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); | |
290 | mutex_lock(&inode->i_mutex); | |
291 | ret = nfs_file_fsync_commit(file, start, end, datasync); | |
02c24a82 | 292 | mutex_unlock(&inode->i_mutex); |
a5c58892 | 293 | |
af7fa165 | 294 | return ret; |
1da177e4 LT |
295 | } |
296 | ||
38c73044 PS |
297 | /* |
298 | * Decide whether a read/modify/write cycle may be more efficient | |
299 | * then a modify/write/read cycle when writing to a page in the | |
300 | * page cache. | |
301 | * | |
302 | * The modify/write/read cycle may occur if a page is read before | |
303 | * being completely filled by the writer. In this situation, the | |
304 | * page must be completely written to stable storage on the server | |
305 | * before it can be refilled by reading in the page from the server. | |
306 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
307 | * done. | |
308 | * | |
309 | * It may be more efficient to read the page first if the file is | |
310 | * open for reading in addition to writing, the page is not marked | |
311 | * as Uptodate, it is not dirty or waiting to be committed, | |
312 | * indicating that it was previously allocated and then modified, | |
313 | * that there were valid bytes of data in that range of the file, | |
314 | * and that the new data won't completely replace the old data in | |
315 | * that range of the file. | |
316 | */ | |
317 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
318 | loff_t pos, unsigned len) | |
319 | { | |
320 | unsigned int pglen = nfs_page_length(page); | |
321 | unsigned int offset = pos & (PAGE_CACHE_SIZE - 1); | |
322 | unsigned int end = offset + len; | |
323 | ||
324 | if ((file->f_mode & FMODE_READ) && /* open for read? */ | |
325 | !PageUptodate(page) && /* Uptodate? */ | |
326 | !PagePrivate(page) && /* i/o request already? */ | |
327 | pglen && /* valid bytes of file? */ | |
328 | (end < pglen || offset)) /* replace all valid bytes? */ | |
329 | return 1; | |
330 | return 0; | |
331 | } | |
332 | ||
1da177e4 | 333 | /* |
4899f9c8 NP |
334 | * This does the "real" work of the write. We must allocate and lock the |
335 | * page to be sent back to the generic routine, which then copies the | |
336 | * data from user space. | |
1da177e4 LT |
337 | * |
338 | * If the writer ends up delaying the write, the writer needs to | |
339 | * increment the page use counts until he is done with the page. | |
340 | */ | |
4899f9c8 NP |
341 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
342 | loff_t pos, unsigned len, unsigned flags, | |
343 | struct page **pagep, void **fsdata) | |
1da177e4 | 344 | { |
4899f9c8 | 345 | int ret; |
38c73044 | 346 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
4899f9c8 | 347 | struct page *page; |
38c73044 | 348 | int once_thru = 0; |
4899f9c8 | 349 | |
b7eaefaa CL |
350 | dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n", |
351 | file->f_path.dentry->d_parent->d_name.name, | |
352 | file->f_path.dentry->d_name.name, | |
353 | mapping->host->i_ino, len, (long long) pos); | |
354 | ||
38c73044 | 355 | start: |
72cb77f4 TM |
356 | /* |
357 | * Prevent starvation issues if someone is doing a consistency | |
358 | * sync-to-disk | |
359 | */ | |
360 | ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, | |
361 | nfs_wait_bit_killable, TASK_KILLABLE); | |
362 | if (ret) | |
363 | return ret; | |
364 | ||
54566b2c | 365 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
366 | if (!page) |
367 | return -ENOMEM; | |
368 | *pagep = page; | |
369 | ||
370 | ret = nfs_flush_incompatible(file, page); | |
371 | if (ret) { | |
372 | unlock_page(page); | |
373 | page_cache_release(page); | |
38c73044 PS |
374 | } else if (!once_thru && |
375 | nfs_want_read_modify_write(file, page, pos, len)) { | |
376 | once_thru = 1; | |
377 | ret = nfs_readpage(file, page); | |
378 | page_cache_release(page); | |
379 | if (!ret) | |
380 | goto start; | |
4899f9c8 NP |
381 | } |
382 | return ret; | |
