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_splice_read(struct file *filp, loff_t *ppos,
187 struct pipe_inode_info *pipe, size_t count,
190 struct inode *inode = file_inode(filp);
193 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
194 filp, (unsigned long) count, (unsigned long long) *ppos);
196 nfs_start_io_read(inode);
197 res = nfs_revalidate_mapping(inode, filp->f_mapping);
199 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
201 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
203 nfs_end_io_read(inode);
206 EXPORT_SYMBOL_GPL(nfs_file_splice_read);
209 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
211 struct inode *inode = file_inode(file);
214 dprintk("NFS: mmap(%pD2)\n", file);
216 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
217 * so we call that before revalidating the mapping
219 status = generic_file_mmap(file, vma);
221 vma->vm_ops = &nfs_file_vm_ops;
222 status = nfs_revalidate_mapping(inode, file->f_mapping);
226 EXPORT_SYMBOL_GPL(nfs_file_mmap);
229 * Flush any dirty pages for this process, and check for write errors.
230 * The return status from this call provides a reliable indication of
231 * whether any write errors occurred for this process.
233 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
234 * disk, but it retrieves and clears ctx->error after synching, despite
235 * the two being set at the same time in nfs_context_set_write_error().
236 * This is because the former is used to notify the _next_ call to
237 * nfs_file_write() that a write error occurred, and hence cause it to
238 * fall back to doing a synchronous write.
241 nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
243 struct nfs_open_context *ctx = nfs_file_open_context(file);
244 struct inode *inode = file_inode(file);
245 int have_error, do_resend, status;
248 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
250 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
251 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
252 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
253 status = nfs_commit_inode(inode, FLUSH_SYNC);
254 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
256 ret = xchg(&ctx->error, 0);
264 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
272 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
275 struct inode *inode = file_inode(file);
277 trace_nfs_fsync_enter(inode);
280 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
283 ret = nfs_file_fsync_commit(file, start, end, datasync);
285 ret = pnfs_sync_inode(inode, !!datasync);
287 * If nfs_file_fsync_commit detected a server reboot, then
288 * resend all dirty pages that might have been covered by
289 * the NFS_CONTEXT_RESEND_WRITES flag
293 } while (ret == -EAGAIN);
295 trace_nfs_fsync_exit(inode, ret);
298 EXPORT_SYMBOL_GPL(nfs_file_fsync);
301 * Decide whether a read/modify/write cycle may be more efficient
302 * then a modify/write/read cycle when writing to a page in the
305 * The modify/write/read cycle may occur if a page is read before
306 * being completely filled by the writer. In this situation, the
307 * page must be completely written to stable storage on the server
308 * before it can be refilled by reading in the page from the server.
309 * This can lead to expensive, small, FILE_SYNC mode writes being
312 * It may be more efficient to read the page first if the file is
313 * open for reading in addition to writing, the page is not marked
314 * as Uptodate, it is not dirty or waiting to be committed,
315 * indicating that it was previously allocated and then modified,
316 * that there were valid bytes of data in that range of the file,
317 * and that the new data won't completely replace the old data in
318 * that range of the file.
320 static int nfs_want_read_modify_write(struct file *file, struct page *page,
321 loff_t pos, unsigned len)
323 unsigned int pglen = nfs_page_length(page);
324 unsigned int offset = pos & (PAGE_SIZE - 1);
325 unsigned int end = offset + len;
327 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
328 if (!PageUptodate(page))
333 if ((file->f_mode & FMODE_READ) && /* open for read? */
334 !PageUptodate(page) && /* Uptodate? */
335 !PagePrivate(page) && /* i/o request already? */
336 pglen && /* valid bytes of file? */
337 (end < pglen || offset)) /* replace all valid bytes? */
343 * This does the "real" work of the write. We must allocate and lock the
344 * page to be sent back to the generic routine, which then copies the
345 * data from user space.
