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 | ||
ddda8e0a | 19 | #include <linux/module.h> |
1da177e4 LT |
20 | #include <linux/time.h> |
21 | #include <linux/kernel.h> | |
22 | #include <linux/errno.h> | |
23 | #include <linux/fcntl.h> | |
24 | #include <linux/stat.h> | |
25 | #include <linux/nfs_fs.h> | |
26 | #include <linux/nfs_mount.h> | |
27 | #include <linux/mm.h> | |
1da177e4 | 28 | #include <linux/pagemap.h> |
e8edc6e0 | 29 | #include <linux/aio.h> |
5a0e3ad6 | 30 | #include <linux/gfp.h> |
b608b283 | 31 | #include <linux/swap.h> |
1da177e4 LT |
32 | |
33 | #include <asm/uaccess.h> | |
1da177e4 LT |
34 | |
35 | #include "delegation.h" | |
94387fb1 | 36 | #include "internal.h" |
91d5b470 | 37 | #include "iostat.h" |
545db45f | 38 | #include "fscache.h" |
1da177e4 LT |
39 | |
40 | #define NFSDBG_FACILITY NFSDBG_FILE | |
41 | ||
f0f37e2f | 42 | static const struct vm_operations_struct nfs_file_vm_ops; |
94387fb1 | 43 | |
1da177e4 LT |
44 | /* Hack for future NFS swap support */ |
45 | #ifndef IS_SWAPFILE | |
46 | # define IS_SWAPFILE(inode) (0) | |
47 | #endif | |
48 | ||
ce4ef7c0 | 49 | int nfs_check_flags(int flags) |
1da177e4 LT |
50 | { |
51 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
52 | return -EINVAL; | |
53 | ||
54 | return 0; | |
55 | } | |
89d77c8f | 56 | EXPORT_SYMBOL_GPL(nfs_check_flags); |
1da177e4 LT |
57 | |
58 | /* | |
59 | * Open file | |
60 | */ | |
61 | static int | |
62 | nfs_file_open(struct inode *inode, struct file *filp) | |
63 | { | |
1da177e4 LT |
64 | int res; |
65 | ||
6da24bc9 | 66 | dprintk("NFS: open file(%s/%s)\n", |
cc0dd2d1 CL |
67 | filp->f_path.dentry->d_parent->d_name.name, |
68 | filp->f_path.dentry->d_name.name); | |
69 | ||
c2459dc4 | 70 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
1da177e4 LT |
71 | res = nfs_check_flags(filp->f_flags); |
72 | if (res) | |
73 | return res; | |
74 | ||
46cb650c | 75 | res = nfs_open(inode, filp); |
1da177e4 LT |
76 | return res; |
77 | } | |
78 | ||
ce4ef7c0 | 79 | int |
1da177e4 LT |
80 | nfs_file_release(struct inode *inode, struct file *filp) |
81 | { | |
6da24bc9 | 82 | dprintk("NFS: release(%s/%s)\n", |
6f276e49 RM |
83 | filp->f_path.dentry->d_parent->d_name.name, |
84 | filp->f_path.dentry->d_name.name); | |
6da24bc9 | 85 | |
91d5b470 | 86 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
46cb650c | 87 | return nfs_release(inode, filp); |
1da177e4 | 88 | } |
89d77c8f | 89 | EXPORT_SYMBOL_GPL(nfs_file_release); |
1da177e4 | 90 | |
980802e3 TM |
91 | /** |
92 | * nfs_revalidate_size - Revalidate the file size | |
93 | * @inode - pointer to inode struct | |
94 | * @file - pointer to struct file | |
95 | * | |
96 | * Revalidates the file length. This is basically a wrapper around | |
97 | * nfs_revalidate_inode() that takes into account the fact that we may | |
98 | * have cached writes (in which case we don't care about the server's | |
99 | * idea of what the file length is), or O_DIRECT (in which case we | |
100 | * shouldn't trust the cache). | |
101 | */ | |
102 | static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) | |
103 | { | |
104 | struct nfs_server *server = NFS_SERVER(inode); | |
105 | struct nfs_inode *nfsi = NFS_I(inode); | |
106 | ||
d7cf8dd0 TM |
107 | if (nfs_have_delegated_attributes(inode)) |
108 | goto out_noreval; | |
109 | ||
980802e3 TM |
110 | if (filp->f_flags & O_DIRECT) |
111 | goto force_reval; | |
d7cf8dd0 TM |
112 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
113 | goto force_reval; | |
114 | if (nfs_attribute_timeout(inode)) | |
115 | goto force_reval; | |
116 | out_noreval: | |
117 | return 0; | |
980802e3 TM |
118 | force_reval: |
119 | return __nfs_revalidate_inode(server, inode); | |
120 | } | |
121 | ||
ce4ef7c0 | 122 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin) |
980802e3 | 123 | { |
6da24bc9 | 124 | dprintk("NFS: llseek file(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
125 | filp->f_path.dentry->d_parent->d_name.name, |
126 | filp->f_path.dentry->d_name.name, | |
127 | offset, origin); | |
128 | ||
06222e49 JB |
129 | /* |
130 | * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate | |
131 | * the cached file length | |
132 | */ | |
6c529617 | 133 | if (origin != SEEK_SET && origin != SEEK_CUR) { |
980802e3 | 134 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 135 | |
980802e3 TM |
