]>
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; |
1da177e4 | 49 | |
e18b890b | 50 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 51 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 52 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
53 | static mempool_t *nfs_commit_mempool; |
54 | ||
0b7c0153 | 55 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 56 | { |
192e501b | 57 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); |
40859d7e | 58 | |
1da177e4 LT |
59 | if (p) { |
60 | memset(p, 0, sizeof(*p)); | |
61 | INIT_LIST_HEAD(&p->pages); | |
62 | } | |
63 | return p; | |
64 | } | |
e0c2b380 | 65 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 66 | |
0b7c0153 | 67 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
68 | { |
69 | mempool_free(p, nfs_commit_mempool); | |
70 | } | |
e0c2b380 | 71 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 72 | |
4a0de55c | 73 | static struct nfs_rw_header *nfs_writehdr_alloc(void) |
3feb2d49 | 74 | { |
c0752cdf | 75 | struct nfs_rw_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
3feb2d49 | 76 | |
4a0de55c | 77 | if (p) |
3feb2d49 | 78 | memset(p, 0, sizeof(*p)); |
3feb2d49 TM |
79 | return p; |
80 | } | |
81 | ||
4a0de55c | 82 | static void nfs_writehdr_free(struct nfs_rw_header *whdr) |
3feb2d49 | 83 | { |
cd841605 | 84 | mempool_free(whdr, nfs_wdata_mempool); |
3feb2d49 TM |
85 | } |
86 | ||
7b159fc1 TM |
87 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
88 | { | |
89 | ctx->error = error; | |
90 | smp_wmb(); | |
91 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
92 | } | |
93 | ||
29418aa4 MG |
94 | static struct nfs_page * |
95 | nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page) | |
277459d2 TM |
96 | { |
97 | struct nfs_page *req = NULL; | |
98 | ||
29418aa4 | 99 | if (PagePrivate(page)) |
277459d2 | 100 | req = (struct nfs_page *)page_private(page); |
29418aa4 MG |
101 | else if (unlikely(PageSwapCache(page))) { |
102 | struct nfs_page *freq, *t; | |
103 | ||
104 | /* Linearly search the commit list for the correct req */ | |
105 | list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { | |
106 | if (freq->wb_page == page) { | |
107 | req = freq; | |
108 | break; | |
109 | } | |
110 | } | |
277459d2 | 111 | } |
29418aa4 MG |
112 | |
113 | if (req) | |
114 | kref_get(&req->wb_kref); | |
115 | ||
277459d2 TM |
116 | return req; |
117 | } | |
118 | ||
119 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
120 | { | |
d56b4ddf | 121 | struct inode *inode = page_file_mapping(page)->host; |
277459d2 | 122 | struct nfs_page *req = NULL; |
277459d2 | 123 | |
587142f8 | 124 | spin_lock(&inode->i_lock); |
29418aa4 | 125 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
587142f8 | 126 | spin_unlock(&inode->i_lock); |
277459d2 TM |
127 | return req; |
128 | } | |
129 | ||
1da177e4 LT |
130 | /* Adjust the file length if we're writing beyond the end */ |
131 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
132 | { | |
d56b4ddf | 133 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
134 | loff_t end, i_size; |
135 | pgoff_t end_index; | |
1da177e4 | 136 | |
a3d01454 TM |
137 | spin_lock(&inode->i_lock); |
138 | i_size = i_size_read(inode); | |
139 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
d56b4ddf | 140 | if (i_size > 0 && page_file_index(page) < end_index) |
a3d01454 | 141 | goto out; |
d56b4ddf | 142 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 143 | if (i_size >= end) |
a3d01454 | 144 | goto out; |
1da177e4 | 145 | i_size_write(inode, end); |
a3d01454 TM |
146 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
147 | out: | |
148 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
149 | } |
150 | ||
a301b777 TM |
151 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
152 | static void nfs_set_pageerror(struct page *page) | |
153 | { | |
d56b4ddf | 154 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
155 | } |
156 | ||
1da177e4 LT |
157 | /* We can set the PG_uptodate flag if we see that a write request |
158 | * covers the full page. | |
159 | */ | |
160 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
161 | { | |
1da177e4 LT |
162 | if (PageUptodate(page)) |
163 | return; | |
164 | if (base != 0) | |
165 | return; | |
49a70f27 | 166 | if (count != nfs_page_length(page)) |
1da177e4 | 167 | return; |
49a70f27 | 168 | SetPageUptodate(page); |
1da177e4 LT |
169 | } |
170 | ||
1da177e4 LT |
171 | static int wb_priority(struct writeback_control *wbc) |
172 | { | |
173 | if (wbc->for_reclaim) | |
c63c7b05 | 174 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 175 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
176 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
177 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
178 | } |
179 | ||
89a09141 PZ |
180 | /* |
181 | * NFS congestion control | |
182 | */ | |
183 | ||
184 | int nfs_congestion_kb; | |
185 | ||
186 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
187 | #define NFS_CONGESTION_OFF_THRESH \ | |
188 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
189 | ||
deed85e7 | 190 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 191 | { |
deed85e7 | 192 | struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); |
5a6d41b3 TM |
193 | int ret = test_set_page_writeback(page); |
194 | ||
deed85e7 | 195 | WARN_ON_ONCE(ret != 0); |
89a09141 | 196 | |
deed85e7 TM |
197 | if (atomic_long_inc_return(&nfss->writeback) > |
198 | NFS_CONGESTION_ON_THRESH) { | |
199 | set_bdi_congested(&nfss->backing_dev_info, | |
200 | BLK_RW_ASYNC); | |
89a09141 PZ |
201 | } |
202 | } | |
203 | ||
204 | static void nfs_end_page_writeback(struct page *page) | |
205 | { | |
d56b4ddf | 206 | struct inode *inode = page_file_mapping(page)->host; |
89a09141 PZ |
207 | struct nfs_server *nfss = NFS_SERVER(inode); |
208 | ||
209 | end_page_writeback(page); | |
c4dc4bee | 210 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 211 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
212 | } |
213 | ||
cfb506e1 | 214 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 215 | { |
d56b4ddf | 216 | struct inode *inode = page_file_mapping(page)->host; |
e261f51f | 217 | struct nfs_page *req; |
e261f51f TM |
218 | int ret; |
219 | ||
587142f8 | 220 | spin_lock(&inode->i_lock); |
074cc1de | 221 | for (;;) { |
29418aa4 | 222 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
074cc1de TM |
223 | if (req == NULL) |
224 | break; | |
7ad84aa9 | 225 | if (nfs_lock_request(req)) |
e261f51f TM |
226 | break; |
227 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
7ad84aa9 | 228 | * then the call to nfs_lock_request() will always |
e261f51f TM |
229 | * succeed provided that someone hasn't already marked the |
230 | * request as dirty (in which case we don't care). | |
231 | */ | |
587142f8 | 232 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
233 | if (!nonblock) |
234 | ret = nfs_wait_on_request(req); | |
235 | else | |
236 | ret = -EAGAIN; | |
e261f51f TM |
237 | nfs_release_request(req); |
238 | if (ret != 0) | |
074cc1de | 239 | return ERR_PTR(ret); |
587142f8 | 240 | spin_lock(&inode->i_lock); |
e261f51f | 241 | } |
587142f8 | 242 | spin_unlock(&inode->i_lock); |
074cc1de TM |
243 | return req; |
244 | } | |
245 | ||
246 | /* | |
247 | * Find an associated nfs write request, and prepare to flush it out | |
248 | * May return an error if the user signalled nfs_wait_on_request(). | |
249 | */ | |
250 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 251 | struct page *page, bool nonblock) |
074cc1de TM |
252 | { |
253 | struct nfs_page *req; | |
254 | int ret = 0; | |
255 | ||