1da177e4 LT |
383 | } |
384 | ||
4899f9c8 NP |
385 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
386 | loff_t pos, unsigned len, unsigned copied, | |
387 | struct page *page, void *fsdata) | |
1da177e4 | 388 | { |
4899f9c8 NP |
389 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
390 | int status; | |
1da177e4 | 391 | |
b7eaefaa CL |
392 | dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n", |
393 | file->f_path.dentry->d_parent->d_name.name, | |
394 | file->f_path.dentry->d_name.name, | |
395 | mapping->host->i_ino, len, (long long) pos); | |
396 | ||
efc91ed0 TM |
397 | /* |
398 | * Zero any uninitialised parts of the page, and then mark the page | |
399 | * as up to date if it turns out that we're extending the file. | |
400 | */ | |
401 | if (!PageUptodate(page)) { | |
402 | unsigned pglen = nfs_page_length(page); | |
403 | unsigned end = offset + len; | |
404 | ||
405 | if (pglen == 0) { | |
406 | zero_user_segments(page, 0, offset, | |
407 | end, PAGE_CACHE_SIZE); | |
408 | SetPageUptodate(page); | |
409 | } else if (end >= pglen) { | |
410 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
411 | if (offset == 0) | |
412 | SetPageUptodate(page); | |
413 | } else | |
414 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
415 | } | |
416 | ||
4899f9c8 | 417 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
418 | |
419 | unlock_page(page); | |
420 | page_cache_release(page); | |
421 | ||
3d509e54 CL |
422 | if (status < 0) |
423 | return status; | |
2701d086 | 424 | NFS_I(mapping->host)->write_io += copied; |
3d509e54 | 425 | return copied; |
1da177e4 LT |
426 | } |
427 | ||
6b9b3514 DH |
428 | /* |
429 | * Partially or wholly invalidate a page | |
430 | * - Release the private state associated with a page if undergoing complete | |
431 | * page invalidation | |
545db45f | 432 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
433 | * - Caller holds page lock |
434 | */ | |
2ff28e22 | 435 | static void nfs_invalidate_page(struct page *page, unsigned long offset) |
cd52ed35 | 436 | { |
b7eaefaa CL |
437 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset); |
438 | ||
1c75950b TM |
439 | if (offset != 0) |
440 | return; | |
d2ccddf0 | 441 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 442 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
443 | |
444 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
445 | } |
446 | ||
6b9b3514 DH |
447 | /* |
448 | * Attempt to release the private state associated with a page | |
545db45f | 449 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
450 | * - Caller holds page lock |
451 | * - Return true (may release page) or false (may not) | |
452 | */ | |
cd52ed35 TM |
453 | static int nfs_release_page(struct page *page, gfp_t gfp) |
454 | { | |
b608b283 TM |
455 | struct address_space *mapping = page->mapping; |
456 | ||
b7eaefaa CL |
457 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
458 | ||
5cf02d09 JL |
459 | /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not |
460 | * doing this memory reclaim for a fs-related allocation. | |
461 | */ | |
462 | if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL && | |
463 | !(current->flags & PF_FSTRANS)) { | |
b608b283 TM |
464 | int how = FLUSH_SYNC; |
465 | ||
466 | /* Don't let kswapd deadlock waiting for OOM RPC calls */ | |
467 | if (current_is_kswapd()) | |
468 | how = 0; | |
469 | nfs_commit_inode(mapping->host, how); | |
470 | } | |
e3db7691 | 471 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
472 | if (PagePrivate(page)) |
473 | return 0; | |
474 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
475 | } |
476 | ||
6b9b3514 DH |
477 | /* |
478 | * Attempt to clear the private state associated with a page when an error | |
479 | * occurs that requires the cached contents of an inode to be written back or | |
480 | * destroyed | |
545db45f | 481 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
482 | * - Caller holds page lock |
