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_splice_write(struct pipe_inode_info *pipe,
52 struct file *filp, loff_t *ppos,
53 size_t count, unsigned int flags);
54 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
55 unsigned long nr_segs, loff_t pos);
56 static int nfs_file_flush(struct file *, fl_owner_t id);
57 static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
58 static int nfs_check_flags(int flags);
59 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
60 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
61 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
63 static struct vm_operations_struct nfs_file_vm_ops;
65 const struct file_operations nfs_file_operations = {
66 .llseek = nfs_file_llseek,
68 .write = do_sync_write,
69 .aio_read = nfs_file_read,
70 .aio_write = nfs_file_write,
71 .mmap = nfs_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 .splice_write = nfs_file_splice_write,
80 .check_flags = nfs_check_flags,
81 .setlease = nfs_setlease,
84 const struct inode_operations nfs_file_inode_operations = {
85 .permission = nfs_permission,
86 .getattr = nfs_getattr,
87 .setattr = nfs_setattr,
91 const struct inode_operations nfs3_file_inode_operations = {
92 .permission = nfs_permission,
93 .getattr = nfs_getattr,
94 .setattr = nfs_setattr,
95 .listxattr = nfs3_listxattr,
96 .getxattr = nfs3_getxattr,
97 .setxattr = nfs3_setxattr,
98 .removexattr = nfs3_removexattr,
100 #endif /* CONFIG_NFS_v3 */
102 /* Hack for future NFS swap support */
104 # define IS_SWAPFILE(inode) (0)
107 static int nfs_check_flags(int flags)
109 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
119 nfs_file_open(struct inode *inode, struct file *filp)
123 dprintk("NFS: open file(%s/%s)\n",
124 filp->f_path.dentry->d_parent->d_name.name,
125 filp->f_path.dentry->d_name.name);
127 res = nfs_check_flags(filp->f_flags);
131 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
132 res = nfs_open(inode, filp);
137 nfs_file_release(struct inode *inode, struct file *filp)
139 struct dentry *dentry = filp->f_path.dentry;
141 dprintk("NFS: release(%s/%s)\n",
142 dentry->d_parent->d_name.name,
143 dentry->d_name.name);
145 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
146 return nfs_release(inode, filp);
150 * nfs_revalidate_size - Revalidate the file size
151 * @inode - pointer to inode struct
152 * @file - pointer to struct file
154 * Revalidates the file length. This is basically a wrapper around
155 * nfs_revalidate_inode() that takes into account the fact that we may
156 * have cached writes (in which case we don't care about the server's
157 * idea of what the file length is), or O_DIRECT (in which case we
158 * shouldn't trust the cache).
160 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
162 struct nfs_server *server = NFS_SERVER(inode);
163 struct nfs_inode *nfsi = NFS_I(inode);
165 if (server->flags & NFS_MOUNT_NOAC)
167 if (filp->f_flags & O_DIRECT)
169 if (nfsi->npages != 0)
171 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
174 return __nfs_revalidate_inode(server, inode);
177 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
181 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
182 filp->f_path.dentry->d_parent->d_name.name,
183 filp->f_path.dentry->d_name.name,
186 /* origin == SEEK_END => we must revalidate the cached file length */
187 if (origin == SEEK_END) {
188 struct inode *inode = filp->f_mapping->host;
190 int retval = nfs_revalidate_file_size(inode, filp);
192 return (loff_t)retval;
194 spin_lock(&inode->i_lock);
195 loff = generic_file_llseek_unlocked(filp, offset, origin);
196 spin_unlock(&inode->i_lock);
198 loff = generic_file_llseek_unlocked(filp, offset, origin);
203 * Helper for nfs_file_flush() and nfs_file_fsync()
205 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
206 * disk, but it retrieves and clears ctx->error after synching, despite
207 * the two being set at the same time in nfs_context_set_write_error().
208 * This is because the former is used to notify the _next_ call to
209 * nfs_file_write() that a write error occured, and hence cause it to
210 * fall back to doing a synchronous write.
212 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
214 int have_error, status;
217 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
218 status = nfs_wb_all(inode);
219 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
221 ret = xchg(&ctx->error, 0);
228 * Flush all dirty pages, and check for write errors.
231 nfs_file_flush(struct file *file, fl_owner_t id)
233 struct nfs_open_context *ctx = nfs_file_open_context(file);
234 struct dentry *dentry = file->f_path.dentry;
235 struct inode *inode = dentry->d_inode;
237 dprintk("NFS: flush(%s/%s)\n",
238 dentry->d_parent->d_name.name,
239 dentry->d_name.name);
241 if ((file->f_mode & FMODE_WRITE) == 0)
243 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
245 /* Flush writes to the server and return any errors */
246 return nfs_do_fsync(ctx, 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 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
302 * so we call that before revalidating the mapping
304 status = generic_file_mmap(file, vma);
306 vma->vm_ops = &nfs_file_vm_ops;
307 status = nfs_revalidate_mapping(inode, file->f_mapping);
313 * Flush any dirty pages for this process, and check for write errors.
