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 | ||
518662e0 | 73 | struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) |
1da177e4 | 74 | { |
518662e0 | 75 | struct nfs_commit_data *p; |
40859d7e | 76 | |
518662e0 N |
77 | if (never_fail) |
78 | p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); | |
79 | else { | |
80 | /* It is OK to do some reclaim, not no safe to wait | |
81 | * for anything to be returned to the pool. | |
82 | * mempool_alloc() cannot handle that particular combination, | |
83 | * so we need two separate attempts. | |
84 | */ | |
85 | p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); | |
86 | if (!p) | |
87 | p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | | |
88 | __GFP_NOWARN | __GFP_NORETRY); | |
89 | if (!p) | |
90 | return NULL; | |
1da177e4 | 91 | } |
518662e0 N |
92 | |
93 | memset(p, 0, sizeof(*p)); | |
94 | INIT_LIST_HEAD(&p->pages); | |
1da177e4 LT |
95 | return p; |
96 | } | |
e0c2b380 | 97 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 98 | |
0b7c0153 | 99 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
100 | { |
101 | mempool_free(p, nfs_commit_mempool); | |
102 | } | |
e0c2b380 | 103 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 104 | |
1e7f3a48 | 105 | static struct nfs_pgio_header *nfs_writehdr_alloc(void) |
3feb2d49 | 106 | { |
2b17d725 | 107 | struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_KERNEL); |
cd841605 | 108 | |
237f8306 N |
109 | memset(p, 0, sizeof(*p)); |
110 | p->rw_mode = FMODE_WRITE; | |
3feb2d49 TM |
111 | return p; |
112 | } | |
6c75dc0d | 113 | |
1e7f3a48 | 114 | static void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 115 | { |
1e7f3a48 | 116 | mempool_free(hdr, nfs_wdata_mempool); |
3feb2d49 | 117 | } |
1da177e4 | 118 | |
919e3bd9 TM |
119 | static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) |
120 | { | |
121 | return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); | |
122 | } | |
123 | ||
124 | static void nfs_io_completion_init(struct nfs_io_completion *ioc, | |
125 | void (*complete)(void *), void *data) | |
126 | { | |
127 | ioc->complete = complete; | |
128 | ioc->data = data; | |
129 | kref_init(&ioc->refcount); | |
130 | } | |
131 | ||
132 | static void nfs_io_completion_release(struct kref *kref) | |
133 | { | |
134 | struct nfs_io_completion *ioc = container_of(kref, | |
135 | struct nfs_io_completion, refcount); | |
136 | ioc->complete(ioc->data); | |
137 | kfree(ioc); | |
138 | } | |
139 | ||
140 | static void nfs_io_completion_get(struct nfs_io_completion *ioc) | |
141 | { | |
142 | if (ioc != NULL) | |
143 | kref_get(&ioc->refcount); | |
144 | } | |
145 | ||
146 | static void nfs_io_completion_put(struct nfs_io_completion *ioc) | |
147 | { | |
148 | if (ioc != NULL) | |
149 | kref_put(&ioc->refcount, nfs_io_completion_release); | |
150 | } | |
151 | ||
e00ed89d TM |
152 | static void |
153 | nfs_page_set_inode_ref(struct nfs_page *req, struct inode *inode) | |
154 | { | |
155 | if (!test_and_set_bit(PG_INODE_REF, &req->wb_flags)) { | |
156 | kref_get(&req->wb_kref); | |
157 | atomic_long_inc(&NFS_I(inode)->nrequests); | |
158 | } | |
159 | } | |
160 | ||
161 | static int | |
162 | nfs_cancel_remove_inode(struct nfs_page *req, struct inode *inode) | |
163 | { | |
164 | int ret; | |
165 | ||
166 | if (!test_bit(PG_REMOVE, &req->wb_flags)) | |
167 | return 0; | |
168 | ret = nfs_page_group_lock(req); | |
169 | if (ret) | |
170 | return ret; | |
171 | if (test_and_clear_bit(PG_REMOVE, &req->wb_flags)) | |
172 | nfs_page_set_inode_ref(req, inode); | |
173 | nfs_page_group_unlock(req); | |
174 | return 0; | |
175 | } | |
176 | ||
bd37d6fc TM |
177 | static struct nfs_page * |
178 | nfs_page_private_request(struct page *page) | |
179 | { | |
180 | if (!PagePrivate(page)) | |
181 | return NULL; | |
182 | return (struct nfs_page *)page_private(page); | |
183 | } | |
184 | ||
84d3a9a9 WAA |
185 | /* |
186 | * nfs_page_find_head_request_locked - find head request associated with @page | |
187 | * | |
188 | * must be called while holding the inode lock. | |
189 | * | |
190 | * returns matching head request with reference held, or NULL if not found. | |
191 | */ | |
29418aa4 | 192 | static struct nfs_page * |
b30d2f04 | 193 | nfs_page_find_private_request(struct page *page) |
277459d2 | 194 | { |
4b9bb25b | 195 | struct address_space *mapping = page_file_mapping(page); |
bd37d6fc | 196 | struct nfs_page *req; |
277459d2 | 197 | |
b30d2f04 TM |
198 | if (!PagePrivate(page)) |
199 | return NULL; | |
4b9bb25b | 200 | spin_lock(&mapping->private_lock); |
bd37d6fc | 201 | req = nfs_page_private_request(page); |
84d3a9a9 WAA |
202 | if (req) { |
203 | WARN_ON_ONCE(req->wb_head != req); | |
29418aa4 | 204 | kref_get(&req->wb_kref); |
84d3a9a9 | 205 | } |
4b9bb25b | 206 | spin_unlock(&mapping->private_lock); |
b30d2f04 TM |
207 | return req; |
208 | } | |
29418aa4 | 209 | |
b30d2f04 TM |
210 | static struct nfs_page * |
211 | nfs_page_find_swap_request(struct page *page) | |
212 | { | |
213 | struct inode *inode = page_file_mapping(page)->host; | |
214 | struct nfs_inode *nfsi = NFS_I(inode); | |
215 | struct nfs_page *req = NULL; | |
216 | if (!PageSwapCache(page)) | |
217 | return NULL; | |
e824f99a | 218 | mutex_lock(&nfsi->commit_mutex); |
b30d2f04 TM |
219 | if (PageSwapCache(page)) { |
220 | req = nfs_page_search_commits_for_head_request_locked(nfsi, | |
221 | page); | |
222 | if (req) { | |
223 | WARN_ON_ONCE(req->wb_head != req); | |
224 | kref_get(&req->wb_kref); | |
225 | } | |
226 | } | |
e824f99a | 227 | mutex_unlock(&nfsi->commit_mutex); |
277459d2 TM |
228 | return req; |
229 | } | |
230 | ||
84d3a9a9 WAA |
231 | /* |
232 | * nfs_page_find_head_request - find head request associated with @page | |
233 | * | |
234 | * returns matching head request with reference held, or NULL if not found. | |
235 | */ | |
236 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 237 | { |
b30d2f04 | 238 | struct nfs_page *req; |
277459d2 | 239 | |
b30d2f04 TM |
240 | req = nfs_page_find_private_request(page); |
241 | if (!req) | |
242 | req = nfs_page_find_swap_request(page); | |
277459d2 TM |
243 | return req; |
244 | } | |
245 | ||
e00ed89d TM |
246 | static struct nfs_page *nfs_find_and_lock_page_request(struct page *page) |
247 | { | |
248 | struct inode *inode = page_file_mapping(page)->host; | |
249 | struct nfs_page *req, *head; | |
250 | int ret; | |
251 | ||
252 | for (;;) { | |
253 | req = nfs_page_find_head_request(page); | |
254 | if (!req) | |
255 | return req; | |
256 | head = nfs_page_group_lock_head(req); | |
257 | if (head != req) | |
258 | nfs_release_request(req); | |
259 | if (IS_ERR(head)) | |
260 | return head; | |
261 | ret = nfs_cancel_remove_inode(head, inode); | |
262 | if (ret < 0) { | |
263 | nfs_unlock_and_release_request(head); | |
264 | return ERR_PTR(ret); | |
265 | } | |
266 | /* Ensure that nobody removed the request before we locked it */ | |
267 | if (head == nfs_page_private_request(page)) | |
268 | break; | |
269 | if (PageSwapCache(page)) | |
270 | break; | |
271 | nfs_unlock_and_release_request(head); | |
272 | } | |
273 | return head; | |
274 | } | |
275 | ||
1da177e4 LT |
276 | /* Adjust the file length if we're writing beyond the end */ |
277 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
278 | { | |
d56b4ddf | 279 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
280 | loff_t end, i_size; |
281 | pgoff_t end_index; | |
1da177e4 | 282 | |
a3d01454 TM |
283 | spin_lock(&inode->i_lock); |
284 | i_size = i_size_read(inode); | |
09cbfeaf | 285 | end_index = (i_size - 1) >> PAGE_SHIFT; |
8cd79788 | 286 | if (i_size > 0 && page_index(page) < end_index) |
a3d01454 | 287 | goto out; |
d56b4ddf | 288 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 289 | if (i_size >= end) |
a3d01454 | 290 | goto out; |
1da177e4 | 291 | i_size_write(inode, end); |
f6cdfa6d | 292 | NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; |
a3d01454 TM |
293 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
294 | out: | |
295 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
296 | } |
297 | ||
a301b777 | 298 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
d2ceb7e5 | 299 | static void nfs_set_pageerror(struct address_space *mapping) |
a301b777 | 300 | { |
0df68ced TM |
301 | struct inode *inode = mapping->host; |
302 | ||
d2ceb7e5 | 303 | nfs_zap_mapping(mapping->host, mapping); |
0df68ced TM |
304 | /* Force file size revalidation */ |
305 | spin_lock(&inode->i_lock); | |
ac46b3d7 TM |
306 | nfs_set_cache_invalid(inode, NFS_INO_REVAL_FORCED | |
307 | NFS_INO_REVAL_PAGECACHE | | |
308 | NFS_INO_INVALID_SIZE); | |
0df68ced | 309 | spin_unlock(&inode->i_lock); |
a301b777 TM |
310 | } |
311 | ||
6fbda89b TM |
312 | static void nfs_mapping_set_error(struct page *page, int error) |
313 | { | |
b8946d7b TM |