483 | * - Return 0 if successful, -error otherwise | |
484 | */ | |
e3db7691 TM |
485 | static int nfs_launder_page(struct page *page) |
486 | { | |
d56b4ddf | 487 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 488 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
489 | |
490 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
491 | inode->i_ino, (long long)page_offset(page)); | |
492 | ||
545db45f | 493 | nfs_fscache_wait_on_page_write(nfsi, page); |
b7eaefaa | 494 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
495 | } |
496 | ||
a564b8f0 MG |
497 | #ifdef CONFIG_NFS_SWAP |
498 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
499 | sector_t *span) | |
500 | { | |
501 | *span = sis->pages; | |
502 | return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1); | |
503 | } | |
504 | ||
505 | static void nfs_swap_deactivate(struct file *file) | |
506 | { | |
507 | xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0); | |
508 | } | |
509 | #endif | |
510 | ||
f5e54d6e | 511 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
512 | .readpage = nfs_readpage, |
513 | .readpages = nfs_readpages, | |
9cccef95 | 514 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
515 | .writepage = nfs_writepage, |
516 | .writepages = nfs_writepages, | |
4899f9c8 NP |
517 | .write_begin = nfs_write_begin, |
518 | .write_end = nfs_write_end, | |
cd52ed35 TM |
519 | .invalidatepage = nfs_invalidate_page, |
520 | .releasepage = nfs_release_page, | |
1da177e4 | 521 | .direct_IO = nfs_direct_IO, |
074cc1de | 522 | .migratepage = nfs_migrate_page, |
e3db7691 | 523 | .launder_page = nfs_launder_page, |
f590f333 | 524 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
525 | #ifdef CONFIG_NFS_SWAP |
526 | .swap_activate = nfs_swap_activate, | |
527 | .swap_deactivate = nfs_swap_deactivate, | |
528 | #endif | |
1da177e4 LT |
529 | }; |
530 | ||
6b9b3514 DH |
531 | /* |
532 | * Notification that a PTE pointing to an NFS page is about to be made | |
533 | * writable, implying that someone is about to modify the page through a | |
534 | * shared-writable mapping | |
535 | */ | |
c2ec175c | 536 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 537 | { |
c2ec175c | 538 | struct page *page = vmf->page; |
94387fb1 | 539 | struct file *filp = vma->vm_file; |
b7eaefaa | 540 | struct dentry *dentry = filp->f_path.dentry; |
94387fb1 | 541 | unsigned pagelen; |
bc4866b6 | 542 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 543 | struct address_space *mapping; |
94387fb1 | 544 | |
b7eaefaa CL |
545 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n", |
546 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
547 | filp->f_mapping->host->i_ino, | |
548 | (long long)page_offset(page)); | |
549 | ||
545db45f DH |
550 | /* make sure the cache has finished storing the page */ |
551 | nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page); | |
552 | ||
94387fb1 | 553 | lock_page(page); |
d56b4ddf | 554 | mapping = page_file_mapping(page); |
b7eaefaa | 555 | if (mapping != dentry->d_inode->i_mapping) |
8b1f9ee5 TM |
556 | goto out_unlock; |
557 | ||
2aeb98f4 TM |
558 | wait_on_page_writeback(page); |
559 | ||
94387fb1 | 560 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
561 | if (pagelen == 0) |
562 | goto out_unlock; | |
4899f9c8 | 563 | |
bc4866b6 TM |
564 | ret = VM_FAULT_LOCKED; |
565 | if (nfs_flush_incompatible(filp, page) == 0 && | |
566 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
567 | goto out; | |
8b1f9ee5 | 568 | |
bc4866b6 | 569 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
570 | out_unlock: |
571 | unlock_page(page); | |
bc4866b6 TM |
572 | out: |
573 | return ret; | |
94387fb1 TM |
574 | } |
575 | ||
f0f37e2f | 576 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 TM |
577 | .fault = filemap_fault, |
578 | .page_mkwrite = nfs_vm_page_mkwrite, | |
579 | }; | |
580 | ||
7b159fc1 TM |
581 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
