]>
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> |
e8edc6e0 | 29 | #include <linux/aio.h> |
5a0e3ad6 | 30 | #include <linux/gfp.h> |
b608b283 | 31 | #include <linux/swap.h> |
1da177e4 LT |
32 | |
33 | #include <asm/uaccess.h> | |
1da177e4 LT |
34 | |
35 | #include "delegation.h" | |
94387fb1 | 36 | #include "internal.h" |
91d5b470 | 37 | #include "iostat.h" |
545db45f | 38 | #include "fscache.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); |
46cb650c | 85 | return nfs_release(inode, filp); |
1da177e4 | 86 | } |
89d77c8f | 87 | EXPORT_SYMBOL_GPL(nfs_file_release); |
1da177e4 | 88 | |
980802e3 TM |
89 | /** |
90 | * nfs_revalidate_size - Revalidate the file size | |
91 | * @inode - pointer to inode struct | |
92 | * @file - pointer to struct file | |
93 | * | |
94 | * Revalidates the file length. This is basically a wrapper around | |
95 | * nfs_revalidate_inode() that takes into account the fact that we may | |
96 | * have cached writes (in which case we don't care about the server's | |
97 | * idea of what the file length is), or O_DIRECT (in which case we | |
98 | * shouldn't trust the cache). | |
99 | */ | |
100 | static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) | |
101 | { | |
102 | struct nfs_server *server = NFS_SERVER(inode); | |
103 | struct nfs_inode *nfsi = NFS_I(inode); | |
104 | ||
d7cf8dd0 TM |
105 | if (nfs_have_delegated_attributes(inode)) |
106 | goto out_noreval; | |
107 | ||
980802e3 TM |
108 | if (filp->f_flags & O_DIRECT) |
109 | goto force_reval; | |
d7cf8dd0 TM |
110 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
111 | goto force_reval; | |
112 | if (nfs_attribute_timeout(inode)) | |
113 | goto force_reval; | |
114 | out_noreval: | |
115 | return 0; | |
980802e3 TM |
116 | force_reval: |
117 | return __nfs_revalidate_inode(server, inode); | |
118 | } | |
119 | ||
965c8e59 | 120 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence) |
980802e3 | 121 | { |
6de1472f AV |
122 | dprintk("NFS: llseek file(%pD2, %lld, %d)\n", |
123 | filp, offset, whence); | |
b84e06c5 | 124 | |
06222e49 | 125 | /* |
965c8e59 | 126 | * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate |
06222e49 JB |
127 | * the cached file length |
128 | */ | |
965c8e59 | 129 | if (whence != SEEK_SET && whence != SEEK_CUR) { |
980802e3 | 130 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 131 | |
980802e3 TM |
132 | int retval = nfs_revalidate_file_size(inode, filp); |
133 | if (retval < 0) | |
134 | return (loff_t)retval; | |
79835a71 | 135 | } |
d5e66348 | 136 | |
965c8e59 | 137 | return generic_file_llseek(filp, offset, whence); |
980802e3 | 138 | } |
89d77c8f | 139 | EXPORT_SYMBOL_GPL(nfs_file_llseek); |
980802e3 | 140 | |
1da177e4 LT |
141 | /* |
142 | * Flush all dirty pages, and check for write errors. | |
1da177e4 | 143 | */ |
ce4ef7c0 | 144 | int |
75e1fcc0 | 145 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 146 | { |
6de1472f | 147 | struct inode *inode = file_inode(file); |
1da177e4 | 148 | |
6de1472f | 149 | dprintk("NFS: flush(%pD2)\n", file); |
1da177e4 | 150 | |
c2459dc4 | 151 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
152 | if ((file->f_mode & FMODE_WRITE) == 0) |
153 | return 0; | |
7b159fc1 | 154 | |
14546c33 TM |
155 | /* |
156 | * If we're holding a write delegation, then just start the i/o | |
157 | * but don't wait for completion (or send a commit). | |
158 | */ | |
011e2a7f | 159 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
14546c33 TM |
160 | return filemap_fdatawrite(file->f_mapping); |
161 | ||
7fe5c398 | 162 | /* Flush writes to the server and return any errors */ |
af7fa165 | 163 | return vfs_fsync(file, 0); |
1da177e4 | 164 | } |
89d77c8f | 165 | EXPORT_SYMBOL_GPL(nfs_file_flush); |
1da177e4 | 166 | |
ce4ef7c0 | 167 | ssize_t |
3aa2d199 | 168 | nfs_file_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 169 | { |
6de1472f | 170 | struct inode *inode = file_inode(iocb->ki_filp); |
1da177e4 LT |
171 | ssize_t result; |
172 | ||
1da177e4 | 173 | if (iocb->ki_filp->f_flags & O_DIRECT) |
3aa2d199 | 174 | return nfs_file_direct_read(iocb, to, iocb->ki_pos, true); |
1da177e4 | 175 | |
