]>
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 CL |
577 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
578 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
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); | |
011e2a7f | 673 | if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 674 | inode->i_version++; |
29418aa4 MG |
675 | /* |
676 | * Swap-space should not get truncated. Hence no need to plug the race | |
677 | * with invalidate/truncate. | |
678 | */ | |
679 | if (likely(!PageSwapCache(req->wb_page))) { | |
680 | set_bit(PG_MAPPED, &req->wb_flags); | |
681 | SetPagePrivate(req->wb_page); | |
682 | set_page_private(req->wb_page, (unsigned long)req); | |
683 | } | |
1da177e4 | 684 | nfsi->npages++; |
17089a29 WAA |
685 | /* this a head request for a page group - mark it as having an |
686 | * extra reference so sub groups can follow suit */ | |
687 | WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); | |
c03b4024 | 688 | kref_get(&req->wb_kref); |
e7d39069 | 689 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
690 | } |
691 | ||
692 | /* | |
89a09141 | 693 | * Remove a write request from an inode |
1da177e4 LT |
694 | */ |
695 | static void nfs_inode_remove_request(struct nfs_page *req) | |
696 | { | |
3d4ff43d | 697 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 | 698 | struct nfs_inode *nfsi = NFS_I(inode); |
20633f04 | 699 | struct nfs_page *head; |
1da177e4 | 700 | |
20633f04 WAA |
701 | if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { |
702 | head = req->wb_head; | |
703 | ||
704 | spin_lock(&inode->i_lock); | |
705 | if (likely(!PageSwapCache(head->wb_page))) { | |
706 | set_page_private(head->wb_page, 0); | |
707 | ClearPagePrivate(head->wb_page); | |
95905446 N |
708 | smp_mb__after_atomic(); |
709 | wake_up_page(head->wb_page, PG_private); | |
20633f04 WAA |
710 | clear_bit(PG_MAPPED, &head->wb_flags); |
711 | } | |
712 | nfsi->npages--; | |
713 | spin_unlock(&inode->i_lock); | |
29418aa4 | 714 | } |
17089a29 WAA |
715 | |
716 | if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) | |
717 | nfs_release_request(req); | |
b412ddf0 WAA |
718 | else |
719 | WARN_ON_ONCE(1); | |
1da177e4 LT |
720 | } |
721 | ||
61822ab5 | 722 | static void |
6d884e8f | 723 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 724 | { |
61822ab5 TM |
725 | __set_page_dirty_nobuffers(req->wb_page); |
726 | } | |
727 | ||
3a3908c8 TM |
728 | /* |
729 | * nfs_page_search_commits_for_head_request_locked | |
730 | * | |
731 | * Search through commit lists on @inode for the head request for @page. | |
732 | * Must be called while holding the inode (which is cinfo) lock. | |
733 | * | |
734 | * Returns the head request if found, or NULL if not found. | |
735 | */ | |
736 | static struct nfs_page * | |
737 | nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, | |
738 | struct page *page) | |
739 | { | |
740 | struct nfs_page *freq, *t; | |
741 | struct nfs_commit_info cinfo; | |
742 | struct inode *inode = &nfsi->vfs_inode; | |
743 | ||
744 | nfs_init_cinfo_from_inode(&cinfo, inode); | |
745 | ||
746 | /* search through pnfs commit lists */ | |
747 | freq = pnfs_search_commit_reqs(inode, &cinfo, page); | |
748 | if (freq) | |
749 | return freq->wb_head; | |
750 | ||
751 | /* Linearly search the commit list for the correct request */ | |
752 | list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { | |
753 | if (freq->wb_page == page) | |
754 | return freq->wb_head; | |
755 | } | |
756 | ||
757 | return NULL; | |
758 | } | |
759 | ||
8dd37758 TM |
760 | /** |
761 | * nfs_request_add_commit_list - add request to a commit list | |
762 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
763 | * @dst: commit list head |
764 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 765 | * |
ea2cf228 | 766 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
767 | * number of outstanding requests requiring a commit as well as |
768 | * the MM page stats. | |
769 | * | |
ea2cf228 | 770 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 771 | * holding the nfs_page lock. |
1da177e4 | 772 | */ |
8dd37758 | 773 | void |
ea2cf228 FI |
774 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
775 | struct nfs_commit_info *cinfo) | |
1da177e4 | 776 | { |
e468bae9 | 777 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
778 | spin_lock(cinfo->lock); |
779 | nfs_list_add_request(req, dst); | |
780 | cinfo->mds->ncommit++; | |
781 | spin_unlock(cinfo->lock); | |
56f9cd68 FI |
782 | if (!cinfo->dreq) { |
783 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 784 | inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
785 | BDI_RECLAIMABLE); |
786 | __mark_inode_dirty(req->wb_context->dentry->d_inode, | |
787 | I_DIRTY_DATASYNC); | |
788 | } | |
1da177e4 | 789 | } |
8dd37758 TM |
790 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
791 | ||
792 | /** | |
793 | * nfs_request_remove_commit_list - Remove request from a commit list | |
794 | * @req: pointer to a nfs_page | |
ea2cf228 | 795 | * @cinfo: holds list lock and accounting info |
8dd37758 | 796 | * |
ea2cf228 | 797 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
798 | * number of outstanding requests requiring a commit |
799 | * It does not update the MM page stats. | |
800 | * | |
ea2cf228 | 801 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
802 | */ |
803 | void | |
ea2cf228 FI |
804 | nfs_request_remove_commit_list(struct nfs_page *req, |
805 | struct nfs_commit_info *cinfo) | |
8dd37758 | 806 | { |
8dd37758 TM |
807 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
808 | return; | |
809 | nfs_list_remove_request(req); | |
ea2cf228 | 810 | cinfo->mds->ncommit--; |
8dd37758 TM |
811 | } |
812 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
813 | ||
ea2cf228 FI |
814 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
815 | struct inode *inode) | |
816 | { | |
817 | cinfo->lock = &inode->i_lock; | |
818 | cinfo->mds = &NFS_I(inode)->commit_info; | |
819 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 820 | cinfo->dreq = NULL; |
f453a54a | 821 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
822 | } |
823 | ||
824 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
825 | struct inode *inode, | |
826 | struct nfs_direct_req *dreq) | |
827 | { | |
1763da12 FI |
