]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/file.c | |
3 | * | |
4 | * Copyright (C) 1992 Rick Sladkey | |
5 | * | |
6 | * Changes Copyright (C) 1994 by Florian La Roche | |
7 | * - Do not copy data too often around in the kernel. | |
8 | * - In nfs_file_read the return value of kmalloc wasn't checked. | |
9 | * - Put in a better version of read look-ahead buffering. Original idea | |
10 | * and implementation by Wai S Kok elekokws@ee.nus.sg. | |
11 | * | |
12 | * Expire cache on write to a file by Wai S Kok (Oct 1994). | |
13 | * | |
14 | * Total rewrite of read side for new NFS buffer cache.. Linus. | |
15 | * | |
16 | * nfs regular file handling functions | |
17 | */ | |
18 | ||
19 | #include <linux/time.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/fcntl.h> | |
23 | #include <linux/stat.h> | |
24 | #include <linux/nfs_fs.h> | |
25 | #include <linux/nfs_mount.h> | |
26 | #include <linux/mm.h> | |
1da177e4 | 27 | #include <linux/pagemap.h> |
e8edc6e0 | 28 | #include <linux/aio.h> |
5a0e3ad6 | 29 | #include <linux/gfp.h> |
b608b283 | 30 | #include <linux/swap.h> |
1da177e4 LT |
31 | |
32 | #include <asm/uaccess.h> | |
1da177e4 LT |
33 | |
34 | #include "delegation.h" | |
94387fb1 | 35 | #include "internal.h" |
91d5b470 | 36 | #include "iostat.h" |
545db45f | 37 | #include "fscache.h" |
1da177e4 LT |
38 | |
39 | #define NFSDBG_FACILITY NFSDBG_FILE | |
40 | ||
f0f37e2f | 41 | static const struct vm_operations_struct nfs_file_vm_ops; |
94387fb1 | 42 | |
1da177e4 LT |
43 | /* Hack for future NFS swap support */ |
44 | #ifndef IS_SWAPFILE | |
45 | # define IS_SWAPFILE(inode) (0) | |
46 | #endif | |
47 | ||
ce4ef7c0 | 48 | int nfs_check_flags(int flags) |
1da177e4 LT |
49 | { |
50 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
51 | return -EINVAL; | |
52 | ||
53 | return 0; | |
54 | } | |
55 | ||
56 | /* | |
57 | * Open file | |
58 | */ | |
59 | static int | |
60 | nfs_file_open(struct inode *inode, struct file *filp) | |
61 | { | |
1da177e4 LT |
62 | int res; |
63 | ||
6da24bc9 | 64 | dprintk("NFS: open file(%s/%s)\n", |
cc0dd2d1 CL |
65 | filp->f_path.dentry->d_parent->d_name.name, |
66 | filp->f_path.dentry->d_name.name); | |
67 | ||
c2459dc4 | 68 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
1da177e4 LT |
69 | res = nfs_check_flags(filp->f_flags); |
70 | if (res) | |
71 | return res; | |
72 | ||
46cb650c | 73 | res = nfs_open(inode, filp); |
1da177e4 LT |
74 | return res; |
75 | } | |
76 | ||
ce4ef7c0 | 77 | int |
1da177e4 LT |
78 | nfs_file_release(struct inode *inode, struct file *filp) |
79 | { | |
6da24bc9 | 80 | dprintk("NFS: release(%s/%s)\n", |
6f276e49 RM |
81 | filp->f_path.dentry->d_parent->d_name.name, |
82 | filp->f_path.dentry->d_name.name); | |
6da24bc9 | 83 | |
91d5b470 | 84 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
46cb650c | 85 | return nfs_release(inode, filp); |
1da177e4 LT |
86 | } |
87 | ||
980802e3 TM |
88 | /** |
89 | * nfs_revalidate_size - Revalidate the file size | |
90 | * @inode - pointer to inode struct | |
91 | * @file - pointer to struct file | |
92 | * | |
93 | * Revalidates the file length. This is basically a wrapper around | |
94 | * nfs_revalidate_inode() that takes into account the fact that we may | |
95 | * have cached writes (in which case we don't care about the server's | |
96 | * idea of what the file length is), or O_DIRECT (in which case we | |
97 | * shouldn't trust the cache). | |
98 | */ | |
99 | static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) | |
100 | { | |
101 | struct nfs_server *server = NFS_SERVER(inode); | |
102 | struct nfs_inode *nfsi = NFS_I(inode); | |
103 | ||
d7cf8dd0 TM |
104 | if (nfs_have_delegated_attributes(inode)) |
105 | goto out_noreval; | |
106 | ||
980802e3 TM |
107 | if (filp->f_flags & O_DIRECT) |
108 | goto force_reval; | |
d7cf8dd0 TM |
109 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
110 | goto force_reval; | |
111 | if (nfs_attribute_timeout(inode)) | |
112 | goto force_reval; | |
113 | out_noreval: | |
114 | return 0; | |
980802e3 TM |
115 | force_reval: |
116 | return __nfs_revalidate_inode(server, inode); | |
117 | } | |
118 | ||
