Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/nfs/file.c | |
4 | * | |
5 | * Copyright (C) 1992 Rick Sladkey | |
6 | * | |
7 | * Changes Copyright (C) 1994 by Florian La Roche | |
8 | * - Do not copy data too often around in the kernel. | |
9 | * - In nfs_file_read the return value of kmalloc wasn't checked. | |
10 | * - Put in a better version of read look-ahead buffering. Original idea | |
11 | * and implementation by Wai S Kok elekokws@ee.nus.sg. | |
12 | * | |
13 | * Expire cache on write to a file by Wai S Kok (Oct 1994). | |
14 | * | |
15 | * Total rewrite of read side for new NFS buffer cache.. Linus. | |
16 | * | |
17 | * nfs regular file handling functions | |
18 | */ | |
19 | ||
ddda8e0a | 20 | #include <linux/module.h> |
1da177e4 LT |
21 | #include <linux/time.h> |
22 | #include <linux/kernel.h> | |
23 | #include <linux/errno.h> | |
24 | #include <linux/fcntl.h> | |
25 | #include <linux/stat.h> | |
26 | #include <linux/nfs_fs.h> | |
27 | #include <linux/nfs_mount.h> | |
28 | #include <linux/mm.h> | |
1da177e4 | 29 | #include <linux/pagemap.h> |
5a0e3ad6 | 30 | #include <linux/gfp.h> |
b608b283 | 31 | #include <linux/swap.h> |
1da177e4 | 32 | |
7c0f6ba6 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | |
35 | #include "delegation.h" | |
94387fb1 | 36 | #include "internal.h" |
91d5b470 | 37 | #include "iostat.h" |
545db45f | 38 | #include "fscache.h" |
612aa983 | 39 | #include "pnfs.h" |
1da177e4 | 40 | |
f4ce1299 TM |
41 | #include "nfstrace.h" |
42 | ||
1da177e4 LT |
43 | #define NFSDBG_FACILITY NFSDBG_FILE |
44 | ||
f0f37e2f | 45 | static const struct vm_operations_struct nfs_file_vm_ops; |
94387fb1 | 46 | |
1da177e4 LT |
47 | /* Hack for future NFS swap support */ |
48 | #ifndef IS_SWAPFILE | |
49 | # define IS_SWAPFILE(inode) (0) | |
50 | #endif | |
51 | ||
ce4ef7c0 | 52 | int nfs_check_flags(int flags) |
1da177e4 LT |
53 | { |
54 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
55 | return -EINVAL; | |
56 | ||
57 | return 0; | |
58 | } | |
89d77c8f | 59 | EXPORT_SYMBOL_GPL(nfs_check_flags); |
1da177e4 LT |
60 | |
61 | /* | |
62 | * Open file | |
63 | */ | |
64 | static int | |
65 | nfs_file_open(struct inode *inode, struct file *filp) | |
66 | { | |
1da177e4 LT |
67 | int res; |
68 | ||
6de1472f | 69 | dprintk("NFS: open file(%pD2)\n", filp); |
cc0dd2d1 | 70 | |
c2459dc4 | 71 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
1da177e4 LT |
72 | res = nfs_check_flags(filp->f_flags); |
73 | if (res) | |
74 | return res; | |
75 | ||
46cb650c | 76 | res = nfs_open(inode, filp); |
1da177e4 LT |
77 | return res; |
78 | } | |
79 | ||
ce4ef7c0 | 80 | int |
1da177e4 LT |
81 | nfs_file_release(struct inode *inode, struct file *filp) |
82 | { | |
6de1472f | 83 | dprintk("NFS: release(%pD2)\n", filp); |
6da24bc9 | 84 | |
91d5b470 | 85 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
aff8d8dc AS |
86 | nfs_file_clear_open_context(filp); |
87 | return 0; | |
1da177e4 | 88 | } |
89d77c8f | 89 | EXPORT_SYMBOL_GPL(nfs_file_release); |
1da177e4 | 90 | |
980802e3 TM |
91 | /** |
92 | * nfs_revalidate_size - Revalidate the file size | |
302fad7b TM |
93 | * @inode: pointer to inode struct |
94 | * @filp: pointer to struct file | |
980802e3 TM |
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); | |
d7cf8dd0 | 105 | |
980802e3 TM |
106 | if (filp->f_flags & O_DIRECT) |
107 | goto force_reval; | |
61540bf6 | 108 | if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE)) |
d7cf8dd0 | 109 | goto force_reval; |
d7cf8dd0 | 110 | return 0; |
980802e3 TM |
111 | force_reval: |
112 | return __nfs_revalidate_inode(server, inode); | |
113 | } | |
114 | ||
965c8e59 | 115 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence) |
980802e3 | 116 | { |
6de1472f AV |
117 | dprintk("NFS: llseek file(%pD2, %lld, %d)\n", |
118 | filp, offset, whence); | |
b84e06c5 | 119 | |
06222e49 | 120 | /* |
965c8e59 | 121 | * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate |
06222e49 JB |
122 | * the cached file length |
123 | */ | |
965c8e59 | 124 | if (whence != SEEK_SET && whence != SEEK_CUR) { |
980802e3 | 125 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 126 | |
