]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/file.c | |
3 | * | |
4 | * Copyright (C) 1992 Rick Sladkey | |
5 | * | |
6 | * Changes Copyright (C) 1994 by Florian La Roche | |
7 | * - Do not copy data too often around in the kernel. | |
8 | * - In nfs_file_read the return value of kmalloc wasn't checked. | |
9 | * - Put in a better version of read look-ahead buffering. Original idea | |
10 | * and implementation by Wai S Kok elekokws@ee.nus.sg. | |
11 | * | |
12 | * Expire cache on write to a file by Wai S Kok (Oct 1994). | |
13 | * | |
14 | * Total rewrite of read side for new NFS buffer cache.. Linus. | |
15 | * | |
16 | * nfs regular file handling functions | |
17 | */ | |
18 | ||
19 | #include <linux/time.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/fcntl.h> | |
23 | #include <linux/stat.h> | |
24 | #include <linux/nfs_fs.h> | |
25 | #include <linux/nfs_mount.h> | |
26 | #include <linux/mm.h> | |
1da177e4 | 27 | #include <linux/pagemap.h> |
e8edc6e0 | 28 | #include <linux/aio.h> |
5a0e3ad6 | 29 | #include <linux/gfp.h> |
b608b283 | 30 | #include <linux/swap.h> |
1da177e4 LT |
31 | |
32 | #include <asm/uaccess.h> | |
33 | #include <asm/system.h> | |
34 | ||
35 | #include "delegation.h" | |
94387fb1 | 36 | #include "internal.h" |
91d5b470 | 37 | #include "iostat.h" |
545db45f | 38 | #include "fscache.h" |
e5e94017 | 39 | #include "pnfs.h" |
1da177e4 LT |
40 | |
41 | #define NFSDBG_FACILITY NFSDBG_FILE | |
42 | ||
f0f37e2f | 43 | static const struct vm_operations_struct nfs_file_vm_ops; |
94387fb1 | 44 | |
92e1d5be | 45 | const struct inode_operations nfs_file_inode_operations = { |
1da177e4 LT |
46 | .permission = nfs_permission, |
47 | .getattr = nfs_getattr, | |
48 | .setattr = nfs_setattr, | |
49 | }; | |
50 | ||
b7fa0554 | 51 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 52 | const struct inode_operations nfs3_file_inode_operations = { |
b7fa0554 AG |
53 | .permission = nfs_permission, |
54 | .getattr = nfs_getattr, | |
55 | .setattr = nfs_setattr, | |
56 | .listxattr = nfs3_listxattr, | |
57 | .getxattr = nfs3_getxattr, | |
58 | .setxattr = nfs3_setxattr, | |
59 | .removexattr = nfs3_removexattr, | |
60 | }; | |
61 | #endif /* CONFIG_NFS_v3 */ | |
62 | ||
1da177e4 LT |
63 | /* Hack for future NFS swap support */ |
64 | #ifndef IS_SWAPFILE | |
65 | # define IS_SWAPFILE(inode) (0) | |
66 | #endif | |
67 | ||
68 | static int nfs_check_flags(int flags) | |
69 | { | |
70 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
71 | return -EINVAL; | |
72 | ||
73 | return 0; | |
74 | } | |
75 | ||
76 | /* | |
77 | * Open file | |
78 | */ | |
79 | static int | |
80 | nfs_file_open(struct inode *inode, struct file *filp) | |
81 | { | |
1da177e4 LT |
82 | int res; |
83 | ||
6da24bc9 | 84 | dprintk("NFS: open file(%s/%s)\n", |
cc0dd2d1 CL |
85 | filp->f_path.dentry->d_parent->d_name.name, |
86 | filp->f_path.dentry->d_name.name); | |
87 | ||
c2459dc4 | 88 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
1da177e4 LT |
89 | res = nfs_check_flags(filp->f_flags); |
90 | if (res) | |
91 | return res; | |
92 | ||
46cb650c | 93 | res = nfs_open(inode, filp); |
1da177e4 LT |
94 | return res; |
95 | } | |
96 | ||
97 | static int | |
98 | nfs_file_release(struct inode *inode, struct file *filp) | |
99 | { | |
6da24bc9 | 100 | dprintk("NFS: release(%s/%s)\n", |
6f276e49 RM |
101 | filp->f_path.dentry->d_parent->d_name.name, |
102 | filp->f_path.dentry->d_name.name); | |
6da24bc9 | 103 | |
91d5b470 | 104 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
46cb650c | 105 | return nfs_release(inode, filp); |
1da177e4 LT |
106 | } |
107 | ||
980802e3 TM |
108 | /** |
109 | * nfs_revalidate_size - Revalidate the file size | |
110 | * @inode - pointer to inode struct | |
111 | * @file - pointer to struct file | |
112 | * | |
113 | * Revalidates the file length. This is basically a wrapper around | |
114 | * nfs_revalidate_inode() that takes into account the fact that we may | |
115 | * have cached writes (in which case we don't care about the server's | |
116 | * idea of what the file length is), or O_DIRECT (in which case we | |
117 | * shouldn't trust the cache). | |
118 | */ | |
119 | static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) | |
120 | { | |
121 | struct nfs_server *server = NFS_SERVER(inode); | |
122 | struct nfs_inode *nfsi = NFS_I(inode); | |