347 * If the writer ends up delaying the write, the writer needs to
348 * increment the page use counts until he is done with the page.
350 static int nfs_write_begin(struct file *file, struct address_space *mapping,
351 loff_t pos, unsigned len, unsigned flags,
352 struct page **pagep, void **fsdata)
355 pgoff_t index = pos >> PAGE_SHIFT;
359 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
360 file, mapping->host->i_ino, len, (long long) pos);
363 page = grab_cache_page_write_begin(mapping, index, flags);
368 ret = nfs_flush_incompatible(file, page);
372 } else if (!once_thru &&
373 nfs_want_read_modify_write(file, page, pos, len)) {
375 ret = nfs_readpage(file, page);
383 static int nfs_write_end(struct file *file, struct address_space *mapping,
384 loff_t pos, unsigned len, unsigned copied,
385 struct page *page, void *fsdata)
387 unsigned offset = pos & (PAGE_SIZE - 1);
388 struct nfs_open_context *ctx = nfs_file_open_context(file);
391 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
392 file, mapping->host->i_ino, len, (long long) pos);
395 * Zero any uninitialised parts of the page, and then mark the page
396 * as up to date if it turns out that we're extending the file.
398 if (!PageUptodate(page)) {
399 unsigned pglen = nfs_page_length(page);
400 unsigned end = offset + len;
403 zero_user_segments(page, 0, offset,
405 SetPageUptodate(page);
406 } else if (end >= pglen) {
407 zero_user_segment(page, end, PAGE_SIZE);
409 SetPageUptodate(page);
411 zero_user_segment(page, pglen, PAGE_SIZE);
414 status = nfs_updatepage(file, page, offset, copied);
421 NFS_I(mapping->host)->write_io += copied;
423 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
424 status = nfs_wb_all(mapping->host);
433 * Partially or wholly invalidate a page
434 * - Release the private state associated with a page if undergoing complete
436 * - Called if either PG_private or PG_fscache is set on the page
437 * - Caller holds page lock
439 static void nfs_invalidate_page(struct page *page, unsigned int offset,
442 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
443 page, offset, length);
445 if (offset != 0 || length < PAGE_SIZE)
447 /* Cancel any unstarted writes on this page */
448 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
450 nfs_fscache_invalidate_page(page, page->mapping->host);
454 * Attempt to release the private state associated with a page
455 * - Called if either PG_private or PG_fscache is set on the page
456 * - Caller holds page lock
457 * - Return true (may release page) or false (may not)
459 static int nfs_release_page(struct page *page, gfp_t gfp)
461 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
463 /* If PagePrivate() is set, then the page is not freeable */
464 if (PagePrivate(page))
466 return nfs_fscache_release_page(page, gfp);
469 static void nfs_check_dirty_writeback(struct page *page,
470 bool *dirty, bool *writeback)
472 struct nfs_inode *nfsi;
473 struct address_space *mapping = page_file_mapping(page);
475 if (!mapping || PageSwapCache(page))
479 * Check if an unstable page is currently being committed and
480 * if so, have the VM treat it as if the page is under writeback
481 * so it will not block due to pages that will shortly be freeable.
