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/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 <asm/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);
104 struct nfs_inode *nfsi = NFS_I(inode);
106 if (nfs_have_delegated_attributes(inode))
109 if (filp->f_flags & O_DIRECT)
111 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
113 if (nfs_attribute_timeout(inode))
118 return __nfs_revalidate_inode(server, inode);
121 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
123 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
124 filp, offset, whence);
127 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
128 * the cached file length
130 if (whence != SEEK_SET && whence != 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, whence);
140 EXPORT_SYMBOL_GPL(nfs_file_llseek);
143 * Flush all dirty pages, and check for write errors.
146 nfs_file_flush(struct file *file, fl_owner_t id)
148 struct inode *inode = file_inode(file);
150 dprintk("NFS: flush(%pD2)\n", file);
152 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
153 if ((file->f_mode & FMODE_WRITE) == 0)
156 /* Flush writes to the server and return any errors */
157 return vfs_fsync(file, 0);
161 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
163 struct inode *inode = file_inode(iocb->ki_filp);
166 if (iocb->ki_flags & IOCB_DIRECT)
167 return nfs_file_direct_read(iocb, to);
169 dprintk("NFS: read(%pD2, %zu@%lu)\n",
171 iov_iter_count(to), (unsigned long) iocb->ki_pos);
173 nfs_start_io_read(inode);
174 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
176 result = generic_file_read_iter(iocb, to);
178 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
180 nfs_end_io_read(inode);
183 EXPORT_SYMBOL_GPL(nfs_file_read);
186 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
188 struct inode *inode = file_inode(file);
191 dprintk("NFS: mmap(%pD2)\n", file);
193 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
194 * so we call that before revalidating the mapping
196 status = generic_file_mmap(file, vma);
198 vma->vm_ops = &nfs_file_vm_ops;
199 status = nfs_revalidate_mapping(inode, file->f_mapping);
203 EXPORT_SYMBOL_GPL(nfs_file_mmap);
206 * Flush any dirty pages for this process, and check for write errors.
207 * The return status from this call provides a reliable indication of
208 * whether any write errors occurred for this process.
210 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
211 * disk, but it retrieves and clears ctx->error after synching, despite
212 * the two being set at the same time in nfs_context_set_write_error().
213 * This is because the former is used to notify the _next_ call to
214 * nfs_file_write() that a write error occurred, and hence cause it to
215 * fall back to doing a synchronous write.
218 nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
220 struct nfs_open_context *ctx = nfs_file_open_context(file);
221 struct inode *inode = file_inode(file);
222 int have_error, do_resend, status;
225 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
227 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
228 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
229 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
230 status = nfs_commit_inode(inode, FLUSH_SYNC);
231 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
233 ret = xchg(&ctx->error, 0);
241 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
249 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
252 struct inode *inode = file_inode(file);
254 trace_nfs_fsync_enter(inode);
257 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
260 ret = nfs_file_fsync_commit(file, start, end, datasync);
262 ret = pnfs_sync_inode(inode, !!datasync);
264 * If nfs_file_fsync_commit detected a server reboot, then
265 * resend all dirty pages that might have been covered by
266 * the NFS_CONTEXT_RESEND_WRITES flag
270 } while (ret == -EAGAIN);
272 trace_nfs_fsync_exit(inode, ret);
275 EXPORT_SYMBOL_GPL(nfs_file_fsync);
278 * Decide whether a read/modify/write cycle may be more efficient
279 * then a modify/write/read cycle when writing to a page in the
282 * The modify/write/read cycle may occur if a page is read before
283 * being completely filled by the writer. In this situation, the
284 * page must be completely written to stable storage on the server
285 * before it can be refilled by reading in the page from the server.
286 * This can lead to expensive, small, FILE_SYNC mode writes being
289 * It may be more efficient to read the page first if the file is
290 * open for reading in addition to writing, the page is not marked
291 * as Uptodate, it is not dirty or waiting to be committed,
292 * indicating that it was previously allocated and then modified,
293 * that there were valid bytes of data in that range of the file,
294 * and that the new data won't completely replace the old data in
295 * that range of the file.
