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>
30 #include <linux/aio.h>
32 #include <asm/uaccess.h>
33 #include <asm/system.h>
35 #include "delegation.h"
39 #define NFSDBG_FACILITY NFSDBG_FILE
41 static int nfs_file_open(struct inode
*, struct file
*);
42 static int nfs_file_release(struct inode
*, struct file
*);
43 static loff_t
nfs_file_llseek(struct file
*file
, loff_t offset
, int origin
);
44 static int nfs_file_mmap(struct file
*, struct vm_area_struct
*);
45 static ssize_t
nfs_file_splice_read(struct file
*filp
, loff_t
*ppos
,
46 struct pipe_inode_info
*pipe
,
47 size_t count
, unsigned int flags
);
48 static ssize_t
nfs_file_read(struct kiocb
*, const struct iovec
*iov
,
49 unsigned long nr_segs
, loff_t pos
);
50 static ssize_t
nfs_file_write(struct kiocb
*, const struct iovec
*iov
,
51 unsigned long nr_segs
, loff_t pos
);
52 static int nfs_file_flush(struct file
*, fl_owner_t id
);
53 static int nfs_file_fsync(struct file
*, struct dentry
*dentry
, int datasync
);
54 static int nfs_check_flags(int flags
);
55 static int nfs_lock(struct file
*filp
, int cmd
, struct file_lock
*fl
);
56 static int nfs_flock(struct file
*filp
, int cmd
, struct file_lock
*fl
);
57 static int nfs_setlease(struct file
*file
, long arg
, struct file_lock
**fl
);
59 static struct vm_operations_struct nfs_file_vm_ops
;
61 const struct file_operations nfs_file_operations
= {
62 .llseek
= nfs_file_llseek
,
64 .write
= do_sync_write
,
65 .aio_read
= nfs_file_read
,
66 .aio_write
= nfs_file_write
,
68 .mmap
= nfs_file_mmap
,
70 .mmap
= generic_file_mmap
,
72 .open
= nfs_file_open
,
73 .flush
= nfs_file_flush
,
74 .release
= nfs_file_release
,
75 .fsync
= nfs_file_fsync
,
78 .splice_read
= nfs_file_splice_read
,
79 .check_flags
= nfs_check_flags
,
80 .setlease
= nfs_setlease
,
83 const struct inode_operations nfs_file_inode_operations
= {
84 .permission
= nfs_permission
,
85 .getattr
= nfs_getattr
,
86 .setattr
= nfs_setattr
,
90 const struct inode_operations nfs3_file_inode_operations
= {
91 .permission
= nfs_permission
,
92 .getattr
= nfs_getattr
,
93 .setattr
= nfs_setattr
,
94 .listxattr
= nfs3_listxattr
,
95 .getxattr
= nfs3_getxattr
,
96 .setxattr
= nfs3_setxattr
,
97 .removexattr
= nfs3_removexattr
,
99 #endif /* CONFIG_NFS_v3 */
101 /* Hack for future NFS swap support */
103 # define IS_SWAPFILE(inode) (0)
106 static int nfs_check_flags(int flags
)
108 if ((flags
& (O_APPEND
| O_DIRECT
)) == (O_APPEND
| O_DIRECT
))
118 nfs_file_open(struct inode
*inode
, struct file
*filp
)
122 dprintk("NFS: open file(%s/%s)\n",
123 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
124 filp
->f_path
.dentry
->d_name
.name
);
126 res
= nfs_check_flags(filp
->f_flags
);
130 nfs_inc_stats(inode
, NFSIOS_VFSOPEN
);
131 res
= nfs_open(inode
, filp
);
136 nfs_file_release(struct inode
*inode
, struct file
*filp
)
138 struct dentry
*dentry
= filp
->f_path
.dentry
;
140 dprintk("NFS: release(%s/%s)\n",
141 dentry
->d_parent
->d_name
.name
,
142 dentry
->d_name
.name
);
144 /* Ensure that dirty pages are flushed out with the right creds */
145 if (filp
->f_mode
& FMODE_WRITE
)
146 nfs_wb_all(dentry
->d_inode
);
147 nfs_inc_stats(inode
, NFSIOS_VFSRELEASE
);
148 return nfs_release(inode
, filp
);
152 * nfs_revalidate_size - Revalidate the file size
153 * @inode - pointer to inode struct
154 * @file - pointer to struct file
156 * Revalidates the file length. This is basically a wrapper around
157 * nfs_revalidate_inode() that takes into account the fact that we may
158 * have cached writes (in which case we don't care about the server's
159 * idea of what the file length is), or O_DIRECT (in which case we
160 * shouldn't trust the cache).
