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
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);
215 ret2 = file_check_and_advance_wb_err(file);
222 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
224 struct nfs_open_context *ctx = nfs_file_open_context(file);
225 struct inode *inode = file_inode(file);
228 trace_nfs_fsync_enter(inode);
231 ret = file_write_and_wait_range(file, start, end);
234 ret = nfs_file_fsync_commit(file, datasync);
237 ret = pnfs_sync_inode(inode, !!datasync);
240 if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
243 * If nfs_file_fsync_commit detected a server reboot, then
244 * resend all dirty pages that might have been covered by
245 * the NFS_CONTEXT_RESEND_WRITES flag
251 trace_nfs_fsync_exit(inode, ret);
254 EXPORT_SYMBOL_GPL(nfs_file_fsync);
257 * Decide whether a read/modify/write cycle may be more efficient
258 * then a modify/write/read cycle when writing to a page in the
261 * Some pNFS layout drivers can only read/write at a certain block
262 * granularity like all block devices and therefore we must perform
263 * read/modify/write whenever a page hasn't read yet and the data
264 * to be written there is not aligned to a block boundary and/or
265 * smaller than the block size.
267 * The modify/write/read cycle may occur if a page is read before
268 * being completely filled by the writer. In this situation, the
269 * page must be completely written to stable storage on the server
270 * before it can be refilled by reading in the page from the server.
271 * This can lead to expensive, small, FILE_SYNC mode writes being
274 * It may be more efficient to read the page first if the file is
275 * open for reading in addition to writing, the page is not marked
276 * as Uptodate, it is not dirty or waiting to be committed,
277 * indicating that it was previously allocated and then modified,
278 * that there were valid bytes of data in that range of the file,
279 * and that the new data won't completely replace the old data in
280 * that range of the file.
282 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
284 unsigned int pglen = nfs_page_length(page);
285 unsigned int offset = pos & (PAGE_SIZE - 1);
286 unsigned int end = offset + len;
288 return !pglen || (end >= pglen && !offset);
291 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
292 loff_t pos, unsigned int len)
295 * Up-to-date pages, those with ongoing or full-page write
296 * don't need read/modify/write
298 if (PageUptodate(page) || PagePrivate(page) ||
299 nfs_full_page_write(page, pos, len))
302 if (pnfs_ld_read_whole_page(file->f_mapping->host))
304 /* Open for reading too? */
305 if (file->f_mode & FMODE_READ)
311 * This does the "real" work of the write. We must allocate and lock the
312 * page to be sent back to the generic routine, which then copies the
313 * data from user space.
315 * If the writer ends up delaying the write, the writer needs to
316 * increment the page use counts until he is done with the page.
318 static int nfs_write_begin(struct file *file, struct address_space *mapping,
319 loff_t pos, unsigned len,
320 struct page **pagep, void **fsdata)
323 pgoff_t index = pos >> PAGE_SHIFT;
327 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
328 file, mapping->host->i_ino, len, (long long) pos);
331 page = grab_cache_page_write_begin(mapping, index);
336 ret = nfs_flush_incompatible(file, page);
340 } else if (!once_thru &&
341 nfs_want_read_modify_write(file, page, pos, len)) {
343 ret = nfs_read_folio(file, page_folio(page));
351 static int nfs_write_end(struct file *file, struct address_space *mapping,
352 loff_t pos, unsigned len, unsigned copied,
353 struct page *page, void *fsdata)
355 unsigned offset = pos & (PAGE_SIZE - 1);
356 struct nfs_open_context *ctx = nfs_file_open_context(file);
359 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
360 file, mapping->host->i_ino, len, (long long) pos);
363 * Zero any uninitialised parts of the page, and then mark the page
364 * as up to date if it turns out that we're extending the file.
366 if (!PageUptodate(page)) {
367 unsigned pglen = nfs_page_length(page);
368 unsigned end = offset + copied;
371 zero_user_segments(page, 0, offset,
373 SetPageUptodate(page);
374 } else if (end >= pglen) {
375 zero_user_segment(page, end, PAGE_SIZE);
377 SetPageUptodate(page);
379 zero_user_segment(page, pglen, PAGE_SIZE);
382 status = nfs_updatepage(file, page, offset, copied);
389 NFS_I(mapping->host)->write_io += copied;
391 if (nfs_ctx_key_to_expire(ctx, mapping->host))
392 nfs_wb_all(mapping->host);
398 * Partially or wholly invalidate a page
399 * - Release the private state associated with a page if undergoing complete
401 * - Called if either PG_private or PG_fscache is set on the page
402 * - Caller holds page lock
404 static void nfs_invalidate_folio(struct folio *folio, size_t offset,
407 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
408 folio->index, offset, length);
410 if (offset != 0 || length < folio_size(folio))
412 /* Cancel any unstarted writes on this page */
413 nfs_wb_folio_cancel(folio->mapping->host, folio);
414 folio_wait_fscache(folio);
418 * Attempt to release the private state associated with a folio
419 * - Called if either private or fscache flags are set on the folio
420 * - Caller holds folio lock
421 * - Return true (may release folio) or false (may not)
423 static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
425 dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
427 /* If the private flag is set, then the folio is not freeable */
428 if (folio_test_private(folio))
430 return nfs_fscache_release_folio(folio, gfp);
433 static void nfs_check_dirty_writeback(struct folio *folio,
434 bool *dirty, bool *writeback)
436 struct nfs_inode *nfsi;
437 struct address_space *mapping = folio->mapping;
440 * Check if an unstable folio is currently being committed and
441 * if so, have the VM treat it as if the folio is under writeback
442 * so it will not block due to folios that will shortly be freeable.
