]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
afeacc8c | 23 | #include <linux/export.h> |
af7cf057 TM |
24 | #include <linux/freezer.h> |
25 | #include <linux/wait.h> | |
3fcfab16 | 26 | |
7c0f6ba6 | 27 | #include <linux/uaccess.h> |
1da177e4 LT |
28 | |
29 | #include "delegation.h" | |
49a70f27 | 30 | #include "internal.h" |
91d5b470 | 31 | #include "iostat.h" |
def6ed7e | 32 | #include "nfs4_fs.h" |
074cc1de | 33 | #include "fscache.h" |
94ad1c80 | 34 | #include "pnfs.h" |
1da177e4 | 35 | |
f4ce1299 TM |
36 | #include "nfstrace.h" |
37 | ||
1da177e4 LT |
38 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
39 | ||
40 | #define MIN_POOL_WRITE (32) | |
41 | #define MIN_POOL_COMMIT (4) | |
42 | ||
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 | ||
518662e0 | 63 | struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) |
1da177e4 | 64 | { |
518662e0 | 65 | struct nfs_commit_data *p; |
40859d7e | 66 | |
518662e0 N |
67 | if (never_fail) |
68 | p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); | |
69 | else { | |
70 | /* It is OK to do some reclaim, not no safe to wait | |
71 | * for anything to be returned to the pool. | |
72 | * mempool_alloc() cannot handle that particular combination, | |
73 | * so we need two separate attempts. | |
74 | */ | |
75 | p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); | |
76 | if (!p) | |
77 | p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | | |
78 | __GFP_NOWARN | __GFP_NORETRY); | |
79 | if (!p) | |
80 | return NULL; | |
1da177e4 | 81 | } |
518662e0 N |
82 | |
83 | memset(p, 0, sizeof(*p)); | |
84 | INIT_LIST_HEAD(&p->pages); | |
1da177e4 LT |
85 | return p; |
86 | } | |
e0c2b380 | 87 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 88 | |
0b7c0153 | 89 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
90 | { |
91 | mempool_free(p, nfs_commit_mempool); | |
92 | } | |
e0c2b380 | 93 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 94 | |
1e7f3a48 | 95 | static struct nfs_pgio_header *nfs_writehdr_alloc(void) |
3feb2d49 | 96 | { |
1e7f3a48 | 97 | struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
cd841605 | 98 | |
fbe77c30 BC |
99 | if (p) { |
100 | memset(p, 0, sizeof(*p)); | |
101 | p->rw_mode = FMODE_WRITE; | |
102 | } | |
3feb2d49 TM |
103 | return p; |
104 | } | |
6c75dc0d | 105 | |
1e7f3a48 | 106 | static void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 107 | { |
1e7f3a48 | 108 | mempool_free(hdr, nfs_wdata_mempool); |
3feb2d49 | 109 | } |
1da177e4 | 110 | |
7b159fc1 TM |
111 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
112 | { | |
113 | ctx->error = error; | |
114 | smp_wmb(); | |
115 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
116 | } | |
117 | ||
84d3a9a9 WAA |
118 | /* |
119 | * nfs_page_find_head_request_locked - find head request associated with @page | |
120 | * | |
121 | * must be called while holding the inode lock. | |
122 | * | |
123 | * returns matching head request with reference held, or NULL if not found. | |
124 | */ | |
29418aa4 | 125 | static struct nfs_page * |
84d3a9a9 | 126 | nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) |
277459d2 TM |
127 | { |
128 | struct nfs_page *req = NULL; | |
129 | ||
29418aa4 | 130 | if (PagePrivate(page)) |
277459d2 | 131 | req = (struct nfs_page *)page_private(page); |
02d1426c WAA |
132 | else if (unlikely(PageSwapCache(page))) |
133 | req = nfs_page_search_commits_for_head_request_locked(nfsi, | |
134 | page); | |
29418aa4 | 135 | |
84d3a9a9 WAA |
136 | if (req) { |
137 | WARN_ON_ONCE(req->wb_head != req); | |
29418aa4 | 138 | kref_get(&req->wb_kref); |
84d3a9a9 | 139 | } |
29418aa4 | 140 | |
277459d2 TM |
141 | return req; |
142 | } | |
143 | ||
84d3a9a9 WAA |
144 | /* |
145 | * nfs_page_find_head_request - find head request associated with @page | |
146 | * | |
147 | * returns matching head request with reference held, or NULL if not found. | |
148 | */ | |
149 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 150 | { |
d56b4ddf | 151 | struct inode *inode = page_file_mapping(page)->host; |
277459d2 | 152 | struct nfs_page *req = NULL; |
277459d2 | 153 | |
587142f8 | 154 | spin_lock(&inode->i_lock); |
84d3a9a9 | 155 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
587142f8 | 156 | spin_unlock(&inode->i_lock); |
277459d2 TM |
157 | return req; |
158 | } | |
159 | ||
1da177e4 LT |
160 | /* Adjust the file length if we're writing beyond the end */ |
161 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
162 | { | |
d56b4ddf | 163 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
164 | loff_t end, i_size; |
165 | pgoff_t end_index; | |
1da177e4 | 166 | |
a3d01454 TM |
167 | spin_lock(&inode->i_lock); |
168 | i_size = i_size_read(inode); | |
09cbfeaf | 169 | end_index = (i_size - 1) >> PAGE_SHIFT; |
8cd79788 | 170 | if (i_size > 0 && page_index(page) < end_index) |
a3d01454 | 171 | goto out; |
d56b4ddf | 172 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 173 | if (i_size >= end) |
a3d01454 | 174 | goto out; |
1da177e4 | 175 | i_size_write(inode, end); |
a3d01454 TM |
176 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
177 | out: | |
178 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
179 | } |
180 | ||
a301b777 TM |
181 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
182 | static void nfs_set_pageerror(struct page *page) | |
183 | { | |
d56b4ddf | 184 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
185 | } |
186 | ||
d72ddcba WAA |
187 | /* |
188 | * nfs_page_group_search_locked | |
189 | * @head - head request of page group | |
190 | * @page_offset - offset into page | |
191 | * | |
192 | * Search page group with head @head to find a request that contains the | |
193 | * page offset @page_offset. | |
194 | * | |
195 | * Returns a pointer to the first matching nfs request, or NULL if no | |
196 | * match is found. | |
197 | * | |
198 | * Must be called with the page group lock held | |
199 | */ | |
200 | static struct nfs_page * | |
201 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
202 | { | |
203 | struct nfs_page *req; | |
204 | ||
205 | WARN_ON_ONCE(head != head->wb_head); | |
206 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); | |
207 | ||
208 | req = head; | |
209 | do { | |
210 | if (page_offset >= req->wb_pgbase && | |
211 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
212 | return req; | |
213 | ||
214 | req = req->wb_this_page; | |
215 | } while (req != head); | |
216 | ||
217 | return NULL; | |
218 | } | |
219 | ||
220 | /* | |
221 | * nfs_page_group_covers_page | |
222 | * @head - head request of page group | |
223 | * | |
224 | * Return true if the page group with head @head covers the whole page, | |
225 | * returns false otherwise | |
226 | */ | |
227 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
228 | { | |
229 | struct nfs_page *tmp; | |
230 | unsigned int pos = 0; | |
231 | unsigned int len = nfs_page_length(req->wb_page); | |
232 | ||
fd2f3a06 | 233 | nfs_page_group_lock(req, false); |
d72ddcba WAA |
234 | |
235 | do { | |
236 | tmp = nfs_page_group_search_locked(req->wb_head, pos); | |
237 | if (tmp) { | |
238 | /* no way this should happen */ | |
239 | WARN_ON_ONCE(tmp->wb_pgbase != pos); | |
240 | pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); | |
241 | } | |
242 | } while (tmp && pos < len); | |
243 | ||
244 | nfs_page_group_unlock(req); | |
245 | WARN_ON_ONCE(pos > len); | |
246 | return pos == len; | |
247 | } | |
248 | ||
1da177e4 LT |
249 | /* We can set the PG_uptodate flag if we see that a write request |
250 | * covers the full page. | |
251 | */ | |
d72ddcba | 252 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 253 | { |
d72ddcba | 254 | if (PageUptodate(req->wb_page)) |
1da177e4 | 255 | return; |
d72ddcba | 256 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 257 | return; |
d72ddcba | 258 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
259 | } |
260 | ||
1da177e4 LT |
261 | static int wb_priority(struct writeback_control *wbc) |
262 | { | |
e87b4c7a | 263 | int ret = 0; |
cca588d6 | 264 | |
e87b4c7a N |
265 | if (wbc->sync_mode == WB_SYNC_ALL) |
266 | ret = FLUSH_COND_STABLE; | |
e87b4c7a | 267 | return ret; |
1da177e4 LT |
268 | } |
269 | ||
89a09141 PZ |
270 | /* |
271 | * NFS congestion control | |
272 | */ | |
273 | ||
274 | int nfs_congestion_kb; | |
275 | ||
276 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
277 | #define NFS_CONGESTION_OFF_THRESH \ | |
278 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
279 | ||
deed85e7 | 280 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 281 | { |
deed85e7 | 282 | struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); |
5a6d41b3 TM |
283 | int ret = test_set_page_writeback(page); |
284 | ||
deed85e7 | 285 | WARN_ON_ONCE(ret != 0); |
89a09141 | 286 | |
deed85e7 TM |
287 | if (atomic_long_inc_return(&nfss->writeback) > |
288 | NFS_CONGESTION_ON_THRESH) { | |
289 | set_bdi_congested(&nfss->backing_dev_info, | |
290 | BLK_RW_ASYNC); | |
89a09141 PZ |
291 | } |
292 | } | |
293 | ||
20633f04 | 294 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 295 | { |
20633f04 | 296 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 PZ |
297 | struct nfs_server *nfss = NFS_SERVER(inode); |
298 | ||
20633f04 WAA |
299 | if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) |
300 | return; | |
301 | ||
302 | end_page_writeback(req->wb_page); | |
c4dc4bee | 303 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 304 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
305 | } |
306 | ||
d4581383 WAA |
307 | |
308 | /* nfs_page_group_clear_bits | |
309 | * @req - an nfs request | |
310 | * clears all page group related bits from @req | |
311 | */ | |
312 | static void | |
313 | nfs_page_group_clear_bits(struct nfs_page *req) | |
314 | { | |
315 | clear_bit(PG_TEARDOWN, &req->wb_flags); | |
316 | clear_bit(PG_UNLOCKPAGE, &req->wb_flags); | |
317 | clear_bit(PG_UPTODATE, &req->wb_flags); | |
318 | clear_bit(PG_WB_END, &req->wb_flags); | |
319 | clear_bit(PG_REMOVE, &req->wb_flags); | |
320 | } | |
321 | ||
322 | ||
323 | /* | |
324 | * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req | |
325 | * | |