cfb506e1 | 256 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
257 | if (!req) |
258 | goto out; | |
259 | ret = PTR_ERR(req); | |
260 | if (IS_ERR(req)) | |
261 | goto out; | |
262 | ||
deed85e7 TM |
263 | nfs_set_page_writeback(page); |
264 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 265 | |
deed85e7 | 266 | ret = 0; |
f8512ad0 FI |
267 | if (!nfs_pageio_add_request(pgio, req)) { |
268 | nfs_redirty_request(req); | |
074cc1de | 269 | ret = pgio->pg_error; |
f8512ad0 | 270 | } |
074cc1de TM |
271 | out: |
272 | return ret; | |
e261f51f TM |
273 | } |
274 | ||
f758c885 | 275 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 276 | { |
d56b4ddf | 277 | struct inode *inode = page_file_mapping(page)->host; |
cfb506e1 | 278 | int ret; |
1da177e4 | 279 | |
91d5b470 CL |
280 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
281 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
282 | ||
d56b4ddf | 283 | nfs_pageio_cond_complete(pgio, page_file_index(page)); |
1b430bee | 284 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
285 | if (ret == -EAGAIN) { |
286 | redirty_page_for_writepage(wbc, page); | |
287 | ret = 0; | |
288 | } | |
289 | return ret; | |
f758c885 | 290 | } |
7fe7f848 | 291 | |
f758c885 TM |
292 | /* |
293 | * Write an mmapped page to the server. | |
294 | */ | |
295 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
296 | { | |
297 | struct nfs_pageio_descriptor pgio; | |
298 | int err; | |
49a70f27 | 299 | |
a20c93e3 CH |
300 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), |
301 | false, &nfs_async_write_completion_ops); | |
f758c885 TM |
302 | err = nfs_do_writepage(page, wbc, &pgio); |
303 | nfs_pageio_complete(&pgio); | |
304 | if (err < 0) | |
305 | return err; | |
306 | if (pgio.pg_error < 0) | |
307 | return pgio.pg_error; | |
308 | return 0; | |
4d770ccf TM |
309 | } |
310 | ||
311 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
312 | { | |
f758c885 | 313 | int ret; |
4d770ccf | 314 | |
f758c885 | 315 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 316 | unlock_page(page); |
f758c885 TM |
317 | return ret; |
318 | } | |
319 | ||
320 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
321 | { | |
322 | int ret; | |
323 | ||
324 | ret = nfs_do_writepage(page, wbc, data); | |
325 | unlock_page(page); | |
326 | return ret; | |
1da177e4 LT |
327 | } |
328 | ||
1da177e4 LT |
329 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
330 | { | |
1da177e4 | 331 | struct inode *inode = mapping->host; |
72cb77f4 | 332 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 333 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
334 | int err; |
335 | ||
72cb77f4 TM |
336 | /* Stop dirtying of new pages while we sync */ |
337 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
338 | nfs_wait_bit_killable, TASK_KILLABLE); | |
339 | if (err) | |
340 | goto out_err; | |
341 | ||
91d5b470 CL |
342 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
343 | ||
a20c93e3 CH |
344 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, |
345 | &nfs_async_write_completion_ops); | |
f758c885 | 346 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 347 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
348 | |
349 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
350 | smp_mb__after_clear_bit(); | |
351 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
352 | ||
f758c885 | 353 | if (err < 0) |
72cb77f4 TM |
354 | goto out_err; |
355 | err = pgio.pg_error; | |
356 | if (err < 0) | |
357 | goto out_err; | |
c63c7b05 | 358 | return 0; |
72cb77f4 TM |
359 | out_err: |
360 | return err; | |
1da177e4 LT |
361 | } |
362 | ||
363 | /* | |
364 | * Insert a write request into an inode | |
365 | */ | |
d6d6dc7c | 366 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
367 | { |
368 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 TM |
369 | |
370 | /* Lock the request! */ | |
7ad84aa9 | 371 | nfs_lock_request(req); |
e7d39069 TM |
372 | |
373 | spin_lock(&inode->i_lock); | |
011e2a7f | 374 | if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 375 | inode->i_version++; |
29418aa4 MG |
376 | /* |
377 | * Swap-space should not get truncated. Hence no need to plug the race | |
378 | * with invalidate/truncate. | |
379 | */ | |
380 | if (likely(!PageSwapCache(req->wb_page))) { | |
381 | set_bit(PG_MAPPED, &req->wb_flags); | |
382 | SetPagePrivate(req->wb_page); | |
383 | set_page_private(req->wb_page, (unsigned long)req); | |
384 | } | |
1da177e4 | 385 | nfsi->npages++; |
c03b4024 | 386 | kref_get(&req->wb_kref); |
e7d39069 | 387 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
388 | } |
389 | ||
390 | /* | |
89a09141 | 391 | * Remove a write request from an inode |
1da177e4 LT |
392 | */ |
393 | static void nfs_inode_remove_request(struct nfs_page *req) | |
394 | { | |
3d4ff43d | 395 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
396 | struct nfs_inode *nfsi = NFS_I(inode); |
397 | ||
587142f8 | 398 | spin_lock(&inode->i_lock); |
29418aa4 MG |
399 | if (likely(!PageSwapCache(req->wb_page))) { |
400 | set_page_private(req->wb_page, 0); | |
401 | ClearPagePrivate(req->wb_page); | |
402 | clear_bit(PG_MAPPED, &req->wb_flags); | |
403 | } | |
1da177e4 | 404 | nfsi->npages--; |
4d65c520 | 405 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
406 | nfs_release_request(req); |
407 | } | |
408 | ||
61822ab5 | 409 | static void |
6d884e8f | 410 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 411 | { |
61822ab5 TM |
412 | __set_page_dirty_nobuffers(req->wb_page); |
413 | } | |
414 | ||
89d77c8f | 415 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
8dd37758 TM |
416 | /** |
417 | * nfs_request_add_commit_list - add request to a commit list | |
418 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
419 | * @dst: commit list head |
420 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 421 | * |
ea2cf228 | 422 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
423 | * number of outstanding requests requiring a commit as well as |
424 | * the MM page stats. | |
425 | * | |
ea2cf228 | 426 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 427 | * holding the nfs_page lock. |
1da177e4 | 428 | */ |
8dd37758 | 429 | void |
ea2cf228 FI |
430 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
431 | struct nfs_commit_info *cinfo) | |
1da177e4 | 432 | { |
e468bae9 | 433 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
434 | spin_lock(cinfo->lock); |
435 | nfs_list_add_request(req, dst); | |
436 | cinfo->mds->ncommit++; | |
437 | spin_unlock(cinfo->lock); | |
56f9cd68 FI |
438 | if (!cinfo->dreq) { |
439 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 440 | inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
441 | BDI_RECLAIMABLE); |
442 | __mark_inode_dirty(req->wb_context->dentry->d_inode, | |
443 | I_DIRTY_DATASYNC); | |
444 | } | |
1da177e4 | 445 | } |
8dd37758 TM |
446 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
447 | ||
448 | /** | |
449 | * nfs_request_remove_commit_list - Remove request from a commit list | |
450 | * @req: pointer to a nfs_page | |
ea2cf228 | 451 | * @cinfo: holds list lock and accounting info |
8dd37758 | 452 | * |
ea2cf228 | 453 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
454 | * number of outstanding requests requiring a commit |
455 | * It does not update the MM page stats. | |
456 | * | |
ea2cf228 | 457 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
458 | */ |
459 | void | |
ea2cf228 FI |
460 | nfs_request_remove_commit_list(struct nfs_page *req, |
461 | struct nfs_commit_info *cinfo) | |
8dd37758 | 462 | { |
8dd37758 TM |
463 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
