Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/file.c | |
3 | * | |
4 | * Copyright (C) 1992 Rick Sladkey | |
5 | * | |
6 | * Changes Copyright (C) 1994 by Florian La Roche | |
7 | * - Do not copy data too often around in the kernel. | |
8 | * - In nfs_file_read the return value of kmalloc wasn't checked. | |
9 | * - Put in a better version of read look-ahead buffering. Original idea | |
10 | * and implementation by Wai S Kok elekokws@ee.nus.sg. | |
11 | * | |
12 | * Expire cache on write to a file by Wai S Kok (Oct 1994). | |
13 | * | |
14 | * Total rewrite of read side for new NFS buffer cache.. Linus. | |
15 | * | |
16 | * nfs regular file handling functions | |
17 | */ | |
18 | ||
19 | #include <linux/time.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/fcntl.h> | |
23 | #include <linux/stat.h> | |
24 | #include <linux/nfs_fs.h> | |
25 | #include <linux/nfs_mount.h> | |
26 | #include <linux/mm.h> | |
1da177e4 | 27 | #include <linux/pagemap.h> |
e8edc6e0 | 28 | #include <linux/aio.h> |
5a0e3ad6 | 29 | #include <linux/gfp.h> |
b608b283 | 30 | #include <linux/swap.h> |
1da177e4 LT |
31 | |
32 | #include <asm/uaccess.h> | |
33 | #include <asm/system.h> | |
34 | ||
35 | #include "delegation.h" | |
94387fb1 | 36 | #include "internal.h" |
91d5b470 | 37 | #include "iostat.h" |
545db45f | 38 | #include "fscache.h" |
e5e94017 | 39 | #include "pnfs.h" |
1da177e4 LT |
40 | |
41 | #define NFSDBG_FACILITY NFSDBG_FILE | |
42 | ||
43 | static int nfs_file_open(struct inode *, struct file *); | |
44 | static int nfs_file_release(struct inode *, struct file *); | |
980802e3 | 45 | static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin); |
1da177e4 | 46 | static int nfs_file_mmap(struct file *, struct vm_area_struct *); |
f0930fff JA |
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); | |
027445c3 BP |
50 | static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov, |
51 | unsigned long nr_segs, loff_t pos); | |
bf40d343 SJ |
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); | |
027445c3 BP |
55 | static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov, |
56 | unsigned long nr_segs, loff_t pos); | |
75e1fcc0 | 57 | static int nfs_file_flush(struct file *, fl_owner_t id); |
7ea80859 | 58 | static int nfs_file_fsync(struct file *, int datasync); |
1da177e4 LT |
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); | |
370f6599 | 62 | static int nfs_setlease(struct file *file, long arg, struct file_lock **fl); |
1da177e4 | 63 | |
f0f37e2f | 64 | static const struct vm_operations_struct nfs_file_vm_ops; |
94387fb1 | 65 | |
4b6f5d20 | 66 | const struct file_operations nfs_file_operations = { |
980802e3 | 67 | .llseek = nfs_file_llseek, |
1da177e4 LT |
68 | .read = do_sync_read, |
69 | .write = do_sync_write, | |
027445c3 BP |
70 | .aio_read = nfs_file_read, |
71 | .aio_write = nfs_file_write, | |
1da177e4 LT |
72 | .mmap = nfs_file_mmap, |
73 | .open = nfs_file_open, | |
74 | .flush = nfs_file_flush, | |
75 | .release = nfs_file_release, | |
54917786 | 76 | .fsync = nfs_file_fsync, |
1da177e4 LT |
77 | .lock = nfs_lock, |
78 | .flock = nfs_flock, | |
f0930fff | 79 | .splice_read = nfs_file_splice_read, |
bf40d343 | 80 | .splice_write = nfs_file_splice_write, |
1da177e4 | 81 | .check_flags = nfs_check_flags, |
370f6599 | 82 | .setlease = nfs_setlease, |
1da177e4 LT |
83 | }; |
84 | ||
92e1d5be | 85 | const struct inode_operations nfs_file_inode_operations = { |
1da177e4 LT |
86 | .permission = nfs_permission, |
87 | .getattr = nfs_getattr, | |
88 | .setattr = nfs_setattr, | |
89 | }; | |
90 | ||
b7fa0554 | 91 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 92 | const struct inode_operations nfs3_file_inode_operations = { |
