]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/write.c | |
3 | * | |
4 | * Writing file data over NFS. | |
5 | * | |
6 | * We do it like this: When a (user) process wishes to write data to an | |
7 | * NFS file, a write request is allocated that contains the RPC task data | |
8 | * plus some info on the page to be written, and added to the inode's | |
9 | * write chain. If the process writes past the end of the page, an async | |
10 | * RPC call to write the page is scheduled immediately; otherwise, the call | |
11 | * is delayed for a few seconds. | |
12 | * | |
13 | * Just like readahead, no async I/O is performed if wsize < PAGE_SIZE. | |
14 | * | |
15 | * Write requests are kept on the inode's writeback list. Each entry in | |
16 | * that list references the page (portion) to be written. When the | |
17 | * cache timeout has expired, the RPC task is woken up, and tries to | |
18 | * lock the page. As soon as it manages to do so, the request is moved | |
19 | * from the writeback list to the writelock list. | |
20 | * | |
21 | * Note: we must make sure never to confuse the inode passed in the | |
22 | * write_page request with the one in page->inode. As far as I understand | |
23 | * it, these are different when doing a swap-out. | |
24 | * | |
25 | * To understand everything that goes on here and in the NFS read code, | |
26 | * one should be aware that a page is locked in exactly one of the following | |
27 | * cases: | |
28 | * | |
29 | * - A write request is in progress. | |
30 | * - A user process is in generic_file_write/nfs_update_page | |
31 | * - A user process is in generic_file_read | |
32 | * | |
33 | * Also note that because of the way pages are invalidated in | |
34 | * nfs_revalidate_inode, the following assertions hold: | |
35 | * | |
36 | * - If a page is dirty, there will be no read requests (a page will | |
37 | * not be re-read unless invalidated by nfs_revalidate_inode). | |
38 | * - If the page is not uptodate, there will be no pending write | |
39 | * requests, and no process will be in nfs_update_page. | |
40 | * | |
41 | * FIXME: Interaction with the vmscan routines is not optimal yet. | |
42 | * Either vmscan must be made nfs-savvy, or we need a different page | |
43 | * reclaim concept that supports something like FS-independent | |
44 | * buffer_heads with a b_ops-> field. | |
45 | * | |
46 | * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> | |
47 | */ | |
48 | ||
1da177e4 LT |
49 | #include <linux/types.h> |
50 | #include <linux/slab.h> | |
51 | #include <linux/mm.h> | |
52 | #include <linux/pagemap.h> | |
53 | #include <linux/file.h> | |
1da177e4 LT |
54 | #include <linux/writeback.h> |
55 | ||
56 | #include <linux/sunrpc/clnt.h> | |
57 | #include <linux/nfs_fs.h> | |
58 | #include <linux/nfs_mount.h> | |
59 | #include <linux/nfs_page.h> | |
3fcfab16 AM |
60 | #include <linux/backing-dev.h> |
61 | ||
1da177e4 LT |
62 | #include <asm/uaccess.h> |
63 | #include <linux/smp_lock.h> | |
64 | ||
65 | #include "delegation.h" | |
49a70f27 | 66 | #include "internal.h" |
91d5b470 | 67 | #include "iostat.h" |
1da177e4 LT |
68 | |
69 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
70 | ||
71 | #define MIN_POOL_WRITE (32) | |
72 | #define MIN_POOL_COMMIT (4) | |
73 | ||
74 | /* | |
75 | * Local function declarations | |
76 | */ | |
77 | static struct nfs_page * nfs_update_request(struct nfs_open_context*, | |
1da177e4 LT |
78 | struct page *, |
79 | unsigned int, unsigned int); | |
1da177e4 LT |
80 | static int nfs_wait_on_write_congestion(struct address_space *, int); |
81 | static int nfs_wait_on_requests(struct inode *, unsigned long, unsigned int); | |
3f442547 | 82 | static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how); |
1c75950b | 83 | static int nfs_wb_page_priority(struct inode *inode, struct page *page, int how); |
788e7a89 TM |
84 | static const struct rpc_call_ops nfs_write_partial_ops; |
85 | static const struct rpc_call_ops nfs_write_full_ops; | |
86 | static const struct rpc_call_ops nfs_commit_ops; | |
1da177e4 LT |
87 | |
88 | static kmem_cache_t *nfs_wdata_cachep; | |
3feb2d49 | 89 | static mempool_t *nfs_wdata_mempool; |
1da177e4 LT |
90 | static mempool_t *nfs_commit_mempool; |
91 | ||
92 | static DECLARE_WAIT_QUEUE_HEAD(nfs_write_congestion); | |
93 | ||
e9f7bee1 | 94 | struct nfs_write_data *nfs_commit_alloc(void) |
1da177e4 LT |
95 | { |
96 | struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, SLAB_NOFS); | |
40859d7e | 97 | |
1da177e4 LT |
98 | if (p) { |
99 | memset(p, 0, sizeof(*p)); | |
100 | INIT_LIST_HEAD(&p->pages); | |
101 | } | |
102 | return p; | |
103 | } | |
104 | ||
8aca67f0 | 105 | void nfs_commit_rcu_free(struct rcu_head *head) |
1da177e4 | 106 | { |
8aca67f0 | 107 | struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); |
40859d7e CL |
108 | if (p && (p->pagevec != &p->page_array[0])) |
109 | kfree(p->pagevec); | |
1da177e4 LT |
110 | mempool_free(p, nfs_commit_mempool); |
111 | } | |
112 | ||
8aca67f0 TM |
113 | void nfs_commit_free(struct nfs_write_data *wdata) |
114 | { | |
115 | call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free); | |
116 | } | |
117 | ||
e9f7bee1 | 118 | struct nfs_write_data *nfs_writedata_alloc(size_t len) |
3feb2d49 | 119 | { |
e9f7bee1 | 120 | unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
3feb2d49 TM |
121 | struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, SLAB_NOFS); |
122 | ||
123 | if (p) { | |
124 | memset(p, 0, sizeof(*p)); | |
125 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 126 | p->npages = pagecount; |
0d0b5cb3 CL |
127 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
128 | p->pagevec = p->page_array; | |
3feb2d49 | 129 | else { |
0d0b5cb3 CL |
130 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
131 | if (!p->pagevec) { | |
3feb2d49 TM |
132 | mempool_free(p, nfs_wdata_mempool); |
133 | p = NULL; | |
134 | } | |
135 | } | |
136 | } | |
137 | return p; | |
138 | } | |
139 | ||
8aca67f0 | 140 | static void nfs_writedata_rcu_free(struct rcu_head *head) |
3feb2d49 | 141 | { |
8aca67f0 | 142 | struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); |
3feb2d49 TM |
143 | if (p && (p->pagevec != &p->page_array[0])) |
144 | kfree(p->pagevec); | |
145 | mempool_free(p, nfs_wdata_mempool); | |
146 | } | |
147 | ||
8aca67f0 TM |
148 | static void nfs_writedata_free(struct nfs_write_data *wdata) |
149 | { | |
150 | call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free); | |
151 | } | |
152 | ||
963d8fe5 | 153 | void nfs_writedata_release(void *wdata) |
1da177e4 | 154 | { |
1da177e4 LT |
155 | nfs_writedata_free(wdata); |
156 | } | |
157 | ||
277459d2 TM |
158 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
159 | { | |
160 | struct nfs_page *req = NULL; | |
161 | ||
162 | if (PagePrivate(page)) { | |
