]>
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> |
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" |
612aa983 | 38 | #include "pnfs.h" |
1da177e4 | 39 | |
f4ce1299 TM |
40 | #include "nfstrace.h" |
41 | ||
1da177e4 LT |
42 | #define NFSDBG_FACILITY NFSDBG_FILE |
43 | ||
f0f37e2f | 44 | static const struct vm_operations_struct nfs_file_vm_ops; |
94387fb1 | 45 | |
1da177e4 LT |
46 | /* Hack for future NFS swap support */ |
47 | #ifndef IS_SWAPFILE | |
48 | # define IS_SWAPFILE(inode) (0) | |
49 | #endif | |
50 | ||
ce4ef7c0 | 51 | int nfs_check_flags(int flags) |
1da177e4 LT |
52 | { |
53 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
54 | return -EINVAL; | |
55 | ||
56 | return 0; | |
57 | } | |
89d77c8f | 58 | EXPORT_SYMBOL_GPL(nfs_check_flags); |
1da177e4 LT |
59 | |
60 | /* | |
61 | * Open file | |
62 | */ | |
63 | static int | |
64 | nfs_file_open(struct inode *inode, struct file *filp) | |
65 | { | |
1da177e4 LT |
66 | int res; |
67 | ||
6de1472f | 68 | dprintk("NFS: open file(%pD2)\n", filp); |
cc0dd2d1 | 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 | { | |
6de1472f | 82 | dprintk("NFS: release(%pD2)\n", filp); |
6da24bc9 | 83 | |
91d5b470 | 84 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
aff8d8dc AS |
85 | nfs_file_clear_open_context(filp); |
86 | return 0; | |
1da177e4 | 87 | } |
89d77c8f | 88 | EXPORT_SYMBOL_GPL(nfs_file_release); |
1da177e4 | 89 | |
980802e3 TM |
90 | /** |
91 | * nfs_revalidate_size - Revalidate the file size | |
92 | * @inode - pointer to inode struct | |
93 | * @file - pointer to struct file | |
94 | * | |
95 | * Revalidates the file length. This is basically a wrapper around | |
96 | * nfs_revalidate_inode() that takes into account the fact that we may | |
97 | * have cached writes (in which case we don't care about the server's | |
98 | * idea of what the file length is), or O_DIRECT (in which case we | |
99 | * shouldn't trust the cache). | |
100 | */ | |
101 | static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) | |
102 | { | |
103 | struct nfs_server *server = NFS_SERVER(inode); | |
104 | struct nfs_inode *nfsi = NFS_I(inode); | |
9310b224 TM |
105 | const unsigned long force_reval = NFS_INO_REVAL_PAGECACHE|NFS_INO_REVAL_FORCED; |
106 | unsigned long cache_validity = nfsi->cache_validity; | |
980802e3 | 107 | |
9310b224 TM |
108 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) && |
109 | (cache_validity & force_reval) != force_reval) | |
d7cf8dd0 TM |
110 | goto out_noreval; |
111 | ||
980802e3 TM |
112 | if (filp->f_flags & O_DIRECT) |
113 | goto force_reval; | |
d7cf8dd0 TM |
114 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
115 | goto force_reval; | |
116 | if (nfs_attribute_timeout(inode)) | |
117 | goto force_reval; | |
118 | out_noreval: | |
119 | return 0; | |
980802e3 TM |
120 | force_reval: |
121 | return __nfs_revalidate_inode(server, inode); | |
122 | } | |
123 | ||
965c8e59 | 124 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence) |
980802e3 | 125 | { |
6de1472f AV |
126 | dprintk("NFS: llseek file(%pD2, %lld, %d)\n", |
127 | filp, offset, whence); | |
b84e06c5 | 128 | |
06222e49 | 129 | /* |
965c8e59 | 130 | * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate |
06222e49 JB |
131 | * the cached file length |
132 | */ | |
965c8e59 | 133 | if (whence != SEEK_SET && whence != 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 | |
965c8e59 | 141 | return generic_file_llseek(filp, offset, whence); |
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 | */ |
5445b1fb | 148 | static int |
75e1fcc0 | 149 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 150 | { |
6de1472f | 151 | struct inode *inode = file_inode(file); |
1da177e4 | 152 | |
6de1472f | 153 | dprintk("NFS: flush(%pD2)\n", file); |
1da177e4 | 154 | |