444 nfsi = NFS_I(mapping->host);
445 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
451 * If the private flag is set, then the folio is not freeable
452 * and as the inode is not being committed, it's not going to
453 * be cleaned in the near future so treat it as dirty
455 if (folio_test_private(folio))
460 * Attempt to clear the private state associated with a page when an error
461 * occurs that requires the cached contents of an inode to be written back or
463 * - Called if either PG_private or fscache is set on the page
464 * - Caller holds page lock
465 * - Return 0 if successful, -error otherwise
467 static int nfs_launder_folio(struct folio *folio)
469 struct inode *inode = folio->mapping->host;
471 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
472 inode->i_ino, folio_pos(folio));
474 folio_wait_fscache(folio);
475 return nfs_wb_page(inode, &folio->page);
478 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
481 unsigned long blocks;
484 struct inode *inode = file_inode(file);
485 struct rpc_clnt *clnt = NFS_CLIENT(inode);
486 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
488 spin_lock(&inode->i_lock);
489 blocks = inode->i_blocks;
490 isize = inode->i_size;
491 spin_unlock(&inode->i_lock);
492 if (blocks*512 < isize) {
493 pr_warn("swap activate: swapfile has holes\n");
497 ret = rpc_clnt_swap_activate(clnt);
500 ret = add_swap_extent(sis, 0, sis->max, 0);
502 rpc_clnt_swap_deactivate(clnt);
508 if (cl->rpc_ops->enable_swap)
509 cl->rpc_ops->enable_swap(inode);
511 sis->flags |= SWP_FS_OPS;
515 static void nfs_swap_deactivate(struct file *file)
517 struct inode *inode = file_inode(file);
518 struct rpc_clnt *clnt = NFS_CLIENT(inode);
519 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
521 rpc_clnt_swap_deactivate(clnt);
522 if (cl->rpc_ops->disable_swap)
523 cl->rpc_ops->disable_swap(file_inode(file));
526 const struct address_space_operations nfs_file_aops = {
527 .read_folio = nfs_read_folio,
528 .readahead = nfs_readahead,
529 .dirty_folio = filemap_dirty_folio,
530 .writepage = nfs_writepage,
531 .writepages = nfs_writepages,
532 .write_begin = nfs_write_begin,
533 .write_end = nfs_write_end,
534 .invalidate_folio = nfs_invalidate_folio,
535 .release_folio = nfs_release_folio,
536 .migrate_folio = nfs_migrate_folio,
537 .launder_folio = nfs_launder_folio,
538 .is_dirty_writeback = nfs_check_dirty_writeback,
539 .error_remove_page = generic_error_remove_page,
540 .swap_activate = nfs_swap_activate,
541 .swap_deactivate = nfs_swap_deactivate,
542 .swap_rw = nfs_swap_rw,
546 * Notification that a PTE pointing to an NFS page is about to be made
547 * writable, implying that someone is about to modify the page through a
548 * shared-writable mapping
550 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
552 struct page *page = vmf->page;
553 struct file *filp = vmf->vma->vm_file;
554 struct inode *inode = file_inode(filp);
556 vm_fault_t ret = VM_FAULT_NOPAGE;
557 struct address_space *mapping;
559 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
560 filp, filp->f_mapping->host->i_ino,
561 (long long)page_offset(page));
563 sb_start_pagefault(inode->i_sb);
565 /* make sure the cache has finished storing the page */
566 if (PageFsCache(page) &&
567 wait_on_page_fscache_killable(vmf->page) < 0) {
568 ret = VM_FAULT_RETRY;
572 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
573 nfs_wait_bit_killable, TASK_KILLABLE);
576 mapping = page_file_mapping(page);
577 if (mapping != inode->i_mapping)
580 wait_on_page_writeback(page);
582 pagelen = nfs_page_length(page);
586 ret = VM_FAULT_LOCKED;
587 if (nfs_flush_incompatible(filp, page) == 0 &&
588 nfs_updatepage(filp, page, 0, pagelen) == 0)
591 ret = VM_FAULT_SIGBUS;
595 sb_end_pagefault(inode->i_sb);
599 static const struct vm_operations_struct nfs_file_vm_ops = {
600 .fault = filemap_fault,
601 .map_pages = filemap_map_pages,
602 .page_mkwrite = nfs_vm_page_mkwrite,
605 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
607 struct file *file = iocb->ki_filp;
608 struct inode *inode = file_inode(file);
609 unsigned int mntflags = NFS_SERVER(inode)->flags;
610 ssize_t result, written;
614 result = nfs_key_timeout_notify(file, inode);
618 if (iocb->ki_flags & IOCB_DIRECT)
619 return nfs_file_direct_write(iocb, from, false);
621 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
622 file, iov_iter_count(from), (long long) iocb->ki_pos);
624 if (IS_SWAPFILE(inode))
627 * O_APPEND implies that we must revalidate the file length.