326 | * this is a helper function for nfs_lock_and_join_requests | |
327 | * | |
328 | * @inode - inode associated with request page group, must be holding inode lock | |
329 | * @head - head request of page group, must be holding head lock | |
330 | * @req - request that couldn't lock and needs to wait on the req bit lock | |
331 | * @nonblock - if true, don't actually wait | |
332 | * | |
333 | * NOTE: this must be called holding page_group bit lock and inode spin lock | |
334 | * and BOTH will be released before returning. | |
335 | * | |
336 | * returns 0 on success, < 0 on error. | |
337 | */ | |
338 | static int | |
339 | nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, | |
340 | struct nfs_page *req, bool nonblock) | |
341 | __releases(&inode->i_lock) | |
342 | { | |
343 | struct nfs_page *tmp; | |
344 | int ret; | |
345 | ||
346 | /* relinquish all the locks successfully grabbed this run */ | |
347 | for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) | |
348 | nfs_unlock_request(tmp); | |
349 | ||
350 | WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); | |
351 | ||
352 | /* grab a ref on the request that will be waited on */ | |
353 | kref_get(&req->wb_kref); | |
354 | ||
355 | nfs_page_group_unlock(head); | |
356 | spin_unlock(&inode->i_lock); | |
357 | ||
358 | /* release ref from nfs_page_find_head_request_locked */ | |
359 | nfs_release_request(head); | |
360 | ||
361 | if (!nonblock) | |
362 | ret = nfs_wait_on_request(req); | |
363 | else | |
364 | ret = -EAGAIN; | |
365 | nfs_release_request(req); | |
366 | ||
367 | return ret; | |
368 | } | |
369 | ||
370 | /* | |
371 | * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests | |
372 | * | |
373 | * @destroy_list - request list (using wb_this_page) terminated by @old_head | |
374 | * @old_head - the old head of the list | |
375 | * | |
376 | * All subrequests must be locked and removed from all lists, so at this point | |
377 | * they are only "active" in this function, and possibly in nfs_wait_on_request | |
378 | * with a reference held by some other context. | |
379 | */ | |
380 | static void | |
381 | nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, | |
382 | struct nfs_page *old_head) | |
383 | { | |
384 | while (destroy_list) { | |
385 | struct nfs_page *subreq = destroy_list; | |
386 | ||
387 | destroy_list = (subreq->wb_this_page == old_head) ? | |
388 | NULL : subreq->wb_this_page; | |
389 | ||
390 | WARN_ON_ONCE(old_head != subreq->wb_head); | |
391 | ||
392 | /* make sure old group is not used */ | |
393 | subreq->wb_head = subreq; | |
394 | subreq->wb_this_page = subreq; | |
395 | ||
d4581383 WAA |
396 | /* subreq is now totally disconnected from page group or any |
397 | * write / commit lists. last chance to wake any waiters */ | |
398 | nfs_unlock_request(subreq); | |
399 | ||
400 | if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { | |
401 | /* release ref on old head request */ | |
402 | nfs_release_request(old_head); | |
403 | ||
404 | nfs_page_group_clear_bits(subreq); | |
405 | ||
406 | /* release the PG_INODE_REF reference */ | |
407 | if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) | |
408 | nfs_release_request(subreq); | |
409 | else | |
410 | WARN_ON_ONCE(1); | |
411 | } else { | |
412 | WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); | |
413 | /* zombie requests have already released the last | |
414 | * reference and were waiting on the rest of the | |
415 | * group to complete. Since it's no longer part of a | |
416 | * group, simply free the request */ | |
417 | nfs_page_group_clear_bits(subreq); | |
418 | nfs_free_request(subreq); | |
419 | } | |
420 | } | |
421 | } | |
422 | ||
423 | /* | |
424 | * nfs_lock_and_join_requests - join all subreqs to the head req and return | |
425 | * a locked reference, cancelling any pending | |
426 | * operations for this page. | |
427 | * | |
428 | * @page - the page used to lookup the "page group" of nfs_page structures | |
429 | * @nonblock - if true, don't block waiting for request locks | |
430 | * | |
431 | * This function joins all sub requests to the head request by first | |
432 | * locking all requests in the group, cancelling any pending operations | |
433 | * and finally updating the head request to cover the whole range covered by | |
434 | * the (former) group. All subrequests are removed from any write or commit | |
435 | * lists, unlinked from the group and destroyed. | |
436 | * | |
437 | * Returns a locked, referenced pointer to the head request - which after | |
438 | * this call is guaranteed to be the only request associated with the page. | |
439 | * Returns NULL if no requests are found for @page, or a ERR_PTR if an | |
440 | * error was encountered. | |
441 | */ | |
442 | static struct nfs_page * | |
443 | nfs_lock_and_join_requests(struct page *page, bool nonblock) | |
e261f51f | 444 | { |
d56b4ddf | 445 | struct inode *inode = page_file_mapping(page)->host; |
d4581383 WAA |
446 | struct nfs_page *head, *subreq; |
447 | struct nfs_page *destroy_list = NULL; | |
448 | unsigned int total_bytes; | |
e261f51f TM |
449 | int ret; |
450 | ||
d4581383 WAA |
451 | try_again: |
452 | total_bytes = 0; | |
453 | ||
454 | WARN_ON_ONCE(destroy_list); | |
455 | ||
587142f8 | 456 | spin_lock(&inode->i_lock); |
d4581383 WAA |
457 | |
458 | /* | |
459 | * A reference is taken only on the head request which acts as a | |
460 | * reference to the whole page group - the group will not be destroyed | |
461 | * until the head reference is released. | |
462 | */ | |
463 | head = nfs_page_find_head_request_locked(NFS_I(inode), page); | |
464 | ||
465 | if (!head) { | |
587142f8 | 466 | spin_unlock(&inode->i_lock); |
d4581383 WAA |
467 | return NULL; |
468 | } | |
469 | ||
7c3af975 WAA |
470 | /* holding inode lock, so always make a non-blocking call to try the |
471 | * page group lock */ | |
fd2f3a06 | 472 | ret = nfs_page_group_lock(head, true); |
94970014 WAA |
473 | if (ret < 0) { |
474 | spin_unlock(&inode->i_lock); | |
7c3af975 WAA |
475 | |
476 | if (!nonblock && ret == -EAGAIN) { | |
477 | nfs_page_group_lock_wait(head); | |
478 | nfs_release_request(head); | |
479 | goto try_again; | |
480 | } | |
481 | ||
94970014 | 482 | nfs_release_request(head); |
e7029206 | 483 | return ERR_PTR(ret); |
94970014 | 484 | } |
7c3af975 WAA |
485 | |
486 | /* lock each request in the page group */ | |
d4581383 WAA |
487 | subreq = head; |
488 | do { | |
489 | /* | |
490 | * Subrequests are always contiguous, non overlapping | |
309a1d65 | 491 | * and in order - but may be repeated (mirrored writes). |
d4581383 | 492 | */ |
309a1d65 WAA |
493 | if (subreq->wb_offset == (head->wb_offset + total_bytes)) { |
494 | /* keep track of how many bytes this group covers */ | |
495 | total_bytes += subreq->wb_bytes; | |
496 | } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || | |
497 | ((subreq->wb_offset + subreq->wb_bytes) > | |
498 | (head->wb_offset + total_bytes)))) { | |
499 | nfs_page_group_unlock(head); | |
500 | spin_unlock(&inode->i_lock); | |
501 | return ERR_PTR(-EIO); | |
502 | } | |
d4581383 WAA |
503 | |
504 | if (!nfs_lock_request(subreq)) { | |
505 | /* releases page group bit lock and | |
506 | * inode spin lock and all references */ | |
507 | ret = nfs_unroll_locks_and_wait(inode, head, | |
508 | subreq, nonblock); | |
509 | ||
510 | if (ret == 0) | |
511 | goto try_again; | |
512 | ||
074cc1de | 513 | return ERR_PTR(ret); |
d4581383 WAA |
514 | } |
515 | ||
516 | subreq = subreq->wb_this_page; | |
517 | } while (subreq != head); | |
518 | ||
519 | /* Now that all requests are locked, make sure they aren't on any list. | |
520 | * Commit list removal accounting is done after locks are dropped */ | |
521 | subreq = head; | |
522 | do { | |
411a99ad | 523 | nfs_clear_request_commit(subreq); |
d4581383 WAA |
524 | subreq = subreq->wb_this_page; |
525 | } while (subreq != head); | |
526 | ||
527 | /* unlink subrequests from head, destroy them later */ | |
528 | if (head->wb_this_page != head) { | |
529 | /* destroy list will be terminated by head */ | |
530 | destroy_list = head->wb_this_page; | |
531 | head->wb_this_page = head; | |
532 | ||
533 | /* change head request to cover whole range that | |
534 | * the former page group covered */ | |
535 | head->wb_bytes = total_bytes; | |
e261f51f | 536 | } |
d4581383 WAA |
537 | |
538 | /* | |
539 | * prepare head request to be added to new pgio descriptor | |
540 | */ | |
541 | nfs_page_group_clear_bits(head); | |
542 | ||
543 | /* | |
544 | * some part of the group was still on the inode list - otherwise | |
545 | * the group wouldn't be involved in async write. | |
546 | * grab a reference for the head request, iff it needs one. | |
547 | */ | |
548 | if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) | |
549 | kref_get(&head->wb_kref); | |
550 | ||
551 | nfs_page_group_unlock(head); | |
552 | ||
411a99ad | 553 | /* drop lock to clean uprequests on destroy list */ |
587142f8 | 554 | spin_unlock(&inode->i_lock); |
d4581383 WAA |
555 | |
556 | nfs_destroy_unlinked_subrequests(destroy_list, head); | |
557 | ||
d4581383 WAA |
558 | /* still holds ref on head from nfs_page_find_head_request_locked |
559 | * and still has lock on head from lock loop */ | |
560 | return head; | |
074cc1de TM |
561 | } |
562 | ||
0bcbf039 PT |
563 | static void nfs_write_error_remove_page(struct nfs_page *req) |
564 | { | |
565 | nfs_unlock_request(req); | |
566 | nfs_end_page_writeback(req); | |
0bcbf039 PT |
567 | generic_error_remove_page(page_file_mapping(req->wb_page), |
568 | req->wb_page); | |
1f84ccdf | 569 | nfs_release_request(req); |
0bcbf039 PT |
570 | } |
571 | ||
a6598813 TM |
572 | static bool |
573 | nfs_error_is_fatal_on_server(int err) | |
574 | { | |
575 | switch (err) { | |
576 | case 0: | |
577 | case -ERESTARTSYS: | |
578 | case -EINTR: | |
579 | return false; | |
580 | } | |
581 | return nfs_error_is_fatal(err); | |
582 | } | |
583 | ||
074cc1de TM |
584 | /* |
585 | * Find an associated nfs write request, and prepare to flush it out | |
586 | * May return an error if the user signalled nfs_wait_on_request(). | |
587 | */ | |
