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/buffer_head.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_sendfile(struct file *, loff_t *, size_t, read_actor_t, void *);
46 static ssize_t nfs_file_read(struct kiocb *, char __user *, size_t, loff_t);
47 static ssize_t nfs_file_write(struct kiocb *, const char __user *, size_t, loff_t);
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 */
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, char __user * buf, size_t count, loff_t pos)
203 struct dentry * dentry = iocb->ki_filp->f_dentry;
204 struct inode * inode = dentry->d_inode;
207 #ifdef CONFIG_NFS_DIRECTIO
208 if (iocb->ki_filp->f_flags & O_DIRECT)
209 return nfs_file_direct_read(iocb, buf, count, pos);
212 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
213 dentry->d_parent->d_name.name, dentry->d_name.name,
214 (unsigned long) count, (unsigned long) pos);
216 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
217 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
219 result = generic_file_aio_read(iocb, buf, count, pos);
224 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
225 read_actor_t actor, void *target)
227 struct dentry *dentry = filp->f_dentry;
228 struct inode *inode = dentry->d_inode;
231 dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
232 dentry->d_parent->d_name.name, dentry->d_name.name,
233 (unsigned long) count, (unsigned long long) *ppos);
235 res = nfs_revalidate_mapping(inode, filp->f_mapping);
237 res = generic_file_sendfile(filp, ppos, count, actor, target);
242 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
244 struct dentry *dentry = file->f_dentry;
245 struct inode *inode = dentry->d_inode;
248 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
249 dentry->d_parent->d_name.name, dentry->d_name.name);
251 status = nfs_revalidate_mapping(inode, file->f_mapping);
253 status = generic_file_mmap(file, vma);
256 nfs_fscache_install_vm_ops(inode, vma);
262 * Flush any dirty pages for this process, and check for write errors.
263 * The return status from this call provides a reliable indication of
264 * whether any write errors occurred for this process.
267 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
269 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
270 struct inode *inode = dentry->d_inode;
273 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
275 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
277 status = nfs_wb_all(inode);
287 * This does the "real" work of the write. The generic routine has
288 * allocated the page, locked it, done all the page alignment stuff
289 * calculations etc. Now we should just copy the data from user
290 * space and write it back to the real medium..
292 * If the writer ends up delaying the write, the writer needs to
293 * increment the page use counts until he is done with the page.
295 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
297 return nfs_flush_incompatible(file, page);
300 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
305 status = nfs_updatepage(file, page, offset, to-offset);
310 static void nfs_invalidate_page(struct page *page, unsigned long offset)
312 struct inode *inode = page->mapping->host;
314 /* Cancel any unstarted writes on this page */
316 nfs_sync_inode_wait(inode, page->index, 1, FLUSH_INVALIDATE);
318 nfs_fscache_invalidate_page(page, inode, offset);
320 /* we can do this here as the bits are only set with the page lock
321 * held, and our caller is holding that */
323 ClearPagePrivate(page);
326 static int nfs_release_page(struct page *page, gfp_t gfp)
328 if (gfp & __GFP_FS) {
329 int error = nfs_wb_page(page->mapping->host, page);
332 nfs_fscache_release_page(page);
334 /* may have been set due to either caching or writing */
335 ClearPagePrivate(page);
338 /* releasepage() returns true/false */
339 return (error == 0) ? 1 : 0;
343 * Avoid deadlock on nfs_wait_on_request().
349 * Since we use page->private for our own nefarious purposes when using
350 * fscache, we have to override extra address space ops to prevent fs/buffer.c
351 * from getting confused, even though we may not have asked its opinion
353 const struct address_space_operations nfs_file_aops = {
354 .readpage = nfs_readpage,
355 .readpages = nfs_readpages,
356 .set_page_dirty = __set_page_dirty_nobuffers,
357 .writepage = nfs_writepage,
358 .writepages = nfs_writepages,
359 .prepare_write = nfs_prepare_write,
360 .commit_write = nfs_commit_write,
361 .invalidatepage = nfs_invalidate_page,
362 .releasepage = nfs_release_page,
363 #ifdef CONFIG_NFS_DIRECTIO
364 .direct_IO = nfs_direct_IO,
366 #ifdef CONFIG_NFS_FSCACHE
367 .sync_page = block_sync_page,
372 * Write to a file (through the page cache).
