xen64: register callbacks in arch-independent way
[linux-2.6/mini2440.git] / fs / nfs / file.c
blob4e98a56a17776cc1ea484db0c5d24f9e2a7490a0
1 /*
2 * linux/fs/nfs/file.c
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>
26 #include <linux/mm.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"
36 #include "internal.h"
37 #include "iostat.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_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,
63 .read = do_sync_read,
64 .write = do_sync_write,
65 .aio_read = nfs_file_read,
66 .aio_write = nfs_file_write,
67 #ifdef CONFIG_MMU
68 .mmap = nfs_file_mmap,
69 #else
70 .mmap = generic_file_mmap,
71 #endif
72 .open = nfs_file_open,
73 .flush = nfs_file_flush,
74 .release = nfs_file_release,
75 .fsync = nfs_fsync,
76 .lock = nfs_lock,
77 .flock = nfs_flock,
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,
89 #ifdef CONFIG_NFS_V3
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 */
102 #ifndef IS_SWAPFILE
103 # define IS_SWAPFILE(inode) (0)
104 #endif
106 static int nfs_check_flags(int flags)
108 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
109 return -EINVAL;
111 return 0;
115 * Open file
117 static int
118 nfs_file_open(struct inode *inode, struct file *filp)
120 int res;
122 res = nfs_check_flags(filp->f_flags);
123 if (res)
124 return res;
126 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
127 lock_kernel();
128 res = NFS_PROTO(inode)->file_open(inode, filp);
129 unlock_kernel();
130 return res;
133 static int
134 nfs_file_release(struct inode *inode, struct file *filp)
136 /* Ensure that dirty pages are flushed out with the right creds */
137 if (filp->f_mode & FMODE_WRITE)
138 nfs_wb_all(filp->f_path.dentry->d_inode);
139 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
140 return NFS_PROTO(inode)->file_release(inode, filp);
144 * nfs_revalidate_size - Revalidate the file size
145 * @inode - pointer to inode struct
146 * @file - pointer to struct file
148 * Revalidates the file length. This is basically a wrapper around
149 * nfs_revalidate_inode() that takes into account the fact that we may
150 * have cached writes (in which case we don't care about the server's
151 * idea of what the file length is), or O_DIRECT (in which case we
152 * shouldn't trust the cache).
154 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
156 struct nfs_server *server = NFS_SERVER(inode);
157 struct nfs_inode *nfsi = NFS_I(inode);
159 if (server->flags & NFS_MOUNT_NOAC)
160 goto force_reval;
161 if (filp->f_flags & O_DIRECT)
162 goto force_reval;
163 if (nfsi->npages != 0)
164 return 0;
165 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
166 return 0;
167 force_reval:
168 return __nfs_revalidate_inode(server, inode);
171 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
173 loff_t loff;
174 /* origin == SEEK_END => we must revalidate the cached file length */
175 if (origin == SEEK_END) {
176 struct inode *inode = filp->f_mapping->host;
177 int retval = nfs_revalidate_file_size(inode, filp);
178 if (retval < 0)
179 return (loff_t)retval;
181 lock_kernel(); /* BKL needed? */
182 loff = generic_file_llseek_unlocked(filp, offset, origin);
183 unlock_kernel();
184 return loff;
188 * Helper for nfs_file_flush() and nfs_fsync()
190 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
191 * disk, but it retrieves and clears ctx->error after synching, despite
192 * the two being set at the same time in nfs_context_set_write_error().
193 * This is because the former is used to notify the _next_ call to
194 * nfs_file_write() that a write error occured, and hence cause it to
195 * fall back to doing a synchronous write.
197 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
199 int have_error, status;
200 int ret = 0;
202 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
203 status = nfs_wb_all(inode);
204 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
205 if (have_error)
206 ret = xchg(&ctx->error, 0);
207 if (!ret)
208 ret = status;
209 return ret;
213 * Flush all dirty pages, and check for write errors.