b7fa0554 AG |
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 | ||
1da177e4 LT |
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 | { | |
1da177e4 LT |
122 | int res; |
123 | ||
6da24bc9 | 124 | dprintk("NFS: open file(%s/%s)\n", |
cc0dd2d1 CL |
125 | filp->f_path.dentry->d_parent->d_name.name, |
126 | filp->f_path.dentry->d_name.name); | |
127 | ||
c2459dc4 | 128 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
1da177e4 LT |
129 | res = nfs_check_flags(filp->f_flags); |
130 | if (res) | |
131 | return res; | |
132 | ||
46cb650c | 133 | res = nfs_open(inode, filp); |
1da177e4 LT |
134 | return res; |
135 | } | |
136 | ||
137 | static int | |
138 | nfs_file_release(struct inode *inode, struct file *filp) | |
139 | { | |
6da24bc9 CL |
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 | ||
91d5b470 | 146 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
46cb650c | 147 | return nfs_release(inode, filp); |
1da177e4 LT |
148 | } |
149 | ||
980802e3 TM |
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 | ||
d7cf8dd0 TM |
166 | if (nfs_have_delegated_attributes(inode)) |
167 | goto out_noreval; | |
168 | ||
980802e3 TM |
169 | if (filp->f_flags & O_DIRECT) |
170 | goto force_reval; | |
d7cf8dd0 TM |
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; | |
980802e3 TM |
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 | { | |
9465efc9 | 183 | loff_t loff; |
e89e896d | 184 | |
6da24bc9 | 185 | dprintk("NFS: llseek file(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
186 | filp->f_path.dentry->d_parent->d_name.name, |
187 | filp->f_path.dentry->d_name.name, | |
188 | offset, origin); | |
189 | ||
980802e3 | 190 | /* origin == SEEK_END => we must revalidate the cached file length */ |
aec5e175 | 191 | if (origin == SEEK_END) { |
980802e3 | 192 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 193 | |
980802e3 TM |
194 | int retval = nfs_revalidate_file_size(inode, filp); |
195 | if (retval < 0) | |
196 | return (loff_t)retval; | |
d5e66348 TM |
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); | |
9465efc9 | 203 | return loff; |
980802e3 TM |
204 | } |
205 | ||
1da177e4 LT |
206 | /* |
207 | * Flush all dirty pages, and check for write errors. | |
1da177e4 LT |
208 | */ |
209 | static int | |
75e1fcc0 | 210 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 211 | { |
6da24bc9 CL |
212 | struct dentry *dentry = file->f_path.dentry; |
213 | struct inode *inode = dentry->d_inode; | |
1da177e4 | 214 | |
6da24bc9 CL |
215 | dprintk("NFS: flush(%s/%s)\n", |
216 | dentry->d_parent->d_name.name, | |
217 | dentry->d_name.name); | |
1da177e4 | 218 | |
c2459dc4 | 219 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
220 | if ((file->f_mode & FMODE_WRITE) == 0) |
221 | return 0; | |
7b159fc1 | 222 | |
7fe5c398 | 223 | /* Flush writes to the server and return any errors */ |
af7fa165 | 224 | return vfs_fsync(file, 0); |
1da177e4 LT |
225 | } |
226 | ||
227 | static ssize_t | |
027445c3 BP |
228 | nfs_file_read(struct kiocb *iocb, const struct iovec *iov, |
229 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 230 | { |
01cce933 | 231 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 LT |
232 | struct inode * inode = dentry->d_inode; |
233 | ssize_t result; | |
027445c3 | 234 | size_t count = iov_length(iov, nr_segs); |
1da177e4 | 235 | |
1da177e4 | 236 | if (iocb->ki_filp->f_flags & O_DIRECT) |
027445c3 | 237 | return nfs_file_direct_read(iocb, iov, nr_segs, pos); |
1da177e4 | 238 | |
6da24bc9 | 239 | dprintk("NFS: read(%s/%s, %lu@%lu)\n", |
1da177e4 LT |
240 | dentry->d_parent->d_name.name, dentry->d_name.name, |
241 | (unsigned long) count, (unsigned long) pos); | |