123 | ||
d7cf8dd0 TM |
124 | if (nfs_have_delegated_attributes(inode)) |
125 | goto out_noreval; | |
126 | ||
980802e3 TM |
127 | if (filp->f_flags & O_DIRECT) |
128 | goto force_reval; | |
d7cf8dd0 TM |
129 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
130 | goto force_reval; | |
131 | if (nfs_attribute_timeout(inode)) | |
132 | goto force_reval; | |
133 | out_noreval: | |
134 | return 0; | |
980802e3 TM |
135 | force_reval: |
136 | return __nfs_revalidate_inode(server, inode); | |
137 | } | |
138 | ||
139 | static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin) | |
140 | { | |
6da24bc9 | 141 | dprintk("NFS: llseek file(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
142 | filp->f_path.dentry->d_parent->d_name.name, |
143 | filp->f_path.dentry->d_name.name, | |
144 | offset, origin); | |
145 | ||
06222e49 JB |
146 | /* |
147 | * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate | |
148 | * the cached file length | |
149 | */ | |
6c529617 | 150 | if (origin != SEEK_SET && origin != SEEK_CUR) { |
980802e3 | 151 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 152 | |
980802e3 TM |
153 | int retval = nfs_revalidate_file_size(inode, filp); |
154 | if (retval < 0) | |
155 | return (loff_t)retval; | |
79835a71 | 156 | } |
d5e66348 | 157 | |
79835a71 | 158 | return generic_file_llseek(filp, offset, origin); |
980802e3 TM |
159 | } |
160 | ||
1da177e4 LT |
161 | /* |
162 | * Flush all dirty pages, and check for write errors. | |
1da177e4 LT |
163 | */ |
164 | static int | |
75e1fcc0 | 165 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 166 | { |
6da24bc9 CL |
167 | struct dentry *dentry = file->f_path.dentry; |
168 | struct inode *inode = dentry->d_inode; | |
1da177e4 | 169 | |
6da24bc9 CL |
170 | dprintk("NFS: flush(%s/%s)\n", |
171 | dentry->d_parent->d_name.name, | |
172 | dentry->d_name.name); | |
1da177e4 | 173 | |
c2459dc4 | 174 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
175 | if ((file->f_mode & FMODE_WRITE) == 0) |
176 | return 0; | |
7b159fc1 | 177 | |
7fe5c398 | 178 | /* Flush writes to the server and return any errors */ |
af7fa165 | 179 | return vfs_fsync(file, 0); |
1da177e4 LT |
180 | } |
181 | ||
182 | static ssize_t | |
027445c3 BP |
183 | nfs_file_read(struct kiocb *iocb, const struct iovec *iov, |
184 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 185 | { |
01cce933 | 186 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 LT |
187 | struct inode * inode = dentry->d_inode; |
188 | ssize_t result; | |
189 | ||
1da177e4 | 190 | if (iocb->ki_filp->f_flags & O_DIRECT) |
027445c3 | 191 | return nfs_file_direct_read(iocb, iov, nr_segs, pos); |
1da177e4 | 192 | |
6da24bc9 | 193 | dprintk("NFS: read(%s/%s, %lu@%lu)\n", |
1da177e4 | 194 | dentry->d_parent->d_name.name, dentry->d_name.name, |
6f276e49 | 195 | (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos); |
1da177e4 | 196 | |
44b11874 | 197 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); |
4184dcf2 | 198 | if (!result) { |
027445c3 | 199 | result = generic_file_aio_read(iocb, iov, nr_segs, pos); |
4184dcf2 CL |
200 | if (result > 0) |
201 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
202 | } | |
1da177e4 LT |
203 | return result; |
204 | } | |
205 | ||
206 | static ssize_t | |
f0930fff JA |
207 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
208 | struct pipe_inode_info *pipe, size_t count, | |
209 | unsigned int flags) | |
1da177e4 | 210 | { |
01cce933 | 211 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
212 | struct inode *inode = dentry->d_inode; |
213 | ssize_t res; | |
214 | ||
6da24bc9 | 215 | dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n", |
1da177e4 LT |
216 | dentry->d_parent->d_name.name, dentry->d_name.name, |
217 | (unsigned long) count, (unsigned long long) *ppos); | |
218 | ||
44b11874 | 219 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
aa2f1ef1 | 220 | if (!res) { |
f0930fff | 221 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
aa2f1ef1 CL |
222 | if (res > 0) |
223 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); | |
224 | } | |
1da177e4 LT |
225 | return res; |
226 | } | |
227 | ||
228 | static int | |
229 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) | |
230 | { | |
01cce933 | 231 | struct dentry *dentry = file->f_path.dentry; |
