sub groups can follow suit. |
695 | * This flag also informs pgio layer when to bump nrequests when | |
696 | * adding subrequests. */ | |
17089a29 | 697 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); |
c03b4024 | 698 | kref_get(&req->wb_kref); |
e7d39069 | 699 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
700 | } |
701 | ||
702 | /* | |
89a09141 | 703 | * Remove a write request from an inode |
1da177e4 LT |
704 | */ |
705 | static void nfs_inode_remove_request(struct nfs_page *req) | |
706 | { | |
2b0143b5 | 707 | struct inode *inode = d_inode(req->wb_context->dentry); |
1da177e4 | 708 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 709 | struct nfs_page *head; |
1da177e4 | 710 | |
20633f04 WAA |
711 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
712 | head = req->wb_head; | |
713 | ||
714 | spin_lock(&inode->i_lock); | |
715 | if (likely(!PageSwapCache(head->wb_page))) { | |
716 | set_page_private(head->wb_page, 0); | |
717 | ClearPagePrivate(head->wb_page); | |
95905446 N |
718 | smp_mb__after_atomic(); |
719 | wake_up_page(head->wb_page, PG_private); | |
20633f04 WAA |
720 | clear_bit(PG_MAPPED, &head->wb_flags); |
721 | } | |
cb1410c7 WAA |
722 | nfsi->nrequests--; |
723 | spin_unlock(&inode->i_lock); | |
724 | } else { | |
725 | spin_lock(&inode->i_lock); | |
726 | nfsi->nrequests--; | |
20633f04 | 727 | spin_unlock(&inode->i_lock); |
29418aa4 | 728 | } |
17089a29 WAA |
729 | |
730 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) | |
731 | nfs_release_request(req); | |
1da177e4 LT |
732 | } |
733 | ||
61822ab5 | 734 | static void |
6d884e8f | 735 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 736 | { |
61822ab5 TM |
737 | __set_page_dirty_nobuffers(req->wb_page); |
738 | } | |
739 | ||
3a3908c8 TM |
740 | /* |
741 | * nfs_page_search_commits_for_head_request_locked | |
742 | * | |
743 | * Search through commit lists on @inode for the head request for @page. | |
744 | * Must be called while holding the inode (which is cinfo) lock. | |
745 | * | |
746 | * Returns the head request if found, or NULL if not found. | |
747 | */ | |
748 | static struct nfs_page * | |
749 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
750 | struct page *page) | |
751 | { | |
752 | struct nfs_page *freq, *t; | |
753 | struct nfs_commit_info cinfo; | |
754 | struct inode *inode = &nfsi->vfs_inode; | |
755 | ||
756 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
757 | ||
758 | /* search through pnfs commit lists */ | |
759 | freq = pnfs_search_commit_reqs(inode, &cinfo, page); | |
760 | if (freq) | |
761 | return freq->wb_head; | |
762 | ||
763 | /* Linearly search the commit list for the correct request */ | |
764 | list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { | |
765 | if (freq->wb_page == page) | |
766 | return freq->wb_head; | |
767 | } | |
768 | ||
769 | return NULL; | |
770 | } | |
771 | ||
86d80f97 TM |
772 | /** |
773 | * nfs_request_add_commit_list_locked - add request to a commit list | |
774 | * @req: pointer to a struct nfs_page | |
775 | * @dst: commit list head | |
776 | * @cinfo: holds list lock and accounting info | |
777 | * | |
778 | * This sets the PG_CLEAN bit, updates the cinfo count of | |
779 | * number of outstanding requests requiring a commit as well as | |
780 | * the MM page stats. | |
781 | * | |
782 | * The caller must hold the cinfo->lock, and the nfs_page lock. | |
783 | */ | |
784 | void | |
785 | nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, | |
786 | struct nfs_commit_info *cinfo) | |
787 | { | |
788 | set_bit(PG_CLEAN, &req->wb_flags); | |
789 | nfs_list_add_request(req, dst); | |
790 | cinfo->mds->ncommit++; | |
791 | } | |
792 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); | |
793 | ||
8dd37758 TM |
794 | /** |
795 | * nfs_request_add_commit_list - add request to a commit list | |
796 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
797 | * @dst: commit list head |
798 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 799 | * |
ea2cf228 | 800 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
801 | * number of outstanding requests requiring a commit as well as |
802 | * the MM page stats. | |
803 | * | |
ea2cf228 | 804 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 805 | * holding the nfs_page lock. |
1da177e4 | 806 | */ |
8dd37758 | 807 | void |
ea2cf228 FI |
808 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
809 | struct nfs_commit_info *cinfo) | |
1da177e4 | 810 | { |
ea2cf228 | 811 | spin_lock(cinfo->lock); |
86d80f97 | 812 | nfs_request_add_commit_list_locked(req, dst, cinfo); |
ea2cf228 | 813 | spin_unlock(cinfo->lock); |
86d80f97 | 814 | nfs_mark_page_unstable(req->wb_page, cinfo); |
1da177e4 | 815 | } |
8dd37758 TM |
816 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
817 | ||
818 | /** | |
819 | * nfs_request_remove_commit_list - Remove request from a commit list | |
820 | * @req: pointer to a nfs_page | |
ea2cf228 | 821 | * @cinfo: holds list lock and accounting info |
8dd37758 | 822 | * |
ea2cf228 | 823 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
824 | * number of outstanding requests requiring a commit |
825 | * It does not update the MM page stats. | |
826 | * | |
ea2cf228 | 827 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
828 | */ |
829 | void | |
ea2cf228 FI |
830 | nfs_request_remove_commit_list(struct nfs_page *req, |
831 | struct nfs_commit_info *cinfo) | |
8dd37758 | 832 | { |
8dd37758 TM |
833 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
834 | return; | |
835 | nfs_list_remove_request(req); | |
ea2cf228 | 836 | cinfo->mds->ncommit--; |
8dd37758 TM |
837 | } |
838 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
839 | ||
ea2cf228 FI |
840 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
841 | struct inode *inode) | |
842 | { | |
843 | cinfo->lock = &inode->i_lock; | |
844 | cinfo->mds = &NFS_I(inode)->commit_info; | |
845 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 846 | cinfo->dreq = NULL; |
f453a54a | 847 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
848 | } |
849 | ||
850 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
851 | struct inode *inode, | |
852 | struct nfs_direct_req *dreq) | |
853 | { | |
1763da12 FI |
854 | if (dreq) |
855 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
856 | else | |
857 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
858 | } |
859 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
860 | |
861 | /* | |
862 | * Add a request to the inode's commit list. | |
863 | */ | |
1763da12 | 864 | void |
ea2cf228 | 865 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
b57ff130 | 866 | struct nfs_commit_info *cinfo, u32 ds_commit_idx) |