163 | req = (struct nfs_page *)page_private(page); | |
164 | if (req != NULL) | |
165 | atomic_inc(&req->wb_count); | |
166 | } | |
167 | return req; | |
168 | } | |
169 | ||
170 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
171 | { | |
172 | struct nfs_page *req = NULL; | |
173 | spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock; | |
174 | ||
175 | spin_lock(req_lock); | |
176 | req = nfs_page_find_request_locked(page); | |
177 | spin_unlock(req_lock); | |
178 | return req; | |
179 | } | |
180 | ||
1da177e4 LT |
181 | /* Adjust the file length if we're writing beyond the end */ |
182 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
183 | { | |
184 | struct inode *inode = page->mapping->host; | |
185 | loff_t end, i_size = i_size_read(inode); | |
186 | unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
187 | ||
188 | if (i_size > 0 && page->index < end_index) | |
189 | return; | |
190 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); | |
191 | if (i_size >= end) | |
192 | return; | |
91d5b470 | 193 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
1da177e4 LT |
194 | i_size_write(inode, end); |
195 | } | |
196 | ||
197 | /* We can set the PG_uptodate flag if we see that a write request | |
198 | * covers the full page. | |
199 | */ | |
200 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
201 | { | |
1da177e4 LT |
202 | if (PageUptodate(page)) |
203 | return; | |
204 | if (base != 0) | |
205 | return; | |
49a70f27 | 206 | if (count != nfs_page_length(page)) |
1da177e4 | 207 | return; |
49a70f27 | 208 | if (count != PAGE_CACHE_SIZE) |
1da177e4 | 209 | memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count); |
49a70f27 | 210 | SetPageUptodate(page); |
1da177e4 LT |
211 | } |
212 | ||
e21195a7 | 213 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
1da177e4 LT |
214 | unsigned int offset, unsigned int count) |
215 | { | |
216 | struct nfs_page *req; | |
e21195a7 | 217 | int ret; |
1da177e4 | 218 | |
e21195a7 TM |
219 | for (;;) { |
220 | req = nfs_update_request(ctx, page, offset, count); | |
221 | if (!IS_ERR(req)) | |
222 | break; | |
223 | ret = PTR_ERR(req); | |
224 | if (ret != -EBUSY) | |
225 | return ret; | |
226 | ret = nfs_wb_page(page->mapping->host, page); | |
227 | if (ret != 0) | |
228 | return ret; | |
229 | } | |
1da177e4 LT |
230 | /* Update file length */ |
231 | nfs_grow_file(page, offset, count); | |
232 | /* Set the PG_uptodate flag? */ | |
233 | nfs_mark_uptodate(page, offset, count); | |
234 | nfs_unlock_request(req); | |
abd3e641 | 235 | return 0; |
1da177e4 LT |
236 | } |
237 | ||
238 | static int wb_priority(struct writeback_control *wbc) | |
239 | { | |
240 | if (wbc->for_reclaim) | |
241 | return FLUSH_HIGHPRI; | |
242 | if (wbc->for_kupdate) | |
243 | return FLUSH_LOWPRI; | |
244 | return 0; | |
245 | } | |
246 | ||
247 | /* | |
248 | * Write an mmapped page to the server. | |
249 | */ | |
250 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
251 | { | |
252 | struct nfs_open_context *ctx; | |
253 | struct inode *inode = page->mapping->host; | |
49a70f27 | 254 | unsigned offset; |
1da177e4 LT |
255 | int err; |
256 | ||
91d5b470 CL |
257 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
258 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
259 | ||
1da177e4 | 260 | /* Ensure we've flushed out any previous writes */ |
e21195a7 | 261 | nfs_wb_page_priority(inode, page, wb_priority(wbc)); |
1da177e4 | 262 | |
49a70f27 TM |
263 | err = 0; |
264 | offset = nfs_page_length(page); | |
265 | if (!offset) | |
1da177e4 | 266 | goto out; |
49a70f27 | 267 | |
d530838b | 268 | ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE); |
1da177e4 LT |
269 | if (ctx == NULL) { |
270 | err = -EBADF; | |
271 | goto out; | |
272 | } | |
200baa21 | 273 | err = nfs_writepage_setup(ctx, page, 0, offset); |
1da177e4 | 274 | put_nfs_open_context(ctx); |
200baa21 | 275 | |
1da177e4 | 276 | out: |
200baa21 TM |
277 | if (!wbc->for_writepages) |
278 | nfs_flush_mapping(page->mapping, wbc, wb_priority(wbc)); | |
1da177e4 | 279 | unlock_page(page); |
1da177e4 LT |
280 | return err; |
281 | } | |
282 | ||
283 | /* | |
284 | * Note: causes nfs_update_request() to block on the assumption | |
285 | * that the writeback is generated due to memory pressure. | |
286 | */ | |
287 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) | |
288 | { | |
289 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
290 | struct inode *inode = mapping->host; | |
291 | int err; | |
292 | ||
91d5b470 CL |
293 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
294 | ||
1da177e4 LT |
295 | err = generic_writepages(mapping, wbc); |
296 | if (err) | |
297 | return err; | |
298 | while (test_and_set_bit(BDI_write_congested, &bdi->state) != 0) { | |
299 | if (wbc->nonblocking) | |
300 | return 0; | |
301 | nfs_wait_on_write_congestion(mapping, 0); | |
302 | } | |
28c6925f | 303 | err = nfs_flush_mapping(mapping, wbc, wb_priority(wbc)); |
1da177e4 LT |
304 | if (err < 0) |
305 | goto out; | |
91d5b470 | 306 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, err); |
1da177e4 LT |
307 | if (!wbc->nonblocking && wbc->sync_mode == WB_SYNC_ALL) { |
308 | err = nfs_wait_on_requests(inode, 0, 0); | |
309 | if (err < 0) | |
310 | goto out; | |
311 | } | |
3da28eb1 | 312 | err = nfs_commit_inode(inode, wb_priority(wbc)); |
3f442547 | 313 | if (err > 0) |
1da177e4 | 314 | err = 0; |
1da177e4 LT |
315 | out: |
316 | clear_bit(BDI_write_congested, &bdi->state); | |
317 | wake_up_all(&nfs_write_congestion); | |
3fcfab16 | 318 | congestion_end(WRITE); |
1da177e4 LT |
319 | return err; |
320 | } | |
321 | ||
322 | /* | |
323 | * Insert a write request into an inode | |
324 | */ | |
325 | static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) | |
326 | { | |
327 | struct nfs_inode *nfsi = NFS_I(inode); | |
328 | int error; | |
329 | ||
330 | error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); | |
331 | BUG_ON(error == -EEXIST); | |
332 | if (error) | |
333 | return error; | |
334 | if (!nfsi->npages) { | |
335 | igrab(inode); | |
336 | nfs_begin_data_update(inode); | |
337 | if (nfs_have_delegation(inode, FMODE_WRITE)) | |
338 | nfsi->change_attr++; | |
339 | } | |
deb7d638 | 340 | SetPagePrivate(req->wb_page); |
277459d2 | 341 | set_page_private(req->wb_page, (unsigned long)req); |
1da177e4 LT |
342 | nfsi->npages++; |
343 | atomic_inc(&req->wb_count); | |
344 | return 0; | |
345 | } | |
346 | ||
347 | /* | |
348 | * Insert a write request into an inode | |
349 | */ | |
350 | static void nfs_inode_remove_request(struct nfs_page *req) | |
351 | { | |
352 | struct inode *inode = req->wb_context->dentry->d_inode; | |
353 | struct nfs_inode *nfsi = NFS_I(inode); | |
354 | ||
355 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
356 | ||
357 | spin_lock(&nfsi->req_lock); | |