1da177e4 LT |
232 | struct inode *inode = dentry->d_inode; |
233 | int status; | |
234 | ||
6da24bc9 | 235 | dprintk("NFS: mmap(%s/%s)\n", |
1da177e4 LT |
236 | dentry->d_parent->d_name.name, dentry->d_name.name); |
237 | ||
e1ebfd33 TM |
238 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
239 | * so we call that before revalidating the mapping | |
240 | */ | |
241 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
242 | if (!status) { |
243 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 244 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 245 | } |
1da177e4 LT |
246 | return status; |
247 | } | |
248 | ||
249 | /* | |
250 | * Flush any dirty pages for this process, and check for write errors. | |
251 | * The return status from this call provides a reliable indication of | |
252 | * whether any write errors occurred for this process. | |
af7fa165 TM |
253 | * |
254 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
255 | * disk, but it retrieves and clears ctx->error after synching, despite | |
256 | * the two being set at the same time in nfs_context_set_write_error(). | |
257 | * This is because the former is used to notify the _next_ call to | |
25985edc | 258 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 259 | * fall back to doing a synchronous write. |
1da177e4 LT |
260 | */ |
261 | static int | |
02c24a82 | 262 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 263 | { |
7ea80859 | 264 | struct dentry *dentry = file->f_path.dentry; |
cd3758e3 | 265 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 266 | struct inode *inode = dentry->d_inode; |
af7fa165 TM |
267 | int have_error, status; |
268 | int ret = 0; | |
269 | ||
6da24bc9 | 270 | dprintk("NFS: fsync file(%s/%s) datasync %d\n", |
54917786 CL |
271 | dentry->d_parent->d_name.name, dentry->d_name.name, |
272 | datasync); | |
1da177e4 | 273 | |
02c24a82 | 274 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); |
02c24a82 JB |
275 | mutex_lock(&inode->i_mutex); |
276 | ||
91d5b470 | 277 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
af7fa165 TM |
278 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
279 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
2edb6bc3 N |
280 | if (status >= 0 && ret < 0) |
281 | status = ret; | |
af7fa165 TM |
282 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
283 | if (have_error) | |
284 | ret = xchg(&ctx->error, 0); | |
0702099b | 285 | if (!ret && status < 0) |
af7fa165 | 286 | ret = status; |
863a3c6c AA |
287 | if (!ret && !datasync) |
288 | /* application has asked for meta-data sync */ | |
ef311537 | 289 | ret = pnfs_layoutcommit_inode(inode, true); |
02c24a82 | 290 | mutex_unlock(&inode->i_mutex); |
af7fa165 | 291 | return ret; |
1da177e4 LT |
292 | } |
293 | ||
38c73044 PS |
294 | /* |
295 | * Decide whether a read/modify/write cycle may be more efficient | |
296 | * then a modify/write/read cycle when writing to a page in the | |
297 | * page cache. | |
298 | * | |
299 | * The modify/write/read cycle may occur if a page is read before | |
300 | * being completely filled by the writer. In this situation, the | |
301 | * page must be completely written to stable storage on the server | |
302 | * before it can be refilled by reading in the page from the server. | |
303 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
304 | * done. | |
305 | * | |
306 | * It may be more efficient to read the page first if the file is | |
307 | * open for reading in addition to writing, the page is not marked | |
308 | * as Uptodate, it is not dirty or waiting to be committed, | |
309 | * indicating that it was previously allocated and then modified, | |
310 | * that there were valid bytes of data in that range of the file, | |
311 | * and that the new data won't completely replace the old data in | |
312 | * that range of the file. | |
313 | */ | |
314 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
315 | loff_t pos, unsigned len) | |
316 | { | |
317 | unsigned int pglen = nfs_page_length(page); | |
318 | unsigned int offset = pos & (PAGE_CACHE_SIZE - 1); | |
319 | unsigned int end = offset + len; | |
320 | ||
321 | if ((file->f_mode & FMODE_READ) && /* open for read? */ | |
322 | !PageUptodate(page) && /* Uptodate? */ | |
323 | !PagePrivate(page) && /* i/o request already? */ | |
324 | pglen && /* valid bytes of file? */ | |