464 | return; | |
465 | nfs_list_remove_request(req); | |
ea2cf228 | 466 | cinfo->mds->ncommit--; |
8dd37758 TM |
467 | } |
468 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
469 | ||
ea2cf228 FI |
470 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
471 | struct inode *inode) | |
472 | { | |
473 | cinfo->lock = &inode->i_lock; | |
474 | cinfo->mds = &NFS_I(inode)->commit_info; | |
475 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 476 | cinfo->dreq = NULL; |
f453a54a | 477 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
478 | } |
479 | ||
480 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
481 | struct inode *inode, | |
482 | struct nfs_direct_req *dreq) | |
483 | { | |
1763da12 FI |
484 | if (dreq) |
485 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
486 | else | |
487 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
488 | } |
489 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
490 | |
491 | /* | |
492 | * Add a request to the inode's commit list. | |
493 | */ | |
1763da12 | 494 | void |
ea2cf228 FI |
495 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
496 | struct nfs_commit_info *cinfo) | |
8dd37758 | 497 | { |
ea2cf228 | 498 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 499 | return; |
ea2cf228 | 500 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 501 | } |
8e821cad | 502 | |
d6d6dc7c FI |
503 | static void |
504 | nfs_clear_page_commit(struct page *page) | |
505 | { | |
506 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
d56b4ddf | 507 | dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); |
d6d6dc7c FI |
508 | } |
509 | ||
8dd37758 | 510 | static void |
e468bae9 TM |
511 | nfs_clear_request_commit(struct nfs_page *req) |
512 | { | |
8dd37758 TM |
513 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
514 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 515 | struct nfs_commit_info cinfo; |
e468bae9 | 516 | |
ea2cf228 FI |
517 | nfs_init_cinfo_from_inode(&cinfo, inode); |
518 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
519 | spin_lock(cinfo.lock); | |
520 | nfs_request_remove_commit_list(req, &cinfo); | |
521 | spin_unlock(cinfo.lock); | |
8dd37758 | 522 | } |
d6d6dc7c | 523 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 524 | } |
e468bae9 TM |
525 | } |
526 | ||
8e821cad | 527 | static inline |
9c7e1b3d | 528 | int nfs_write_need_commit(struct nfs_pgio_data *data) |
8e821cad | 529 | { |
465d5243 | 530 | if (data->verf.committed == NFS_DATA_SYNC) |
cd841605 FI |
531 | return data->header->lseg == NULL; |
532 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
533 | } |
534 | ||
8e821cad | 535 | #else |
68cd6fa4 BS |
536 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
537 | struct inode *inode) | |
538 | { | |
539 | } | |
540 | ||
541 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
542 | struct inode *inode, | |
543 | struct nfs_direct_req *dreq) | |
544 | { | |
545 | } | |
546 | ||
1763da12 | 547 | void |
ea2cf228 FI |
548 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
549 | struct nfs_commit_info *cinfo) | |
8e821cad TM |
550 | { |
551 | } | |
552 | ||
8dd37758 | 553 | static void |
e468bae9 TM |
554 | nfs_clear_request_commit(struct nfs_page *req) |
555 | { | |
e468bae9 TM |
556 | } |
557 | ||
8e821cad | 558 | static inline |
9c7e1b3d | 559 | int nfs_write_need_commit(struct nfs_pgio_data *data) |
8e821cad TM |
560 | { |
561 | return 0; | |
562 | } | |
563 | ||
6c75dc0d FI |
564 | #endif |
565 | ||
061ae2ed | 566 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 567 | { |
ea2cf228 | 568 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
569 | unsigned long bytes = 0; |
570 | ||
571 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
572 | goto out; | |
ea2cf228 | 573 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
574 | while (!list_empty(&hdr->pages)) { |
575 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
576 | |
577 | bytes += req->wb_bytes; | |
578 | nfs_list_remove_request(req); | |
579 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
580 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 581 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
582 | nfs_context_set_write_error(req->wb_context, hdr->error); |
583 | goto remove_req; | |
584 | } | |
585 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
586 | nfs_mark_request_dirty(req); | |
587 | goto next; | |
588 | } | |
589 | if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
f79d06f5 | 590 | memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); |
ea2cf228 | 591 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
592 | goto next; |
593 | } | |
594 | remove_req: | |
595 | nfs_inode_remove_request(req); | |
596 | next: | |
1d1afcbc | 597 | nfs_unlock_request(req); |
d1182b33 | 598 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 599 | nfs_release_request(req); |
6c75dc0d FI |
600 | } |
601 | out: | |
602 | hdr->release(hdr); | |
8e821cad | 603 | } |
1da177e4 | 604 | |
89d77c8f | 605 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
ce59515c | 606 | unsigned long |
ea2cf228 | 607 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 | 608 | { |
ea2cf228 | 609 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
610 | } |
611 | ||
ea2cf228 | 612 | /* cinfo->lock held by caller */ |
1763da12 | 613 | int |
ea2cf228 FI |
614 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
615 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
616 | { |
617 | struct nfs_page *req, *tmp; | |
618 | int ret = 0; | |
619 | ||
620 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
621 | if (!nfs_lock_request(req)) |
622 | continue; | |
7ad84aa9 | 623 | kref_get(&req->wb_kref); |
ea2cf228 | 624 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 625 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 626 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
627 | nfs_list_add_request(req, dst); |
628 | ret++; | |
1763da12 | 629 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 630 | break; |
d6d6dc7c FI |
631 | } |
632 | return ret; | |
fb8a1f11 TM |
633 | } |
634 | ||
1da177e4 LT |
635 | /* |
636 | * nfs_scan_commit - Scan an inode for commit requests | |
637 | * @inode: NFS inode to scan | |
ea2cf228 FI |
638 | * @dst: mds destination list |
639 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
640 | * |
641 | * Moves requests from the inode's 'commit' request list. | |
642 | * The requests are *not* checked to ensure that they form a contiguous set. | |
643 | */ | |
1763da12 | 644 | int |
ea2cf228 FI |
645 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
646 | struct nfs_commit_info *cinfo) | |
1da177e4 | 647 | { |
d6d6dc7c | 648 | int ret = 0; |
fb8a1f11 | 649 | |
ea2cf228 FI |
650 | spin_lock(cinfo->lock); |
651 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 652 | const int max = INT_MAX; |
d6d6dc7c | 653 | |
ea2cf228 FI |
654 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
655 | cinfo, max); | |
656 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 657 | } |
ea2cf228 | 658 | spin_unlock(cinfo->lock); |
ff778d02 | 659 | return ret; |
1da177e4 | 660 | } |
d6d6dc7c | 661 | |
c42de9dd | 662 | #else |
ce59515c | 663 | unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 TM |
664 | { |
665 | return 0; | |
666 | } | |
667 | ||
1763da12 FI |
668 | int nfs_scan_commit(struct inode *inode, struct list_head *dst, |
669 | struct nfs_commit_info *cinfo) | |
c42de9dd TM |
670 | { |
671 | return 0; | |
672 | } | |
1da177e4 LT |
673 | #endif |
674 | ||
1da177e4 | 675 | /* |
e7d39069 TM |
676 | * Search for an existing write request, and attempt to update |
