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 LT |
33 | |
34 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE | |
35 | ||
36 | #define MIN_POOL_WRITE (32) | |
37 | #define MIN_POOL_COMMIT (4) | |
38 | ||
39 | /* | |
40 | * Local function declarations | |
41 | */ | |
c63c7b05 TM |
42 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, |
43 | struct inode *inode, int ioflags); | |
f8512ad0 | 44 | static void nfs_redirty_request(struct nfs_page *req); |
6c75dc0d | 45 | static const struct rpc_call_ops nfs_write_common_ops; |
788e7a89 | 46 | static const struct rpc_call_ops nfs_commit_ops; |
1da177e4 | 47 | |
e18b890b | 48 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 49 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 50 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
51 | static mempool_t *nfs_commit_mempool; |
52 | ||
0b7c0153 | 53 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 54 | { |
0b7c0153 | 55 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); |
40859d7e | 56 | |
1da177e4 LT |
57 | if (p) { |
58 | memset(p, 0, sizeof(*p)); | |
59 | INIT_LIST_HEAD(&p->pages); | |
60 | } | |
61 | return p; | |
62 | } | |
e0c2b380 | 63 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 64 | |
0b7c0153 | 65 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
66 | { |
67 | mempool_free(p, nfs_commit_mempool); | |
68 | } | |
e0c2b380 | 69 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 70 | |
6c75dc0d | 71 | struct nfs_write_header *nfs_writehdr_alloc(void) |
3feb2d49 | 72 | { |
cd841605 | 73 | struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); |
3feb2d49 TM |
74 | |
75 | if (p) { | |
cd841605 | 76 | struct nfs_pgio_header *hdr = &p->header; |
cd841605 | 77 | |
3feb2d49 | 78 | memset(p, 0, sizeof(*p)); |
cd841605 | 79 | INIT_LIST_HEAD(&hdr->pages); |
6c75dc0d FI |
80 | INIT_LIST_HEAD(&hdr->rpc_list); |
81 | spin_lock_init(&hdr->lock); | |
82 | atomic_set(&hdr->refcnt, 0); | |
3feb2d49 TM |
83 | } |
84 | return p; | |
85 | } | |
86 | ||
6c75dc0d FI |
87 | struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr, |
88 | unsigned int pagecount) | |
89 | { | |
90 | struct nfs_write_data *data, *prealloc; | |
91 | ||
92 | prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data; | |
93 | if (prealloc->header == NULL) | |
94 | data = prealloc; | |
95 | else | |
96 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
97 | if (!data) | |
98 | goto out; | |
99 | ||
100 | if (nfs_pgarray_set(&data->pages, pagecount)) { | |
101 | data->header = hdr; | |
102 | atomic_inc(&hdr->refcnt); | |
103 | } else { | |
104 | if (data != prealloc) | |
105 | kfree(data); | |
106 | data = NULL; | |
107 | } | |
108 | out: | |
109 | return data; | |
110 | } | |
111 | ||
cd841605 | 112 | void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 113 | { |
cd841605 FI |
114 | struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header); |
115 | mempool_free(whdr, nfs_wdata_mempool); | |
3feb2d49 TM |
116 | } |
117 | ||
dce81290 | 118 | void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 119 | { |
6c75dc0d FI |
120 | struct nfs_pgio_header *hdr = wdata->header; |
121 | struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header); | |
122 | ||
383ba719 | 123 | put_nfs_open_context(wdata->args.context); |
30dd374f FI |
124 | if (wdata->pages.pagevec != wdata->pages.page_array) |
125 | kfree(wdata->pages.pagevec); | |
6c75dc0d FI |
126 | if (wdata != &write_header->rpc_data) |
127 | kfree(wdata); | |
128 | else | |
129 | wdata->header = NULL; | |
130 | if (atomic_dec_and_test(&hdr->refcnt)) | |
131 | nfs_write_completion(hdr); | |
1da177e4 LT |
132 | } |
133 | ||
7b159fc1 TM |
134 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
135 | { | |
136 | ctx->error = error; | |
137 | smp_wmb(); | |
138 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
139 | } | |
140 | ||
277459d2 TM |
141 | static struct nfs_page *nfs_page_find_request_locked(struct page *page) |
142 | { | |
143 | struct nfs_page *req = NULL; | |
144 | ||
145 | if (PagePrivate(page)) { | |
146 | req = (struct nfs_page *)page_private(page); | |
147 | if (req != NULL) | |
c03b4024 | 148 | kref_get(&req->wb_kref); |
277459d2 TM |
149 | } |
150 | return req; | |
151 | } | |
152 | ||
153 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
154 | { | |
587142f8 | 155 | struct inode *inode = page->mapping->host; |
277459d2 | 156 | struct nfs_page *req = NULL; |
277459d2 | 157 | |
587142f8 | 158 | spin_lock(&inode->i_lock); |
277459d2 | 159 | req = nfs_page_find_request_locked(page); |
587142f8 | 160 | spin_unlock(&inode->i_lock); |
277459d2 TM |
161 | return req; |
162 | } | |
163 | ||
1da177e4 LT |
164 | /* Adjust the file length if we're writing beyond the end */ |
165 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
166 | { | |
167 | struct inode *inode = page->mapping->host; | |
a3d01454 TM |
168 | loff_t end, i_size; |
169 | pgoff_t end_index; | |
1da177e4 | 170 | |
a3d01454 TM |
171 | spin_lock(&inode->i_lock); |
172 | i_size = i_size_read(inode); | |
173 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 | 174 | if (i_size > 0 && page->index < end_index) |
a3d01454 | 175 | goto out; |
1da177e4 LT |
176 | end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); |
177 | if (i_size >= end) | |
a3d01454 | 178 | goto out; |
1da177e4 | 179 | i_size_write(inode, end); |
a3d01454 TM |
180 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
181 | out: | |
182 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
183 | } |
184 | ||
a301b777 TM |
185 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
186 | static void nfs_set_pageerror(struct page *page) | |
187 | { | |
188 | SetPageError(page); | |
189 | nfs_zap_mapping(page->mapping->host, page->mapping); | |
190 | } | |
191 | ||
1da177e4 LT |
192 | /* We can set the PG_uptodate flag if we see that a write request |
193 | * covers the full page. | |
194 | */ | |
195 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
196 | { | |
1da177e4 LT |
197 | if (PageUptodate(page)) |
198 | return; | |
199 | if (base != 0) | |
200 | return; | |
49a70f27 | 201 | if (count != nfs_page_length(page)) |
1da177e4 | 202 | return; |
49a70f27 | 203 | SetPageUptodate(page); |
1da177e4 LT |
204 | } |
205 | ||
1da177e4 LT |
206 | static int wb_priority(struct writeback_control *wbc) |
207 | { | |
208 | if (wbc->for_reclaim) | |
c63c7b05 | 209 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 210 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
211 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
212 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
213 | } |
214 | ||
89a09141 PZ |
215 | /* |
216 | * NFS congestion control | |
217 | */ | |
218 | ||
219 | int nfs_congestion_kb; | |
220 | ||
221 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
222 | #define NFS_CONGESTION_OFF_THRESH \ | |
223 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
224 | ||
5a6d41b3 | 225 | static int nfs_set_page_writeback(struct page *page) |
89a09141 | 226 | { |
5a6d41b3 TM |
227 | int ret = test_set_page_writeback(page); |
228 | ||
229 | if (!ret) { | |
89a09141 PZ |
230 | struct inode *inode = page->mapping->host; |
231 | struct nfs_server *nfss = NFS_SERVER(inode); | |
232 | ||
a6305ddb | 233 | page_cache_get(page); |
277866a0 | 234 | if (atomic_long_inc_return(&nfss->writeback) > |
8aa7e847 JA |
235 | NFS_CONGESTION_ON_THRESH) { |
236 | set_bdi_congested(&nfss->backing_dev_info, | |
237 | BLK_RW_ASYNC); | |
238 | } | |
89a09141 | 239 | } |
5a6d41b3 | 240 | return ret; |
89a09141 PZ |
241 | } |
242 | ||
243 | static void nfs_end_page_writeback(struct page *page) | |
244 | { | |
245 | struct inode *inode = page->mapping->host; | |
246 | struct nfs_server *nfss = NFS_SERVER(inode); | |
247 | ||
248 | end_page_writeback(page); | |
a6305ddb | 249 | page_cache_release(page); |
c4dc4bee | 250 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 251 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
252 | } |
253 | ||
cfb506e1 | 254 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 255 | { |
587142f8 | 256 | struct inode *inode = page->mapping->host; |
e261f51f | 257 | struct nfs_page *req; |
e261f51f TM |
258 | int ret; |