242 | ||
44b11874 | 243 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); |
4184dcf2 | 244 | if (!result) { |
027445c3 | 245 | result = generic_file_aio_read(iocb, iov, nr_segs, pos); |
4184dcf2 CL |
246 | if (result > 0) |
247 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
248 | } | |
1da177e4 LT |
249 | return result; |
250 | } | |
251 | ||
252 | static ssize_t | |
f0930fff JA |
253 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
254 | struct pipe_inode_info *pipe, size_t count, | |
255 | unsigned int flags) | |
1da177e4 | 256 | { |
01cce933 | 257 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
258 | struct inode *inode = dentry->d_inode; |
259 | ssize_t res; | |
260 | ||
6da24bc9 | 261 | dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n", |
1da177e4 LT |
262 | dentry->d_parent->d_name.name, dentry->d_name.name, |
263 | (unsigned long) count, (unsigned long long) *ppos); | |
264 | ||
44b11874 | 265 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
aa2f1ef1 | 266 | if (!res) { |
f0930fff | 267 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
aa2f1ef1 CL |
268 | if (res > 0) |
269 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); | |
270 | } | |
1da177e4 LT |
271 | return res; |
272 | } | |
273 | ||
274 | static int | |
275 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) | |
276 | { | |
01cce933 | 277 | struct dentry *dentry = file->f_path.dentry; |
1da177e4 LT |
278 | struct inode *inode = dentry->d_inode; |
279 | int status; | |
280 | ||
6da24bc9 | 281 | dprintk("NFS: mmap(%s/%s)\n", |
1da177e4 LT |
282 | dentry->d_parent->d_name.name, dentry->d_name.name); |
283 | ||
e1ebfd33 TM |
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); | |
94387fb1 TM |
288 | if (!status) { |
289 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 290 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 291 | } |
1da177e4 LT |
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. | |
af7fa165 TM |
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. | |
1da177e4 LT |
306 | */ |
307 | static int | |
7ea80859 | 308 | nfs_file_fsync(struct file *file, int datasync) |
1da177e4 | 309 | { |
7ea80859 | 310 | struct dentry *dentry = file->f_path.dentry; |
cd3758e3 | 311 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 312 | struct inode *inode = dentry->d_inode; |
af7fa165 TM |
313 | int have_error, status; |
314 | int ret = 0; | |
315 | ||
1da177e4 | 316 | |
6da24bc9 | 317 | dprintk("NFS: fsync file(%s/%s) datasync %d\n", |
54917786 CL |
318 | dentry->d_parent->d_name.name, dentry->d_name.name, |
319 | datasync); | |
1da177e4 | 320 | |
91d5b470 | 321 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
af7fa165 TM |
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); | |
0702099b | 327 | if (!ret && status < 0) |
af7fa165 TM |
328 | ret = status; |
329 | return ret; | |
1da177e4 LT |
330 | } |
331 | ||
38c73044 PS |
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 | ||
1da177e4 | 368 | /* |
4899f9c8 NP |
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. | |
1da177e4 LT |
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 | */ | |
4899f9c8 NP |
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) | |
1da177e4 | 379 | { |
4899f9c8 | 380 | int ret; |
38c73044 | 381 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
4899f9c8 | 382 | struct page *page; |
38c73044 | 383 | int once_thru = 0; |
4899f9c8 | 384 | |
b7eaefaa CL |
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 | ||
38c73044 | 390 | start: |
72cb77f4 TM |
391 | /* |
392 | * Prevent starvation issues if someone is doing a consistency | |
393 | * sync-to-disk | |
394 | */ | |
