cpuset: limit the input of cpuset.sched_relax_domain_level
[linux-2.6/kmemtrace.git] / fs / nfs / file.c
blobd84a3d8f32af395f8109b58dcdd232fc3d0ddaa3
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 /* origin == SEEK_END => we must revalidate the cached file length */
174 if (origin == SEEK_END) {
175 struct inode *inode = filp->f_mapping->host;
176 int retval = nfs_revalidate_file_size(inode, filp);
177 if (retval < 0)
178 return (loff_t)retval;
180 return remote_llseek(filp, offset, origin);
184 * Helper for nfs_file_flush() and nfs_fsync()
186 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
187 * disk, but it retrieves and clears ctx->error after synching, despite
188 * the two being set at the same time in nfs_context_set_write_error().
189 * This is because the former is used to notify the _next_ call to
190 * nfs_file_write() that a write error occured, and hence cause it to
191 * fall back to doing a synchronous write.
193 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
195 int have_error, status;
196 int ret = 0;
198 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
199 status = nfs_wb_all(inode);
200 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
201 if (have_error)
202 ret = xchg(&ctx->error, 0);
203 if (!ret)
204 ret = status;
205 return ret;
209 * Flush all dirty pages, and check for write errors.
212 static int
213 nfs_file_flush(struct file *file, fl_owner_t id)
215 struct nfs_open_context *ctx = nfs_file_open_context(file);
216 struct inode *inode = file->f_path.dentry->d_inode;
217 int status;
219 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
221 if ((file->f_mode & FMODE_WRITE) == 0)
222 return 0;
223 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
225 /* Ensure that data+attribute caches are up to date after close() */
226 status = nfs_do_fsync(ctx, inode);
227 if (!status)
228 nfs_revalidate_inode(NFS_SERVER(inode), inode);
229 return status;
232 static ssize_t
233 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
234 unsigned long nr_segs, loff_t pos)
236 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
237 struct inode * inode = dentry->d_inode;
238 ssize_t result;
239 size_t count = iov_length(iov, nr_segs);
241 if (iocb->ki_filp->f_flags & O_DIRECT)
242 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
244 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
245 dentry->d_parent->d_name.name, dentry->d_name.name,
246 (unsigned long) count, (unsigned long) pos);
248 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
249 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
250 if (!result)
251 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
252 return result;
255 static ssize_t
256 nfs_file_splice_read(struct file *filp, loff_t *ppos,
257 struct pipe_inode_info *pipe, size_t count,
258 unsigned int flags)
260 struct dentry *dentry = filp->f_path.dentry;
261 struct inode *inode = dentry->d_inode;
262 ssize_t res;
264 dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
265 dentry->d_parent->d_name.name, dentry->d_name.name,
266 (unsigned long) count, (unsigned long long) *ppos);
268 res = nfs_revalidate_mapping(inode, filp->f_mapping);
269 if (!res)
270 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
271 return res;
274 static int
275 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
277 struct dentry *dentry = file->f_path.dentry;
278 struct inode *inode = dentry->d_inode;
279 int status;
281 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
282 dentry->d_parent->d_name.name, dentry->d_name.name);
284 status = nfs_revalidate_mapping(inode, file->f_mapping);
285 if (!status) {
286 vma->vm_ops = &nfs_file_vm_ops;
287 vma->vm_flags |= VM_CAN_NONLINEAR;
288 file_accessed(file);
290 return status;
294 * Flush any dirty pages for this process, and check for write errors.
295 * The return status from this call provides a reliable indication of
296 * whether any write errors occurred for this process.
298 static int
299 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
301 struct nfs_open_context *ctx = nfs_file_open_context(file);
302 struct inode *inode = dentry->d_inode;
304 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
306 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
307 return nfs_do_fsync(ctx, inode);
311 * This does the "real" work of the write. We must allocate and lock the
312 * page to be sent back to the generic routine, which then copies the
313 * data from user space.
315 * If the writer ends up delaying the write, the writer needs to
316 * increment the page use counts until he is done with the page.
