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/module.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/fcntl.h>
24 #include <linux/stat.h>
25 #include <linux/nfs_fs.h>
26 #include <linux/nfs_mount.h>
28 #include <linux/pagemap.h>
29 #include <linux/gfp.h>
30 #include <linux/swap.h>
32 #include <linux/uaccess.h>
34 #include "delegation.h"
42 #define NFSDBG_FACILITY NFSDBG_FILE
44 static const struct vm_operations_struct nfs_file_vm_ops;
46 /* Hack for future NFS swap support */
48 # define IS_SWAPFILE(inode) (0)
51 int nfs_check_flags(int flags)
53 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
58 EXPORT_SYMBOL_GPL(nfs_check_flags);
64 nfs_file_open(struct inode *inode, struct file *filp)
68 dprintk("NFS: open file(%pD2)\n", filp);
70 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
71 res = nfs_check_flags(filp->f_flags);
75 res = nfs_open(inode, filp);
80 nfs_file_release(struct inode *inode, struct file *filp)
82 dprintk("NFS: release(%pD2)\n", filp);
84 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
85 nfs_file_clear_open_context(filp);
88 EXPORT_SYMBOL_GPL(nfs_file_release);
91 * nfs_revalidate_size - Revalidate the file size
92 * @inode - pointer to inode struct
93 * @file - pointer to struct file
95 * Revalidates the file length. This is basically a wrapper around
96 * nfs_revalidate_inode() that takes into account the fact that we may
97 * have cached writes (in which case we don't care about the server's
98 * idea of what the file length is), or O_DIRECT (in which case we
99 * shouldn't trust the cache).
101 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
103 struct nfs_server *server = NFS_SERVER(inode);
105 if (filp->f_flags & O_DIRECT)
107 if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE))
111 return __nfs_revalidate_inode(server, inode);
114 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
116 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
117 filp, offset, whence);
120 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
121 * the cached file length
123 if (whence != SEEK_SET && whence != SEEK_CUR) {
124 struct inode *inode = filp->f_mapping->host;
126 int retval = nfs_revalidate_file_size(inode, filp);
128 return (loff_t)retval;
131 return generic_file_llseek(filp, offset, whence);
133 EXPORT_SYMBOL_GPL(nfs_file_llseek);
136 * Flush all dirty pages, and check for write errors.
139 nfs_file_flush(struct file *file, fl_owner_t id)
141 struct inode *inode = file_inode(file);
143 dprintk("NFS: flush(%pD2)\n", file);
145 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
146 if ((file->f_mode & FMODE_WRITE) == 0)
149 /* Flush writes to the server and return any errors */
150 return vfs_fsync(file, 0);
154 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
156 struct inode *inode = file_inode(iocb->ki_filp);
159 if (iocb->ki_flags & IOCB_DIRECT)
160 return nfs_file_direct_read(iocb, to);
162 dprintk("NFS: read(%pD2, %zu@%lu)\n",
164 iov_iter_count(to), (unsigned long) iocb->ki_pos);
166 nfs_start_io_read(inode);
167 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
169 result = generic_file_read_iter(iocb, to);
171 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
173 nfs_end_io_read(inode);
176 EXPORT_SYMBOL_GPL(nfs_file_read);
179 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
181 struct inode *inode = file_inode(file);
184 dprintk("NFS: mmap(%pD2)\n", file);
186 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
187 * so we call that before revalidating the mapping
189 status = generic_file_mmap(file, vma);
191 vma->vm_ops = &nfs_file_vm_ops;
192 status = nfs_revalidate_mapping(inode, file->f_mapping);
196 EXPORT_SYMBOL_GPL(nfs_file_mmap);
199 * Flush any dirty pages for this process, and check for write errors.
200 * The return status from this call provides a reliable indication of
201 * whether any write errors occurred for this process.
203 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
204 * disk, but it retrieves and clears ctx->error after synching, despite
205 * the two being set at the same time in nfs_context_set_write_error().
206 * This is because the former is used to notify the _next_ call to
207 * nfs_file_write() that a write error occurred, and hence cause it to
208 * fall back to doing a synchronous write.
