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 *, const struct iovec *iov,
45 unsigned long nr_segs, loff_t pos);
46 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
47 unsigned long nr_segs, loff_t pos);
48 static int nfs_file_flush(struct file *, fl_owner_t id);
49 static int nfs_fsync(struct file *, struct dentry *dentry, int datasync);
50 static int nfs_check_flags(int flags);
51 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
52 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
54 const struct file_operations nfs_file_operations = {
55 .llseek = nfs_file_llseek,
57 .write = do_sync_write,
58 .aio_read = nfs_file_read,
59 .aio_write = nfs_file_write,
60 .mmap = nfs_file_mmap,
61 .open = nfs_file_open,
62 .flush = nfs_file_flush,
63 .release = nfs_file_release,
67 .sendfile = nfs_file_sendfile,
68 .check_flags = nfs_check_flags,
71 struct inode_operations nfs_file_inode_operations = {
72 .permission = nfs_permission,
73 .getattr = nfs_getattr,
74 .setattr = nfs_setattr,
78 struct inode_operations nfs3_file_inode_operations = {
79 .permission = nfs_permission,
80 .getattr = nfs_getattr,
81 .setattr = nfs_setattr,
82 .listxattr = nfs3_listxattr,
83 .getxattr = nfs3_getxattr,
84 .setxattr = nfs3_setxattr,
85 .removexattr = nfs3_removexattr,
87 #endif /* CONFIG_NFS_v3 */
89 /* Hack for future NFS swap support */
91 # define IS_SWAPFILE(inode) (0)
94 static int nfs_check_flags(int flags)
96 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
106 nfs_file_open(struct inode *inode, struct file *filp)
110 res = nfs_check_flags(filp->f_flags);
114 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
116 res = NFS_PROTO(inode)->file_open(inode, filp);
122 nfs_file_release(struct inode *inode, struct file *filp)
124 /* Ensure that dirty pages are flushed out with the right creds */
125 if (filp->f_mode & FMODE_WRITE)
126 filemap_fdatawrite(filp->f_mapping);
127 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
128 return NFS_PROTO(inode)->file_release(inode, filp);
132 * nfs_revalidate_size - Revalidate the file size
133 * @inode - pointer to inode struct
134 * @file - pointer to struct file
136 * Revalidates the file length. This is basically a wrapper around
137 * nfs_revalidate_inode() that takes into account the fact that we may
138 * have cached writes (in which case we don't care about the server's
139 * idea of what the file length is), or O_DIRECT (in which case we
140 * shouldn't trust the cache).
142 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
144 struct nfs_server *server = NFS_SERVER(inode);
145 struct nfs_inode *nfsi = NFS_I(inode);
147 if (server->flags & NFS_MOUNT_NOAC)
149 if (filp->f_flags & O_DIRECT)
151 if (nfsi->npages != 0)
153 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
156 return __nfs_revalidate_inode(server, inode);
159 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
161 /* origin == SEEK_END => we must revalidate the cached file length */
162 if (origin == SEEK_END) {
163 struct inode *inode = filp->f_mapping->host;
164 int retval = nfs_revalidate_file_size(inode, filp);
166 return (loff_t)retval;
168 return remote_llseek(filp, offset, origin);
172 * Flush all dirty pages, and check for write errors.
176 nfs_file_flush(struct file *file, fl_owner_t id)
178 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
179 struct inode *inode = file->f_dentry->d_inode;
182 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
184 if ((file->f_mode & FMODE_WRITE) == 0)
186 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
188 /* Ensure that data+attribute caches are up to date after close() */
189 status = nfs_wb_all(inode);
194 nfs_revalidate_inode(NFS_SERVER(inode), inode);
201 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
202 unsigned long nr_segs, loff_t pos)
204 struct dentry * dentry = iocb->ki_filp->f_dentry;
205 struct inode * inode = dentry->d_inode;
207 size_t count = iov_length(iov, nr_segs);
209 #ifdef CONFIG_NFS_DIRECTIO
210 if (iocb->ki_filp->f_flags & O_DIRECT)
211 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
214 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
215 dentry->d_parent->d_name.name, dentry->d_name.name,
216 (unsigned long) count, (unsigned long) pos);
218 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
219 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
221 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
226 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
227 read_actor_t actor, void *target)
229 struct dentry *dentry = filp->f_dentry;
230 struct inode *inode = dentry->d_inode;
233 dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
234 dentry->d_parent->d_name.name, dentry->d_name.name,
235 (unsigned long) count, (unsigned long long) *ppos);
237 res = nfs_revalidate_mapping(inode, filp->f_mapping);
239 res = generic_file_sendfile(filp, ppos, count, actor, target);
244 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
246 struct dentry *dentry = file->f_dentry;
247 struct inode *inode = dentry->d_inode;
250 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
251 dentry->d_parent->d_name.name, dentry->d_name.name);
253 status = nfs_revalidate_mapping(inode, file->f_mapping);
255 status = generic_file_mmap(file, vma);
260 * Flush any dirty pages for this process, and check for write errors.
261 * The return status from this call provides a reliable indication of
262 * whether any write errors occurred for this process.
265 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
267 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
268 struct inode *inode = dentry->d_inode;
271 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
273 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
275 status = nfs_wb_all(inode);
285 * This does the "real" work of the write. The generic routine has
286 * allocated the page, locked it, done all the page alignment stuff
287 * calculations etc. Now we should just copy the data from user
288 * space and write it back to the real medium..
