]>
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 | ssize_t |
f0930fff JA |
186 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
187 | struct pipe_inode_info *pipe, size_t count, | |
188 | unsigned int flags) | |
1da177e4 | 189 | { |
6de1472f | 190 | struct inode *inode = file_inode(filp); |
1da177e4 LT |
191 | ssize_t res; |
192 | ||
6de1472f AV |
193 | dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n", |
194 | filp, (unsigned long) count, (unsigned long long) *ppos); | |
1da177e4 | 195 | |
a5864c99 TM |
196 | nfs_start_io_read(inode); |
197 | res = nfs_revalidate_mapping(inode, filp->f_mapping); | |
aa2f1ef1 | 198 | if (!res) { |
f0930fff | 199 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
aa2f1ef1 CL |
200 | if (res > 0) |
201 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); | |
202 | } | |
a5864c99 | 203 | nfs_end_io_read(inode); |
1da177e4 LT |
204 | return res; |
205 | } | |
89d77c8f | 206 | EXPORT_SYMBOL_GPL(nfs_file_splice_read); |
1da177e4 | 207 | |
ce4ef7c0 | 208 | int |
1da177e4 LT |
209 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
210 | { | |
6de1472f | 211 | struct inode *inode = file_inode(file); |
1da177e4 LT |
212 | int status; |
213 | ||
6de1472f | 214 | dprintk("NFS: mmap(%pD2)\n", file); |
1da177e4 | 215 | |
e1ebfd33 TM |
216 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
217 | * so we call that before revalidating the mapping | |
218 | */ | |
219 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
220 | if (!status) { |
221 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 222 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 223 | } |
1da177e4 LT |
224 | return status; |
225 | } | |
89d77c8f | 226 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
227 | |
228 | /* | |
229 | * Flush any dirty pages for this process, and check for write errors. | |
230 | * The return status from this call provides a reliable indication of | |
231 | * whether any write errors occurred for this process. | |
af7fa165 TM |
232 | * |
233 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
234 | * disk, but it retrieves and clears ctx->error after synching, despite | |
235 | * the two being set at the same time in nfs_context_set_write_error(). | |
236 | * This is because the former is used to notify the _next_ call to | |
25985edc | 237 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 238 | * fall back to doing a synchronous write. |
1da177e4 | 239 | */ |
4ff79bc7 | 240 | static int |
a5c58892 | 241 | nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 242 | { |
cd3758e3 | 243 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
6de1472f | 244 | struct inode *inode = file_inode(file); |
05990d1b | 245 | int have_error, do_resend, status; |
af7fa165 TM |
246 | int ret = 0; |
247 | ||
6de1472f | 248 | dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); |
1da177e4 | 249 | |
91d5b470 | 250 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
05990d1b | 251 | do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); |
af7fa165 TM |
252 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
253 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
254 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
05990d1b | 255 | if (have_error) { |
af7fa165 | 256 | ret = xchg(&ctx->error, 0); |
05990d1b TM |
257 | if (ret) |
258 | goto out; | |
259 | } | |
260 | if (status < 0) { | |
af7fa165 | 261 | ret = status; |
05990d1b TM |
262 | goto out; |
263 | } | |
264 | do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); | |
265 | if (do_resend) | |
266 | ret = -EAGAIN; | |
267 | out: | |
a5c58892 BS |
268 | return ret; |
269 | } | |
270 | ||
4ff79bc7 | 271 | int |
a5c58892 BS |
272 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
273 | { | |
274 | int ret; | |
496ad9aa | 275 | struct inode *inode = file_inode(file); |
a5c58892 | 276 | |
f4ce1299 TM |
277 | trace_nfs_fsync_enter(inode); |
