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,
74 /* Hack for future NFS swap support */
76 # define IS_SWAPFILE(inode) (0)
79 static int nfs_check_flags(int flags)
81 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
91 nfs_file_open(struct inode *inode, struct file *filp)
93 struct nfs_server *server = NFS_SERVER(inode);
94 int (*open)(struct inode *, struct file *);
97 res = nfs_check_flags(filp->f_flags);
102 /* Do NFSv4 open() call */
103 if ((open = server->rpc_ops->file_open) != NULL)
104 res = open(inode, filp);
110 nfs_file_release(struct inode *inode, struct file *filp)
112 /* Ensure that dirty pages are flushed out with the right creds */
113 if (filp->f_mode & FMODE_WRITE)
114 filemap_fdatawrite(filp->f_mapping);
115 return NFS_PROTO(inode)->file_release(inode, filp);
119 * nfs_revalidate_size - Revalidate the file size
120 * @inode - pointer to inode struct
121 * @file - pointer to struct file
123 * Revalidates the file length. This is basically a wrapper around
124 * nfs_revalidate_inode() that takes into account the fact that we may
125 * have cached writes (in which case we don't care about the server's
126 * idea of what the file length is), or O_DIRECT (in which case we
127 * shouldn't trust the cache).
129 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
131 struct nfs_server *server = NFS_SERVER(inode);
132 struct nfs_inode *nfsi = NFS_I(inode);
134 if (server->flags & NFS_MOUNT_NOAC)
136 if (filp->f_flags & O_DIRECT)
138 if (nfsi->npages != 0)
140 return nfs_revalidate_inode(server, inode);
142 return __nfs_revalidate_inode(server, inode);
145 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
147 /* origin == SEEK_END => we must revalidate the cached file length */
149 struct inode *inode = filp->f_mapping->host;
150 int retval = nfs_revalidate_file_size(inode, filp);
152 return (loff_t)retval;
154 return remote_llseek(filp, offset, origin);
158 * Flush all dirty pages, and check for write errors.
162 nfs_file_flush(struct file *file)
164 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
165 struct inode *inode = file->f_dentry->d_inode;
168 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
170 if ((file->f_mode & FMODE_WRITE) == 0)
173 /* Ensure that data+attribute caches are up to date after close() */
174 status = nfs_wb_all(inode);
178 if (!status && !nfs_have_delegation(inode, FMODE_READ))
179 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
186 nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos)
188 struct dentry * dentry = iocb->ki_filp->f_dentry;
189 struct inode * inode = dentry->d_inode;
192 #ifdef CONFIG_NFS_DIRECTIO
193 if (iocb->ki_filp->f_flags & O_DIRECT)
194 return nfs_file_direct_read(iocb, buf, count, pos);
197 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
198 dentry->d_parent->d_name.name, dentry->d_name.name,
199 (unsigned long) count, (unsigned long) pos);
201 result = nfs_revalidate_inode(NFS_SERVER(inode), inode);
203 result = generic_file_aio_read(iocb, buf, count, pos);
208 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
209 read_actor_t actor, void *target)
211 struct dentry *dentry = filp->f_dentry;
212 struct inode *inode = dentry->d_inode;
215 dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
216 dentry->d_parent->d_name.name, dentry->d_name.name,
217 (unsigned long) count, (unsigned long long) *ppos);
219 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
221 res = generic_file_sendfile(filp, ppos, count, actor, target);
226 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
228 struct dentry *dentry = file->f_dentry;
229 struct inode *inode = dentry->d_inode;
232 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
233 dentry->d_parent->d_name.name, dentry->d_name.name);
235 status = nfs_revalidate_inode(NFS_SERVER(inode), inode);
237 status = generic_file_mmap(file, vma);
242 * Flush any dirty pages for this process, and check for write errors.
243 * The return status from this call provides a reliable indication of
244 * whether any write errors occurred for this process.
247 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
249 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
250 struct inode *inode = dentry->d_inode;
253 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
256 status = nfs_wb_all(inode);
266 * This does the "real" work of the write. The generic routine has
267 * allocated the page, locked it, done all the page alignment stuff
268 * calculations etc. Now we should just copy the data from user
269 * space and write it back to the real medium..
271 * If the writer ends up delaying the write, the writer needs to
272 * increment the page use counts until he is done with the page.