136 | int retval = nfs_revalidate_file_size(inode, filp); |
137 | if (retval < 0) | |
138 | return (loff_t)retval; | |
79835a71 | 139 | } |
d5e66348 | 140 | |
79835a71 | 141 | return generic_file_llseek(filp, offset, origin); |
980802e3 | 142 | } |
89d77c8f | 143 | EXPORT_SYMBOL_GPL(nfs_file_llseek); |
980802e3 | 144 | |
1da177e4 LT |
145 | /* |
146 | * Flush all dirty pages, and check for write errors. | |
1da177e4 | 147 | */ |
ce4ef7c0 | 148 | int |
75e1fcc0 | 149 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 150 | { |
6da24bc9 CL |
151 | struct dentry *dentry = file->f_path.dentry; |
152 | struct inode *inode = dentry->d_inode; | |
1da177e4 | 153 | |
6da24bc9 CL |
154 | dprintk("NFS: flush(%s/%s)\n", |
155 | dentry->d_parent->d_name.name, | |
156 | dentry->d_name.name); | |
1da177e4 | 157 | |
c2459dc4 | 158 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
159 | if ((file->f_mode & FMODE_WRITE) == 0) |
160 | return 0; | |
7b159fc1 | 161 | |
14546c33 TM |
162 | /* |
163 | * If we're holding a write delegation, then just start the i/o | |
164 | * but don't wait for completion (or send a commit). | |
165 | */ | |
011e2a7f | 166 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
14546c33 TM |
167 | return filemap_fdatawrite(file->f_mapping); |
168 | ||
7fe5c398 | 169 | /* Flush writes to the server and return any errors */ |
af7fa165 | 170 | return vfs_fsync(file, 0); |
1da177e4 | 171 | } |
89d77c8f | 172 | EXPORT_SYMBOL_GPL(nfs_file_flush); |
1da177e4 | 173 | |
ce4ef7c0 | 174 | ssize_t |
027445c3 BP |
175 | nfs_file_read(struct kiocb *iocb, const struct iovec *iov, |
176 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 177 | { |
01cce933 | 178 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 LT |
179 | struct inode * inode = dentry->d_inode; |
180 | ssize_t result; | |
181 | ||
1da177e4 | 182 | if (iocb->ki_filp->f_flags & O_DIRECT) |
a564b8f0 | 183 | return nfs_file_direct_read(iocb, iov, nr_segs, pos, true); |
1da177e4 | 184 | |
6da24bc9 | 185 | dprintk("NFS: read(%s/%s, %lu@%lu)\n", |
1da177e4 | 186 | dentry->d_parent->d_name.name, dentry->d_name.name, |
6f276e49 | 187 | (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos); |
1da177e4 | 188 | |
44b11874 | 189 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); |
4184dcf2 | 190 | if (!result) { |
027445c3 | 191 | result = generic_file_aio_read(iocb, iov, nr_segs, pos); |
4184dcf2 CL |
192 | if (result > 0) |
193 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
194 | } | |
1da177e4 LT |
195 | return result; |
196 | } | |
89d77c8f | 197 | EXPORT_SYMBOL_GPL(nfs_file_read); |
1da177e4 | 198 | |
ce4ef7c0 | 199 | ssize_t |
f0930fff JA |
200 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
201 | struct pipe_inode_info *pipe, size_t count, | |
202 | unsigned int flags) | |
1da177e4 | 203 | { |
01cce933 | 204 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
205 | struct inode *inode = dentry->d_inode; |
206 | ssize_t res; | |
207 | ||
6da24bc9 | 208 | dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n", |
1da177e4 LT |
209 | dentry->d_parent->d_name.name, dentry->d_name.name, |
210 | (unsigned long) count, (unsigned long long) *ppos); | |
211 | ||
44b11874 | 212 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
aa2f1ef1 | 213 | if (!res) { |
f0930fff | 214 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
aa2f1ef1 CL |
215 | if (res > 0) |
216 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); | |
217 | } | |
1da177e4 LT |
218 | return res; |
219 | } | |
89d77c8f | 220 | EXPORT_SYMBOL_GPL(nfs_file_splice_read); |
1da177e4 | 221 | |
ce4ef7c0 | 222 | int |
1da177e4 LT |
223 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
224 | { | |
01cce933 | 225 | struct dentry *dentry = file->f_path.dentry; |
1da177e4 LT |
226 | struct inode *inode = dentry->d_inode; |
227 | int status; | |
228 | ||
6da24bc9 | 229 | dprintk("NFS: mmap(%s/%s)\n", |
1da177e4 LT |
230 | dentry->d_parent->d_name.name, dentry->d_name.name); |
231 | ||
e1ebfd33 TM |
232 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
233 | * so we call that before revalidating the mapping | |
234 | */ | |
235 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
236 | if (!status) { |
237 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 238 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 239 | } |
1da177e4 LT |
240 | return status; |
