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 | 31 | |
7c0f6ba6 | 32 | #include <linux/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); | |
d7cf8dd0 | 104 | |
980802e3 TM |
105 | if (filp->f_flags & O_DIRECT) |
106 | goto force_reval; | |
61540bf6 | 107 | if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE)) |
d7cf8dd0 | 108 | goto force_reval; |
d7cf8dd0 | 109 | return 0; |
980802e3 TM |
110 | force_reval: |
111 | return __nfs_revalidate_inode(server, inode); | |
112 | } | |
113 | ||
965c8e59 | 114 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence) |
980802e3 | 115 | { |
6de1472f AV |
116 | dprintk("NFS: llseek file(%pD2, %lld, %d)\n", |
117 | filp, offset, whence); | |
b84e06c5 | 118 | |
06222e49 | 119 | /* |
965c8e59 | 120 | * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate |
06222e49 JB |
121 | * the cached file length |
122 | */ | |
965c8e59 | 123 | if (whence != SEEK_SET && whence != SEEK_CUR) { |
980802e3 | 124 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 125 | |
980802e3 TM |
126 | int retval = nfs_revalidate_file_size(inode, filp); |
127 | if (retval < 0) | |
128 | return (loff_t)retval; | |
79835a71 | 129 | } |
d5e66348 | 130 | |
965c8e59 | 131 | return generic_file_llseek(filp, offset, whence); |
980802e3 | 132 | } |
89d77c8f | 133 | EXPORT_SYMBOL_GPL(nfs_file_llseek); |
980802e3 | 134 | |
1da177e4 LT |
135 | /* |
136 | * Flush all dirty pages, and check for write errors. | |
1da177e4 | 137 | */ |
5445b1fb | 138 | static int |
75e1fcc0 | 139 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 140 | { |
6de1472f | 141 | struct inode *inode = file_inode(file); |
1da177e4 | 142 | |
6de1472f | 143 | dprintk("NFS: flush(%pD2)\n", file); |
1da177e4 | 144 | |
c2459dc4 | 145 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
146 | if ((file->f_mode & FMODE_WRITE) == 0) |
147 | return 0; | |
7b159fc1 | 148 | |
7fe5c398 | 149 | /* Flush writes to the server and return any errors */ |
af7fa165 | 150 | return vfs_fsync(file, 0); |
1da177e4 LT |
151 | } |
152 | ||
ce4ef7c0 | 153 | ssize_t |
3aa2d199 | 154 | nfs_file_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 155 | { |
6de1472f | 156 | struct inode *inode = file_inode(iocb->ki_filp); |
1da177e4 LT |
157 | ssize_t result; |
158 | ||
2ba48ce5 | 159 | if (iocb->ki_flags & IOCB_DIRECT) |
c8b8e32d | 160 | return nfs_file_direct_read(iocb, to); |
1da177e4 | 161 | |
619d30b4 | 162 | dprintk("NFS: read(%pD2, %zu@%lu)\n", |
6de1472f | 163 | iocb->ki_filp, |
3aa2d199 | 164 | iov_iter_count(to), (unsigned long) iocb->ki_pos); |
1da177e4 | 165 | |
a5864c99 TM |
166 | nfs_start_io_read(inode); |
167 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); | |
4184dcf2 | 168 | if (!result) { |
3aa2d199 | 169 | result = generic_file_read_iter(iocb, to); |
4184dcf2 CL |
170 | if (result > 0) |
171 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
172 | } | |
a5864c99 | 173 | nfs_end_io_read(inode); |
1da177e4 LT |
174 | return result; |
175 | } | |
89d77c8f | 176 | EXPORT_SYMBOL_GPL(nfs_file_read); |
1da177e4 | 177 | |
ce4ef7c0 | 178 | int |
1da177e4 LT |
179 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
180 | { | |
6de1472f | 181 | struct inode *inode = file_inode(file); |
1da177e4 LT |
182 | int status; |
183 | ||
6de1472f | 184 | dprintk("NFS: mmap(%pD2)\n", file); |
1da177e4 | 185 | |
e1ebfd33 TM |
186 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
187 | * so we call that before revalidating the mapping | |
188 | */ | |
189 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
190 | if (!status) { |
191 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 192 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 193 | } |
1da177e4 LT |