828 | if (dreq) |
829 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
830 | else | |
831 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
832 | } |
833 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
834 | |
835 | /* | |
836 | * Add a request to the inode's commit list. | |
837 | */ | |
1763da12 | 838 | void |
ea2cf228 FI |
839 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
840 | struct nfs_commit_info *cinfo) | |
8dd37758 | 841 | { |
ea2cf228 | 842 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 843 | return; |
ea2cf228 | 844 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 845 | } |
8e821cad | 846 | |
d6d6dc7c FI |
847 | static void |
848 | nfs_clear_page_commit(struct page *page) | |
849 | { | |
850 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
d56b4ddf | 851 | dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); |
d6d6dc7c FI |
852 | } |
853 | ||
411a99ad | 854 | /* Called holding inode (/cinfo) lock */ |
8dd37758 | 855 | static void |
e468bae9 TM |
856 | nfs_clear_request_commit(struct nfs_page *req) |
857 | { | |
8dd37758 TM |
858 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
859 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 860 | struct nfs_commit_info cinfo; |
e468bae9 | 861 | |
ea2cf228 FI |
862 | nfs_init_cinfo_from_inode(&cinfo, inode); |
863 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
ea2cf228 | 864 | nfs_request_remove_commit_list(req, &cinfo); |
8dd37758 | 865 | } |
d6d6dc7c | 866 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 867 | } |
e468bae9 TM |
868 | } |
869 | ||
d45f60c6 | 870 | int nfs_write_need_commit(struct nfs_pgio_header *hdr) |
8e821cad | 871 | { |
c65e6254 | 872 | if (hdr->verf.committed == NFS_DATA_SYNC) |
d45f60c6 | 873 | return hdr->lseg == NULL; |
c65e6254 | 874 | return hdr->verf.committed != NFS_FILE_SYNC; |
8e821cad TM |
875 | } |
876 | ||
061ae2ed | 877 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 878 | { |
ea2cf228 | 879 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
880 | unsigned long bytes = 0; |
881 | ||
882 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
883 | goto out; | |
ea2cf228 | 884 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
885 | while (!list_empty(&hdr->pages)) { |
886 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
887 | |
888 | bytes += req->wb_bytes; | |
889 | nfs_list_remove_request(req); | |
890 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
891 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 892 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
893 | nfs_context_set_write_error(req->wb_context, hdr->error); |
894 | goto remove_req; | |
895 | } | |
c65e6254 | 896 | if (nfs_write_need_commit(hdr)) { |
f79d06f5 | 897 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
ea2cf228 | 898 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
899 | goto next; |
900 | } | |
901 | remove_req: | |
902 | nfs_inode_remove_request(req); | |
903 | next: | |
1d1afcbc | 904 | nfs_unlock_request(req); |
20633f04 | 905 | nfs_end_page_writeback(req); |
3aff4ebb | 906 | nfs_release_request(req); |
6c75dc0d FI |
907 | } |
908 | out: | |
909 | hdr->release(hdr); | |
8e821cad | 910 | } |
1da177e4 | 911 | |
ce59515c | 912 | unsigned long |
ea2cf228 | 913 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 914 | { |
ea2cf228 | 915 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
916 | } |
917 | ||
ea2cf228 | 918 | /* cinfo->lock held by caller */ |
1763da12 | 919 | int |
ea2cf228 FI |
920 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
921 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
922 | { |
923 | struct nfs_page *req, *tmp; | |
924 | int ret = 0; | |
925 | ||
926 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
927 | if (!nfs_lock_request(req)) |
928 | continue; | |
7ad84aa9 | 929 | kref_get(&req->wb_kref); |
ea2cf228 | 930 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 931 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 932 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
933 | nfs_list_add_request(req, dst); |
934 | ret++; | |
1763da12 | 935 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 936 | break; |
d6d6dc7c FI |
937 | } |
938 | return ret; | |
fb8a1f11 TM |
939 | } |
940 | ||
1da177e4 LT |
941 | /* |
942 | * nfs_scan_commit - Scan an inode for commit requests | |
943 | * @inode: NFS inode to scan | |
ea2cf228 FI |
944 | * @dst: mds destination list |
945 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
946 | * |
947 | * Moves requests from the inode's 'commit' request list. | |
948 | * The requests are *not* checked to ensure that they form a contiguous set. | |
949 | */ | |
1763da12 | 950 | int |
ea2cf228 FI |
951 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
952 | struct nfs_commit_info *cinfo) | |
1da177e4 | 953 | { |
d6d6dc7c | 954 | int ret = 0; |
fb8a1f11 | 955 | |
ea2cf228 FI |
956 | spin_lock(cinfo->lock); |
957 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 958 | const int max = INT_MAX; |
d6d6dc7c | 959 | |
ea2cf228 FI |
960 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
961 | cinfo, max); | |
962 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 963 | } |
ea2cf228 | 964 | spin_unlock(cinfo->lock); |
ff778d02 | 965 | return ret; |
1da177e4 | 966 | } |
d6d6dc7c | 967 | |
1da177e4 | 968 | /* |
e7d39069 TM |
969 | * Search for an existing write request, and attempt to update |
970 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 971 | * |
e7d39069 TM |
972 | * If the attempt fails, then the existing request is flushed out |
973 | * to disk. | |
1da177e4 | 974 | */ |
e7d39069 TM |
975 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
976 | struct page *page, | |
977 | unsigned int offset, | |
978 | unsigned int bytes) | |
1da177e4 | 979 | { |
e7d39069 TM |
980 | struct nfs_page *req; |
981 | unsigned int rqend; | |
982 | unsigned int end; | |
983 | int error; | |
984 | ||
985 | if (!PagePrivate(page)) | |
986 | return NULL; | |
1da177e4 LT |
987 | |
988 | end = offset + bytes; | |
e7d39069 | 989 | spin_lock(&inode->i_lock); |
1da177e4 | 990 | |
1da177e4 | 991 | for (;;) { |
84d3a9a9 | 992 | req = nfs_page_find_head_request_locked(NFS_I(inode), page); |
e7d39069 TM |
993 | if (req == NULL) |
994 | goto out_unlock; | |
995 | ||