582 | { | |
583 | struct nfs_open_context *ctx; | |
584 | ||
6b2f3d1f | 585 | if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) |
7b159fc1 | 586 | return 1; |
cd3758e3 | 587 | ctx = nfs_file_open_context(filp); |
7b159fc1 TM |
588 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) |
589 | return 1; | |
590 | return 0; | |
591 | } | |
592 | ||
ce4ef7c0 BS |
593 | ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, |
594 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 595 | { |
01cce933 | 596 | struct dentry * dentry = iocb->ki_filp->f_path.dentry; |
1da177e4 | 597 | struct inode * inode = dentry->d_inode; |
7e381172 | 598 | unsigned long written = 0; |
1da177e4 | 599 | ssize_t result; |
027445c3 | 600 | size_t count = iov_length(iov, nr_segs); |
1da177e4 | 601 | |
1da177e4 | 602 | if (iocb->ki_filp->f_flags & O_DIRECT) |
a564b8f0 | 603 | return nfs_file_direct_write(iocb, iov, nr_segs, pos, true); |
1da177e4 | 604 | |
6da24bc9 | 605 | dprintk("NFS: write(%s/%s, %lu@%Ld)\n", |
1da177e4 | 606 | dentry->d_parent->d_name.name, dentry->d_name.name, |
6da24bc9 | 607 | (unsigned long) count, (long long) pos); |
1da177e4 LT |
608 | |
609 | result = -EBUSY; | |
610 | if (IS_SWAPFILE(inode)) | |
611 | goto out_swapfile; | |
7d52e862 TM |
612 | /* |
613 | * O_APPEND implies that we must revalidate the file length. | |
614 | */ | |
615 | if (iocb->ki_filp->f_flags & O_APPEND) { | |
616 | result = nfs_revalidate_file_size(inode, iocb->ki_filp); | |
617 | if (result) | |
618 | goto out; | |
fe51beec | 619 | } |
1da177e4 LT |
620 | |
621 | result = count; | |
622 | if (!count) | |
623 | goto out; | |
624 | ||
027445c3 | 625 | result = generic_file_aio_write(iocb, iov, nr_segs, pos); |
7e381172 CL |
626 | if (result > 0) |
627 | written = result; | |
628 | ||
6b2f3d1f | 629 | /* Return error values for O_DSYNC and IS_SYNC() */ |
7b159fc1 | 630 | if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { |
af7fa165 | 631 | int err = vfs_fsync(iocb->ki_filp, 0); |
200baa21 TM |
632 | if (err < 0) |
633 | result = err; | |
634 | } | |
7e381172 CL |
635 | if (result > 0) |
636 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
1da177e4 LT |
637 | out: |
638 | return result; | |
639 | ||
640 | out_swapfile: | |
641 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
642 | goto out; | |
643 | } | |
89d77c8f | 644 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 645 | |
ce4ef7c0 BS |
646 | ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, |
647 | struct file *filp, loff_t *ppos, | |
648 | size_t count, unsigned int flags) | |
bf40d343 SJ |
649 | { |
650 | struct dentry *dentry = filp->f_path.dentry; | |
651 | struct inode *inode = dentry->d_inode; | |
7e381172 | 652 | unsigned long written = 0; |
bf40d343 SJ |
653 | ssize_t ret; |
654 | ||
655 | dprintk("NFS splice_write(%s/%s, %lu@%llu)\n", | |
656 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
657 | (unsigned long) count, (unsigned long long) *ppos); | |
658 | ||
659 | /* | |
660 | * The combination of splice and an O_APPEND destination is disallowed. | |
661 | */ | |
662 | ||
bf40d343 | 663 | ret = generic_file_splice_write(pipe, filp, ppos, count, flags); |
7e381172 CL |
664 | if (ret > 0) |
665 | written = ret; | |
666 | ||
bf40d343 | 667 | if (ret >= 0 && nfs_need_sync_write(filp, inode)) { |
af7fa165 | 668 | int err = vfs_fsync(filp, 0); |
bf40d343 SJ |
669 | if (err < 0) |
670 | ret = err; | |
671 | } | |
7e381172 CL |
672 | if (ret > 0) |
673 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
bf40d343 SJ |
674 | return ret; |
675 | } | |
89d77c8f | 676 | EXPORT_SYMBOL_GPL(nfs_file_splice_write); |
bf40d343 | 677 | |
5eebde23 SJ |
678 | static int |
679 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
680 | { |
681 | struct inode *inode = filp->f_mapping->host; | |
682 | int status = 0; | |
21ac19d4 | 683 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 684 | |
039c4d7a | 685 | /* Try local locking first */ |
