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,
68 .mmap = nfs_file_mmap,
70 .mmap = generic_file_mmap,
72 .open = nfs_file_open,
73 .flush = nfs_file_flush,
74 .release = nfs_file_release,
78 .splice_read = nfs_file_splice_read,
79 .check_flags = nfs_check_flags,
80 .setlease = nfs_setlease,
83 const struct inode_operations nfs_file_inode_operations = {
84 .permission = nfs_permission,
85 .getattr = nfs_getattr,
86 .setattr = nfs_setattr,
90 const struct inode_operations nfs3_file_inode_operations = {
91 .permission = nfs_permission,
92 .getattr = nfs_getattr,
93 .setattr = nfs_setattr,
94 .listxattr = nfs3_listxattr,
95 .getxattr = nfs3_getxattr,
96 .setxattr = nfs3_setxattr,
97 .removexattr = nfs3_removexattr,
99 #endif /* CONFIG_NFS_v3 */
101 /* Hack for future NFS swap support */
103 # define IS_SWAPFILE(inode) (0)
106 static int nfs_check_flags(int flags)
108 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
118 nfs_file_open(struct inode *inode, struct file *filp)
122 res = nfs_check_flags(filp->f_flags);
126 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
128 res = NFS_PROTO(inode)->file_open(inode, filp);
134 nfs_file_release(struct inode *inode, struct file *filp)
136 /* Ensure that dirty pages are flushed out with the right creds */
137 if (filp->f_mode & FMODE_WRITE)
138 nfs_wb_all(filp->f_path.dentry->d_inode);
139 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
140 return NFS_PROTO(inode)->file_release(inode, filp);
144 * nfs_revalidate_size - Revalidate the file size
145 * @inode - pointer to inode struct
146 * @file - pointer to struct file
148 * Revalidates the file length. This is basically a wrapper around
149 * nfs_revalidate_inode() that takes into account the fact that we may
150 * have cached writes (in which case we don't care about the server's
151 * idea of what the file length is), or O_DIRECT (in which case we
152 * shouldn't trust the cache).
154 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
156 struct nfs_server *server = NFS_SERVER(inode);
157 struct nfs_inode *nfsi = NFS_I(inode);
159 if (server->flags & NFS_MOUNT_NOAC)
161 if (filp->f_flags & O_DIRECT)
163 if (nfsi->npages != 0)
165 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
168 return __nfs_revalidate_inode(server, inode);
171 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
173 /* origin == SEEK_END => we must revalidate the cached file length */
174 if (origin == SEEK_END) {
175 struct inode *inode = filp->f_mapping->host;
176 int retval = nfs_revalidate_file_size(inode, filp);
178 return (loff_t)retval;
180 return remote_llseek(filp, offset, origin);
184 * Helper for nfs_file_flush() and nfs_fsync()
186 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
187 * disk, but it retrieves and clears ctx->error after synching, despite
188 * the two being set at the same time in nfs_context_set_write_error().
189 * This is because the former is used to notify the _next_ call to
190 * nfs_file_write() that a write error occured, and hence cause it to
191 * fall back to doing a synchronous write.
193 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
195 int have_error, status;
198 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
199 status = nfs_wb_all(inode);
200 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
202 ret = xchg(&ctx->error, 0);
209 * Flush all dirty pages, and check for write errors.