c2459dc4 | 155 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
156 | if ((file->f_mode & FMODE_WRITE) == 0) |
157 | return 0; | |
7b159fc1 | 158 | |
7fe5c398 | 159 | /* Flush writes to the server and return any errors */ |
af7fa165 | 160 | return vfs_fsync(file, 0); |
1da177e4 LT |
161 | } |
162 | ||
ce4ef7c0 | 163 | ssize_t |
3aa2d199 | 164 | nfs_file_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 165 | { |
6de1472f | 166 | struct inode *inode = file_inode(iocb->ki_filp); |
1da177e4 LT |
167 | ssize_t result; |
168 | ||
2ba48ce5 | 169 | if (iocb->ki_flags & IOCB_DIRECT) |
c8b8e32d | 170 | return nfs_file_direct_read(iocb, to); |
1da177e4 | 171 | |
619d30b4 | 172 | dprintk("NFS: read(%pD2, %zu@%lu)\n", |
6de1472f | 173 | iocb->ki_filp, |
3aa2d199 | 174 | iov_iter_count(to), (unsigned long) iocb->ki_pos); |
1da177e4 | 175 | |
a5864c99 TM |
176 | nfs_start_io_read(inode); |
177 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); | |
4184dcf2 | 178 | if (!result) { |
3aa2d199 | 179 | result = generic_file_read_iter(iocb, to); |
4184dcf2 CL |
180 | if (result > 0) |
181 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
182 | } | |
a5864c99 | 183 | nfs_end_io_read(inode); |
1da177e4 LT |
184 | return result; |
185 | } | |
89d77c8f | 186 | EXPORT_SYMBOL_GPL(nfs_file_read); |
1da177e4 | 187 | |
ce4ef7c0 | 188 | int |
1da177e4 LT |
189 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
190 | { | |
6de1472f | 191 | struct inode *inode = file_inode(file); |
1da177e4 LT |
192 | int status; |
193 | ||
6de1472f | 194 | dprintk("NFS: mmap(%pD2)\n", file); |
1da177e4 | 195 | |
e1ebfd33 TM |
196 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
197 | * so we call that before revalidating the mapping | |
198 | */ | |
199 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
200 | if (!status) { |
201 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 202 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 203 | } |
1da177e4 LT |
204 | return status; |
205 | } | |
89d77c8f | 206 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
207 | |
208 | /* | |
209 | * Flush any dirty pages for this process, and check for write errors. | |
210 | * The return status from this call provides a reliable indication of | |
211 | * whether any write errors occurred for this process. | |
af7fa165 TM |
212 | * |
213 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
214 | * disk, but it retrieves and clears ctx->error after synching, despite | |
215 | * the two being set at the same time in nfs_context_set_write_error(). | |
216 | * This is because the former is used to notify the _next_ call to | |
25985edc | 217 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 218 | * fall back to doing a synchronous write. |
1da177e4 | 219 | */ |
4ff79bc7 | 220 | static int |
a5c58892 | 221 | nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 222 | { |
cd3758e3 | 223 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
6de1472f | 224 | struct inode *inode = file_inode(file); |
05990d1b | 225 | int have_error, do_resend, status; |
af7fa165 TM |
226 | int ret = 0; |
227 | ||
6de1472f | 228 | dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); |
1da177e4 | 229 | |
91d5b470 | 230 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
05990d1b | 231 | do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); |
af7fa165 TM |
232 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
233 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
234 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
05990d1b | 235 | if (have_error) { |
af7fa165 | 236 | ret = xchg(&ctx->error, 0); |
05990d1b TM |
237 | if (ret) |
238 | goto out; | |
239 | } | |
240 | if (status < 0) { | |
af7fa165 | 241 | ret = status; |
05990d1b TM |
242 | goto out; |
243 | } | |
