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