314 | struct address_space *mapping = page_file_mapping(page); |
315 | ||
6fbda89b | 316 | SetPageError(page); |
b8946d7b TM |
317 | mapping_set_error(mapping, error); |
318 | nfs_set_pageerror(mapping); | |
6fbda89b TM |
319 | } |
320 | ||
d72ddcba WAA |
321 | /* |
322 | * nfs_page_group_search_locked | |
323 | * @head - head request of page group | |
324 | * @page_offset - offset into page | |
325 | * | |
326 | * Search page group with head @head to find a request that contains the | |
327 | * page offset @page_offset. | |
328 | * | |
329 | * Returns a pointer to the first matching nfs request, or NULL if no | |
330 | * match is found. | |
331 | * | |
332 | * Must be called with the page group lock held | |
333 | */ | |
334 | static struct nfs_page * | |
335 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
336 | { | |
337 | struct nfs_page *req; | |
338 | ||
d72ddcba WAA |
339 | req = head; |
340 | do { | |
341 | if (page_offset >= req->wb_pgbase && | |
342 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
343 | return req; | |
344 | ||
345 | req = req->wb_this_page; | |
346 | } while (req != head); | |
347 | ||
348 | return NULL; | |
349 | } | |
350 | ||
351 | /* | |
352 | * nfs_page_group_covers_page | |
353 | * @head - head request of page group | |
354 | * | |
355 | * Return true if the page group with head @head covers the whole page, | |
356 | * returns false otherwise | |
357 | */ | |
358 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
359 | { | |
360 | struct nfs_page *tmp; | |
361 | unsigned int pos = 0; | |
362 | unsigned int len = nfs_page_length(req->wb_page); | |
363 | ||
1344b7ea | 364 | nfs_page_group_lock(req); |
d72ddcba | 365 | |
7e8a30f8 | 366 | for (;;) { |
d72ddcba | 367 | tmp = nfs_page_group_search_locked(req->wb_head, pos); |
7e8a30f8 TM |
368 | if (!tmp) |
369 | break; | |
370 | pos = tmp->wb_pgbase + tmp->wb_bytes; | |
371 | } | |
d72ddcba WAA |
372 | |
373 | nfs_page_group_unlock(req); | |
7e8a30f8 | 374 | return pos >= len; |
d72ddcba WAA |
375 | } |
376 | ||
1da177e4 LT |
377 | /* We can set the PG_uptodate flag if we see that a write request |
378 | * covers the full page. | |
379 | */ | |
d72ddcba | 380 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 381 | { |
d72ddcba | 382 | if (PageUptodate(req->wb_page)) |
1da177e4 | 383 | return; |
d72ddcba | 384 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 385 | return; |
d72ddcba | 386 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
387 | } |
388 | ||
1da177e4 LT |
389 | static int wb_priority(struct writeback_control *wbc) |
390 | { | |
e87b4c7a | 391 | int ret = 0; |
cca588d6 | 392 | |
e87b4c7a N |
393 | if (wbc->sync_mode == WB_SYNC_ALL) |
394 | ret = FLUSH_COND_STABLE; | |
e87b4c7a | 395 | return ret; |
1da177e4 LT |
396 | } |
397 | ||
89a09141 PZ |
398 | /* |
399 | * NFS congestion control | |
400 | */ | |
401 | ||
402 | int nfs_congestion_kb; | |
403 | ||
404 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
405 | #define NFS_CONGESTION_OFF_THRESH \ | |
406 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
407 | ||
deed85e7 | 408 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 409 | { |
0db10944 JK |
410 | struct inode *inode = page_file_mapping(page)->host; |
411 | struct nfs_server *nfss = NFS_SERVER(inode); | |
5a6d41b3 TM |
412 | int ret = test_set_page_writeback(page); |
413 | ||
deed85e7 | 414 | WARN_ON_ONCE(ret != 0); |
89a09141 | 415 | |
deed85e7 | 416 | if (atomic_long_inc_return(&nfss->writeback) > |
0db10944 JK |
417 | NFS_CONGESTION_ON_THRESH) |
418 | set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); | |
89a09141 PZ |
419 | } |
420 | ||
20633f04 | 421 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 422 | { |
20633f04 | 423 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 | 424 | struct nfs_server *nfss = NFS_SERVER(inode); |
31a01f09 | 425 | bool is_done; |
89a09141 | 426 | |
31a01f09 TM |
427 | is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); |
428 | nfs_unlock_request(req); | |
429 | if (!is_done) | |
20633f04 WAA |
430 | return; |
431 | ||
432 | end_page_writeback(req->wb_page); | |
c4dc4bee | 433 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
0db10944 | 434 | clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); |
89a09141 PZ |
435 | } |
436 | ||
d4581383 WAA |
437 | /* |
438 | * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests | |
439 | * | |
440 | * @destroy_list - request list (using wb_this_page) terminated by @old_head | |
441 | * @old_head - the old head of the list | |
442 | * | |
443 | * All subrequests must be locked and removed from all lists, so at this point | |
444 | * they are only "active" in this function, and possibly in nfs_wait_on_request | |
445 | * with a reference held by some other context. | |
446 | */ | |
447 | static void | |
448 | nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, | |
b66aaa8d TM |
449 | struct nfs_page *old_head, |
450 | struct inode *inode) | |
d4581383 WAA |
451 | { |
452 | while (destroy_list) { | |
453 | struct nfs_page *subreq = destroy_list; | |
454 | ||
455 | destroy_list = (subreq->wb_this_page == old_head) ? | |
456 | NULL : subreq->wb_this_page; | |
457 | ||
08ca8b21 TM |
458 | /* Note: lock subreq in order to change subreq->wb_head */ |
459 | nfs_page_set_headlock(subreq); | |
d4581383 WAA |
460 | WARN_ON_ONCE(old_head != subreq->wb_head); |
461 | ||
462 | /* make sure old group is not used */ | |
d4581383 | 463 | subreq->wb_this_page = subreq; |
08ca8b21 | 464 | subreq->wb_head = subreq; |
d4581383 | 465 | |
902a4c00 TM |
466 | clear_bit(PG_REMOVE, &subreq->wb_flags); |
467 | ||
5b2b5187 TM |
468 | /* Note: races with nfs_page_group_destroy() */ |
469 | if (!kref_read(&subreq->wb_kref)) { | |
5b2b5187 | 470 | /* Check if we raced with nfs_page_group_destroy() */ |
08ca8b21 TM |
471 | if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) { |
472 | nfs_page_clear_headlock(subreq); | |
5b2b5187 | 473 | nfs_free_request(subreq); |
08ca8b21 TM |
474 | } else |
475 | nfs_page_clear_headlock(subreq); | |
5b2b5187 TM |
476 | continue; |
477 | } | |
08ca8b21 | 478 | nfs_page_clear_headlock(subreq); |
d4581383 | 479 | |
add42de3 | 480 | nfs_release_request(old_head); |
5b2b5187 TM |
481 | |
482 | if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { | |
483 | nfs_release_request(subreq); | |
a6b6d5b8 | 484 | atomic_long_dec(&NFS_I(inode)->nrequests); |
d4581383 | 485 | } |
5b2b5187 | 486 | |
5b2b5187 TM |
487 | /* subreq is now totally disconnected from page group or any |
488 | * write / commit lists. last chance to wake any waiters */ | |
489 | nfs_unlock_and_release_request(subreq); | |
d4581383 WAA |
490 | } |
491 | } | |
492 | ||
493 | /* | |
e00ed89d TM |
494 | * nfs_join_page_group - destroy subrequests of the head req |
495 | * @head: the page used to lookup the "page group" of nfs_page structures | |
496 | * @inode: Inode to which the request belongs. | |
d4581383 WAA |
497 | * |
498 | * This function joins all sub requests to the head request by first | |
499 | * locking all requests in the group, cancelling any pending operations | |
500 | * and finally updating the head request to cover the whole range covered by | |
501 | * the (former) group. All subrequests are removed from any write or commit | |
502 | * lists, unlinked from the group and destroyed. | |
d4581383 | 503 | */ |
ed5d588f | 504 | void |
e00ed89d | 505 | nfs_join_page_group(struct nfs_page *head, struct inode *inode) |
e261f51f | 506 | { |
e00ed89d | 507 | struct nfs_page *subreq; |
d4581383 | 508 | struct nfs_page *destroy_list = NULL; |
a62f8e3b | 509 | unsigned int pgbase, off, bytes; |
a62f8e3b TM |
510 | |
511 | pgbase = head->wb_pgbase; | |
512 | bytes = head->wb_bytes; | |
513 | off = head->wb_offset; | |
a0e265bc TM |
514 | for (subreq = head->wb_this_page; subreq != head; |
515 | subreq = subreq->wb_this_page) { | |
a62f8e3b TM |
516 | /* Subrequests should always form a contiguous range */ |
517 | if (pgbase > subreq->wb_pgbase) { | |
518 | off -= pgbase - subreq->wb_pgbase; | |
519 | bytes += pgbase - subreq->wb_pgbase; | |
520 | pgbase = subreq->wb_pgbase; | |
309a1d65 | 521 | } |
a62f8e3b TM |
522 | bytes = max(subreq->wb_pgbase + subreq->wb_bytes |
523 | - pgbase, bytes); | |
a0e265bc | 524 | } |
d4581383 | 525 | |
a62f8e3b TM |
526 | /* Set the head request's range to cover the former page group */ |
527 | head->wb_pgbase = pgbase; | |
528 | head->wb_bytes = bytes; | |
529 | head->wb_offset = off; | |
530 | ||
d4581383 WAA |
531 | /* Now that all requests are locked, make sure they aren't on any list. |
532 | * Commit list removal accounting is done after locks are dropped */ | |
533 | subreq = head; | |
534 | do { | |
411a99ad | 535 | nfs_clear_request_commit(subreq); |
d4581383 WAA |
536 | subreq = subreq->wb_this_page; |
