1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1992 Rick Sladkey
7 * Changes Copyright (C) 1994 by Florian La Roche
8 * - Do not copy data too often around in the kernel.
9 * - In nfs_file_read the return value of kmalloc wasn't checked.
10 * - Put in a better version of read look-ahead buffering. Original idea
11 * and implementation by Wai S Kok elekokws@ee.nus.sg.
13 * Expire cache on write to a file by Wai S Kok (Oct 1994).
15 * Total rewrite of read side for new NFS buffer cache.. Linus.
17 * nfs regular file handling functions
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/fcntl.h>
25 #include <linux/stat.h>
26 #include <linux/nfs_fs.h>
27 #include <linux/nfs_mount.h>
29 #include <linux/pagemap.h>
30 #include <linux/gfp.h>
31 #include <linux/swap.h>
33 #include <linux/uaccess.h>
35 #include "delegation.h"
43 #define NFSDBG_FACILITY NFSDBG_FILE
45 static const struct vm_operations_struct nfs_file_vm_ops;
47 int nfs_check_flags(int flags)
49 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
54 EXPORT_SYMBOL_GPL(nfs_check_flags);
60 nfs_file_open(struct inode *inode, struct file *filp)
64 dprintk("NFS: open file(%pD2)\n", filp);
66 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
67 res = nfs_check_flags(filp->f_flags);
71 res = nfs_open(inode, filp);
73 filp->f_mode |= FMODE_CAN_ODIRECT;
78 nfs_file_release(struct inode *inode, struct file *filp)
80 dprintk("NFS: release(%pD2)\n", filp);
82 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
83 nfs_file_clear_open_context(filp);
84 nfs_fscache_release_file(inode, filp);
87 EXPORT_SYMBOL_GPL(nfs_file_release);
90 * nfs_revalidate_file_size - Revalidate the file size
91 * @inode: pointer to inode struct
92 * @filp: pointer to struct file
94 * Revalidates the file length. This is basically a wrapper around
95 * nfs_revalidate_inode() that takes into account the fact that we may
96 * have cached writes (in which case we don't care about the server's
97 * idea of what the file length is), or O_DIRECT (in which case we
98 * shouldn't trust the cache).
100 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
102 struct nfs_server *server = NFS_SERVER(inode);
104 if (filp->f_flags & O_DIRECT)
106 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE))
110 return __nfs_revalidate_inode(server, inode);
113 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
115 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
116 filp, offset, whence);
119 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
120 * the cached file length
122 if (whence != SEEK_SET && whence != SEEK_CUR) {
123 struct inode *inode = filp->f_mapping->host;
125 int retval = nfs_revalidate_file_size(inode, filp);
127 return (loff_t)retval;
130 return generic_file_llseek(filp, offset, whence);
132 EXPORT_SYMBOL_GPL(nfs_file_llseek);
135 * Flush all dirty pages, and check for write errors.
138 nfs_file_flush(struct file *file, fl_owner_t id)
140 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 since = filemap_sample_wb_err(file->f_mapping);
152 return filemap_check_wb_err(file->f_mapping, since);
156 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
158 struct inode *inode = file_inode(iocb->ki_filp);
161 if (iocb->ki_flags & IOCB_DIRECT)
162 return nfs_file_direct_read(iocb, to, false);
164 dprintk("NFS: read(%pD2, %zu@%lu)\n",
166 iov_iter_count(to), (unsigned long) iocb->ki_pos);
168 nfs_start_io_read(inode);
169 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
171 result = generic_file_read_iter(iocb, to);
173 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
175 nfs_end_io_read(inode);
178 EXPORT_SYMBOL_GPL(nfs_file_read);
181 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
183 struct inode *inode = file_inode(file);
186 dprintk("NFS: mmap(%pD2)\n", file);
188 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
189 * so we call that before revalidating the mapping
191 status = generic_file_mmap(file, vma);
193 vma->vm_ops = &nfs_file_vm_ops;
194 status = nfs_revalidate_mapping(inode, file->f_mapping);
198 EXPORT_SYMBOL_GPL(nfs_file_mmap);
201 * Flush any dirty pages for this process, and check for write errors.