277459d2 | 358 | set_page_private(req->wb_page, 0); |
deb7d638 | 359 | ClearPagePrivate(req->wb_page); |
1da177e4 LT |
360 | radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); |
361 | nfsi->npages--; | |
362 | if (!nfsi->npages) { | |
363 | spin_unlock(&nfsi->req_lock); | |
951a143b | 364 | nfs_end_data_update(inode); |
1da177e4 LT |
365 | iput(inode); |
366 | } else | |
367 | spin_unlock(&nfsi->req_lock); | |
368 | nfs_clear_request(req); | |
369 | nfs_release_request(req); | |
370 | } | |
371 | ||
1da177e4 LT |
372 | /* |
373 | * Add a request to the inode's dirty list. | |
374 | */ | |
375 | static void | |
376 | nfs_mark_request_dirty(struct nfs_page *req) | |
377 | { | |
378 | struct inode *inode = req->wb_context->dentry->d_inode; | |
379 | struct nfs_inode *nfsi = NFS_I(inode); | |
380 | ||
381 | spin_lock(&nfsi->req_lock); | |
3da28eb1 TM |
382 | radix_tree_tag_set(&nfsi->nfs_page_tree, |
383 | req->wb_index, NFS_PAGE_TAG_DIRTY); | |
1da177e4 LT |
384 | nfs_list_add_request(req, &nfsi->dirty); |
385 | nfsi->ndirty++; | |
386 | spin_unlock(&nfsi->req_lock); | |
b1e7a8fd | 387 | inc_zone_page_state(req->wb_page, NR_FILE_DIRTY); |
1da177e4 LT |
388 | mark_inode_dirty(inode); |
389 | } | |
390 | ||
391 | /* | |
392 | * Check if a request is dirty | |
393 | */ | |
394 | static inline int | |
395 | nfs_dirty_request(struct nfs_page *req) | |
396 | { | |
397 | struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode); | |
398 | return !list_empty(&req->wb_list) && req->wb_list_head == &nfsi->dirty; | |
399 | } | |
400 | ||
401 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
402 | /* | |
403 | * Add a request to the inode's commit list. | |
404 | */ | |
405 | static void | |
406 | nfs_mark_request_commit(struct nfs_page *req) | |
407 | { | |
408 | struct inode *inode = req->wb_context->dentry->d_inode; | |
409 | struct nfs_inode *nfsi = NFS_I(inode); | |
410 | ||
411 | spin_lock(&nfsi->req_lock); | |
412 | nfs_list_add_request(req, &nfsi->commit); | |
413 | nfsi->ncommit++; | |
414 | spin_unlock(&nfsi->req_lock); | |
fd39fc85 | 415 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
1da177e4 LT |
416 | mark_inode_dirty(inode); |
417 | } | |
418 | #endif | |
419 | ||
420 | /* | |
421 | * Wait for a request to complete. | |
422 | * | |
423 | * Interruptible by signals only if mounted with intr flag. | |
424 | */ | |
c42de9dd | 425 | static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages) |
1da177e4 LT |
426 | { |
427 | struct nfs_inode *nfsi = NFS_I(inode); | |
428 | struct nfs_page *req; | |
429 | unsigned long idx_end, next; | |
430 | unsigned int res = 0; | |
431 | int error; | |
432 | ||
433 | if (npages == 0) | |
434 | idx_end = ~0; | |
435 | else | |
436 | idx_end = idx_start + npages - 1; | |
437 | ||
1da177e4 | 438 | next = idx_start; |
c6a556b8 | 439 | while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) { |
1da177e4 LT |
440 | if (req->wb_index > idx_end) |
441 | break; | |
442 | ||
443 | next = req->wb_index + 1; | |
c6a556b8 | 444 | BUG_ON(!NFS_WBACK_BUSY(req)); |
1da177e4 LT |
445 | |
446 | atomic_inc(&req->wb_count); | |
447 | spin_unlock(&nfsi->req_lock); | |
448 | error = nfs_wait_on_request(req); | |
449 | nfs_release_request(req); | |
c42de9dd | 450 | spin_lock(&nfsi->req_lock); |
1da177e4 LT |
451 | if (error < 0) |
452 | return error; | |
1da177e4 LT |
453 | res++; |
454 | } | |
1da177e4 LT |
455 | return res; |
456 | } | |
457 | ||
c42de9dd TM |
458 | static int nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, unsigned int npages) |
459 | { | |
460 | struct nfs_inode *nfsi = NFS_I(inode); | |
461 | int ret; | |
462 | ||
463 | spin_lock(&nfsi->req_lock); | |
464 | ret = nfs_wait_on_requests_locked(inode, idx_start, npages); | |
465 | spin_unlock(&nfsi->req_lock); | |
466 | return ret; | |
467 | } | |
468 | ||
83715ad5 | 469 | static void nfs_cancel_dirty_list(struct list_head *head) |
d2ccddf0 TM |
470 | { |
471 | struct nfs_page *req; | |
472 | while(!list_empty(head)) { | |
473 | req = nfs_list_entry(head->next); | |
474 | nfs_list_remove_request(req); | |
475 | nfs_inode_remove_request(req); | |
476 | nfs_clear_page_writeback(req); | |
477 | } | |
478 | } | |
479 | ||
83715ad5 TM |
480 | static void nfs_cancel_commit_list(struct list_head *head) |
481 | { | |
482 | struct nfs_page *req; | |
483 | ||
484 | while(!list_empty(head)) { | |
485 | req = nfs_list_entry(head->next); | |
b6dff26a | 486 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
83715ad5 TM |
487 | nfs_list_remove_request(req); |
488 | nfs_inode_remove_request(req); | |
b6dff26a | 489 | nfs_unlock_request(req); |
83715ad5 TM |
490 | } |
491 | } | |
492 | ||
1da177e4 LT |
493 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
494 | /* | |
495 | * nfs_scan_commit - Scan an inode for commit requests | |
496 | * @inode: NFS inode to scan | |
497 | * @dst: destination list | |
498 | * @idx_start: lower bound of page->index to scan. | |
499 | * @npages: idx_start + npages sets the upper bound to scan. | |
500 | * | |
501 | * Moves requests from the inode's 'commit' request list. | |
502 | * The requests are *not* checked to ensure that they form a contiguous set. | |
503 | */ | |
504 | static int | |
505 | nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) | |
506 | { | |
507 | struct nfs_inode *nfsi = NFS_I(inode); | |
3da28eb1 TM |
508 | int res = 0; |
509 | ||
510 | if (nfsi->ncommit != 0) { | |
d2ccddf0 | 511 | res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages); |
3da28eb1 TM |
512 | nfsi->ncommit -= res; |
513 | if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit)) | |
514 | printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n"); | |
515 | } | |
1da177e4 LT |
516 | return res; |
517 | } | |
c42de9dd TM |
518 | #else |
519 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) | |
520 | { | |
521 | return 0; | |
522 | } | |
1da177e4 LT |
523 | #endif |
524 | ||
525 | static int nfs_wait_on_write_congestion(struct address_space *mapping, int intr) | |
526 | { | |
527 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
528 | DEFINE_WAIT(wait); | |
529 | int ret = 0; | |
530 | ||
531 | might_sleep(); | |
532 | ||
533 | if (!bdi_write_congested(bdi)) | |
534 | return 0; | |
91d5b470 CL |
535 | |
536 | nfs_inc_stats(mapping->host, NFSIOS_CONGESTIONWAIT); | |
537 | ||
1da177e4 LT |
538 | if (intr) { |
539 | struct rpc_clnt *clnt = NFS_CLIENT(mapping->host); | |
540 | sigset_t oldset; | |
541 | ||
542 | rpc_clnt_sigmask(clnt, &oldset); | |
543 | prepare_to_wait(&nfs_write_congestion, &wait, TASK_INTERRUPTIBLE); | |
544 | if (bdi_write_congested(bdi)) { | |
545 | if (signalled()) | |
546 | ret = -ERESTARTSYS; | |
547 | else | |
548 | schedule(); | |
549 | } | |
550 | rpc_clnt_sigunmask(clnt, &oldset); | |
551 | } else { | |