588 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
d6c843b9 PT |
589 | struct page *page, bool nonblock, |
590 | bool launder) | |
074cc1de TM |
591 | { |
592 | struct nfs_page *req; | |
593 | int ret = 0; | |
594 | ||
d4581383 | 595 | req = nfs_lock_and_join_requests(page, nonblock); |
074cc1de TM |
596 | if (!req) |
597 | goto out; | |
598 | ret = PTR_ERR(req); | |
599 | if (IS_ERR(req)) | |
600 | goto out; | |
601 | ||
deed85e7 TM |
602 | nfs_set_page_writeback(page); |
603 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 604 | |
deed85e7 | 605 | ret = 0; |
a6598813 TM |
606 | /* If there is a fatal error that covers this write, just exit */ |
607 | if (nfs_error_is_fatal_on_server(req->wb_context->error)) | |
608 | goto out_launder; | |
609 | ||
f8512ad0 | 610 | if (!nfs_pageio_add_request(pgio, req)) { |
074cc1de | 611 | ret = pgio->pg_error; |
0bcbf039 | 612 | /* |
d6c843b9 PT |
613 | * Remove the problematic req upon fatal errors |
614 | * in launder case, while other dirty pages can | |
615 | * still be around until they get flushed. | |
0bcbf039 PT |
616 | */ |
617 | if (nfs_error_is_fatal(ret)) { | |
618 | nfs_context_set_write_error(req->wb_context, ret); | |
a6598813 TM |
619 | if (launder) |
620 | goto out_launder; | |
0bcbf039 | 621 | } |
d6c843b9 PT |
622 | nfs_redirty_request(req); |
623 | ret = -EAGAIN; | |
40f90271 TM |
624 | } else |
625 | nfs_add_stats(page_file_mapping(page)->host, | |
626 | NFSIOS_WRITEPAGES, 1); | |
074cc1de TM |
627 | out: |
628 | return ret; | |
a6598813 TM |
629 | out_launder: |
630 | nfs_write_error_remove_page(req); | |
631 | return ret; | |
e261f51f TM |
632 | } |
633 | ||
d6c843b9 PT |
634 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, |
635 | struct nfs_pageio_descriptor *pgio, bool launder) | |
1da177e4 | 636 | { |
cfb506e1 | 637 | int ret; |
1da177e4 | 638 | |
8cd79788 | 639 | nfs_pageio_cond_complete(pgio, page_index(page)); |
d6c843b9 PT |
640 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE, |
641 | launder); | |
cfb506e1 TM |
642 | if (ret == -EAGAIN) { |
643 | redirty_page_for_writepage(wbc, page); | |
644 | ret = 0; | |
645 | } | |
646 | return ret; | |
f758c885 | 647 | } |
7fe7f848 | 648 | |
f758c885 TM |
649 | /* |
650 | * Write an mmapped page to the server. | |
651 | */ | |
d6c843b9 PT |
652 | static int nfs_writepage_locked(struct page *page, |
653 | struct writeback_control *wbc, | |
654 | bool launder) | |
f758c885 TM |
655 | { |
656 | struct nfs_pageio_descriptor pgio; | |
40f90271 | 657 | struct inode *inode = page_file_mapping(page)->host; |
f758c885 | 658 | int err; |
49a70f27 | 659 | |
40f90271 | 660 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
811ed92e | 661 | nfs_pageio_init_write(&pgio, inode, 0, |
a20c93e3 | 662 | false, &nfs_async_write_completion_ops); |
d6c843b9 | 663 | err = nfs_do_writepage(page, wbc, &pgio, launder); |
f758c885 TM |
664 | nfs_pageio_complete(&pgio); |
665 | if (err < 0) | |
666 | return err; | |
667 | if (pgio.pg_error < 0) | |
668 | return pgio.pg_error; | |
669 | return 0; | |
4d770ccf TM |
670 | } |
671 | ||
672 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
673 | { | |
f758c885 | 674 | int ret; |
4d770ccf | 675 | |
d6c843b9 | 676 | ret = nfs_writepage_locked(page, wbc, false); |
1da177e4 | 677 | unlock_page(page); |
f758c885 TM |
678 | return ret; |
679 | } | |
680 | ||
681 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
682 | { | |
683 | int ret; | |
684 | ||
d6c843b9 | 685 | ret = nfs_do_writepage(page, wbc, data, false); |
f758c885 TM |
686 | unlock_page(page); |
687 | return ret; | |
1da177e4 LT |
688 | } |
689 | ||
1da177e4 LT |
690 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
691 | { | |
1da177e4 | 692 | struct inode *inode = mapping->host; |
c63c7b05 | 693 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
694 | int err; |
695 | ||
91d5b470 CL |
696 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
697 | ||
a20c93e3 CH |
698 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
699 | &nfs_async_write_completion_ops); | |
f758c885 | 700 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 701 | nfs_pageio_complete(&pgio); |
72cb77f4 | 702 | |
f758c885 | 703 | if (err < 0) |
72cb77f4 TM |
704 | goto out_err; |
705 | err = pgio.pg_error; | |
706 | if (err < 0) | |
707 | goto out_err; | |
c63c7b05 | 708 | return 0; |
72cb77f4 TM |
709 | out_err: |
710 | return err; | |
1da177e4 LT |
711 | } |
712 | ||
713 | /* | |
714 | * Insert a write request into an inode | |
715 | */ | |
d6d6dc7c | 716 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
717 | { |
718 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 | 719 | |
2bfc6e56 WAA |
720 | WARN_ON_ONCE(req->wb_this_page != req); |
721 | ||
e7d39069 | 722 | /* Lock the request! */ |
7ad84aa9 | 723 | nfs_lock_request(req); |
e7d39069 TM |
724 | |
725 | spin_lock(&inode->i_lock); | |
cb1410c7 WAA |
726 | if (!nfsi->nrequests && |
727 | NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) | |
a9a4a87a | 728 | inode->i_version++; |
29418aa4 MG |
729 | /* |
730 | * Swap-space should not get truncated. Hence no need to plug the race | |
731 | * with invalidate/truncate. | |
732 | */ | |
733 | if (likely(!PageSwapCache(req->wb_page))) { | |
734 | set_bit(PG_MAPPED, &req->wb_flags); | |
735 | SetPagePrivate(req->wb_page); | |
736 | set_page_private(req->wb_page, (unsigned long)req); | |
737 | } | |
cb1410c7 | 738 | nfsi->nrequests++; |
17089a29 | 739 | /* this a head request for a page group - mark it as having an |
cb1410c7 WAA |
740 | * extra reference so sub groups can follow suit. |
741 | * This flag also informs pgio layer when to bump nrequests when | |
742 | * adding subrequests. */ | |
17089a29 | 743 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); |
c03b4024 | 744 | kref_get(&req->wb_kref); |
e7d39069 | 745 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
746 | } |
747 | ||
748 | /* | |
89a09141 | 749 | * Remove a write request from an inode |
1da177e4 LT |
750 | */ |
751 | static void nfs_inode_remove_request(struct nfs_page *req) | |
752 | { | |
2b0143b5 | 753 | struct inode *inode = d_inode(req->wb_context->dentry); |
1da177e4 | 754 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 755 | struct nfs_page *head; |
1da177e4 | 756 | |
20633f04 WAA |
757 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
758 | head = req->wb_head; | |
759 | ||
760 | spin_lock(&inode->i_lock); | |
67911c8f | 761 | if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { |
20633f04 WAA |
762 | set_page_private(head->wb_page, 0); |
763 | ClearPagePrivate(head->wb_page); | |
764 | clear_bit(PG_MAPPED, &head->wb_flags); | |
765 | } | |
cb1410c7 WAA |
766 | nfsi->nrequests--; |
767 | spin_unlock(&inode->i_lock); | |
768 | } else { | |
769 | spin_lock(&inode->i_lock); | |
770 | nfsi->nrequests--; | |
20633f04 | 771 | spin_unlock(&inode->i_lock); |
29418aa4 | 772 | } |
17089a29 WAA |
773 | |
774 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) | |
775 | nfs_release_request(req); | |
1da177e4 LT |
776 | } |
777 | ||
61822ab5 | 778 | static void |
6d884e8f | 779 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 780 | { |
67911c8f AS |
781 | if (req->wb_page) |
782 | __set_page_dirty_nobuffers(req->wb_page); | |
61822ab5 TM |
783 | } |
784 | ||
3a3908c8 TM |
785 | /* |
786 | * nfs_page_search_commits_for_head_request_locked | |
787 | * | |
788 | * Search through commit lists on @inode for the head request for @page. | |
789 | * Must be called while holding the inode (which is cinfo) lock. | |
790 | * | |
791 | * Returns the head request if found, or NULL if not found. | |
792 | */ | |
793 | static struct nfs_page * | |
794 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
795 | struct page *page) | |
796 | { | |
797 | struct nfs_page *freq, *t; | |
798 | struct nfs_commit_info cinfo; | |
799 | struct inode *inode = &nfsi->vfs_inode; | |
800 | ||
801 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
802 | ||
803 | /* search through pnfs commit lists */ | |
804 | freq = pnfs_search_commit_reqs(inode, &cinfo, page); | |
805 | if (freq) | |
806 | return freq->wb_head; | |
807 | ||
808 | /* Linearly search the commit list for the correct request */ | |
809 | list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { | |
810 | if (freq->wb_page == page) | |
811 | return freq->wb_head; | |
812 | } | |
813 | ||
814 | return NULL; | |
815 | } | |
816 | ||
86d80f97 TM |
817 | /** |
818 | * nfs_request_add_commit_list_locked - add request to a commit list | |
819 | * @req: pointer to a struct nfs_page | |
820 | * @dst: commit list head | |
821 | * @cinfo: holds list lock and accounting info | |
822 | * | |
823 | * This sets the PG_CLEAN bit, updates the cinfo count of | |
824 | * number of outstanding requests requiring a commit as well as | |
825 | * the MM page stats. | |
826 | * | |
fe238e60 | 827 | * The caller must hold cinfo->inode->i_lock, and the nfs_page lock. |
86d80f97 TM |
828 | */ |
829 | void | |
830 | nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, | |
831 | struct nfs_commit_info *cinfo) | |
832 | { | |
833 | set_bit(PG_CLEAN, &req->wb_flags); | |
834 | nfs_list_add_request(req, dst); | |
835 | cinfo->mds->ncommit++; | |
836 | } | |
837 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); | |
838 | ||
8dd37758 TM |
839 | /** |
840 | * nfs_request_add_commit_list - add request to a commit list | |
841 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
842 | * @dst: commit list head |
843 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 844 | * |
ea2cf228 | 845 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
846 | * number of outstanding requests requiring a commit as well as |
847 | * the MM page stats. | |
848 | * | |
ea2cf228 | 849 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 850 | * holding the nfs_page lock. |
1da177e4 | 851 | */ |
8dd37758 | 852 | void |
6272dcc6 | 853 | nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) |
1da177e4 | 854 | { |
fe238e60 | 855 | spin_lock(&cinfo->inode->i_lock); |
6272dcc6 | 856 | nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); |
fe238e60 | 857 | spin_unlock(&cinfo->inode->i_lock); |
