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 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
141 return nfs_release(inode, filp);
145 * nfs_revalidate_size - Revalidate the file size
146 * @inode - pointer to inode struct
147 * @file - pointer to struct file
149 * Revalidates the file length. This is basically a wrapper around
150 * nfs_revalidate_inode() that takes into account the fact that we may
151 * have cached writes (in which case we don't care about the server's
152 * idea of what the file length is), or O_DIRECT (in which case we
153 * shouldn't trust the cache).
155 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
157 struct nfs_server *server = NFS_SERVER(inode);
158 struct nfs_inode *nfsi = NFS_I(inode);
160 if (server->flags & NFS_MOUNT_NOAC)
162 if (filp->f_flags & O_DIRECT)
164 if (nfsi->npages != 0)
166 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
169 return __nfs_revalidate_inode(server, inode);
172 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
176 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
177 filp->f_path.dentry->d_parent->d_name.name,
178 filp->f_path.dentry->d_name.name,
181 /* origin == SEEK_END => we must revalidate the cached file length */
182 if (origin == SEEK_END) {
183 struct inode *inode = filp->f_mapping->host;
185 int retval = nfs_revalidate_file_size(inode, filp);
187 return (loff_t)retval;
189 spin_lock(&inode->i_lock);
190 loff = generic_file_llseek_unlocked(filp, offset, origin);
191 spin_unlock(&inode->i_lock);
193 loff = generic_file_llseek_unlocked(filp, offset, origin);
198 * Helper for nfs_file_flush() and nfs_file_fsync()
200 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
201 * disk, but it retrieves and clears ctx->error after synching, despite
202 * the two being set at the same time in nfs_context_set_write_error().
203 * This is because the former is used to notify the _next_ call to
204 * nfs_file_write() that a write error occured, and hence cause it to
205 * fall back to doing a synchronous write.
207 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
209 int have_error, status;
212 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
213 status = nfs_wb_all(inode);
214 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
216 ret = xchg(&ctx->error, 0);
223 * Flush all dirty pages, and check for write errors.
226 nfs_file_flush(struct file *file, fl_owner_t id)
228 struct nfs_open_context *ctx = nfs_file_open_context(file);
229 struct dentry *dentry = file->f_path.dentry;
230 struct inode *inode = dentry->d_inode;
232 dprintk("NFS: flush(%s/%s)\n",
233 dentry->d_parent->d_name.name,
234 dentry->d_name.name);
236 if ((file->f_mode & FMODE_WRITE) == 0)
238 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
240 /* Flush writes to the server and return any errors */
241 return nfs_do_fsync(ctx, inode);
245 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
246 unsigned long nr_segs, loff_t pos)
248 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
249 struct inode * inode = dentry->d_inode;
251 size_t count = iov_length(iov, nr_segs);
253 if (iocb->ki_filp->f_flags & O_DIRECT)
254 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
256 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
257 dentry->d_parent->d_name.name, dentry->d_name.name,
258 (unsigned long) count, (unsigned long) pos);
260 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
261 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
263 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
268 nfs_file_splice_read(struct file *filp, loff_t *ppos,
269 struct pipe_inode_info *pipe, size_t count,
272 struct dentry *dentry = filp->f_path.dentry;
273 struct inode *inode = dentry->d_inode;
276 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
277 dentry->d_parent->d_name.name, dentry->d_name.name,
278 (unsigned long) count, (unsigned long long) *ppos);
280 res = nfs_revalidate_mapping(inode, filp->f_mapping);
282 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
287 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
289 struct dentry *dentry = file->f_path.dentry;
290 struct inode *inode = dentry->d_inode;
293 dprintk("NFS: mmap(%s/%s)\n",
294 dentry->d_parent->d_name.name, dentry->d_name.name);
296 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
297 * so we call that before revalidating the mapping
299 status = generic_file_mmap(file, vma);
301 vma->vm_ops = &nfs_file_vm_ops;
302 status = nfs_revalidate_mapping(inode, file->f_mapping);
308 * Flush any dirty pages for this process, and check for write errors.
309 * The return status from this call provides a reliable indication of
310 * whether any write errors occurred for this process.