297 static int nfs_want_read_modify_write(struct file *file, struct page *page,
298 loff_t pos, unsigned len)
300 unsigned int pglen = nfs_page_length(page);
301 unsigned int offset = pos & (PAGE_SIZE - 1);
302 unsigned int end = offset + len;
304 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
305 if (!PageUptodate(page))
310 if ((file->f_mode & FMODE_READ) && /* open for read? */
311 !PageUptodate(page) && /* Uptodate? */
312 !PagePrivate(page) && /* i/o request already? */
313 pglen && /* valid bytes of file? */
314 (end < pglen || offset)) /* replace all valid bytes? */
320 * This does the "real" work of the write. We must allocate and lock the
321 * page to be sent back to the generic routine, which then copies the
322 * data from user space.
324 * If the writer ends up delaying the write, the writer needs to
325 * increment the page use counts until he is done with the page.
327 static int nfs_write_begin(struct file *file, struct address_space *mapping,
328 loff_t pos, unsigned len, unsigned flags,
329 struct page **pagep, void **fsdata)
332 pgoff_t index = pos >> PAGE_SHIFT;
336 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
337 file, mapping->host->i_ino, len, (long long) pos);
340 page = grab_cache_page_write_begin(mapping, index, flags);
345 ret = nfs_flush_incompatible(file, page);
349 } else if (!once_thru &&
350 nfs_want_read_modify_write(file, page, pos, len)) {
352 ret = nfs_readpage(file, page);
360 static int nfs_write_end(struct file *file, struct address_space *mapping,
361 loff_t pos, unsigned len, unsigned copied,
362 struct page *page, void *fsdata)
364 unsigned offset = pos & (PAGE_SIZE - 1);
365 struct nfs_open_context *ctx = nfs_file_open_context(file);
368 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
369 file, mapping->host->i_ino, len, (long long) pos);
372 * Zero any uninitialised parts of the page, and then mark the page
373 * as up to date if it turns out that we're extending the file.
375 if (!PageUptodate(page)) {
376 unsigned pglen = nfs_page_length(page);
377 unsigned end = offset + len;
380 zero_user_segments(page, 0, offset,
382 SetPageUptodate(page);
383 } else if (end >= pglen) {
384 zero_user_segment(page, end, PAGE_SIZE);
386 SetPageUptodate(page);
388 zero_user_segment(page, pglen, PAGE_SIZE);
391 status = nfs_updatepage(file, page, offset, copied);
398 NFS_I(mapping->host)->write_io += copied;
400 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
401 status = nfs_wb_all(mapping->host);
410 * Partially or wholly invalidate a page
411 * - Release the private state associated with a page if undergoing complete
413 * - Called if either PG_private or PG_fscache is set on the page
414 * - Caller holds page lock
416 static void nfs_invalidate_page(struct page *page, unsigned int offset,
419 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
420 page, offset, length);
422 if (offset != 0 || length < PAGE_SIZE)
424 /* Cancel any unstarted writes on this page */
425 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
427 nfs_fscache_invalidate_page(page, page->mapping->host);
431 * Attempt to release the private state associated with a page
432 * - Called if either PG_private or PG_fscache is set on the page
433 * - Caller holds page lock
434 * - Return true (may release page) or false (may not)
436 static int nfs_release_page(struct page *page, gfp_t gfp)
438 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
440 /* If PagePrivate() is set, then the page is not freeable */
441 if (PagePrivate(page))
443 return nfs_fscache_release_page(page, gfp);
446 static void nfs_check_dirty_writeback(struct page *page,
447 bool *dirty, bool *writeback)
449 struct nfs_inode *nfsi;
450 struct address_space *mapping = page_file_mapping(page);
452 if (!mapping || PageSwapCache(page))
456 * Check if an unstable page is currently being committed and
457 * if so, have the VM treat it as if the page is under writeback
458 * so it will not block due to pages that will shortly be freeable.