677 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 678 | * |
e7d39069 TM |
679 | * If the attempt fails, then the existing request is flushed out |
680 | * to disk. | |
1da177e4 | 681 | */ |
e7d39069 TM |
682 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
683 | struct page *page, | |
684 | unsigned int offset, | |
685 | unsigned int bytes) | |
1da177e4 | 686 | { |
e7d39069 TM |
687 | struct nfs_page *req; |
688 | unsigned int rqend; | |
689 | unsigned int end; | |
690 | int error; | |
691 | ||
692 | if (!PagePrivate(page)) | |
693 | return NULL; | |
1da177e4 LT |
694 | |
695 | end = offset + bytes; | |
e7d39069 | 696 | spin_lock(&inode->i_lock); |
1da177e4 | 697 | |
1da177e4 | 698 | for (;;) { |
29418aa4 | 699 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
e7d39069 TM |
700 | if (req == NULL) |
701 | goto out_unlock; | |
702 | ||
703 | rqend = req->wb_offset + req->wb_bytes; | |
704 | /* | |
705 | * Tell the caller to flush out the request if | |
706 | * the offsets are non-contiguous. | |
707 | * Note: nfs_flush_incompatible() will already | |
708 | * have flushed out requests having wrong owners. | |
709 | */ | |
e468bae9 | 710 | if (offset > rqend |
e7d39069 TM |
711 | || end < req->wb_offset) |
712 | goto out_flushme; | |
713 | ||
7ad84aa9 | 714 | if (nfs_lock_request(req)) |
1da177e4 | 715 | break; |
1da177e4 | 716 | |
e7d39069 | 717 | /* The request is locked, so wait and then retry */ |
587142f8 | 718 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
719 | error = nfs_wait_on_request(req); |
720 | nfs_release_request(req); | |
721 | if (error != 0) | |
722 | goto out_err; | |
723 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
724 | } |
725 | ||
726 | /* Okay, the request matches. Update the region */ | |
727 | if (offset < req->wb_offset) { | |
728 | req->wb_offset = offset; | |
729 | req->wb_pgbase = offset; | |
1da177e4 | 730 | } |
1da177e4 LT |
731 | if (end > rqend) |
732 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
733 | else |
734 | req->wb_bytes = rqend - req->wb_offset; | |
735 | out_unlock: | |
736 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
737 | if (req) |
738 | nfs_clear_request_commit(req); | |
e7d39069 TM |
739 | return req; |
740 | out_flushme: | |
741 | spin_unlock(&inode->i_lock); | |
742 | nfs_release_request(req); | |
743 | error = nfs_wb_page(inode, page); | |
744 | out_err: | |
745 | return ERR_PTR(error); | |
746 | } | |
747 | ||
748 | /* | |
749 | * Try to update an existing write request, or create one if there is none. | |
750 | * | |
751 | * Note: Should always be called with the Page Lock held to prevent races | |
752 | * if we have to add a new request. Also assumes that the caller has | |
753 | * already called nfs_flush_incompatible() if necessary. | |
754 | */ | |
755 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
756 | struct page *page, unsigned int offset, unsigned int bytes) | |
757 | { | |
d56b4ddf | 758 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 759 | struct nfs_page *req; |
1da177e4 | 760 | |
e7d39069 TM |
761 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
762 | if (req != NULL) | |
763 | goto out; | |
8c8f1ac1 | 764 | req = nfs_create_request(ctx, page, offset, bytes); |
e7d39069 TM |
765 | if (IS_ERR(req)) |
766 | goto out; | |
d6d6dc7c | 767 | nfs_inode_add_request(inode, req); |
efc91ed0 | 768 | out: |
61e930a9 | 769 | return req; |
1da177e4 LT |
770 | } |
771 | ||
e7d39069 TM |
772 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
773 | unsigned int offset, unsigned int count) | |
774 | { | |
775 | struct nfs_page *req; | |
776 | ||
777 | req = nfs_setup_write_request(ctx, page, offset, count); | |
778 | if (IS_ERR(req)) | |
779 | return PTR_ERR(req); | |
780 | /* Update file length */ | |
781 | nfs_grow_file(page, offset, count); | |
782 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 783 | nfs_mark_request_dirty(req); |
1d1afcbc | 784 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
785 | return 0; |
786 | } | |
787 | ||
1da177e4 LT |
788 | int nfs_flush_incompatible(struct file *file, struct page *page) |
789 | { | |
cd3758e3 | 790 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 791 | struct nfs_lock_context *l_ctx; |
1da177e4 | 792 | struct nfs_page *req; |
1a54533e | 793 | int do_flush, status; |
1da177e4 LT |
794 | /* |
795 | * Look for a request corresponding to this page. If there | |
796 | * is one, and it belongs to another file, we flush it out | |
797 | * before we try to copy anything into the page. Do this | |
798 | * due to the lack of an ACCESS-type call in NFSv2. | |
799 | * Also do the same if we find a request from an existing | |
800 | * dropped page. | |
801 | */ | |
1a54533e TM |
802 | do { |
803 | req = nfs_page_find_request(page); | |
804 | if (req == NULL) | |
805 | return 0; | |
2a369153 TM |
806 | l_ctx = req->wb_lock_context; |
807 | do_flush = req->wb_page != page || req->wb_context != ctx; | |
0f1d2605 | 808 | if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { |
2a369153 TM |
809 | do_flush |= l_ctx->lockowner.l_owner != current->files |
810 | || l_ctx->lockowner.l_pid != current->tgid; | |
811 | } | |
1da177e4 | 812 | nfs_release_request(req); |
1a54533e TM |
813 | if (!do_flush) |
814 | return 0; | |
d56b4ddf | 815 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
816 | } while (status == 0); |
817 | return status; | |
1da177e4 LT |
818 | } |
819 | ||
dc24826b AA |
820 | /* |
821 | * Avoid buffered writes when a open context credential's key would | |
822 | * expire soon. | |
823 | * | |
824 | * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. | |
825 | * | |
826 | * Return 0 and set a credential flag which triggers the inode to flush | |
827 | * and performs NFS_FILE_SYNC writes if the key will expired within | |
828 | * RPC_KEY_EXPIRE_TIMEO. | |
829 | */ | |
830 | int | |
831 | nfs_key_timeout_notify(struct file *filp, struct inode *inode) | |
832 | { | |
833 | struct nfs_open_context *ctx = nfs_file_open_context(filp); | |
834 | struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; | |
835 | ||
836 | return rpcauth_key_timeout_notify(auth, ctx->cred); | |
837 | } | |
838 | ||
839 | /* | |
840 | * Test if the open context credential key is marked to expire soon. | |
841 | */ | |
842 | bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) | |
843 | { | |
844 | return rpcauth_cred_key_to_expire(ctx->cred); | |
845 | } | |
846 | ||
5d47a356 TM |
847 | /* |
848 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
849 | * the PageUptodate() flag. In this case, we will need to turn off | |
850 | * write optimisations that depend on the page contents being correct. | |
851 | */ | |
8d197a56 | 852 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 853 | { |
d529ef83 JL |
854 | struct nfs_inode *nfsi = NFS_I(inode); |
855 | ||
8d197a56 TM |
856 | if (nfs_have_delegated_attributes(inode)) |
857 | goto out; | |
d529ef83 JL |
858 | if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE)) |
859 | return false; | |
4db72b40 | 860 | smp_rmb(); |
d529ef83 | 861 | if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) |
8d197a56 TM |
862 | return false; |
863 | out: | |
864 | return PageUptodate(page) != 0; | |
5d47a356 TM |
865 | } |
866 | ||
c7559663 SM |
867 | /* If we know the page is up to date, and we're not using byte range locks (or |
868 | * if we have the whole file locked for writing), it may be more efficient to | |
869 | * extend the write to cover the entire page in order to avoid fragmentation | |
870 | * inefficiencies. | |
871 | * | |
263b4509 SM |
872 | * If the file is opened for synchronous writes then we can just skip the rest |
