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
*, char __user
*, size_t, loff_t
);
45 static ssize_t
nfs_file_write(struct kiocb
*, const char __user
*, size_t, loff_t
);
46 static int nfs_file_flush(struct file
*);
47 static int nfs_fsync(struct file
*, struct dentry
*dentry
, int datasync
);
48 static int nfs_check_flags(int flags
);
49 static int nfs_lock(struct file
*filp
, int cmd
, struct file_lock
*fl
);
50 static int nfs_flock(struct file
*filp
, int cmd
, struct file_lock
*fl
);
52 const struct file_operations nfs_file_operations
= {
53 .llseek
= nfs_file_llseek
,
55 .write
= do_sync_write
,
56 .aio_read
= nfs_file_read
,
57 .aio_write
= nfs_file_write
,
58 .mmap
= nfs_file_mmap
,
59 .open
= nfs_file_open
,
60 .flush
= nfs_file_flush
,
61 .release
= nfs_file_release
,
65 .sendfile
= nfs_file_sendfile
,
66 .check_flags
= nfs_check_flags
,
69 struct inode_operations nfs_file_inode_operations
= {
70 .permission
= nfs_permission
,
71 .getattr
= nfs_getattr
,
72 .setattr
= nfs_setattr
,
76 struct inode_operations nfs3_file_inode_operations
= {
77 .permission
= nfs_permission
,
78 .getattr
= nfs_getattr
,
79 .setattr
= nfs_setattr
,
80 .listxattr
= nfs3_listxattr
,
81 .getxattr
= nfs3_getxattr
,
82 .setxattr
= nfs3_setxattr
,
83 .removexattr
= nfs3_removexattr
,
85 #endif /* CONFIG_NFS_v3 */
87 /* Hack for future NFS swap support */
89 # define IS_SWAPFILE(inode) (0)
92 static int nfs_check_flags(int flags
)
94 if ((flags
& (O_APPEND
| O_DIRECT
)) == (O_APPEND
| O_DIRECT
))
104 nfs_file_open(struct inode
*inode
, struct file
*filp
)
108 res
= nfs_check_flags(filp
->f_flags
);
112 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
114 res
= NFS_SERVER(inode
)->rpc_ops
->file_open(inode
, filp
);
120 nfs_file_release(struct inode
*inode
, struct file
*filp
)
122 /* Ensure that dirty pages are flushed out with the right creds */
123 if (filp
->f_mode
& FMODE_WRITE
)
124 filemap_fdatawrite(filp
->f_mapping
);
125 nfs_inc_stats(inode
, NFSIOS_VFSRELEASE
);
126 return NFS_PROTO(inode
)->file_release(inode
, filp
);
130 * nfs_revalidate_size - Revalidate the file size
131 * @inode - pointer to inode struct
132 * @file - pointer to struct file
134 * Revalidates the file length. This is basically a wrapper around
135 * nfs_revalidate_inode() that takes into account the fact that we may
136 * have cached writes (in which case we don't care about the server's
137 * idea of what the file length is), or O_DIRECT (in which case we
138 * shouldn't trust the cache).
140 static int nfs_revalidate_file_size(struct inode
*inode
, struct file
*filp
)
142 struct nfs_server
*server
= NFS_SERVER(inode
);
143 struct nfs_inode
*nfsi
= NFS_I(inode
);
145 if (server
->flags
& NFS_MOUNT_NOAC
)
147 if (filp
->f_flags
& O_DIRECT
)
149 if (nfsi
->npages
!= 0)
151 if (!(nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
) && !nfs_attribute_timeout(inode
))
154 return __nfs_revalidate_inode(server
, inode
);
157 static loff_t
nfs_file_llseek(struct file
*filp
, loff_t offset
, int origin
)
159 /* origin == SEEK_END => we must revalidate the cached file length */
161 struct inode
*inode
= filp
->f_mapping
->host
;
162 int retval
= nfs_revalidate_file_size(inode
, filp
);
164 return (loff_t
)retval
;
166 return remote_llseek(filp
, offset
, origin
);
170 * Flush all dirty pages, and check for write errors.