313 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
315 struct nfs_open_context *ctx = nfs_file_open_context(file);
316 struct inode *inode = dentry->d_inode;
318 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
319 dentry->d_parent->d_name.name, dentry->d_name.name,
322 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
323 return nfs_do_fsync(ctx, inode);
327 * This does the "real" work of the write. We must allocate and lock the
328 * page to be sent back to the generic routine, which then copies the
329 * data from user space.
331 * If the writer ends up delaying the write, the writer needs to
332 * increment the page use counts until he is done with the page.
334 static int nfs_write_begin(struct file *file, struct address_space *mapping,
335 loff_t pos, unsigned len, unsigned flags,
336 struct page **pagep, void **fsdata)
341 index = pos >> PAGE_CACHE_SHIFT;
343 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
344 file->f_path.dentry->d_parent->d_name.name,
345 file->f_path.dentry->d_name.name,
346 mapping->host->i_ino, len, (long long) pos);
349 * Prevent starvation issues if someone is doing a consistency
352 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
353 nfs_wait_bit_killable, TASK_KILLABLE);
357 page = grab_cache_page_write_begin(mapping, index, flags);
362 ret = nfs_flush_incompatible(file, page);
365 page_cache_release(page);
370 static int nfs_write_end(struct file *file, struct address_space *mapping,
371 loff_t pos, unsigned len, unsigned copied,
372 struct page *page, void *fsdata)
374 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
377 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
378 file->f_path.dentry->d_parent->d_name.name,
379 file->f_path.dentry->d_name.name,
380 mapping->host->i_ino, len, (long long) pos);
383 * Zero any uninitialised parts of the page, and then mark the page
384 * as up to date if it turns out that we're extending the file.
386 if (!PageUptodate(page)) {
387 unsigned pglen = nfs_page_length(page);
388 unsigned end = offset + len;
391 zero_user_segments(page, 0, offset,
392 end, PAGE_CACHE_SIZE);
393 SetPageUptodate(page);
394 } else if (end >= pglen) {
395 zero_user_segment(page, end, PAGE_CACHE_SIZE);
397 SetPageUptodate(page);
399 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
402 status = nfs_updatepage(file, page, offset, copied);
405 page_cache_release(page);
413 * Partially or wholly invalidate a page
414 * - Release the private state associated with a page if undergoing complete
416 * - Called if either PG_private or PG_private_2 is set on the page
417 * - Caller holds page lock
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);
430 * Attempt to release the private state associated with a page
431 * - Called if either PG_private or PG_private_2 is set on the page
432 * - Caller holds page lock
433 * - Return true (may release page) or false (may not)
435 static int nfs_release_page(struct page *page, gfp_t gfp)
437 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
439 /* If PagePrivate() is set, then the page is not freeable */
444 * Attempt to clear the private state associated with a page when an error
445 * occurs that requires the cached contents of an inode to be written back or
447 * - Called if either PG_private or PG_private_2 is set on the page
448 * - Caller holds page lock
449 * - Return 0 if successful, -error otherwise
451 static int nfs_launder_page(struct page *page)
453 struct inode *inode = page->mapping->host;
455 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
456 inode->i_ino, (long long)page_offset(page));
458 return nfs_wb_page(inode, page);
461 const struct address_space_operations nfs_file_aops = {
462 .readpage = nfs_readpage,
463 .readpages = nfs_readpages,
464 .set_page_dirty = __set_page_dirty_nobuffers,
465 .writepage = nfs_writepage,
466 .writepages = nfs_writepages,
467 .write_begin = nfs_write_begin,
468 .write_end = nfs_write_end,
469 .invalidatepage = nfs_invalidate_page,
470 .releasepage = nfs_release_page,
471 .direct_IO = nfs_direct_IO,
472 .launder_page = nfs_launder_page,
476 * Notification that a PTE pointing to an NFS page is about to be made
477 * writable, implying that someone is about to modify the page through a
478 * shared-writable mapping
480 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
482 struct page *page = vmf->page;
483 struct file *filp = vma->vm_file;
484 struct dentry *dentry = filp->f_path.dentry;
487 struct address_space *mapping;
489 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
490 dentry->d_parent->d_name.name, dentry->d_name.name,
491 filp->f_mapping->host->i_ino,
492 (long long)page_offset(page));
495 mapping = page->mapping;
496 if (mapping != dentry->d_inode->i_mapping)
500 pagelen = nfs_page_length(page);
504 ret = nfs_flush_incompatible(filp, page);
508 ret = nfs_updatepage(filp, page, 0, pagelen);
514 ret = VM_FAULT_SIGBUS;
518 static struct vm_operations_struct nfs_file_vm_ops = {
519 .fault = filemap_fault,
520 .page_mkwrite = nfs_vm_page_mkwrite,
523 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
525 struct nfs_open_context *ctx;
527 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
529 ctx = nfs_file_open_context(filp);
530 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
535 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
536 unsigned long nr_segs, loff_t pos)
538 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
539 struct inode * inode = dentry->d_inode;
541 size_t count = iov_length(iov, nr_segs);
543 if (iocb->ki_filp->f_flags & O_DIRECT)
544 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
546 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
547 dentry->d_parent->d_name.name, dentry->d_name.name,
548 (unsigned long) count, (long long) pos);
551 if (IS_SWAPFILE(inode))
554 * O_APPEND implies that we must revalidate the file length.
