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,
67 .mmap = nfs_file_mmap,
68 .open = nfs_file_open,
69 .flush = nfs_file_flush,
70 .release = nfs_file_release,
71 .fsync = nfs_file_fsync,
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 dprintk("NFS: open file(%s/%s)\n",
119 filp->f_path.dentry->d_parent->d_name.name,
120 filp->f_path.dentry->d_name.name);
122 res = nfs_check_flags(filp->f_flags);
126 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
127 res = nfs_open(inode, filp);
132 nfs_file_release(struct inode *inode, struct file *filp)
134 struct dentry *dentry = filp->f_path.dentry;
136 dprintk("NFS: release(%s/%s)\n",
137 dentry->d_parent->d_name.name,
138 dentry->d_name.name);
140 /* Ensure that dirty pages are flushed out with the right creds */
141 if (filp->f_mode & FMODE_WRITE)
142 nfs_wb_all(dentry->d_inode);
143 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
144 return nfs_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)
179 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
180 filp->f_path.dentry->d_parent->d_name.name,
181 filp->f_path.dentry->d_name.name,
184 /* origin == SEEK_END => we must revalidate the cached file length */
185 if (origin == SEEK_END) {
186 struct inode *inode = filp->f_mapping->host;
188 int retval = nfs_revalidate_file_size(inode, filp);
190 return (loff_t)retval;
192 spin_lock(&inode->i_lock);
193 loff = generic_file_llseek_unlocked(filp, offset, origin);
194 spin_unlock(&inode->i_lock);
196 loff = generic_file_llseek_unlocked(filp, offset, origin);
201 * Helper for nfs_file_flush() and nfs_file_fsync()
203 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
204 * disk, but it retrieves and clears ctx->error after synching, despite
205 * the two being set at the same time in nfs_context_set_write_error().
206 * This is because the former is used to notify the _next_ call to
207 * nfs_file_write() that a write error occured, and hence cause it to
208 * fall back to doing a synchronous write.
210 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
212 int have_error, status;
215 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
216 status = nfs_wb_all(inode);
217 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
219 ret = xchg(&ctx->error, 0);
226 * Flush all dirty pages, and check for write errors.
229 nfs_file_flush(struct file *file, fl_owner_t id)
231 struct nfs_open_context *ctx = nfs_file_open_context(file);
232 struct dentry *dentry = file->f_path.dentry;
233 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 /* Ensure that data+attribute caches are up to date after close() */
245 status = nfs_do_fsync(ctx, inode);
247 nfs_revalidate_inode(NFS_SERVER(inode), inode);
252 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
253 unsigned long nr_segs, loff_t pos)
255 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
256 struct inode * inode = dentry->d_inode;
258 size_t count = iov_length(iov, nr_segs);
260 if (iocb->ki_filp->f_flags & O_DIRECT)
261 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
263 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
264 dentry->d_parent->d_name.name, dentry->d_name.name,
265 (unsigned long) count, (unsigned long) pos);
267 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
268 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
270 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
275 nfs_file_splice_read(struct file *filp, loff_t *ppos,
276 struct pipe_inode_info *pipe, size_t count,
279 struct dentry *dentry = filp->f_path.dentry;
280 struct inode *inode = dentry->d_inode;
283 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
284 dentry->d_parent->d_name.name, dentry->d_name.name,
285 (unsigned long) count, (unsigned long long) *ppos);
287 res = nfs_revalidate_mapping(inode, filp->f_mapping);
289 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
294 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
296 struct dentry *dentry = file->f_path.dentry;
297 struct inode *inode = dentry->d_inode;
300 dprintk("NFS: mmap(%s/%s)\n",
301 dentry->d_parent->d_name.name, dentry->d_name.name);
303 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
304 * so we call that before revalidating the mapping
306 status = generic_file_mmap(file, vma);
308 vma->vm_ops = &nfs_file_vm_ops;
309 status = nfs_revalidate_mapping(inode, file->f_mapping);
315 * Flush any dirty pages for this process, and check for write errors.