395 | ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, | |
396 | nfs_wait_bit_killable, TASK_KILLABLE); | |
397 | if (ret) | |
398 | return ret; | |
399 | ||
54566b2c | 400 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
401 | if (!page) |
402 | return -ENOMEM; | |
403 | *pagep = page; | |
404 | ||
405 | ret = nfs_flush_incompatible(file, page); | |
406 | if (ret) { | |
407 | unlock_page(page); | |
408 | page_cache_release(page); | |
38c73044 PS |
409 | } else if (!once_thru && |
410 | nfs_want_read_modify_write(file, page, pos, len)) { | |
411 | once_thru = 1; | |
412 | ret = nfs_readpage(file, page); | |
413 | page_cache_release(page); | |
414 | if (!ret) | |
415 | goto start; | |
4899f9c8 NP |
416 | } |
417 | return ret; | |
1da177e4 LT |
418 | } |
419 | ||
4899f9c8 NP |
420 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
421 | loff_t pos, unsigned len, unsigned copied, | |
422 | struct page *page, void *fsdata) | |
1da177e4 | 423 | { |
4899f9c8 NP |
424 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
425 | int status; | |
1da177e4 | 426 | |
b7eaefaa CL |
427 | dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n", |
428 | file->f_path.dentry->d_parent->d_name.name, | |
429 | file->f_path.dentry->d_name.name, | |
430 | mapping->host->i_ino, len, (long long) pos); | |
431 | ||
efc91ed0 TM |
432 | /* |
433 | * Zero any uninitialised parts of the page, and then mark the page | |
434 | * as up to date if it turns out that we're extending the file. | |
435 | */ | |
436 | if (!PageUptodate(page)) { | |
437 | unsigned pglen = nfs_page_length(page); | |
438 | unsigned end = offset + len; | |
439 | ||
440 | if (pglen == 0) { | |
441 | zero_user_segments(page, 0, offset, | |
442 | end, PAGE_CACHE_SIZE); | |
443 | SetPageUptodate(page); | |
444 | } else if (end >= pglen) { | |
445 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
446 | if (offset == 0) | |
447 | SetPageUptodate(page); | |
448 | } else | |
449 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
450 | } | |
451 | ||
4899f9c8 | 452 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
453 | |
454 | unlock_page(page); | |
455 | page_cache_release(page); | |
456 | ||
3d509e54 CL |
457 | if (status < 0) |
458 | return status; | |
459 | return copied; | |
1da177e4 LT |
460 | } |
461 | ||
6b9b3514 DH |
462 | /* |
463 | * Partially or wholly invalidate a page | |
464 | * - Release the private state associated with a page if undergoing complete | |
465 | * page invalidation | |
545db45f | 466 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
467 | * - Caller holds page lock |
468 | */ | |
2ff28e22 | 469 | static void nfs_invalidate_page(struct page *page, unsigned long offset) |
cd52ed35 | 470 | { |
b7eaefaa CL |
471 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset); |
472 | ||
1c75950b TM |
473 | if (offset != 0) |
474 | return; | |
d2ccddf0 | 475 | /* Cancel any unstarted writes on this page */ |
1b3b4a1a | 476 | nfs_wb_page_cancel(page->mapping->host, page); |
545db45f DH |
477 | |
478 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
479 | } |
480 | ||
6b9b3514 DH |
481 | /* |
482 | * Attempt to release the private state associated with a page | |
545db45f | 483 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
484 | * - Caller holds page lock |
485 | * - Return true (may release page) or false (may not) | |
486 | */ | |
cd52ed35 TM |
487 | static int nfs_release_page(struct page *page, gfp_t gfp) |
488 | { | |
b608b283 TM |
489 | struct address_space *mapping = page->mapping; |
490 | ||
b7eaefaa CL |
491 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
492 | ||
d812e575 | 493 | /* Only do I/O if gfp is a superset of GFP_KERNEL */ |