556 if (iocb->ki_filp->f_flags & O_APPEND) {
557 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
566 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
567 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
568 /* Return error values for O_SYNC and IS_SYNC() */
569 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
570 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
578 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
582 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
584 struct inode *inode = filp->f_mapping->host;
588 /* Try local locking first */
589 posix_test_lock(filp, fl);
590 if (fl->fl_type != F_UNLCK) {
591 /* found a conflict */
595 if (nfs_have_delegation(inode, FMODE_READ))
598 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
601 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
606 fl->fl_type = F_UNLCK;
610 static int do_vfs_lock(struct file *file, struct file_lock *fl)
613 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
615 res = posix_lock_file_wait(file, fl);
618 res = flock_lock_file_wait(file, fl);
624 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
630 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
632 struct inode *inode = filp->f_mapping->host;
636 * Flush all pending writes before doing anything
639 nfs_sync_mapping(filp->f_mapping);
641 /* NOTE: special case
642 * If we're signalled while cleaning up locks on process exit, we
643 * still need to complete the unlock.
646 /* Use local locking if mounted with "-onolock" */
647 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
648 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
650 status = do_vfs_lock(filp, fl);
655 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
657 struct inode *inode = filp->f_mapping->host;
661 * Flush all pending writes before doing anything
664 status = nfs_sync_mapping(filp->f_mapping);
669 /* Use local locking if mounted with "-onolock" */
670 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
671 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
673 status = do_vfs_lock(filp, fl);
678 * Make sure we clear the cache whenever we try to get the lock.
679 * This makes locking act as a cache coherency point.
681 nfs_sync_mapping(filp->f_mapping);
682 if (!nfs_have_delegation(inode, FMODE_READ))
683 nfs_zap_caches(inode);
689 * Lock a (portion of) a file
691 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
693 struct inode *inode = filp->f_mapping->host;
696 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\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,
700 (long long)fl->fl_start, (long long)fl->fl_end);
702 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
704 /* No mandatory locks over NFS */
705 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
708 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
709 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
715 ret = do_getlk(filp, cmd, fl);
716 else if (fl->fl_type == F_UNLCK)
717 ret = do_unlk(filp, cmd, fl);
719 ret = do_setlk(filp, cmd, fl);
725 * Lock a (portion of) a file
727 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
729 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\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);
734 if (!(fl->fl_flags & FL_FLOCK))
737 /* We're simulating flock() locks using posix locks on the server */
738 fl->fl_owner = (fl_owner_t)filp;
740 fl->fl_end = OFFSET_MAX;
742 if (fl->fl_type == F_UNLCK)
743 return do_unlk(filp, cmd, fl);
744 return do_setlk(filp, cmd, fl);
748 * There is no protocol support for leases, so we have no way to implement
749 * them correctly in the face of opens by other clients.
751 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
753 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
754 file->f_path.dentry->d_parent->d_name.name,
755 file->f_path.dentry->d_name.name, arg);