873 | * of the checks. | |
c7559663 SM |
874 | */ |
875 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
876 | { | |
877 | if (file->f_flags & O_DSYNC) | |
878 | return 0; | |
263b4509 SM |
879 | if (!nfs_write_pageuptodate(page, inode)) |
880 | return 0; | |
c7559663 SM |
881 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
882 | return 1; | |
263b4509 | 883 | if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && |
c7559663 | 884 | inode->i_flock->fl_end == OFFSET_MAX && |
263b4509 | 885 | inode->i_flock->fl_type != F_RDLCK)) |
c7559663 SM |
886 | return 1; |
887 | return 0; | |
888 | } | |
889 | ||
1da177e4 LT |
890 | /* |
891 | * Update and possibly write a cached page of an NFS file. | |
892 | * | |
893 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
894 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
895 | */ | |
896 | int nfs_updatepage(struct file *file, struct page *page, | |
897 | unsigned int offset, unsigned int count) | |
898 | { | |
cd3758e3 | 899 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 900 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
901 | int status = 0; |
902 | ||
91d5b470 CL |
903 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
904 | ||
6de1472f AV |
905 | dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", |
906 | file, count, (long long)(page_file_offset(page) + offset)); | |
1da177e4 | 907 | |
c7559663 | 908 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 909 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 910 | offset = 0; |
1da177e4 LT |
911 | } |
912 | ||
e21195a7 | 913 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
914 | if (status < 0) |
915 | nfs_set_pageerror(page); | |
59b7c05f TM |
916 | else |
917 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 918 | |
48186c7d | 919 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 920 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
921 | return status; |
922 | } | |
923 | ||
3ff7576d | 924 | static int flush_task_priority(int how) |
1da177e4 LT |
925 | { |
926 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
927 | case FLUSH_HIGHPRI: | |
928 | return RPC_PRIORITY_HIGH; | |
929 | case FLUSH_LOWPRI: | |
930 | return RPC_PRIORITY_LOW; | |
931 | } | |
932 | return RPC_PRIORITY_NORMAL; | |
933 | } | |
934 | ||
1ed26f33 AS |
935 | static void nfs_initiate_write(struct nfs_pgio_data *data, struct rpc_message *msg, |
936 | struct rpc_task_setup *task_setup_data, int how) | |
1da177e4 | 937 | { |
cd841605 | 938 | struct inode *inode = data->header->inode; |
3ff7576d | 939 | int priority = flush_task_priority(how); |
1da177e4 | 940 | |
1ed26f33 AS |
941 | task_setup_data->priority = priority; |
942 | NFS_PROTO(inode)->write_setup(data, msg); | |
d138d5d1 | 943 | |
8c21c62c | 944 | nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, |
1ed26f33 | 945 | &task_setup_data->rpc_client, msg, data); |
d138d5d1 AA |
946 | } |
947 | ||
6d884e8f F |
948 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
949 | * call this on each, which will prepare them to be retried on next | |
950 | * writeback using standard nfs. | |
951 | */ | |
952 | static void nfs_redirty_request(struct nfs_page *req) | |
953 | { | |
954 | nfs_mark_request_dirty(req); | |
1d1afcbc | 955 | nfs_unlock_request(req); |
d1182b33 | 956 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 957 | nfs_release_request(req); |
6d884e8f F |
958 | } |
959 | ||
061ae2ed | 960 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
961 | { |
962 | struct nfs_page *req; | |
963 | ||
964 | while (!list_empty(head)) { | |
965 | req = nfs_list_entry(head->next); | |
966 | nfs_list_remove_request(req); | |
967 | nfs_redirty_request(req); | |
968 | } | |
969 | } | |
970 | ||
061ae2ed FI |
971 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
972 | .error_cleanup = nfs_async_write_error, | |
973 | .completion = nfs_write_completion, | |
974 | }; | |
975 | ||
57208fa7 | 976 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
a20c93e3 | 977 | struct inode *inode, int ioflags, bool force_mds, |
061ae2ed | 978 | const struct nfs_pgio_completion_ops *compl_ops) |
1da177e4 | 979 | { |
a20c93e3 | 980 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 981 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
a20c93e3 CH |
982 | |
983 | #ifdef CONFIG_NFS_V4_1 | |
984 | if (server->pnfs_curr_ld && !force_mds) | |
985 | pg_ops = server->pnfs_curr_ld->pg_write_ops; | |
986 | #endif | |
4a0de55c AS |
987 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, |
988 | server->wsize, ioflags); | |
1751c363 | 989 | } |
ddda8e0a | 990 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 991 | |
dce81290 TM |
992 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
993 | { | |
41d8d5b7 | 994 | pgio->pg_ops = &nfs_pgio_rw_ops; |
dce81290 TM |
995 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; |
996 | } | |
1f945357 | 997 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 998 | |
1da177e4 | 999 | |
0b7c0153 FI |
1000 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1001 | { | |
1002 | struct nfs_commit_data *data = calldata; | |
1003 | ||
1004 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1005 | } | |
1006 | ||
a4cdda59 | 1007 | static void nfs_writeback_release_common(struct nfs_pgio_data *data) |
c9d8f89d | 1008 | { |
cd841605 | 1009 | struct nfs_pgio_header *hdr = data->header; |
e2fecb21 | 1010 | int status = data->task.tk_status; |
788e7a89 | 1011 | |
6c75dc0d FI |
1012 | if ((status >= 0) && nfs_write_need_commit(data)) { |
1013 | spin_lock(&hdr->lock); | |
1014 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) | |
1015 | ; /* Do nothing */ | |
1016 | else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) | |
f79d06f5 AS |
1017 | memcpy(&hdr->verf, &data->verf, sizeof(hdr->verf)); |
1018 | else if (memcmp(&hdr->verf, &data->verf, sizeof(hdr->verf))) | |
6c75dc0d FI |
1019 | set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); |
1020 | spin_unlock(&hdr->lock); | |
1da177e4 LT |
1021 | } |
1022 | } | |
1023 | ||
1f2edbe3 TM |
1024 | /* |
1025 | * Special version of should_remove_suid() that ignores capabilities. | |
1026 | */ | |
1027 | static int nfs_should_remove_suid(const struct inode *inode) | |
1028 | { | |
1029 | umode_t mode = inode->i_mode; | |
1030 | int kill = 0; | |
1031 | ||
1032 | /* suid always must be killed */ | |
1033 | if (unlikely(mode & S_ISUID)) | |
1034 | kill = ATTR_KILL_SUID; | |
1035 | ||
1036 | /* | |
1037 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1038 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1039 | */ | |
1040 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1041 | kill |= ATTR_KILL_SGID; | |
1042 | ||
1043 | if (unlikely(kill && S_ISREG(mode))) | |
1044 | return kill; | |
1045 | ||
1046 | return 0; | |
1047 | } | |
788e7a89 | 1048 | |
1da177e4 LT |
1049 | /* |
1050 | * This function is called when the WRITE call is complete. | |
1051 | */ | |
0eecb214 AS |
1052 | static int nfs_writeback_done(struct rpc_task *task, struct nfs_pgio_data *data, |
1053 | struct inode *inode) | |
1da177e4 | 1054 | { |
788e7a89 | 1055 | int status; |
1da177e4 | 1056 | |
f551e44f CL |
1057 | /* |
1058 | * ->write_done will attempt to use post-op attributes to detect | |
1059 | * conflicting writes by other clients. A strict interpretation | |
1060 | * of close-to-open would allow us to continue caching even if | |
1061 | * another writer had changed the file, but some applications | |
1062 | * depend on tighter cache coherency when writing. | |
1063 | */ | |
cd841605 | 1064 | status = NFS_PROTO(inode)->write_done(task, data); |
