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/aio.h>
32 #include <asm/uaccess.h>
33 #include <asm/system.h>
35 #include "delegation.h"
38 #define NFSDBG_FACILITY NFSDBG_FILE
40 static int nfs_file_open(struct inode *, struct file *);
41 static int nfs_file_release(struct inode *, struct file *);
42 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
43 static int nfs_file_mmap(struct file *, struct vm_area_struct *);
44 static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
45 struct pipe_inode_info *pipe,
46 size_t count, unsigned int flags);
47 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
48 unsigned long nr_segs, loff_t pos);
49 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
50 unsigned long nr_segs, loff_t pos);
51 static int nfs_file_flush(struct file *, fl_owner_t id);
52 static int nfs_fsync(struct file *, struct dentry *dentry, int datasync);
53 static int nfs_check_flags(int flags);
54 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
55 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
57 const struct file_operations nfs_file_operations = {
58 .llseek = nfs_file_llseek,
60 .write = do_sync_write,
61 .aio_read = nfs_file_read,
62 .aio_write = nfs_file_write,
63 .mmap = nfs_file_mmap,
64 .open = nfs_file_open,
65 .flush = nfs_file_flush,
66 .release = nfs_file_release,
70 .splice_read = nfs_file_splice_read,
71 .check_flags = nfs_check_flags,
74 const struct inode_operations nfs_file_inode_operations = {
75 .permission = nfs_permission,
76 .getattr = nfs_getattr,
77 .setattr = nfs_setattr,
81 const struct inode_operations nfs3_file_inode_operations = {
82 .permission = nfs_permission,
83 .getattr = nfs_getattr,
84 .setattr = nfs_setattr,
85 .listxattr = nfs3_listxattr,
86 .getxattr = nfs3_getxattr,
87 .setxattr = nfs3_setxattr,
88 .removexattr = nfs3_removexattr,
90 #endif /* CONFIG_NFS_v3 */
92 /* Hack for future NFS swap support */
94 # define IS_SWAPFILE(inode) (0)
97 static int nfs_check_flags(int flags)
99 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
109 nfs_file_open(struct inode *inode, struct file *filp)
113 res = nfs_check_flags(filp->f_flags);
117 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
119 res = NFS_PROTO(inode)->file_open(inode, filp);
125 nfs_file_release(struct inode *inode, struct file *filp)
127 /* Ensure that dirty pages are flushed out with the right creds */
128 if (filp->f_mode & FMODE_WRITE)
129 filemap_fdatawrite(filp->f_mapping);
130 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
131 return NFS_PROTO(inode)->file_release(inode, filp);
135 * nfs_revalidate_size - Revalidate the file size
136 * @inode - pointer to inode struct
137 * @file - pointer to struct file
139 * Revalidates the file length. This is basically a wrapper around
140 * nfs_revalidate_inode() that takes into account the fact that we may
141 * have cached writes (in which case we don't care about the server's
142 * idea of what the file length is), or O_DIRECT (in which case we
143 * shouldn't trust the cache).
145 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
147 struct nfs_server *server = NFS_SERVER(inode);
148 struct nfs_inode *nfsi = NFS_I(inode);
150 if (server->flags & NFS_MOUNT_NOAC)
152 if (filp->f_flags & O_DIRECT)
154 if (nfsi->npages != 0)
156 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
159 return __nfs_revalidate_inode(server, inode);
162 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
164 /* origin == SEEK_END => we must revalidate the cached file length */
165 if (origin == SEEK_END) {
166 struct inode *inode = filp->f_mapping->host;
167 int retval = nfs_revalidate_file_size(inode, filp);
169 return (loff_t)retval;
171 return remote_llseek(filp, offset, origin);
175 * Flush all dirty pages, and check for write errors.
179 nfs_file_flush(struct file *file, fl_owner_t id)
181 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
182 struct inode *inode = file->f_path.dentry->d_inode;
185 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
187 if ((file->f_mode & FMODE_WRITE) == 0)
189 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
191 /* Ensure that data+attribute caches are up to date after close() */
192 status = nfs_wb_all(inode);
197 nfs_revalidate_inode(NFS_SERVER(inode), inode);
204 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
205 unsigned long nr_segs, loff_t pos)
207 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
208 struct inode * inode = dentry->d_inode;
210 size_t count = iov_length(iov, nr_segs);
212 #ifdef CONFIG_NFS_DIRECTIO
213 if (iocb->ki_filp->f_flags & O_DIRECT)
214 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
217 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
218 dentry->d_parent->d_name.name, dentry->d_name.name,
219 (unsigned long) count, (unsigned long) pos);
221 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
222 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
224 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
229 nfs_file_splice_read(struct file *filp, loff_t *ppos,
230 struct pipe_inode_info *pipe, size_t count,
233 struct dentry *dentry = filp->f_path.dentry;
234 struct inode *inode = dentry->d_inode;
237 dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
238 dentry->d_parent->d_name.name, dentry->d_name.name,
239 (unsigned long) count, (unsigned long long) *ppos);
241 res = nfs_revalidate_mapping(inode, filp->f_mapping);
243 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
248 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
250 struct dentry *dentry = file->f_path.dentry;
251 struct inode *inode = dentry->d_inode;
254 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
255 dentry->d_parent->d_name.name, dentry->d_name.name);
257 status = nfs_revalidate_mapping(inode, file->f_mapping);
259 status = generic_file_mmap(file, vma);
264 * Flush any dirty pages for this process, and check for write errors.
265 * The return status from this call provides a reliable indication of
266 * whether any write errors occurred for this process.