ce4ef7c0 | 119 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin) |
980802e3 | 120 | { |
6da24bc9 | 121 | dprintk("NFS: llseek file(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
122 | filp->f_path.dentry->d_parent->d_name.name, |
123 | filp->f_path.dentry->d_name.name, | |
124 | offset, origin); | |
125 | ||
06222e49 JB |
126 | /* |
127 | * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate | |
128 | * the cached file length | |
129 | */ | |
6c529617 | 130 | if (origin != SEEK_SET && origin != SEEK_CUR) { |
980802e3 | 131 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 132 | |
980802e3 TM |
133 | int retval = nfs_revalidate_file_size(inode, filp); |
134 | if (retval < 0) | |
135 | return (loff_t)retval; | |
79835a71 | 136 | } |
d5e66348 | 137 | |
79835a71 | 138 | return generic_file_llseek(filp, offset, origin); |
980802e3 TM |
139 | } |
140 | ||
1da177e4 LT |
141 | /* |
142 | * Flush all dirty pages, and check for write errors. | |
1da177e4 | 143 | */ |
ce4ef7c0 | 144 | int |
75e1fcc0 | 145 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 146 | { |
6da24bc9 CL |
147 | struct dentry *dentry = file->f_path.dentry; |
148 | struct inode *inode = dentry->d_inode; | |
1da177e4 | 149 | |
6da24bc9 CL |
150 | dprintk("NFS: flush(%s/%s)\n", |
151 | dentry->d_parent->d_name.name, | |
152 | dentry->d_name.name); | |
1da177e4 | 153 | |
c2459dc4 | 154 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
155 | if ((file->f_mode & FMODE_WRITE) == 0) |
156 | return 0; | |
7b159fc1 | 157 | |
14546c33 TM |
158 | /* |
159 | * If we're holding a write delegation, then just start the i/o | |
160 | * but don't wait for completion (or send a commit). | |
161 | */ | |
011e2a7f | 162 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
14546c33 TM |
163 | return filemap_fdatawrite(file->f_mapping); |
164 | ||
7fe5c398 | 165 | /* Flush writes to the server and return any errors */ |
af7fa165 | 166 | return vfs_fsync(file, 0); |
1da177e4 LT |
167 | } |
168 | ||
ce4ef7c0 | 169 | ssize_t |
027445c3 BP |
170 | nfs_file_read(struct kiocb *iocb, const struct iovec *iov, |
171 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 172 | { |
01cce933 | 173 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 LT |
174 | struct inode * inode = dentry->d_inode; |
175 | ssize_t result; | |
176 | ||
1da177e4 | 177 | if (iocb->ki_filp->f_flags & O_DIRECT) |
027445c3 | 178 | return nfs_file_direct_read(iocb, iov, nr_segs, pos); |
1da177e4 | 179 | |
6da24bc9 | 180 | dprintk("NFS: read(%s/%s, %lu@%lu)\n", |
1da177e4 | 181 | dentry->d_parent->d_name.name, dentry->d_name.name, |
6f276e49 | 182 | (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos); |
1da177e4 | 183 | |
44b11874 | 184 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); |
4184dcf2 | 185 | if (!result) { |
027445c3 | 186 | result = generic_file_aio_read(iocb, iov, nr_segs, pos); |
4184dcf2 CL |
187 | if (result > 0) |
188 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
189 | } | |
1da177e4 LT |
190 | return result; |
191 | } | |
192 | ||
ce4ef7c0 | 193 | ssize_t |
f0930fff JA |
194 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
195 | struct pipe_inode_info *pipe, size_t count, | |
196 | unsigned int flags) | |
1da177e4 | 197 | { |
01cce933 | 198 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
199 | struct inode *inode = dentry->d_inode; |
200 | ssize_t res; | |
201 | ||
6da24bc9 | 202 | dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n", |
1da177e4 LT |
203 | dentry->d_parent->d_name.name, dentry->d_name.name, |
204 | (unsigned long) count, (unsigned long long) *ppos); | |
205 | ||
44b11874 | 206 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
aa2f1ef1 | 207 | if (!res) { |
f0930fff | 208 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
aa2f1ef1 CL |
209 | if (res > 0) |
210 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); | |
211 | } | |
1da177e4 LT |
212 | return res; |
213 | } | |
214 | ||
ce4ef7c0 | 215 | int |
1da177e4 LT |
216 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
217 | { | |
01cce933 | 218 | struct dentry *dentry = file->f_path.dentry; |
1da177e4 LT |
219 | struct inode *inode = dentry->d_inode; |