162 static int nfs_revalidate_file_size(struct inode
*inode
, struct file
*filp
)
164 struct nfs_server
*server
= NFS_SERVER(inode
);
165 struct nfs_inode
*nfsi
= NFS_I(inode
);
167 if (server
->flags
& NFS_MOUNT_NOAC
)
169 if (filp
->f_flags
& O_DIRECT
)
171 if (nfsi
->npages
!= 0)
173 if (!(nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
) && !nfs_attribute_timeout(inode
))
176 return __nfs_revalidate_inode(server
, inode
);
179 static loff_t
nfs_file_llseek(struct file
*filp
, loff_t offset
, int origin
)
183 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
184 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
185 filp
->f_path
.dentry
->d_name
.name
,
188 /* origin == SEEK_END => we must revalidate the cached file length */
189 if (origin
== SEEK_END
) {
190 struct inode
*inode
= filp
->f_mapping
->host
;
191 int retval
= nfs_revalidate_file_size(inode
, filp
);
193 return (loff_t
)retval
;
195 lock_kernel(); /* BKL needed? */
196 loff
= generic_file_llseek_unlocked(filp
, offset
, origin
);
202 * Helper for nfs_file_flush() and nfs_file_fsync()
204 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
205 * disk, but it retrieves and clears ctx->error after synching, despite
206 * the two being set at the same time in nfs_context_set_write_error().
207 * This is because the former is used to notify the _next_ call to
208 * nfs_file_write() that a write error occured, and hence cause it to
209 * fall back to doing a synchronous write.
211 static int nfs_do_fsync(struct nfs_open_context
*ctx
, struct inode
*inode
)
213 int have_error
, status
;
216 have_error
= test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
217 status
= nfs_wb_all(inode
);
218 have_error
|= test_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
);
220 ret
= xchg(&ctx
->error
, 0);
227 * Flush all dirty pages, and check for write errors.
230 nfs_file_flush(struct file
*file
, fl_owner_t id
)
232 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
233 struct dentry
*dentry
= file
->f_path
.dentry
;
234 struct inode
*inode
= dentry
->d_inode
;
237 dprintk("NFS: flush(%s/%s)\n",
238 dentry
->d_parent
->d_name
.name
,
239 dentry
->d_name
.name
);
241 if ((file
->f_mode
& FMODE_WRITE
) == 0)
243 nfs_inc_stats(inode
, NFSIOS_VFSFLUSH
);
245 /* Ensure that data+attribute caches are up to date after close() */
246 status
= nfs_do_fsync(ctx
, inode
);
248 nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
253 nfs_file_read(struct kiocb
*iocb
, const struct iovec
*iov
,
254 unsigned long nr_segs
, loff_t pos
)
256 struct dentry
* dentry
= iocb
->ki_filp
->f_path
.dentry
;
257 struct inode
* inode
= dentry
->d_inode
;
259 size_t count
= iov_length(iov
, nr_segs
);
261 if (iocb
->ki_filp
->f_flags
& O_DIRECT
)
262 return nfs_file_direct_read(iocb
, iov
, nr_segs
, pos
);
264 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
265 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
266 (unsigned long) count
, (unsigned long) pos
);
268 result
= nfs_revalidate_mapping(inode
, iocb
->ki_filp
->f_mapping
);
269 nfs_add_stats(inode
, NFSIOS_NORMALREADBYTES
, count
);
271 result
= generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
276 nfs_file_splice_read(struct file
*filp
, loff_t
*ppos
,
277 struct pipe_inode_info
*pipe
, size_t count
,
280 struct dentry
*dentry
= filp
->f_path
.dentry
;
281 struct inode
*inode
= dentry
->d_inode
;
284 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
285 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
286 (unsigned long) count
, (unsigned long long) *ppos
);
288 res
= nfs_revalidate_mapping(inode
, filp
->f_mapping
);
290 res
= generic_file_splice_read(filp
, ppos
, pipe
, count
, flags
);
295 nfs_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
297 struct dentry
*dentry
= file
->f_path
.dentry
;
298 struct inode
*inode
= dentry
->d_inode
;
301 dprintk("NFS: mmap(%s/%s)\n",
302 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
304 status
= nfs_revalidate_mapping(inode
, file
->f_mapping
);
306 vma
->vm_ops
= &nfs_file_vm_ops
;
307 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
314 * Flush any dirty pages for this process, and check for write errors.
