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/aio.h>
31 #include <asm/uaccess.h>
32 #include <asm/system.h>
34 #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_splice_write(struct pipe_inode_info *pipe,
51 struct file *filp, loff_t *ppos,
52 size_t count, unsigned int flags);
53 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
54 unsigned long nr_segs, loff_t pos);
55 static int nfs_file_flush(struct file *, fl_owner_t id);
56 static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
57 static int nfs_check_flags(int flags);
58 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
59 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
60 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
62 static struct vm_operations_struct nfs_file_vm_ops;
64 const struct file_operations nfs_file_operations = {
65 .llseek = nfs_file_llseek,
67 .write = do_sync_write,
68 .aio_read = nfs_file_read,
69 .aio_write = nfs_file_write,
70 .mmap = nfs_file_mmap,
71 .open = nfs_file_open,
72 .flush = nfs_file_flush,
73 .release = nfs_file_release,
74 .fsync = nfs_file_fsync,
77 .splice_read = nfs_file_splice_read,
78 .splice_write = nfs_file_splice_write,
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 dprintk("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);
131 res = nfs_open(inode, filp);
136 nfs_file_release(struct inode *inode, struct file *filp)
138 struct dentry *dentry = filp->f_path.dentry;
140 dprintk("NFS: release(%s/%s)\n",
141 dentry->d_parent->d_name.name,
142 dentry->d_name.name);
144 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
145 return nfs_release(inode, filp);
149 * nfs_revalidate_size - Revalidate the file size
150 * @inode - pointer to inode struct
151 * @file - pointer to struct file
153 * Revalidates the file length. This is basically a wrapper around
154 * nfs_revalidate_inode() that takes into account the fact that we may
155 * have cached writes (in which case we don't care about the server's
156 * idea of what the file length is), or O_DIRECT (in which case we
157 * shouldn't trust the cache).
159 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
161 struct nfs_server *server = NFS_SERVER(inode);
162 struct nfs_inode *nfsi = NFS_I(inode);
164 if (server->flags & NFS_MOUNT_NOAC)
166 if (filp->f_flags & O_DIRECT)
168 if (nfsi->npages != 0)
170 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
173 return __nfs_revalidate_inode(server, inode);
176 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
180 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
181 filp->f_path.dentry->d_parent->d_name.name,
182 filp->f_path.dentry->d_name.name,
185 /* origin == SEEK_END => we must revalidate the cached file length */
186 if (origin == SEEK_END) {
187 struct inode *inode = filp->f_mapping->host;
189 int retval = nfs_revalidate_file_size(inode, filp);
191 return (loff_t)retval;
193 spin_lock(&inode->i_lock);
194 loff = generic_file_llseek_unlocked(filp, offset, origin);
195 spin_unlock(&inode->i_lock);
197 loff = generic_file_llseek_unlocked(filp, offset, origin);
202 * Helper for nfs_file_flush() and nfs_file_fsync()
204 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
205 * disk, but it retrieves and clears ctx->error after synching, despite
206 * the two being set at the same time in nfs_context_set_write_error().
207 * This is because the former is used to notify the _next_ call to
208 * nfs_file_write() that a write error occured, and hence cause it to
209 * fall back to doing a synchronous write.
211 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
213 int have_error, status;
216 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
217 status = nfs_wb_all(inode);
218 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
220 ret = xchg(&ctx->error, 0);
227 * Flush all dirty pages, and check for write errors.
230 nfs_file_flush(struct file *file, fl_owner_t id)
232 struct nfs_open_context *ctx = nfs_file_open_context(file);
233 struct dentry *dentry = file->f_path.dentry;
234 struct inode *inode = dentry->d_inode;
236 dprintk("NFS: flush(%s/%s)\n",
237 dentry->d_parent->d_name.name,
238 dentry->d_name.name);
240 if ((file->f_mode & FMODE_WRITE) == 0)
242 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
244 /* Flush writes to the server and return any errors */
245 return nfs_do_fsync(ctx, inode);
249 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
250 unsigned long nr_segs, loff_t pos)
252 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
253 struct inode * inode = dentry->d_inode;
255 size_t count = iov_length(iov, nr_segs);
257 if (iocb->ki_filp->f_flags & O_DIRECT)
258 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
260 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
261 dentry->d_parent->d_name.name, dentry->d_name.name,
262 (unsigned long) count, (unsigned long) pos);
264 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
265 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
267 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
272 nfs_file_splice_read(struct file *filp, loff_t *ppos,
273 struct pipe_inode_info *pipe, size_t count,
276 struct dentry *dentry = filp->f_path.dentry;
277 struct inode *inode = dentry->d_inode;
280 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
281 dentry->d_parent->d_name.name, dentry->d_name.name,
282 (unsigned long) count, (unsigned long long) *ppos);
284 res = nfs_revalidate_mapping(inode, filp->f_mapping);
286 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
291 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
293 struct dentry *dentry = file->f_path.dentry;
294 struct inode *inode = dentry->d_inode;
297 dprintk("NFS: mmap(%s/%s)\n",
298 dentry->d_parent->d_name.name, dentry->d_name.name);
300 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
301 * so we call that before revalidating the mapping
303 status = generic_file_mmap(file, vma);
305 vma->vm_ops = &nfs_file_vm_ops;
306 status = nfs_revalidate_mapping(inode, file->f_mapping);
312 * Flush any dirty pages for this process, and check for write errors.
