]>
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> | |
e8edc6e0 | 30 | #include <linux/aio.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" |
1da177e4 LT |
38 | |
39 | #define NFSDBG_FACILITY NFSDBG_FILE | |
40 | ||
41 | static int nfs_file_open(struct inode *, struct file *); | |
42 | static int nfs_file_release(struct inode *, struct file *); | |
980802e3 | 43 | static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin); |
1da177e4 | 44 | static int nfs_file_mmap(struct file *, struct vm_area_struct *); |
f0930fff JA |
45 | static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos, |
46 | struct pipe_inode_info *pipe, | |
47 | size_t count, unsigned int flags); | |
027445c3 BP |
48 | static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov, |
49 | unsigned long nr_segs, loff_t pos); | |
50 | static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov, | |
51 | unsigned long nr_segs, loff_t pos); | |
75e1fcc0 | 52 | static int nfs_file_flush(struct file *, fl_owner_t id); |
54917786 | 53 | static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync); |
1da177e4 LT |
54 | static int nfs_check_flags(int flags); |
55 | static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl); | |
56 | static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl); | |
370f6599 | 57 | static int nfs_setlease(struct file *file, long arg, struct file_lock **fl); |
1da177e4 | 58 | |
94387fb1 TM |
59 | static struct vm_operations_struct nfs_file_vm_ops; |
60 | ||
4b6f5d20 | 61 | const struct file_operations nfs_file_operations = { |
980802e3 | 62 | .llseek = nfs_file_llseek, |
1da177e4 LT |
63 | .read = do_sync_read, |
64 | .write = do_sync_write, | |
027445c3 BP |
65 | .aio_read = nfs_file_read, |
66 | .aio_write = nfs_file_write, | |
240ee831 | 67 | #ifdef CONFIG_MMU |
1da177e4 | 68 | .mmap = nfs_file_mmap, |
240ee831 BW |
69 | #else |
70 | .mmap = generic_file_mmap, | |
71 | #endif | |
1da177e4 LT |
72 | .open = nfs_file_open, |
73 | .flush = nfs_file_flush, | |
74 | .release = nfs_file_release, | |
54917786 | 75 | .fsync = nfs_file_fsync, |
1da177e4 LT |
76 | .lock = nfs_lock, |
77 | .flock = nfs_flock, | |
f0930fff | 78 | .splice_read = nfs_file_splice_read, |
1da177e4 | 79 | .check_flags = nfs_check_flags, |
370f6599 | 80 | .setlease = nfs_setlease, |
1da177e4 LT |
81 | }; |
82 | ||
92e1d5be | 83 | const struct inode_operations nfs_file_inode_operations = { |
1da177e4 LT |
84 | .permission = nfs_permission, |
85 | .getattr = nfs_getattr, | |
86 | .setattr = nfs_setattr, | |
87 | }; | |
88 | ||
b7fa0554 | 89 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 90 | const struct inode_operations nfs3_file_inode_operations = { |
b7fa0554 AG |
91 | .permission = nfs_permission, |
92 | .getattr = nfs_getattr, | |
93 | .setattr = nfs_setattr, | |
94 | .listxattr = nfs3_listxattr, | |
95 | .getxattr = nfs3_getxattr, | |
96 | .setxattr = nfs3_setxattr, | |
97 | .removexattr = nfs3_removexattr, | |
98 | }; | |
99 | #endif /* CONFIG_NFS_v3 */ | |
100 | ||
1da177e4 LT |
101 | /* Hack for future NFS swap support */ |
102 | #ifndef IS_SWAPFILE | |
103 | # define IS_SWAPFILE(inode) (0) | |
104 | #endif | |
105 | ||
106 | static int nfs_check_flags(int flags) | |
107 | { | |
108 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
109 | return -EINVAL; | |
110 | ||
111 | return 0; | |
112 | } | |
113 | ||
114 | /* | |
115 | * Open file | |
116 | */ | |
117 | static int | |
118 | nfs_file_open(struct inode *inode, struct file *filp) | |
119 | { | |
1da177e4 LT |
120 | int res; |
121 | ||
cc0dd2d1 CL |