980802e3 TM |
127 | int retval = nfs_revalidate_file_size(inode, filp); |
128 | if (retval < 0) | |
129 | return (loff_t)retval; | |
79835a71 | 130 | } |
d5e66348 | 131 | |
965c8e59 | 132 | return generic_file_llseek(filp, offset, whence); |
980802e3 | 133 | } |
89d77c8f | 134 | EXPORT_SYMBOL_GPL(nfs_file_llseek); |
980802e3 | 135 | |
1da177e4 LT |
136 | /* |
137 | * Flush all dirty pages, and check for write errors. | |
1da177e4 | 138 | */ |
5445b1fb | 139 | static int |
75e1fcc0 | 140 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 141 | { |
6de1472f | 142 | struct inode *inode = file_inode(file); |
1da177e4 | 143 | |
6de1472f | 144 | dprintk("NFS: flush(%pD2)\n", file); |
1da177e4 | 145 | |
c2459dc4 | 146 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
147 | if ((file->f_mode & FMODE_WRITE) == 0) |
148 | return 0; | |
7b159fc1 | 149 | |
7fe5c398 | 150 | /* Flush writes to the server and return any errors */ |
aded8d7b | 151 | return nfs_wb_all(inode); |
1da177e4 LT |
152 | } |
153 | ||
ce4ef7c0 | 154 | ssize_t |
3aa2d199 | 155 | nfs_file_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 156 | { |
6de1472f | 157 | struct inode *inode = file_inode(iocb->ki_filp); |
1da177e4 LT |
158 | ssize_t result; |
159 | ||
2ba48ce5 | 160 | if (iocb->ki_flags & IOCB_DIRECT) |
c8b8e32d | 161 | return nfs_file_direct_read(iocb, to); |
1da177e4 | 162 | |
619d30b4 | 163 | dprintk("NFS: read(%pD2, %zu@%lu)\n", |
6de1472f | 164 | iocb->ki_filp, |
3aa2d199 | 165 | iov_iter_count(to), (unsigned long) iocb->ki_pos); |
1da177e4 | 166 | |
a5864c99 TM |
167 | nfs_start_io_read(inode); |
168 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); | |
4184dcf2 | 169 | if (!result) { |
3aa2d199 | 170 | result = generic_file_read_iter(iocb, to); |
4184dcf2 CL |
171 | if (result > 0) |
172 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
173 | } | |
a5864c99 | 174 | nfs_end_io_read(inode); |
1da177e4 LT |
175 | return result; |
176 | } | |
89d77c8f | 177 | EXPORT_SYMBOL_GPL(nfs_file_read); |
1da177e4 | 178 | |
ce4ef7c0 | 179 | int |
1da177e4 LT |
180 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
181 | { | |
6de1472f | 182 | struct inode *inode = file_inode(file); |
1da177e4 LT |
183 | int status; |
184 | ||
6de1472f | 185 | dprintk("NFS: mmap(%pD2)\n", file); |
1da177e4 | 186 | |
e1ebfd33 TM |
187 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
188 | * so we call that before revalidating the mapping | |
189 | */ | |
190 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
191 | if (!status) { |
192 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 193 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 194 | } |
1da177e4 LT |
195 | return status; |
196 | } | |
89d77c8f | 197 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
198 | |
199 | /* | |
200 | * Flush any dirty pages for this process, and check for write errors. | |
201 | * The return status from this call provides a reliable indication of | |
202 | * whether any write errors occurred for this process. | |
203 | */ | |
4ff79bc7 | 204 | static int |
bf4b4905 | 205 | nfs_file_fsync_commit(struct file *file, int datasync) |
1da177e4 | 206 | { |
cd3758e3 | 207 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
6de1472f | 208 | struct inode *inode = file_inode(file); |
bf4b4905 | 209 | int do_resend, status; |
af7fa165 TM |
210 | int ret = 0; |
211 | ||
6de1472f | 212 | dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); |
1da177e4 | 213 | |
91d5b470 | 214 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
05990d1b | 215 | do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); |
af7fa165 | 216 | status = nfs_commit_inode(inode, FLUSH_SYNC); |
6fbda89b TM |
217 | if (status == 0) |
218 | status = file_check_and_advance_wb_err(file); | |
05990d1b | 219 | if (status < 0) { |
af7fa165 | 220 | ret = status; |
05990d1b TM |
221 | goto out; |
222 | } | |
223 | do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); | |
224 | if (do_resend) | |
225 | ret = -EAGAIN; | |
226 | out: | |
a5c58892 BS |
227 | return ret; |
228 | } | |
229 | ||
4ff79bc7 | 230 | int |
