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