Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
7c85d900 | 4 | * Write file data over NFS. |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/mm.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/file.h> | |
1da177e4 | 14 | #include <linux/writeback.h> |
89a09141 | 15 | #include <linux/swap.h> |
074cc1de | 16 | #include <linux/migrate.h> |
1da177e4 LT |
17 | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | #include <linux/nfs_page.h> | |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
afeacc8c | 23 | #include <linux/export.h> |
3fcfab16 | 24 | |
1da177e4 | 25 | #include <asm/uaccess.h> |
1da177e4 LT |
26 | |
27 | #include "delegation.h" | |
49a70f27 | 28 | #include "internal.h" |
91d5b470 | 29 | #include "iostat.h" |
def6ed7e | 30 | #include "nfs4_fs.h" |
074cc1de | 31 | #include "fscache.h" |
94ad1c80 | 32 | #include "pnfs.h" |
1da177e4 | 33 | |
f4ce1299 TM |
34 | #include "nfstrace.h" |
35 | ||
1da177e4 LT |
36 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
37 | ||
38 | #define MIN_POOL_WRITE (32) | |
39 | #define MIN_POOL_COMMIT (4) | |
40 | ||
41 | /* | |
42 | * Local function declarations | |
43 | */ | |
f8512ad0 | 44 | static void nfs_redirty_request(struct nfs_page *req); |
788e7a89 | 45 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 46 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 47 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
4a0de55c | 48 | static const struct nfs_rw_ops nfs_rw_write_ops; |
d4581383 | 49 | static void nfs_clear_request_commit(struct nfs_page *req); |
02d1426c WAA |
50 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
51 | struct inode *inode); | |
3a3908c8 TM |
52 | static struct nfs_page * |
53 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
54 | struct page *page); | |
1da177e4 | 55 | |
e18b890b | 56 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 57 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 58 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
59 | static mempool_t *nfs_commit_mempool; |
60 | ||
0b7c0153 | 61 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 62 | { |
192e501b | 63 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); |
40859d7e | 64 | |
1da177e4 LT |
65 | if (p) { |
66 | memset(p, 0, sizeof(*p)); | |
67 | INIT_LIST_HEAD(&p->pages); | |
68 | } | |
69 | return p; | |
70 | } | |
e0c2b380 | 71 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 72 | |
0b7c0153 | 73 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
74 | { |
75 | mempool_free(p, nfs_commit_mempool); | |
76 | } | |
e0c2b380 | 77 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 78 | |
1e7f3a48 | 79 | static struct nfs_pgio_header *nfs_writehdr_alloc(void) |
3feb2d49 | 80 | { |
1e7f3a48 | 81 | struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
cd841605 | 82 | |
4a0de55c | 83 | if (p) |
3feb2d49 | 84 | memset(p, 0, sizeof(*p)); |
3feb2d49 TM |
85 | return p; |
86 | } | |
6c75dc0d | 87 | |
1e7f3a48 | 88 | static void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 89 | { |
1e7f3a48 | 90 | mempool_free(hdr, nfs_wdata_mempool); |
3feb2d49 | 91 | } |
1da177e4 | 92 | |
7b159fc1 TM |
93 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
94 | { | |
95 | ctx->error = error; | |
96 | smp_wmb(); | |
97 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
98 | } | |
99 | ||
84d3a9a9 WAA |
100 | /* |
101 | * nfs_page_find_head_request_locked - find head request associated with @page | |
102 | * | |
103 | * must be called while holding the inode lock. | |
104 | * | |
105 | * returns matching head request with reference held, or NULL if not found. | |
106 | */ | |
29418aa4 | 107 | static struct nfs_page * |
84d3a9a9 | 108 | nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) |
277459d2 TM |
109 | { |
110 | struct nfs_page *req = NULL; | |
111 | ||
29418aa4 | 112 | if (PagePrivate(page)) |
277459d2 | 113 | req = (struct nfs_page *)page_private(page); |
02d1426c WAA |
114 | else if (unlikely(PageSwapCache(page))) |
115 | req = nfs_page_search_commits_for_head_request_locked(nfsi, | |
116 | page); | |
29418aa4 | 117 | |
84d3a9a9 WAA |
118 | if (req) { |
119 | WARN_ON_ONCE(req->wb_head != req); | |
29418aa4 | 120 | kref_get(&req->wb_kref); |
84d3a9a9 | 121 | } |
29418aa4 | 122 | |
277459d2 TM |
123 | return req; |
124 | } | |
125 | ||
84d3a9a9 WAA |
126 | /* |
127 | * nfs_page_find_head_request - find head request associated with @page | |
128 | * | |
129 | * returns matching head request with reference held, or NULL if not found. | |
130 | */ | |
131 | static struct nfs_page *nfs_page_find_head_request(struct page *page) | |
277459d2 | 132 | { |
d56b4ddf | 133 | struct inode *inode = page_file_mapping(page)->host; |
277459d2 | 134 | struct nfs_page *req = NULL; |
277459d2 | 135 | |
587142f8 | 136 | spin_lock(&inode->i_lock); |
84d3a9a9 | 137 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
587142f8 | 138 | spin_unlock(&inode->i_lock); |
277459d2 TM |
139 | return req; |
140 | } | |
141 | ||
1da177e4 LT |
142 | /* Adjust the file length if we're writing beyond the end */ |
143 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
144 | { | |
d56b4ddf | 145 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
146 | loff_t end, i_size; |
147 | pgoff_t end_index; | |
1da177e4 | 148 | |
a3d01454 TM |
149 | spin_lock(&inode->i_lock); |
150 | i_size = i_size_read(inode); | |
151 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
d56b4ddf | 152 | if (i_size > 0 && page_file_index(page) < end_index) |
a3d01454 | 153 | goto out; |
d56b4ddf | 154 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 155 | if (i_size >= end) |
a3d01454 | 156 | goto out; |
1da177e4 | 157 | i_size_write(inode, end); |
a3d01454 TM |
158 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
159 | out: | |
160 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
161 | } |
162 | ||
a301b777 TM |
163 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
164 | static void nfs_set_pageerror(struct page *page) | |
165 | { | |
d56b4ddf | 166 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
167 | } |
168 | ||
d72ddcba WAA |
169 | /* |
170 | * nfs_page_group_search_locked | |
171 | * @head - head request of page group | |
172 | * @page_offset - offset into page | |
173 | * | |
174 | * Search page group with head @head to find a request that contains the | |
175 | * page offset @page_offset. | |
176 | * | |
177 | * Returns a pointer to the first matching nfs request, or NULL if no | |
178 | * match is found. | |
179 | * | |
180 | * Must be called with the page group lock held | |
181 | */ | |
182 | static struct nfs_page * | |
183 | nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) | |
184 | { | |
185 | struct nfs_page *req; | |
186 | ||
187 | WARN_ON_ONCE(head != head->wb_head); | |
188 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); | |
189 | ||
190 | req = head; | |
191 | do { | |
192 | if (page_offset >= req->wb_pgbase && | |
193 | page_offset < (req->wb_pgbase + req->wb_bytes)) | |
194 | return req; | |
195 | ||
196 | req = req->wb_this_page; | |
197 | } while (req != head); | |
198 | ||
199 | return NULL; | |
200 | } | |
201 | ||
202 | /* | |
203 | * nfs_page_group_covers_page | |
204 | * @head - head request of page group | |
205 | * | |
206 | * Return true if the page group with head @head covers the whole page, | |
207 | * returns false otherwise | |
208 | */ | |
209 | static bool nfs_page_group_covers_page(struct nfs_page *req) | |
210 | { | |
211 | struct nfs_page *tmp; | |
212 | unsigned int pos = 0; | |
213 | unsigned int len = nfs_page_length(req->wb_page); | |
214 | ||
fd2f3a06 | 215 | nfs_page_group_lock(req, false); |
d72ddcba WAA |
216 | |
217 | do { | |
218 | tmp = nfs_page_group_search_locked(req->wb_head, pos); | |
219 | if (tmp) { | |
220 | /* no way this should happen */ | |
221 | WARN_ON_ONCE(tmp->wb_pgbase != pos); | |
222 | pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); | |
223 | } | |
224 | } while (tmp && pos < len); | |
225 | ||
226 | nfs_page_group_unlock(req); | |
227 | WARN_ON_ONCE(pos > len); | |
228 | return pos == len; | |
229 | } | |
230 | ||
1da177e4 LT |
231 | /* We can set the PG_uptodate flag if we see that a write request |
232 | * covers the full page. | |
233 | */ | |
d72ddcba | 234 | static void nfs_mark_uptodate(struct nfs_page *req) |
1da177e4 | 235 | { |
d72ddcba | 236 | if (PageUptodate(req->wb_page)) |
1da177e4 | 237 | return; |
d72ddcba | 238 | if (!nfs_page_group_covers_page(req)) |
1da177e4 | 239 | return; |
d72ddcba | 240 | SetPageUptodate(req->wb_page); |
1da177e4 LT |
241 | } |
242 | ||
1da177e4 LT |
243 | static int wb_priority(struct writeback_control *wbc) |
244 | { | |
e87b4c7a | 245 | int ret = 0; |
1da177e4 | 246 | if (wbc->for_reclaim) |
c63c7b05 | 247 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
e87b4c7a N |
248 | if (wbc->sync_mode == WB_SYNC_ALL) |
249 | ret = FLUSH_COND_STABLE; | |
b17621fe | 250 | if (wbc->for_kupdate || wbc->for_background) |
e87b4c7a N |
251 | ret |= FLUSH_LOWPRI; |
252 | return ret; | |
1da177e4 LT |
253 | } |
254 | ||
89a09141 PZ |
255 | /* |
256 | * NFS congestion control | |
257 | */ | |
258 | ||
259 | int nfs_congestion_kb; | |
260 | ||
261 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
262 | #define NFS_CONGESTION_OFF_THRESH \ | |
263 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
264 | ||