314 * The return status from this call provides a reliable indication of
315 * whether any write errors occurred for this process.
318 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
320 struct nfs_open_context *ctx = nfs_file_open_context(file);
321 struct inode *inode = dentry->d_inode;
323 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
324 dentry->d_parent->d_name.name, dentry->d_name.name,
327 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
328 return nfs_do_fsync(ctx, inode);
332 * This does the "real" work of the write. We must allocate and lock the
333 * page to be sent back to the generic routine, which then copies the
334 * data from user space.
336 * If the writer ends up delaying the write, the writer needs to
337 * increment the page use counts until he is done with the page.
339 static int nfs_write_begin(struct file *file, struct address_space *mapping,
340 loff_t pos, unsigned len, unsigned flags,
341 struct page **pagep, void **fsdata)
346 index = pos >> PAGE_CACHE_SHIFT;
348 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
349 file->f_path.dentry->d_parent->d_name.name,
350 file->f_path.dentry->d_name.name,
351 mapping->host->i_ino, len, (long long) pos);
354 * Prevent starvation issues if someone is doing a consistency
357 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
358 nfs_wait_bit_killable, TASK_KILLABLE);
362 page = grab_cache_page_write_begin(mapping, index, flags);
367 ret = nfs_flush_incompatible(file, page);
370 page_cache_release(page);
375 static int nfs_write_end(struct file *file, struct address_space *mapping,
376 loff_t pos, unsigned len, unsigned copied,
377 struct page *page, void *fsdata)
379 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
382 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
383 file->f_path.dentry->d_parent->d_name.name,
384 file->f_path.dentry->d_name.name,
385 mapping->host->i_ino, len, (long long) pos);
388 * Zero any uninitialised parts of the page, and then mark the page
389 * as up to date if it turns out that we're extending the file.
391 if (!PageUptodate(page)) {
392 unsigned pglen = nfs_page_length(page);
393 unsigned end = offset + len;
396 zero_user_segments(page, 0, offset,
397 end, PAGE_CACHE_SIZE);
398 SetPageUptodate(page);
399 } else if (end >= pglen) {
400 zero_user_segment(page, end, PAGE_CACHE_SIZE);
402 SetPageUptodate(page);
404 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
407 status = nfs_updatepage(file, page, offset, copied);
410 page_cache_release(page);
418 * Partially or wholly invalidate a page
419 * - Release the private state associated with a page if undergoing complete
421 * - Called if either PG_private or PG_fscache is set on the page
422 * - Caller holds page lock
424 static void nfs_invalidate_page(struct page *page, unsigned long offset)
426 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
430 /* Cancel any unstarted writes on this page */
431 nfs_wb_page_cancel(page->mapping->host, page);
433 nfs_fscache_invalidate_page(page, page->mapping->host);
437 * Attempt to release the private state associated with a page
438 * - Called if either PG_private or PG_fscache is set on the page
439 * - Caller holds page lock
440 * - Return true (may release page) or false (may not)
442 static int nfs_release_page(struct page *page, gfp_t gfp)
444 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
446 /* If PagePrivate() is set, then the page is not freeable */
447 if (PagePrivate(page))
449 return nfs_fscache_release_page(page, gfp);
453 * Attempt to clear the private state associated with a page when an error
454 * occurs that requires the cached contents of an inode to be written back or
456 * - Called if either PG_private or fscache is set on the page
457 * - Caller holds page lock
458 * - Return 0 if successful, -error otherwise
460 static int nfs_launder_page(struct page *page)
462 struct inode *inode = page->mapping->host;
463 struct nfs_inode *nfsi = NFS_I(inode);
465 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
466 inode->i_ino, (long long)page_offset(page));
468 nfs_fscache_wait_on_page_write(nfsi, page);
469 return nfs_wb_page(inode, page);
472 const struct address_space_operations nfs_file_aops = {
473 .readpage = nfs_readpage,
474 .readpages = nfs_readpages,
475 .set_page_dirty = __set_page_dirty_nobuffers,
476 .writepage = nfs_writepage,
477 .writepages = nfs_writepages,
478 .write_begin = nfs_write_begin,
479 .write_end = nfs_write_end,
480 .invalidatepage = nfs_invalidate_page,
481 .releasepage = nfs_release_page,
482 .direct_IO = nfs_direct_IO,
483 .launder_page = nfs_launder_page,
487 * Notification that a PTE pointing to an NFS page is about to be made
488 * writable, implying that someone is about to modify the page through a
489 * shared-writable mapping
491 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
493 struct page *page = vmf->page;
494 struct file *filp = vma->vm_file;
495 struct dentry *dentry = filp->f_path.dentry;
498 struct address_space *mapping;
500 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
501 dentry->d_parent->d_name.name, dentry->d_name.name,
502 filp->f_mapping->host->i_ino,
503 (long long)page_offset(page));
505 /* make sure the cache has finished storing the page */
506 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
509 mapping = page->mapping;
510 if (mapping != dentry->d_inode->i_mapping)
514 pagelen = nfs_page_length(page);
518 ret = nfs_flush_incompatible(filp, page);
522 ret = nfs_updatepage(filp, page, 0, pagelen);
525 return VM_FAULT_LOCKED;
527 return VM_FAULT_SIGBUS;
530 static struct vm_operations_struct nfs_file_vm_ops = {
531 .fault = filemap_fault,
532 .page_mkwrite = nfs_vm_page_mkwrite,
535 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
537 struct nfs_open_context *ctx;
539 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
541 ctx = nfs_file_open_context(filp);
542 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
547 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
548 unsigned long nr_segs, loff_t pos)
550 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
551 struct inode * inode = dentry->d_inode;
553 size_t count = iov_length(iov, nr_segs);
555 if (iocb->ki_filp->f_flags & O_DIRECT)
556 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
558 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
559 dentry->d_parent->d_name.name, dentry->d_name.name,
560 (unsigned long) count, (long long) pos);
563 if (IS_SWAPFILE(inode))
566 * O_APPEND implies that we must revalidate the file length.
