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 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);
132 res = nfs_open(inode, filp);
138 nfs_file_release(struct inode *inode, struct file *filp)
140 struct dentry *dentry = filp->f_path.dentry;
142 dprintk("NFS: release(%s/%s)\n",
143 dentry->d_parent->d_name.name,
144 dentry->d_name.name);
146 /* Ensure that dirty pages are flushed out with the right creds */
147 if (filp->f_mode & FMODE_WRITE)
148 nfs_wb_all(dentry->d_inode);
149 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
150 return nfs_release(inode, filp);
154 * nfs_revalidate_size - Revalidate the file size
155 * @inode - pointer to inode struct
156 * @file - pointer to struct file
158 * Revalidates the file length. This is basically a wrapper around
159 * nfs_revalidate_inode() that takes into account the fact that we may
160 * have cached writes (in which case we don't care about the server's
161 * idea of what the file length is), or O_DIRECT (in which case we
162 * shouldn't trust the cache).
164 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
166 struct nfs_server *server = NFS_SERVER(inode);
167 struct nfs_inode *nfsi = NFS_I(inode);
169 if (server->flags & NFS_MOUNT_NOAC)
171 if (filp->f_flags & O_DIRECT)
173 if (nfsi->npages != 0)
175 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
178 return __nfs_revalidate_inode(server, inode);
181 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
183 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
184 filp->f_path.dentry->d_parent->d_name.name,
185 filp->f_path.dentry->d_name.name,
188 /* origin == SEEK_END => we must revalidate the cached file length */
189 if (origin == SEEK_END) {
190 struct inode *inode = filp->f_mapping->host;
191 int retval = nfs_revalidate_file_size(inode, filp);
193 return (loff_t)retval;
195 return remote_llseek(filp, offset, origin);
199 * Helper for nfs_file_flush() and nfs_file_fsync()
201 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
202 * disk, but it retrieves and clears ctx->error after synching, despite
203 * the two being set at the same time in nfs_context_set_write_error().
204 * This is because the former is used to notify the _next_ call to
205 * nfs_file_write() that a write error occured, and hence cause it to
206 * fall back to doing a synchronous write.
208 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
210 int have_error, status;
213 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
214 status = nfs_wb_all(inode);
215 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
217 ret = xchg(&ctx->error, 0);
224 * Flush all dirty pages, and check for write errors.
227 nfs_file_flush(struct file *file, fl_owner_t id)
229 struct nfs_open_context *ctx = nfs_file_open_context(file);
230 struct dentry *dentry = file->f_path.dentry;
231 struct inode *inode = dentry->d_inode;
234 dprintk("NFS: flush(%s/%s)\n",
235 dentry->d_parent->d_name.name,
236 dentry->d_name.name);
238 if ((file->f_mode & FMODE_WRITE) == 0)
240 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
242 /* Ensure that data+attribute caches are up to date after close() */
243 status = nfs_do_fsync(ctx, inode);
245 nfs_revalidate_inode(NFS_SERVER(inode), inode);
250 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
251 unsigned long nr_segs, loff_t pos)
253 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
254 struct inode * inode = dentry->d_inode;
256 size_t count = iov_length(iov, nr_segs);
258 if (iocb->ki_filp->f_flags & O_DIRECT)
259 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
261 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
262 dentry->d_parent->d_name.name, dentry->d_name.name,
263 (unsigned long) count, (unsigned long) pos);
265 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
266 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
268 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
273 nfs_file_splice_read(struct file *filp, loff_t *ppos,
274 struct pipe_inode_info *pipe, size_t count,
277 struct dentry *dentry = filp->f_path.dentry;
278 struct inode *inode = dentry->d_inode;
281 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
282 dentry->d_parent->d_name.name, dentry->d_name.name,
283 (unsigned long) count, (unsigned long long) *ppos);
285 res = nfs_revalidate_mapping(inode, filp->f_mapping);
287 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
292 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
294 struct dentry *dentry = file->f_path.dentry;
295 struct inode *inode = dentry->d_inode;
298 dprintk("NFS: mmap(%s/%s)\n",
299 dentry->d_parent->d_name.name, dentry->d_name.name);
301 status = nfs_revalidate_mapping(inode, file->f_mapping);
303 vma->vm_ops = &nfs_file_vm_ops;
304 vma->vm_flags |= VM_CAN_NONLINEAR;
311 * Flush any dirty pages for this process, and check for write errors.