8dd37758 | 867 | { |
b57ff130 | 868 | if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) |
8dd37758 | 869 | return; |
ea2cf228 | 870 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 871 | } |
8e821cad | 872 | |
d6d6dc7c FI |
873 | static void |
874 | nfs_clear_page_commit(struct page *page) | |
875 | { | |
876 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
93f78d88 TH |
877 | dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, |
878 | WB_RECLAIMABLE); | |
d6d6dc7c FI |
879 | } |
880 | ||
411a99ad | 881 | /* Called holding inode (/cinfo) lock */ |
8dd37758 | 882 | static void |
e468bae9 TM |
883 | nfs_clear_request_commit(struct nfs_page *req) |
884 | { | |
8dd37758 | 885 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
2b0143b5 | 886 | struct inode *inode = d_inode(req->wb_context->dentry); |
ea2cf228 | 887 | struct nfs_commit_info cinfo; |
e468bae9 | 888 | |
ea2cf228 FI |
889 | nfs_init_cinfo_from_inode(&cinfo, inode); |
890 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
ea2cf228 | 891 | nfs_request_remove_commit_list(req, &cinfo); |
8dd37758 | 892 | } |
d6d6dc7c | 893 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 894 | } |
e468bae9 TM |
895 | } |
896 | ||
d45f60c6 | 897 | int nfs_write_need_commit(struct nfs_pgio_header *hdr) |
8e821cad | 898 | { |
c65e6254 | 899 | if (hdr->verf.committed == NFS_DATA_SYNC) |
d45f60c6 | 900 | return hdr->lseg == NULL; |
c65e6254 | 901 | return hdr->verf.committed != NFS_FILE_SYNC; |
8e821cad TM |
902 | } |
903 | ||
061ae2ed | 904 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 905 | { |
ea2cf228 | 906 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
907 | unsigned long bytes = 0; |
908 | ||
909 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
910 | goto out; | |
ea2cf228 | 911 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
912 | while (!list_empty(&hdr->pages)) { |
913 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
914 | |
915 | bytes += req->wb_bytes; | |
916 | nfs_list_remove_request(req); | |
917 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
918 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 919 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
920 | nfs_context_set_write_error(req->wb_context, hdr->error); |
921 | goto remove_req; | |
922 | } | |
c65e6254 | 923 | if (nfs_write_need_commit(hdr)) { |
f79d06f5 | 924 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
b57ff130 | 925 | nfs_mark_request_commit(req, hdr->lseg, &cinfo, |
a7d42ddb | 926 | hdr->pgio_mirror_idx); |
6c75dc0d FI |
927 | goto next; |
928 | } | |
929 | remove_req: | |
930 | nfs_inode_remove_request(req); | |
931 | next: | |
1d1afcbc | 932 | nfs_unlock_request(req); |
20633f04 | 933 | nfs_end_page_writeback(req); |
3aff4ebb | 934 | nfs_release_request(req); |
6c75dc0d FI |
935 | } |
936 | out: | |
937 | hdr->release(hdr); | |
8e821cad | 938 | } |
1da177e4 | 939 | |
ce59515c | 940 | unsigned long |
ea2cf228 | 941 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 942 | { |
ea2cf228 | 943 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
944 | } |
945 | ||
ea2cf228 | 946 | /* cinfo->lock held by caller */ |
1763da12 | 947 | int |
ea2cf228 FI |
948 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
949 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
950 | { |
951 | struct nfs_page *req, *tmp; | |
952 | int ret = 0; | |
953 | ||
954 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
955 | if (!nfs_lock_request(req)) |
956 | continue; | |
7ad84aa9 | 957 | kref_get(&req->wb_kref); |
ea2cf228 | 958 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 959 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 960 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
961 | nfs_list_add_request(req, dst); |
962 | ret++; | |
1763da12 | 963 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 964 | break; |
d6d6dc7c FI |
965 | } |
966 | return ret; | |
fb8a1f11 TM |
967 | } |
968 | ||
1da177e4 LT |
969 | /* |
970 | * nfs_scan_commit - Scan an inode for commit requests | |
971 | * @inode: NFS inode to scan | |
ea2cf228 FI |
972 | * @dst: mds destination list |
973 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
974 | * |
975 | * Moves requests from the inode's 'commit' request list. | |
976 | * The requests are *not* checked to ensure that they form a contiguous set. | |
977 | */ | |
1763da12 | 978 | int |
ea2cf228 FI |
979 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
980 | struct nfs_commit_info *cinfo) | |
1da177e4 | 981 | { |
d6d6dc7c | 982 | int ret = 0; |
fb8a1f11 | 983 | |
ea2cf228 FI |
984 | spin_lock(cinfo->lock); |
985 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 986 | const int max = INT_MAX; |
d6d6dc7c | 987 | |
ea2cf228 FI |
988 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
989 | cinfo, max); | |
990 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 991 | } |
ea2cf228 | 992 | spin_unlock(cinfo->lock); |
ff778d02 | 993 | return ret; |
1da177e4 | 994 | } |
d6d6dc7c | 995 | |
1da177e4 | 996 | /* |
e7d39069 TM |
997 | * Search for an existing write request, and attempt to update |
998 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 999 | * |
e7d39069 TM |
1000 | * If the attempt fails, then the existing request is flushed out |
1001 | * to disk. | |
1da177e4 | 1002 | */ |
e7d39069 TM |
1003 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
1004 | struct page *page, | |
1005 | unsigned int offset, | |
1006 | unsigned int bytes) | |
1da177e4 | 1007 | { |
e7d39069 TM |
1008 | struct nfs_page *req; |
1009 | unsigned int rqend; | |
1010 | unsigned int end; | |
1011 | int error; | |
1012 | ||
1013 | if (!PagePrivate(page)) | |
1014 | return NULL; | |
1da177e4 LT |
1015 | |
1016 | end = offset + bytes; | |
e7d39069 | 1017 | spin_lock(&inode->i_lock); |
1da177e4 | 1018 | |
1da177e4 | 1019 | for (;;) { |
84d3a9a9 | 1020 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
e7d39069 TM |
1021 | if (req == NULL) |
1022 | goto out_unlock; | |
1023 | ||
2bfc6e56 WAA |
1024 | /* should be handled by nfs_flush_incompatible */ |
1025 | WARN_ON_ONCE(req->wb_head != req); | |
1026 | WARN_ON_ONCE(req->wb_this_page != req); | |
1027 | ||
e7d39069 TM |
1028 | rqend = req->wb_offset + req->wb_bytes; |
1029 | /* | |
1030 | * Tell the caller to flush out the request if | |
1031 | * the offsets are non-contiguous. | |
1032 | * Note: nfs_flush_incompatible() will already | |
1033 | * have flushed out requests having wrong owners. | |
1034 | */ | |
e468bae9 | 1035 | if (offset > rqend |
e7d39069 TM |
1036 | || end < req->wb_offset) |
1037 | goto out_flushme; | |
1038 | ||
7ad84aa9 | 1039 | if (nfs_lock_request(req)) |