483 nfsi = NFS_I(mapping->host);
484 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
490 * If PagePrivate() is set, then the page is not freeable and as the
491 * inode is not being committed, it's not going to be cleaned in the
492 * near future so treat it as dirty
494 if (PagePrivate(page))
499 * Attempt to clear the private state associated with a page when an error
500 * occurs that requires the cached contents of an inode to be written back or
502 * - Called if either PG_private or fscache is set on the page
503 * - Caller holds page lock
504 * - Return 0 if successful, -error otherwise
506 static int nfs_launder_page(struct page *page)
508 struct inode *inode = page_file_mapping(page)->host;
509 struct nfs_inode *nfsi = NFS_I(inode);
511 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
512 inode->i_ino, (long long)page_offset(page));
514 nfs_fscache_wait_on_page_write(nfsi, page);
515 return nfs_wb_launder_page(inode, page);
518 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
521 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
525 return rpc_clnt_swap_activate(clnt);
528 static void nfs_swap_deactivate(struct file *file)
530 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
532 rpc_clnt_swap_deactivate(clnt);
535 const struct address_space_operations nfs_file_aops = {
536 .readpage = nfs_readpage,
537 .readpages = nfs_readpages,
538 .set_page_dirty = __set_page_dirty_nobuffers,
539 .writepage = nfs_writepage,
540 .writepages = nfs_writepages,
541 .write_begin = nfs_write_begin,
542 .write_end = nfs_write_end,
543 .invalidatepage = nfs_invalidate_page,
544 .releasepage = nfs_release_page,
545 .direct_IO = nfs_direct_IO,
546 .migratepage = nfs_migrate_page,
547 .launder_page = nfs_launder_page,
548 .is_dirty_writeback = nfs_check_dirty_writeback,
549 .error_remove_page = generic_error_remove_page,
550 .swap_activate = nfs_swap_activate,
551 .swap_deactivate = nfs_swap_deactivate,
555 * Notification that a PTE pointing to an NFS page is about to be made
556 * writable, implying that someone is about to modify the page through a
557 * shared-writable mapping
559 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
561 struct page *page = vmf->page;
562 struct file *filp = vma->vm_file;
563 struct inode *inode = file_inode(filp);
565 int ret = VM_FAULT_NOPAGE;
566 struct address_space *mapping;
568 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
569 filp, filp->f_mapping->host->i_ino,
570 (long long)page_offset(page));
572 sb_start_pagefault(inode->i_sb);
574 /* make sure the cache has finished storing the page */
575 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
577 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
578 nfs_wait_bit_killable, TASK_KILLABLE);
581 mapping = page_file_mapping(page);
582 if (mapping != inode->i_mapping)
585 wait_on_page_writeback(page);
587 pagelen = nfs_page_length(page);
591 ret = VM_FAULT_LOCKED;
592 if (nfs_flush_incompatible(filp, page) == 0 &&
593 nfs_updatepage(filp, page, 0, pagelen) == 0)
596 ret = VM_FAULT_SIGBUS;
600 sb_end_pagefault(inode->i_sb);
604 static const struct vm_operations_struct nfs_file_vm_ops = {
605 .fault = filemap_fault,
606 .map_pages = filemap_map_pages,
607 .page_mkwrite = nfs_vm_page_mkwrite,
610 static int nfs_need_check_write(struct file *filp, struct inode *inode)
612 struct nfs_open_context *ctx;
614 ctx = nfs_file_open_context(filp);
615 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
616 nfs_ctx_key_to_expire(ctx, inode))
621 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
623 struct file *file = iocb->ki_filp;
624 struct inode *inode = file_inode(file);
625 unsigned long written = 0;
628 result = nfs_key_timeout_notify(file, inode);
632 if (iocb->ki_flags & IOCB_DIRECT)
633 return nfs_file_direct_write(iocb, from);
635 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
636 file, iov_iter_count(from), (long long) iocb->ki_pos);
638 if (IS_SWAPFILE(inode))
641 * O_APPEND implies that we must revalidate the file length.