244 | do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); | |
245 | if (do_resend) | |
246 | ret = -EAGAIN; | |
247 | out: | |
a5c58892 BS |
248 | return ret; |
249 | } | |
250 | ||
4ff79bc7 | 251 | int |
a5c58892 BS |
252 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
253 | { | |
254 | int ret; | |
496ad9aa | 255 | struct inode *inode = file_inode(file); |
a5c58892 | 256 | |
f4ce1299 TM |
257 | trace_nfs_fsync_enter(inode); |
258 | ||
05990d1b TM |
259 | do { |
260 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
261 | if (ret != 0) | |
262 | break; | |
05990d1b | 263 | ret = nfs_file_fsync_commit(file, start, end, datasync); |
4ff79bc7 CH |
264 | if (!ret) |
265 | ret = pnfs_sync_inode(inode, !!datasync); | |
dcfc4f25 TM |
266 | /* |
267 | * If nfs_file_fsync_commit detected a server reboot, then | |
268 | * resend all dirty pages that might have been covered by | |
269 | * the NFS_CONTEXT_RESEND_WRITES flag | |
270 | */ | |
271 | start = 0; | |
272 | end = LLONG_MAX; | |
05990d1b TM |
273 | } while (ret == -EAGAIN); |
274 | ||
f4ce1299 | 275 | trace_nfs_fsync_exit(inode, ret); |
af7fa165 | 276 | return ret; |
1da177e4 | 277 | } |
4ff79bc7 | 278 | EXPORT_SYMBOL_GPL(nfs_file_fsync); |
1da177e4 | 279 | |
38c73044 PS |
280 | /* |
281 | * Decide whether a read/modify/write cycle may be more efficient | |
282 | * then a modify/write/read cycle when writing to a page in the | |
283 | * page cache. | |
284 | * | |
285 | * The modify/write/read cycle may occur if a page is read before | |
286 | * being completely filled by the writer. In this situation, the | |
287 | * page must be completely written to stable storage on the server | |
288 | * before it can be refilled by reading in the page from the server. | |
289 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
290 | * done. | |
291 | * | |
292 | * It may be more efficient to read the page first if the file is | |
293 | * open for reading in addition to writing, the page is not marked | |
294 | * as Uptodate, it is not dirty or waiting to be committed, | |
295 | * indicating that it was previously allocated and then modified, | |
296 | * that there were valid bytes of data in that range of the file, | |
297 | * and that the new data won't completely replace the old data in | |
298 | * that range of the file. | |
299 | */ | |
300 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
301 | loff_t pos, unsigned len) | |
302 | { | |
303 | unsigned int pglen = nfs_page_length(page); | |
09cbfeaf | 304 | unsigned int offset = pos & (PAGE_SIZE - 1); |
38c73044 PS |
305 | unsigned int end = offset + len; |
306 | ||
612aa983 CH |
307 | if (pnfs_ld_read_whole_page(file->f_mapping->host)) { |
308 | if (!PageUptodate(page)) | |
309 | return 1; | |
310 | return 0; | |
311 | } | |
312 | ||
38c73044 PS |
313 | if ((file->f_mode & FMODE_READ) && /* open for read? */ |
314 | !PageUptodate(page) && /* Uptodate? */ | |
315 | !PagePrivate(page) && /* i/o request already? */ | |
316 | pglen && /* valid bytes of file? */ | |
317 | (end < pglen || offset)) /* replace all valid bytes? */ | |
318 | return 1; | |
319 | return 0; | |
320 | } | |
321 | ||
1da177e4 | 322 | /* |
4899f9c8 NP |
323 | * This does the "real" work of the write. We must allocate and lock the |
324 | * page to be sent back to the generic routine, which then copies the | |
325 | * data from user space. | |
1da177e4 LT |
326 | * |
327 | * If the writer ends up delaying the write, the writer needs to | |
328 | * increment the page use counts until he is done with the page. | |
329 | */ | |
4899f9c8 NP |
330 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
331 | loff_t pos, unsigned len, unsigned flags, | |
332 | struct page **pagep, void **fsdata) | |
1da177e4 | 333 | { |
4899f9c8 | 334 | int ret; |
09cbfeaf | 335 | pgoff_t index = pos >> PAGE_SHIFT; |
