]>
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); | |
105 | ||
d7cf8dd0 TM |
106 | if (nfs_have_delegated_attributes(inode)) |
107 | goto out_noreval; | |
108 | ||
980802e3 TM |
109 | if (filp->f_flags & O_DIRECT) |
110 | goto force_reval; | |
d7cf8dd0 TM |
111 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
112 | goto force_reval; | |
113 | if (nfs_attribute_timeout(inode)) | |
114 | goto force_reval; | |
115 | out_noreval: | |
116 | return 0; | |
980802e3 TM |
117 | force_reval: |
118 | return __nfs_revalidate_inode(server, inode); | |
119 | } | |
120 | ||
965c8e59 | 121 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence) |
980802e3 | 122 | { |
6de1472f AV |
123 | dprintk("NFS: llseek file(%pD2, %lld, %d)\n", |
124 | filp, offset, whence); | |
b84e06c5 | 125 | |
06222e49 | 126 | /* |
965c8e59 | 127 | * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate |
06222e49 JB |
128 | * the cached file length |
129 | */ | |
965c8e59 | 130 | if (whence != SEEK_SET && whence != SEEK_CUR) { |
980802e3 | 131 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 132 | |
980802e3 TM |
133 | int retval = nfs_revalidate_file_size(inode, filp); |
134 | if (retval < 0) | |
135 | return (loff_t)retval; | |
79835a71 | 136 | } |
d5e66348 | 137 | |
965c8e59 | 138 | return generic_file_llseek(filp, offset, whence); |
980802e3 | 139 | } |
89d77c8f | 140 | EXPORT_SYMBOL_GPL(nfs_file_llseek); |
980802e3 | 141 | |
1da177e4 LT |
142 | /* |
143 | * Flush all dirty pages, and check for write errors. | |
1da177e4 | 144 | */ |
5445b1fb | 145 | static int |
75e1fcc0 | 146 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 147 | { |
6de1472f | 148 | struct inode *inode = file_inode(file); |
1da177e4 | 149 | |
6de1472f | 150 | dprintk("NFS: flush(%pD2)\n", file); |
1da177e4 | 151 | |
c2459dc4 | 152 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
153 | if ((file->f_mode & FMODE_WRITE) == 0) |
154 | return 0; | |
7b159fc1 | 155 | |
7fe5c398 | 156 | /* Flush writes to the server and return any errors */ |
af7fa165 | 157 | return vfs_fsync(file, 0); |
1da177e4 LT |
158 | } |
159 | ||
ce4ef7c0 | 160 | ssize_t |
3aa2d199 | 161 | nfs_file_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 162 | { |
6de1472f | 163 | struct inode *inode = file_inode(iocb->ki_filp); |
1da177e4 LT |
164 | ssize_t result; |
165 | ||
2ba48ce5 | 166 | if (iocb->ki_flags & IOCB_DIRECT) |
c8b8e32d | 167 | return nfs_file_direct_read(iocb, to); |
1da177e4 | 168 | |
619d30b4 | 169 | dprintk("NFS: read(%pD2, %zu@%lu)\n", |
6de1472f | 170 | iocb->ki_filp, |
3aa2d199 | 171 | iov_iter_count(to), (unsigned long) iocb->ki_pos); |
1da177e4 | 172 | |
a5864c99 TM |
173 | nfs_start_io_read(inode); |
174 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); | |
4184dcf2 | 175 | if (!result) { |
3aa2d199 | 176 | result = generic_file_read_iter(iocb, to); |
4184dcf2 CL |
177 | if (result > 0) |
178 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
179 | } | |
a5864c99 | 180 | nfs_end_io_read(inode); |
1da177e4 LT |
181 | return result; |
182 | } | |
89d77c8f | 183 | EXPORT_SYMBOL_GPL(nfs_file_read); |
1da177e4 | 184 | |
ce4ef7c0 | 185 | int |
1da177e4 LT |
186 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
187 | { | |
6de1472f | 188 | struct inode *inode = file_inode(file); |
1da177e4 LT |
189 | int status; |
190 | ||
6de1472f | 191 | dprintk("NFS: mmap(%pD2)\n", file); |
1da177e4 | 192 | |
e1ebfd33 TM |
193 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
194 | * so we call that before revalidating the mapping | |
195 | */ | |
196 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
197 | if (!status) { |
198 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 199 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 200 | } |
1da177e4 LT |
201 | return status; |
202 | } | |
89d77c8f | 203 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
204 | |
205 | /* | |
