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
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/pagemap.h>
28 #include <linux/aio.h>
29 #include <linux/gfp.h>
30 #include <linux/swap.h>
32 #include <asm/uaccess.h>
34 #include "delegation.h"
39 #define NFSDBG_FACILITY NFSDBG_FILE
41 static const struct vm_operations_struct nfs_file_vm_ops;
43 /* Hack for future NFS swap support */
45 # define IS_SWAPFILE(inode) (0)
48 int nfs_check_flags(int flags)
50 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
60 nfs_file_open(struct inode *inode, struct file *filp)
64 dprintk("NFS: open file(%s/%s)\n",
65 filp->f_path.dentry->d_parent->d_name.name,
66 filp->f_path.dentry->d_name.name);
68 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
69 res = nfs_check_flags(filp->f_flags);
73 res = nfs_open(inode, filp);
78 nfs_file_release(struct inode *inode, struct file *filp)
80 dprintk("NFS: release(%s/%s)\n",
81 filp->f_path.dentry->d_parent->d_name.name,
82 filp->f_path.dentry->d_name.name);
84 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
85 return nfs_release(inode, filp);
89 * nfs_revalidate_size - Revalidate the file size
90 * @inode - pointer to inode struct
91 * @file - pointer to struct file
93 * Revalidates the file length. This is basically a wrapper around
94 * nfs_revalidate_inode() that takes into account the fact that we may
95 * have cached writes (in which case we don't care about the server's
96 * idea of what the file length is), or O_DIRECT (in which case we
97 * shouldn't trust the cache).
99 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
101 struct nfs_server *server = NFS_SERVER(inode);
102 struct nfs_inode *nfsi = NFS_I(inode);
104 if (nfs_have_delegated_attributes(inode))
107 if (filp->f_flags & O_DIRECT)
109 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
111 if (nfs_attribute_timeout(inode))
116 return __nfs_revalidate_inode(server, inode);
119 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
121 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
122 filp->f_path.dentry->d_parent->d_name.name,
123 filp->f_path.dentry->d_name.name,
127 * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
128 * the cached file length
130 if (origin != SEEK_SET && origin != SEEK_CUR) {
131 struct inode *inode = filp->f_mapping->host;
133 int retval = nfs_revalidate_file_size(inode, filp);
135 return (loff_t)retval;
138 return generic_file_llseek(filp, offset, origin);
142 * Flush all dirty pages, and check for write errors.
145 nfs_file_flush(struct file *file, fl_owner_t id)
147 struct dentry *dentry = file->f_path.dentry;
148 struct inode *inode = dentry->d_inode;
150 dprintk("NFS: flush(%s/%s)\n",
151 dentry->d_parent->d_name.name,
152 dentry->d_name.name);
154 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
155 if ((file->f_mode & FMODE_WRITE) == 0)
159 * If we're holding a write delegation, then just start the i/o
160 * but don't wait for completion (or send a commit).
162 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
163 return filemap_fdatawrite(file->f_mapping);
165 /* Flush writes to the server and return any errors */
166 return vfs_fsync(file, 0);
170 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
171 unsigned long nr_segs, loff_t pos)
173 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
174 struct inode * inode = dentry->d_inode;
177 if (iocb->ki_filp->f_flags & O_DIRECT)
178 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
180 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
181 dentry->d_parent->d_name.name, dentry->d_name.name,
182 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
184 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
186 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
188 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
194 nfs_file_splice_read(struct file *filp, loff_t *ppos,
195 struct pipe_inode_info *pipe, size_t count,
198 struct dentry *dentry = filp->f_path.dentry;
199 struct inode *inode = dentry->d_inode;
202 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
203 dentry->d_parent->d_name.name, dentry->d_name.name,
204 (unsigned long) count, (unsigned long long) *ppos);
206 res = nfs_revalidate_mapping(inode, filp->f_mapping);
208 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
210 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
216 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
218 struct dentry *dentry = file->f_path.dentry;
219 struct inode *inode = dentry->d_inode;
222 dprintk("NFS: mmap(%s/%s)\n",
223 dentry->d_parent->d_name.name, dentry->d_name.name);
225 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
226 * so we call that before revalidating the mapping
228 status = generic_file_mmap(file, vma);
230 vma->vm_ops = &nfs_file_vm_ops;
231 status = nfs_revalidate_mapping(inode, file->f_mapping);
237 * Flush any dirty pages for this process, and check for write errors.
238 * The return status from this call provides a reliable indication of
239 * whether any write errors occurred for this process.
