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"
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_splice_read(struct file *filp, loff_t *ppos,
46 struct pipe_inode_info *pipe,
47 size_t count, unsigned int flags);
48 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
49 unsigned long nr_segs, loff_t pos);
50 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
51 unsigned long nr_segs, loff_t pos);
52 static int nfs_file_flush(struct file *, fl_owner_t id);
53 static int nfs_fsync(struct file *, struct dentry *dentry, int datasync);
54 static int nfs_check_flags(int flags);
55 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
56 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
57 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
59 static struct vm_operations_struct nfs_file_vm_ops;
61 const struct file_operations nfs_file_operations = {
62 .llseek = nfs_file_llseek,
64 .write = do_sync_write,
65 .aio_read = nfs_file_read,
66 .aio_write = nfs_file_write,
67 .mmap = nfs_file_mmap,
68 .open = nfs_file_open,
69 .flush = nfs_file_flush,
70 .release = nfs_file_release,
74 .splice_read = nfs_file_splice_read,
75 .check_flags = nfs_check_flags,
76 .setlease = nfs_setlease,
79 const struct inode_operations nfs_file_inode_operations = {
80 .permission = nfs_permission,
81 .getattr = nfs_getattr,
82 .setattr = nfs_setattr,
86 const struct inode_operations nfs3_file_inode_operations = {
87 .permission = nfs_permission,
88 .getattr = nfs_getattr,
89 .setattr = nfs_setattr,
90 .listxattr = nfs3_listxattr,
91 .getxattr = nfs3_getxattr,
92 .setxattr = nfs3_setxattr,
93 .removexattr = nfs3_removexattr,
95 #endif /* CONFIG_NFS_v3 */
97 /* Hack for future NFS swap support */
99 # define IS_SWAPFILE(inode) (0)
102 static int nfs_check_flags(int flags)
104 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
114 nfs_file_open(struct inode *inode, struct file *filp)
118 res = nfs_check_flags(filp->f_flags);
122 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
124 res = NFS_PROTO(inode)->file_open(inode, filp);
130 nfs_file_release(struct inode *inode, struct file *filp)
132 /* Ensure that dirty pages are flushed out with the right creds */
133 if (filp->f_mode & FMODE_WRITE)
134 nfs_wb_all(filp->f_path.dentry->d_inode);
135 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
136 return NFS_PROTO(inode)->file_release(inode, filp);
140 * nfs_revalidate_size - Revalidate the file size
141 * @inode - pointer to inode struct
142 * @file - pointer to struct file
144 * Revalidates the file length. This is basically a wrapper around
145 * nfs_revalidate_inode() that takes into account the fact that we may
146 * have cached writes (in which case we don't care about the server's
147 * idea of what the file length is), or O_DIRECT (in which case we
148 * shouldn't trust the cache).
150 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
152 struct nfs_server *server = NFS_SERVER(inode);
153 struct nfs_inode *nfsi = NFS_I(inode);
155 if (server->flags & NFS_MOUNT_NOAC)
157 if (filp->f_flags & O_DIRECT)
159 if (nfsi->npages != 0)
161 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
164 return __nfs_revalidate_inode(server, inode);
167 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
169 /* origin == SEEK_END => we must revalidate the cached file length */
170 if (origin == SEEK_END) {
171 struct inode *inode = filp->f_mapping->host;
172 int retval = nfs_revalidate_file_size(inode, filp);
174 return (loff_t)retval;
176 return remote_llseek(filp, offset, origin);
180 * Helper for nfs_file_flush() and nfs_fsync()
182 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
183 * disk, but it retrieves and clears ctx->error after synching, despite
184 * the two being set at the same time in nfs_context_set_write_error().
185 * This is because the former is used to notify the _next_ call to
186 * nfs_file_write() that a write error occured, and hence cause it to
187 * fall back to doing a synchronous write.
189 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
191 int have_error, status;
194 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
195 status = nfs_wb_all(inode);
196 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
198 ret = xchg(&ctx->error, 0);
205 * Flush all dirty pages, and check for write errors.