619d30b4 | 176 | dprintk("NFS: read(%pD2, %zu@%lu)\n", |
6de1472f | 177 | iocb->ki_filp, |
3aa2d199 | 178 | iov_iter_count(to), (unsigned long) iocb->ki_pos); |
1da177e4 | 179 | |
44b11874 | 180 | result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); |
4184dcf2 | 181 | if (!result) { |
3aa2d199 | 182 | result = generic_file_read_iter(iocb, to); |
4184dcf2 CL |
183 | if (result > 0) |
184 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
185 | } | |
1da177e4 LT |
186 | return result; |
187 | } | |
89d77c8f | 188 | EXPORT_SYMBOL_GPL(nfs_file_read); |
1da177e4 | 189 | |
ce4ef7c0 | 190 | ssize_t |
f0930fff JA |
191 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
192 | struct pipe_inode_info *pipe, size_t count, | |
193 | unsigned int flags) | |
1da177e4 | 194 | { |
6de1472f | 195 | struct inode *inode = file_inode(filp); |
1da177e4 LT |
196 | ssize_t res; |
197 | ||
6de1472f AV |
198 | dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n", |
199 | filp, (unsigned long) count, (unsigned long long) *ppos); | |
1da177e4 | 200 | |
44b11874 | 201 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
aa2f1ef1 | 202 | if (!res) { |
f0930fff | 203 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
aa2f1ef1 CL |
204 | if (res > 0) |
205 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); | |
206 | } | |
1da177e4 LT |
207 | return res; |
208 | } | |
89d77c8f | 209 | EXPORT_SYMBOL_GPL(nfs_file_splice_read); |
1da177e4 | 210 | |
ce4ef7c0 | 211 | int |
1da177e4 LT |
212 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
213 | { | |
6de1472f | 214 | struct inode *inode = file_inode(file); |
1da177e4 LT |
215 | int status; |
216 | ||
6de1472f | 217 | dprintk("NFS: mmap(%pD2)\n", file); |
1da177e4 | 218 | |
e1ebfd33 TM |
219 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
220 | * so we call that before revalidating the mapping | |
221 | */ | |
222 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
223 | if (!status) { |
224 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 225 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 226 | } |
1da177e4 LT |
227 | return status; |
228 | } | |
89d77c8f | 229 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
230 | |
231 | /* | |
232 | * Flush any dirty pages for this process, and check for write errors. | |
233 | * The return status from this call provides a reliable indication of | |
234 | * whether any write errors occurred for this process. | |
af7fa165 TM |
235 | * |
236 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
237 | * disk, but it retrieves and clears ctx->error after synching, despite | |
238 | * the two being set at the same time in nfs_context_set_write_error(). | |
239 | * This is because the former is used to notify the _next_ call to | |
25985edc | 240 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 241 | * fall back to doing a synchronous write. |
1da177e4 | 242 | */ |
ce4ef7c0 | 243 | int |
a5c58892 | 244 | nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 245 | { |
cd3758e3 | 246 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
6de1472f | 247 | struct inode *inode = file_inode(file); |
05990d1b | 248 | int have_error, do_resend, status; |
af7fa165 TM |
249 | int ret = 0; |
250 | ||
6de1472f | 251 | dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); |
1da177e4 | 252 | |
91d5b470 | 253 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
05990d1b | 254 | do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); |
af7fa165 TM |
255 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
256 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
257 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
05990d1b | 258 | if (have_error) { |
af7fa165 | 259 | ret = xchg(&ctx->error, 0); |
05990d1b TM |
260 | if (ret) |
261 | goto out; | |
262 | } | |
263 | if (status < 0) { | |
af7fa165 | 264 | ret = status; |
05990d1b TM |
265 | goto out; |
266 | } | |
267 | do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); | |
268 | if (do_resend) | |
269 | ret = -EAGAIN; | |
270 | out: | |
a5c58892 BS |
271 | return ret; |
272 | } | |
89d77c8f | 273 | EXPORT_SYMBOL_GPL(nfs_file_fsync_commit); |