a5c58892 BS |
231 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
232 | { | |
233 | int ret; | |
496ad9aa | 234 | struct inode *inode = file_inode(file); |
a5c58892 | 235 | |
f4ce1299 TM |
236 | trace_nfs_fsync_enter(inode); |
237 | ||
05990d1b | 238 | do { |
6fbda89b | 239 | ret = file_write_and_wait_range(file, start, end); |
05990d1b TM |
240 | if (ret != 0) |
241 | break; | |
bf4b4905 | 242 | ret = nfs_file_fsync_commit(file, datasync); |
4ff79bc7 CH |
243 | if (!ret) |
244 | ret = pnfs_sync_inode(inode, !!datasync); | |
dcfc4f25 TM |
245 | /* |
246 | * If nfs_file_fsync_commit detected a server reboot, then | |
247 | * resend all dirty pages that might have been covered by | |
248 | * the NFS_CONTEXT_RESEND_WRITES flag | |
249 | */ | |
250 | start = 0; | |
251 | end = LLONG_MAX; | |
05990d1b TM |
252 | } while (ret == -EAGAIN); |
253 | ||
f4ce1299 | 254 | trace_nfs_fsync_exit(inode, ret); |
af7fa165 | 255 | return ret; |
1da177e4 | 256 | } |
4ff79bc7 | 257 | EXPORT_SYMBOL_GPL(nfs_file_fsync); |
1da177e4 | 258 | |
38c73044 PS |
259 | /* |
260 | * Decide whether a read/modify/write cycle may be more efficient | |
261 | * then a modify/write/read cycle when writing to a page in the | |
262 | * page cache. | |
263 | * | |
2cde04e9 KI |
264 | * Some pNFS layout drivers can only read/write at a certain block |
265 | * granularity like all block devices and therefore we must perform | |
266 | * read/modify/write whenever a page hasn't read yet and the data | |
267 | * to be written there is not aligned to a block boundary and/or | |
268 | * smaller than the block size. | |
269 | * | |
38c73044 PS |
270 | * The modify/write/read cycle may occur if a page is read before |
271 | * being completely filled by the writer. In this situation, the | |
272 | * page must be completely written to stable storage on the server | |
273 | * before it can be refilled by reading in the page from the server. | |
274 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
275 | * done. | |
276 | * | |
277 | * It may be more efficient to read the page first if the file is | |
278 | * open for reading in addition to writing, the page is not marked | |
279 | * as Uptodate, it is not dirty or waiting to be committed, | |
280 | * indicating that it was previously allocated and then modified, | |
281 | * that there were valid bytes of data in that range of the file, | |
282 | * and that the new data won't completely replace the old data in | |
283 | * that range of the file. | |
284 | */ | |
2cde04e9 | 285 | static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len) |
38c73044 PS |
286 | { |
287 | unsigned int pglen = nfs_page_length(page); | |
09cbfeaf | 288 | unsigned int offset = pos & (PAGE_SIZE - 1); |
38c73044 PS |
289 | unsigned int end = offset + len; |
290 | ||
2cde04e9 KI |
291 | return !pglen || (end >= pglen && !offset); |
292 | } | |
612aa983 | 293 | |
2cde04e9 KI |
294 | static bool nfs_want_read_modify_write(struct file *file, struct page *page, |
295 | loff_t pos, unsigned int len) | |
296 | { | |
297 | /* | |
298 | * Up-to-date pages, those with ongoing or full-page write | |
299 | * don't need read/modify/write | |
300 | */ | |
301 | if (PageUptodate(page) || PagePrivate(page) || | |
302 | nfs_full_page_write(page, pos, len)) | |
303 | return false; | |
304 | ||
305 | if (pnfs_ld_read_whole_page(file->f_mapping->host)) | |
306 | return true; | |
307 | /* Open for reading too? */ | |
308 | if (file->f_mode & FMODE_READ) | |
309 | return true; | |
310 | return false; | |
38c73044 PS |
311 | } |
312 | ||
1da177e4 | 313 | /* |
4899f9c8 NP |
314 | * This does the "real" work of the write. We must allocate and lock the |
315 | * page to be sent back to the generic routine, which then copies the | |
316 | * data from user space. | |
1da177e4 LT |
317 | * |
318 | * If the writer ends up delaying the write, the writer needs to | |
319 | * increment the page use counts until he is done with the page. | |
320 | */ | |
4899f9c8 NP |
321 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
322 | loff_t pos, unsigned len, unsigned flags, | |
