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"
36 #define NFSDBG_FACILITY NFSDBG_FILE
38 static int nfs_file_open(struct inode *, struct file *);
39 static int nfs_file_release(struct inode *, struct file *);
40 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
41 static int nfs_file_mmap(struct file *, struct vm_area_struct *);
42 static ssize_t nfs_file_sendfile(struct file *, loff_t *, size_t, read_actor_t, void *);
43 static ssize_t nfs_file_read(struct kiocb *, char __user *, size_t, loff_t);
44 static ssize_t nfs_file_write(struct kiocb *, const char __user *, size_t, loff_t);
45 static int nfs_file_flush(struct file *);
46 static int nfs_fsync(struct file *, struct dentry *dentry, int datasync);
47 static int nfs_check_flags(int flags);
48 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
49 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
51 struct file_operations nfs_file_operations = {
52 .llseek = nfs_file_llseek,
54 .write = do_sync_write,
55 .aio_read = nfs_file_read,
56 .aio_write = nfs_file_write,
57 .mmap = nfs_file_mmap,
58 .open = nfs_file_open,
59 .flush = nfs_file_flush,
60 .release = nfs_file_release,
64 .sendfile = nfs_file_sendfile,
65 .check_flags = nfs_check_flags,
68 struct inode_operations nfs_file_inode_operations = {
69 .permission = nfs_permission,
70 .getattr = nfs_getattr,
71 .setattr = nfs_setattr,
75 struct inode_operations nfs3_file_inode_operations = {
76 .permission = nfs_permission,
77 .getattr = nfs_getattr,
78 .setattr = nfs_setattr,
79 .listxattr = nfs3_listxattr,
80 .getxattr = nfs3_getxattr,
81 .setxattr = nfs3_setxattr,
82 .removexattr = nfs3_removexattr,
84 #endif /* CONFIG_NFS_v3 */
86 /* Hack for future NFS swap support */
88 # define IS_SWAPFILE(inode) (0)
91 static int nfs_check_flags(int flags)
93 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
103 nfs_file_open(struct inode *inode, struct file *filp)
105 struct nfs_server *server = NFS_SERVER(inode);
106 int (*open)(struct inode *, struct file *);
109 res = nfs_check_flags(filp->f_flags);
114 /* Do NFSv4 open() call */
115 if ((open = server->rpc_ops->file_open) != NULL)
116 res = 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 return NFS_PROTO(inode)->file_release(inode, filp);
131 * nfs_revalidate_file - Revalidate the page cache & related metadata
132 * @inode - pointer to inode struct
133 * @file - pointer to file
135 static int nfs_revalidate_file(struct inode *inode, struct file *filp)
139 if ((NFS_FLAGS(inode) & NFS_INO_REVAL_PAGECACHE) || nfs_attribute_timeout(inode))
140 retval = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
141 nfs_revalidate_mapping(inode, filp->f_mapping);
146 * nfs_revalidate_size - Revalidate the file size
147 * @inode - pointer to inode struct
148 * @file - pointer to struct file
150 * Revalidates the file length. This is basically a wrapper around
151 * nfs_revalidate_inode() that takes into account the fact that we may
152 * have cached writes (in which case we don't care about the server's
153 * idea of what the file length is), or O_DIRECT (in which case we
154 * shouldn't trust the cache).
156 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
158 struct nfs_server *server = NFS_SERVER(inode);
159 struct nfs_inode *nfsi = NFS_I(inode);
161 if (server->flags & NFS_MOUNT_NOAC)
163 if (filp->f_flags & O_DIRECT)
165 if (nfsi->npages != 0)
167 if (!(NFS_FLAGS(inode) & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
170 return __nfs_revalidate_inode(server, inode);
173 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
175 /* origin == SEEK_END => we must revalidate the cached file length */
177 struct inode *inode = filp->f_mapping->host;
178 int retval = nfs_revalidate_file_size(inode, filp);
180 return (loff_t)retval;
182 return remote_llseek(filp, offset, origin);
186 * Flush all dirty pages, and check for write errors.
190 nfs_file_flush(struct file *file)
192 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
193 struct inode *inode = file->f_dentry->d_inode;
196 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
198 if ((file->f_mode & FMODE_WRITE) == 0)
201 /* Ensure that data+attribute caches are up to date after close() */
202 status = nfs_wb_all(inode);
206 if (!status && !nfs_have_delegation(inode, FMODE_READ))
207 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
214 nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos)
216 struct dentry * dentry = iocb->ki_filp->f_dentry;
217 struct inode * inode = dentry->d_inode;
220 #ifdef CONFIG_NFS_DIRECTIO
221 if (iocb->ki_filp->f_flags & O_DIRECT)
222 return nfs_file_direct_read(iocb, buf, count, pos);
225 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
226 dentry->d_parent->d_name.name, dentry->d_name.name,
227 (unsigned long) count, (unsigned long) pos);
229 result = nfs_revalidate_file(inode, iocb->ki_filp);
231 result = generic_file_aio_read(iocb, buf, count, pos);
236 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
237 read_actor_t actor, void *target)
239 struct dentry *dentry = filp->f_dentry;
240 struct inode *inode = dentry->d_inode;
243 dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
244 dentry->d_parent->d_name.name, dentry->d_name.name,
245 (unsigned long) count, (unsigned long long) *ppos);
247 res = nfs_revalidate_file(inode, filp);
249 res = generic_file_sendfile(filp, ppos, count, actor, target);
254 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
256 struct dentry *dentry = file->f_dentry;
257 struct inode *inode = dentry->d_inode;
260 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
261 dentry->d_parent->d_name.name, dentry->d_name.name);
263 status = nfs_revalidate_file(inode, file);
265 status = generic_file_mmap(file, vma);
270 * Flush any dirty pages for this process, and check for write errors.