4899f9c8 | 336 | struct page *page; |
38c73044 | 337 | int once_thru = 0; |
4899f9c8 | 338 | |
1e8968c5 | 339 | dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", |
6de1472f | 340 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 341 | |
38c73044 | 342 | start: |
54566b2c | 343 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
344 | if (!page) |
345 | return -ENOMEM; | |
346 | *pagep = page; | |
347 | ||
348 | ret = nfs_flush_incompatible(file, page); | |
349 | if (ret) { | |
350 | unlock_page(page); | |
09cbfeaf | 351 | put_page(page); |
38c73044 PS |
352 | } else if (!once_thru && |
353 | nfs_want_read_modify_write(file, page, pos, len)) { | |
354 | once_thru = 1; | |
355 | ret = nfs_readpage(file, page); | |
09cbfeaf | 356 | put_page(page); |
38c73044 PS |
357 | if (!ret) |
358 | goto start; | |
4899f9c8 NP |
359 | } |
360 | return ret; | |
1da177e4 LT |
361 | } |
362 | ||
4899f9c8 NP |
363 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
364 | loff_t pos, unsigned len, unsigned copied, | |
365 | struct page *page, void *fsdata) | |
1da177e4 | 366 | { |
09cbfeaf | 367 | unsigned offset = pos & (PAGE_SIZE - 1); |
dc24826b | 368 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
4899f9c8 | 369 | int status; |
1da177e4 | 370 | |
1e8968c5 | 371 | dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", |
6de1472f | 372 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 373 | |
efc91ed0 TM |
374 | /* |
375 | * Zero any uninitialised parts of the page, and then mark the page | |
376 | * as up to date if it turns out that we're extending the file. | |
377 | */ | |
378 | if (!PageUptodate(page)) { | |
379 | unsigned pglen = nfs_page_length(page); | |
380 | unsigned end = offset + len; | |
381 | ||
382 | if (pglen == 0) { | |
383 | zero_user_segments(page, 0, offset, | |
09cbfeaf | 384 | end, PAGE_SIZE); |
efc91ed0 TM |
385 | SetPageUptodate(page); |
386 | } else if (end >= pglen) { | |
09cbfeaf | 387 | zero_user_segment(page, end, PAGE_SIZE); |
efc91ed0 TM |
388 | if (offset == 0) |
389 | SetPageUptodate(page); | |
390 | } else | |
09cbfeaf | 391 | zero_user_segment(page, pglen, PAGE_SIZE); |
efc91ed0 TM |
392 | } |
393 | ||
4899f9c8 | 394 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
395 | |
396 | unlock_page(page); | |
09cbfeaf | 397 | put_page(page); |
4899f9c8 | 398 | |
3d509e54 CL |
399 | if (status < 0) |
400 | return status; | |
2701d086 | 401 | NFS_I(mapping->host)->write_io += copied; |
dc24826b | 402 | |
ce52914e | 403 | if (nfs_ctx_key_to_expire(ctx, mapping->host)) { |
dc24826b AA |
404 | status = nfs_wb_all(mapping->host); |
405 | if (status < 0) | |
406 | return status; | |
407 | } | |
408 | ||
3d509e54 | 409 | return copied; |
1da177e4 LT |
410 | } |
411 | ||
6b9b3514 DH |
412 | /* |
413 | * Partially or wholly invalidate a page | |
414 | * - Release the private state associated with a page if undergoing complete | |
415 | * page invalidation | |
545db45f | 416 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
417 | * - Caller holds page lock |
418 | */ | |
d47992f8 LC |
419 | static void nfs_invalidate_page(struct page *page, unsigned int offset, |
420 | unsigned int length) | |
cd52ed35 | 421 | { |
d47992f8 LC |
422 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n", |
423 | page, offset, length); | |
b7eaefaa | 424 | |
09cbfeaf | 425 | if (offset != 0 || length < PAGE_SIZE) |
1c75950b | 426 | return; |
d2ccddf0 | 427 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 428 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
429 | |
430 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
431 | } |
432 | ||
6b9b3514 DH |
433 | /* |
434 | * Attempt to release the private state associated with a page | |
545db45f | 435 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