211 nfs_file_fsync_commit(struct file *file, int datasync)
213 struct nfs_open_context *ctx = nfs_file_open_context(file);
214 struct inode *inode = file_inode(file);
215 int do_resend, status;
218 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
220 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
221 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
222 status = nfs_commit_inode(inode, FLUSH_SYNC);
223 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) {
224 ret = xchg(&ctx->error, 0);
232 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
240 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
243 struct inode *inode = file_inode(file);
245 trace_nfs_fsync_enter(inode);
248 struct nfs_open_context *ctx = nfs_file_open_context(file);
249 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
250 if (test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags)) {
251 int ret2 = xchg(&ctx->error, 0);
257 ret = nfs_file_fsync_commit(file, datasync);
259 ret = pnfs_sync_inode(inode, !!datasync);
261 * If nfs_file_fsync_commit detected a server reboot, then
262 * resend all dirty pages that might have been covered by
263 * the NFS_CONTEXT_RESEND_WRITES flag
267 } while (ret == -EAGAIN);
269 trace_nfs_fsync_exit(inode, ret);
272 EXPORT_SYMBOL_GPL(nfs_file_fsync);
275 * Decide whether a read/modify/write cycle may be more efficient
276 * then a modify/write/read cycle when writing to a page in the
279 * The modify/write/read cycle may occur if a page is read before
280 * being completely filled by the writer. In this situation, the
281 * page must be completely written to stable storage on the server
282 * before it can be refilled by reading in the page from the server.
283 * This can lead to expensive, small, FILE_SYNC mode writes being
286 * It may be more efficient to read the page first if the file is
287 * open for reading in addition to writing, the page is not marked
288 * as Uptodate, it is not dirty or waiting to be committed,
289 * indicating that it was previously allocated and then modified,
290 * that there were valid bytes of data in that range of the file,
291 * and that the new data won't completely replace the old data in
292 * that range of the file.
294 static int nfs_want_read_modify_write(struct file *file, struct page *page,
295 loff_t pos, unsigned len)
297 unsigned int pglen = nfs_page_length(page);
298 unsigned int offset = pos & (PAGE_SIZE - 1);
299 unsigned int end = offset + len;
301 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
302 if (!PageUptodate(page))
307 if ((file->f_mode & FMODE_READ) && /* open for read? */
308 !PageUptodate(page) && /* Uptodate? */
309 !PagePrivate(page) && /* i/o request already? */
310 pglen && /* valid bytes of file? */
311 (end < pglen || offset)) /* replace all valid bytes? */
317 * This does the "real" work of the write. We must allocate and lock the
318 * page to be sent back to the generic routine, which then copies the
319 * data from user space.
321 * If the writer ends up delaying the write, the writer needs to
322 * increment the page use counts until he is done with the page.
324 static int nfs_write_begin(struct file *file, struct address_space *mapping,
325 loff_t pos, unsigned len, unsigned flags,
326 struct page **pagep, void **fsdata)
329 pgoff_t index = pos >> PAGE_SHIFT;
333 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
334 file, mapping->host->i_ino, len, (long long) pos);
337 page = grab_cache_page_write_begin(mapping, index, flags);
342 ret = nfs_flush_incompatible(file, page);
346 } else if (!once_thru &&
347 nfs_want_read_modify_write(file, page, pos, len)) {
349 ret = nfs_readpage(file, page);
357 static int nfs_write_end(struct file *file, struct address_space *mapping,
358 loff_t pos, unsigned len, unsigned copied,
359 struct page *page, void *fsdata)
361 unsigned offset = pos & (PAGE_SIZE - 1);
362 struct nfs_open_context *ctx = nfs_file_open_context(file);
365 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
366 file, mapping->host->i_ino, len, (long long) pos);
369 * Zero any uninitialised parts of the page, and then mark the page
370 * as up to date if it turns out that we're extending the file.
372 if (!PageUptodate(page)) {
373 unsigned pglen = nfs_page_length(page);
374 unsigned end = offset + copied;
377 zero_user_segments(page, 0, offset,
379 SetPageUptodate(page);
380 } else if (end >= pglen) {
381 zero_user_segment(page, end, PAGE_SIZE);
383 SetPageUptodate(page);
385 zero_user_segment(page, pglen, PAGE_SIZE);
388 status = nfs_updatepage(file, page, offset, copied);
395 NFS_I(mapping->host)->write_io += copied;
397 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
398 status = nfs_wb_all(mapping->host);
407 * Partially or wholly invalidate a page
408 * - Release the private state associated with a page if undergoing complete
410 * - Called if either PG_private or PG_fscache is set on the page
411 * - Caller holds page lock
413 static void nfs_invalidate_page(struct page *page, unsigned int offset,
416 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
417 page, offset, length);
419 if (offset != 0 || length < PAGE_SIZE)
421 /* Cancel any unstarted writes on this page */
422 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
424 nfs_fscache_invalidate_page(page, page->mapping->host);
428 * Attempt to release the private state associated with a page
429 * - Called if either PG_private or PG_fscache is set on the page
430 * - Caller holds page lock
431 * - Return true (may release page) or false (may not)
433 static int nfs_release_page(struct page *page, gfp_t gfp)
435 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
437 /* If PagePrivate() is set, then the page is not freeable */
438 if (PagePrivate(page))
440 return nfs_fscache_release_page(page, gfp);
443 static void nfs_check_dirty_writeback(struct page *page,
444 bool *dirty, bool *writeback)
446 struct nfs_inode *nfsi;
447 struct address_space *mapping = page_file_mapping(page);
449 if (!mapping || PageSwapCache(page))
453 * Check if an unstable page is currently being committed and
454 * if so, have the VM treat it as if the page is under writeback
455 * so it will not block due to pages that will shortly be freeable.