2bfc6e56 WAA |
996 | /* should be handled by nfs_flush_incompatible */ |
997 | WARN_ON_ONCE(req->wb_head != req); | |
998 | WARN_ON_ONCE(req->wb_this_page != req); | |
999 | ||
e7d39069 TM |
1000 | rqend = req->wb_offset + req->wb_bytes; |
1001 | /* | |
1002 | * Tell the caller to flush out the request if | |
1003 | * the offsets are non-contiguous. | |
1004 | * Note: nfs_flush_incompatible() will already | |
1005 | * have flushed out requests having wrong owners. | |
1006 | */ | |
e468bae9 | 1007 | if (offset > rqend |
e7d39069 TM |
1008 | || end < req->wb_offset) |
1009 | goto out_flushme; | |
1010 | ||
7ad84aa9 | 1011 | if (nfs_lock_request(req)) |
1da177e4 | 1012 | break; |
1da177e4 | 1013 | |
e7d39069 | 1014 | /* The request is locked, so wait and then retry */ |
587142f8 | 1015 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
1016 | error = nfs_wait_on_request(req); |
1017 | nfs_release_request(req); | |
1018 | if (error != 0) | |
1019 | goto out_err; | |
1020 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
1021 | } |
1022 | ||
1023 | /* Okay, the request matches. Update the region */ | |
1024 | if (offset < req->wb_offset) { | |
1025 | req->wb_offset = offset; | |
1026 | req->wb_pgbase = offset; | |
1da177e4 | 1027 | } |
1da177e4 LT |
1028 | if (end > rqend) |
1029 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
1030 | else |
1031 | req->wb_bytes = rqend - req->wb_offset; | |
1032 | out_unlock: | |
ca138f36 FI |
1033 | if (req) |
1034 | nfs_clear_request_commit(req); | |
411a99ad | 1035 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
1036 | return req; |
1037 | out_flushme: | |
1038 | spin_unlock(&inode->i_lock); | |
1039 | nfs_release_request(req); | |
1040 | error = nfs_wb_page(inode, page); | |
1041 | out_err: | |
1042 | return ERR_PTR(error); | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * Try to update an existing write request, or create one if there is none. | |
1047 | * | |
1048 | * Note: Should always be called with the Page Lock held to prevent races | |
1049 | * if we have to add a new request. Also assumes that the caller has | |
1050 | * already called nfs_flush_incompatible() if necessary. | |
1051 | */ | |
1052 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
1053 | struct page *page, unsigned int offset, unsigned int bytes) | |
1054 | { | |
d56b4ddf | 1055 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 1056 | struct nfs_page *req; |
1da177e4 | 1057 | |
e7d39069 TM |
1058 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
1059 | if (req != NULL) | |
1060 | goto out; | |
2bfc6e56 | 1061 | req = nfs_create_request(ctx, page, NULL, offset, bytes); |
e7d39069 TM |
1062 | if (IS_ERR(req)) |
1063 | goto out; | |
d6d6dc7c | 1064 | nfs_inode_add_request(inode, req); |
efc91ed0 | 1065 | out: |
61e930a9 | 1066 | return req; |
1da177e4 LT |
1067 | } |
1068 | ||
e7d39069 TM |
1069 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1070 | unsigned int offset, unsigned int count) | |
1071 | { | |
1072 | struct nfs_page *req; | |
1073 | ||
1074 | req = nfs_setup_write_request(ctx, page, offset, count); | |
1075 | if (IS_ERR(req)) | |
1076 | return PTR_ERR(req); | |
1077 | /* Update file length */ | |
1078 | nfs_grow_file(page, offset, count); | |
d72ddcba | 1079 | nfs_mark_uptodate(req); |
a6305ddb | 1080 | nfs_mark_request_dirty(req); |
1d1afcbc | 1081 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
1082 | return 0; |
1083 | } | |
1084 | ||
1da177e4 LT |
1085 | int nfs_flush_incompatible(struct file *file, struct page *page) |
1086 | { | |
cd3758e3 | 1087 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 1088 | struct nfs_lock_context *l_ctx; |
1da177e4 | 1089 | struct nfs_page *req; |
1a54533e | 1090 | int do_flush, status; |
1da177e4 LT |
1091 | /* |
1092 | * Look for a request corresponding to this page. If there | |
1093 | * is one, and it belongs to another file, we flush it out | |
1094 | * before we try to copy anything into the page. Do this | |
1095 | * due to the lack of an ACCESS-type call in NFSv2. | |
1096 | * Also do the same if we find a request from an existing | |
1097 | * dropped page. | |
1098 | */ | |
1a54533e | 1099 | do { |
84d3a9a9 | 1100 | req = nfs_page_find_head_request(page); |
1a54533e TM |
1101 | if (req == NULL) |
1102 | return 0; | |
2a369153 TM |
1103 | l_ctx = req->wb_lock_context; |
1104 | do_flush = req->wb_page != page || req->wb_context != ctx; | |
2bfc6e56 WAA |
1105 | /* for now, flush if more than 1 request in page_group */ |
1106 | do_flush |= req->wb_this_page != req; | |
0f1d2605 | 1107 | if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { |
2a369153 TM |
1108 | do_flush |= l_ctx->lockowner.l_owner != current->files |
1109 | || l_ctx->lockowner.l_pid != current->tgid; | |
1110 | } | |
1da177e4 | 1111 | nfs_release_request(req); |
1a54533e TM |
1112 | if (!do_flush) |
1113 | return 0; | |
d56b4ddf | 1114 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
1115 | } while (status == 0); |
1116 | return status; | |
1da177e4 LT |
1117 | } |
1118 | ||
dc24826b AA |
1119 | /* |
1120 | * Avoid buffered writes when a open context credential's key would | |
1121 | * expire soon. | |
1122 | * | |
1123 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
1124 | * | |
1125 | * Return 0 and set a credential flag which triggers the inode to flush | |
1126 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
1127 | * RPC_KEY_EXPIRE_TIMEO. | |
1128 | */ | |
1129 | int | |
1130 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
1131 | { | |
1132 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
1133 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
1134 | ||
1135 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Test if the open context credential key is marked to expire soon. | |
1140 | */ | |
1141 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) | |
1142 | { | |
1143 | return rpcauth_cred_key_to_expire(ctx->cred); | |
1144 | } | |
1145 | ||
5d47a356 TM |
1146 | /* |
1147 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
1148 | * the PageUptodate() flag. In this case, we will need to turn off | |
1149 | * write optimisations that depend on the page contents being correct. | |
1150 | */ | |
8d197a56 | 1151 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 1152 | { |
