NFS: Add debugging facility for NFS aops
[linux-2.6/cjktty.git] / fs / nfs / file.c
blob6c447a37edefa6bfd94ac11383d086b8de1c58de
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_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,
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_file_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 dfprintk(VFS, "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);
127 if (res)
128 return res;
130 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
131 lock_kernel();
132 res = NFS_PROTO(inode)->file_open(inode, filp);
133 unlock_kernel();
134 return res;
137 static int
138 nfs_file_release(struct inode *inode, struct file *filp)
140 /* Ensure that dirty pages are flushed out with the right creds */
141 if (filp->f_mode & FMODE_WRITE)
142 nfs_wb_all(filp->f_path.dentry->d_inode);
143 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
144 return NFS_PROTO(inode)->file_release(inode, filp);
148 * nfs_revalidate_size - Revalidate the file size
149 * @inode - pointer to inode struct
150 * @file - pointer to struct file
152 * Revalidates the file length. This is basically a wrapper around
153 * nfs_revalidate_inode() that takes into account the fact that we may
154 * have cached writes (in which case we don't care about the server's
155 * idea of what the file length is), or O_DIRECT (in which case we
156 * shouldn't trust the cache).
158 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
160 struct nfs_server *server = NFS_SERVER(inode);
161 struct nfs_inode *nfsi = NFS_I(inode);
163 if (server->flags & NFS_MOUNT_NOAC)
164 goto force_reval;
165 if (filp->f_flags & O_DIRECT)
166 goto force_reval;
167 if (nfsi->npages != 0)
168 return 0;
169 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
170 return 0;
171 force_reval:
172 return __nfs_revalidate_inode(server, inode);
175 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
177 dfprintk(VFS, "NFS: llseek file(%s/%s, %lld, %d)\n",
178 filp->f_path.dentry->d_parent->d_name.name,
179 filp->f_path.dentry->d_name.name,
180 offset, origin);
182 /* origin == SEEK_END => we must revalidate the cached file length */
183 if (origin == SEEK_END) {
184 struct inode *inode = filp->f_mapping->host;
185 int retval = nfs_revalidate_file_size(inode, filp);
186 if (retval < 0)
187 return (loff_t)retval;
189 return remote_llseek(filp, offset, origin);
193 * Helper for nfs_file_flush() and nfs_file_fsync()
195 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
196 * disk, but it retrieves and clears ctx->error after synching, despite
197 * the two being set at the same time in nfs_context_set_write_error().
198 * This is because the former is used to notify the _next_ call to
199 * nfs_file_write() that a write error occured, and hence cause it to
200 * fall back to doing a synchronous write.
202 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
204 int have_error, status;
205 int ret = 0;
207 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
208 status = nfs_wb_all(inode);
209 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
210 if (have_error)
211 ret = xchg(&ctx->error, 0);
212 if (!ret)
213 ret = status;
214 return ret;
218 * Flush all dirty pages, and check for write errors.
