]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/nfs/write.c | |
4 | * | |
7c85d900 | 5 | * Write file data over NFS. |
1da177e4 LT |
6 | * |
7 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
8 | */ | |
9 | ||
1da177e4 LT |
10 | #include <linux/types.h> |
11 | #include <linux/slab.h> | |
12 | #include <linux/mm.h> | |
13 | #include <linux/pagemap.h> | |
14 | #include <linux/file.h> | |
1da177e4 | 15 | #include <linux/writeback.h> |
89a09141 | 16 | #include <linux/swap.h> |
074cc1de | 17 | #include <linux/migrate.h> |
1da177e4 LT |
18 | |
19 | #include <linux/sunrpc/clnt.h> | |
20 | #include <linux/nfs_fs.h> | |
21 | #include <linux/nfs_mount.h> | |
22 | #include <linux/nfs_page.h> | |
3fcfab16 | 23 | #include <linux/backing-dev.h> |
afeacc8c | 24 | #include <linux/export.h> |
af7cf057 TM |
25 | #include <linux/freezer.h> |
26 | #include <linux/wait.h> | |
1eb5d98f | 27 | #include <linux/iversion.h> |
3fcfab16 | 28 | |
7c0f6ba6 | 29 | #include <linux/uaccess.h> |
875bc3fb | 30 | #include <linux/sched/mm.h> |
1da177e4 LT |
31 | |
32 | #include "delegation.h" | |
49a70f27 | 33 | #include "internal.h" |
91d5b470 | 34 | #include "iostat.h" |
def6ed7e | 35 | #include "nfs4_fs.h" |
074cc1de | 36 | #include "fscache.h" |
94ad1c80 | 37 | #include "pnfs.h" |
1da177e4 | 38 | |
f4ce1299 TM |
39 | #include "nfstrace.h" |
40 | ||
1da177e4 LT |
41 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
42 | ||
43 | #define MIN_POOL_WRITE (32) | |
44 | #define MIN_POOL_COMMIT (4) | |
45 | ||
919e3bd9 TM |
46 | struct nfs_io_completion { |
47 | void (*complete)(void *data); | |
48 | void *data; | |
49 | struct kref refcount; | |
50 | }; | |
51 | ||
1da177e4 LT |
52 | /* |
53 | * Local function declarations | |
54 | */ | |
f8512ad0 | 55 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 | 56 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 57 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 58 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
4a0de55c | 59 | static const struct nfs_rw_ops nfs_rw_write_ops; |
06c9fdf3 | 60 | static void nfs_inode_remove_request(struct nfs_page *req); |
d4581383 | 61 | static void nfs_clear_request_commit(struct nfs_page *req); |
02d1426c WAA |
62 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
63 | struct inode *inode); | |
3a3908c8 TM |
64 | static struct nfs_page * |
65 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
66 | struct page *page); | |
1da177e4 | 67 | |
e18b890b | 68 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 69 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 70 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
71 | static mempool_t *nfs_commit_mempool; |
72 | ||
515dcdcd | 73 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 74 | { |
518662e0 | 75 | struct nfs_commit_data *p; |
40859d7e | 76 | |
515dcdcd TM |
77 | p = kmem_cache_zalloc(nfs_cdata_cachep, nfs_io_gfp_mask()); |
78 | if (!p) { | |
518662e0 | 79 | p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); |
518662e0 N |
80 | if (!p) |
81 | return NULL; | |
515dcdcd | 82 | memset(p, 0, sizeof(*p)); |
1da177e4 | 83 | } |
518662e0 | 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 | { |
0bae835b | 97 | struct nfs_pgio_header *p; |
cd841605 | 98 | |
0bae835b TM |
99 | p = kmem_cache_zalloc(nfs_wdata_cachep, nfs_io_gfp_mask()); |
100 | if (!p) { | |
101 | p = mempool_alloc(nfs_wdata_mempool, GFP_NOWAIT); | |
102 | if (!p) | |
103 | return NULL; | |
104 | memset(p, 0, sizeof(*p)); | |
105 | } | |
237f8306 | 106 | p->rw_mode = FMODE_WRITE; |
3feb2d49 TM |
107 | return p; |
108 | } | |
6c75dc0d | 109 | |
1e7f3a48 | 110 | static void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 111 | { |
1e7f3a48 | 112 | mempool_free(hdr, nfs_wdata_mempool); |
3feb2d49 | 113 | } |
1da177e4 | 114 | |
919e3bd9 TM |
115 | static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) |
116 | { | |
117 | return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); | |
118 | } | |
119 | ||
120 | static void nfs_io_completion_init(struct nfs_io_completion *ioc, | |
121 | void (*complete)(void *), void *data) | |
122 | { | |
123 | ioc->complete = complete; | |
124 | ioc->data = data; | |
125 | kref_init(&ioc->refcount); | |
126 | } | |
127 | ||
128 | static void nfs_io_completion_release(struct kref *kref) | |
129 | { | |
130 | struct nfs_io_completion *ioc = container_of(kref, | |
131 | struct nfs_io_completion, refcount); | |
132 | ioc->complete(ioc->data); | |
133 | kfree(ioc); | |
134 | } | |
135 | ||
136 | static void nfs_io_completion_get(struct nfs_io_completion *ioc) | |
137 | { | |
138 | if (ioc != NULL) | |
139 | kref_get(&ioc->refcount); | |
140 | } | |
141 | ||
142 | static void nfs_io_completion_put(struct nfs_io_completion *ioc) | |
143 | { | |
144 | if (ioc != NULL) | |
145 | kref_put(&ioc->refcount, nfs_io_completion_release); | |
146 | } | |
147 | ||
e00ed89d TM |
148 | static void |
149 | nfs_page_set_inode_ref(struct nfs_page *req, struct inode *inode) | |
150 | { | |
151 | if (!test_and_set_bit(PG_INODE_REF, &req->wb_flags)) { | |
152 | kref_get(&req->wb_kref); | |
153 | atomic_long_inc(&NFS_I(inode)->nrequests); | |
154 | } | |
155 | } | |
156 | ||
157 | static int | |
158 | nfs_cancel_remove_inode(struct nfs_page *req, struct inode *inode) | |
159 | { | |
160 | int ret; | |
161 | ||
162 | if (!test_bit(PG_REMOVE, &req->wb_flags)) | |
163 | return 0; | |
164 | ret = nfs_page_group_lock(req); | |
165 | if (ret) | |
166 | return ret; | |
167 | if (test_and_clear_bit(PG_REMOVE, &req->wb_flags)) | |
168 | nfs_page_set_inode_ref(req, inode); | |
169 | nfs_page_group_unlock(req); | |
170 | return 0; | |
171 | } | |
172 | ||
bd37d6fc TM |
173 | static struct nfs_page * |
174 | nfs_page_private_request(struct page *page) | |
175 | { | |
176 | if (!PagePrivate(page)) | |
177 | return NULL; | |
178 | return (struct nfs_page *)page_private(page); | |
179 | } | |
180 | ||
84d3a9a9 WAA |
181 | /* |
182 | * nfs_page_find_head_request_locked - find head request associated with @page | |
183 | * | |
184 | * must be called while holding the inode lock. | |
185 | * | |
186 | * returns matching head request with reference held, or NULL if not found. | |
187 | */ | |
29418aa4 | 188 | static struct nfs_page * |
b30d2f04 | 189 | nfs_page_find_private_request(struct page *page) |
277459d2 | 190 | { |
4b9bb25b | 191 | struct address_space *mapping = page_file_mapping(page); |
bd37d6fc | 192 | struct nfs_page *req; |
277459d2 | 193 | |
b30d2f04 TM |
194 | if (!PagePrivate(page)) |
195 | return NULL; | |
4b9bb25b | 196 | spin_lock(&mapping->private_lock); |
bd37d6fc | 197 | req = nfs_page_private_request(page); |
84d3a9a9 WAA |
198 | if (req) { |
199 | WARN_ON_ONCE(req->wb_head != req); | |
29418aa4 | 200 | kref_get(&req->wb_kref); |
84d3a9a9 | 201 | } |
4b9bb25b | 202 | spin_unlock(&mapping->private_lock); |
b30d2f04 TM |
203 | return req; |
204 | } | |
29418aa4 | 205 | |
b30d2f04 TM |
206 | static struct nfs_page * |
207 | nfs_page_find_swap_request(struct page *page) | |
208 | { | |
209 | struct inode *inode = page_file_mapping(page)->host; | |
210 | struct nfs_inode *nfsi = NFS_I(inode); | |
211 | struct nfs_page *req = NULL; | |
212 | if (!PageSwapCache(page)) | |
213 | return NULL; | |
e824f99a | 214 | mutex_lock(&nfsi->commit_mutex); |
b30d2f04 TM |
215 | if (PageSwapCache(page)) { |
216 | req = nfs_page_search_commits_for_head_request_locked(nfsi, | |
217 | page); | |
218 | if (req) { | |
219 | WARN_ON_ONCE(req->wb_head != req); | |
220 | kref_get(&req->wb_kref); | |
221 | } | |
222 | } | |
e824f99a | 223 | mutex_unlock(&nfsi->commit_mutex); |
277459d2 TM |
224 | return req; |
225 | } | |
226 | ||
84d3a9a9 WAA |
227 | /* |
228 | * nfs_page_find_head_request - find head request associated with @page | |
229 | * | |
230 | * returns matching head request with reference held, or NULL if not found. | |
231 | */ | |
232 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 233 | { |
b30d2f04 | 234 | struct nfs_page *req; |
277459d2 | 235 | |
b30d2f04 TM |
236 | req = nfs_page_find_private_request(page); |
237 | if (!req) | |
238 | req = nfs_page_find_swap_request(page); | |
277459d2 TM |
239 | return req; |
240 | } | |
241 | ||
e00ed89d TM |
242 | static struct nfs_page *nfs_find_and_lock_page_request(struct page *page) |
243 | { | |
244 | struct inode *inode = page_file_mapping(page)->host; | |
245 | struct nfs_page *req, *head; | |
246 | int ret; | |
247 | ||
248 | for (;;) { | |
249 | req = nfs_page_find_head_request(page); | |
250 | if (!req) | |
251 | return req; | |
252 | head = nfs_page_group_lock_head(req); | |
253 | if (head != req) | |
254 | nfs_release_request(req); | |
255 | if (IS_ERR(head)) | |
256 | return head; | |
257 | ret = nfs_cancel_remove_inode(head, inode); | |
258 | if (ret < 0) { | |
259 | nfs_unlock_and_release_request(head); | |
260 | return ERR_PTR(ret); | |
261 | } | |
262 | /* Ensure that nobody removed the request before we locked it */ | |
263 | if (head == nfs_page_private_request(page)) | |
264 | break; | |
265 | if (PageSwapCache(page)) | |
266 | break; | |
267 | nfs_unlock_and_release_request(head); | |
268 | } | |
269 | return head; | |
270 | } | |
271 | ||
1da177e4 LT |
272 | /* Adjust the file length if we're writing beyond the end */ |
273 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
274 | { | |
d56b4ddf | 275 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
276 | loff_t end, i_size; |
277 | pgoff_t end_index; | |
1da177e4 | 278 | |
a3d01454 TM |
279 | spin_lock(&inode->i_lock); |
280 | i_size = i_size_read(inode); | |
09cbfeaf | 281 | end_index = (i_size - 1) >> PAGE_SHIFT; |
8cd79788 | 282 | if (i_size > 0 && page_index(page) < end_index) |
a3d01454 | 283 | goto out; |
d56b4ddf | 284 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 285 | if (i_size >= end) |
a3d01454 | 286 | goto out; |
110cb2d2 | 287 | trace_nfs_size_grow(inode, end); |
1da177e4 | 288 | i_size_write(inode, end); |
f6cdfa6d | 289 | NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; |
a3d01454 TM |
290 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
291 | out: | |
292 | spin_unlock(&inode->i_lock); | |
a6b5a28e | 293 | nfs_fscache_invalidate(inode, 0); |
1da177e4 LT |
294 | } |
295 | ||
a301b777 | 296 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
d2ceb7e5 | 297 | static void nfs_set_pageerror(struct address_space *mapping) |
a301b777 | 298 | { |
0df68ced TM |
299 | struct inode *inode = mapping->host; |
300 | ||
d2ceb7e5 | 301 | nfs_zap_mapping(mapping->host, mapping); |
0df68ced TM |
302 | /* Force file size revalidation */ |
303 | spin_lock(&inode->i_lock); | |
ac46b3d7 | 304 | nfs_set_cache_invalid(inode, NFS_INO_REVAL_FORCED | |
88a6099f | 305 | NFS_INO_INVALID_CHANGE | |
ac46b3d7 | 306 | NFS_INO_INVALID_SIZE); |
0df68ced | 307 | spin_unlock(&inode->i_lock); |
a301b777 TM |
308 | } |