278 | ||
05990d1b TM |
279 | do { |
280 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
281 | if (ret != 0) | |
282 | break; | |
05990d1b | 283 | ret = nfs_file_fsync_commit(file, start, end, datasync); |
4ff79bc7 CH |
284 | if (!ret) |
285 | ret = pnfs_sync_inode(inode, !!datasync); | |
dcfc4f25 TM |
286 | /* |
287 | * If nfs_file_fsync_commit detected a server reboot, then | |
288 | * resend all dirty pages that might have been covered by | |
289 | * the NFS_CONTEXT_RESEND_WRITES flag | |
290 | */ | |
291 | start = 0; | |
292 | end = LLONG_MAX; | |
05990d1b TM |
293 | } while (ret == -EAGAIN); |
294 | ||
f4ce1299 | 295 | trace_nfs_fsync_exit(inode, ret); |
af7fa165 | 296 | return ret; |
1da177e4 | 297 | } |
4ff79bc7 | 298 | EXPORT_SYMBOL_GPL(nfs_file_fsync); |
1da177e4 | 299 | |
38c73044 PS |
300 | /* |
301 | * Decide whether a read/modify/write cycle may be more efficient | |
302 | * then a modify/write/read cycle when writing to a page in the | |
303 | * page cache. | |
304 | * | |
305 | * The modify/write/read cycle may occur if a page is read before | |
306 | * being completely filled by the writer. In this situation, the | |
307 | * page must be completely written to stable storage on the server | |
308 | * before it can be refilled by reading in the page from the server. | |
309 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
310 | * done. | |
311 | * | |
312 | * It may be more efficient to read the page first if the file is | |
313 | * open for reading in addition to writing, the page is not marked | |
314 | * as Uptodate, it is not dirty or waiting to be committed, | |
315 | * indicating that it was previously allocated and then modified, | |
316 | * that there were valid bytes of data in that range of the file, | |
317 | * and that the new data won't completely replace the old data in | |
318 | * that range of the file. | |
319 | */ | |
320 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
321 | loff_t pos, unsigned len) | |
322 | { | |
323 | unsigned int pglen = nfs_page_length(page); | |
09cbfeaf | 324 | unsigned int offset = pos & (PAGE_SIZE - 1); |
38c73044 PS |
325 | unsigned int end = offset + len; |
326 | ||
612aa983 CH |
327 | if (pnfs_ld_read_whole_page(file->f_mapping->host)) { |
328 | if (!PageUptodate(page)) | |
329 | return 1; | |
330 | return 0; | |
331 | } | |
332 | ||
38c73044 PS |
333 | if ((file->f_mode & FMODE_READ) && /* open for read? */ |
334 | !PageUptodate(page) && /* Uptodate? */ | |
335 | !PagePrivate(page) && /* i/o request already? */ | |
336 | pglen && /* valid bytes of file? */ | |
337 | (end < pglen || offset)) /* replace all valid bytes? */ | |
338 | return 1; | |
339 | return 0; | |
340 | } | |
341 | ||
1da177e4 | 342 | /* |
4899f9c8 NP |
343 | * This does the "real" work of the write. We must allocate and lock the |
344 | * page to be sent back to the generic routine, which then copies the | |
345 | * data from user space. | |
1da177e4 LT |
346 | * |
347 | * If the writer ends up delaying the write, the writer needs to | |
348 | * increment the page use counts until he is done with the page. | |
349 | */ | |
4899f9c8 NP |
350 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
351 | loff_t pos, unsigned len, unsigned flags, | |
352 | struct page **pagep, void **fsdata) | |
1da177e4 | 353 | { |
4899f9c8 | 354 | int ret; |
09cbfeaf | 355 | pgoff_t index = pos >> PAGE_SHIFT; |
4899f9c8 | 356 | struct page *page; |
38c73044 | 357 | int once_thru = 0; |
4899f9c8 | 358 | |
1e8968c5 | 359 | dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", |
6de1472f | 360 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 361 | |
38c73044 | 362 | start: |
54566b2c | 363 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
364 | if (!page) |
365 | return -ENOMEM; | |
366 | *pagep = page; | |
367 | ||
368 | ret = nfs_flush_incompatible(file, page); | |
369 | if (ret) { | |