312 * The return status from this call provides a reliable indication of
313 * whether any write errors occurred for this process.
316 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
318 struct nfs_open_context *ctx = nfs_file_open_context(file);
319 struct inode *inode = dentry->d_inode;
321 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
322 dentry->d_parent->d_name.name, dentry->d_name.name,
325 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
326 return nfs_do_fsync(ctx, inode);
330 * This does the "real" work of the write. We must allocate and lock the
331 * page to be sent back to the generic routine, which then copies the
332 * data from user space.
334 * If the writer ends up delaying the write, the writer needs to
335 * increment the page use counts until he is done with the page.
337 static int nfs_write_begin(struct file *file, struct address_space *mapping,
338 loff_t pos, unsigned len, unsigned flags,
339 struct page **pagep, void **fsdata)
344 index = pos >> PAGE_CACHE_SHIFT;
346 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
347 file->f_path.dentry->d_parent->d_name.name,
348 file->f_path.dentry->d_name.name,
349 mapping->host->i_ino, len, (long long) pos);
351 page = __grab_cache_page(mapping, index);
356 ret = nfs_flush_incompatible(file, page);
359 page_cache_release(page);
364 static int nfs_write_end(struct file *file, struct address_space *mapping,
365 loff_t pos, unsigned len, unsigned copied,
366 struct page *page, void *fsdata)
368 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
371 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
372 file->f_path.dentry->d_parent->d_name.name,
373 file->f_path.dentry->d_name.name,
374 mapping->host->i_ino, len, (long long) pos);
377 * Zero any uninitialised parts of the page, and then mark the page
378 * as up to date if it turns out that we're extending the file.
380 if (!PageUptodate(page)) {
381 unsigned pglen = nfs_page_length(page);
382 unsigned end = offset + len;
385 zero_user_segments(page, 0, offset,
386 end, PAGE_CACHE_SIZE);
387 SetPageUptodate(page);
388 } else if (end >= pglen) {
389 zero_user_segment(page, end, PAGE_CACHE_SIZE);
391 SetPageUptodate(page);
393 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
397 status = nfs_updatepage(file, page, offset, copied);
401 page_cache_release(page);
408 static void nfs_invalidate_page(struct page *page, unsigned long offset)
410 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
414 /* Cancel any unstarted writes on this page */
415 nfs_wb_page_cancel(page->mapping->host, page);
418 static int nfs_release_page(struct page *page, gfp_t gfp)
420 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
422 /* If PagePrivate() is set, then the page is not freeable */
426 static int nfs_launder_page(struct page *page)
428 struct inode *inode = page->mapping->host;
430 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
431 inode->i_ino, (long long)page_offset(page));
433 return nfs_wb_page(inode, page);
436 const struct address_space_operations nfs_file_aops = {
437 .readpage = nfs_readpage,
438 .readpages = nfs_readpages,
439 .set_page_dirty = __set_page_dirty_nobuffers,
440 .writepage = nfs_writepage,
441 .writepages = nfs_writepages,
442 .write_begin = nfs_write_begin,
443 .write_end = nfs_write_end,
444 .invalidatepage = nfs_invalidate_page,
445 .releasepage = nfs_release_page,
446 .direct_IO = nfs_direct_IO,
447 .launder_page = nfs_launder_page,
450 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
452 struct file *filp = vma->vm_file;
453 struct dentry *dentry = filp->f_path.dentry;
456 struct address_space *mapping;
458 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
459 dentry->d_parent->d_name.name, dentry->d_name.name,
460 filp->f_mapping->host->i_ino,
461 (long long)page_offset(page));
464 mapping = page->mapping;
465 if (mapping != dentry->d_inode->i_mapping)
469 pagelen = nfs_page_length(page);
473 ret = nfs_flush_incompatible(filp, page);
477 ret = nfs_updatepage(filp, page, 0, pagelen);
485 static struct vm_operations_struct nfs_file_vm_ops = {
486 .fault = filemap_fault,
487 .page_mkwrite = nfs_vm_page_mkwrite,
490 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
492 struct nfs_open_context *ctx;
494 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
496 ctx = nfs_file_open_context(filp);
497 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
502 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
503 unsigned long nr_segs, loff_t pos)
505 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
506 struct inode * inode = dentry->d_inode;
508 size_t count = iov_length(iov, nr_segs);
510 if (iocb->ki_filp->f_flags & O_DIRECT)
511 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
513 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
514 dentry->d_parent->d_name.name, dentry->d_name.name,
515 (unsigned long) count, (long long) pos);
518 if (IS_SWAPFILE(inode))
521 * O_APPEND implies that we must revalidate the file length.