241 | } | |
89d77c8f | 242 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
243 | |
244 | /* | |
245 | * Flush any dirty pages for this process, and check for write errors. | |
246 | * The return status from this call provides a reliable indication of | |
247 | * whether any write errors occurred for this process. | |
af7fa165 TM |
248 | * |
249 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
250 | * disk, but it retrieves and clears ctx->error after synching, despite | |
251 | * the two being set at the same time in nfs_context_set_write_error(). | |
252 | * This is because the former is used to notify the _next_ call to | |
25985edc | 253 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 254 | * fall back to doing a synchronous write. |
1da177e4 | 255 | */ |
ce4ef7c0 | 256 | int |
a5c58892 | 257 | nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 258 | { |
7ea80859 | 259 | struct dentry *dentry = file->f_path.dentry; |
cd3758e3 | 260 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
1da177e4 | 261 | struct inode *inode = dentry->d_inode; |
af7fa165 TM |
262 | int have_error, status; |
263 | int ret = 0; | |
264 | ||
6da24bc9 | 265 | dprintk("NFS: fsync file(%s/%s) datasync %d\n", |
54917786 CL |
266 | dentry->d_parent->d_name.name, dentry->d_name.name, |
267 | datasync); | |
1da177e4 | 268 | |
91d5b470 | 269 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
af7fa165 TM |
270 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
271 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
2edb6bc3 N |
272 | if (status >= 0 && ret < 0) |
273 | status = ret; | |
af7fa165 TM |
274 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
275 | if (have_error) | |
276 | ret = xchg(&ctx->error, 0); | |
0702099b | 277 | if (!ret && status < 0) |
af7fa165 | 278 | ret = status; |
a5c58892 BS |
279 | return ret; |
280 | } | |
89d77c8f | 281 | EXPORT_SYMBOL_GPL(nfs_file_fsync_commit); |
a5c58892 BS |
282 | |
283 | static int | |
284 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) | |
285 | { | |
286 | int ret; | |
287 | struct inode *inode = file->f_path.dentry->d_inode; | |
288 | ||
289 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
7b281ee0 TM |
290 | if (ret != 0) |
291 | goto out; | |
a5c58892 BS |
292 | mutex_lock(&inode->i_mutex); |
293 | ret = nfs_file_fsync_commit(file, start, end, datasync); | |
02c24a82 | 294 | mutex_unlock(&inode->i_mutex); |
7b281ee0 | 295 | out: |
af7fa165 | 296 | return ret; |
1da177e4 LT |
297 | } |
298 | ||
38c73044 PS |
299 | /* |
300 | * Decide whether a read/modify/write cycle may be more efficient | |
301 | * then a modify/write/read cycle when writing to a page in the | |
302 | * page cache. | |
303 | * | |
304 | * The modify/write/read cycle may occur if a page is read before | |
305 | * being completely filled by the writer. In this situation, the | |
306 | * page must be completely written to stable storage on the server | |
307 | * before it can be refilled by reading in the page from the server. | |
308 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
309 | * done. | |
310 | * | |
311 | * It may be more efficient to read the page first if the file is | |
312 | * open for reading in addition to writing, the page is not marked | |
313 | * as Uptodate, it is not dirty or waiting to be committed, | |
314 | * indicating that it was previously allocated and then modified, | |
315 | * that there were valid bytes of data in that range of the file, | |
316 | * and that the new data won't completely replace the old data in | |
317 | * that range of the file. | |
318 | */ | |
319 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
320 | loff_t pos, unsigned len) | |
321 | { | |
322 | unsigned int pglen = nfs_page_length(page); | |
323 | unsigned int offset = pos & (PAGE_CACHE_SIZE - 1); | |
324 | unsigned int end = offset + len; | |
325 | ||
326 | if ((file->f_mode & FMODE_READ) && /* open for read? */ | |
327 | !PageUptodate(page) && /* Uptodate? */ | |
328 | !PagePrivate(page) && /* i/o request already? */ | |
329 | pglen && /* valid bytes of file? */ | |
330 | (end < pglen || offset)) /* replace all valid bytes? */ | |
331 | return 1; | |
332 | return 0; | |
333 | } | |
334 | ||
1da177e4 | 335 | /* |
4899f9c8 NP |
336 | * This does the "real" work of the write. We must allocate and lock the |