122 | dfprintk(VFS, "NFS: open file(%s/%s)\n", |
123 | filp->f_path.dentry->d_parent->d_name.name, | |
124 | filp->f_path.dentry->d_name.name); | |
125 | ||
1da177e4 LT |
126 | res = nfs_check_flags(filp->f_flags); |
127 | if (res) | |
128 | return res; | |
129 | ||
91d5b470 | 130 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
1da177e4 | 131 | lock_kernel(); |
1f163415 | 132 | res = NFS_PROTO(inode)->file_open(inode, filp); |
1da177e4 LT |
133 | unlock_kernel(); |
134 | return res; | |
135 | } | |
136 | ||
137 | static int | |
138 | nfs_file_release(struct inode *inode, struct file *filp) | |
139 | { | |
140 | /* Ensure that dirty pages are flushed out with the right creds */ | |
141 | if (filp->f_mode & FMODE_WRITE) | |
a49c3c77 | 142 | nfs_wb_all(filp->f_path.dentry->d_inode); |
91d5b470 | 143 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
1da177e4 LT |
144 | return NFS_PROTO(inode)->file_release(inode, filp); |
145 | } | |
146 | ||
980802e3 TM |
147 | /** |
148 | * nfs_revalidate_size - Revalidate the file size | |
149 | * @inode - pointer to inode struct | |
150 | * @file - pointer to struct file | |
151 | * | |
152 | * Revalidates the file length. This is basically a wrapper around | |
153 | * nfs_revalidate_inode() that takes into account the fact that we may | |
154 | * have cached writes (in which case we don't care about the server's | |
155 | * idea of what the file length is), or O_DIRECT (in which case we | |
156 | * shouldn't trust the cache). | |
157 | */ | |
158 | static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) | |
159 | { | |
160 | struct nfs_server *server = NFS_SERVER(inode); | |
161 | struct nfs_inode *nfsi = NFS_I(inode); | |
162 | ||
163 | if (server->flags & NFS_MOUNT_NOAC) | |
164 | goto force_reval; | |
165 | if (filp->f_flags & O_DIRECT) | |
166 | goto force_reval; | |
167 | if (nfsi->npages != 0) | |
168 | return 0; | |
55296809 | 169 | if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode)) |
fe51beec | 170 | return 0; |
980802e3 TM |
171 | force_reval: |
172 | return __nfs_revalidate_inode(server, inode); | |
173 | } | |
174 | ||
175 | static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin) | |
176 | { | |
b84e06c5 CL |
177 | dfprintk(VFS, "NFS: llseek file(%s/%s, %lld, %d)\n", |
178 | filp->f_path.dentry->d_parent->d_name.name, | |
179 | filp->f_path.dentry->d_name.name, | |
180 | offset, origin); | |
181 | ||
980802e3 | 182 | /* origin == SEEK_END => we must revalidate the cached file length */ |
aec5e175 | 183 | if (origin == SEEK_END) { |
980802e3 TM |
184 | struct inode *inode = filp->f_mapping->host; |
185 | int retval = nfs_revalidate_file_size(inode, filp); | |
186 | if (retval < 0) | |
187 | return (loff_t)retval; | |
188 | } | |
189 | return remote_llseek(filp, offset, origin); | |
190 | } | |
191 | ||
7b159fc1 | 192 | /* |
54917786 | 193 | * Helper for nfs_file_flush() and nfs_file_fsync() |
7b159fc1 TM |
194 | * |
195 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
196 | * disk, but it retrieves and clears ctx->error after synching, despite | |
197 | * the two being set at the same time in nfs_context_set_write_error(). | |
198 | * This is because the former is used to notify the _next_ call to | |
199 | * nfs_file_write() that a write error occured, and hence cause it to | |
200 | * fall back to doing a synchronous write. | |
201 | */ | |
202 | static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode) | |
203 | { | |
204 | int have_error, status; | |
205 | int ret = 0; | |
206 | ||
207 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