370 | unlock_page(page); | |
09cbfeaf | 371 | put_page(page); |
38c73044 PS |
372 | } else if (!once_thru && |
373 | nfs_want_read_modify_write(file, page, pos, len)) { | |
374 | once_thru = 1; | |
375 | ret = nfs_readpage(file, page); | |
09cbfeaf | 376 | put_page(page); |
38c73044 PS |
377 | if (!ret) |
378 | goto start; | |
4899f9c8 NP |
379 | } |
380 | return ret; | |
1da177e4 LT |
381 | } |
382 | ||
4899f9c8 NP |
383 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
384 | loff_t pos, unsigned len, unsigned copied, | |
385 | struct page *page, void *fsdata) | |
1da177e4 | 386 | { |
09cbfeaf | 387 | unsigned offset = pos & (PAGE_SIZE - 1); |
dc24826b | 388 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
4899f9c8 | 389 | int status; |
1da177e4 | 390 | |
1e8968c5 | 391 | dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", |
6de1472f | 392 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 393 | |
efc91ed0 TM |
394 | /* |
395 | * Zero any uninitialised parts of the page, and then mark the page | |
396 | * as up to date if it turns out that we're extending the file. | |
397 | */ | |
398 | if (!PageUptodate(page)) { | |
399 | unsigned pglen = nfs_page_length(page); | |
400 | unsigned end = offset + len; | |
401 | ||
402 | if (pglen == 0) { | |
403 | zero_user_segments(page, 0, offset, | |
09cbfeaf | 404 | end, PAGE_SIZE); |
efc91ed0 TM |
405 | SetPageUptodate(page); |
406 | } else if (end >= pglen) { | |
09cbfeaf | 407 | zero_user_segment(page, end, PAGE_SIZE); |
efc91ed0 TM |
408 | if (offset == 0) |
409 | SetPageUptodate(page); | |
410 | } else | |
09cbfeaf | 411 | zero_user_segment(page, pglen, PAGE_SIZE); |
efc91ed0 TM |
412 | } |
413 | ||
4899f9c8 | 414 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
415 | |
416 | unlock_page(page); | |
09cbfeaf | 417 | put_page(page); |
4899f9c8 | 418 | |
3d509e54 CL |
419 | if (status < 0) |
420 | return status; | |
2701d086 | 421 | NFS_I(mapping->host)->write_io += copied; |
dc24826b | 422 | |
ce52914e | 423 | if (nfs_ctx_key_to_expire(ctx, mapping->host)) { |
dc24826b AA |
424 | status = nfs_wb_all(mapping->host); |
425 | if (status < 0) | |
426 | return status; | |
427 | } | |
428 | ||
3d509e54 | 429 | return copied; |
1da177e4 LT |
430 | } |
431 | ||
6b9b3514 DH |
432 | /* |
433 | * Partially or wholly invalidate a page | |
434 | * - Release the private state associated with a page if undergoing complete | |
435 | * page invalidation | |
545db45f | 436 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
437 | * - Caller holds page lock |
438 | */ | |
d47992f8 LC |
439 | static void nfs_invalidate_page(struct page *page, unsigned int offset, |
440 | unsigned int length) | |
cd52ed35 | 441 | { |
d47992f8 LC |
442 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n", |
443 | page, offset, length); | |
b7eaefaa | 444 | |
09cbfeaf | 445 | if (offset != 0 || length < PAGE_SIZE) |
1c75950b | 446 | return; |
d2ccddf0 | 447 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 448 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
449 | |
450 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
451 | } |
452 | ||
6b9b3514 DH |
453 | /* |
454 | * Attempt to release the private state associated with a page | |
545db45f | 455 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
456 | * - Caller holds page lock |
457 | * - Return true (may release page) or false (may not) | |
458 | */ | |
cd52ed35 TM |
459 | static int nfs_release_page(struct page *page, gfp_t gfp) |
460 | { | |
b7eaefaa CL |
461 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
462 | ||
e3db7691 | 463 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
464 | if (PagePrivate(page)) |
465 | return 0; | |
466 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
467 | } |
468 | ||
f919b196 MG |
469 | static void nfs_check_dirty_writeback(struct page *page, |
470 | bool *dirty, bool *writeback) | |
471 | { | |