788e7a89 | 1065 | if (status != 0) |
0eecb214 AS |
1066 | return status; |
1067 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, data->res.count); | |
91d5b470 | 1068 | |
89d77c8f | 1069 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
0eecb214 | 1070 | if (data->res.verf->committed < data->args.stable && task->tk_status >= 0) { |
1da177e4 LT |
1071 | /* We tried a write call, but the server did not |
1072 | * commit data to stable storage even though we | |
1073 | * requested it. | |
1074 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1075 | * the server reports NFS_DATA_SYNC, but performs | |
1076 | * NFS_FILE_SYNC. We therefore implement this checking | |
1077 | * as a dprintk() in order to avoid filling syslog. | |
1078 | */ | |
1079 | static unsigned long complain; | |
1080 | ||
a69aef14 | 1081 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1082 | if (time_before(complain, jiffies)) { |
48186c7d | 1083 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1084 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1085 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
0eecb214 | 1086 | data->res.verf->committed, data->args.stable); |
1da177e4 LT |
1087 | complain = jiffies + 300 * HZ; |
1088 | } | |
1089 | } | |
1090 | #endif | |
1f2edbe3 TM |
1091 | |
1092 | /* Deal with the suid/sgid bit corner case */ | |
1093 | if (nfs_should_remove_suid(inode)) | |
1094 | nfs_mark_for_revalidate(inode); | |
0eecb214 AS |
1095 | return 0; |
1096 | } | |
1097 | ||
1098 | /* | |
1099 | * This function is called when the WRITE call is complete. | |
1100 | */ | |
1101 | static void nfs_writeback_result(struct rpc_task *task, struct nfs_pgio_data *data) | |
1102 | { | |
1103 | struct nfs_pgio_args *argp = &data->args; | |
1104 | struct nfs_pgio_res *resp = &data->res; | |
1f2edbe3 TM |
1105 | |
1106 | if (resp->count < argp->count) { | |
1da177e4 LT |
1107 | static unsigned long complain; |
1108 | ||
6c75dc0d | 1109 | /* This a short write! */ |
0eecb214 | 1110 | nfs_inc_stats(data->header->inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1111 | |
1da177e4 | 1112 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1113 | if (resp->count == 0) { |
1114 | if (time_before(complain, jiffies)) { | |
1115 | printk(KERN_WARNING | |
1116 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1117 | argp->count); | |
1118 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1119 | } |
6c75dc0d FI |
1120 | nfs_set_pgio_error(data->header, -EIO, argp->offset); |
1121 | task->tk_status = -EIO; | |
13602896 | 1122 | return; |
1da177e4 | 1123 | } |
6c75dc0d FI |
1124 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1125 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1126 | /* Resend from where the server left off */ | |
1127 | data->mds_offset += resp->count; | |
1128 | argp->offset += resp->count; | |
1129 | argp->pgbase += resp->count; | |
1130 | argp->count -= resp->count; | |
1131 | } else { | |
1132 | /* Resend as a stable write in order to avoid | |
1133 | * headaches in the case of a server crash. | |
1134 | */ | |
1135 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1136 | } |
6c75dc0d | 1137 | rpc_restart_call_prepare(task); |
1da177e4 | 1138 | } |
1da177e4 LT |
1139 | } |
1140 | ||
1141 | ||
89d77c8f | 1142 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
71d0a611 TM |
1143 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1144 | { | |
b8413f98 TM |
1145 | int ret; |
1146 | ||
71d0a611 TM |
1147 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1148 | return 1; | |
b8413f98 TM |
1149 | if (!may_wait) |
1150 | return 0; | |
1151 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1152 | NFS_INO_COMMIT, | |
1153 | nfs_wait_bit_killable, | |
1154 | TASK_KILLABLE); | |
1155 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1156 | } |
1157 | ||
f453a54a | 1158 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1159 | { |
1160 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1161 | smp_mb__after_clear_bit(); | |
1162 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1163 | } | |
1164 | ||
0b7c0153 | 1165 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1166 | { |
0b7c0153 FI |
1167 | put_nfs_open_context(data->context); |
1168 | nfs_commit_free(data); | |
1da177e4 | 1169 | } |
e0c2b380 | 1170 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1171 | |
0b7c0153 | 1172 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd | 1173 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1174 | int how, int flags) |
1da177e4 | 1175 | { |
07737691 | 1176 | struct rpc_task *task; |
9ace33cd | 1177 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1178 | struct rpc_message msg = { |
1179 | .rpc_argp = &data->args, | |
1180 | .rpc_resp = &data->res, | |
9ace33cd | 1181 | .rpc_cred = data->cred, |
bdc7f021 | 1182 | }; |
84115e1c | 1183 | struct rpc_task_setup task_setup_data = { |
07737691 | 1184 | .task = &data->task, |
9ace33cd | 1185 | .rpc_client = clnt, |
bdc7f021 | 1186 | .rpc_message = &msg, |
9ace33cd | 1187 | .callback_ops = call_ops, |
84115e1c | 1188 | .callback_data = data, |
101070ca | 1189 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1190 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1191 | .priority = priority, |
84115e1c | 1192 | }; |
9ace33cd FI |
1193 | /* Set up the initial task struct. */ |
1194 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1195 | ||
1196 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1197 | ||
8c21c62c WAA |
1198 | nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, |
1199 | NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); | |
1200 | ||
9ace33cd FI |
1201 | task = rpc_run_task(&task_setup_data); |
1202 | if (IS_ERR(task)) | |
1203 | return PTR_ERR(task); | |
1204 | if (how & FLUSH_SYNC) | |
1205 | rpc_wait_for_completion_task(task); | |
1206 | rpc_put_task(task); | |
1207 | return 0; | |
1208 | } | |
e0c2b380 | 1209 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1210 | |
1211 | /* | |
1212 | * Set up the argument/result storage required for the RPC call. | |
1213 | */ | |
0b7c0153 | 1214 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1215 | struct list_head *head, |
1216 | struct pnfs_layout_segment *lseg, | |
1217 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1218 | { |
1219 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1220 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1221 | |
1222 | /* Set up the RPC argument and reply structs | |
1223 | * NB: take care not to mess about with data->commit et al. */ | |
1224 | ||
1225 | list_splice_init(head, &data->pages); | |
1da177e4 | 1226 | |
1da177e4 | 1227 | data->inode = inode; |
9ace33cd | 1228 | data->cred = first->wb_context->cred; |
988b6dce | 1229 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1230 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1231 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1232 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1233 | |
1234 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1235 | /* Note: we always request a commit of the entire inode */ |
1236 | data->args.offset = 0; | |
1237 | data->args.count = 0; | |
0b7c0153 | 1238 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1239 | data->res.fattr = &data->fattr; |
1240 | data->res.verf = &data->verf; | |
0e574af1 | 1241 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1242 | } |
e0c2b380 | 1243 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1244 | |
e0c2b380 | 1245 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1246 | struct pnfs_layout_segment *lseg, |