220 | int status; | |
221 | ||
6da24bc9 | 222 | dprintk("NFS: mmap(%s/%s)\n", |
1da177e4 LT |
223 | dentry->d_parent->d_name.name, dentry->d_name.name); |
224 | ||
e1ebfd33 TM |
225 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
226 | * so we call that before revalidating the mapping | |
227 | */ | |
228 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
229 | if (!status) { |
230 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 231 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 232 | } |
1da177e4 LT |
233 | return status; |
234 | } | |
235 | ||
236 | /* | |
237 | * Flush any dirty pages for this process, and check for write errors. | |
238 | * The return status from this call provides a reliable indication of | |
239 | * whether any write errors occurred for this process. | |
af7fa165 TM |
240 | * |
241 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
242 | * disk, but it retrieves and clears ctx->error after synching, despite | |
243 | * the two being set at the same time in nfs_context_set_write_error(). | |
244 | * This is because the former is used to notify the _next_ call to | |
25985edc | 245 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 246 | * fall back to doing a synchronous write. |
1da177e4 | 247 | */ |
ce4ef7c0 | 248 | int |
a5c58892 | 249 | nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 250 | { |
7ea80859 | 251 | struct dentry *dentry = file->f_path.dentry; |
cd3758e3 | 252 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 253 | struct inode *inode = dentry->d_inode; |
af7fa165 TM |
254 | int have_error, status; |
255 | int ret = 0; | |
256 | ||
6da24bc9 | 257 | dprintk("NFS: fsync file(%s/%s) datasync %d\n", |
54917786 CL |
258 | dentry->d_parent->d_name.name, dentry->d_name.name, |
259 | datasync); | |
1da177e4 | 260 | |
91d5b470 | 261 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
af7fa165 TM |
262 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
263 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
2edb6bc3 N |
264 | if (status >= 0 && ret < 0) |
265 | status = ret; | |
af7fa165 TM |
266 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
267 | if (have_error) | |
268 | ret = xchg(&ctx->error, 0); | |
0702099b | 269 | if (!ret && status < 0) |
af7fa165 | 270 | ret = status; |
a5c58892 BS |
271 | return ret; |
272 | } | |
273 | ||
274 | static int | |
275 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) | |
276 | { | |
277 | int ret; | |
278 | struct inode *inode = file->f_path.dentry->d_inode; | |
279 | ||
280 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
281 | mutex_lock(&inode->i_mutex); | |
282 | ret = nfs_file_fsync_commit(file, start, end, datasync); | |
02c24a82 | 283 | mutex_unlock(&inode->i_mutex); |
a5c58892 | 284 | |
af7fa165 | 285 | return ret; |
1da177e4 LT |
286 | } |
287 | ||
38c73044 PS |
288 | /* |
289 | * Decide whether a read/modify/write cycle may be more efficient | |
290 | * then a modify/write/read cycle when writing to a page in the | |
291 | * page cache. | |
292 | * | |
293 | * The modify/write/read cycle may occur if a page is read before | |
294 | * being completely filled by the writer. In this situation, the | |
295 | * page must be completely written to stable storage on the server | |
296 | * before it can be refilled by reading in the page from the server. | |
297 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
298 | * done. | |
299 | * | |
300 | * It may be more efficient to read the page first if the file is | |
301 | * open for reading in addition to writing, the page is not marked | |
302 | * as Uptodate, it is not dirty or waiting to be committed, | |
303 | * indicating that it was previously allocated and then modified, | |
304 | * that there were valid bytes of data in that range of the file, | |
305 | * and that the new data won't completely replace the old data in | |
306 | * that range of the file. | |
307 | */ | |
308 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
309 | loff_t pos, unsigned len) | |
310 | { | |
311 | unsigned int pglen = nfs_page_length(page); | |
312 | unsigned int offset = pos & (PAGE_CACHE_SIZE - 1); | |
313 | unsigned int end = offset + len; | |