325 | (end < pglen || offset)) /* replace all valid bytes? */ | |
326 | return 1; | |
327 | return 0; | |
328 | } | |
329 | ||
1da177e4 | 330 | /* |
4899f9c8 NP |
331 | * This does the "real" work of the write. We must allocate and lock the |
332 | * page to be sent back to the generic routine, which then copies the | |
333 | * data from user space. | |
1da177e4 LT |
334 | * |
335 | * If the writer ends up delaying the write, the writer needs to | |
336 | * increment the page use counts until he is done with the page. | |
337 | */ | |
4899f9c8 NP |
338 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
339 | loff_t pos, unsigned len, unsigned flags, | |
340 | struct page **pagep, void **fsdata) | |
1da177e4 | 341 | { |
4899f9c8 | 342 | int ret; |
38c73044 | 343 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
4899f9c8 | 344 | struct page *page; |
38c73044 | 345 | int once_thru = 0; |
4899f9c8 | 346 | |
b7eaefaa CL |
347 | dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n", |
348 | file->f_path.dentry->d_parent->d_name.name, | |
349 | file->f_path.dentry->d_name.name, | |
350 | mapping->host->i_ino, len, (long long) pos); | |
351 | ||
38c73044 | 352 | start: |
72cb77f4 TM |
353 | /* |
354 | * Prevent starvation issues if someone is doing a consistency | |
355 | * sync-to-disk | |
356 | */ | |
357 | ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, | |
358 | nfs_wait_bit_killable, TASK_KILLABLE); | |
359 | if (ret) | |
360 | return ret; | |
361 | ||
54566b2c | 362 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
363 | if (!page) |
364 | return -ENOMEM; | |
365 | *pagep = page; | |
366 | ||
367 | ret = nfs_flush_incompatible(file, page); | |
368 | if (ret) { | |
369 | unlock_page(page); | |
370 | page_cache_release(page); | |
38c73044 PS |
371 | } else if (!once_thru && |
372 | nfs_want_read_modify_write(file, page, pos, len)) { | |
373 | once_thru = 1; | |
374 | ret = nfs_readpage(file, page); | |
375 | page_cache_release(page); | |
376 | if (!ret) | |
377 | goto start; | |
4899f9c8 NP |
378 | } |
379 | return ret; | |
1da177e4 LT |
380 | } |
381 | ||
4899f9c8 NP |
382 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
383 | loff_t pos, unsigned len, unsigned copied, | |
384 | struct page *page, void *fsdata) | |
1da177e4 | 385 | { |
4899f9c8 NP |
386 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
387 | int status; | |
1da177e4 | 388 | |
b7eaefaa CL |
389 | dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n", |
390 | file->f_path.dentry->d_parent->d_name.name, | |
391 | file->f_path.dentry->d_name.name, | |
392 | mapping->host->i_ino, len, (long long) pos); | |
393 | ||
efc91ed0 TM |
394 | /* |
395 | * Zero any uninitialised parts of the page, and then mark the page | |
396 | * as up to date if it turns out that we're extending the file. | |
397 | */ | |
398 | if (!PageUptodate(page)) { | |
399 | unsigned pglen = nfs_page_length(page); | |
400 | unsigned end = offset + len; | |
401 | ||
402 | if (pglen == 0) { | |
403 | zero_user_segments(page, 0, offset, | |
404 | end, PAGE_CACHE_SIZE); | |
405 | SetPageUptodate(page); | |
406 | } else if (end >= pglen) { | |
407 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
408 | if (offset == 0) | |
409 | SetPageUptodate(page); | |
410 | } else | |
411 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
412 | } | |
413 | ||
4899f9c8 | 414 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
415 | |
416 | unlock_page(page); | |
417 | page_cache_release(page); | |
418 | ||
3d509e54 CL |
419 | if (status < 0) |
420 | return status; | |
421 | return copied; | |
1da177e4 LT |
422 | } |
423 | ||
6b9b3514 DH |
424 | /* |
425 | * Partially or wholly invalidate a page | |
426 | * - Release the private state associated with a page if undergoing complete | |
427 | * page invalidation | |
545db45f | 428 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
429 | * - Caller holds page lock |
430 | */ | |
2ff28e22 | 431 | static void nfs_invalidate_page(struct page *page, unsigned long offset) |
cd52ed35 | 432 | { |
b7eaefaa CL |
433 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset); |
434 | ||
1c75950b TM |
435 | if (offset != 0) |
436 | return; | |
d2ccddf0 | 437 | /* Cancel any unstarted writes on this page */ |
1b3b4a1a | 438 | nfs_wb_page_cancel(page->mapping->host, page); |