457 nfsi = NFS_I(mapping->host);
458 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
464 * If PagePrivate() is set, then the page is not freeable and as the
465 * inode is not being committed, it's not going to be cleaned in the
466 * near future so treat it as dirty
468 if (PagePrivate(page))
473 * Attempt to clear the private state associated with a page when an error
474 * occurs that requires the cached contents of an inode to be written back or
476 * - Called if either PG_private or fscache is set on the page
477 * - Caller holds page lock
478 * - Return 0 if successful, -error otherwise
480 static int nfs_launder_page(struct page *page)
482 struct inode *inode = page_file_mapping(page)->host;
483 struct nfs_inode *nfsi = NFS_I(inode);
485 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
486 inode->i_ino, (long long)page_offset(page));
488 nfs_fscache_wait_on_page_write(nfsi, page);
489 return nfs_wb_page(inode, page);
492 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
495 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
499 return rpc_clnt_swap_activate(clnt);
502 static void nfs_swap_deactivate(struct file *file)
504 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
506 rpc_clnt_swap_deactivate(clnt);
509 const struct address_space_operations nfs_file_aops = {
510 .readpage = nfs_readpage,
511 .readpages = nfs_readpages,
512 .set_page_dirty = __set_page_dirty_nobuffers,
513 .writepage = nfs_writepage,
514 .writepages = nfs_writepages,
515 .write_begin = nfs_write_begin,
516 .write_end = nfs_write_end,
517 .invalidatepage = nfs_invalidate_page,
518 .releasepage = nfs_release_page,
519 .direct_IO = nfs_direct_IO,
520 #ifdef CONFIG_MIGRATION
521 .migratepage = nfs_migrate_page,
523 .launder_page = nfs_launder_page,
524 .is_dirty_writeback = nfs_check_dirty_writeback,
525 .error_remove_page = generic_error_remove_page,
526 .swap_activate = nfs_swap_activate,
527 .swap_deactivate = nfs_swap_deactivate,
531 * Notification that a PTE pointing to an NFS page is about to be made
532 * writable, implying that someone is about to modify the page through a
533 * shared-writable mapping
535 static int nfs_vm_page_mkwrite(struct vm_fault *vmf)
537 struct page *page = vmf->page;
538 struct file *filp = vmf->vma->vm_file;
539 struct inode *inode = file_inode(filp);
541 int ret = VM_FAULT_NOPAGE;
542 struct address_space *mapping;
544 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
545 filp, filp->f_mapping->host->i_ino,
546 (long long)page_offset(page));
548 sb_start_pagefault(inode->i_sb);
550 /* make sure the cache has finished storing the page */
551 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
553 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
554 nfs_wait_bit_killable, TASK_KILLABLE);
557 mapping = page_file_mapping(page);
558 if (mapping != inode->i_mapping)
561 wait_on_page_writeback(page);
563 pagelen = nfs_page_length(page);
567 ret = VM_FAULT_LOCKED;
568 if (nfs_flush_incompatible(filp, page) == 0 &&
569 nfs_updatepage(filp, page, 0, pagelen) == 0)
572 ret = VM_FAULT_SIGBUS;
576 sb_end_pagefault(inode->i_sb);
580 static const struct vm_operations_struct nfs_file_vm_ops = {
581 .fault = filemap_fault,
582 .map_pages = filemap_map_pages,
583 .page_mkwrite = nfs_vm_page_mkwrite,
586 static int nfs_need_check_write(struct file *filp, struct inode *inode)
588 struct nfs_open_context *ctx;
590 ctx = nfs_file_open_context(filp);
591 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
592 nfs_ctx_key_to_expire(ctx, inode))
597 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
599 struct file *file = iocb->ki_filp;
600 struct inode *inode = file_inode(file);
601 unsigned long written = 0;
604 result = nfs_key_timeout_notify(file, inode);
608 if (iocb->ki_flags & IOCB_DIRECT)
609 return nfs_file_direct_write(iocb, from);
611 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
612 file, iov_iter_count(from), (long long) iocb->ki_pos);
614 if (IS_SWAPFILE(inode))
617 * O_APPEND implies that we must revalidate the file length.