a5c58892 BS |
274 | |
275 | static int | |
276 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) | |
277 | { | |
278 | int ret; | |
496ad9aa | 279 | struct inode *inode = file_inode(file); |
a5c58892 | 280 | |
f4ce1299 TM |
281 | trace_nfs_fsync_enter(inode); |
282 | ||
05990d1b TM |
283 | do { |
284 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
285 | if (ret != 0) | |
286 | break; | |
287 | mutex_lock(&inode->i_mutex); | |
288 | ret = nfs_file_fsync_commit(file, start, end, datasync); | |
289 | mutex_unlock(&inode->i_mutex); | |
dcfc4f25 TM |
290 | /* |
291 | * If nfs_file_fsync_commit detected a server reboot, then | |
292 | * resend all dirty pages that might have been covered by | |
293 | * the NFS_CONTEXT_RESEND_WRITES flag | |
294 | */ | |
295 | start = 0; | |
296 | end = LLONG_MAX; | |
05990d1b TM |
297 | } while (ret == -EAGAIN); |
298 | ||
f4ce1299 | 299 | trace_nfs_fsync_exit(inode, ret); |
af7fa165 | 300 | return ret; |
1da177e4 LT |
301 | } |
302 | ||
38c73044 PS |
303 | /* |
304 | * Decide whether a read/modify/write cycle may be more efficient | |
305 | * then a modify/write/read cycle when writing to a page in the | |
306 | * page cache. | |
307 | * | |
308 | * The modify/write/read cycle may occur if a page is read before | |
309 | * being completely filled by the writer. In this situation, the | |
310 | * page must be completely written to stable storage on the server | |
311 | * before it can be refilled by reading in the page from the server. | |
312 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
313 | * done. | |
314 | * | |
315 | * It may be more efficient to read the page first if the file is | |
316 | * open for reading in addition to writing, the page is not marked | |
317 | * as Uptodate, it is not dirty or waiting to be committed, | |
318 | * indicating that it was previously allocated and then modified, | |
319 | * that there were valid bytes of data in that range of the file, | |
320 | * and that the new data won't completely replace the old data in | |
321 | * that range of the file. | |
322 | */ | |
323 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
324 | loff_t pos, unsigned len) | |
325 | { | |
326 | unsigned int pglen = nfs_page_length(page); | |
327 | unsigned int offset = pos & (PAGE_CACHE_SIZE - 1); | |
328 | unsigned int end = offset + len; | |
329 | ||
330 | if ((file->f_mode & FMODE_READ) && /* open for read? */ | |
331 | !PageUptodate(page) && /* Uptodate? */ | |
332 | !PagePrivate(page) && /* i/o request already? */ | |
333 | pglen && /* valid bytes of file? */ | |
334 | (end < pglen || offset)) /* replace all valid bytes? */ | |
335 | return 1; | |
336 | return 0; | |
337 | } | |
338 | ||
1da177e4 | 339 | /* |
4899f9c8 NP |
340 | * This does the "real" work of the write. We must allocate and lock the |
341 | * page to be sent back to the generic routine, which then copies the | |
342 | * data from user space. | |
1da177e4 LT |
343 | * |
344 | * If the writer ends up delaying the write, the writer needs to | |
345 | * increment the page use counts until he is done with the page. | |
346 | */ | |
4899f9c8 NP |
347 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
348 | loff_t pos, unsigned len, unsigned flags, | |
349 | struct page **pagep, void **fsdata) | |
1da177e4 | 350 | { |
4899f9c8 | 351 | int ret; |
38c73044 | 352 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
4899f9c8 | 353 | struct page *page; |
38c73044 | 354 | int once_thru = 0; |
4899f9c8 | 355 | |
1e8968c5 | 356 | dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", |
6de1472f | 357 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 358 | |
38c73044 | 359 | start: |
72cb77f4 TM |
360 | /* |
361 | * Prevent starvation issues if someone is doing a consistency | |
362 | * sync-to-disk | |
363 | */ | |
74316201 N |
364 | ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, |
365 | nfs_wait_bit_killable, TASK_KILLABLE); | |
72cb77f4 TM |
366 | if (ret) |
367 | return ret; | |
368 | ||
54566b2c | 369 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
370 | if (!page) |
371 | return -ENOMEM; | |