b608b283 TM |
494 | if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL) { |
495 | int how = FLUSH_SYNC; | |
496 | ||
497 | /* Don't let kswapd deadlock waiting for OOM RPC calls */ | |
498 | if (current_is_kswapd()) | |
499 | how = 0; | |
500 | nfs_commit_inode(mapping->host, how); | |
501 | } | |
e3db7691 | 502 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
503 | if (PagePrivate(page)) |
504 | return 0; | |
505 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
506 | } |
507 | ||
6b9b3514 DH |
508 | /* |
509 | * Attempt to clear the private state associated with a page when an error | |
510 | * occurs that requires the cached contents of an inode to be written back or | |
511 | * destroyed | |
545db45f | 512 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
513 | * - Caller holds page lock |
514 | * - Return 0 if successful, -error otherwise | |
515 | */ | |
e3db7691 TM |
516 | static int nfs_launder_page(struct page *page) |
517 | { | |
b7eaefaa | 518 | struct inode *inode = page->mapping->host; |
545db45f | 519 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
520 | |
521 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
522 | inode->i_ino, (long long)page_offset(page)); | |
523 | ||
545db45f | 524 | nfs_fscache_wait_on_page_write(nfsi, page); |
b7eaefaa | 525 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
526 | } |
527 | ||
f5e54d6e | 528 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
529 | .readpage = nfs_readpage, |
530 | .readpages = nfs_readpages, | |
9cccef95 | 531 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
532 | .writepage = nfs_writepage, |
533 | .writepages = nfs_writepages, | |
4899f9c8 NP |
534 | .write_begin = nfs_write_begin, |
535 | .write_end = nfs_write_end, | |
cd52ed35 TM |
536 | .invalidatepage = nfs_invalidate_page, |
537 | .releasepage = nfs_release_page, | |
1da177e4 | 538 | .direct_IO = nfs_direct_IO, |
074cc1de | 539 | .migratepage = nfs_migrate_page, |
e3db7691 | 540 | .launder_page = nfs_launder_page, |
f590f333 | 541 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
542 | }; |
543 | ||
6b9b3514 DH |
544 | /* |
545 | * Notification that a PTE pointing to an NFS page is about to be made | |
546 | * writable, implying that someone is about to modify the page through a | |
547 | * shared-writable mapping | |
548 | */ | |
c2ec175c | 549 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 550 | { |
c2ec175c | 551 | struct page *page = vmf->page; |
94387fb1 | 552 | struct file *filp = vma->vm_file; |
b7eaefaa | 553 | struct dentry *dentry = filp->f_path.dentry; |
94387fb1 | 554 | unsigned pagelen; |
bc4866b6 | 555 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 556 | struct address_space *mapping; |
94387fb1 | 557 | |
b7eaefaa CL |
558 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n", |
559 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
560 | filp->f_mapping->host->i_ino, | |
561 | (long long)page_offset(page)); | |
562 | ||
545db45f DH |
563 | /* make sure the cache has finished storing the page */ |
564 | nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page); | |
565 | ||
94387fb1 | 566 | lock_page(page); |
4899f9c8 | 567 | mapping = page->mapping; |
b7eaefaa | 568 | if (mapping != dentry->d_inode->i_mapping) |
8b1f9ee5 TM |
569 | goto out_unlock; |
570 | ||
94387fb1 | 571 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
572 | if (pagelen == 0) |
573 | goto out_unlock; | |
4899f9c8 | 574 | |
bc4866b6 TM |
575 | ret = VM_FAULT_LOCKED; |
576 | if (nfs_flush_incompatible(filp, page) == 0 && | |
577 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
578 | goto out; | |
8b1f9ee5 | 579 | |