337 | * page to be sent back to the generic routine, which then copies the | |
338 | * data from user space. | |
1da177e4 LT |
339 | * |
340 | * If the writer ends up delaying the write, the writer needs to | |
341 | * increment the page use counts until he is done with the page. | |
342 | */ | |
4899f9c8 NP |
343 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
344 | loff_t pos, unsigned len, unsigned flags, | |
345 | struct page **pagep, void **fsdata) | |
1da177e4 | 346 | { |
4899f9c8 | 347 | int ret; |
38c73044 | 348 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
4899f9c8 | 349 | struct page *page; |
38c73044 | 350 | int once_thru = 0; |
4899f9c8 | 351 | |
b7eaefaa CL |
352 | dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n", |
353 | file->f_path.dentry->d_parent->d_name.name, | |
354 | file->f_path.dentry->d_name.name, | |
355 | mapping->host->i_ino, len, (long long) pos); | |
356 | ||
38c73044 | 357 | start: |
72cb77f4 TM |
358 | /* |
359 | * Prevent starvation issues if someone is doing a consistency | |
360 | * sync-to-disk | |
361 | */ | |
362 | ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, | |
363 | nfs_wait_bit_killable, TASK_KILLABLE); | |
364 | if (ret) | |
365 | return ret; | |
366 | ||
54566b2c | 367 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
368 | if (!page) |
369 | return -ENOMEM; | |
370 | *pagep = page; | |
371 | ||
372 | ret = nfs_flush_incompatible(file, page); | |
373 | if (ret) { | |
374 | unlock_page(page); | |
375 | page_cache_release(page); | |
38c73044 PS |
376 | } else if (!once_thru && |
377 | nfs_want_read_modify_write(file, page, pos, len)) { | |
378 | once_thru = 1; | |
379 | ret = nfs_readpage(file, page); | |
380 | page_cache_release(page); | |
381 | if (!ret) | |
382 | goto start; | |
4899f9c8 NP |
383 | } |
384 | return ret; | |
1da177e4 LT |
385 | } |
386 | ||
4899f9c8 NP |
387 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
388 | loff_t pos, unsigned len, unsigned copied, | |
389 | struct page *page, void *fsdata) | |
1da177e4 | 390 | { |
4899f9c8 NP |
391 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
392 | int status; | |
1da177e4 | 393 | |
b7eaefaa CL |
394 | dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n", |
395 | file->f_path.dentry->d_parent->d_name.name, | |
396 | file->f_path.dentry->d_name.name, | |
397 | mapping->host->i_ino, len, (long long) pos); | |
398 | ||
efc91ed0 TM |
399 | /* |
400 | * Zero any uninitialised parts of the page, and then mark the page | |
401 | * as up to date if it turns out that we're extending the file. | |
402 | */ | |
403 | if (!PageUptodate(page)) { | |
404 | unsigned pglen = nfs_page_length(page); | |
405 | unsigned end = offset + len; | |
406 | ||
407 | if (pglen == 0) { | |
408 | zero_user_segments(page, 0, offset, | |
409 | end, PAGE_CACHE_SIZE); | |
410 | SetPageUptodate(page); | |
411 | } else if (end >= pglen) { | |
412 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
413 | if (offset == 0) | |
414 | SetPageUptodate(page); | |
415 | } else | |
416 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
417 | } | |
418 | ||
4899f9c8 | 419 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
420 | |
421 | unlock_page(page); | |
422 | page_cache_release(page); | |
423 | ||
3d509e54 CL |
424 | if (status < 0) |
425 | return status; | |
2701d086 | 426 | NFS_I(mapping->host)->write_io += copied; |
3d509e54 | 427 | return copied; |
1da177e4 LT |
428 | } |
429 | ||
6b9b3514 DH |
430 | /* |
431 | * Partially or wholly invalidate a page | |
432 | * - Release the private state associated with a page if undergoing complete | |
433 | * page invalidation | |
545db45f | 434 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
435 | * - Caller holds page lock |
436 | */ | |
2ff28e22 | 437 | static void nfs_invalidate_page(struct page *page, unsigned long offset) |
cd52ed35 | 438 | { |
b7eaefaa CL |
439 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset); |
440 | ||
1c75950b TM |
441 | if (offset != 0) |
442 | return; | |
d2ccddf0 | 443 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 444 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
445 | |
446 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
447 | } |
448 | ||
6b9b3514 DH |
449 | /* |
450 | * Attempt to release the private state associated with a page | |
545db45f | 451 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