d529ef83 JL |
1153 | struct nfs_inode *nfsi = NFS_I(inode); |
1154 | ||
8d197a56 TM |
1155 | if (nfs_have_delegated_attributes(inode)) |
1156 | goto out; | |
18dd78c4 | 1157 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
d529ef83 | 1158 | return false; |
4db72b40 | 1159 | smp_rmb(); |
d529ef83 | 1160 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
1161 | return false; |
1162 | out: | |
18dd78c4 SM |
1163 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
1164 | return false; | |
8d197a56 | 1165 | return PageUptodate(page) != 0; |
5d47a356 TM |
1166 | } |
1167 | ||
c7559663 SM |
1168 | /* If we know the page is up to date, and we're not using byte range locks (or |
1169 | * if we have the whole file locked for writing), it may be more efficient to | |
1170 | * extend the write to cover the entire page in order to avoid fragmentation | |
1171 | * inefficiencies. | |
1172 | * | |
263b4509 SM |
1173 | * If the file is opened for synchronous writes then we can just skip the rest |
1174 | * of the checks. | |
c7559663 SM |
1175 | */ |
1176 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
1177 | { | |
1178 | if (file->f_flags & O_DSYNC) | |
1179 | return 0; | |
263b4509 SM |
1180 | if (!nfs_write_pageuptodate(page, inode)) |
1181 | return 0; | |
c7559663 SM |
1182 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
1183 | return 1; | |
263b4509 | 1184 | if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && |
c7559663 | 1185 | inode->i_flock->fl_end == OFFSET_MAX && |
263b4509 | 1186 | inode->i_flock->fl_type != F_RDLCK)) |
c7559663 SM |
1187 | return 1; |
1188 | return 0; | |
1189 | } | |
1190 | ||
1da177e4 LT |
1191 | /* |
1192 | * Update and possibly write a cached page of an NFS file. | |
1193 | * | |
1194 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
1195 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
1196 | */ | |
1197 | int nfs_updatepage(struct file *file, struct page *page, | |
1198 | unsigned int offset, unsigned int count) | |
1199 | { | |
cd3758e3 | 1200 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 1201 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
1202 | int status = 0; |
1203 | ||
91d5b470 CL |
1204 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
1205 | ||
6de1472f AV |
1206 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
1207 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 1208 | |
c7559663 | 1209 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 1210 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 1211 | offset = 0; |
1da177e4 LT |
1212 | } |
1213 | ||
e21195a7 | 1214 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
1215 | if (status < 0) |
1216 | nfs_set_pageerror(page); | |
59b7c05f TM |
1217 | else |
1218 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 1219 | |
48186c7d | 1220 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 1221 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
1222 | return status; |
1223 | } | |
1224 | ||
3ff7576d | 1225 | static int flush_task_priority(int how) |
1da177e4 LT |
1226 | { |
1227 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
1228 | case FLUSH_HIGHPRI: | |
1229 | return RPC_PRIORITY_HIGH; | |
1230 | case FLUSH_LOWPRI: | |
1231 | return RPC_PRIORITY_LOW; | |
1232 | } | |
1233 | return RPC_PRIORITY_NORMAL; | |
1234 | } | |
1235 | ||
d45f60c6 WAA |
1236 | static void nfs_initiate_write(struct nfs_pgio_header *hdr, |
1237 | struct rpc_message *msg, | |
1ed26f33 | 1238 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 1239 | { |
d45f60c6 | 1240 | struct inode *inode = hdr->inode; |
3ff7576d | 1241 | int priority = flush_task_priority(how); |
d138d5d1 | 1242 | |
1ed26f33 | 1243 | task_setup_data->priority = priority; |
d45f60c6 | 1244 | NFS_PROTO(inode)->write_setup(hdr, msg); |
d138d5d1 | 1245 | |
8c21c62c | 1246 | nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, |
d45f60c6 | 1247 | &task_setup_data->rpc_client, msg, hdr); |
275acaaf TM |
1248 | } |
1249 | ||
6d884e8f F |
1250 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1251 | * call this on each, which will prepare them to be retried on next | |
1252 | * writeback using standard nfs. | |
1253 | */ | |
1254 | static void nfs_redirty_request(struct nfs_page *req) | |
1255 | { | |
1256 | nfs_mark_request_dirty(req); | |
1d1afcbc | 1257 | nfs_unlock_request(req); |
20633f04 | 1258 | nfs_end_page_writeback(req); |
3aff4ebb | 1259 | nfs_release_request(req); |
6d884e8f F |
1260 | } |
1261 | ||
061ae2ed | 1262 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1263 | { |
1264 | struct nfs_page *req; | |
1265 | ||
1266 | while (!list_empty(head)) { | |
1267 | req = nfs_list_entry(head->next); | |
1268 | nfs_list_remove_request(req); | |
1269 | nfs_redirty_request(req); | |
1270 | } | |
1271 | } | |
1272 | ||
061ae2ed FI |
1273 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1274 | .error_cleanup = nfs_async_write_error, | |
1275 | .completion = nfs_write_completion, | |
1276 | }; | |
1277 | ||
57208fa7 | 1278 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 1279 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 1280 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 1281 | { |
a20c93e3 | 1282 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 1283 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
1284 | |
1285 | #ifdef CONFIG_NFS_V4_1 | |
1286 | if (server->pnfs_curr_ld && !force_mds) | |
1287 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
1288 | #endif | |
4a0de55c AS |
1289 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
1290 | server->wsize, ioflags); | |
1751c363 | 1291 | } |
ddda8e0a | 1292 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1293 | |
dce81290 TM |
1294 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1295 | { | |
41d8d5b7 | 1296 | pgio->pg_ops = &nfs_pgio_rw_ops; |
dce81290 TM |
1297 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; |
1298 | } | |
1f945357 | 1299 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1300 | |
1da177e4 | 1301 | |
0b7c0153 FI |
1302 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1303 | { | |
1304 | struct nfs_commit_data *data = calldata; | |
1305 | ||
1306 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1307 | } | |
1308 | ||
d45f60c6 | 1309 | static void nfs_writeback_release_common(struct nfs_pgio_header *hdr) |