269 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
271 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
272 struct inode *inode = dentry->d_inode;
275 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
277 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
279 status = nfs_wb_all(inode);
289 * This does the "real" work of the write. The generic routine has
290 * allocated the page, locked it, done all the page alignment stuff
291 * calculations etc. Now we should just copy the data from user
292 * space and write it back to the real medium..
294 * If the writer ends up delaying the write, the writer needs to
295 * increment the page use counts until he is done with the page.
297 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
299 return nfs_flush_incompatible(file, page);
302 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
307 status = nfs_updatepage(file, page, offset, to-offset);
312 static void nfs_invalidate_page(struct page *page, unsigned long offset)
316 /* Cancel any unstarted writes on this page */
317 nfs_wb_page_priority(page->mapping->host, page, FLUSH_INVALIDATE);
320 static int nfs_release_page(struct page *page, gfp_t gfp)
322 /* If PagePrivate() is set, then the page is not freeable */
326 static int nfs_launder_page(struct page *page)
328 return nfs_wb_page(page->mapping->host, page);
331 const struct address_space_operations nfs_file_aops = {
332 .readpage = nfs_readpage,
333 .readpages = nfs_readpages,
334 .set_page_dirty = nfs_set_page_dirty,
335 .writepage = nfs_writepage,
336 .writepages = nfs_writepages,
337 .prepare_write = nfs_prepare_write,
338 .commit_write = nfs_commit_write,
339 .invalidatepage = nfs_invalidate_page,
340 .releasepage = nfs_release_page,
341 #ifdef CONFIG_NFS_DIRECTIO
342 .direct_IO = nfs_direct_IO,
344 .launder_page = nfs_launder_page,
347 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
348 unsigned long nr_segs, loff_t pos)
350 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
351 struct inode * inode = dentry->d_inode;
353 size_t count = iov_length(iov, nr_segs);
355 #ifdef CONFIG_NFS_DIRECTIO
356 if (iocb->ki_filp->f_flags & O_DIRECT)
357 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
360 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
361 dentry->d_parent->d_name.name, dentry->d_name.name,
362 inode->i_ino, (unsigned long) count, (long long) pos);
365 if (IS_SWAPFILE(inode))
368 * O_APPEND implies that we must revalidate the file length.
370 if (iocb->ki_filp->f_flags & O_APPEND) {
371 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
380 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
381 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
382 /* Return error values for O_SYNC and IS_SYNC() */
383 if (result >= 0 && (IS_SYNC(inode) || (iocb->ki_filp->f_flags & O_SYNC))) {
384 int err = nfs_fsync(iocb->ki_filp, dentry, 1);
392 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
396 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
398 struct inode *inode = filp->f_mapping->host;
402 /* Try local locking first */
403 if (posix_test_lock(filp, fl)) {
407 if (nfs_have_delegation(inode, FMODE_READ))
410 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
413 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
418 fl->fl_type = F_UNLCK;
422 static int do_vfs_lock(struct file *file, struct file_lock *fl)
425 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
427 res = posix_lock_file_wait(file, fl);
430 res = flock_lock_file_wait(file, fl);
436 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
442 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
444 struct inode *inode = filp->f_mapping->host;
448 * Flush all pending writes before doing anything
451 nfs_sync_mapping(filp->f_mapping);
453 /* NOTE: special case
454 * If we're signalled while cleaning up locks on process exit, we
455 * still need to complete the unlock.
458 /* Use local locking if mounted with "-onolock" */
459 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
460 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
462 status = do_vfs_lock(filp, fl);
467 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
469 struct inode *inode = filp->f_mapping->host;
473 * Flush all pending writes before doing anything
476 status = nfs_sync_mapping(filp->f_mapping);
481 /* Use local locking if mounted with "-onolock" */
482 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
483 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
484 /* If we were signalled we still need to ensure that
485 * we clean up any state on the server. We therefore
486 * record the lock call as having succeeded in order to
487 * ensure that locks_remove_posix() cleans it out when
490 if (status == -EINTR || status == -ERESTARTSYS)
491 do_vfs_lock(filp, fl);
493 status = do_vfs_lock(filp, fl);
498 * Make sure we clear the cache whenever we try to get the lock.
499 * This makes locking act as a cache coherency point.
501 nfs_sync_mapping(filp->f_mapping);
502 nfs_zap_caches(inode);
508 * Lock a (portion of) a file
510 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
512 struct inode * inode = filp->f_mapping->host;
514 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
515 inode->i_sb->s_id, inode->i_ino,
516 fl->fl_type, fl->fl_flags,
517 (long long)fl->fl_start, (long long)fl->fl_end);
518 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
520 /* No mandatory locks over NFS */
521 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
522 fl->fl_type != F_UNLCK)
526 return do_getlk(filp, cmd, fl);
527 if (fl->fl_type == F_UNLCK)
528 return do_unlk(filp, cmd, fl);
529 return do_setlk(filp, cmd, fl);
533 * Lock a (portion of) a file
535 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
537 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
538 filp->f_path.dentry->d_inode->i_sb->s_id,
539 filp->f_path.dentry->d_inode->i_ino,
540 fl->fl_type, fl->fl_flags);
543 * No BSD flocks over NFS allowed.
544 * Note: we could try to fake a POSIX lock request here by
545 * using ((u32) filp | 0x80000000) or some such as the pid.
546 * Not sure whether that would be unique, though, or whether
547 * that would break in other places.
549 if (!(fl->fl_flags & FL_FLOCK))
552 /* We're simulating flock() locks using posix locks on the server */
553 fl->fl_owner = (fl_owner_t)filp;
555 fl->fl_end = OFFSET_MAX;
557 if (fl->fl_type == F_UNLCK)
558 return do_unlk(filp, cmd, fl);
559 return do_setlk(filp, cmd, fl);