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)
137 struct nfs_inode *nfsi = NFS_I(inode);
140 if ((nfsi->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATTR))
141 || nfs_attribute_timeout(inode))
142 retval = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
143 nfs_revalidate_mapping(inode, filp->f_mapping);
148 * nfs_revalidate_size - Revalidate the file size
149 * @inode - pointer to inode struct
150 * @file - pointer to struct file
152 * Revalidates the file length. This is basically a wrapper around
153 * nfs_revalidate_inode() that takes into account the fact that we may
154 * have cached writes (in which case we don't care about the server's
155 * idea of what the file length is), or O_DIRECT (in which case we
156 * shouldn't trust the cache).
158 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
160 struct nfs_server *server = NFS_SERVER(inode);
161 struct nfs_inode *nfsi = NFS_I(inode);
163 if (server->flags & NFS_MOUNT_NOAC)
165 if (filp->f_flags & O_DIRECT)
167 if (nfsi->npages != 0)
169 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
172 return __nfs_revalidate_inode(server, inode);
175 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
177 /* origin == SEEK_END => we must revalidate the cached file length */
179 struct inode *inode = filp->f_mapping->host;
180 int retval = nfs_revalidate_file_size(inode, filp);
182 return (loff_t)retval;
184 return remote_llseek(filp, offset, origin);
188 * Flush all dirty pages, and check for write errors.
192 nfs_file_flush(struct file *file)
194 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
195 struct inode *inode = file->f_dentry->d_inode;
198 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
200 if ((file->f_mode & FMODE_WRITE) == 0)
203 /* Ensure that data+attribute caches are up to date after close() */
204 status = nfs_wb_all(inode);
209 nfs_revalidate_inode(NFS_SERVER(inode), inode);
216 nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos)
218 struct dentry * dentry = iocb->ki_filp->f_dentry;
219 struct inode * inode = dentry->d_inode;
222 #ifdef CONFIG_NFS_DIRECTIO
223 if (iocb->ki_filp->f_flags & O_DIRECT)
224 return nfs_file_direct_read(iocb, buf, count, pos);
227 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
228 dentry->d_parent->d_name.name, dentry->d_name.name,
229 (unsigned long) count, (unsigned long) pos);
231 result = nfs_revalidate_file(inode, iocb->ki_filp);
233 result = generic_file_aio_read(iocb, buf, count, pos);
238 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
239 read_actor_t actor, void *target)
241 struct dentry *dentry = filp->f_dentry;
242 struct inode *inode = dentry->d_inode;
245 dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
246 dentry->d_parent->d_name.name, dentry->d_name.name,
247 (unsigned long) count, (unsigned long long) *ppos);
249 res = nfs_revalidate_file(inode, filp);
251 res = generic_file_sendfile(filp, ppos, count, actor, target);
256 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
258 struct dentry *dentry = file->f_dentry;
259 struct inode *inode = dentry->d_inode;
262 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
263 dentry->d_parent->d_name.name, dentry->d_name.name);
265 status = nfs_revalidate_file(inode, file);
267 status = generic_file_mmap(file, vma);
272 * Flush any dirty pages for this process, and check for write errors.
273 * The return status from this call provides a reliable indication of
274 * whether any write errors occurred for this process.
277 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
279 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
280 struct inode *inode = dentry->d_inode;
283 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
286 status = nfs_wb_all(inode);
296 * This does the "real" work of the write. The generic routine has
297 * allocated the page, locked it, done all the page alignment stuff
298 * calculations etc. Now we should just copy the data from user
299 * space and write it back to the real medium..
301 * If the writer ends up delaying the write, the writer needs to
302 * increment the page use counts until he is done with the page.
304 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
306 return nfs_flush_incompatible(file, page);
309 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
314 status = nfs_updatepage(file, page, offset, to-offset);
319 struct address_space_operations nfs_file_aops = {
320 .readpage = nfs_readpage,
321 .readpages = nfs_readpages,
322 .set_page_dirty = __set_page_dirty_nobuffers,
323 .writepage = nfs_writepage,
324 .writepages = nfs_writepages,
325 .prepare_write = nfs_prepare_write,
326 .commit_write = nfs_commit_write,
327 #ifdef CONFIG_NFS_DIRECTIO
328 .direct_IO = nfs_direct_IO,
333 * Write to a file (through the page cache).