545db45f DH |
439 | |
440 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
441 | } |
442 | ||
6b9b3514 DH |
443 | /* |
444 | * Attempt to release the private state associated with a page | |
545db45f | 445 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
446 | * - Caller holds page lock |
447 | * - Return true (may release page) or false (may not) | |
448 | */ | |
cd52ed35 TM |
449 | static int nfs_release_page(struct page *page, gfp_t gfp) |
450 | { | |
b608b283 TM |
451 | struct address_space *mapping = page->mapping; |
452 | ||
b7eaefaa CL |
453 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
454 | ||
d812e575 | 455 | /* Only do I/O if gfp is a superset of GFP_KERNEL */ |
b608b283 TM |
456 | if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL) { |
457 | int how = FLUSH_SYNC; | |
458 | ||
459 | /* Don't let kswapd deadlock waiting for OOM RPC calls */ | |
460 | if (current_is_kswapd()) | |
461 | how = 0; | |
462 | nfs_commit_inode(mapping->host, how); | |
463 | } | |
e3db7691 | 464 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
465 | if (PagePrivate(page)) |
466 | return 0; | |
467 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
468 | } |
469 | ||
6b9b3514 DH |
470 | /* |
471 | * Attempt to clear the private state associated with a page when an error | |
472 | * occurs that requires the cached contents of an inode to be written back or | |
473 | * destroyed | |
545db45f | 474 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
475 | * - Caller holds page lock |
476 | * - Return 0 if successful, -error otherwise | |
477 | */ | |
e3db7691 TM |
478 | static int nfs_launder_page(struct page *page) |
479 | { | |
b7eaefaa | 480 | struct inode *inode = page->mapping->host; |
545db45f | 481 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
482 | |
483 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
484 | inode->i_ino, (long long)page_offset(page)); | |
485 | ||
545db45f | 486 | nfs_fscache_wait_on_page_write(nfsi, page); |
b7eaefaa | 487 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
488 | } |
489 | ||
f5e54d6e | 490 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
491 | .readpage = nfs_readpage, |
492 | .readpages = nfs_readpages, | |
9cccef95 | 493 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
494 | .writepage = nfs_writepage, |
495 | .writepages = nfs_writepages, | |
4899f9c8 NP |
496 | .write_begin = nfs_write_begin, |
497 | .write_end = nfs_write_end, | |
cd52ed35 TM |
498 | .invalidatepage = nfs_invalidate_page, |
499 | .releasepage = nfs_release_page, | |
1da177e4 | 500 | .direct_IO = nfs_direct_IO, |
074cc1de | 501 | .migratepage = nfs_migrate_page, |
e3db7691 | 502 | .launder_page = nfs_launder_page, |
f590f333 | 503 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
504 | }; |
505 | ||
6b9b3514 DH |
506 | /* |
507 | * Notification that a PTE pointing to an NFS page is about to be made | |
508 | * writable, implying that someone is about to modify the page through a | |
509 | * shared-writable mapping | |
510 | */ | |
c2ec175c | 511 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 512 | { |
c2ec175c | 513 | struct page *page = vmf->page; |
94387fb1 | 514 | struct file *filp = vma->vm_file; |
b7eaefaa | 515 | struct dentry *dentry = filp->f_path.dentry; |
94387fb1 | 516 | unsigned pagelen; |
bc4866b6 | 517 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 518 | struct address_space *mapping; |
94387fb1 | 519 | |
b7eaefaa CL |
520 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n", |
521 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
522 | filp->f_mapping->host->i_ino, | |
523 | (long long)page_offset(page)); | |
524 | ||
545db45f DH |
525 | /* make sure the cache has finished storing the page */ |
526 | nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page); | |
527 | ||
94387fb1 | 528 | lock_page(page); |
4899f9c8 | 529 | mapping = page->mapping; |
b7eaefaa | 530 | if (mapping != dentry->d_inode->i_mapping) |
8b1f9ee5 TM |
531 | goto out_unlock; |
532 | ||
2aeb98f4 TM |
533 | wait_on_page_writeback(page); |
534 | ||
94387fb1 | 535 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
536 | if (pagelen == 0) |
537 | goto out_unlock; | |
4899f9c8 | 538 | |
bc4866b6 TM |
539 | ret = VM_FAULT_LOCKED; |
540 | if (nfs_flush_incompatible(filp, page) == 0 && | |