314 | ||
315 | if ((file->f_mode & FMODE_READ) && /* open for read? */ | |
316 | !PageUptodate(page) && /* Uptodate? */ | |
317 | !PagePrivate(page) && /* i/o request already? */ | |
318 | pglen && /* valid bytes of file? */ | |
319 | (end < pglen || offset)) /* replace all valid bytes? */ | |
320 | return 1; | |
321 | return 0; | |
322 | } | |
323 | ||
1da177e4 | 324 | /* |
4899f9c8 NP |
325 | * This does the "real" work of the write. We must allocate and lock the |
326 | * page to be sent back to the generic routine, which then copies the | |
327 | * data from user space. | |
1da177e4 LT |
328 | * |
329 | * If the writer ends up delaying the write, the writer needs to | |
330 | * increment the page use counts until he is done with the page. | |
331 | */ | |
4899f9c8 NP |
332 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
333 | loff_t pos, unsigned len, unsigned flags, | |
334 | struct page **pagep, void **fsdata) | |
1da177e4 | 335 | { |
4899f9c8 | 336 | int ret; |
38c73044 | 337 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
4899f9c8 | 338 | struct page *page; |
38c73044 | 339 | int once_thru = 0; |
4899f9c8 | 340 | |
b7eaefaa CL |
341 | dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n", |
342 | file->f_path.dentry->d_parent->d_name.name, | |
343 | file->f_path.dentry->d_name.name, | |
344 | mapping->host->i_ino, len, (long long) pos); | |
345 | ||
38c73044 | 346 | start: |
72cb77f4 TM |
347 | /* |
348 | * Prevent starvation issues if someone is doing a consistency | |
349 | * sync-to-disk | |
350 | */ | |
351 | ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, | |
352 | nfs_wait_bit_killable, TASK_KILLABLE); | |
353 | if (ret) | |
354 | return ret; | |
355 | ||
54566b2c | 356 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
357 | if (!page) |
358 | return -ENOMEM; | |
359 | *pagep = page; | |
360 | ||
361 | ret = nfs_flush_incompatible(file, page); | |
362 | if (ret) { | |
363 | unlock_page(page); | |
364 | page_cache_release(page); | |
38c73044 PS |
365 | } else if (!once_thru && |
366 | nfs_want_read_modify_write(file, page, pos, len)) { | |
367 | once_thru = 1; | |
368 | ret = nfs_readpage(file, page); | |
369 | page_cache_release(page); | |
370 | if (!ret) | |
371 | goto start; | |
4899f9c8 NP |
372 | } |
373 | return ret; | |
1da177e4 LT |
374 | } |
375 | ||
4899f9c8 NP |
376 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
377 | loff_t pos, unsigned len, unsigned copied, | |
378 | struct page *page, void *fsdata) | |
1da177e4 | 379 | { |
4899f9c8 NP |
380 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
381 | int status; | |
1da177e4 | 382 | |
b7eaefaa CL |
383 | dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n", |
384 | file->f_path.dentry->d_parent->d_name.name, | |
385 | file->f_path.dentry->d_name.name, | |
386 | mapping->host->i_ino, len, (long long) pos); | |
387 | ||
efc91ed0 TM |
388 | /* |
389 | * Zero any uninitialised parts of the page, and then mark the page | |
390 | * as up to date if it turns out that we're extending the file. | |
391 | */ | |
392 | if (!PageUptodate(page)) { | |
393 | unsigned pglen = nfs_page_length(page); | |
394 | unsigned end = offset + len; | |
395 | ||
396 | if (pglen == 0) { | |
397 | zero_user_segments(page, 0, offset, | |
398 | end, PAGE_CACHE_SIZE); | |
399 | SetPageUptodate(page); | |
400 | } else if (end >= pglen) { | |
401 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
402 | if (offset == 0) | |
403 | SetPageUptodate(page); | |
404 | } else | |
405 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
406 | } | |
407 | ||
4899f9c8 | 408 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
409 | |
410 | unlock_page(page); | |
411 | page_cache_release(page); | |
412 | ||
3d509e54 CL |
413 | if (status < 0) |
414 | return status; | |
2701d086 | 415 | NFS_I(mapping->host)->write_io += copied; |
3d509e54 | 416 | return copied; |
1da177e4 LT |
417 | } |
418 | ||
6b9b3514 DH |
419 | /* |
420 | * Partially or wholly invalidate a page | |
421 | * - Release the private state associated with a page if undergoing complete | |
422 | * page invalidation | |