472 | struct nfs_inode *nfsi; | |
473 | struct address_space *mapping = page_file_mapping(page); | |
474 | ||
475 | if (!mapping || PageSwapCache(page)) | |
476 | return; | |
477 | ||
478 | /* | |
479 | * Check if an unstable page is currently being committed and | |
480 | * if so, have the VM treat it as if the page is under writeback | |
481 | * so it will not block due to pages that will shortly be freeable. | |
482 | */ | |
483 | nfsi = NFS_I(mapping->host); | |
af7cf057 | 484 | if (atomic_read(&nfsi->commit_info.rpcs_out)) { |
f919b196 MG |
485 | *writeback = true; |
486 | return; | |
487 | } | |
488 | ||
489 | /* | |
490 | * If PagePrivate() is set, then the page is not freeable and as the | |
491 | * inode is not being committed, it's not going to be cleaned in the | |
492 | * near future so treat it as dirty | |
493 | */ | |
494 | if (PagePrivate(page)) | |
495 | *dirty = true; | |
496 | } | |
497 | ||
6b9b3514 DH |
498 | /* |
499 | * Attempt to clear the private state associated with a page when an error | |
500 | * occurs that requires the cached contents of an inode to be written back or | |
501 | * destroyed | |
545db45f | 502 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
503 | * - Caller holds page lock |
504 | * - Return 0 if successful, -error otherwise | |
505 | */ | |
e3db7691 TM |
506 | static int nfs_launder_page(struct page *page) |
507 | { | |
d56b4ddf | 508 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 509 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
510 | |
511 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
512 | inode->i_ino, (long long)page_offset(page)); | |
513 | ||
545db45f | 514 | nfs_fscache_wait_on_page_write(nfsi, page); |
d6c843b9 | 515 | return nfs_wb_launder_page(inode, page); |
cd52ed35 TM |
516 | } |
517 | ||
a564b8f0 MG |
518 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, |
519 | sector_t *span) | |
520 | { | |
dad2b015 JL |
521 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
522 | ||
a564b8f0 | 523 | *span = sis->pages; |
dad2b015 | 524 | |
3c87ef6e | 525 | return rpc_clnt_swap_activate(clnt); |
a564b8f0 MG |
526 | } |
527 | ||
528 | static void nfs_swap_deactivate(struct file *file) | |
529 | { | |
dad2b015 JL |
530 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
531 | ||
3c87ef6e | 532 | rpc_clnt_swap_deactivate(clnt); |
a564b8f0 | 533 | } |
a564b8f0 | 534 | |
f5e54d6e | 535 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
536 | .readpage = nfs_readpage, |
537 | .readpages = nfs_readpages, | |
9cccef95 | 538 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
539 | .writepage = nfs_writepage, |
540 | .writepages = nfs_writepages, | |
4899f9c8 NP |
541 | .write_begin = nfs_write_begin, |
542 | .write_end = nfs_write_end, | |
cd52ed35 TM |
543 | .invalidatepage = nfs_invalidate_page, |
544 | .releasepage = nfs_release_page, | |
1da177e4 | 545 | .direct_IO = nfs_direct_IO, |
f844cd0d | 546 | #ifdef CONFIG_MIGRATION |
074cc1de | 547 | .migratepage = nfs_migrate_page, |
f844cd0d | 548 | #endif |
e3db7691 | 549 | .launder_page = nfs_launder_page, |
f919b196 | 550 | .is_dirty_writeback = nfs_check_dirty_writeback, |
f590f333 | 551 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
552 | .swap_activate = nfs_swap_activate, |
553 | .swap_deactivate = nfs_swap_deactivate, | |
1da177e4 LT |
554 | }; |
555 | ||
6b9b3514 DH |
556 | /* |
557 | * Notification that a PTE pointing to an NFS page is about to be made | |
558 | * writable, implying that someone is about to modify the page through a | |
559 | * shared-writable mapping | |
560 | */ | |
c2ec175c | 561 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 562 | { |
c2ec175c | 563 | struct page *page = vmf->page; |
94387fb1 | 564 | struct file *filp = vma->vm_file; |
6de1472f | 565 | struct inode *inode = file_inode(filp); |
94387fb1 | 566 | unsigned pagelen; |