208 | status = nfs_wb_all(inode); | |
209 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
210 | if (have_error) | |
211 | ret = xchg(&ctx->error, 0); | |
212 | if (!ret) | |
213 | ret = status; | |
214 | return ret; | |
215 | } | |
216 | ||
1da177e4 LT |
217 | /* |
218 | * Flush all dirty pages, and check for write errors. | |
219 | * | |
220 | */ | |
221 | static int | |
75e1fcc0 | 222 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 223 | { |
cd3758e3 | 224 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
01cce933 | 225 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
226 | int status; |
227 | ||
228 | dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino); | |
229 | ||
230 | if ((file->f_mode & FMODE_WRITE) == 0) | |
231 | return 0; | |
91d5b470 | 232 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
7b159fc1 | 233 | |
1da177e4 | 234 | /* Ensure that data+attribute caches are up to date after close() */ |
7b159fc1 TM |
235 | status = nfs_do_fsync(ctx, inode); |
236 | if (!status) | |
237 | nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
1da177e4 LT |
238 | return status; |
239 | } | |
240 | ||
241 | static ssize_t | |
027445c3 BP |
242 | nfs_file_read(struct kiocb *iocb, const struct iovec *iov, |
243 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 244 | { |
01cce933 | 245 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 LT |
246 | struct inode * inode = dentry->d_inode; |
247 | ssize_t result; | |
027445c3 | 248 | size_t count = iov_length(iov, nr_segs); |
1da177e4 | 249 | |
1da177e4 | 250 | if (iocb->ki_filp->f_flags & O_DIRECT) |
027445c3 | 251 | return nfs_file_direct_read(iocb, iov, nr_segs, pos); |
1da177e4 LT |
252 | |
253 | dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n", | |
254 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
255 | (unsigned long) count, (unsigned long) pos); | |
256 | ||
44b11874 | 257 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); |
91d5b470 | 258 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count); |
1da177e4 | 259 | if (!result) |
027445c3 | 260 | result = generic_file_aio_read(iocb, iov, nr_segs, pos); |
1da177e4 LT |
261 | return result; |
262 | } | |
263 | ||
264 | static ssize_t | |
f0930fff JA |
265 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
266 | struct pipe_inode_info *pipe, size_t count, | |
267 | unsigned int flags) | |
1da177e4 | 268 | { |
01cce933 | 269 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
270 | struct inode *inode = dentry->d_inode; |
271 | ssize_t res; | |
272 | ||
f0930fff | 273 | dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n", |
1da177e4 LT |
274 | dentry->d_parent->d_name.name, dentry->d_name.name, |
275 | (unsigned long) count, (unsigned long long) *ppos); | |
276 | ||
44b11874 | 277 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
1da177e4 | 278 | if (!res) |
f0930fff | 279 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
1da177e4 LT |
280 | return res; |
281 | } | |
282 | ||
283 | static int | |
284 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) | |
285 | { | |
01cce933 | 286 | struct dentry *dentry = file->f_path.dentry; |
1da177e4 LT |
287 | struct inode *inode = dentry->d_inode; |
288 | int status; | |
289 | ||
290 | dfprintk(VFS, "nfs: mmap(%s/%s)\n", | |
291 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
292 | ||
44b11874 | 293 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 TM |
294 | if (!status) { |
295 | vma->vm_ops = &nfs_file_vm_ops; | |
296 | vma->vm_flags |= VM_CAN_NONLINEAR; | |
297 | file_accessed(file); | |
298 | } | |