174 nfs_file_flush(struct file
*file
)
176 struct nfs_open_context
*ctx
= (struct nfs_open_context
*)file
->private_data
;
177 struct inode
*inode
= file
->f_dentry
->d_inode
;
180 dfprintk(VFS
, "nfs: flush(%s/%ld)\n", inode
->i_sb
->s_id
, inode
->i_ino
);
182 if ((file
->f_mode
& FMODE_WRITE
) == 0)
184 nfs_inc_stats(inode
, NFSIOS_VFSFLUSH
);
186 /* Ensure that data+attribute caches are up to date after close() */
187 status
= nfs_wb_all(inode
);
192 nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
199 nfs_file_read(struct kiocb
*iocb
, char __user
* buf
, size_t count
, loff_t pos
)
201 struct dentry
* dentry
= iocb
->ki_filp
->f_dentry
;
202 struct inode
* inode
= dentry
->d_inode
;
205 #ifdef CONFIG_NFS_DIRECTIO
206 if (iocb
->ki_filp
->f_flags
& O_DIRECT
)
207 return nfs_file_direct_read(iocb
, buf
, count
, pos
);
210 dfprintk(VFS
, "nfs: read(%s/%s, %lu@%lu)\n",
211 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
212 (unsigned long) count
, (unsigned long) pos
);
214 result
= nfs_revalidate_mapping(inode
, iocb
->ki_filp
->f_mapping
);
215 nfs_add_stats(inode
, NFSIOS_NORMALREADBYTES
, count
);
217 result
= generic_file_aio_read(iocb
, buf
, count
, pos
);
222 nfs_file_sendfile(struct file
*filp
, loff_t
*ppos
, size_t count
,
223 read_actor_t actor
, void *target
)
225 struct dentry
*dentry
= filp
->f_dentry
;
226 struct inode
*inode
= dentry
->d_inode
;
229 dfprintk(VFS
, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
230 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
231 (unsigned long) count
, (unsigned long long) *ppos
);
233 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
235 res
= generic_file_sendfile(filp
, ppos
, count
, actor
, target
);
240 nfs_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
242 struct dentry
*dentry
= file
->f_dentry
;
243 struct inode
*inode
= dentry
->d_inode
;
246 dfprintk(VFS
, "nfs: mmap(%s/%s)\n",
247 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
249 status
= nfs_revalidate_mapping(inode
, file
->f_mapping
);
251 status
= generic_file_mmap(file
, vma
);
256 * Flush any dirty pages for this process, and check for write errors.
257 * The return status from this call provides a reliable indication of
258 * whether any write errors occurred for this process.
261 nfs_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
263 struct nfs_open_context
*ctx
= (struct nfs_open_context
*)file
->private_data
;
264 struct inode
*inode
= dentry
->d_inode
;
267 dfprintk(VFS
, "nfs: fsync(%s/%ld)\n", inode
->i_sb
->s_id
, inode
->i_ino
);
269 nfs_inc_stats(inode
, NFSIOS_VFSFSYNC
);
271 status
= nfs_wb_all(inode
);
281 * This does the "real" work of the write. The generic routine has
282 * allocated the page, locked it, done all the page alignment stuff
283 * calculations etc. Now we should just copy the data from user
284 * space and write it back to the real medium..
286 * If the writer ends up delaying the write, the writer needs to
287 * increment the page use counts until he is done with the page.
289 static int nfs_prepare_write(struct file
*file
, struct page
*page
, unsigned offset
, unsigned to
)
291 return nfs_flush_incompatible(file
, page
);
294 static int nfs_commit_write(struct file
*file
, struct page
*page
, unsigned offset
, unsigned to
)
299 status
= nfs_updatepage(file
, page
, offset
, to
-offset
);
304 static void nfs_invalidate_page(struct page
*page
, unsigned long offset
)
306 /* FIXME: we really should cancel any unstarted writes on this page */
309 static int nfs_release_page(struct page
*page
, gfp_t gfp
)
311 return !nfs_wb_page(page
->mapping
->host
, page
);
314 struct address_space_operations nfs_file_aops
= {
315 .readpage
= nfs_readpage
,
316 .readpages
= nfs_readpages
,
317 .set_page_dirty
= __set_page_dirty_nobuffers
,
318 .writepage
= nfs_writepage
,
319 .writepages
= nfs_writepages
,
320 .prepare_write
= nfs_prepare_write
,
321 .commit_write
= nfs_commit_write
,
322 .invalidatepage
= nfs_invalidate_page
,
323 .releasepage
= nfs_release_page
,
324 #ifdef CONFIG_NFS_DIRECTIO
325 .direct_IO
= nfs_direct_IO
,
330 * Write to a file (through the page cache).
333 nfs_file_write(struct kiocb
*iocb
, const char __user
*buf
, size_t count
, loff_t pos
)
335 struct dentry
* dentry
= iocb
->ki_filp
->f_dentry
;
336 struct inode
* inode
= dentry
->d_inode
;
339 #ifdef CONFIG_NFS_DIRECTIO
340 if (iocb
->ki_filp
->f_flags
& O_DIRECT
)
341 return nfs_file_direct_write(iocb
, buf
, count
, pos
);
344 dfprintk(VFS
, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
345 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
346 inode
->i_ino
, (unsigned long) count
, (unsigned long) pos
);
349 if (IS_SWAPFILE(inode
))
352 * O_APPEND implies that we must revalidate the file length.