552 | prepare_to_wait(&nfs_write_congestion, &wait, TASK_UNINTERRUPTIBLE); | |
553 | if (bdi_write_congested(bdi)) | |
554 | schedule(); | |
555 | } | |
556 | finish_wait(&nfs_write_congestion, &wait); | |
557 | return ret; | |
558 | } | |
559 | ||
560 | ||
561 | /* | |
562 | * Try to update any existing write request, or create one if there is none. | |
563 | * In order to match, the request's credentials must match those of | |
564 | * the calling process. | |
565 | * | |
566 | * Note: Should always be called with the Page Lock held! | |
567 | */ | |
568 | static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, | |
e21195a7 | 569 | struct page *page, unsigned int offset, unsigned int bytes) |
1da177e4 | 570 | { |
e21195a7 | 571 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
572 | struct nfs_inode *nfsi = NFS_I(inode); |
573 | struct nfs_page *req, *new = NULL; | |
574 | unsigned long rqend, end; | |
575 | ||
576 | end = offset + bytes; | |
577 | ||
e21195a7 | 578 | if (nfs_wait_on_write_congestion(page->mapping, NFS_SERVER(inode)->flags & NFS_MOUNT_INTR)) |
1da177e4 LT |
579 | return ERR_PTR(-ERESTARTSYS); |
580 | for (;;) { | |
581 | /* Loop over all inode entries and see if we find | |
582 | * A request for the page we wish to update | |
583 | */ | |
584 | spin_lock(&nfsi->req_lock); | |
277459d2 | 585 | req = nfs_page_find_request_locked(page); |
1da177e4 LT |
586 | if (req) { |
587 | if (!nfs_lock_request_dontget(req)) { | |
588 | int error; | |
277459d2 | 589 | |
1da177e4 LT |
590 | spin_unlock(&nfsi->req_lock); |
591 | error = nfs_wait_on_request(req); | |
592 | nfs_release_request(req); | |
1dd594b2 NB |
593 | if (error < 0) { |
594 | if (new) | |
595 | nfs_release_request(new); | |
1da177e4 | 596 | return ERR_PTR(error); |
1dd594b2 | 597 | } |
1da177e4 LT |
598 | continue; |
599 | } | |
600 | spin_unlock(&nfsi->req_lock); | |
601 | if (new) | |
602 | nfs_release_request(new); | |
603 | break; | |
604 | } | |
605 | ||
606 | if (new) { | |
607 | int error; | |
608 | nfs_lock_request_dontget(new); | |
609 | error = nfs_inode_add_request(inode, new); | |
610 | if (error) { | |
611 | spin_unlock(&nfsi->req_lock); | |
612 | nfs_unlock_request(new); | |
613 | return ERR_PTR(error); | |
614 | } | |
615 | spin_unlock(&nfsi->req_lock); | |
616 | nfs_mark_request_dirty(new); | |
617 | return new; | |
618 | } | |
619 | spin_unlock(&nfsi->req_lock); | |
620 | ||
621 | new = nfs_create_request(ctx, inode, page, offset, bytes); | |
622 | if (IS_ERR(new)) | |
623 | return new; | |
624 | } | |
625 | ||
626 | /* We have a request for our page. | |
627 | * If the creds don't match, or the | |
628 | * page addresses don't match, | |
629 | * tell the caller to wait on the conflicting | |
630 | * request. | |
631 | */ | |
632 | rqend = req->wb_offset + req->wb_bytes; | |
633 | if (req->wb_context != ctx | |
634 | || req->wb_page != page | |
635 | || !nfs_dirty_request(req) | |
636 | || offset > rqend || end < req->wb_offset) { | |
637 | nfs_unlock_request(req); | |
638 | return ERR_PTR(-EBUSY); | |
639 | } | |
640 | ||
641 | /* Okay, the request matches. Update the region */ | |
642 | if (offset < req->wb_offset) { | |
643 | req->wb_offset = offset; | |
644 | req->wb_pgbase = offset; | |
645 | req->wb_bytes = rqend - req->wb_offset; | |
646 | } | |
647 | ||
648 | if (end > rqend) | |
649 | req->wb_bytes = end - req->wb_offset; | |
650 | ||
651 | return req; | |
652 | } | |
653 | ||
654 | int nfs_flush_incompatible(struct file *file, struct page *page) | |
655 | { | |
656 | struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; | |
1da177e4 LT |
657 | struct nfs_page *req; |
658 | int status = 0; | |
659 | /* | |
660 | * Look for a request corresponding to this page. If there | |
661 | * is one, and it belongs to another file, we flush it out | |
662 | * before we try to copy anything into the page. Do this | |
663 | * due to the lack of an ACCESS-type call in NFSv2. | |
664 | * Also do the same if we find a request from an existing | |
665 | * dropped page. | |
666 | */ | |
277459d2 TM |
667 | req = nfs_page_find_request(page); |
668 | if (req != NULL) { | |
669 | int do_flush = req->wb_page != page || req->wb_context != ctx; | |
670 | ||
1da177e4 | 671 | nfs_release_request(req); |
277459d2 TM |
672 | if (do_flush) |
673 | status = nfs_wb_page(page->mapping->host, page); | |
1da177e4 LT |
674 | } |
675 | return (status < 0) ? status : 0; | |
676 | } | |
677 | ||
678 | /* | |
679 | * Update and possibly write a cached page of an NFS file. | |
680 | * | |
681 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
682 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
683 | */ | |
684 | int nfs_updatepage(struct file *file, struct page *page, | |
685 | unsigned int offset, unsigned int count) | |
686 | { | |
687 | struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; | |
1da177e4 | 688 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
689 | int status = 0; |
690 | ||
91d5b470 CL |
691 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
692 | ||
1da177e4 | 693 | dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", |
0bbacc40 CL |
694 | file->f_dentry->d_parent->d_name.name, |
695 | file->f_dentry->d_name.name, count, | |
696 | (long long)(page_offset(page) +offset)); | |
1da177e4 | 697 | |
1da177e4 LT |
698 | /* If we're not using byte range locks, and we know the page |
699 | * is entirely in cache, it may be more efficient to avoid | |
700 | * fragmenting write requests. | |
701 | */ | |
ab0a3dbe | 702 | if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { |
49a70f27 | 703 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 704 | offset = 0; |
1da177e4 LT |
705 | } |
706 | ||
e21195a7 | 707 | status = nfs_writepage_setup(ctx, page, offset, count); |
1da177e4 | 708 | |
1da177e4 LT |
709 | dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", |
710 | status, (long long)i_size_read(inode)); | |
711 | if (status < 0) | |
712 | ClearPageUptodate(page); | |
713 | return status; | |
714 | } | |
715 | ||
716 | static void nfs_writepage_release(struct nfs_page *req) | |
717 | { | |
718 | end_page_writeback(req->wb_page); | |
719 | ||
720 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
721 | if (!PageError(req->wb_page)) { | |
722 | if (NFS_NEED_RESCHED(req)) { | |
723 | nfs_mark_request_dirty(req); | |
724 | goto out; | |
725 | } else if (NFS_NEED_COMMIT(req)) { | |
726 | nfs_mark_request_commit(req); | |
727 | goto out; | |
728 | } | |
729 | } | |
730 | nfs_inode_remove_request(req); | |
731 | ||
732 | out: | |
733 | nfs_clear_commit(req); | |
734 | nfs_clear_reschedule(req); | |
735 | #else | |
736 | nfs_inode_remove_request(req); | |
737 | #endif | |
c6a556b8 | 738 | nfs_clear_page_writeback(req); |
1da177e4 LT |
739 | } |
740 | ||
741 | static inline int flush_task_priority(int how) | |
742 | { | |