1da177e4 | 1040 | break; |
1da177e4 | 1041 | |
e7d39069 | 1042 | /* The request is locked, so wait and then retry */ |
587142f8 | 1043 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
1044 | error = nfs_wait_on_request(req); |
1045 | nfs_release_request(req); | |
1046 | if (error != 0) | |
1047 | goto out_err; | |
1048 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
1049 | } |
1050 | ||
1051 | /* Okay, the request matches. Update the region */ | |
1052 | if (offset < req->wb_offset) { | |
1053 | req->wb_offset = offset; | |
1054 | req->wb_pgbase = offset; | |
1da177e4 | 1055 | } |
1da177e4 LT |
1056 | if (end > rqend) |
1057 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1058 | else |
1059 | req->wb_bytes = rqend - req->wb_offset; | |
1060 | out_unlock: | |
ca138f36 FI |
1061 | if (req) |
1062 | nfs_clear_request_commit(req); | |
411a99ad | 1063 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
1064 | return req; |
1065 | out_flushme: | |
1066 | spin_unlock(&inode->i_lock); | |
1067 | nfs_release_request(req); | |
1068 | error = nfs_wb_page(inode, page); | |
1069 | out_err: | |
1070 | return ERR_PTR(error); | |
1071 | } | |
1072 | ||
1073 | /* | |
1074 | * Try to update an existing write request, or create one if there is none. | |
1075 | * | |
1076 | * Note: Should always be called with the Page Lock held to prevent races | |
1077 | * if we have to add a new request. Also assumes that the caller has | |
1078 | * already called nfs_flush_incompatible() if necessary. | |
1079 | */ | |
1080 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1081 | struct page *page, unsigned int offset, unsigned int bytes) | |
1082 | { | |
d56b4ddf | 1083 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1084 | struct nfs_page *req; |
1da177e4 | 1085 | |
e7d39069 TM |
1086 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1087 | if (req != NULL) | |
1088 | goto out; | |
2bfc6e56 | 1089 | req = nfs_create_request(ctx, page, NULL, offset, bytes); |
e7d39069 TM |
1090 | if (IS_ERR(req)) |
1091 | goto out; | |
d6d6dc7c | 1092 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1093 | out: |
61e930a9 | 1094 | return req; |
1da177e4 LT |
1095 | } |
1096 | ||
e7d39069 TM |
1097 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1098 | unsigned int offset, unsigned int count) | |
1099 | { | |
1100 | struct nfs_page *req; | |
1101 | ||
1102 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1103 | if (IS_ERR(req)) | |
1104 | return PTR_ERR(req); | |
1105 | /* Update file length */ | |
1106 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1107 | nfs_mark_uptodate(req); |
a6305ddb | 1108 | nfs_mark_request_dirty(req); |
1d1afcbc | 1109 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1110 | return 0; |
1111 | } | |
1112 | ||
1da177e4 LT |
1113 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1114 | { | |
cd3758e3 | 1115 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1116 | struct nfs_lock_context *l_ctx; |
bd61e0a9 | 1117 | struct file_lock_context *flctx = file_inode(file)->i_flctx; |
1da177e4 | 1118 | struct nfs_page *req; |
1a54533e | 1119 | int do_flush, status; |
1da177e4 LT |
1120 | /* |
1121 | * Look for a request corresponding to this page. If there | |
1122 | * is one, and it belongs to another file, we flush it out | |
1123 | * before we try to copy anything into the page. Do this | |
1124 | * due to the lack of an ACCESS-type call in NFSv2. | |
1125 | * Also do the same if we find a request from an existing | |
1126 | * dropped page. | |
1127 | */ | |
1a54533e | 1128 | do { |
84d3a9a9 | 1129 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1130 | if (req == NULL) |
1131 | return 0; | |
2a369153 TM |
1132 | l_ctx = req->wb_lock_context; |
1133 | do_flush = req->wb_page != page || req->wb_context != ctx; | |
2bfc6e56 WAA |
1134 | /* for now, flush if more than 1 request in page_group */ |
1135 | do_flush |= req->wb_this_page != req; | |
bd61e0a9 JL |
1136 | if (l_ctx && flctx && |
1137 | !(list_empty_careful(&flctx->flc_posix) && | |
1138 | list_empty_careful(&flctx->flc_flock))) { | |
5263e31e JL |
1139 | do_flush |= l_ctx->lockowner.l_owner != current->files |
1140 | || l_ctx->lockowner.l_pid != current->tgid; | |
1141 | } | |
1da177e4 | 1142 | nfs_release_request(req); |
1a54533e TM |
1143 | if (!do_flush) |
1144 | return 0; | |
d56b4ddf | 1145 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1146 | } while (status == 0); |
1147 | return status; | |
1da177e4 LT |
1148 | } |
1149 | ||
dc24826b AA |
1150 | /* |
1151 | * Avoid buffered writes when a open context credential's key would | |
1152 | * expire soon. | |
1153 | * | |
1154 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1155 | * | |
1156 | * Return 0 and set a credential flag which triggers the inode to flush | |
1157 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1158 | * RPC_KEY_EXPIRE_TIMEO. | |
1159 | */ | |
1160 | int | |
1161 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1162 | { | |
1163 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
1164 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
1165 | ||
1166 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * Test if the open context credential key is marked to expire soon. | |
1171 | */ | |
1172 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) | |
1173 | { | |
1174 | return rpcauth_cred_key_to_expire(ctx->cred); | |
1175 | } | |
1176 | ||
5d47a356 TM |
1177 | /* |
1178 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1179 | * the PageUptodate() flag. In this case, we will need to turn off | |
1180 | * write optimisations that depend on the page contents being correct. | |
1181 | */ | |
8d197a56 | 1182 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 1183 | { |
d529ef83 JL |
1184 | struct nfs_inode *nfsi = NFS_I(inode); |
1185 | ||
8d197a56 TM |
1186 | if (nfs_have_delegated_attributes(inode)) |
1187 | goto out; | |
18dd78c4 | 1188 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
d529ef83 | 1189 | return false; |
4db72b40 | 1190 | smp_rmb(); |
d529ef83 | 1191 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
1192 | return false; |
1193 | out: | |
18dd78c4 SM |
1194 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
1195 | return false; | |
8d197a56 | 1196 | return PageUptodate(page) != 0; |
5d47a356 TM |
1197 | } |
1198 | ||
5263e31e JL |
1199 | static bool |
1200 | is_whole_file_wrlock(struct file_lock *fl) | |
1201 | { | |
1202 | return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && | |
1203 | fl->fl_type == F_WRLCK; | |
1204 | } | |
1205 | ||
c7559663 SM |
1206 | /* If we know the page is up to date, and we're not using byte range locks (or |