67911c8f AS |
858 | if (req->wb_page) |
859 | nfs_mark_page_unstable(req->wb_page, cinfo); | |
1da177e4 | 860 | } |
8dd37758 TM |
861 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
862 | ||
863 | /** | |
864 | * nfs_request_remove_commit_list - Remove request from a commit list | |
865 | * @req: pointer to a nfs_page | |
ea2cf228 | 866 | * @cinfo: holds list lock and accounting info |
8dd37758 | 867 | * |
ea2cf228 | 868 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
869 | * number of outstanding requests requiring a commit |
870 | * It does not update the MM page stats. | |
871 | * | |
ea2cf228 | 872 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
873 | */ |
874 | void | |
ea2cf228 FI |
875 | nfs_request_remove_commit_list(struct nfs_page *req, |
876 | struct nfs_commit_info *cinfo) | |
8dd37758 | 877 | { |
8dd37758 TM |
878 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
879 | return; | |
880 | nfs_list_remove_request(req); | |
ea2cf228 | 881 | cinfo->mds->ncommit--; |
8dd37758 TM |
882 | } |
883 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
884 | ||
ea2cf228 FI |
885 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
886 | struct inode *inode) | |
887 | { | |
fe238e60 | 888 | cinfo->inode = inode; |
ea2cf228 FI |
889 | cinfo->mds = &NFS_I(inode)->commit_info; |
890 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 891 | cinfo->dreq = NULL; |
f453a54a | 892 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
893 | } |
894 | ||
895 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
896 | struct inode *inode, | |
897 | struct nfs_direct_req *dreq) | |
898 | { | |
1763da12 FI |
899 | if (dreq) |
900 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
901 | else | |
902 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
903 | } |
904 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
905 | |
906 | /* | |
907 | * Add a request to the inode's commit list. | |
908 | */ | |
1763da12 | 909 | void |
ea2cf228 | 910 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
b57ff130 | 911 | struct nfs_commit_info *cinfo, u32 ds_commit_idx) |
8dd37758 | 912 | { |
b57ff130 | 913 | if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) |
8dd37758 | 914 | return; |
6272dcc6 | 915 | nfs_request_add_commit_list(req, cinfo); |
8dd37758 | 916 | } |
8e821cad | 917 | |
d6d6dc7c FI |
918 | static void |
919 | nfs_clear_page_commit(struct page *page) | |
920 | { | |
11fb9989 | 921 | dec_node_page_state(page, NR_UNSTABLE_NFS); |
93f78d88 TH |
922 | dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, |
923 | WB_RECLAIMABLE); | |
d6d6dc7c FI |
924 | } |
925 | ||
411a99ad | 926 | /* Called holding inode (/cinfo) lock */ |
8dd37758 | 927 | static void |
e468bae9 TM |
928 | nfs_clear_request_commit(struct nfs_page *req) |
929 | { | |
8dd37758 | 930 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
2b0143b5 | 931 | struct inode *inode = d_inode(req->wb_context->dentry); |
ea2cf228 | 932 | struct nfs_commit_info cinfo; |
e468bae9 | 933 | |
ea2cf228 FI |
934 | nfs_init_cinfo_from_inode(&cinfo, inode); |
935 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
ea2cf228 | 936 | nfs_request_remove_commit_list(req, &cinfo); |
8dd37758 | 937 | } |
d6d6dc7c | 938 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 939 | } |
e468bae9 TM |
940 | } |
941 | ||
d45f60c6 | 942 | int nfs_write_need_commit(struct nfs_pgio_header *hdr) |
8e821cad | 943 | { |
c65e6254 | 944 | if (hdr->verf.committed == NFS_DATA_SYNC) |
d45f60c6 | 945 | return hdr->lseg == NULL; |
c65e6254 | 946 | return hdr->verf.committed != NFS_FILE_SYNC; |
8e821cad TM |
947 | } |
948 | ||
061ae2ed | 949 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 950 | { |
ea2cf228 | 951 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
952 | unsigned long bytes = 0; |
953 | ||
954 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
955 | goto out; | |
ea2cf228 | 956 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
957 | while (!list_empty(&hdr->pages)) { |
958 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
959 | |
960 | bytes += req->wb_bytes; | |
961 | nfs_list_remove_request(req); | |
962 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
963 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 964 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
965 | nfs_context_set_write_error(req->wb_context, hdr->error); |
966 | goto remove_req; | |
967 | } | |
c65e6254 | 968 | if (nfs_write_need_commit(hdr)) { |
f79d06f5 | 969 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
b57ff130 | 970 | nfs_mark_request_commit(req, hdr->lseg, &cinfo, |
a7d42ddb | 971 | hdr->pgio_mirror_idx); |
6c75dc0d FI |
972 | goto next; |
973 | } | |
974 | remove_req: | |
975 | nfs_inode_remove_request(req); | |
976 | next: | |
1d1afcbc | 977 | nfs_unlock_request(req); |
20633f04 | 978 | nfs_end_page_writeback(req); |
3aff4ebb | 979 | nfs_release_request(req); |
6c75dc0d FI |
980 | } |
981 | out: | |
982 | hdr->release(hdr); | |
8e821cad | 983 | } |
1da177e4 | 984 | |
ce59515c | 985 | unsigned long |
ea2cf228 | 986 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 987 | { |
ea2cf228 | 988 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
989 | } |
990 | ||
fe238e60 | 991 | /* cinfo->inode->i_lock held by caller */ |
1763da12 | 992 | int |
ea2cf228 FI |
993 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
994 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
995 | { |
996 | struct nfs_page *req, *tmp; | |
997 | int ret = 0; | |
998 | ||
999 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
1000 | if (!nfs_lock_request(req)) |
1001 | continue; | |
7ad84aa9 | 1002 | kref_get(&req->wb_kref); |
fe238e60 | 1003 | if (cond_resched_lock(&cinfo->inode->i_lock)) |
3b3be88d | 1004 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 1005 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
1006 | nfs_list_add_request(req, dst); |
1007 | ret++; | |
1763da12 | 1008 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 1009 | break; |
d6d6dc7c FI |
1010 | } |
1011 | return ret; | |
fb8a1f11 TM |
1012 | } |
1013 | ||
1da177e4 LT |
1014 | /* |
1015 | * nfs_scan_commit - Scan an inode for commit requests | |
1016 | * @inode: NFS inode to scan | |
ea2cf228 FI |
1017 | * @dst: mds destination list |
1018 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
1019 | * |
1020 | * Moves requests from the inode's 'commit' request list. | |
1021 | * The requests are *not* checked to ensure that they form a contiguous set. | |
1022 | */ | |
1763da12 | 1023 | int |
ea2cf228 FI |
1024 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
1025 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1026 | { |
d6d6dc7c | 1027 | int ret = 0; |
fb8a1f11 | 1028 | |
fe238e60 | 1029 | spin_lock(&cinfo->inode->i_lock); |
ea2cf228 | 1030 | if (cinfo->mds->ncommit > 0) { |
8dd37758 | 1031 | const int max = INT_MAX; |
d6d6dc7c | 1032 | |
ea2cf228 FI |
1033 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
1034 | cinfo, max); | |
1035 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 1036 | } |
fe238e60 | 1037 | spin_unlock(&cinfo->inode->i_lock); |
ff778d02 | 1038 | return ret; |
1da177e4 | 1039 | } |
d6d6dc7c | 1040 | |
1da177e4 | 1041 | /* |
e7d39069 TM |
1042 | * Search for an existing write request, and attempt to update |
1043 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 1044 | * |
e7d39069 TM |
1045 | * If the attempt fails, then the existing request is flushed out |
1046 | * to disk. | |
1da177e4 | 1047 | */ |
e7d39069 TM |
1048 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
1049 | struct page *page, | |
1050 | unsigned int offset, | |
1051 | unsigned int bytes) | |
1da177e4 | 1052 | { |
e7d39069 TM |
1053 | struct nfs_page *req; |
1054 | unsigned int rqend; | |
1055 | unsigned int end; | |
1056 | int error; | |
1057 | ||
1058 | if (!PagePrivate(page)) | |
1059 | return NULL; | |
1da177e4 LT |
1060 | |
1061 | end = offset + bytes; | |
e7d39069 | 1062 | spin_lock(&inode->i_lock); |
1da177e4 | 1063 | |
1da177e4 | 1064 | for (;;) { |
84d3a9a9 | 1065 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
e7d39069 TM |
1066 | if (req == NULL) |
1067 | goto out_unlock; | |
1068 | ||
2bfc6e56 WAA |
1069 | /* should be handled by nfs_flush_incompatible */ |
1070 | WARN_ON_ONCE(req->wb_head != req); | |
1071 | WARN_ON_ONCE(req->wb_this_page != req); | |
1072 | ||
e7d39069 TM |
1073 | rqend = req->wb_offset + req->wb_bytes; |
1074 | /* | |
1075 | * Tell the caller to flush out the request if | |
1076 | * the offsets are non-contiguous. | |
1077 | * Note: nfs_flush_incompatible() will already | |
1078 | * have flushed out requests having wrong owners. | |
1079 | */ | |
e468bae9 | 1080 | if (offset > rqend |
e7d39069 TM |
1081 | || end < req->wb_offset) |
1082 | goto out_flushme; | |
1083 | ||
7ad84aa9 | 1084 | if (nfs_lock_request(req)) |
1da177e4 | 1085 | break; |
1da177e4 | 1086 | |
e7d39069 | 1087 | /* The request is locked, so wait and then retry */ |
587142f8 | 1088 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
1089 | error = nfs_wait_on_request(req); |
1090 | nfs_release_request(req); | |
1091 | if (error != 0) | |
1092 | goto out_err; | |
1093 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
1094 | } |
1095 | ||
1096 | /* Okay, the request matches. Update the region */ | |
1097 | if (offset < req->wb_offset) { | |
1098 | req->wb_offset = offset; | |
1099 | req->wb_pgbase = offset; | |
1da177e4 | 1100 | } |
1da177e4 LT |
1101 | if (end > rqend) |
1102 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1103 | else |
1104 | req->wb_bytes = rqend - req->wb_offset; | |
1105 | out_unlock: | |
ca138f36 FI |
1106 | if (req) |
1107 | nfs_clear_request_commit(req); | |
411a99ad | 1108 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
1109 | return req; |
1110 | out_flushme: | |
1111 | spin_unlock(&inode->i_lock); | |
1112 | nfs_release_request(req); | |
1113 | error = nfs_wb_page(inode, page); | |
1114 | out_err: | |
1115 | return ERR_PTR(error); | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * Try to update an existing write request, or create one if there is none. | |
1120 | * | |
