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"
40 #define NFSDBG_FACILITY NFSDBG_FILE
42 static int nfs_file_open(struct inode *, struct file *);
43 static int nfs_file_release(struct inode *, struct file *);
44 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
45 static int nfs_file_mmap(struct file *, struct vm_area_struct *);
46 static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
47 struct pipe_inode_info *pipe,
48 size_t count, unsigned int flags);
49 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
50 unsigned long nr_segs, loff_t pos);
51 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
52 unsigned long nr_segs, loff_t pos);
53 static int nfs_file_flush(struct file *, fl_owner_t id);
54 static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
55 static int nfs_check_flags(int flags);
56 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
57 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
58 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
60 static struct vm_operations_struct nfs_file_vm_ops;
62 const struct file_operations nfs_file_operations = {
63 .llseek = nfs_file_llseek,
65 .write = do_sync_write,
66 .aio_read = nfs_file_read,
67 .aio_write = nfs_file_write,
68 .mmap = nfs_file_mmap,
69 .open = nfs_file_open,
70 .flush = nfs_file_flush,
71 .release = nfs_file_release,
72 .fsync = nfs_file_fsync,
75 .splice_read = nfs_file_splice_read,
76 .check_flags = nfs_check_flags,
77 .setlease = nfs_setlease,
80 const struct inode_operations nfs_file_inode_operations = {
81 .permission = nfs_permission,
82 .getattr = nfs_getattr,
83 .setattr = nfs_setattr,
87 const struct inode_operations nfs3_file_inode_operations = {
88 .permission = nfs_permission,
89 .getattr = nfs_getattr,
90 .setattr = nfs_setattr,
91 .listxattr = nfs3_listxattr,
92 .getxattr = nfs3_getxattr,
93 .setxattr = nfs3_setxattr,
94 .removexattr = nfs3_removexattr,
96 #endif /* CONFIG_NFS_v3 */
98 /* Hack for future NFS swap support */
100 # define IS_SWAPFILE(inode) (0)
103 static int nfs_check_flags(int flags)
105 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
115 nfs_file_open(struct inode *inode, struct file *filp)
119 dprintk("NFS: open file(%s/%s)\n",
120 filp->f_path.dentry->d_parent->d_name.name,
121 filp->f_path.dentry->d_name.name);
123 res = nfs_check_flags(filp->f_flags);
127 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
128 res = nfs_open(inode, filp);
133 nfs_file_release(struct inode *inode, struct file *filp)
135 struct dentry *dentry = filp->f_path.dentry;
137 dprintk("NFS: release(%s/%s)\n",
138 dentry->d_parent->d_name.name,
139 dentry->d_name.name);
141 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
142 return nfs_release(inode, filp);
146 * nfs_revalidate_size - Revalidate the file size
147 * @inode - pointer to inode struct
148 * @file - pointer to struct file
150 * Revalidates the file length. This is basically a wrapper around
151 * nfs_revalidate_inode() that takes into account the fact that we may
152 * have cached writes (in which case we don't care about the server's
153 * idea of what the file length is), or O_DIRECT (in which case we
154 * shouldn't trust the cache).
156 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
158 struct nfs_server *server = NFS_SERVER(inode);
159 struct nfs_inode *nfsi = NFS_I(inode);
161 if (server->flags & NFS_MOUNT_NOAC)
163 if (filp->f_flags & O_DIRECT)
165 if (nfsi->npages != 0)
167 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
170 return __nfs_revalidate_inode(server, inode);
173 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
177 dprintk("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;
186 int retval = nfs_revalidate_file_size(inode, filp);
188 return (loff_t)retval;
190 spin_lock(&inode->i_lock);
191 loff = generic_file_llseek_unlocked(filp, offset, origin);
192 spin_unlock(&inode->i_lock);
194 loff = generic_file_llseek_unlocked(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;
233 dprintk("NFS: flush(%s/%s)\n",
234 dentry->d_parent->d_name.name,
235 dentry->d_name.name);
237 if ((file->f_mode & FMODE_WRITE) == 0)
239 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
241 /* Flush writes to the server and return any errors */
242 return nfs_do_fsync(ctx, inode);
246 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
247 unsigned long nr_segs, loff_t pos)
249 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
250 struct inode * inode = dentry->d_inode;
252 size_t count = iov_length(iov, nr_segs);
254 if (iocb->ki_filp->f_flags & O_DIRECT)
255 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
257 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
258 dentry->d_parent->d_name.name, dentry->d_name.name,
259 (unsigned long) count, (unsigned long) pos);
261 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
262 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
264 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
269 nfs_file_splice_read(struct file *filp, loff_t *ppos,
270 struct pipe_inode_info *pipe, size_t count,
273 struct dentry *dentry = filp->f_path.dentry;
274 struct inode *inode = dentry->d_inode;
277 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
278 dentry->d_parent->d_name.name, dentry->d_name.name,
279 (unsigned long) count, (unsigned long long) *ppos);
281 res = nfs_revalidate_mapping(inode, filp->f_mapping);
283 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
288 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
290 struct dentry *dentry = file->f_path.dentry;
291 struct inode *inode = dentry->d_inode;
294 dprintk("NFS: mmap(%s/%s)\n",
295 dentry->d_parent->d_name.name, dentry->d_name.name);
297 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
298 * so we call that before revalidating the mapping
300 status = generic_file_mmap(file, vma);
302 vma->vm_ops = &nfs_file_vm_ops;
303 status = nfs_revalidate_mapping(inode, file->f_mapping);
309 * Flush any dirty pages for this process, and check for write errors.
