]>
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> | |
27 | #include <linux/slab.h> | |
28 | #include <linux/pagemap.h> | |
29 | #include <linux/smp_lock.h> | |
30 | ||
31 | #include <asm/uaccess.h> | |
32 | #include <asm/system.h> | |
33 | ||
34 | #include "delegation.h" | |
35 | ||
36 | #define NFSDBG_FACILITY NFSDBG_FILE | |
37 | ||
38 | static int nfs_file_open(struct inode *, struct file *); | |
39 | static int nfs_file_release(struct inode *, struct file *); | |
40 | static int nfs_file_mmap(struct file *, struct vm_area_struct *); | |
41 | static ssize_t nfs_file_sendfile(struct file *, loff_t *, size_t, read_actor_t, void *); | |
42 | static ssize_t nfs_file_read(struct kiocb *, char __user *, size_t, loff_t); | |
43 | static ssize_t nfs_file_write(struct kiocb *, const char __user *, size_t, loff_t); | |
44 | static int nfs_file_flush(struct file *); | |
45 | static int nfs_fsync(struct file *, struct dentry *dentry, int datasync); | |
46 | static int nfs_check_flags(int flags); | |
47 | static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl); | |
48 | static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl); | |
49 | ||
50 | struct file_operations nfs_file_operations = { | |
51 | .llseek = remote_llseek, | |
52 | .read = do_sync_read, | |
53 | .write = do_sync_write, | |
54 | .aio_read = nfs_file_read, | |
55 | .aio_write = nfs_file_write, | |
56 | .mmap = nfs_file_mmap, | |
57 | .open = nfs_file_open, | |
58 | .flush = nfs_file_flush, | |
59 | .release = nfs_file_release, | |
60 | .fsync = nfs_fsync, | |
61 | .lock = nfs_lock, | |
62 | .flock = nfs_flock, | |
63 | .sendfile = nfs_file_sendfile, | |
64 | .check_flags = nfs_check_flags, | |
65 | }; | |
66 | ||
67 | struct inode_operations nfs_file_inode_operations = { | |
68 | .permission = nfs_permission, | |
69 | .getattr = nfs_getattr, | |
70 | .setattr = nfs_setattr, | |
71 | }; | |
72 | ||
73 | /* Hack for future NFS swap support */ | |
74 | #ifndef IS_SWAPFILE | |
75 | # define IS_SWAPFILE(inode) (0) | |
76 | #endif | |
77 | ||
78 | static int nfs_check_flags(int flags) | |
79 | { | |
80 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
81 | return -EINVAL; | |
82 | ||
83 | return 0; | |
84 | } | |
85 | ||
86 | /* | |
87 | * Open file | |
88 | */ | |
89 | static int | |
90 | nfs_file_open(struct inode *inode, struct file *filp) | |
91 | { | |
92 | struct nfs_server *server = NFS_SERVER(inode); | |
93 | int (*open)(struct inode *, struct file *); | |
94 | int res; | |
95 | ||
96 | res = nfs_check_flags(filp->f_flags); | |
97 | if (res) | |
98 | return res; | |
99 | ||
100 | lock_kernel(); | |
101 | /* Do NFSv4 open() call */ | |
102 | if ((open = server->rpc_ops->file_open) != NULL) | |
103 | res = open(inode, filp); | |
104 | unlock_kernel(); | |
105 | return res; | |
106 | } | |
107 | ||
108 | static int | |
109 | nfs_file_release(struct inode *inode, struct file *filp) | |
110 | { | |
111 | /* Ensure that dirty pages are flushed out with the right creds */ | |
112 | if (filp->f_mode & FMODE_WRITE) | |
113 | filemap_fdatawrite(filp->f_mapping); | |
114 | return NFS_PROTO(inode)->file_release(inode, filp); | |
115 | } | |
116 | ||
117 | /* | |
118 | * Flush all dirty pages, and check for write errors. | |
119 | * | |
120 | */ | |
121 | static int | |
122 | nfs_file_flush(struct file *file) | |
123 | { | |
124 | struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; | |
125 | struct inode *inode = file->f_dentry->d_inode; | |
126 | int status; | |
127 | ||
128 | dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino); | |
129 | ||
130 | if ((file->f_mode & FMODE_WRITE) == 0) | |
131 | return 0; | |
132 | lock_kernel(); | |
133 | /* Ensure that data+attribute caches are up to date after close() */ | |
134 | status = nfs_wb_all(inode); | |
135 | if (!status) { | |
136 | status = ctx->error; | |
137 | ctx->error = 0; | |
138 | if (!status && !nfs_have_delegation(inode, FMODE_READ)) | |