241 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
242 * disk, but it retrieves and clears ctx->error after synching, despite
243 * the two being set at the same time in nfs_context_set_write_error().
244 * This is because the former is used to notify the _next_ call to
245 * nfs_file_write() that a write error occurred, and hence cause it to
246 * fall back to doing a synchronous write.
249 nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
251 struct dentry *dentry = file->f_path.dentry;
252 struct nfs_open_context *ctx = nfs_file_open_context(file);
253 struct inode *inode = dentry->d_inode;
254 int have_error, status;
257 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
258 dentry->d_parent->d_name.name, dentry->d_name.name,
261 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
262 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
263 status = nfs_commit_inode(inode, FLUSH_SYNC);
264 if (status >= 0 && ret < 0)
266 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
268 ret = xchg(&ctx->error, 0);
269 if (!ret && status < 0)
275 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
278 struct inode *inode = file->f_path.dentry->d_inode;
280 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
281 mutex_lock(&inode->i_mutex);
282 ret = nfs_file_fsync_commit(file, start, end, datasync);
283 mutex_unlock(&inode->i_mutex);
289 * Decide whether a read/modify/write cycle may be more efficient
290 * then a modify/write/read cycle when writing to a page in the
293 * The modify/write/read cycle may occur if a page is read before
294 * being completely filled by the writer. In this situation, the
295 * page must be completely written to stable storage on the server
296 * before it can be refilled by reading in the page from the server.
297 * This can lead to expensive, small, FILE_SYNC mode writes being
300 * It may be more efficient to read the page first if the file is
301 * open for reading in addition to writing, the page is not marked
302 * as Uptodate, it is not dirty or waiting to be committed,
303 * indicating that it was previously allocated and then modified,
304 * that there were valid bytes of data in that range of the file,
305 * and that the new data won't completely replace the old data in
306 * that range of the file.
308 static int nfs_want_read_modify_write(struct file *file, struct page *page,
309 loff_t pos, unsigned len)
311 unsigned int pglen = nfs_page_length(page);
312 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
313 unsigned int end = offset + len;
315 if ((file->f_mode & FMODE_READ) && /* open for read? */
316 !PageUptodate(page) && /* Uptodate? */
317 !PagePrivate(page) && /* i/o request already? */
318 pglen && /* valid bytes of file? */
319 (end < pglen || offset)) /* replace all valid bytes? */
325 * This does the "real" work of the write. We must allocate and lock the
326 * page to be sent back to the generic routine, which then copies the
327 * data from user space.
329 * If the writer ends up delaying the write, the writer needs to
330 * increment the page use counts until he is done with the page.
332 static int nfs_write_begin(struct file *file, struct address_space *mapping,
333 loff_t pos, unsigned len, unsigned flags,
334 struct page **pagep, void **fsdata)
337 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
341 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
342 file->f_path.dentry->d_parent->d_name.name,
343 file->f_path.dentry->d_name.name,
344 mapping->host->i_ino, len, (long long) pos);
348 * Prevent starvation issues if someone is doing a consistency
351 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
352 nfs_wait_bit_killable, TASK_KILLABLE);
356 page = grab_cache_page_write_begin(mapping, index, flags);
361 ret = nfs_flush_incompatible(file, page);
364 page_cache_release(page);
365 } else if (!once_thru &&
366 nfs_want_read_modify_write(file, page, pos, len)) {
368 ret = nfs_readpage(file, page);
369 page_cache_release(page);
376 static int nfs_write_end(struct file *file, struct address_space *mapping,
377 loff_t pos, unsigned len, unsigned copied,
378 struct page *page, void *fsdata)
380 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
383 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
384 file->f_path.dentry->d_parent->d_name.name,
385 file->f_path.dentry->d_name.name,
386 mapping->host->i_ino, len, (long long) pos);
389 * Zero any uninitialised parts of the page, and then mark the page
390 * as up to date if it turns out that we're extending the file.
