4 * Copyright (C) 1992 Rick Sladkey
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.
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
16 * nfs regular file handling functions
19 #include <linux/time.h>
20 #include <linux/kernel.h>
21 #include <linux/errno.h>
22 #include <linux/fcntl.h>
23 #include <linux/stat.h>
24 #include <linux/nfs_fs.h>
25 #include <linux/nfs_mount.h>
27 #include <linux/slab.h>
28 #include <linux/pagemap.h>
29 #include <linux/smp_lock.h>
31 #include <asm/uaccess.h>
32 #include <asm/system.h>
34 #include "delegation.h"
37 #define NFSDBG_FACILITY NFSDBG_FILE
39 static int nfs_file_open(struct inode
*, struct file
*);
40 static int nfs_file_release(struct inode
*, struct file
*);
41 static loff_t
nfs_file_llseek(struct file
*file
, loff_t offset
, int origin
);
42 static int nfs_file_mmap(struct file
*, struct vm_area_struct
*);
43 static ssize_t
nfs_file_sendfile(struct file
*, loff_t
*, size_t, read_actor_t
, void *);
44 static ssize_t
nfs_file_read(struct kiocb
*, const struct iovec
*iov
,
45 unsigned long nr_segs
, loff_t pos
);
46 static ssize_t
nfs_file_write(struct kiocb
*, const struct iovec
*iov
,
47 unsigned long nr_segs
, loff_t pos
);
48 static int nfs_file_flush(struct file
*, fl_owner_t id
);
49 static int nfs_fsync(struct file
*, struct dentry
*dentry
, int datasync
);
50 static int nfs_check_flags(int flags
);
51 static int nfs_lock(struct file
*filp
, int cmd
, struct file_lock
*fl
);
52 static int nfs_flock(struct file
*filp
, int cmd
, struct file_lock
*fl
);
54 const struct file_operations nfs_file_operations
= {
55 .llseek
= nfs_file_llseek
,
57 .write
= do_sync_write
,
58 .aio_read
= nfs_file_read
,
59 .aio_write
= nfs_file_write
,
60 .mmap
= nfs_file_mmap
,
61 .open
= nfs_file_open
,
62 .flush
= nfs_file_flush
,
63 .release
= nfs_file_release
,
67 .sendfile
= nfs_file_sendfile
,
68 .check_flags
= nfs_check_flags
,
71 struct inode_operations nfs_file_inode_operations
= {
72 .permission
= nfs_permission
,
73 .getattr
= nfs_getattr
,
74 .setattr
= nfs_setattr
,
78 struct inode_operations nfs3_file_inode_operations
= {
79 .permission
= nfs_permission
,
80 .getattr
= nfs_getattr
,
81 .setattr
= nfs_setattr
,
82 .listxattr
= nfs3_listxattr
,
83 .getxattr
= nfs3_getxattr
,
84 .setxattr
= nfs3_setxattr
,
85 .removexattr
= nfs3_removexattr
,
87 #endif /* CONFIG_NFS_v3 */
89 /* Hack for future NFS swap support */
91 # define IS_SWAPFILE(inode) (0)
94 static int nfs_check_flags(int flags
)
96 if ((flags
& (O_APPEND
| O_DIRECT
)) == (O_APPEND
| O_DIRECT
))
106 nfs_file_open(struct inode
*inode
, struct file
*filp
)
110 res
= nfs_check_flags(filp
->f_flags
);
114 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
116 res
= NFS_PROTO(inode
)->file_open(inode
, filp
);
122 nfs_file_release(struct inode
*inode
, struct file
*filp
)
124 /* Ensure that dirty pages are flushed out with the right creds */
125 if (filp
->f_mode
& FMODE_WRITE
)
126 filemap_fdatawrite(filp
->f_mapping
);
127 nfs_inc_stats(inode
, NFSIOS_VFSRELEASE
);
128 return NFS_PROTO(inode
)->file_release(inode
, filp
);
132 * nfs_revalidate_size - Revalidate the file size
133 * @inode - pointer to inode struct
134 * @file - pointer to struct file
136 * Revalidates the file length. This is basically a wrapper around
137 * nfs_revalidate_inode() that takes into account the fact that we may
138 * have cached writes (in which case we don't care about the server's
139 * idea of what the file length is), or O_DIRECT (in which case we
140 * shouldn't trust the cache).