315 * The return status from this call provides a reliable indication of
316 * whether any write errors occurred for this process.
319 nfs_file_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
321 struct nfs_open_context
*ctx
= nfs_file_open_context(file
);
322 struct inode
*inode
= dentry
->d_inode
;
324 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
325 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
328 nfs_inc_stats(inode
, NFSIOS_VFSFSYNC
);
329 return nfs_do_fsync(ctx
, inode
);
333 * This does the "real" work of the write. We must allocate and lock the
334 * page to be sent back to the generic routine, which then copies the
335 * data from user space.
337 * If the writer ends up delaying the write, the writer needs to
338 * increment the page use counts until he is done with the page.
340 static int nfs_write_begin(struct file
*file
, struct address_space
*mapping
,
341 loff_t pos
, unsigned len
, unsigned flags
,
342 struct page
**pagep
, void **fsdata
)
347 index
= pos
>> PAGE_CACHE_SHIFT
;
349 dfprintk(PAGECACHE
, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
350 file
->f_path
.dentry
->d_parent
->d_name
.name
,
351 file
->f_path
.dentry
->d_name
.name
,
352 mapping
->host
->i_ino
, len
, (long long) pos
);
354 page
= __grab_cache_page(mapping
, index
);
359 ret
= nfs_flush_incompatible(file
, page
);
362 page_cache_release(page
);
367 static int nfs_write_end(struct file
*file
, struct address_space
*mapping
,
368 loff_t pos
, unsigned len
, unsigned copied
,
369 struct page
*page
, void *fsdata
)
371 unsigned offset
= pos
& (PAGE_CACHE_SIZE
- 1);
374 dfprintk(PAGECACHE
, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
375 file
->f_path
.dentry
->d_parent
->d_name
.name
,
376 file
->f_path
.dentry
->d_name
.name
,
377 mapping
->host
->i_ino
, len
, (long long) pos
);
380 * Zero any uninitialised parts of the page, and then mark the page
381 * as up to date if it turns out that we're extending the file.
383 if (!PageUptodate(page
)) {
384 unsigned pglen
= nfs_page_length(page
);
385 unsigned end
= offset
+ len
;
388 zero_user_segments(page
, 0, offset
,
389 end
, PAGE_CACHE_SIZE
);
390 SetPageUptodate(page
);
391 } else if (end
>= pglen
) {
392 zero_user_segment(page
, end
, PAGE_CACHE_SIZE
);
394 SetPageUptodate(page
);
396 zero_user_segment(page
, pglen
, PAGE_CACHE_SIZE
);
399 status
= nfs_updatepage(file
, page
, offset
, copied
);
402 page_cache_release(page
);
409 static void nfs_invalidate_page(struct page
*page
, unsigned long offset
)
411 dfprintk(PAGECACHE
, "NFS: invalidate_page(%p, %lu)\n", page
, offset
);
415 /* Cancel any unstarted writes on this page */
416 nfs_wb_page_cancel(page
->mapping
->host
, page
);
419 static int nfs_release_page(struct page
*page
, gfp_t gfp
)
421 dfprintk(PAGECACHE
, "NFS: release_page(%p)\n", page
);
423 /* If PagePrivate() is set, then the page is not freeable */
427 static int nfs_launder_page(struct page
*page
)
429 struct inode
*inode
= page
->mapping
->host
;
431 dfprintk(PAGECACHE
, "NFS: launder_page(%ld, %llu)\n",
432 inode
->i_ino
, (long long)page_offset(page
));
434 return nfs_wb_page(inode
, page
);
437 const struct address_space_operations nfs_file_aops
= {
438 .readpage
= nfs_readpage
,
439 .readpages
= nfs_readpages
,
440 .set_page_dirty
= __set_page_dirty_nobuffers
,
441 .writepage
= nfs_writepage
,
442 .writepages
= nfs_writepages
,
443 .write_begin
= nfs_write_begin
,
444 .write_end