1da177e4 LT |
299 | return status; |
300 | } | |
301 | ||
302 | /* | |
303 | * Flush any dirty pages for this process, and check for write errors. | |
304 | * The return status from this call provides a reliable indication of | |
305 | * whether any write errors occurred for this process. | |
306 | */ | |
307 | static int | |
54917786 | 308 | nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync) |
1da177e4 | 309 | { |
cd3758e3 | 310 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 311 | struct inode *inode = dentry->d_inode; |
1da177e4 | 312 | |
54917786 CL |
313 | dfprintk(VFS, "NFS: fsync file(%s/%s) datasync %d\n", |
314 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
315 | datasync); | |
1da177e4 | 316 | |
91d5b470 | 317 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
7b159fc1 | 318 | return nfs_do_fsync(ctx, inode); |
1da177e4 LT |
319 | } |
320 | ||
321 | /* | |
4899f9c8 NP |
322 | * This does the "real" work of the write. We must allocate and lock the |
323 | * page to be sent back to the generic routine, which then copies the | |
324 | * data from user space. | |
1da177e4 LT |
325 | * |
326 | * If the writer ends up delaying the write, the writer needs to | |
327 | * increment the page use counts until he is done with the page. | |
328 | */ | |
4899f9c8 NP |
329 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
330 | loff_t pos, unsigned len, unsigned flags, | |
331 | struct page **pagep, void **fsdata) | |
1da177e4 | 332 | { |
4899f9c8 NP |
333 | int ret; |
334 | pgoff_t index; | |
335 | struct page *page; | |
336 | index = pos >> PAGE_CACHE_SHIFT; | |
337 | ||
338 | page = __grab_cache_page(mapping, index); | |
339 | if (!page) | |
340 | return -ENOMEM; | |
341 | *pagep = page; | |
342 | ||
343 | ret = nfs_flush_incompatible(file, page); | |
344 | if (ret) { | |
345 | unlock_page(page); | |
346 | page_cache_release(page); | |
347 | } | |
348 | return ret; | |
1da177e4 LT |
349 | } |
350 | ||
4899f9c8 NP |
351 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
352 | loff_t pos, unsigned len, unsigned copied, | |
353 | struct page *page, void *fsdata) | |
1da177e4 | 354 | { |
4899f9c8 NP |
355 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
356 | int status; | |
1da177e4 | 357 | |
efc91ed0 TM |
358 | /* |
359 | * Zero any uninitialised parts of the page, and then mark the page | |
360 | * as up to date if it turns out that we're extending the file. | |
361 | */ | |
362 | if (!PageUptodate(page)) { | |
363 | unsigned pglen = nfs_page_length(page); | |
364 | unsigned end = offset + len; | |
365 | ||
366 | if (pglen == 0) { | |
367 | zero_user_segments(page, 0, offset, | |
368 | end, PAGE_CACHE_SIZE); | |
369 | SetPageUptodate(page); | |
370 | } else if (end >= pglen) { | |
371 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
372 | if (offset == 0) | |
373 | SetPageUptodate(page); | |
374 | } else | |
375 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
376 | } | |
377 | ||
1da177e4 | 378 | lock_kernel(); |
4899f9c8 | 379 | status = nfs_updatepage(file, page, offset, copied); |
1da177e4 | 380 | unlock_kernel(); |
4899f9c8 NP |
381 | |
382 | unlock_page(page); | |
383 | page_cache_release(page); | |
384 | ||
3d509e54 CL |
385 | if (status < 0) |
386 | return status; | |
387 | return copied; | |
1da177e4 LT |
388 | } |
389 | ||
2ff28e22 | 390 | static void nfs_invalidate_page(struct page *page, unsigned long offset) |
cd52ed35 | 391 | { |
1c75950b TM |
392 | if (offset != 0) |
393 | return; | |
d2ccddf0 | 394 | /* Cancel any unstarted writes on this page */ |