1207 | * if we have the whole file locked for writing), it may be more efficient to | |
1208 | * extend the write to cover the entire page in order to avoid fragmentation | |
1209 | * inefficiencies. | |
1210 | * | |
263b4509 SM |
1211 | * If the file is opened for synchronous writes then we can just skip the rest |
1212 | * of the checks. | |
c7559663 SM |
1213 | */ |
1214 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
1215 | { | |
5263e31e JL |
1216 | int ret; |
1217 | struct file_lock_context *flctx = inode->i_flctx; | |
1218 | struct file_lock *fl; | |
1219 | ||
c7559663 SM |
1220 | if (file->f_flags & O_DSYNC) |
1221 | return 0; | |
263b4509 SM |
1222 | if (!nfs_write_pageuptodate(page, inode)) |
1223 | return 0; | |
c7559663 SM |
1224 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1225 | return 1; | |
bd61e0a9 JL |
1226 | if (!flctx || (list_empty_careful(&flctx->flc_flock) && |
1227 | list_empty_careful(&flctx->flc_posix))) | |
8fa4592a | 1228 | return 1; |
5263e31e JL |
1229 | |
1230 | /* Check to see if there are whole file write locks */ | |
5263e31e | 1231 | ret = 0; |
6109c850 | 1232 | spin_lock(&flctx->flc_lock); |
bd61e0a9 JL |
1233 | if (!list_empty(&flctx->flc_posix)) { |
1234 | fl = list_first_entry(&flctx->flc_posix, struct file_lock, | |
1235 | fl_list); | |
1236 | if (is_whole_file_wrlock(fl)) | |
1237 | ret = 1; | |
1238 | } else if (!list_empty(&flctx->flc_flock)) { | |
5263e31e JL |
1239 | fl = list_first_entry(&flctx->flc_flock, struct file_lock, |
1240 | fl_list); | |
1241 | if (fl->fl_type == F_WRLCK) | |
1242 | ret = 1; | |
1243 | } | |
6109c850 | 1244 | spin_unlock(&flctx->flc_lock); |
5263e31e | 1245 | return ret; |
c7559663 SM |
1246 | } |
1247 | ||
1da177e4 LT |
1248 | /* |
1249 | * Update and possibly write a cached page of an NFS file. | |
1250 | * | |
1251 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1252 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1253 | */ | |
1254 | int nfs_updatepage(struct file *file, struct page *page, | |
1255 | unsigned int offset, unsigned int count) | |
1256 | { | |
cd3758e3 | 1257 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 1258 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
1259 | int status = 0; |
1260 | ||
91d5b470 CL |
1261 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1262 | ||
6de1472f AV |
1263 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1264 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1265 | |
c7559663 | 1266 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 1267 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 1268 | offset = 0; |
1da177e4 LT |
1269 | } |
1270 | ||
e21195a7 | 1271 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
1272 | if (status < 0) |
1273 | nfs_set_pageerror(page); | |
59b7c05f TM |
1274 | else |
1275 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 1276 | |
48186c7d | 1277 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1278 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1279 | return status; |
1280 | } | |
1281 | ||
3ff7576d | 1282 | static int flush_task_priority(int how) |
1da177e4 LT |
1283 | { |
1284 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1285 | case FLUSH_HIGHPRI: | |
1286 | return RPC_PRIORITY_HIGH; | |
1287 | case FLUSH_LOWPRI: | |
1288 | return RPC_PRIORITY_LOW; | |
1289 | } | |
1290 | return RPC_PRIORITY_NORMAL; | |
1291 | } | |
1292 | ||
d45f60c6 WAA |
1293 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1294 | struct rpc_message *msg, | |
abde71f4 | 1295 | const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 | 1296 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1297 | { |
3ff7576d | 1298 | int priority = flush_task_priority(how); |
d138d5d1 | 1299 | |
1ed26f33 | 1300 | task_setup_data->priority = priority; |
abde71f4 | 1301 | rpc_ops->write_setup(hdr, msg); |
d138d5d1 | 1302 | |
abde71f4 | 1303 | nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, |
d45f60c6 | 1304 | &task_setup_data->rpc_client, msg, hdr); |
275acaaf TM |
1305 | } |
1306 | ||
6d884e8f F |
1307 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1308 | * call this on each, which will prepare them to be retried on next | |
1309 | * writeback using standard nfs. | |
1310 | */ | |
1311 | static void nfs_redirty_request(struct nfs_page *req) | |
1312 | { | |
1313 | nfs_mark_request_dirty(req); | |
c7070113 | 1314 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1d1afcbc | 1315 | nfs_unlock_request(req); |
20633f04 | 1316 | nfs_end_page_writeback(req); |
3aff4ebb | 1317 | nfs_release_request(req); |
6d884e8f F |
1318 | } |
1319 | ||
061ae2ed | 1320 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1321 | { |
1322 | struct nfs_page *req; | |
1323 | ||
1324 | while (!list_empty(head)) { | |
1325 | req = nfs_list_entry(head->next); | |
1326 | nfs_list_remove_request(req); | |
1327 | nfs_redirty_request(req); | |
1328 | } | |
1329 | } | |
1330 | ||
dc602dd7 TM |
1331 | static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) |
1332 | { | |
1333 | nfs_async_write_error(&hdr->pages); | |
1334 | } | |
1335 | ||
061ae2ed FI |
1336 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1337 | .error_cleanup = nfs_async_write_error, | |
1338 | .completion = nfs_write_completion, | |
dc602dd7 | 1339 | .reschedule_io = nfs_async_write_reschedule_io, |
061ae2ed FI |
1340 | }; |
1341 | ||
57208fa7 | 1342 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1343 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1344 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1345 | { |
a20c93e3 | 1346 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1347 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1348 | |
1349 | #ifdef CONFIG_NFS_V4_1 | |
1350 | if (server->pnfs_curr_ld && !force_mds) | |
1351 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1352 | #endif | |
4a0de55c AS |
1353 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
1354 | server->wsize, ioflags); | |
1751c363 | 1355 | } |
ddda8e0a | 1356 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1357 | |
dce81290 TM |
1358 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1359 | { | |
a7d42ddb WAA |
1360 | struct nfs_pgio_mirror *mirror; |
1361 | ||
6f29b9bb KM |
1362 | if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) |
1363 | pgio->pg_ops->pg_cleanup(pgio); | |
1364 | ||
41d8d5b7 | 1365 | pgio->pg_ops = &nfs_pgio_rw_ops; |
a7d42ddb WAA |
1366 | |
1367 | nfs_pageio_stop_mirroring(pgio); | |
1368 | ||
1369 | mirror = &pgio->pg_mirrors[0]; | |
1370 | mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
dce81290 | 1371 | } |
1f945357 | 1372 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1373 | |