202 * The return status from this call provides a reliable indication of
203 * whether any write errors occurred for this process.
206 nfs_file_fsync_commit(struct file *file, int datasync)
208 struct inode *inode = file_inode(file);
211 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
213 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
214 ret = nfs_commit_inode(inode, FLUSH_SYNC);
217 return file_check_and_advance_wb_err(file);
221 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
223 struct nfs_open_context *ctx = nfs_file_open_context(file);
224 struct inode *inode = file_inode(file);
227 trace_nfs_fsync_enter(inode);
230 ret = file_write_and_wait_range(file, start, end);
233 ret = nfs_file_fsync_commit(file, datasync);
236 ret = pnfs_sync_inode(inode, !!datasync);
239 if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
242 * If nfs_file_fsync_commit detected a server reboot, then
243 * resend all dirty pages that might have been covered by
244 * the NFS_CONTEXT_RESEND_WRITES flag
250 trace_nfs_fsync_exit(inode, ret);
253 EXPORT_SYMBOL_GPL(nfs_file_fsync);
256 * Decide whether a read/modify/write cycle may be more efficient
257 * then a modify/write/read cycle when writing to a page in the
260 * Some pNFS layout drivers can only read/write at a certain block
261 * granularity like all block devices and therefore we must perform
262 * read/modify/write whenever a page hasn't read yet and the data
263 * to be written there is not aligned to a block boundary and/or
264 * smaller than the block size.
266 * The modify/write/read cycle may occur if a page is read before
267 * being completely filled by the writer. In this situation, the
268 * page must be completely written to stable storage on the server
269 * before it can be refilled by reading in the page from the server.
270 * This can lead to expensive, small, FILE_SYNC mode writes being
273 * It may be more efficient to read the page first if the file is
274 * open for reading in addition to writing, the page is not marked
275 * as Uptodate, it is not dirty or waiting to be committed,
276 * indicating that it was previously allocated and then modified,
277 * that there were valid bytes of data in that range of the file,
278 * and that the new data won't completely replace the old data in
279 * that range of the file.
281 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
283 unsigned int pglen = nfs_page_length(page);
284 unsigned int offset = pos & (PAGE_SIZE - 1);
285 unsigned int end = offset + len;
287 return !pglen || (end >= pglen && !offset);
290 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
291 loff_t pos, unsigned int len)
294 * Up-to-date pages, those with ongoing or full-page write
295 * don't need read/modify/write
297 if (PageUptodate(page) || PagePrivate(page) ||
298 nfs_full_page_write(page, pos, len))
301 if (pnfs_ld_read_whole_page(file->f_mapping->host))
303 /* Open for reading too? */
304 if (file->f_mode & FMODE_READ)
310 * This does the "real" work of the write. We must allocate and lock the
311 * page to be sent back to the generic routine, which then copies the
312 * data from user space.
314 * If the writer ends up delaying the write, the writer needs to
315 * increment the page use counts until he is done with the page.
317 static int nfs_write_begin(struct file *file, struct address_space *mapping,
318 loff_t pos, unsigned len, unsigned flags,
319 struct page **pagep, void **fsdata)
322 pgoff_t index = pos >> PAGE_SHIFT;
326 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
327 file, mapping->host->i_ino, len, (long long) pos);
330 page = grab_cache_page_write_begin(mapping, index, flags);
335 ret = nfs_flush_incompatible(file, page);
339 } else if (!once_thru &&
340 nfs_want_read_modify_write(file, page, pos, len)) {
342 ret = nfs_readpage(file, page);
350 static int nfs_write_end(struct file *file, struct address_space *mapping,
351 loff_t pos, unsigned len, unsigned copied,
352 struct page *page, void *fsdata)
354 unsigned offset = pos & (PAGE_SIZE - 1);
355 struct nfs_open_context *ctx = nfs_file_open_context(file);
358 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
359 file, mapping->host->i_ino, len, (long long) pos);
362 * Zero any uninitialised parts of the page, and then mark the page
363 * as up to date if it turns out that we're extending the file.