545db45f | 423 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
424 | * - Caller holds page lock |
425 | */ | |
2ff28e22 | 426 | static void nfs_invalidate_page(struct page *page, unsigned long offset) |
cd52ed35 | 427 | { |
b7eaefaa CL |
428 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset); |
429 | ||
1c75950b TM |
430 | if (offset != 0) |
431 | return; | |
d2ccddf0 | 432 | /* Cancel any unstarted writes on this page */ |
1b3b4a1a | 433 | nfs_wb_page_cancel(page->mapping->host, page); |
545db45f DH |
434 | |
435 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
436 | } |
437 | ||
6b9b3514 DH |
438 | /* |
439 | * Attempt to release the private state associated with a page | |
545db45f | 440 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
441 | * - Caller holds page lock |
442 | * - Return true (may release page) or false (may not) | |
443 | */ | |
cd52ed35 TM |
444 | static int nfs_release_page(struct page *page, gfp_t gfp) |
445 | { | |
b608b283 TM |
446 | struct address_space *mapping = page->mapping; |
447 | ||
b7eaefaa CL |
448 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
449 | ||
d812e575 | 450 | /* Only do I/O if gfp is a superset of GFP_KERNEL */ |
b608b283 TM |
451 | if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL) { |
452 | int how = FLUSH_SYNC; | |
453 | ||
454 | /* Don't let kswapd deadlock waiting for OOM RPC calls */ | |
455 | if (current_is_kswapd()) | |
456 | how = 0; | |
457 | nfs_commit_inode(mapping->host, how); | |
458 | } | |
e3db7691 | 459 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
460 | if (PagePrivate(page)) |
461 | return 0; | |
462 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
463 | } |
464 | ||
6b9b3514 DH |
465 | /* |
466 | * Attempt to clear the private state associated with a page when an error | |
467 | * occurs that requires the cached contents of an inode to be written back or | |
468 | * destroyed | |
545db45f | 469 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
470 | * - Caller holds page lock |
471 | * - Return 0 if successful, -error otherwise | |
472 | */ | |
e3db7691 TM |
473 | static int nfs_launder_page(struct page *page) |
474 | { | |
b7eaefaa | 475 | struct inode *inode = page->mapping->host; |
545db45f | 476 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
477 | |
478 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
479 | inode->i_ino, (long long)page_offset(page)); | |
480 | ||
545db45f | 481 | nfs_fscache_wait_on_page_write(nfsi, page); |
b7eaefaa | 482 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
483 | } |
484 | ||
f5e54d6e | 485 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
486 | .readpage = nfs_readpage, |
487 | .readpages = nfs_readpages, | |
9cccef95 | 488 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
489 | .writepage = nfs_writepage, |
490 | .writepages = nfs_writepages, | |
4899f9c8 NP |
491 | .write_begin = nfs_write_begin, |
492 | .write_end = nfs_write_end, | |
cd52ed35 TM |
493 | .invalidatepage = nfs_invalidate_page, |
494 | .releasepage = nfs_release_page, | |
1da177e4 | 495 | .direct_IO = nfs_direct_IO, |
074cc1de | 496 | .migratepage = nfs_migrate_page, |
e3db7691 | 497 | .launder_page = nfs_launder_page, |
f590f333 | 498 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
499 | }; |
500 | ||
6b9b3514 DH |
501 | /* |
502 | * Notification that a PTE pointing to an NFS page is about to be made | |
503 | * writable, implying that someone is about to modify the page through a | |
504 | * shared-writable mapping | |
505 | */ | |
c2ec175c | 506 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 507 | { |
c2ec175c | 508 | struct page *page = vmf->page; |
94387fb1 | 509 | struct file *filp = vma->vm_file; |
b7eaefaa | 510 | struct dentry *dentry = filp->f_path.dentry; |
94387fb1 | 511 | unsigned pagelen; |
bc4866b6 | 512 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 513 | struct address_space *mapping; |
94387fb1 | 514 | |
b7eaefaa CL |
515 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n", |
516 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
517 | filp->f_mapping->host->i_ino, | |
518 | (long long)page_offset(page)); | |
519 | ||
545db45f DH |