142 static int nfs_revalidate_file_size(struct inode
*inode
, struct file
*filp
)
144 struct nfs_server
*server
= NFS_SERVER(inode
);
145 struct nfs_inode
*nfsi
= NFS_I(inode
);
147 if (server
->flags
& NFS_MOUNT_NOAC
)
149 if (filp
->f_flags
& O_DIRECT
)
151 if (nfsi
->npages
!= 0)
153 if (!(nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
) && !nfs_attribute_timeout(inode
))
156 return __nfs_revalidate_inode(server
, inode
);
159 static loff_t
nfs_file_llseek(struct file
*filp
, loff_t offset
, int origin
)
161 /* origin == SEEK_END => we must revalidate the cached file length */
162 if (origin
== SEEK_END
) {
163 struct inode
*inode
= filp
->f_mapping
->host
;
164 int retval
= nfs_revalidate_file_size(inode
, filp
);
166 return (loff_t
)retval
;
168 return remote_llseek(filp
, offset
, origin
);
172 * Flush all dirty pages, and check for write errors.
176 nfs_file_flush(struct file
*file
, fl_owner_t id
)
178 struct nfs_open_context
*ctx
= (struct nfs_open_context
*)file
->private_data
;
179 struct inode
*inode
= file
->f_dentry
->d_inode
;
182 dfprintk(VFS
, "nfs: flush(%s/%ld)\n", inode
->i_sb
->s_id
, inode
->i_ino
);
184 if ((file
->f_mode
& FMODE_WRITE
) == 0)
186 nfs_inc_stats(inode
, NFSIOS_VFSFLUSH
);
188 /* Ensure that data+attribute caches are up to date after close() */
189 status
= nfs_wb_all(inode
);
194 nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
201 nfs_file_read(struct kiocb
*iocb
, const struct iovec
*iov
,
202 unsigned long nr_segs
, loff_t pos
)
204 struct dentry
* dentry
= iocb
->ki_filp
->f_dentry
;
205 struct inode
* inode
= dentry
->d_inode
;
207 size_t count
= iov_length(iov
, nr_segs
);
209 #ifdef CONFIG_NFS_DIRECTIO
210 if (iocb
->ki_filp
->f_flags
& O_DIRECT
)
211 return nfs_file_direct_read(iocb
, iov
, nr_segs
, pos
);
214 dfprintk(VFS
, "nfs: read(%s/%s, %lu@%lu)\n",
215 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
216 (unsigned long) count
, (unsigned long) pos
);
218 result
= nfs_revalidate_mapping(inode
, iocb
->ki_filp
->f_mapping
);
219 nfs_add_stats(inode
, NFSIOS_NORMALREADBYTES
, count
);
221 result
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
226 nfs_file_sendfile(struct file
*filp
, loff_t
*ppos
, size_t count
,
227 read_actor_t actor
, void *target
)
229 struct dentry
*dentry
= filp
->f_dentry
;
230 struct inode
*inode
= dentry
->d_inode
;
233 dfprintk(VFS
, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
234 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
235 (unsigned long) count
, (unsigned long long) *ppos
);
237 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
239 res
= generic_file_sendfile(filp
, ppos
, count
, actor
, target
);
244 nfs_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
246 struct dentry
*dentry
= file
->f_dentry
;
247 struct inode
*inode
= dentry
->d_inode
;
250 dfprintk(VFS
, "nfs: mmap(%s/%s)\n",
251 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
253 status
= nfs_revalidate_mapping(inode
, file
->f_mapping
);
255 status
= generic_file_mmap(file
, vma
);
260 * Flush any dirty pages for this process, and check for write errors.
261 * The return status from this call provides a reliable indication of
262 * whether any write errors occurred for this process.