323 | struct page **pagep, void **fsdata) | |
1da177e4 | 324 | { |
4899f9c8 | 325 | int ret; |
09cbfeaf | 326 | pgoff_t index = pos >> PAGE_SHIFT; |
4899f9c8 | 327 | struct page *page; |
38c73044 | 328 | int once_thru = 0; |
4899f9c8 | 329 | |
1e8968c5 | 330 | dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", |
6de1472f | 331 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 332 | |
38c73044 | 333 | start: |
54566b2c | 334 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
335 | if (!page) |
336 | return -ENOMEM; | |
337 | *pagep = page; | |
338 | ||
339 | ret = nfs_flush_incompatible(file, page); | |
340 | if (ret) { | |
341 | unlock_page(page); | |
09cbfeaf | 342 | put_page(page); |
38c73044 PS |
343 | } else if (!once_thru && |
344 | nfs_want_read_modify_write(file, page, pos, len)) { | |
345 | once_thru = 1; | |
346 | ret = nfs_readpage(file, page); | |
09cbfeaf | 347 | put_page(page); |
38c73044 PS |
348 | if (!ret) |
349 | goto start; | |
4899f9c8 NP |
350 | } |
351 | return ret; | |
1da177e4 LT |
352 | } |
353 | ||
4899f9c8 NP |
354 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
355 | loff_t pos, unsigned len, unsigned copied, | |
356 | struct page *page, void *fsdata) | |
1da177e4 | 357 | { |
09cbfeaf | 358 | unsigned offset = pos & (PAGE_SIZE - 1); |
dc24826b | 359 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
4899f9c8 | 360 | int status; |
1da177e4 | 361 | |
1e8968c5 | 362 | dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", |
6de1472f | 363 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 364 | |
efc91ed0 TM |
365 | /* |
366 | * Zero any uninitialised parts of the page, and then mark the page | |
367 | * as up to date if it turns out that we're extending the file. | |
368 | */ | |
369 | if (!PageUptodate(page)) { | |
370 | unsigned pglen = nfs_page_length(page); | |
c0cf3ef5 | 371 | unsigned end = offset + copied; |
efc91ed0 TM |
372 | |
373 | if (pglen == 0) { | |
374 | zero_user_segments(page, 0, offset, | |
09cbfeaf | 375 | end, PAGE_SIZE); |
efc91ed0 TM |
376 | SetPageUptodate(page); |
377 | } else if (end >= pglen) { | |
09cbfeaf | 378 | zero_user_segment(page, end, PAGE_SIZE); |
efc91ed0 TM |
379 | if (offset == 0) |
380 | SetPageUptodate(page); | |
381 | } else | |
09cbfeaf | 382 | zero_user_segment(page, pglen, PAGE_SIZE); |
efc91ed0 TM |
383 | } |
384 | ||
4899f9c8 | 385 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
386 | |
387 | unlock_page(page); | |
09cbfeaf | 388 | put_page(page); |
4899f9c8 | 389 | |
3d509e54 CL |
390 | if (status < 0) |
391 | return status; | |
2701d086 | 392 | NFS_I(mapping->host)->write_io += copied; |
dc24826b | 393 | |
ce52914e | 394 | if (nfs_ctx_key_to_expire(ctx, mapping->host)) { |
dc24826b AA |
395 | status = nfs_wb_all(mapping->host); |
396 | if (status < 0) | |
397 | return status; | |
398 | } | |
399 | ||
3d509e54 | 400 | return copied; |
1da177e4 LT |
401 | } |
402 | ||
6b9b3514 DH |
403 | /* |
404 | * Partially or wholly invalidate a page | |
405 | * - Release the private state associated with a page if undergoing complete | |
406 | * page invalidation | |
545db45f | 407 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
408 | * - Caller holds page lock |
409 | */ | |
d47992f8 LC |
410 | static void nfs_invalidate_page(struct page *page, unsigned int offset, |
411 | unsigned int length) | |
cd52ed35 | 412 | { |
d47992f8 LC |
413 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n", |
414 | page, offset, length); | |
b7eaefaa | 415 | |
09cbfeaf | 416 | if (offset != 0 || length < PAGE_SIZE) |
1c75950b | 417 | return; |
d2ccddf0 | 418 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 419 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
420 | |
421 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
422 | } |
423 | ||
6b9b3514 DH |
424 | /* |
425 | * Attempt to release the private state associated with a page | |
545db45f | 426 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
427 | * - Caller holds page lock |
428 | * - Return true (may release page) or false (may not) | |
429 | */ | |
cd52ed35 TM |
430 | static int nfs_release_page(struct page *page, gfp_t gfp) |