206 | * Flush any dirty pages for this process, and check for write errors. | |
207 | * The return status from this call provides a reliable indication of | |
208 | * whether any write errors occurred for this process. | |
af7fa165 TM |
209 | * |
210 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
211 | * disk, but it retrieves and clears ctx->error after synching, despite | |
212 | * the two being set at the same time in nfs_context_set_write_error(). | |
213 | * This is because the former is used to notify the _next_ call to | |
25985edc | 214 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 215 | * fall back to doing a synchronous write. |
1da177e4 | 216 | */ |
4ff79bc7 | 217 | static int |
a5c58892 | 218 | nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 219 | { |
cd3758e3 | 220 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
6de1472f | 221 | struct inode *inode = file_inode(file); |
05990d1b | 222 | int have_error, do_resend, status; |
af7fa165 TM |
223 | int ret = 0; |
224 | ||
6de1472f | 225 | dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); |
1da177e4 | 226 | |
91d5b470 | 227 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
05990d1b | 228 | do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); |
af7fa165 TM |
229 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
230 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
231 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
05990d1b | 232 | if (have_error) { |
af7fa165 | 233 | ret = xchg(&ctx->error, 0); |
05990d1b TM |
234 | if (ret) |
235 | goto out; | |
236 | } | |
237 | if (status < 0) { | |
af7fa165 | 238 | ret = status; |
05990d1b TM |
239 | goto out; |
240 | } | |
241 | do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); | |
242 | if (do_resend) | |
243 | ret = -EAGAIN; | |
244 | out: | |
a5c58892 BS |
245 | return ret; |
246 | } | |
247 | ||
4ff79bc7 | 248 | int |
a5c58892 BS |
249 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
250 | { | |
251 | int ret; | |
496ad9aa | 252 | struct inode *inode = file_inode(file); |
a5c58892 | 253 | |
f4ce1299 TM |
254 | trace_nfs_fsync_enter(inode); |
255 | ||
05990d1b TM |
256 | do { |
257 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
258 | if (ret != 0) | |
259 | break; | |
05990d1b | 260 | ret = nfs_file_fsync_commit(file, start, end, datasync); |
4ff79bc7 CH |
261 | if (!ret) |
262 | ret = pnfs_sync_inode(inode, !!datasync); | |
dcfc4f25 TM |
263 | /* |
264 | * If nfs_file_fsync_commit detected a server reboot, then | |
265 | * resend all dirty pages that might have been covered by | |
266 | * the NFS_CONTEXT_RESEND_WRITES flag | |
267 | */ | |
268 | start = 0; | |
269 | end = LLONG_MAX; | |
05990d1b TM |
270 | } while (ret == -EAGAIN); |
271 | ||
f4ce1299 | 272 | trace_nfs_fsync_exit(inode, ret); |
af7fa165 | 273 | return ret; |
1da177e4 | 274 | } |
4ff79bc7 | 275 | EXPORT_SYMBOL_GPL(nfs_file_fsync); |
1da177e4 | 276 | |
38c73044 PS |
277 | /* |
278 | * Decide whether a read/modify/write cycle may be more efficient | |
279 | * then a modify/write/read cycle when writing to a page in the | |
280 | * page cache. | |
281 | * | |
282 | * The modify/write/read cycle may occur if a page is read before | |
283 | * being completely filled by the writer. In this situation, the | |
284 | * page must be completely written to stable storage on the server | |
285 | * before it can be refilled by reading in the page from the server. | |
286 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
287 | * done. | |
288 | * | |
289 | * It may be more efficient to read the page first if the file is | |
290 | * open for reading in addition to writing, the page is not marked | |
291 | * as Uptodate, it is not dirty or waiting to be committed, | |
292 | * indicating that it was previously allocated and then modified, | |
293 | * that there were valid bytes of data in that range of the file, | |