221 static int
222 nfs_file_flush(struct file *file, fl_owner_t id)
224 struct nfs_open_context *ctx = nfs_file_open_context(file);
225 struct inode *inode = file->f_path.dentry->d_inode;
226 int status;
228 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
230 if ((file->f_mode & FMODE_WRITE) == 0)
231 return 0;
232 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
234 /* Ensure that data+attribute caches are up to date after close() */
235 status = nfs_do_fsync(ctx, inode);
236 if (!status)
237 nfs_revalidate_inode(NFS_SERVER(inode), inode);
238 return status;
241 static ssize_t
242 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
243 unsigned long nr_segs, loff_t pos)
245 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
246 struct inode * inode = dentry->d_inode;
247 ssize_t result;
248 size_t count = iov_length(iov, nr_segs);
250 if (iocb->ki_filp->f_flags & O_DIRECT)
251 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
253 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
254 dentry->d_parent->d_name.name, dentry->d_name.name,
255 (unsigned long) count, (unsigned long) pos);
257 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
258 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
259 if (!result)
260 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
261 return result;
264 static ssize_t
265 nfs_file_splice_read(struct file *filp, loff_t *ppos,
266 struct pipe_inode_info *pipe, size_t count,
267 unsigned int flags)
269 struct dentry *dentry = filp->f_path.dentry;
270 struct inode *inode = dentry->d_inode;
271 ssize_t res;
273 dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
274 dentry->d_parent->d_name.name, dentry->d_name.name,
275 (unsigned long) count, (unsigned long long) *ppos);
277 res = nfs_revalidate_mapping(inode, filp->f_mapping);
278 if (!res)
279 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
280 return res;
283 static int
284 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
286 struct dentry *dentry = file->f_path.dentry;
287 struct inode *inode = dentry->d_inode;
288 int status;
290 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
291 dentry->d_parent->d_name.name, dentry->d_name.name);
293 status = nfs_revalidate_mapping(inode, file->f_mapping);
294 if (!status) {
295 vma->vm_ops = &nfs_file_vm_ops;
296 vma->vm_flags |= VM_CAN_NONLINEAR;
297 file_accessed(file);
299 return status;
303 * Flush any dirty pages for this process, and check for write errors.
304 * The return status from this call provides a reliable indication of
305 * whether any write errors occurred for this process.
307 static int
308 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
310 struct nfs_open_context *ctx = nfs_file_open_context(file);
311 struct inode *inode = dentry->d_inode;
313 dfprintk(VFS, "NFS: fsync file(%s/%s) datasync %d\n",
314 dentry->d_parent->d_name.name, dentry->d_name.name,
315 datasync);
317 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
318 return nfs_do_fsync(ctx, inode);
322 * This does the "real" work of the write. We must allocate and lock the
323 * page to be sent back to the generic routine, which then copies the
324 * data from user space.
326 * If the writer ends up delaying the write, the writer needs to
327 * increment the page use counts until he is done with the page.
329 static int nfs_write_begin(struct file *file, struct address_space *mapping,
330 loff_t pos, unsigned len, unsigned flags,
331 struct page **pagep, void **fsdata)
333 int ret;
334 pgoff_t index;
335 struct page *page;
336 index = pos >> PAGE_CACHE_SHIFT;
338 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
339 file->f_path.dentry->d_parent->d_name.name,
340 file->f_path.dentry->d_name.name,
341 mapping->host->i_ino, len, (long long) pos);
343 page = __grab_cache_page(mapping, index);
344 if (!page)
345 return -ENOMEM;
346 *pagep = page;
348 ret = nfs_flush_incompatible(file, page);
349 if (ret) {
350 unlock_page(page);
351 page_cache_release(page);
353 return ret;
356 static int nfs_write_end(struct file *file, struct address_space *mapping,
357 loff_t pos, unsigned len, unsigned copied,
358 struct page *page, void *fsdata)
360 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
361 int status;
363 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
364 file->f_path.dentry->d_parent->d_name.name,
365 file->f_path.dentry->d_name.name,
366 mapping->host->i_ino, len, (long long) pos);
369 * Zero any uninitialised parts of the page, and then mark the page
370 * as up to date if it turns out that we're extending the file.