452 | * - Caller holds page lock |
453 | * - Return true (may release page) or false (may not) | |
454 | */ | |
cd52ed35 TM |
455 | static int nfs_release_page(struct page *page, gfp_t gfp) |
456 | { | |
b608b283 TM |
457 | struct address_space *mapping = page->mapping; |
458 | ||
b7eaefaa CL |
459 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
460 | ||
5cf02d09 JL |
461 | /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not |
462 | * doing this memory reclaim for a fs-related allocation. | |
463 | */ | |
464 | if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL && | |
465 | !(current->flags & PF_FSTRANS)) { | |
b608b283 TM |
466 | int how = FLUSH_SYNC; |
467 | ||
468 | /* Don't let kswapd deadlock waiting for OOM RPC calls */ | |
469 | if (current_is_kswapd()) | |
470 | how = 0; | |
471 | nfs_commit_inode(mapping->host, how); | |
472 | } | |
e3db7691 | 473 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
474 | if (PagePrivate(page)) |
475 | return 0; | |
476 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
477 | } |
478 | ||
6b9b3514 DH |
479 | /* |
480 | * Attempt to clear the private state associated with a page when an error | |
481 | * occurs that requires the cached contents of an inode to be written back or | |
482 | * destroyed | |
545db45f | 483 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
484 | * - Caller holds page lock |
485 | * - Return 0 if successful, -error otherwise | |
486 | */ | |
e3db7691 TM |
487 | static int nfs_launder_page(struct page *page) |
488 | { | |
d56b4ddf | 489 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 490 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
491 | |
492 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
493 | inode->i_ino, (long long)page_offset(page)); | |
494 | ||
545db45f | 495 | nfs_fscache_wait_on_page_write(nfsi, page); |
b7eaefaa | 496 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
497 | } |
498 | ||
a564b8f0 MG |
499 | #ifdef CONFIG_NFS_SWAP |
500 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
501 | sector_t *span) | |
502 | { | |
503 | *span = sis->pages; | |
504 | return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1); | |
505 | } | |
506 | ||
507 | static void nfs_swap_deactivate(struct file *file) | |
508 | { | |
509 | xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0); | |
510 | } | |
511 | #endif | |
512 | ||
f5e54d6e | 513 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
514 | .readpage = nfs_readpage, |
515 | .readpages = nfs_readpages, | |
9cccef95 | 516 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
517 | .writepage = nfs_writepage, |
518 | .writepages = nfs_writepages, | |
4899f9c8 NP |
519 | .write_begin = nfs_write_begin, |
520 | .write_end = nfs_write_end, | |
cd52ed35 TM |
521 | .invalidatepage = nfs_invalidate_page, |
522 | .releasepage = nfs_release_page, | |
1da177e4 | 523 | .direct_IO = nfs_direct_IO, |
074cc1de | 524 | .migratepage = nfs_migrate_page, |
e3db7691 | 525 | .launder_page = nfs_launder_page, |
f590f333 | 526 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
527 | #ifdef CONFIG_NFS_SWAP |
528 | .swap_activate = nfs_swap_activate, | |
529 | .swap_deactivate = nfs_swap_deactivate, | |
530 | #endif | |
1da177e4 LT |
531 | }; |
532 | ||
6b9b3514 DH |
533 | /* |
534 | * Notification that a PTE pointing to an NFS page is about to be made | |
535 | * writable, implying that someone is about to modify the page through a | |
536 | * shared-writable mapping | |
537 | */ | |
c2ec175c | 538 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 539 | { |
c2ec175c | 540 | struct page *page = vmf->page; |
94387fb1 | 541 | struct file *filp = vma->vm_file; |
b7eaefaa | 542 | struct dentry *dentry = filp->f_path.dentry; |
94387fb1 | 543 | unsigned pagelen; |
bc4866b6 | 544 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 545 | struct address_space *mapping; |
94387fb1 | 546 | |
b7eaefaa CL |
547 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n", |
548 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
549 | filp->f_mapping->host->i_ino, | |
550 | (long long)page_offset(page)); | |
551 | ||
545db45f DH |
552 | /* make sure the cache has finished storing the page */ |
553 | nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page); | |
554 | ||
94387fb1 | 555 | lock_page(page); |
d56b4ddf | 556 | mapping = page_file_mapping(page); |