1da177e4 | 1374 | |
0b7c0153 FI |
1375 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1376 | { | |
1377 | struct nfs_commit_data *data = calldata; | |
1378 | ||
1379 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1380 | } | |
1381 | ||
1f2edbe3 TM |
1382 | /* |
1383 | * Special version of should_remove_suid() that ignores capabilities. | |
1384 | */ | |
1385 | static int nfs_should_remove_suid(const struct inode *inode) | |
1386 | { | |
1387 | umode_t mode = inode->i_mode; | |
1388 | int kill = 0; | |
1389 | ||
1390 | /* suid always must be killed */ | |
1391 | if (unlikely(mode & S_ISUID)) | |
1392 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1393 | |
1f2edbe3 TM |
1394 | /* |
1395 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1396 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1397 | */ | |
1398 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1399 | kill |= ATTR_KILL_SGID; | |
1400 | ||
1401 | if (unlikely(kill && S_ISREG(mode))) | |
1402 | return kill; | |
1403 | ||
1404 | return 0; | |
1405 | } | |
788e7a89 | 1406 | |
a08a8cd3 TM |
1407 | static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, |
1408 | struct nfs_fattr *fattr) | |
1409 | { | |
1410 | struct nfs_pgio_args *argp = &hdr->args; | |
1411 | struct nfs_pgio_res *resp = &hdr->res; | |
2b83d3de | 1412 | u64 size = argp->offset + resp->count; |
a08a8cd3 TM |
1413 | |
1414 | if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) | |
2b83d3de TM |
1415 | fattr->size = size; |
1416 | if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { | |
1417 | fattr->valid &= ~NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 | 1418 | return; |
2b83d3de TM |
1419 | } |
1420 | if (size != fattr->size) | |
a08a8cd3 TM |
1421 | return; |
1422 | /* Set attribute barrier */ | |
1423 | nfs_fattr_set_barrier(fattr); | |
2b83d3de TM |
1424 | /* ...and update size */ |
1425 | fattr->valid |= NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 TM |
1426 | } |
1427 | ||
1428 | void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) | |
1429 | { | |
2b83d3de | 1430 | struct nfs_fattr *fattr = &hdr->fattr; |
a08a8cd3 TM |
1431 | struct inode *inode = hdr->inode; |
1432 | ||
a08a8cd3 TM |
1433 | spin_lock(&inode->i_lock); |
1434 | nfs_writeback_check_extend(hdr, fattr); | |
1435 | nfs_post_op_update_inode_force_wcc_locked(inode, fattr); | |
1436 | spin_unlock(&inode->i_lock); | |
1437 | } | |
1438 | EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); | |
1439 | ||
1da177e4 LT |
1440 | /* |
1441 | * This function is called when the WRITE call is complete. | |
1442 | */ | |
d45f60c6 WAA |
1443 | static int nfs_writeback_done(struct rpc_task *task, |
1444 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1445 | struct inode *inode) |
1da177e4 | 1446 | { |
788e7a89 | 1447 | int status; |
1da177e4 | 1448 | |
f551e44f CL |
1449 | /* |
1450 | * ->write_done will attempt to use post-op attributes to detect | |
1451 | * conflicting writes by other clients. A strict interpretation | |
1452 | * of close-to-open would allow us to continue caching even if | |
1453 | * another writer had changed the file, but some applications | |
1454 | * depend on tighter cache coherency when writing. | |
1455 | */ | |
d45f60c6 | 1456 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1457 | if (status != 0) |
0eecb214 | 1458 | return status; |
d45f60c6 | 1459 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
91d5b470 | 1460 | |
d45f60c6 WAA |
1461 | if (hdr->res.verf->committed < hdr->args.stable && |
1462 | task->tk_status >= 0) { | |
1da177e4 LT |
1463 | /* We tried a write call, but the server did not |
1464 | * commit data to stable storage even though we | |
1465 | * requested it. | |
1466 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1467 | * the server reports NFS_DATA_SYNC, but performs | |
1468 | * NFS_FILE_SYNC. We therefore implement this checking | |
1469 | * as a dprintk() in order to avoid filling syslog. | |
1470 | */ | |
1471 | static unsigned long complain; | |
1472 | ||
a69aef14 | 1473 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1474 | if (time_before(complain, jiffies)) { |
48186c7d | 1475 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1476 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1477 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
d45f60c6 | 1478 | hdr->res.verf->committed, hdr->args.stable); |
1da177e4 LT |
1479 | complain = jiffies + 300 * HZ; |
1480 | } | |
1481 | } | |
1f2edbe3 TM |
1482 | |
1483 | /* Deal with the suid/sgid bit corner case */ | |
1484 | if (nfs_should_remove_suid(inode)) | |
1485 | nfs_mark_for_revalidate(inode); | |
0eecb214 AS |
1486 | return 0; |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * This function is called when the WRITE call is complete. | |
1491 | */ | |
d45f60c6 WAA |
1492 | static void nfs_writeback_result(struct rpc_task *task, |
1493 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1494 | { |
d45f60c6 WAA |
1495 | struct nfs_pgio_args *argp = &hdr->args; |
1496 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1497 | |
1498 | if (resp->count < argp->count) { | |
1da177e4 LT |
1499 | static unsigned long complain; |
1500 | ||
6c75dc0d | 1501 | /* This a short write! */ |
d45f60c6 | 1502 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1503 | |
1da177e4 | 1504 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1505 | if (resp->count == 0) { |
1506 | if (time_before(complain, jiffies)) { | |
1507 | printk(KERN_WARNING | |
1508 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1509 | argp->count); | |
1510 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1511 | } |
d45f60c6 | 1512 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1513 | task->tk_status = -EIO; |
13602896 | 1514 | return; |
1da177e4 | 1515 | } |
f8417b48 KM |
1516 | |
1517 | /* For non rpc-based layout drivers, retry-through-MDS */ | |
1518 | if (!task->tk_ops) { | |
1519 | hdr->pnfs_error = -EAGAIN; | |
1520 | return; | |
1521 | } | |
1522 | ||
6c75dc0d FI |
1523 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1524 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1525 | /* Resend from where the server left off */ | |
d45f60c6 | 1526 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1527 | argp->offset += resp->count; |
1528 | argp->pgbase += resp->count; | |
1529 | argp->count -= resp->count; | |
1530 | } else { | |
1531 | /* Resend as a stable write in order to avoid | |
1532 | * headaches in the case of a server crash. | |
1533 | */ | |
1534 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1535 | } |
6c75dc0d | 1536 | rpc_restart_call_prepare(task); |
1da177e4 | 1537 | } |