1247 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1248 | { |
1249 | struct nfs_page *req; | |
1250 | ||
1251 | while (!list_empty(page_list)) { | |
1252 | req = nfs_list_entry(page_list->next); | |
1253 | nfs_list_remove_request(req); | |
ea2cf228 | 1254 | nfs_mark_request_commit(req, lseg, cinfo); |
56f9cd68 FI |
1255 | if (!cinfo->dreq) { |
1256 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 1257 | dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
1258 | BDI_RECLAIMABLE); |
1259 | } | |
1d1afcbc | 1260 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1261 | } |
1262 | } | |
e0c2b380 | 1263 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1264 | |
1da177e4 LT |
1265 | /* |
1266 | * Commit dirty pages | |
1267 | */ | |
1268 | static int | |
ea2cf228 FI |
1269 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1270 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1271 | { |
0b7c0153 | 1272 | struct nfs_commit_data *data; |
1da177e4 | 1273 | |
c9d8f89d | 1274 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1275 | |
1276 | if (!data) | |
1277 | goto out_bad; | |
1278 | ||
1279 | /* Set up the argument struct */ | |
f453a54a FI |
1280 | nfs_init_commit(data, head, NULL, cinfo); |
1281 | atomic_inc(&cinfo->mds->rpcs_out); | |
9f0ec176 AA |
1282 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, |
1283 | how, 0); | |
1da177e4 | 1284 | out_bad: |
ea2cf228 | 1285 | nfs_retry_commit(head, NULL, cinfo); |
f453a54a | 1286 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
1da177e4 LT |
1287 | return -ENOMEM; |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * COMMIT call returned | |
1292 | */ | |
788e7a89 | 1293 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1294 | { |
0b7c0153 | 1295 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1296 | |
a3f565b1 | 1297 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1298 | task->tk_pid, task->tk_status); |
1299 | ||
788e7a89 | 1300 | /* Call the NFS version-specific code */ |
c0d0e96b | 1301 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1302 | } |
1303 | ||
f453a54a | 1304 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1305 | { |
5917ce84 | 1306 | struct nfs_page *req; |
c9d8f89d | 1307 | int status = data->task.tk_status; |
f453a54a | 1308 | struct nfs_commit_info cinfo; |
788e7a89 | 1309 | |
1da177e4 LT |
1310 | while (!list_empty(&data->pages)) { |
1311 | req = nfs_list_entry(data->pages.next); | |
1312 | nfs_list_remove_request(req); | |
d6d6dc7c | 1313 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1314 | |
1e8968c5 | 1315 | dprintk("NFS: commit (%s/%llu %d@%lld)", |
3d4ff43d | 1316 | req->wb_context->dentry->d_sb->s_id, |
1e8968c5 | 1317 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
1da177e4 LT |
1318 | req->wb_bytes, |
1319 | (long long)req_offset(req)); | |
c9d8f89d TM |
1320 | if (status < 0) { |
1321 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1322 | nfs_inode_remove_request(req); |
c9d8f89d | 1323 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1324 | goto next; |
1325 | } | |
1326 | ||
1327 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1328 | * returned by the server against all stored verfs. */ | |
2f2c63bc | 1329 | if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { |
1da177e4 LT |
1330 | /* We have a match */ |
1331 | nfs_inode_remove_request(req); | |
1332 | dprintk(" OK\n"); | |
1333 | goto next; | |
1334 | } | |
1335 | /* We have a mismatch. Write the page again */ | |
1336 | dprintk(" mismatch\n"); | |
6d884e8f | 1337 | nfs_mark_request_dirty(req); |
05990d1b | 1338 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1339 | next: |
1d1afcbc | 1340 | nfs_unlock_and_release_request(req); |
1da177e4 | 1341 | } |
f453a54a FI |
1342 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
1343 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
1344 | nfs_commit_clear_lock(NFS_I(data->inode)); | |
5917ce84 FI |
1345 | } |
1346 | ||
1347 | static void nfs_commit_release(void *calldata) | |
1348 | { | |
0b7c0153 | 1349 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1350 | |
f453a54a | 1351 | data->completion_ops->completion(data); |
c9d8f89d | 1352 | nfs_commitdata_release(calldata); |
1da177e4 | 1353 | } |
788e7a89 TM |
1354 | |
1355 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1356 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1357 | .rpc_call_done = nfs_commit_done, |
1358 | .rpc_release = nfs_commit_release, | |
1359 | }; | |
1da177e4 | 1360 | |
f453a54a FI |
1361 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1362 | .completion = nfs_commit_release_pages, | |
1363 | .error_cleanup = nfs_commit_clear_lock, | |
1364 | }; | |
1365 | ||
1763da12 FI |
1366 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1367 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1368 | { |
1369 | int status; | |
1370 | ||
ea2cf228 | 1371 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1372 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1373 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1374 | return status; |
1375 | } | |
1376 | ||
b608b283 | 1377 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1378 | { |
1da177e4 | 1379 | LIST_HEAD(head); |
ea2cf228 | 1380 | struct nfs_commit_info cinfo; |
71d0a611 | 1381 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1382 | int res; |
1da177e4 | 1383 | |
b8413f98 TM |
1384 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1385 | if (res <= 0) | |
c5efa5fc | 1386 | goto out_mark_dirty; |
ea2cf228 FI |
1387 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1388 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1389 | if (res) { |
a861a1e1 FI |
1390 | int error; |
1391 | ||
ea2cf228 | 1392 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1393 | if (error < 0) |
1394 | return error; | |
b8413f98 | 1395 | if (!may_wait) |
c5efa5fc | 1396 | goto out_mark_dirty; |
b8413f98 TM |
1397 | error = wait_on_bit(&NFS_I(inode)->flags, |
1398 | NFS_INO_COMMIT, | |
1399 | nfs_wait_bit_killable, | |
1400 | TASK_KILLABLE); | |
1401 | if (error < 0) | |
1402 | return error; | |
71d0a611 TM |
1403 | } else |
1404 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1405 | return res; |
1406 | /* Note: If we exit without ensuring that the commit is complete, | |
1407 | * we must mark the inode as dirty. Otherwise, future calls to | |
1408 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1409 | * that the data is on the disk. | |
1410 | */ | |
1411 | out_mark_dirty: | |
1412 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1413 | return res; |
1414 | } | |
8fc795f7 TM |
1415 | |
1416 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1417 | { | |
420e3646 TM |
1418 | struct nfs_inode *nfsi = NFS_I(inode); |
1419 | int flags = FLUSH_SYNC; | |
1420 | int ret = 0; | |
8fc795f7 | 1421 | |
3236c3e1 | 1422 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1423 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1424 | return ret; |
1425 | ||
a00dd6c0 JL |
1426 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1427 | /* Don't commit yet if this is a non-blocking flush and there | |
1428 | * are a lot of outstanding writes for this mapping. | |
1429 | */ | |
ea2cf228 | 1430 | if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) |
a00dd6c0 | 1431 | goto out_mark_dirty; |
420e3646 | 1432 | |
a00dd6c0 | 1433 | /* don't wait for the COMMIT response */ |
420e3646 | 1434 | flags = 0; |
a00dd6c0 JL |
1435 | } |
1436 | ||
420e3646 TM |
1437 | ret = nfs_commit_inode(inode, flags); |
1438 | if (ret >= 0) { | |
1439 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1440 | if (ret < wbc->nr_to_write) | |