318 static int nfs_write_begin(struct file *file, struct address_space *mapping,
319 loff_t pos, unsigned len, unsigned flags,
320 struct page **pagep, void **fsdata)
322 int ret;
323 pgoff_t index;
324 struct page *page;
325 index = pos >> PAGE_CACHE_SHIFT;
327 page = __grab_cache_page(mapping, index);
328 if (!page)
329 return -ENOMEM;
330 *pagep = page;
332 ret = nfs_flush_incompatible(file, page);
333 if (ret) {
334 unlock_page(page);
335 page_cache_release(page);
337 return ret;
340 static int nfs_write_end(struct file *file, struct address_space *mapping,
341 loff_t pos, unsigned len, unsigned copied,
342 struct page *page, void *fsdata)
344 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
345 int status;
347 lock_kernel();
348 status = nfs_updatepage(file, page, offset, copied);
349 unlock_kernel();
351 unlock_page(page);
352 page_cache_release(page);
354 if (status < 0)
355 return status;
356 return copied;
359 static void nfs_invalidate_page(struct page *page, unsigned long offset)
361 if (offset != 0)
362 return;
363 /* Cancel any unstarted writes on this page */
364 nfs_wb_page_cancel(page->mapping->host, page);
367 static int nfs_release_page(struct page *page, gfp_t gfp)
369 /* If PagePrivate() is set, then the page is not freeable */
370 return 0;
373 static int nfs_launder_page(struct page *page)
375 return nfs_wb_page(page->mapping->host, page);
378 const struct address_space_operations nfs_file_aops = {
379 .readpage = nfs_readpage,
380 .readpages = nfs_readpages,
381 .set_page_dirty = __set_page_dirty_nobuffers,
382 .writepage = nfs_writepage,
383 .writepages = nfs_writepages,
384 .write_begin = nfs_write_begin,
385 .write_end = nfs_write_end,
386 .invalidatepage = nfs_invalidate_page,
387 .releasepage = nfs_release_page,
388 .direct_IO = nfs_direct_IO,
389 .launder_page = nfs_launder_page,
392 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
394 struct file *filp = vma->vm_file;
395 unsigned pagelen;
396 int ret = -EINVAL;
397 struct address_space *mapping;
399 lock_page(page);
400 mapping = page->mapping;
401 if (mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping)
402 goto out_unlock;
404 ret = 0;
405 pagelen = nfs_page_length(page);
406 if (pagelen == 0)
407 goto out_unlock;
409 ret = nfs_flush_incompatible(filp, page);
410 if (ret != 0)
411 goto out_unlock;
413 ret = nfs_updatepage(filp, page, 0, pagelen);
414 if (ret == 0)
415 ret = pagelen;
416 out_unlock:
417 unlock_page(page);
418 return ret;
421 static struct vm_operations_struct nfs_file_vm_ops = {
422 .fault = filemap_fault,
423 .page_mkwrite = nfs_vm_page_mkwrite,
426 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
428 struct nfs_open_context *ctx;
430 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
431 return 1;
432 ctx = nfs_file_open_context(filp);
433 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
434 return 1;
435 return 0;
438 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
439 unsigned long nr_segs, loff_t pos)
441 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
442 struct inode * inode = dentry->d_inode;
443 ssize_t result;
444 size_t count = iov_length(iov, nr_segs);
446 if (iocb->ki_filp->f_flags & O_DIRECT)
447 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
449 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
450 dentry->d_parent->d_name.name, dentry->d_name.name,
451 inode->i_ino, (unsigned long) count, (long long) pos);
453 result = -EBUSY;
454 if (IS_SWAPFILE(inode))
455 goto out_swapfile;
457 * O_APPEND implies that we must revalidate the file length.