c9d8f89d | 1310 | { |
c65e6254 | 1311 | /* do nothing! */ |
1da177e4 LT |
1312 | } |
1313 | ||
1f2edbe3 TM |
1314 | /* |
1315 | * Special version of should_remove_suid() that ignores capabilities. | |
1316 | */ | |
1317 | static int nfs_should_remove_suid(const struct inode *inode) | |
1318 | { | |
1319 | umode_t mode = inode->i_mode; | |
1320 | int kill = 0; | |
1321 | ||
1322 | /* suid always must be killed */ | |
1323 | if (unlikely(mode & S_ISUID)) | |
1324 | kill = ATTR_KILL_SUID; | |
788e7a89 | 1325 | |
1f2edbe3 TM |
1326 | /* |
1327 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1328 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1329 | */ | |
1330 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1331 | kill |= ATTR_KILL_SGID; | |
1332 | ||
1333 | if (unlikely(kill && S_ISREG(mode))) | |
1334 | return kill; | |
1335 | ||
1336 | return 0; | |
1337 | } | |
788e7a89 | 1338 | |
1da177e4 LT |
1339 | /* |
1340 | * This function is called when the WRITE call is complete. | |
1341 | */ | |
d45f60c6 WAA |
1342 | static int nfs_writeback_done(struct rpc_task *task, |
1343 | struct nfs_pgio_header *hdr, | |
0eecb214 | 1344 | struct inode *inode) |
1da177e4 | 1345 | { |
788e7a89 | 1346 | int status; |
1da177e4 | 1347 | |
f551e44f CL |
1348 | /* |
1349 | * ->write_done will attempt to use post-op attributes to detect | |
1350 | * conflicting writes by other clients. A strict interpretation | |
1351 | * of close-to-open would allow us to continue caching even if | |
1352 | * another writer had changed the file, but some applications | |
1353 | * depend on tighter cache coherency when writing. | |
1354 | */ | |
d45f60c6 | 1355 | status = NFS_PROTO(inode)->write_done(task, hdr); |
788e7a89 | 1356 | if (status != 0) |
0eecb214 | 1357 | return status; |
d45f60c6 | 1358 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); |
91d5b470 | 1359 | |
d45f60c6 WAA |
1360 | if (hdr->res.verf->committed < hdr->args.stable && |
1361 | task->tk_status >= 0) { | |
1da177e4 LT |
1362 | /* We tried a write call, but the server did not |
1363 | * commit data to stable storage even though we | |
1364 | * requested it. | |
1365 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1366 | * the server reports NFS_DATA_SYNC, but performs | |
1367 | * NFS_FILE_SYNC. We therefore implement this checking | |
1368 | * as a dprintk() in order to avoid filling syslog. | |
1369 | */ | |
1370 | static unsigned long complain; | |
1371 | ||
a69aef14 | 1372 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1373 | if (time_before(complain, jiffies)) { |
48186c7d | 1374 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1375 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1376 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
d45f60c6 | 1377 | hdr->res.verf->committed, hdr->args.stable); |
1da177e4 LT |
1378 | complain = jiffies + 300 * HZ; |
1379 | } | |
1380 | } | |
1f2edbe3 TM |
1381 | |
1382 | /* Deal with the suid/sgid bit corner case */ | |
1383 | if (nfs_should_remove_suid(inode)) | |
1384 | nfs_mark_for_revalidate(inode); | |
0eecb214 AS |
1385 | return 0; |
1386 | } | |
1387 | ||
1388 | /* | |
1389 | * This function is called when the WRITE call is complete. | |
1390 | */ | |
d45f60c6 WAA |
1391 | static void nfs_writeback_result(struct rpc_task *task, |
1392 | struct nfs_pgio_header *hdr) | |
0eecb214 | 1393 | { |
d45f60c6 WAA |
1394 | struct nfs_pgio_args *argp = &hdr->args; |
1395 | struct nfs_pgio_res *resp = &hdr->res; | |
1f2edbe3 TM |
1396 | |
1397 | if (resp->count < argp->count) { | |
1da177e4 LT |
1398 | static unsigned long complain; |
1399 | ||
6c75dc0d | 1400 | /* This a short write! */ |
d45f60c6 | 1401 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1402 | |
1da177e4 | 1403 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1404 | if (resp->count == 0) { |
1405 | if (time_before(complain, jiffies)) { | |
1406 | printk(KERN_WARNING | |
1407 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1408 | argp->count); | |
1409 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1410 | } |
d45f60c6 | 1411 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
6c75dc0d | 1412 | task->tk_status = -EIO; |
13602896 | 1413 | return; |
1da177e4 | 1414 | } |
6c75dc0d FI |
1415 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1416 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1417 | /* Resend from where the server left off */ | |
d45f60c6 | 1418 | hdr->mds_offset += resp->count; |
6c75dc0d FI |
1419 | argp->offset += resp->count; |
1420 | argp->pgbase += resp->count; | |
1421 | argp->count -= resp->count; | |
1422 | } else { | |
1423 | /* Resend as a stable write in order to avoid | |
1424 | * headaches in the case of a server crash. | |
1425 | */ | |
1426 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1427 | } |
6c75dc0d | 1428 | rpc_restart_call_prepare(task); |
1da177e4 | 1429 | } |
1da177e4 LT |
1430 | } |
1431 | ||
1432 | ||
71d0a611 TM |
1433 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1434 | { | |
b8413f98 TM |
1435 | int ret; |
1436 | ||
71d0a611 TM |
1437 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1438 | return 1; | |
b8413f98 TM |
1439 | if (!may_wait) |
1440 | return 0; | |
1441 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1442 | NFS_INO_COMMIT, | |
1443 | nfs_wait_bit_killable, | |
1444 | TASK_KILLABLE); | |
1445 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1446 | } |
1447 | ||
f453a54a | 1448 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1449 | { |
1450 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
4e857c58 | 1451 | smp_mb__after_atomic(); |
71d0a611 TM |
1452 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); |
1453 | } | |
1454 | ||
0b7c0153 | 1455 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1456 | { |
0b7c0153 FI |
1457 | put_nfs_open_context(data->context); |
1458 | nfs_commit_free(data); | |
1da177e4 | 1459 | } |
e0c2b380 | 1460 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1461 | |
0b7c0153 | 1462 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd | 1463 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1464 | int how, int flags) |
1da177e4 | 1465 | { |
07737691 | 1466 | struct rpc_task *task; |