372 | *pagep = page; | |
373 | ||
374 | ret = nfs_flush_incompatible(file, page); | |
375 | if (ret) { | |
376 | unlock_page(page); | |
377 | page_cache_release(page); | |
38c73044 PS |
378 | } else if (!once_thru && |
379 | nfs_want_read_modify_write(file, page, pos, len)) { | |
380 | once_thru = 1; | |
381 | ret = nfs_readpage(file, page); | |
382 | page_cache_release(page); | |
383 | if (!ret) | |
384 | goto start; | |
4899f9c8 NP |
385 | } |
386 | return ret; | |
1da177e4 LT |
387 | } |
388 | ||
4899f9c8 NP |
389 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
390 | loff_t pos, unsigned len, unsigned copied, | |
391 | struct page *page, void *fsdata) | |
1da177e4 | 392 | { |
4899f9c8 | 393 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
dc24826b | 394 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
4899f9c8 | 395 | int status; |
1da177e4 | 396 | |
1e8968c5 | 397 | dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", |
6de1472f | 398 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 399 | |
efc91ed0 TM |
400 | /* |
401 | * Zero any uninitialised parts of the page, and then mark the page | |
402 | * as up to date if it turns out that we're extending the file. | |
403 | */ | |
404 | if (!PageUptodate(page)) { | |
405 | unsigned pglen = nfs_page_length(page); | |
406 | unsigned end = offset + len; | |
407 | ||
408 | if (pglen == 0) { | |
409 | zero_user_segments(page, 0, offset, | |
410 | end, PAGE_CACHE_SIZE); | |
411 | SetPageUptodate(page); | |
412 | } else if (end >= pglen) { | |
413 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
414 | if (offset == 0) | |
415 | SetPageUptodate(page); | |
416 | } else | |
417 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
418 | } | |
419 | ||
4899f9c8 | 420 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
421 | |
422 | unlock_page(page); | |
423 | page_cache_release(page); | |
424 | ||
3d509e54 CL |
425 | if (status < 0) |
426 | return status; | |
2701d086 | 427 | NFS_I(mapping->host)->write_io += copied; |
dc24826b AA |
428 | |
429 | if (nfs_ctx_key_to_expire(ctx)) { | |
430 | status = nfs_wb_all(mapping->host); | |
431 | if (status < 0) | |
432 | return status; | |
433 | } | |
434 | ||
3d509e54 | 435 | return copied; |
1da177e4 LT |
436 | } |
437 | ||
6b9b3514 DH |
438 | /* |
439 | * Partially or wholly invalidate a page | |
440 | * - Release the private state associated with a page if undergoing complete | |
441 | * page invalidation | |
545db45f | 442 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
443 | * - Caller holds page lock |
444 | */ | |
d47992f8 LC |
445 | static void nfs_invalidate_page(struct page *page, unsigned int offset, |
446 | unsigned int length) | |
cd52ed35 | 447 | { |
d47992f8 LC |
448 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n", |
449 | page, offset, length); | |
b7eaefaa | 450 | |
d47992f8 | 451 | if (offset != 0 || length < PAGE_CACHE_SIZE) |
1c75950b | 452 | return; |
d2ccddf0 | 453 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 454 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
455 | |
456 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
457 | } |
458 | ||
6b9b3514 DH |
459 | /* |
460 | * Attempt to release the private state associated with a page | |
545db45f | 461 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
462 | * - Caller holds page lock |
463 | * - Return true (may release page) or false (may not) | |
464 | */ | |
cd52ed35 TM |
465 | static int nfs_release_page(struct page *page, gfp_t gfp) |
466 | { | |
b608b283 TM |
467 | struct address_space *mapping = page->mapping; |
468 | ||
b7eaefaa CL |
469 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
470 | ||
5cf02d09 JL |
471 | /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not |
472 | * doing this memory reclaim for a fs-related allocation. | |
473 | */ | |
474 | if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL && | |
475 | !(current->flags & PF_FSTRANS)) { | |
b608b283 TM |
476 | int how = FLUSH_SYNC; |
477 | ||
478 | /* Don't let kswapd deadlock waiting for OOM RPC calls */ | |
479 | if (current_is_kswapd()) | |
480 | how = 0; | |