294 | * and that the new data won't completely replace the old data in | |
295 | * that range of the file. | |
296 | */ | |
297 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
298 | loff_t pos, unsigned len) | |
299 | { | |
300 | unsigned int pglen = nfs_page_length(page); | |
09cbfeaf | 301 | unsigned int offset = pos & (PAGE_SIZE - 1); |
38c73044 PS |
302 | unsigned int end = offset + len; |
303 | ||
612aa983 CH |
304 | if (pnfs_ld_read_whole_page(file->f_mapping->host)) { |
305 | if (!PageUptodate(page)) | |
306 | return 1; | |
307 | return 0; | |
308 | } | |
309 | ||
38c73044 PS |
310 | if ((file->f_mode & FMODE_READ) && /* open for read? */ |
311 | !PageUptodate(page) && /* Uptodate? */ | |
312 | !PagePrivate(page) && /* i/o request already? */ | |
313 | pglen && /* valid bytes of file? */ | |
314 | (end < pglen || offset)) /* replace all valid bytes? */ | |
315 | return 1; | |
316 | return 0; | |
317 | } | |
318 | ||
1da177e4 | 319 | /* |
4899f9c8 NP |
320 | * This does the "real" work of the write. We must allocate and lock the |
321 | * page to be sent back to the generic routine, which then copies the | |
322 | * data from user space. | |
1da177e4 LT |
323 | * |
324 | * If the writer ends up delaying the write, the writer needs to | |
325 | * increment the page use counts until he is done with the page. | |
326 | */ | |
4899f9c8 NP |
327 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
328 | loff_t pos, unsigned len, unsigned flags, | |
329 | struct page **pagep, void **fsdata) | |
1da177e4 | 330 | { |
4899f9c8 | 331 | int ret; |
09cbfeaf | 332 | pgoff_t index = pos >> PAGE_SHIFT; |
4899f9c8 | 333 | struct page *page; |
38c73044 | 334 | int once_thru = 0; |
4899f9c8 | 335 | |
1e8968c5 | 336 | dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", |
6de1472f | 337 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 338 | |
38c73044 | 339 | start: |
54566b2c | 340 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
341 | if (!page) |
342 | return -ENOMEM; | |
343 | *pagep = page; | |
344 | ||
345 | ret = nfs_flush_incompatible(file, page); | |
346 | if (ret) { | |
347 | unlock_page(page); | |
09cbfeaf | 348 | put_page(page); |
38c73044 PS |
349 | } else if (!once_thru && |
350 | nfs_want_read_modify_write(file, page, pos, len)) { | |
351 | once_thru = 1; | |
352 | ret = nfs_readpage(file, page); | |
09cbfeaf | 353 | put_page(page); |
38c73044 PS |
354 | if (!ret) |
355 | goto start; | |
4899f9c8 NP |
356 | } |
357 | return ret; | |
1da177e4 LT |
358 | } |
359 | ||
4899f9c8 NP |
360 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
361 | loff_t pos, unsigned len, unsigned copied, | |
362 | struct page *page, void *fsdata) | |
1da177e4 | 363 | { |
09cbfeaf | 364 | unsigned offset = pos & (PAGE_SIZE - 1); |
dc24826b | 365 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
4899f9c8 | 366 | int status; |
1da177e4 | 367 | |
1e8968c5 | 368 | dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", |
6de1472f | 369 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 370 | |
efc91ed0 TM |
371 | /* |
372 | * Zero any uninitialised parts of the page, and then mark the page | |
373 | * as up to date if it turns out that we're extending the file. | |
374 | */ | |
375 | if (!PageUptodate(page)) { | |
376 | unsigned pglen = nfs_page_length(page); | |
377 | unsigned end = offset + len; | |
378 | ||
379 | if (pglen == 0) { | |
380 | zero_user_segments(page, 0, offset, | |
09cbfeaf | 381 | end, PAGE_SIZE); |
efc91ed0 TM |
382 | SetPageUptodate(page); |
383 | } else if (end >= pglen) { | |
09cbfeaf | 384 | zero_user_segment(page, end, PAGE_SIZE); |
efc91ed0 TM |
385 | if (offset == 0) |
386 | SetPageUptodate(page); | |
387 | } else | |
09cbfeaf | 388 | zero_user_segment(page, pglen, PAGE_SIZE); |
efc91ed0 TM |
389 | } |
390 | ||
4899f9c8 | 391 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
392 | |
393 | unlock_page(page); | |