537 | } while (subreq != head); | |
538 | ||
539 | /* unlink subrequests from head, destroy them later */ | |
540 | if (head->wb_this_page != head) { | |
541 | /* destroy list will be terminated by head */ | |
542 | destroy_list = head->wb_this_page; | |
543 | head->wb_this_page = head; | |
e261f51f | 544 | } |
d4581383 | 545 | |
e00ed89d TM |
546 | nfs_destroy_unlinked_subrequests(destroy_list, head, inode); |
547 | } | |
b66aaa8d | 548 | |
e00ed89d TM |
549 | /* |
550 | * nfs_lock_and_join_requests - join all subreqs to the head req | |
551 | * @page: the page used to lookup the "page group" of nfs_page structures | |
552 | * | |
553 | * This function joins all sub requests to the head request by first | |
554 | * locking all requests in the group, cancelling any pending operations | |
555 | * and finally updating the head request to cover the whole range covered by | |
556 | * the (former) group. All subrequests are removed from any write or commit | |
557 | * lists, unlinked from the group and destroyed. | |
558 | * | |
559 | * Returns a locked, referenced pointer to the head request - which after | |
560 | * this call is guaranteed to be the only request associated with the page. | |
561 | * Returns NULL if no requests are found for @page, or a ERR_PTR if an | |
562 | * error was encountered. | |
563 | */ | |
564 | static struct nfs_page * | |
565 | nfs_lock_and_join_requests(struct page *page) | |
566 | { | |
567 | struct inode *inode = page_file_mapping(page)->host; | |
568 | struct nfs_page *head; | |
569 | int ret; | |
d4581383 | 570 | |
e00ed89d TM |
571 | /* |
572 | * A reference is taken only on the head request which acts as a | |
573 | * reference to the whole page group - the group will not be destroyed | |
574 | * until the head reference is released. | |
575 | */ | |
576 | head = nfs_find_and_lock_page_request(page); | |
577 | if (IS_ERR_OR_NULL(head)) | |
578 | return head; | |
d4581383 | 579 | |
e00ed89d TM |
580 | /* lock each request in the page group */ |
581 | ret = nfs_page_group_lock_subrequests(head); | |
582 | if (ret < 0) { | |
b5bab9bf | 583 | nfs_unlock_and_release_request(head); |
e00ed89d | 584 | return ERR_PTR(ret); |
b5bab9bf TM |
585 | } |
586 | ||
e00ed89d | 587 | nfs_join_page_group(head, inode); |
0671d8f1 | 588 | |
e00ed89d | 589 | return head; |
074cc1de TM |
590 | } |
591 | ||
6fbda89b | 592 | static void nfs_write_error(struct nfs_page *req, int error) |
0bcbf039 | 593 | { |
861e1671 | 594 | trace_nfs_write_error(req, error); |
6fbda89b | 595 | nfs_mapping_set_error(req->wb_page, error); |
06c9fdf3 | 596 | nfs_inode_remove_request(req); |
0bcbf039 | 597 | nfs_end_page_writeback(req); |
1f84ccdf | 598 | nfs_release_request(req); |
0bcbf039 PT |
599 | } |
600 | ||
074cc1de TM |
601 | /* |
602 | * Find an associated nfs write request, and prepare to flush it out | |
603 | * May return an error if the user signalled nfs_wait_on_request(). | |
604 | */ | |
605 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
6d17d653 | 606 | struct page *page) |
074cc1de TM |
607 | { |
608 | struct nfs_page *req; | |
609 | int ret = 0; | |
610 | ||
6d17d653 | 611 | req = nfs_lock_and_join_requests(page); |
074cc1de TM |
612 | if (!req) |
613 | goto out; | |
614 | ret = PTR_ERR(req); | |
615 | if (IS_ERR(req)) | |
616 | goto out; | |
617 | ||
deed85e7 TM |
618 | nfs_set_page_writeback(page); |
619 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 620 | |
a6598813 | 621 | /* If there is a fatal error that covers this write, just exit */ |
96c41455 TM |
622 | ret = pgio->pg_error; |
623 | if (nfs_error_is_fatal_on_server(ret)) | |
a6598813 TM |
624 | goto out_launder; |
625 | ||
96c41455 | 626 | ret = 0; |
f8512ad0 | 627 | if (!nfs_pageio_add_request(pgio, req)) { |
074cc1de | 628 | ret = pgio->pg_error; |
0bcbf039 | 629 | /* |
c373fff7 | 630 | * Remove the problematic req upon fatal errors on the server |
0bcbf039 PT |
631 | */ |
632 | if (nfs_error_is_fatal(ret)) { | |
c373fff7 | 633 | if (nfs_error_is_fatal_on_server(ret)) |
a6598813 | 634 | goto out_launder; |
8fc75bed TM |
635 | } else |
636 | ret = -EAGAIN; | |
d6c843b9 | 637 | nfs_redirty_request(req); |
96c41455 | 638 | pgio->pg_error = 0; |
40f90271 TM |
639 | } else |
640 | nfs_add_stats(page_file_mapping(page)->host, | |
641 | NFSIOS_WRITEPAGES, 1); | |
074cc1de TM |
642 | out: |
643 | return ret; | |
a6598813 | 644 | out_launder: |
6fbda89b | 645 | nfs_write_error(req, ret); |
14bebe3c | 646 | return 0; |
e261f51f TM |
647 | } |
648 | ||
d6c843b9 | 649 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, |
c373fff7 | 650 | struct nfs_pageio_descriptor *pgio) |
1da177e4 | 651 | { |
cfb506e1 | 652 | int ret; |
1da177e4 | 653 | |
8cd79788 | 654 | nfs_pageio_cond_complete(pgio, page_index(page)); |
6d17d653 | 655 | ret = nfs_page_async_flush(pgio, page); |
cfb506e1 TM |
656 | if (ret == -EAGAIN) { |
657 | redirty_page_for_writepage(wbc, page); | |
96c41455 | 658 | ret = AOP_WRITEPAGE_ACTIVATE; |
cfb506e1 TM |
659 | } |
660 | return ret; | |
f758c885 | 661 | } |
7fe7f848 | 662 | |
f758c885 TM |
663 | /* |
664 | * Write an mmapped page to the server. | |
665 | */ | |
d6c843b9 | 666 | static int nfs_writepage_locked(struct page *page, |
c373fff7 | 667 | struct writeback_control *wbc) |
f758c885 TM |
668 | { |
669 | struct nfs_pageio_descriptor pgio; | |
40f90271 | 670 | struct inode *inode = page_file_mapping(page)->host; |
f758c885 | 671 | int err; |
49a70f27 | 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 TM |
678 | nfs_pageio_complete(&pgio); |
679 | if (err < 0) | |
680 | return err; | |
96c41455 | 681 | if (nfs_error_is_fatal(pgio.pg_error)) |
f758c885 TM |
682 | return pgio.pg_error; |
683 | return 0; | |
4d770ccf TM |
684 | } |
685 | ||
686 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
687 | { | |
f758c885 | 688 | int ret; |
4d770ccf | 689 | |
c373fff7 | 690 | ret = nfs_writepage_locked(page, wbc); |
96c41455 TM |
691 | if (ret != AOP_WRITEPAGE_ACTIVATE) |
692 | unlock_page(page); | |
f758c885 TM |
693 | return ret; |
694 | } | |
695 | ||
696 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
697 | { | |
698 | int ret; | |
699 | ||
c373fff7 | 700 | ret = nfs_do_writepage(page, wbc, data); |
96c41455 TM |
701 | if (ret != AOP_WRITEPAGE_ACTIVATE) |
702 | unlock_page(page); | |
f758c885 | 703 | return ret; |
1da177e4 LT |
704 | } |
705 | ||
919e3bd9 TM |
706 | static void nfs_io_completion_commit(void *inode) |
707 | { | |
708 | nfs_commit_inode(inode, 0); | |
709 | } | |
710 | ||
1da177e4 LT |
711 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
712 | { | |
1da177e4 | 713 | struct inode *inode = mapping->host; |
c63c7b05 | 714 | struct nfs_pageio_descriptor pgio; |
ed7bcdb3 TM |
715 | struct nfs_io_completion *ioc = NULL; |
716 | unsigned int mntflags = NFS_SERVER(inode)->flags; | |
717 | int priority = 0; | |
1da177e4 LT |
718 | int err; |
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 | |
ed7bcdb3 | 731 | nfs_pageio_init_write(&pgio, inode, priority, false, |
a20c93e3 | 732 | &nfs_async_write_completion_ops); |
919e3bd9 | 733 | pgio.pg_io_completion = ioc; |
f758c885 | 734 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
96c41455 | 735 | pgio.pg_error = 0; |
c63c7b05 | 736 | nfs_pageio_complete(&pgio); |
919e3bd9 | 737 | nfs_io_completion_put(ioc); |
72cb77f4 | 738 | |
f758c885 | 739 | if (err < 0) |
72cb77f4 TM |
740 | goto out_err; |
741 | err = pgio.pg_error; | |
96c41455 | 742 | if (nfs_error_is_fatal(err)) |
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)) { | |
861e1671 | 1004 | trace_nfs_comp_error(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 TM |
1041 | restart: |
1042 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
7ad84aa9 | 1043 | kref_get(&req->wb_kref); |
2ce209c4 TM |
1044 | if (!nfs_lock_request(req)) { |
1045 | int status; | |
137da553 TM |
1046 | |
1047 | /* Prevent deadlock with nfs_lock_and_join_requests */ | |
1048 | if (!list_empty(dst)) { | |
1049 | nfs_release_request(req); | |
1050 | continue; | |
1051 | } | |
1052 | /* Ensure we make progress to prevent livelock */ | |
2ce209c4 TM |
1053 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
1054 | status = nfs_wait_on_request(req); | |
1055 | nfs_release_request(req); | |
1056 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); | |
1057 | if (status < 0) | |
1058 | break; | |
137da553 | 1059 | goto restart; |
2ce209c4 | 1060 | } |
ea2cf228 | 1061 | nfs_request_remove_commit_list(req, cinfo); |
5d2a9d9d | 1062 | clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); |
8dd37758 TM |
1063 | nfs_list_add_request(req, dst); |
1064 | ret++; | |
1763da12 | 1065 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 1066 | break; |
e824f99a | 1067 | cond_resched(); |
d6d6dc7c FI |
1068 | } |
1069 | return ret; | |
fb8a1f11 | 1070 | } |