392 if (!PageUptodate(page)) {
393 unsigned pglen = nfs_page_length(page);
394 unsigned end = offset + len;
397 zero_user_segments(page, 0, offset,
398 end, PAGE_CACHE_SIZE);
399 SetPageUptodate(page);
400 } else if (end >= pglen) {
401 zero_user_segment(page, end, PAGE_CACHE_SIZE);
403 SetPageUptodate(page);
405 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
408 status = nfs_updatepage(file, page, offset, copied);
411 page_cache_release(page);
415 NFS_I(mapping->host)->write_io += copied;
420 * Partially or wholly invalidate a page
421 * - Release the private state associated with a page if undergoing complete
423 * - Called if either PG_private or PG_fscache is set on the page
424 * - Caller holds page lock
426 static void nfs_invalidate_page(struct page *page, unsigned long offset)
428 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
432 /* Cancel any unstarted writes on this page */
433 nfs_wb_page_cancel(page->mapping->host, page);
435 nfs_fscache_invalidate_page(page, page->mapping->host);
439 * Attempt to release the private state associated with a page
440 * - Called if either PG_private or PG_fscache is set on the page
441 * - Caller holds page lock
442 * - Return true (may release page) or false (may not)
444 static int nfs_release_page(struct page *page, gfp_t gfp)
446 struct address_space *mapping = page->mapping;
448 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
450 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
451 * doing this memory reclaim for a fs-related allocation.
453 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
454 !(current->flags & PF_FSTRANS)) {
455 int how = FLUSH_SYNC;
457 /* Don't let kswapd deadlock waiting for OOM RPC calls */
458 if (current_is_kswapd())
460 nfs_commit_inode(mapping->host, how);
462 /* If PagePrivate() is set, then the page is not freeable */
463 if (PagePrivate(page))
465 return nfs_fscache_release_page(page, gfp);
469 * Attempt to clear the private state associated with a page when an error
470 * occurs that requires the cached contents of an inode to be written back or
472 * - Called if either PG_private or fscache is set on the page
473 * - Caller holds page lock
474 * - Return 0 if successful, -error otherwise
476 static int nfs_launder_page(struct page *page)
478 struct inode *inode = page->mapping->host;
479 struct nfs_inode *nfsi = NFS_I(inode);
481 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
482 inode->i_ino, (long long)page_offset(page));
484 nfs_fscache_wait_on_page_write(nfsi, page);
485 return nfs_wb_page(inode, page);
488 const struct address_space_operations nfs_file_aops = {
489 .readpage = nfs_readpage,
490 .readpages = nfs_readpages,
491 .set_page_dirty = __set_page_dirty_nobuffers,
492 .writepage = nfs_writepage,
493 .writepages = nfs_writepages,
494 .write_begin = nfs_write_begin,
495 .write_end = nfs_write_end,
496 .invalidatepage = nfs_invalidate_page,
497 .releasepage = nfs_release_page,
498 .direct_IO = nfs_direct_IO,
499 .migratepage = nfs_migrate_page,
500 .launder_page = nfs_launder_page,
501 .error_remove_page = generic_error_remove_page,
505 * Notification that a PTE pointing to an NFS page is about to be made
506 * writable, implying that someone is about to modify the page through a
507 * shared-writable mapping
509 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
511 struct page *page = vmf->page;
512 struct file *filp = vma->vm_file;
513 struct dentry *dentry = filp->f_path.dentry;
515 int ret = VM_FAULT_NOPAGE;
516 struct address_space *mapping;
518 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
519 dentry->d_parent->d_name.name, dentry->d_name.name,
520 filp->f_mapping->host->i_ino,
521 (long long)page_offset(page));
523 /* make sure the cache has finished storing the page */
524 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
527 mapping = page->mapping;
528 if (mapping != dentry->d_inode->i_mapping)
531 wait_on_page_writeback(page);
533 pagelen = nfs_page_length(page);
537 ret = VM_FAULT_LOCKED;
538 if (nfs_flush_incompatible(filp, page) == 0 &&
539 nfs_updatepage(filp, page, 0, pagelen) == 0)
542 ret = VM_FAULT_SIGBUS;
549 static const struct vm_operations_struct nfs_file_vm_ops = {
550 .fault = filemap_fault,
551 .page_mkwrite = nfs_vm_page_mkwrite,
554 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
556 struct nfs_open_context *ctx;
558 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
560 ctx = nfs_file_open_context(filp);
561 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
566 ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
567 unsigned long nr_segs, loff_t pos)
569 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
570 struct inode * inode = dentry->d_inode;
571 unsigned long written = 0;
573 size_t count = iov_length(iov, nr_segs);
575 if (iocb->ki_filp->f_flags & O_DIRECT)
576 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
578 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
579 dentry->d_parent->d_name.name, dentry->d_name.name,
580 (unsigned long) count, (long long) pos);
583 if (IS_SWAPFILE(inode))
586 * O_APPEND implies that we must revalidate the file length.