139 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
140 | } | |
141 | unlock_kernel(); | |
142 | return status; | |
143 | } | |
144 | ||
145 | static ssize_t | |
146 | nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos) | |
147 | { | |
148 | struct dentry * dentry = iocb->ki_filp->f_dentry; | |
149 | struct inode * inode = dentry->d_inode; | |
150 | ssize_t result; | |
151 | ||
152 | #ifdef CONFIG_NFS_DIRECTIO | |
153 | if (iocb->ki_filp->f_flags & O_DIRECT) | |
154 | return nfs_file_direct_read(iocb, buf, count, pos); | |
155 | #endif | |
156 | ||
157 | dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n", | |
158 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
159 | (unsigned long) count, (unsigned long) pos); | |
160 | ||
161 | result = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
162 | if (!result) | |
163 | result = generic_file_aio_read(iocb, buf, count, pos); | |
164 | return result; | |
165 | } | |
166 | ||
167 | static ssize_t | |
168 | nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count, | |
169 | read_actor_t actor, void *target) | |
170 | { | |
171 | struct dentry *dentry = filp->f_dentry; | |
172 | struct inode *inode = dentry->d_inode; | |
173 | ssize_t res; | |
174 | ||
175 | dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n", | |
176 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
177 | (unsigned long) count, (unsigned long long) *ppos); | |
178 | ||
179 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
180 | if (!res) | |
181 | res = generic_file_sendfile(filp, ppos, count, actor, target); | |
182 | return res; | |
183 | } | |
184 | ||
185 | static int | |
186 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) | |
187 | { | |
188 | struct dentry *dentry = file->f_dentry; | |
189 | struct inode *inode = dentry->d_inode; | |
190 | int status; | |
191 | ||
192 | dfprintk(VFS, "nfs: mmap(%s/%s)\n", | |
193 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
194 | ||
195 | status = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
196 | if (!status) | |
197 | status = generic_file_mmap(file, vma); | |
198 | return status; | |
199 | } | |
200 | ||
201 | /* | |
202 | * Flush any dirty pages for this process, and check for write errors. | |
203 | * The return status from this call provides a reliable indication of | |
204 | * whether any write errors occurred for this process. | |
205 | */ | |
206 | static int | |
207 | nfs_fsync(struct file *file, struct dentry *dentry, int datasync) | |
208 | { | |
209 | struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; | |
210 | struct inode *inode = dentry->d_inode; | |
211 | int status; | |
212 | ||
213 | dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino); | |
214 | ||
215 | lock_kernel(); | |
216 | status = nfs_wb_all(inode); | |
217 | if (!status) { | |
218 | status = ctx->error; | |
219 | ctx->error = 0; | |
220 | } | |
221 | unlock_kernel(); | |
222 | return status; | |
223 | } | |
224 | ||
225 | /* | |
226 | * This does the "real" work of the write. The generic routine has | |
227 | * allocated the page, locked it, done all the page alignment stuff | |
228 | * calculations etc. Now we should just copy the data from user | |
229 | * space and write it back to the real medium.. | |
230 | * | |
231 | * If the writer ends up delaying the write, the writer needs to | |
232 | * increment the page use counts until he is done with the page. | |
233 | */ | |
234 | static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to) | |
235 | { | |
236 | return nfs_flush_incompatible(file, page); | |
237 | } | |
238 | ||
239 | static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to) | |
240 | { | |
241 | long status; | |
242 | ||
243 | lock_kernel(); | |
244 | status = nfs_updatepage(file, page, offset, to-offset); | |
245 | unlock_kernel(); | |
246 | return status; | |
247 | } | |
248 | ||
249 | struct address_space_operations nfs_file_aops = { | |
250 | .readpage = nfs_readpage, | |