194 | return status; |
195 | } | |
89d77c8f | 196 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
197 | |
198 | /* | |
199 | * Flush any dirty pages for this process, and check for write errors. | |
200 | * The return status from this call provides a reliable indication of | |
201 | * whether any write errors occurred for this process. | |
af7fa165 TM |
202 | * |
203 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
204 | * disk, but it retrieves and clears ctx->error after synching, despite | |
205 | * the two being set at the same time in nfs_context_set_write_error(). | |
206 | * This is because the former is used to notify the _next_ call to | |
25985edc | 207 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 208 | * fall back to doing a synchronous write. |
1da177e4 | 209 | */ |
4ff79bc7 | 210 | static int |
bf4b4905 | 211 | nfs_file_fsync_commit(struct file *file, int datasync) |
1da177e4 | 212 | { |
cd3758e3 | 213 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
6de1472f | 214 | struct inode *inode = file_inode(file); |
bf4b4905 | 215 | int do_resend, status; |
af7fa165 TM |
216 | int ret = 0; |
217 | ||
6de1472f | 218 | dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); |
1da177e4 | 219 | |
91d5b470 | 220 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
05990d1b | 221 | do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); |
af7fa165 | 222 | status = nfs_commit_inode(inode, FLUSH_SYNC); |
bf4b4905 | 223 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) { |
af7fa165 | 224 | ret = xchg(&ctx->error, 0); |
05990d1b TM |
225 | if (ret) |
226 | goto out; | |
227 | } | |
228 | if (status < 0) { | |
af7fa165 | 229 | ret = status; |
05990d1b TM |
230 | goto out; |
231 | } | |
232 | do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); | |
233 | if (do_resend) | |
234 | ret = -EAGAIN; | |
235 | out: | |
a5c58892 BS |
236 | return ret; |
237 | } | |
238 | ||
4ff79bc7 | 239 | int |
a5c58892 BS |
240 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
241 | { | |
242 | int ret; | |
496ad9aa | 243 | struct inode *inode = file_inode(file); |
a5c58892 | 244 | |
f4ce1299 TM |
245 | trace_nfs_fsync_enter(inode); |
246 | ||
05990d1b | 247 | do { |
bf4b4905 | 248 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
05990d1b | 249 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); |
bf4b4905 N |
250 | if (test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) { |
251 | int ret2 = xchg(&ctx->error, 0); | |
252 | if (ret2) | |
253 | ret = ret2; | |
254 | } | |
05990d1b TM |
255 | if (ret != 0) |
256 | break; | |
bf4b4905 | 257 | ret = nfs_file_fsync_commit(file, datasync); |
4ff79bc7 CH |
258 | if (!ret) |
259 | ret = pnfs_sync_inode(inode, !!datasync); | |
dcfc4f25 TM |
260 | /* |
261 | * If nfs_file_fsync_commit detected a server reboot, then | |
262 | * resend all dirty pages that might have been covered by | |
263 | * the NFS_CONTEXT_RESEND_WRITES flag | |
264 | */ | |
265 | start = 0; | |
266 | end = LLONG_MAX; | |
05990d1b TM |
267 | } while (ret == -EAGAIN); |
268 | ||
f4ce1299 | 269 | trace_nfs_fsync_exit(inode, ret); |
af7fa165 | 270 | return ret; |
1da177e4 | 271 | } |
4ff79bc7 | 272 | EXPORT_SYMBOL_GPL(nfs_file_fsync); |
1da177e4 | 273 | |
38c73044 PS |
274 | /* |
275 | * Decide whether a read/modify/write cycle may be more efficient | |
276 | * then a modify/write/read cycle when writing to a page in the | |
277 | * page cache. | |
278 | * | |
279 | * The modify/write/read cycle may occur if a page is read before | |
280 | * being completely filled by the writer. In this situation, the | |
281 | * page must be completely written to stable storage on the server | |
282 | * before it can be refilled by reading in the page from the server. | |