365 if (!PageUptodate(page)) {
366 unsigned pglen = nfs_page_length(page);
367 unsigned end = offset + copied;
370 zero_user_segments(page, 0, offset,
372 SetPageUptodate(page);
373 } else if (end >= pglen) {
374 zero_user_segment(page, end, PAGE_SIZE);
376 SetPageUptodate(page);
378 zero_user_segment(page, pglen, PAGE_SIZE);
381 status = nfs_updatepage(file, page, offset, copied);
388 NFS_I(mapping->host)->write_io += copied;
390 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
391 status = nfs_wb_all(mapping->host);
400 * Partially or wholly invalidate a page
401 * - Release the private state associated with a page if undergoing complete
403 * - Called if either PG_private or PG_fscache is set on the page
404 * - Caller holds page lock
406 static void nfs_invalidate_folio(struct folio *folio, size_t offset,
409 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
410 folio->index, offset, length);
412 if (offset != 0 || length < folio_size(folio))
414 /* Cancel any unstarted writes on this page */
415 nfs_wb_folio_cancel(folio->mapping->host, folio);
416 folio_wait_fscache(folio);
420 * Attempt to release the private state associated with a page
421 * - Called if either PG_private or PG_fscache is set on the page
422 * - Caller holds page lock
423 * - Return true (may release page) or false (may not)
425 static int nfs_release_page(struct page *page, gfp_t gfp)
427 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
429 /* If PagePrivate() is set, then the page is not freeable */
430 if (PagePrivate(page))
432 return nfs_fscache_release_page(page, gfp);
435 static void nfs_check_dirty_writeback(struct page *page,
436 bool *dirty, bool *writeback)
438 struct nfs_inode *nfsi;
439 struct address_space *mapping = page_file_mapping(page);
441 if (!mapping || PageSwapCache(page))
445 * Check if an unstable page is currently being committed and
446 * if so, have the VM treat it as if the page is under writeback
447 * so it will not block due to pages that will shortly be freeable.
449 nfsi = NFS_I(mapping->host);
450 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
456 * If PagePrivate() is set, then the page is not freeable and as the
457 * inode is not being committed, it's not going to be cleaned in the
458 * near future so treat it as dirty
460 if (PagePrivate(page))
465 * Attempt to clear the private state associated with a page when an error
466 * occurs that requires the cached contents of an inode to be written back or
468 * - Called if either PG_private or fscache is set on the page
469 * - Caller holds page lock
470 * - Return 0 if successful, -error otherwise
472 static int nfs_launder_folio(struct folio *folio)
474 struct inode *inode = folio->mapping->host;
476 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
477 inode->i_ino, folio_pos(folio));
479 folio_wait_fscache(folio);
480 return nfs_wb_page(inode, &folio->page);
483 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
486 unsigned long blocks;
489 struct inode *inode = file_inode(file);
490 struct rpc_clnt *clnt = NFS_CLIENT(inode);
491 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
493 spin_lock(&inode->i_lock);
494 blocks = inode->i_blocks;
495 isize = inode->i_size;
496 spin_unlock(&inode->i_lock);
497 if (blocks*512 < isize) {
498 pr_warn("swap activate: swapfile has holes\n");
502 ret = rpc_clnt_swap_activate(clnt);