541 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
542 | goto out; | |
8b1f9ee5 | 543 | |
bc4866b6 | 544 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
545 | out_unlock: |
546 | unlock_page(page); | |
bc4866b6 TM |
547 | out: |
548 | return ret; | |
94387fb1 TM |
549 | } |
550 | ||
f0f37e2f | 551 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 TM |
552 | .fault = filemap_fault, |
553 | .page_mkwrite = nfs_vm_page_mkwrite, | |
554 | }; | |
555 | ||
7b159fc1 TM |
556 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
557 | { | |
558 | struct nfs_open_context *ctx; | |
559 | ||
6b2f3d1f | 560 | if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) |
7b159fc1 | 561 | return 1; |
cd3758e3 | 562 | ctx = nfs_file_open_context(filp); |
7b159fc1 TM |
563 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) |
564 | return 1; | |
565 | return 0; | |
566 | } | |
567 | ||
027445c3 BP |
568 | static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, |
569 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 570 | { |
01cce933 | 571 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 | 572 | struct inode * inode = dentry->d_inode; |
7e381172 | 573 | unsigned long written = 0; |
1da177e4 | 574 | ssize_t result; |
027445c3 | 575 | size_t count = iov_length(iov, nr_segs); |
1da177e4 | 576 | |
1da177e4 | 577 | if (iocb->ki_filp->f_flags & O_DIRECT) |
027445c3 | 578 | return nfs_file_direct_write(iocb, iov, nr_segs, pos); |
1da177e4 | 579 | |
6da24bc9 | 580 | dprintk("NFS: write(%s/%s, %lu@%Ld)\n", |
1da177e4 | 581 | dentry->d_parent->d_name.name, dentry->d_name.name, |
6da24bc9 | 582 | (unsigned long) count, (long long) pos); |
1da177e4 LT |
583 | |
584 | result = -EBUSY; | |
585 | if (IS_SWAPFILE(inode)) | |
586 | goto out_swapfile; | |
7d52e862 TM |
587 | /* |
588 | * O_APPEND implies that we must revalidate the file length. | |
589 | */ | |
590 | if (iocb->ki_filp->f_flags & O_APPEND) { | |
591 | result = nfs_revalidate_file_size(inode, iocb->ki_filp); | |
592 | if (result) | |
593 | goto out; | |
fe51beec | 594 | } |
1da177e4 LT |
595 | |
596 | result = count; | |
597 | if (!count) | |
598 | goto out; | |
599 | ||
027445c3 | 600 | result = generic_file_aio_write(iocb, iov, nr_segs, pos); |
7e381172 CL |
601 | if (result > 0) |
602 | written = result; | |
603 | ||
6b2f3d1f | 604 | /* Return error values for O_DSYNC and IS_SYNC() */ |
7b159fc1 | 605 | if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { |
af7fa165 | 606 | int err = vfs_fsync(iocb->ki_filp, 0); |
200baa21 TM |
607 | if (err < 0) |
608 | result = err; | |
609 | } | |
7e381172 CL |
610 | if (result > 0) |
611 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
1da177e4 LT |
612 | out: |
613 | return result; | |
614 | ||
615 | out_swapfile: | |
616 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
617 | goto out; | |
618 | } | |
619 | ||
bf40d343 SJ |
620 | static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, |
621 | struct file *filp, loff_t *ppos, | |
622 | size_t count, unsigned int flags) | |
623 | { | |
624 | struct dentry *dentry = filp->f_path.dentry; | |
625 | struct inode *inode = dentry->d_inode; | |
7e381172 | 626 | unsigned long written = 0; |
bf40d343 SJ |
627 | ssize_t ret; |
628 | ||
629 | dprintk("NFS splice_write(%s/%s, %lu@%llu)\n", | |
630 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
631 | (unsigned long) count, (unsigned long long) *ppos); | |
632 | ||
633 | /* | |
634 | * The combination of splice and an O_APPEND destination is disallowed. | |
635 | */ | |
636 | ||
bf40d343 | 637 | ret = generic_file_splice_write(pipe, filp, ppos, count, flags); |
7e381172 CL |
638 | if (ret > 0) |
639 | written = ret; | |
640 | ||
bf40d343 | 641 | if (ret >= 0 && nfs_need_sync_write(filp, inode)) { |
af7fa165 | 642 | int err = vfs_fsync(filp, 0); |
bf40d343 SJ |
643 | if (err < 0) |
644 | ret = err; | |
645 | } | |
7e381172 CL |
646 | if (ret > 0) |
647 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
bf40d343 SJ |
648 | return ret; |
649 | } | |
650 | ||
5eebde23 SJ |
651 | static int |
652 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
653 | { |
654 | struct inode *inode = filp->f_mapping->host; | |
655 | int status = 0; | |
21ac19d4 | 656 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 657 | |
039c4d7a | 658 | /* Try local locking first */ |