bc4866b6 | 567 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 568 | struct address_space *mapping; |
94387fb1 | 569 | |
1e8968c5 | 570 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", |
6de1472f | 571 | filp, filp->f_mapping->host->i_ino, |
b7eaefaa CL |
572 | (long long)page_offset(page)); |
573 | ||
9a773e7c TM |
574 | sb_start_pagefault(inode->i_sb); |
575 | ||
545db45f | 576 | /* make sure the cache has finished storing the page */ |
6de1472f | 577 | nfs_fscache_wait_on_page_write(NFS_I(inode), page); |
545db45f | 578 | |
ef070dcb TM |
579 | wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING, |
580 | nfs_wait_bit_killable, TASK_KILLABLE); | |
581 | ||
94387fb1 | 582 | lock_page(page); |
d56b4ddf | 583 | mapping = page_file_mapping(page); |
6de1472f | 584 | if (mapping != inode->i_mapping) |
8b1f9ee5 TM |
585 | goto out_unlock; |
586 | ||
2aeb98f4 TM |
587 | wait_on_page_writeback(page); |
588 | ||
94387fb1 | 589 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
590 | if (pagelen == 0) |
591 | goto out_unlock; | |
4899f9c8 | 592 | |
bc4866b6 TM |
593 | ret = VM_FAULT_LOCKED; |
594 | if (nfs_flush_incompatible(filp, page) == 0 && | |
595 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
596 | goto out; | |
8b1f9ee5 | 597 | |
bc4866b6 | 598 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
599 | out_unlock: |
600 | unlock_page(page); | |
bc4866b6 | 601 | out: |
9a773e7c | 602 | sb_end_pagefault(inode->i_sb); |
bc4866b6 | 603 | return ret; |
94387fb1 TM |
604 | } |
605 | ||
f0f37e2f | 606 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 | 607 | .fault = filemap_fault, |
f1820361 | 608 | .map_pages = filemap_map_pages, |
94387fb1 TM |
609 | .page_mkwrite = nfs_vm_page_mkwrite, |
610 | }; | |
611 | ||
7e94d6c4 | 612 | static int nfs_need_check_write(struct file *filp, struct inode *inode) |
7b159fc1 TM |
613 | { |
614 | struct nfs_open_context *ctx; | |
615 | ||
cd3758e3 | 616 | ctx = nfs_file_open_context(filp); |
dc24826b | 617 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) || |
ce52914e | 618 | nfs_ctx_key_to_expire(ctx, inode)) |
7b159fc1 TM |
619 | return 1; |
620 | return 0; | |
621 | } | |
622 | ||
edaf4369 | 623 | ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 624 | { |
6de1472f AV |
625 | struct file *file = iocb->ki_filp; |
626 | struct inode *inode = file_inode(file); | |
7e381172 | 627 | unsigned long written = 0; |
1da177e4 LT |
628 | ssize_t result; |
629 | ||
6de1472f | 630 | result = nfs_key_timeout_notify(file, inode); |
dc24826b AA |
631 | if (result) |
632 | return result; | |
633 | ||
89698b24 | 634 | if (iocb->ki_flags & IOCB_DIRECT) |
65a4a1ca | 635 | return nfs_file_direct_write(iocb, from); |
1da177e4 | 636 | |
619d30b4 | 637 | dprintk("NFS: write(%pD2, %zu@%Ld)\n", |
18290650 | 638 | file, iov_iter_count(from), (long long) iocb->ki_pos); |
1da177e4 | 639 | |
1da177e4 LT |
640 | if (IS_SWAPFILE(inode)) |
641 | goto out_swapfile; | |
7d52e862 TM |
642 | /* |
643 | * O_APPEND implies that we must revalidate the file length. | |
644 | */ | |
2ba48ce5 | 645 | if (iocb->ki_flags & IOCB_APPEND) { |
6de1472f | 646 | result = nfs_revalidate_file_size(inode, file); |
7d52e862 TM |
647 | if (result) |
648 | goto out; | |
fe51beec | 649 | } |
1da177e4 | 650 | |
a5864c99 | 651 | nfs_start_io_write(inode); |
18290650 TM |
652 | result = generic_write_checks(iocb, from); |
653 | if (result > 0) { | |
654 | current->backing_dev_info = inode_to_bdi(inode); | |
655 | result = generic_perform_write(file, from, iocb->ki_pos); | |
656 | current->backing_dev_info = NULL; | |
657 | } | |
a5864c99 | 658 | nfs_end_io_write(inode); |
18290650 | 659 | if (result <= 0) |
1da177e4 LT |
660 | goto out; |
661 | ||
c49edecd TM |
662 | result = generic_write_sync(iocb, result); |
663 | if (result < 0) | |
664 | goto out; | |
665 | written = result; | |
18290650 | 666 | iocb->ki_pos += written; |
7e381172 | 667 | |
7e94d6c4 | 668 | /* Return error values */ |