431 | { | |
b7eaefaa CL |
432 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
433 | ||
e3db7691 | 434 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
435 | if (PagePrivate(page)) |
436 | return 0; | |
437 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
438 | } |
439 | ||
f919b196 MG |
440 | static void nfs_check_dirty_writeback(struct page *page, |
441 | bool *dirty, bool *writeback) | |
442 | { | |
443 | struct nfs_inode *nfsi; | |
444 | struct address_space *mapping = page_file_mapping(page); | |
445 | ||
446 | if (!mapping || PageSwapCache(page)) | |
447 | return; | |
448 | ||
449 | /* | |
450 | * Check if an unstable page is currently being committed and | |
451 | * if so, have the VM treat it as if the page is under writeback | |
452 | * so it will not block due to pages that will shortly be freeable. | |
453 | */ | |
454 | nfsi = NFS_I(mapping->host); | |
af7cf057 | 455 | if (atomic_read(&nfsi->commit_info.rpcs_out)) { |
f919b196 MG |
456 | *writeback = true; |
457 | return; | |
458 | } | |
459 | ||
460 | /* | |
461 | * If PagePrivate() is set, then the page is not freeable and as the | |
462 | * inode is not being committed, it's not going to be cleaned in the | |
463 | * near future so treat it as dirty | |
464 | */ | |
465 | if (PagePrivate(page)) | |
466 | *dirty = true; | |
467 | } | |
468 | ||
6b9b3514 DH |
469 | /* |
470 | * Attempt to clear the private state associated with a page when an error | |
471 | * occurs that requires the cached contents of an inode to be written back or | |
472 | * destroyed | |
545db45f | 473 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
474 | * - Caller holds page lock |
475 | * - Return 0 if successful, -error otherwise | |
476 | */ | |
e3db7691 TM |
477 | static int nfs_launder_page(struct page *page) |
478 | { | |
d56b4ddf | 479 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 480 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
481 | |
482 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
483 | inode->i_ino, (long long)page_offset(page)); | |
484 | ||
545db45f | 485 | nfs_fscache_wait_on_page_write(nfsi, page); |
c373fff7 | 486 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
487 | } |
488 | ||
a564b8f0 MG |
489 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, |
490 | sector_t *span) | |
491 | { | |
bd89bc67 MZ |
492 | unsigned long blocks; |
493 | long long isize; | |
dad2b015 | 494 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
bd89bc67 MZ |
495 | struct inode *inode = file->f_mapping->host; |
496 | ||
497 | spin_lock(&inode->i_lock); | |
498 | blocks = inode->i_blocks; | |
499 | isize = inode->i_size; | |
500 | spin_unlock(&inode->i_lock); | |
501 | if (blocks*512 < isize) { | |
502 | pr_warn("swap activate: swapfile has holes\n"); | |
503 | return -EINVAL; | |
504 | } | |
dad2b015 | 505 | |
a564b8f0 | 506 | *span = sis->pages; |
dad2b015 | 507 | |
3c87ef6e | 508 | return rpc_clnt_swap_activate(clnt); |
a564b8f0 MG |
509 | } |
510 | ||
511 | static void nfs_swap_deactivate(struct file *file) | |
512 | { | |
dad2b015 JL |
513 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
514 | ||
3c87ef6e | 515 | rpc_clnt_swap_deactivate(clnt); |
a564b8f0 | 516 | } |
a564b8f0 | 517 | |
f5e54d6e | 518 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
519 | .readpage = nfs_readpage, |
520 | .readpages = nfs_readpages, | |
9cccef95 | 521 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
522 | .writepage = nfs_writepage, |
523 | .writepages = nfs_writepages, | |
4899f9c8 NP |
524 | .write_begin = nfs_write_begin, |
525 | .write_end = nfs_write_end, | |
cd52ed35 TM |
526 | .invalidatepage = nfs_invalidate_page, |
527 | .releasepage = nfs_release_page, | |
1da177e4 | 528 | .direct_IO = nfs_direct_IO, |
f844cd0d | 529 | #ifdef CONFIG_MIGRATION |
074cc1de | 530 | .migratepage = nfs_migrate_page, |
f844cd0d | 531 | #endif |
e3db7691 | 532 | .launder_page = nfs_launder_page, |
f919b196 | 533 | .is_dirty_writeback = nfs_check_dirty_writeback, |
f590f333 | 534 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
535 | .swap_activate = nfs_swap_activate, |