520 | /* make sure the cache has finished storing the page */ |
521 | nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page); | |
522 | ||
94387fb1 | 523 | lock_page(page); |
4899f9c8 | 524 | mapping = page->mapping; |
b7eaefaa | 525 | if (mapping != dentry->d_inode->i_mapping) |
8b1f9ee5 TM |
526 | goto out_unlock; |
527 | ||
2aeb98f4 TM |
528 | wait_on_page_writeback(page); |
529 | ||
94387fb1 | 530 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
531 | if (pagelen == 0) |
532 | goto out_unlock; | |
4899f9c8 | 533 | |
bc4866b6 TM |
534 | ret = VM_FAULT_LOCKED; |
535 | if (nfs_flush_incompatible(filp, page) == 0 && | |
536 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
537 | goto out; | |
8b1f9ee5 | 538 | |
bc4866b6 | 539 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
540 | out_unlock: |
541 | unlock_page(page); | |
bc4866b6 TM |
542 | out: |
543 | return ret; | |
94387fb1 TM |
544 | } |
545 | ||
f0f37e2f | 546 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 TM |
547 | .fault = filemap_fault, |
548 | .page_mkwrite = nfs_vm_page_mkwrite, | |
549 | }; | |
550 | ||
7b159fc1 TM |
551 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
552 | { | |
553 | struct nfs_open_context *ctx; | |
554 | ||
6b2f3d1f | 555 | if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) |
7b159fc1 | 556 | return 1; |
cd3758e3 | 557 | ctx = nfs_file_open_context(filp); |
7b159fc1 TM |
558 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) |
559 | return 1; | |
560 | return 0; | |
561 | } | |
562 | ||
ce4ef7c0 BS |
563 | ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, |
564 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 565 | { |
01cce933 | 566 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 | 567 | struct inode * inode = dentry->d_inode; |
7e381172 | 568 | unsigned long written = 0; |
1da177e4 | 569 | ssize_t result; |
027445c3 | 570 | size_t count = iov_length(iov, nr_segs); |
1da177e4 | 571 | |
1da177e4 | 572 | if (iocb->ki_filp->f_flags & O_DIRECT) |
027445c3 | 573 | return nfs_file_direct_write(iocb, iov, nr_segs, pos); |
1da177e4 | 574 | |
6da24bc9 | 575 | dprintk("NFS: write(%s/%s, %lu@%Ld)\n", |
1da177e4 | 576 | dentry->d_parent->d_name.name, dentry->d_name.name, |
6da24bc9 | 577 | (unsigned long) count, (long long) pos); |
1da177e4 LT |
578 | |
579 | result = -EBUSY; | |
580 | if (IS_SWAPFILE(inode)) | |
581 | goto out_swapfile; | |
7d52e862 TM |
582 | /* |
583 | * O_APPEND implies that we must revalidate the file length. | |
584 | */ | |
585 | if (iocb->ki_filp->f_flags & O_APPEND) { | |
586 | result = nfs_revalidate_file_size(inode, iocb->ki_filp); | |
587 | if (result) | |
588 | goto out; | |
fe51beec | 589 | } |
1da177e4 LT |
590 | |
591 | result = count; | |
592 | if (!count) | |
593 | goto out; | |
594 | ||
027445c3 | 595 | result = generic_file_aio_write(iocb, iov, nr_segs, pos); |
7e381172 CL |
596 | if (result > 0) |
597 | written = result; | |
598 | ||
6b2f3d1f | 599 | /* Return error values for O_DSYNC and IS_SYNC() */ |
7b159fc1 | 600 | if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { |
af7fa165 | 601 | int err = vfs_fsync(iocb->ki_filp, 0); |
200baa21 TM |
602 | if (err < 0) |
603 | result = err; | |
604 | } | |
7e381172 CL |
605 | if (result > 0) |
606 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
1da177e4 LT |
607 | out: |
608 | return result; | |
609 | ||
610 | out_swapfile: | |
611 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
612 | goto out; | |
613 | } | |
614 | ||
ce4ef7c0 BS |
615 | ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, |
616 | struct file *filp, loff_t *ppos, | |
617 | size_t count, unsigned int flags) | |
bf40d343 SJ |
618 | { |
619 | struct dentry *dentry = filp->f_path.dentry; | |
620 | struct inode *inode = dentry->d_inode; | |
7e381172 | 621 | unsigned long written = 0; |
bf40d343 SJ |
622 | ssize_t ret; |
623 | ||
624 | dprintk("NFS splice_write(%s/%s, %lu@%llu)\n", | |
625 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
626 | (unsigned long) count, (unsigned long long) *ppos); | |
627 | ||
628 | /* | |
629 | * The combination of splice and an O_APPEND destination is disallowed. | |