743 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
744 | case FLUSH_HIGHPRI: | |
745 | return RPC_PRIORITY_HIGH; | |
746 | case FLUSH_LOWPRI: | |
747 | return RPC_PRIORITY_LOW; | |
748 | } | |
749 | return RPC_PRIORITY_NORMAL; | |
750 | } | |
751 | ||
752 | /* | |
753 | * Set up the argument/result storage required for the RPC call. | |
754 | */ | |
755 | static void nfs_write_rpcsetup(struct nfs_page *req, | |
756 | struct nfs_write_data *data, | |
788e7a89 | 757 | const struct rpc_call_ops *call_ops, |
1da177e4 LT |
758 | unsigned int count, unsigned int offset, |
759 | int how) | |
760 | { | |
1da177e4 | 761 | struct inode *inode; |
788e7a89 | 762 | int flags; |
1da177e4 LT |
763 | |
764 | /* Set up the RPC argument and reply structs | |
765 | * NB: take care not to mess about with data->commit et al. */ | |
766 | ||
767 | data->req = req; | |
768 | data->inode = inode = req->wb_context->dentry->d_inode; | |
769 | data->cred = req->wb_context->cred; | |
770 | ||
771 | data->args.fh = NFS_FH(inode); | |
772 | data->args.offset = req_offset(req) + offset; | |
773 | data->args.pgbase = req->wb_pgbase + offset; | |
774 | data->args.pages = data->pagevec; | |
775 | data->args.count = count; | |
776 | data->args.context = req->wb_context; | |
777 | ||
778 | data->res.fattr = &data->fattr; | |
779 | data->res.count = count; | |
780 | data->res.verf = &data->verf; | |
0e574af1 | 781 | nfs_fattr_init(&data->fattr); |
1da177e4 | 782 | |
788e7a89 TM |
783 | /* Set up the initial task struct. */ |
784 | flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; | |
785 | rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); | |
1da177e4 LT |
786 | NFS_PROTO(inode)->write_setup(data, how); |
787 | ||
788 | data->task.tk_priority = flush_task_priority(how); | |
789 | data->task.tk_cookie = (unsigned long)inode; | |
1da177e4 LT |
790 | |
791 | dprintk("NFS: %4d initiated write call (req %s/%Ld, %u bytes @ offset %Lu)\n", | |
0bbacc40 | 792 | data->task.tk_pid, |
1da177e4 LT |
793 | inode->i_sb->s_id, |
794 | (long long)NFS_FILEID(inode), | |
795 | count, | |
796 | (unsigned long long)data->args.offset); | |
797 | } | |
798 | ||
799 | static void nfs_execute_write(struct nfs_write_data *data) | |
800 | { | |
801 | struct rpc_clnt *clnt = NFS_CLIENT(data->inode); | |
802 | sigset_t oldset; | |
803 | ||
804 | rpc_clnt_sigmask(clnt, &oldset); | |
1da177e4 | 805 | rpc_execute(&data->task); |
1da177e4 LT |
806 | rpc_clnt_sigunmask(clnt, &oldset); |
807 | } | |
808 | ||
809 | /* | |
810 | * Generate multiple small requests to write out a single | |
811 | * contiguous dirty area on one page. | |
812 | */ | |
7d46a49f | 813 | static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
814 | { |
815 | struct nfs_page *req = nfs_list_entry(head->next); | |
816 | struct page *page = req->wb_page; | |
817 | struct nfs_write_data *data; | |
e9f7bee1 TM |
818 | size_t wsize = NFS_SERVER(inode)->wsize, nbytes; |
819 | unsigned int offset; | |
1da177e4 LT |
820 | int requests = 0; |
821 | LIST_HEAD(list); | |
822 | ||
823 | nfs_list_remove_request(req); | |
824 | ||
825 | nbytes = req->wb_bytes; | |
e9f7bee1 TM |
826 | do { |
827 | size_t len = min(nbytes, wsize); | |
828 | ||
829 | data = nfs_writedata_alloc(len); | |
1da177e4 LT |
830 | if (!data) |
831 | goto out_bad; | |
832 | list_add(&data->pages, &list); | |
833 | requests++; | |
e9f7bee1 TM |
834 | nbytes -= len; |
835 | } while (nbytes != 0); | |
1da177e4 LT |
836 | atomic_set(&req->wb_complete, requests); |
837 | ||
838 | ClearPageError(page); | |
bb713d6d | 839 | set_page_writeback(page); |
1da177e4 LT |
840 | offset = 0; |
841 | nbytes = req->wb_bytes; | |
842 | do { | |
843 | data = list_entry(list.next, struct nfs_write_data, pages); | |
844 | list_del_init(&data->pages); | |
845 | ||
846 | data->pagevec[0] = page; | |
1da177e4 LT |
847 | |
848 | if (nbytes > wsize) { | |
788e7a89 TM |
849 | nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, |
850 | wsize, offset, how); | |
1da177e4 LT |
851 | offset += wsize; |
852 | nbytes -= wsize; | |
853 | } else { | |
788e7a89 TM |
854 | nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, |
855 | nbytes, offset, how); | |
1da177e4 LT |
856 | nbytes = 0; |
857 | } | |
858 | nfs_execute_write(data); | |
859 | } while (nbytes != 0); | |
860 | ||
861 | return 0; | |
862 | ||
863 | out_bad: | |
864 | while (!list_empty(&list)) { | |
865 | data = list_entry(list.next, struct nfs_write_data, pages); | |
866 | list_del(&data->pages); | |
8aca67f0 | 867 | nfs_writedata_release(data); |
1da177e4 LT |
868 | } |
869 | nfs_mark_request_dirty(req); | |
c6a556b8 | 870 | nfs_clear_page_writeback(req); |
1da177e4 LT |
871 | return -ENOMEM; |
872 | } | |
873 | ||
874 | /* | |
875 | * Create an RPC task for the given write request and kick it. | |
876 | * The page must have been locked by the caller. | |
877 | * | |
878 | * It may happen that the page we're passed is not marked dirty. | |
879 | * This is the case if nfs_updatepage detects a conflicting request | |
880 | * that has been written but not committed. | |
881 | */ | |
7d46a49f | 882 | static int nfs_flush_one(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
883 | { |
884 | struct nfs_page *req; | |
885 | struct page **pages; | |
886 | struct nfs_write_data *data; | |
887 | unsigned int count; | |
888 | ||
e9f7bee1 | 889 | data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize); |
1da177e4 LT |
890 | if (!data) |
891 | goto out_bad; | |
892 | ||
893 | pages = data->pagevec; | |
894 | count = 0; | |
895 | while (!list_empty(head)) { | |
896 | req = nfs_list_entry(head->next); | |
897 | nfs_list_remove_request(req); | |
898 | nfs_list_add_request(req, &data->pages); | |
899 | ClearPageError(req->wb_page); | |
bb713d6d | 900 | set_page_writeback(req->wb_page); |
1da177e4 LT |
901 | *pages++ = req->wb_page; |
902 | count += req->wb_bytes; | |
903 | } | |
904 | req = nfs_list_entry(data->pages.next); | |
905 | ||
1da177e4 | 906 | /* Set up the argument struct */ |
788e7a89 | 907 | nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); |
1da177e4 LT |
908 | |
909 | nfs_execute_write(data); | |
910 | return 0; | |
911 | out_bad: | |
912 | while (!list_empty(head)) { | |
913 | struct nfs_page *req = nfs_list_entry(head->next); | |
914 | nfs_list_remove_request(req); | |
915 | nfs_mark_request_dirty(req); | |
c6a556b8 | 916 | nfs_clear_page_writeback(req); |
1da177e4 LT |
917 | } |
918 | return -ENOMEM; | |
919 | } | |
920 | ||
7d46a49f | 921 | static int nfs_flush_list(struct inode *inode, struct list_head *head, int npages, int how) |
1da177e4 LT |
922 | { |
923 | LIST_HEAD(one_request); | |
7d46a49f TM |
924 | int (*flush_one)(struct inode *, struct list_head *, int); |
925 | struct nfs_page *req; | |
926 | int wpages = NFS_SERVER(inode)->wpages; | |
927 | int wsize = NFS_SERVER(inode)->wsize; | |