460 nfsi = NFS_I(mapping->host);
461 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
467 * If PagePrivate() is set, then the page is not freeable and as the
468 * inode is not being committed, it's not going to be cleaned in the
469 * near future so treat it as dirty
471 if (PagePrivate(page))
476 * Attempt to clear the private state associated with a page when an error
477 * occurs that requires the cached contents of an inode to be written back or
479 * - Called if either PG_private or fscache is set on the page
480 * - Caller holds page lock
481 * - Return 0 if successful, -error otherwise
483 static int nfs_launder_page(struct page *page)
485 struct inode *inode = page_file_mapping(page)->host;
486 struct nfs_inode *nfsi = NFS_I(inode);
488 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
489 inode->i_ino, (long long)page_offset(page));
491 nfs_fscache_wait_on_page_write(nfsi, page);
492 return nfs_wb_launder_page(inode, page);
495 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
498 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
502 return rpc_clnt_swap_activate(clnt);
505 static void nfs_swap_deactivate(struct file *file)
507 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
509 rpc_clnt_swap_deactivate(clnt);
512 const struct address_space_operations nfs_file_aops = {
513 .readpage = nfs_readpage,
514 .readpages = nfs_readpages,
515 .set_page_dirty = __set_page_dirty_nobuffers,
516 .writepage = nfs_writepage,
517 .writepages = nfs_writepages,
518 .write_begin = nfs_write_begin,
519 .write_end = nfs_write_end,
520 .invalidatepage = nfs_invalidate_page,
521 .releasepage = nfs_release_page,
522 .direct_IO = nfs_direct_IO,
523 #ifdef CONFIG_MIGRATION
524 .migratepage = nfs_migrate_page,
526 .launder_page = nfs_launder_page,
527 .is_dirty_writeback = nfs_check_dirty_writeback,
528 .error_remove_page = generic_error_remove_page,
529 .swap_activate = nfs_swap_activate,
530 .swap_deactivate = nfs_swap_deactivate,
534 * Notification that a PTE pointing to an NFS page is about to be made
535 * writable, implying that someone is about to modify the page through a
536 * shared-writable mapping
538 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
540 struct page *page = vmf->page;
541 struct file *filp = vma->vm_file;
542 struct inode *inode = file_inode(filp);
544 int ret = VM_FAULT_NOPAGE;
545 struct address_space *mapping;
547 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
548 filp, filp->f_mapping->host->i_ino,
549 (long long)page_offset(page));
551 sb_start_pagefault(inode->i_sb);
553 /* make sure the cache has finished storing the page */
554 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
556 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
557 nfs_wait_bit_killable, TASK_KILLABLE);
560 mapping = page_file_mapping(page);
561 if (mapping != inode->i_mapping)
564 wait_on_page_writeback(page);
566 pagelen = nfs_page_length(page);
570 ret = VM_FAULT_LOCKED;
571 if (nfs_flush_incompatible(filp, page) == 0 &&
572 nfs_updatepage(filp, page, 0, pagelen) == 0)
575 ret = VM_FAULT_SIGBUS;
579 sb_end_pagefault(inode->i_sb);
583 static const struct vm_operations_struct nfs_file_vm_ops = {
584 .fault = filemap_fault,
585 .map_pages = filemap_map_pages,
586 .page_mkwrite = nfs_vm_page_mkwrite,
589 static int nfs_need_check_write(struct file *filp, struct inode *inode)
591 struct nfs_open_context *ctx;
593 ctx = nfs_file_open_context(filp);
594 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
595 nfs_ctx_key_to_expire(ctx, inode))
600 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
602 struct file *file = iocb->ki_filp;
603 struct inode *inode = file_inode(file);
604 unsigned long written = 0;
607 result = nfs_key_timeout_notify(file, inode);
611 if (iocb->ki_flags & IOCB_DIRECT)
612 return nfs_file_direct_write(iocb, from);
614 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
615 file, iov_iter_count(from), (long long) iocb->ki_pos);
617 if (IS_SWAPFILE(inode))
620 * O_APPEND implies that we must revalidate the file length.