b7eaefaa | 557 | if (mapping != dentry->d_inode->i_mapping) |
8b1f9ee5 TM |
558 | goto out_unlock; |
559 | ||
2aeb98f4 TM |
560 | wait_on_page_writeback(page); |
561 | ||
94387fb1 | 562 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
563 | if (pagelen == 0) |
564 | goto out_unlock; | |
4899f9c8 | 565 | |
bc4866b6 TM |
566 | ret = VM_FAULT_LOCKED; |
567 | if (nfs_flush_incompatible(filp, page) == 0 && | |
568 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
569 | goto out; | |
8b1f9ee5 | 570 | |
bc4866b6 | 571 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
572 | out_unlock: |
573 | unlock_page(page); | |
bc4866b6 TM |
574 | out: |
575 | return ret; | |
94387fb1 TM |
576 | } |
577 | ||
f0f37e2f | 578 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 TM |
579 | .fault = filemap_fault, |
580 | .page_mkwrite = nfs_vm_page_mkwrite, | |
0b173bc4 | 581 | .remap_pages = generic_file_remap_pages, |
94387fb1 TM |
582 | }; |
583 | ||
7b159fc1 TM |
584 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
585 | { | |
586 | struct nfs_open_context *ctx; | |
587 | ||
6b2f3d1f | 588 | if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) |
7b159fc1 | 589 | return 1; |
cd3758e3 | 590 | ctx = nfs_file_open_context(filp); |
7b159fc1 TM |
591 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) |
592 | return 1; | |
593 | return 0; | |
594 | } | |
595 | ||
ce4ef7c0 BS |
596 | ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, |
597 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 598 | { |
01cce933 | 599 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 | 600 | struct inode * inode = dentry->d_inode; |
7e381172 | 601 | unsigned long written = 0; |
1da177e4 | 602 | ssize_t result; |
027445c3 | 603 | size_t count = iov_length(iov, nr_segs); |
1da177e4 | 604 | |
1da177e4 | 605 | if (iocb->ki_filp->f_flags & O_DIRECT) |
a564b8f0 | 606 | return nfs_file_direct_write(iocb, iov, nr_segs, pos, true); |
1da177e4 | 607 | |
6da24bc9 | 608 | dprintk("NFS: write(%s/%s, %lu@%Ld)\n", |
1da177e4 | 609 | dentry->d_parent->d_name.name, dentry->d_name.name, |
6da24bc9 | 610 | (unsigned long) count, (long long) pos); |
1da177e4 LT |
611 | |
612 | result = -EBUSY; | |
613 | if (IS_SWAPFILE(inode)) | |
614 | goto out_swapfile; | |
7d52e862 TM |
615 | /* |
616 | * O_APPEND implies that we must revalidate the file length. | |
617 | */ | |
618 | if (iocb->ki_filp->f_flags & O_APPEND) { | |
619 | result = nfs_revalidate_file_size(inode, iocb->ki_filp); | |
620 | if (result) | |
621 | goto out; | |
fe51beec | 622 | } |
1da177e4 LT |
623 | |
624 | result = count; | |
625 | if (!count) | |
626 | goto out; | |
627 | ||
027445c3 | 628 | result = generic_file_aio_write(iocb, iov, nr_segs, pos); |
7e381172 CL |
629 | if (result > 0) |
630 | written = result; | |
631 | ||
6b2f3d1f | 632 | /* Return error values for O_DSYNC and IS_SYNC() */ |
7b159fc1 | 633 | if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { |
af7fa165 | 634 | int err = vfs_fsync(iocb->ki_filp, 0); |
200baa21 TM |
635 | if (err < 0) |
636 | result = err; | |
637 | } | |
7e381172 CL |
638 | if (result > 0) |
639 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
1da177e4 LT |
640 | out: |
641 | return result; | |
642 | ||
643 | out_swapfile: | |
644 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
645 | goto out; | |
646 | } | |
89d77c8f | 647 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 648 | |
ce4ef7c0 BS |
649 | ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, |
650 | struct file *filp, loff_t *ppos, | |
651 | size_t count, unsigned int flags) | |
bf40d343 SJ |
652 | { |
653 | struct dentry *dentry = filp->f_path.dentry; | |
654 | struct inode *inode = dentry->d_inode; | |
7e381172 | 655 | unsigned long written = 0; |
bf40d343 SJ |
656 | ssize_t ret; |
657 | ||
658 | dprintk("NFS splice_write(%s/%s, %lu@%llu)\n", | |
659 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
660 | (unsigned long) count, (unsigned long long) *ppos); | |
661 | ||
662 | /* | |
663 | * The combination of splice and an O_APPEND destination is disallowed. | |
664 | */ | |
665 | ||
bf40d343 | 666 | ret = generic_file_splice_write(pipe, filp, ppos, count, flags); |
7e381172 CL |
667 | if (ret > 0) |
668 | written = ret; | |
669 | ||
bf40d343 | 670 | if (ret >= 0 && nfs_need_sync_write(filp, inode)) { |