481 | nfs_commit_inode(mapping->host, how); | |
482 | } | |
e3db7691 | 483 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
484 | if (PagePrivate(page)) |
485 | return 0; | |
486 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
487 | } |
488 | ||
f919b196 MG |
489 | static void nfs_check_dirty_writeback(struct page *page, |
490 | bool *dirty, bool *writeback) | |
491 | { | |
492 | struct nfs_inode *nfsi; | |
493 | struct address_space *mapping = page_file_mapping(page); | |
494 | ||
495 | if (!mapping || PageSwapCache(page)) | |
496 | return; | |
497 | ||
498 | /* | |
499 | * Check if an unstable page is currently being committed and | |
500 | * if so, have the VM treat it as if the page is under writeback | |
501 | * so it will not block due to pages that will shortly be freeable. | |
502 | */ | |
503 | nfsi = NFS_I(mapping->host); | |
504 | if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) { | |
505 | *writeback = true; | |
506 | return; | |
507 | } | |
508 | ||
509 | /* | |
510 | * If PagePrivate() is set, then the page is not freeable and as the | |
511 | * inode is not being committed, it's not going to be cleaned in the | |
512 | * near future so treat it as dirty | |
513 | */ | |
514 | if (PagePrivate(page)) | |
515 | *dirty = true; | |
516 | } | |
517 | ||
6b9b3514 DH |
518 | /* |
519 | * Attempt to clear the private state associated with a page when an error | |
520 | * occurs that requires the cached contents of an inode to be written back or | |
521 | * destroyed | |
545db45f | 522 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
523 | * - Caller holds page lock |
524 | * - Return 0 if successful, -error otherwise | |
525 | */ | |
e3db7691 TM |
526 | static int nfs_launder_page(struct page *page) |
527 | { | |
d56b4ddf | 528 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 529 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
530 | |
531 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
532 | inode->i_ino, (long long)page_offset(page)); | |
533 | ||
545db45f | 534 | nfs_fscache_wait_on_page_write(nfsi, page); |
b7eaefaa | 535 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
536 | } |
537 | ||
a564b8f0 MG |
538 | #ifdef CONFIG_NFS_SWAP |
539 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
540 | sector_t *span) | |
541 | { | |
542 | *span = sis->pages; | |
543 | return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1); | |
544 | } | |
545 | ||
546 | static void nfs_swap_deactivate(struct file *file) | |
547 | { | |
548 | xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0); | |
549 | } | |
550 | #endif | |
551 | ||
f5e54d6e | 552 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
553 | .readpage = nfs_readpage, |
554 | .readpages = nfs_readpages, | |
9cccef95 | 555 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
556 | .writepage = nfs_writepage, |
557 | .writepages = nfs_writepages, | |
4899f9c8 NP |
558 | .write_begin = nfs_write_begin, |
559 | .write_end = nfs_write_end, | |
cd52ed35 TM |
560 | .invalidatepage = nfs_invalidate_page, |
561 | .releasepage = nfs_release_page, | |
1da177e4 | 562 | .direct_IO = nfs_direct_IO, |
074cc1de | 563 | .migratepage = nfs_migrate_page, |
e3db7691 | 564 | .launder_page = nfs_launder_page, |
f919b196 | 565 | .is_dirty_writeback = nfs_check_dirty_writeback, |
f590f333 | 566 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
567 | #ifdef CONFIG_NFS_SWAP |
568 | .swap_activate = nfs_swap_activate, | |
569 | .swap_deactivate = nfs_swap_deactivate, | |
570 | #endif | |
1da177e4 LT |
571 | }; |
572 | ||
6b9b3514 DH |
573 | /* |
574 | * Notification that a PTE pointing to an NFS page is about to be made | |
575 | * writable, implying that someone is about to modify the page through a | |
576 | * shared-writable mapping | |
577 | */ | |
c2ec175c | 578 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 579 | { |
c2ec175c | 580 | struct page *page = vmf->page; |
94387fb1 | 581 | struct file *filp = vma->vm_file; |
6de1472f | 582 | struct inode *inode = file_inode(filp); |
94387fb1 | 583 | unsigned pagelen; |
bc4866b6 | 584 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 585 | struct address_space *mapping; |
