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_file_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,
75 .fsync = nfs_file_fsync,
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 dfprintk(VFS, "NFS: open file(%s/%s)\n",
123 filp->f_path.dentry->d_parent->d_name.name,
124 filp->f_path.dentry->d_name.name);
126 res = nfs_check_flags(filp->f_flags);
130 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
132 res = NFS_PROTO(inode)->file_open(inode, filp);
138 nfs_file_release(struct inode *inode, struct file *filp)
140 /* Ensure that dirty pages are flushed out with the right creds */
141 if (filp->f_mode & FMODE_WRITE)
142 nfs_wb_all(filp->f_path.dentry->d_inode);
143 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
144 return NFS_PROTO(inode)->file_release(inode, filp);
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 dfprintk(VFS, "NFS: llseek file(%s/%s, %lld, %d)\n",
178 filp->f_path.dentry->d_parent->d_name.name,
179 filp->f_path.dentry->d_name.name,
182 /* origin == SEEK_END => we must revalidate the cached file length */
183 if (origin == SEEK_END) {
184 struct inode *inode = filp->f_mapping->host;
185 int retval = nfs_revalidate_file_size(inode, filp);
187 return (loff_t)retval;
189 return remote_llseek(filp, offset, origin);
193 * Helper for nfs_file_flush() and nfs_file_fsync()
195 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
196 * disk, but it retrieves and clears ctx->error after synching, despite
197 * the two being set at the same time in nfs_context_set_write_error().
198 * This is because the former is used to notify the _next_ call to
199 * nfs_file_write() that a write error occured, and hence cause it to
200 * fall back to doing a synchronous write.
202 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
204 int have_error, status;
207 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
208 status = nfs_wb_all(inode);
209 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
211 ret = xchg(&ctx->error, 0);
218 * Flush all dirty pages, and check for write errors.
222 nfs_file_flush(struct file *file, fl_owner_t id)
224 struct nfs_open_context *ctx = nfs_file_open_context(file);
225 struct inode *inode = file->f_path.dentry->d_inode;
228 dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
230 if ((file->f_mode & FMODE_WRITE) == 0)
232 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
234 /* Ensure that data+attribute caches are up to date after close() */
235 status = nfs_do_fsync(ctx, inode);
237 nfs_revalidate_inode(NFS_SERVER(inode), inode);
242 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
243 unsigned long nr_segs, loff_t pos)
245 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
246 struct inode * inode = dentry->d_inode;
248 size_t count = iov_length(iov, nr_segs);
250 if (iocb->ki_filp->f_flags & O_DIRECT)
251 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
253 dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
254 dentry->d_parent->d_name.name, dentry->d_name.name,
255 (unsigned long) count, (unsigned long) pos);
257 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
258 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
260 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
265 nfs_file_splice_read(struct file *filp, loff_t *ppos,
266 struct pipe_inode_info *pipe, size_t count,
269 struct dentry *dentry = filp->f_path.dentry;
270 struct inode *inode = dentry->d_inode;
273 dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
274 dentry->d_parent->d_name.name, dentry->d_name.name,
275 (unsigned long) count, (unsigned long long) *ppos);
277 res = nfs_revalidate_mapping(inode, filp->f_mapping);
279 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
284 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
286 struct dentry *dentry = file->f_path.dentry;
287 struct inode *inode = dentry->d_inode;
290 dfprintk(VFS, "nfs: mmap(%s/%s)\n",
291 dentry->d_parent->d_name.name, dentry->d_name.name);
293 status = nfs_revalidate_mapping(inode, file->f_mapping);
295 vma->vm_ops = &nfs_file_vm_ops;
296 vma->vm_flags |= VM_CAN_NONLINEAR;
303 * Flush any dirty pages for this process, and check for write errors.
304 * The return status from this call provides a reliable indication of
305 * whether any write errors occurred for this process.
308 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
310 struct nfs_open_context *ctx = nfs_file_open_context(file);
311 struct inode *inode = dentry->d_inode;
313 dfprintk(VFS, "NFS: fsync file(%s/%s) datasync %d\n",
314 dentry->d_parent->d_name.name, dentry->d_name.name,
317 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
318 return nfs_do_fsync(ctx, inode);
322 * This does the "real" work of the write. We must allocate and lock the
323 * page to be sent back to the generic routine, which then copies the
324 * data from user space.
326 * If the writer ends up delaying the write, the writer needs to
327 * increment the page use counts until he is done with the page.
329 static int nfs_write_begin(struct file *file, struct address_space *mapping,
330 loff_t pos, unsigned len, unsigned flags,
331 struct page **pagep, void **fsdata)
336 index = pos >> PAGE_CACHE_SHIFT;
338 page = __grab_cache_page(mapping, index);
343 ret = nfs_flush_incompatible(file, page);
346 page_cache_release(page);
351 static int nfs_write_end(struct file *file, struct address_space *mapping,
352 loff_t pos, unsigned len, unsigned copied,
353 struct page *page, void *fsdata)
355 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
359 * Zero any uninitialised parts of the page, and then mark the page
360 * as up to date if it turns out that we're extending the file.