316 * The return status from this call provides a reliable indication of
317 * whether any write errors occurred for this process.
320 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
322 struct nfs_open_context *ctx = nfs_file_open_context(file);
323 struct inode *inode = dentry->d_inode;
325 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
326 dentry->d_parent->d_name.name, dentry->d_name.name,
329 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
330 return nfs_do_fsync(ctx, inode);
334 * This does the "real" work of the write. We must allocate and lock the
335 * page to be sent back to the generic routine, which then copies the
336 * data from user space.
338 * If the writer ends up delaying the write, the writer needs to
339 * increment the page use counts until he is done with the page.
341 static int nfs_write_begin(struct file *file, struct address_space *mapping,
342 loff_t pos, unsigned len, unsigned flags,
343 struct page **pagep, void **fsdata)
348 index = pos >> PAGE_CACHE_SHIFT;
350 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
351 file->f_path.dentry->d_parent->d_name.name,
352 file->f_path.dentry->d_name.name,
353 mapping->host->i_ino, len, (long long) pos);
356 * Prevent starvation issues if someone is doing a consistency
359 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
360 nfs_wait_bit_killable, TASK_KILLABLE);
364 page = grab_cache_page_write_begin(mapping, index, flags);
369 ret = nfs_flush_incompatible(file, page);
372 page_cache_release(page);
377 static int nfs_write_end(struct file *file, struct address_space *mapping,
378 loff_t pos, unsigned len, unsigned copied,
379 struct page *page, void *fsdata)
381 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
384 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
385 file->f_path.dentry->d_parent->d_name.name,
386 file->f_path.dentry->d_name.name,
387 mapping->host->i_ino, len, (long long) pos);
390 * Zero any uninitialised parts of the page, and then mark the page
391 * as up to date if it turns out that we're extending the file.
393 if (!PageUptodate(page)) {
394 unsigned pglen = nfs_page_length(page);
395 unsigned end = offset + len;
398 zero_user_segments(page, 0, offset,
399 end, PAGE_CACHE_SIZE);
400 SetPageUptodate(page);
401 } else if (end >= pglen) {
402 zero_user_segment(page, end, PAGE_CACHE_SIZE);
404 SetPageUptodate(page);
406 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
409 status = nfs_updatepage(file, page, offset, copied);
412 page_cache_release(page);
419 static void nfs_invalidate_page(struct page *page, unsigned long offset)
421 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
425 /* Cancel any unstarted writes on this page */
426 nfs_wb_page_cancel(page->mapping->host, page);
429 static int nfs_release_page(struct page *page, gfp_t gfp)
431 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
433 /* If PagePrivate() is set, then the page is not freeable */
437 static int nfs_launder_page(struct page *page)
439 struct inode *inode = page->mapping->host;
441 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
442 inode->i_ino, (long long)page_offset(page));
444 return nfs_wb_page(inode, page);
447 const struct address_space_operations nfs_file_aops = {
448 .readpage = nfs_readpage,
449 .readpages = nfs_readpages,
450 .set_page_dirty = __set_page_dirty_nobuffers,
451 .writepage = nfs_writepage,
452 .writepages = nfs_writepages,
453 .write_begin = nfs_write_begin,
454 .write_end = nfs_write_end,
455 .invalidatepage = nfs_invalidate_page,
456 .releasepage = nfs_release_page,
457 .direct_IO = nfs_direct_IO,
458 .launder_page = nfs_launder_page,
461 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
463 struct file *filp = vma->vm_file;
464 struct dentry *dentry = filp->f_path.dentry;
467 struct address_space *mapping;
469 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
470 dentry->d_parent->d_name.name, dentry->d_name.name,
471 filp->f_mapping->host->i_ino,
472 (long long)page_offset(page));
475 mapping = page->mapping;
476 if (mapping != dentry->d_inode->i_mapping)
480 pagelen = nfs_page_length(page);
484 ret = nfs_flush_incompatible(filp, page);
488 ret = nfs_updatepage(filp, page, 0, pagelen);
496 static struct vm_operations_struct nfs_file_vm_ops = {
497 .fault = filemap_fault,
498 .page_mkwrite = nfs_vm_page_mkwrite,
501 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
503 struct nfs_open_context *ctx;
505 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
507 ctx = nfs_file_open_context(filp);
508 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
513 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
514 unsigned long nr_segs, loff_t pos)
516 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
517 struct inode * inode = dentry->d_inode;
519 size_t count = iov_length(iov, nr_segs);
521 if (iocb->ki_filp->f_flags & O_DIRECT)
522 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
524 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
525 dentry->d_parent->d_name.name, dentry->d_name.name,
526 (unsigned long) count, (long long) pos);
529 if (IS_SWAPFILE(inode))
532 * O_APPEND implies that we must revalidate the file length.