372 if (!PageUptodate(page)) {
373 unsigned pglen = nfs_page_length(page);
374 unsigned end = offset + len;
376 if (pglen == 0) {
377 zero_user_segments(page, 0, offset,
378 end, PAGE_CACHE_SIZE);
379 SetPageUptodate(page);
380 } else if (end >= pglen) {
381 zero_user_segment(page, end, PAGE_CACHE_SIZE);
382 if (offset == 0)
383 SetPageUptodate(page);
384 } else
385 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
388 lock_kernel();
389 status = nfs_updatepage(file, page, offset, copied);
390 unlock_kernel();
392 unlock_page(page);
393 page_cache_release(page);
395 if (status < 0)
396 return status;
397 return copied;
400 static void nfs_invalidate_page(struct page *page, unsigned long offset)
402 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
404 if (offset != 0)
405 return;
406 /* Cancel any unstarted writes on this page */
407 nfs_wb_page_cancel(page->mapping->host, page);
410 static int nfs_release_page(struct page *page, gfp_t gfp)
412 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
414 /* If PagePrivate() is set, then the page is not freeable */
415 return 0;
418 static int nfs_launder_page(struct page *page)
420 struct inode *inode = page->mapping->host;
422 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
423 inode->i_ino, (long long)page_offset(page));
425 return nfs_wb_page(inode, page);
428 const struct address_space_operations nfs_file_aops = {
429 .readpage = nfs_readpage,
430 .readpages = nfs_readpages,
431 .set_page_dirty = __set_page_dirty_nobuffers,
432 .writepage = nfs_writepage,
433 .writepages = nfs_writepages,
434 .write_begin = nfs_write_begin,
435 .write_end = nfs_write_end,
436 .invalidatepage = nfs_invalidate_page,
437 .releasepage = nfs_release_page,
438 .direct_IO = nfs_direct_IO,
439 .launder_page = nfs_launder_page,
442 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
444 struct file *filp = vma->vm_file;
445 struct dentry *dentry = filp->f_path.dentry;
446 unsigned pagelen;
447 int ret = -EINVAL;
448 struct address_space *mapping;
450 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
451 dentry->d_parent->d_name.name, dentry->d_name.name,
452 filp->f_mapping->host->i_ino,
453 (long long)page_offset(page));
455 lock_page(page);
456 mapping = page->mapping;
457 if (mapping != dentry->d_inode->i_mapping)
458 goto out_unlock;
460 ret = 0;
461 pagelen = nfs_page_length(page);
462 if (pagelen == 0)
463 goto out_unlock;
465 ret = nfs_flush_incompatible(filp, page);
466 if (ret != 0)
467 goto out_unlock;
469 ret = nfs_updatepage(filp, page, 0, pagelen);
470 if (ret == 0)
471 ret = pagelen;
472 out_unlock:
473 unlock_page(page);
474 return ret;
477 static struct vm_operations_struct nfs_file_vm_ops = {
478 .fault = filemap_fault,
479 .page_mkwrite = nfs_vm_page_mkwrite,
482 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
484 struct nfs_open_context *ctx;
486 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
487 return 1;
488 ctx = nfs_file_open_context(filp);
489 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
490 return 1;
491 return 0;
494 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
495 unsigned long nr_segs, loff_t pos)
497 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
498 struct inode * inode = dentry->d_inode;
499 ssize_t result;
500 size_t count = iov_length(iov, nr_segs);
502 if (iocb->ki_filp->f_flags & O_DIRECT)
503 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
505 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
506 dentry->d_parent->d_name.name, dentry->d_name.name,
507 inode->i_ino, (unsigned long) count, (long long) pos);
509 result = -EBUSY;
510 if (IS_SWAPFILE(inode))
511 goto out_swapfile;
513 * O_APPEND implies that we must revalidate the file length.