1da177e4 LT |
1538 | } |
1539 | ||
af7cf057 TM |
1540 | static int nfs_wait_atomic_killable(atomic_t *key) |
1541 | { | |
1542 | if (fatal_signal_pending(current)) | |
1543 | return -ERESTARTSYS; | |
1544 | freezable_schedule_unsafe(); | |
1545 | return 0; | |
1546 | } | |
1da177e4 | 1547 | |
af7cf057 | 1548 | static int wait_on_commit(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1549 | { |
af7cf057 TM |
1550 | return wait_on_atomic_t(&cinfo->rpcs_out, |
1551 | nfs_wait_atomic_killable, TASK_KILLABLE); | |
1552 | } | |
b8413f98 | 1553 | |
af7cf057 TM |
1554 | static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) |
1555 | { | |
1556 | atomic_inc(&cinfo->rpcs_out); | |
71d0a611 TM |
1557 | } |
1558 | ||
af7cf057 | 1559 | static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1560 | { |
af7cf057 TM |
1561 | if (atomic_dec_and_test(&cinfo->rpcs_out)) |
1562 | wake_up_atomic_t(&cinfo->rpcs_out); | |
71d0a611 TM |
1563 | } |
1564 | ||
0b7c0153 | 1565 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1566 | { |
0b7c0153 FI |
1567 | put_nfs_open_context(data->context); |
1568 | nfs_commit_free(data); | |
1da177e4 | 1569 | } |
e0c2b380 | 1570 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1571 | |
0b7c0153 | 1572 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
c36aae9a | 1573 | const struct nfs_rpc_ops *nfs_ops, |
9ace33cd | 1574 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1575 | int how, int flags) |
1da177e4 | 1576 | { |
07737691 | 1577 | struct rpc_task *task; |
9ace33cd | 1578 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1579 | struct rpc_message msg = { |
1580 | .rpc_argp = &data->args, | |
1581 | .rpc_resp = &data->res, | |
9ace33cd | 1582 | .rpc_cred = data->cred, |
bdc7f021 | 1583 | }; |
84115e1c | 1584 | struct rpc_task_setup task_setup_data = { |
07737691 | 1585 | .task = &data->task, |
9ace33cd | 1586 | .rpc_client = clnt, |
bdc7f021 | 1587 | .rpc_message = &msg, |
9ace33cd | 1588 | .callback_ops = call_ops, |
84115e1c | 1589 | .callback_data = data, |
101070ca | 1590 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1591 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1592 | .priority = priority, |
84115e1c | 1593 | }; |
9ace33cd | 1594 | /* Set up the initial task struct. */ |
c36aae9a | 1595 | nfs_ops->commit_setup(data, &msg); |
9ace33cd | 1596 | |
b4839ebe | 1597 | dprintk("NFS: initiated commit call\n"); |
9ace33cd | 1598 | |
8c21c62c WAA |
1599 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1600 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1601 | ||
9ace33cd FI |
1602 | task = rpc_run_task(&task_setup_data); |
1603 | if (IS_ERR(task)) | |
1604 | return PTR_ERR(task); | |
1605 | if (how & FLUSH_SYNC) | |
1606 | rpc_wait_for_completion_task(task); | |
1607 | rpc_put_task(task); | |
1608 | return 0; | |
1609 | } | |
e0c2b380 | 1610 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1611 | |
378520b8 PT |
1612 | static loff_t nfs_get_lwb(struct list_head *head) |
1613 | { | |
1614 | loff_t lwb = 0; | |
1615 | struct nfs_page *req; | |
1616 | ||
1617 | list_for_each_entry(req, head, wb_list) | |
1618 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1619 | lwb = req_offset(req) + req->wb_bytes; | |
1620 | ||
1621 | return lwb; | |
1622 | } | |
1623 | ||
9ace33cd FI |
1624 | /* |
1625 | * Set up the argument/result storage required for the RPC call. | |
1626 | */ | |
0b7c0153 | 1627 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1628 | struct list_head *head, |
1629 | struct pnfs_layout_segment *lseg, | |
1630 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1631 | { |
1632 | struct nfs_page *first = nfs_list_entry(head->next); | |
2b0143b5 | 1633 | struct inode *inode = d_inode(first->wb_context->dentry); |
1da177e4 LT |
1634 | |
1635 | /* Set up the RPC argument and reply structs | |
1636 | * NB: take care not to mess about with data->commit et al. */ | |
1637 | ||
1638 | list_splice_init(head, &data->pages); | |
1da177e4 | 1639 | |
1da177e4 | 1640 | data->inode = inode; |
9ace33cd | 1641 | data->cred = first->wb_context->cred; |
988b6dce | 1642 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1643 | /* only set lwb for pnfs commit */ |
1644 | if (lseg) | |
1645 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1646 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1647 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1648 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1649 | |
1650 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1651 | /* Note: we always request a commit of the entire inode */ |
1652 | data->args.offset = 0; | |
1653 | data->args.count = 0; | |
0b7c0153 | 1654 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1655 | data->res.fattr = &data->fattr; |
1656 | data->res.verf = &data->verf; | |
0e574af1 | 1657 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1658 | } |
e0c2b380 | 1659 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1660 | |
e0c2b380 | 1661 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 | 1662 | struct pnfs_layout_segment *lseg, |
b57ff130 WAA |
1663 | struct nfs_commit_info *cinfo, |
1664 | u32 ds_commit_idx) | |
64bfeb49 FI |
1665 | { |
1666 | struct nfs_page *req; | |
1667 | ||
1668 | while (!list_empty(page_list)) { | |
1669 | req = nfs_list_entry(page_list->next); | |
1670 | nfs_list_remove_request(req); | |
b57ff130 | 1671 | nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); |
487b9b8a TH |
1672 | if (!cinfo->dreq) |
1673 | nfs_clear_page_commit(req->wb_page); | |
1d1afcbc | 1674 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1675 | } |
1676 | } | |
e0c2b380 | 1677 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1678 | |
b20135d0 TM |
1679 | static void |
1680 | nfs_commit_resched_write(struct nfs_commit_info *cinfo, | |
1681 | struct nfs_page *req) | |
1682 | { | |
1683 | __set_page_dirty_nobuffers(req->wb_page); | |
1684 | } | |
1685 | ||
1da177e4 LT |
1686 | /* |
1687 | * Commit dirty pages | |
1688 | */ | |
1689 | static int | |
ea2cf228 FI |
1690 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1691 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1692 | { |
0b7c0153 | 1693 | struct nfs_commit_data *data; |
1da177e4 | 1694 | |
c9d8f89d | 1695 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1696 | |
1697 | if (!data) | |
1698 | goto out_bad; | |
1699 | ||
1700 | /* Set up the argument struct */ | |
f453a54a FI |
1701 | nfs_init_commit(data, head, NULL, cinfo); |
1702 | atomic_inc(&cinfo->mds->rpcs_out); | |
c36aae9a PT |
1703 | return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), |
1704 | data->mds_ops, how, 0); | |
1da177e4 | 1705 | out_bad: |