283 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
284 | * done. | |
285 | * | |
286 | * It may be more efficient to read the page first if the file is | |
287 | * open for reading in addition to writing, the page is not marked | |
288 | * as Uptodate, it is not dirty or waiting to be committed, | |
289 | * indicating that it was previously allocated and then modified, | |
290 | * that there were valid bytes of data in that range of the file, | |
291 | * and that the new data won't completely replace the old data in | |
292 | * that range of the file. | |
293 | */ | |
294 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
295 | loff_t pos, unsigned len) | |
296 | { | |
297 | unsigned int pglen = nfs_page_length(page); | |
09cbfeaf | 298 | unsigned int offset = pos & (PAGE_SIZE - 1); |
38c73044 PS |
299 | unsigned int end = offset + len; |
300 | ||
612aa983 CH |
301 | if (pnfs_ld_read_whole_page(file->f_mapping->host)) { |
302 | if (!PageUptodate(page)) | |
303 | return 1; | |
304 | return 0; | |
305 | } | |
306 | ||
38c73044 PS |
307 | if ((file->f_mode & FMODE_READ) && /* open for read? */ |
308 | !PageUptodate(page) && /* Uptodate? */ | |
309 | !PagePrivate(page) && /* i/o request already? */ | |
310 | pglen && /* valid bytes of file? */ | |
311 | (end < pglen || offset)) /* replace all valid bytes? */ | |
312 | return 1; | |
313 | return 0; | |
314 | } | |
315 | ||
1da177e4 | 316 | /* |
4899f9c8 NP |
317 | * This does the "real" work of the write. We must allocate and lock the |
318 | * page to be sent back to the generic routine, which then copies the | |
319 | * data from user space. | |
1da177e4 LT |
320 | * |
321 | * If the writer ends up delaying the write, the writer needs to | |
322 | * increment the page use counts until he is done with the page. | |
323 | */ | |
4899f9c8 NP |
324 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
325 | loff_t pos, unsigned len, unsigned flags, | |
326 | struct page **pagep, void **fsdata) | |
1da177e4 | 327 | { |
4899f9c8 | 328 | int ret; |
09cbfeaf | 329 | pgoff_t index = pos >> PAGE_SHIFT; |
4899f9c8 | 330 | struct page *page; |
38c73044 | 331 | int once_thru = 0; |
4899f9c8 | 332 | |
1e8968c5 | 333 | dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", |
6de1472f | 334 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 335 | |
38c73044 | 336 | start: |
54566b2c | 337 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
338 | if (!page) |
339 | return -ENOMEM; | |
340 | *pagep = page; | |
341 | ||
342 | ret = nfs_flush_incompatible(file, page); | |
343 | if (ret) { | |
344 | unlock_page(page); | |
09cbfeaf | 345 | put_page(page); |
38c73044 PS |
346 | } else if (!once_thru && |
347 | nfs_want_read_modify_write(file, page, pos, len)) { | |
348 | once_thru = 1; | |
349 | ret = nfs_readpage(file, page); | |
09cbfeaf | 350 | put_page(page); |
38c73044 PS |
351 | if (!ret) |
352 | goto start; | |
4899f9c8 NP |
353 | } |
354 | return ret; | |
1da177e4 LT |
355 | } |
356 | ||
4899f9c8 NP |
357 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
358 | loff_t pos, unsigned len, unsigned copied, | |
359 | struct page *page, void *fsdata) | |
1da177e4 | 360 | { |
09cbfeaf | 361 | unsigned offset = pos & (PAGE_SIZE - 1); |
dc24826b | 362 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
4899f9c8 | 363 | int status; |
1da177e4 | 364 | |
1e8968c5 | 365 | dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", |
6de1472f | 366 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 367 | |
efc91ed0 TM |
368 | /* |
369 | * Zero any uninitialised parts of the page, and then mark the page | |
370 | * as up to date if it turns out that we're extending the file. | |
371 | */ | |