536 | .swap_deactivate = nfs_swap_deactivate, | |
1da177e4 LT |
537 | }; |
538 | ||
6b9b3514 DH |
539 | /* |
540 | * Notification that a PTE pointing to an NFS page is about to be made | |
541 | * writable, implying that someone is about to modify the page through a | |
542 | * shared-writable mapping | |
543 | */ | |
01a36844 | 544 | static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf) |
94387fb1 | 545 | { |
c2ec175c | 546 | struct page *page = vmf->page; |
11bac800 | 547 | struct file *filp = vmf->vma->vm_file; |
6de1472f | 548 | struct inode *inode = file_inode(filp); |
94387fb1 | 549 | unsigned pagelen; |
01a36844 | 550 | vm_fault_t ret = VM_FAULT_NOPAGE; |
4899f9c8 | 551 | struct address_space *mapping; |
94387fb1 | 552 | |
1e8968c5 | 553 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", |
6de1472f | 554 | filp, filp->f_mapping->host->i_ino, |
b7eaefaa CL |
555 | (long long)page_offset(page)); |
556 | ||
9a773e7c TM |
557 | sb_start_pagefault(inode->i_sb); |
558 | ||
545db45f | 559 | /* make sure the cache has finished storing the page */ |
6de1472f | 560 | nfs_fscache_wait_on_page_write(NFS_I(inode), page); |
545db45f | 561 | |
ef070dcb TM |
562 | wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING, |
563 | nfs_wait_bit_killable, TASK_KILLABLE); | |
564 | ||
94387fb1 | 565 | lock_page(page); |
d56b4ddf | 566 | mapping = page_file_mapping(page); |
6de1472f | 567 | if (mapping != inode->i_mapping) |
8b1f9ee5 TM |
568 | goto out_unlock; |
569 | ||
2aeb98f4 TM |
570 | wait_on_page_writeback(page); |
571 | ||
94387fb1 | 572 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
573 | if (pagelen == 0) |
574 | goto out_unlock; | |
4899f9c8 | 575 | |
bc4866b6 TM |
576 | ret = VM_FAULT_LOCKED; |
577 | if (nfs_flush_incompatible(filp, page) == 0 && | |
578 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
579 | goto out; | |
8b1f9ee5 | 580 | |
bc4866b6 | 581 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
582 | out_unlock: |
583 | unlock_page(page); | |
bc4866b6 | 584 | out: |
9a773e7c | 585 | sb_end_pagefault(inode->i_sb); |
bc4866b6 | 586 | return ret; |
94387fb1 TM |
587 | } |
588 | ||
f0f37e2f | 589 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 | 590 | .fault = filemap_fault, |
f1820361 | 591 | .map_pages = filemap_map_pages, |
94387fb1 TM |
592 | .page_mkwrite = nfs_vm_page_mkwrite, |
593 | }; | |
594 | ||
7e94d6c4 | 595 | static int nfs_need_check_write(struct file *filp, struct inode *inode) |
7b159fc1 TM |
596 | { |
597 | struct nfs_open_context *ctx; | |
598 | ||
cd3758e3 | 599 | ctx = nfs_file_open_context(filp); |
6fbda89b | 600 | if (nfs_ctx_key_to_expire(ctx, inode)) |
7b159fc1 TM |
601 | return 1; |
602 | return 0; | |
603 | } | |
604 | ||
edaf4369 | 605 | ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 606 | { |
6de1472f AV |
607 | struct file *file = iocb->ki_filp; |
608 | struct inode *inode = file_inode(file); | |
7e381172 | 609 | unsigned long written = 0; |
1da177e4 LT |
610 | ssize_t result; |
611 | ||
6de1472f | 612 | result = nfs_key_timeout_notify(file, inode); |
dc24826b AA |
613 | if (result) |
614 | return result; | |
615 | ||
89698b24 | 616 | if (iocb->ki_flags & IOCB_DIRECT) |
65a4a1ca | 617 | return nfs_file_direct_write(iocb, from); |
1da177e4 | 618 | |
619d30b4 | 619 | dprintk("NFS: write(%pD2, %zu@%Ld)\n", |
18290650 | 620 | file, iov_iter_count(from), (long long) iocb->ki_pos); |
1da177e4 | 621 | |
1da177e4 LT |
622 | if (IS_SWAPFILE(inode)) |
623 | goto out_swapfile; | |
7d52e862 TM |
624 | /* |
625 | * O_APPEND implies that we must revalidate the file length. | |
626 | */ | |
2ba48ce5 | 627 | if (iocb->ki_flags & IOCB_APPEND) { |
6de1472f | 628 | result = nfs_revalidate_file_size(inode, file); |
7d52e862 TM |
629 | if (result) |
630 | goto out; | |
fe51beec | 631 | } |
6ba80d43 N |
632 | if (iocb->ki_pos > i_size_read(inode)) |
633 | nfs_revalidate_mapping(inode, file->f_mapping); | |
1da177e4 | 634 | |
a5864c99 | 635 | nfs_start_io_write(inode); |
18290650 TM |
636 | result = generic_write_checks(iocb, from); |
637 | if (result > 0) { | |