af7fa165 | 671 | int err = vfs_fsync(filp, 0); |
bf40d343 SJ |
672 | if (err < 0) |
673 | ret = err; | |
674 | } | |
7e381172 CL |
675 | if (ret > 0) |
676 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
bf40d343 SJ |
677 | return ret; |
678 | } | |
89d77c8f | 679 | EXPORT_SYMBOL_GPL(nfs_file_splice_write); |
bf40d343 | 680 | |
5eebde23 SJ |
681 | static int |
682 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
683 | { |
684 | struct inode *inode = filp->f_mapping->host; | |
685 | int status = 0; | |
21ac19d4 | 686 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 687 | |
039c4d7a | 688 | /* Try local locking first */ |
6d34ac19 BF |
689 | posix_test_lock(filp, fl); |
690 | if (fl->fl_type != F_UNLCK) { | |
691 | /* found a conflict */ | |
039c4d7a | 692 | goto out; |
1da177e4 | 693 | } |
21ac19d4 | 694 | fl->fl_type = saved_type; |
039c4d7a | 695 | |
011e2a7f | 696 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
697 | goto out_noconflict; |
698 | ||
5eebde23 | 699 | if (is_local) |
039c4d7a TM |
700 | goto out_noconflict; |
701 | ||
702 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
703 | out: | |
1da177e4 | 704 | return status; |
039c4d7a TM |
705 | out_noconflict: |
706 | fl->fl_type = F_UNLCK; | |
707 | goto out; | |
1da177e4 LT |
708 | } |
709 | ||
710 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
711 | { | |
712 | int res = 0; | |
713 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
714 | case FL_POSIX: | |
715 | res = posix_lock_file_wait(file, fl); | |
716 | break; | |
717 | case FL_FLOCK: | |
718 | res = flock_lock_file_wait(file, fl); | |
719 | break; | |
720 | default: | |
721 | BUG(); | |
722 | } | |
1da177e4 LT |
723 | return res; |
724 | } | |
725 | ||
5eebde23 SJ |
726 | static int |
727 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
728 | { |
729 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
730 | int status; |
731 | ||
1da177e4 LT |
732 | /* |
733 | * Flush all pending writes before doing anything | |
734 | * with locks.. | |
735 | */ | |
29884df0 | 736 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 LT |
737 | |
738 | /* NOTE: special case | |
739 | * If we're signalled while cleaning up locks on process exit, we | |
740 | * still need to complete the unlock. | |
741 | */ | |
5eebde23 SJ |
742 | /* |
743 | * Use local locking if mounted with "-onolock" or with appropriate | |
744 | * "-olocal_lock=" | |
745 | */ | |
746 | if (!is_local) | |
1da177e4 LT |
747 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
748 | else | |
749 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
750 | return status; |
751 | } | |
752 | ||
6b96724e RL |
753 | static int |
754 | is_time_granular(struct timespec *ts) { | |
755 | return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000)); | |
756 | } | |
757 | ||
5eebde23 SJ |
758 | static int |
759 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
760 | { |
761 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
762 | int status; |
763 | ||
1da177e4 LT |
764 | /* |
765 | * Flush all pending writes before doing anything | |
766 | * with locks.. | |
767 | */ | |
29884df0 TM |
768 | status = nfs_sync_mapping(filp->f_mapping); |
769 | if (status != 0) | |
1da177e4 LT |
770 | goto out; |
771 | ||
5eebde23 SJ |
772 | /* |
773 | * Use local locking if mounted with "-onolock" or with appropriate | |
774 | * "-olocal_lock=" | |
775 | */ | |
776 | if (!is_local) | |
1da177e4 | 777 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 778 | else |
1da177e4 | 779 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
780 | if (status < 0) |
781 | goto out; | |
6b96724e | 782 | |
1da177e4 | 783 | /* |
6b96724e RL |
784 | * Revalidate the cache if the server has time stamps granular |
785 | * enough to detect subsecond changes. Otherwise, clear the | |
786 | * cache to prevent missing any changes. | |
787 | * | |
1da177e4 LT |
788 | * This makes locking act as a cache coherency point. |
789 | */ | |
29884df0 | 790 | nfs_sync_mapping(filp->f_mapping); |
011e2a7f | 791 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { |
6b96724e RL |
792 | if (is_time_granular(&NFS_SERVER(inode)->time_delta)) |
793 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
794 | else | |
795 | nfs_zap_caches(inode); | |
796 | } | |
1da177e4 | 797 | out: |
1da177e4 LT |
798 | return status; |
799 | } | |
800 | ||