bc4866b6 | 580 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
581 | out_unlock: |
582 | unlock_page(page); | |
bc4866b6 TM |
583 | out: |
584 | return ret; | |
94387fb1 TM |
585 | } |
586 | ||
f0f37e2f | 587 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 TM |
588 | .fault = filemap_fault, |
589 | .page_mkwrite = nfs_vm_page_mkwrite, | |
590 | }; | |
591 | ||
7b159fc1 TM |
592 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
593 | { | |
594 | struct nfs_open_context *ctx; | |
595 | ||
6b2f3d1f | 596 | if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) |
7b159fc1 | 597 | return 1; |
cd3758e3 | 598 | ctx = nfs_file_open_context(filp); |
7b159fc1 TM |
599 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) |
600 | return 1; | |
601 | return 0; | |
602 | } | |
603 | ||
027445c3 BP |
604 | static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, |
605 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 606 | { |
01cce933 | 607 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 | 608 | struct inode * inode = dentry->d_inode; |
7e381172 | 609 | unsigned long written = 0; |
1da177e4 | 610 | ssize_t result; |
027445c3 | 611 | size_t count = iov_length(iov, nr_segs); |
1da177e4 | 612 | |
1da177e4 | 613 | if (iocb->ki_filp->f_flags & O_DIRECT) |
027445c3 | 614 | return nfs_file_direct_write(iocb, iov, nr_segs, pos); |
1da177e4 | 615 | |
6da24bc9 | 616 | dprintk("NFS: write(%s/%s, %lu@%Ld)\n", |
1da177e4 | 617 | dentry->d_parent->d_name.name, dentry->d_name.name, |
6da24bc9 | 618 | (unsigned long) count, (long long) pos); |
1da177e4 LT |
619 | |
620 | result = -EBUSY; | |
621 | if (IS_SWAPFILE(inode)) | |
622 | goto out_swapfile; | |
7d52e862 TM |
623 | /* |
624 | * O_APPEND implies that we must revalidate the file length. | |
625 | */ | |
626 | if (iocb->ki_filp->f_flags & O_APPEND) { | |
627 | result = nfs_revalidate_file_size(inode, iocb->ki_filp); | |
628 | if (result) | |
629 | goto out; | |
fe51beec | 630 | } |
1da177e4 LT |
631 | |
632 | result = count; | |
633 | if (!count) | |
634 | goto out; | |
635 | ||
027445c3 | 636 | result = generic_file_aio_write(iocb, iov, nr_segs, pos); |
7e381172 CL |
637 | if (result > 0) |
638 | written = result; | |
639 | ||
6b2f3d1f | 640 | /* Return error values for O_DSYNC and IS_SYNC() */ |
7b159fc1 | 641 | if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { |
af7fa165 | 642 | int err = vfs_fsync(iocb->ki_filp, 0); |
200baa21 TM |
643 | if (err < 0) |
644 | result = err; | |
645 | } | |
7e381172 CL |
646 | if (result > 0) |
647 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
1da177e4 LT |
648 | out: |
649 | return result; | |
650 | ||
651 | out_swapfile: | |
652 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
653 | goto out; | |
654 | } | |
655 | ||
bf40d343 SJ |
656 | static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, |
657 | struct file *filp, loff_t *ppos, | |
658 | size_t count, unsigned int flags) | |
659 | { | |
660 | struct dentry *dentry = filp->f_path.dentry; | |
661 | struct inode *inode = dentry->d_inode; | |
7e381172 | 662 | unsigned long written = 0; |
bf40d343 SJ |
663 | ssize_t ret; |
664 | ||
665 | dprintk("NFS splice_write(%s/%s, %lu@%llu)\n", | |
666 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
667 | (unsigned long) count, (unsigned long long) *ppos); | |
668 | ||
669 | /* | |
670 | * The combination of splice and an O_APPEND destination is disallowed. | |
671 | */ | |
672 | ||
bf40d343 | 673 | ret = generic_file_splice_write(pipe, filp, ppos, count, flags); |
7e381172 CL |
674 | if (ret > 0) |
675 | written = ret; | |
676 | ||
bf40d343 | 677 | if (ret >= 0 && nfs_need_sync_write(filp, inode)) { |