313 * The return status from this call provides a reliable indication of
314 * whether any write errors occurred for this process.
317 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
319 struct nfs_open_context *ctx = nfs_file_open_context(file);
320 struct inode *inode = dentry->d_inode;
322 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
323 dentry->d_parent->d_name.name, dentry->d_name.name,
326 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
327 return nfs_do_fsync(ctx, inode);
331 * This does the "real" work of the write. We must allocate and lock the
332 * page to be sent back to the generic routine, which then copies the
333 * data from user space.
335 * If the writer ends up delaying the write, the writer needs to
336 * increment the page use counts until he is done with the page.
338 static int nfs_write_begin(struct file *file, struct address_space *mapping,
339 loff_t pos, unsigned len, unsigned flags,
340 struct page **pagep, void **fsdata)
345 index = pos >> PAGE_CACHE_SHIFT;
347 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
348 file->f_path.dentry->d_parent->d_name.name,
349 file->f_path.dentry->d_name.name,
350 mapping->host->i_ino, len, (long long) pos);
353 * Prevent starvation issues if someone is doing a consistency
356 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
357 nfs_wait_bit_killable, TASK_KILLABLE);
361 page = grab_cache_page_write_begin(mapping, index, flags);
366 ret = nfs_flush_incompatible(file, page);
369 page_cache_release(page);
374 static int nfs_write_end(struct file *file, struct address_space *mapping,
375 loff_t pos, unsigned len, unsigned copied,
376 struct page *page, void *fsdata)
378 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
381 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
382 file->f_path.dentry->d_parent->d_name.name,
383 file->f_path.dentry->d_name.name,
384 mapping->host->i_ino, len, (long long) pos);
387 * Zero any uninitialised parts of the page, and then mark the page
388 * as up to date if it turns out that we're extending the file.
390 if (!PageUptodate(page)) {
391 unsigned pglen = nfs_page_length(page);
392 unsigned end = offset + len;
395 zero_user_segments(page, 0, offset,
396 end, PAGE_CACHE_SIZE);
397 SetPageUptodate(page);
398 } else if (end >= pglen) {
399 zero_user_segment(page, end, PAGE_CACHE_SIZE);
401 SetPageUptodate(page);
403 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
406 status = nfs_updatepage(file, page, offset, copied);
409 page_cache_release(page);
417 * Partially or wholly invalidate a page
418 * - Release the private state associated with a page if undergoing complete
420 * - Called if either PG_private or PG_fscache is set on the page
421 * - Caller holds page lock
423 static void nfs_invalidate_page(struct page *page, unsigned long offset)
425 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
429 /* Cancel any unstarted writes on this page */
430 nfs_wb_page_cancel(page->mapping->host, page);
432 nfs_fscache_invalidate_page(page, page->mapping->host);
436 * Attempt to release the private state associated with a page
437 * - Called if either PG_private or PG_fscache is set on the page
438 * - Caller holds page lock
439 * - Return true (may release page) or false (may not)
441 static int nfs_release_page(struct page *page, gfp_t gfp)
443 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
445 /* If PagePrivate() is set, then the page is not freeable */
446 if (PagePrivate(page))
448 return nfs_fscache_release_page(page, gfp);
452 * Attempt to clear the private state associated with a page when an error
453 * occurs that requires the cached contents of an inode to be written back or
455 * - Called if either PG_private or fscache is set on the page
456 * - Caller holds page lock
457 * - Return 0 if successful, -error otherwise
459 static int nfs_launder_page(struct page *page)
461 struct inode *inode = page->mapping->host;
462 struct nfs_inode *nfsi = NFS_I(inode);
464 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
465 inode->i_ino, (long long)page_offset(page));
467 nfs_fscache_wait_on_page_write(nfsi, page);
468 return nfs_wb_page(inode, page);
471 const struct address_space_operations nfs_file_aops = {
472 .readpage = nfs_readpage,
473 .readpages = nfs_readpages,
474 .set_page_dirty = __set_page_dirty_nobuffers,
475 .writepage = nfs_writepage,
476 .writepages = nfs_writepages,
477 .write_begin = nfs_write_begin,
478 .write_end = nfs_write_end,
479 .invalidatepage = nfs_invalidate_page,
480 .releasepage = nfs_release_page,
481 .direct_IO = nfs_direct_IO,
482 .launder_page = nfs_launder_page,
486 * Notification that a PTE pointing to an NFS page is about to be made
487 * writable, implying that someone is about to modify the page through a
488 * shared-writable mapping
490 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
492 struct page *page = vmf->page;
493 struct file *filp = vma->vm_file;
494 struct dentry *dentry = filp->f_path.dentry;
497 struct address_space *mapping;
499 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
500 dentry->d_parent->d_name.name, dentry->d_name.name,
501 filp->f_mapping->host->i_ino,
502 (long long)page_offset(page));
504 /* make sure the cache has finished storing the page */
505 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
508 mapping = page->mapping;
509 if (mapping != dentry->d_inode->i_mapping)
513 pagelen = nfs_page_length(page);
517 ret = nfs_flush_incompatible(filp, page);
521 ret = nfs_updatepage(filp, page, 0, pagelen);
524 return VM_FAULT_LOCKED;
526 return VM_FAULT_SIGBUS;
529 static struct vm_operations_struct nfs_file_vm_ops = {
530 .fault = filemap_fault,
531 .page_mkwrite = nfs_vm_page_mkwrite,
534 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
536 struct nfs_open_context *ctx;
538 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
540 ctx = nfs_file_open_context(filp);
541 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
546 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
547 unsigned long nr_segs, loff_t pos)
549 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
550 struct inode * inode = dentry->d_inode;
552 size_t count = iov_length(iov, nr_segs);
554 if (iocb->ki_filp->f_flags & O_DIRECT)
555 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
557 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
558 dentry->d_parent->d_name.name, dentry->d_name.name,
559 (unsigned long) count, (long long) pos);
562 if (IS_SWAPFILE(inode))
565 * O_APPEND implies that we must revalidate the file length.