259 | ||
587142f8 | 260 | spin_lock(&inode->i_lock); |
074cc1de | 261 | for (;;) { |
e261f51f | 262 | req = nfs_page_find_request_locked(page); |
074cc1de TM |
263 | if (req == NULL) |
264 | break; | |
9994b62b | 265 | if (nfs_lock_request_dontget(req)) |
e261f51f TM |
266 | break; |
267 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
9994b62b | 268 | * then the call to nfs_lock_request_dontget() will always |
e261f51f TM |
269 | * succeed provided that someone hasn't already marked the |
270 | * request as dirty (in which case we don't care). | |
271 | */ | |
587142f8 | 272 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
273 | if (!nonblock) |
274 | ret = nfs_wait_on_request(req); | |
275 | else | |
276 | ret = -EAGAIN; | |
e261f51f TM |
277 | nfs_release_request(req); |
278 | if (ret != 0) | |
074cc1de | 279 | return ERR_PTR(ret); |
587142f8 | 280 | spin_lock(&inode->i_lock); |
e261f51f | 281 | } |
587142f8 | 282 | spin_unlock(&inode->i_lock); |
074cc1de TM |
283 | return req; |
284 | } | |
285 | ||
286 | /* | |
287 | * Find an associated nfs write request, and prepare to flush it out | |
288 | * May return an error if the user signalled nfs_wait_on_request(). | |
289 | */ | |
290 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 291 | struct page *page, bool nonblock) |
074cc1de TM |
292 | { |
293 | struct nfs_page *req; | |
294 | int ret = 0; | |
295 | ||
cfb506e1 | 296 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
297 | if (!req) |
298 | goto out; | |
299 | ret = PTR_ERR(req); | |
300 | if (IS_ERR(req)) | |
301 | goto out; | |
302 | ||
303 | ret = nfs_set_page_writeback(page); | |
304 | BUG_ON(ret != 0); | |
305 | BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); | |
306 | ||
f8512ad0 FI |
307 | if (!nfs_pageio_add_request(pgio, req)) { |
308 | nfs_redirty_request(req); | |
074cc1de | 309 | ret = pgio->pg_error; |
f8512ad0 | 310 | } |
074cc1de TM |
311 | out: |
312 | return ret; | |
e261f51f TM |
313 | } |
314 | ||
f758c885 | 315 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 316 | { |
1da177e4 | 317 | struct inode *inode = page->mapping->host; |
cfb506e1 | 318 | int ret; |
1da177e4 | 319 | |
91d5b470 CL |
320 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
321 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
322 | ||
7fe7f848 | 323 | nfs_pageio_cond_complete(pgio, page->index); |
1b430bee | 324 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
325 | if (ret == -EAGAIN) { |
326 | redirty_page_for_writepage(wbc, page); | |
327 | ret = 0; | |
328 | } | |
329 | return ret; | |
f758c885 | 330 | } |
7fe7f848 | 331 | |
f758c885 TM |
332 | /* |
333 | * Write an mmapped page to the server. | |
334 | */ | |
335 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
336 | { | |
337 | struct nfs_pageio_descriptor pgio; | |
338 | int err; | |
49a70f27 | 339 | |
f758c885 TM |
340 | nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); |
341 | err = nfs_do_writepage(page, wbc, &pgio); | |
342 | nfs_pageio_complete(&pgio); | |
343 | if (err < 0) | |
344 | return err; | |
345 | if (pgio.pg_error < 0) | |
346 | return pgio.pg_error; | |
347 | return 0; | |
4d770ccf TM |
348 | } |
349 | ||
350 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
351 | { | |
f758c885 | 352 | int ret; |
4d770ccf | 353 | |
f758c885 | 354 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 355 | unlock_page(page); |
f758c885 TM |
356 | return ret; |
357 | } | |
358 | ||
359 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
360 | { | |
361 | int ret; | |
362 | ||
363 | ret = nfs_do_writepage(page, wbc, data); | |
364 | unlock_page(page); | |
365 | return ret; | |
1da177e4 LT |
366 | } |
367 | ||
1da177e4 LT |
368 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
369 | { | |
1da177e4 | 370 | struct inode *inode = mapping->host; |
72cb77f4 | 371 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 372 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
373 | int err; |
374 | ||
72cb77f4 TM |
375 | /* Stop dirtying of new pages while we sync */ |
376 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
377 | nfs_wait_bit_killable, TASK_KILLABLE); | |
378 | if (err) | |
379 | goto out_err; | |
380 | ||
91d5b470 CL |
381 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
382 | ||
c63c7b05 | 383 | nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); |
f758c885 | 384 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 385 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
386 | |
387 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
388 | smp_mb__after_clear_bit(); | |
389 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
390 | ||
f758c885 | 391 | if (err < 0) |
72cb77f4 TM |
392 | goto out_err; |
393 | err = pgio.pg_error; | |
394 | if (err < 0) | |
395 | goto out_err; | |
c63c7b05 | 396 | return 0; |
72cb77f4 TM |
397 | out_err: |
398 | return err; | |
1da177e4 LT |
399 | } |
400 | ||
401 | /* | |
402 | * Insert a write request into an inode | |
403 | */ | |
d6d6dc7c | 404 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
405 | { |
406 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 TM |
407 | |
408 | /* Lock the request! */ | |
409 | nfs_lock_request_dontget(req); | |
410 | ||
411 | spin_lock(&inode->i_lock); | |
4d65c520 | 412 | if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 413 | inode->i_version++; |
2df485a7 | 414 | set_bit(PG_MAPPED, &req->wb_flags); |
deb7d638 | 415 | SetPagePrivate(req->wb_page); |
277459d2 | 416 | set_page_private(req->wb_page, (unsigned long)req); |
1da177e4 | 417 | nfsi->npages++; |
c03b4024 | 418 | kref_get(&req->wb_kref); |
e7d39069 | 419 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
420 | } |
421 | ||
422 | /* | |
89a09141 | 423 | * Remove a write request from an inode |
1da177e4 LT |
424 | */ |
425 | static void nfs_inode_remove_request(struct nfs_page *req) | |
426 | { | |
3d4ff43d | 427 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
428 | struct nfs_inode *nfsi = NFS_I(inode); |
429 | ||
430 | BUG_ON (!NFS_WBACK_BUSY(req)); | |
431 | ||
587142f8 | 432 | spin_lock(&inode->i_lock); |
277459d2 | 433 | set_page_private(req->wb_page, 0); |
deb7d638 | 434 | ClearPagePrivate(req->wb_page); |
2df485a7 | 435 | clear_bit(PG_MAPPED, &req->wb_flags); |
1da177e4 | 436 | nfsi->npages--; |
4d65c520 | 437 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
438 | nfs_release_request(req); |
439 | } | |
440 | ||
61822ab5 | 441 | static void |
6d884e8f | 442 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 443 | { |
61822ab5 TM |
444 | __set_page_dirty_nobuffers(req->wb_page); |
445 | } | |
446 | ||
1da177e4 | 447 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
8dd37758 TM |
448 | /** |
449 | * nfs_request_add_commit_list - add request to a commit list | |
450 | * @req: pointer to a struct nfs_page | |
451 | * @head: commit list head | |
452 | * | |
453 | * This sets the PG_CLEAN bit, updates the inode global count of | |
454 | * number of outstanding requests requiring a commit as well as | |
455 | * the MM page stats. | |
456 | * | |
457 | * The caller must _not_ hold the inode->i_lock, but must be | |
458 | * holding the nfs_page lock. | |
1da177e4 | 459 | */ |
8dd37758 TM |
460 | void |
461 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *head) | |
1da177e4 | 462 | { |
3d4ff43d | 463 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 | 464 | |
e468bae9 | 465 | set_bit(PG_CLEAN, &(req)->wb_flags); |
8dd37758 TM |
466 | spin_lock(&inode->i_lock); |
467 | nfs_list_add_request(req, head); | |
468 | NFS_I(inode)->ncommit++; | |
587142f8 | 469 | spin_unlock(&inode->i_lock); |
fd39fc85 | 470 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
c9e51e41 | 471 | inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); |
a1803044 | 472 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1da177e4 | 473 | } |
8dd37758 TM |
474 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
475 | ||
476 | /** | |
477 | * nfs_request_remove_commit_list - Remove request from a commit list | |
478 | * @req: pointer to a nfs_page | |
479 | * | |
480 | * This clears the PG_CLEAN bit, and updates the inode global count of | |
481 | * number of outstanding requests requiring a commit | |