5d2a9d9d | 1071 | EXPORT_SYMBOL_GPL(nfs_scan_commit_list); |
fb8a1f11 | 1072 | |
1da177e4 LT |
1073 | /* |
1074 | * nfs_scan_commit - Scan an inode for commit requests | |
1075 | * @inode: NFS inode to scan | |
ea2cf228 FI |
1076 | * @dst: mds destination list |
1077 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
1078 | * |
1079 | * Moves requests from the inode's 'commit' request list. | |
1080 | * The requests are *not* checked to ensure that they form a contiguous set. | |
1081 | */ | |
1763da12 | 1082 | int |
ea2cf228 FI |
1083 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
1084 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1085 | { |
d6d6dc7c | 1086 | int ret = 0; |
fb8a1f11 | 1087 | |
5cb953d4 TM |
1088 | if (!atomic_long_read(&cinfo->mds->ncommit)) |
1089 | return 0; | |
e824f99a | 1090 | mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); |
5cb953d4 | 1091 | if (atomic_long_read(&cinfo->mds->ncommit) > 0) { |
8dd37758 | 1092 | const int max = INT_MAX; |
d6d6dc7c | 1093 | |
ea2cf228 FI |
1094 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
1095 | cinfo, max); | |
1096 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 1097 | } |
e824f99a | 1098 | mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); |
ff778d02 | 1099 | return ret; |
1da177e4 | 1100 | } |
d6d6dc7c | 1101 | |
1da177e4 | 1102 | /* |
e7d39069 TM |
1103 | * Search for an existing write request, and attempt to update |
1104 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 1105 | * |
e7d39069 TM |
1106 | * If the attempt fails, then the existing request is flushed out |
1107 | * to disk. | |
1da177e4 | 1108 | */ |
e7d39069 TM |
1109 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
1110 | struct page *page, | |
1111 | unsigned int offset, | |
1112 | unsigned int bytes) | |
1da177e4 | 1113 | { |
e7d39069 TM |
1114 | struct nfs_page *req; |
1115 | unsigned int rqend; | |
1116 | unsigned int end; | |
1117 | int error; | |
1118 | ||
1da177e4 LT |
1119 | end = offset + bytes; |
1120 | ||
f6032f21 TM |
1121 | req = nfs_lock_and_join_requests(page); |
1122 | if (IS_ERR_OR_NULL(req)) | |
1123 | return req; | |
1da177e4 | 1124 | |
f6032f21 TM |
1125 | rqend = req->wb_offset + req->wb_bytes; |
1126 | /* | |
1127 | * Tell the caller to flush out the request if | |
1128 | * the offsets are non-contiguous. | |
1129 | * Note: nfs_flush_incompatible() will already | |
1130 | * have flushed out requests having wrong owners. | |
1131 | */ | |
1132 | if (offset > rqend || end < req->wb_offset) | |
1133 | goto out_flushme; | |
1da177e4 LT |
1134 | |
1135 | /* Okay, the request matches. Update the region */ | |
1136 | if (offset < req->wb_offset) { | |
1137 | req->wb_offset = offset; | |
1138 | req->wb_pgbase = offset; | |
1da177e4 | 1139 | } |
1da177e4 LT |
1140 | if (end > rqend) |
1141 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1142 | else |
1143 | req->wb_bytes = rqend - req->wb_offset; | |
33344e0f | 1144 | req->wb_nio = 0; |
e7d39069 TM |
1145 | return req; |
1146 | out_flushme: | |
f6032f21 TM |
1147 | /* |
1148 | * Note: we mark the request dirty here because | |
1149 | * nfs_lock_and_join_requests() cannot preserve | |
1150 | * commit flags, so we have to replay the write. | |
1151 | */ | |
1152 | nfs_mark_request_dirty(req); | |
1153 | nfs_unlock_and_release_request(req); | |
e7d39069 | 1154 | error = nfs_wb_page(inode, page); |
f6032f21 | 1155 | return (error < 0) ? ERR_PTR(error) : NULL; |
e7d39069 TM |
1156 | } |
1157 | ||
1158 | /* | |
1159 | * Try to update an existing write request, or create one if there is none. | |
1160 | * | |
1161 | * Note: Should always be called with the Page Lock held to prevent races | |
1162 | * if we have to add a new request. Also assumes that the caller has | |
1163 | * already called nfs_flush_incompatible() if necessary. | |
1164 | */ | |
1165 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1166 | struct page *page, unsigned int offset, unsigned int bytes) | |
1167 | { | |
d56b4ddf | 1168 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1169 | struct nfs_page *req; |
1da177e4 | 1170 | |
e7d39069 TM |
1171 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1172 | if (req != NULL) | |
1173 | goto out; | |
28b1d3f5 | 1174 | req = nfs_create_request(ctx, page, offset, bytes); |
e7d39069 TM |
1175 | if (IS_ERR(req)) |
1176 | goto out; | |
d6d6dc7c | 1177 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1178 | out: |
61e930a9 | 1179 | return req; |
1da177e4 LT |
1180 | } |
1181 | ||
e7d39069 TM |
1182 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1183 | unsigned int offset, unsigned int count) | |
1184 | { | |
1185 | struct nfs_page *req; | |
1186 | ||
1187 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1188 | if (IS_ERR(req)) | |
1189 | return PTR_ERR(req); | |
1190 | /* Update file length */ | |
1191 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1192 | nfs_mark_uptodate(req); |
a6305ddb | 1193 | nfs_mark_request_dirty(req); |
1d1afcbc | 1194 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1195 | return 0; |
1196 | } | |
1197 | ||
1da177e4 LT |
1198 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1199 | { | |
cd3758e3 | 1200 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1201 | struct nfs_lock_context *l_ctx; |
bd61e0a9 | 1202 | struct file_lock_context *flctx = file_inode(file)->i_flctx; |
1da177e4 | 1203 | struct nfs_page *req; |
1a54533e | 1204 | int do_flush, status; |
1da177e4 LT |
1205 | /* |
1206 | * Look for a request corresponding to this page. If there | |
1207 | * is one, and it belongs to another file, we flush it out | |
1208 | * before we try to copy anything into the page. Do this | |
1209 | * due to the lack of an ACCESS-type call in NFSv2. | |
1210 | * Also do the same if we find a request from an existing | |
1211 | * dropped page. | |
1212 | */ | |
1a54533e | 1213 | do { |
84d3a9a9 | 1214 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1215 | if (req == NULL) |
1216 | return 0; | |
2a369153 | 1217 | l_ctx = req->wb_lock_context; |
138a2935 | 1218 | do_flush = req->wb_page != page || |
9fcd5960 | 1219 | !nfs_match_open_context(nfs_req_openctx(req), ctx); |
bd61e0a9 JL |
1220 | if (l_ctx && flctx && |
1221 | !(list_empty_careful(&flctx->flc_posix) && | |
1222 | list_empty_careful(&flctx->flc_flock))) { | |
d51fdb87 | 1223 | do_flush |= l_ctx->lockowner != current->files; |
5263e31e | 1224 | } |
1da177e4 | 1225 | nfs_release_request(req); |
1a54533e TM |
1226 | if (!do_flush) |
1227 | return 0; | |
d56b4ddf | 1228 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1229 | } while (status == 0); |
1230 | return status; | |
1da177e4 LT |
1231 | } |
1232 | ||
dc24826b AA |
1233 | /* |
1234 | * Avoid buffered writes when a open context credential's key would | |
1235 | * expire soon. | |
1236 | * | |
1237 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1238 | * | |
1239 | * Return 0 and set a credential flag which triggers the inode to flush | |
1240 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1241 | * RPC_KEY_EXPIRE_TIMEO. | |
1242 | */ | |
1243 | int | |
1244 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1245 | { | |
1246 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
dc24826b | 1247 | |
ddf529ee N |
1248 | if (nfs_ctx_key_to_expire(ctx, inode) && |
1249 | !ctx->ll_cred) | |
1250 | /* Already expired! */ | |
1251 | return -EACCES; | |
1252 | return 0; | |
dc24826b AA |
1253 | } |
1254 | ||
1255 | /* | |
1256 | * Test if the open context credential key is marked to expire soon. | |
1257 | */ | |
ce52914e | 1258 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) |
dc24826b | 1259 | { |
ce52914e | 1260 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; |
ddf529ee N |
1261 | struct rpc_cred *cred = ctx->ll_cred; |
1262 | struct auth_cred acred = { | |
a52458b4 | 1263 | .cred = ctx->cred, |
ddf529ee N |
1264 | }; |
1265 | ||
1266 | if (cred && !cred->cr_ops->crmatch(&acred, cred, 0)) { | |
1267 | put_rpccred(cred); | |
1268 | ctx->ll_cred = NULL; | |
1269 | cred = NULL; | |
1270 | } | |
1271 | if (!cred) | |
1272 | cred = auth->au_ops->lookup_cred(auth, &acred, 0); | |
1273 | if (!cred || IS_ERR(cred)) | |
1274 | return true; | |
1275 | ctx->ll_cred = cred; | |
1276 | return !!(cred->cr_ops->crkey_timeout && | |
1277 | cred->cr_ops->crkey_timeout(cred)); | |
dc24826b AA |
1278 | } |
1279 | ||
5d47a356 TM |
1280 | /* |
1281 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1282 | * the PageUptodate() flag. In this case, we will need to turn off | |
1283 | * write optimisations that depend on the page contents being correct. | |
1284 | */ | |
fc9dc401 TM |
1285 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode, |
1286 | unsigned int pagelen) | |
5d47a356 | 1287 | { |
d529ef83 JL |
1288 | struct nfs_inode *nfsi = NFS_I(inode); |
1289 | ||
8d197a56 TM |