251 | .readpages = nfs_readpages, | |
252 | .set_page_dirty = __set_page_dirty_nobuffers, | |
253 | .writepage = nfs_writepage, | |
254 | .writepages = nfs_writepages, | |
255 | .prepare_write = nfs_prepare_write, | |
256 | .commit_write = nfs_commit_write, | |
257 | #ifdef CONFIG_NFS_DIRECTIO | |
258 | .direct_IO = nfs_direct_IO, | |
259 | #endif | |
260 | }; | |
261 | ||
262 | /* | |
263 | * Write to a file (through the page cache). | |
264 | */ | |
265 | static ssize_t | |
266 | nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos) | |
267 | { | |
268 | struct dentry * dentry = iocb->ki_filp->f_dentry; | |
269 | struct inode * inode = dentry->d_inode; | |
270 | ssize_t result; | |
271 | ||
272 | #ifdef CONFIG_NFS_DIRECTIO | |
273 | if (iocb->ki_filp->f_flags & O_DIRECT) | |
274 | return nfs_file_direct_write(iocb, buf, count, pos); | |
275 | #endif | |
276 | ||
277 | dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n", | |
278 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
279 | inode->i_ino, (unsigned long) count, (unsigned long) pos); | |
280 | ||
281 | result = -EBUSY; | |
282 | if (IS_SWAPFILE(inode)) | |
283 | goto out_swapfile; | |
284 | result = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
285 | if (result) | |
286 | goto out; | |
287 | ||
288 | result = count; | |
289 | if (!count) | |
290 | goto out; | |
291 | ||
292 | result = generic_file_aio_write(iocb, buf, count, pos); | |
293 | out: | |
294 | return result; | |
295 | ||
296 | out_swapfile: | |
297 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
298 | goto out; | |
299 | } | |
300 | ||
301 | static int do_getlk(struct file *filp, int cmd, struct file_lock *fl) | |
302 | { | |
303 | struct inode *inode = filp->f_mapping->host; | |
304 | int status = 0; | |
305 | ||
306 | lock_kernel(); | |
307 | /* Use local locking if mounted with "-onolock" */ | |
308 | if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) | |
309 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
310 | else { | |
311 | struct file_lock *cfl = posix_test_lock(filp, fl); | |
312 | ||
313 | fl->fl_type = F_UNLCK; | |
314 | if (cfl != NULL) | |
315 | memcpy(fl, cfl, sizeof(*fl)); | |
316 | } | |
317 | unlock_kernel(); | |
318 | return status; | |
319 | } | |
320 | ||
321 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
322 | { | |
323 | int res = 0; | |
324 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
325 | case FL_POSIX: | |
326 | res = posix_lock_file_wait(file, fl); | |
327 | break; | |
328 | case FL_FLOCK: | |
329 | res = flock_lock_file_wait(file, fl); | |
330 | break; | |
331 | default: | |
332 | BUG(); | |
333 | } | |
334 | if (res < 0) | |
335 | printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n", | |
336 | __FUNCTION__); | |
337 | return res; | |
338 | } | |
339 | ||
340 | static int do_unlk(struct file *filp, int cmd, struct file_lock *fl) | |
341 | { | |
342 | struct inode *inode = filp->f_mapping->host; | |
343 | sigset_t oldset; | |
344 | int status; | |
345 | ||
346 | rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset); | |
347 | /* | |
348 | * Flush all pending writes before doing anything | |
349 | * with locks.. | |
350 | */ | |
351 | filemap_fdatawrite(filp->f_mapping); | |
352 | down(&inode->i_sem); | |
353 | nfs_wb_all(inode); | |
354 | up(&inode->i_sem); | |
355 | filemap_fdatawait(filp->f_mapping); | |
356 | ||
357 | /* NOTE: special case | |
358 | * If we're signalled while cleaning up locks on process exit, we | |
359 | * still need to complete the unlock. | |
360 | */ | |
361 | lock_kernel(); | |
362 | /* Use local locking if mounted with "-onolock" */ | |
363 | if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) | |
364 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
365 | else | |
366 | status = do_vfs_lock(filp, fl); | |
367 | unlock_kernel(); | |
368 | rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset); | |
369 | return status; | |
370 | } | |
371 | ||