482 | * It does not update the MM page stats. | |
483 | * | |
484 | * The caller _must_ hold the inode->i_lock and the nfs_page lock. | |
485 | */ | |
486 | void | |
487 | nfs_request_remove_commit_list(struct nfs_page *req) | |
488 | { | |
489 | struct inode *inode = req->wb_context->dentry->d_inode; | |
490 | ||
491 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) | |
492 | return; | |
493 | nfs_list_remove_request(req); | |
494 | NFS_I(inode)->ncommit--; | |
495 | } | |
496 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
497 | ||
498 | ||
499 | /* | |
500 | * Add a request to the inode's commit list. | |
501 | */ | |
502 | static void | |
503 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) | |
504 | { | |
505 | struct inode *inode = req->wb_context->dentry->d_inode; | |
506 | ||
507 | if (pnfs_mark_request_commit(req, lseg)) | |
508 | return; | |
509 | nfs_request_add_commit_list(req, &NFS_I(inode)->commit_list); | |
510 | } | |
8e821cad | 511 | |
d6d6dc7c FI |
512 | static void |
513 | nfs_clear_page_commit(struct page *page) | |
514 | { | |
515 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
516 | dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); | |
517 | } | |
518 | ||
8dd37758 | 519 | static void |
e468bae9 TM |
520 | nfs_clear_request_commit(struct nfs_page *req) |
521 | { | |
8dd37758 TM |
522 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
523 | struct inode *inode = req->wb_context->dentry->d_inode; | |
e468bae9 | 524 | |
8dd37758 TM |
525 | if (!pnfs_clear_request_commit(req)) { |
526 | spin_lock(&inode->i_lock); | |
527 | nfs_request_remove_commit_list(req); | |
528 | spin_unlock(&inode->i_lock); | |
529 | } | |
d6d6dc7c | 530 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 531 | } |
e468bae9 TM |
532 | } |
533 | ||
8e821cad TM |
534 | static inline |
535 | int nfs_write_need_commit(struct nfs_write_data *data) | |
536 | { | |
465d5243 | 537 | if (data->verf.committed == NFS_DATA_SYNC) |
cd841605 FI |
538 | return data->header->lseg == NULL; |
539 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
540 | } |
541 | ||
8e821cad | 542 | #else |
8dd37758 | 543 | static void |
a861a1e1 | 544 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) |
8e821cad TM |
545 | { |
546 | } | |
547 | ||
8dd37758 | 548 | static void |
e468bae9 TM |
549 | nfs_clear_request_commit(struct nfs_page *req) |
550 | { | |
e468bae9 TM |
551 | } |
552 | ||
8e821cad TM |
553 | static inline |
554 | int nfs_write_need_commit(struct nfs_write_data *data) | |
555 | { | |
556 | return 0; | |
557 | } | |
558 | ||
6c75dc0d FI |
559 | #endif |
560 | ||
561 | void nfs_write_completion(struct nfs_pgio_header *hdr) | |
8e821cad | 562 | { |
6c75dc0d FI |
563 | unsigned long bytes = 0; |
564 | ||
565 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
566 | goto out; | |
567 | while (!list_empty(&hdr->pages)) { | |
568 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
569 | struct page *page = req->wb_page; | |
570 | ||
571 | bytes += req->wb_bytes; | |
572 | nfs_list_remove_request(req); | |
573 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
574 | (hdr->good_bytes < bytes)) { | |
575 | nfs_set_pageerror(page); | |
576 | nfs_context_set_write_error(req->wb_context, hdr->error); | |
577 | goto remove_req; | |
578 | } | |
579 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
580 | nfs_mark_request_dirty(req); | |
581 | goto next; | |
582 | } | |
583 | if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
584 | nfs_mark_request_commit(req, hdr->lseg); | |
585 | goto next; | |
586 | } | |
587 | remove_req: | |
588 | nfs_inode_remove_request(req); | |
589 | next: | |
590 | nfs_unlock_request(req); | |
591 | nfs_end_page_writeback(page); | |
592 | } | |
593 | out: | |
594 | hdr->release(hdr); | |
8e821cad | 595 | } |
1da177e4 | 596 | |
47c62564 | 597 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
fb8a1f11 TM |
598 | static int |
599 | nfs_need_commit(struct nfs_inode *nfsi) | |
600 | { | |
d6d6dc7c FI |
601 | return nfsi->ncommit > 0; |
602 | } | |
603 | ||
604 | /* i_lock held by caller */ | |
8dd37758 | 605 | static int |
3b3be88d TM |
606 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, int max, |
607 | spinlock_t *lock) | |
d6d6dc7c FI |
608 | { |
609 | struct nfs_page *req, *tmp; | |
610 | int ret = 0; | |
611 | ||
612 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
613 | if (!nfs_lock_request(req)) |
614 | continue; | |
3b3be88d TM |
615 | if (cond_resched_lock(lock)) |
616 | list_safe_reset_next(req, tmp, wb_list); | |
8dd37758 TM |
617 | nfs_request_remove_commit_list(req); |
618 | nfs_list_add_request(req, dst); | |
619 | ret++; | |
620 | if (ret == max) | |
621 | break; | |
d6d6dc7c FI |
622 | } |
623 | return ret; | |
fb8a1f11 TM |
624 | } |
625 | ||
1da177e4 LT |
626 | /* |
627 | * nfs_scan_commit - Scan an inode for commit requests | |
628 | * @inode: NFS inode to scan | |
629 | * @dst: destination list | |
1da177e4 LT |
630 | * |
631 | * Moves requests from the inode's 'commit' request list. | |
632 | * The requests are *not* checked to ensure that they form a contiguous set. | |
633 | */ | |
634 | static int | |
d6d6dc7c | 635 | nfs_scan_commit(struct inode *inode, struct list_head *dst) |
1da177e4 LT |
636 | { |
637 | struct nfs_inode *nfsi = NFS_I(inode); | |
d6d6dc7c | 638 | int ret = 0; |
fb8a1f11 | 639 | |
0d88f6e8 | 640 | spin_lock(&inode->i_lock); |
d6d6dc7c | 641 | if (nfsi->ncommit > 0) { |
8dd37758 | 642 | const int max = INT_MAX; |
d6d6dc7c | 643 | |
3b3be88d TM |
644 | ret = nfs_scan_commit_list(&nfsi->commit_list, dst, max, |
645 | &inode->i_lock); | |
646 | ret += pnfs_scan_commit_lists(inode, max - ret, | |
647 | &inode->i_lock); | |
d6d6dc7c | 648 | } |
0d88f6e8 | 649 | spin_unlock(&inode->i_lock); |
ff778d02 | 650 | return ret; |
1da177e4 | 651 | } |
d6d6dc7c | 652 | |
c42de9dd | 653 | #else |
fb8a1f11 TM |
654 | static inline int nfs_need_commit(struct nfs_inode *nfsi) |
655 | { | |
656 | return 0; | |
657 | } | |
658 | ||
d6d6dc7c | 659 | static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst) |
c42de9dd TM |
660 | { |
661 | return 0; | |
662 | } | |
1da177e4 LT |
663 | #endif |
664 | ||
1da177e4 | 665 | /* |
e7d39069 TM |
666 | * Search for an existing write request, and attempt to update |
667 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 668 | * |
e7d39069 TM |
669 | * If the attempt fails, then the existing request is flushed out |
670 | * to disk. | |
1da177e4 | 671 | */ |
e7d39069 TM |
672 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
673 | struct page *page, | |
674 | unsigned int offset, | |
675 | unsigned int bytes) | |
1da177e4 | 676 | { |
e7d39069 TM |
677 | struct nfs_page *req; |
678 | unsigned int rqend; | |
679 | unsigned int end; | |
680 | int error; | |
681 | ||
682 | if (!PagePrivate(page)) | |
683 | return NULL; | |
1da177e4 LT |
684 | |
685 | end = offset + bytes; | |
e7d39069 | 686 | spin_lock(&inode->i_lock); |
1da177e4 | 687 | |
1da177e4 | 688 | for (;;) { |
277459d2 | 689 | req = nfs_page_find_request_locked(page); |
e7d39069 TM |
690 | if (req == NULL) |
691 | goto out_unlock; | |
692 | ||
693 | rqend = req->wb_offset + req->wb_bytes; | |
694 | /* | |
695 | * Tell the caller to flush out the request if | |
696 | * the offsets are non-contiguous. | |
697 | * Note: nfs_flush_incompatible() will already | |
698 | * have flushed out requests having wrong owners. | |
699 | */ | |
e468bae9 | 700 | if (offset > rqend |
e7d39069 TM |
701 | || end < req->wb_offset) |
702 | goto out_flushme; | |
703 | ||
9994b62b | 704 | if (nfs_lock_request_dontget(req)) |
1da177e4 | 705 | break; |
1da177e4 | 706 | |
e7d39069 | 707 | /* The request is locked, so wait and then retry */ |
587142f8 | 708 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
709 | error = nfs_wait_on_request(req); |
710 | nfs_release_request(req); | |
711 | if (error != 0) | |
712 | goto out_err; | |
713 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
714 | } |
715 | ||
716 | /* Okay, the request matches. Update the region */ | |
717 | if (offset < req->wb_offset) { | |
718 | req->wb_offset = offset; | |
719 | req->wb_pgbase = offset; | |