309 | ||
6fbda89b TM |
310 | static void nfs_mapping_set_error(struct page *page, int error) |
311 | { | |
b8946d7b TM |
312 | struct address_space *mapping = page_file_mapping(page); |
313 | ||
6fbda89b | 314 | SetPageError(page); |
6c984083 TM |
315 | filemap_set_wb_err(mapping, error); |
316 | if (mapping->host) | |
317 | errseq_set(&mapping->host->i_sb->s_wb_err, | |
318 | error == -ENOSPC ? -ENOSPC : -EIO); | |
b8946d7b | 319 | nfs_set_pageerror(mapping); |
6fbda89b TM |
320 | } |
321 | ||
d72ddcba WAA |
322 | /* |
323 | * nfs_page_group_search_locked | |
324 | * @head - head request of page group | |
325 | * @page_offset - offset into page | |
326 | * | |
327 | * Search page group with head @head to find a request that contains the | |
328 | * page offset @page_offset. | |
329 | * | |
330 | * Returns a pointer to the first matching nfs request, or NULL if no | |
331 | * match is found. | |
332 | * | |
333 | * Must be called with the page group lock held | |
334 | */ | |
335 | static struct nfs_page * | |
336 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
337 | { | |
338 | struct nfs_page *req; | |
339 | ||
d72ddcba WAA |
340 | req = head; |
341 | do { | |
342 | if (page_offset >= req->wb_pgbase && | |
343 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
344 | return req; | |
345 | ||
346 | req = req->wb_this_page; | |
347 | } while (req != head); | |
348 | ||
349 | return NULL; | |
350 | } | |
351 | ||
352 | /* | |
353 | * nfs_page_group_covers_page | |
354 | * @head - head request of page group | |
355 | * | |
356 | * Return true if the page group with head @head covers the whole page, | |
357 | * returns false otherwise | |
358 | */ | |
359 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
360 | { | |
361 | struct nfs_page *tmp; | |
362 | unsigned int pos = 0; | |
363 | unsigned int len = nfs_page_length(req->wb_page); | |
364 | ||
1344b7ea | 365 | nfs_page_group_lock(req); |
d72ddcba | 366 | |
7e8a30f8 | 367 | for (;;) { |
d72ddcba | 368 | tmp = nfs_page_group_search_locked(req->wb_head, pos); |
7e8a30f8 TM |
369 | if (!tmp) |
370 | break; | |
371 | pos = tmp->wb_pgbase + tmp->wb_bytes; | |
372 | } | |
d72ddcba WAA |
373 | |
374 | nfs_page_group_unlock(req); | |
7e8a30f8 | 375 | return pos >= len; |
d72ddcba WAA |
376 | } |
377 | ||
1da177e4 LT |
378 | /* We can set the PG_uptodate flag if we see that a write request |
379 | * covers the full page. | |
380 | */ | |
d72ddcba | 381 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 382 | { |
d72ddcba | 383 | if (PageUptodate(req->wb_page)) |
1da177e4 | 384 | return; |
d72ddcba | 385 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 386 | return; |
d72ddcba | 387 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
388 | } |
389 | ||
1da177e4 LT |
390 | static int wb_priority(struct writeback_control *wbc) |
391 | { | |
e87b4c7a | 392 | int ret = 0; |
cca588d6 | 393 | |
e87b4c7a N |
394 | if (wbc->sync_mode == WB_SYNC_ALL) |
395 | ret = FLUSH_COND_STABLE; | |
e87b4c7a | 396 | return ret; |
1da177e4 LT |
397 | } |
398 | ||
89a09141 PZ |
399 | /* |
400 | * NFS congestion control | |
401 | */ | |
402 | ||
403 | int nfs_congestion_kb; | |
404 | ||
405 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
406 | #define NFS_CONGESTION_OFF_THRESH \ | |
407 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
408 | ||
deed85e7 | 409 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 410 | { |
0db10944 JK |
411 | struct inode *inode = page_file_mapping(page)->host; |
412 | struct nfs_server *nfss = NFS_SERVER(inode); | |
5a6d41b3 TM |
413 | int ret = test_set_page_writeback(page); |
414 | ||
deed85e7 | 415 | WARN_ON_ONCE(ret != 0); |
89a09141 | 416 | |
deed85e7 | 417 | if (atomic_long_inc_return(&nfss->writeback) > |
0db10944 | 418 | NFS_CONGESTION_ON_THRESH) |
6df25e58 | 419 | nfss->write_congested = 1; |
89a09141 PZ |
420 | } |
421 | ||
20633f04 | 422 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 423 | { |
20633f04 | 424 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 | 425 | struct nfs_server *nfss = NFS_SERVER(inode); |
31a01f09 | 426 | bool is_done; |
89a09141 | 427 | |
31a01f09 TM |
428 | is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); |
429 | nfs_unlock_request(req); | |
430 | if (!is_done) | |
20633f04 WAA |
431 | return; |
432 | ||
433 | end_page_writeback(req->wb_page); | |
c4dc4bee | 434 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
6df25e58 | 435 | nfss->write_congested = 0; |
89a09141 PZ |
436 | } |
437 | ||
d4581383 WAA |
438 | /* |
439 | * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests | |
440 | * | |
441 | * @destroy_list - request list (using wb_this_page) terminated by @old_head | |
442 | * @old_head - the old head of the list | |
443 | * | |
444 | * All subrequests must be locked and removed from all lists, so at this point | |
445 | * they are only "active" in this function, and possibly in nfs_wait_on_request | |
446 | * with a reference held by some other context. | |
447 | */ | |
448 | static void | |
449 | nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, | |
b66aaa8d TM |
450 | struct nfs_page *old_head, |
451 | struct inode *inode) | |
d4581383 WAA |
452 | { |
453 | while (destroy_list) { | |
454 | struct nfs_page *subreq = destroy_list; | |
455 | ||
456 | destroy_list = (subreq->wb_this_page == old_head) ? | |
457 | NULL : subreq->wb_this_page; | |
458 | ||
08ca8b21 TM |
459 | /* Note: lock subreq in order to change subreq->wb_head */ |
460 | nfs_page_set_headlock(subreq); | |
d4581383 WAA |
461 | WARN_ON_ONCE(old_head != subreq->wb_head); |
462 | ||
463 | /* make sure old group is not used */ | |
d4581383 | 464 | subreq->wb_this_page = subreq; |
08ca8b21 | 465 | subreq->wb_head = subreq; |
d4581383 | 466 | |
902a4c00 TM |
467 | clear_bit(PG_REMOVE, &subreq->wb_flags); |
468 | ||
5b2b5187 TM |
469 | /* Note: races with nfs_page_group_destroy() */ |
470 | if (!kref_read(&subreq->wb_kref)) { | |
5b2b5187 | 471 | /* Check if we raced with nfs_page_group_destroy() */ |
08ca8b21 TM |
472 | if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) { |
473 | nfs_page_clear_headlock(subreq); | |
5b2b5187 | 474 | nfs_free_request(subreq); |
08ca8b21 TM |
475 | } else |
476 | nfs_page_clear_headlock(subreq); | |
5b2b5187 TM |
477 | continue; |
478 | } | |
08ca8b21 | 479 | nfs_page_clear_headlock(subreq); |
d4581383 | 480 | |
add42de3 | 481 | nfs_release_request(old_head); |
5b2b5187 TM |
482 | |
483 | if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { | |
484 | nfs_release_request(subreq); | |
a6b6d5b8 | 485 | atomic_long_dec(&NFS_I(inode)->nrequests); |
d4581383 | 486 | } |
5b2b5187 | 487 | |
5b2b5187 TM |
488 | /* subreq is now totally disconnected from page group or any |
489 | * write / commit lists. last chance to wake any waiters */ | |
490 | nfs_unlock_and_release_request(subreq); | |
d4581383 WAA |
491 | } |
492 | } | |
493 | ||
494 | /* | |
e00ed89d TM |
495 | * nfs_join_page_group - destroy subrequests of the head req |
496 | * @head: the page used to lookup the "page group" of nfs_page structures | |
497 | * @inode: Inode to which the request belongs. | |
d4581383 WAA |
498 | * |
499 | * This function joins all sub requests to the head request by first | |
500 | * locking all requests in the group, cancelling any pending operations | |
501 | * and finally updating the head request to cover the whole range covered by | |
502 | * the (former) group. All subrequests are removed from any write or commit | |
503 | * lists, unlinked from the group and destroyed. | |
d4581383 | 504 | */ |
ed5d588f | 505 | void |
e00ed89d | 506 | nfs_join_page_group(struct nfs_page *head, struct inode *inode) |
e261f51f | 507 | { |
e00ed89d | 508 | struct nfs_page *subreq; |
d4581383 | 509 | struct nfs_page *destroy_list = NULL; |
a62f8e3b | 510 | unsigned int pgbase, off, bytes; |
a62f8e3b TM |
511 | |
512 | pgbase = head->wb_pgbase; | |
513 | bytes = head->wb_bytes; | |
514 | off = head->wb_offset; | |
a0e265bc TM |
515 | for (subreq = head->wb_this_page; subreq != head; |
516 | subreq = subreq->wb_this_page) { | |
a62f8e3b TM |
517 | /* Subrequests should always form a contiguous range */ |
518 | if (pgbase > subreq->wb_pgbase) { | |
519 | off -= pgbase - subreq->wb_pgbase; | |
520 | bytes += pgbase - subreq->wb_pgbase; | |
521 | pgbase = subreq->wb_pgbase; | |
309a1d65 | 522 | } |
a62f8e3b TM |
523 | bytes = max(subreq->wb_pgbase + subreq->wb_bytes |
524 | - pgbase, bytes); | |
a0e265bc | 525 | } |
d4581383 | 526 | |
a62f8e3b TM |
527 | /* Set the head request's range to cover the former page group */ |
528 | head->wb_pgbase = pgbase; | |
529 | head->wb_bytes = bytes; | |
530 | head->wb_offset = off; | |
531 | ||
d4581383 WAA |
532 | /* Now that all requests are locked, make sure they aren't on any list. |
533 | * Commit list removal accounting is done after locks are dropped */ | |
534 | subreq = head; | |
535 | do { | |
411a99ad | 536 | nfs_clear_request_commit(subreq); |
d4581383 WAA |
537 | subreq = subreq->wb_this_page; |
538 | } while (subreq != head); | |
539 | ||
540 | /* unlink subrequests from head, destroy them later */ | |
541 | if (head->wb_this_page != head) { | |
542 | /* destroy list will be terminated by head */ | |
543 | destroy_list = head->wb_this_page; | |
544 | head->wb_this_page = head; | |
e261f51f | 545 | } |
d4581383 | 546 | |
e00ed89d TM |
547 | nfs_destroy_unlinked_subrequests(destroy_list, head, inode); |
548 | } | |
b66aaa8d | 549 | |
e00ed89d TM |
550 | /* |
551 | * nfs_lock_and_join_requests - join all subreqs to the head req | |
552 | * @page: the page used to lookup the "page group" of nfs_page structures | |
553 | * | |
554 | * This function joins all sub requests to the head request by first | |
555 | * locking all requests in the group, cancelling any pending operations | |
556 | * and finally updating the head request to cover the whole range covered by | |
557 | * the (former) group. All subrequests are removed from any write or commit | |
558 | * lists, unlinked from the group and destroyed. | |
559 | * | |
560 | * Returns a locked, referenced pointer to the head request - which after | |
561 | * this call is guaranteed to be the only request associated with the page. | |
562 | * Returns NULL if no requests are found for @page, or a ERR_PTR if an | |
563 | * error was encountered. | |
564 | */ | |
565 | static struct nfs_page * | |
566 | nfs_lock_and_join_requests(struct page *page) | |
567 | { | |
568 | struct inode *inode = page_file_mapping(page)->host; | |
569 | struct nfs_page *head; | |
570 | int ret; | |
d4581383 | 571 | |
e00ed89d TM |
572 | /* |
573 | * A reference is taken only on the head request which acts as a | |
574 | * reference to the whole page group - the group will not be destroyed | |
575 | * until the head reference is released. | |
576 | */ | |