515 if (iocb->ki_filp->f_flags & O_APPEND) {
516 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
517 if (result)
518 goto out;
521 result = count;
522 if (!count)
523 goto out;
525 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
526 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
527 /* Return error values for O_SYNC and IS_SYNC() */
528 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
529 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
530 if (err < 0)
531 result = err;
533 out:
534 return result;
536 out_swapfile:
537 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
538 goto out;
541 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
543 struct inode *inode = filp->f_mapping->host;
544 int status = 0;
546 lock_kernel();
547 /* Try local locking first */
548 posix_test_lock(filp, fl);
549 if (fl->fl_type != F_UNLCK) {
550 /* found a conflict */
551 goto out;
554 if (nfs_have_delegation(inode, FMODE_READ))
555 goto out_noconflict;
557 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
558 goto out_noconflict;
560 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
561 out:
562 unlock_kernel();
563 return status;
564 out_noconflict:
565 fl->fl_type = F_UNLCK;
566 goto out;
569 static int do_vfs_lock(struct file *file, struct file_lock *fl)
571 int res = 0;
572 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
573 case FL_POSIX:
574 res = posix_lock_file_wait(file, fl);
575 break;
576 case FL_FLOCK:
577 res = flock_lock_file_wait(file, fl);
578 break;
579 default:
580 BUG();
582 if (res < 0)
583 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
584 " - error %d!\n",
585 __func__, res);
586 return res;
589 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
591 struct inode *inode = filp->f_mapping->host;
592 int status;
595 * Flush all pending writes before doing anything
596 * with locks..
598 nfs_sync_mapping(filp->f_mapping);
600 /* NOTE: special case
601 * If we're signalled while cleaning up locks on process exit, we
602 * still need to complete the unlock.
604 lock_kernel();
605 /* Use local locking if mounted with "-onolock" */
606 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
607 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
608 else
609 status = do_vfs_lock(filp, fl);
610 unlock_kernel();
611 return status;
614 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
616 struct inode *inode = filp->f_mapping->host;
617 int status;
620 * Flush all pending writes before doing anything
621 * with locks..
623 status = nfs_sync_mapping(filp->f_mapping);
624 if (status != 0)
625 goto out;
627 lock_kernel();
628 /* Use local locking if mounted with "-onolock" */
629 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
630 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
631 else
632 status = do_vfs_lock(filp, fl);
633 unlock_kernel();
634 if (status < 0)
635 goto out;
637 * Make sure we clear the cache whenever we try to get the lock.
638 * This makes locking act as a cache coherency point.
640 nfs_sync_mapping(filp->f_mapping);
641 if (!nfs_have_delegation(inode, FMODE_READ))
642 nfs_zap_caches(inode);
643 out:
644 return status;
648 * Lock a (portion of) a file
650 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
652 struct inode * inode = filp->f_mapping->host;
653 int ret = -ENOLCK;
655 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
656 inode->i_sb->s_id, inode->i_ino,
657 fl->fl_type, fl->fl_flags,
658 (long long)fl->fl_start, (long long)fl->fl_end);
659 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
661 /* No mandatory locks over NFS */
662 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
663 goto out_err;
665 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
666 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
667 if (ret < 0)
668 goto out_err;
671 if (IS_GETLK(cmd))
672 ret = do_getlk(filp, cmd, fl);
673 else if (fl->fl_type == F_UNLCK)
674 ret = do_unlk(filp, cmd, fl);
675 else
676 ret = do_setlk(filp, cmd, fl);
677 out_err:
678 return ret;
682 * Lock a (portion of) a file
684 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
686 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
687 filp->f_path.dentry->d_inode->i_sb->s_id,
688 filp->f_path.dentry->d_inode->i_ino,
689 fl->fl_type, fl->fl_flags);
692 * No BSD flocks over NFS allowed.
693 * Note: we could try to fake a POSIX lock request here by
694 * using ((u32) filp | 0x80000000) or some such as the pid.
695 * Not sure whether that would be unique, though, or whether
696 * that would break in other places.
698 if (!(fl->fl_flags & FL_FLOCK))
699 return -ENOLCK;
701 /* We're simulating flock() locks using posix locks on the server */
702 fl->fl_owner = (fl_owner_t)filp;
703 fl->fl_start = 0;
704 fl->fl_end = OFFSET_MAX;
706 if (fl->fl_type == F_UNLCK)
707 return do_unlk(filp, cmd, fl);
708 return do_setlk(filp, cmd, fl);
711 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
714 * There is no protocol support for leases, so we have no way
715 * to implement them correctly in the face of opens by other
716 * clients.
718 return -EINVAL;