1121 | * Note: Should always be called with the Page Lock held to prevent races | |
1122 | * if we have to add a new request. Also assumes that the caller has | |
1123 | * already called nfs_flush_incompatible() if necessary. | |
1124 | */ | |
1125 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1126 | struct page *page, unsigned int offset, unsigned int bytes) | |
1127 | { | |
d56b4ddf | 1128 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1129 | struct nfs_page *req; |
1da177e4 | 1130 | |
e7d39069 TM |
1131 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1132 | if (req != NULL) | |
1133 | goto out; | |
2bfc6e56 | 1134 | req = nfs_create_request(ctx, page, NULL, offset, bytes); |
e7d39069 TM |
1135 | if (IS_ERR(req)) |
1136 | goto out; | |
d6d6dc7c | 1137 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1138 | out: |
61e930a9 | 1139 | return req; |
1da177e4 LT |
1140 | } |
1141 | ||
e7d39069 TM |
1142 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1143 | unsigned int offset, unsigned int count) | |
1144 | { | |
1145 | struct nfs_page *req; | |
1146 | ||
1147 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1148 | if (IS_ERR(req)) | |
1149 | return PTR_ERR(req); | |
1150 | /* Update file length */ | |
1151 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1152 | nfs_mark_uptodate(req); |
a6305ddb | 1153 | nfs_mark_request_dirty(req); |
1d1afcbc | 1154 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1155 | return 0; |
1156 | } | |
1157 | ||
1da177e4 LT |
1158 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1159 | { | |
cd3758e3 | 1160 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1161 | struct nfs_lock_context *l_ctx; |
bd61e0a9 | 1162 | struct file_lock_context *flctx = file_inode(file)->i_flctx; |
1da177e4 | 1163 | struct nfs_page *req; |
1a54533e | 1164 | int do_flush, status; |
1da177e4 LT |
1165 | /* |
1166 | * Look for a request corresponding to this page. If there | |
1167 | * is one, and it belongs to another file, we flush it out | |
1168 | * before we try to copy anything into the page. Do this | |
1169 | * due to the lack of an ACCESS-type call in NFSv2. | |
1170 | * Also do the same if we find a request from an existing | |
1171 | * dropped page. | |
1172 | */ | |
1a54533e | 1173 | do { |
84d3a9a9 | 1174 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1175 | if (req == NULL) |
1176 | return 0; | |
2a369153 | 1177 | l_ctx = req->wb_lock_context; |
138a2935 TM |
1178 | do_flush = req->wb_page != page || |
1179 | !nfs_match_open_context(req->wb_context, ctx); | |
2bfc6e56 WAA |
1180 | /* for now, flush if more than 1 request in page_group */ |
1181 | do_flush |= req->wb_this_page != req; | |
bd61e0a9 JL |
1182 | if (l_ctx && flctx && |
1183 | !(list_empty_careful(&flctx->flc_posix) && | |
1184 | list_empty_careful(&flctx->flc_flock))) { | |
d51fdb87 | 1185 | do_flush |= l_ctx->lockowner != current->files; |
5263e31e | 1186 | } |
1da177e4 | 1187 | nfs_release_request(req); |
1a54533e TM |
1188 | if (!do_flush) |
1189 | return 0; | |
d56b4ddf | 1190 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1191 | } while (status == 0); |
1192 | return status; | |
1da177e4 LT |
1193 | } |
1194 | ||
dc24826b AA |
1195 | /* |
1196 | * Avoid buffered writes when a open context credential's key would | |
1197 | * expire soon. | |
1198 | * | |
1199 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1200 | * | |
1201 | * Return 0 and set a credential flag which triggers the inode to flush | |
1202 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1203 | * RPC_KEY_EXPIRE_TIMEO. | |
1204 | */ | |
1205 | int | |
1206 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1207 | { | |
1208 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
1209 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
1210 | ||
1211 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
1212 | } | |
1213 | ||
1214 | /* | |
1215 | * Test if the open context credential key is marked to expire soon. | |
1216 | */ | |
ce52914e | 1217 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) |
dc24826b | 1218 | { |
ce52914e SM |
1219 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; |
1220 | ||
1221 | return rpcauth_cred_key_to_expire(auth, ctx->cred); | |
dc24826b AA |
1222 | } |
1223 | ||
5d47a356 TM |
1224 | /* |
1225 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1226 | * the PageUptodate() flag. In this case, we will need to turn off | |
1227 | * write optimisations that depend on the page contents being correct. | |
1228 | */ | |
8d197a56 | 1229 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 1230 | { |
d529ef83 JL |
1231 | struct nfs_inode *nfsi = NFS_I(inode); |
1232 | ||
8d197a56 TM |
1233 | if (nfs_have_delegated_attributes(inode)) |
1234 | goto out; | |
18dd78c4 | 1235 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
d529ef83 | 1236 | return false; |
4db72b40 | 1237 | smp_rmb(); |
d529ef83 | 1238 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
1239 | return false; |
1240 | out: | |
18dd78c4 SM |
1241 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
1242 | return false; | |
8d197a56 | 1243 | return PageUptodate(page) != 0; |
5d47a356 TM |
1244 | } |
1245 | ||
5263e31e JL |
1246 | static bool |
1247 | is_whole_file_wrlock(struct file_lock *fl) | |
1248 | { | |
1249 | return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && | |
1250 | fl->fl_type == F_WRLCK; | |
1251 | } | |
1252 | ||
c7559663 SM |
1253 | /* If we know the page is up to date, and we're not using byte range locks (or |
1254 | * if we have the whole file locked for writing), it may be more efficient to | |
1255 | * extend the write to cover the entire page in order to avoid fragmentation | |
1256 | * inefficiencies. | |
1257 | * | |
263b4509 SM |
1258 | * If the file is opened for synchronous writes then we can just skip the rest |
1259 | * of the checks. | |
c7559663 SM |
1260 | */ |
1261 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
1262 | { | |
5263e31e JL |
1263 | int ret; |
1264 | struct file_lock_context *flctx = inode->i_flctx; | |
1265 | struct file_lock *fl; | |
1266 | ||
c7559663 SM |
1267 | if (file->f_flags & O_DSYNC) |
1268 | return 0; | |
263b4509 SM |
1269 | if (!nfs_write_pageuptodate(page, inode)) |
1270 | return 0; | |
c7559663 SM |
1271 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1272 | return 1; | |
bd61e0a9 JL |
1273 | if (!flctx || (list_empty_careful(&flctx->flc_flock) && |
1274 | list_empty_careful(&flctx->flc_posix))) | |
8fa4592a | 1275 | return 1; |
5263e31e JL |
1276 | |
1277 | /* Check to see if there are whole file write locks */ | |
5263e31e | 1278 | ret = 0; |
6109c850 | 1279 | spin_lock(&flctx->flc_lock); |
bd61e0a9 JL |
1280 | if (!list_empty(&flctx->flc_posix)) { |
1281 | fl = list_first_entry(&flctx->flc_posix, struct file_lock, | |
1282 | fl_list); | |
1283 | if (is_whole_file_wrlock(fl)) | |
1284 | ret = 1; | |
1285 | } else if (!list_empty(&flctx->flc_flock)) { | |
5263e31e JL |
1286 | fl = list_first_entry(&flctx->flc_flock, struct file_lock, |
1287 | fl_list); | |
1288 | if (fl->fl_type == F_WRLCK) | |
1289 | ret = 1; | |
1290 | } | |
6109c850 | 1291 | spin_unlock(&flctx->flc_lock); |
5263e31e | 1292 | return ret; |
c7559663 SM |
1293 | } |
1294 | ||
1da177e4 LT |
1295 | /* |
1296 | * Update and possibly write a cached page of an NFS file. | |
1297 | * | |
1298 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1299 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1300 | */ | |
1301 | int nfs_updatepage(struct file *file, struct page *page, | |
1302 | unsigned int offset, unsigned int count) | |
1303 | { | |
cd3758e3 | 1304 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 1305 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
1306 | int status = 0; |
1307 | ||
91d5b470 CL |
1308 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1309 | ||
6de1472f AV |
1310 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1311 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1312 | |
149a4fdd BC |
1313 | if (!count) |
1314 | goto out; | |
1315 | ||
c7559663 | 1316 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 1317 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 1318 | offset = 0; |
1da177e4 LT |
1319 | } |
1320 | ||
e21195a7 | 1321 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
1322 | if (status < 0) |
1323 | nfs_set_pageerror(page); | |
59b7c05f TM |
1324 | else |
1325 | __set_page_dirty_nobuffers(page); | |
149a4fdd | 1326 | out: |
48186c7d | 1327 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1328 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1329 | return status; |
1330 | } | |
1331 | ||
3ff7576d | 1332 | static int flush_task_priority(int how) |
1da177e4 LT |
1333 | { |
1334 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1335 | case FLUSH_HIGHPRI: | |
1336 | return RPC_PRIORITY_HIGH; | |
1337 | case FLUSH_LOWPRI: | |
1338 | return RPC_PRIORITY_LOW; | |
1339 | } | |
1340 | return RPC_PRIORITY_NORMAL; | |
1341 | } | |
1342 | ||
d45f60c6 WAA |
1343 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1344 | struct rpc_message *msg, | |
abde71f4 | 1345 | const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 | 1346 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1347 | { |
3ff7576d | 1348 | int priority = flush_task_priority(how); |
d138d5d1 | 1349 | |
1ed26f33 | 1350 | task_setup_data->priority = priority; |
abde71f4 | 1351 | rpc_ops->write_setup(hdr, msg); |
d138d5d1 | 1352 | |
abde71f4 | 1353 | nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, |
d45f60c6 | 1354 | &task_setup_data->rpc_client, msg, hdr); |
275acaaf TM |
1355 | } |
1356 | ||
6d884e8f F |
1357 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1358 | * call this on each, which will prepare them to be retried on next | |
1359 | * writeback using standard nfs. | |
1360 | */ | |
1361 | static void nfs_redirty_request(struct nfs_page *req) | |
1362 | { | |
1363 | nfs_mark_request_dirty(req); | |
c7070113 | 1364 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1d1afcbc | 1365 | nfs_unlock_request(req); |
20633f04 | 1366 | nfs_end_page_writeback(req); |