290 * If the writer ends up delaying the write, the writer needs to
291 * increment the page use counts until he is done with the page.
293 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
295 return nfs_flush_incompatible(file, page);
298 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
303 status = nfs_updatepage(file, page, offset, to-offset);
308 static void nfs_invalidate_page(struct page *page, unsigned long offset)
310 struct inode *inode = page->mapping->host;
312 /* Cancel any unstarted writes on this page */
314 nfs_sync_inode_wait(inode, page->index, 1, FLUSH_INVALIDATE);
317 static int nfs_release_page(struct page *page, gfp_t gfp)
320 return !nfs_wb_page(page->mapping->host, page);
323 * Avoid deadlock on nfs_wait_on_request().
328 const struct address_space_operations nfs_file_aops = {
329 .readpage = nfs_readpage,
330 .readpages = nfs_readpages,
331 .set_page_dirty = __set_page_dirty_nobuffers,
332 .writepage = nfs_writepage,
333 .writepages = nfs_writepages,
334 .prepare_write = nfs_prepare_write,
335 .commit_write = nfs_commit_write,
336 .invalidatepage = nfs_invalidate_page,
337 .releasepage = nfs_release_page,
338 #ifdef CONFIG_NFS_DIRECTIO
339 .direct_IO = nfs_direct_IO,
343 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
344 unsigned long nr_segs, loff_t pos)
346 struct dentry * dentry = iocb->ki_filp->f_dentry;
347 struct inode * inode = dentry->d_inode;
349 size_t count = iov_length(iov, nr_segs);
351 #ifdef CONFIG_NFS_DIRECTIO
352 if (iocb->ki_filp->f_flags & O_DIRECT)
353 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
356 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
357 dentry->d_parent->d_name.name, dentry->d_name.name,
358 inode->i_ino, (unsigned long) count, (long long) pos);
361 if (IS_SWAPFILE(inode))
364 * O_APPEND implies that we must revalidate the file length.
366 if (iocb->ki_filp->f_flags & O_APPEND) {
367 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
376 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
377 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
382 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
386 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
388 struct file_lock cfl;
389 struct inode *inode = filp->f_mapping->host;
393 /* Try local locking first */
394 if (posix_test_lock(filp, fl, &cfl)) {
395 fl->fl_start = cfl.fl_start;
396 fl->fl_end = cfl.fl_end;
397 fl->fl_type = cfl.fl_type;
398 fl->fl_pid = cfl.fl_pid;
402 if (nfs_have_delegation(inode, FMODE_READ))
405 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
408 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
413 fl->fl_type = F_UNLCK;
417 static int do_vfs_lock(struct file *file, struct file_lock *fl)
420 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
422 res = posix_lock_file_wait(file, fl);
425 res = flock_lock_file_wait(file, fl);
431 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
436 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
438 struct inode *inode = filp->f_mapping->host;
442 * Flush all pending writes before doing anything
445 nfs_sync_mapping(filp->f_mapping);
447 /* NOTE: special case
448 * If we're signalled while cleaning up locks on process exit, we
449 * still need to complete the unlock.
452 /* Use local locking if mounted with "-onolock" */
453 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
454 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
456 status = do_vfs_lock(filp, fl);
461 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
463 struct inode *inode = filp->f_mapping->host;
467 * Flush all pending writes before doing anything
470 status = nfs_sync_mapping(filp->f_mapping);
475 /* Use local locking if mounted with "-onolock" */
476 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
477 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
478 /* If we were signalled we still need to ensure that
479 * we clean up any state on the server. We therefore
480 * record the lock call as having succeeded in order to
481 * ensure that locks_remove_posix() cleans it out when
484 if (status == -EINTR || status == -ERESTARTSYS)
485 do_vfs_lock(filp, fl);
487 status = do_vfs_lock(filp, fl);
492 * Make sure we clear the cache whenever we try to get the lock.
493 * This makes locking act as a cache coherency point.
495 nfs_sync_mapping(filp->f_mapping);
496 nfs_zap_caches(inode);
502 * Lock a (portion of) a file
504 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
506 struct inode * inode = filp->f_mapping->host;
508 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
509 inode->i_sb->s_id, inode->i_ino,
510 fl->fl_type, fl->fl_flags,
511 (long long)fl->fl_start, (long long)fl->fl_end);
512 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
514 /* No mandatory locks over NFS */
515 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
516 fl->fl_type != F_UNLCK)
520 return do_getlk(filp, cmd, fl);
521 if (fl->fl_type == F_UNLCK)
522 return do_unlk(filp, cmd, fl);
523 return do_setlk(filp, cmd, fl);
527 * Lock a (portion of) a file
529 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
531 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
532 filp->f_dentry->d_inode->i_sb->s_id,
533 filp->f_dentry->d_inode->i_ino,
534 fl->fl_type, fl->fl_flags);
537 * No BSD flocks over NFS allowed.
538 * Note: we could try to fake a POSIX lock request here by
539 * using ((u32) filp | 0x80000000) or some such as the pid.
540 * Not sure whether that would be unique, though, or whether
541 * that would break in other places.
543 if (!(fl->fl_flags & FL_FLOCK))
546 /* We're simulating flock() locks using posix locks on the server */
547 fl->fl_owner = (fl_owner_t)filp;
549 fl->fl_end = OFFSET_MAX;
551 if (fl->fl_type == F_UNLCK)
552 return do_unlk(filp, cmd, fl);
553 return do_setlk(filp, cmd, fl);