375 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
377 struct dentry * dentry = iocb->ki_filp->f_dentry;
378 struct inode * inode = dentry->d_inode;
381 #ifdef CONFIG_NFS_DIRECTIO
382 if (iocb->ki_filp->f_flags & O_DIRECT)
383 return nfs_file_direct_write(iocb, buf, count, pos);
386 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
387 dentry->d_parent->d_name.name, dentry->d_name.name,
388 inode->i_ino, (unsigned long) count, (unsigned long) pos);
391 if (IS_SWAPFILE(inode))
394 * O_APPEND implies that we must revalidate the file length.
396 if (iocb->ki_filp->f_flags & O_APPEND) {
397 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
406 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
407 result = generic_file_aio_write(iocb, buf, count, pos);
412 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
416 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
418 struct file_lock cfl;
419 struct inode *inode = filp->f_mapping->host;
423 /* Try local locking first */
424 if (posix_test_lock(filp, fl, &cfl)) {
425 fl->fl_start = cfl.fl_start;
426 fl->fl_end = cfl.fl_end;
427 fl->fl_type = cfl.fl_type;
428 fl->fl_pid = cfl.fl_pid;
432 if (nfs_have_delegation(inode, FMODE_READ))
435 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
438 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
443 fl->fl_type = F_UNLCK;
447 static int do_vfs_lock(struct file *file, struct file_lock *fl)
450 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
452 res = posix_lock_file_wait(file, fl);
455 res = flock_lock_file_wait(file, fl);
461 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
466 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
468 struct inode *inode = filp->f_mapping->host;
472 * Flush all pending writes before doing anything
475 nfs_sync_mapping(filp->f_mapping);
477 /* NOTE: special case
478 * If we're signalled while cleaning up locks on process exit, we
479 * still need to complete the unlock.
482 /* Use local locking if mounted with "-onolock" */
483 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
484 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
486 status = do_vfs_lock(filp, fl);
491 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
493 struct inode *inode = filp->f_mapping->host;
497 * Flush all pending writes before doing anything
500 status = nfs_sync_mapping(filp->f_mapping);
505 /* Use local locking if mounted with "-onolock" */
506 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
507 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
508 /* If we were signalled we still need to ensure that
509 * we clean up any state on the server. We therefore
510 * record the lock call as having succeeded in order to
511 * ensure that locks_remove_posix() cleans it out when
514 if (status == -EINTR || status == -ERESTARTSYS)
515 do_vfs_lock(filp, fl);
517 status = do_vfs_lock(filp, fl);
522 * Make sure we clear the cache whenever we try to get the lock.
523 * This makes locking act as a cache coherency point.
525 nfs_sync_mapping(filp->f_mapping);
526 nfs_zap_caches(inode);
532 * Lock a (portion of) a file
534 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
536 struct inode * inode = filp->f_mapping->host;
538 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
539 inode->i_sb->s_id, inode->i_ino,
540 fl->fl_type, fl->fl_flags,
541 (long long)fl->fl_start, (long long)fl->fl_end);
542 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
544 /* No mandatory locks over NFS */
545 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
546 fl->fl_type != F_UNLCK)
550 return do_getlk(filp, cmd, fl);
551 if (fl->fl_type == F_UNLCK)
552 return do_unlk(filp, cmd, fl);
553 return do_setlk(filp, cmd, fl);
557 * Lock a (portion of) a file
559 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
561 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
562 filp->f_dentry->d_inode->i_sb->s_id,
563 filp->f_dentry->d_inode->i_ino,
564 fl->fl_type, fl->fl_flags);
567 * No BSD flocks over NFS allowed.
568 * Note: we could try to fake a POSIX lock request here by
569 * using ((u32) filp | 0x80000000) or some such as the pid.
570 * Not sure whether that would be unique, though, or whether
571 * that would break in other places.
573 if (!(fl->fl_flags & FL_FLOCK))
576 /* We're simulating flock() locks using posix locks on the server */
577 fl->fl_owner = (fl_owner_t)filp;
579 fl->fl_end = OFFSET_MAX;
581 if (fl->fl_type == F_UNLCK)
582 return do_unlk(filp, cmd, fl);
583 return do_setlk(filp, cmd, fl);