577 | head = nfs_find_and_lock_page_request(page); | |
578 | if (IS_ERR_OR_NULL(head)) | |
579 | return head; | |
d4581383 | 580 | |
e00ed89d TM |
581 | /* lock each request in the page group */ |
582 | ret = nfs_page_group_lock_subrequests(head); | |
583 | if (ret < 0) { | |
b5bab9bf | 584 | nfs_unlock_and_release_request(head); |
e00ed89d | 585 | return ERR_PTR(ret); |
b5bab9bf TM |
586 | } |
587 | ||
e00ed89d | 588 | nfs_join_page_group(head, inode); |
0671d8f1 | 589 | |
e00ed89d | 590 | return head; |
074cc1de TM |
591 | } |
592 | ||
6fbda89b | 593 | static void nfs_write_error(struct nfs_page *req, int error) |
0bcbf039 | 594 | { |
af887e43 TM |
595 | trace_nfs_write_error(page_file_mapping(req->wb_page)->host, req, |
596 | error); | |
6fbda89b | 597 | nfs_mapping_set_error(req->wb_page, error); |
06c9fdf3 | 598 | nfs_inode_remove_request(req); |
0bcbf039 | 599 | nfs_end_page_writeback(req); |
1f84ccdf | 600 | nfs_release_request(req); |
0bcbf039 PT |
601 | } |
602 | ||
074cc1de TM |
603 | /* |
604 | * Find an associated nfs write request, and prepare to flush it out | |
605 | * May return an error if the user signalled nfs_wait_on_request(). | |
606 | */ | |
c6fd3511 TM |
607 | static int nfs_page_async_flush(struct page *page, |
608 | struct writeback_control *wbc, | |
609 | struct nfs_pageio_descriptor *pgio) | |
074cc1de TM |
610 | { |
611 | struct nfs_page *req; | |
612 | int ret = 0; | |
613 | ||
6d17d653 | 614 | req = nfs_lock_and_join_requests(page); |
074cc1de TM |
615 | if (!req) |
616 | goto out; | |
617 | ret = PTR_ERR(req); | |
618 | if (IS_ERR(req)) | |
619 | goto out; | |
620 | ||
deed85e7 TM |
621 | nfs_set_page_writeback(page); |
622 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 623 | |
a6598813 | 624 | /* If there is a fatal error that covers this write, just exit */ |
96c41455 TM |
625 | ret = pgio->pg_error; |
626 | if (nfs_error_is_fatal_on_server(ret)) | |
a6598813 TM |
627 | goto out_launder; |
628 | ||
96c41455 | 629 | ret = 0; |
f8512ad0 | 630 | if (!nfs_pageio_add_request(pgio, req)) { |
074cc1de | 631 | ret = pgio->pg_error; |
0bcbf039 | 632 | /* |
c373fff7 | 633 | * Remove the problematic req upon fatal errors on the server |
0bcbf039 | 634 | */ |
c6fd3511 TM |
635 | if (nfs_error_is_fatal_on_server(ret)) |
636 | goto out_launder; | |
637 | if (wbc->sync_mode == WB_SYNC_NONE) | |
638 | ret = AOP_WRITEPAGE_ACTIVATE; | |
639 | redirty_page_for_writepage(wbc, page); | |
d6c843b9 | 640 | nfs_redirty_request(req); |
96c41455 | 641 | pgio->pg_error = 0; |
40f90271 TM |
642 | } else |
643 | nfs_add_stats(page_file_mapping(page)->host, | |
644 | NFSIOS_WRITEPAGES, 1); | |
074cc1de TM |
645 | out: |
646 | return ret; | |
a6598813 | 647 | out_launder: |
6fbda89b | 648 | nfs_write_error(req, ret); |
14bebe3c | 649 | return 0; |
e261f51f TM |
650 | } |
651 | ||
d6c843b9 | 652 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, |
c373fff7 | 653 | struct nfs_pageio_descriptor *pgio) |
1da177e4 | 654 | { |
8cd79788 | 655 | nfs_pageio_cond_complete(pgio, page_index(page)); |
c6fd3511 | 656 | return nfs_page_async_flush(page, wbc, pgio); |
f758c885 | 657 | } |
7fe7f848 | 658 | |
f758c885 TM |
659 | /* |
660 | * Write an mmapped page to the server. | |
661 | */ | |
d6c843b9 | 662 | static int nfs_writepage_locked(struct page *page, |
c373fff7 | 663 | struct writeback_control *wbc) |
f758c885 TM |
664 | { |
665 | struct nfs_pageio_descriptor pgio; | |
40f90271 | 666 | struct inode *inode = page_file_mapping(page)->host; |
f758c885 | 667 | int err; |
49a70f27 | 668 | |
6df25e58 N |
669 | if (wbc->sync_mode == WB_SYNC_NONE && |
670 | NFS_SERVER(inode)->write_congested) | |
671 | return AOP_WRITEPAGE_ACTIVATE; | |
672 | ||
40f90271 | 673 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
811ed92e | 674 | nfs_pageio_init_write(&pgio, inode, 0, |
a20c93e3 | 675 | false, &nfs_async_write_completion_ops); |
c373fff7 | 676 | err = nfs_do_writepage(page, wbc, &pgio); |
96c41455 | 677 | pgio.pg_error = 0; |
f758c885 | 678 | nfs_pageio_complete(&pgio); |
c5e483b7 | 679 | return err; |
4d770ccf TM |
680 | } |
681 | ||
682 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
683 | { | |
f758c885 | 684 | int ret; |
4d770ccf | 685 | |
c373fff7 | 686 | ret = nfs_writepage_locked(page, wbc); |
96c41455 TM |
687 | if (ret != AOP_WRITEPAGE_ACTIVATE) |
688 | unlock_page(page); | |
f758c885 TM |
689 | return ret; |
690 | } | |
691 | ||
692 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
693 | { | |
694 | int ret; | |
695 | ||
c373fff7 | 696 | ret = nfs_do_writepage(page, wbc, data); |
96c41455 TM |
697 | if (ret != AOP_WRITEPAGE_ACTIVATE) |
698 | unlock_page(page); | |
f758c885 | 699 | return ret; |
1da177e4 LT |
700 | } |
701 | ||
919e3bd9 TM |
702 | static void nfs_io_completion_commit(void *inode) |
703 | { | |
704 | nfs_commit_inode(inode, 0); | |
705 | } | |
706 | ||
1da177e4 LT |
707 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
708 | { | |
1da177e4 | 709 | struct inode *inode = mapping->host; |
c63c7b05 | 710 | struct nfs_pageio_descriptor pgio; |
ed7bcdb3 TM |
711 | struct nfs_io_completion *ioc = NULL; |
712 | unsigned int mntflags = NFS_SERVER(inode)->flags; | |
713 | int priority = 0; | |
1da177e4 LT |
714 | int err; |
715 | ||
6df25e58 N |
716 | if (wbc->sync_mode == WB_SYNC_NONE && |
717 | NFS_SERVER(inode)->write_congested) | |
718 | return 0; | |
719 | ||
91d5b470 CL |
720 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
721 | ||
ed7bcdb3 TM |
722 | if (!(mntflags & NFS_MOUNT_WRITE_EAGER) || wbc->for_kupdate || |
723 | wbc->for_background || wbc->for_sync || wbc->for_reclaim) { | |
724 | ioc = nfs_io_completion_alloc(GFP_KERNEL); | |
725 | if (ioc) | |
726 | nfs_io_completion_init(ioc, nfs_io_completion_commit, | |
727 | inode); | |
728 | priority = wb_priority(wbc); | |
729 | } | |
919e3bd9 | 730 | |
c6fd3511 TM |
731 | do { |
732 | nfs_pageio_init_write(&pgio, inode, priority, false, | |
733 | &nfs_async_write_completion_ops); | |
734 | pgio.pg_io_completion = ioc; | |
735 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, | |
736 | &pgio); | |
737 | pgio.pg_error = 0; | |
738 | nfs_pageio_complete(&pgio); | |
739 | } while (err < 0 && !nfs_error_is_fatal(err)); | |
919e3bd9 | 740 | nfs_io_completion_put(ioc); |
72cb77f4 | 741 | |
f758c885 | 742 | if (err < 0) |
72cb77f4 | 743 | goto out_err; |
c63c7b05 | 744 | return 0; |
72cb77f4 TM |
745 | out_err: |
746 | return err; | |
1da177e4 LT |
747 | } |
748 | ||
749 | /* | |
750 | * Insert a write request into an inode | |
751 | */ | |
d6d6dc7c | 752 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 | 753 | { |
4b9bb25b | 754 | struct address_space *mapping = page_file_mapping(req->wb_page); |
1da177e4 | 755 | struct nfs_inode *nfsi = NFS_I(inode); |
e7d39069 | 756 | |
2bfc6e56 WAA |
757 | WARN_ON_ONCE(req->wb_this_page != req); |
758 | ||
e7d39069 | 759 | /* Lock the request! */ |
7ad84aa9 | 760 | nfs_lock_request(req); |
e7d39069 | 761 | |
29418aa4 MG |
762 | /* |
763 | * Swap-space should not get truncated. Hence no need to plug the race | |
764 | * with invalidate/truncate. | |
765 | */ | |
4b9bb25b | 766 | spin_lock(&mapping->private_lock); |
29418aa4 MG |
767 | if (likely(!PageSwapCache(req->wb_page))) { |
768 | set_bit(PG_MAPPED, &req->wb_flags); | |
769 | SetPagePrivate(req->wb_page); | |
770 | set_page_private(req->wb_page, (unsigned long)req); | |
771 | } | |
4b9bb25b | 772 | spin_unlock(&mapping->private_lock); |
a6b6d5b8 | 773 | atomic_long_inc(&nfsi->nrequests); |
17089a29 | 774 | /* this a head request for a page group - mark it as having an |
cb1410c7 WAA |
775 | * extra reference so sub groups can follow suit. |
776 | * This flag also informs pgio layer when to bump nrequests when | |
777 | * adding subrequests. */ | |
17089a29 | 778 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); |
c03b4024 | 779 | kref_get(&req->wb_kref); |
1da177e4 LT |
780 | } |
781 | ||
782 | /* | |
89a09141 | 783 | * Remove a write request from an inode |
1da177e4 LT |
784 | */ |
785 | static void nfs_inode_remove_request(struct nfs_page *req) | |
786 | { | |
4b9bb25b TM |
787 | struct address_space *mapping = page_file_mapping(req->wb_page); |
788 | struct inode *inode = mapping->host; | |
1da177e4 | 789 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 790 | struct nfs_page *head; |
1da177e4 | 791 | |
20633f04 WAA |
792 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
793 | head = req->wb_head; | |
794 | ||
4b9bb25b | 795 | spin_lock(&mapping->private_lock); |
67911c8f | 796 | if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { |
20633f04 WAA |
797 | set_page_private(head->wb_page, 0); |
798 | ClearPagePrivate(head->wb_page); | |
799 | clear_bit(PG_MAPPED, &head->wb_flags); | |
800 | } | |
4b9bb25b | 801 | spin_unlock(&mapping->private_lock); |
29418aa4 | 802 | } |
17089a29 | 803 | |
33ea5aaa | 804 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) { |
17089a29 | 805 | nfs_release_request(req); |
33ea5aaa Z |
806 | atomic_long_dec(&nfsi->nrequests); |
807 | } | |
1da177e4 LT |
808 | } |
809 | ||
61822ab5 | 810 | static void |
6d884e8f | 811 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 812 | { |
67911c8f AS |
813 | if (req->wb_page) |
814 | __set_page_dirty_nobuffers(req->wb_page); | |
61822ab5 TM |
815 | } |
816 | ||
3a3908c8 TM |
817 | /* |
818 | * nfs_page_search_commits_for_head_request_locked | |
819 | * | |
820 | * Search through commit lists on @inode for the head request for @page. | |
821 | * Must be called while holding the inode (which is cinfo) lock. | |
822 | * | |
823 | * Returns the head request if found, or NULL if not found. | |
824 | */ | |
825 | static struct nfs_page * | |
826 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
827 | struct page *page) | |
828 | { | |
829 | struct nfs_page *freq, *t; | |
830 | struct nfs_commit_info cinfo; | |
831 | struct inode *inode = &nfsi->vfs_inode; | |
832 | ||
833 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
834 | ||
835 | /* search through pnfs commit lists */ | |
836 | freq = pnfs_search_commit_reqs(inode, &cinfo, page); | |
837 | if (freq) | |
838 | return freq->wb_head; | |
839 | ||
840 | /* Linearly search the commit list for the correct request */ | |
841 | list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { | |
842 | if (freq->wb_page == page) | |
843 | return freq->wb_head; | |
844 | } | |
845 | ||
846 | return NULL; | |
847 | } | |
848 | ||
86d80f97 TM |
849 | /** |
850 | * nfs_request_add_commit_list_locked - add request to a commit list | |
851 | * @req: pointer to a struct nfs_page | |
852 | * @dst: commit list head | |
853 | * @cinfo: holds list lock and accounting info | |
854 | * | |
855 | * This sets the PG_CLEAN bit, updates the cinfo count of | |
856 | * number of outstanding requests requiring a commit as well as | |
857 | * the MM page stats. | |
858 | * | |
e824f99a TM |