1da177e4 | 720 | } |
1da177e4 LT |
721 | if (end > rqend) |
722 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
723 | else |
724 | req->wb_bytes = rqend - req->wb_offset; | |
725 | out_unlock: | |
726 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
727 | if (req) |
728 | nfs_clear_request_commit(req); | |
e7d39069 TM |
729 | return req; |
730 | out_flushme: | |
731 | spin_unlock(&inode->i_lock); | |
732 | nfs_release_request(req); | |
733 | error = nfs_wb_page(inode, page); | |
734 | out_err: | |
735 | return ERR_PTR(error); | |
736 | } | |
737 | ||
738 | /* | |
739 | * Try to update an existing write request, or create one if there is none. | |
740 | * | |
741 | * Note: Should always be called with the Page Lock held to prevent races | |
742 | * if we have to add a new request. Also assumes that the caller has | |
743 | * already called nfs_flush_incompatible() if necessary. | |
744 | */ | |
745 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
746 | struct page *page, unsigned int offset, unsigned int bytes) | |
747 | { | |
748 | struct inode *inode = page->mapping->host; | |
749 | struct nfs_page *req; | |
1da177e4 | 750 | |
e7d39069 TM |
751 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
752 | if (req != NULL) | |
753 | goto out; | |
754 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
755 | if (IS_ERR(req)) | |
756 | goto out; | |
d6d6dc7c | 757 | nfs_inode_add_request(inode, req); |
efc91ed0 | 758 | out: |
61e930a9 | 759 | return req; |
1da177e4 LT |
760 | } |
761 | ||
e7d39069 TM |
762 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
763 | unsigned int offset, unsigned int count) | |
764 | { | |
765 | struct nfs_page *req; | |
766 | ||
767 | req = nfs_setup_write_request(ctx, page, offset, count); | |
768 | if (IS_ERR(req)) | |
769 | return PTR_ERR(req); | |
770 | /* Update file length */ | |
771 | nfs_grow_file(page, offset, count); | |
772 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 773 | nfs_mark_request_dirty(req); |
9994b62b | 774 | nfs_unlock_request(req); |
e7d39069 TM |
775 | return 0; |
776 | } | |
777 | ||
1da177e4 LT |
778 | int nfs_flush_incompatible(struct file *file, struct page *page) |
779 | { | |
cd3758e3 | 780 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 781 | struct nfs_page *req; |
1a54533e | 782 | int do_flush, status; |
1da177e4 LT |
783 | /* |
784 | * Look for a request corresponding to this page. If there | |
785 | * is one, and it belongs to another file, we flush it out | |
786 | * before we try to copy anything into the page. Do this | |
787 | * due to the lack of an ACCESS-type call in NFSv2. | |
788 | * Also do the same if we find a request from an existing | |
789 | * dropped page. | |
790 | */ | |
1a54533e TM |
791 | do { |
792 | req = nfs_page_find_request(page); | |
793 | if (req == NULL) | |
794 | return 0; | |
f11ac8db TM |
795 | do_flush = req->wb_page != page || req->wb_context != ctx || |
796 | req->wb_lock_context->lockowner != current->files || | |
797 | req->wb_lock_context->pid != current->tgid; | |
1da177e4 | 798 | nfs_release_request(req); |
1a54533e TM |
799 | if (!do_flush) |
800 | return 0; | |
801 | status = nfs_wb_page(page->mapping->host, page); | |
802 | } while (status == 0); | |
803 | return status; | |
1da177e4 LT |
804 | } |
805 | ||
5d47a356 TM |
806 | /* |
807 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
808 | * the PageUptodate() flag. In this case, we will need to turn off | |
809 | * write optimisations that depend on the page contents being correct. | |
810 | */ | |
811 | static int nfs_write_pageuptodate(struct page *page, struct inode *inode) | |
812 | { | |
813 | return PageUptodate(page) && | |
814 | !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); | |
815 | } | |
816 | ||
1da177e4 LT |
817 | /* |
818 | * Update and possibly write a cached page of an NFS file. | |
819 | * | |
820 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
821 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
822 | */ | |
823 | int nfs_updatepage(struct file *file, struct page *page, | |
824 | unsigned int offset, unsigned int count) | |
825 | { | |
cd3758e3 | 826 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 827 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
828 | int status = 0; |
829 | ||
91d5b470 CL |
830 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
831 | ||
48186c7d | 832 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
833 | file->f_path.dentry->d_parent->d_name.name, |
834 | file->f_path.dentry->d_name.name, count, | |
48186c7d | 835 | (long long)(page_offset(page) + offset)); |
1da177e4 | 836 | |
1da177e4 | 837 | /* If we're not using byte range locks, and we know the page |
5d47a356 TM |
838 | * is up to date, it may be more efficient to extend the write |
839 | * to cover the entire page in order to avoid fragmentation | |
840 | * inefficiencies. | |
1da177e4 | 841 | */ |
5d47a356 TM |
842 | if (nfs_write_pageuptodate(page, inode) && |
843 | inode->i_flock == NULL && | |
6b2f3d1f | 844 | !(file->f_flags & O_DSYNC)) { |
49a70f27 | 845 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 846 | offset = 0; |
1da177e4 LT |
847 | } |
848 | ||
e21195a7 | 849 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
850 | if (status < 0) |
851 | nfs_set_pageerror(page); | |
59b7c05f TM |
852 | else |
853 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 854 | |
48186c7d | 855 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 856 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
857 | return status; |
858 | } | |
859 | ||
3ff7576d | 860 | static int flush_task_priority(int how) |
1da177e4 LT |
861 | { |
862 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
863 | case FLUSH_HIGHPRI: | |
864 | return RPC_PRIORITY_HIGH; | |
865 | case FLUSH_LOWPRI: | |
866 | return RPC_PRIORITY_LOW; | |
867 | } | |
868 | return RPC_PRIORITY_NORMAL; | |
869 | } | |
870 | ||
c5996c4e FI |
871 | int nfs_initiate_write(struct rpc_clnt *clnt, |
872 | struct nfs_write_data *data, | |
d138d5d1 AA |
873 | const struct rpc_call_ops *call_ops, |
874 | int how) | |
1da177e4 | 875 | { |
cd841605 | 876 | struct inode *inode = data->header->inode; |
3ff7576d | 877 | int priority = flush_task_priority(how); |
07737691 | 878 | struct rpc_task *task; |
bdc7f021 TM |
879 | struct rpc_message msg = { |
880 | .rpc_argp = &data->args, | |
881 | .rpc_resp = &data->res, | |
cd841605 | 882 | .rpc_cred = data->header->cred, |
bdc7f021 | 883 | }; |
84115e1c | 884 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 885 | .rpc_client = clnt, |
07737691 | 886 | .task = &data->task, |
bdc7f021 | 887 | .rpc_message = &msg, |
84115e1c TM |
888 | .callback_ops = call_ops, |
889 | .callback_data = data, | |
101070ca | 890 | .workqueue = nfsiod_workqueue, |
2c61be0a | 891 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 892 | .priority = priority, |
84115e1c | 893 | }; |
2c61be0a | 894 | int ret = 0; |
1da177e4 | 895 | |
d138d5d1 AA |
896 | /* Set up the initial task struct. */ |
897 | NFS_PROTO(inode)->write_setup(data, &msg); | |
898 | ||
899 | dprintk("NFS: %5u initiated write call " | |
900 | "(req %s/%lld, %u bytes @ offset %llu)\n", | |
901 | data->task.tk_pid, | |
902 | inode->i_sb->s_id, | |
903 | (long long)NFS_FILEID(inode), | |
904 | data->args.count, | |
905 | (unsigned long long)data->args.offset); | |
906 | ||
907 | task = rpc_run_task(&task_setup_data); | |
908 | if (IS_ERR(task)) { | |
909 | ret = PTR_ERR(task); | |
910 | goto out; | |
911 | } | |
912 | if (how & FLUSH_SYNC) { | |
913 | ret = rpc_wait_for_completion_task(task); | |
914 | if (ret == 0) | |
915 | ret = task->tk_status; | |
916 | } | |
917 | rpc_put_task(task); | |
918 | out: | |
919 | return ret; | |
920 | } | |
a69aef14 | 921 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
922 | |
923 | /* | |
924 | * Set up the argument/result storage required for the RPC call. | |
925 | */ | |
6c75dc0d | 926 | static void nfs_write_rpcsetup(struct nfs_write_data *data, |
d138d5d1 AA |
927 | unsigned int count, unsigned int offset, |
928 | int how) | |
929 | { | |
6c75dc0d | 930 | struct nfs_page *req = data->header->req; |
d138d5d1 | 931 | |
1da177e4 LT |
932 | /* Set up the RPC argument and reply structs |