6d34ac19 BF |
659 | posix_test_lock(filp, fl); |
660 | if (fl->fl_type != F_UNLCK) { | |
661 | /* found a conflict */ | |
039c4d7a | 662 | goto out; |
1da177e4 | 663 | } |
21ac19d4 | 664 | fl->fl_type = saved_type; |
039c4d7a TM |
665 | |
666 | if (nfs_have_delegation(inode, FMODE_READ)) | |
667 | goto out_noconflict; | |
668 | ||
5eebde23 | 669 | if (is_local) |
039c4d7a TM |
670 | goto out_noconflict; |
671 | ||
672 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
673 | out: | |
1da177e4 | 674 | return status; |
039c4d7a TM |
675 | out_noconflict: |
676 | fl->fl_type = F_UNLCK; | |
677 | goto out; | |
1da177e4 LT |
678 | } |
679 | ||
680 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
681 | { | |
682 | int res = 0; | |
683 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
684 | case FL_POSIX: | |
685 | res = posix_lock_file_wait(file, fl); | |
686 | break; | |
687 | case FL_FLOCK: | |
688 | res = flock_lock_file_wait(file, fl); | |
689 | break; | |
690 | default: | |
691 | BUG(); | |
692 | } | |
1da177e4 LT |
693 | return res; |
694 | } | |
695 | ||
5eebde23 SJ |
696 | static int |
697 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
698 | { |
699 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
700 | int status; |
701 | ||
1da177e4 LT |
702 | /* |
703 | * Flush all pending writes before doing anything | |
704 | * with locks.. | |
705 | */ | |
29884df0 | 706 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 LT |
707 | |
708 | /* NOTE: special case | |
709 | * If we're signalled while cleaning up locks on process exit, we | |
710 | * still need to complete the unlock. | |
711 | */ | |
5eebde23 SJ |
712 | /* |
713 | * Use local locking if mounted with "-onolock" or with appropriate | |
714 | * "-olocal_lock=" | |
715 | */ | |
716 | if (!is_local) | |
1da177e4 LT |
717 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
718 | else | |
719 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
720 | return status; |
721 | } | |
722 | ||
6b96724e RL |
723 | static int |
724 | is_time_granular(struct timespec *ts) { | |
725 | return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000)); | |
726 | } | |
727 | ||
5eebde23 SJ |
728 | static int |
729 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
730 | { |
731 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
732 | int status; |
733 | ||
1da177e4 LT |
734 | /* |
735 | * Flush all pending writes before doing anything | |
736 | * with locks.. | |
737 | */ | |
29884df0 TM |
738 | status = nfs_sync_mapping(filp->f_mapping); |
739 | if (status != 0) | |
1da177e4 LT |
740 | goto out; |
741 | ||
5eebde23 SJ |
742 | /* |
743 | * Use local locking if mounted with "-onolock" or with appropriate | |
744 | * "-olocal_lock=" | |
745 | */ | |
746 | if (!is_local) | |
1da177e4 | 747 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 748 | else |
1da177e4 | 749 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
750 | if (status < 0) |
751 | goto out; | |
6b96724e | 752 | |
1da177e4 | 753 | /* |
6b96724e RL |
754 | * Revalidate the cache if the server has time stamps granular |
755 | * enough to detect subsecond changes. Otherwise, clear the | |
756 | * cache to prevent missing any changes. | |
757 | * | |
1da177e4 LT |
758 | * This makes locking act as a cache coherency point. |
759 | */ | |
29884df0 | 760 | nfs_sync_mapping(filp->f_mapping); |
6b96724e RL |
761 | if (!nfs_have_delegation(inode, FMODE_READ)) { |
762 | if (is_time_granular(&NFS_SERVER(inode)->time_delta)) | |
763 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
764 | else | |
765 | nfs_zap_caches(inode); | |
766 | } | |
1da177e4 | 767 | out: |
1da177e4 LT |
768 | return status; |
769 | } | |
770 | ||
771 | /* | |
772 | * Lock a (portion of) a file | |
773 | */ | |
774 | static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) | |
775 | { | |
6da24bc9 | 776 | struct inode *inode = filp->f_mapping->host; |
2116271a | 777 | int ret = -ENOLCK; |
5eebde23 | 778 | int is_local = 0; |
1da177e4 | 779 | |
6da24bc9 CL |
780 | dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n", |
781 | filp->f_path.dentry->d_parent->d_name.name, | |
782 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
783 | fl->fl_type, fl->fl_flags, |
784 | (long long)fl->fl_start, (long long)fl->fl_end); | |