9ace33cd | 1467 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1468 | struct rpc_message msg = { |
1469 | .rpc_argp = &data->args, | |
1470 | .rpc_resp = &data->res, | |
9ace33cd | 1471 | .rpc_cred = data->cred, |
bdc7f021 | 1472 | }; |
84115e1c | 1473 | struct rpc_task_setup task_setup_data = { |
07737691 | 1474 | .task = &data->task, |
9ace33cd | 1475 | .rpc_client = clnt, |
bdc7f021 | 1476 | .rpc_message = &msg, |
9ace33cd | 1477 | .callback_ops = call_ops, |
84115e1c | 1478 | .callback_data = data, |
101070ca | 1479 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1480 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1481 | .priority = priority, |
84115e1c | 1482 | }; |
9ace33cd FI |
1483 | /* Set up the initial task struct. */ |
1484 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1485 | ||
1486 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1487 | ||
8c21c62c WAA |
1488 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1489 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1490 | ||
9ace33cd FI |
1491 | task = rpc_run_task(&task_setup_data); |
1492 | if (IS_ERR(task)) | |
1493 | return PTR_ERR(task); | |
1494 | if (how & FLUSH_SYNC) | |
1495 | rpc_wait_for_completion_task(task); | |
1496 | rpc_put_task(task); | |
1497 | return 0; | |
1498 | } | |
e0c2b380 | 1499 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd | 1500 | |
378520b8 PT |
1501 | static loff_t nfs_get_lwb(struct list_head *head) |
1502 | { | |
1503 | loff_t lwb = 0; | |
1504 | struct nfs_page *req; | |
1505 | ||
1506 | list_for_each_entry(req, head, wb_list) | |
1507 | if (lwb < (req_offset(req) + req->wb_bytes)) | |
1508 | lwb = req_offset(req) + req->wb_bytes; | |
1509 | ||
1510 | return lwb; | |
1511 | } | |
1512 | ||
9ace33cd FI |
1513 | /* |
1514 | * Set up the argument/result storage required for the RPC call. | |
1515 | */ | |
0b7c0153 | 1516 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1517 | struct list_head *head, |
1518 | struct pnfs_layout_segment *lseg, | |
1519 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1520 | { |
1521 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1522 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1523 | |
1524 | /* Set up the RPC argument and reply structs | |
1525 | * NB: take care not to mess about with data->commit et al. */ | |
1526 | ||
1527 | list_splice_init(head, &data->pages); | |
1da177e4 | 1528 | |
1da177e4 | 1529 | data->inode = inode; |
9ace33cd | 1530 | data->cred = first->wb_context->cred; |
988b6dce | 1531 | data->lseg = lseg; /* reference transferred */ |
378520b8 PT |
1532 | /* only set lwb for pnfs commit */ |
1533 | if (lseg) | |
1534 | data->lwb = nfs_get_lwb(&data->pages); | |
9ace33cd | 1535 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1536 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1537 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1538 | |
1539 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1540 | /* Note: we always request a commit of the entire inode */ |
1541 | data->args.offset = 0; | |
1542 | data->args.count = 0; | |
0b7c0153 | 1543 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1544 | data->res.fattr = &data->fattr; |
1545 | data->res.verf = &data->verf; | |
0e574af1 | 1546 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1547 | } |
e0c2b380 | 1548 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1549 | |
e0c2b380 | 1550 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1551 | struct pnfs_layout_segment *lseg, |
1552 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1553 | { |
1554 | struct nfs_page *req; | |
1555 | ||
1556 | while (!list_empty(page_list)) { | |
1557 | req = nfs_list_entry(page_list->next); | |
1558 | nfs_list_remove_request(req); | |
ea2cf228 | 1559 | nfs_mark_request_commit(req, lseg, cinfo); |
56f9cd68 FI |
1560 | if (!cinfo->dreq) { |
1561 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 1562 | dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
1563 | BDI_RECLAIMABLE); |
1564 | } | |
1d1afcbc | 1565 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1566 | } |
1567 | } | |
e0c2b380 | 1568 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1569 | |
1da177e4 LT |
1570 | /* |
1571 | * Commit dirty pages | |
1572 | */ | |
1573 | static int | |
ea2cf228 FI |
1574 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1575 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1576 | { |
0b7c0153 | 1577 | struct nfs_commit_data *data; |
1da177e4 | 1578 | |
c9d8f89d | 1579 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1580 | |
1581 | if (!data) | |
1582 | goto out_bad; | |
1583 | ||
1584 | /* Set up the argument struct */ | |
f453a54a FI |
1585 | nfs_init_commit(data, head, NULL, cinfo); |
1586 | atomic_inc(&cinfo->mds->rpcs_out); | |
9f0ec176 AA |
1587 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, |
1588 | how, 0); | |
1da177e4 | 1589 | out_bad: |
ea2cf228 | 1590 | nfs_retry_commit(head, NULL, cinfo); |
f453a54a | 1591 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
1da177e4 LT |
1592 | return -ENOMEM; |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * COMMIT call returned | |
1597 | */ | |
788e7a89 | 1598 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1599 | { |
0b7c0153 | 1600 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1601 | |
a3f565b1 | 1602 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1603 | task->tk_pid, task->tk_status); |
1604 | ||
788e7a89 | 1605 | /* Call the NFS version-specific code */ |
c0d0e96b | 1606 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1607 | } |
1608 | ||
f453a54a | 1609 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1610 | { |
5917ce84 | 1611 | struct nfs_page *req; |
c9d8f89d | 1612 | int status = data->task.tk_status; |
f453a54a | 1613 | struct nfs_commit_info cinfo; |
353db796 | 1614 | struct nfs_server *nfss; |
788e7a89 | 1615 | |
1da177e4 LT |
1616 | while (!list_empty(&data->pages)) { |
1617 | req = nfs_list_entry(data->pages.next); | |
1618 | nfs_list_remove_request(req); | |
d6d6dc7c | 1619 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1620 | |
1e8968c5 | 1621 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1622 | req->wb_context->dentry->d_sb->s_id, |
1e8968c5 | 1623 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
1da177e4 LT |
1624 | req->wb_bytes, |