588 if (iocb->ki_filp->f_flags & O_APPEND) {
589 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
598 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
602 /* Return error values for O_DSYNC and IS_SYNC() */
603 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
604 int err = vfs_fsync(iocb->ki_filp, 0);
609 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
614 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
618 ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
619 struct file *filp, loff_t *ppos,
620 size_t count, unsigned int flags)
622 struct dentry *dentry = filp->f_path.dentry;
623 struct inode *inode = dentry->d_inode;
624 unsigned long written = 0;
627 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
628 dentry->d_parent->d_name.name, dentry->d_name.name,
629 (unsigned long) count, (unsigned long long) *ppos);
632 * The combination of splice and an O_APPEND destination is disallowed.
635 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
639 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
640 int err = vfs_fsync(filp, 0);
645 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
650 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
652 struct inode *inode = filp->f_mapping->host;
654 unsigned int saved_type = fl->fl_type;
656 /* Try local locking first */
657 posix_test_lock(filp, fl);
658 if (fl->fl_type != F_UNLCK) {
659 /* found a conflict */
662 fl->fl_type = saved_type;
664 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
670 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
674 fl->fl_type = F_UNLCK;
678 static int do_vfs_lock(struct file *file, struct file_lock *fl)
681 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
683 res = posix_lock_file_wait(file, fl);
686 res = flock_lock_file_wait(file, fl);
695 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
697 struct inode *inode = filp->f_mapping->host;
701 * Flush all pending writes before doing anything
704 nfs_sync_mapping(filp->f_mapping);
706 /* NOTE: special case
707 * If we're signalled while cleaning up locks on process exit, we
708 * still need to complete the unlock.
711 * Use local locking if mounted with "-onolock" or with appropriate
715 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
717 status = do_vfs_lock(filp, fl);
722 is_time_granular(struct timespec *ts) {
723 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
727 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
729 struct inode *inode = filp->f_mapping->host;
733 * Flush all pending writes before doing anything
736 status = nfs_sync_mapping(filp->f_mapping);
741 * Use local locking if mounted with "-onolock" or with appropriate
745 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
747 status = do_vfs_lock(filp, fl);
752 * Revalidate the cache if the server has time stamps granular
753 * enough to detect subsecond changes. Otherwise, clear the
754 * cache to prevent missing any changes.
756 * This makes locking act as a cache coherency point.
758 nfs_sync_mapping(filp->f_mapping);
759 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
760 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
761 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
763 nfs_zap_caches(inode);
770 * Lock a (portion of) a file
772 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
774 struct inode *inode = filp->f_mapping->host;
778 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
779 filp->f_path.dentry->d_parent->d_name.name,
780 filp->f_path.dentry->d_name.name,
781 fl->fl_type, fl->fl_flags,
782 (long long)fl->fl_start, (long long)fl->fl_end);
784 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
786 /* No mandatory locks over NFS */
787 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
790 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
793 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
794 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
800 ret = do_getlk(filp, cmd, fl, is_local);
801 else if (fl->fl_type == F_UNLCK)
802 ret = do_unlk(filp, cmd, fl, is_local);
804 ret = do_setlk(filp, cmd, fl, is_local);
810 * Lock a (portion of) a file
812 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
814 struct inode *inode = filp->f_mapping->host;
817 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
818 filp->f_path.dentry->d_parent->d_name.name,
819 filp->f_path.dentry->d_name.name,
820 fl->fl_type, fl->fl_flags);
822 if (!(fl->fl_flags & FL_FLOCK))
825 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
828 /* We're simulating flock() locks using posix locks on the server */
829 fl->fl_owner = (fl_owner_t)filp;
831 fl->fl_end = OFFSET_MAX;
833 if (fl->fl_type == F_UNLCK)
834 return do_unlk(filp, cmd, fl, is_local);
835 return do_setlk(filp, cmd, fl, is_local);
839 * There is no protocol support for leases, so we have no way to implement
840 * them correctly in the face of opens by other clients.
842 int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
844 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
845 file->f_path.dentry->d_parent->d_name.name,
846 file->f_path.dentry->d_name.name, arg);
850 const struct file_operations nfs_file_operations = {
851 .llseek = nfs_file_llseek,
852 .read = do_sync_read,
853 .write = do_sync_write,
854 .aio_read = nfs_file_read,
855 .aio_write = nfs_file_write,
856 .mmap = nfs_file_mmap,
857 .open = nfs_file_open,
858 .flush = nfs_file_flush,
859 .release = nfs_file_release,
860 .fsync = nfs_file_fsync,
863 .splice_read = nfs_file_splice_read,
864 .splice_write = nfs_file_splice_write,
865 .check_flags = nfs_check_flags,
866 .setlease = nfs_setlease,