619 if (iocb->ki_flags & IOCB_APPEND) {
620 result = nfs_revalidate_file_size(inode, file);
624 if (iocb->ki_pos > i_size_read(inode))
625 nfs_revalidate_mapping(inode, file->f_mapping);
627 nfs_start_io_write(inode);
628 result = generic_write_checks(iocb, from);
630 current->backing_dev_info = inode_to_bdi(inode);
631 result = generic_perform_write(file, from, iocb->ki_pos);
632 current->backing_dev_info = NULL;
634 nfs_end_io_write(inode);
639 iocb->ki_pos += written;
640 result = generic_write_sync(iocb, written);
644 /* Return error values */
645 if (nfs_need_check_write(file, inode)) {
646 int err = vfs_fsync(file, 0);
650 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
655 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
658 EXPORT_SYMBOL_GPL(nfs_file_write);
661 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
663 struct inode *inode = filp->f_mapping->host;
665 unsigned int saved_type = fl->fl_type;
667 /* Try local locking first */
668 posix_test_lock(filp, fl);
669 if (fl->fl_type != F_UNLCK) {
670 /* found a conflict */
673 fl->fl_type = saved_type;
675 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
681 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
685 fl->fl_type = F_UNLCK;
690 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
692 struct inode *inode = filp->f_mapping->host;
693 struct nfs_lock_context *l_ctx;
697 * Flush all pending writes before doing anything
702 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
703 if (!IS_ERR(l_ctx)) {
704 status = nfs_iocounter_wait(l_ctx);
705 nfs_put_lock_context(l_ctx);
706 /* NOTE: special case
707 * If we're signalled while cleaning up locks on process exit, we
708 * still need to complete the unlock.
710 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
715 * Use local locking if mounted with "-onolock" or with appropriate
719 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
721 status = locks_lock_file_wait(filp, fl);
726 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
728 struct inode *inode = filp->f_mapping->host;
732 * Flush all pending writes before doing anything
735 status = nfs_sync_mapping(filp->f_mapping);
740 * Use local locking if mounted with "-onolock" or with appropriate
744 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
746 status = locks_lock_file_wait(filp, fl);
751 * Invalidate cache to prevent missing any changes. If
752 * the file is mapped, clear the page cache as well so
753 * those mappings will be loaded.
755 * This makes locking act as a cache coherency point.
757 nfs_sync_mapping(filp->f_mapping);
758 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
759 nfs_zap_caches(inode);
760 if (mapping_mapped(filp->f_mapping))
761 nfs_revalidate_mapping(inode, filp->f_mapping);
768 * Lock a (portion of) a file
770 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
772 struct inode *inode = filp->f_mapping->host;
776 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
777 filp, fl->fl_type, fl->fl_flags,
778 (long long)fl->fl_start, (long long)fl->fl_end);
780 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
782 /* No mandatory locks over NFS */
783 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
786 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
789 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
790 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
796 ret = do_getlk(filp, cmd, fl, is_local);
797 else if (fl->fl_type == F_UNLCK)
798 ret = do_unlk(filp, cmd, fl, is_local);
800 ret = do_setlk(filp, cmd, fl, is_local);
804 EXPORT_SYMBOL_GPL(nfs_lock);
807 * Lock a (portion of) a file
809 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
811 struct inode *inode = filp->f_mapping->host;
814 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
815 filp, fl->fl_type, fl->fl_flags);
817 if (!(fl->fl_flags & FL_FLOCK))
821 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
822 * any standard. In principle we might be able to support LOCK_MAND
823 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
824 * NFS code is not set up for it.
826 if (fl->fl_type & LOCK_MAND)
829 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
833 * VFS doesn't require the open mode to match a flock() lock's type.
834 * NFS, however, may simulate flock() locking with posix locking which
835 * requires the open mode to match the lock type.
837 switch (fl->fl_type) {
839 return do_unlk(filp, cmd, fl, is_local);
841 if (!(filp->f_mode & FMODE_READ))
845 if (!(filp->f_mode & FMODE_WRITE))
849 return do_setlk(filp, cmd, fl, is_local);
851 EXPORT_SYMBOL_GPL(nfs_flock);
853 const struct file_operations nfs_file_operations = {
854 .llseek = nfs_file_llseek,
855 .read_iter = nfs_file_read,
856 .write_iter = nfs_file_write,
857 .mmap = nfs_file_mmap,
858 .open = nfs_file_open,
859 .flush = nfs_file_flush,
860 .release = nfs_file_release,
861 .fsync = nfs_file_fsync,
864 .splice_read = generic_file_splice_read,
865 .splice_write = iter_file_splice_write,
866 .check_flags = nfs_check_flags,
867 .setlease = simple_nosetlease,
869 EXPORT_SYMBOL_GPL(nfs_file_operations);