09cbfeaf | 394 | put_page(page); |
4899f9c8 | 395 | |
3d509e54 CL |
396 | if (status < 0) |
397 | return status; | |
2701d086 | 398 | NFS_I(mapping->host)->write_io += copied; |
dc24826b | 399 | |
ce52914e | 400 | if (nfs_ctx_key_to_expire(ctx, mapping->host)) { |
dc24826b AA |
401 | status = nfs_wb_all(mapping->host); |
402 | if (status < 0) | |
403 | return status; | |
404 | } | |
405 | ||
3d509e54 | 406 | return copied; |
1da177e4 LT |
407 | } |
408 | ||
6b9b3514 DH |
409 | /* |
410 | * Partially or wholly invalidate a page | |
411 | * - Release the private state associated with a page if undergoing complete | |
412 | * page invalidation | |
545db45f | 413 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
414 | * - Caller holds page lock |
415 | */ | |
d47992f8 LC |
416 | static void nfs_invalidate_page(struct page *page, unsigned int offset, |
417 | unsigned int length) | |
cd52ed35 | 418 | { |
d47992f8 LC |
419 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n", |
420 | page, offset, length); | |
b7eaefaa | 421 | |
09cbfeaf | 422 | if (offset != 0 || length < PAGE_SIZE) |
1c75950b | 423 | return; |
d2ccddf0 | 424 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 425 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
426 | |
427 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
428 | } |
429 | ||
6b9b3514 DH |
430 | /* |
431 | * Attempt to release the private state associated with a page | |
545db45f | 432 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
433 | * - Caller holds page lock |
434 | * - Return true (may release page) or false (may not) | |
435 | */ | |
cd52ed35 TM |
436 | static int nfs_release_page(struct page *page, gfp_t gfp) |
437 | { | |
b7eaefaa CL |
438 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
439 | ||
e3db7691 | 440 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
441 | if (PagePrivate(page)) |
442 | return 0; | |
443 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
444 | } |
445 | ||
f919b196 MG |
446 | static void nfs_check_dirty_writeback(struct page *page, |
447 | bool *dirty, bool *writeback) | |
448 | { | |
449 | struct nfs_inode *nfsi; | |
450 | struct address_space *mapping = page_file_mapping(page); | |
451 | ||
452 | if (!mapping || PageSwapCache(page)) | |
453 | return; | |
454 | ||
455 | /* | |
456 | * Check if an unstable page is currently being committed and | |
457 | * if so, have the VM treat it as if the page is under writeback | |
458 | * so it will not block due to pages that will shortly be freeable. | |
459 | */ | |
460 | nfsi = NFS_I(mapping->host); | |
af7cf057 | 461 | if (atomic_read(&nfsi->commit_info.rpcs_out)) { |
f919b196 MG |
462 | *writeback = true; |
463 | return; | |
464 | } | |
465 | ||
466 | /* | |
467 | * If PagePrivate() is set, then the page is not freeable and as the | |
468 | * inode is not being committed, it's not going to be cleaned in the | |
469 | * near future so treat it as dirty | |
470 | */ | |
471 | if (PagePrivate(page)) | |
472 | *dirty = true; | |
473 | } | |
474 | ||
6b9b3514 DH |
475 | /* |
476 | * Attempt to clear the private state associated with a page when an error | |
477 | * occurs that requires the cached contents of an inode to be written back or | |
478 | * destroyed | |
545db45f | 479 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
480 | * - Caller holds page lock |
481 | * - Return 0 if successful, -error otherwise | |
482 | */ | |
e3db7691 TM |
483 | static int nfs_launder_page(struct page *page) |
484 | { | |
d56b4ddf | 485 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 486 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
487 | |
488 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
489 | inode->i_ino, (long long)page_offset(page)); | |
490 | ||
545db45f | 491 | nfs_fscache_wait_on_page_write(nfsi, page); |
d6c843b9 | 492 | return nfs_wb_launder_page(inode, page); |
cd52ed35 TM |
493 | } |
494 | ||