213 nfs_file_flush(struct file *file, fl_owner_t id)
215 struct nfs_open_context *ctx = nfs_file_open_context(file);
216 struct inode *inode = file->f_path.dentry->d_inode;
219 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
221 if ((file->f_mode & FMODE_WRITE) == 0)
223 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
225 /* Ensure that data+attribute caches are up to date after close() */
226 status = nfs_do_fsync(ctx, inode);
228 nfs_revalidate_inode(NFS_SERVER(inode), inode);
233 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
234 unsigned long nr_segs, loff_t pos)
236 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
237 struct inode * inode = dentry->d_inode;
239 size_t count = iov_length(iov, nr_segs);
241 #ifdef CONFIG_NFS_DIRECTIO
242 if (iocb->ki_filp->f_flags & O_DIRECT)
243 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
246 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
247 dentry->d_parent->d_name.name, dentry->d_name.name,
248 (unsigned long) count, (unsigned long) pos);
250 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
251 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
253 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
258 nfs_file_splice_read(struct file *filp, loff_t *ppos,
259 struct pipe_inode_info *pipe, size_t count,
262 struct dentry *dentry = filp->f_path.dentry;
263 struct inode *inode = dentry->d_inode;
266 dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
267 dentry->d_parent->d_name.name, dentry->d_name.name,
268 (unsigned long) count, (unsigned long long) *ppos);
270 res = nfs_revalidate_mapping(inode, filp->f_mapping);
272 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
277 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
279 struct dentry *dentry = file->f_path.dentry;
280 struct inode *inode = dentry->d_inode;
283 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
284 dentry->d_parent->d_name.name, dentry->d_name.name);
286 status = nfs_revalidate_mapping(inode, file->f_mapping);
288 vma->vm_ops = &nfs_file_vm_ops;
289 vma->vm_flags |= VM_CAN_NONLINEAR;
296 * Flush any dirty pages for this process, and check for write errors.
297 * The return status from this call provides a reliable indication of
298 * whether any write errors occurred for this process.
301 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
303 struct nfs_open_context *ctx = nfs_file_open_context(file);
304 struct inode *inode = dentry->d_inode;
306 dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
308 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
309 return nfs_do_fsync(ctx, inode);
313 * This does the "real" work of the write. We must allocate and lock the
314 * page to be sent back to the generic routine, which then copies the
315 * data from user space.
317 * If the writer ends up delaying the write, the writer needs to
318 * increment the page use counts until he is done with the page.
320 static int nfs_write_begin(struct file *file, struct address_space *mapping,
321 loff_t pos, unsigned len, unsigned flags,
322 struct page **pagep, void **fsdata)
327 index = pos >> PAGE_CACHE_SHIFT;
329 page = __grab_cache_page(mapping, index);
334 ret = nfs_flush_incompatible(file, page);
337 page_cache_release(page);
342 static int nfs_write_end(struct file *file, struct address_space *mapping,
343 loff_t pos, unsigned len, unsigned copied,
344 struct page *page, void *fsdata)
346 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
350 status = nfs_updatepage(file, page, offset, copied);
354 page_cache_release(page);
361 static void nfs_invalidate_page(struct page *page, unsigned long offset)
365 /* Cancel any unstarted writes on this page */
366 nfs_wb_page_cancel(page->mapping->host, page);
369 static int nfs_release_page(struct page *page, gfp_t gfp)
371 /* If PagePrivate() is set, then the page is not freeable */
375 static int nfs_launder_page(struct page *page)
377 return nfs_wb_page(page->mapping->host, page);
380 const struct address_space_operations nfs_file_aops = {
381 .readpage = nfs_readpage,
382 .readpages = nfs_readpages,
383 .set_page_dirty = __set_page_dirty_nobuffers,
384 .writepage = nfs_writepage,
385 .writepages = nfs_writepages,
386 .write_begin = nfs_write_begin,
387 .write_end = nfs_write_end,
388 .invalidatepage = nfs_invalidate_page,
389 .releasepage = nfs_release_page,
390 #ifdef CONFIG_NFS_DIRECTIO
391 .direct_IO = nfs_direct_IO,
393 .launder_page = nfs_launder_page,
396 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
398 struct file *filp = vma->vm_file;
401 struct address_space *mapping;
404 mapping = page->mapping;
405 if (mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping)
409 pagelen = nfs_page_length(page);
413 ret = nfs_flush_incompatible(filp, page);
417 ret = nfs_updatepage(filp, page, 0, pagelen);
425 static struct vm_operations_struct nfs_file_vm_ops = {
426 .fault = filemap_fault,
427 .page_mkwrite = nfs_vm_page_mkwrite,
430 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
432 struct nfs_open_context *ctx;
434 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
436 ctx = nfs_file_open_context(filp);
437 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
442 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
443 unsigned long nr_segs, loff_t pos)
445 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
446 struct inode * inode = dentry->d_inode;
448 size_t count = iov_length(iov, nr_segs);
450 #ifdef CONFIG_NFS_DIRECTIO
451 if (iocb->ki_filp->f_flags & O_DIRECT)
452 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
455 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
456 dentry->d_parent->d_name.name, dentry->d_name.name,
457 inode->i_ino, (unsigned long) count, (long long) pos);
460 if (IS_SWAPFILE(inode))
463 * O_APPEND implies that we must revalidate the file length.