1625 | (long long)req_offset(req)); | |
c9d8f89d TM |
1626 | if (status < 0) { |
1627 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1628 | nfs_inode_remove_request(req); |
c9d8f89d | 1629 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1630 | goto next; |
1631 | } | |
1632 | ||
1633 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1634 | * returned by the server against all stored verfs. */ | |
2f2c63bc | 1635 | if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { |
1da177e4 LT |
1636 | /* We have a match */ |
1637 | nfs_inode_remove_request(req); | |
1638 | dprintk(" OK\n"); | |
1639 | goto next; | |
1640 | } | |
1641 | /* We have a mismatch. Write the page again */ | |
1642 | dprintk(" mismatch\n"); | |
6d884e8f | 1643 | nfs_mark_request_dirty(req); |
05990d1b | 1644 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1645 | next: |
1d1afcbc | 1646 | nfs_unlock_and_release_request(req); |
1da177e4 | 1647 | } |
353db796 N |
1648 | nfss = NFS_SERVER(data->inode); |
1649 | if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) | |
1650 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); | |
1651 | ||
f453a54a FI |
1652 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
1653 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
1654 | nfs_commit_clear_lock(NFS_I(data->inode)); | |
5917ce84 FI |
1655 | } |
1656 | ||
1657 | static void nfs_commit_release(void *calldata) | |
1658 | { | |
0b7c0153 | 1659 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1660 | |
f453a54a | 1661 | data->completion_ops->completion(data); |
c9d8f89d | 1662 | nfs_commitdata_release(calldata); |
1da177e4 | 1663 | } |
788e7a89 TM |
1664 | |
1665 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1666 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1667 | .rpc_call_done = nfs_commit_done, |
1668 | .rpc_release = nfs_commit_release, | |
1669 | }; | |
1da177e4 | 1670 | |
f453a54a FI |
1671 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1672 | .completion = nfs_commit_release_pages, | |
1673 | .error_cleanup = nfs_commit_clear_lock, | |
1674 | }; | |
1675 | ||
1763da12 FI |
1676 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1677 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1678 | { |
1679 | int status; | |
1680 | ||
ea2cf228 | 1681 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1682 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1683 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1684 | return status; |
1685 | } | |
1686 | ||
b608b283 | 1687 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1688 | { |
1da177e4 | 1689 | LIST_HEAD(head); |
ea2cf228 | 1690 | struct nfs_commit_info cinfo; |
71d0a611 | 1691 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1692 | int res; |
1da177e4 | 1693 | |
b8413f98 TM |
1694 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1695 | if (res <= 0) | |
c5efa5fc | 1696 | goto out_mark_dirty; |
ea2cf228 FI |
1697 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1698 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1699 | if (res) { |
a861a1e1 FI |
1700 | int error; |
1701 | ||
ea2cf228 | 1702 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1703 | if (error < 0) |
1704 | return error; | |
b8413f98 | 1705 | if (!may_wait) |
c5efa5fc | 1706 | goto out_mark_dirty; |
74316201 | 1707 | error = wait_on_bit_action(&NFS_I(inode)->flags, |
b8413f98 TM |
1708 | NFS_INO_COMMIT, |
1709 | nfs_wait_bit_killable, | |
1710 | TASK_KILLABLE); | |
1711 | if (error < 0) | |
1712 | return error; | |
71d0a611 TM |
1713 | } else |
1714 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1715 | return res; |
1716 | /* Note: If we exit without ensuring that the commit is complete, | |
1717 | * we must mark the inode as dirty. Otherwise, future calls to | |
1718 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1719 | * that the data is on the disk. | |
1720 | */ | |
1721 | out_mark_dirty: | |
1722 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1723 | return res; |
1724 | } | |
8fc795f7 TM |
1725 | |
1726 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1727 | { | |
420e3646 TM |
1728 | struct nfs_inode *nfsi = NFS_I(inode); |
1729 | int flags = FLUSH_SYNC; | |
1730 | int ret = 0; | |
8fc795f7 | 1731 | |
3236c3e1 | 1732 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1733 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1734 | return ret; |
1735 | ||
a00dd6c0 JL |
1736 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1737 | /* Don't commit yet if this is a non-blocking flush and there | |
1738 | * are a lot of outstanding writes for this mapping. | |
1739 | */ | |
ea2cf228 | 1740 | if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) |
a00dd6c0 | 1741 | goto out_mark_dirty; |
420e3646 | 1742 | |
a00dd6c0 | 1743 | /* don't wait for the COMMIT response */ |
420e3646 | 1744 | flags = 0; |
a00dd6c0 JL |
1745 | } |
1746 | ||
420e3646 TM |
1747 | ret = nfs_commit_inode(inode, flags); |
1748 | if (ret >= 0) { | |
1749 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1750 | if (ret < wbc->nr_to_write) | |
1751 | wbc->nr_to_write -= ret; | |
1752 | else | |
1753 | wbc->nr_to_write = 0; | |
1754 | } | |
8fc795f7 | 1755 | return 0; |
420e3646 TM |
1756 | } |
1757 | out_mark_dirty: | |
8fc795f7 TM |
1758 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1759 | return ret; | |
1760 | } | |
1da177e4 | 1761 | |
8fc795f7 TM |
1762 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1763 | { | |
a8d8f02c BS |
1764 | return nfs_commit_unstable_pages(inode, wbc); |
1765 | } | |
89d77c8f | 1766 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1767 | |
acdc53b2 TM |
1768 | /* |
1769 | * flush the inode to disk. | |
1770 | */ | |
1771 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1772 | { |
1773 | struct writeback_control wbc = { | |
72cb77f4 | 1774 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1775 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1776 | .range_start = 0, |
1777 | .range_end = LLONG_MAX, | |
34901f70 | 1778 | }; |
f4ce1299 TM |
1779 | int ret; |
1780 | ||
1781 | trace_nfs_writeback_inode_enter(inode); | |
1782 | ||
1783 | ret = sync_inode(inode, &wbc); | |
34901f70 | 1784 | |
f4ce1299 TM |
1785 | trace_nfs_writeback_inode_exit(inode, ret); |
1786 | return ret; | |