af7fa165 | 678 | int err = vfs_fsync(filp, 0); |
bf40d343 SJ |
679 | if (err < 0) |
680 | ret = err; | |
681 | } | |
7e381172 CL |
682 | if (ret > 0) |
683 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
bf40d343 SJ |
684 | return ret; |
685 | } | |
686 | ||
5eebde23 SJ |
687 | static int |
688 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
689 | { |
690 | struct inode *inode = filp->f_mapping->host; | |
691 | int status = 0; | |
21ac19d4 | 692 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 693 | |
039c4d7a | 694 | /* Try local locking first */ |
6d34ac19 BF |
695 | posix_test_lock(filp, fl); |
696 | if (fl->fl_type != F_UNLCK) { | |
697 | /* found a conflict */ | |
039c4d7a | 698 | goto out; |
1da177e4 | 699 | } |
21ac19d4 | 700 | fl->fl_type = saved_type; |
039c4d7a TM |
701 | |
702 | if (nfs_have_delegation(inode, FMODE_READ)) | |
703 | goto out_noconflict; | |
704 | ||
5eebde23 | 705 | if (is_local) |
039c4d7a TM |
706 | goto out_noconflict; |
707 | ||
708 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
709 | out: | |
1da177e4 | 710 | return status; |
039c4d7a TM |
711 | out_noconflict: |
712 | fl->fl_type = F_UNLCK; | |
713 | goto out; | |
1da177e4 LT |
714 | } |
715 | ||
716 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
717 | { | |
718 | int res = 0; | |
719 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
720 | case FL_POSIX: | |
721 | res = posix_lock_file_wait(file, fl); | |
722 | break; | |
723 | case FL_FLOCK: | |
724 | res = flock_lock_file_wait(file, fl); | |
725 | break; | |
726 | default: | |
727 | BUG(); | |
728 | } | |
1da177e4 LT |
729 | return res; |
730 | } | |
731 | ||
5eebde23 SJ |
732 | static int |
733 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
734 | { |
735 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
736 | int status; |
737 | ||
1da177e4 LT |
738 | /* |
739 | * Flush all pending writes before doing anything | |
740 | * with locks.. | |
741 | */ | |
29884df0 | 742 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 LT |
743 | |
744 | /* NOTE: special case | |
745 | * If we're signalled while cleaning up locks on process exit, we | |
746 | * still need to complete the unlock. | |
747 | */ | |
5eebde23 SJ |
748 | /* |
749 | * Use local locking if mounted with "-onolock" or with appropriate | |
750 | * "-olocal_lock=" | |
751 | */ | |
752 | if (!is_local) | |
1da177e4 LT |
753 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
754 | else | |
755 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
756 | return status; |
757 | } | |
758 | ||
6b96724e RL |
759 | static int |
760 | is_time_granular(struct timespec *ts) { | |
761 | return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000)); | |
762 | } | |
763 | ||
5eebde23 SJ |
764 | static int |
765 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
766 | { |
767 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
768 | int status; |
769 | ||
1da177e4 LT |
770 | /* |
771 | * Flush all pending writes before doing anything | |
772 | * with locks.. | |
773 | */ | |
29884df0 TM |
774 | status = nfs_sync_mapping(filp->f_mapping); |
775 | if (status != 0) | |
1da177e4 LT |
776 | goto out; |
777 | ||
5eebde23 SJ |
778 | /* |
779 | * Use local locking if mounted with "-onolock" or with appropriate | |
780 | * "-olocal_lock=" | |
781 | */ | |
782 | if (!is_local) | |
1da177e4 | 783 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 784 | else |
1da177e4 | 785 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
786 | if (status < 0) |
787 | goto out; | |
6b96724e | 788 | |
1da177e4 | 789 | /* |
6b96724e RL |
790 | * Revalidate the cache if the server has time stamps granular |