6d34ac19 BF |
686 | posix_test_lock(filp, fl); |
687 | if (fl->fl_type != F_UNLCK) { | |
688 | /* found a conflict */ | |
039c4d7a | 689 | goto out; |
1da177e4 | 690 | } |
21ac19d4 | 691 | fl->fl_type = saved_type; |
039c4d7a | 692 | |
011e2a7f | 693 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
694 | goto out_noconflict; |
695 | ||
5eebde23 | 696 | if (is_local) |
039c4d7a TM |
697 | goto out_noconflict; |
698 | ||
699 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
700 | out: | |
1da177e4 | 701 | return status; |
039c4d7a TM |
702 | out_noconflict: |
703 | fl->fl_type = F_UNLCK; | |
704 | goto out; | |
1da177e4 LT |
705 | } |
706 | ||
707 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
708 | { | |
709 | int res = 0; | |
710 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
711 | case FL_POSIX: | |
712 | res = posix_lock_file_wait(file, fl); | |
713 | break; | |
714 | case FL_FLOCK: | |
715 | res = flock_lock_file_wait(file, fl); | |
716 | break; | |
717 | default: | |
718 | BUG(); | |
719 | } | |
1da177e4 LT |
720 | return res; |
721 | } | |
722 | ||
5eebde23 SJ |
723 | static int |
724 | do_unlk(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 | 733 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 LT |
734 | |
735 | /* NOTE: special case | |
736 | * If we're signalled while cleaning up locks on process exit, we | |
737 | * still need to complete the unlock. | |
738 | */ | |
5eebde23 SJ |
739 | /* |
740 | * Use local locking if mounted with "-onolock" or with appropriate | |
741 | * "-olocal_lock=" | |
742 | */ | |
743 | if (!is_local) | |
1da177e4 LT |
744 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
745 | else | |
746 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
747 | return status; |
748 | } | |
749 | ||
6b96724e RL |
750 | static int |
751 | is_time_granular(struct timespec *ts) { | |
752 | return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000)); | |
753 | } | |
754 | ||
5eebde23 SJ |
755 | static int |
756 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
757 | { |
758 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
759 | int status; |
760 | ||
1da177e4 LT |
761 | /* |
762 | * Flush all pending writes before doing anything | |
763 | * with locks.. | |
764 | */ | |
29884df0 TM |
765 | status = nfs_sync_mapping(filp->f_mapping); |
766 | if (status != 0) | |
1da177e4 LT |
767 | goto out; |
768 | ||
5eebde23 SJ |
769 | /* |
770 | * Use local locking if mounted with "-onolock" or with appropriate | |
771 | * "-olocal_lock=" | |
772 | */ | |
773 | if (!is_local) | |
1da177e4 | 774 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 775 | else |
1da177e4 | 776 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
777 | if (status < 0) |
778 | goto out; | |
6b96724e | 779 | |
1da177e4 | 780 | /* |
6b96724e RL |
781 | * Revalidate the cache if the server has time stamps granular |
782 | * enough to detect subsecond changes. Otherwise, clear the | |
783 | * cache to prevent missing any changes. | |
784 | * | |
1da177e4 LT |
785 | * This makes locking act as a cache coherency point. |
786 | */ | |
29884df0 | 787 | nfs_sync_mapping(filp->f_mapping); |
011e2a7f | 788 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { |
6b96724e RL |
789 | if (is_time_granular(&NFS_SERVER(inode)->time_delta)) |
790 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
791 | else | |
792 | nfs_zap_caches(inode); | |
793 | } | |
1da177e4 | 794 | out: |
1da177e4 LT |
795 | return status; |
796 | } | |
797 | ||
798 | /* | |
799 | * Lock a (portion of) a file | |
800 | */ | |
ce4ef7c0 | 801 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 802 | { |
6da24bc9 | 803 | struct inode *inode = filp->f_mapping->host; |
2116271a | 804 | int ret = -ENOLCK; |
5eebde23 | 805 | int is_local = 0; |
1da177e4 | 806 | |
6da24bc9 CL |
807 | dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n", |
808 | filp->f_path.dentry->d_parent->d_name.name, | |