b57ff130 | 1706 | nfs_retry_commit(head, NULL, cinfo, 0); |
1da177e4 LT |
1707 | return -ENOMEM; |
1708 | } | |
1709 | ||
1710 | /* | |
1711 | * COMMIT call returned | |
1712 | */ | |
788e7a89 | 1713 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1714 | { |
0b7c0153 | 1715 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1716 | |
a3f565b1 | 1717 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1718 | task->tk_pid, task->tk_status); |
1719 | ||
788e7a89 | 1720 | /* Call the NFS version-specific code */ |
c0d0e96b | 1721 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1722 | } |
1723 | ||
f453a54a | 1724 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1725 | { |
5917ce84 | 1726 | struct nfs_page *req; |
c9d8f89d | 1727 | int status = data->task.tk_status; |
f453a54a | 1728 | struct nfs_commit_info cinfo; |
353db796 | 1729 | struct nfs_server *nfss; |
788e7a89 | 1730 | |
1da177e4 LT |
1731 | while (!list_empty(&data->pages)) { |
1732 | req = nfs_list_entry(data->pages.next); | |
1733 | nfs_list_remove_request(req); | |
d6d6dc7c | 1734 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1735 | |
1e8968c5 | 1736 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1737 | req->wb_context->dentry->d_sb->s_id, |
2b0143b5 | 1738 | (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), |
1da177e4 LT |
1739 | req->wb_bytes, |
1740 | (long long)req_offset(req)); | |
c9d8f89d TM |
1741 | if (status < 0) { |
1742 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1743 | nfs_inode_remove_request(req); |
c9d8f89d | 1744 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1745 | goto next; |
1746 | } | |
1747 | ||
1748 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1749 | * returned by the server against all stored verfs. */ | |
2f2c63bc | 1750 | if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { |
1da177e4 LT |
1751 | /* We have a match */ |
1752 | nfs_inode_remove_request(req); | |
1753 | dprintk(" OK\n"); | |
1754 | goto next; | |
1755 | } | |
1756 | /* We have a mismatch. Write the page again */ | |
1757 | dprintk(" mismatch\n"); | |
6d884e8f | 1758 | nfs_mark_request_dirty(req); |
05990d1b | 1759 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1760 | next: |
1d1afcbc | 1761 | nfs_unlock_and_release_request(req); |
1da177e4 | 1762 | } |
353db796 N |
1763 | nfss = NFS_SERVER(data->inode); |
1764 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
1765 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); | |
1766 | ||
f453a54a | 1767 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
af7cf057 | 1768 | nfs_commit_end(cinfo.mds); |
5917ce84 FI |
1769 | } |
1770 | ||
1771 | static void nfs_commit_release(void *calldata) | |
1772 | { | |
0b7c0153 | 1773 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1774 | |
f453a54a | 1775 | data->completion_ops->completion(data); |
c9d8f89d | 1776 | nfs_commitdata_release(calldata); |
1da177e4 | 1777 | } |
788e7a89 TM |
1778 | |
1779 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1780 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1781 | .rpc_call_done = nfs_commit_done, |
1782 | .rpc_release = nfs_commit_release, | |
1783 | }; | |
1da177e4 | 1784 | |
f453a54a FI |
1785 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1786 | .completion = nfs_commit_release_pages, | |
b20135d0 | 1787 | .resched_write = nfs_commit_resched_write, |
f453a54a FI |
1788 | }; |
1789 | ||
1763da12 FI |
1790 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1791 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1792 | { |
1793 | int status; | |
1794 | ||
ea2cf228 | 1795 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1796 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1797 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1798 | return status; |
1799 | } | |
1800 | ||
b608b283 | 1801 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1802 | { |
1da177e4 | 1803 | LIST_HEAD(head); |
ea2cf228 | 1804 | struct nfs_commit_info cinfo; |
71d0a611 | 1805 | int may_wait = how & FLUSH_SYNC; |
af7cf057 | 1806 | int error = 0; |
b8413f98 | 1807 | int res; |
1da177e4 | 1808 | |
ea2cf228 | 1809 | nfs_init_cinfo_from_inode(&cinfo, inode); |
af7cf057 | 1810 | nfs_commit_begin(cinfo.mds); |
ea2cf228 | 1811 | res = nfs_scan_commit(inode, &head, &cinfo); |
af7cf057 | 1812 | if (res) |
ea2cf228 | 1813 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
af7cf057 TM |
1814 | nfs_commit_end(cinfo.mds); |
1815 | if (error < 0) | |
1816 | goto out_error; | |
1817 | if (!may_wait) | |
1818 | goto out_mark_dirty; | |
1819 | error = wait_on_commit(cinfo.mds); | |
1820 | if (error < 0) | |
1821 | return error; | |
c5efa5fc | 1822 | return res; |
af7cf057 TM |
1823 | out_error: |
1824 | res = error; | |
c5efa5fc TM |
1825 | /* Note: If we exit without ensuring that the commit is complete, |
1826 | * we must mark the inode as dirty. Otherwise, future calls to | |
1827 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1828 | * that the data is on the disk. | |
1829 | */ | |
1830 | out_mark_dirty: | |
1831 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1832 | return res; |
1833 | } | |
b20135d0 | 1834 | EXPORT_SYMBOL_GPL(nfs_commit_inode); |
8fc795f7 | 1835 | |
ae09c31f | 1836 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
8fc795f7 | 1837 | { |
420e3646 TM |
1838 | struct nfs_inode *nfsi = NFS_I(inode); |
1839 | int flags = FLUSH_SYNC; | |
1840 | int ret = 0; | |
8fc795f7 | 1841 | |
3236c3e1 | 1842 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1843 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1844 | return ret; |
1845 | ||
a00dd6c0 JL |
1846 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1847 | /* Don't commit yet if this is a non-blocking flush and there | |
1848 | * are a lot of outstanding writes for this mapping. | |
1849 | */ | |
cb1410c7 | 1850 | if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) |
a00dd6c0 | 1851 | goto out_mark_dirty; |
420e3646 | 1852 | |
a00dd6c0 | 1853 | /* don't wait for the COMMIT response */ |
420e3646 | 1854 | flags = 0; |
a00dd6c0 JL |
1855 | } |
1856 | ||
420e3646 TM |
1857 | ret = nfs_commit_inode(inode, flags); |
1858 | if (ret >= 0) { | |
1859 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1860 | if (ret < wbc->nr_to_write) | |
1861 | wbc->nr_to_write -= ret; | |
1862 | else | |
1863 | wbc->nr_to_write = 0; | |
1864 | } | |
8fc795f7 | 1865 | return 0; |
420e3646 TM |
1866 | } |
1867 | out_mark_dirty: | |
8fc795f7 TM |
1868 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1869 | return ret; | |
1870 | } | |
89d77c8f | 1871 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1872 | |
acdc53b2 TM |
1873 | /* |
1874 | * flush the inode to disk. | |