1b3b4a1a | 395 | nfs_wb_page_cancel(page->mapping->host, page); |
cd52ed35 TM |
396 | } |
397 | ||
398 | static int nfs_release_page(struct page *page, gfp_t gfp) | |
399 | { | |
e3db7691 TM |
400 | /* If PagePrivate() is set, then the page is not freeable */ |
401 | return 0; | |
402 | } | |
403 | ||
404 | static int nfs_launder_page(struct page *page) | |
405 | { | |
406 | return nfs_wb_page(page->mapping->host, page); | |
cd52ed35 TM |
407 | } |
408 | ||
f5e54d6e | 409 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
410 | .readpage = nfs_readpage, |
411 | .readpages = nfs_readpages, | |
9cccef95 | 412 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
413 | .writepage = nfs_writepage, |
414 | .writepages = nfs_writepages, | |
4899f9c8 NP |
415 | .write_begin = nfs_write_begin, |
416 | .write_end = nfs_write_end, | |
cd52ed35 TM |
417 | .invalidatepage = nfs_invalidate_page, |
418 | .releasepage = nfs_release_page, | |
1da177e4 | 419 | .direct_IO = nfs_direct_IO, |
e3db7691 | 420 | .launder_page = nfs_launder_page, |
1da177e4 LT |
421 | }; |
422 | ||
94387fb1 TM |
423 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page) |
424 | { | |
425 | struct file *filp = vma->vm_file; | |
426 | unsigned pagelen; | |
427 | int ret = -EINVAL; | |
4899f9c8 | 428 | struct address_space *mapping; |
94387fb1 TM |
429 | |
430 | lock_page(page); | |
4899f9c8 | 431 | mapping = page->mapping; |
8b1f9ee5 TM |
432 | if (mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping) |
433 | goto out_unlock; | |
434 | ||
435 | ret = 0; | |
94387fb1 | 436 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
437 | if (pagelen == 0) |
438 | goto out_unlock; | |
4899f9c8 | 439 | |
8b1f9ee5 TM |
440 | ret = nfs_flush_incompatible(filp, page); |
441 | if (ret != 0) | |
442 | goto out_unlock; | |
443 | ||
444 | ret = nfs_updatepage(filp, page, 0, pagelen); | |
445 | if (ret == 0) | |
446 | ret = pagelen; | |
447 | out_unlock: | |
448 | unlock_page(page); | |
94387fb1 TM |
449 | return ret; |
450 | } | |
451 | ||
452 | static struct vm_operations_struct nfs_file_vm_ops = { | |
453 | .fault = filemap_fault, | |
454 | .page_mkwrite = nfs_vm_page_mkwrite, | |
455 | }; | |
456 | ||
7b159fc1 TM |
457 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
458 | { | |
459 | struct nfs_open_context *ctx; | |
460 | ||
461 | if (IS_SYNC(inode) || (filp->f_flags & O_SYNC)) | |
462 | return 1; | |
cd3758e3 | 463 | ctx = nfs_file_open_context(filp); |
7b159fc1 TM |
464 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) |
465 | return 1; | |
466 | return 0; | |
467 | } | |
468 | ||
027445c3 BP |
469 | static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, |
470 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 471 | { |
01cce933 | 472 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 LT |
473 | struct inode * inode = dentry->d_inode; |
474 | ssize_t result; | |
027445c3 | 475 | size_t count = iov_length(iov, nr_segs); |
1da177e4 | 476 | |
1da177e4 | 477 | if (iocb->ki_filp->f_flags & O_DIRECT) |
027445c3 | 478 | return nfs_file_direct_write(iocb, iov, nr_segs, pos); |
1da177e4 | 479 | |
027445c3 | 480 | dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n", |
1da177e4 | 481 | dentry->d_parent->d_name.name, dentry->d_name.name, |
027445c3 | 482 | inode->i_ino, (unsigned long) count, (long long) pos); |
1da177e4 LT |
483 | |
484 | result = -EBUSY; | |
485 | if (IS_SWAPFILE(inode)) | |
486 | goto out_swapfile; | |
7d52e862 TM |
487 | /* |
488 | * O_APPEND implies that we must revalidate the file length. | |
489 | */ | |
490 | if (iocb->ki_filp->f_flags & O_APPEND) { | |