630 | */ | |
631 | ||
bf40d343 | 632 | ret = generic_file_splice_write(pipe, filp, ppos, count, flags); |
7e381172 CL |
633 | if (ret > 0) |
634 | written = ret; | |
635 | ||
bf40d343 | 636 | if (ret >= 0 && nfs_need_sync_write(filp, inode)) { |
af7fa165 | 637 | int err = vfs_fsync(filp, 0); |
bf40d343 SJ |
638 | if (err < 0) |
639 | ret = err; | |
640 | } | |
7e381172 CL |
641 | if (ret > 0) |
642 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
bf40d343 SJ |
643 | return ret; |
644 | } | |
645 | ||
5eebde23 SJ |
646 | static int |
647 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
648 | { |
649 | struct inode *inode = filp->f_mapping->host; | |
650 | int status = 0; | |
21ac19d4 | 651 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 652 | |
039c4d7a | 653 | /* Try local locking first */ |
6d34ac19 BF |
654 | posix_test_lock(filp, fl); |
655 | if (fl->fl_type != F_UNLCK) { | |
656 | /* found a conflict */ | |
039c4d7a | 657 | goto out; |
1da177e4 | 658 | } |
21ac19d4 | 659 | fl->fl_type = saved_type; |
039c4d7a | 660 | |
011e2a7f | 661 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
662 | goto out_noconflict; |
663 | ||
5eebde23 | 664 | if (is_local) |
039c4d7a TM |
665 | goto out_noconflict; |
666 | ||
667 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
668 | out: | |
1da177e4 | 669 | return status; |
039c4d7a TM |
670 | out_noconflict: |
671 | fl->fl_type = F_UNLCK; | |
672 | goto out; | |
1da177e4 LT |
673 | } |
674 | ||
675 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
676 | { | |
677 | int res = 0; | |
678 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
679 | case FL_POSIX: | |
680 | res = posix_lock_file_wait(file, fl); | |
681 | break; | |
682 | case FL_FLOCK: | |
683 | res = flock_lock_file_wait(file, fl); | |
684 | break; | |
685 | default: | |
686 | BUG(); | |
687 | } | |
1da177e4 LT |
688 | return res; |
689 | } | |
690 | ||
5eebde23 SJ |
691 | static int |
692 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
693 | { |
694 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
695 | int status; |
696 | ||
1da177e4 LT |
697 | /* |
698 | * Flush all pending writes before doing anything | |
699 | * with locks.. | |
700 | */ | |
29884df0 | 701 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 LT |
702 | |
703 | /* NOTE: special case | |
704 | * If we're signalled while cleaning up locks on process exit, we | |
705 | * still need to complete the unlock. | |
706 | */ | |
5eebde23 SJ |
707 | /* |
708 | * Use local locking if mounted with "-onolock" or with appropriate | |
709 | * "-olocal_lock=" | |
710 | */ | |
711 | if (!is_local) | |
1da177e4 LT |
712 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
713 | else | |
714 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
715 | return status; |
716 | } | |
717 | ||
6b96724e RL |
718 | static int |
719 | is_time_granular(struct timespec *ts) { | |
720 | return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000)); | |
721 | } | |
722 | ||
5eebde23 SJ |
723 | static int |
724 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
725 | { |
726 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
727 | int status; |
728 | ||
1da177e4 LT |
729 | /* |
730 | * Flush all pending writes before doing anything | |
731 | * with locks.. | |
732 | */ | |
29884df0 TM |
733 | status = nfs_sync_mapping(filp->f_mapping); |
734 | if (status != 0) | |
1da177e4 LT |
735 | goto out; |
736 | ||
5eebde23 SJ |
737 | /* |
738 | * Use local locking if mounted with "-onolock" or with appropriate | |
739 | * "-olocal_lock=" | |
740 | */ | |
741 | if (!is_local) | |
1da177e4 | 742 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 743 | else |
1da177e4 | 744 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
745 | if (status < 0) |
746 | goto out; | |
6b96724e | 747 | |
1da177e4 | 748 | /* |
6b96724e RL |
749 | * Revalidate the cache if the server has time stamps granular |
750 | * enough to detect subsecond changes. Otherwise, clear the | |
751 | * cache to prevent missing any changes. | |
752 | * | |
1da177e4 LT |
753 | * This makes locking act as a cache coherency point. |
754 | */ | |