809 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
810 | fl->fl_type, fl->fl_flags, |
811 | (long long)fl->fl_start, (long long)fl->fl_end); | |
6da24bc9 | 812 | |
91d5b470 | 813 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
814 | |
815 | /* No mandatory locks over NFS */ | |
dfad9441 | 816 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
817 | goto out_err; |
818 | ||
5eebde23 SJ |
819 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
820 | is_local = 1; | |
821 | ||
2116271a TM |
822 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
823 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
824 | if (ret < 0) | |
825 | goto out_err; | |
826 | } | |
1da177e4 LT |
827 | |
828 | if (IS_GETLK(cmd)) | |
5eebde23 | 829 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 830 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 831 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 832 | else |
5eebde23 | 833 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
834 | out_err: |
835 | return ret; | |
1da177e4 | 836 | } |
89d77c8f | 837 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
838 | |
839 | /* | |
840 | * Lock a (portion of) a file | |
841 | */ | |
ce4ef7c0 | 842 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 843 | { |
5eebde23 SJ |
844 | struct inode *inode = filp->f_mapping->host; |
845 | int is_local = 0; | |
846 | ||
6da24bc9 CL |
847 | dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n", |
848 | filp->f_path.dentry->d_parent->d_name.name, | |
849 | filp->f_path.dentry->d_name.name, | |
1da177e4 LT |
850 | fl->fl_type, fl->fl_flags); |
851 | ||
1da177e4 LT |
852 | if (!(fl->fl_flags & FL_FLOCK)) |
853 | return -ENOLCK; | |
854 | ||
ad0fcd4e JL |
855 | /* |
856 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
857 | * any standard. In principle we might be able to support LOCK_MAND | |
858 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
859 | * NFS code is not set up for it. | |
860 | */ | |
861 | if (fl->fl_type & LOCK_MAND) | |
862 | return -EINVAL; | |
863 | ||
5eebde23 SJ |
864 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
865 | is_local = 1; | |
866 | ||
1da177e4 LT |
867 | /* We're simulating flock() locks using posix locks on the server */ |
868 | fl->fl_owner = (fl_owner_t)filp; | |
869 | fl->fl_start = 0; | |
870 | fl->fl_end = OFFSET_MAX; | |
871 | ||
872 | if (fl->fl_type == F_UNLCK) | |
5eebde23 SJ |
873 | return do_unlk(filp, cmd, fl, is_local); |
874 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 875 | } |
89d77c8f | 876 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 877 | |
6da24bc9 CL |
878 | /* |
879 | * There is no protocol support for leases, so we have no way to implement | |
880 | * them correctly in the face of opens by other clients. | |
881 | */ | |
ce4ef7c0 | 882 | int nfs_setlease(struct file *file, long arg, struct file_lock **fl) |
370f6599 | 883 | { |
6da24bc9 CL |
884 | dprintk("NFS: setlease(%s/%s, arg=%ld)\n", |
885 | file->f_path.dentry->d_parent->d_name.name, | |
886 | file->f_path.dentry->d_name.name, arg); | |
370f6599 BF |
887 | return -EINVAL; |
888 | } | |
89d77c8f | 889 | EXPORT_SYMBOL_GPL(nfs_setlease); |
1788ea6e | 890 | |
0486958f JL |
891 | const struct file_operations nfs_file_operations = { |
892 | .llseek = nfs_file_llseek, | |
893 | .read = do_sync_read, | |
894 | .write = do_sync_write, | |
895 | .aio_read = nfs_file_read, | |
896 | .aio_write = nfs_file_write, | |
897 | .mmap = nfs_file_mmap, | |
898 | .open = nfs_file_open, | |
899 | .flush = nfs_file_flush, | |
900 | .release = nfs_file_release, | |
901 | .fsync = nfs_file_fsync, | |
902 | .lock = nfs_lock, | |
903 | .flock = nfs_flock, | |
904 | .splice_read = nfs_file_splice_read, | |
905 | .splice_write = nfs_file_splice_write, | |
906 | .check_flags = nfs_check_flags, | |
907 | .setlease = nfs_setlease, | |
908 | }; | |
ddda8e0a | 909 | EXPORT_SYMBOL_GPL(nfs_file_operations); |