791 | * enough to detect subsecond changes. Otherwise, clear the | |
792 | * cache to prevent missing any changes. | |
793 | * | |
1da177e4 LT |
794 | * This makes locking act as a cache coherency point. |
795 | */ | |
29884df0 | 796 | nfs_sync_mapping(filp->f_mapping); |
6b96724e RL |
797 | if (!nfs_have_delegation(inode, FMODE_READ)) { |
798 | if (is_time_granular(&NFS_SERVER(inode)->time_delta)) | |
799 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
800 | else | |
801 | nfs_zap_caches(inode); | |
802 | } | |
1da177e4 | 803 | out: |
1da177e4 LT |
804 | return status; |
805 | } | |
806 | ||
807 | /* | |
808 | * Lock a (portion of) a file | |
809 | */ | |
810 | static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) | |
811 | { | |
6da24bc9 | 812 | struct inode *inode = filp->f_mapping->host; |
2116271a | 813 | int ret = -ENOLCK; |
5eebde23 | 814 | int is_local = 0; |
1da177e4 | 815 | |
6da24bc9 CL |
816 | dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n", |
817 | filp->f_path.dentry->d_parent->d_name.name, | |
818 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
819 | fl->fl_type, fl->fl_flags, |
820 | (long long)fl->fl_start, (long long)fl->fl_end); | |
6da24bc9 | 821 | |
91d5b470 | 822 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
823 | |
824 | /* No mandatory locks over NFS */ | |
dfad9441 | 825 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
826 | goto out_err; |
827 | ||
5eebde23 SJ |
828 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
829 | is_local = 1; | |
830 | ||
2116271a TM |
831 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
832 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
833 | if (ret < 0) | |
834 | goto out_err; | |
835 | } | |
1da177e4 LT |
836 | |
837 | if (IS_GETLK(cmd)) | |
5eebde23 | 838 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 839 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 840 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 841 | else |
5eebde23 | 842 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
843 | out_err: |
844 | return ret; | |
1da177e4 LT |
845 | } |
846 | ||
847 | /* | |
848 | * Lock a (portion of) a file | |
849 | */ | |
850 | static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) | |
851 | { | |
5eebde23 SJ |
852 | struct inode *inode = filp->f_mapping->host; |
853 | int is_local = 0; | |
854 | ||
6da24bc9 CL |
855 | dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n", |
856 | filp->f_path.dentry->d_parent->d_name.name, | |
857 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
858 | fl->fl_type, fl->fl_flags); |
859 | ||
1da177e4 LT |
860 | if (!(fl->fl_flags & FL_FLOCK)) |
861 | return -ENOLCK; | |
862 | ||
5eebde23 SJ |
863 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
864 | is_local = 1; | |
865 | ||
1da177e4 LT |
866 | /* We're simulating flock() locks using posix locks on the server */ |
867 | fl->fl_owner = (fl_owner_t)filp; | |
868 | fl->fl_start = 0; | |
869 | fl->fl_end = OFFSET_MAX; | |
870 | ||
871 | if (fl->fl_type == F_UNLCK) | |
5eebde23 SJ |
872 | return do_unlk(filp, cmd, fl, is_local); |
873 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 874 | } |
370f6599 | 875 | |
6da24bc9 CL |
876 | /* |
877 | * There is no protocol support for leases, so we have no way to implement | |
878 | * them correctly in the face of opens by other clients. | |
879 | */ | |
370f6599 BF |
880 | static int nfs_setlease(struct file *file, long arg, struct file_lock **fl) |
881 | { | |
6da24bc9 CL |
882 | dprintk("NFS: setlease(%s/%s, arg=%ld)\n", |
883 | file->f_path.dentry->d_parent->d_name.name, | |
884 | file->f_path.dentry->d_name.name, arg); | |
370f6599 BF |
885 | return -EINVAL; |
886 | } |