265 nfs_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
267 struct nfs_open_context
*ctx
= (struct nfs_open_context
*)file
->private_data
;
268 struct inode
*inode
= dentry
->d_inode
;
271 dfprintk(VFS
, "nfs: fsync(%s/%ld)\n", inode
->i_sb
->s_id
, inode
->i_ino
);
273 nfs_inc_stats(inode
, NFSIOS_VFSFSYNC
);
275 status
= nfs_wb_all(inode
);
285 * This does the "real" work of the write. The generic routine has
286 * allocated the page, locked it, done all the page alignment stuff
287 * calculations etc. Now we should just copy the data from user
288 * space and write it back to the real medium..
290 * If the writer ends up delaying the write, the writer needs to
291 * increment the page use counts until he is done with the page.
293 static int nfs_prepare_write(struct file
*file
, struct page
*page
, unsigned offset
, unsigned to
)
295 return nfs_flush_incompatible(file
, page
);
298 static int nfs_commit_write(struct file
*file
, struct page
*page
, unsigned offset
, unsigned to
)
303 status
= nfs_updatepage(file
, page
, offset
, to
-offset
);
308 static void nfs_invalidate_page(struct page
*page
, unsigned long offset
)
310 #if 0 // mask by Victor Yu. 02-12-2007
311 struct inode
*inode
= page
->mapping
->host
;
313 struct inode
*inode
= page
->u
.xx
.mapping
->host
;
316 /* Cancel any unstarted writes on this page */
318 nfs_sync_inode_wait(inode
, page
->index
, 1, FLUSH_INVALIDATE
);
321 static int nfs_release_page(struct page
*page
, gfp_t gfp
)
324 #if 0 // mask by Victor Yu. 02-12-2007
325 return !nfs_wb_page(page
->mapping
->host
, page
);
327 return !nfs_wb_page(page
->u
.xx
.mapping
->host
, page
);
331 * Avoid deadlock on nfs_wait_on_request().
336 const struct address_space_operations nfs_file_aops
= {
337 .readpage
= nfs_readpage
,
338 .readpages
= nfs_readpages
,
339 .set_page_dirty
= __set_page_dirty_nobuffers
,
340 .writepage
= nfs_writepage
,
341 .writepages
= nfs_writepages
,
342 .prepare_write
= nfs_prepare_write
,
343 .commit_write
= nfs_commit_write
,
344 .invalidatepage
= nfs_invalidate_page
,
345 .releasepage
= nfs_release_page
,
346 #ifdef CONFIG_NFS_DIRECTIO
347 .direct_IO
= nfs_direct_IO
,
351 static ssize_t
nfs_file_write(struct kiocb
*iocb
, const struct iovec
*iov
,
352 unsigned long nr_segs
, loff_t pos
)
354 struct dentry
* dentry
= iocb
->ki_filp
->f_dentry
;
355 struct inode
* inode
= dentry
->d_inode
;
357 size_t count
= iov_length(iov
, nr_segs
);
359 #ifdef CONFIG_NFS_DIRECTIO
360 if (iocb
->ki_filp
->f_flags
& O_DIRECT
)
361 return nfs_file_direct_write(iocb
, iov
, nr_segs
, pos
);
364 dfprintk(VFS
, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
365 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
366 inode
->i_ino
, (unsigned long) count
, (long long) pos
);
369 if (IS_SWAPFILE(inode
))
372 * O_APPEND implies that we must revalidate the file length.