491 | result = nfs_revalidate_file_size(inode, iocb->ki_filp); | |
492 | if (result) | |
493 | goto out; | |
fe51beec | 494 | } |
1da177e4 LT |
495 | |
496 | result = count; | |
497 | if (!count) | |
498 | goto out; | |
499 | ||
91d5b470 | 500 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count); |
027445c3 | 501 | result = generic_file_aio_write(iocb, iov, nr_segs, pos); |
200baa21 | 502 | /* Return error values for O_SYNC and IS_SYNC() */ |
7b159fc1 | 503 | if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { |
cd3758e3 | 504 | int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode); |
200baa21 TM |
505 | if (err < 0) |
506 | result = err; | |
507 | } | |
1da177e4 LT |
508 | out: |
509 | return result; | |
510 | ||
511 | out_swapfile: | |
512 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
513 | goto out; | |
514 | } | |
515 | ||
516 | static int do_getlk(struct file *filp, int cmd, struct file_lock *fl) | |
517 | { | |
518 | struct inode *inode = filp->f_mapping->host; | |
519 | int status = 0; | |
520 | ||
521 | lock_kernel(); | |
039c4d7a | 522 | /* Try local locking first */ |
6d34ac19 BF |
523 | posix_test_lock(filp, fl); |
524 | if (fl->fl_type != F_UNLCK) { | |
525 | /* found a conflict */ | |
039c4d7a | 526 | goto out; |
1da177e4 | 527 | } |
039c4d7a TM |
528 | |
529 | if (nfs_have_delegation(inode, FMODE_READ)) | |
530 | goto out_noconflict; | |
531 | ||
532 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM) | |
533 | goto out_noconflict; | |
534 | ||
535 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
536 | out: | |
1da177e4 LT |
537 | unlock_kernel(); |
538 | return status; | |
039c4d7a TM |
539 | out_noconflict: |
540 | fl->fl_type = F_UNLCK; | |
541 | goto out; | |
1da177e4 LT |
542 | } |
543 | ||
544 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
545 | { | |
546 | int res = 0; | |
547 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
548 | case FL_POSIX: | |
549 | res = posix_lock_file_wait(file, fl); | |
550 | break; | |
551 | case FL_FLOCK: | |
552 | res = flock_lock_file_wait(file, fl); | |
553 | break; | |
554 | default: | |
555 | BUG(); | |
556 | } | |
557 | if (res < 0) | |
46bae1a9 NB |
558 | dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager" |
559 | " - error %d!\n", | |
3110ff80 | 560 | __func__, res); |
1da177e4 LT |
561 | return res; |
562 | } | |
563 | ||
564 | static int do_unlk(struct file *filp, int cmd, struct file_lock *fl) | |
565 | { | |
566 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
567 | int status; |
568 | ||
1da177e4 LT |
569 | /* |
570 | * Flush all pending writes before doing anything | |
571 | * with locks.. | |
572 | */ | |
29884df0 | 573 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 LT |
574 | |
575 | /* NOTE: special case | |
576 | * If we're signalled while cleaning up locks on process exit, we | |
577 | * still need to complete the unlock. | |
578 | */ | |
579 | lock_kernel(); | |
580 | /* Use local locking if mounted with "-onolock" */ | |
581 | if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) | |
582 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
583 | else | |
584 | status = do_vfs_lock(filp, fl); | |
585 | unlock_kernel(); | |
1da177e4 LT |
586 | return status; |
587 | } | |
588 | ||
589 | static int do_setlk(struct file *filp, int cmd, struct file_lock *fl) | |
590 | { | |
591 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
592 | int status; |
593 | ||
1da177e4 LT |
594 | /* |
595 | * Flush all pending writes before doing anything | |
596 | * with locks.. | |