372 | if (!PageUptodate(page)) { | |
373 | unsigned pglen = nfs_page_length(page); | |
c0cf3ef5 | 374 | unsigned end = offset + copied; |
efc91ed0 TM |
375 | |
376 | if (pglen == 0) { | |
377 | zero_user_segments(page, 0, offset, | |
09cbfeaf | 378 | end, PAGE_SIZE); |
efc91ed0 TM |
379 | SetPageUptodate(page); |
380 | } else if (end >= pglen) { | |
09cbfeaf | 381 | zero_user_segment(page, end, PAGE_SIZE); |
efc91ed0 TM |
382 | if (offset == 0) |
383 | SetPageUptodate(page); | |
384 | } else | |
09cbfeaf | 385 | zero_user_segment(page, pglen, PAGE_SIZE); |
efc91ed0 TM |
386 | } |
387 | ||
4899f9c8 | 388 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
389 | |
390 | unlock_page(page); | |
09cbfeaf | 391 | put_page(page); |
4899f9c8 | 392 | |
3d509e54 CL |
393 | if (status < 0) |
394 | return status; | |
2701d086 | 395 | NFS_I(mapping->host)->write_io += copied; |
dc24826b | 396 | |
ce52914e | 397 | if (nfs_ctx_key_to_expire(ctx, mapping->host)) { |
dc24826b AA |
398 | status = nfs_wb_all(mapping->host); |
399 | if (status < 0) | |
400 | return status; | |
401 | } | |
402 | ||
3d509e54 | 403 | return copied; |
1da177e4 LT |
404 | } |
405 | ||
6b9b3514 DH |
406 | /* |
407 | * Partially or wholly invalidate a page | |
408 | * - Release the private state associated with a page if undergoing complete | |
409 | * page invalidation | |
545db45f | 410 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
411 | * - Caller holds page lock |
412 | */ | |
d47992f8 LC |
413 | static void nfs_invalidate_page(struct page *page, unsigned int offset, |
414 | unsigned int length) | |
cd52ed35 | 415 | { |
d47992f8 LC |
416 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n", |
417 | page, offset, length); | |
b7eaefaa | 418 | |
09cbfeaf | 419 | if (offset != 0 || length < PAGE_SIZE) |
1c75950b | 420 | return; |
d2ccddf0 | 421 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 422 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
423 | |
424 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
425 | } |
426 | ||
6b9b3514 DH |
427 | /* |
428 | * Attempt to release the private state associated with a page | |
545db45f | 429 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
430 | * - Caller holds page lock |
431 | * - Return true (may release page) or false (may not) | |
432 | */ | |
cd52ed35 TM |
433 | static int nfs_release_page(struct page *page, gfp_t gfp) |
434 | { | |
b7eaefaa CL |
435 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
436 | ||
e3db7691 | 437 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
438 | if (PagePrivate(page)) |
439 | return 0; | |
440 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
441 | } |
442 | ||
f919b196 MG |
443 | static void nfs_check_dirty_writeback(struct page *page, |
444 | bool *dirty, bool *writeback) | |
445 | { | |
446 | struct nfs_inode *nfsi; | |
447 | struct address_space *mapping = page_file_mapping(page); | |
448 | ||
449 | if (!mapping || PageSwapCache(page)) | |
450 | return; | |
451 | ||
452 | /* | |
453 | * Check if an unstable page is currently being committed and | |
454 | * if so, have the VM treat it as if the page is under writeback | |
455 | * so it will not block due to pages that will shortly be freeable. | |
456 | */ | |
457 | nfsi = NFS_I(mapping->host); | |
af7cf057 | 458 | if (atomic_read(&nfsi->commit_info.rpcs_out)) { |
f919b196 MG |
459 | *writeback = true; |
460 | return; | |
461 | } | |
462 | ||
463 | /* | |
464 | * If PagePrivate() is set, then the page is not freeable and as the | |
465 | * inode is not being committed, it's not going to be cleaned in the | |
466 | * near future so treat it as dirty | |
467 | */ | |
468 | if (PagePrivate(page)) | |
469 | *dirty = true; | |
470 | } | |