1c75950b | 1787 | } |
ddda8e0a | 1788 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1789 | |
1b3b4a1a TM |
1790 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1791 | { | |
1792 | struct nfs_page *req; | |
1b3b4a1a TM |
1793 | int ret = 0; |
1794 | ||
3e217045 WAA |
1795 | wait_on_page_writeback(page); |
1796 | ||
1797 | /* blocking call to cancel all requests and join to a single (head) | |
1798 | * request */ | |
1799 | req = nfs_lock_and_join_requests(page, false); | |
1800 | ||
1801 | if (IS_ERR(req)) { | |
1802 | ret = PTR_ERR(req); | |
1803 | } else if (req) { | |
1804 | /* all requests from this page have been cancelled by | |
1805 | * nfs_lock_and_join_requests, so just remove the head | |
1806 | * request from the inode / page_private pointer and | |
1807 | * release it */ | |
1808 | nfs_inode_remove_request(req); | |
1809 | /* | |
1810 | * In case nfs_inode_remove_request has marked the | |
1811 | * page as being dirty | |
1812 | */ | |
1813 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1814 | nfs_unlock_and_release_request(req); | |
1b3b4a1a | 1815 | } |
3e217045 | 1816 | |
1b3b4a1a TM |
1817 | return ret; |
1818 | } | |
1819 | ||
7f2f12d9 TM |
1820 | /* |
1821 | * Write back all requests on one page - we do this before reading it. | |
1822 | */ | |
1823 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b | 1824 | { |
29418aa4 | 1825 | loff_t range_start = page_file_offset(page); |
1c75950b | 1826 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); |
4d770ccf | 1827 | struct writeback_control wbc = { |
4d770ccf | 1828 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1829 | .nr_to_write = 0, |
4d770ccf TM |
1830 | .range_start = range_start, |
1831 | .range_end = range_end, | |
1832 | }; | |
1833 | int ret; | |
1c75950b | 1834 | |
f4ce1299 TM |
1835 | trace_nfs_writeback_page_enter(inode); |
1836 | ||
0522f6ad | 1837 | for (;;) { |
ba8b06e6 | 1838 | wait_on_page_writeback(page); |
73e3302f TM |
1839 | if (clear_page_dirty_for_io(page)) { |
1840 | ret = nfs_writepage_locked(page, &wbc); | |
1841 | if (ret < 0) | |
1842 | goto out_error; | |
0522f6ad | 1843 | continue; |
7f2f12d9 | 1844 | } |
f4ce1299 | 1845 | ret = 0; |
0522f6ad TM |
1846 | if (!PagePrivate(page)) |
1847 | break; | |
1848 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1849 | if (ret < 0) |
73e3302f | 1850 | goto out_error; |
7f2f12d9 | 1851 | } |
73e3302f | 1852 | out_error: |
f4ce1299 | 1853 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 1854 | return ret; |
1c75950b TM |
1855 | } |
1856 | ||
074cc1de TM |
1857 | #ifdef CONFIG_MIGRATION |
1858 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1859 | struct page *page, enum migrate_mode mode) |
074cc1de | 1860 | { |
2da95652 JL |
1861 | /* |
1862 | * If PagePrivate is set, then the page is currently associated with | |
1863 | * an in-progress read or write request. Don't try to migrate it. | |
1864 | * | |
1865 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1866 | * that we can safely release the inode reference while holding | |
1867 | * the page lock. | |
1868 | */ | |
1869 | if (PagePrivate(page)) | |
1870 | return -EBUSY; | |
074cc1de | 1871 | |
8c209ce7 DH |
1872 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
1873 | return -EBUSY; | |
074cc1de | 1874 | |
a6bc32b8 | 1875 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1876 | } |
1877 | #endif | |
1878 | ||
f7b422b1 | 1879 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1880 | { |
1881 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1e7f3a48 | 1882 | sizeof(struct nfs_pgio_header), |
1da177e4 | 1883 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1884 | NULL); |
1da177e4 LT |
1885 | if (nfs_wdata_cachep == NULL) |
1886 | return -ENOMEM; | |
1887 | ||
93d2341c MD |
1888 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1889 | nfs_wdata_cachep); | |
1da177e4 | 1890 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 1891 | goto out_destroy_write_cache; |
1da177e4 | 1892 | |
0b7c0153 FI |
1893 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1894 | sizeof(struct nfs_commit_data), | |
1895 | 0, SLAB_HWCACHE_ALIGN, | |
1896 | NULL); | |
1897 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 1898 | goto out_destroy_write_mempool; |
0b7c0153 | 1899 | |
93d2341c | 1900 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 1901 | nfs_cdata_cachep); |
1da177e4 | 1902 | if (nfs_commit_mempool == NULL) |
3dd4765f | 1903 | goto out_destroy_commit_cache; |
1da177e4 | 1904 | |
89a09141 PZ |
1905 | /* |
1906 | * NFS congestion size, scale with available memory. | |
1907 | * | |
1908 | * 64MB: 8192k | |
1909 | * 128MB: 11585k | |
1910 | * 256MB: 16384k | |
1911 | * 512MB: 23170k | |
1912 | * 1GB: 32768k | |
1913 | * 2GB: 46340k | |
1914 | * 4GB: 65536k | |
1915 | * 8GB: 92681k | |
1916 | * 16GB: 131072k | |
1917 | * | |
1918 | * This allows larger machines to have larger/more transfers. | |
1919 | * Limit the default to 256M | |
1920 | */ | |
1921 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1922 | if (nfs_congestion_kb > 256*1024) | |
1923 | nfs_congestion_kb = 256*1024; | |
1924 | ||
1da177e4 | 1925 | return 0; |
3dd4765f JL |
1926 | |
1927 | out_destroy_commit_cache: | |
1928 | kmem_cache_destroy(nfs_cdata_cachep); | |
1929 | out_destroy_write_mempool: | |
1930 | mempool_destroy(nfs_wdata_mempool); | |
1931 | out_destroy_write_cache: | |
1932 | kmem_cache_destroy(nfs_wdata_cachep); | |
1933 | return -ENOMEM; | |
1da177e4 LT |
1934 | } |
1935 | ||
266bee88 | 1936 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1937 | { |
1938 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 1939 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 1940 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 1941 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1942 | } |
1943 | ||
4a0de55c | 1944 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
a4cdda59 | 1945 | .rw_mode = FMODE_WRITE, |
4a0de55c AS |
1946 | .rw_alloc_header = nfs_writehdr_alloc, |
1947 | .rw_free_header = nfs_writehdr_free, | |
a4cdda59 | 1948 | .rw_release = nfs_writeback_release_common, |
0eecb214 AS |
1949 | .rw_done = nfs_writeback_done, |
1950 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 1951 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 1952 | }; |