336 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
338 struct dentry * dentry = iocb->ki_filp->f_dentry;
339 struct inode * inode = dentry->d_inode;
342 #ifdef CONFIG_NFS_DIRECTIO
343 if (iocb->ki_filp->f_flags & O_DIRECT)
344 return nfs_file_direct_write(iocb, buf, count, pos);
347 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
348 dentry->d_parent->d_name.name, dentry->d_name.name,
349 inode->i_ino, (unsigned long) count, (unsigned long) pos);
352 if (IS_SWAPFILE(inode))
355 * O_APPEND implies that we must revalidate the file length.
357 if (iocb->ki_filp->f_flags & O_APPEND) {
358 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
362 nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
368 result = generic_file_aio_write(iocb, buf, count, pos);
373 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
377 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
379 struct file_lock *cfl;
380 struct inode *inode = filp->f_mapping->host;
384 /* Try local locking first */
385 cfl = posix_test_lock(filp, fl);
387 locks_copy_lock(fl, cfl);
391 if (nfs_have_delegation(inode, FMODE_READ))
394 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
397 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
402 fl->fl_type = F_UNLCK;
406 static int do_vfs_lock(struct file *file, struct file_lock *fl)
409 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
411 res = posix_lock_file_wait(file, fl);
414 res = flock_lock_file_wait(file, fl);
420 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
425 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
427 struct inode *inode = filp->f_mapping->host;
431 rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
433 * Flush all pending writes before doing anything
436 filemap_fdatawrite(filp->f_mapping);
440 filemap_fdatawait(filp->f_mapping);
442 /* NOTE: special case
443 * If we're signalled while cleaning up locks on process exit, we
444 * still need to complete the unlock.
447 /* Use local locking if mounted with "-onolock" */
448 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
449 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
451 status = do_vfs_lock(filp, fl);
453 rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
457 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
459 struct inode *inode = filp->f_mapping->host;
463 rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
465 * Flush all pending writes before doing anything
468 status = filemap_fdatawrite(filp->f_mapping);
471 status = nfs_wb_all(inode);
474 status = filemap_fdatawait(filp->f_mapping);
480 /* Use local locking if mounted with "-onolock" */
481 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
482 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
483 /* If we were signalled we still need to ensure that
484 * we clean up any state on the server. We therefore
485 * record the lock call as having succeeded in order to
486 * ensure that locks_remove_posix() cleans it out when
489 if (status == -EINTR || status == -ERESTARTSYS)
490 do_vfs_lock(filp, fl);
492 status = do_vfs_lock(filp, fl);
497 * Make sure we clear the cache whenever we try to get the lock.
498 * This makes locking act as a cache coherency point.
500 filemap_fdatawrite(filp->f_mapping);
502 nfs_wb_all(inode); /* we may have slept */
504 filemap_fdatawait(filp->f_mapping);
505 nfs_zap_caches(inode);
507 rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
512 * Lock a (portion of) a file
514 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
516 struct inode * inode = filp->f_mapping->host;
518 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
519 inode->i_sb->s_id, inode->i_ino,
520 fl->fl_type, fl->fl_flags,
521 (long long)fl->fl_start, (long long)fl->fl_end);
526 /* No mandatory locks over NFS */
527 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
531 return do_getlk(filp, cmd, fl);
532 if (fl->fl_type == F_UNLCK)
533 return do_unlk(filp, cmd, fl);
534 return do_setlk(filp, cmd, fl);
538 * Lock a (portion of) a file
540 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
542 struct inode * inode = filp->f_mapping->host;
544 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
545 inode->i_sb->s_id, inode->i_ino,
546 fl->fl_type, fl->fl_flags);
552 * No BSD flocks over NFS allowed.
553 * Note: we could try to fake a POSIX lock request here by
554 * using ((u32) filp | 0x80000000) or some such as the pid.
555 * Not sure whether that would be unique, though, or whether
556 * that would break in other places.
558 if (!(fl->fl_flags & FL_FLOCK))
561 /* We're simulating flock() locks using posix locks on the server */
562 fl->fl_owner = (fl_owner_t)filp;
564 fl->fl_end = OFFSET_MAX;
566 if (fl->fl_type == F_UNLCK)
567 return do_unlk(filp, cmd, fl);
568 return do_setlk(filp, cmd, fl);