859 | * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the |
860 | * nfs_page lock. | |
86d80f97 TM |
861 | */ |
862 | void | |
863 | nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, | |
864 | struct nfs_commit_info *cinfo) | |
865 | { | |
866 | set_bit(PG_CLEAN, &req->wb_flags); | |
867 | nfs_list_add_request(req, dst); | |
5cb953d4 | 868 | atomic_long_inc(&cinfo->mds->ncommit); |
86d80f97 TM |
869 | } |
870 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); | |
871 | ||
8dd37758 TM |
872 | /** |
873 | * nfs_request_add_commit_list - add request to a commit list | |
874 | * @req: pointer to a struct nfs_page | |
ea2cf228 | 875 | * @cinfo: holds list lock and accounting info |
8dd37758 | 876 | * |
ea2cf228 | 877 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
878 | * number of outstanding requests requiring a commit as well as |
879 | * the MM page stats. | |
880 | * | |
ea2cf228 | 881 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 882 | * holding the nfs_page lock. |
1da177e4 | 883 | */ |
8dd37758 | 884 | void |
6272dcc6 | 885 | nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) |
1da177e4 | 886 | { |
e824f99a | 887 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
6272dcc6 | 888 | nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); |
e824f99a | 889 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
67911c8f AS |
890 | if (req->wb_page) |
891 | nfs_mark_page_unstable(req->wb_page, cinfo); | |
1da177e4 | 892 | } |
8dd37758 TM |
893 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
894 | ||
895 | /** | |
896 | * nfs_request_remove_commit_list - Remove request from a commit list | |
897 | * @req: pointer to a nfs_page | |
ea2cf228 | 898 | * @cinfo: holds list lock and accounting info |
8dd37758 | 899 | * |
ea2cf228 | 900 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
901 | * number of outstanding requests requiring a commit |
902 | * It does not update the MM page stats. | |
903 | * | |
ea2cf228 | 904 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
905 | */ |
906 | void | |
ea2cf228 FI |
907 | nfs_request_remove_commit_list(struct nfs_page *req, |
908 | struct nfs_commit_info *cinfo) | |
8dd37758 | 909 | { |
8dd37758 TM |
910 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
911 | return; | |
912 | nfs_list_remove_request(req); | |
5cb953d4 | 913 | atomic_long_dec(&cinfo->mds->ncommit); |
8dd37758 TM |
914 | } |
915 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
916 | ||
ea2cf228 FI |
917 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
918 | struct inode *inode) | |
919 | { | |
fe238e60 | 920 | cinfo->inode = inode; |
ea2cf228 FI |
921 | cinfo->mds = &NFS_I(inode)->commit_info; |
922 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 923 | cinfo->dreq = NULL; |
f453a54a | 924 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
925 | } |
926 | ||
927 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
928 | struct inode *inode, | |
929 | struct nfs_direct_req *dreq) | |
930 | { | |
1763da12 FI |
931 | if (dreq) |
932 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
933 | else | |
934 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
935 | } |
936 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
937 | |
938 | /* | |
939 | * Add a request to the inode's commit list. | |
940 | */ | |
1763da12 | 941 | void |
ea2cf228 | 942 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
b57ff130 | 943 | struct nfs_commit_info *cinfo, u32 ds_commit_idx) |
8dd37758 | 944 | { |
b57ff130 | 945 | if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) |
8dd37758 | 946 | return; |
6272dcc6 | 947 | nfs_request_add_commit_list(req, cinfo); |
8dd37758 | 948 | } |
8e821cad | 949 | |
d6d6dc7c FI |
950 | static void |
951 | nfs_clear_page_commit(struct page *page) | |
952 | { | |
8d92890b | 953 | dec_node_page_state(page, NR_WRITEBACK); |
93f78d88 | 954 | dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, |
8d92890b | 955 | WB_WRITEBACK); |
d6d6dc7c FI |
956 | } |
957 | ||
b5bab9bf | 958 | /* Called holding the request lock on @req */ |
8dd37758 | 959 | static void |
e468bae9 TM |
960 | nfs_clear_request_commit(struct nfs_page *req) |
961 | { | |
8dd37758 | 962 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
9fcd5960 TM |
963 | struct nfs_open_context *ctx = nfs_req_openctx(req); |
964 | struct inode *inode = d_inode(ctx->dentry); | |
ea2cf228 | 965 | struct nfs_commit_info cinfo; |
e468bae9 | 966 | |
ea2cf228 | 967 | nfs_init_cinfo_from_inode(&cinfo, inode); |
e824f99a | 968 | mutex_lock(&NFS_I(inode)->commit_mutex); |
ea2cf228 | 969 | if (!pnfs_clear_request_commit(req, &cinfo)) { |
ea2cf228 | 970 | nfs_request_remove_commit_list(req, &cinfo); |
8dd37758 | 971 | } |
e824f99a | 972 | mutex_unlock(&NFS_I(inode)->commit_mutex); |
d6d6dc7c | 973 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 974 | } |
e468bae9 TM |
975 | } |
976 | ||
d45f60c6 | 977 | int nfs_write_need_commit(struct nfs_pgio_header *hdr) |
8e821cad | 978 | { |
c65e6254 | 979 | if (hdr->verf.committed == NFS_DATA_SYNC) |
d45f60c6 | 980 | return hdr->lseg == NULL; |
c65e6254 | 981 | return hdr->verf.committed != NFS_FILE_SYNC; |
8e821cad TM |
982 | } |
983 | ||
919e3bd9 TM |
984 | static void nfs_async_write_init(struct nfs_pgio_header *hdr) |
985 | { | |
986 | nfs_io_completion_get(hdr->io_completion); | |
987 | } | |
988 | ||
061ae2ed | 989 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 990 | { |
ea2cf228 | 991 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
992 | unsigned long bytes = 0; |
993 | ||
994 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
995 | goto out; | |
ea2cf228 | 996 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
997 | while (!list_empty(&hdr->pages)) { |
998 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
999 | |
1000 | bytes += req->wb_bytes; | |
1001 | nfs_list_remove_request(req); | |
1002 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
1003 | (hdr->good_bytes < bytes)) { | |
af887e43 | 1004 | trace_nfs_comp_error(hdr->inode, req, hdr->error); |
6fbda89b | 1005 | nfs_mapping_set_error(req->wb_page, hdr->error); |
6c75dc0d FI |
1006 | goto remove_req; |
1007 | } | |
c65e6254 | 1008 | if (nfs_write_need_commit(hdr)) { |
33344e0f TM |
1009 | /* Reset wb_nio, since the write was successful. */ |
1010 | req->wb_nio = 0; | |
f79d06f5 | 1011 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
b57ff130 | 1012 | nfs_mark_request_commit(req, hdr->lseg, &cinfo, |
a7d42ddb | 1013 | hdr->pgio_mirror_idx); |
6c75dc0d FI |
1014 | goto next; |
1015 | } | |
1016 | remove_req: | |
1017 | nfs_inode_remove_request(req); | |
1018 | next: | |
20633f04 | 1019 | nfs_end_page_writeback(req); |
3aff4ebb | 1020 | nfs_release_request(req); |
6c75dc0d FI |
1021 | } |
1022 | out: | |
919e3bd9 | 1023 | nfs_io_completion_put(hdr->io_completion); |
6c75dc0d | 1024 | hdr->release(hdr); |
8e821cad | 1025 | } |
1da177e4 | 1026 | |
ce59515c | 1027 | unsigned long |
ea2cf228 | 1028 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 1029 | { |
5cb953d4 | 1030 | return atomic_long_read(&cinfo->mds->ncommit); |
d6d6dc7c FI |
1031 | } |
1032 | ||
e824f99a | 1033 | /* NFS_I(cinfo->inode)->commit_mutex held by caller */ |
1763da12 | 1034 | int |
ea2cf228 FI |
1035 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
1036 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c | 1037 | { |
137da553 | 1038 | struct nfs_page *req, *tmp; |
d6d6dc7c FI |
1039 | int ret = 0; |
1040 | ||
137da553 | 1041 | list_for_each_entry_safe(req, tmp, src, wb_list) { |
7ad84aa9 | 1042 | kref_get(&req->wb_kref); |
2ce209c4 | 1043 | if (!nfs_lock_request(req)) { |
2ce209c4 | 1044 | nfs_release_request(req); |
64a93dbf | 1045 | continue; |
2ce209c4 | 1046 | } |
ea2cf228 | 1047 | nfs_request_remove_commit_list(req, cinfo); |
5d2a9d9d | 1048 | clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); |
8dd37758 TM |
1049 | nfs_list_add_request(req, dst); |
1050 | ret++; | |
1763da12 | 1051 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 1052 | break; |
e824f99a | 1053 | cond_resched(); |
d6d6dc7c FI |
1054 | } |
1055 | return ret; | |
fb8a1f11 | 1056 | } |
5d2a9d9d | 1057 | EXPORT_SYMBOL_GPL(nfs_scan_commit_list); |
fb8a1f11 | 1058 | |
1da177e4 LT |
1059 | /* |
1060 | * nfs_scan_commit - Scan an inode for commit requests | |
1061 | * @inode: NFS inode to scan | |
ea2cf228 FI |
1062 | * @dst: mds destination list |
1063 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
1064 | * |
1065 | * Moves requests from the inode's 'commit' request list. | |
1066 | * The requests are *not* checked to ensure that they form a contiguous set. | |
1067 | */ | |
1763da12 | 1068 | int |
ea2cf228 FI |
1069 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
1070 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1071 | { |
d6d6dc7c | 1072 | int ret = 0; |
fb8a1f11 | 1073 | |
5cb953d4 TM |
1074 | if (!atomic_long_read(&cinfo->mds->ncommit)) |
1075 | return 0; | |
e824f99a | 1076 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
5cb953d4 | 1077 | if (atomic_long_read(&cinfo->mds->ncommit) > 0) { |
8dd37758 | 1078 | const int max = INT_MAX; |
d6d6dc7c | 1079 | |
ea2cf228 FI |
1080 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
1081 | cinfo, max); | |
1082 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 1083 | } |
e824f99a | 1084 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
ff778d02 | 1085 | return ret; |
1da177e4 | 1086 | } |
d6d6dc7c | 1087 | |
1da177e4 | 1088 | /* |
e7d39069 TM |
1089 | * Search for an existing write request, and attempt to update |
1090 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 1091 | * |
e7d39069 TM |
1092 | * If the attempt fails, then the existing request is flushed out |
1093 | * to disk. | |
1da177e4 | 1094 | */ |
e7d39069 TM |
1095 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
1096 | struct page *page, | |
1097 | unsigned int offset, | |
1098 | unsigned int bytes) | |
1da177e4 | 1099 | { |
e7d39069 TM |
1100 | struct nfs_page *req; |
1101 | unsigned int rqend; | |
1102 | unsigned int end; | |
1103 | int error; | |
1104 | ||
1da177e4 LT |
1105 | end = offset + bytes; |
1106 | ||
f6032f21 TM |
1107 | req = nfs_lock_and_join_requests(page); |
1108 | if (IS_ERR_OR_NULL(req)) | |
1109 | return req; | |
1da177e4 | 1110 | |
f6032f21 TM |
1111 | rqend = req->wb_offset + req->wb_bytes; |
1112 | /* | |
1113 | * Tell the caller to flush out the request if | |
1114 | * the offsets are non-contiguous. | |
1115 | * Note: nfs_flush_incompatible() will already | |
1116 | * have flushed out requests having wrong owners. | |
1117 | */ | |
1118 | if (offset > rqend || end < req->wb_offset) | |
1119 | goto out_flushme; | |
1da177e4 LT |
1120 | |
1121 | /* Okay, the request matches. Update the region */ | |
1122 | if (offset < req->wb_offset) { | |