3aff4ebb | 1367 | nfs_release_request(req); |
6d884e8f F |
1368 | } |
1369 | ||
061ae2ed | 1370 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1371 | { |
1372 | struct nfs_page *req; | |
1373 | ||
1374 | while (!list_empty(head)) { | |
1375 | req = nfs_list_entry(head->next); | |
1376 | nfs_list_remove_request(req); | |
1377 | nfs_redirty_request(req); | |
1378 | } | |
1379 | } | |
1380 | ||
dc602dd7 TM |
1381 | static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) |
1382 | { | |
1383 | nfs_async_write_error(&hdr->pages); | |
1384 | } | |
1385 | ||
061ae2ed FI |
1386 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1387 | .error_cleanup = nfs_async_write_error, | |
1388 | .completion = nfs_write_completion, | |
dc602dd7 | 1389 | .reschedule_io = nfs_async_write_reschedule_io, |
061ae2ed FI |
1390 | }; |
1391 | ||
57208fa7 | 1392 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1393 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1394 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1395 | { |
a20c93e3 | 1396 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1397 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1398 | |
1399 | #ifdef CONFIG_NFS_V4_1 | |
1400 | if (server->pnfs_curr_ld && !force_mds) | |
1401 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1402 | #endif | |
4a0de55c | 1403 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
fbe77c30 | 1404 | server->wsize, ioflags, GFP_NOIO); |
1751c363 | 1405 | } |
ddda8e0a | 1406 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1407 | |
dce81290 TM |
1408 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1409 | { | |
a7d42ddb WAA |
1410 | struct nfs_pgio_mirror *mirror; |
1411 | ||
6f29b9bb KM |
1412 | if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) |
1413 | pgio->pg_ops->pg_cleanup(pgio); | |
1414 | ||
41d8d5b7 | 1415 | pgio->pg_ops = &nfs_pgio_rw_ops; |
a7d42ddb WAA |
1416 | |
1417 | nfs_pageio_stop_mirroring(pgio); | |
1418 | ||
1419 | mirror = &pgio->pg_mirrors[0]; | |
1420 | mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
dce81290 | 1421 | } |
1f945357 | 1422 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1423 | |
1da177e4 | 1424 | |
0b7c0153 FI |
1425 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1426 | { | |
1427 | struct nfs_commit_data *data = calldata; | |
1428 | ||
1429 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1430 | } | |
1431 | ||
1f2edbe3 TM |
1432 | /* |
1433 | * Special version of should_remove_suid() that ignores capabilities. | |
1434 | */ | |
1435 | static int nfs_should_remove_suid(const struct inode *inode) | |
1436 | { | |
1437 | umode_t mode = inode->i_mode; | |
1438 | int kill = 0; | |
1439 | ||
1440 | /* suid always must be killed */ | |
1441 | if (unlikely(mode & S_ISUID)) | |
1442 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1443 | |
1f2edbe3 TM |
1444 | /* |
1445 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1446 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1447 | */ | |
1448 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1449 | kill |= ATTR_KILL_SGID; | |
1450 | ||
1451 | if (unlikely(kill && S_ISREG(mode))) | |
1452 | return kill; | |
1453 | ||
1454 | return 0; | |
1455 | } | |
788e7a89 | 1456 | |
a08a8cd3 TM |
1457 | static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, |
1458 | struct nfs_fattr *fattr) | |
1459 | { | |
1460 | struct nfs_pgio_args *argp = &hdr->args; | |
1461 | struct nfs_pgio_res *resp = &hdr->res; | |
2b83d3de | 1462 | u64 size = argp->offset + resp->count; |
a08a8cd3 TM |
1463 | |
1464 | if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) | |
2b83d3de TM |
1465 | fattr->size = size; |
1466 | if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { | |
1467 | fattr->valid &= ~NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 | 1468 | return; |
2b83d3de TM |
1469 | } |
1470 | if (size != fattr->size) | |
a08a8cd3 TM |
1471 | return; |
1472 | /* Set attribute barrier */ | |
1473 | nfs_fattr_set_barrier(fattr); | |
2b83d3de TM |
1474 | /* ...and update size */ |
1475 | fattr->valid |= NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 TM |
1476 | } |
1477 | ||
1478 | void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) | |
1479 | { | |
2b83d3de | 1480 | struct nfs_fattr *fattr = &hdr->fattr; |
a08a8cd3 TM |
1481 | struct inode *inode = hdr->inode; |
1482 | ||
a08a8cd3 TM |
1483 | spin_lock(&inode->i_lock); |
1484 | nfs_writeback_check_extend(hdr, fattr); | |
1485 | nfs_post_op_update_inode_force_wcc_locked(inode, fattr); | |
1486 | spin_unlock(&inode->i_lock); | |
1487 | } | |
1488 | EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); | |
1489 | ||
1da177e4 LT |
1490 | /* |
1491 | * This function is called when the WRITE call is complete. | |
1492 | */ | |
d45f60c6 WAA |
1493 | static int nfs_writeback_done(struct rpc_task *task, |
1494 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1495 | struct inode *inode) |
1da177e4 | 1496 | { |
788e7a89 | 1497 | int status; |
1da177e4 | 1498 | |
f551e44f CL |
1499 | /* |
1500 | * ->write_done will attempt to use post-op attributes to detect | |
1501 | * conflicting writes by other clients. A strict interpretation | |
1502 | * of close-to-open would allow us to continue caching even if | |
1503 | * another writer had changed the file, but some applications | |
1504 | * depend on tighter cache coherency when writing. | |
1505 | */ | |
d45f60c6 | 1506 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1507 | if (status != 0) |
0eecb214 | 1508 | return status; |
d45f60c6 | 1509 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
91d5b470 | 1510 | |
d45f60c6 WAA |
1511 | if (hdr->res.verf->committed < hdr->args.stable && |
1512 | task->tk_status >= 0) { | |
1da177e4 LT |
1513 | /* We tried a write call, but the server did not |
1514 | * commit data to stable storage even though we | |
1515 | * requested it. | |
1516 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1517 | * the server reports NFS_DATA_SYNC, but performs | |
1518 | * NFS_FILE_SYNC. We therefore implement this checking | |
1519 | * as a dprintk() in order to avoid filling syslog. | |
1520 | */ | |
1521 | static unsigned long complain; | |
1522 | ||
a69aef14 | 1523 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1524 | if (time_before(complain, jiffies)) { |
48186c7d | 1525 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1526 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1527 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
d45f60c6 | 1528 | hdr->res.verf->committed, hdr->args.stable); |
1da177e4 LT |
1529 | complain = jiffies + 300 * HZ; |
1530 | } | |
1531 | } | |
1f2edbe3 TM |
1532 | |
1533 | /* Deal with the suid/sgid bit corner case */ | |
1534 | if (nfs_should_remove_suid(inode)) | |
1535 | nfs_mark_for_revalidate(inode); | |
0eecb214 AS |
1536 | return 0; |
1537 | } | |
1538 | ||
1539 | /* | |
1540 | * This function is called when the WRITE call is complete. | |
1541 | */ | |
d45f60c6 WAA |
1542 | static void nfs_writeback_result(struct rpc_task *task, |
1543 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1544 | { |
d45f60c6 WAA |
1545 | struct nfs_pgio_args *argp = &hdr->args; |
1546 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1547 | |
1548 | if (resp->count < argp->count) { | |
1da177e4 LT |
1549 | static unsigned long complain; |
1550 | ||
6c75dc0d | 1551 | /* This a short write! */ |
d45f60c6 | 1552 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1553 | |
1da177e4 | 1554 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1555 | if (resp->count == 0) { |
1556 | if (time_before(complain, jiffies)) { | |
1557 | printk(KERN_WARNING | |
1558 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1559 | argp->count); | |
1560 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1561 | } |
d45f60c6 | 1562 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1563 | task->tk_status = -EIO; |
13602896 | 1564 | return; |
1da177e4 | 1565 | } |
f8417b48 KM |
1566 | |
1567 | /* For non rpc-based layout drivers, retry-through-MDS */ | |
1568 | if (!task->tk_ops) { | |
1569 | hdr->pnfs_error = -EAGAIN; | |
1570 | return; | |
1571 | } | |
1572 | ||
6c75dc0d FI |
1573 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1574 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1575 | /* Resend from where the server left off */ | |
d45f60c6 | 1576 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1577 | argp->offset += resp->count; |
1578 | argp->pgbase += resp->count; | |
1579 | argp->count -= resp->count; | |
1580 | } else { | |
1581 | /* Resend as a stable write in order to avoid | |
1582 | * headaches in the case of a server crash. | |
1583 | */ | |
1584 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1585 | } |
6c75dc0d | 1586 | rpc_restart_call_prepare(task); |
1da177e4 | 1587 | } |
1da177e4 LT |
1588 | } |
1589 | ||
af7cf057 | 1590 | static int wait_on_commit(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1591 | { |
af7cf057 TM |
1592 | return wait_on_atomic_t(&cinfo->rpcs_out, |
1593 | nfs_wait_atomic_killable, TASK_KILLABLE); | |
1594 | } | |
b8413f98 | 1595 | |
af7cf057 TM |
1596 | static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) |
1597 | { | |
1598 | atomic_inc(&cinfo->rpcs_out); | |
71d0a611 TM |
1599 | } |
1600 | ||
af7cf057 | 1601 | static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1602 | { |
af7cf057 TM |
1603 | if (atomic_dec_and_test(&cinfo->rpcs_out)) |
1604 | wake_up_atomic_t(&cinfo->rpcs_out); | |
71d0a611 TM |
1605 | } |
1606 | ||
0b7c0153 | 1607 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1608 | { |
0b7c0153 FI |
1609 | put_nfs_open_context(data->context); |
1610 | nfs_commit_free(data); | |
1da177e4 | 1611 | } |
e0c2b380 | 1612 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1613 | |
0b7c0153 | 1614 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
c36aae9a | 1615 | const struct nfs_rpc_ops *nfs_ops, |
9ace33cd | 1616 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1617 | int how, int flags) |
1da177e4 | 1618 | { |
07737691 | 1619 | struct rpc_task *task; |
9ace33cd | 1620 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1621 | struct rpc_message msg = { |