928 | int error; | |
1da177e4 | 929 | |
7d46a49f TM |
930 | flush_one = nfs_flush_one; |
931 | if (wsize < PAGE_CACHE_SIZE) | |
932 | flush_one = nfs_flush_multi; | |
933 | /* For single writes, FLUSH_STABLE is more efficient */ | |
934 | if (npages <= wpages && npages == NFS_I(inode)->npages | |
935 | && nfs_list_entry(head->next)->wb_bytes <= wsize) | |
936 | how |= FLUSH_STABLE; | |
937 | ||
938 | do { | |
939 | nfs_coalesce_requests(head, &one_request, wpages); | |
1da177e4 | 940 | req = nfs_list_entry(one_request.next); |
7d46a49f | 941 | error = flush_one(inode, &one_request, how); |
1da177e4 | 942 | if (error < 0) |
7d46a49f TM |
943 | goto out_err; |
944 | } while (!list_empty(head)); | |
945 | return 0; | |
946 | out_err: | |
1da177e4 LT |
947 | while (!list_empty(head)) { |
948 | req = nfs_list_entry(head->next); | |
949 | nfs_list_remove_request(req); | |
950 | nfs_mark_request_dirty(req); | |
c6a556b8 | 951 | nfs_clear_page_writeback(req); |
1da177e4 LT |
952 | } |
953 | return error; | |
954 | } | |
955 | ||
956 | /* | |
957 | * Handle a write reply that flushed part of a page. | |
958 | */ | |
788e7a89 | 959 | static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 960 | { |
788e7a89 | 961 | struct nfs_write_data *data = calldata; |
1da177e4 LT |
962 | struct nfs_page *req = data->req; |
963 | struct page *page = req->wb_page; | |
964 | ||
965 | dprintk("NFS: write (%s/%Ld %d@%Ld)", | |
966 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
967 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
968 | req->wb_bytes, | |
969 | (long long)req_offset(req)); | |
970 | ||
788e7a89 TM |
971 | if (nfs_writeback_done(task, data) != 0) |
972 | return; | |
973 | ||
974 | if (task->tk_status < 0) { | |
1da177e4 LT |
975 | ClearPageUptodate(page); |
976 | SetPageError(page); | |
788e7a89 TM |
977 | req->wb_context->error = task->tk_status; |
978 | dprintk(", error = %d\n", task->tk_status); | |
1da177e4 LT |
979 | } else { |
980 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
981 | if (data->verf.committed < NFS_FILE_SYNC) { | |
982 | if (!NFS_NEED_COMMIT(req)) { | |
983 | nfs_defer_commit(req); | |
984 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
985 | dprintk(" defer commit\n"); | |
986 | } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { | |
987 | nfs_defer_reschedule(req); | |
988 | dprintk(" server reboot detected\n"); | |
989 | } | |
990 | } else | |
991 | #endif | |
992 | dprintk(" OK\n"); | |
993 | } | |
994 | ||
995 | if (atomic_dec_and_test(&req->wb_complete)) | |
996 | nfs_writepage_release(req); | |
997 | } | |
998 | ||
788e7a89 TM |
999 | static const struct rpc_call_ops nfs_write_partial_ops = { |
1000 | .rpc_call_done = nfs_writeback_done_partial, | |
1001 | .rpc_release = nfs_writedata_release, | |
1002 | }; | |
1003 | ||
1da177e4 LT |
1004 | /* |
1005 | * Handle a write reply that flushes a whole page. | |
1006 | * | |
1007 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1008 | * writebacks since the page->count is kept > 1 for as long | |
1009 | * as the page has a write request pending. | |
1010 | */ | |
788e7a89 | 1011 | static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) |
1da177e4 | 1012 | { |
788e7a89 | 1013 | struct nfs_write_data *data = calldata; |
1da177e4 LT |
1014 | struct nfs_page *req; |
1015 | struct page *page; | |
1016 | ||
788e7a89 TM |
1017 | if (nfs_writeback_done(task, data) != 0) |
1018 | return; | |
1019 | ||
1da177e4 LT |
1020 | /* Update attributes as result of writeback. */ |
1021 | while (!list_empty(&data->pages)) { | |
1022 | req = nfs_list_entry(data->pages.next); | |
1023 | nfs_list_remove_request(req); | |
1024 | page = req->wb_page; | |
1025 | ||
1026 | dprintk("NFS: write (%s/%Ld %d@%Ld)", | |
1027 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
1028 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1029 | req->wb_bytes, | |
1030 | (long long)req_offset(req)); | |
1031 | ||
788e7a89 | 1032 | if (task->tk_status < 0) { |
1da177e4 LT |
1033 | ClearPageUptodate(page); |
1034 | SetPageError(page); | |
788e7a89 | 1035 | req->wb_context->error = task->tk_status; |
1da177e4 LT |
1036 | end_page_writeback(page); |
1037 | nfs_inode_remove_request(req); | |
788e7a89 | 1038 | dprintk(", error = %d\n", task->tk_status); |
1da177e4 LT |
1039 | goto next; |
1040 | } | |
1041 | end_page_writeback(page); | |
1042 | ||
1043 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
1044 | if (data->args.stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) { | |
1045 | nfs_inode_remove_request(req); | |
1046 | dprintk(" OK\n"); | |
1047 | goto next; | |
1048 | } | |
1049 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); | |
1050 | nfs_mark_request_commit(req); | |
1051 | dprintk(" marked for commit\n"); | |
1052 | #else | |
1053 | nfs_inode_remove_request(req); | |
1054 | #endif | |
1055 | next: | |
c6a556b8 | 1056 | nfs_clear_page_writeback(req); |
1da177e4 LT |
1057 | } |
1058 | } | |
1059 | ||
788e7a89 TM |
1060 | static const struct rpc_call_ops nfs_write_full_ops = { |
1061 | .rpc_call_done = nfs_writeback_done_full, | |
1062 | .rpc_release = nfs_writedata_release, | |
1063 | }; | |
1064 | ||
1065 | ||
1da177e4 LT |
1066 | /* |
1067 | * This function is called when the WRITE call is complete. | |
1068 | */ | |
462d5b32 | 1069 | int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1070 | { |
1da177e4 LT |
1071 | struct nfs_writeargs *argp = &data->args; |
1072 | struct nfs_writeres *resp = &data->res; | |
788e7a89 | 1073 | int status; |
1da177e4 LT |
1074 | |
1075 | dprintk("NFS: %4d nfs_writeback_done (status %d)\n", | |
1076 | task->tk_pid, task->tk_status); | |
1077 | ||
f551e44f CL |
1078 | /* |
1079 | * ->write_done will attempt to use post-op attributes to detect | |
1080 | * conflicting writes by other clients. A strict interpretation | |
1081 | * of close-to-open would allow us to continue caching even if | |
1082 | * another writer had changed the file, but some applications | |
1083 | * depend on tighter cache coherency when writing. | |
1084 | */ | |
788e7a89 TM |
1085 | status = NFS_PROTO(data->inode)->write_done(task, data); |
1086 | if (status != 0) | |
1087 | return status; | |
91d5b470 CL |
1088 | nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
1089 | ||
1da177e4 LT |
1090 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1091 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1092 | /* We tried a write call, but the server did not | |
1093 | * commit data to stable storage even though we | |
1094 | * requested it. | |
1095 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1096 | * the server reports NFS_DATA_SYNC, but performs | |
1097 | * NFS_FILE_SYNC. We therefore implement this checking | |
1098 | * as a dprintk() in order to avoid filling syslog. | |