471 | ||
6b9b3514 DH |
472 | /* |
473 | * Attempt to clear the private state associated with a page when an error | |
474 | * occurs that requires the cached contents of an inode to be written back or | |
475 | * destroyed | |
545db45f | 476 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
477 | * - Caller holds page lock |
478 | * - Return 0 if successful, -error otherwise | |
479 | */ | |
e3db7691 TM |
480 | static int nfs_launder_page(struct page *page) |
481 | { | |
d56b4ddf | 482 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 483 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
484 | |
485 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
486 | inode->i_ino, (long long)page_offset(page)); | |
487 | ||
545db45f | 488 | nfs_fscache_wait_on_page_write(nfsi, page); |
c373fff7 | 489 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
490 | } |
491 | ||
a564b8f0 MG |
492 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, |
493 | sector_t *span) | |
494 | { | |
dad2b015 JL |
495 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
496 | ||
a564b8f0 | 497 | *span = sis->pages; |
dad2b015 | 498 | |
3c87ef6e | 499 | return rpc_clnt_swap_activate(clnt); |
a564b8f0 MG |
500 | } |
501 | ||
502 | static void nfs_swap_deactivate(struct file *file) | |
503 | { | |
dad2b015 JL |
504 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
505 | ||
3c87ef6e | 506 | rpc_clnt_swap_deactivate(clnt); |
a564b8f0 | 507 | } |
a564b8f0 | 508 | |
f5e54d6e | 509 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
510 | .readpage = nfs_readpage, |
511 | .readpages = nfs_readpages, | |
9cccef95 | 512 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
513 | .writepage = nfs_writepage, |
514 | .writepages = nfs_writepages, | |
4899f9c8 NP |
515 | .write_begin = nfs_write_begin, |
516 | .write_end = nfs_write_end, | |
cd52ed35 TM |
517 | .invalidatepage = nfs_invalidate_page, |
518 | .releasepage = nfs_release_page, | |
1da177e4 | 519 | .direct_IO = nfs_direct_IO, |
f844cd0d | 520 | #ifdef CONFIG_MIGRATION |
074cc1de | 521 | .migratepage = nfs_migrate_page, |
f844cd0d | 522 | #endif |
e3db7691 | 523 | .launder_page = nfs_launder_page, |
f919b196 | 524 | .is_dirty_writeback = nfs_check_dirty_writeback, |
f590f333 | 525 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
526 | .swap_activate = nfs_swap_activate, |
527 | .swap_deactivate = nfs_swap_deactivate, | |
1da177e4 LT |
528 | }; |
529 | ||
6b9b3514 DH |
530 | /* |
531 | * Notification that a PTE pointing to an NFS page is about to be made | |
532 | * writable, implying that someone is about to modify the page through a | |
533 | * shared-writable mapping | |
534 | */ | |
11bac800 | 535 | static int nfs_vm_page_mkwrite(struct vm_fault *vmf) |
94387fb1 | 536 | { |
c2ec175c | 537 | struct page *page = vmf->page; |
11bac800 | 538 | struct file *filp = vmf->vma->vm_file; |
6de1472f | 539 | struct inode *inode = file_inode(filp); |
94387fb1 | 540 | unsigned pagelen; |
bc4866b6 | 541 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 542 | struct address_space *mapping; |
94387fb1 | 543 | |
1e8968c5 | 544 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", |
6de1472f | 545 | filp, filp->f_mapping->host->i_ino, |
b7eaefaa CL |
546 | (long long)page_offset(page)); |
547 | ||
9a773e7c TM |
548 | sb_start_pagefault(inode->i_sb); |
549 | ||
545db45f | 550 | /* make sure the cache has finished storing the page */ |
6de1472f | 551 | nfs_fscache_wait_on_page_write(NFS_I(inode), page); |
545db45f | 552 | |
ef070dcb TM |
553 | wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING, |
554 | nfs_wait_bit_killable, TASK_KILLABLE); | |
555 | ||
94387fb1 | 556 | lock_page(page); |
d56b4ddf | 557 | mapping = page_file_mapping(page); |