94387fb1 | 586 | |
1e8968c5 | 587 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", |
6de1472f | 588 | filp, filp->f_mapping->host->i_ino, |
b7eaefaa CL |
589 | (long long)page_offset(page)); |
590 | ||
545db45f | 591 | /* make sure the cache has finished storing the page */ |
6de1472f | 592 | nfs_fscache_wait_on_page_write(NFS_I(inode), page); |
545db45f | 593 | |
94387fb1 | 594 | lock_page(page); |
d56b4ddf | 595 | mapping = page_file_mapping(page); |
6de1472f | 596 | if (mapping != inode->i_mapping) |
8b1f9ee5 TM |
597 | goto out_unlock; |
598 | ||
2aeb98f4 TM |
599 | wait_on_page_writeback(page); |
600 | ||
94387fb1 | 601 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
602 | if (pagelen == 0) |
603 | goto out_unlock; | |
4899f9c8 | 604 | |
bc4866b6 TM |
605 | ret = VM_FAULT_LOCKED; |
606 | if (nfs_flush_incompatible(filp, page) == 0 && | |
607 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
608 | goto out; | |
8b1f9ee5 | 609 | |
bc4866b6 | 610 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
611 | out_unlock: |
612 | unlock_page(page); | |
bc4866b6 TM |
613 | out: |
614 | return ret; | |
94387fb1 TM |
615 | } |
616 | ||
f0f37e2f | 617 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 | 618 | .fault = filemap_fault, |
f1820361 | 619 | .map_pages = filemap_map_pages, |
94387fb1 | 620 | .page_mkwrite = nfs_vm_page_mkwrite, |
0b173bc4 | 621 | .remap_pages = generic_file_remap_pages, |
94387fb1 TM |
622 | }; |
623 | ||
7b159fc1 TM |
624 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
625 | { | |
626 | struct nfs_open_context *ctx; | |
627 | ||
6b2f3d1f | 628 | if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) |
7b159fc1 | 629 | return 1; |
cd3758e3 | 630 | ctx = nfs_file_open_context(filp); |
dc24826b AA |
631 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) || |
632 | nfs_ctx_key_to_expire(ctx)) | |
7b159fc1 TM |
633 | return 1; |
634 | return 0; | |
635 | } | |
636 | ||
edaf4369 | 637 | ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 638 | { |
6de1472f AV |
639 | struct file *file = iocb->ki_filp; |
640 | struct inode *inode = file_inode(file); | |
7e381172 | 641 | unsigned long written = 0; |
1da177e4 | 642 | ssize_t result; |
edaf4369 AV |
643 | size_t count = iov_iter_count(from); |
644 | loff_t pos = iocb->ki_pos; | |
1da177e4 | 645 | |
6de1472f | 646 | result = nfs_key_timeout_notify(file, inode); |
dc24826b AA |
647 | if (result) |
648 | return result; | |
649 | ||
6de1472f | 650 | if (file->f_flags & O_DIRECT) |
edaf4369 | 651 | return nfs_file_direct_write(iocb, from, pos, true); |
1da177e4 | 652 | |
619d30b4 AV |
653 | dprintk("NFS: write(%pD2, %zu@%Ld)\n", |
654 | file, count, (long long) pos); | |
1da177e4 LT |
655 | |
656 | result = -EBUSY; | |
657 | if (IS_SWAPFILE(inode)) | |
658 | goto out_swapfile; | |
7d52e862 TM |
659 | /* |
660 | * O_APPEND implies that we must revalidate the file length. | |
661 | */ | |
6de1472f AV |
662 | if (file->f_flags & O_APPEND) { |
663 | result = nfs_revalidate_file_size(inode, file); | |
7d52e862 TM |
664 | if (result) |
665 | goto out; | |
fe51beec | 666 | } |
1da177e4 LT |
667 | |
668 | result = count; | |
669 | if (!count) | |
670 | goto out; | |
671 | ||
edaf4369 | 672 | result = generic_file_write_iter(iocb, from); |
7e381172 CL |
673 | if (result > 0) |
674 | written = result; | |
675 | ||
6b2f3d1f | 676 | /* Return error values for O_DSYNC and IS_SYNC() */ |
6de1472f AV |
677 | if (result >= 0 && nfs_need_sync_write(file, inode)) { |
678 | int err = vfs_fsync(file, 0); | |
200baa21 TM |
679 | if (err < 0) |
680 | result = err; | |
681 | } | |
7e381172 CL |
682 | if (result > 0) |
683 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
1da177e4 LT |
684 | out: |
685 | return result; | |
686 | ||
687 | out_swapfile: | |
688 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
689 | goto out; | |
690 | } | |
89d77c8f | 691 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 692 | |
5eebde23 SJ |
693 | static int |
694 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
695 | { |
696 | struct inode *inode = filp->f_mapping->host; | |