933 | * NB: take care not to mess about with data->commit et al. */ | |
934 | ||
6c75dc0d | 935 | data->args.fh = NFS_FH(data->header->inode); |
1da177e4 | 936 | data->args.offset = req_offset(req) + offset; |
2bea038c BH |
937 | /* pnfs_set_layoutcommit needs this */ |
938 | data->mds_offset = data->args.offset; | |
1da177e4 | 939 | data->args.pgbase = req->wb_pgbase + offset; |
30dd374f | 940 | data->args.pages = data->pages.pagevec; |
1da177e4 | 941 | data->args.count = count; |
383ba719 | 942 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 943 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 944 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
945 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
946 | case 0: | |
947 | break; | |
948 | case FLUSH_COND_STABLE: | |
6c75dc0d | 949 | if (nfs_need_commit(NFS_I(data->header->inode))) |
87ed5eb4 TM |
950 | break; |
951 | default: | |
952 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 953 | } |
1da177e4 LT |
954 | |
955 | data->res.fattr = &data->fattr; | |
956 | data->res.count = count; | |
957 | data->res.verf = &data->verf; | |
0e574af1 | 958 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 959 | } |
1da177e4 | 960 | |
6e4efd56 TM |
961 | static int nfs_do_write(struct nfs_write_data *data, |
962 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
963 | int how) |
964 | { | |
cd841605 | 965 | struct inode *inode = data->header->inode; |
0382b744 | 966 | |
c5996c4e | 967 | return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how); |
1da177e4 LT |
968 | } |
969 | ||
275acaaf TM |
970 | static int nfs_do_multiple_writes(struct list_head *head, |
971 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
972 | int how) |
973 | { | |
974 | struct nfs_write_data *data; | |
975 | int ret = 0; | |
976 | ||
977 | while (!list_empty(head)) { | |
978 | int ret2; | |
979 | ||
6c75dc0d | 980 | data = list_first_entry(head, struct nfs_write_data, list); |
275acaaf TM |
981 | list_del_init(&data->list); |
982 | ||
dce81290 | 983 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
984 | if (ret == 0) |
985 | ret = ret2; | |
986 | } | |
987 | return ret; | |
988 | } | |
989 | ||
6d884e8f F |
990 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
991 | * call this on each, which will prepare them to be retried on next | |
992 | * writeback using standard nfs. | |
993 | */ | |
994 | static void nfs_redirty_request(struct nfs_page *req) | |
995 | { | |
a6305ddb TM |
996 | struct page *page = req->wb_page; |
997 | ||
6d884e8f | 998 | nfs_mark_request_dirty(req); |
9994b62b | 999 | nfs_unlock_request(req); |
a6305ddb | 1000 | nfs_end_page_writeback(page); |
6d884e8f F |
1001 | } |
1002 | ||
6c75dc0d FI |
1003 | void nfs_async_write_error(struct list_head *head) |
1004 | { | |
1005 | struct nfs_page *req; | |
1006 | ||
1007 | while (!list_empty(head)) { | |
1008 | req = nfs_list_entry(head->next); | |
1009 | nfs_list_remove_request(req); | |
1010 | nfs_redirty_request(req); | |
1011 | } | |
1012 | } | |
1013 | ||
1da177e4 LT |
1014 | /* |
1015 | * Generate multiple small requests to write out a single | |
1016 | * contiguous dirty area on one page. | |
1017 | */ | |
6c75dc0d FI |
1018 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, |
1019 | struct nfs_pgio_header *hdr) | |
1da177e4 | 1020 | { |
6c75dc0d | 1021 | struct nfs_page *req = hdr->req; |
1da177e4 LT |
1022 | struct page *page = req->wb_page; |
1023 | struct nfs_write_data *data; | |
d097971d | 1024 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 1025 | unsigned int offset; |
1da177e4 | 1026 | int requests = 0; |
dbae4c73 | 1027 | int ret = 0; |
1da177e4 LT |
1028 | |
1029 | nfs_list_remove_request(req); | |
6c75dc0d | 1030 | nfs_list_add_request(req, &hdr->pages); |
1da177e4 | 1031 | |
b31268ac TM |
1032 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
1033 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit || | |
1034 | desc->pg_count > wsize)) | |
1035 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1036 | ||
1037 | ||
275acaaf | 1038 | offset = 0; |
c76069bd | 1039 | nbytes = desc->pg_count; |
e9f7bee1 TM |
1040 | do { |
1041 | size_t len = min(nbytes, wsize); | |
1042 | ||
6c75dc0d FI |
1043 | data = nfs_writedata_alloc(hdr, 1); |
1044 | if (!data) | |
1da177e4 | 1045 | goto out_bad; |
30dd374f | 1046 | data->pages.pagevec[0] = page; |
6c75dc0d FI |
1047 | nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags); |
1048 | list_add(&data->list, &hdr->rpc_list); | |
1da177e4 | 1049 | requests++; |
e9f7bee1 | 1050 | nbytes -= len; |
275acaaf | 1051 | offset += len; |
e9f7bee1 | 1052 | } while (nbytes != 0); |
1da177e4 | 1053 | atomic_set(&req->wb_complete, requests); |
6c75dc0d | 1054 | desc->pg_rpc_callops = &nfs_write_common_ops; |
dbae4c73 | 1055 | return ret; |
1da177e4 LT |
1056 | |
1057 | out_bad: | |
6c75dc0d FI |
1058 | while (!list_empty(&hdr->rpc_list)) { |
1059 | data = list_first_entry(&hdr->rpc_list, struct nfs_write_data, list); | |
6e4efd56 | 1060 | list_del(&data->list); |
8ccd271f | 1061 | nfs_writedata_release(data); |
1da177e4 | 1062 | } |
6c75dc0d | 1063 | nfs_async_write_error(&hdr->pages); |
1da177e4 LT |
1064 | return -ENOMEM; |
1065 | } | |
1066 | ||
1067 | /* | |
1068 | * Create an RPC task for the given write request and kick it. | |
1069 | * The page must have been locked by the caller. | |
1070 | * | |
1071 | * It may happen that the page we're passed is not marked dirty. | |
1072 | * This is the case if nfs_updatepage detects a conflicting request | |
1073 | * that has been written but not committed. | |
1074 | */ | |
6c75dc0d FI |
1075 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, |
1076 | struct nfs_pgio_header *hdr) | |
1da177e4 LT |
1077 | { |
1078 | struct nfs_page *req; | |
1079 | struct page **pages; | |
1080 | struct nfs_write_data *data; | |
c76069bd | 1081 | struct list_head *head = &desc->pg_list; |
3b609184 | 1082 | int ret = 0; |
1da177e4 | 1083 | |
6c75dc0d FI |
1084 | data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base, |
1085 | desc->pg_count)); | |
1086 | if (!data) { | |
1087 | nfs_async_write_error(head); | |
44b83799 FI |
1088 | ret = -ENOMEM; |
1089 | goto out; | |
1090 | } | |
6c75dc0d | 1091 | |
30dd374f | 1092 | pages = data->pages.pagevec; |
1da177e4 LT |
1093 | while (!list_empty(head)) { |
1094 | req = nfs_list_entry(head->next); | |
1095 | nfs_list_remove_request(req); | |
6c75dc0d | 1096 | nfs_list_add_request(req, &hdr->pages); |
1da177e4 | 1097 | *pages++ = req->wb_page; |
1da177e4 | 1098 | } |
1da177e4 | 1099 | |
b31268ac TM |
1100 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
1101 | (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) | |
1102 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1103 | ||
1da177e4 | 1104 | /* Set up the argument struct */ |
6c75dc0d FI |
1105 | nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags); |
1106 | list_add(&data->list, &hdr->rpc_list); | |
1107 | desc->pg_rpc_callops = &nfs_write_common_ops; | |
44b83799 | 1108 | out: |
44b83799 | 1109 | return ret; |
1da177e4 LT |
1110 | } |
1111 | ||
6c75dc0d FI |
1112 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, |
1113 | struct nfs_pgio_header *hdr) | |
dce81290 TM |
1114 | { |
1115 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
6c75dc0d FI |
1116 | return nfs_flush_multi(desc, hdr); |
1117 | return nfs_flush_one(desc, hdr); | |
dce81290 TM |
1118 | } |
1119 | ||
1120 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1121 | { |
6c75dc0d FI |
1122 | struct nfs_write_header *whdr; |
1123 | struct nfs_pgio_header *hdr; | |
275acaaf TM |
1124 | int ret; |
1125 | ||
6c75dc0d FI |
1126 | whdr = nfs_writehdr_alloc(); |
1127 | if (!whdr) { | |
1128 | nfs_async_write_error(&desc->pg_list); | |
1129 | return -ENOMEM; | |
1130 | } | |
1131 | hdr = &whdr->header; | |
1132 | nfs_pgheader_init(desc, hdr, nfs_writehdr_free); | |
1133 | atomic_inc(&hdr->refcnt); | |
1134 | ret = nfs_generic_flush(desc, hdr); | |
50828d7e | 1135 | if (ret == 0) |
6c75dc0d FI |
1136 | ret = nfs_do_multiple_writes(&hdr->rpc_list, |
1137 | desc->pg_rpc_callops, | |
1138 | desc->pg_ioflags); | |
1139 | else | |
