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 #ifdef CONFIG_NFS_DIRECTIO
238 if (iocb->ki_filp->f_flags & O_DIRECT)
239 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
242 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
243 dentry->d_parent->d_name.name, dentry->d_name.name,
244 (unsigned long) count, (unsigned long) pos);
246 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
247 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
249 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
254 nfs_file_splice_read(struct file *filp, loff_t *ppos,
255 struct pipe_inode_info *pipe, size_t count,
258 struct dentry *dentry = filp->f_path.dentry;
259 struct inode *inode = dentry->d_inode;
262 dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
263 dentry->d_parent->d_name.name, dentry->d_name.name,
264 (unsigned long) count, (unsigned long long) *ppos);
266 res = nfs_revalidate_mapping(inode, filp->f_mapping);
268 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
273 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
275 struct dentry *dentry = file->f_path.dentry;
276 struct inode *inode = dentry->d_inode;
279 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
280 dentry->d_parent->d_name.name, dentry->d_name.name);
282 status = nfs_revalidate_mapping(inode, file->f_mapping);
284 vma->vm_ops = &nfs_file_vm_ops;
285 vma->vm_flags |= VM_CAN_NONLINEAR;
292 * Flush any dirty pages for this process, and check for write errors.
293 * The return status from this call provides a reliable indication of
294 * whether any write errors occurred for this process.
297 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
299 struct nfs_open_context *ctx = nfs_file_open_context(file);
300 struct inode *inode = dentry->d_inode;
302 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
304 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
305 return nfs_do_fsync(ctx, inode);
309 * This does the "real" work of the write. We must allocate and lock the
310 * page to be sent back to the generic routine, which then copies the
311 * data from user space.
313 * If the writer ends up delaying the write, the writer needs to
314 * increment the page use counts until he is done with the page.
316 static int nfs_write_begin(struct file *file, struct address_space *mapping,
317 loff_t pos, unsigned len, unsigned flags,
318 struct page **pagep, void **fsdata)
323 index = pos >> PAGE_CACHE_SHIFT;
325 page = __grab_cache_page(mapping, index);
330 ret = nfs_flush_incompatible(file, page);
333 page_cache_release(page);
338 static int nfs_write_end(struct file *file, struct address_space *mapping,
339 loff_t pos, unsigned len, unsigned copied,
340 struct page *page, void *fsdata)
342 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
346 status = nfs_updatepage(file, page, offset, copied);
350 page_cache_release(page);
352 return status < 0 ? status : copied;
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 #ifdef CONFIG_NFS_DIRECTIO
385 .direct_IO = nfs_direct_IO,
387 .launder_page = nfs_launder_page,
390 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
392 struct file *filp = vma->vm_file;
396 struct address_space *mapping;
400 mapping = page->mapping;
401 if (mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping) {
405 pagelen = nfs_page_length(page);
406 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
410 * we can use mapping after releasing the page lock, because:
411 * we hold mmap_sem on the fault path, which should pin the vma
412 * which should pin the file, which pins the dentry which should
413 * hold a reference on inode.
417 struct page *page2 = NULL;
418 ret = nfs_write_begin(filp, mapping, offset, pagelen,
421 ret = nfs_write_end(filp, mapping, offset, pagelen,
422 pagelen, page2, fsdata);
427 static struct vm_operations_struct nfs_file_vm_ops = {
428 .fault = filemap_fault,
429 .page_mkwrite = nfs_vm_page_mkwrite,
432 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
434 struct nfs_open_context *ctx;
436 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
438 ctx = nfs_file_open_context(filp);
439 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
444 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
445 unsigned long nr_segs, loff_t pos)
447 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
448 struct inode * inode = dentry->d_inode;
450 size_t count = iov_length(iov, nr_segs);
452 #ifdef CONFIG_NFS_DIRECTIO
453 if (iocb->ki_filp->f_flags & O_DIRECT)
454 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
457 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
458 dentry->d_parent->d_name.name, dentry->d_name.name,
459 inode->i_ino, (unsigned long) count, (long long) pos);
462 if (IS_SWAPFILE(inode))
465 * O_APPEND implies that we must revalidate the file length.
467 if (iocb->ki_filp->f_flags & O_APPEND) {
468 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
477 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
478 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
479 /* Return error values for O_SYNC and IS_SYNC() */
480 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
481 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
489 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
493 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
495 struct inode *inode = filp->f_mapping->host;
499 /* Try local locking first */
500 posix_test_lock(filp, fl);
501 if (fl->fl_type != F_UNLCK) {
502 /* found a conflict */
506 if (nfs_have_delegation(inode, FMODE_READ))
509 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
512 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
517 fl->fl_type = F_UNLCK;
521 static int do_vfs_lock(struct file *file, struct file_lock *fl)
524 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
526 res = posix_lock_file_wait(file, fl);
529 res = flock_lock_file_wait(file, fl);
535 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
541 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
543 struct inode *inode = filp->f_mapping->host;
547 * Flush all pending writes before doing anything
550 nfs_sync_mapping(filp->f_mapping);
552 /* NOTE: special case
553 * If we're signalled while cleaning up locks on process exit, we
554 * still need to complete the unlock.
557 /* Use local locking if mounted with "-onolock" */
558 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
559 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
561 status = do_vfs_lock(filp, fl);
566 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
568 struct inode *inode = filp->f_mapping->host;
572 * Flush all pending writes before doing anything
575 status = nfs_sync_mapping(filp->f_mapping);
580 /* Use local locking if mounted with "-onolock" */
581 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
582 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
583 /* If we were signalled we still need to ensure that
584 * we clean up any state on the server. We therefore
585 * record the lock call as having succeeded in order to
586 * ensure that locks_remove_posix() cleans it out when
589 if (status == -EINTR || status == -ERESTARTSYS)
590 do_vfs_lock(filp, fl);
592 status = do_vfs_lock(filp, fl);
597 * Make sure we clear the cache whenever we try to get the lock.
598 * This makes locking act as a cache coherency point.
600 nfs_sync_mapping(filp->f_mapping);
601 nfs_zap_caches(inode);
607 * Lock a (portion of) a file
609 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
611 struct inode * inode = filp->f_mapping->host;
613 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
614 inode->i_sb->s_id, inode->i_ino,
615 fl->fl_type, fl->fl_flags,
616 (long long)fl->fl_start, (long long)fl->fl_end);
617 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
619 /* No mandatory locks over NFS */
620 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
624 return do_getlk(filp, cmd, fl);
625 if (fl->fl_type == F_UNLCK)
626 return do_unlk(filp, cmd, fl);
627 return do_setlk(filp, cmd, fl);
631 * Lock a (portion of) a file
633 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
635 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
636 filp->f_path.dentry->d_inode->i_sb->s_id,
637 filp->f_path.dentry->d_inode->i_ino,
638 fl->fl_type, fl->fl_flags);
641 * No BSD flocks over NFS allowed.
642 * Note: we could try to fake a POSIX lock request here by
643 * using ((u32) filp | 0x80000000) or some such as the pid.
644 * Not sure whether that would be unique, though, or whether
645 * that would break in other places.
647 if (!(fl->fl_flags & FL_FLOCK))
650 /* We're simulating flock() locks using posix locks on the server */
651 fl->fl_owner = (fl_owner_t)filp;
653 fl->fl_end = OFFSET_MAX;
655 if (fl->fl_type == F_UNLCK)
656 return do_unlk(filp, cmd, fl);
657 return do_setlk(filp, cmd, fl);
660 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
663 * There is no protocol support for leases, so we have no way
664 * to implement them correctly in the face of opens by other