697 | int status = 0; | |
21ac19d4 | 698 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 699 | |
039c4d7a | 700 | /* Try local locking first */ |
6d34ac19 BF |
701 | posix_test_lock(filp, fl); |
702 | if (fl->fl_type != F_UNLCK) { | |
703 | /* found a conflict */ | |
039c4d7a | 704 | goto out; |
1da177e4 | 705 | } |
21ac19d4 | 706 | fl->fl_type = saved_type; |
039c4d7a | 707 | |
011e2a7f | 708 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
709 | goto out_noconflict; |
710 | ||
5eebde23 | 711 | if (is_local) |
039c4d7a TM |
712 | goto out_noconflict; |
713 | ||
714 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
715 | out: | |
1da177e4 | 716 | return status; |
039c4d7a TM |
717 | out_noconflict: |
718 | fl->fl_type = F_UNLCK; | |
719 | goto out; | |
1da177e4 LT |
720 | } |
721 | ||
722 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
723 | { | |
724 | int res = 0; | |
725 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
726 | case FL_POSIX: | |
727 | res = posix_lock_file_wait(file, fl); | |
728 | break; | |
729 | case FL_FLOCK: | |
730 | res = flock_lock_file_wait(file, fl); | |
731 | break; | |
732 | default: | |
733 | BUG(); | |
734 | } | |
1da177e4 LT |
735 | return res; |
736 | } | |
737 | ||
5eebde23 SJ |
738 | static int |
739 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
740 | { |
741 | struct inode *inode = filp->f_mapping->host; | |
7a8203d8 | 742 | struct nfs_lock_context *l_ctx; |
1da177e4 LT |
743 | int status; |
744 | ||
1da177e4 LT |
745 | /* |
746 | * Flush all pending writes before doing anything | |
747 | * with locks.. | |
748 | */ | |
29884df0 | 749 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 | 750 | |
7a8203d8 TM |
751 | l_ctx = nfs_get_lock_context(nfs_file_open_context(filp)); |
752 | if (!IS_ERR(l_ctx)) { | |
753 | status = nfs_iocounter_wait(&l_ctx->io_count); | |
754 | nfs_put_lock_context(l_ctx); | |
755 | if (status < 0) | |
756 | return status; | |
757 | } | |
758 | ||
1da177e4 LT |
759 | /* NOTE: special case |
760 | * If we're signalled while cleaning up locks on process exit, we | |
761 | * still need to complete the unlock. | |
762 | */ | |
5eebde23 SJ |
763 | /* |
764 | * Use local locking if mounted with "-onolock" or with appropriate | |
765 | * "-olocal_lock=" | |
766 | */ | |
767 | if (!is_local) | |
1da177e4 LT |
768 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
769 | else | |
770 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
771 | return status; |
772 | } | |
773 | ||
6b96724e RL |
774 | static int |
775 | is_time_granular(struct timespec *ts) { | |
776 | return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000)); | |
777 | } | |
778 | ||
5eebde23 SJ |
779 | static int |
780 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
781 | { |
782 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
783 | int status; |
784 | ||
1da177e4 LT |
785 | /* |
786 | * Flush all pending writes before doing anything | |
787 | * with locks.. | |
788 | */ | |
29884df0 TM |
789 | status = nfs_sync_mapping(filp->f_mapping); |
790 | if (status != 0) | |
1da177e4 LT |
791 | goto out; |
792 | ||
5eebde23 SJ |
793 | /* |
794 | * Use local locking if mounted with "-onolock" or with appropriate | |
795 | * "-olocal_lock=" | |
796 | */ | |
797 | if (!is_local) | |
1da177e4 | 798 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 799 | else |
1da177e4 | 800 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
801 | if (status < 0) |
802 | goto out; | |
6b96724e | 803 | |
1da177e4 | 804 | /* |
6b96724e RL |
805 | * Revalidate the cache if the server has time stamps granular |
806 | * enough to detect subsecond changes. Otherwise, clear the | |
807 | * cache to prevent missing any changes. | |
808 | * | |
1da177e4 LT |
809 | * This makes locking act as a cache coherency point. |
810 | */ | |
29884df0 | 811 | nfs_sync_mapping(filp->f_mapping); |
011e2a7f | 812 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { |
6b96724e RL |
813 | if (is_time_granular(&NFS_SERVER(inode)->time_delta)) |
814 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
815 | else | |
816 | nfs_zap_caches(inode); | |