216 static int
217 nfs_file_flush(struct file *file, fl_owner_t id)
219 struct nfs_open_context *ctx = nfs_file_open_context(file);
220 struct inode *inode = file->f_path.dentry->d_inode;
221 int status;
223 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
225 if ((file->f_mode & FMODE_WRITE) == 0)
226 return 0;
227 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
229 /* Ensure that data+attribute caches are up to date after close() */
230 status = nfs_do_fsync(ctx, inode);
231 if (!status)
232 nfs_revalidate_inode(NFS_SERVER(inode), inode);
233 return status;
236 static ssize_t
237 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
238 unsigned long nr_segs, loff_t pos)
240 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
241 struct inode * inode = dentry->d_inode;
242 ssize_t result;
243 size_t count = iov_length(iov, nr_segs);
245 if (iocb->ki_filp->f_flags & O_DIRECT)
246 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
248 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
249 dentry->d_parent->d_name.name, dentry->d_name.name,
250 (unsigned long) count, (unsigned long) pos);
252 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
253 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
254 if (!result)
255 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
256 return result;
259 static ssize_t
260 nfs_file_splice_read(struct file *filp, loff_t *ppos,
261 struct pipe_inode_info *pipe, size_t count,
262 unsigned int flags)
264 struct dentry *dentry = filp->f_path.dentry;
265 struct inode *inode = dentry->d_inode;
266 ssize_t res;
268 dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
269 dentry->d_parent->d_name.name, dentry->d_name.name,
270 (unsigned long) count, (unsigned long long) *ppos);
272 res = nfs_revalidate_mapping(inode, filp->f_mapping);
273 if (!res)
274 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
275 return res;
278 static int
279 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
281 struct dentry *dentry = file->f_path.dentry;
282 struct inode *inode = dentry->d_inode;
283 int status;
285 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
286 dentry->d_parent->d_name.name, dentry->d_name.name);
288 status = nfs_revalidate_mapping(inode, file->f_mapping);
289 if (!status) {
290 vma->vm_ops = &nfs_file_vm_ops;
291 vma->vm_flags |= VM_CAN_NONLINEAR;
292 file_accessed(file);
294 return status;
298 * Flush any dirty pages for this process, and check for write errors.
299 * The return status from this call provides a reliable indication of
300 * whether any write errors occurred for this process.
302 static int
303 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
305 struct nfs_open_context *ctx = nfs_file_open_context(file);
306 struct inode *inode = dentry->d_inode;
308 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
310 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
311 return nfs_do_fsync(ctx, inode);
315 * This does the "real" work of the write. We must allocate and lock the
316 * page to be sent back to the generic routine, which then copies the
317 * data from user space.
319 * If the writer ends up delaying the write, the writer needs to
320 * increment the page use counts until he is done with the page.
322 static int nfs_write_begin(struct file *file, struct address_space *mapping,
323 loff_t pos, unsigned len, unsigned flags,
324 struct page **pagep, void **fsdata)
326 int ret;
327 pgoff_t index;
328 struct page *page;
329 index = pos >> PAGE_CACHE_SHIFT;
331 page = __grab_cache_page(mapping, index);
332 if (!page)
333 return -ENOMEM;
334 *pagep = page;
336 ret = nfs_flush_incompatible(file, page);
337 if (ret) {
338 unlock_page(page);
339 page_cache_release(page);
341 return ret;
344 static int nfs_write_end(struct file *file, struct address_space *mapping,
345 loff_t pos, unsigned len, unsigned copied,
346 struct page *page, void *fsdata)
348 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
349 int status;
351 lock_kernel();
352 status = nfs_updatepage(file, page, offset, copied);
353 unlock_kernel();
355 unlock_page(page);
356 page_cache_release(page);
358 if (status < 0)
359 return status;
360 return copied;
363 static void nfs_invalidate_page(struct page *page, unsigned long offset)
365 if (offset != 0)
366 return;
367 /* Cancel any unstarted writes on this page */
368 nfs_wb_page_cancel(page->mapping->host, page);
371 static int nfs_release_page(struct page *page, gfp_t gfp)
373 /* If PagePrivate() is set, then the page is not freeable */
374 return 0;
377 static int nfs_launder_page(struct page *page)
379 return nfs_wb_page(page->mapping->host, page);
382 const struct address_space_operations nfs_file_aops = {
383 .readpage = nfs_readpage,
384 .readpages = nfs_readpages,
385 .set_page_dirty = __set_page_dirty_nobuffers,
386 .writepage = nfs_writepage,
387 .writepages = nfs_writepages,
388 .write_begin = nfs_write_begin,
389 .write_end = nfs_write_end,
390 .invalidatepage = nfs_invalidate_page,
391 .releasepage = nfs_release_page,
392 .direct_IO = nfs_direct_IO,
393 .launder_page = nfs_launder_page,
396 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
398 struct file *filp = vma->vm_file;
399 unsigned pagelen;
400 int ret = -EINVAL;
401 struct address_space *mapping;
403 lock_page(page);
404 mapping = page->mapping;
405 if (mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping)
406 goto out_unlock;
408 ret = 0;
409 pagelen = nfs_page_length(page);
410 if (pagelen == 0)
411 goto out_unlock;
413 ret = nfs_flush_incompatible(filp, page);
414 if (ret != 0)
415 goto out_unlock;
417 ret = nfs_updatepage(filp, page, 0, pagelen);
418 if (ret == 0)
419 ret = pagelen;
420 out_unlock:
421 unlock_page(page);
422 return ret;
425 static struct vm_operations_struct nfs_file_vm_ops = {
426 .fault = filemap_fault,
427 .page_mkwrite = nfs_vm_page_mkwrite,
430 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
432 struct nfs_open_context *ctx;
434 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
435 return 1;
436 ctx = nfs_file_open_context(filp);
437 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
438 return 1;
439 return 0;
442 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
443 unsigned long nr_segs, loff_t pos)
445 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
446 struct inode * inode = dentry->d_inode;
447 ssize_t result;
448 size_t count = iov_length(iov, nr_segs);
450 if (iocb->ki_filp->f_flags & O_DIRECT)
451 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
453 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
454 dentry->d_parent->d_name.name, dentry->d_name.name,
455 inode->i_ino, (unsigned long) count, (long long) pos);
457 result = -EBUSY;
458 if (IS_SWAPFILE(inode))
459 goto out_swapfile;
461 * O_APPEND implies that we must revalidate the file length.