1441 | wbc->nr_to_write -= ret; | |
1442 | else | |
1443 | wbc->nr_to_write = 0; | |
1444 | } | |
8fc795f7 | 1445 | return 0; |
420e3646 TM |
1446 | } |
1447 | out_mark_dirty: | |
8fc795f7 TM |
1448 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1449 | return ret; | |
1450 | } | |
c63c7b05 | 1451 | #else |
8fc795f7 TM |
1452 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1453 | { | |
1454 | return 0; | |
1455 | } | |
1da177e4 LT |
1456 | #endif |
1457 | ||
8fc795f7 TM |
1458 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1459 | { | |
a8d8f02c BS |
1460 | return nfs_commit_unstable_pages(inode, wbc); |
1461 | } | |
89d77c8f | 1462 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1463 | |
acdc53b2 TM |
1464 | /* |
1465 | * flush the inode to disk. | |
1466 | */ | |
1467 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1468 | { |
1469 | struct writeback_control wbc = { | |
72cb77f4 | 1470 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1471 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1472 | .range_start = 0, |
1473 | .range_end = LLONG_MAX, | |
34901f70 | 1474 | }; |
f4ce1299 TM |
1475 | int ret; |
1476 | ||
1477 | trace_nfs_writeback_inode_enter(inode); | |
1478 | ||
1479 | ret = sync_inode(inode, &wbc); | |
34901f70 | 1480 | |
f4ce1299 TM |
1481 | trace_nfs_writeback_inode_exit(inode, ret); |
1482 | return ret; | |
1c75950b | 1483 | } |
ddda8e0a | 1484 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1485 | |
1b3b4a1a TM |
1486 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1487 | { | |
1488 | struct nfs_page *req; | |
1b3b4a1a TM |
1489 | int ret = 0; |
1490 | ||
1b3b4a1a | 1491 | for (;;) { |
ba8b06e6 | 1492 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1493 | req = nfs_page_find_request(page); |
1494 | if (req == NULL) | |
1b3b4a1a | 1495 | break; |
7ad84aa9 | 1496 | if (nfs_lock_request(req)) { |
8dd37758 | 1497 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1498 | nfs_inode_remove_request(req); |
1499 | /* | |
1500 | * In case nfs_inode_remove_request has marked the | |
1501 | * page as being dirty | |
1502 | */ | |
1503 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1d1afcbc | 1504 | nfs_unlock_and_release_request(req); |
1b3b4a1a TM |
1505 | break; |
1506 | } | |
1507 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1508 | nfs_release_request(req); |
1b3b4a1a | 1509 | if (ret < 0) |
c988950e | 1510 | break; |
1b3b4a1a | 1511 | } |
1b3b4a1a TM |
1512 | return ret; |
1513 | } | |
1514 | ||
7f2f12d9 TM |
1515 | /* |
1516 | * Write back all requests on one page - we do this before reading it. | |
1517 | */ | |
1518 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b | 1519 | { |
29418aa4 | 1520 | loff_t range_start = page_file_offset(page); |
1c75950b | 1521 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); |
4d770ccf | 1522 | struct writeback_control wbc = { |
4d770ccf | 1523 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1524 | .nr_to_write = 0, |
4d770ccf TM |
1525 | .range_start = range_start, |
1526 | .range_end = range_end, | |
1527 | }; | |
1528 | int ret; | |
1c75950b | 1529 | |
f4ce1299 TM |
1530 | trace_nfs_writeback_page_enter(inode); |
1531 | ||
0522f6ad | 1532 | for (;;) { |
ba8b06e6 | 1533 | wait_on_page_writeback(page); |
73e3302f TM |
1534 | if (clear_page_dirty_for_io(page)) { |
1535 | ret = nfs_writepage_locked(page, &wbc); | |
1536 | if (ret < 0) | |
1537 | goto out_error; | |
0522f6ad | 1538 | continue; |
7f2f12d9 | 1539 | } |
f4ce1299 | 1540 | ret = 0; |
0522f6ad TM |
1541 | if (!PagePrivate(page)) |
1542 | break; | |
1543 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1544 | if (ret < 0) |
73e3302f | 1545 | goto out_error; |
7f2f12d9 | 1546 | } |
73e3302f | 1547 | out_error: |
f4ce1299 | 1548 | trace_nfs_writeback_page_exit(inode, ret); |
4d770ccf | 1549 | return ret; |
1c75950b TM |
1550 | } |
1551 | ||
074cc1de TM |
1552 | #ifdef CONFIG_MIGRATION |
1553 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1554 | struct page *page, enum migrate_mode mode) |
074cc1de | 1555 | { |
2da95652 JL |
1556 | /* |
1557 | * If PagePrivate is set, then the page is currently associated with | |
1558 | * an in-progress read or write request. Don't try to migrate it. | |
1559 | * | |
1560 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1561 | * that we can safely release the inode reference while holding | |
1562 | * the page lock. | |
1563 | */ | |
1564 | if (PagePrivate(page)) | |
1565 | return -EBUSY; | |
074cc1de | 1566 | |
8c209ce7 DH |
1567 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
1568 | return -EBUSY; | |
074cc1de | 1569 | |
a6bc32b8 | 1570 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1571 | } |
1572 | #endif | |
1573 | ||
f7b422b1 | 1574 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1575 | { |
1576 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
c0752cdf | 1577 | sizeof(struct nfs_rw_header), |
1da177e4 | 1578 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1579 | NULL); |
1da177e4 LT |
1580 | if (nfs_wdata_cachep == NULL) |
1581 | return -ENOMEM; | |
1582 | ||
93d2341c MD |
1583 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1584 | nfs_wdata_cachep); | |
1da177e4 | 1585 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 1586 | goto out_destroy_write_cache; |
1da177e4 | 1587 | |
0b7c0153 FI |
1588 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1589 | sizeof(struct nfs_commit_data), | |
1590 | 0, SLAB_HWCACHE_ALIGN, | |
1591 | NULL); | |
1592 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 1593 | goto out_destroy_write_mempool; |
0b7c0153 | 1594 | |
93d2341c | 1595 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 1596 | nfs_cdata_cachep); |
1da177e4 | 1597 | if (nfs_commit_mempool == NULL) |
3dd4765f | 1598 | goto out_destroy_commit_cache; |
1da177e4 | 1599 | |
89a09141 PZ |
1600 | /* |
1601 | * NFS congestion size, scale with available memory. | |
1602 | * | |
1603 | * 64MB: 8192k | |
1604 | * 128MB: 11585k | |
1605 | * 256MB: 16384k | |
1606 | * 512MB: 23170k | |
1607 | * 1GB: 32768k | |
1608 | * 2GB: 46340k | |
1609 | * 4GB: 65536k | |
1610 | * 8GB: 92681k | |
1611 | * 16GB: 131072k | |
1612 | * | |
1613 | * This allows larger machines to have larger/more transfers. | |
1614 | * Limit the default to 256M | |
1615 | */ | |
1616 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1617 | if (nfs_congestion_kb > 256*1024) | |
1618 | nfs_congestion_kb = 256*1024; | |
1619 | ||
1da177e4 | 1620 | return 0; |
3dd4765f JL |
1621 | |
1622 | out_destroy_commit_cache: | |
1623 | kmem_cache_destroy(nfs_cdata_cachep); | |
1624 | out_destroy_write_mempool: | |
1625 | mempool_destroy(nfs_wdata_mempool); | |
1626 | out_destroy_write_cache: | |
1627 | kmem_cache_destroy(nfs_wdata_cachep); | |
1628 | return -ENOMEM; | |
1da177e4 LT |
1629 | } |
1630 | ||
266bee88 | 1631 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1632 | { |
1633 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 1634 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 1635 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 1636 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1637 | } |
1638 | ||
4a0de55c | 1639 | static const struct nfs_rw_ops nfs_rw_write_ops = { |
a4cdda59 | 1640 | .rw_mode = FMODE_WRITE, |
4a0de55c AS |
1641 | .rw_alloc_header = nfs_writehdr_alloc, |
1642 | .rw_free_header = nfs_writehdr_free, | |
a4cdda59 | 1643 | .rw_release = nfs_writeback_release_common, |
0eecb214 AS |
1644 | .rw_done = nfs_writeback_done, |
1645 | .rw_result = nfs_writeback_result, | |
1ed26f33 | 1646 | .rw_initiate = nfs_initiate_write, |
4a0de55c | 1647 | }; |