6de1472f | 558 | if (mapping != inode->i_mapping) |
8b1f9ee5 TM |
559 | goto out_unlock; |
560 | ||
2aeb98f4 TM |
561 | wait_on_page_writeback(page); |
562 | ||
94387fb1 | 563 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
564 | if (pagelen == 0) |
565 | goto out_unlock; | |
4899f9c8 | 566 | |
bc4866b6 TM |
567 | ret = VM_FAULT_LOCKED; |
568 | if (nfs_flush_incompatible(filp, page) == 0 && | |
569 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
570 | goto out; | |
8b1f9ee5 | 571 | |
bc4866b6 | 572 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
573 | out_unlock: |
574 | unlock_page(page); | |
bc4866b6 | 575 | out: |
9a773e7c | 576 | sb_end_pagefault(inode->i_sb); |
bc4866b6 | 577 | return ret; |
94387fb1 TM |
578 | } |
579 | ||
f0f37e2f | 580 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 | 581 | .fault = filemap_fault, |
f1820361 | 582 | .map_pages = filemap_map_pages, |
94387fb1 TM |
583 | .page_mkwrite = nfs_vm_page_mkwrite, |
584 | }; | |
585 | ||
7e94d6c4 | 586 | static int nfs_need_check_write(struct file *filp, struct inode *inode) |
7b159fc1 TM |
587 | { |
588 | struct nfs_open_context *ctx; | |
589 | ||
cd3758e3 | 590 | ctx = nfs_file_open_context(filp); |
dc24826b | 591 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) || |
ce52914e | 592 | nfs_ctx_key_to_expire(ctx, inode)) |
7b159fc1 TM |
593 | return 1; |
594 | return 0; | |
595 | } | |
596 | ||
edaf4369 | 597 | ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 598 | { |
6de1472f AV |
599 | struct file *file = iocb->ki_filp; |
600 | struct inode *inode = file_inode(file); | |
7e381172 | 601 | unsigned long written = 0; |
1da177e4 LT |
602 | ssize_t result; |
603 | ||
6de1472f | 604 | result = nfs_key_timeout_notify(file, inode); |
dc24826b AA |
605 | if (result) |
606 | return result; | |
607 | ||
89698b24 | 608 | if (iocb->ki_flags & IOCB_DIRECT) |
65a4a1ca | 609 | return nfs_file_direct_write(iocb, from); |
1da177e4 | 610 | |
619d30b4 | 611 | dprintk("NFS: write(%pD2, %zu@%Ld)\n", |
18290650 | 612 | file, iov_iter_count(from), (long long) iocb->ki_pos); |
1da177e4 | 613 | |
1da177e4 LT |
614 | if (IS_SWAPFILE(inode)) |
615 | goto out_swapfile; | |
7d52e862 TM |
616 | /* |
617 | * O_APPEND implies that we must revalidate the file length. | |
618 | */ | |
2ba48ce5 | 619 | if (iocb->ki_flags & IOCB_APPEND) { |
6de1472f | 620 | result = nfs_revalidate_file_size(inode, file); |
7d52e862 TM |
621 | if (result) |
622 | goto out; | |
fe51beec | 623 | } |
6ba80d43 N |
624 | if (iocb->ki_pos > i_size_read(inode)) |
625 | nfs_revalidate_mapping(inode, file->f_mapping); | |
1da177e4 | 626 | |
a5864c99 | 627 | nfs_start_io_write(inode); |
18290650 TM |
628 | result = generic_write_checks(iocb, from); |
629 | if (result > 0) { | |
630 | current->backing_dev_info = inode_to_bdi(inode); | |
631 | result = generic_perform_write(file, from, iocb->ki_pos); | |
632 | current->backing_dev_info = NULL; | |
633 | } | |
a5864c99 | 634 | nfs_end_io_write(inode); |
18290650 | 635 | if (result <= 0) |
1da177e4 LT |
636 | goto out; |
637 | ||
c49edecd | 638 | written = result; |
18290650 | 639 | iocb->ki_pos += written; |
e973b1a5 | 640 | result = generic_write_sync(iocb, written); |
641 | if (result < 0) | |
642 | goto out; | |
7e381172 | 643 | |
7e94d6c4 | 644 | /* Return error values */ |
18290650 | 645 | if (nfs_need_check_write(file, inode)) { |
6de1472f | 646 | int err = vfs_fsync(file, 0); |
200baa21 TM |
647 | if (err < 0) |
648 | result = err; | |
649 | } | |
18290650 | 650 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); |
1da177e4 LT |
651 | out: |
652 | return result; | |
653 | ||
654 | out_swapfile: | |
655 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