1140 | set_bit(NFS_IOHDR_REDO, &hdr->flags); | |
1141 | if (atomic_dec_and_test(&hdr->refcnt)) | |
1142 | nfs_write_completion(hdr); | |
275acaaf | 1143 | return ret; |
1751c363 | 1144 | } |
1751c363 TM |
1145 | |
1146 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1147 | .pg_test = nfs_generic_pg_test, | |
1148 | .pg_doio = nfs_generic_pg_writepages, | |
1149 | }; | |
1150 | ||
e2fecb21 | 1151 | void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, |
c63c7b05 | 1152 | struct inode *inode, int ioflags) |
1da177e4 | 1153 | { |
1751c363 TM |
1154 | nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, |
1155 | NFS_SERVER(inode)->wsize, ioflags); | |
1156 | } | |
1da177e4 | 1157 | |
dce81290 TM |
1158 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1159 | { | |
1160 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1161 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1162 | } | |
1f945357 | 1163 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1164 | |
1751c363 TM |
1165 | static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
1166 | struct inode *inode, int ioflags) | |
1167 | { | |
1168 | if (!pnfs_pageio_init_write(pgio, inode, ioflags)) | |
1169 | nfs_pageio_init_write_mds(pgio, inode, ioflags); | |
1da177e4 LT |
1170 | } |
1171 | ||
def6ed7e AA |
1172 | void nfs_write_prepare(struct rpc_task *task, void *calldata) |
1173 | { | |
1174 | struct nfs_write_data *data = calldata; | |
cd841605 | 1175 | NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); |
def6ed7e | 1176 | } |
def6ed7e | 1177 | |
0b7c0153 FI |
1178 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1179 | { | |
1180 | struct nfs_commit_data *data = calldata; | |
1181 | ||
1182 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1183 | } | |
1184 | ||
1da177e4 LT |
1185 | /* |
1186 | * Handle a write reply that flushes a whole page. | |
1187 | * | |
1188 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1189 | * writebacks since the page->count is kept > 1 for as long | |
1190 | * as the page has a write request pending. | |
1191 | */ | |
6c75dc0d | 1192 | static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) |
1da177e4 | 1193 | { |
788e7a89 | 1194 | struct nfs_write_data *data = calldata; |
1da177e4 | 1195 | |
c9d8f89d TM |
1196 | nfs_writeback_done(task, data); |
1197 | } | |
1198 | ||
6c75dc0d | 1199 | static void nfs_writeback_release_common(void *calldata) |
c9d8f89d TM |
1200 | { |
1201 | struct nfs_write_data *data = calldata; | |
cd841605 | 1202 | struct nfs_pgio_header *hdr = data->header; |
e2fecb21 | 1203 | int status = data->task.tk_status; |
6c75dc0d | 1204 | struct nfs_page *req = hdr->req; |
788e7a89 | 1205 | |
6c75dc0d FI |
1206 | if ((status >= 0) && nfs_write_need_commit(data)) { |
1207 | spin_lock(&hdr->lock); | |
1208 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) | |
1209 | ; /* Do nothing */ | |
1210 | else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) | |
8e821cad | 1211 | memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); |
6c75dc0d FI |
1212 | else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) |
1213 | set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); | |
1214 | spin_unlock(&hdr->lock); | |
1da177e4 | 1215 | } |
cd841605 | 1216 | nfs_writedata_release(data); |
1da177e4 LT |
1217 | } |
1218 | ||
6c75dc0d | 1219 | static const struct rpc_call_ops nfs_write_common_ops = { |
def6ed7e | 1220 | .rpc_call_prepare = nfs_write_prepare, |
6c75dc0d FI |
1221 | .rpc_call_done = nfs_writeback_done_common, |
1222 | .rpc_release = nfs_writeback_release_common, | |
788e7a89 TM |
1223 | }; |
1224 | ||
1225 | ||
1da177e4 LT |
1226 | /* |
1227 | * This function is called when the WRITE call is complete. | |
1228 | */ | |
13602896 | 1229 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1230 | { |
1da177e4 LT |
1231 | struct nfs_writeargs *argp = &data->args; |
1232 | struct nfs_writeres *resp = &data->res; | |
cd841605 | 1233 | struct inode *inode = data->header->inode; |
788e7a89 | 1234 | int status; |
1da177e4 | 1235 | |
a3f565b1 | 1236 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1237 | task->tk_pid, task->tk_status); |
1238 | ||
f551e44f CL |
1239 | /* |
1240 | * ->write_done will attempt to use post-op attributes to detect | |
1241 | * conflicting writes by other clients. A strict interpretation | |
1242 | * of close-to-open would allow us to continue caching even if | |
1243 | * another writer had changed the file, but some applications | |
1244 | * depend on tighter cache coherency when writing. | |
1245 | */ | |
cd841605 | 1246 | status = NFS_PROTO(inode)->write_done(task, data); |
788e7a89 | 1247 | if (status != 0) |
13602896 | 1248 | return; |
cd841605 | 1249 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
91d5b470 | 1250 | |
1da177e4 LT |
1251 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) |
1252 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { | |
1253 | /* We tried a write call, but the server did not | |
1254 | * commit data to stable storage even though we | |
1255 | * requested it. | |
1256 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1257 | * the server reports NFS_DATA_SYNC, but performs | |
1258 | * NFS_FILE_SYNC. We therefore implement this checking | |
1259 | * as a dprintk() in order to avoid filling syslog. | |
1260 | */ | |
1261 | static unsigned long complain; | |
1262 | ||
a69aef14 | 1263 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1264 | if (time_before(complain, jiffies)) { |
48186c7d | 1265 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1266 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1267 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1268 | resp->verf->committed, argp->stable); |
1269 | complain = jiffies + 300 * HZ; | |
1270 | } | |
1271 | } | |
1272 | #endif | |
6c75dc0d FI |
1273 | if (task->tk_status < 0) |
1274 | nfs_set_pgio_error(data->header, task->tk_status, argp->offset); | |
1275 | else if (resp->count < argp->count) { | |
1da177e4 LT |
1276 | static unsigned long complain; |
1277 | ||
6c75dc0d | 1278 | /* This a short write! */ |
cd841605 | 1279 | nfs_inc_stats(inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1280 | |
1da177e4 | 1281 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1282 | if (resp->count == 0) { |
1283 | if (time_before(complain, jiffies)) { | |
1284 | printk(KERN_WARNING | |
1285 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1286 | argp->count); | |
1287 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1288 | } |
6c75dc0d FI |
1289 | nfs_set_pgio_error(data->header, -EIO, argp->offset); |
1290 | task->tk_status = -EIO; | |
13602896 | 1291 | return; |
1da177e4 | 1292 | } |
6c75dc0d FI |
1293 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1294 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1295 | /* Resend from where the server left off */ | |
1296 | data->mds_offset += resp->count; | |
1297 | argp->offset += resp->count; | |
1298 | argp->pgbase += resp->count; | |
1299 | argp->count -= resp->count; | |
1300 | } else { | |
1301 | /* Resend as a stable write in order to avoid | |
1302 | * headaches in the case of a server crash. | |
1303 | */ | |
1304 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1305 | } |
6c75dc0d | 1306 | rpc_restart_call_prepare(task); |
1da177e4 | 1307 | } |
1da177e4 LT |
1308 | } |
1309 | ||
1310 | ||
1311 | #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) | |
71d0a611 TM |
1312 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1313 | { | |
b8413f98 TM |
1314 | int ret; |
1315 | ||
71d0a611 TM |
1316 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1317 | return 1; | |
b8413f98 TM |
1318 | if (!may_wait) |
1319 | return 0; | |
1320 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1321 | NFS_INO_COMMIT, | |
1322 | nfs_wait_bit_killable, | |
1323 | TASK_KILLABLE); | |
1324 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1325 | } |
1326 | ||
e0c2b380 | 1327 | void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1328 | { |
1329 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1330 | smp_mb__after_clear_bit(); | |
1331 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1332 | } | |