1099 | */ | |
1100 | static unsigned long complain; | |
1101 | ||
1102 | if (time_before(complain, jiffies)) { | |
1103 | dprintk("NFS: faulty NFS server %s:" | |
1104 | " (committed = %d) != (stable = %d)\n", | |
54ceac45 | 1105 | NFS_SERVER(data->inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1106 | resp->verf->committed, argp->stable); |
1107 | complain = jiffies + 300 * HZ; | |
1108 | } | |
1109 | } | |
1110 | #endif | |
1111 | /* Is this a short write? */ | |
1112 | if (task->tk_status >= 0 && resp->count < argp->count) { | |
1113 | static unsigned long complain; | |
1114 | ||
91d5b470 CL |
1115 | nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); |
1116 | ||
1da177e4 LT |
1117 | /* Has the server at least made some progress? */ |
1118 | if (resp->count != 0) { | |
1119 | /* Was this an NFSv2 write or an NFSv3 stable write? */ | |
1120 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1121 | /* Resend from where the server left off */ | |
1122 | argp->offset += resp->count; | |
1123 | argp->pgbase += resp->count; | |
1124 | argp->count -= resp->count; | |
1125 | } else { | |
1126 | /* Resend as a stable write in order to avoid | |
1127 | * headaches in the case of a server crash. | |
1128 | */ | |
1129 | argp->stable = NFS_FILE_SYNC; | |
1130 | } | |
1131 | rpc_restart_call(task); | |
788e7a89 | 1132 | return -EAGAIN; |
1da177e4 LT |
1133 | } |
1134 | if (time_before(complain, jiffies)) { | |
1135 | printk(KERN_WARNING | |
1136 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1137 | argp->count); | |
1138 | complain = jiffies + 300 * HZ; | |
1139 | } | |
1140 | /* Can't do anything about it except throw an error. */ | |
1141 | task->tk_status = -EIO; | |
1142 | } | |
788e7a89 | 1143 | return 0; |
1da177e4 LT |
1144 | } |
1145 | ||
1146 | ||
1147 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
963d8fe5 | 1148 | void nfs_commit_release(void *wdata) |
1da177e4 | 1149 | { |
1da177e4 LT |
1150 | nfs_commit_free(wdata); |
1151 | } | |
1152 | ||
1153 | /* | |
1154 | * Set up the argument/result storage required for the RPC call. | |
1155 | */ | |
1156 | static void nfs_commit_rpcsetup(struct list_head *head, | |
788e7a89 TM |
1157 | struct nfs_write_data *data, |
1158 | int how) | |
1da177e4 | 1159 | { |
3da28eb1 | 1160 | struct nfs_page *first; |
1da177e4 | 1161 | struct inode *inode; |
788e7a89 | 1162 | int flags; |
1da177e4 LT |
1163 | |
1164 | /* Set up the RPC argument and reply structs | |
1165 | * NB: take care not to mess about with data->commit et al. */ | |
1166 | ||
1167 | list_splice_init(head, &data->pages); | |
1168 | first = nfs_list_entry(data->pages.next); | |
1da177e4 LT |
1169 | inode = first->wb_context->dentry->d_inode; |
1170 | ||
1da177e4 LT |
1171 | data->inode = inode; |
1172 | data->cred = first->wb_context->cred; | |
1173 | ||
1174 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1175 | /* Note: we always request a commit of the entire inode */ |
1176 | data->args.offset = 0; | |
1177 | data->args.count = 0; | |
1178 | data->res.count = 0; | |
1da177e4 LT |
1179 | data->res.fattr = &data->fattr; |
1180 | data->res.verf = &data->verf; | |
0e574af1 | 1181 | nfs_fattr_init(&data->fattr); |
788e7a89 TM |
1182 | |
1183 | /* Set up the initial task struct. */ | |
1184 | flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; | |
1185 | rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data); | |
1da177e4 LT |
1186 | NFS_PROTO(inode)->commit_setup(data, how); |
1187 | ||
1188 | data->task.tk_priority = flush_task_priority(how); | |
1189 | data->task.tk_cookie = (unsigned long)inode; | |
1da177e4 | 1190 | |
0bbacc40 | 1191 | dprintk("NFS: %4d initiated commit call\n", data->task.tk_pid); |
1da177e4 LT |
1192 | } |
1193 | ||
1194 | /* | |
1195 | * Commit dirty pages | |
1196 | */ | |
1197 | static int | |
40859d7e | 1198 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 LT |
1199 | { |
1200 | struct nfs_write_data *data; | |
1201 | struct nfs_page *req; | |
1202 | ||
e9f7bee1 | 1203 | data = nfs_commit_alloc(); |
1da177e4 LT |
1204 | |
1205 | if (!data) | |
1206 | goto out_bad; | |
1207 | ||
1208 | /* Set up the argument struct */ | |
1209 | nfs_commit_rpcsetup(head, data, how); | |
1210 | ||
1211 | nfs_execute_write(data); | |
1212 | return 0; | |
1213 | out_bad: | |
1214 | while (!list_empty(head)) { | |
1215 | req = nfs_list_entry(head->next); | |
1216 | nfs_list_remove_request(req); | |
1217 | nfs_mark_request_commit(req); | |
83715ad5 | 1218 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
5c2d97cb | 1219 | nfs_clear_page_writeback(req); |
1da177e4 LT |
1220 | } |
1221 | return -ENOMEM; | |
1222 | } | |
1223 | ||
1224 | /* | |
1225 | * COMMIT call returned | |
1226 | */ | |
788e7a89 | 1227 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1228 | { |
963d8fe5 | 1229 | struct nfs_write_data *data = calldata; |
1da177e4 | 1230 | struct nfs_page *req; |
1da177e4 LT |
1231 | |
1232 | dprintk("NFS: %4d nfs_commit_done (status %d)\n", | |
1233 | task->tk_pid, task->tk_status); | |
1234 | ||
788e7a89 TM |
1235 | /* Call the NFS version-specific code */ |
1236 | if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) | |
1237 | return; | |
1238 | ||
1da177e4 LT |
1239 | while (!list_empty(&data->pages)) { |
1240 | req = nfs_list_entry(data->pages.next); | |
1241 | nfs_list_remove_request(req); | |
fd39fc85 | 1242 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
1da177e4 LT |
1243 | |
1244 | dprintk("NFS: commit (%s/%Ld %d@%Ld)", | |
1245 | req->wb_context->dentry->d_inode->i_sb->s_id, | |
1246 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1247 | req->wb_bytes, | |
1248 | (long long)req_offset(req)); | |
1249 | if (task->tk_status < 0) { | |
1250 | req->wb_context->error = task->tk_status; | |
1251 | nfs_inode_remove_request(req); | |
1252 | dprintk(", error = %d\n", task->tk_status); | |
1253 | goto next; | |
1254 | } | |
1255 | ||
1256 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1257 | * returned by the server against all stored verfs. */ | |
1258 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1259 | /* We have a match */ | |
1260 | nfs_inode_remove_request(req); | |
1261 | dprintk(" OK\n"); | |
1262 | goto next; | |
1263 | } | |
1264 | /* We have a mismatch. Write the page again */ | |
1265 | dprintk(" mismatch\n"); | |
1266 | nfs_mark_request_dirty(req); | |
1267 | next: | |
c6a556b8 | 1268 | nfs_clear_page_writeback(req); |
1da177e4 | 1269 | } |
1da177e4 | 1270 | } |
788e7a89 TM |
1271 | |
1272 | static const struct rpc_call_ops nfs_commit_ops = { | |
1273 | .rpc_call_done = nfs_commit_done, | |
1274 | .rpc_release = nfs_commit_release, | |
1275 | }; | |
c42de9dd TM |
1276 | #else |
1277 | static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) | |
1278 | { | |
1279 | return 0; | |
1280 | } | |
1da177e4 LT |