1622 | .rpc_argp = &data->args, | |
1623 | .rpc_resp = &data->res, | |
9ace33cd | 1624 | .rpc_cred = data->cred, |
bdc7f021 | 1625 | }; |
84115e1c | 1626 | struct rpc_task_setup task_setup_data = { |
07737691 | 1627 | .task = &data->task, |
9ace33cd | 1628 | .rpc_client = clnt, |
bdc7f021 | 1629 | .rpc_message = &msg, |
9ace33cd | 1630 | .callback_ops = call_ops, |
84115e1c | 1631 | .callback_data = data, |
101070ca | 1632 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1633 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1634 | .priority = priority, |
84115e1c | 1635 | }; |
9ace33cd | 1636 | /* Set up the initial task struct. */ |
c36aae9a | 1637 | nfs_ops->commit_setup(data, &msg); |
9ace33cd | 1638 | |
b4839ebe | 1639 | dprintk("NFS: initiated commit call\n"); |
9ace33cd | 1640 | |
8c21c62c WAA |
1641 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1642 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1643 | ||
9ace33cd FI |
1644 | task = rpc_run_task(&task_setup_data); |
1645 | if (IS_ERR(task)) | |
1646 | return PTR_ERR(task); | |
1647 | if (how & FLUSH_SYNC) | |
1648 | rpc_wait_for_completion_task(task); | |
1649 | rpc_put_task(task); | |
1650 | return 0; | |
1651 | } | |
e0c2b380 | 1652 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1653 | |
378520b8 PT |
1654 | static loff_t nfs_get_lwb(struct list_head *head) |
1655 | { | |
1656 | loff_t lwb = 0; | |
1657 | struct nfs_page *req; | |
1658 | ||
1659 | list_for_each_entry(req, head, wb_list) | |
1660 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1661 | lwb = req_offset(req) + req->wb_bytes; | |
1662 | ||
1663 | return lwb; | |
1664 | } | |
1665 | ||
9ace33cd FI |
1666 | /* |
1667 | * Set up the argument/result storage required for the RPC call. | |
1668 | */ | |
0b7c0153 | 1669 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1670 | struct list_head *head, |
1671 | struct pnfs_layout_segment *lseg, | |
1672 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1673 | { |
1674 | struct nfs_page *first = nfs_list_entry(head->next); | |
2b0143b5 | 1675 | struct inode *inode = d_inode(first->wb_context->dentry); |
1da177e4 LT |
1676 | |
1677 | /* Set up the RPC argument and reply structs | |
1678 | * NB: take care not to mess about with data->commit et al. */ | |
1679 | ||
1680 | list_splice_init(head, &data->pages); | |
1da177e4 | 1681 | |
1da177e4 | 1682 | data->inode = inode; |
9ace33cd | 1683 | data->cred = first->wb_context->cred; |
988b6dce | 1684 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1685 | /* only set lwb for pnfs commit */ |
1686 | if (lseg) | |
1687 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1688 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1689 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1690 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1691 | |
1692 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1693 | /* Note: we always request a commit of the entire inode */ |
1694 | data->args.offset = 0; | |
1695 | data->args.count = 0; | |
0b7c0153 | 1696 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1697 | data->res.fattr = &data->fattr; |
1698 | data->res.verf = &data->verf; | |
0e574af1 | 1699 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1700 | } |
e0c2b380 | 1701 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1702 | |
e0c2b380 | 1703 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 | 1704 | struct pnfs_layout_segment *lseg, |
b57ff130 WAA |
1705 | struct nfs_commit_info *cinfo, |
1706 | u32 ds_commit_idx) | |
64bfeb49 FI |
1707 | { |
1708 | struct nfs_page *req; | |
1709 | ||
1710 | while (!list_empty(page_list)) { | |
1711 | req = nfs_list_entry(page_list->next); | |
1712 | nfs_list_remove_request(req); | |
b57ff130 | 1713 | nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); |
487b9b8a TH |
1714 | if (!cinfo->dreq) |
1715 | nfs_clear_page_commit(req->wb_page); | |
1d1afcbc | 1716 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1717 | } |
1718 | } | |
e0c2b380 | 1719 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1720 | |
b20135d0 TM |
1721 | static void |
1722 | nfs_commit_resched_write(struct nfs_commit_info *cinfo, | |
1723 | struct nfs_page *req) | |
1724 | { | |
1725 | __set_page_dirty_nobuffers(req->wb_page); | |
1726 | } | |
1727 | ||
1da177e4 LT |
1728 | /* |
1729 | * Commit dirty pages | |
1730 | */ | |
1731 | static int | |
ea2cf228 FI |
1732 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1733 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1734 | { |
0b7c0153 | 1735 | struct nfs_commit_data *data; |
1da177e4 | 1736 | |
ade8febd WAA |
1737 | /* another commit raced with us */ |
1738 | if (list_empty(head)) | |
1739 | return 0; | |
1740 | ||
518662e0 | 1741 | data = nfs_commitdata_alloc(true); |
1da177e4 LT |
1742 | |
1743 | /* Set up the argument struct */ | |
f453a54a FI |
1744 | nfs_init_commit(data, head, NULL, cinfo); |
1745 | atomic_inc(&cinfo->mds->rpcs_out); | |
c36aae9a PT |
1746 | return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), |
1747 | data->mds_ops, how, 0); | |
1da177e4 LT |
1748 | } |
1749 | ||
67911c8f AS |
1750 | int nfs_commit_file(struct file *file, struct nfs_write_verifier *verf) |
1751 | { | |
1752 | struct inode *inode = file_inode(file); | |
1753 | struct nfs_open_context *open; | |
1754 | struct nfs_commit_info cinfo; | |
1755 | struct nfs_page *req; | |
1756 | int ret; | |
1757 | ||
1758 | open = get_nfs_open_context(nfs_file_open_context(file)); | |
1759 | req = nfs_create_request(open, NULL, NULL, 0, i_size_read(inode)); | |
2997bfd0 DC |
1760 | if (IS_ERR(req)) { |
1761 | ret = PTR_ERR(req); | |
67911c8f AS |
1762 | goto out_put; |
1763 | } | |
1764 | ||
1765 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
1766 | ||
1767 | memcpy(&req->wb_verf, verf, sizeof(struct nfs_write_verifier)); | |
1768 | nfs_request_add_commit_list(req, &cinfo); | |
1769 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
1770 | if (ret > 0) | |
1771 | ret = 0; | |
1772 | ||
1773 | nfs_free_request(req); | |
1774 | out_put: | |
1775 | put_nfs_open_context(open); | |
1776 | return ret; | |
1777 | } | |
1778 | EXPORT_SYMBOL_GPL(nfs_commit_file); | |
1779 | ||
1da177e4 LT |
1780 | /* |
1781 | * COMMIT call returned | |
1782 | */ | |
788e7a89 | 1783 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1784 | { |
0b7c0153 | 1785 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1786 | |
a3f565b1 | 1787 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1788 | task->tk_pid, task->tk_status); |
1789 | ||
788e7a89 | 1790 | /* Call the NFS version-specific code */ |
c0d0e96b | 1791 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1792 | } |
1793 | ||
f453a54a | 1794 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1795 | { |
5917ce84 | 1796 | struct nfs_page *req; |
c9d8f89d | 1797 | int status = data->task.tk_status; |
f453a54a | 1798 | struct nfs_commit_info cinfo; |
353db796 | 1799 | struct nfs_server *nfss; |
788e7a89 | 1800 | |
1da177e4 LT |
1801 | while (!list_empty(&data->pages)) { |
1802 | req = nfs_list_entry(data->pages.next); | |
1803 | nfs_list_remove_request(req); | |
67911c8f AS |
1804 | if (req->wb_page) |
1805 | nfs_clear_page_commit(req->wb_page); | |
1da177e4 | 1806 | |
1e8968c5 | 1807 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1808 | req->wb_context->dentry->d_sb->s_id, |
2b0143b5 | 1809 | (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), |
1da177e4 LT |
1810 | req->wb_bytes, |
1811 | (long long)req_offset(req)); | |
c9d8f89d TM |
1812 | if (status < 0) { |
1813 | nfs_context_set_write_error(req->wb_context, status); | |
38a33101 KM |
1814 | if (req->wb_page) |
1815 | nfs_inode_remove_request(req); | |
ddeaa637 | 1816 | dprintk_cont(", error = %d\n", status); |
1da177e4 LT |
1817 | goto next; |
1818 | } | |
1819 | ||
1820 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1821 | * returned by the server against all stored verfs. */ | |
8fc3c386 | 1822 | if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { |
1da177e4 | 1823 | /* We have a match */ |
38a33101 KM |
1824 | if (req->wb_page) |
1825 | nfs_inode_remove_request(req); | |
ddeaa637 | 1826 | dprintk_cont(" OK\n"); |
1da177e4 LT |
1827 | goto next; |
1828 | } | |
1829 | /* We have a mismatch. Write the page again */ | |
ddeaa637 | 1830 | dprintk_cont(" mismatch\n"); |
6d884e8f | 1831 | nfs_mark_request_dirty(req); |
05990d1b | 1832 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1833 | next: |
1d1afcbc | 1834 | nfs_unlock_and_release_request(req); |
1da177e4 | 1835 | } |
353db796 N |
1836 | nfss = NFS_SERVER(data->inode); |
1837 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
1838 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); | |
1839 | ||
f453a54a | 1840 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
af7cf057 | 1841 | nfs_commit_end(cinfo.mds); |
5917ce84 FI |
1842 | } |
1843 | ||
1844 | static void nfs_commit_release(void *calldata) | |
1845 | { | |
0b7c0153 | 1846 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1847 | |
f453a54a | 1848 | data->completion_ops->completion(data); |
c9d8f89d | 1849 | nfs_commitdata_release(calldata); |
1da177e4 | 1850 | } |
788e7a89 TM |
1851 | |
1852 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1853 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1854 | .rpc_call_done = nfs_commit_done, |
1855 | .rpc_release = nfs_commit_release, | |
1856 | }; | |
1da177e4 | 1857 | |
f453a54a FI |
1858 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1859 | .completion = nfs_commit_release_pages, | |
b20135d0 | 1860 | .resched_write = nfs_commit_resched_write, |
f453a54a FI |
1861 | }; |
1862 | ||
1763da12 FI |
1863 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1864 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1865 | { |
1866 | int status; | |
1867 | ||
ea2cf228 | 1868 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1869 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1870 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1871 | return status; |
1872 | } | |
1873 | ||
b608b283 | 1874 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1875 | { |