deed85e7 | 265 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 266 | { |
deed85e7 | 267 | struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); |
5a6d41b3 TM |
268 | int ret = test_set_page_writeback(page); |
269 | ||
deed85e7 | 270 | WARN_ON_ONCE(ret != 0); |
89a09141 | 271 | |
deed85e7 TM |
272 | if (atomic_long_inc_return(&nfss->writeback) > |
273 | NFS_CONGESTION_ON_THRESH) { | |
274 | set_bdi_congested(&nfss->backing_dev_info, | |
275 | BLK_RW_ASYNC); | |
89a09141 PZ |
276 | } |
277 | } | |
278 | ||
20633f04 | 279 | static void nfs_end_page_writeback(struct nfs_page *req) |
89a09141 | 280 | { |
20633f04 | 281 | struct inode *inode = page_file_mapping(req->wb_page)->host; |
89a09141 PZ |
282 | struct nfs_server *nfss = NFS_SERVER(inode); |
283 | ||
20633f04 WAA |
284 | if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) |
285 | return; | |
286 | ||
287 | end_page_writeback(req->wb_page); | |
c4dc4bee | 288 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 289 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
290 | } |
291 | ||
d4581383 WAA |
292 | |
293 | /* nfs_page_group_clear_bits | |
294 | * @req - an nfs request | |
295 | * clears all page group related bits from @req | |
296 | */ | |
297 | static void | |
298 | nfs_page_group_clear_bits(struct nfs_page *req) | |
299 | { | |
300 | clear_bit(PG_TEARDOWN, &req->wb_flags); | |
301 | clear_bit(PG_UNLOCKPAGE, &req->wb_flags); | |
302 | clear_bit(PG_UPTODATE, &req->wb_flags); | |
303 | clear_bit(PG_WB_END, &req->wb_flags); | |
304 | clear_bit(PG_REMOVE, &req->wb_flags); | |
305 | } | |
306 | ||
307 | ||
308 | /* | |
309 | * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req | |
310 | * | |
311 | * this is a helper function for nfs_lock_and_join_requests | |
312 | * | |
313 | * @inode - inode associated with request page group, must be holding inode lock | |
314 | * @head - head request of page group, must be holding head lock | |
315 | * @req - request that couldn't lock and needs to wait on the req bit lock | |
316 | * @nonblock - if true, don't actually wait | |
317 | * | |
318 | * NOTE: this must be called holding page_group bit lock and inode spin lock | |
319 | * and BOTH will be released before returning. | |
320 | * | |
321 | * returns 0 on success, < 0 on error. | |
322 | */ | |
323 | static int | |
324 | nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, | |
325 | struct nfs_page *req, bool nonblock) | |
326 | __releases(&inode->i_lock) | |
327 | { | |
328 | struct nfs_page *tmp; | |
329 | int ret; | |
330 | ||
331 | /* relinquish all the locks successfully grabbed this run */ | |
332 | for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) | |
333 | nfs_unlock_request(tmp); | |
334 | ||
335 | WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); | |
336 | ||
337 | /* grab a ref on the request that will be waited on */ | |
338 | kref_get(&req->wb_kref); | |
339 | ||
340 | nfs_page_group_unlock(head); | |
341 | spin_unlock(&inode->i_lock); | |
342 | ||
343 | /* release ref from nfs_page_find_head_request_locked */ | |
344 | nfs_release_request(head); | |
345 | ||
346 | if (!nonblock) | |
347 | ret = nfs_wait_on_request(req); | |
348 | else | |
349 | ret = -EAGAIN; | |
350 | nfs_release_request(req); | |
351 | ||
352 | return ret; | |
353 | } | |
354 | ||
355 | /* | |
356 | * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests | |
357 | * | |
358 | * @destroy_list - request list (using wb_this_page) terminated by @old_head | |
359 | * @old_head - the old head of the list | |
360 | * | |
361 | * All subrequests must be locked and removed from all lists, so at this point | |
362 | * they are only "active" in this function, and possibly in nfs_wait_on_request | |
363 | * with a reference held by some other context. | |
364 | */ | |
365 | static void | |
366 | nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, | |
367 | struct nfs_page *old_head) | |
368 | { | |
369 | while (destroy_list) { | |
370 | struct nfs_page *subreq = destroy_list; | |
371 | ||
372 | destroy_list = (subreq->wb_this_page == old_head) ? | |
373 | NULL : subreq->wb_this_page; | |
374 | ||
375 | WARN_ON_ONCE(old_head != subreq->wb_head); | |
376 | ||
377 | /* make sure old group is not used */ | |
378 | subreq->wb_head = subreq; | |
379 | subreq->wb_this_page = subreq; | |
380 | ||
d4581383 WAA |
381 | /* subreq is now totally disconnected from page group or any |
382 | * write / commit lists. last chance to wake any waiters */ | |
383 | nfs_unlock_request(subreq); | |
384 | ||
385 | if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { | |
386 | /* release ref on old head request */ | |
387 | nfs_release_request(old_head); | |
388 | ||
389 | nfs_page_group_clear_bits(subreq); | |
390 | ||
391 | /* release the PG_INODE_REF reference */ | |
392 | if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) | |
393 | nfs_release_request(subreq); | |
394 | else | |
395 | WARN_ON_ONCE(1); | |
396 | } else { | |
397 | WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); | |
398 | /* zombie requests have already released the last | |
399 | * reference and were waiting on the rest of the | |
400 | * group to complete. Since it's no longer part of a | |
401 | * group, simply free the request */ | |
402 | nfs_page_group_clear_bits(subreq); | |
403 | nfs_free_request(subreq); | |
404 | } | |
405 | } | |
406 | } | |
407 | ||
408 | /* | |
409 | * nfs_lock_and_join_requests - join all subreqs to the head req and return | |
410 | * a locked reference, cancelling any pending | |
411 | * operations for this page. | |
412 | * | |
413 | * @page - the page used to lookup the "page group" of nfs_page structures | |
414 | * @nonblock - if true, don't block waiting for request locks | |
415 | * | |
416 | * This function joins all sub requests to the head request by first | |
417 | * locking all requests in the group, cancelling any pending operations | |
418 | * and finally updating the head request to cover the whole range covered by | |
419 | * the (former) group. All subrequests are removed from any write or commit | |
420 | * lists, unlinked from the group and destroyed. | |
421 | * | |
422 | * Returns a locked, referenced pointer to the head request - which after | |
423 | * this call is guaranteed to be the only request associated with the page. | |
424 | * Returns NULL if no requests are found for @page, or a ERR_PTR if an | |
425 | * error was encountered. | |
426 | */ | |
427 | static struct nfs_page * | |
428 | nfs_lock_and_join_requests(struct page *page, bool nonblock) | |
e261f51f | 429 | { |
d56b4ddf | 430 | struct inode *inode = page_file_mapping(page)->host; |
d4581383 WAA |
431 | struct nfs_page *head, *subreq; |
432 | struct nfs_page *destroy_list = NULL; | |
433 | unsigned int total_bytes; | |
e261f51f TM |
434 | int ret; |
435 | ||
d4581383 WAA |
436 | try_again: |
437 | total_bytes = 0; | |
438 | ||
439 | WARN_ON_ONCE(destroy_list); | |
440 | ||
587142f8 | 441 | spin_lock(&inode->i_lock); |
d4581383 WAA |
442 | |
443 | /* | |
444 | * A reference is taken only on the head request which acts as a | |
445 | * reference to the whole page group - the group will not be destroyed | |
446 | * until the head reference is released. | |
447 | */ | |
448 | head = nfs_page_find_head_request_locked(NFS_I(inode), page); | |
449 | ||
450 | if (!head) { | |
587142f8 | 451 | spin_unlock(&inode->i_lock); |
d4581383 WAA |
452 | return NULL; |
453 | } | |
454 | ||
7c3af975 WAA |
455 | /* holding inode lock, so always make a non-blocking call to try the |
456 | * page group lock */ | |
fd2f3a06 | 457 | ret = nfs_page_group_lock(head, true); |
94970014 WAA |
458 | if (ret < 0) { |
459 | spin_unlock(&inode->i_lock); | |
7c3af975 WAA |
460 | |
461 | if (!nonblock && ret == -EAGAIN) { | |
462 | nfs_page_group_lock_wait(head); | |
463 | nfs_release_request(head); | |
464 | goto try_again; | |
465 | } | |
466 | ||
94970014 | 467 | nfs_release_request(head); |
e7029206 | 468 | return ERR_PTR(ret); |
94970014 | 469 | } |
7c3af975 WAA |
470 | |
471 | /* lock each request in the page group */ | |
d4581383 WAA |
472 | subreq = head; |
473 | do { | |
474 | /* | |
475 | * Subrequests are always contiguous, non overlapping | |
476 | * and in order. If not, it's a programming error. | |
477 | */ | |
478 | WARN_ON_ONCE(subreq->wb_offset != | |
479 | (head->wb_offset + total_bytes)); | |
480 | ||
481 | /* keep track of how many bytes this group covers */ | |
482 | total_bytes += subreq->wb_bytes; | |
483 | ||
484 | if (!nfs_lock_request(subreq)) { | |
485 | /* releases page group bit lock and | |
486 | * inode spin lock and all references */ | |
487 | ret = nfs_unroll_locks_and_wait(inode, head, | |
488 | subreq, nonblock); | |
489 | ||
490 | if (ret == 0) | |
491 | goto try_again; | |
492 | ||
074cc1de | 493 | return ERR_PTR(ret); |
d4581383 WAA |
494 | } |
495 | ||
496 | subreq = subreq->wb_this_page; | |
497 | } while (subreq != head); | |
498 | ||
499 | /* Now that all requests are locked, make sure they aren't on any list. | |
500 | * Commit list removal accounting is done after locks are dropped */ | |
501 | subreq = head; | |
502 | do { | |
411a99ad | 503 | nfs_clear_request_commit(subreq); |
d4581383 WAA |
504 | subreq = subreq->wb_this_page; |
505 | } while (subreq != head); | |
506 | ||
507 | /* unlink subrequests from head, destroy them later */ | |