e0c2b380 | 1333 | EXPORT_SYMBOL_GPL(nfs_commit_clear_lock); |
71d0a611 | 1334 | |
0b7c0153 | 1335 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1336 | { |
0b7c0153 FI |
1337 | put_nfs_open_context(data->context); |
1338 | nfs_commit_free(data); | |
1da177e4 | 1339 | } |
e0c2b380 | 1340 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1341 | |
0b7c0153 | 1342 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd FI |
1343 | const struct rpc_call_ops *call_ops, |
1344 | int how) | |
1da177e4 | 1345 | { |
07737691 | 1346 | struct rpc_task *task; |
9ace33cd | 1347 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1348 | struct rpc_message msg = { |
1349 | .rpc_argp = &data->args, | |
1350 | .rpc_resp = &data->res, | |
9ace33cd | 1351 | .rpc_cred = data->cred, |
bdc7f021 | 1352 | }; |
84115e1c | 1353 | struct rpc_task_setup task_setup_data = { |
07737691 | 1354 | .task = &data->task, |
9ace33cd | 1355 | .rpc_client = clnt, |
bdc7f021 | 1356 | .rpc_message = &msg, |
9ace33cd | 1357 | .callback_ops = call_ops, |
84115e1c | 1358 | .callback_data = data, |
101070ca | 1359 | .workqueue = nfsiod_workqueue, |
2c61be0a | 1360 | .flags = RPC_TASK_ASYNC, |
3ff7576d | 1361 | .priority = priority, |
84115e1c | 1362 | }; |
9ace33cd FI |
1363 | /* Set up the initial task struct. */ |
1364 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1365 | ||
1366 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1367 | ||
1368 | task = rpc_run_task(&task_setup_data); | |
1369 | if (IS_ERR(task)) | |
1370 | return PTR_ERR(task); | |
1371 | if (how & FLUSH_SYNC) | |
1372 | rpc_wait_for_completion_task(task); | |
1373 | rpc_put_task(task); | |
1374 | return 0; | |
1375 | } | |
e0c2b380 | 1376 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1377 | |
1378 | /* | |
1379 | * Set up the argument/result storage required for the RPC call. | |
1380 | */ | |
0b7c0153 | 1381 | void nfs_init_commit(struct nfs_commit_data *data, |
988b6dce FI |
1382 | struct list_head *head, |
1383 | struct pnfs_layout_segment *lseg) | |
9ace33cd FI |
1384 | { |
1385 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1386 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1387 | |
1388 | /* Set up the RPC argument and reply structs | |
1389 | * NB: take care not to mess about with data->commit et al. */ | |
1390 | ||
1391 | list_splice_init(head, &data->pages); | |
1da177e4 | 1392 | |
1da177e4 | 1393 | data->inode = inode; |
9ace33cd | 1394 | data->cred = first->wb_context->cred; |
988b6dce | 1395 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1396 | data->mds_ops = &nfs_commit_ops; |
1da177e4 LT |
1397 | |
1398 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1399 | /* Note: we always request a commit of the entire inode */ |
1400 | data->args.offset = 0; | |
1401 | data->args.count = 0; | |
0b7c0153 | 1402 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1403 | data->res.fattr = &data->fattr; |
1404 | data->res.verf = &data->verf; | |
0e574af1 | 1405 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1406 | } |
e0c2b380 | 1407 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1408 | |
e0c2b380 | 1409 | void nfs_retry_commit(struct list_head *page_list, |
a861a1e1 | 1410 | struct pnfs_layout_segment *lseg) |
64bfeb49 FI |
1411 | { |
1412 | struct nfs_page *req; | |
1413 | ||
1414 | while (!list_empty(page_list)) { | |
1415 | req = nfs_list_entry(page_list->next); | |
1416 | nfs_list_remove_request(req); | |
a861a1e1 | 1417 | nfs_mark_request_commit(req, lseg); |
64bfeb49 FI |
1418 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); |
1419 | dec_bdi_stat(req->wb_page->mapping->backing_dev_info, | |
1420 | BDI_RECLAIMABLE); | |
9994b62b | 1421 | nfs_unlock_request(req); |
64bfeb49 FI |
1422 | } |
1423 | } | |
e0c2b380 | 1424 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1425 | |
1da177e4 LT |
1426 | /* |
1427 | * Commit dirty pages | |
1428 | */ | |
1429 | static int | |
40859d7e | 1430 | nfs_commit_list(struct inode *inode, struct list_head *head, int how) |
1da177e4 | 1431 | { |
0b7c0153 | 1432 | struct nfs_commit_data *data; |
1da177e4 | 1433 | |
c9d8f89d | 1434 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1435 | |
1436 | if (!data) | |
1437 | goto out_bad; | |
1438 | ||
1439 | /* Set up the argument struct */ | |
988b6dce | 1440 | nfs_init_commit(data, head, NULL); |
0b7c0153 | 1441 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, how); |
1da177e4 | 1442 | out_bad: |
a861a1e1 | 1443 | nfs_retry_commit(head, NULL); |
71d0a611 | 1444 | nfs_commit_clear_lock(NFS_I(inode)); |
1da177e4 LT |
1445 | return -ENOMEM; |
1446 | } | |
1447 | ||
1448 | /* | |
1449 | * COMMIT call returned | |
1450 | */ | |
788e7a89 | 1451 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1452 | { |
0b7c0153 | 1453 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1454 | |
a3f565b1 | 1455 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1456 | task->tk_pid, task->tk_status); |
1457 | ||
788e7a89 | 1458 | /* Call the NFS version-specific code */ |
c0d0e96b | 1459 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1460 | } |
1461 | ||
0b7c0153 | 1462 | void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1463 | { |
5917ce84 | 1464 | struct nfs_page *req; |
c9d8f89d | 1465 | int status = data->task.tk_status; |
788e7a89 | 1466 | |
1da177e4 LT |
1467 | while (!list_empty(&data->pages)) { |
1468 | req = nfs_list_entry(data->pages.next); | |
1469 | nfs_list_remove_request(req); | |
d6d6dc7c | 1470 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1471 | |
48186c7d | 1472 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
3d4ff43d AV |
1473 | req->wb_context->dentry->d_sb->s_id, |
1474 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1475 | req->wb_bytes, |
1476 | (long long)req_offset(req)); | |
c9d8f89d TM |
1477 | if (status < 0) { |
1478 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1479 | nfs_inode_remove_request(req); |
c9d8f89d | 1480 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1481 | goto next; |
1482 | } | |
1483 | ||
1484 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1485 | * returned by the server against all stored verfs. */ | |
1486 | if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { | |
1487 | /* We have a match */ | |
1488 | nfs_inode_remove_request(req); | |
1489 | dprintk(" OK\n"); | |
1490 | goto next; | |
1491 | } | |
1492 | /* We have a mismatch. Write the page again */ | |
1493 | dprintk(" mismatch\n"); | |
6d884e8f | 1494 | nfs_mark_request_dirty(req); |
1da177e4 | 1495 | next: |
9994b62b | 1496 | nfs_unlock_request(req); |
1da177e4 | 1497 | } |
5917ce84 | 1498 | } |
e0c2b380 | 1499 | EXPORT_SYMBOL_GPL(nfs_commit_release_pages); |
5917ce84 FI |
1500 | |
1501 | static void nfs_commit_release(void *calldata) | |
1502 | { | |
0b7c0153 | 1503 | struct nfs_commit_data *data = calldata; |
5917ce84 FI |
1504 | |
1505 | nfs_commit_release_pages(data); | |
71d0a611 | 1506 | nfs_commit_clear_lock(NFS_I(data->inode)); |
c9d8f89d | 1507 | nfs_commitdata_release(calldata); |
1da177e4 | 1508 | } |
788e7a89 TM |
1509 | |
1510 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1511 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1512 | .rpc_call_done = nfs_commit_done, |
1513 | .rpc_release = nfs_commit_release, | |
1514 | }; | |
1da177e4 | 1515 | |
b608b283 | 1516 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1517 | { |
1da177e4 | 1518 | LIST_HEAD(head); |
71d0a611 | 1519 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1520 | int res; |
1da177e4 | 1521 | |
b8413f98 TM |
1522 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1523 | if (res <= 0) | |
c5efa5fc | 1524 | goto out_mark_dirty; |
d6d6dc7c | 1525 | res = nfs_scan_commit(inode, &head); |
1da177e4 | 1526 | if (res) { |
a861a1e1 FI |
1527 | int error; |
1528 | ||
1529 | error = pnfs_commit_list(inode, &head, how); | |
1530 | if (error == PNFS_NOT_ATTEMPTED) | |
1531 | error = nfs_commit_list(inode, &head, how); | |
3da28eb1 TM |
1532 | if (error < 0) |
1533 | return error; | |
b8413f98 | 1534 | if (!may_wait) |
c5efa5fc | 1535 | goto out_mark_dirty; |
b8413f98 TM |
1536 | error = wait_on_bit(&NFS_I(inode)->flags, |
1537 | NFS_INO_COMMIT, | |
1538 | nfs_wait_bit_killable, | |