18290650 | 669 | if (nfs_need_check_write(file, inode)) { |
6de1472f | 670 | int err = vfs_fsync(file, 0); |
200baa21 TM |
671 | if (err < 0) |
672 | result = err; | |
673 | } | |
18290650 | 674 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); |
1da177e4 LT |
675 | out: |
676 | return result; | |
677 | ||
678 | out_swapfile: | |
679 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
18290650 | 680 | return -EBUSY; |
1da177e4 | 681 | } |
89d77c8f | 682 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 683 | |
5eebde23 SJ |
684 | static int |
685 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
686 | { |
687 | struct inode *inode = filp->f_mapping->host; | |
688 | int status = 0; | |
21ac19d4 | 689 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 690 | |
039c4d7a | 691 | /* Try local locking first */ |
6d34ac19 BF |
692 | posix_test_lock(filp, fl); |
693 | if (fl->fl_type != F_UNLCK) { | |
694 | /* found a conflict */ | |
039c4d7a | 695 | goto out; |
1da177e4 | 696 | } |
21ac19d4 | 697 | fl->fl_type = saved_type; |
039c4d7a | 698 | |
011e2a7f | 699 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
700 | goto out_noconflict; |
701 | ||
5eebde23 | 702 | if (is_local) |
039c4d7a TM |
703 | goto out_noconflict; |
704 | ||
705 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
706 | out: | |
1da177e4 | 707 | return status; |
039c4d7a TM |
708 | out_noconflict: |
709 | fl->fl_type = F_UNLCK; | |
710 | goto out; | |
1da177e4 LT |
711 | } |
712 | ||
713 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
714 | { | |
4f656367 | 715 | return locks_lock_file_wait(file, fl); |
1da177e4 LT |
716 | } |
717 | ||
5eebde23 SJ |
718 | static int |
719 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
720 | { |
721 | struct inode *inode = filp->f_mapping->host; | |
7a8203d8 | 722 | struct nfs_lock_context *l_ctx; |
1da177e4 LT |
723 | int status; |
724 | ||
1da177e4 LT |
725 | /* |
726 | * Flush all pending writes before doing anything | |
727 | * with locks.. | |
728 | */ | |
d9dabc1a | 729 | vfs_fsync(filp, 0); |
1da177e4 | 730 | |
7a8203d8 TM |
731 | l_ctx = nfs_get_lock_context(nfs_file_open_context(filp)); |
732 | if (!IS_ERR(l_ctx)) { | |
210c7c17 | 733 | status = nfs_iocounter_wait(l_ctx); |
7a8203d8 TM |
734 | nfs_put_lock_context(l_ctx); |
735 | if (status < 0) | |
736 | return status; | |
737 | } | |
738 | ||
1da177e4 LT |
739 | /* NOTE: special case |
740 | * If we're signalled while cleaning up locks on process exit, we | |
741 | * still need to complete the unlock. | |
742 | */ | |
5eebde23 SJ |
743 | /* |
744 | * Use local locking if mounted with "-onolock" or with appropriate | |
745 | * "-olocal_lock=" | |
746 | */ | |
747 | if (!is_local) | |
1da177e4 LT |
748 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
749 | else | |
750 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
751 | return status; |
752 | } | |
753 | ||
5eebde23 SJ |
754 | static int |
755 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
756 | { |
757 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
758 | int status; |
759 | ||
1da177e4 LT |
760 | /* |
761 | * Flush all pending writes before doing anything | |
762 | * with locks.. | |
763 | */ | |
29884df0 TM |
764 | status = nfs_sync_mapping(filp->f_mapping); |
765 | if (status != 0) | |
1da177e4 LT |
766 | goto out; |
767 | ||
5eebde23 SJ |
768 | /* |
769 | * Use local locking if mounted with "-onolock" or with appropriate | |
770 | * "-olocal_lock=" | |
771 | */ | |
772 | if (!is_local) | |
1da177e4 | 773 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 774 | else |
1da177e4 | 775 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
776 | if (status < 0) |
777 | goto out; | |
6b96724e | 778 | |
1da177e4 | 779 | /* |
6b96724e RL |
780 | * Revalidate the cache if the server has time stamps granular |
781 | * enough to detect subsecond changes. Otherwise, clear the | |