209 nfs_file_flush(struct file *file, fl_owner_t id)
211 struct nfs_open_context *ctx = nfs_file_open_context(file);
212 struct inode *inode = file->f_path.dentry->d_inode;
215 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
217 if ((file->f_mode & FMODE_WRITE) == 0)
219 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
221 /* Ensure that data+attribute caches are up to date after close() */
222 status = nfs_do_fsync(ctx, inode);
224 nfs_revalidate_inode(NFS_SERVER(inode), inode);
229 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
230 unsigned long nr_segs, loff_t pos)
232 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
233 struct inode * inode = dentry->d_inode;
235 size_t count = iov_length(iov, nr_segs);
237 if (iocb->ki_filp->f_flags & O_DIRECT)
238 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
240 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
241 dentry->d_parent->d_name.name, dentry->d_name.name,
242 (unsigned long) count, (unsigned long) pos);
244 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
245 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
247 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
252 nfs_file_splice_read(struct file *filp, loff_t *ppos,
253 struct pipe_inode_info *pipe, size_t count,
256 struct dentry *dentry = filp->f_path.dentry;
257 struct inode *inode = dentry->d_inode;
260 dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
261 dentry->d_parent->d_name.name, dentry->d_name.name,
262 (unsigned long) count, (unsigned long long) *ppos);
264 res = nfs_revalidate_mapping(inode, filp->f_mapping);
266 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
271 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
273 struct dentry *dentry = file->f_path.dentry;
274 struct inode *inode = dentry->d_inode;
277 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
278 dentry->d_parent->d_name.name, dentry->d_name.name);
280 status = nfs_revalidate_mapping(inode, file->f_mapping);
282 vma->vm_ops = &nfs_file_vm_ops;
283 vma->vm_flags |= VM_CAN_NONLINEAR;
290 * Flush any dirty pages for this process, and check for write errors.
291 * The return status from this call provides a reliable indication of
292 * whether any write errors occurred for this process.
295 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
297 struct nfs_open_context *ctx = nfs_file_open_context(file);
298 struct inode *inode = dentry->d_inode;
300 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
302 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
303 return nfs_do_fsync(ctx, inode);
307 * This does the "real" work of the write. We must allocate and lock the
308 * page to be sent back to the generic routine, which then copies the
309 * data from user space.
311 * If the writer ends up delaying the write, the writer needs to
312 * increment the page use counts until he is done with the page.
314 static int nfs_write_begin(struct file *file, struct address_space *mapping,
315 loff_t pos, unsigned len, unsigned flags,
316 struct page **pagep, void **fsdata)
321 index = pos >> PAGE_CACHE_SHIFT;
323 page = __grab_cache_page(mapping, index);
328 ret = nfs_flush_incompatible(file, page);
331 page_cache_release(page);
336 static int nfs_write_end(struct file *file, struct address_space *mapping,
337 loff_t pos, unsigned len, unsigned copied,
338 struct page *page, void *fsdata)
340 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
344 status = nfs_updatepage(file, page, offset, copied);
348 page_cache_release(page);
355 static void nfs_invalidate_page(struct page *page, unsigned long offset)
359 /* Cancel any unstarted writes on this page */
360 nfs_wb_page_cancel(page->mapping->host, page);
363 static int nfs_release_page(struct page *page, gfp_t gfp)
365 /* If PagePrivate() is set, then the page is not freeable */
369 static int nfs_launder_page(struct page *page)
371 return nfs_wb_page(page->mapping->host, page);
374 const struct address_space_operations nfs_file_aops = {
375 .readpage = nfs_readpage,
376 .readpages = nfs_readpages,
377 .set_page_dirty = __set_page_dirty_nobuffers,
378 .writepage = nfs_writepage,
379 .writepages = nfs_writepages,
380 .write_begin = nfs_write_begin,
381 .write_end = nfs_write_end,
382 .invalidatepage = nfs_invalidate_page,
383 .releasepage = nfs_release_page,
384 .direct_IO = nfs_direct_IO,
385 .launder_page = nfs_launder_page,
388 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
390 struct file *filp = vma->vm_file;
393 struct address_space *mapping;
396 mapping = page->mapping;
397 if (mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping)
401 pagelen = nfs_page_length(page);
405 ret = nfs_flush_incompatible(filp, page);
409 ret = nfs_updatepage(filp, page, 0, pagelen);
417 static struct vm_operations_struct nfs_file_vm_ops = {
418 .fault = filemap_fault,
419 .page_mkwrite = nfs_vm_page_mkwrite,
422 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
424 struct nfs_open_context *ctx;
426 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
428 ctx = nfs_file_open_context(filp);
429 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
434 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
435 unsigned long nr_segs, loff_t pos)
437 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
438 struct inode * inode = dentry->d_inode;
440 size_t count = iov_length(iov, nr_segs);
442 if (iocb->ki_filp->f_flags & O_DIRECT)
443 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
445 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
446 dentry->d_parent->d_name.name, dentry->d_name.name,
447 inode->i_ino, (unsigned long) count, (long long) pos);
450 if (IS_SWAPFILE(inode))
453 * O_APPEND implies that we must revalidate the file length.