1281 | #endif |
1282 | ||
3f442547 | 1283 | static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) |
1da177e4 | 1284 | { |
28c6925f | 1285 | struct nfs_inode *nfsi = NFS_I(mapping->host); |
1da177e4 | 1286 | LIST_HEAD(head); |
3f442547 | 1287 | long res; |
1da177e4 LT |
1288 | |
1289 | spin_lock(&nfsi->req_lock); | |
3f442547 | 1290 | res = nfs_scan_dirty(mapping, wbc, &head); |
1da177e4 | 1291 | spin_unlock(&nfsi->req_lock); |
ab0a3dbe | 1292 | if (res) { |
28c6925f | 1293 | int error = nfs_flush_list(mapping->host, &head, res, how); |
7d46a49f TM |
1294 | if (error < 0) |
1295 | return error; | |
ab0a3dbe | 1296 | } |
1da177e4 LT |
1297 | return res; |
1298 | } | |
1299 | ||
1300 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
3da28eb1 | 1301 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 LT |
1302 | { |
1303 | struct nfs_inode *nfsi = NFS_I(inode); | |
1304 | LIST_HEAD(head); | |
7d46a49f | 1305 | int res; |
1da177e4 LT |
1306 | |
1307 | spin_lock(&nfsi->req_lock); | |
3da28eb1 TM |
1308 | res = nfs_scan_commit(inode, &head, 0, 0); |
1309 | spin_unlock(&nfsi->req_lock); | |
1da177e4 | 1310 | if (res) { |
7d46a49f | 1311 | int error = nfs_commit_list(inode, &head, how); |
3da28eb1 TM |
1312 | if (error < 0) |
1313 | return error; | |
1314 | } | |
1da177e4 LT |
1315 | return res; |
1316 | } | |
1317 | #endif | |
1318 | ||
1c75950b | 1319 | long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) |
1da177e4 | 1320 | { |
1c75950b | 1321 | struct inode *inode = mapping->host; |
c42de9dd | 1322 | struct nfs_inode *nfsi = NFS_I(inode); |
1c75950b TM |
1323 | unsigned long idx_start, idx_end; |
1324 | unsigned int npages = 0; | |
c42de9dd | 1325 | LIST_HEAD(head); |
70b9ecbd | 1326 | int nocommit = how & FLUSH_NOCOMMIT; |
3f442547 | 1327 | long pages, ret; |
1da177e4 | 1328 | |
1c75950b TM |
1329 | /* FIXME */ |
1330 | if (wbc->range_cyclic) | |
1331 | idx_start = 0; | |
1332 | else { | |
1333 | idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
1334 | idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
1335 | if (idx_end > idx_start) { | |
1336 | unsigned long l_npages = 1 + idx_end - idx_start; | |
1337 | npages = l_npages; | |
1338 | if (sizeof(npages) != sizeof(l_npages) && | |
1339 | (unsigned long)npages != l_npages) | |
1340 | npages = 0; | |
1341 | } | |
1342 | } | |
c42de9dd TM |
1343 | how &= ~FLUSH_NOCOMMIT; |
1344 | spin_lock(&nfsi->req_lock); | |
1da177e4 | 1345 | do { |
1c75950b | 1346 | wbc->pages_skipped = 0; |
c42de9dd TM |
1347 | ret = nfs_wait_on_requests_locked(inode, idx_start, npages); |
1348 | if (ret != 0) | |
70b9ecbd | 1349 | continue; |
1c75950b | 1350 | pages = nfs_scan_dirty(mapping, wbc, &head); |
c42de9dd TM |
1351 | if (pages != 0) { |
1352 | spin_unlock(&nfsi->req_lock); | |
e8e058e8 | 1353 | if (how & FLUSH_INVALIDATE) { |
83715ad5 | 1354 | nfs_cancel_dirty_list(&head); |
e8e058e8 TM |
1355 | ret = pages; |
1356 | } else | |
d2ccddf0 | 1357 | ret = nfs_flush_list(inode, &head, pages, how); |
c42de9dd TM |
1358 | spin_lock(&nfsi->req_lock); |
1359 | continue; | |
1360 | } | |
1c75950b TM |
1361 | if (wbc->pages_skipped != 0) |
1362 | continue; | |
c42de9dd TM |
1363 | if (nocommit) |
1364 | break; | |
d2ccddf0 | 1365 | pages = nfs_scan_commit(inode, &head, idx_start, npages); |
1c75950b TM |
1366 | if (pages == 0) { |
1367 | if (wbc->pages_skipped != 0) | |
1368 | continue; | |
c42de9dd | 1369 | break; |
1c75950b | 1370 | } |
d2ccddf0 TM |
1371 | if (how & FLUSH_INVALIDATE) { |
1372 | spin_unlock(&nfsi->req_lock); | |
83715ad5 | 1373 | nfs_cancel_commit_list(&head); |
e8e058e8 | 1374 | ret = pages; |
d2ccddf0 TM |
1375 | spin_lock(&nfsi->req_lock); |
1376 | continue; | |
1377 | } | |
1378 | pages += nfs_scan_commit(inode, &head, 0, 0); | |
c42de9dd TM |
1379 | spin_unlock(&nfsi->req_lock); |
1380 | ret = nfs_commit_list(inode, &head, how); | |
1381 | spin_lock(&nfsi->req_lock); | |
1382 | } while (ret >= 0); | |
1383 | spin_unlock(&nfsi->req_lock); | |
1384 | return ret; | |
1da177e4 LT |
1385 | } |
1386 | ||
1c75950b TM |
1387 | /* |
1388 | * flush the inode to disk. | |
1389 | */ | |
1390 | int nfs_wb_all(struct inode *inode) | |
1391 | { | |
1392 | struct address_space *mapping = inode->i_mapping; | |
1393 | struct writeback_control wbc = { | |
1394 | .bdi = mapping->backing_dev_info, | |
1395 | .sync_mode = WB_SYNC_ALL, | |
1396 | .nr_to_write = LONG_MAX, | |
1397 | .range_cyclic = 1, | |
1398 | }; | |
1399 | int ret; | |
1400 | ||
1401 | ret = nfs_sync_mapping_wait(mapping, &wbc, 0); | |
1402 | if (ret >= 0) | |
1403 | return 0; | |
1404 | return ret; | |
1405 | } | |
1406 | ||
1407 | int nfs_sync_mapping_range(struct address_space *mapping, loff_t range_start, loff_t range_end, int how) | |
1408 | { | |
1409 | struct writeback_control wbc = { | |
1410 | .bdi = mapping->backing_dev_info, | |
1411 | .sync_mode = WB_SYNC_ALL, | |
1412 | .nr_to_write = LONG_MAX, | |
1413 | .range_start = range_start, | |
1414 | .range_end = range_end, | |
1415 | }; | |
1416 | int ret; | |
1417 | ||
1418 | ret = nfs_sync_mapping_wait(mapping, &wbc, how); | |
1419 | if (ret >= 0) | |
1420 | return 0; | |
1421 | return ret; | |
1422 | } | |
1423 | ||
1424 | static int nfs_wb_page_priority(struct inode *inode, struct page *page, int how) | |
1425 | { | |
1426 | loff_t range_start = page_offset(page); | |
1427 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
1428 | ||
1429 | return nfs_sync_mapping_range(inode->i_mapping, range_start, range_end, how | FLUSH_STABLE); | |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * Write back all requests on one page - we do this before reading it. | |
1434 | */ | |
1435 | int nfs_wb_page(struct inode *inode, struct page* page) | |
1436 | { | |
1437 | return nfs_wb_page_priority(inode, page, 0); | |
1438 | } | |
1439 | ||
1440 | ||
f7b422b1 | 1441 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1442 | { |
1443 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
1444 | sizeof(struct nfs_write_data), | |
1445 | 0, SLAB_HWCACHE_ALIGN, | |
1446 | NULL, NULL); | |
1447 | if (nfs_wdata_cachep == NULL) | |
1448 | return -ENOMEM; | |
1449 | ||
93d2341c MD |
1450 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1451 | nfs_wdata_cachep); | |
1da177e4 LT |
1452 | if (nfs_wdata_mempool == NULL) |
1453 | return -ENOMEM; | |
1454 | ||
93d2341c MD |
1455 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1456 | nfs_wdata_cachep); | |
1da177e4 LT |
1457 | if (nfs_commit_mempool == NULL) |
1458 | return -ENOMEM; | |
1459 | ||
1460 | return 0; | |
1461 | } | |
1462 | ||
266bee88 | 1463 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1464 | { |
1465 | mempool_destroy(nfs_commit_mempool); | |
1466 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1467 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1468 | } |
1469 |