1290 | if (nfs_have_delegated_attributes(inode)) |
1291 | goto out; | |
fc9dc401 | 1292 | if (nfsi->cache_validity & |
13c0b082 | 1293 | (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE)) |
d529ef83 | 1294 | return false; |
4db72b40 | 1295 | smp_rmb(); |
fc9dc401 | 1296 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags) && pagelen != 0) |
8d197a56 TM |
1297 | return false; |
1298 | out: | |
fc9dc401 | 1299 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA && pagelen != 0) |
18dd78c4 | 1300 | return false; |
8d197a56 | 1301 | return PageUptodate(page) != 0; |
5d47a356 TM |
1302 | } |
1303 | ||
5263e31e JL |
1304 | static bool |
1305 | is_whole_file_wrlock(struct file_lock *fl) | |
1306 | { | |
1307 | return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && | |
1308 | fl->fl_type == F_WRLCK; | |
1309 | } | |
1310 | ||
c7559663 SM |
1311 | /* If we know the page is up to date, and we're not using byte range locks (or |
1312 | * if we have the whole file locked for writing), it may be more efficient to | |
1313 | * extend the write to cover the entire page in order to avoid fragmentation | |
1314 | * inefficiencies. | |
1315 | * | |
263b4509 SM |
1316 | * If the file is opened for synchronous writes then we can just skip the rest |
1317 | * of the checks. | |
c7559663 | 1318 | */ |
fc9dc401 TM |
1319 | static int nfs_can_extend_write(struct file *file, struct page *page, |
1320 | struct inode *inode, unsigned int pagelen) | |
c7559663 | 1321 | { |
5263e31e JL |
1322 | int ret; |
1323 | struct file_lock_context *flctx = inode->i_flctx; | |
1324 | struct file_lock *fl; | |
1325 | ||
c7559663 SM |
1326 | if (file->f_flags & O_DSYNC) |
1327 | return 0; | |
fc9dc401 | 1328 | if (!nfs_write_pageuptodate(page, inode, pagelen)) |
263b4509 | 1329 | return 0; |
c7559663 SM |
1330 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1331 | return 1; | |
bd61e0a9 JL |
1332 | if (!flctx || (list_empty_careful(&flctx->flc_flock) && |
1333 | list_empty_careful(&flctx->flc_posix))) | |
8fa4592a | 1334 | return 1; |
5263e31e JL |
1335 | |
1336 | /* Check to see if there are whole file write locks */ | |
5263e31e | 1337 | ret = 0; |
6109c850 | 1338 | spin_lock(&flctx->flc_lock); |
bd61e0a9 JL |
1339 | if (!list_empty(&flctx->flc_posix)) { |
1340 | fl = list_first_entry(&flctx->flc_posix, struct file_lock, | |
1341 | fl_list); | |
1342 | if (is_whole_file_wrlock(fl)) | |
1343 | ret = 1; | |
1344 | } else if (!list_empty(&flctx->flc_flock)) { | |
5263e31e JL |
1345 | fl = list_first_entry(&flctx->flc_flock, struct file_lock, |
1346 | fl_list); | |
1347 | if (fl->fl_type == F_WRLCK) | |
1348 | ret = 1; | |
1349 | } | |
6109c850 | 1350 | spin_unlock(&flctx->flc_lock); |
5263e31e | 1351 | return ret; |
c7559663 SM |
1352 | } |
1353 | ||
1da177e4 LT |
1354 | /* |
1355 | * Update and possibly write a cached page of an NFS file. | |
1356 | * | |
1357 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1358 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1359 | */ | |
1360 | int nfs_updatepage(struct file *file, struct page *page, | |
1361 | unsigned int offset, unsigned int count) | |
1362 | { | |
cd3758e3 | 1363 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d2ceb7e5 BC |
1364 | struct address_space *mapping = page_file_mapping(page); |
1365 | struct inode *inode = mapping->host; | |
fc9dc401 | 1366 | unsigned int pagelen = nfs_page_length(page); |
1da177e4 LT |
1367 | int status = 0; |
1368 | ||
91d5b470 CL |
1369 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1370 | ||
6de1472f AV |
1371 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1372 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1373 | |
149a4fdd BC |
1374 | if (!count) |
1375 | goto out; | |
1376 | ||
fc9dc401 TM |
1377 | if (nfs_can_extend_write(file, page, inode, pagelen)) { |
1378 | count = max(count + offset, pagelen); | |
1da177e4 | 1379 | offset = 0; |
1da177e4 LT |
1380 | } |
1381 | ||
e21195a7 | 1382 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 | 1383 | if (status < 0) |
d2ceb7e5 | 1384 | nfs_set_pageerror(mapping); |
59b7c05f TM |
1385 | else |
1386 | __set_page_dirty_nobuffers(page); | |
149a4fdd | 1387 | out: |
48186c7d | 1388 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1389 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1390 | return status; |
1391 | } | |
1392 | ||
3ff7576d | 1393 | static int flush_task_priority(int how) |
1da177e4 LT |
1394 | { |
1395 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1396 | case FLUSH_HIGHPRI: | |
1397 | return RPC_PRIORITY_HIGH; | |
1398 | case FLUSH_LOWPRI: | |
1399 | return RPC_PRIORITY_LOW; | |
1400 | } | |
1401 | return RPC_PRIORITY_NORMAL; | |
1402 | } | |
1403 | ||
d45f60c6 WAA |
1404 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1405 | struct rpc_message *msg, | |
abde71f4 | 1406 | const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 | 1407 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1408 | { |
3ff7576d | 1409 | int priority = flush_task_priority(how); |
d138d5d1 | 1410 | |
1ed26f33 | 1411 | task_setup_data->priority = priority; |
fb91fb0e | 1412 | rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); |
5bb2a7cb | 1413 | trace_nfs_initiate_write(hdr); |
275acaaf TM |
1414 | } |
1415 | ||
6d884e8f F |
1416 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1417 | * call this on each, which will prepare them to be retried on next | |
1418 | * writeback using standard nfs. | |
1419 | */ | |
1420 | static void nfs_redirty_request(struct nfs_page *req) | |
1421 | { | |
33344e0f TM |
1422 | /* Bump the transmission count */ |
1423 | req->wb_nio++; | |
6d884e8f | 1424 | nfs_mark_request_dirty(req); |
9fcd5960 | 1425 | set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); |
20633f04 | 1426 | nfs_end_page_writeback(req); |
3aff4ebb | 1427 | nfs_release_request(req); |
6d884e8f F |
1428 | } |
1429 | ||
df3accb8 | 1430 | static void nfs_async_write_error(struct list_head *head, int error) |
6c75dc0d FI |
1431 | { |
1432 | struct nfs_page *req; | |
1433 | ||
1434 | while (!list_empty(head)) { | |
1435 | req = nfs_list_entry(head->next); | |
1436 | nfs_list_remove_request(req); | |
6fbda89b TM |
1437 | if (nfs_error_is_fatal(error)) |
1438 | nfs_write_error(req, error); | |
1439 | else | |
1440 | nfs_redirty_request(req); | |
6c75dc0d FI |
1441 | } |
1442 | } | |
1443 | ||
dc602dd7 TM |
1444 | static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) |
1445 | { | |
df3accb8 | 1446 | nfs_async_write_error(&hdr->pages, 0); |
7be7b3ca TM |
1447 | filemap_fdatawrite_range(hdr->inode->i_mapping, hdr->args.offset, |
1448 | hdr->args.offset + hdr->args.count - 1); | |
dc602dd7 TM |
1449 | } |
1450 | ||
061ae2ed | 1451 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
919e3bd9 | 1452 | .init_hdr = nfs_async_write_init, |
061ae2ed FI |
1453 | .error_cleanup = nfs_async_write_error, |
1454 | .completion = nfs_write_completion, | |
dc602dd7 | 1455 | .reschedule_io = nfs_async_write_reschedule_io, |
061ae2ed FI |
1456 | }; |
1457 | ||
57208fa7 | 1458 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1459 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1460 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1461 | { |
a20c93e3 | 1462 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1463 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1464 | |
1465 | #ifdef CONFIG_NFS_V4_1 | |
1466 | if (server->pnfs_curr_ld && !force_mds) | |
1467 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1468 | #endif | |
4a0de55c | 1469 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
3bde7afd | 1470 | server->wsize, ioflags); |
1751c363 | 1471 | } |
ddda8e0a | 1472 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1473 | |
dce81290 TM |
1474 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1475 | { | |
a7d42ddb WAA |
1476 | struct nfs_pgio_mirror *mirror; |
1477 | ||
6f29b9bb KM |
1478 | if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) |
1479 | pgio->pg_ops->pg_cleanup(pgio); | |
1480 | ||
41d8d5b7 | 1481 | pgio->pg_ops = &nfs_pgio_rw_ops; |
a7d42ddb WAA |
1482 | |
1483 | nfs_pageio_stop_mirroring(pgio); | |
1484 | ||
1485 | mirror = &pgio->pg_mirrors[0]; | |
1486 | mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
dce81290 | 1487 | } |
1f945357 | 1488 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1489 | |
1da177e4 | 1490 | |
0b7c0153 FI |
1491 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1492 | { | |
1493 | struct nfs_commit_data *data = calldata; | |
1494 | ||
1495 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1496 | } | |
1497 | ||
1f2edbe3 TM |
1498 | /* |
1499 | * Special version of should_remove_suid() that ignores capabilities. | |
1500 | */ | |
1501 | static int nfs_should_remove_suid(const struct inode *inode) | |
1502 | { | |