801 | /* | |
802 | * Lock a (portion of) a file | |
803 | */ | |
ce4ef7c0 | 804 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 805 | { |
6da24bc9 | 806 | struct inode *inode = filp->f_mapping->host; |
2116271a | 807 | int ret = -ENOLCK; |
5eebde23 | 808 | int is_local = 0; |
1da177e4 | 809 | |
6da24bc9 CL |
810 | dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n", |
811 | filp->f_path.dentry->d_parent->d_name.name, | |
812 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
813 | fl->fl_type, fl->fl_flags, |
814 | (long long)fl->fl_start, (long long)fl->fl_end); | |
6da24bc9 | 815 | |
91d5b470 | 816 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
817 | |
818 | /* No mandatory locks over NFS */ | |
dfad9441 | 819 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
820 | goto out_err; |
821 | ||
5eebde23 SJ |
822 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
823 | is_local = 1; | |
824 | ||
2116271a TM |
825 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
826 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
827 | if (ret < 0) | |
828 | goto out_err; | |
829 | } | |
1da177e4 LT |
830 | |
831 | if (IS_GETLK(cmd)) | |
5eebde23 | 832 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 833 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 834 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 835 | else |
5eebde23 | 836 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
837 | out_err: |
838 | return ret; | |
1da177e4 | 839 | } |
89d77c8f | 840 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
841 | |
842 | /* | |
843 | * Lock a (portion of) a file | |
844 | */ | |
ce4ef7c0 | 845 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 846 | { |
5eebde23 SJ |
847 | struct inode *inode = filp->f_mapping->host; |
848 | int is_local = 0; | |
849 | ||
6da24bc9 CL |
850 | dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n", |
851 | filp->f_path.dentry->d_parent->d_name.name, | |
852 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
853 | fl->fl_type, fl->fl_flags); |
854 | ||
1da177e4 LT |
855 | if (!(fl->fl_flags & FL_FLOCK)) |
856 | return -ENOLCK; | |
857 | ||
ad0fcd4e JL |
858 | /* |
859 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
860 | * any standard. In principle we might be able to support LOCK_MAND | |
861 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
862 | * NFS code is not set up for it. | |
863 | */ | |
864 | if (fl->fl_type & LOCK_MAND) | |
865 | return -EINVAL; | |
866 | ||
5eebde23 SJ |
867 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
868 | is_local = 1; | |
869 | ||
1da177e4 LT |
870 | /* We're simulating flock() locks using posix locks on the server */ |
871 | fl->fl_owner = (fl_owner_t)filp; | |
872 | fl->fl_start = 0; | |
873 | fl->fl_end = OFFSET_MAX; | |
874 | ||
875 | if (fl->fl_type == F_UNLCK) | |
5eebde23 SJ |
876 | return do_unlk(filp, cmd, fl, is_local); |
877 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 878 | } |
89d77c8f | 879 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 880 | |
6da24bc9 CL |
881 | /* |
882 | * There is no protocol support for leases, so we have no way to implement | |
883 | * them correctly in the face of opens by other clients. | |
884 | */ | |
ce4ef7c0 | 885 | int nfs_setlease(struct file *file, long arg, struct file_lock **fl) |
370f6599 | 886 | { |
6da24bc9 CL |
887 | dprintk("NFS: setlease(%s/%s, arg=%ld)\n", |
888 | file->f_path.dentry->d_parent->d_name.name, | |
889 | file->f_path.dentry->d_name.name, arg); | |
370f6599 BF |
890 | return -EINVAL; |
891 | } | |
89d77c8f | 892 | EXPORT_SYMBOL_GPL(nfs_setlease); |
1788ea6e | 893 | |
0486958f JL |
894 | const struct file_operations nfs_file_operations = { |
895 | .llseek = nfs_file_llseek, | |
896 | .read = do_sync_read, | |
897 | .write = do_sync_write, | |
898 | .aio_read = nfs_file_read, | |
899 | .aio_write = nfs_file_write, | |
900 | .mmap = nfs_file_mmap, | |
901 | .open = nfs_file_open, | |
902 | .flush = nfs_file_flush, | |
903 | .release = nfs_file_release, | |
904 | .fsync = nfs_file_fsync, | |
905 | .lock = nfs_lock, | |
906 | .flock = nfs_flock, | |
907 | .splice_read = nfs_file_splice_read, | |
908 | .splice_write = nfs_file_splice_write, | |
909 | .check_flags = nfs_check_flags, | |
910 | .setlease = nfs_setlease, | |
911 | }; | |
ddda8e0a | 912 | EXPORT_SYMBOL_GPL(nfs_file_operations); |