622 if (iocb->ki_flags & IOCB_APPEND) {
623 result = nfs_revalidate_file_size(inode, file);
628 nfs_start_io_write(inode);
629 result = generic_write_checks(iocb, from);
631 current->backing_dev_info = inode_to_bdi(inode);
632 result = generic_perform_write(file, from, iocb->ki_pos);
633 current->backing_dev_info = NULL;
635 nfs_end_io_write(inode);
639 result = generic_write_sync(iocb, result);
643 iocb->ki_pos += written;
645 /* Return error values */
646 if (nfs_need_check_write(file, inode)) {
647 int err = vfs_fsync(file, 0);
651 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
656 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
659 EXPORT_SYMBOL_GPL(nfs_file_write);
662 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
664 struct inode *inode = filp->f_mapping->host;
666 unsigned int saved_type = fl->fl_type;
668 /* Try local locking first */
669 posix_test_lock(filp, fl);
670 if (fl->fl_type != F_UNLCK) {
671 /* found a conflict */
674 fl->fl_type = saved_type;
676 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
682 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
686 fl->fl_type = F_UNLCK;
691 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
693 struct inode *inode = filp->f_mapping->host;
694 struct nfs_lock_context *l_ctx;
698 * Flush all pending writes before doing anything
703 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
704 if (!IS_ERR(l_ctx)) {
705 status = nfs_iocounter_wait(l_ctx);
706 nfs_put_lock_context(l_ctx);
711 /* NOTE: special case
712 * If we're signalled while cleaning up locks on process exit, we
713 * still need to complete the unlock.
716 * Use local locking if mounted with "-onolock" or with appropriate
720 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
722 status = locks_lock_file_wait(filp, fl);
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 = locks_lock_file_wait(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 nfs_zap_mapping(inode, filp->f_mapping);
766 * Lock a (portion of) a file
768 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
770 struct inode *inode = filp->f_mapping->host;
774 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
775 filp, fl->fl_type, fl->fl_flags,
776 (long long)fl->fl_start, (long long)fl->fl_end);
778 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
780 /* No mandatory locks over NFS */
781 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
784 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
787 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
788 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
794 ret = do_getlk(filp, cmd, fl, is_local);
795 else if (fl->fl_type == F_UNLCK)
796 ret = do_unlk(filp, cmd, fl, is_local);
798 ret = do_setlk(filp, cmd, fl, is_local);
802 EXPORT_SYMBOL_GPL(nfs_lock);
805 * Lock a (portion of) a file
807 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
809 struct inode *inode = filp->f_mapping->host;
812 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
813 filp, fl->fl_type, fl->fl_flags);
815 if (!(fl->fl_flags & FL_FLOCK))
819 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
820 * any standard. In principle we might be able to support LOCK_MAND
821 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
822 * NFS code is not set up for it.
824 if (fl->fl_type & LOCK_MAND)
827 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
830 /* We're simulating flock() locks using posix locks on the server */
831 if (fl->fl_type == F_UNLCK)
832 return do_unlk(filp, cmd, fl, is_local);
833 return do_setlk(filp, cmd, fl, is_local);
835 EXPORT_SYMBOL_GPL(nfs_flock);
837 const struct file_operations nfs_file_operations = {
838 .llseek = nfs_file_llseek,
839 .read_iter = nfs_file_read,
840 .write_iter = nfs_file_write,
841 .mmap = nfs_file_mmap,
842 .open = nfs_file_open,
843 .flush = nfs_file_flush,
844 .release = nfs_file_release,
845 .fsync = nfs_file_fsync,
848 .splice_read = generic_file_splice_read,
849 .splice_write = iter_file_splice_write,
850 .check_flags = nfs_check_flags,
851 .setlease = simple_nosetlease,
853 EXPORT_SYMBOL_GPL(nfs_file_operations);