455 if (iocb->ki_filp->f_flags & O_APPEND) {
456 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
465 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
466 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
467 /* Return error values for O_SYNC and IS_SYNC() */
468 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
469 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
477 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
481 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
483 struct inode *inode = filp->f_mapping->host;
487 /* Try local locking first */
488 posix_test_lock(filp, fl);
489 if (fl->fl_type != F_UNLCK) {
490 /* found a conflict */
494 if (nfs_have_delegation(inode, FMODE_READ))
497 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
500 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
505 fl->fl_type = F_UNLCK;
509 static int do_vfs_lock(struct file *file, struct file_lock *fl)
512 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
514 res = posix_lock_file_wait(file, fl);
517 res = flock_lock_file_wait(file, fl);
523 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
529 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
531 struct inode *inode = filp->f_mapping->host;
535 * Flush all pending writes before doing anything
538 nfs_sync_mapping(filp->f_mapping);
540 /* NOTE: special case
541 * If we're signalled while cleaning up locks on process exit, we
542 * still need to complete the unlock.
545 /* Use local locking if mounted with "-onolock" */
546 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
547 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
549 status = do_vfs_lock(filp, fl);
554 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
556 struct inode *inode = filp->f_mapping->host;
560 * Flush all pending writes before doing anything
563 status = nfs_sync_mapping(filp->f_mapping);
568 /* Use local locking if mounted with "-onolock" */
569 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
570 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
571 /* If we were signalled we still need to ensure that
572 * we clean up any state on the server. We therefore
573 * record the lock call as having succeeded in order to
574 * ensure that locks_remove_posix() cleans it out when
577 if (status == -EINTR || status == -ERESTARTSYS)
578 do_vfs_lock(filp, fl);
580 status = do_vfs_lock(filp, fl);
585 * Make sure we clear the cache whenever we try to get the lock.
586 * This makes locking act as a cache coherency point.
588 nfs_sync_mapping(filp->f_mapping);
589 nfs_zap_caches(inode);
595 * Lock a (portion of) a file
597 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
599 struct inode * inode = filp->f_mapping->host;
601 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
602 inode->i_sb->s_id, inode->i_ino,
603 fl->fl_type, fl->fl_flags,
604 (long long)fl->fl_start, (long long)fl->fl_end);
605 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
607 /* No mandatory locks over NFS */
608 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
612 return do_getlk(filp, cmd, fl);
613 if (fl->fl_type == F_UNLCK)
614 return do_unlk(filp, cmd, fl);
615 return do_setlk(filp, cmd, fl);
619 * Lock a (portion of) a file
621 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
623 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
624 filp->f_path.dentry->d_inode->i_sb->s_id,
625 filp->f_path.dentry->d_inode->i_ino,
626 fl->fl_type, fl->fl_flags);
629 * No BSD flocks over NFS allowed.
630 * Note: we could try to fake a POSIX lock request here by
631 * using ((u32) filp | 0x80000000) or some such as the pid.
632 * Not sure whether that would be unique, though, or whether
633 * that would break in other places.
635 if (!(fl->fl_flags & FL_FLOCK))
638 /* We're simulating flock() locks using posix locks on the server */
639 fl->fl_owner = (fl_owner_t)filp;
641 fl->fl_end = OFFSET_MAX;
643 if (fl->fl_type == F_UNLCK)
644 return do_unlk(filp, cmd, fl);
645 return do_setlk(filp, cmd, fl);
648 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
651 * There is no protocol support for leases, so we have no way
652 * to implement them correctly in the face of opens by other