508 | if (head->wb_this_page != head) { | |
509 | /* destroy list will be terminated by head */ | |
510 | destroy_list = head->wb_this_page; | |
511 | head->wb_this_page = head; | |
512 | ||
513 | /* change head request to cover whole range that | |
514 | * the former page group covered */ | |
515 | head->wb_bytes = total_bytes; | |
e261f51f | 516 | } |
d4581383 WAA |
517 | |
518 | /* | |
519 | * prepare head request to be added to new pgio descriptor | |
520 | */ | |
521 | nfs_page_group_clear_bits(head); | |
522 | ||
523 | /* | |
524 | * some part of the group was still on the inode list - otherwise | |
525 | * the group wouldn't be involved in async write. | |
526 | * grab a reference for the head request, iff it needs one. | |
527 | */ | |
528 | if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) | |
529 | kref_get(&head->wb_kref); | |
530 | ||
531 | nfs_page_group_unlock(head); | |
532 | ||
411a99ad | 533 | /* drop lock to clean uprequests on destroy list */ |
587142f8 | 534 | spin_unlock(&inode->i_lock); |
d4581383 WAA |
535 | |
536 | nfs_destroy_unlinked_subrequests(destroy_list, head); | |
537 | ||
d4581383 WAA |
538 | /* still holds ref on head from nfs_page_find_head_request_locked |
539 | * and still has lock on head from lock loop */ | |
540 | return head; | |
074cc1de TM |
541 | } |
542 | ||
543 | /* | |
544 | * Find an associated nfs write request, and prepare to flush it out | |
545 | * May return an error if the user signalled nfs_wait_on_request(). | |
546 | */ | |
547 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 548 | struct page *page, bool nonblock) |
074cc1de TM |
549 | { |
550 | struct nfs_page *req; | |
551 | int ret = 0; | |
552 | ||
d4581383 | 553 | req = nfs_lock_and_join_requests(page, nonblock); |
074cc1de TM |
554 | if (!req) |
555 | goto out; | |
556 | ret = PTR_ERR(req); | |
557 | if (IS_ERR(req)) | |
558 | goto out; | |
559 | ||
deed85e7 TM |
560 | nfs_set_page_writeback(page); |
561 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 562 | |
deed85e7 | 563 | ret = 0; |
f8512ad0 FI |
564 | if (!nfs_pageio_add_request(pgio, req)) { |
565 | nfs_redirty_request(req); | |
074cc1de | 566 | ret = pgio->pg_error; |
f8512ad0 | 567 | } |
074cc1de TM |
568 | out: |
569 | return ret; | |
e261f51f TM |
570 | } |
571 | ||
f758c885 | 572 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 573 | { |
d56b4ddf | 574 | struct inode *inode = page_file_mapping(page)->host; |
cfb506e1 | 575 | int ret; |
1da177e4 | 576 | |
91d5b470 | 577 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
5a254d08 | 578 | nfs_inc_stats(inode, NFSIOS_WRITEPAGES); |
91d5b470 | 579 | |
d56b4ddf | 580 | nfs_pageio_cond_complete(pgio, page_file_index(page)); |
1b430bee | 581 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
582 | if (ret == -EAGAIN) { |
583 | redirty_page_for_writepage(wbc, page); | |
584 | ret = 0; | |
585 | } | |
586 | return ret; | |
f758c885 | 587 | } |
7fe7f848 | 588 | |
f758c885 TM |
589 | /* |
590 | * Write an mmapped page to the server. | |
591 | */ | |
592 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
593 | { | |
594 | struct nfs_pageio_descriptor pgio; | |
595 | int err; | |
49a70f27 | 596 | |
a20c93e3 CH |
597 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), |
598 | false, &nfs_async_write_completion_ops); | |
f758c885 TM |
599 | err = nfs_do_writepage(page, wbc, &pgio); |
600 | nfs_pageio_complete(&pgio); | |
601 | if (err < 0) | |
602 | return err; | |
603 | if (pgio.pg_error < 0) | |
604 | return pgio.pg_error; | |
605 | return 0; | |
4d770ccf TM |
606 | } |
607 | ||
608 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
609 | { | |
f758c885 | 610 | int ret; |
4d770ccf | 611 | |
f758c885 | 612 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 613 | unlock_page(page); |
f758c885 TM |
614 | return ret; |
615 | } | |
616 | ||
617 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
618 | { | |
619 | int ret; | |
620 | ||
621 | ret = nfs_do_writepage(page, wbc, data); | |
622 | unlock_page(page); | |
623 | return ret; | |
1da177e4 LT |
624 | } |
625 | ||
1da177e4 LT |
626 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
627 | { | |
1da177e4 | 628 | struct inode *inode = mapping->host; |
72cb77f4 | 629 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 630 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
631 | int err; |
632 | ||
72cb77f4 | 633 | /* Stop dirtying of new pages while we sync */ |
74316201 | 634 | err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING, |
72cb77f4 TM |
635 | nfs_wait_bit_killable, TASK_KILLABLE); |
636 | if (err) | |
637 | goto out_err; | |
638 | ||
91d5b470 CL |
639 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
640 | ||
a20c93e3 CH |
641 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
642 | &nfs_async_write_completion_ops); | |
f758c885 | 643 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 644 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
645 | |
646 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
4e857c58 | 647 | smp_mb__after_atomic(); |
72cb77f4 TM |
648 | wake_up_bit(bitlock, NFS_INO_FLUSHING); |
649 | ||
f758c885 | 650 | if (err < 0) |
72cb77f4 TM |
651 | goto out_err; |
652 | err = pgio.pg_error; | |
653 | if (err < 0) | |
654 | goto out_err; | |
c63c7b05 | 655 | return 0; |
72cb77f4 TM |
656 | out_err: |
657 | return err; | |
1da177e4 LT |
658 | } |
659 | ||
660 | /* | |
661 | * Insert a write request into an inode | |
662 | */ | |
d6d6dc7c | 663 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
664 | { |
665 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 | 666 | |
2bfc6e56 WAA |
667 | WARN_ON_ONCE(req->wb_this_page != req); |
668 | ||
e7d39069 | 669 | /* Lock the request! */ |
7ad84aa9 | 670 | nfs_lock_request(req); |
e7d39069 TM |
671 | |
672 | spin_lock(&inode->i_lock); | |
cb1410c7 WAA |
673 | if (!nfsi->nrequests && |
674 | NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) | |
a9a4a87a | 675 | inode->i_version++; |
29418aa4 MG |
676 | /* |
677 | * Swap-space should not get truncated. Hence no need to plug the race | |
678 | * with invalidate/truncate. | |
679 | */ | |
680 | if (likely(!PageSwapCache(req->wb_page))) { | |
681 | set_bit(PG_MAPPED, &req->wb_flags); | |
682 | SetPagePrivate(req->wb_page); | |
683 | set_page_private(req->wb_page, (unsigned long)req); | |
684 | } | |
cb1410c7 | 685 | nfsi->nrequests++; |
17089a29 | 686 | /* this a head request for a page group - mark it as having an |
cb1410c7 WAA |
687 | * extra reference so sub groups can follow suit. |
688 | * This flag also informs pgio layer when to bump nrequests when | |
689 | * adding subrequests. */ | |
17089a29 | 690 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); |
c03b4024 | 691 | kref_get(&req->wb_kref); |
e7d39069 | 692 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
693 | } |
694 | ||
695 | /* | |
89a09141 | 696 | * Remove a write request from an inode |
1da177e4 LT |
697 | */ |
698 | static void nfs_inode_remove_request(struct nfs_page *req) | |
699 | { | |
3d4ff43d | 700 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 | 701 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 702 | struct nfs_page *head; |
1da177e4 | 703 | |
20633f04 WAA |
704 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
705 | head = req->wb_head; | |
706 | ||
707 | spin_lock(&inode->i_lock); | |
708 | if (likely(!PageSwapCache(head->wb_page))) { | |
709 | set_page_private(head->wb_page, 0); | |
710 | ClearPagePrivate(head->wb_page); | |
95905446 N |
711 | smp_mb__after_atomic(); |
712 | wake_up_page(head->wb_page, PG_private); | |
20633f04 WAA |
713 | clear_bit(PG_MAPPED, &head->wb_flags); |
714 | } | |
cb1410c7 WAA |
715 | nfsi->nrequests--; |
716 | spin_unlock(&inode->i_lock); | |
717 | } else { | |
718 | spin_lock(&inode->i_lock); | |
719 | nfsi->nrequests--; | |
20633f04 | 720 | spin_unlock(&inode->i_lock); |
29418aa4 | 721 | } |
17089a29 WAA |
722 | |
723 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) | |
724 | nfs_release_request(req); | |
1da177e4 LT |
725 | } |
726 | ||
61822ab5 | 727 | static void |
6d884e8f | 728 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 729 | { |
61822ab5 TM |
730 | __set_page_dirty_nobuffers(req->wb_page); |
731 | } | |
732 | ||
3a3908c8 TM |
733 | /* |
734 | * nfs_page_search_commits_for_head_request_locked | |
735 | * | |
736 | * Search through commit lists on @inode for the head request for @page. | |
737 | * Must be called while holding the inode (which is cinfo) lock. | |
738 | * | |
739 | * Returns the head request if found, or NULL if not found. | |
740 | */ | |
741 | static struct nfs_page * | |
742 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
743 | struct page *page) | |
744 | { | |
745 | struct nfs_page *freq, *t; | |
746 | struct nfs_commit_info cinfo; | |
747 | struct inode *inode = &nfsi->vfs_inode; | |
748 | ||
749 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
750 | ||
751 | /* search through pnfs commit lists */ | |
752 | freq = pnfs_search_commit_reqs(inode, &cinfo, page); | |
753 | if (freq) | |
754 | return freq->wb_head; | |
755 | ||
756 | /* Linearly search the commit list for the correct request */ | |
757 | list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { | |
758 | if (freq->wb_page == page) | |
759 | return freq->wb_head; | |
760 | } | |
761 | ||
762 | return NULL; | |