310 * The return status from this call provides a reliable indication of
311 * whether any write errors occurred for this process.
314 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
316 struct nfs_open_context *ctx = nfs_file_open_context(file);
317 struct inode *inode = dentry->d_inode;
319 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
320 dentry->d_parent->d_name.name, dentry->d_name.name,
323 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
324 return nfs_do_fsync(ctx, inode);
328 * This does the "real" work of the write. We must allocate and lock the
329 * page to be sent back to the generic routine, which then copies the
330 * data from user space.
332 * If the writer ends up delaying the write, the writer needs to
333 * increment the page use counts until he is done with the page.
335 static int nfs_write_begin(struct file *file, struct address_space *mapping,
336 loff_t pos, unsigned len, unsigned flags,
337 struct page **pagep, void **fsdata)
342 index = pos >> PAGE_CACHE_SHIFT;
344 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
345 file->f_path.dentry->d_parent->d_name.name,
346 file->f_path.dentry->d_name.name,
347 mapping->host->i_ino, len, (long long) pos);
350 * Prevent starvation issues if someone is doing a consistency
353 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
354 nfs_wait_bit_killable, TASK_KILLABLE);
358 page = grab_cache_page_write_begin(mapping, index, flags);
363 ret = nfs_flush_incompatible(file, page);
366 page_cache_release(page);
371 static int nfs_write_end(struct file *file, struct address_space *mapping,
372 loff_t pos, unsigned len, unsigned copied,
373 struct page *page, void *fsdata)
375 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
378 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
379 file->f_path.dentry->d_parent->d_name.name,
380 file->f_path.dentry->d_name.name,
381 mapping->host->i_ino, len, (long long) pos);
384 * Zero any uninitialised parts of the page, and then mark the page
385 * as up to date if it turns out that we're extending the file.
387 if (!PageUptodate(page)) {
388 unsigned pglen = nfs_page_length(page);
389 unsigned end = offset + len;
392 zero_user_segments(page, 0, offset,
393 end, PAGE_CACHE_SIZE);
394 SetPageUptodate(page);
395 } else if (end >= pglen) {
396 zero_user_segment(page, end, PAGE_CACHE_SIZE);
398 SetPageUptodate(page);
400 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
403 status = nfs_updatepage(file, page, offset, copied);
406 page_cache_release(page);
414 * Partially or wholly invalidate a page
415 * - Release the private state associated with a page if undergoing complete
417 * - Called if either PG_private or PG_fscache is set on the page
418 * - Caller holds page lock
420 static void nfs_invalidate_page(struct page *page, unsigned long offset)
422 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
426 /* Cancel any unstarted writes on this page */
427 nfs_wb_page_cancel(page->mapping->host, page);
429 nfs_fscache_invalidate_page(page, page->mapping->host);
433 * Attempt to release the private state associated with a page
434 * - Called if either PG_private or PG_fscache is set on the page
435 * - Caller holds page lock
436 * - Return true (may release page) or false (may not)
438 static int nfs_release_page(struct page *page, gfp_t gfp)
440 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
442 /* If PagePrivate() is set, then the page is not freeable */
443 if (PagePrivate(page))
445 return nfs_fscache_release_page(page, gfp);
449 * Attempt to clear the private state associated with a page when an error
450 * occurs that requires the cached contents of an inode to be written back or
452 * - Called if either PG_private or fscache is set on the page
453 * - Caller holds page lock
454 * - Return 0 if successful, -error otherwise
456 static int nfs_launder_page(struct page *page)
458 struct inode *inode = page->mapping->host;
459 struct nfs_inode *nfsi = NFS_I(inode);
461 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
462 inode->i_ino, (long long)page_offset(page));
464 nfs_fscache_wait_on_page_write(nfsi, page);
465 return nfs_wb_page(inode, page);
468 const struct address_space_operations nfs_file_aops = {
469 .readpage = nfs_readpage,
470 .readpages = nfs_readpages,
471 .set_page_dirty = __set_page_dirty_nobuffers,
472 .writepage = nfs_writepage,
473 .writepages = nfs_writepages,
474 .write_begin = nfs_write_begin,
475 .write_end = nfs_write_end,
476 .invalidatepage = nfs_invalidate_page,
477 .releasepage = nfs_release_page,
478 .direct_IO = nfs_direct_IO,
479 .launder_page = nfs_launder_page,
483 * Notification that a PTE pointing to an NFS page is about to be made
484 * writable, implying that someone is about to modify the page through a
485 * shared-writable mapping
487 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
489 struct page *page = vmf->page;
490 struct file *filp = vma->vm_file;
491 struct dentry *dentry = filp->f_path.dentry;
494 struct address_space *mapping;
496 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
497 dentry->d_parent->d_name.name, dentry->d_name.name,
498 filp->f_mapping->host->i_ino,
499 (long long)page_offset(page));
501 /* make sure the cache has finished storing the page */