567 if (iocb->ki_filp->f_flags & O_APPEND) {
568 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
577 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
578 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
579 /* Return error values for O_SYNC and IS_SYNC() */
580 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
581 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
589 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
593 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
594 struct file *filp, loff_t *ppos,
595 size_t count, unsigned int flags)
597 struct dentry *dentry = filp->f_path.dentry;
598 struct inode *inode = dentry->d_inode;
601 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
602 dentry->d_parent->d_name.name, dentry->d_name.name,
603 (unsigned long) count, (unsigned long long) *ppos);
606 * The combination of splice and an O_APPEND destination is disallowed.
609 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
611 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
612 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
613 int err = nfs_do_fsync(nfs_file_open_context(filp), inode);
620 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
622 struct inode *inode = filp->f_mapping->host;
625 /* Try local locking first */
626 posix_test_lock(filp, fl);
627 if (fl->fl_type != F_UNLCK) {
628 /* found a conflict */
632 if (nfs_have_delegation(inode, FMODE_READ))
635 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
638 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
642 fl->fl_type = F_UNLCK;
646 static int do_vfs_lock(struct file *file, struct file_lock *fl)
649 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
651 res = posix_lock_file_wait(file, fl);
654 res = flock_lock_file_wait(file, fl);
660 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
666 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
668 struct inode *inode = filp->f_mapping->host;
672 * Flush all pending writes before doing anything
675 nfs_sync_mapping(filp->f_mapping);
677 /* NOTE: special case
678 * If we're signalled while cleaning up locks on process exit, we
679 * still need to complete the unlock.
681 /* Use local locking if mounted with "-onolock" */
682 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
683 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
685 status = do_vfs_lock(filp, fl);
689 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
691 struct inode *inode = filp->f_mapping->host;
695 * Flush all pending writes before doing anything
698 status = nfs_sync_mapping(filp->f_mapping);
702 /* Use local locking if mounted with "-onolock" */
703 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
704 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
706 status = do_vfs_lock(filp, fl);
710 * Make sure we clear the cache whenever we try to get the lock.
711 * This makes locking act as a cache coherency point.
713 nfs_sync_mapping(filp->f_mapping);
714 if (!nfs_have_delegation(inode, FMODE_READ))
715 nfs_zap_caches(inode);
721 * Lock a (portion of) a file
723 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
725 struct inode *inode = filp->f_mapping->host;
728 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
729 filp->f_path.dentry->d_parent->d_name.name,
730 filp->f_path.dentry->d_name.name,
731 fl->fl_type, fl->fl_flags,
732 (long long)fl->fl_start, (long long)fl->fl_end);
734 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
736 /* No mandatory locks over NFS */
737 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
740 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
741 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
747 ret = do_getlk(filp, cmd, fl);
748 else if (fl->fl_type == F_UNLCK)
749 ret = do_unlk(filp, cmd, fl);
751 ret = do_setlk(filp, cmd, fl);
757 * Lock a (portion of) a file
759 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
761 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
762 filp->f_path.dentry->d_parent->d_name.name,
763 filp->f_path.dentry->d_name.name,
764 fl->fl_type, fl->fl_flags);
766 if (!(fl->fl_flags & FL_FLOCK))
769 /* We're simulating flock() locks using posix locks on the server */
770 fl->fl_owner = (fl_owner_t)filp;
772 fl->fl_end = OFFSET_MAX;
774 if (fl->fl_type == F_UNLCK)
775 return do_unlk(filp, cmd, fl);
776 return do_setlk(filp, cmd, fl);
780 * There is no protocol support for leases, so we have no way to implement
781 * them correctly in the face of opens by other clients.
783 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
785 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
786 file->f_path.dentry->d_parent->d_name.name,
787 file->f_path.dentry->d_name.name, arg);