1123 | req->wb_offset = offset; | |
1124 | req->wb_pgbase = offset; | |
1da177e4 | 1125 | } |
1da177e4 LT |
1126 | if (end > rqend) |
1127 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1128 | else |
1129 | req->wb_bytes = rqend - req->wb_offset; | |
33344e0f | 1130 | req->wb_nio = 0; |
e7d39069 TM |
1131 | return req; |
1132 | out_flushme: | |
f6032f21 TM |
1133 | /* |
1134 | * Note: we mark the request dirty here because | |
1135 | * nfs_lock_and_join_requests() cannot preserve | |
1136 | * commit flags, so we have to replay the write. | |
1137 | */ | |
1138 | nfs_mark_request_dirty(req); | |
1139 | nfs_unlock_and_release_request(req); | |
e7d39069 | 1140 | error = nfs_wb_page(inode, page); |
f6032f21 | 1141 | return (error < 0) ? ERR_PTR(error) : NULL; |
e7d39069 TM |
1142 | } |
1143 | ||
1144 | /* | |
1145 | * Try to update an existing write request, or create one if there is none. | |
1146 | * | |
1147 | * Note: Should always be called with the Page Lock held to prevent races | |
1148 | * if we have to add a new request. Also assumes that the caller has | |
1149 | * already called nfs_flush_incompatible() if necessary. | |
1150 | */ | |
1151 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1152 | struct page *page, unsigned int offset, unsigned int bytes) | |
1153 | { | |
d56b4ddf | 1154 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1155 | struct nfs_page *req; |
1da177e4 | 1156 | |
e7d39069 TM |
1157 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1158 | if (req != NULL) | |
1159 | goto out; | |
28b1d3f5 | 1160 | req = nfs_create_request(ctx, page, offset, bytes); |
e7d39069 TM |
1161 | if (IS_ERR(req)) |
1162 | goto out; | |
d6d6dc7c | 1163 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1164 | out: |
61e930a9 | 1165 | return req; |
1da177e4 LT |
1166 | } |
1167 | ||
e7d39069 TM |
1168 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1169 | unsigned int offset, unsigned int count) | |
1170 | { | |
1171 | struct nfs_page *req; | |
1172 | ||
1173 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1174 | if (IS_ERR(req)) | |
1175 | return PTR_ERR(req); | |
1176 | /* Update file length */ | |
1177 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1178 | nfs_mark_uptodate(req); |
a6305ddb | 1179 | nfs_mark_request_dirty(req); |
1d1afcbc | 1180 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1181 | return 0; |
1182 | } | |
1183 | ||
1da177e4 LT |
1184 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1185 | { | |
cd3758e3 | 1186 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1187 | struct nfs_lock_context *l_ctx; |
bd61e0a9 | 1188 | struct file_lock_context *flctx = file_inode(file)->i_flctx; |
1da177e4 | 1189 | struct nfs_page *req; |
1a54533e | 1190 | int do_flush, status; |
1da177e4 LT |
1191 | /* |
1192 | * Look for a request corresponding to this page. If there | |
1193 | * is one, and it belongs to another file, we flush it out | |
1194 | * before we try to copy anything into the page. Do this | |
1195 | * due to the lack of an ACCESS-type call in NFSv2. | |
1196 | * Also do the same if we find a request from an existing | |
1197 | * dropped page. | |
1198 | */ | |
1a54533e | 1199 | do { |
84d3a9a9 | 1200 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1201 | if (req == NULL) |
1202 | return 0; | |
2a369153 | 1203 | l_ctx = req->wb_lock_context; |
138a2935 | 1204 | do_flush = req->wb_page != page || |
9fcd5960 | 1205 | !nfs_match_open_context(nfs_req_openctx(req), ctx); |
bd61e0a9 JL |
1206 | if (l_ctx && flctx && |
1207 | !(list_empty_careful(&flctx->flc_posix) && | |
1208 | list_empty_careful(&flctx->flc_flock))) { | |
d51fdb87 | 1209 | do_flush |= l_ctx->lockowner != current->files; |
5263e31e | 1210 | } |
1da177e4 | 1211 | nfs_release_request(req); |
1a54533e TM |
1212 | if (!do_flush) |
1213 | return 0; | |
d56b4ddf | 1214 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1215 | } while (status == 0); |
1216 | return status; | |
1da177e4 LT |
1217 | } |
1218 | ||
dc24826b AA |
1219 | /* |
1220 | * Avoid buffered writes when a open context credential's key would | |
1221 | * expire soon. | |
1222 | * | |
1223 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1224 | * | |
1225 | * Return 0 and set a credential flag which triggers the inode to flush | |
1226 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1227 | * RPC_KEY_EXPIRE_TIMEO. | |
1228 | */ | |
1229 | int | |
1230 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1231 | { | |
1232 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
dc24826b | 1233 | |
ddf529ee | 1234 | if (nfs_ctx_key_to_expire(ctx, inode) && |
ca05cbae | 1235 | !rcu_access_pointer(ctx->ll_cred)) |
ddf529ee N |
1236 | /* Already expired! */ |
1237 | return -EACCES; | |
1238 | return 0; | |
dc24826b AA |
1239 | } |
1240 | ||
1241 | /* | |
1242 | * Test if the open context credential key is marked to expire soon. | |
1243 | */ | |
ce52914e | 1244 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) |
dc24826b | 1245 | { |
ce52914e | 1246 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; |
ca05cbae | 1247 | struct rpc_cred *cred, *new, *old = NULL; |
ddf529ee | 1248 | struct auth_cred acred = { |
a52458b4 | 1249 | .cred = ctx->cred, |
ddf529ee | 1250 | }; |
ca05cbae | 1251 | bool ret = false; |
ddf529ee | 1252 | |
ca05cbae TM |
1253 | rcu_read_lock(); |
1254 | cred = rcu_dereference(ctx->ll_cred); | |
1255 | if (cred && !(cred->cr_ops->crkey_timeout && | |
1256 | cred->cr_ops->crkey_timeout(cred))) | |
1257 | goto out; | |
1258 | rcu_read_unlock(); | |
1259 | ||
1260 | new = auth->au_ops->lookup_cred(auth, &acred, 0); | |
1261 | if (new == cred) { | |
1262 | put_rpccred(new); | |
ddf529ee | 1263 | return true; |
ca05cbae TM |
1264 | } |
1265 | if (IS_ERR_OR_NULL(new)) { | |
1266 | new = NULL; | |
1267 | ret = true; | |
1268 | } else if (new->cr_ops->crkey_timeout && | |
1269 | new->cr_ops->crkey_timeout(new)) | |
1270 | ret = true; | |
1271 | ||
1272 | rcu_read_lock(); | |
1273 | old = rcu_dereference_protected(xchg(&ctx->ll_cred, | |
1274 | RCU_INITIALIZER(new)), 1); | |
1275 | out: | |
1276 | rcu_read_unlock(); | |
1277 | put_rpccred(old); | |
1278 | return ret; | |
dc24826b AA |
1279 | } |
1280 | ||
5d47a356 TM |
1281 | /* |
1282 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1283 | * the PageUptodate() flag. In this case, we will need to turn off | |
1284 | * write optimisations that depend on the page contents being correct. | |
1285 | */ | |
fc9dc401 TM |
1286 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode, |
1287 | unsigned int pagelen) | |
5d47a356 | 1288 | { |
d529ef83 JL |
1289 | struct nfs_inode *nfsi = NFS_I(inode); |
1290 | ||
8d197a56 TM |
1291 | if (nfs_have_delegated_attributes(inode)) |
1292 | goto out; | |
fc9dc401 | 1293 | if (nfsi->cache_validity & |
13c0b082 | 1294 | (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE)) |
d529ef83 | 1295 | return false; |
4db72b40 | 1296 | smp_rmb(); |
fc9dc401 | 1297 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags) && pagelen != 0) |
8d197a56 TM |
1298 | return false; |
1299 | out: | |
fc9dc401 | 1300 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA && pagelen != 0) |
18dd78c4 | 1301 | return false; |
8d197a56 | 1302 | return PageUptodate(page) != 0; |
5d47a356 TM |
1303 | } |
1304 | ||
5263e31e JL |
1305 | static bool |
1306 | is_whole_file_wrlock(struct file_lock *fl) | |
1307 | { | |
1308 | return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && | |
1309 | fl->fl_type == F_WRLCK; | |
1310 | } | |
1311 | ||
c7559663 SM |
1312 | /* If we know the page is up to date, and we're not using byte range locks (or |
1313 | * if we have the whole file locked for writing), it may be more efficient to | |
1314 | * extend the write to cover the entire page in order to avoid fragmentation | |
1315 | * inefficiencies. | |
1316 | * | |
263b4509 SM |
1317 | * If the file is opened for synchronous writes then we can just skip the rest |
1318 | * of the checks. | |
c7559663 | 1319 | */ |
fc9dc401 TM |
1320 | static int nfs_can_extend_write(struct file *file, struct page *page, |
1321 | struct inode *inode, unsigned int pagelen) | |
c7559663 | 1322 | { |
5263e31e JL |
1323 | int ret; |
1324 | struct file_lock_context *flctx = inode->i_flctx; | |
1325 | struct file_lock *fl; | |
1326 | ||
c7559663 SM |
1327 | if (file->f_flags & O_DSYNC) |
1328 | return 0; | |
fc9dc401 | 1329 | if (!nfs_write_pageuptodate(page, inode, pagelen)) |
263b4509 | 1330 | return 0; |
c7559663 SM |
1331 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1332 | return 1; | |
bd61e0a9 JL |
1333 | if (!flctx || (list_empty_careful(&flctx->flc_flock) && |
1334 | list_empty_careful(&flctx->flc_posix))) | |
8fa4592a | 1335 | return 1; |
5263e31e JL |
1336 | |
1337 | /* Check to see if there are whole file write locks */ | |
5263e31e | 1338 | ret = 0; |
6109c850 | 1339 | spin_lock(&flctx->flc_lock); |
bd61e0a9 JL |
1340 | if (!list_empty(&flctx->flc_posix)) { |
1341 | fl = list_first_entry(&flctx->flc_posix, struct file_lock, | |
1342 | fl_list); | |
1343 | if (is_whole_file_wrlock(fl)) | |
1344 | ret = 1; | |
1345 | } else if (!list_empty(&flctx->flc_flock)) { | |
5263e31e JL |
1346 | fl = list_first_entry(&flctx->flc_flock, struct file_lock, |
1347 | fl_list); | |
1348 | if (fl->fl_type == F_WRLCK) | |
1349 | ret = 1; | |
1350 | } | |
6109c850 | 1351 | spin_unlock(&flctx->flc_lock); |
5263e31e | 1352 | return ret; |
c7559663 SM |
1353 | } |
1354 | ||
1da177e4 LT |
1355 | /* |
1356 | * Update and possibly write a cached page of an NFS file. | |
1357 | * | |
1358 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1359 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1360 | */ | |
1361 | int nfs_updatepage(struct file *file, struct page *page, | |
1362 | unsigned int offset, unsigned int count) | |
1363 | { | |
cd3758e3 | 1364 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d2ceb7e5 BC |
1365 | struct address_space *mapping = page_file_mapping(page); |
1366 | struct inode *inode = mapping->host; | |
fc9dc401 | 1367 | unsigned int pagelen = nfs_page_length(page); |
1da177e4 LT |
1368 | int status = 0; |
1369 | ||
91d5b470 CL |
1370 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1371 | ||
6de1472f AV |
1372 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1373 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1374 | |
149a4fdd BC |
1375 | if (!count) |
1376 | goto out; | |
1377 | ||
fc9dc401 TM |
1378 | if (nfs_can_extend_write(file, page, inode, pagelen)) { |
1379 | count = max(count + offset, pagelen); | |
1da177e4 | 1380 | offset = 0; |
1da177e4 LT |
1381 | } |
1382 | ||
e21195a7 | 1383 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 | 1384 | if (status < 0) |
d2ceb7e5 | 1385 | nfs_set_pageerror(mapping); |
149a4fdd | 1386 | out: |