= nfs_write_end
,
445 .invalidatepage
= nfs_invalidate_page
,
446 .releasepage
= nfs_release_page
,
447 .direct_IO
= nfs_direct_IO
,
448 .launder_page
= nfs_launder_page
,
451 static int nfs_vm_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
453 struct file
*filp
= vma
->vm_file
;
454 struct dentry
*dentry
= filp
->f_path
.dentry
;
457 struct address_space
*mapping
;
459 dfprintk(PAGECACHE
, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
460 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
461 filp
->f_mapping
->host
->i_ino
,
462 (long long)page_offset(page
));
465 mapping
= page
->mapping
;
466 if (mapping
!= dentry
->d_inode
->i_mapping
)
470 pagelen
= nfs_page_length(page
);
474 ret
= nfs_flush_incompatible(filp
, page
);
478 ret
= nfs_updatepage(filp
, page
, 0, pagelen
);
486 static struct vm_operations_struct nfs_file_vm_ops
= {
487 .fault
= filemap_fault
,
488 .page_mkwrite
= nfs_vm_page_mkwrite
,
491 static int nfs_need_sync_write(struct file
*filp
, struct inode
*inode
)
493 struct nfs_open_context
*ctx
;
495 if (IS_SYNC(inode
) || (filp
->f_flags
& O_SYNC
))
497 ctx
= nfs_file_open_context(filp
);
498 if (test_bit(NFS_CONTEXT_ERROR_WRITE
, &ctx
->flags
))
503 static ssize_t
nfs_file_write(struct kiocb
*iocb
, const struct iovec
*iov
,
504 unsigned long nr_segs
, loff_t pos
)
506 struct dentry
* dentry
= iocb
->ki_filp
->f_path
.dentry
;
507 struct inode
* inode
= dentry
->d_inode
;
509 size_t count
= iov_length(iov
, nr_segs
);
511 if (iocb
->ki_filp
->f_flags
& O_DIRECT
)
512 return nfs_file_direct_write(iocb
, iov
, nr_segs
, pos
);
514 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
515 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
,
516 (unsigned long) count
, (long long) pos
);
519 if (IS_SWAPFILE(inode
))
522 * O_APPEND implies that we must revalidate the file length.
524 if (iocb
->ki_filp
->f_flags
& O_APPEND
) {
525 result
= nfs_revalidate_file_size(inode
, iocb
->ki_filp
);
534 nfs_add_stats(inode
, NFSIOS_NORMALWRITTENBYTES
, count
);
535 result
= generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
536 /* Return error values for O_SYNC and IS_SYNC() */
537 if (result
>= 0 && nfs_need_sync_write(iocb
->ki_filp
, inode
)) {
538 int err
= nfs_do_fsync(nfs_file_open_context(iocb
->ki_filp
), inode
);
546 printk(KERN_INFO
"NFS: attempt to write to active swap file!\n");
550 static int do_getlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
552 struct inode
*inode
= filp
->f_mapping
->host
;
556 /* Try local locking first */
557 posix_test_lock(filp
, fl
);
558 if (fl
->fl_type
!= F_UNLCK
) {
559 /* found a conflict */
563 if (nfs_have_delegation(inode
, FMODE_READ
))
566 if (NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
)
569 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
574 fl
->fl_type
= F_UNLCK
;
578 static int do_vfs_lock(struct file
*file
, struct file_lock
*fl
)
581 switch (fl
->fl_flags
& (FL_POSIX
|FL_FLOCK
)) {
583 res
= posix_lock_file_wait(file
, fl
);
586 res
= flock_lock_file_wait(file
, fl
);
592 dprintk(KERN_WARNING
"%s: VFS is out of sync with lock manager"
598 static int do_unlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
600 struct inode
*inode
= filp
->f_mapping
->host
;
604 * Flush all pending writes before doing anything
607 nfs_sync_mapping(filp
->f_mapping
);
609 /* NOTE: special case
610 * If we're signalled while cleaning up locks on process exit, we
611 * still need to complete the unlock.