763 | } | |
764 | ||
8dd37758 TM |
765 | /** |
766 | * nfs_request_add_commit_list - add request to a commit list | |
767 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
768 | * @dst: commit list head |
769 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 770 | * |
ea2cf228 | 771 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
772 | * number of outstanding requests requiring a commit as well as |
773 | * the MM page stats. | |
774 | * | |
ea2cf228 | 775 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 776 | * holding the nfs_page lock. |
1da177e4 | 777 | */ |
8dd37758 | 778 | void |
ea2cf228 FI |
779 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
780 | struct nfs_commit_info *cinfo) | |
1da177e4 | 781 | { |
e468bae9 | 782 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
783 | spin_lock(cinfo->lock); |
784 | nfs_list_add_request(req, dst); | |
785 | cinfo->mds->ncommit++; | |
786 | spin_unlock(cinfo->lock); | |
56f9cd68 FI |
787 | if (!cinfo->dreq) { |
788 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 789 | inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
790 | BDI_RECLAIMABLE); |
791 | __mark_inode_dirty(req->wb_context->dentry->d_inode, | |
792 | I_DIRTY_DATASYNC); | |
793 | } | |
1da177e4 | 794 | } |
8dd37758 TM |
795 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
796 | ||
797 | /** | |
798 | * nfs_request_remove_commit_list - Remove request from a commit list | |
799 | * @req: pointer to a nfs_page | |
ea2cf228 | 800 | * @cinfo: holds list lock and accounting info |
8dd37758 | 801 | * |
ea2cf228 | 802 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
803 | * number of outstanding requests requiring a commit |
804 | * It does not update the MM page stats. | |
805 | * | |
ea2cf228 | 806 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
807 | */ |
808 | void | |
ea2cf228 FI |
809 | nfs_request_remove_commit_list(struct nfs_page *req, |
810 | struct nfs_commit_info *cinfo) | |
8dd37758 | 811 | { |
8dd37758 TM |
812 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
813 | return; | |
814 | nfs_list_remove_request(req); | |
ea2cf228 | 815 | cinfo->mds->ncommit--; |
8dd37758 TM |
816 | } |
817 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
818 | ||
ea2cf228 FI |
819 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
820 | struct inode *inode) | |
821 | { | |
822 | cinfo->lock = &inode->i_lock; | |
823 | cinfo->mds = &NFS_I(inode)->commit_info; | |
824 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 825 | cinfo->dreq = NULL; |
f453a54a | 826 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
827 | } |
828 | ||
829 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
830 | struct inode *inode, | |
831 | struct nfs_direct_req *dreq) | |
832 | { | |
1763da12 FI |
833 | if (dreq) |
834 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
835 | else | |
836 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
837 | } |
838 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
839 | |
840 | /* | |
841 | * Add a request to the inode's commit list. | |
842 | */ | |
1763da12 | 843 | void |
ea2cf228 FI |
844 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
845 | struct nfs_commit_info *cinfo) | |
8dd37758 | 846 | { |
ea2cf228 | 847 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 848 | return; |
ea2cf228 | 849 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 850 | } |
8e821cad | 851 | |
d6d6dc7c FI |
852 | static void |
853 | nfs_clear_page_commit(struct page *page) | |
854 | { | |
855 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
d56b4ddf | 856 | dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); |
d6d6dc7c FI |
857 | } |
858 | ||
411a99ad | 859 | /* Called holding inode (/cinfo) lock */ |
8dd37758 | 860 | static void |
e468bae9 TM |
861 | nfs_clear_request_commit(struct nfs_page *req) |
862 | { | |
8dd37758 TM |
863 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
864 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 865 | struct nfs_commit_info cinfo; |
e468bae9 | 866 | |
ea2cf228 FI |
867 | nfs_init_cinfo_from_inode(&cinfo, inode); |
868 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
ea2cf228 | 869 | nfs_request_remove_commit_list(req, &cinfo); |
8dd37758 | 870 | } |
d6d6dc7c | 871 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 872 | } |
e468bae9 TM |
873 | } |
874 | ||
d45f60c6 | 875 | int nfs_write_need_commit(struct nfs_pgio_header *hdr) |
8e821cad | 876 | { |
c65e6254 | 877 | if (hdr->verf.committed == NFS_DATA_SYNC) |
d45f60c6 | 878 | return hdr->lseg == NULL; |
c65e6254 | 879 | return hdr->verf.committed != NFS_FILE_SYNC; |
8e821cad TM |
880 | } |
881 | ||
061ae2ed | 882 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 883 | { |
ea2cf228 | 884 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
885 | unsigned long bytes = 0; |
886 | ||
887 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
888 | goto out; | |
ea2cf228 | 889 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
890 | while (!list_empty(&hdr->pages)) { |
891 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
892 | |
893 | bytes += req->wb_bytes; | |
894 | nfs_list_remove_request(req); | |
895 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
896 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 897 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
898 | nfs_context_set_write_error(req->wb_context, hdr->error); |
899 | goto remove_req; | |
900 | } | |
c65e6254 | 901 | if (nfs_write_need_commit(hdr)) { |
f79d06f5 | 902 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
ea2cf228 | 903 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
904 | goto next; |
905 | } | |
906 | remove_req: | |
907 | nfs_inode_remove_request(req); | |
908 | next: | |
1d1afcbc | 909 | nfs_unlock_request(req); |
20633f04 | 910 | nfs_end_page_writeback(req); |
3aff4ebb | 911 | nfs_release_request(req); |
6c75dc0d FI |
912 | } |
913 | out: | |
914 | hdr->release(hdr); | |
8e821cad | 915 | } |
1da177e4 | 916 | |
ce59515c | 917 | unsigned long |
ea2cf228 | 918 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 919 | { |
ea2cf228 | 920 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
921 | } |
922 | ||
ea2cf228 | 923 | /* cinfo->lock held by caller */ |
1763da12 | 924 | int |
ea2cf228 FI |
925 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
926 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
927 | { |
928 | struct nfs_page *req, *tmp; | |
929 | int ret = 0; | |
930 | ||
931 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
932 | if (!nfs_lock_request(req)) |
933 | continue; | |
7ad84aa9 | 934 | kref_get(&req->wb_kref); |
ea2cf228 | 935 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 936 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 937 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
938 | nfs_list_add_request(req, dst); |
939 | ret++; | |
1763da12 | 940 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 941 | break; |
d6d6dc7c FI |
942 | } |
943 | return ret; | |
fb8a1f11 TM |
944 | } |
945 | ||
1da177e4 LT |
946 | /* |
947 | * nfs_scan_commit - Scan an inode for commit requests | |
948 | * @inode: NFS inode to scan | |
ea2cf228 FI |
949 | * @dst: mds destination list |
950 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
951 | * |
952 | * Moves requests from the inode's 'commit' request list. | |
953 | * The requests are *not* checked to ensure that they form a contiguous set. | |
954 | */ | |
1763da12 | 955 | int |
ea2cf228 FI |
956 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
957 | struct nfs_commit_info *cinfo) | |
1da177e4 | 958 | { |
d6d6dc7c | 959 | int ret = 0; |
fb8a1f11 | 960 | |
ea2cf228 FI |
961 | spin_lock(cinfo->lock); |
962 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 963 | const int max = INT_MAX; |
d6d6dc7c | 964 | |
ea2cf228 FI |
965 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
966 | cinfo, max); | |
967 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 968 | } |
ea2cf228 | 969 | spin_unlock(cinfo->lock); |
ff778d02 | 970 | return ret; |
1da177e4 | 971 | } |
d6d6dc7c | 972 | |
1da177e4 | 973 | /* |
e7d39069 TM |
974 | * Search for an existing write request, and attempt to update |
975 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 976 | * |
e7d39069 TM |
977 | * If the attempt fails, then the existing request is flushed out |
978 | * to disk. | |
1da177e4 | 979 | */ |
e7d39069 TM |
980 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
981 | struct page *page, | |
982 | unsigned int offset, | |
983 | unsigned int bytes) | |
1da177e4 | 984 | { |
e7d39069 TM |
985 | struct nfs_page *req; |
986 | unsigned int rqend; | |
987 | unsigned int end; | |
988 | int error; | |
989 | ||
990 | if (!PagePrivate(page)) | |
991 | return NULL; | |
1da177e4 LT |
992 | |
993 | end = offset + bytes; | |
e7d39069 | 994 | spin_lock(&inode->i_lock); |
1da177e4 | 995 | |
1da177e4 | 996 | for (;;) { |
84d3a9a9 | 997 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
e7d39069 TM |
998 | if (req == NULL) |
999 | goto out_unlock; | |
1000 | ||
2bfc6e56 WAA |
1001 | /* should be handled by nfs_flush_incompatible */ |
1002 | WARN_ON_ONCE(req->wb_head != req); | |
1003 | WARN_ON_ONCE(req->wb_this_page != req); | |
1004 | ||
e7d39069 TM |
1005 | rqend = req->wb_offset + req->wb_bytes; |
1006 | /* | |
1007 | * Tell the caller to flush out the request if | |