505 ret = add_swap_extent(sis, 0, sis->max, 0);
507 rpc_clnt_swap_deactivate(clnt);
513 if (cl->rpc_ops->enable_swap)
514 cl->rpc_ops->enable_swap(inode);
516 sis->flags |= SWP_FS_OPS;
520 static void nfs_swap_deactivate(struct file *file)
522 struct inode *inode = file_inode(file);
523 struct rpc_clnt *clnt = NFS_CLIENT(inode);
524 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
526 rpc_clnt_swap_deactivate(clnt);
527 if (cl->rpc_ops->disable_swap)
528 cl->rpc_ops->disable_swap(file_inode(file));
531 const struct address_space_operations nfs_file_aops = {
532 .readpage = nfs_readpage,
533 .readahead = nfs_readahead,
534 .dirty_folio = filemap_dirty_folio,
535 .writepage = nfs_writepage,
536 .writepages = nfs_writepages,
537 .write_begin = nfs_write_begin,
538 .write_end = nfs_write_end,
539 .invalidate_folio = nfs_invalidate_folio,
540 .releasepage = nfs_release_page,
541 #ifdef CONFIG_MIGRATION
542 .migratepage = nfs_migrate_page,
544 .launder_folio = nfs_launder_folio,
545 .is_dirty_writeback = nfs_check_dirty_writeback,
546 .error_remove_page = generic_error_remove_page,
547 .swap_activate = nfs_swap_activate,
548 .swap_deactivate = nfs_swap_deactivate,
549 .swap_rw = nfs_swap_rw,
553 * Notification that a PTE pointing to an NFS page is about to be made
554 * writable, implying that someone is about to modify the page through a
555 * shared-writable mapping
557 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
559 struct page *page = vmf->page;
560 struct file *filp = vmf->vma->vm_file;
561 struct inode *inode = file_inode(filp);
563 vm_fault_t ret = VM_FAULT_NOPAGE;
564 struct address_space *mapping;
566 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
567 filp, filp->f_mapping->host->i_ino,
568 (long long)page_offset(page));
570 sb_start_pagefault(inode->i_sb);
572 /* make sure the cache has finished storing the page */
573 if (PageFsCache(page) &&
574 wait_on_page_fscache_killable(vmf->page) < 0) {
575 ret = VM_FAULT_RETRY;
579 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
580 nfs_wait_bit_killable, TASK_KILLABLE);
583 mapping = page_file_mapping(page);
584 if (mapping != inode->i_mapping)
587 wait_on_page_writeback(page);
589 pagelen = nfs_page_length(page);
593 ret = VM_FAULT_LOCKED;
594 if (nfs_flush_incompatible(filp, page) == 0 &&
595 nfs_updatepage(filp, page, 0, pagelen) == 0)
598 ret = VM_FAULT_SIGBUS;
602 sb_end_pagefault(inode->i_sb);
606 static const struct vm_operations_struct nfs_file_vm_ops = {
607 .fault = filemap_fault,
608 .map_pages = filemap_map_pages,
609 .page_mkwrite = nfs_vm_page_mkwrite,
612 static int nfs_need_check_write(struct file *filp, struct inode *inode,
615 struct nfs_open_context *ctx;
617 ctx = nfs_file_open_context(filp);
618 if (nfs_error_is_fatal_on_server(error) ||
619 nfs_ctx_key_to_expire(ctx, inode))
624 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
626 struct file *file = iocb->ki_filp;
627 struct inode *inode = file_inode(file);
628 unsigned int mntflags = NFS_SERVER(inode)->flags;
629 ssize_t result, written;
633 result = nfs_key_timeout_notify(file, inode);
637 if (iocb->ki_flags & IOCB_DIRECT)
638 return nfs_file_direct_write(iocb, from, false);
640 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
641 file, iov_iter_count(from), (long long) iocb->ki_pos);
643 if (IS_SWAPFILE(inode))
646 * O_APPEND implies that we must revalidate the file length.