436 | * - Caller holds page lock |
437 | * - Return true (may release page) or false (may not) | |
438 | */ | |
cd52ed35 TM |
439 | static int nfs_release_page(struct page *page, gfp_t gfp) |
440 | { | |
b7eaefaa CL |
441 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
442 | ||
e3db7691 | 443 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
444 | if (PagePrivate(page)) |
445 | return 0; | |
446 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
447 | } |
448 | ||
f919b196 MG |
449 | static void nfs_check_dirty_writeback(struct page *page, |
450 | bool *dirty, bool *writeback) | |
451 | { | |
452 | struct nfs_inode *nfsi; | |
453 | struct address_space *mapping = page_file_mapping(page); | |
454 | ||
455 | if (!mapping || PageSwapCache(page)) | |
456 | return; | |
457 | ||
458 | /* | |
459 | * Check if an unstable page is currently being committed and | |
460 | * if so, have the VM treat it as if the page is under writeback | |
461 | * so it will not block due to pages that will shortly be freeable. | |
462 | */ | |
463 | nfsi = NFS_I(mapping->host); | |
af7cf057 | 464 | if (atomic_read(&nfsi->commit_info.rpcs_out)) { |
f919b196 MG |
465 | *writeback = true; |
466 | return; | |
467 | } | |
468 | ||
469 | /* | |
470 | * If PagePrivate() is set, then the page is not freeable and as the | |
471 | * inode is not being committed, it's not going to be cleaned in the | |
472 | * near future so treat it as dirty | |
473 | */ | |
474 | if (PagePrivate(page)) | |
475 | *dirty = true; | |
476 | } | |
477 | ||
6b9b3514 DH |
478 | /* |
479 | * Attempt to clear the private state associated with a page when an error | |
480 | * occurs that requires the cached contents of an inode to be written back or | |
481 | * destroyed | |
545db45f | 482 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
483 | * - Caller holds page lock |
484 | * - Return 0 if successful, -error otherwise | |
485 | */ | |
e3db7691 TM |
486 | static int nfs_launder_page(struct page *page) |
487 | { | |
d56b4ddf | 488 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 489 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
490 | |
491 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
492 | inode->i_ino, (long long)page_offset(page)); | |
493 | ||
545db45f | 494 | nfs_fscache_wait_on_page_write(nfsi, page); |
d6c843b9 | 495 | return nfs_wb_launder_page(inode, page); |
cd52ed35 TM |
496 | } |
497 | ||
a564b8f0 MG |
498 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, |
499 | sector_t *span) | |
500 | { | |
dad2b015 JL |
501 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
502 | ||
a564b8f0 | 503 | *span = sis->pages; |
dad2b015 | 504 | |
3c87ef6e | 505 | return rpc_clnt_swap_activate(clnt); |
a564b8f0 MG |
506 | } |
507 | ||
508 | static void nfs_swap_deactivate(struct file *file) | |
509 | { | |
dad2b015 JL |
510 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
511 | ||
3c87ef6e | 512 | rpc_clnt_swap_deactivate(clnt); |
a564b8f0 | 513 | } |
a564b8f0 | 514 | |
f5e54d6e | 515 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
516 | .readpage = nfs_readpage, |
517 | .readpages = nfs_readpages, | |
9cccef95 | 518 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
519 | .writepage = nfs_writepage, |
520 | .writepages = nfs_writepages, | |
4899f9c8 NP |
521 | .write_begin = nfs_write_begin, |
522 | .write_end = nfs_write_end, | |
cd52ed35 TM |
523 | .invalidatepage = nfs_invalidate_page, |
524 | .releasepage = nfs_release_page, | |
1da177e4 | 525 | .direct_IO = nfs_direct_IO, |
f844cd0d | 526 | #ifdef CONFIG_MIGRATION |
074cc1de | 527 | .migratepage = nfs_migrate_page, |
f844cd0d | 528 | #endif |
e3db7691 | 529 | .launder_page = nfs_launder_page, |
f919b196 | 530 | .is_dirty_writeback = nfs_check_dirty_writeback, |
f590f333 | 531 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