638 | current->backing_dev_info = inode_to_bdi(inode); | |
639 | result = generic_perform_write(file, from, iocb->ki_pos); | |
640 | current->backing_dev_info = NULL; | |
641 | } | |
a5864c99 | 642 | nfs_end_io_write(inode); |
18290650 | 643 | if (result <= 0) |
1da177e4 LT |
644 | goto out; |
645 | ||
c49edecd | 646 | written = result; |
18290650 | 647 | iocb->ki_pos += written; |
e973b1a5 | 648 | result = generic_write_sync(iocb, written); |
649 | if (result < 0) | |
650 | goto out; | |
7e381172 | 651 | |
7e94d6c4 | 652 | /* Return error values */ |
18290650 | 653 | if (nfs_need_check_write(file, inode)) { |
aded8d7b | 654 | int err = nfs_wb_all(inode); |
200baa21 TM |
655 | if (err < 0) |
656 | result = err; | |
657 | } | |
18290650 | 658 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); |
1da177e4 LT |
659 | out: |
660 | return result; | |
661 | ||
662 | out_swapfile: | |
663 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
89658c4d | 664 | return -ETXTBSY; |
1da177e4 | 665 | } |
89d77c8f | 666 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 667 | |
5eebde23 SJ |
668 | static int |
669 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
670 | { |
671 | struct inode *inode = filp->f_mapping->host; | |
672 | int status = 0; | |
21ac19d4 | 673 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 674 | |
039c4d7a | 675 | /* Try local locking first */ |
6d34ac19 BF |
676 | posix_test_lock(filp, fl); |
677 | if (fl->fl_type != F_UNLCK) { | |
678 | /* found a conflict */ | |
039c4d7a | 679 | goto out; |
1da177e4 | 680 | } |
21ac19d4 | 681 | fl->fl_type = saved_type; |
039c4d7a | 682 | |
011e2a7f | 683 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
684 | goto out_noconflict; |
685 | ||
5eebde23 | 686 | if (is_local) |
039c4d7a TM |
687 | goto out_noconflict; |
688 | ||
689 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
690 | out: | |
1da177e4 | 691 | return status; |
039c4d7a TM |
692 | out_noconflict: |
693 | fl->fl_type = F_UNLCK; | |
694 | goto out; | |
1da177e4 LT |
695 | } |
696 | ||
5eebde23 SJ |
697 | static int |
698 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
699 | { |
700 | struct inode *inode = filp->f_mapping->host; | |
7a8203d8 | 701 | struct nfs_lock_context *l_ctx; |
1da177e4 LT |
702 | int status; |
703 | ||
1da177e4 LT |
704 | /* |
705 | * Flush all pending writes before doing anything | |
706 | * with locks.. | |
707 | */ | |
aded8d7b | 708 | nfs_wb_all(inode); |
1da177e4 | 709 | |
7a8203d8 TM |
710 | l_ctx = nfs_get_lock_context(nfs_file_open_context(filp)); |
711 | if (!IS_ERR(l_ctx)) { | |
210c7c17 | 712 | status = nfs_iocounter_wait(l_ctx); |
7a8203d8 | 713 | nfs_put_lock_context(l_ctx); |
f30cb757 BC |
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 | */ | |
718 | if (status < 0 && !(fl->fl_flags & FL_CLOSE)) | |
7a8203d8 TM |
719 | return status; |
720 | } | |
721 | ||
5eebde23 SJ |
722 | /* |
723 | * Use local locking if mounted with "-onolock" or with appropriate | |
724 | * "-olocal_lock=" | |
725 | */ | |
726 | if (!is_local) | |
1da177e4 LT |
727 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
728 | else | |
75575ddf | 729 | status = locks_lock_file_wait(filp, fl); |
1da177e4 LT |
730 | return status; |
731 | } | |
732 | ||
5eebde23 SJ |
733 | static int |
734 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
735 | { |
736 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
737 | int status; |
738 | ||
1da177e4 LT |
739 | /* |
740 | * Flush all pending writes before doing anything | |
741 | * with locks.. | |
742 | */ | |
29884df0 TM |
743 | status = nfs_sync_mapping(filp->f_mapping); |
744 | if (status != 0) | |
1da177e4 LT |
745 | goto out; |
746 | ||
5eebde23 SJ |
747 | /* |
748 | * Use local locking if mounted with "-onolock" or with appropriate | |
749 | * "-olocal_lock=" | |
750 | */ | |
751 | if (!is_local) | |
1da177e4 | 752 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 753 | else |
75575ddf | 754 | status = locks_lock_file_wait(filp, fl); |
1da177e4 LT |
755 | if (status < 0) |
756 | goto out; | |
6b96724e | 757 | |
1da177e4 | 758 | /* |
779eafab N |