817 | } | |
1da177e4 | 818 | out: |
1da177e4 LT |
819 | return status; |
820 | } | |
821 | ||
822 | /* | |
823 | * Lock a (portion of) a file | |
824 | */ | |
ce4ef7c0 | 825 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 826 | { |
6da24bc9 | 827 | struct inode *inode = filp->f_mapping->host; |
2116271a | 828 | int ret = -ENOLCK; |
5eebde23 | 829 | int is_local = 0; |
1da177e4 | 830 | |
6de1472f AV |
831 | dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n", |
832 | filp, fl->fl_type, fl->fl_flags, | |
1da177e4 | 833 | (long long)fl->fl_start, (long long)fl->fl_end); |
6da24bc9 | 834 | |
91d5b470 | 835 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
836 | |
837 | /* No mandatory locks over NFS */ | |
dfad9441 | 838 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
839 | goto out_err; |
840 | ||
5eebde23 SJ |
841 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
842 | is_local = 1; | |
843 | ||
2116271a TM |
844 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
845 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
846 | if (ret < 0) | |
847 | goto out_err; | |
848 | } | |
1da177e4 LT |
849 | |
850 | if (IS_GETLK(cmd)) | |
5eebde23 | 851 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 852 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 853 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 854 | else |
5eebde23 | 855 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
856 | out_err: |
857 | return ret; | |
1da177e4 | 858 | } |
89d77c8f | 859 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
860 | |
861 | /* | |
862 | * Lock a (portion of) a file | |
863 | */ | |
ce4ef7c0 | 864 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 865 | { |
5eebde23 SJ |
866 | struct inode *inode = filp->f_mapping->host; |
867 | int is_local = 0; | |
868 | ||
6de1472f AV |
869 | dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n", |
870 | filp, fl->fl_type, fl->fl_flags); | |
1da177e4 | 871 | |
1da177e4 LT |
872 | if (!(fl->fl_flags & FL_FLOCK)) |
873 | return -ENOLCK; | |
874 | ||
ad0fcd4e JL |
875 | /* |
876 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
877 | * any standard. In principle we might be able to support LOCK_MAND | |
878 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
879 | * NFS code is not set up for it. | |
880 | */ | |
881 | if (fl->fl_type & LOCK_MAND) | |
882 | return -EINVAL; | |
883 | ||
5eebde23 SJ |
884 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
885 | is_local = 1; | |
886 | ||
1da177e4 | 887 | /* We're simulating flock() locks using posix locks on the server */ |
1da177e4 | 888 | if (fl->fl_type == F_UNLCK) |
5eebde23 SJ |
889 | return do_unlk(filp, cmd, fl, is_local); |
890 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 891 | } |
89d77c8f | 892 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 893 | |
6da24bc9 CL |
894 | /* |
895 | * There is no protocol support for leases, so we have no way to implement | |
896 | * them correctly in the face of opens by other clients. | |
897 | */ | |
ce4ef7c0 | 898 | int nfs_setlease(struct file *file, long arg, struct file_lock **fl) |
370f6599 | 899 | { |
6de1472f | 900 | dprintk("NFS: setlease(%pD2, arg=%ld)\n", file, arg); |
370f6599 BF |
901 | return -EINVAL; |
902 | } | |
89d77c8f | 903 | EXPORT_SYMBOL_GPL(nfs_setlease); |
1788ea6e | 904 | |
0486958f JL |
905 | const struct file_operations nfs_file_operations = { |
906 | .llseek = nfs_file_llseek, | |
3aa2d199 | 907 | .read = new_sync_read, |
edaf4369 | 908 | .write = new_sync_write, |
3aa2d199 | 909 | .read_iter = nfs_file_read, |
edaf4369 | 910 | .write_iter = nfs_file_write, |
0486958f JL |
911 | .mmap = nfs_file_mmap, |
912 | .open = nfs_file_open, | |
913 | .flush = nfs_file_flush, | |
914 | .release = nfs_file_release, | |
915 | .fsync = nfs_file_fsync, | |
916 | .lock = nfs_lock, | |
917 | .flock = nfs_flock, | |
918 | .splice_read = nfs_file_splice_read, | |
4da54c21 | 919 | .splice_write = iter_file_splice_write, |
0486958f JL |
920 | .check_flags = nfs_check_flags, |
921 | .setlease = nfs_setlease, | |
922 | }; | |
ddda8e0a | 923 | EXPORT_SYMBOL_GPL(nfs_file_operations); |