1503 | umode_t mode = inode->i_mode; | |
1504 | int kill = 0; | |
1505 | ||
1506 | /* suid always must be killed */ | |
1507 | if (unlikely(mode & S_ISUID)) | |
1508 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1509 | |
1f2edbe3 TM |
1510 | /* |
1511 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1512 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1513 | */ | |
1514 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1515 | kill |= ATTR_KILL_SGID; | |
1516 | ||
1517 | if (unlikely(kill && S_ISREG(mode))) | |
1518 | return kill; | |
1519 | ||
1520 | return 0; | |
1521 | } | |
788e7a89 | 1522 | |
a08a8cd3 TM |
1523 | static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, |
1524 | struct nfs_fattr *fattr) | |
1525 | { | |
1526 | struct nfs_pgio_args *argp = &hdr->args; | |
1527 | struct nfs_pgio_res *resp = &hdr->res; | |
2b83d3de | 1528 | u64 size = argp->offset + resp->count; |
a08a8cd3 TM |
1529 | |
1530 | if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) | |
2b83d3de TM |
1531 | fattr->size = size; |
1532 | if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { | |
1533 | fattr->valid &= ~NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 | 1534 | return; |
2b83d3de TM |
1535 | } |
1536 | if (size != fattr->size) | |
a08a8cd3 TM |
1537 | return; |
1538 | /* Set attribute barrier */ | |
1539 | nfs_fattr_set_barrier(fattr); | |
2b83d3de TM |
1540 | /* ...and update size */ |
1541 | fattr->valid |= NFS_ATTR_FATTR_SIZE; | |
a08a8cd3 TM |
1542 | } |
1543 | ||
1544 | void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) | |
1545 | { | |
2b83d3de | 1546 | struct nfs_fattr *fattr = &hdr->fattr; |
a08a8cd3 TM |
1547 | struct inode *inode = hdr->inode; |
1548 | ||
a08a8cd3 TM |
1549 | spin_lock(&inode->i_lock); |
1550 | nfs_writeback_check_extend(hdr, fattr); | |
1551 | nfs_post_op_update_inode_force_wcc_locked(inode, fattr); | |
1552 | spin_unlock(&inode->i_lock); | |
1553 | } | |
1554 | EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); | |
1555 | ||
1da177e4 LT |
1556 | /* |
1557 | * This function is called when the WRITE call is complete. | |
1558 | */ | |
d45f60c6 WAA |
1559 | static int nfs_writeback_done(struct rpc_task *task, |
1560 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1561 | struct inode *inode) |
1da177e4 | 1562 | { |
788e7a89 | 1563 | int status; |
1da177e4 | 1564 | |
f551e44f CL |
1565 | /* |
1566 | * ->write_done will attempt to use post-op attributes to detect | |
1567 | * conflicting writes by other clients. A strict interpretation | |
1568 | * of close-to-open would allow us to continue caching even if | |
1569 | * another writer had changed the file, but some applications | |
1570 | * depend on tighter cache coherency when writing. | |
1571 | */ | |
d45f60c6 | 1572 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1573 | if (status != 0) |
0eecb214 | 1574 | return status; |
8224b273 | 1575 | |
d45f60c6 | 1576 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
5bb2a7cb | 1577 | trace_nfs_writeback_done(task, hdr); |
91d5b470 | 1578 | |
d45f60c6 WAA |
1579 | if (hdr->res.verf->committed < hdr->args.stable && |
1580 | task->tk_status >= 0) { | |
1da177e4 LT |
1581 | /* We tried a write call, but the server did not |
1582 | * commit data to stable storage even though we | |
1583 | * requested it. | |
1584 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1585 | * the server reports NFS_DATA_SYNC, but performs | |
1586 | * NFS_FILE_SYNC. We therefore implement this checking | |
1587 | * as a dprintk() in order to avoid filling syslog. | |
1588 | */ | |
1589 | static unsigned long complain; | |
1590 | ||
a69aef14 | 1591 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1592 | if (time_before(complain, jiffies)) { |
48186c7d | 1593 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1594 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1595 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
d45f60c6 | 1596 | hdr->res.verf->committed, hdr->args.stable); |
1da177e4 LT |
1597 | complain = jiffies + 300 * HZ; |
1598 | } | |
1599 | } | |
1f2edbe3 TM |
1600 | |
1601 | /* Deal with the suid/sgid bit corner case */ | |
16e14375 TM |
1602 | if (nfs_should_remove_suid(inode)) { |
1603 | spin_lock(&inode->i_lock); | |
720869eb | 1604 | nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE); |
16e14375 TM |
1605 | spin_unlock(&inode->i_lock); |
1606 | } | |
0eecb214 AS |
1607 | return 0; |
1608 | } | |
1609 | ||
1610 | /* | |
1611 | * This function is called when the WRITE call is complete. | |
1612 | */ | |
d45f60c6 WAA |
1613 | static void nfs_writeback_result(struct rpc_task *task, |
1614 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1615 | { |
d45f60c6 WAA |
1616 | struct nfs_pgio_args *argp = &hdr->args; |
1617 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1618 | |
1619 | if (resp->count < argp->count) { | |
1da177e4 LT |
1620 | static unsigned long complain; |
1621 | ||
6c75dc0d | 1622 | /* This a short write! */ |
d45f60c6 | 1623 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1624 | |
1da177e4 | 1625 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1626 | if (resp->count == 0) { |
1627 | if (time_before(complain, jiffies)) { | |
1628 | printk(KERN_WARNING | |
1629 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1630 | argp->count); | |
1631 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1632 | } |
d45f60c6 | 1633 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1634 | task->tk_status = -EIO; |
13602896 | 1635 | return; |
1da177e4 | 1636 | } |
f8417b48 KM |
1637 | |
1638 | /* For non rpc-based layout drivers, retry-through-MDS */ | |
1639 | if (!task->tk_ops) { | |
1640 | hdr->pnfs_error = -EAGAIN; | |
1641 | return; | |
1642 | } | |
1643 | ||
6c75dc0d FI |
1644 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1645 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1646 | /* Resend from where the server left off */ | |
d45f60c6 | 1647 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1648 | argp->offset += resp->count; |
1649 | argp->pgbase += resp->count; | |
1650 | argp->count -= resp->count; | |
1651 | } else { | |
1652 | /* Resend as a stable write in order to avoid | |
1653 | * headaches in the case of a server crash. | |
1654 | */ | |
1655 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1656 | } |
8c9cb714 TM |
1657 | resp->count = 0; |
1658 | resp->verf->committed = 0; | |
6c75dc0d | 1659 | rpc_restart_call_prepare(task); |
1da177e4 | 1660 | } |
1da177e4 LT |
1661 | } |
1662 | ||
af7cf057 | 1663 | static int wait_on_commit(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1664 | { |
723c921e PZ |
1665 | return wait_var_event_killable(&cinfo->rpcs_out, |
1666 | !atomic_read(&cinfo->rpcs_out)); | |
af7cf057 | 1667 | } |
b8413f98 | 1668 | |
af7cf057 TM |
1669 | static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) |
1670 | { | |
1671 | atomic_inc(&cinfo->rpcs_out); | |
71d0a611 TM |
1672 | } |
1673 | ||
af7cf057 | 1674 | static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) |
71d0a611 | 1675 | { |
af7cf057 | 1676 | if (atomic_dec_and_test(&cinfo->rpcs_out)) |
723c921e | 1677 | wake_up_var(&cinfo->rpcs_out); |
71d0a611 TM |
1678 | } |
1679 | ||
0b7c0153 | 1680 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1681 | { |
0b7c0153 FI |
1682 | put_nfs_open_context(data->context); |
1683 | nfs_commit_free(data); | |
1da177e4 | 1684 | } |
e0c2b380 | 1685 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1686 | |
0b7c0153 | 1687 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
c36aae9a | 1688 | const struct nfs_rpc_ops *nfs_ops, |
9ace33cd | 1689 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1690 | int how, int flags) |
1da177e4 | 1691 | { |
07737691 | 1692 | struct rpc_task *task; |
9ace33cd | 1693 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1694 | struct rpc_message msg = { |
1695 | .rpc_argp = &data->args, | |
1696 | .rpc_resp = &data->res, | |
9ace33cd | 1697 | .rpc_cred = data->cred, |
bdc7f021 | 1698 | }; |
84115e1c | 1699 | struct rpc_task_setup task_setup_data = { |
07737691 | 1700 | .task = &data->task, |
9ace33cd | 1701 | .rpc_client = clnt, |
bdc7f021 | 1702 | .rpc_message = &msg, |
9ace33cd | 1703 | .callback_ops = call_ops, |
84115e1c | 1704 | .callback_data = data, |
101070ca | 1705 | .workqueue = nfsiod_workqueue, |
4fa7ef69 | 1706 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1707 | .priority = priority, |
84115e1c | 1708 | }; |
9ace33cd | 1709 | /* Set up the initial task struct. */ |
e9ae1ee2 | 1710 | nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); |
8224b273 | 1711 | trace_nfs_initiate_commit(data); |
9ace33cd | 1712 | |
b4839ebe | 1713 | dprintk("NFS: initiated commit call\n"); |
9ace33cd FI |
1714 | |
1715 | task = rpc_run_task(&task_setup_data); | |
1716 | if (IS_ERR(task)) | |