48186c7d | 1387 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1388 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1389 | return status; |
1390 | } | |
1391 | ||
3ff7576d | 1392 | static int flush_task_priority(int how) |
1da177e4 LT |
1393 | { |
1394 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1395 | case FLUSH_HIGHPRI: | |
1396 | return RPC_PRIORITY_HIGH; | |
1397 | case FLUSH_LOWPRI: | |
1398 | return RPC_PRIORITY_LOW; | |
1399 | } | |
1400 | return RPC_PRIORITY_NORMAL; | |
1401 | } | |
1402 | ||
d45f60c6 WAA |
1403 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1404 | struct rpc_message *msg, | |
abde71f4 | 1405 | const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 | 1406 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1407 | { |
3ff7576d | 1408 | int priority = flush_task_priority(how); |
d138d5d1 | 1409 | |
8db55a03 N |
1410 | if (IS_SWAPFILE(hdr->inode)) |
1411 | task_setup_data->flags |= RPC_TASK_SWAPPER; | |
1ed26f33 | 1412 | task_setup_data->priority = priority; |
fb91fb0e | 1413 | rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); |
5bb2a7cb | 1414 | trace_nfs_initiate_write(hdr); |
275acaaf TM |
1415 | } |
1416 | ||
6d884e8f F |
1417 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1418 | * call this on each, which will prepare them to be retried on next | |
1419 | * writeback using standard nfs. | |
1420 | */ | |
1421 | static void nfs_redirty_request(struct nfs_page *req) | |
1422 | { | |
67f4b5dc TM |
1423 | struct nfs_inode *nfsi = NFS_I(page_file_mapping(req->wb_page)->host); |
1424 | ||
33344e0f TM |
1425 | /* Bump the transmission count */ |
1426 | req->wb_nio++; | |
6d884e8f | 1427 | nfs_mark_request_dirty(req); |
67f4b5dc | 1428 | atomic_long_inc(&nfsi->redirtied_pages); |
20633f04 | 1429 | nfs_end_page_writeback(req); |
3aff4ebb | 1430 | nfs_release_request(req); |
6d884e8f F |
1431 | } |
1432 | ||
df3accb8 | 1433 | static void nfs_async_write_error(struct list_head *head, int error) |
6c75dc0d FI |
1434 | { |
1435 | struct nfs_page *req; | |
1436 | ||
1437 | while (!list_empty(head)) { | |
1438 | req = nfs_list_entry(head->next); | |
1439 | nfs_list_remove_request(req); | |
cea9ba72 | 1440 | if (nfs_error_is_fatal_on_server(error)) |
6fbda89b TM |
1441 | nfs_write_error(req, error); |
1442 | else | |
1443 | nfs_redirty_request(req); | |
6c75dc0d FI |
1444 | } |
1445 | } | |
1446 | ||
dc602dd7 TM |
1447 | static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) |
1448 | { | |
df3accb8 | 1449 | nfs_async_write_error(&hdr->pages, 0); |
dc602dd7 TM |
1450 | } |
1451 | ||
061ae2ed | 1452 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
919e3bd9 | 1453 | .init_hdr = nfs_async_write_init, |
061ae2ed FI |
1454 | .error_cleanup = nfs_async_write_error, |
1455 | .completion = nfs_write_completion, | |
dc602dd7 | 1456 | .reschedule_io = nfs_async_write_reschedule_io, |
061ae2ed FI |
1457 | }; |
1458 | ||
57208fa7 | 1459 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1460 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1461 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1462 | { |
a20c93e3 | 1463 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1464 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1465 | |
1466 | #ifdef CONFIG_NFS_V4_1 | |
1467 | if (server->pnfs_curr_ld && !force_mds) | |
1468 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1469 | #endif | |
4a0de55c | 1470 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
3bde7afd | 1471 | server->wsize, ioflags); |
1751c363 | 1472 | } |
ddda8e0a | 1473 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1474 | |
dce81290 TM |
1475 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1476 | { | |
a7d42ddb WAA |
1477 | struct nfs_pgio_mirror *mirror; |
1478 | ||
6f29b9bb KM |
1479 | if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) |
1480 | pgio->pg_ops->pg_cleanup(pgio); | |
1481 | ||
41d8d5b7 | 1482 | pgio->pg_ops = &nfs_pgio_rw_ops; |
a7d42ddb WAA |
1483 | |
1484 | nfs_pageio_stop_mirroring(pgio); | |
1485 | ||
1486 | mirror = &pgio->pg_mirrors[0]; | |
1487 | mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
dce81290 | 1488 | } |
1f945357 | 1489 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1490 | |
1da177e4 | 1491 | |
0b7c0153 FI |
1492 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1493 | { | |
1494 | struct nfs_commit_data *data = calldata; | |
1495 | ||
1496 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1497 | } | |
1498 | ||
a08a8cd3 TM |
1499 | static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, |
1500 | struct nfs_fattr *fattr) | |
1501 | { | |
1502 | struct nfs_pgio_args *argp = &hdr->args; | |
1503 | struct nfs_pgio_res *resp = &hdr->res; | |
2b83d3de | 1504 | u64 size = argp->offset + resp->count; |
a08a8cd3 TM |
1505 | |
1506 | if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) | |
2b83d3de TM |
1507 | fattr->size = size; |
1508 | if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { | |
1509 | fattr->valid &= ~NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 | 1510 | return; |
2b83d3de TM |
1511 | } |
1512 | if (size != fattr->size) | |
a08a8cd3 TM |
1513 | return; |
1514 | /* Set attribute barrier */ | |
1515 | nfs_fattr_set_barrier(fattr); | |
2b83d3de TM |
1516 | /* ...and update size */ |
1517 | fattr->valid |= NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 TM |
1518 | } |
1519 | ||
1520 | void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) | |
1521 | { | |
2b83d3de | 1522 | struct nfs_fattr *fattr = &hdr->fattr; |
a08a8cd3 TM |
1523 | struct inode *inode = hdr->inode; |
1524 | ||
a08a8cd3 TM |
1525 | spin_lock(&inode->i_lock); |
1526 | nfs_writeback_check_extend(hdr, fattr); | |
1527 | nfs_post_op_update_inode_force_wcc_locked(inode, fattr); | |
1528 | spin_unlock(&inode->i_lock); | |
1529 | } | |
1530 | EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); | |
1531 | ||
1da177e4 LT |
1532 | /* |
1533 | * This function is called when the WRITE call is complete. | |
1534 | */ | |
d45f60c6 WAA |
1535 | static int nfs_writeback_done(struct rpc_task *task, |
1536 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1537 | struct inode *inode) |
1da177e4 | 1538 | { |
788e7a89 | 1539 | int status; |
1da177e4 | 1540 | |
f551e44f CL |
1541 | /* |
1542 | * ->write_done will attempt to use post-op attributes to detect | |
1543 | * conflicting writes by other clients. A strict interpretation | |
1544 | * of close-to-open would allow us to continue caching even if | |
1545 | * another writer had changed the file, but some applications | |
1546 | * depend on tighter cache coherency when writing. | |
1547 | */ | |
d45f60c6 | 1548 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1549 | if (status != 0) |
0eecb214 | 1550 | return status; |
8224b273 | 1551 | |
d45f60c6 | 1552 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
5bb2a7cb | 1553 | trace_nfs_writeback_done(task, hdr); |
91d5b470 | 1554 | |
69d96651 JL |
1555 | if (task->tk_status >= 0) { |
1556 | enum nfs3_stable_how committed = hdr->res.verf->committed; | |
1557 | ||
1558 | if (committed == NFS_UNSTABLE) { | |
1559 | /* | |
1560 | * We have some uncommitted data on the server at | |
1561 | * this point, so ensure that we keep track of that | |
1562 | * fact irrespective of what later writes do. | |
1563 | */ | |
1564 | set_bit(NFS_IOHDR_UNSTABLE_WRITES, &hdr->flags); | |
1565 | } | |
1da177e4 | 1566 | |
69d96651 JL |
1567 | if (committed < hdr->args.stable) { |
1568 | /* We tried a write call, but the server did not | |
1569 | * commit data to stable storage even though we | |
1570 | * requested it. | |
1571 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1572 | * the server reports NFS_DATA_SYNC, but performs | |
1573 | * NFS_FILE_SYNC. We therefore implement this checking | |
1574 | * as a dprintk() in order to avoid filling syslog. | |
1575 | */ | |
1576 | static unsigned long complain; | |
1577 | ||
1578 | /* Note this will print the MDS for a DS write */ | |
1579 | if (time_before(complain, jiffies)) { | |
1580 | dprintk("NFS: faulty NFS server %s:" | |
1581 | " (committed = %d) != (stable = %d)\n", | |
1582 | NFS_SERVER(inode)->nfs_client->cl_hostname, | |
1583 | committed, hdr->args.stable); | |
1584 | complain = jiffies + 300 * HZ; | |
1585 | } | |
1da177e4 LT |
1586 | } |
1587 | } | |
1f2edbe3 TM |
1588 | |
1589 | /* Deal with the suid/sgid bit corner case */ | |
16e14375 TM |
1590 | if (nfs_should_remove_suid(inode)) { |
1591 | spin_lock(&inode->i_lock); | |
720869eb | 1592 | nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); |
16e14375 TM |
1593 | spin_unlock(&inode->i_lock); |
1594 | } | |
0eecb214 AS |
1595 | return 0; |
1596 | } | |
1597 | ||
1598 | /* | |
1599 | * This function is called when the WRITE call is complete. | |
1600 | */ | |
d45f60c6 WAA |
1601 | static void nfs_writeback_result(struct rpc_task *task, |
1602 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1603 | { |
d45f60c6 WAA |
1604 | struct nfs_pgio_args *argp = &hdr->args; |
1605 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1606 | |
1607 | if (resp->count < argp->count) { | |
1da177e4 LT |
1608 | static unsigned long complain; |
1609 | ||
6c75dc0d | 1610 | /* This a short write! */ |
d45f60c6 | 1611 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1612 | |
1da177e4 | 1613 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1614 | if (resp->count == 0) { |
1615 | if (time_before(complain, jiffies)) { | |
1616 | printk(KERN_WARNING | |
1617 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1618 | argp->count); | |
1619 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1620 | } |
d45f60c6 | 1621 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1622 | task->tk_status = -EIO; |
13602896 | 1623 | return; |
1da177e4 | 1624 | } |
f8417b48 KM |
1625 | |
1626 | /* For non rpc-based layout drivers, retry-through-MDS */ | |
1627 | if (!task->tk_ops) { | |
1628 | hdr->pnfs_error = -EAGAIN; | |
1629 | return; | |
1630 | } | |
1631 | ||
6c75dc0d FI |
1632 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1633 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1634 | /* Resend from where the server left off */ | |
d45f60c6 | 1635 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1636 | argp->offset += resp->count; |
1637 | argp->pgbase += resp->count; | |
1638 | argp->count -= resp->count; | |
1639 | } else { | |
1640 | /* Resend as a stable write in order to avoid | |
1641 | * headaches in the case of a server crash. | |
1642 | */ | |
1643 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1644 | } |
8c9cb714 TM |
1645 | resp->count = 0; |
1646 | resp->verf->committed = 0; | |
6c75dc0d | 1647 | rpc_restart_call_prepare(task); |
1da177e4 | 1648 | } |
1da177e4 LT |
1649 | } |
1650 | ||
af7cf057 | 1651 | static int wait_on_commit(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1652 | { |
723c921e PZ |