1008 | * the offsets are non-contiguous. | |
1009 | * Note: nfs_flush_incompatible() will already | |
1010 | * have flushed out requests having wrong owners. | |
1011 | */ | |
e468bae9 | 1012 | if (offset > rqend |
e7d39069 TM |
1013 | || end < req->wb_offset) |
1014 | goto out_flushme; | |
1015 | ||
7ad84aa9 | 1016 | if (nfs_lock_request(req)) |
1da177e4 | 1017 | break; |
1da177e4 | 1018 | |
e7d39069 | 1019 | /* The request is locked, so wait and then retry */ |
587142f8 | 1020 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
1021 | error = nfs_wait_on_request(req); |
1022 | nfs_release_request(req); | |
1023 | if (error != 0) | |
1024 | goto out_err; | |
1025 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
1026 | } |
1027 | ||
1028 | /* Okay, the request matches. Update the region */ | |
1029 | if (offset < req->wb_offset) { | |
1030 | req->wb_offset = offset; | |
1031 | req->wb_pgbase = offset; | |
1da177e4 | 1032 | } |
1da177e4 LT |
1033 | if (end > rqend) |
1034 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1035 | else |
1036 | req->wb_bytes = rqend - req->wb_offset; | |
1037 | out_unlock: | |
ca138f36 FI |
1038 | if (req) |
1039 | nfs_clear_request_commit(req); | |
411a99ad | 1040 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
1041 | return req; |
1042 | out_flushme: | |
1043 | spin_unlock(&inode->i_lock); | |
1044 | nfs_release_request(req); | |
1045 | error = nfs_wb_page(inode, page); | |
1046 | out_err: | |
1047 | return ERR_PTR(error); | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * Try to update an existing write request, or create one if there is none. | |
1052 | * | |
1053 | * Note: Should always be called with the Page Lock held to prevent races | |
1054 | * if we have to add a new request. Also assumes that the caller has | |
1055 | * already called nfs_flush_incompatible() if necessary. | |
1056 | */ | |
1057 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1058 | struct page *page, unsigned int offset, unsigned int bytes) | |
1059 | { | |
d56b4ddf | 1060 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1061 | struct nfs_page *req; |
1da177e4 | 1062 | |
e7d39069 TM |
1063 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1064 | if (req != NULL) | |
1065 | goto out; | |
2bfc6e56 | 1066 | req = nfs_create_request(ctx, page, NULL, offset, bytes); |
e7d39069 TM |
1067 | if (IS_ERR(req)) |
1068 | goto out; | |
d6d6dc7c | 1069 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1070 | out: |
61e930a9 | 1071 | return req; |
1da177e4 LT |
1072 | } |
1073 | ||
e7d39069 TM |
1074 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1075 | unsigned int offset, unsigned int count) | |
1076 | { | |
1077 | struct nfs_page *req; | |
1078 | ||
1079 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1080 | if (IS_ERR(req)) | |
1081 | return PTR_ERR(req); | |
1082 | /* Update file length */ | |
1083 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1084 | nfs_mark_uptodate(req); |
a6305ddb | 1085 | nfs_mark_request_dirty(req); |
1d1afcbc | 1086 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1087 | return 0; |
1088 | } | |
1089 | ||
1da177e4 LT |
1090 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1091 | { | |
cd3758e3 | 1092 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1093 | struct nfs_lock_context *l_ctx; |
1da177e4 | 1094 | struct nfs_page *req; |
1a54533e | 1095 | int do_flush, status; |
1da177e4 LT |
1096 | /* |
1097 | * Look for a request corresponding to this page. If there | |
1098 | * is one, and it belongs to another file, we flush it out | |
1099 | * before we try to copy anything into the page. Do this | |
1100 | * due to the lack of an ACCESS-type call in NFSv2. | |
1101 | * Also do the same if we find a request from an existing | |
1102 | * dropped page. | |
1103 | */ | |
1a54533e | 1104 | do { |
84d3a9a9 | 1105 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1106 | if (req == NULL) |
1107 | return 0; | |
2a369153 TM |
1108 | l_ctx = req->wb_lock_context; |
1109 | do_flush = req->wb_page != page || req->wb_context != ctx; | |
2bfc6e56 WAA |
1110 | /* for now, flush if more than 1 request in page_group */ |
1111 | do_flush |= req->wb_this_page != req; | |
0f1d2605 | 1112 | if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { |
2a369153 TM |
1113 | do_flush |= l_ctx->lockowner.l_owner != current->files |
1114 | || l_ctx->lockowner.l_pid != current->tgid; | |
1115 | } | |
1da177e4 | 1116 | nfs_release_request(req); |
1a54533e TM |
1117 | if (!do_flush) |
1118 | return 0; | |
d56b4ddf | 1119 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1120 | } while (status == 0); |
1121 | return status; | |
1da177e4 LT |
1122 | } |
1123 | ||
dc24826b AA |
1124 | /* |
1125 | * Avoid buffered writes when a open context credential's key would | |
1126 | * expire soon. | |
1127 | * | |
1128 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1129 | * | |
1130 | * Return 0 and set a credential flag which triggers the inode to flush | |
1131 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1132 | * RPC_KEY_EXPIRE_TIMEO. | |
1133 | */ | |
1134 | int | |
1135 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1136 | { | |
1137 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
1138 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
1139 | ||
1140 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
1141 | } | |
1142 | ||
1143 | /* | |
1144 | * Test if the open context credential key is marked to expire soon. | |
1145 | */ | |
1146 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) | |
1147 | { | |
1148 | return rpcauth_cred_key_to_expire(ctx->cred); | |
1149 | } | |
1150 | ||
5d47a356 TM |
1151 | /* |
1152 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1153 | * the PageUptodate() flag. In this case, we will need to turn off | |
1154 | * write optimisations that depend on the page contents being correct. | |
1155 | */ | |
8d197a56 | 1156 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 1157 | { |
d529ef83 JL |
1158 | struct nfs_inode *nfsi = NFS_I(inode); |
1159 | ||
8d197a56 TM |
1160 | if (nfs_have_delegated_attributes(inode)) |
1161 | goto out; | |
18dd78c4 | 1162 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
d529ef83 | 1163 | return false; |
4db72b40 | 1164 | smp_rmb(); |
d529ef83 | 1165 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
1166 | return false; |
1167 | out: | |
18dd78c4 SM |
1168 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
1169 | return false; | |
8d197a56 | 1170 | return PageUptodate(page) != 0; |
5d47a356 TM |
1171 | } |
1172 | ||
c7559663 SM |
1173 | /* If we know the page is up to date, and we're not using byte range locks (or |
1174 | * if we have the whole file locked for writing), it may be more efficient to | |
1175 | * extend the write to cover the entire page in order to avoid fragmentation | |
1176 | * inefficiencies. | |
1177 | * | |
263b4509 SM |
1178 | * If the file is opened for synchronous writes then we can just skip the rest |
1179 | * of the checks. | |
c7559663 SM |
1180 | */ |
1181 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
1182 | { | |
1183 | if (file->f_flags & O_DSYNC) | |
1184 | return 0; | |
263b4509 SM |
1185 | if (!nfs_write_pageuptodate(page, inode)) |
1186 | return 0; | |
c7559663 SM |
1187 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1188 | return 1; | |
263b4509 | 1189 | if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && |
c7559663 | 1190 | inode->i_flock->fl_end == OFFSET_MAX && |
263b4509 | 1191 | inode->i_flock->fl_type != F_RDLCK)) |
c7559663 SM |
1192 | return 1; |
1193 | return 0; | |
1194 | } | |
1195 | ||
1da177e4 LT |
1196 | /* |
1197 | * Update and possibly write a cached page of an NFS file. | |
1198 | * | |
1199 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1200 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1201 | */ | |
1202 | int nfs_updatepage(struct file *file, struct page *page, | |
1203 | unsigned int offset, unsigned int count) | |
1204 | { | |
cd3758e3 | 1205 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 1206 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
1207 | int status = 0; |
1208 | ||
91d5b470 CL |
1209 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1210 | ||
6de1472f AV |
1211 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1212 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1213 | |
c7559663 | 1214 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 1215 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 1216 | offset = 0; |
1da177e4 LT |
1217 | } |
1218 | ||
e21195a7 | 1219 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
1220 | if (status < 0) |
1221 | nfs_set_pageerror(page); | |
59b7c05f TM |
1222 | else |
1223 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 1224 | |
48186c7d | 1225 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1226 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1227 | return status; |
1228 | } | |
1229 | ||
3ff7576d | 1230 | static int flush_task_priority(int how) |
1da177e4 LT |
1231 | { |
1232 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1233 | case FLUSH_HIGHPRI: | |
1234 | return RPC_PRIORITY_HIGH; | |
1235 | case FLUSH_LOWPRI: | |
1236 | return RPC_PRIORITY_LOW; | |
1237 | } | |
1238 | return RPC_PRIORITY_NORMAL; | |
1239 | } | |
1240 | ||
d45f60c6 WAA |