532 | .swap_activate = nfs_swap_activate, |
533 | .swap_deactivate = nfs_swap_deactivate, | |
1da177e4 LT |
534 | }; |
535 | ||
6b9b3514 DH |
536 | /* |
537 | * Notification that a PTE pointing to an NFS page is about to be made | |
538 | * writable, implying that someone is about to modify the page through a | |
539 | * shared-writable mapping | |
540 | */ | |
c2ec175c | 541 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 542 | { |
c2ec175c | 543 | struct page *page = vmf->page; |
94387fb1 | 544 | struct file *filp = vma->vm_file; |
6de1472f | 545 | struct inode *inode = file_inode(filp); |
94387fb1 | 546 | unsigned pagelen; |
bc4866b6 | 547 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 548 | struct address_space *mapping; |
94387fb1 | 549 | |
1e8968c5 | 550 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", |
6de1472f | 551 | filp, filp->f_mapping->host->i_ino, |
b7eaefaa CL |
552 | (long long)page_offset(page)); |
553 | ||
9a773e7c TM |
554 | sb_start_pagefault(inode->i_sb); |
555 | ||
545db45f | 556 | /* make sure the cache has finished storing the page */ |
6de1472f | 557 | nfs_fscache_wait_on_page_write(NFS_I(inode), page); |
545db45f | 558 | |
ef070dcb TM |
559 | wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING, |
560 | nfs_wait_bit_killable, TASK_KILLABLE); | |
561 | ||
94387fb1 | 562 | lock_page(page); |
d56b4ddf | 563 | mapping = page_file_mapping(page); |
6de1472f | 564 | if (mapping != inode->i_mapping) |
8b1f9ee5 TM |
565 | goto out_unlock; |
566 | ||
2aeb98f4 TM |
567 | wait_on_page_writeback(page); |
568 | ||
94387fb1 | 569 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
570 | if (pagelen == 0) |
571 | goto out_unlock; | |
4899f9c8 | 572 | |
bc4866b6 TM |
573 | ret = VM_FAULT_LOCKED; |
574 | if (nfs_flush_incompatible(filp, page) == 0 && | |
575 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
576 | goto out; | |
8b1f9ee5 | 577 | |
bc4866b6 | 578 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
579 | out_unlock: |
580 | unlock_page(page); | |
bc4866b6 | 581 | out: |
9a773e7c | 582 | sb_end_pagefault(inode->i_sb); |
bc4866b6 | 583 | return ret; |
94387fb1 TM |
584 | } |
585 | ||
f0f37e2f | 586 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 | 587 | .fault = filemap_fault, |
f1820361 | 588 | .map_pages = filemap_map_pages, |
94387fb1 TM |
589 | .page_mkwrite = nfs_vm_page_mkwrite, |
590 | }; | |
591 | ||
7e94d6c4 | 592 | static int nfs_need_check_write(struct file *filp, struct inode *inode) |
7b159fc1 TM |
593 | { |
594 | struct nfs_open_context *ctx; | |
595 | ||
cd3758e3 | 596 | ctx = nfs_file_open_context(filp); |
dc24826b | 597 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) || |
ce52914e | 598 | nfs_ctx_key_to_expire(ctx, inode)) |
7b159fc1 TM |
599 | return 1; |
600 | return 0; | |
601 | } | |
602 | ||
edaf4369 | 603 | ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 604 | { |
6de1472f AV |
605 | struct file *file = iocb->ki_filp; |
606 | struct inode *inode = file_inode(file); | |
7e381172 | 607 | unsigned long written = 0; |
1da177e4 LT |
608 | ssize_t result; |
609 | ||
6de1472f | 610 | result = nfs_key_timeout_notify(file, inode); |
dc24826b AA |
611 | if (result) |
612 | return result; | |
613 | ||
89698b24 | 614 | if (iocb->ki_flags & IOCB_DIRECT) |
65a4a1ca | 615 | return nfs_file_direct_write(iocb, from); |
1da177e4 | 616 | |
619d30b4 | 617 | dprintk("NFS: write(%pD2, %zu@%Ld)\n", |
18290650 | 618 | file, iov_iter_count(from), (long long) iocb->ki_pos); |
1da177e4 | 619 | |
1da177e4 LT |
620 | if (IS_SWAPFILE(inode)) |
621 | goto out_swapfile; | |
7d52e862 TM |
622 | /* |
623 | * O_APPEND implies that we must revalidate the file length. | |
624 | */ | |
2ba48ce5 | 625 | if (iocb->ki_flags & IOCB_APPEND) { |