463 if (iocb->ki_filp->f_flags & O_APPEND) {
464 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
465 if (result)
466 goto out;
469 result = count;
470 if (!count)
471 goto out;
473 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
474 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
475 /* Return error values for O_SYNC and IS_SYNC() */
476 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
477 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
478 if (err < 0)
479 result = err;
481 out:
482 return result;
484 out_swapfile:
485 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
486 goto out;
489 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
491 struct inode *inode = filp->f_mapping->host;
492 int status = 0;
494 lock_kernel();
495 /* Try local locking first */
496 posix_test_lock(filp, fl);
497 if (fl->fl_type != F_UNLCK) {
498 /* found a conflict */
499 goto out;
502 if (nfs_have_delegation(inode, FMODE_READ))
503 goto out_noconflict;
505 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
506 goto out_noconflict;
508 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
509 out:
510 unlock_kernel();
511 return status;
512 out_noconflict:
513 fl->fl_type = F_UNLCK;
514 goto out;
517 static int do_vfs_lock(struct file *file, struct file_lock *fl)
519 int res = 0;
520 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
521 case FL_POSIX:
522 res = posix_lock_file_wait(file, fl);
523 break;
524 case FL_FLOCK:
525 res = flock_lock_file_wait(file, fl);
526 break;
527 default:
528 BUG();
530 if (res < 0)
531 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
532 " - error %d!\n",
533 __func__, res);
534 return res;
537 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
539 struct inode *inode = filp->f_mapping->host;
540 int status;
543 * Flush all pending writes before doing anything
544 * with locks..
546 nfs_sync_mapping(filp->f_mapping);
548 /* NOTE: special case
549 * If we're signalled while cleaning up locks on process exit, we
550 * still need to complete the unlock.
552 lock_kernel();
553 /* Use local locking if mounted with "-onolock" */
554 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
555 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
556 else
557 status = do_vfs_lock(filp, fl);
558 unlock_kernel();
559 return status;
562 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
564 struct inode *inode = filp->f_mapping->host;
565 int status;
568 * Flush all pending writes before doing anything
569 * with locks..
571 status = nfs_sync_mapping(filp->f_mapping);
572 if (status != 0)
573 goto out;
575 lock_kernel();
576 /* Use local locking if mounted with "-onolock" */
577 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
578 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
579 else
580 status = do_vfs_lock(filp, fl);
581 unlock_kernel();
582 if (status < 0)
583 goto out;
585 * Make sure we clear the cache whenever we try to get the lock.
586 * This makes locking act as a cache coherency point.
588 nfs_sync_mapping(filp->f_mapping);
589 nfs_zap_caches(inode);
590 out:
591 return status;
595 * Lock a (portion of) a file
597 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
599 struct inode * inode = filp->f_mapping->host;
601 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
602 inode->i_sb->s_id, inode->i_ino,
603 fl->fl_type, fl->fl_flags,
604 (long long)fl->fl_start, (long long)fl->fl_end);
605 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
607 /* No mandatory locks over NFS */
608 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
609 return -ENOLCK;
611 if (IS_GETLK(cmd))
612 return do_getlk(filp, cmd, fl);
613 if (fl->fl_type == F_UNLCK)
614 return do_unlk(filp, cmd, fl);
615 return do_setlk(filp, cmd, fl);
619 * Lock a (portion of) a file
621 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
623 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
624 filp->f_path.dentry->d_inode->i_sb->s_id,
625 filp->f_path.dentry->d_inode->i_ino,
626 fl->fl_type, fl->fl_flags);
629 * No BSD flocks over NFS allowed.
630 * Note: we could try to fake a POSIX lock request here by
631 * using ((u32) filp | 0x80000000) or some such as the pid.
632 * Not sure whether that would be unique, though, or whether
633 * that would break in other places.
635 if (!(fl->fl_flags & FL_FLOCK))
636 return -ENOLCK;
638 /* We're simulating flock() locks using posix locks on the server */
639 fl->fl_owner = (fl_owner_t)filp;
640 fl->fl_start = 0;
641 fl->fl_end = OFFSET_MAX;
643 if (fl->fl_type == F_UNLCK)
644 return do_unlk(filp, cmd, fl);
645 return do_setlk(filp, cmd, fl);
648 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
651 * There is no protocol support for leases, so we have no way
652 * to implement them correctly in the face of opens by other
653 * clients.
655 return -EINVAL;