1539 | TASK_KILLABLE); | |
1540 | if (error < 0) | |
1541 | return error; | |
71d0a611 TM |
1542 | } else |
1543 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1544 | return res; |
1545 | /* Note: If we exit without ensuring that the commit is complete, | |
1546 | * we must mark the inode as dirty. Otherwise, future calls to | |
1547 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1548 | * that the data is on the disk. | |
1549 | */ | |
1550 | out_mark_dirty: | |
1551 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1552 | return res; |
1553 | } | |
8fc795f7 TM |
1554 | |
1555 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1556 | { | |
420e3646 TM |
1557 | struct nfs_inode *nfsi = NFS_I(inode); |
1558 | int flags = FLUSH_SYNC; | |
1559 | int ret = 0; | |
8fc795f7 | 1560 | |
3236c3e1 JL |
1561 | /* no commits means nothing needs to be done */ |
1562 | if (!nfsi->ncommit) | |
1563 | return ret; | |
1564 | ||
a00dd6c0 JL |
1565 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1566 | /* Don't commit yet if this is a non-blocking flush and there | |
1567 | * are a lot of outstanding writes for this mapping. | |
1568 | */ | |
1569 | if (nfsi->ncommit <= (nfsi->npages >> 1)) | |
1570 | goto out_mark_dirty; | |
420e3646 | 1571 | |
a00dd6c0 | 1572 | /* don't wait for the COMMIT response */ |
420e3646 | 1573 | flags = 0; |
a00dd6c0 JL |
1574 | } |
1575 | ||
420e3646 TM |
1576 | ret = nfs_commit_inode(inode, flags); |
1577 | if (ret >= 0) { | |
1578 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1579 | if (ret < wbc->nr_to_write) | |
1580 | wbc->nr_to_write -= ret; | |
1581 | else | |
1582 | wbc->nr_to_write = 0; | |
1583 | } | |
8fc795f7 | 1584 | return 0; |
420e3646 TM |
1585 | } |
1586 | out_mark_dirty: | |
8fc795f7 TM |
1587 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1588 | return ret; | |
1589 | } | |
c63c7b05 | 1590 | #else |
8fc795f7 TM |
1591 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1592 | { | |
1593 | return 0; | |
1594 | } | |
1da177e4 LT |
1595 | #endif |
1596 | ||
8fc795f7 TM |
1597 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1598 | { | |
863a3c6c AA |
1599 | int ret; |
1600 | ||
1601 | ret = nfs_commit_unstable_pages(inode, wbc); | |
1602 | if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { | |
ef311537 AA |
1603 | int status; |
1604 | bool sync = true; | |
863a3c6c | 1605 | |
846d5a09 | 1606 | if (wbc->sync_mode == WB_SYNC_NONE) |
ef311537 | 1607 | sync = false; |
863a3c6c AA |
1608 | |
1609 | status = pnfs_layoutcommit_inode(inode, sync); | |
1610 | if (status < 0) | |
1611 | return status; | |
1612 | } | |
1613 | return ret; | |
8fc795f7 TM |
1614 | } |
1615 | ||
acdc53b2 TM |
1616 | /* |
1617 | * flush the inode to disk. | |
1618 | */ | |
1619 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1620 | { |
1621 | struct writeback_control wbc = { | |
72cb77f4 | 1622 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1623 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1624 | .range_start = 0, |
1625 | .range_end = LLONG_MAX, | |
34901f70 | 1626 | }; |
34901f70 | 1627 | |
acdc53b2 | 1628 | return sync_inode(inode, &wbc); |
1c75950b TM |
1629 | } |
1630 | ||
1b3b4a1a TM |
1631 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1632 | { | |
1633 | struct nfs_page *req; | |
1b3b4a1a TM |
1634 | int ret = 0; |
1635 | ||
1636 | BUG_ON(!PageLocked(page)); | |
1637 | for (;;) { | |
ba8b06e6 | 1638 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1639 | req = nfs_page_find_request(page); |
1640 | if (req == NULL) | |
1b3b4a1a | 1641 | break; |
1b3b4a1a | 1642 | if (nfs_lock_request_dontget(req)) { |
8dd37758 | 1643 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1644 | nfs_inode_remove_request(req); |
1645 | /* | |
1646 | * In case nfs_inode_remove_request has marked the | |
1647 | * page as being dirty | |
1648 | */ | |
1649 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1650 | nfs_unlock_request(req); | |
1651 | break; | |
1652 | } | |
1653 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1654 | nfs_release_request(req); |
1b3b4a1a | 1655 | if (ret < 0) |
c988950e | 1656 | break; |
1b3b4a1a | 1657 | } |
1b3b4a1a TM |
1658 | return ret; |
1659 | } | |
1660 | ||
7f2f12d9 TM |
1661 | /* |
1662 | * Write back all requests on one page - we do this before reading it. | |
1663 | */ | |
1664 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b TM |
1665 | { |
1666 | loff_t range_start = page_offset(page); | |
1667 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); | |
4d770ccf | 1668 | struct writeback_control wbc = { |
4d770ccf | 1669 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1670 | .nr_to_write = 0, |
4d770ccf TM |
1671 | .range_start = range_start, |
1672 | .range_end = range_end, | |
1673 | }; | |
1674 | int ret; | |
1c75950b | 1675 | |
0522f6ad | 1676 | for (;;) { |
ba8b06e6 | 1677 | wait_on_page_writeback(page); |
73e3302f TM |
1678 | if (clear_page_dirty_for_io(page)) { |
1679 | ret = nfs_writepage_locked(page, &wbc); | |
1680 | if (ret < 0) | |
1681 | goto out_error; | |
0522f6ad | 1682 | continue; |
7f2f12d9 | 1683 | } |
0522f6ad TM |
1684 | if (!PagePrivate(page)) |
1685 | break; | |
1686 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1687 | if (ret < 0) |
73e3302f | 1688 | goto out_error; |
7f2f12d9 | 1689 | } |
73e3302f TM |
1690 | return 0; |
1691 | out_error: | |
4d770ccf | 1692 | return ret; |
1c75950b TM |
1693 | } |
1694 | ||
074cc1de TM |
1695 | #ifdef CONFIG_MIGRATION |
1696 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1697 | struct page *page, enum migrate_mode mode) |
074cc1de | 1698 | { |
2da95652 JL |
1699 | /* |
1700 | * If PagePrivate is set, then the page is currently associated with | |
1701 | * an in-progress read or write request. Don't try to migrate it. | |
1702 | * | |
1703 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1704 | * that we can safely release the inode reference while holding | |
1705 | * the page lock. | |
1706 | */ | |
1707 | if (PagePrivate(page)) | |
1708 | return -EBUSY; | |
074cc1de | 1709 | |
7549ad5f | 1710 | nfs_fscache_release_page(page, GFP_KERNEL); |
074cc1de | 1711 | |
a6bc32b8 | 1712 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1713 | } |
1714 | #endif | |
1715 | ||
f7b422b1 | 1716 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1717 | { |
1718 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
cd841605 | 1719 | sizeof(struct nfs_write_header), |
1da177e4 | 1720 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1721 | NULL); |
1da177e4 LT |
1722 | if (nfs_wdata_cachep == NULL) |
1723 | return -ENOMEM; | |
1724 | ||
93d2341c MD |
1725 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1726 | nfs_wdata_cachep); | |
1da177e4 LT |
1727 | if (nfs_wdata_mempool == NULL) |
1728 | return -ENOMEM; | |
1729 | ||
0b7c0153 FI |
1730 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1731 | sizeof(struct nfs_commit_data), | |
1732 | 0, SLAB_HWCACHE_ALIGN, | |
1733 | NULL); | |
1734 | if (nfs_cdata_cachep == NULL) | |
1735 | return -ENOMEM; | |
1736 | ||
93d2341c MD |
1737 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
1738 | nfs_wdata_cachep); | |
1da177e4 LT |
1739 | if (nfs_commit_mempool == NULL) |
1740 | return -ENOMEM; | |
1741 | ||
89a09141 PZ |
1742 | /* |
1743 | * NFS congestion size, scale with available memory. | |
1744 | * | |
1745 | * 64MB: 8192k | |
1746 | * 128MB: 11585k | |
1747 | * 256MB: 16384k | |
1748 | * 512MB: 23170k | |
1749 | * 1GB: 32768k | |
1750 | * 2GB: 46340k | |
1751 | * 4GB: 65536k | |
1752 | * 8GB: 92681k | |
1753 | * 16GB: 131072k | |
1754 | * | |
1755 | * This allows larger machines to have larger/more transfers. | |
1756 | * Limit the default to 256M | |
1757 | */ | |
1758 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1759 | if (nfs_congestion_kb > 256*1024) | |
1760 | nfs_congestion_kb = 256*1024; | |
1761 | ||
1da177e4 LT |
1762 | return 0; |
1763 | } | |
1764 | ||
266bee88 | 1765 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1766 | { |
1767 | mempool_destroy(nfs_commit_mempool); | |
1768 | mempool_destroy(nfs_wdata_mempool); | |
1a1d92c1 | 1769 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1770 | } |
1771 |