6da24bc9 | 785 | |
91d5b470 | 786 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
787 | |
788 | /* No mandatory locks over NFS */ | |
dfad9441 | 789 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
790 | goto out_err; |
791 | ||
5eebde23 SJ |
792 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
793 | is_local = 1; | |
794 | ||
2116271a TM |
795 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
796 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
797 | if (ret < 0) | |
798 | goto out_err; | |
799 | } | |
1da177e4 LT |
800 | |
801 | if (IS_GETLK(cmd)) | |
5eebde23 | 802 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 803 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 804 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 805 | else |
5eebde23 | 806 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
807 | out_err: |
808 | return ret; | |
1da177e4 LT |
809 | } |
810 | ||
811 | /* | |
812 | * Lock a (portion of) a file | |
813 | */ | |
814 | static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) | |
815 | { | |
5eebde23 SJ |
816 | struct inode *inode = filp->f_mapping->host; |
817 | int is_local = 0; | |
818 | ||
6da24bc9 CL |
819 | dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n", |
820 | filp->f_path.dentry->d_parent->d_name.name, | |
821 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
822 | fl->fl_type, fl->fl_flags); |
823 | ||
1da177e4 LT |
824 | if (!(fl->fl_flags & FL_FLOCK)) |
825 | return -ENOLCK; | |
826 | ||
5eebde23 SJ |
827 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
828 | is_local = 1; | |
829 | ||
1da177e4 LT |
830 | /* We're simulating flock() locks using posix locks on the server */ |
831 | fl->fl_owner = (fl_owner_t)filp; | |
832 | fl->fl_start = 0; | |
833 | fl->fl_end = OFFSET_MAX; | |
834 | ||
835 | if (fl->fl_type == F_UNLCK) | |
5eebde23 SJ |
836 | return do_unlk(filp, cmd, fl, is_local); |
837 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 838 | } |
370f6599 | 839 | |
6da24bc9 CL |
840 | /* |
841 | * There is no protocol support for leases, so we have no way to implement | |
842 | * them correctly in the face of opens by other clients. | |
843 | */ | |
370f6599 BF |
844 | static int nfs_setlease(struct file *file, long arg, struct file_lock **fl) |
845 | { | |
6da24bc9 CL |
846 | dprintk("NFS: setlease(%s/%s, arg=%ld)\n", |
847 | file->f_path.dentry->d_parent->d_name.name, | |
848 | file->f_path.dentry->d_name.name, arg); | |
370f6599 BF |
849 | return -EINVAL; |
850 | } | |
1788ea6e | 851 | |
0486958f JL |
852 | const struct file_operations nfs_file_operations = { |
853 | .llseek = nfs_file_llseek, | |
854 | .read = do_sync_read, | |
855 | .write = do_sync_write, | |
856 | .aio_read = nfs_file_read, | |
857 | .aio_write = nfs_file_write, | |
858 | .mmap = nfs_file_mmap, | |
859 | .open = nfs_file_open, | |
860 | .flush = nfs_file_flush, | |
861 | .release = nfs_file_release, | |
862 | .fsync = nfs_file_fsync, | |
863 | .lock = nfs_lock, | |
864 | .flock = nfs_flock, | |
865 | .splice_read = nfs_file_splice_read, | |
866 | .splice_write = nfs_file_splice_write, | |
867 | .check_flags = nfs_check_flags, | |
868 | .setlease = nfs_setlease, | |
869 | }; | |
870 | ||
1788ea6e JL |
871 | #ifdef CONFIG_NFS_V4 |
872 | static int | |
873 | nfs4_file_open(struct inode *inode, struct file *filp) | |
874 | { | |
875 | /* | |
876 | * NFSv4 opens are handled in d_lookup and d_revalidate. If we get to | |
877 | * this point, then something is very wrong | |
878 | */ | |
879 | dprintk("NFS: %s called! inode=%p filp=%p\n", __func__, inode, filp); | |
880 | return -ENOTDIR; | |
881 | } | |
882 | ||
883 | const struct file_operations nfs4_file_operations = { | |
884 | .llseek = nfs_file_llseek, | |
885 | .read = do_sync_read, | |
886 | .write = do_sync_write, | |
887 | .aio_read = nfs_file_read, | |
888 | .aio_write = nfs_file_write, | |
889 | .mmap = nfs_file_mmap, | |
890 | .open = nfs4_file_open, | |
891 | .flush = nfs_file_flush, | |
892 | .release = nfs_file_release, | |
893 | .fsync = nfs_file_fsync, | |
894 | .lock = nfs_lock, | |
895 | .flock = nfs_flock, | |
896 | .splice_read = nfs_file_splice_read, | |
897 | .splice_write = nfs_file_splice_write, | |
898 | .check_flags = nfs_check_flags, | |
899 | .setlease = nfs_setlease, | |
900 | }; | |
901 | #endif /* CONFIG_NFS_V4 */ |