1da177e4 | 1876 | LIST_HEAD(head); |
ea2cf228 | 1877 | struct nfs_commit_info cinfo; |
71d0a611 | 1878 | int may_wait = how & FLUSH_SYNC; |
af7cf057 | 1879 | int error = 0; |
b8413f98 | 1880 | int res; |
1da177e4 | 1881 | |
ea2cf228 | 1882 | nfs_init_cinfo_from_inode(&cinfo, inode); |
af7cf057 | 1883 | nfs_commit_begin(cinfo.mds); |
ea2cf228 | 1884 | res = nfs_scan_commit(inode, &head, &cinfo); |
af7cf057 | 1885 | if (res) |
ea2cf228 | 1886 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
af7cf057 TM |
1887 | nfs_commit_end(cinfo.mds); |
1888 | if (error < 0) | |
1889 | goto out_error; | |
1890 | if (!may_wait) | |
1891 | goto out_mark_dirty; | |
1892 | error = wait_on_commit(cinfo.mds); | |
1893 | if (error < 0) | |
1894 | return error; | |
c5efa5fc | 1895 | return res; |
af7cf057 TM |
1896 | out_error: |
1897 | res = error; | |
c5efa5fc TM |
1898 | /* Note: If we exit without ensuring that the commit is complete, |
1899 | * we must mark the inode as dirty. Otherwise, future calls to | |
1900 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1901 | * that the data is on the disk. | |
1902 | */ | |
1903 | out_mark_dirty: | |
1904 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1905 | return res; |
1906 | } | |
b20135d0 | 1907 | EXPORT_SYMBOL_GPL(nfs_commit_inode); |
8fc795f7 | 1908 | |
ae09c31f | 1909 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
8fc795f7 | 1910 | { |
420e3646 TM |
1911 | struct nfs_inode *nfsi = NFS_I(inode); |
1912 | int flags = FLUSH_SYNC; | |
1913 | int ret = 0; | |
8fc795f7 | 1914 | |
3236c3e1 | 1915 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1916 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1917 | return ret; |
1918 | ||
a00dd6c0 JL |
1919 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1920 | /* Don't commit yet if this is a non-blocking flush and there | |
1921 | * are a lot of outstanding writes for this mapping. | |
1922 | */ | |
cb1410c7 | 1923 | if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) |
a00dd6c0 | 1924 | goto out_mark_dirty; |
420e3646 | 1925 | |
a00dd6c0 | 1926 | /* don't wait for the COMMIT response */ |
420e3646 | 1927 | flags = 0; |
a00dd6c0 JL |
1928 | } |
1929 | ||
420e3646 TM |
1930 | ret = nfs_commit_inode(inode, flags); |
1931 | if (ret >= 0) { | |
1932 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1933 | if (ret < wbc->nr_to_write) | |
1934 | wbc->nr_to_write -= ret; | |
1935 | else | |
1936 | wbc->nr_to_write = 0; | |
1937 | } | |
8fc795f7 | 1938 | return 0; |
420e3646 TM |
1939 | } |
1940 | out_mark_dirty: | |
8fc795f7 TM |
1941 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1942 | return ret; | |
1943 | } | |
89d77c8f | 1944 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1945 | |
837bb1d7 TM |
1946 | /* |
1947 | * Wrapper for filemap_write_and_wait_range() | |
1948 | * | |
1949 | * Needed for pNFS in order to ensure data becomes visible to the | |
1950 | * client. | |
1951 | */ | |
1952 | int nfs_filemap_write_and_wait_range(struct address_space *mapping, | |
1953 | loff_t lstart, loff_t lend) | |
1954 | { | |
1955 | int ret; | |
1956 | ||
1957 | ret = filemap_write_and_wait_range(mapping, lstart, lend); | |
1958 | if (ret == 0) | |
1959 | ret = pnfs_sync_inode(mapping->host, true); | |
1960 | return ret; | |
1961 | } | |
1962 | EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); | |
1963 | ||
acdc53b2 TM |
1964 | /* |
1965 | * flush the inode to disk. | |
1966 | */ | |
1967 | int nfs_wb_all(struct inode *inode) | |
34901f70 | 1968 | { |
f4ce1299 TM |
1969 | int ret; |
1970 | ||
1971 | trace_nfs_writeback_inode_enter(inode); | |
1972 | ||
5bb89b47 | 1973 | ret = filemap_write_and_wait(inode->i_mapping); |
6b196875 CL |
1974 | if (ret) |
1975 | goto out; | |
1976 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
1977 | if (ret < 0) | |
1978 | goto out; | |
1979 | pnfs_sync_inode(inode, true); | |
1980 | ret = 0; | |
34901f70 | 1981 | |
6b196875 | 1982 | out: |
f4ce1299 TM |
1983 | trace_nfs_writeback_inode_exit(inode, ret); |
1984 | return ret; | |
1c75950b | 1985 | } |
ddda8e0a | 1986 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1987 | |
1b3b4a1a TM |
1988 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1989 | { | |
1990 | struct nfs_page *req; | |
1b3b4a1a TM |
1991 | int ret = 0; |
1992 | ||
3e217045 WAA |
1993 | wait_on_page_writeback(page); |
1994 | ||
1995 | /* blocking call to cancel all requests and join to a single (head) | |
1996 | * request */ | |
1997 | req = nfs_lock_and_join_requests(page, false); | |
1998 | ||
1999 | if (IS_ERR(req)) { | |
2000 | ret = PTR_ERR(req); | |
2001 | } else if (req) { | |
2002 | /* all requests from this page have been cancelled by | |
2003 | * nfs_lock_and_join_requests, so just remove the head | |
2004 | * request from the inode / page_private pointer and | |
2005 | * release it */ | |
2006 | nfs_inode_remove_request(req); | |
3e217045 | 2007 | nfs_unlock_and_release_request(req); |
1b3b4a1a | 2008 | } |
3e217045 | 2009 | |
1b3b4a1a TM |
2010 | return ret; |
2011 | } | |
2012 | ||
7f2f12d9 TM |
2013 | /* |
2014 | * Write back all requests on one page - we do this before reading it. | |
2015 | */ | |
d6c843b9 | 2016 | int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder) |
1c75950b | 2017 | { |
29418aa4 | 2018 | loff_t range_start = page_file_offset(page); |
09cbfeaf | 2019 | loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); |
4d770ccf | 2020 | struct writeback_control wbc = { |
4d770ccf | 2021 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 2022 | .nr_to_write = 0, |
4d770ccf TM |
2023 | .range_start = range_start, |
2024 | .range_end = range_end, | |
2025 | }; | |
2026 | int ret; | |
1c75950b | 2027 | |
f4ce1299 TM |
2028 | trace_nfs_writeback_page_enter(inode); |
2029 | ||
0522f6ad | 2030 | for (;;) { |
ba8b06e6 | 2031 | wait_on_page_writeback(page); |
73e3302f | 2032 | if (clear_page_dirty_for_io(page)) { |
d6c843b9 | 2033 | ret = nfs_writepage_locked(page, &wbc, launder); |
73e3302f TM |
2034 | if (ret < 0) |
2035 | goto out_error; | |
0522f6ad | 2036 | continue; |
7f2f12d9 | 2037 | } |
f4ce1299 | 2038 | ret = 0; |
0522f6ad TM |
2039 | if (!PagePrivate(page)) |
2040 | break; | |
2041 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 2042 | if (ret < 0) |
73e3302f | 2043 | goto out_error; |
7f2f12d9 | 2044 | } |
73e3302f | 2045 | out_error: |
f4ce1299 | 2046 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 2047 | return ret; |
1c75950b TM |
2048 | } |
2049 | ||
074cc1de TM |
2050 | #ifdef CONFIG_MIGRATION |
2051 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 2052 | struct page *page, enum migrate_mode mode) |
074cc1de | 2053 | { |
2da95652 JL |
2054 | /* |
2055 | * If PagePrivate is set, then the page is currently associated with | |
2056 | * an in-progress read or write request. Don't try to migrate it. | |
2057 | * | |
2058 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
2059 | * that we can safely release the inode reference while holding | |
2060 | * the page lock. | |
2061 | */ | |
2062 | if (PagePrivate(page)) | |
2063 | return -EBUSY; | |
074cc1de | 2064 | |
8c209ce7 DH |
2065 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
2066 | return -EBUSY; | |
074cc1de | 2067 | |
a6bc32b8 | 2068 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
2069 | } |
2070 | #endif | |
2071 | ||
f7b422b1 | 2072 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
2073 | { |
2074 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 2075 | sizeof(struct nfs_pgio_header), |
1da177e4 | 2076 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 2077 | NULL); |
1da177e4 LT |
2078 | if (nfs_wdata_cachep == NULL) |
2079 | return -ENOMEM; | |
2080 | ||
93d2341c MD |
2081 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
2082 | nfs_wdata_cachep); | |
1da177e4 | 2083 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 2084 | goto out_destroy_write_cache; |
1da177e4 | 2085 | |
0b7c0153 FI |
2086 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
2087 | sizeof(struct nfs_commit_data), | |
2088 | 0, SLAB_HWCACHE_ALIGN, | |
2089 | NULL); | |
2090 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 2091 | goto out_destroy_write_mempool; |
0b7c0153 | 2092 | |
93d2341c | 2093 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 2094 | nfs_cdata_cachep); |
1da177e4 | 2095 | if (nfs_commit_mempool == NULL) |
3dd4765f | 2096 | goto out_destroy_commit_cache; |
1da177e4 | 2097 | |
89a09141 PZ |
2098 | /* |
2099 | * NFS congestion size, scale with available memory. | |
2100 | * | |
2101 | * 64MB: 8192k | |
2102 | * 128MB: 11585k | |
2103 | * 256MB: 16384k | |
2104 | * 512MB: 23170k | |
2105 | * 1GB: 32768k | |
2106 | * 2GB: 46340k | |
2107 | * 4GB: 65536k | |
2108 | * 8GB: 92681k | |
2109 | * 16GB: 131072k | |
2110 | * | |
2111 | * This allows larger machines to have larger/more transfers. | |
2112 | * Limit the default to 256M | |
2113 | */ | |
2114 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
2115 | if (nfs_congestion_kb > 256*1024) | |
2116 | nfs_congestion_kb = 256*1024; | |
2117 | ||
1da177e4 | 2118 | return 0; |
3dd4765f JL |
2119 | |
2120 | out_destroy_commit_cache: | |
2121 | kmem_cache_destroy(nfs_cdata_cachep); | |
2122 | out_destroy_write_mempool: | |
2123 | mempool_destroy(nfs_wdata_mempool); | |
2124 | out_destroy_write_cache: | |
2125 | kmem_cache_destroy(nfs_wdata_cachep); | |
2126 | return -ENOMEM; | |
1da177e4 LT |
2127 | } |
2128 | ||
266bee88 | 2129 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
2130 | { |
2131 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 2132 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 2133 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 2134 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
2135 | } |
2136 | ||
4a0de55c AS |
2137 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
2138 | .rw_alloc_header = nfs_writehdr_alloc, | |
2139 | .rw_free_header = nfs_writehdr_free, | |
0eecb214 AS |
2140 | .rw_done = nfs_writeback_done, |
2141 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 2142 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 2143 | }; |