1875 | */ | |
1876 | int nfs_wb_all(struct inode *inode) | |
34901f70 | 1877 | { |
f4ce1299 TM |
1878 | int ret; |
1879 | ||
1880 | trace_nfs_writeback_inode_enter(inode); | |
1881 | ||
5bb89b47 | 1882 | ret = filemap_write_and_wait(inode->i_mapping); |
6b196875 CL |
1883 | if (ret) |
1884 | goto out; | |
1885 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
1886 | if (ret < 0) | |
1887 | goto out; | |
1888 | pnfs_sync_inode(inode, true); | |
1889 | ret = 0; | |
34901f70 | 1890 | |
6b196875 | 1891 | out: |
f4ce1299 TM |
1892 | trace_nfs_writeback_inode_exit(inode, ret); |
1893 | return ret; | |
1c75950b | 1894 | } |
ddda8e0a | 1895 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1896 | |
1b3b4a1a TM |
1897 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1898 | { | |
1899 | struct nfs_page *req; | |
1b3b4a1a TM |
1900 | int ret = 0; |
1901 | ||
3e217045 WAA |
1902 | wait_on_page_writeback(page); |
1903 | ||
1904 | /* blocking call to cancel all requests and join to a single (head) | |
1905 | * request */ | |
1906 | req = nfs_lock_and_join_requests(page, false); | |
1907 | ||
1908 | if (IS_ERR(req)) { | |
1909 | ret = PTR_ERR(req); | |
1910 | } else if (req) { | |
1911 | /* all requests from this page have been cancelled by | |
1912 | * nfs_lock_and_join_requests, so just remove the head | |
1913 | * request from the inode / page_private pointer and | |
1914 | * release it */ | |
1915 | nfs_inode_remove_request(req); | |
3e217045 | 1916 | nfs_unlock_and_release_request(req); |
1b3b4a1a | 1917 | } |
3e217045 | 1918 | |
1b3b4a1a TM |
1919 | return ret; |
1920 | } | |
1921 | ||
7f2f12d9 TM |
1922 | /* |
1923 | * Write back all requests on one page - we do this before reading it. | |
1924 | */ | |
1925 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b | 1926 | { |
29418aa4 | 1927 | loff_t range_start = page_file_offset(page); |
1c75950b | 1928 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); |
4d770ccf | 1929 | struct writeback_control wbc = { |
4d770ccf | 1930 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1931 | .nr_to_write = 0, |
4d770ccf TM |
1932 | .range_start = range_start, |
1933 | .range_end = range_end, | |
1934 | }; | |
1935 | int ret; | |
1c75950b | 1936 | |
f4ce1299 TM |
1937 | trace_nfs_writeback_page_enter(inode); |
1938 | ||
0522f6ad | 1939 | for (;;) { |
ba8b06e6 | 1940 | wait_on_page_writeback(page); |
73e3302f TM |
1941 | if (clear_page_dirty_for_io(page)) { |
1942 | ret = nfs_writepage_locked(page, &wbc); | |
1943 | if (ret < 0) | |
1944 | goto out_error; | |
0522f6ad | 1945 | continue; |
7f2f12d9 | 1946 | } |
f4ce1299 | 1947 | ret = 0; |
0522f6ad TM |
1948 | if (!PagePrivate(page)) |
1949 | break; | |
1950 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1951 | if (ret < 0) |
73e3302f | 1952 | goto out_error; |
7f2f12d9 | 1953 | } |
73e3302f | 1954 | out_error: |
f4ce1299 | 1955 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 1956 | return ret; |
1c75950b TM |
1957 | } |
1958 | ||
074cc1de TM |
1959 | #ifdef CONFIG_MIGRATION |
1960 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1961 | struct page *page, enum migrate_mode mode) |
074cc1de | 1962 | { |
2da95652 JL |
1963 | /* |
1964 | * If PagePrivate is set, then the page is currently associated with | |
1965 | * an in-progress read or write request. Don't try to migrate it. | |
1966 | * | |
1967 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1968 | * that we can safely release the inode reference while holding | |
1969 | * the page lock. | |
1970 | */ | |
1971 | if (PagePrivate(page)) | |
1972 | return -EBUSY; | |
074cc1de | 1973 | |
8c209ce7 DH |
1974 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
1975 | return -EBUSY; | |
074cc1de | 1976 | |
a6bc32b8 | 1977 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1978 | } |
1979 | #endif | |
1980 | ||
f7b422b1 | 1981 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1982 | { |
1983 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 1984 | sizeof(struct nfs_pgio_header), |
1da177e4 | 1985 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1986 | NULL); |
1da177e4 LT |
1987 | if (nfs_wdata_cachep == NULL) |
1988 | return -ENOMEM; | |
1989 | ||
93d2341c MD |
1990 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1991 | nfs_wdata_cachep); | |
1da177e4 | 1992 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 1993 | goto out_destroy_write_cache; |
1da177e4 | 1994 | |
0b7c0153 FI |
1995 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1996 | sizeof(struct nfs_commit_data), | |
1997 | 0, SLAB_HWCACHE_ALIGN, | |
1998 | NULL); | |
1999 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 2000 | goto out_destroy_write_mempool; |
0b7c0153 | 2001 | |
93d2341c | 2002 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 2003 | nfs_cdata_cachep); |
1da177e4 | 2004 | if (nfs_commit_mempool == NULL) |
3dd4765f | 2005 | goto out_destroy_commit_cache; |
1da177e4 | 2006 | |
89a09141 PZ |
2007 | /* |
2008 | * NFS congestion size, scale with available memory. | |
2009 | * | |
2010 | * 64MB: 8192k | |
2011 | * 128MB: 11585k | |
2012 | * 256MB: 16384k | |
2013 | * 512MB: 23170k | |
2014 | * 1GB: 32768k | |
2015 | * 2GB: 46340k | |
2016 | * 4GB: 65536k | |
2017 | * 8GB: 92681k | |
2018 | * 16GB: 131072k | |
2019 | * | |
2020 | * This allows larger machines to have larger/more transfers. | |
2021 | * Limit the default to 256M | |
2022 | */ | |
2023 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
2024 | if (nfs_congestion_kb > 256*1024) | |
2025 | nfs_congestion_kb = 256*1024; | |
2026 | ||
1da177e4 | 2027 | return 0; |
3dd4765f JL |
2028 | |
2029 | out_destroy_commit_cache: | |
2030 | kmem_cache_destroy(nfs_cdata_cachep); | |
2031 | out_destroy_write_mempool: | |
2032 | mempool_destroy(nfs_wdata_mempool); | |
2033 | out_destroy_write_cache: | |
2034 | kmem_cache_destroy(nfs_wdata_cachep); | |
2035 | return -ENOMEM; | |
1da177e4 LT |
2036 | } |
2037 | ||
266bee88 | 2038 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
2039 | { |
2040 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 2041 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 2042 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 2043 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
2044 | } |
2045 | ||
4a0de55c | 2046 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
a4cdda59 | 2047 | .rw_mode = FMODE_WRITE, |
4a0de55c AS |
2048 | .rw_alloc_header = nfs_writehdr_alloc, |
2049 | .rw_free_header = nfs_writehdr_free, | |
0eecb214 AS |
2050 | .rw_done = nfs_writeback_done, |
2051 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 2052 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 2053 | }; |