18290650 | 656 | return -EBUSY; |
1da177e4 | 657 | } |
89d77c8f | 658 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 659 | |
5eebde23 SJ |
660 | static int |
661 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
662 | { |
663 | struct inode *inode = filp->f_mapping->host; | |
664 | int status = 0; | |
21ac19d4 | 665 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 666 | |
039c4d7a | 667 | /* Try local locking first */ |
6d34ac19 BF |
668 | posix_test_lock(filp, fl); |
669 | if (fl->fl_type != F_UNLCK) { | |
670 | /* found a conflict */ | |
039c4d7a | 671 | goto out; |
1da177e4 | 672 | } |
21ac19d4 | 673 | fl->fl_type = saved_type; |
039c4d7a | 674 | |
011e2a7f | 675 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
676 | goto out_noconflict; |
677 | ||
5eebde23 | 678 | if (is_local) |
039c4d7a TM |
679 | goto out_noconflict; |
680 | ||
681 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
682 | out: | |
1da177e4 | 683 | return status; |
039c4d7a TM |
684 | out_noconflict: |
685 | fl->fl_type = F_UNLCK; | |
686 | goto out; | |
1da177e4 LT |
687 | } |
688 | ||
5eebde23 SJ |
689 | static int |
690 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
691 | { |
692 | struct inode *inode = filp->f_mapping->host; | |
7a8203d8 | 693 | struct nfs_lock_context *l_ctx; |
1da177e4 LT |
694 | int status; |
695 | ||
1da177e4 LT |
696 | /* |
697 | * Flush all pending writes before doing anything | |
698 | * with locks.. | |
699 | */ | |
d9dabc1a | 700 | vfs_fsync(filp, 0); |
1da177e4 | 701 | |
7a8203d8 TM |
702 | l_ctx = nfs_get_lock_context(nfs_file_open_context(filp)); |
703 | if (!IS_ERR(l_ctx)) { | |
210c7c17 | 704 | status = nfs_iocounter_wait(l_ctx); |
7a8203d8 | 705 | nfs_put_lock_context(l_ctx); |
f30cb757 BC |
706 | /* NOTE: special case |
707 | * If we're signalled while cleaning up locks on process exit, we | |
708 | * still need to complete the unlock. | |
709 | */ | |
710 | if (status < 0 && !(fl->fl_flags & FL_CLOSE)) | |
7a8203d8 TM |
711 | return status; |
712 | } | |
713 | ||
5eebde23 SJ |
714 | /* |
715 | * Use local locking if mounted with "-onolock" or with appropriate | |
716 | * "-olocal_lock=" | |
717 | */ | |
718 | if (!is_local) | |
1da177e4 LT |
719 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
720 | else | |
75575ddf | 721 | status = locks_lock_file_wait(filp, fl); |
1da177e4 LT |
722 | return status; |
723 | } | |
724 | ||
5eebde23 SJ |
725 | static int |
726 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
727 | { |
728 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
729 | int status; |
730 | ||
1da177e4 LT |
731 | /* |
732 | * Flush all pending writes before doing anything | |
733 | * with locks.. | |
734 | */ | |
29884df0 TM |
735 | status = nfs_sync_mapping(filp->f_mapping); |
736 | if (status != 0) | |
1da177e4 LT |
737 | goto out; |
738 | ||
5eebde23 SJ |
739 | /* |
740 | * Use local locking if mounted with "-onolock" or with appropriate | |
741 | * "-olocal_lock=" | |
742 | */ | |
743 | if (!is_local) | |
1da177e4 | 744 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 745 | else |
75575ddf | 746 | status = locks_lock_file_wait(filp, fl); |
1da177e4 LT |
747 | if (status < 0) |
748 | goto out; | |
6b96724e | 749 | |
1da177e4 | 750 | /* |
779eafab N |
751 | * Invalidate cache to prevent missing any changes. If |
752 | * the file is mapped, clear the page cache as well so | |
753 | * those mappings will be loaded. | |
6b96724e | 754 | * |
1da177e4 LT |
755 | * This makes locking act as a cache coherency point. |
756 | */ | |
29884df0 | 757 | nfs_sync_mapping(filp->f_mapping); |
779eafab | 758 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { |
442ce049 | 759 | nfs_zap_caches(inode); |