1717 | return PTR_ERR(task); | |
1718 | if (how & FLUSH_SYNC) | |
1719 | rpc_wait_for_completion_task(task); | |
1720 | rpc_put_task(task); | |
1721 | return 0; | |
1722 | } | |
e0c2b380 | 1723 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1724 | |
378520b8 PT |
1725 | static loff_t nfs_get_lwb(struct list_head *head) |
1726 | { | |
1727 | loff_t lwb = 0; | |
1728 | struct nfs_page *req; | |
1729 | ||
1730 | list_for_each_entry(req, head, wb_list) | |
1731 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1732 | lwb = req_offset(req) + req->wb_bytes; | |
1733 | ||
1734 | return lwb; | |
1735 | } | |
1736 | ||
9ace33cd FI |
1737 | /* |
1738 | * Set up the argument/result storage required for the RPC call. | |
1739 | */ | |
0b7c0153 | 1740 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1741 | struct list_head *head, |
1742 | struct pnfs_layout_segment *lseg, | |
1743 | struct nfs_commit_info *cinfo) | |
9ace33cd | 1744 | { |
19573c93 TM |
1745 | struct nfs_page *first; |
1746 | struct nfs_open_context *ctx; | |
1747 | struct inode *inode; | |
1da177e4 LT |
1748 | |
1749 | /* Set up the RPC argument and reply structs | |
1750 | * NB: take care not to mess about with data->commit et al. */ | |
1751 | ||
19573c93 TM |
1752 | if (head) |
1753 | list_splice_init(head, &data->pages); | |
1754 | ||
1755 | first = nfs_list_entry(data->pages.next); | |
1756 | ctx = nfs_req_openctx(first); | |
1757 | inode = d_inode(ctx->dentry); | |
1da177e4 | 1758 | |
1da177e4 | 1759 | data->inode = inode; |
9fcd5960 | 1760 | data->cred = ctx->cred; |
988b6dce | 1761 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1762 | /* only set lwb for pnfs commit */ |
1763 | if (lseg) | |
1764 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1765 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1766 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1767 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1768 | |
1769 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1770 | /* Note: we always request a commit of the entire inode */ |
1771 | data->args.offset = 0; | |
1772 | data->args.count = 0; | |
9fcd5960 | 1773 | data->context = get_nfs_open_context(ctx); |
1da177e4 LT |
1774 | data->res.fattr = &data->fattr; |
1775 | data->res.verf = &data->verf; | |
0e574af1 | 1776 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1777 | } |
e0c2b380 | 1778 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1779 | |
e0c2b380 | 1780 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 | 1781 | struct pnfs_layout_segment *lseg, |
b57ff130 WAA |
1782 | struct nfs_commit_info *cinfo, |
1783 | u32 ds_commit_idx) | |
64bfeb49 FI |
1784 | { |
1785 | struct nfs_page *req; | |
1786 | ||
1787 | while (!list_empty(page_list)) { | |
1788 | req = nfs_list_entry(page_list->next); | |
1789 | nfs_list_remove_request(req); | |
b57ff130 | 1790 | nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); |
487b9b8a TH |
1791 | if (!cinfo->dreq) |
1792 | nfs_clear_page_commit(req->wb_page); | |
1d1afcbc | 1793 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1794 | } |
1795 | } | |
e0c2b380 | 1796 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1797 | |
b20135d0 TM |
1798 | static void |
1799 | nfs_commit_resched_write(struct nfs_commit_info *cinfo, | |
1800 | struct nfs_page *req) | |
1801 | { | |
1802 | __set_page_dirty_nobuffers(req->wb_page); | |
1803 | } | |
1804 | ||
1da177e4 LT |
1805 | /* |
1806 | * Commit dirty pages | |
1807 | */ | |
1808 | static int | |
ea2cf228 FI |
1809 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1810 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1811 | { |
0b7c0153 | 1812 | struct nfs_commit_data *data; |
1da177e4 | 1813 | |
ade8febd WAA |
1814 | /* another commit raced with us */ |
1815 | if (list_empty(head)) | |
1816 | return 0; | |
1817 | ||
518662e0 | 1818 | data = nfs_commitdata_alloc(true); |
1da177e4 LT |
1819 | |
1820 | /* Set up the argument struct */ | |
f453a54a FI |
1821 | nfs_init_commit(data, head, NULL, cinfo); |
1822 | atomic_inc(&cinfo->mds->rpcs_out); | |
c36aae9a | 1823 | return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), |
4fa7ef69 | 1824 | data->mds_ops, how, RPC_TASK_CRED_NOREF); |
67911c8f | 1825 | } |
67911c8f | 1826 | |
1da177e4 LT |
1827 | /* |
1828 | * COMMIT call returned | |
1829 | */ | |
788e7a89 | 1830 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1831 | { |
0b7c0153 | 1832 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1833 | |
a3f565b1 | 1834 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1835 | task->tk_pid, task->tk_status); |
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) { |
861e1671 | 1863 | trace_nfs_commit_error(req, status); |
6fbda89b | 1864 | nfs_mapping_set_error(req->wb_page, status); |
38a33101 | 1865 | nfs_inode_remove_request(req); |
6fbda89b | 1866 | } |
ddeaa637 | 1867 | dprintk_cont(", error = %d\n", status); |
1da177e4 LT |
1868 | goto next; |
1869 | } | |
1870 | ||
1871 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1872 | * returned by the server against all stored verfs. */ | |
1f28476d | 1873 | if (nfs_write_match_verf(verf, req)) { |
1da177e4 | 1874 | /* We have a match */ |
38a33101 KM |
1875 | if (req->wb_page) |
1876 | nfs_inode_remove_request(req); | |
ddeaa637 | 1877 | dprintk_cont(" OK\n"); |
1da177e4 LT |
1878 | goto next; |
1879 | } | |
1880 | /* We have a mismatch. Write the page again */ | |
ddeaa637 | 1881 | dprintk_cont(" mismatch\n"); |
6d884e8f | 1882 | nfs_mark_request_dirty(req); |
9fcd5960 | 1883 | set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); |
1da177e4 | 1884 | next: |
1d1afcbc | 1885 | nfs_unlock_and_release_request(req); |
7f1bda44 TM |
1886 | /* Latency breaker */ |
1887 | cond_resched(); | |
1da177e4 | 1888 | } |
353db796 N |
1889 | nfss = NFS_SERVER(data->inode); |
1890 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
0db10944 | 1891 | clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); |
353db796 | 1892 | |
f453a54a | 1893 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
af7cf057 | 1894 | nfs_commit_end(cinfo.mds); |
5917ce84 FI |
1895 | } |
1896 | ||
1897 | static void nfs_commit_release(void *calldata) | |
1898 | { | |
0b7c0153 | 1899 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1900 | |
f453a54a | 1901 | data->completion_ops->completion(data); |
c9d8f89d | 1902 | nfs_commitdata_release(calldata); |
1da177e4 | 1903 | } |
788e7a89 TM |
1904 | |
1905 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1906 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1907 | .rpc_call_done = nfs_commit_done, |
1908 | .rpc_release = nfs_commit_release, | |
1909 | }; | |
1da177e4 | 1910 | |
f453a54a FI |
1911 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1912 | .completion = nfs_commit_release_pages, | |
b20135d0 | 1913 | .resched_write = nfs_commit_resched_write, |
f453a54a FI |
1914 | }; |
1915 | ||
1763da12 FI |
1916 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1917 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1918 | { |
1919 | int status; | |
1920 | ||
ea2cf228 | 1921 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1922 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1923 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1924 | return status; |
1925 | } | |
1926 | ||
c4f24df9 TM |
1927 | static int __nfs_commit_inode(struct inode *inode, int how, |
1928 | struct writeback_control *wbc) | |
1da177e4 | 1929 | { |
1da177e4 | 1930 | LIST_HEAD(head); |
ea2cf228 | 1931 | struct nfs_commit_info cinfo; |
71d0a611 | 1932 | int may_wait = how & FLUSH_SYNC; |
c4f24df9 | 1933 | int ret, nscan; |
1da177e4 | 1934 | |
ea2cf228 | 1935 | nfs_init_cinfo_from_inode(&cinfo, inode); |
af7cf057 | 1936 | nfs_commit_begin(cinfo.mds); |
c4f24df9 TM |
1937 | for (;;) { |
1938 | ret = nscan = nfs_scan_commit(inode, &head, &cinfo); | |
1939 | if (ret <= 0) | |
1940 | break; | |
1941 | ret = nfs_generic_commit_list(inode, &head, how, &cinfo); | |
1942 | if (ret < 0) | |
1943 | break; | |
1944 | ret = 0; | |
1945 | if (wbc && wbc->sync_mode == WB_SYNC_NONE) { | |
1946 | if (nscan < wbc->nr_to_write) | |
1947 | wbc->nr_to_write -= nscan; | |
1948 | else | |
1949 | wbc->nr_to_write = 0; | |
1950 | } | |
1951 | if (nscan < INT_MAX) | |
1952 | break; | |
1953 | cond_resched(); | |
1954 | } | |
af7cf057 | 1955 | nfs_commit_end(cinfo.mds); |
c4f24df9 TM |
1956 | if (ret || !may_wait) |
1957 | return ret; | |
1958 | return wait_on_commit(cinfo.mds); | |
1959 | } | |
1960 | ||
1961 | int nfs_commit_inode(struct inode *inode, int how) | |
1962 | { | |
1963 | return __nfs_commit_inode(inode, how, NULL); | |
1da177e4 | 1964 | } |
b20135d0 | 1965 | EXPORT_SYMBOL_GPL(nfs_commit_inode); |
8fc795f7 | 1966 | |
ae09c31f | 1967 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
8fc795f7 | 1968 | { |