6de1472f | 626 | result = nfs_revalidate_file_size(inode, file); |
7d52e862 TM |
627 | if (result) |
628 | goto out; | |
fe51beec | 629 | } |
1da177e4 | 630 | |
a5864c99 | 631 | nfs_start_io_write(inode); |
18290650 TM |
632 | result = generic_write_checks(iocb, from); |
633 | if (result > 0) { | |
634 | current->backing_dev_info = inode_to_bdi(inode); | |
635 | result = generic_perform_write(file, from, iocb->ki_pos); | |
636 | current->backing_dev_info = NULL; | |
637 | } | |
a5864c99 | 638 | nfs_end_io_write(inode); |
18290650 | 639 | if (result <= 0) |
1da177e4 LT |
640 | goto out; |
641 | ||
c49edecd TM |
642 | result = generic_write_sync(iocb, result); |
643 | if (result < 0) | |
644 | goto out; | |
645 | written = result; | |
18290650 | 646 | iocb->ki_pos += written; |
7e381172 | 647 | |
7e94d6c4 | 648 | /* Return error values */ |
18290650 | 649 | if (nfs_need_check_write(file, inode)) { |
6de1472f | 650 | int err = vfs_fsync(file, 0); |
200baa21 TM |
651 | if (err < 0) |
652 | result = err; | |
653 | } | |
18290650 | 654 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); |
1da177e4 LT |
655 | out: |
656 | return result; | |
657 | ||
658 | out_swapfile: | |
659 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
18290650 | 660 | return -EBUSY; |
1da177e4 | 661 | } |
89d77c8f | 662 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 663 | |
5eebde23 SJ |
664 | static int |
665 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
666 | { |
667 | struct inode *inode = filp->f_mapping->host; | |
668 | int status = 0; | |
21ac19d4 | 669 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 670 | |
039c4d7a | 671 | /* Try local locking first */ |
6d34ac19 BF |
672 | posix_test_lock(filp, fl); |
673 | if (fl->fl_type != F_UNLCK) { | |
674 | /* found a conflict */ | |
039c4d7a | 675 | goto out; |
1da177e4 | 676 | } |
21ac19d4 | 677 | fl->fl_type = saved_type; |
039c4d7a | 678 | |
011e2a7f | 679 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
680 | goto out_noconflict; |
681 | ||
5eebde23 | 682 | if (is_local) |
039c4d7a TM |
683 | goto out_noconflict; |
684 | ||
685 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
686 | out: | |
1da177e4 | 687 | return status; |
039c4d7a TM |
688 | out_noconflict: |
689 | fl->fl_type = F_UNLCK; | |
690 | goto out; | |
1da177e4 LT |
691 | } |
692 | ||
5eebde23 SJ |
693 | static int |
694 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
695 | { |
696 | struct inode *inode = filp->f_mapping->host; | |
7a8203d8 | 697 | struct nfs_lock_context *l_ctx; |
1da177e4 LT |
698 | int status; |
699 | ||
1da177e4 LT |
700 | /* |
701 | * Flush all pending writes before doing anything | |
702 | * with locks.. | |
703 | */ | |
d9dabc1a | 704 | vfs_fsync(filp, 0); |
1da177e4 | 705 | |
7a8203d8 TM |
706 | l_ctx = nfs_get_lock_context(nfs_file_open_context(filp)); |
707 | if (!IS_ERR(l_ctx)) { | |
210c7c17 | 708 | status = nfs_iocounter_wait(l_ctx); |
7a8203d8 TM |
709 | nfs_put_lock_context(l_ctx); |
710 | if (status < 0) | |
711 | return status; | |
712 | } | |
713 | ||
1da177e4 LT |
714 | /* NOTE: special case |
715 | * If we're signalled while cleaning up locks on process exit, we | |
716 | * still need to complete the unlock. | |
717 | */ | |
5eebde23 SJ |
718 | /* |
719 | * Use local locking if mounted with "-onolock" or with appropriate | |
720 | * "-olocal_lock=" | |
721 | */ | |
722 | if (!is_local) | |
1da177e4 LT |
723 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
724 | else | |
75575ddf | 725 | status = locks_lock_file_wait(filp, fl); |
1da177e4 LT |
726 | return status; |
727 | } | |
728 | ||
5eebde23 SJ |
729 | static int |
730 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
731 | { |
732 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
733 | int status; |
734 | ||
1da177e4 LT |
735 | /* |