568 if (iocb->ki_filp->f_flags & O_APPEND) {
569 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
578 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
579 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
580 /* Return error values for O_SYNC and IS_SYNC() */
581 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
582 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
590 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
594 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
595 struct file *filp, loff_t *ppos,
596 size_t count, unsigned int flags)
598 struct dentry *dentry = filp->f_path.dentry;
599 struct inode *inode = dentry->d_inode;
602 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
603 dentry->d_parent->d_name.name, dentry->d_name.name,
604 (unsigned long) count, (unsigned long long) *ppos);
607 * The combination of splice and an O_APPEND destination is disallowed.
610 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
612 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
613 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
614 int err = nfs_do_fsync(nfs_file_open_context(filp), inode);
621 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
623 struct inode *inode = filp->f_mapping->host;
626 /* Try local locking first */
627 posix_test_lock(filp, fl);
628 if (fl->fl_type != F_UNLCK) {
629 /* found a conflict */
633 if (nfs_have_delegation(inode, FMODE_READ))
636 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
639 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
643 fl->fl_type = F_UNLCK;
647 static int do_vfs_lock(struct file *file, struct file_lock *fl)
650 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
652 res = posix_lock_file_wait(file, fl);
655 res = flock_lock_file_wait(file, fl);
661 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
667 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
669 struct inode *inode = filp->f_mapping->host;
673 * Flush all pending writes before doing anything
676 nfs_sync_mapping(filp->f_mapping);
678 /* NOTE: special case
679 * If we're signalled while cleaning up locks on process exit, we
680 * still need to complete the unlock.
682 /* Use local locking if mounted with "-onolock" */
683 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
684 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
686 status = do_vfs_lock(filp, fl);
690 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
692 struct inode *inode = filp->f_mapping->host;
696 * Flush all pending writes before doing anything
699 status = nfs_sync_mapping(filp->f_mapping);
703 /* Use local locking if mounted with "-onolock" */
704 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
705 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
707 status = do_vfs_lock(filp, fl);
711 * Make sure we clear the cache whenever we try to get the lock.
712 * This makes locking act as a cache coherency point.
714 nfs_sync_mapping(filp->f_mapping);
715 if (!nfs_have_delegation(inode, FMODE_READ))
716 nfs_zap_caches(inode);
722 * Lock a (portion of) a file
724 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
726 struct inode *inode = filp->f_mapping->host;
729 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
730 filp->f_path.dentry->d_parent->d_name.name,
731 filp->f_path.dentry->d_name.name,
732 fl->fl_type, fl->fl_flags,
733 (long long)fl->fl_start, (long long)fl->fl_end);
735 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
737 /* No mandatory locks over NFS */
738 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
741 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
742 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
748 ret = do_getlk(filp, cmd, fl);
749 else if (fl->fl_type == F_UNLCK)
750 ret = do_unlk(filp, cmd, fl);
752 ret = do_setlk(filp, cmd, fl);
758 * Lock a (portion of) a file
760 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
762 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
763 filp->f_path.dentry->d_parent->d_name.name,
764 filp->f_path.dentry->d_name.name,
765 fl->fl_type, fl->fl_flags);
767 if (!(fl->fl_flags & FL_FLOCK))
770 /* We're simulating flock() locks using posix locks on the server */
771 fl->fl_owner = (fl_owner_t)filp;
773 fl->fl_end = OFFSET_MAX;
775 if (fl->fl_type == F_UNLCK)
776 return do_unlk(filp, cmd, fl);
777 return do_setlk(filp, cmd, fl);
781 * There is no protocol support for leases, so we have no way to implement
782 * them correctly in the face of opens by other clients.
784 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
786 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
787 file->f_path.dentry->d_parent->d_name.name,
788 file->f_path.dentry->d_name.name, arg);