1241 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1242 | struct rpc_message *msg, | |
1ed26f33 | 1243 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1244 | { |
d45f60c6 | 1245 | struct inode *inode = hdr->inode; |
3ff7576d | 1246 | int priority = flush_task_priority(how); |
d138d5d1 | 1247 | |
1ed26f33 | 1248 | task_setup_data->priority = priority; |
d45f60c6 | 1249 | NFS_PROTO(inode)->write_setup(hdr, msg); |
d138d5d1 | 1250 | |
8c21c62c | 1251 | nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, |
d45f60c6 | 1252 | &task_setup_data->rpc_client, msg, hdr); |
275acaaf TM |
1253 | } |
1254 | ||
6d884e8f F |
1255 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1256 | * call this on each, which will prepare them to be retried on next | |
1257 | * writeback using standard nfs. | |
1258 | */ | |
1259 | static void nfs_redirty_request(struct nfs_page *req) | |
1260 | { | |
1261 | nfs_mark_request_dirty(req); | |
1d1afcbc | 1262 | nfs_unlock_request(req); |
20633f04 | 1263 | nfs_end_page_writeback(req); |
3aff4ebb | 1264 | nfs_release_request(req); |
6d884e8f F |
1265 | } |
1266 | ||
061ae2ed | 1267 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1268 | { |
1269 | struct nfs_page *req; | |
1270 | ||
1271 | while (!list_empty(head)) { | |
1272 | req = nfs_list_entry(head->next); | |
1273 | nfs_list_remove_request(req); | |
1274 | nfs_redirty_request(req); | |
1275 | } | |
1276 | } | |
1277 | ||
061ae2ed FI |
1278 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1279 | .error_cleanup = nfs_async_write_error, | |
1280 | .completion = nfs_write_completion, | |
1281 | }; | |
1282 | ||
57208fa7 | 1283 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1284 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1285 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1286 | { |
a20c93e3 | 1287 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1288 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1289 | |
1290 | #ifdef CONFIG_NFS_V4_1 | |
1291 | if (server->pnfs_curr_ld && !force_mds) | |
1292 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1293 | #endif | |
4a0de55c AS |
1294 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
1295 | server->wsize, ioflags); | |
1751c363 | 1296 | } |
ddda8e0a | 1297 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1298 | |
dce81290 TM |
1299 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1300 | { | |
41d8d5b7 | 1301 | pgio->pg_ops = &nfs_pgio_rw_ops; |
dce81290 TM |
1302 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; |
1303 | } | |
1f945357 | 1304 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1305 | |
1da177e4 | 1306 | |
0b7c0153 FI |
1307 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1308 | { | |
1309 | struct nfs_commit_data *data = calldata; | |
1310 | ||
1311 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1312 | } | |
1313 | ||
d45f60c6 | 1314 | static void nfs_writeback_release_common(struct nfs_pgio_header *hdr) |
c9d8f89d | 1315 | { |
c65e6254 | 1316 | /* do nothing! */ |
1da177e4 LT |
1317 | } |
1318 | ||
1f2edbe3 TM |
1319 | /* |
1320 | * Special version of should_remove_suid() that ignores capabilities. | |
1321 | */ | |
1322 | static int nfs_should_remove_suid(const struct inode *inode) | |
1323 | { | |
1324 | umode_t mode = inode->i_mode; | |
1325 | int kill = 0; | |
1326 | ||
1327 | /* suid always must be killed */ | |
1328 | if (unlikely(mode & S_ISUID)) | |
1329 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1330 | |
1f2edbe3 TM |
1331 | /* |
1332 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1333 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1334 | */ | |
1335 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1336 | kill |= ATTR_KILL_SGID; | |
1337 | ||
1338 | if (unlikely(kill && S_ISREG(mode))) | |
1339 | return kill; | |
1340 | ||
1341 | return 0; | |
1342 | } | |
788e7a89 | 1343 | |
1da177e4 LT |
1344 | /* |
1345 | * This function is called when the WRITE call is complete. | |
1346 | */ | |
d45f60c6 WAA |
1347 | static int nfs_writeback_done(struct rpc_task *task, |
1348 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1349 | struct inode *inode) |
1da177e4 | 1350 | { |
788e7a89 | 1351 | int status; |
1da177e4 | 1352 | |
f551e44f CL |
1353 | /* |
1354 | * ->write_done will attempt to use post-op attributes to detect | |
1355 | * conflicting writes by other clients. A strict interpretation | |
1356 | * of close-to-open would allow us to continue caching even if | |
1357 | * another writer had changed the file, but some applications | |
1358 | * depend on tighter cache coherency when writing. | |
1359 | */ | |
d45f60c6 | 1360 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1361 | if (status != 0) |
0eecb214 | 1362 | return status; |
d45f60c6 | 1363 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
91d5b470 | 1364 | |
d45f60c6 WAA |
1365 | if (hdr->res.verf->committed < hdr->args.stable && |
1366 | task->tk_status >= 0) { | |
1da177e4 LT |
1367 | /* We tried a write call, but the server did not |
1368 | * commit data to stable storage even though we | |
1369 | * requested it. | |
1370 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1371 | * the server reports NFS_DATA_SYNC, but performs | |
1372 | * NFS_FILE_SYNC. We therefore implement this checking | |
1373 | * as a dprintk() in order to avoid filling syslog. | |
1374 | */ | |
1375 | static unsigned long complain; | |
1376 | ||
a69aef14 | 1377 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1378 | if (time_before(complain, jiffies)) { |
48186c7d | 1379 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1380 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1381 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
d45f60c6 | 1382 | hdr->res.verf->committed, hdr->args.stable); |
1da177e4 LT |
1383 | complain = jiffies + 300 * HZ; |
1384 | } | |
1385 | } | |
1f2edbe3 TM |
1386 | |
1387 | /* Deal with the suid/sgid bit corner case */ | |
1388 | if (nfs_should_remove_suid(inode)) | |
1389 | nfs_mark_for_revalidate(inode); | |
0eecb214 AS |
1390 | return 0; |
1391 | } | |
1392 | ||
1393 | /* | |
1394 | * This function is called when the WRITE call is complete. | |
1395 | */ | |
d45f60c6 WAA |
1396 | static void nfs_writeback_result(struct rpc_task *task, |
1397 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1398 | { |
d45f60c6 WAA |
1399 | struct nfs_pgio_args *argp = &hdr->args; |
1400 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1401 | |
1402 | if (resp->count < argp->count) { | |
1da177e4 LT |
1403 | static unsigned long complain; |
1404 | ||
6c75dc0d | 1405 | /* This a short write! */ |
d45f60c6 | 1406 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1407 | |
1da177e4 | 1408 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1409 | if (resp->count == 0) { |
1410 | if (time_before(complain, jiffies)) { | |
1411 | printk(KERN_WARNING | |
1412 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1413 | argp->count); | |
1414 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1415 | } |
d45f60c6 | 1416 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1417 | task->tk_status = -EIO; |
13602896 | 1418 | return; |
1da177e4 | 1419 | } |
6c75dc0d FI |
1420 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1421 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1422 | /* Resend from where the server left off */ | |
d45f60c6 | 1423 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1424 | argp->offset += resp->count; |
1425 | argp->pgbase += resp->count; | |
1426 | argp->count -= resp->count; | |
1427 | } else { | |
1428 | /* Resend as a stable write in order to avoid | |
1429 | * headaches in the case of a server crash. | |
1430 | */ | |
1431 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1432 | } |
6c75dc0d | 1433 | rpc_restart_call_prepare(task); |
1da177e4 | 1434 | } |
1da177e4 LT |
1435 | } |
1436 | ||
1437 | ||
71d0a611 TM |
1438 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1439 | { | |
b8413f98 TM |
1440 | int ret; |
1441 | ||
71d0a611 TM |
1442 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1443 | return 1; | |
b8413f98 TM |
1444 | if (!may_wait) |
1445 | return 0; | |
1446 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1447 | NFS_INO_COMMIT, | |
1448 | nfs_wait_bit_killable, | |
1449 | TASK_KILLABLE); | |
1450 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1451 | } |
1452 | ||
f453a54a | 1453 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1454 | { |
1455 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
4e857c58 | 1456 | smp_mb__after_atomic(); |
71d0a611 TM |
1457 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); |
1458 | } | |
1459 | ||
0b7c0153 | 1460 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1461 | { |
0b7c0153 FI |
1462 | put_nfs_open_context(data->context); |
1463 | nfs_commit_free(data); | |
1da177e4 | 1464 | } |
e0c2b380 | 1465 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1466 | |
0b7c0153 | 1467 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd | 1468 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1469 | int how, int flags) |
1da177e4 | 1470 | { |
07737691 | 1471 | struct rpc_task *task; |
9ace33cd | 1472 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1473 | struct rpc_message msg = { |
1474 | .rpc_argp = &data->args, | |
1475 | .rpc_resp = &data->res, | |