372 | static int do_setlk(struct file *filp, int cmd, struct file_lock *fl) | |
373 | { | |
374 | struct inode *inode = filp->f_mapping->host; | |
375 | sigset_t oldset; | |
376 | int status; | |
377 | ||
378 | rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset); | |
379 | /* | |
380 | * Flush all pending writes before doing anything | |
381 | * with locks.. | |
382 | */ | |
383 | status = filemap_fdatawrite(filp->f_mapping); | |
384 | if (status == 0) { | |
385 | down(&inode->i_sem); | |
386 | status = nfs_wb_all(inode); | |
387 | up(&inode->i_sem); | |
388 | if (status == 0) | |
389 | status = filemap_fdatawait(filp->f_mapping); | |
390 | } | |
391 | if (status < 0) | |
392 | goto out; | |
393 | ||
394 | lock_kernel(); | |
395 | /* Use local locking if mounted with "-onolock" */ | |
396 | if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) { | |
397 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
398 | /* If we were signalled we still need to ensure that | |
399 | * we clean up any state on the server. We therefore | |
400 | * record the lock call as having succeeded in order to | |
401 | * ensure that locks_remove_posix() cleans it out when | |
402 | * the process exits. | |
403 | */ | |
404 | if (status == -EINTR || status == -ERESTARTSYS) | |
405 | do_vfs_lock(filp, fl); | |
406 | } else | |
407 | status = do_vfs_lock(filp, fl); | |
408 | unlock_kernel(); | |
409 | if (status < 0) | |
410 | goto out; | |
411 | /* | |
412 | * Make sure we clear the cache whenever we try to get the lock. | |
413 | * This makes locking act as a cache coherency point. | |
414 | */ | |
415 | filemap_fdatawrite(filp->f_mapping); | |
416 | down(&inode->i_sem); | |
417 | nfs_wb_all(inode); /* we may have slept */ | |
418 | up(&inode->i_sem); | |
419 | filemap_fdatawait(filp->f_mapping); | |
420 | nfs_zap_caches(inode); | |
421 | out: | |
422 | rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset); | |
423 | return status; | |
424 | } | |
425 | ||
426 | /* | |
427 | * Lock a (portion of) a file | |
428 | */ | |
429 | static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) | |
430 | { | |
431 | struct inode * inode = filp->f_mapping->host; | |
432 | ||
433 | dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n", | |
434 | inode->i_sb->s_id, inode->i_ino, | |
435 | fl->fl_type, fl->fl_flags, | |
436 | (long long)fl->fl_start, (long long)fl->fl_end); | |
437 | ||
438 | if (!inode) | |
439 | return -EINVAL; | |
440 | ||
441 | /* No mandatory locks over NFS */ | |
442 | if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID) | |
443 | return -ENOLCK; | |
444 | ||
445 | if (IS_GETLK(cmd)) | |
446 | return do_getlk(filp, cmd, fl); | |
447 | if (fl->fl_type == F_UNLCK) | |
448 | return do_unlk(filp, cmd, fl); | |
449 | return do_setlk(filp, cmd, fl); | |
450 | } | |
451 | ||
452 | /* | |
453 | * Lock a (portion of) a file | |
454 | */ | |
455 | static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) | |
456 | { | |
457 | struct inode * inode = filp->f_mapping->host; | |
458 | ||
459 | dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n", | |
460 | inode->i_sb->s_id, inode->i_ino, | |
461 | fl->fl_type, fl->fl_flags); | |
462 | ||
463 | if (!inode) | |
464 | return -EINVAL; | |
465 | ||
466 | /* | |
467 | * No BSD flocks over NFS allowed. | |
468 | * Note: we could try to fake a POSIX lock request here by | |
469 | * using ((u32) filp | 0x80000000) or some such as the pid. | |
470 | * Not sure whether that would be unique, though, or whether | |
471 | * that would break in other places. | |
472 | */ | |
473 | if (!(fl->fl_flags & FL_FLOCK)) | |
474 | return -ENOLCK; | |
475 | ||
476 | /* We're simulating flock() locks using posix locks on the server */ | |
477 | fl->fl_owner = (fl_owner_t)filp; | |
478 | fl->fl_start = 0; | |
479 | fl->fl_end = OFFSET_MAX; | |
480 | ||
481 | if (fl->fl_type == F_UNLCK) | |
482 | return do_unlk(filp, cmd, fl); | |
483 | return do_setlk(filp, cmd, fl); | |
484 | } |