a564b8f0 MG |
495 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, |
496 | sector_t *span) | |
497 | { | |
dad2b015 JL |
498 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
499 | ||
a564b8f0 | 500 | *span = sis->pages; |
dad2b015 | 501 | |
3c87ef6e | 502 | return rpc_clnt_swap_activate(clnt); |
a564b8f0 MG |
503 | } |
504 | ||
505 | static void nfs_swap_deactivate(struct file *file) | |
506 | { | |
dad2b015 JL |
507 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
508 | ||
3c87ef6e | 509 | rpc_clnt_swap_deactivate(clnt); |
a564b8f0 | 510 | } |
a564b8f0 | 511 | |
f5e54d6e | 512 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
513 | .readpage = nfs_readpage, |
514 | .readpages = nfs_readpages, | |
9cccef95 | 515 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
516 | .writepage = nfs_writepage, |
517 | .writepages = nfs_writepages, | |
4899f9c8 NP |
518 | .write_begin = nfs_write_begin, |
519 | .write_end = nfs_write_end, | |
cd52ed35 TM |
520 | .invalidatepage = nfs_invalidate_page, |
521 | .releasepage = nfs_release_page, | |
1da177e4 | 522 | .direct_IO = nfs_direct_IO, |
074cc1de | 523 | .migratepage = nfs_migrate_page, |
e3db7691 | 524 | .launder_page = nfs_launder_page, |
f919b196 | 525 | .is_dirty_writeback = nfs_check_dirty_writeback, |
f590f333 | 526 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
527 | .swap_activate = nfs_swap_activate, |
528 | .swap_deactivate = nfs_swap_deactivate, | |
1da177e4 LT |
529 | }; |
530 | ||
6b9b3514 DH |
531 | /* |
532 | * Notification that a PTE pointing to an NFS page is about to be made | |
533 | * writable, implying that someone is about to modify the page through a | |
534 | * shared-writable mapping | |
535 | */ | |
c2ec175c | 536 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 537 | { |
c2ec175c | 538 | struct page *page = vmf->page; |
94387fb1 | 539 | struct file *filp = vma->vm_file; |
6de1472f | 540 | struct inode *inode = file_inode(filp); |
94387fb1 | 541 | unsigned pagelen; |
bc4866b6 | 542 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 543 | struct address_space *mapping; |
94387fb1 | 544 | |
1e8968c5 | 545 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", |
6de1472f | 546 | filp, filp->f_mapping->host->i_ino, |
b7eaefaa CL |
547 | (long long)page_offset(page)); |
548 | ||
9a773e7c TM |
549 | sb_start_pagefault(inode->i_sb); |
550 | ||
545db45f | 551 | /* make sure the cache has finished storing the page */ |
6de1472f | 552 | nfs_fscache_wait_on_page_write(NFS_I(inode), page); |
545db45f | 553 | |
ef070dcb TM |
554 | wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING, |
555 | nfs_wait_bit_killable, TASK_KILLABLE); | |
556 | ||
94387fb1 | 557 | lock_page(page); |
d56b4ddf | 558 | mapping = page_file_mapping(page); |
6de1472f | 559 | if (mapping != inode->i_mapping) |
8b1f9ee5 TM |
560 | goto out_unlock; |
561 | ||
2aeb98f4 TM |
562 | wait_on_page_writeback(page); |
563 | ||
94387fb1 | 564 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
565 | if (pagelen == 0) |
566 | goto out_unlock; | |
4899f9c8 | 567 | |
bc4866b6 TM |
568 | ret = VM_FAULT_LOCKED; |
569 | if (nfs_flush_incompatible(filp, page) == 0 && | |
570 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
571 | goto out; | |
8b1f9ee5 | 572 | |
bc4866b6 | 573 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
574 | out_unlock: |
575 | unlock_page(page); | |
bc4866b6 | 576 | out: |
9a773e7c | 577 | sb_end_pagefault(inode->i_sb); |
bc4866b6 | 578 | return ret; |
94387fb1 TM |
579 | } |
580 | ||
f0f37e2f | 581 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 | 582 | .fault = filemap_fault, |
f1820361 | 583 | .map_pages = filemap_map_pages, |
94387fb1 TM |
584 | .page_mkwrite = nfs_vm_page_mkwrite, |
585 | }; | |
586 | ||
7e94d6c4 | 587 | static int nfs_need_check_write(struct file *filp, struct inode *inode) |
7b159fc1 TM |
588 | { |
589 | struct nfs_open_context *ctx; | |