420e3646 TM |
1969 | struct nfs_inode *nfsi = NFS_I(inode); |
1970 | int flags = FLUSH_SYNC; | |
1971 | int ret = 0; | |
8fc795f7 | 1972 | |
a00dd6c0 | 1973 | if (wbc->sync_mode == WB_SYNC_NONE) { |
c4f24df9 TM |
1974 | /* no commits means nothing needs to be done */ |
1975 | if (!atomic_long_read(&nfsi->commit_info.ncommit)) | |
1976 | goto check_requests_outstanding; | |
1977 | ||
a00dd6c0 JL |
1978 | /* Don't commit yet if this is a non-blocking flush and there |
1979 | * are a lot of outstanding writes for this mapping. | |
1980 | */ | |
1a4edf0f | 1981 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) |
a00dd6c0 | 1982 | goto out_mark_dirty; |
420e3646 | 1983 | |
a00dd6c0 | 1984 | /* don't wait for the COMMIT response */ |
420e3646 | 1985 | flags = 0; |
a00dd6c0 JL |
1986 | } |
1987 | ||
c4f24df9 TM |
1988 | ret = __nfs_commit_inode(inode, flags, wbc); |
1989 | if (!ret) { | |
1990 | if (flags & FLUSH_SYNC) | |
1991 | return 0; | |
1992 | } else if (atomic_long_read(&nfsi->commit_info.ncommit)) | |
1993 | goto out_mark_dirty; | |
1994 | ||
1995 | check_requests_outstanding: | |
1996 | if (!atomic_read(&nfsi->commit_info.rpcs_out)) | |
1997 | return ret; | |
420e3646 | 1998 | out_mark_dirty: |
8fc795f7 TM |
1999 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
2000 | return ret; | |
2001 | } | |
89d77c8f | 2002 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 2003 | |
837bb1d7 TM |
2004 | /* |
2005 | * Wrapper for filemap_write_and_wait_range() | |
2006 | * | |
2007 | * Needed for pNFS in order to ensure data becomes visible to the | |
2008 | * client. | |
2009 | */ | |
2010 | int nfs_filemap_write_and_wait_range(struct address_space *mapping, | |
2011 | loff_t lstart, loff_t lend) | |
2012 | { | |
2013 | int ret; | |
2014 | ||
2015 | ret = filemap_write_and_wait_range(mapping, lstart, lend); | |
2016 | if (ret == 0) | |
2017 | ret = pnfs_sync_inode(mapping->host, true); | |
2018 | return ret; | |
2019 | } | |
2020 | EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); | |
2021 | ||
acdc53b2 TM |
2022 | /* |
2023 | * flush the inode to disk. | |
2024 | */ | |
2025 | int nfs_wb_all(struct inode *inode) | |
34901f70 | 2026 | { |
f4ce1299 TM |
2027 | int ret; |
2028 | ||
2029 | trace_nfs_writeback_inode_enter(inode); | |
2030 | ||
5bb89b47 | 2031 | ret = filemap_write_and_wait(inode->i_mapping); |
6b196875 CL |
2032 | if (ret) |
2033 | goto out; | |
2034 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
2035 | if (ret < 0) | |
2036 | goto out; | |
2037 | pnfs_sync_inode(inode, true); | |
2038 | ret = 0; | |
34901f70 | 2039 | |
6b196875 | 2040 | out: |
f4ce1299 TM |
2041 | trace_nfs_writeback_inode_exit(inode, ret); |
2042 | return ret; | |
1c75950b | 2043 | } |
ddda8e0a | 2044 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 2045 | |
1b3b4a1a TM |
2046 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
2047 | { | |
2048 | struct nfs_page *req; | |
1b3b4a1a TM |
2049 | int ret = 0; |
2050 | ||
3e217045 WAA |
2051 | wait_on_page_writeback(page); |
2052 | ||
2053 | /* blocking call to cancel all requests and join to a single (head) | |
2054 | * request */ | |
6d17d653 | 2055 | req = nfs_lock_and_join_requests(page); |
3e217045 WAA |
2056 | |
2057 | if (IS_ERR(req)) { | |
2058 | ret = PTR_ERR(req); | |
2059 | } else if (req) { | |
2060 | /* all requests from this page have been cancelled by | |
2061 | * nfs_lock_and_join_requests, so just remove the head | |
2062 | * request from the inode / page_private pointer and | |
2063 | * release it */ | |
2064 | nfs_inode_remove_request(req); | |
3e217045 | 2065 | nfs_unlock_and_release_request(req); |
1b3b4a1a | 2066 | } |
3e217045 | 2067 | |
1b3b4a1a TM |
2068 | return ret; |
2069 | } | |
2070 | ||
7f2f12d9 TM |
2071 | /* |
2072 | * Write back all requests on one page - we do this before reading it. | |
2073 | */ | |
c373fff7 | 2074 | int nfs_wb_page(struct inode *inode, struct page *page) |
1c75950b | 2075 | { |
29418aa4 | 2076 | loff_t range_start = page_file_offset(page); |
09cbfeaf | 2077 | loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); |
4d770ccf | 2078 | struct writeback_control wbc = { |
4d770ccf | 2079 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 2080 | .nr_to_write = 0, |
4d770ccf TM |
2081 | .range_start = range_start, |
2082 | .range_end = range_end, | |
2083 | }; | |
2084 | int ret; | |
1c75950b | 2085 | |
f4ce1299 TM |
2086 | trace_nfs_writeback_page_enter(inode); |
2087 | ||
0522f6ad | 2088 | for (;;) { |
ba8b06e6 | 2089 | wait_on_page_writeback(page); |
73e3302f | 2090 | if (clear_page_dirty_for_io(page)) { |
c373fff7 | 2091 | ret = nfs_writepage_locked(page, &wbc); |
73e3302f TM |
2092 | if (ret < 0) |
2093 | goto out_error; | |
0522f6ad | 2094 | continue; |
7f2f12d9 | 2095 | } |
f4ce1299 | 2096 | ret = 0; |
0522f6ad TM |
2097 | if (!PagePrivate(page)) |
2098 | break; | |
2099 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 2100 | if (ret < 0) |
73e3302f | 2101 | goto out_error; |
7f2f12d9 | 2102 | } |
73e3302f | 2103 | out_error: |
f4ce1299 | 2104 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 2105 | return ret; |
1c75950b TM |
2106 | } |
2107 | ||
074cc1de TM |
2108 | #ifdef CONFIG_MIGRATION |
2109 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 2110 | struct page *page, enum migrate_mode mode) |
074cc1de | 2111 | { |
2da95652 JL |
2112 | /* |
2113 | * If PagePrivate is set, then the page is currently associated with | |
2114 | * an in-progress read or write request. Don't try to migrate it. | |
2115 | * | |
2116 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
2117 | * that we can safely release the inode reference while holding | |
2118 | * the page lock. | |
2119 | */ | |
2120 | if (PagePrivate(page)) | |
2121 | return -EBUSY; | |
074cc1de | 2122 | |
8c209ce7 DH |
2123 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
2124 | return -EBUSY; | |
074cc1de | 2125 | |
a6bc32b8 | 2126 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
2127 | } |
2128 | #endif | |
2129 | ||
f7b422b1 | 2130 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
2131 | { |
2132 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 2133 | sizeof(struct nfs_pgio_header), |
1da177e4 | 2134 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 2135 | NULL); |
1da177e4 LT |
2136 | if (nfs_wdata_cachep == NULL) |
2137 | return -ENOMEM; | |
2138 | ||
93d2341c MD |
2139 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
2140 | nfs_wdata_cachep); | |
1da177e4 | 2141 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 2142 | goto out_destroy_write_cache; |
1da177e4 | 2143 | |
0b7c0153 FI |
2144 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
2145 | sizeof(struct nfs_commit_data), | |
2146 | 0, SLAB_HWCACHE_ALIGN, | |
2147 | NULL); | |
2148 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 2149 | goto out_destroy_write_mempool; |
0b7c0153 | 2150 | |
93d2341c | 2151 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 2152 | nfs_cdata_cachep); |
1da177e4 | 2153 | if (nfs_commit_mempool == NULL) |
3dd4765f | 2154 | goto out_destroy_commit_cache; |
1da177e4 | 2155 | |
89a09141 PZ |
2156 | /* |
2157 | * NFS congestion size, scale with available memory. | |
2158 | * | |
2159 | * 64MB: 8192k | |
2160 | * 128MB: 11585k | |
2161 | * 256MB: 16384k | |
2162 | * 512MB: 23170k | |
2163 | * 1GB: 32768k | |
2164 | * 2GB: 46340k | |
2165 | * 4GB: 65536k | |
2166 | * 8GB: 92681k | |
2167 | * 16GB: 131072k | |
2168 | * | |
2169 | * This allows larger machines to have larger/more transfers. | |
2170 | * Limit the default to 256M | |
2171 | */ | |
ca79b0c2 | 2172 | nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); |
89a09141 PZ |
2173 | if (nfs_congestion_kb > 256*1024) |
2174 | nfs_congestion_kb = 256*1024; | |
2175 | ||
1da177e4 | 2176 | return 0; |
3dd4765f JL |
2177 | |
2178 | out_destroy_commit_cache: | |
2179 | kmem_cache_destroy(nfs_cdata_cachep); | |
2180 | out_destroy_write_mempool: | |
2181 | mempool_destroy(nfs_wdata_mempool); | |
2182 | out_destroy_write_cache: | |
2183 | kmem_cache_destroy(nfs_wdata_cachep); | |
2184 | return -ENOMEM; | |
1da177e4 LT |
2185 | } |
2186 | ||
266bee88 | 2187 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
2188 | { |
2189 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 2190 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 2191 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 2192 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
2193 | } |
2194 | ||
4a0de55c AS |
2195 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
2196 | .rw_alloc_header = nfs_writehdr_alloc, | |
2197 | .rw_free_header = nfs_writehdr_free, | |
0eecb214 AS |
2198 | .rw_done = nfs_writeback_done, |
2199 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 2200 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 2201 | }; |