782 | * cache to prevent missing any changes. | |
783 | * | |
1da177e4 LT |
784 | * This makes locking act as a cache coherency point. |
785 | */ | |
29884df0 | 786 | nfs_sync_mapping(filp->f_mapping); |
ca0daa27 TM |
787 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
788 | nfs_zap_mapping(inode, filp->f_mapping); | |
1da177e4 | 789 | out: |
1da177e4 LT |
790 | return status; |
791 | } | |
792 | ||
793 | /* | |
794 | * Lock a (portion of) a file | |
795 | */ | |
ce4ef7c0 | 796 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 797 | { |
6da24bc9 | 798 | struct inode *inode = filp->f_mapping->host; |
2116271a | 799 | int ret = -ENOLCK; |
5eebde23 | 800 | int is_local = 0; |
1da177e4 | 801 | |
6de1472f AV |
802 | dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n", |
803 | filp, fl->fl_type, fl->fl_flags, | |
1da177e4 | 804 | (long long)fl->fl_start, (long long)fl->fl_end); |
6da24bc9 | 805 | |
91d5b470 | 806 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
807 | |
808 | /* No mandatory locks over NFS */ | |
dfad9441 | 809 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
810 | goto out_err; |
811 | ||
5eebde23 SJ |
812 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
813 | is_local = 1; | |
814 | ||
2116271a TM |
815 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
816 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
817 | if (ret < 0) | |
818 | goto out_err; | |
819 | } | |
1da177e4 LT |
820 | |
821 | if (IS_GETLK(cmd)) | |
5eebde23 | 822 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 823 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 824 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 825 | else |
5eebde23 | 826 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
827 | out_err: |
828 | return ret; | |
1da177e4 | 829 | } |
89d77c8f | 830 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
831 | |
832 | /* | |
833 | * Lock a (portion of) a file | |
834 | */ | |
ce4ef7c0 | 835 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 836 | { |
5eebde23 SJ |
837 | struct inode *inode = filp->f_mapping->host; |
838 | int is_local = 0; | |
839 | ||
6de1472f AV |
840 | dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n", |
841 | filp, fl->fl_type, fl->fl_flags); | |
1da177e4 | 842 | |
1da177e4 LT |
843 | if (!(fl->fl_flags & FL_FLOCK)) |
844 | return -ENOLCK; | |
845 | ||
ad0fcd4e JL |
846 | /* |
847 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
848 | * any standard. In principle we might be able to support LOCK_MAND | |
849 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
850 | * NFS code is not set up for it. | |
851 | */ | |
852 | if (fl->fl_type & LOCK_MAND) | |
853 | return -EINVAL; | |
854 | ||
5eebde23 SJ |
855 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
856 | is_local = 1; | |
857 | ||
1da177e4 | 858 | /* We're simulating flock() locks using posix locks on the server */ |
1da177e4 | 859 | if (fl->fl_type == F_UNLCK) |
5eebde23 SJ |
860 | return do_unlk(filp, cmd, fl, is_local); |
861 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 862 | } |
89d77c8f | 863 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 864 | |
0486958f JL |
865 | const struct file_operations nfs_file_operations = { |
866 | .llseek = nfs_file_llseek, | |
3aa2d199 | 867 | .read_iter = nfs_file_read, |
edaf4369 | 868 | .write_iter = nfs_file_write, |
0486958f JL |
869 | .mmap = nfs_file_mmap, |
870 | .open = nfs_file_open, | |
871 | .flush = nfs_file_flush, | |
872 | .release = nfs_file_release, | |
873 | .fsync = nfs_file_fsync, | |
874 | .lock = nfs_lock, | |
875 | .flock = nfs_flock, | |
876 | .splice_read = nfs_file_splice_read, | |
4da54c21 | 877 | .splice_write = iter_file_splice_write, |
0486958f | 878 | .check_flags = nfs_check_flags, |
1c994a09 | 879 | .setlease = simple_nosetlease, |
0486958f | 880 | }; |
ddda8e0a | 881 | EXPORT_SYMBOL_GPL(nfs_file_operations); |