29884df0 | 755 | nfs_sync_mapping(filp->f_mapping); |
011e2a7f | 756 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { |
6b96724e RL |
757 | if (is_time_granular(&NFS_SERVER(inode)->time_delta)) |
758 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
759 | else | |
760 | nfs_zap_caches(inode); | |
761 | } | |
1da177e4 | 762 | out: |
1da177e4 LT |
763 | return status; |
764 | } | |
765 | ||
766 | /* | |
767 | * Lock a (portion of) a file | |
768 | */ | |
ce4ef7c0 | 769 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 770 | { |
6da24bc9 | 771 | struct inode *inode = filp->f_mapping->host; |
2116271a | 772 | int ret = -ENOLCK; |
5eebde23 | 773 | int is_local = 0; |
1da177e4 | 774 | |
6da24bc9 CL |
775 | dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n", |
776 | filp->f_path.dentry->d_parent->d_name.name, | |
777 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
778 | fl->fl_type, fl->fl_flags, |
779 | (long long)fl->fl_start, (long long)fl->fl_end); | |
6da24bc9 | 780 | |
91d5b470 | 781 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
782 | |
783 | /* No mandatory locks over NFS */ | |
dfad9441 | 784 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
785 | goto out_err; |
786 | ||
5eebde23 SJ |
787 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
788 | is_local = 1; | |
789 | ||
2116271a TM |
790 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
791 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
792 | if (ret < 0) | |
793 | goto out_err; | |
794 | } | |
1da177e4 LT |
795 | |
796 | if (IS_GETLK(cmd)) | |
5eebde23 | 797 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 798 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 799 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 800 | else |
5eebde23 | 801 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
802 | out_err: |
803 | return ret; | |
1da177e4 LT |
804 | } |
805 | ||
806 | /* | |
807 | * Lock a (portion of) a file | |
808 | */ | |
ce4ef7c0 | 809 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 810 | { |
5eebde23 SJ |
811 | struct inode *inode = filp->f_mapping->host; |
812 | int is_local = 0; | |
813 | ||
6da24bc9 CL |
814 | dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n", |
815 | filp->f_path.dentry->d_parent->d_name.name, | |
816 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
817 | fl->fl_type, fl->fl_flags); |
818 | ||
1da177e4 LT |
819 | if (!(fl->fl_flags & FL_FLOCK)) |
820 | return -ENOLCK; | |
821 | ||
5eebde23 SJ |
822 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
823 | is_local = 1; | |
824 | ||
1da177e4 LT |
825 | /* We're simulating flock() locks using posix locks on the server */ |
826 | fl->fl_owner = (fl_owner_t)filp; | |
827 | fl->fl_start = 0; | |
828 | fl->fl_end = OFFSET_MAX; | |
829 | ||
830 | if (fl->fl_type == F_UNLCK) | |
5eebde23 SJ |
831 | return do_unlk(filp, cmd, fl, is_local); |
832 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 833 | } |
370f6599 | 834 | |
6da24bc9 CL |
835 | /* |
836 | * There is no protocol support for leases, so we have no way to implement | |
837 | * them correctly in the face of opens by other clients. | |
838 | */ | |
ce4ef7c0 | 839 | int nfs_setlease(struct file *file, long arg, struct file_lock **fl) |
370f6599 | 840 | { |
6da24bc9 CL |
841 | dprintk("NFS: setlease(%s/%s, arg=%ld)\n", |
842 | file->f_path.dentry->d_parent->d_name.name, | |
843 | file->f_path.dentry->d_name.name, arg); | |
370f6599 BF |
844 | return -EINVAL; |
845 | } | |
1788ea6e | 846 | |
0486958f JL |
847 | const struct file_operations nfs_file_operations = { |
848 | .llseek = nfs_file_llseek, | |
849 | .read = do_sync_read, | |
850 | .write = do_sync_write, | |
851 | .aio_read = nfs_file_read, | |
852 | .aio_write = nfs_file_write, | |
853 | .mmap = nfs_file_mmap, | |
854 | .open = nfs_file_open, | |
855 | .flush = nfs_file_flush, | |
856 | .release = nfs_file_release, | |
857 | .fsync = nfs_file_fsync, | |
858 | .lock = nfs_lock, | |
859 | .flock = nfs_flock, | |
860 | .splice_read = nfs_file_splice_read, | |
861 | .splice_write = nfs_file_splice_write, | |
862 | .check_flags = nfs_check_flags, | |
863 | .setlease = nfs_setlease, | |
864 | }; |