759 | * Invalidate cache to prevent missing any changes. If |
760 | * the file is mapped, clear the page cache as well so | |
761 | * those mappings will be loaded. | |
6b96724e | 762 | * |
1da177e4 LT |
763 | * This makes locking act as a cache coherency point. |
764 | */ | |
29884df0 | 765 | nfs_sync_mapping(filp->f_mapping); |
779eafab | 766 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { |
442ce049 | 767 | nfs_zap_caches(inode); |
779eafab N |
768 | if (mapping_mapped(filp->f_mapping)) |
769 | nfs_revalidate_mapping(inode, filp->f_mapping); | |
770 | } | |
1da177e4 | 771 | out: |
1da177e4 LT |
772 | return status; |
773 | } | |
774 | ||
775 | /* | |
776 | * Lock a (portion of) a file | |
777 | */ | |
ce4ef7c0 | 778 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 779 | { |
6da24bc9 | 780 | struct inode *inode = filp->f_mapping->host; |
2116271a | 781 | int ret = -ENOLCK; |
5eebde23 | 782 | int is_local = 0; |
1da177e4 | 783 | |
6de1472f AV |
784 | dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n", |
785 | filp, fl->fl_type, fl->fl_flags, | |
1da177e4 | 786 | (long long)fl->fl_start, (long long)fl->fl_end); |
6da24bc9 | 787 | |
91d5b470 | 788 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
789 | |
790 | /* No mandatory locks over NFS */ | |
dfad9441 | 791 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
792 | goto out_err; |
793 | ||
5eebde23 SJ |
794 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
795 | is_local = 1; | |
796 | ||
2116271a TM |
797 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
798 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
799 | if (ret < 0) | |
800 | goto out_err; | |
801 | } | |
1da177e4 LT |
802 | |
803 | if (IS_GETLK(cmd)) | |
5eebde23 | 804 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 805 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 806 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 807 | else |
5eebde23 | 808 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
809 | out_err: |
810 | return ret; | |
1da177e4 | 811 | } |
89d77c8f | 812 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
813 | |
814 | /* | |
815 | * Lock a (portion of) a file | |
816 | */ | |
ce4ef7c0 | 817 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 818 | { |
5eebde23 SJ |
819 | struct inode *inode = filp->f_mapping->host; |
820 | int is_local = 0; | |
821 | ||
6de1472f AV |
822 | dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n", |
823 | filp, fl->fl_type, fl->fl_flags); | |
1da177e4 | 824 | |
1da177e4 LT |
825 | if (!(fl->fl_flags & FL_FLOCK)) |
826 | return -ENOLCK; | |
827 | ||
ad0fcd4e JL |
828 | /* |
829 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
830 | * any standard. In principle we might be able to support LOCK_MAND | |
831 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
832 | * NFS code is not set up for it. | |
833 | */ | |
834 | if (fl->fl_type & LOCK_MAND) | |
835 | return -EINVAL; | |
836 | ||
5eebde23 SJ |
837 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
838 | is_local = 1; | |
839 | ||
fcfa4470 BC |
840 | /* We're simulating flock() locks using posix locks on the server */ |
841 | if (fl->fl_type == F_UNLCK) | |
5eebde23 SJ |
842 | return do_unlk(filp, cmd, fl, is_local); |
843 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 844 | } |
89d77c8f | 845 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 846 | |
0486958f JL |
847 | const struct file_operations nfs_file_operations = { |
848 | .llseek = nfs_file_llseek, | |
3aa2d199 | 849 | .read_iter = nfs_file_read, |
edaf4369 | 850 | .write_iter = nfs_file_write, |
0486958f JL |
851 | .mmap = nfs_file_mmap, |
852 | .open = nfs_file_open, | |
853 | .flush = nfs_file_flush, | |
854 | .release = nfs_file_release, | |
855 | .fsync = nfs_file_fsync, | |
856 | .lock = nfs_lock, | |
857 | .flock = nfs_flock, | |
82c156f8 | 858 | .splice_read = generic_file_splice_read, |
4da54c21 | 859 | .splice_write = iter_file_splice_write, |
0486958f | 860 | .check_flags = nfs_check_flags, |
1c994a09 | 861 | .setlease = simple_nosetlease, |
0486958f | 862 | }; |
ddda8e0a | 863 | EXPORT_SYMBOL_GPL(nfs_file_operations); |