1653 | return wait_var_event_killable(&cinfo->rpcs_out, |
1654 | !atomic_read(&cinfo->rpcs_out)); | |
af7cf057 | 1655 | } |
b8413f98 | 1656 | |
af7cf057 TM |
1657 | static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) |
1658 | { | |
1659 | atomic_inc(&cinfo->rpcs_out); | |
71d0a611 TM |
1660 | } |
1661 | ||
133a48ab | 1662 | bool nfs_commit_end(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1663 | { |
133a48ab | 1664 | if (atomic_dec_and_test(&cinfo->rpcs_out)) { |
723c921e | 1665 | wake_up_var(&cinfo->rpcs_out); |
133a48ab TM |
1666 | return true; |
1667 | } | |
1668 | return false; | |
71d0a611 TM |
1669 | } |
1670 | ||
0b7c0153 | 1671 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1672 | { |
0b7c0153 FI |
1673 | put_nfs_open_context(data->context); |
1674 | nfs_commit_free(data); | |
1da177e4 | 1675 | } |
e0c2b380 | 1676 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1677 | |
0b7c0153 | 1678 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
c36aae9a | 1679 | const struct nfs_rpc_ops *nfs_ops, |
9ace33cd | 1680 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1681 | int how, int flags) |
1da177e4 | 1682 | { |
07737691 | 1683 | struct rpc_task *task; |
9ace33cd | 1684 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1685 | struct rpc_message msg = { |
1686 | .rpc_argp = &data->args, | |
1687 | .rpc_resp = &data->res, | |
9ace33cd | 1688 | .rpc_cred = data->cred, |
bdc7f021 | 1689 | }; |
84115e1c | 1690 | struct rpc_task_setup task_setup_data = { |
07737691 | 1691 | .task = &data->task, |
9ace33cd | 1692 | .rpc_client = clnt, |
bdc7f021 | 1693 | .rpc_message = &msg, |
9ace33cd | 1694 | .callback_ops = call_ops, |
84115e1c | 1695 | .callback_data = data, |
101070ca | 1696 | .workqueue = nfsiod_workqueue, |
4fa7ef69 | 1697 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1698 | .priority = priority, |
84115e1c | 1699 | }; |
118f09ed OK |
1700 | |
1701 | if (nfs_server_capable(data->inode, NFS_CAP_MOVEABLE)) | |
1702 | task_setup_data.flags |= RPC_TASK_MOVEABLE; | |
1703 | ||
9ace33cd | 1704 | /* Set up the initial task struct. */ |
e9ae1ee2 | 1705 | nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); |
8224b273 | 1706 | trace_nfs_initiate_commit(data); |
9ace33cd | 1707 | |
b4839ebe | 1708 | dprintk("NFS: initiated commit call\n"); |
9ace33cd FI |
1709 | |
1710 | task = rpc_run_task(&task_setup_data); | |
1711 | if (IS_ERR(task)) | |
1712 | return PTR_ERR(task); | |
1713 | if (how & FLUSH_SYNC) | |
1714 | rpc_wait_for_completion_task(task); | |
1715 | rpc_put_task(task); | |
1716 | return 0; | |
1717 | } | |
e0c2b380 | 1718 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1719 | |
378520b8 PT |
1720 | static loff_t nfs_get_lwb(struct list_head *head) |
1721 | { | |
1722 | loff_t lwb = 0; | |
1723 | struct nfs_page *req; | |
1724 | ||
1725 | list_for_each_entry(req, head, wb_list) | |
1726 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1727 | lwb = req_offset(req) + req->wb_bytes; | |
1728 | ||
1729 | return lwb; | |
1730 | } | |
1731 | ||
9ace33cd FI |
1732 | /* |
1733 | * Set up the argument/result storage required for the RPC call. | |
1734 | */ | |
0b7c0153 | 1735 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1736 | struct list_head *head, |
1737 | struct pnfs_layout_segment *lseg, | |
1738 | struct nfs_commit_info *cinfo) | |
9ace33cd | 1739 | { |
19573c93 TM |
1740 | struct nfs_page *first; |
1741 | struct nfs_open_context *ctx; | |
1742 | struct inode *inode; | |
1da177e4 LT |
1743 | |
1744 | /* Set up the RPC argument and reply structs | |
1745 | * NB: take care not to mess about with data->commit et al. */ | |
1746 | ||
19573c93 TM |
1747 | if (head) |
1748 | list_splice_init(head, &data->pages); | |
1749 | ||
1750 | first = nfs_list_entry(data->pages.next); | |
1751 | ctx = nfs_req_openctx(first); | |
1752 | inode = d_inode(ctx->dentry); | |
1da177e4 | 1753 | |
1da177e4 | 1754 | data->inode = inode; |
9fcd5960 | 1755 | data->cred = ctx->cred; |
988b6dce | 1756 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1757 | /* only set lwb for pnfs commit */ |
1758 | if (lseg) | |
1759 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1760 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1761 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1762 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1763 | |
1764 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1765 | /* Note: we always request a commit of the entire inode */ |
1766 | data->args.offset = 0; | |
1767 | data->args.count = 0; | |
9fcd5960 | 1768 | data->context = get_nfs_open_context(ctx); |
1da177e4 LT |
1769 | data->res.fattr = &data->fattr; |
1770 | data->res.verf = &data->verf; | |
0e574af1 | 1771 | nfs_fattr_init(&data->fattr); |
133a48ab | 1772 | nfs_commit_begin(cinfo->mds); |
1da177e4 | 1773 | } |
e0c2b380 | 1774 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1775 | |
e0c2b380 | 1776 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 | 1777 | struct pnfs_layout_segment *lseg, |
b57ff130 WAA |
1778 | struct nfs_commit_info *cinfo, |
1779 | u32 ds_commit_idx) | |
64bfeb49 FI |
1780 | { |
1781 | struct nfs_page *req; | |
1782 | ||
1783 | while (!list_empty(page_list)) { | |
1784 | req = nfs_list_entry(page_list->next); | |
1785 | nfs_list_remove_request(req); | |
b57ff130 | 1786 | nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); |
487b9b8a TH |
1787 | if (!cinfo->dreq) |
1788 | nfs_clear_page_commit(req->wb_page); | |
1d1afcbc | 1789 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1790 | } |
1791 | } | |
e0c2b380 | 1792 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1793 | |
b20135d0 TM |
1794 | static void |
1795 | nfs_commit_resched_write(struct nfs_commit_info *cinfo, | |
1796 | struct nfs_page *req) | |
1797 | { | |
1798 | __set_page_dirty_nobuffers(req->wb_page); | |
1799 | } | |
1800 | ||
1da177e4 LT |
1801 | /* |
1802 | * Commit dirty pages | |
1803 | */ | |
1804 | static int | |
ea2cf228 FI |
1805 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1806 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1807 | { |
0b7c0153 | 1808 | struct nfs_commit_data *data; |
85e39fee | 1809 | unsigned short task_flags = 0; |
1da177e4 | 1810 | |
ade8febd WAA |
1811 | /* another commit raced with us */ |
1812 | if (list_empty(head)) | |
1813 | return 0; | |
1814 | ||
515dcdcd TM |
1815 | data = nfs_commitdata_alloc(); |
1816 | if (!data) { | |
1817 | nfs_retry_commit(head, NULL, cinfo, -1); | |
1818 | return -ENOMEM; | |
1819 | } | |
1da177e4 LT |
1820 | |
1821 | /* Set up the argument struct */ | |
f453a54a | 1822 | nfs_init_commit(data, head, NULL, cinfo); |
85e39fee OK |
1823 | if (NFS_SERVER(inode)->nfs_client->cl_minorversion) |
1824 | task_flags = RPC_TASK_MOVEABLE; | |
c36aae9a | 1825 | return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), |
85e39fee OK |
1826 | data->mds_ops, how, |
1827 | RPC_TASK_CRED_NOREF | task_flags); | |
67911c8f | 1828 | } |
67911c8f | 1829 | |
1da177e4 LT |
1830 | /* |
1831 | * COMMIT call returned | |
1832 | */ | |
788e7a89 | 1833 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1834 | { |
0b7c0153 | 1835 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1836 | |
788e7a89 | 1837 | /* Call the NFS version-specific code */ |
c0d0e96b | 1838 | NFS_PROTO(data->inode)->commit_done(task, data); |
7bdd297e | 1839 | trace_nfs_commit_done(task, data); |
c9d8f89d TM |
1840 | } |
1841 | ||
f453a54a | 1842 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1843 | { |
221203ce | 1844 | const struct nfs_writeverf *verf = data->res.verf; |
5917ce84 | 1845 | struct nfs_page *req; |
c9d8f89d | 1846 | int status = data->task.tk_status; |
f453a54a | 1847 | struct nfs_commit_info cinfo; |
353db796 | 1848 | struct nfs_server *nfss; |
788e7a89 | 1849 | |
1da177e4 LT |
1850 | while (!list_empty(&data->pages)) { |
1851 | req = nfs_list_entry(data->pages.next); | |
1852 | nfs_list_remove_request(req); | |
67911c8f AS |
1853 | if (req->wb_page) |
1854 | nfs_clear_page_commit(req->wb_page); | |
1da177e4 | 1855 | |
1e8968c5 | 1856 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
9fcd5960 TM |
1857 | nfs_req_openctx(req)->dentry->d_sb->s_id, |
1858 | (unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)), | |
1da177e4 LT |
1859 | req->wb_bytes, |
1860 | (long long)req_offset(req)); | |
c9d8f89d | 1861 | if (status < 0) { |
6fbda89b | 1862 | if (req->wb_page) { |
af887e43 TM |
1863 | trace_nfs_commit_error(data->inode, req, |
1864 | status); | |
6fbda89b | 1865 | nfs_mapping_set_error(req->wb_page, status); |
38a33101 | 1866 | nfs_inode_remove_request(req); |
6fbda89b | 1867 | } |
ddeaa637 | 1868 | dprintk_cont(", error = %d\n", status); |
1da177e4 LT |
1869 | goto next; |
1870 | } | |
1871 | ||
1872 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1873 | * returned by the server against all stored verfs. */ | |
1f28476d | 1874 | if (nfs_write_match_verf(verf, req)) { |
1da177e4 | 1875 | /* We have a match */ |
38a33101 KM |
1876 | if (req->wb_page) |
1877 | nfs_inode_remove_request(req); | |
ddeaa637 | 1878 | dprintk_cont(" OK\n"); |
1da177e4 LT |
1879 | goto next; |
1880 | } | |
1881 | /* We have a mismatch. Write the page again */ | |
ddeaa637 | 1882 | dprintk_cont(" mismatch\n"); |
6d884e8f | 1883 | nfs_mark_request_dirty(req); |
67f4b5dc | 1884 | atomic_long_inc(&NFS_I(data->inode)->redirtied_pages); |
1da177e4 | 1885 | next: |
1d1afcbc | 1886 | nfs_unlock_and_release_request(req); |
7f1bda44 TM |
1887 | /* Latency breaker */ |
1888 | cond_resched(); | |
1da177e4 | 1889 | } |
353db796 N |
1890 | nfss = NFS_SERVER(data->inode); |
1891 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
6df25e58 | 1892 | nfss->write_congested = 0; |
353db796 | 1893 | |
f453a54a | 1894 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
af7cf057 | 1895 | nfs_commit_end(cinfo.mds); |
5917ce84 FI |
1896 | } |
1897 | ||
1898 | static void nfs_commit_release(void *calldata) | |
1899 | { | |
0b7c0153 | 1900 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1901 | |
f453a54a | 1902 | data->completion_ops->completion(data); |
c9d8f89d | 1903 | nfs_commitdata_release(calldata); |
1da177e4 | 1904 | } |
788e7a89 TM |
1905 | |
1906 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1907 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1908 | .rpc_call_done = nfs_commit_done, |
1909 | .rpc_release = nfs_commit_release, | |
1910 | }; | |
1da177e4 | 1911 | |
f453a54a FI |
1912 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1913 | .completion = nfs_commit_release_pages, | |
b20135d0 | 1914 | .resched_write = nfs_commit_resched_write, |
f453a54a FI |
1915 | }; |
1916 | ||
1763da12 FI |
1917 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1918 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1919 | { |
1920 | int status; | |
1921 | ||