271 * The return status from this call provides a reliable indication of
272 * whether any write errors occurred for this process.
275 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
277 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
278 struct inode *inode = dentry->d_inode;
281 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
284 status = nfs_wb_all(inode);
294 * This does the "real" work of the write. The generic routine has
295 * allocated the page, locked it, done all the page alignment stuff
296 * calculations etc. Now we should just copy the data from user
297 * space and write it back to the real medium..
299 * If the writer ends up delaying the write, the writer needs to
300 * increment the page use counts until he is done with the page.
302 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
304 return nfs_flush_incompatible(file, page);
307 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
312 status = nfs_updatepage(file, page, offset, to-offset);
317 struct address_space_operations nfs_file_aops = {
318 .readpage = nfs_readpage,
319 .readpages = nfs_readpages,
320 .set_page_dirty = __set_page_dirty_nobuffers,
321 .writepage = nfs_writepage,
322 .writepages = nfs_writepages,
323 .prepare_write = nfs_prepare_write,
324 .commit_write = nfs_commit_write,
325 #ifdef CONFIG_NFS_DIRECTIO
326 .direct_IO = nfs_direct_IO,
331 * Write to a file (through the page cache).
334 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
336 struct dentry * dentry = iocb->ki_filp->f_dentry;
337 struct inode * inode = dentry->d_inode;
340 #ifdef CONFIG_NFS_DIRECTIO
341 if (iocb->ki_filp->f_flags & O_DIRECT)
342 return nfs_file_direct_write(iocb, buf, count, pos);
345 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
346 dentry->d_parent->d_name.name, dentry->d_name.name,
347 inode->i_ino, (unsigned long) count, (unsigned long) pos);
350 if (IS_SWAPFILE(inode))
353 * O_APPEND implies that we must revalidate the file length.
355 if (iocb->ki_filp->f_flags & O_APPEND) {
356 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
360 nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
366 result = generic_file_aio_write(iocb, buf, count, pos);
371 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
375 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
377 struct inode *inode = filp->f_mapping->host;
381 /* Use local locking if mounted with "-onolock" */
382 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
383 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
385 struct file_lock *cfl = posix_test_lock(filp, fl);
387 fl->fl_type = F_UNLCK;
389 memcpy(fl, cfl, sizeof(*fl));
395 static int do_vfs_lock(struct file *file, struct file_lock *fl)
398 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
400 res = posix_lock_file_wait(file, fl);
403 res = flock_lock_file_wait(file, fl);
409 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
414 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
416 struct inode *inode = filp->f_mapping->host;
420 rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
422 * Flush all pending writes before doing anything
425 filemap_fdatawrite(filp->f_mapping);
429 filemap_fdatawait(filp->f_mapping);
431 /* NOTE: special case
432 * If we're signalled while cleaning up locks on process exit, we
433 * still need to complete the unlock.
436 /* Use local locking if mounted with "-onolock" */
437 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
438 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
440 status = do_vfs_lock(filp, fl);
442 rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
446 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
448 struct inode *inode = filp->f_mapping->host;
452 rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
454 * Flush all pending writes before doing anything
457 status = filemap_fdatawrite(filp->f_mapping);
460 status = nfs_wb_all(inode);
463 status = filemap_fdatawait(filp->f_mapping);
469 /* Use local locking if mounted with "-onolock" */
470 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
471 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
472 /* If we were signalled we still need to ensure that
473 * we clean up any state on the server. We therefore
474 * record the lock call as having succeeded in order to
475 * ensure that locks_remove_posix() cleans it out when
478 if (status == -EINTR || status == -ERESTARTSYS)
479 do_vfs_lock(filp, fl);
481 status = do_vfs_lock(filp, fl);
486 * Make sure we clear the cache whenever we try to get the lock.
487 * This makes locking act as a cache coherency point.
489 filemap_fdatawrite(filp->f_mapping);
491 nfs_wb_all(inode); /* we may have slept */
493 filemap_fdatawait(filp->f_mapping);
494 nfs_zap_caches(inode);
496 rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
501 * Lock a (portion of) a file
503 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
505 struct inode * inode = filp->f_mapping->host;
507 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
508 inode->i_sb->s_id, inode->i_ino,
509 fl->fl_type, fl->fl_flags,
510 (long long)fl->fl_start, (long long)fl->fl_end);
515 /* No mandatory locks over NFS */
516 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
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 struct inode * inode = filp->f_mapping->host;
533 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
534 inode->i_sb->s_id, inode->i_ino,
535 fl->fl_type, fl->fl_flags);
541 * No BSD flocks over NFS allowed.
542 * Note: we could try to fake a POSIX lock request here by
543 * using ((u32) filp | 0x80000000) or some such as the pid.
544 * Not sure whether that would be unique, though, or whether
545 * that would break in other places.
547 if (!(fl->fl_flags & FL_FLOCK))
550 /* We're simulating flock() locks using posix locks on the server */
551 fl->fl_owner = (fl_owner_t)filp;
553 fl->fl_end = OFFSET_MAX;
555 if (fl->fl_type == F_UNLCK)
556 return do_unlk(filp, cmd, fl);
557 return do_setlk(filp, cmd, fl);