459 if (iocb->ki_filp->f_flags & O_APPEND) {
460 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
461 if (result)
462 goto out;
465 result = count;
466 if (!count)
467 goto out;
469 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
470 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
471 /* Return error values for O_SYNC and IS_SYNC() */
472 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
473 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
474 if (err < 0)
475 result = err;
477 out:
478 return result;
480 out_swapfile:
481 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
482 goto out;
485 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
487 struct inode *inode = filp->f_mapping->host;
488 int status = 0;
490 lock_kernel();
491 /* Try local locking first */
492 posix_test_lock(filp, fl);
493 if (fl->fl_type != F_UNLCK) {
494 /* found a conflict */
495 goto out;
498 if (nfs_have_delegation(inode, FMODE_READ))
499 goto out_noconflict;
501 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
502 goto out_noconflict;
504 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
505 out:
506 unlock_kernel();
507 return status;
508 out_noconflict:
509 fl->fl_type = F_UNLCK;
510 goto out;
513 static int do_vfs_lock(struct file *file, struct file_lock *fl)
515 int res = 0;
516 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
517 case FL_POSIX:
518 res = posix_lock_file_wait(file, fl);
519 break;
520 case FL_FLOCK:
521 res = flock_lock_file_wait(file, fl);
522 break;
523 default:
524 BUG();
526 if (res < 0)
527 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
528 " - error %d!\n",
529 __func__, res);
530 return res;
533 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
535 struct inode *inode = filp->f_mapping->host;
536 int status;
539 * Flush all pending writes before doing anything
540 * with locks..
542 nfs_sync_mapping(filp->f_mapping);
544 /* NOTE: special case
545 * If we're signalled while cleaning up locks on process exit, we
546 * still need to complete the unlock.
548 lock_kernel();
549 /* Use local locking if mounted with "-onolock" */
550 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
551 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
552 else
553 status = do_vfs_lock(filp, fl);
554 unlock_kernel();
555 return status;
558 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
560 struct inode *inode = filp->f_mapping->host;
561 int status;
564 * Flush all pending writes before doing anything
565 * with locks..
567 status = nfs_sync_mapping(filp->f_mapping);
568 if (status != 0)
569 goto out;
571 lock_kernel();
572 /* Use local locking if mounted with "-onolock" */
573 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
574 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
575 else
576 status = do_vfs_lock(filp, fl);
577 unlock_kernel();
578 if (status < 0)
579 goto out;
581 * Make sure we clear the cache whenever we try to get the lock.
582 * This makes locking act as a cache coherency point.
584 nfs_sync_mapping(filp->f_mapping);
585 nfs_zap_caches(inode);
586 out:
587 return status;
591 * Lock a (portion of) a file
593 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
595 struct inode * inode = filp->f_mapping->host;
597 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
598 inode->i_sb->s_id, inode->i_ino,
599 fl->fl_type, fl->fl_flags,
600 (long long)fl->fl_start, (long long)fl->fl_end);
601 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
603 /* No mandatory locks over NFS */
604 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
605 return -ENOLCK;
607 if (IS_GETLK(cmd))
608 return do_getlk(filp, cmd, fl);
609 if (fl->fl_type == F_UNLCK)
610 return do_unlk(filp, cmd, fl);
611 return do_setlk(filp, cmd, fl);
615 * Lock a (portion of) a file
617 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
619 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
620 filp->f_path.dentry->d_inode->i_sb->s_id,
621 filp->f_path.dentry->d_inode->i_ino,
622 fl->fl_type, fl->fl_flags);
625 * No BSD flocks over NFS allowed.
626 * Note: we could try to fake a POSIX lock request here by
627 * using ((u32) filp | 0x80000000) or some such as the pid.
628 * Not sure whether that would be unique, though, or whether
629 * that would break in other places.
631 if (!(fl->fl_flags & FL_FLOCK))
632 return -ENOLCK;
634 /* We're simulating flock() locks using posix locks on the server */
635 fl->fl_owner = (fl_owner_t)filp;
636 fl->fl_start = 0;
637 fl->fl_end = OFFSET_MAX;
639 if (fl->fl_type == F_UNLCK)
640 return do_unlk(filp, cmd, fl);
641 return do_setlk(filp, cmd, fl);
644 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
647 * There is no protocol support for leases, so we have no way
648 * to implement them correctly in the face of opens by other
649 * clients.
651 return -EINVAL;