736 | * Flush all pending writes before doing anything | |
737 | * with locks.. | |
738 | */ | |
29884df0 TM |
739 | status = nfs_sync_mapping(filp->f_mapping); |
740 | if (status != 0) | |
1da177e4 LT |
741 | goto out; |
742 | ||
5eebde23 SJ |
743 | /* |
744 | * Use local locking if mounted with "-onolock" or with appropriate | |
745 | * "-olocal_lock=" | |
746 | */ | |
747 | if (!is_local) | |
1da177e4 | 748 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 749 | else |
75575ddf | 750 | status = locks_lock_file_wait(filp, fl); |
1da177e4 LT |
751 | if (status < 0) |
752 | goto out; | |
6b96724e | 753 | |
1da177e4 | 754 | /* |
6b96724e RL |
755 | * Revalidate the cache if the server has time stamps granular |
756 | * enough to detect subsecond changes. Otherwise, clear the | |
757 | * cache to prevent missing any changes. | |
758 | * | |
1da177e4 LT |
759 | * This makes locking act as a cache coherency point. |
760 | */ | |
29884df0 | 761 | nfs_sync_mapping(filp->f_mapping); |
ca0daa27 TM |
762 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
763 | nfs_zap_mapping(inode, filp->f_mapping); | |
1da177e4 | 764 | out: |
1da177e4 LT |
765 | return status; |
766 | } | |
767 | ||
768 | /* | |
769 | * Lock a (portion of) a file | |
770 | */ | |
ce4ef7c0 | 771 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 772 | { |
6da24bc9 | 773 | struct inode *inode = filp->f_mapping->host; |
2116271a | 774 | int ret = -ENOLCK; |
5eebde23 | 775 | int is_local = 0; |
1da177e4 | 776 | |
6de1472f AV |
777 | dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n", |
778 | filp, fl->fl_type, fl->fl_flags, | |
1da177e4 | 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 | 804 | } |
89d77c8f | 805 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
806 | |
807 | /* | |
808 | * Lock a (portion of) a file | |
809 | */ | |
ce4ef7c0 | 810 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 811 | { |
5eebde23 SJ |
812 | struct inode *inode = filp->f_mapping->host; |
813 | int is_local = 0; | |
814 | ||
6de1472f AV |
815 | dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n", |
816 | filp, fl->fl_type, fl->fl_flags); | |
1da177e4 | 817 | |
1da177e4 LT |
818 | if (!(fl->fl_flags & FL_FLOCK)) |
819 | return -ENOLCK; | |
820 | ||
ad0fcd4e JL |
821 | /* |
822 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
823 | * any standard. In principle we might be able to support LOCK_MAND | |
824 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
825 | * NFS code is not set up for it. | |
826 | */ | |
827 | if (fl->fl_type & LOCK_MAND) | |
828 | return -EINVAL; | |
829 | ||
5eebde23 SJ |
830 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
831 | is_local = 1; | |
832 | ||
1da177e4 | 833 | /* We're simulating flock() locks using posix locks on the server */ |
1da177e4 | 834 | if (fl->fl_type == F_UNLCK) |
5eebde23 SJ |
835 | return do_unlk(filp, cmd, fl, is_local); |
836 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 837 | } |
89d77c8f | 838 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 839 | |
0486958f JL |
840 | const struct file_operations nfs_file_operations = { |
841 | .llseek = nfs_file_llseek, | |
3aa2d199 | 842 | .read_iter = nfs_file_read, |
edaf4369 | 843 | .write_iter = nfs_file_write, |
0486958f JL |
844 | .mmap = nfs_file_mmap, |
845 | .open = nfs_file_open, | |
846 | .flush = nfs_file_flush, | |
847 | .release = nfs_file_release, | |
848 | .fsync = nfs_file_fsync, | |
849 | .lock = nfs_lock, | |
850 | .flock = nfs_flock, | |
82c156f8 | 851 | .splice_read = generic_file_splice_read, |
4da54c21 | 852 | .splice_write = iter_file_splice_write, |
0486958f | 853 | .check_flags = nfs_check_flags, |
1c994a09 | 854 | .setlease = simple_nosetlease, |
0486958f | 855 | }; |
ddda8e0a | 856 | EXPORT_SYMBOL_GPL(nfs_file_operations); |