9ace33cd | 1476 | .rpc_cred = data->cred, |
bdc7f021 | 1477 | }; |
84115e1c | 1478 | struct rpc_task_setup task_setup_data = { |
07737691 | 1479 | .task = &data->task, |
9ace33cd | 1480 | .rpc_client = clnt, |
bdc7f021 | 1481 | .rpc_message = &msg, |
9ace33cd | 1482 | .callback_ops = call_ops, |
84115e1c | 1483 | .callback_data = data, |
101070ca | 1484 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1485 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1486 | .priority = priority, |
84115e1c | 1487 | }; |
9ace33cd FI |
1488 | /* Set up the initial task struct. */ |
1489 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1490 | ||
1491 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1492 | ||
8c21c62c WAA |
1493 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1494 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1495 | ||
9ace33cd FI |
1496 | task = rpc_run_task(&task_setup_data); |
1497 | if (IS_ERR(task)) | |
1498 | return PTR_ERR(task); | |
1499 | if (how & FLUSH_SYNC) | |
1500 | rpc_wait_for_completion_task(task); | |
1501 | rpc_put_task(task); | |
1502 | return 0; | |
1503 | } | |
e0c2b380 | 1504 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1505 | |
378520b8 PT |
1506 | static loff_t nfs_get_lwb(struct list_head *head) |
1507 | { | |
1508 | loff_t lwb = 0; | |
1509 | struct nfs_page *req; | |
1510 | ||
1511 | list_for_each_entry(req, head, wb_list) | |
1512 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1513 | lwb = req_offset(req) + req->wb_bytes; | |
1514 | ||
1515 | return lwb; | |
1516 | } | |
1517 | ||
9ace33cd FI |
1518 | /* |
1519 | * Set up the argument/result storage required for the RPC call. | |
1520 | */ | |
0b7c0153 | 1521 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1522 | struct list_head *head, |
1523 | struct pnfs_layout_segment *lseg, | |
1524 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1525 | { |
1526 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1527 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1528 | |
1529 | /* Set up the RPC argument and reply structs | |
1530 | * NB: take care not to mess about with data->commit et al. */ | |
1531 | ||
1532 | list_splice_init(head, &data->pages); | |
1da177e4 | 1533 | |
1da177e4 | 1534 | data->inode = inode; |
9ace33cd | 1535 | data->cred = first->wb_context->cred; |
988b6dce | 1536 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1537 | /* only set lwb for pnfs commit */ |
1538 | if (lseg) | |
1539 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1540 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1541 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1542 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1543 | |
1544 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1545 | /* Note: we always request a commit of the entire inode */ |
1546 | data->args.offset = 0; | |
1547 | data->args.count = 0; | |
0b7c0153 | 1548 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1549 | data->res.fattr = &data->fattr; |
1550 | data->res.verf = &data->verf; | |
0e574af1 | 1551 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1552 | } |
e0c2b380 | 1553 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1554 | |
e0c2b380 | 1555 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1556 | struct pnfs_layout_segment *lseg, |
1557 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1558 | { |
1559 | struct nfs_page *req; | |
1560 | ||
1561 | while (!list_empty(page_list)) { | |
1562 | req = nfs_list_entry(page_list->next); | |
1563 | nfs_list_remove_request(req); | |
ea2cf228 | 1564 | nfs_mark_request_commit(req, lseg, cinfo); |
56f9cd68 FI |
1565 | if (!cinfo->dreq) { |
1566 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 1567 | dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
1568 | BDI_RECLAIMABLE); |
1569 | } | |
1d1afcbc | 1570 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1571 | } |
1572 | } | |
e0c2b380 | 1573 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1574 | |
1da177e4 LT |
1575 | /* |
1576 | * Commit dirty pages | |
1577 | */ | |
1578 | static int | |
ea2cf228 FI |
1579 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1580 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1581 | { |
0b7c0153 | 1582 | struct nfs_commit_data *data; |
1da177e4 | 1583 | |
c9d8f89d | 1584 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1585 | |
1586 | if (!data) | |
1587 | goto out_bad; | |
1588 | ||
1589 | /* Set up the argument struct */ | |
f453a54a FI |
1590 | nfs_init_commit(data, head, NULL, cinfo); |
1591 | atomic_inc(&cinfo->mds->rpcs_out); | |
9f0ec176 AA |
1592 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, |
1593 | how, 0); | |
1da177e4 | 1594 | out_bad: |
ea2cf228 | 1595 | nfs_retry_commit(head, NULL, cinfo); |
f453a54a | 1596 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
1da177e4 LT |
1597 | return -ENOMEM; |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * COMMIT call returned | |
1602 | */ | |
788e7a89 | 1603 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1604 | { |
0b7c0153 | 1605 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1606 | |
a3f565b1 | 1607 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1608 | task->tk_pid, task->tk_status); |
1609 | ||
788e7a89 | 1610 | /* Call the NFS version-specific code */ |
c0d0e96b | 1611 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1612 | } |
1613 | ||
f453a54a | 1614 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1615 | { |
5917ce84 | 1616 | struct nfs_page *req; |
c9d8f89d | 1617 | int status = data->task.tk_status; |
f453a54a | 1618 | struct nfs_commit_info cinfo; |
353db796 | 1619 | struct nfs_server *nfss; |
788e7a89 | 1620 | |
1da177e4 LT |
1621 | while (!list_empty(&data->pages)) { |
1622 | req = nfs_list_entry(data->pages.next); | |
1623 | nfs_list_remove_request(req); | |
d6d6dc7c | 1624 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1625 | |
1e8968c5 | 1626 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1627 | req->wb_context->dentry->d_sb->s_id, |
1e8968c5 | 1628 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
1da177e4 LT |
1629 | req->wb_bytes, |
1630 | (long long)req_offset(req)); | |
c9d8f89d TM |
1631 | if (status < 0) { |
1632 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1633 | nfs_inode_remove_request(req); |
c9d8f89d | 1634 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1635 | goto next; |
1636 | } | |
1637 | ||
1638 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1639 | * returned by the server against all stored verfs. */ | |
2f2c63bc | 1640 | if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { |
1da177e4 LT |
1641 | /* We have a match */ |
1642 | nfs_inode_remove_request(req); | |
1643 | dprintk(" OK\n"); | |
1644 | goto next; | |
1645 | } | |
1646 | /* We have a mismatch. Write the page again */ | |
1647 | dprintk(" mismatch\n"); | |
6d884e8f | 1648 | nfs_mark_request_dirty(req); |
05990d1b | 1649 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1650 | next: |
1d1afcbc | 1651 | nfs_unlock_and_release_request(req); |
1da177e4 | 1652 | } |
353db796 N |
1653 | nfss = NFS_SERVER(data->inode); |
1654 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
1655 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); | |
1656 | ||
f453a54a FI |
1657 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
1658 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
1659 | nfs_commit_clear_lock(NFS_I(data->inode)); | |
5917ce84 FI |
1660 | } |
1661 | ||
1662 | static void nfs_commit_release(void *calldata) | |
1663 | { | |
0b7c0153 | 1664 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1665 | |
f453a54a | 1666 | data->completion_ops->completion(data); |
c9d8f89d | 1667 | nfs_commitdata_release(calldata); |
1da177e4 | 1668 | } |
788e7a89 TM |
1669 | |
1670 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1671 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1672 | .rpc_call_done = nfs_commit_done, |
1673 | .rpc_release = nfs_commit_release, | |
1674 | }; | |
1da177e4 | 1675 | |
f453a54a FI |
1676 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1677 | .completion = nfs_commit_release_pages, | |
1678 | .error_cleanup = nfs_commit_clear_lock, | |
1679 | }; | |
1680 | ||
1763da12 FI |
1681 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1682 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1683 | { |
1684 | int status; | |
1685 | ||
ea2cf228 | 1686 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1687 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1688 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1689 | return status; |
1690 | } | |
1691 | ||
b608b283 | 1692 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1693 | { |
1da177e4 | 1694 | LIST_HEAD(head); |
ea2cf228 | 1695 | struct nfs_commit_info cinfo; |
71d0a611 | 1696 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1697 | int res; |
1da177e4 | 1698 | |
b8413f98 TM |
1699 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1700 | if (res <= 0) | |
c5efa5fc | 1701 | goto out_mark_dirty; |
ea2cf228 FI |
1702 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1703 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1704 | if (res) { |
a861a1e1 FI |
1705 | int error; |
1706 | ||
ea2cf228 | 1707 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1708 | if (error < 0) |
1709 | return error; | |