597 | */ | |
29884df0 TM |
598 | status = nfs_sync_mapping(filp->f_mapping); |
599 | if (status != 0) | |
1da177e4 LT |
600 | goto out; |
601 | ||
602 | lock_kernel(); | |
603 | /* Use local locking if mounted with "-onolock" */ | |
c4d7c402 | 604 | if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) |
1da177e4 | 605 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 606 | else |
1da177e4 LT |
607 | status = do_vfs_lock(filp, fl); |
608 | unlock_kernel(); | |
609 | if (status < 0) | |
610 | goto out; | |
611 | /* | |
612 | * Make sure we clear the cache whenever we try to get the lock. | |
613 | * This makes locking act as a cache coherency point. | |
614 | */ | |
29884df0 | 615 | nfs_sync_mapping(filp->f_mapping); |
b5418383 TM |
616 | if (!nfs_have_delegation(inode, FMODE_READ)) |
617 | nfs_zap_caches(inode); | |
1da177e4 | 618 | out: |
1da177e4 LT |
619 | return status; |
620 | } | |
621 | ||
622 | /* | |
623 | * Lock a (portion of) a file | |
624 | */ | |
625 | static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) | |
626 | { | |
627 | struct inode * inode = filp->f_mapping->host; | |
2116271a | 628 | int ret = -ENOLCK; |
1da177e4 LT |
629 | |
630 | dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n", | |
631 | inode->i_sb->s_id, inode->i_ino, | |
632 | fl->fl_type, fl->fl_flags, | |
633 | (long long)fl->fl_start, (long long)fl->fl_end); | |
91d5b470 | 634 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
635 | |
636 | /* No mandatory locks over NFS */ | |
dfad9441 | 637 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
638 | goto out_err; |
639 | ||
640 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { | |
641 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
642 | if (ret < 0) | |
643 | goto out_err; | |
644 | } | |
1da177e4 LT |
645 | |
646 | if (IS_GETLK(cmd)) | |
2116271a TM |
647 | ret = do_getlk(filp, cmd, fl); |
648 | else if (fl->fl_type == F_UNLCK) | |
649 | ret = do_unlk(filp, cmd, fl); | |
650 | else | |
651 | ret = do_setlk(filp, cmd, fl); | |
652 | out_err: | |
653 | return ret; | |
1da177e4 LT |
654 | } |
655 | ||
656 | /* | |
657 | * Lock a (portion of) a file | |
658 | */ | |
659 | static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) | |
660 | { | |
1da177e4 | 661 | dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n", |
01cce933 JJS |
662 | filp->f_path.dentry->d_inode->i_sb->s_id, |
663 | filp->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
664 | fl->fl_type, fl->fl_flags); |
665 | ||
1da177e4 LT |
666 | /* |
667 | * No BSD flocks over NFS allowed. | |
668 | * Note: we could try to fake a POSIX lock request here by | |
669 | * using ((u32) filp | 0x80000000) or some such as the pid. | |
670 | * Not sure whether that would be unique, though, or whether | |
671 | * that would break in other places. | |
672 | */ | |
673 | if (!(fl->fl_flags & FL_FLOCK)) | |
674 | return -ENOLCK; | |
675 | ||
676 | /* We're simulating flock() locks using posix locks on the server */ | |
677 | fl->fl_owner = (fl_owner_t)filp; | |
678 | fl->fl_start = 0; | |
679 | fl->fl_end = OFFSET_MAX; | |
680 | ||
681 | if (fl->fl_type == F_UNLCK) | |
682 | return do_unlk(filp, cmd, fl); | |
683 | return do_setlk(filp, cmd, fl); | |
684 | } | |
370f6599 BF |
685 | |
686 | static int nfs_setlease(struct file *file, long arg, struct file_lock **fl) | |
687 | { | |
688 | /* | |
689 | * There is no protocol support for leases, so we have no way | |
690 | * to implement them correctly in the face of opens by other | |
691 | * clients. | |
692 | */ | |
693 | return -EINVAL; | |
694 | } |