590 | ||
cd3758e3 | 591 | ctx = nfs_file_open_context(filp); |
dc24826b | 592 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) || |
ce52914e | 593 | nfs_ctx_key_to_expire(ctx, inode)) |
7b159fc1 TM |
594 | return 1; |
595 | return 0; | |
596 | } | |
597 | ||
edaf4369 | 598 | ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 599 | { |
6de1472f AV |
600 | struct file *file = iocb->ki_filp; |
601 | struct inode *inode = file_inode(file); | |
7e381172 | 602 | unsigned long written = 0; |
1da177e4 LT |
603 | ssize_t result; |
604 | ||
6de1472f | 605 | result = nfs_key_timeout_notify(file, inode); |
dc24826b AA |
606 | if (result) |
607 | return result; | |
608 | ||
89698b24 | 609 | if (iocb->ki_flags & IOCB_DIRECT) |
65a4a1ca | 610 | return nfs_file_direct_write(iocb, from); |
1da177e4 | 611 | |
619d30b4 | 612 | dprintk("NFS: write(%pD2, %zu@%Ld)\n", |
18290650 | 613 | file, iov_iter_count(from), (long long) iocb->ki_pos); |
1da177e4 | 614 | |
1da177e4 LT |
615 | if (IS_SWAPFILE(inode)) |
616 | goto out_swapfile; | |
7d52e862 TM |
617 | /* |
618 | * O_APPEND implies that we must revalidate the file length. | |
619 | */ | |
2ba48ce5 | 620 | if (iocb->ki_flags & IOCB_APPEND) { |
6de1472f | 621 | result = nfs_revalidate_file_size(inode, file); |
7d52e862 TM |
622 | if (result) |
623 | goto out; | |
fe51beec | 624 | } |
1da177e4 | 625 | |
a5864c99 | 626 | nfs_start_io_write(inode); |
18290650 TM |
627 | result = generic_write_checks(iocb, from); |
628 | if (result > 0) { | |
629 | current->backing_dev_info = inode_to_bdi(inode); | |
630 | result = generic_perform_write(file, from, iocb->ki_pos); | |
631 | current->backing_dev_info = NULL; | |
632 | } | |
a5864c99 | 633 | nfs_end_io_write(inode); |
18290650 | 634 | if (result <= 0) |
1da177e4 LT |
635 | goto out; |
636 | ||
c49edecd TM |
637 | result = generic_write_sync(iocb, result); |
638 | if (result < 0) | |
639 | goto out; | |
640 | written = result; | |
18290650 | 641 | iocb->ki_pos += written; |
7e381172 | 642 | |
7e94d6c4 | 643 | /* Return error values */ |
18290650 | 644 | if (nfs_need_check_write(file, inode)) { |
6de1472f | 645 | int err = vfs_fsync(file, 0); |
200baa21 TM |
646 | if (err < 0) |
647 | result = err; | |
648 | } | |
18290650 | 649 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); |
1da177e4 LT |
650 | out: |
651 | return result; | |
652 | ||
653 | out_swapfile: | |
654 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
18290650 | 655 | return -EBUSY; |
1da177e4 | 656 | } |
89d77c8f | 657 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 658 | |
5eebde23 SJ |
659 | static int |
660 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
661 | { |
662 | struct inode *inode = filp->f_mapping->host; | |
663 | int status = 0; | |
21ac19d4 | 664 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 665 | |
039c4d7a | 666 | /* Try local locking first */ |
6d34ac19 BF |
667 | posix_test_lock(filp, fl); |
668 | if (fl->fl_type != F_UNLCK) { | |
669 | /* found a conflict */ | |
039c4d7a | 670 | goto out; |
1da177e4 | 671 | } |
21ac19d4 | 672 | fl->fl_type = saved_type; |
039c4d7a | 673 | |
011e2a7f | 674 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
675 | goto out_noconflict; |
676 | ||
5eebde23 | 677 | if (is_local) |
039c4d7a TM |
678 | goto out_noconflict; |
679 | ||
680 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
681 | out: | |
1da177e4 | 682 | return status; |
039c4d7a TM |
683 | out_noconflict: |
684 | fl->fl_type = F_UNLCK; | |
685 | goto out; | |
1da177e4 LT |
686 | } |
687 | ||
688 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
689 | { | |
4f656367 | 690 | return locks_lock_file_wait(file, fl); |
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 | |
725 | status = do_vfs_lock(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 |
1da177e4 | 750 | status = do_vfs_lock(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); |