643 if (iocb->ki_flags & IOCB_APPEND) {
644 result = nfs_revalidate_file_size(inode, file);
649 nfs_start_io_write(inode);
650 result = generic_write_checks(iocb, from);
652 current->backing_dev_info = inode_to_bdi(inode);
653 result = generic_perform_write(file, from, iocb->ki_pos);
654 current->backing_dev_info = NULL;
656 nfs_end_io_write(inode);
660 result = generic_write_sync(iocb, result);
664 iocb->ki_pos += written;
666 /* Return error values */
667 if (nfs_need_check_write(file, inode)) {
668 int err = vfs_fsync(file, 0);
672 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
677 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
680 EXPORT_SYMBOL_GPL(nfs_file_write);
683 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
685 struct inode *inode = filp->f_mapping->host;
687 unsigned int saved_type = fl->fl_type;
689 /* Try local locking first */
690 posix_test_lock(filp, fl);
691 if (fl->fl_type != F_UNLCK) {
692 /* found a conflict */
695 fl->fl_type = saved_type;
697 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
703 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
707 fl->fl_type = F_UNLCK;
711 static int do_vfs_lock(struct file *file, struct file_lock *fl)
713 return locks_lock_file_wait(file, fl);
717 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
719 struct inode *inode = filp->f_mapping->host;
720 struct nfs_lock_context *l_ctx;
724 * Flush all pending writes before doing anything
729 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
730 if (!IS_ERR(l_ctx)) {
731 status = nfs_iocounter_wait(l_ctx);
732 nfs_put_lock_context(l_ctx);
737 /* NOTE: special case
738 * If we're signalled while cleaning up locks on process exit, we
739 * still need to complete the unlock.
742 * Use local locking if mounted with "-onolock" or with appropriate
746 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
748 status = do_vfs_lock(filp, fl);
753 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
755 struct inode *inode = filp->f_mapping->host;
759 * Flush all pending writes before doing anything
762 status = nfs_sync_mapping(filp->f_mapping);
767 * Use local locking if mounted with "-onolock" or with appropriate
771 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
773 status = do_vfs_lock(filp, fl);
778 * Revalidate the cache if the server has time stamps granular
779 * enough to detect subsecond changes. Otherwise, clear the
780 * cache to prevent missing any changes.
782 * This makes locking act as a cache coherency point.
784 nfs_sync_mapping(filp->f_mapping);
785 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
786 nfs_zap_mapping(inode, filp->f_mapping);
792 * Lock a (portion of) a file
794 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
796 struct inode *inode = filp->f_mapping->host;
800 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
801 filp, fl->fl_type, fl->fl_flags,
802 (long long)fl->fl_start, (long long)fl->fl_end);
804 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
806 /* No mandatory locks over NFS */
807 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
810 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
813 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
814 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
820 ret = do_getlk(filp, cmd, fl, is_local);
821 else if (fl->fl_type == F_UNLCK)
822 ret = do_unlk(filp, cmd, fl, is_local);
824 ret = do_setlk(filp, cmd, fl, is_local);
828 EXPORT_SYMBOL_GPL(nfs_lock);
831 * Lock a (portion of) a file
833 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
835 struct inode *inode = filp->f_mapping->host;
838 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
839 filp, fl->fl_type, fl->fl_flags);
841 if (!(fl->fl_flags & FL_FLOCK))
845 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
846 * any standard. In principle we might be able to support LOCK_MAND
847 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
848 * NFS code is not set up for it.
850 if (fl->fl_type & LOCK_MAND)
853 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
856 /* We're simulating flock() locks using posix locks on the server */
857 if (fl->fl_type == F_UNLCK)
858 return do_unlk(filp, cmd, fl, is_local);
859 return do_setlk(filp, cmd, fl, is_local);
861 EXPORT_SYMBOL_GPL(nfs_flock);
863 const struct file_operations nfs_file_operations = {
864 .llseek = nfs_file_llseek,
865 .read_iter = nfs_file_read,
866 .write_iter = nfs_file_write,
867 .mmap = nfs_file_mmap,
868 .open = nfs_file_open,
869 .flush = nfs_file_flush,
870 .release = nfs_file_release,
871 .fsync = nfs_file_fsync,
874 .splice_read = nfs_file_splice_read,
875 .splice_write = iter_file_splice_write,
876 .check_flags = nfs_check_flags,
877 .setlease = simple_nosetlease,
879 EXPORT_SYMBOL_GPL(nfs_file_operations);