ea2cf228 | 1922 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1923 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1924 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1925 | return status; |
1926 | } | |
1927 | ||
c4f24df9 TM |
1928 | static int __nfs_commit_inode(struct inode *inode, int how, |
1929 | struct writeback_control *wbc) | |
1da177e4 | 1930 | { |
1da177e4 | 1931 | LIST_HEAD(head); |
ea2cf228 | 1932 | struct nfs_commit_info cinfo; |
71d0a611 | 1933 | int may_wait = how & FLUSH_SYNC; |
c4f24df9 | 1934 | int ret, nscan; |
1da177e4 | 1935 | |
64a93dbf | 1936 | how &= ~FLUSH_SYNC; |
ea2cf228 | 1937 | nfs_init_cinfo_from_inode(&cinfo, inode); |
af7cf057 | 1938 | nfs_commit_begin(cinfo.mds); |
c4f24df9 TM |
1939 | for (;;) { |
1940 | ret = nscan = nfs_scan_commit(inode, &head, &cinfo); | |
1941 | if (ret <= 0) | |
1942 | break; | |
1943 | ret = nfs_generic_commit_list(inode, &head, how, &cinfo); | |
1944 | if (ret < 0) | |
1945 | break; | |
1946 | ret = 0; | |
1947 | if (wbc && wbc->sync_mode == WB_SYNC_NONE) { | |
1948 | if (nscan < wbc->nr_to_write) | |
1949 | wbc->nr_to_write -= nscan; | |
1950 | else | |
1951 | wbc->nr_to_write = 0; | |
1952 | } | |
1953 | if (nscan < INT_MAX) | |
1954 | break; | |
1955 | cond_resched(); | |
1956 | } | |
af7cf057 | 1957 | nfs_commit_end(cinfo.mds); |
c4f24df9 TM |
1958 | if (ret || !may_wait) |
1959 | return ret; | |
1960 | return wait_on_commit(cinfo.mds); | |
1961 | } | |
1962 | ||
1963 | int nfs_commit_inode(struct inode *inode, int how) | |
1964 | { | |
1965 | return __nfs_commit_inode(inode, how, NULL); | |
1da177e4 | 1966 | } |
b20135d0 | 1967 | EXPORT_SYMBOL_GPL(nfs_commit_inode); |
8fc795f7 | 1968 | |
ae09c31f | 1969 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
8fc795f7 | 1970 | { |
420e3646 TM |
1971 | struct nfs_inode *nfsi = NFS_I(inode); |
1972 | int flags = FLUSH_SYNC; | |
1973 | int ret = 0; | |
8fc795f7 | 1974 | |
a00dd6c0 | 1975 | if (wbc->sync_mode == WB_SYNC_NONE) { |
c4f24df9 TM |
1976 | /* no commits means nothing needs to be done */ |
1977 | if (!atomic_long_read(&nfsi->commit_info.ncommit)) | |
1978 | goto check_requests_outstanding; | |
1979 | ||
a00dd6c0 JL |
1980 | /* Don't commit yet if this is a non-blocking flush and there |
1981 | * are a lot of outstanding writes for this mapping. | |
1982 | */ | |
1a4edf0f | 1983 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) |
a00dd6c0 | 1984 | goto out_mark_dirty; |
420e3646 | 1985 | |
a00dd6c0 | 1986 | /* don't wait for the COMMIT response */ |
420e3646 | 1987 | flags = 0; |
a00dd6c0 JL |
1988 | } |
1989 | ||
c4f24df9 TM |
1990 | ret = __nfs_commit_inode(inode, flags, wbc); |
1991 | if (!ret) { | |
1992 | if (flags & FLUSH_SYNC) | |
1993 | return 0; | |
1994 | } else if (atomic_long_read(&nfsi->commit_info.ncommit)) | |
1995 | goto out_mark_dirty; | |
1996 | ||
1997 | check_requests_outstanding: | |
1998 | if (!atomic_read(&nfsi->commit_info.rpcs_out)) | |
1999 | return ret; | |
420e3646 | 2000 | out_mark_dirty: |
8fc795f7 TM |
2001 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
2002 | return ret; | |
2003 | } | |
89d77c8f | 2004 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 2005 | |
837bb1d7 TM |
2006 | /* |
2007 | * Wrapper for filemap_write_and_wait_range() | |
2008 | * | |
2009 | * Needed for pNFS in order to ensure data becomes visible to the | |
2010 | * client. | |
2011 | */ | |
2012 | int nfs_filemap_write_and_wait_range(struct address_space *mapping, | |
2013 | loff_t lstart, loff_t lend) | |
2014 | { | |
2015 | int ret; | |
2016 | ||
2017 | ret = filemap_write_and_wait_range(mapping, lstart, lend); | |
2018 | if (ret == 0) | |
2019 | ret = pnfs_sync_inode(mapping->host, true); | |
2020 | return ret; | |
2021 | } | |
2022 | EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); | |
2023 | ||
acdc53b2 TM |
2024 | /* |
2025 | * flush the inode to disk. | |
2026 | */ | |
2027 | int nfs_wb_all(struct inode *inode) | |
34901f70 | 2028 | { |
f4ce1299 TM |
2029 | int ret; |
2030 | ||
2031 | trace_nfs_writeback_inode_enter(inode); | |
2032 | ||
5bb89b47 | 2033 | ret = filemap_write_and_wait(inode->i_mapping); |
6b196875 CL |
2034 | if (ret) |
2035 | goto out; | |
2036 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
2037 | if (ret < 0) | |
2038 | goto out; | |
2039 | pnfs_sync_inode(inode, true); | |
2040 | ret = 0; | |
34901f70 | 2041 | |
6b196875 | 2042 | out: |
f4ce1299 TM |
2043 | trace_nfs_writeback_inode_exit(inode, ret); |
2044 | return ret; | |
1c75950b | 2045 | } |
ddda8e0a | 2046 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 2047 | |
6d740c76 | 2048 | int nfs_wb_folio_cancel(struct inode *inode, struct folio *folio) |
1b3b4a1a TM |
2049 | { |
2050 | struct nfs_page *req; | |
1b3b4a1a TM |
2051 | int ret = 0; |
2052 | ||
6d740c76 | 2053 | folio_wait_writeback(folio); |
3e217045 WAA |
2054 | |
2055 | /* blocking call to cancel all requests and join to a single (head) | |
2056 | * request */ | |
6d740c76 | 2057 | req = nfs_lock_and_join_requests(&folio->page); |
3e217045 WAA |
2058 | |
2059 | if (IS_ERR(req)) { | |
2060 | ret = PTR_ERR(req); | |
2061 | } else if (req) { | |
6d740c76 | 2062 | /* all requests from this folio have been cancelled by |
3e217045 WAA |
2063 | * nfs_lock_and_join_requests, so just remove the head |
2064 | * request from the inode / page_private pointer and | |
2065 | * release it */ | |
2066 | nfs_inode_remove_request(req); | |
3e217045 | 2067 | nfs_unlock_and_release_request(req); |
1b3b4a1a | 2068 | } |
3e217045 | 2069 | |
1b3b4a1a TM |
2070 | return ret; |
2071 | } | |
2072 | ||
7f2f12d9 TM |
2073 | /* |
2074 | * Write back all requests on one page - we do this before reading it. | |
2075 | */ | |
c373fff7 | 2076 | int nfs_wb_page(struct inode *inode, struct page *page) |
1c75950b | 2077 | { |
29418aa4 | 2078 | loff_t range_start = page_file_offset(page); |
09cbfeaf | 2079 | loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); |
4d770ccf | 2080 | struct writeback_control wbc = { |
4d770ccf | 2081 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 2082 | .nr_to_write = 0, |
4d770ccf TM |
2083 | .range_start = range_start, |
2084 | .range_end = range_end, | |
2085 | }; | |
2086 | int ret; | |
1c75950b | 2087 | |
f4ce1299 TM |
2088 | trace_nfs_writeback_page_enter(inode); |
2089 | ||
0522f6ad | 2090 | for (;;) { |
ba8b06e6 | 2091 | wait_on_page_writeback(page); |
73e3302f | 2092 | if (clear_page_dirty_for_io(page)) { |
c373fff7 | 2093 | ret = nfs_writepage_locked(page, &wbc); |
73e3302f TM |
2094 | if (ret < 0) |
2095 | goto out_error; | |
0522f6ad | 2096 | continue; |
7f2f12d9 | 2097 | } |
f4ce1299 | 2098 | ret = 0; |
0522f6ad TM |
2099 | if (!PagePrivate(page)) |
2100 | break; | |
2101 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 2102 | if (ret < 0) |
73e3302f | 2103 | goto out_error; |
7f2f12d9 | 2104 | } |
73e3302f | 2105 | out_error: |
f4ce1299 | 2106 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 2107 | return ret; |
1c75950b TM |
2108 | } |
2109 | ||
074cc1de | 2110 | #ifdef CONFIG_MIGRATION |
4ae84a80 MWO |
2111 | int nfs_migrate_folio(struct address_space *mapping, struct folio *dst, |
2112 | struct folio *src, enum migrate_mode mode) | |
074cc1de | 2113 | { |
2da95652 | 2114 | /* |
4ae84a80 | 2115 | * If the private flag is set, the folio is currently associated with |
2da95652 JL |
2116 | * an in-progress read or write request. Don't try to migrate it. |
2117 | * | |
2118 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
2119 | * that we can safely release the inode reference while holding | |
4ae84a80 | 2120 | * the folio lock. |
2da95652 | 2121 | */ |
4ae84a80 | 2122 | if (folio_test_private(src)) |
2da95652 | 2123 | return -EBUSY; |
074cc1de | 2124 | |
4ae84a80 | 2125 | if (folio_test_fscache(src)) { |
16f2f4e6 DH |
2126 | if (mode == MIGRATE_ASYNC) |
2127 | return -EBUSY; | |
4ae84a80 | 2128 | folio_wait_fscache(src); |
16f2f4e6 | 2129 | } |
074cc1de | 2130 | |
54184650 | 2131 | return migrate_folio(mapping, dst, src, mode); |
074cc1de TM |
2132 | } |
2133 | #endif | |
2134 | ||
f7b422b1 | 2135 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
2136 | { |
2137 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 2138 | sizeof(struct nfs_pgio_header), |
1da177e4 | 2139 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 2140 | NULL); |
1da177e4 LT |
2141 | if (nfs_wdata_cachep == NULL) |
2142 | return -ENOMEM; | |
2143 | ||
93d2341c MD |
2144 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
2145 | nfs_wdata_cachep); | |
1da177e4 | 2146 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 2147 | goto out_destroy_write_cache; |
1da177e4 | 2148 | |
0b7c0153 FI |
2149 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
2150 | sizeof(struct nfs_commit_data), | |
2151 | 0, SLAB_HWCACHE_ALIGN, | |
2152 | NULL); | |
2153 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 2154 | goto out_destroy_write_mempool; |
0b7c0153 | 2155 | |
93d2341c | 2156 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 2157 | nfs_cdata_cachep); |
1da177e4 | 2158 | if (nfs_commit_mempool == NULL) |
3dd4765f | 2159 | goto out_destroy_commit_cache; |
1da177e4 | 2160 | |
89a09141 PZ |
2161 | /* |
2162 | * NFS congestion size, scale with available memory. | |
2163 | * | |
2164 | * 64MB: 8192k | |
2165 | * 128MB: 11585k | |
2166 | * 256MB: 16384k | |
2167 | * 512MB: 23170k | |
2168 | * 1GB: 32768k | |
2169 | * 2GB: 46340k | |
2170 | * 4GB: 65536k | |
2171 | * 8GB: 92681k | |
2172 | * 16GB: 131072k | |
2173 | * | |
2174 | * This allows larger machines to have larger/more transfers. | |
2175 | * Limit the default to 256M | |
2176 | */ | |
ca79b0c2 | 2177 | nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); |
89a09141 PZ |
2178 | if (nfs_congestion_kb > 256*1024) |
2179 | nfs_congestion_kb = 256*1024; | |
2180 | ||
1da177e4 | 2181 | return 0; |
3dd4765f JL |
2182 | |
2183 | out_destroy_commit_cache: | |
2184 | kmem_cache_destroy(nfs_cdata_cachep); | |
2185 | out_destroy_write_mempool: | |
2186 | mempool_destroy(nfs_wdata_mempool); | |
2187 | out_destroy_write_cache: | |
2188 | kmem_cache_destroy(nfs_wdata_cachep); | |
2189 | return -ENOMEM; | |
1da177e4 LT |
2190 | } |
2191 | ||
266bee88 | 2192 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
2193 | { |
2194 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 2195 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 2196 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 2197 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
2198 | } |
2199 | ||
4a0de55c AS |
2200 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
2201 | .rw_alloc_header = nfs_writehdr_alloc, | |
2202 | .rw_free_header = nfs_writehdr_free, | |
0eecb214 AS |
2203 | .rw_done = nfs_writeback_done, |
2204 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 2205 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 2206 | }; |