362 if (!PageUptodate(page)) {
363 unsigned pglen = nfs_page_length(page);
364 unsigned end = offset + len;
367 zero_user_segments(page, 0, offset,
368 end, PAGE_CACHE_SIZE);
369 SetPageUptodate(page);
370 } else if (end >= pglen) {
371 zero_user_segment(page, end, PAGE_CACHE_SIZE);
373 SetPageUptodate(page);
375 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
379 status = nfs_updatepage(file, page, offset, copied);
383 page_cache_release(page);
390 static void nfs_invalidate_page(struct page *page, unsigned long offset)
394 /* Cancel any unstarted writes on this page */
395 nfs_wb_page_cancel(page->mapping->host, page);
398 static int nfs_release_page(struct page *page, gfp_t gfp)
400 /* If PagePrivate() is set, then the page is not freeable */
404 static int nfs_launder_page(struct page *page)
406 return nfs_wb_page(page->mapping->host, page);
409 const struct address_space_operations nfs_file_aops = {
410 .readpage = nfs_readpage,
411 .readpages = nfs_readpages,
412 .set_page_dirty = __set_page_dirty_nobuffers,
413 .writepage = nfs_writepage,
414 .writepages = nfs_writepages,
415 .write_begin = nfs_write_begin,
416 .write_end = nfs_write_end,
417 .invalidatepage = nfs_invalidate_page,
418 .releasepage = nfs_release_page,
419 .direct_IO = nfs_direct_IO,
420 .launder_page = nfs_launder_page,
423 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
425 struct file *filp = vma->vm_file;
428 struct address_space *mapping;
431 mapping = page->mapping;
432 if (mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping)
436 pagelen = nfs_page_length(page);
440 ret = nfs_flush_incompatible(filp, page);
444 ret = nfs_updatepage(filp, page, 0, pagelen);
452 static struct vm_operations_struct nfs_file_vm_ops = {
453 .fault = filemap_fault,
454 .page_mkwrite = nfs_vm_page_mkwrite,
457 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
459 struct nfs_open_context *ctx;
461 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
463 ctx = nfs_file_open_context(filp);
464 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
469 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
470 unsigned long nr_segs, loff_t pos)
472 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
473 struct inode * inode = dentry->d_inode;
475 size_t count = iov_length(iov, nr_segs);
477 if (iocb->ki_filp->f_flags & O_DIRECT)
478 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
480 dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
481 dentry->d_parent->d_name.name, dentry->d_name.name,
482 inode->i_ino, (unsigned long) count, (long long) pos);
485 if (IS_SWAPFILE(inode))
488 * O_APPEND implies that we must revalidate the file length.
490 if (iocb->ki_filp->f_flags & O_APPEND) {
491 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
500 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
501 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
502 /* Return error values for O_SYNC and IS_SYNC() */
503 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
504 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
512 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
516 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
518 struct inode *inode = filp->f_mapping->host;
522 /* Try local locking first */
523 posix_test_lock(filp, fl);
524 if (fl->fl_type != F_UNLCK) {
525 /* found a conflict */
529 if (nfs_have_delegation(inode, FMODE_READ))
532 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
535 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
540 fl->fl_type = F_UNLCK;
544 static int do_vfs_lock(struct file *file, struct file_lock *fl)
547 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
549 res = posix_lock_file_wait(file, fl);
552 res = flock_lock_file_wait(file, fl);
558 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
564 static int do_unlk(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 nfs_sync_mapping(filp->f_mapping);
575 /* NOTE: special case
576 * If we're signalled while cleaning up locks on process exit, we
577 * still need to complete the unlock.
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);
584 status = do_vfs_lock(filp, fl);
589 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
591 struct inode *inode = filp->f_mapping->host;
595 * Flush all pending writes before doing anything
598 status = nfs_sync_mapping(filp->f_mapping);
603 /* Use local locking if mounted with "-onolock" */
604 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
605 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
607 status = do_vfs_lock(filp, fl);
612 * Make sure we clear the cache whenever we try to get the lock.
613 * This makes locking act as a cache coherency point.
615 nfs_sync_mapping(filp->f_mapping);
616 if (!nfs_have_delegation(inode, FMODE_READ))
617 nfs_zap_caches(inode);
623 * Lock a (portion of) a file
625 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
627 struct inode * inode = filp->f_mapping->host;
630 dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
631 inode->i_sb->s_id, inode->i_ino,
632 fl->fl_type, fl->fl_flags,
633 (long long)fl->fl_start, (long long)fl->fl_end);
634 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
636 /* No mandatory locks over NFS */
637 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
640 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
641 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
647 ret = do_getlk(filp, cmd, fl);
648 else if (fl->fl_type == F_UNLCK)
649 ret = do_unlk(filp, cmd, fl);
651 ret = do_setlk(filp, cmd, fl);
657 * Lock a (portion of) a file
659 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
661 dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
662 filp->f_path.dentry->d_inode->i_sb->s_id,
663 filp->f_path.dentry->d_inode->i_ino,
664 fl->fl_type, fl->fl_flags);
667 * No BSD flocks over NFS allowed.
668 * Note: we could try to fake a POSIX lock request here by
669 * using ((u32) filp | 0x80000000) or some such as the pid.
670 * Not sure whether that would be unique, though, or whether
671 * that would break in other places.
673 if (!(fl->fl_flags & FL_FLOCK))
676 /* We're simulating flock() locks using posix locks on the server */
677 fl->fl_owner = (fl_owner_t)filp;
679 fl->fl_end = OFFSET_MAX;
681 if (fl->fl_type == F_UNLCK)
682 return do_unlk(filp, cmd, fl);
683 return do_setlk(filp, cmd, fl);
686 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
689 * There is no protocol support for leases, so we have no way
690 * to implement them correctly in the face of opens by other