374 if (iocb
->ki_filp
->f_flags
& O_APPEND
) {
375 result
= nfs_revalidate_file_size(inode
, iocb
->ki_filp
);
384 nfs_add_stats(inode
, NFSIOS_NORMALWRITTENBYTES
, count
);
385 result
= generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
390 printk(KERN_INFO
"NFS: attempt to write to active swap file!\n");
394 static int do_getlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
396 struct file_lock cfl
;
397 struct inode
*inode
= filp
->f_mapping
->host
;
401 /* Try local locking first */
402 if (posix_test_lock(filp
, fl
, &cfl
)) {
403 fl
->fl_start
= cfl
.fl_start
;
404 fl
->fl_end
= cfl
.fl_end
;
405 fl
->fl_type
= cfl
.fl_type
;
406 fl
->fl_pid
= cfl
.fl_pid
;
410 if (nfs_have_delegation(inode
, FMODE_READ
))
413 if (NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
)
416 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
421 fl
->fl_type
= F_UNLCK
;
425 static int do_vfs_lock(struct file
*file
, struct file_lock
*fl
)
428 switch (fl
->fl_flags
& (FL_POSIX
|FL_FLOCK
)) {
430 res
= posix_lock_file_wait(file
, fl
);
433 res
= flock_lock_file_wait(file
, fl
);
439 printk(KERN_WARNING
"%s: VFS is out of sync with lock manager!\n",
444 static int do_unlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
446 struct inode
*inode
= filp
->f_mapping
->host
;
450 * Flush all pending writes before doing anything
453 nfs_sync_mapping(filp
->f_mapping
);
455 /* NOTE: special case
456 * If we're signalled while cleaning up locks on process exit, we
457 * still need to complete the unlock.
460 /* Use local locking if mounted with "-onolock" */
461 if (!(NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
))
462 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
464 status
= do_vfs_lock(filp
, fl
);
469 static int do_setlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
471 struct inode
*inode
= filp
->f_mapping
->host
;
475 * Flush all pending writes before doing anything
478 status
= nfs_sync_mapping(filp
->f_mapping
);
483 /* Use local locking if mounted with "-onolock" */
484 if (!(NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
)) {
485 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
486 /* If we were signalled we still need to ensure that
487 * we clean up any state on the server. We therefore
488 * record the lock call as having succeeded in order to
489 * ensure that locks_remove_posix() cleans it out when
492 if (status
== -EINTR
|| status
== -ERESTARTSYS
)
493 do_vfs_lock(filp
, fl
);
495 status
= do_vfs_lock(filp
, fl
);
500 * Make sure we clear the cache whenever we try to get the lock.
501 * This makes locking act as a cache coherency point.
503 nfs_sync_mapping(filp
->f_mapping
);
504 nfs_zap_caches(inode
);
510 * Lock a (portion of) a file
512 static int nfs_lock(struct file
*filp
, int cmd
, struct file_lock
*fl
)
514 struct inode
* inode
= filp
->f_mapping
->host
;
516 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
517 inode
->i_sb
->s_id
, inode
->i_ino
,
518 fl
->fl_type
, fl
->fl_flags
,
519 (long long)fl
->fl_start
, (long long)fl
->fl_end
);
520 nfs_inc_stats(inode
, NFSIOS_VFSLOCK
);
522 /* No mandatory locks over NFS */
523 if ((inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
524 fl
->fl_type
!= F_UNLCK
)
528 return do_getlk(filp
, cmd
, fl
);
529 if (fl
->fl_type
== F_UNLCK
)
530 return do_unlk(filp
, cmd
, fl
);
531 return do_setlk(filp
, cmd
, fl
);
535 * Lock a (portion of) a file
537 static int nfs_flock(struct file
*filp
, int cmd
, struct file_lock
*fl
)
539 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
540 filp
->f_dentry
->d_inode
->i_sb
->s_id
,
541 filp
->f_dentry
->d_inode
->i_ino
,
542 fl
->fl_type
, fl
->fl_flags
);
545 * No BSD flocks over NFS allowed.
546 * Note: we could try to fake a POSIX lock request here by
547 * using ((u32) filp | 0x80000000) or some such as the pid.
548 * Not sure whether that would be unique, though, or whether
549 * that would break in other places.
551 if (!(fl
->fl_flags
& FL_FLOCK
))
554 /* We're simulating flock() locks using posix locks on the server */
555 fl
->fl_owner
= (fl_owner_t
)filp
;
557 fl
->fl_end
= OFFSET_MAX
;
559 if (fl
->fl_type
== F_UNLCK
)
560 return do_unlk(filp
, cmd
, fl
);
561 return do_setlk(filp
, cmd
, fl
);