274 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
276 return nfs_flush_incompatible(file, page);
279 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
284 status = nfs_updatepage(file, page, offset, to-offset);
289 struct address_space_operations nfs_file_aops = {
290 .readpage = nfs_readpage,
291 .readpages = nfs_readpages,
292 .set_page_dirty = __set_page_dirty_nobuffers,
293 .writepage = nfs_writepage,
294 .writepages = nfs_writepages,
295 .prepare_write = nfs_prepare_write,
296 .commit_write = nfs_commit_write,
297 #ifdef CONFIG_NFS_DIRECTIO
298 .direct_IO = nfs_direct_IO,
303 * Write to a file (through the page cache).
306 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
308 struct dentry * dentry = iocb->ki_filp->f_dentry;
309 struct inode * inode = dentry->d_inode;
312 #ifdef CONFIG_NFS_DIRECTIO
313 if (iocb->ki_filp->f_flags & O_DIRECT)
314 return nfs_file_direct_write(iocb, buf, count, pos);
317 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
318 dentry->d_parent->d_name.name, dentry->d_name.name,
319 inode->i_ino, (unsigned long) count, (unsigned long) pos);
322 if (IS_SWAPFILE(inode))
324 result = nfs_revalidate_inode(NFS_SERVER(inode), inode);
332 result = generic_file_aio_write(iocb, buf, count, pos);
337 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
341 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
343 struct inode *inode = filp->f_mapping->host;
347 /* Use local locking if mounted with "-onolock" */
348 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
349 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
351 struct file_lock *cfl = posix_test_lock(filp, fl);
353 fl->fl_type = F_UNLCK;
355 memcpy(fl, cfl, sizeof(*fl));
361 static int do_vfs_lock(struct file *file, struct file_lock *fl)
364 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
366 res = posix_lock_file_wait(file, fl);
369 res = flock_lock_file_wait(file, fl);
375 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
380 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
382 struct inode *inode = filp->f_mapping->host;
386 rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
388 * Flush all pending writes before doing anything
391 filemap_fdatawrite(filp->f_mapping);
395 filemap_fdatawait(filp->f_mapping);
397 /* NOTE: special case
398 * If we're signalled while cleaning up locks on process exit, we
399 * still need to complete the unlock.
402 /* Use local locking if mounted with "-onolock" */
403 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
404 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
406 status = do_vfs_lock(filp, fl);
408 rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
412 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
414 struct inode *inode = filp->f_mapping->host;
418 rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
420 * Flush all pending writes before doing anything
423 status = filemap_fdatawrite(filp->f_mapping);
426 status = nfs_wb_all(inode);
429 status = filemap_fdatawait(filp->f_mapping);
435 /* Use local locking if mounted with "-onolock" */
436 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
437 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
438 /* If we were signalled we still need to ensure that
439 * we clean up any state on the server. We therefore
440 * record the lock call as having succeeded in order to
441 * ensure that locks_remove_posix() cleans it out when
444 if (status == -EINTR || status == -ERESTARTSYS)
445 do_vfs_lock(filp, fl);
447 status = do_vfs_lock(filp, fl);
452 * Make sure we clear the cache whenever we try to get the lock.
453 * This makes locking act as a cache coherency point.
455 filemap_fdatawrite(filp->f_mapping);
457 nfs_wb_all(inode); /* we may have slept */
459 filemap_fdatawait(filp->f_mapping);
460 nfs_zap_caches(inode);
462 rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
467 * Lock a (portion of) a file
469 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
471 struct inode * inode = filp->f_mapping->host;
473 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
474 inode->i_sb->s_id, inode->i_ino,
475 fl->fl_type, fl->fl_flags,
476 (long long)fl->fl_start, (long long)fl->fl_end);
481 /* No mandatory locks over NFS */
482 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
486 return do_getlk(filp, cmd, fl);
487 if (fl->fl_type == F_UNLCK)
488 return do_unlk(filp, cmd, fl);
489 return do_setlk(filp, cmd, fl);
493 * Lock a (portion of) a file
495 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
497 struct inode * inode = filp->f_mapping->host;
499 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
500 inode->i_sb->s_id, inode->i_ino,
501 fl->fl_type, fl->fl_flags);
507 * No BSD flocks over NFS allowed.
508 * Note: we could try to fake a POSIX lock request here by
509 * using ((u32) filp | 0x80000000) or some such as the pid.
510 * Not sure whether that would be unique, though, or whether
511 * that would break in other places.
513 if (!(fl->fl_flags & FL_FLOCK))
516 /* We're simulating flock() locks using posix locks on the server */
517 fl->fl_owner = (fl_owner_t)filp;
519 fl->fl_end = OFFSET_MAX;
521 if (fl->fl_type == F_UNLCK)
522 return do_unlk(filp, cmd, fl);
523 return do_setlk(filp, cmd, fl);