465 if (iocb->ki_filp->f_flags & O_APPEND) {
466 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
475 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
476 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
477 /* Return error values for O_SYNC and IS_SYNC() */
478 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
479 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
487 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
491 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
493 struct inode *inode = filp->f_mapping->host;
497 /* Try local locking first */
498 posix_test_lock(filp, fl);
499 if (fl->fl_type != F_UNLCK) {
500 /* found a conflict */
504 if (nfs_have_delegation(inode, FMODE_READ))
507 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
510 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
515 fl->fl_type = F_UNLCK;
519 static int do_vfs_lock(struct file *file, struct file_lock *fl)
522 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
524 res = posix_lock_file_wait(file, fl);
527 res = flock_lock_file_wait(file, fl);
533 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
539 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
541 struct inode *inode = filp->f_mapping->host;
545 * Flush all pending writes before doing anything
548 nfs_sync_mapping(filp->f_mapping);
550 /* NOTE: special case
551 * If we're signalled while cleaning up locks on process exit, we
552 * still need to complete the unlock.
555 /* Use local locking if mounted with "-onolock" */
556 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
557 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
559 status = do_vfs_lock(filp, fl);
564 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
566 struct inode *inode = filp->f_mapping->host;
570 * Flush all pending writes before doing anything
573 status = nfs_sync_mapping(filp->f_mapping);
578 /* Use local locking if mounted with "-onolock" */
579 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
580 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
581 /* If we were signalled we still need to ensure that
582 * we clean up any state on the server. We therefore
583 * record the lock call as having succeeded in order to
584 * ensure that locks_remove_posix() cleans it out when
587 if (status == -EINTR || status == -ERESTARTSYS)
588 do_vfs_lock(filp, fl);
590 status = do_vfs_lock(filp, fl);
595 * Make sure we clear the cache whenever we try to get the lock.
596 * This makes locking act as a cache coherency point.
598 nfs_sync_mapping(filp->f_mapping);
599 nfs_zap_caches(inode);
605 * Lock a (portion of) a file
607 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
609 struct inode * inode = filp->f_mapping->host;
611 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
612 inode->i_sb->s_id, inode->i_ino,
613 fl->fl_type, fl->fl_flags,
614 (long long)fl->fl_start, (long long)fl->fl_end);
615 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
617 /* No mandatory locks over NFS */
618 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
622 return do_getlk(filp, cmd, fl);
623 if (fl->fl_type == F_UNLCK)
624 return do_unlk(filp, cmd, fl);
625 return do_setlk(filp, cmd, fl);
629 * Lock a (portion of) a file
631 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
633 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
634 filp->f_path.dentry->d_inode->i_sb->s_id,
635 filp->f_path.dentry->d_inode->i_ino,
636 fl->fl_type, fl->fl_flags);
639 * No BSD flocks over NFS allowed.
640 * Note: we could try to fake a POSIX lock request here by
641 * using ((u32) filp | 0x80000000) or some such as the pid.
642 * Not sure whether that would be unique, though, or whether
643 * that would break in other places.
645 if (!(fl->fl_flags & FL_FLOCK))
648 /* We're simulating flock() locks using posix locks on the server */
649 fl->fl_owner = (fl_owner_t)filp;
651 fl->fl_end = OFFSET_MAX;
653 if (fl->fl_type == F_UNLCK)
654 return do_unlk(filp, cmd, fl);
655 return do_setlk(filp, cmd, fl);
658 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
661 * There is no protocol support for leases, so we have no way
662 * to implement them correctly in the face of opens by other