502 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
505 mapping = page->mapping;
506 if (mapping != dentry->d_inode->i_mapping)
510 pagelen = nfs_page_length(page);
514 ret = nfs_flush_incompatible(filp, page);
518 ret = nfs_updatepage(filp, page, 0, pagelen);
521 return VM_FAULT_LOCKED;
523 return VM_FAULT_SIGBUS;
526 static struct vm_operations_struct nfs_file_vm_ops = {
527 .fault = filemap_fault,
528 .page_mkwrite = nfs_vm_page_mkwrite,
531 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
533 struct nfs_open_context *ctx;
535 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
537 ctx = nfs_file_open_context(filp);
538 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
543 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
544 unsigned long nr_segs, loff_t pos)
546 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
547 struct inode * inode = dentry->d_inode;
549 size_t count = iov_length(iov, nr_segs);
551 if (iocb->ki_filp->f_flags & O_DIRECT)
552 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
554 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
555 dentry->d_parent->d_name.name, dentry->d_name.name,
556 (unsigned long) count, (long long) pos);
559 if (IS_SWAPFILE(inode))
562 * O_APPEND implies that we must revalidate the file length.
564 if (iocb->ki_filp->f_flags & O_APPEND) {
565 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
574 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
575 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
576 /* Return error values for O_SYNC and IS_SYNC() */
577 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
578 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
586 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
590 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
592 struct inode *inode = filp->f_mapping->host;
596 /* Try local locking first */
597 posix_test_lock(filp, fl);
598 if (fl->fl_type != F_UNLCK) {
599 /* found a conflict */
603 if (nfs_have_delegation(inode, FMODE_READ))
606 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
609 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
614 fl->fl_type = F_UNLCK;
618 static int do_vfs_lock(struct file *file, struct file_lock *fl)
621 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
623 res = posix_lock_file_wait(file, fl);
626 res = flock_lock_file_wait(file, fl);
632 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
638 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
640 struct inode *inode = filp->f_mapping->host;
644 * Flush all pending writes before doing anything
647 nfs_sync_mapping(filp->f_mapping);
649 /* NOTE: special case
650 * If we're signalled while cleaning up locks on process exit, we
651 * still need to complete the unlock.
654 /* Use local locking if mounted with "-onolock" */
655 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
656 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
658 status = do_vfs_lock(filp, fl);
663 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
665 struct inode *inode = filp->f_mapping->host;
669 * Flush all pending writes before doing anything
672 status = nfs_sync_mapping(filp->f_mapping);
677 /* Use local locking if mounted with "-onolock" */
678 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
679 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
681 status = do_vfs_lock(filp, fl);
686 * Make sure we clear the cache whenever we try to get the lock.
687 * This makes locking act as a cache coherency point.
689 nfs_sync_mapping(filp->f_mapping);
690 if (!nfs_have_delegation(inode, FMODE_READ))
691 nfs_zap_caches(inode);
697 * Lock a (portion of) a file
699 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
701 struct inode *inode = filp->f_mapping->host;
704 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
705 filp->f_path.dentry->d_parent->d_name.name,
706 filp->f_path.dentry->d_name.name,
707 fl->fl_type, fl->fl_flags,
708 (long long)fl->fl_start, (long long)fl->fl_end);
710 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
712 /* No mandatory locks over NFS */
713 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
716 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
717 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
723 ret = do_getlk(filp, cmd, fl);
724 else if (fl->fl_type == F_UNLCK)
725 ret = do_unlk(filp, cmd, fl);
727 ret = do_setlk(filp, cmd, fl);
733 * Lock a (portion of) a file
735 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
737 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
738 filp->f_path.dentry->d_parent->d_name.name,
739 filp->f_path.dentry->d_name.name,
740 fl->fl_type, fl->fl_flags);
742 if (!(fl->fl_flags & FL_FLOCK))
745 /* We're simulating flock() locks using posix locks on the server */
746 fl->fl_owner = (fl_owner_t)filp;
748 fl->fl_end = OFFSET_MAX;
750 if (fl->fl_type == F_UNLCK)
751 return do_unlk(filp, cmd, fl);
752 return do_setlk(filp, cmd, fl);
756 * There is no protocol support for leases, so we have no way to implement
757 * them correctly in the face of opens by other clients.
759 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
761 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
762 file->f_path.dentry->d_parent->d_name.name,
763 file->f_path.dentry->d_name.name, arg);