648 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
649 result = nfs_revalidate_file_size(inode, file);
654 nfs_clear_invalid_mapping(file->f_mapping);
656 since = filemap_sample_wb_err(file->f_mapping);
657 nfs_start_io_write(inode);
658 result = generic_write_checks(iocb, from);
660 current->backing_dev_info = inode_to_bdi(inode);
661 result = generic_perform_write(iocb, from);
662 current->backing_dev_info = NULL;
664 nfs_end_io_write(inode);
669 iocb->ki_pos += written;
671 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
672 result = filemap_fdatawrite_range(file->f_mapping,
673 iocb->ki_pos - written,
678 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
679 result = filemap_fdatawait_range(file->f_mapping,
680 iocb->ki_pos - written,
685 result = generic_write_sync(iocb, written);
689 /* Return error values */
690 error = filemap_check_wb_err(file->f_mapping, since);
691 if (nfs_need_check_write(file, inode, error)) {
692 int err = nfs_wb_all(inode);
696 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
701 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
704 EXPORT_SYMBOL_GPL(nfs_file_write);
707 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
709 struct inode *inode = filp->f_mapping->host;
711 unsigned int saved_type = fl->fl_type;
713 /* Try local locking first */
714 posix_test_lock(filp, fl);
715 if (fl->fl_type != F_UNLCK) {
716 /* found a conflict */
719 fl->fl_type = saved_type;
721 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
727 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
731 fl->fl_type = F_UNLCK;
736 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
738 struct inode *inode = filp->f_mapping->host;
739 struct nfs_lock_context *l_ctx;
743 * Flush all pending writes before doing anything
748 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
749 if (!IS_ERR(l_ctx)) {
750 status = nfs_iocounter_wait(l_ctx);
751 nfs_put_lock_context(l_ctx);
752 /* NOTE: special case
753 * If we're signalled while cleaning up locks on process exit, we
754 * still need to complete the unlock.
756 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
761 * Use local locking if mounted with "-onolock" or with appropriate
765 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
767 status = locks_lock_file_wait(filp, fl);
772 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
774 struct inode *inode = filp->f_mapping->host;
778 * Flush all pending writes before doing anything
781 status = nfs_sync_mapping(filp->f_mapping);
786 * Use local locking if mounted with "-onolock" or with appropriate
790 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
792 status = locks_lock_file_wait(filp, fl);
797 * Invalidate cache to prevent missing any changes. If
798 * the file is mapped, clear the page cache as well so
799 * those mappings will be loaded.
801 * This makes locking act as a cache coherency point.
803 nfs_sync_mapping(filp->f_mapping);
804 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
805 nfs_zap_caches(inode);
806 if (mapping_mapped(filp->f_mapping))
807 nfs_revalidate_mapping(inode, filp->f_mapping);
814 * Lock a (portion of) a file
816 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
818 struct inode *inode = filp->f_mapping->host;
822 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
823 filp, fl->fl_type, fl->fl_flags,
824 (long long)fl->fl_start, (long long)fl->fl_end);
826 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
828 if (fl->fl_flags & FL_RECLAIM)
831 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
834 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
835 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
841 ret = do_getlk(filp, cmd, fl, is_local);
842 else if (fl->fl_type == F_UNLCK)
843 ret = do_unlk(filp, cmd, fl, is_local);
845 ret = do_setlk(filp, cmd, fl, is_local);
849 EXPORT_SYMBOL_GPL(nfs_lock);
852 * Lock a (portion of) a file
854 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
856 struct inode *inode = filp->f_mapping->host;
859 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
860 filp, fl->fl_type, fl->fl_flags);
862 if (!(fl->fl_flags & FL_FLOCK))
865 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
868 /* We're simulating flock() locks using posix locks on the server */
869 if (fl->fl_type == F_UNLCK)
870 return do_unlk(filp, cmd, fl, is_local);
871 return do_setlk(filp, cmd, fl, is_local);
873 EXPORT_SYMBOL_GPL(nfs_flock);
875 const struct file_operations nfs_file_operations = {
876 .llseek = nfs_file_llseek,
877 .read_iter = nfs_file_read,
878 .write_iter = nfs_file_write,
879 .mmap = nfs_file_mmap,
880 .open = nfs_file_open,
881 .flush = nfs_file_flush,
882 .release = nfs_file_release,
883 .fsync = nfs_file_fsync,
886 .splice_read = generic_file_splice_read,
887 .splice_write = iter_file_splice_write,
888 .check_flags = nfs_check_flags,
889 .setlease = simple_nosetlease,
891 EXPORT_SYMBOL_GPL(nfs_file_operations);