779eafab N |
760 | if (mapping_mapped(filp->f_mapping)) |
761 | nfs_revalidate_mapping(inode, filp->f_mapping); | |
762 | } | |
1da177e4 | 763 | out: |
1da177e4 LT |
764 | return status; |
765 | } | |
766 | ||
767 | /* | |
768 | * Lock a (portion of) a file | |
769 | */ | |
ce4ef7c0 | 770 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 771 | { |
6da24bc9 | 772 | struct inode *inode = filp->f_mapping->host; |
2116271a | 773 | int ret = -ENOLCK; |
5eebde23 | 774 | int is_local = 0; |
1da177e4 | 775 | |
6de1472f AV |
776 | dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n", |
777 | filp, fl->fl_type, fl->fl_flags, | |
1da177e4 | 778 | (long long)fl->fl_start, (long long)fl->fl_end); |
6da24bc9 | 779 | |
91d5b470 | 780 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
781 | |
782 | /* No mandatory locks over NFS */ | |
dfad9441 | 783 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
784 | goto out_err; |
785 | ||
5eebde23 SJ |
786 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
787 | is_local = 1; | |
788 | ||
2116271a TM |
789 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
790 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
791 | if (ret < 0) | |
792 | goto out_err; | |
793 | } | |
1da177e4 LT |
794 | |
795 | if (IS_GETLK(cmd)) | |
5eebde23 | 796 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 797 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 798 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 799 | else |
5eebde23 | 800 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
801 | out_err: |
802 | return ret; | |
1da177e4 | 803 | } |
89d77c8f | 804 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
805 | |
806 | /* | |
807 | * Lock a (portion of) a file | |
808 | */ | |
ce4ef7c0 | 809 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 810 | { |
5eebde23 SJ |
811 | struct inode *inode = filp->f_mapping->host; |
812 | int is_local = 0; | |
813 | ||
6de1472f AV |
814 | dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n", |
815 | filp, fl->fl_type, fl->fl_flags); | |
1da177e4 | 816 | |
1da177e4 LT |
817 | if (!(fl->fl_flags & FL_FLOCK)) |
818 | return -ENOLCK; | |
819 | ||
ad0fcd4e JL |
820 | /* |
821 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
822 | * any standard. In principle we might be able to support LOCK_MAND | |
823 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
824 | * NFS code is not set up for it. | |
825 | */ | |
826 | if (fl->fl_type & LOCK_MAND) | |
827 | return -EINVAL; | |
828 | ||
5eebde23 SJ |
829 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
830 | is_local = 1; | |
831 | ||
fcfa4470 BC |
832 | /* We're simulating flock() locks using posix locks on the server */ |
833 | if (fl->fl_type == F_UNLCK) | |
5eebde23 SJ |
834 | return do_unlk(filp, cmd, fl, is_local); |
835 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 836 | } |
89d77c8f | 837 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 838 | |
0486958f JL |
839 | const struct file_operations nfs_file_operations = { |
840 | .llseek = nfs_file_llseek, | |
3aa2d199 | 841 | .read_iter = nfs_file_read, |
edaf4369 | 842 | .write_iter = nfs_file_write, |
0486958f JL |
843 | .mmap = nfs_file_mmap, |
844 | .open = nfs_file_open, | |
845 | .flush = nfs_file_flush, | |
846 | .release = nfs_file_release, | |
847 | .fsync = nfs_file_fsync, | |
848 | .lock = nfs_lock, | |
849 | .flock = nfs_flock, | |
82c156f8 | 850 | .splice_read = generic_file_splice_read, |
4da54c21 | 851 | .splice_write = iter_file_splice_write, |
0486958f | 852 | .check_flags = nfs_check_flags, |
1c994a09 | 853 | .setlease = simple_nosetlease, |
0486958f | 854 | }; |
ddda8e0a | 855 | EXPORT_SYMBOL_GPL(nfs_file_operations); |