523 if (iocb->ki_filp->f_flags & O_APPEND) {
524 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
533 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
534 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
535 /* Return error values for O_SYNC and IS_SYNC() */
536 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
537 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
545 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
549 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
551 struct inode *inode = filp->f_mapping->host;
555 /* Try local locking first */
556 posix_test_lock(filp, fl);
557 if (fl->fl_type != F_UNLCK) {
558 /* found a conflict */
562 if (nfs_have_delegation(inode, FMODE_READ))
565 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
568 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
573 fl->fl_type = F_UNLCK;
577 static int do_vfs_lock(struct file *file, struct file_lock *fl)
580 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
582 res = posix_lock_file_wait(file, fl);
585 res = flock_lock_file_wait(file, fl);
591 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
597 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
599 struct inode *inode = filp->f_mapping->host;
603 * Flush all pending writes before doing anything
606 nfs_sync_mapping(filp->f_mapping);
608 /* NOTE: special case
609 * If we're signalled while cleaning up locks on process exit, we
610 * still need to complete the unlock.
613 /* Use local locking if mounted with "-onolock" */
614 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
615 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
617 status = do_vfs_lock(filp, fl);
622 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
624 struct inode *inode = filp->f_mapping->host;
628 * Flush all pending writes before doing anything
631 status = nfs_sync_mapping(filp->f_mapping);
636 /* Use local locking if mounted with "-onolock" */
637 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
638 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
640 status = do_vfs_lock(filp, fl);
645 * Make sure we clear the cache whenever we try to get the lock.
646 * This makes locking act as a cache coherency point.
648 nfs_sync_mapping(filp->f_mapping);
649 if (!nfs_have_delegation(inode, FMODE_READ))
650 nfs_zap_caches(inode);
656 * Lock a (portion of) a file
658 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
660 struct inode *inode = filp->f_mapping->host;
663 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
664 filp->f_path.dentry->d_parent->d_name.name,
665 filp->f_path.dentry->d_name.name,
666 fl->fl_type, fl->fl_flags,
667 (long long)fl->fl_start, (long long)fl->fl_end);
669 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
671 /* No mandatory locks over NFS */
672 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
675 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
676 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
682 ret = do_getlk(filp, cmd, fl);
683 else if (fl->fl_type == F_UNLCK)
684 ret = do_unlk(filp, cmd, fl);
686 ret = do_setlk(filp, cmd, fl);
692 * Lock a (portion of) a file
694 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
696 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
697 filp->f_path.dentry->d_parent->d_name.name,
698 filp->f_path.dentry->d_name.name,
699 fl->fl_type, fl->fl_flags);
702 * No BSD flocks over NFS allowed.
703 * Note: we could try to fake a POSIX lock request here by
704 * using ((u32) filp | 0x80000000) or some such as the pid.
705 * Not sure whether that would be unique, though, or whether
706 * that would break in other places.
708 if (!(fl->fl_flags & FL_FLOCK))
711 /* We're simulating flock() locks using posix locks on the server */
712 fl->fl_owner = (fl_owner_t)filp;
714 fl->fl_end = OFFSET_MAX;
716 if (fl->fl_type == F_UNLCK)
717 return do_unlk(filp, cmd, fl);
718 return do_setlk(filp, cmd, fl);
722 * There is no protocol support for leases, so we have no way to implement
723 * them correctly in the face of opens by other clients.
725 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
727 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
728 file->f_path.dentry->d_parent->d_name.name,
729 file->f_path.dentry->d_name.name, arg);