b8413f98 | 1710 | if (!may_wait) |
c5efa5fc | 1711 | goto out_mark_dirty; |
74316201 | 1712 | error = wait_on_bit_action(&NFS_I(inode)->flags, |
b8413f98 TM |
1713 | NFS_INO_COMMIT, |
1714 | nfs_wait_bit_killable, | |
1715 | TASK_KILLABLE); | |
1716 | if (error < 0) | |
1717 | return error; | |
71d0a611 TM |
1718 | } else |
1719 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1720 | return res; |
1721 | /* Note: If we exit without ensuring that the commit is complete, | |
1722 | * we must mark the inode as dirty. Otherwise, future calls to | |
1723 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1724 | * that the data is on the disk. | |
1725 | */ | |
1726 | out_mark_dirty: | |
1727 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1728 | return res; |
1729 | } | |
8fc795f7 TM |
1730 | |
1731 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1732 | { | |
420e3646 TM |
1733 | struct nfs_inode *nfsi = NFS_I(inode); |
1734 | int flags = FLUSH_SYNC; | |
1735 | int ret = 0; | |
8fc795f7 | 1736 | |
3236c3e1 | 1737 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1738 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1739 | return ret; |
1740 | ||
a00dd6c0 JL |
1741 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1742 | /* Don't commit yet if this is a non-blocking flush and there | |
1743 | * are a lot of outstanding writes for this mapping. | |
1744 | */ | |
cb1410c7 | 1745 | if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) |
a00dd6c0 | 1746 | goto out_mark_dirty; |
420e3646 | 1747 | |
a00dd6c0 | 1748 | /* don't wait for the COMMIT response */ |
420e3646 | 1749 | flags = 0; |
a00dd6c0 JL |
1750 | } |
1751 | ||
420e3646 TM |
1752 | ret = nfs_commit_inode(inode, flags); |
1753 | if (ret >= 0) { | |
1754 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1755 | if (ret < wbc->nr_to_write) | |
1756 | wbc->nr_to_write -= ret; | |
1757 | else | |
1758 | wbc->nr_to_write = 0; | |
1759 | } | |
8fc795f7 | 1760 | return 0; |
420e3646 TM |
1761 | } |
1762 | out_mark_dirty: | |
8fc795f7 TM |
1763 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1764 | return ret; | |
1765 | } | |
1da177e4 | 1766 | |
8fc795f7 TM |
1767 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1768 | { | |
a8d8f02c BS |
1769 | return nfs_commit_unstable_pages(inode, wbc); |
1770 | } | |
89d77c8f | 1771 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1772 | |
acdc53b2 TM |
1773 | /* |
1774 | * flush the inode to disk. | |
1775 | */ | |
1776 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1777 | { |
1778 | struct writeback_control wbc = { | |
72cb77f4 | 1779 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1780 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1781 | .range_start = 0, |
1782 | .range_end = LLONG_MAX, | |
34901f70 | 1783 | }; |
f4ce1299 TM |
1784 | int ret; |
1785 | ||
1786 | trace_nfs_writeback_inode_enter(inode); | |
1787 | ||
1788 | ret = sync_inode(inode, &wbc); | |
34901f70 | 1789 | |
f4ce1299 TM |
1790 | trace_nfs_writeback_inode_exit(inode, ret); |
1791 | return ret; | |
1c75950b | 1792 | } |
ddda8e0a | 1793 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1794 | |
1b3b4a1a TM |
1795 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1796 | { | |
1797 | struct nfs_page *req; | |
1b3b4a1a TM |
1798 | int ret = 0; |
1799 | ||
3e217045 WAA |
1800 | wait_on_page_writeback(page); |
1801 | ||
1802 | /* blocking call to cancel all requests and join to a single (head) | |
1803 | * request */ | |
1804 | req = nfs_lock_and_join_requests(page, false); | |
1805 | ||
1806 | if (IS_ERR(req)) { | |
1807 | ret = PTR_ERR(req); | |
1808 | } else if (req) { | |
1809 | /* all requests from this page have been cancelled by | |
1810 | * nfs_lock_and_join_requests, so just remove the head | |
1811 | * request from the inode / page_private pointer and | |
1812 | * release it */ | |
1813 | nfs_inode_remove_request(req); | |
1814 | /* | |
1815 | * In case nfs_inode_remove_request has marked the | |
1816 | * page as being dirty | |
1817 | */ | |
1818 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1819 | nfs_unlock_and_release_request(req); | |
1b3b4a1a | 1820 | } |
3e217045 | 1821 | |
1b3b4a1a TM |
1822 | return ret; |
1823 | } | |
1824 | ||
7f2f12d9 TM |
1825 | /* |
1826 | * Write back all requests on one page - we do this before reading it. | |
1827 | */ | |
1828 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b | 1829 | { |
29418aa4 | 1830 | loff_t range_start = page_file_offset(page); |
1c75950b | 1831 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); |
4d770ccf | 1832 | struct writeback_control wbc = { |
4d770ccf | 1833 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1834 | .nr_to_write = 0, |
4d770ccf TM |
1835 | .range_start = range_start, |
1836 | .range_end = range_end, | |
1837 | }; | |
1838 | int ret; | |
1c75950b | 1839 | |
f4ce1299 TM |
1840 | trace_nfs_writeback_page_enter(inode); |
1841 | ||
0522f6ad | 1842 | for (;;) { |
ba8b06e6 | 1843 | wait_on_page_writeback(page); |
73e3302f TM |
1844 | if (clear_page_dirty_for_io(page)) { |
1845 | ret = nfs_writepage_locked(page, &wbc); | |
1846 | if (ret < 0) | |
1847 | goto out_error; | |
0522f6ad | 1848 | continue; |
7f2f12d9 | 1849 | } |
f4ce1299 | 1850 | ret = 0; |
0522f6ad TM |
1851 | if (!PagePrivate(page)) |
1852 | break; | |
1853 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1854 | if (ret < 0) |
73e3302f | 1855 | goto out_error; |
7f2f12d9 | 1856 | } |
73e3302f | 1857 | out_error: |
f4ce1299 | 1858 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 1859 | return ret; |
1c75950b TM |
1860 | } |
1861 | ||
074cc1de TM |
1862 | #ifdef CONFIG_MIGRATION |
1863 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1864 | struct page *page, enum migrate_mode mode) |
074cc1de | 1865 | { |
2da95652 JL |
1866 | /* |
1867 | * If PagePrivate is set, then the page is currently associated with | |
1868 | * an in-progress read or write request. Don't try to migrate it. | |
1869 | * | |
1870 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1871 | * that we can safely release the inode reference while holding | |
1872 | * the page lock. | |
1873 | */ | |
1874 | if (PagePrivate(page)) | |
1875 | return -EBUSY; | |
074cc1de | 1876 | |
8c209ce7 DH |
1877 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
1878 | return -EBUSY; | |
074cc1de | 1879 | |
a6bc32b8 | 1880 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1881 | } |
1882 | #endif | |
1883 | ||
f7b422b1 | 1884 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1885 | { |
1886 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 1887 | sizeof(struct nfs_pgio_header), |
1da177e4 | 1888 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1889 | NULL); |
1da177e4 LT |
1890 | if (nfs_wdata_cachep == NULL) |
1891 | return -ENOMEM; | |
1892 | ||
93d2341c MD |
1893 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1894 | nfs_wdata_cachep); | |
1da177e4 | 1895 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 1896 | goto out_destroy_write_cache; |
1da177e4 | 1897 | |
0b7c0153 FI |
1898 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1899 | sizeof(struct nfs_commit_data), | |
1900 | 0, SLAB_HWCACHE_ALIGN, | |
1901 | NULL); | |
1902 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 1903 | goto out_destroy_write_mempool; |
0b7c0153 | 1904 | |
93d2341c | 1905 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 1906 | nfs_cdata_cachep); |
1da177e4 | 1907 | if (nfs_commit_mempool == NULL) |
3dd4765f | 1908 | goto out_destroy_commit_cache; |
1da177e4 | 1909 | |
89a09141 PZ |
1910 | /* |
1911 | * NFS congestion size, scale with available memory. | |
1912 | * | |
1913 | * 64MB: 8192k | |
1914 | * 128MB: 11585k | |
1915 | * 256MB: 16384k | |
1916 | * 512MB: 23170k | |
1917 | * 1GB: 32768k | |
1918 | * 2GB: 46340k | |
1919 | * 4GB: 65536k | |
1920 | * 8GB: 92681k | |
1921 | * 16GB: 131072k | |
1922 | * | |
1923 | * This allows larger machines to have larger/more transfers. | |
1924 | * Limit the default to 256M | |
1925 | */ | |
1926 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1927 | if (nfs_congestion_kb > 256*1024) | |
1928 | nfs_congestion_kb = 256*1024; | |
1929 | ||
1da177e4 | 1930 | return 0; |
3dd4765f JL |
1931 | |
1932 | out_destroy_commit_cache: | |
1933 | kmem_cache_destroy(nfs_cdata_cachep); | |
1934 | out_destroy_write_mempool: | |
1935 | mempool_destroy(nfs_wdata_mempool); | |
1936 | out_destroy_write_cache: | |
1937 | kmem_cache_destroy(nfs_wdata_cachep); | |
1938 | return -ENOMEM; | |
1da177e4 LT |
1939 | } |
1940 | ||
266bee88 | 1941 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1942 | { |
1943 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 1944 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 1945 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 1946 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1947 | } |
1948 | ||
4a0de55c | 1949 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
a4cdda59 | 1950 | .rw_mode = FMODE_WRITE, |
4a0de55c AS |
1951 | .rw_alloc_header = nfs_writehdr_alloc, |
1952 | .rw_free_header = nfs_writehdr_free, | |
a4cdda59 | 1953 | .rw_release = nfs_writeback_release_common, |
0eecb214 AS |
1954 | .rw_done = nfs_writeback_done, |
1955 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 1956 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 1957 | }; |