614 /* Use local locking if mounted with "-onolock" */
615 if (!(NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
))
616 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
618 status
= do_vfs_lock(filp
, fl
);
623 static int do_setlk(struct file
*filp
, int cmd
, struct file_lock
*fl
)
625 struct inode
*inode
= filp
->f_mapping
->host
;
629 * Flush all pending writes before doing anything
632 status
= nfs_sync_mapping(filp
->f_mapping
);
637 /* Use local locking if mounted with "-onolock" */
638 if (!(NFS_SERVER(inode
)->flags
& NFS_MOUNT_NONLM
))
639 status
= NFS_PROTO(inode
)->lock(filp
, cmd
, fl
);
641 status
= do_vfs_lock(filp
, fl
);
646 * Make sure we clear the cache whenever we try to get the lock.
647 * This makes locking act as a cache coherency point.
649 nfs_sync_mapping(filp
->f_mapping
);
650 if (!nfs_have_delegation(inode
, FMODE_READ
))
651 nfs_zap_caches(inode
);
657 * Lock a (portion of) a file
659 static int nfs_lock(struct file
*filp
, int cmd
, struct file_lock
*fl
)
661 struct inode
*inode
= filp
->f_mapping
->host
;
664 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
665 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
666 filp
->f_path
.dentry
->d_name
.name
,
667 fl
->fl_type
, fl
->fl_flags
,
668 (long long)fl
->fl_start
, (long long)fl
->fl_end
);
670 nfs_inc_stats(inode
, NFSIOS_VFSLOCK
);
672 /* No mandatory locks over NFS */
673 if (__mandatory_lock(inode
) && fl
->fl_type
!= F_UNLCK
)
676 if (NFS_PROTO(inode
)->lock_check_bounds
!= NULL
) {
677 ret
= NFS_PROTO(inode
)->lock_check_bounds(fl
);
683 ret
= do_getlk(filp
, cmd
, fl
);
684 else if (fl
->fl_type
== F_UNLCK
)
685 ret
= do_unlk(filp
, cmd
, fl
);
687 ret
= do_setlk(filp
, cmd
, fl
);
693 * Lock a (portion of) a file
695 static int nfs_flock(struct file
*filp
, int cmd
, struct file_lock
*fl
)
697 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
698 filp
->f_path
.dentry
->d_parent
->d_name
.name
,
699 filp
->f_path
.dentry
->d_name
.name
,
700 fl
->fl_type
, fl
->fl_flags
);
703 * No BSD flocks over NFS allowed.
704 * Note: we could try to fake a POSIX lock request here by
705 * using ((u32) filp | 0x80000000) or some such as the pid.
706 * Not sure whether that would be unique, though, or whether
707 * that would break in other places.
709 if (!(fl
->fl_flags
& FL_FLOCK
))
712 /* We're simulating flock() locks using posix locks on the server */
713 fl
->fl_owner
= (fl_owner_t
)filp
;
715 fl
->fl_end
= OFFSET_MAX
;
717 if (fl
->fl_type
== F_UNLCK
)
718 return do_unlk(filp
, cmd
, fl
);
719 return do_setlk(filp
, cmd
, fl
);
723 * There is no protocol support for leases, so we have no way to implement
724 * them correctly in the face of opens by other clients.
726 static int nfs_setlease(struct file
*file
, long arg
, struct file_lock
**fl
)
728 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
729 file
->f_path
.dentry
->d_parent
->d_name
.name
,
730 file
->f_path
.dentry
->d_name
.name
, arg
);