354 if (iocb
->ki_filp
->f_flags
& O_APPEND
) {
355 result
= nfs_revalidate_file_size(inode
, iocb
->ki_filp
);
364 nfs_add_stats(inode
, NFSIOS_NORMALWRITTENBYTES
, count
);
365 result
= generic_file_aio_write(iocb
, buf
, count
, pos
);
370 printk(KERN_INFO
"NFS: attempt to write to active swap file!\n");
374 static int do_getlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
376 struct file_lock cfl
;
377 struct inode
*inode
= filp
->f_mapping
->host
;
381 /* Try local locking first */
382 if (posix_test_lock(filp
, fl
, &cfl
)) {
383 fl
->fl_start
= cfl
.fl_start
;
384 fl
->fl_end
= cfl
.fl_end
;
385 fl
->fl_type
= cfl
.fl_type
;
386 fl
->fl_pid
= cfl
.fl_pid
;
390 if (nfs_have_delegation(inode
, FMODE_READ
))
393 if (NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
)
396 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
401 fl
->fl_type
= F_UNLCK
;
405 static int do_vfs_lock(struct file
*file
, struct file_lock
*fl
)
408 switch (fl
->fl_flags
& (FL_POSIX
|FL_FLOCK
)) {
410 res
= posix_lock_file_wait(file
, fl
);
413 res
= flock_lock_file_wait(file
, fl
);
419 printk(KERN_WARNING
"%s: VFS is out of sync with lock manager!\n",
424 static int do_unlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
426 struct inode
*inode
= filp
->f_mapping
->host
;
430 * Flush all pending writes before doing anything
433 nfs_sync_mapping(filp
->f_mapping
);
435 /* NOTE: special case
436 * If we're signalled while cleaning up locks on process exit, we
437 * still need to complete the unlock.
440 /* Use local locking if mounted with "-onolock" */
441 if (!(NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
))
442 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
444 status
= do_vfs_lock(filp
, fl
);
449 static int do_setlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
451 struct inode
*inode
= filp
->f_mapping
->host
;
455 * Flush all pending writes before doing anything
458 status
= nfs_sync_mapping(filp
->f_mapping
);
463 /* Use local locking if mounted with "-onolock" */
464 if (!(NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
)) {
465 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
466 /* If we were signalled we still need to ensure that
467 * we clean up any state on the server. We therefore
468 * record the lock call as having succeeded in order to
469 * ensure that locks_remove_posix() cleans it out when
472 if (status
== -EINTR
|| status
== -ERESTARTSYS
)
473 do_vfs_lock(filp
, fl
);
475 status
= do_vfs_lock(filp
, fl
);
480 * Make sure we clear the cache whenever we try to get the lock.
481 * This makes locking act as a cache coherency point.
483 nfs_sync_mapping(filp
->f_mapping
);
484 nfs_zap_caches(inode
);
490 * Lock a (portion of) a file
492 static int nfs_lock(struct file
*filp
, int cmd
, struct file_lock
*fl
)
494 struct inode
* inode
= filp
->f_mapping
->host
;
496 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
497 inode
->i_sb
->s_id
, inode
->i_ino
,
498 fl
->fl_type
, fl
->fl_flags
,
499 (long long)fl
->fl_start
, (long long)fl
->fl_end
);
500 nfs_inc_stats(inode
, NFSIOS_VFSLOCK
);
502 /* No mandatory locks over NFS */
503 if ((inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
504 fl
->fl_type
!= F_UNLCK
)
508 return do_getlk(filp
, cmd
, fl
);
509 if (fl
->fl_type
== F_UNLCK
)
510 return do_unlk(filp
, cmd
, fl
);
511 return do_setlk(filp
, cmd
, fl
);
515 * Lock a (portion of) a file
517 static int nfs_flock(struct file
*filp
, int cmd
, struct file_lock
*fl
)
519 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
520 filp
->f_dentry
->d_inode
->i_sb
->s_id
,
521 filp
->f_dentry
->d_inode
->i_ino
,
522 fl
->fl_type
, fl
->fl_flags
);
525 * No BSD flocks over NFS allowed.
526 * Note: we could try to fake a POSIX lock request here by
527 * using ((u32) filp | 0x80000000) or some such as the pid.
528 * Not sure whether that would be unique, though, or whether
529 * that would break in other places.
531 if (!(fl
->fl_flags
& FL_FLOCK
))
534 /* We're simulating flock() locks using posix locks on the server */
535 fl
->fl_owner
= (fl_owner_t
)filp
;
537 fl
->fl_end
= OFFSET_MAX
;
539 if (fl
->fl_type
== F_UNLCK
)
540 return do_unlk(filp
, cmd
, fl
);
541 return do_setlk(filp
, cmd
, fl
);