534 if (iocb->ki_filp->f_flags & O_APPEND) {
535 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
544 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
545 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
546 /* Return error values for O_SYNC and IS_SYNC() */
547 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
548 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
556 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
560 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
562 struct inode *inode = filp->f_mapping->host;
566 /* Try local locking first */
567 posix_test_lock(filp, fl);
568 if (fl->fl_type != F_UNLCK) {
569 /* found a conflict */
573 if (nfs_have_delegation(inode, FMODE_READ))
576 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
579 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
584 fl->fl_type = F_UNLCK;
588 static int do_vfs_lock(struct file *file, struct file_lock *fl)
591 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
593 res = posix_lock_file_wait(file, fl);
596 res = flock_lock_file_wait(file, fl);
602 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
608 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
610 struct inode *inode = filp->f_mapping->host;
614 * Flush all pending writes before doing anything
617 nfs_sync_mapping(filp->f_mapping);
619 /* NOTE: special case
620 * If we're signalled while cleaning up locks on process exit, we
621 * still need to complete the unlock.
624 /* Use local locking if mounted with "-onolock" */
625 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
626 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
628 status = do_vfs_lock(filp, fl);
633 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
635 struct inode *inode = filp->f_mapping->host;
639 * Flush all pending writes before doing anything
642 status = nfs_sync_mapping(filp->f_mapping);
647 /* Use local locking if mounted with "-onolock" */
648 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
649 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
651 status = do_vfs_lock(filp, fl);
656 * Make sure we clear the cache whenever we try to get the lock.
657 * This makes locking act as a cache coherency point.
659 nfs_sync_mapping(filp->f_mapping);
660 if (!nfs_have_delegation(inode, FMODE_READ))
661 nfs_zap_caches(inode);
667 * Lock a (portion of) a file
669 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
671 struct inode *inode = filp->f_mapping->host;
674 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
675 filp->f_path.dentry->d_parent->d_name.name,
676 filp->f_path.dentry->d_name.name,
677 fl->fl_type, fl->fl_flags,
678 (long long)fl->fl_start, (long long)fl->fl_end);
680 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
682 /* No mandatory locks over NFS */
683 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
686 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
687 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
693 ret = do_getlk(filp, cmd, fl);
694 else if (fl->fl_type == F_UNLCK)
695 ret = do_unlk(filp, cmd, fl);
697 ret = do_setlk(filp, cmd, fl);
703 * Lock a (portion of) a file
705 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
707 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
708 filp->f_path.dentry->d_parent->d_name.name,
709 filp->f_path.dentry->d_name.name,
710 fl->fl_type, fl->fl_flags);
712 if (!(fl->fl_flags & FL_FLOCK))
715 /* We're simulating flock() locks using posix locks on the server */
716 fl->fl_owner = (fl_owner_t)filp;
718 fl->fl_end = OFFSET_MAX;
720 if (fl->fl_type == F_UNLCK)
721 return do_unlk(filp, cmd, fl);
722 return do_setlk(filp, cmd, fl);
726 * There is no protocol support for leases, so we have no way to implement
727 * them correctly in the face of opens by other clients.
729 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
731 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
732 file->f_path.dentry->d_parent->d_name.name,
733 file->f_path.dentry->d_name.name, arg);