629 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
630 result = nfs_revalidate_file_size(inode, file);
635 nfs_clear_invalid_mapping(file->f_mapping);
637 since = filemap_sample_wb_err(file->f_mapping);
638 nfs_start_io_write(inode);
639 result = generic_write_checks(iocb, from);
641 current->backing_dev_info = inode_to_bdi(inode);
642 result = generic_perform_write(iocb, from);
643 current->backing_dev_info = NULL;
645 nfs_end_io_write(inode);
650 iocb->ki_pos += written;
651 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
653 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
654 result = filemap_fdatawrite_range(file->f_mapping,
655 iocb->ki_pos - written,
660 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
661 result = filemap_fdatawait_range(file->f_mapping,
662 iocb->ki_pos - written,
667 result = generic_write_sync(iocb, written);
672 /* Return error values */
673 error = filemap_check_wb_err(file->f_mapping, since);
681 error = file_check_and_advance_wb_err(file);
688 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
691 EXPORT_SYMBOL_GPL(nfs_file_write);
694 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
696 struct inode *inode = filp->f_mapping->host;
698 unsigned int saved_type = fl->fl_type;
700 /* Try local locking first */
701 posix_test_lock(filp, fl);
702 if (fl->fl_type != F_UNLCK) {
703 /* found a conflict */
706 fl->fl_type = saved_type;
708 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
714 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
718 fl->fl_type = F_UNLCK;
723 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
725 struct inode *inode = filp->f_mapping->host;
726 struct nfs_lock_context *l_ctx;
730 * Flush all pending writes before doing anything
735 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
736 if (!IS_ERR(l_ctx)) {
737 status = nfs_iocounter_wait(l_ctx);
738 nfs_put_lock_context(l_ctx);
739 /* NOTE: special case
740 * If we're signalled while cleaning up locks on process exit, we
741 * still need to complete the unlock.
743 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
748 * Use local locking if mounted with "-onolock" or with appropriate
752 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
754 status = locks_lock_file_wait(filp, fl);
759 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
761 struct inode *inode = filp->f_mapping->host;
765 * Flush all pending writes before doing anything
768 status = nfs_sync_mapping(filp->f_mapping);
773 * Use local locking if mounted with "-onolock" or with appropriate
777 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
779 status = locks_lock_file_wait(filp, fl);
784 * Invalidate cache to prevent missing any changes. If
785 * the file is mapped, clear the page cache as well so
786 * those mappings will be loaded.
788 * This makes locking act as a cache coherency point.
790 nfs_sync_mapping(filp->f_mapping);
791 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
792 nfs_zap_caches(inode);
793 if (mapping_mapped(filp->f_mapping))
794 nfs_revalidate_mapping(inode, filp->f_mapping);
801 * Lock a (portion of) a file
803 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
805 struct inode *inode = filp->f_mapping->host;
809 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
810 filp, fl->fl_type, fl->fl_flags,
811 (long long)fl->fl_start, (long long)fl->fl_end);
813 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
815 if (fl->fl_flags & FL_RECLAIM)
818 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
821 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
822 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
828 ret = do_getlk(filp, cmd, fl, is_local);
829 else if (fl->fl_type == F_UNLCK)
830 ret = do_unlk(filp, cmd, fl, is_local);
832 ret = do_setlk(filp, cmd, fl, is_local);
836 EXPORT_SYMBOL_GPL(nfs_lock);
839 * Lock a (portion of) a file
841 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
843 struct inode *inode = filp->f_mapping->host;
846 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
847 filp, fl->fl_type, fl->fl_flags);
849 if (!(fl->fl_flags & FL_FLOCK))
852 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
855 /* We're simulating flock() locks using posix locks on the server */
856 if (fl->fl_type == F_UNLCK)
857 return do_unlk(filp, cmd, fl, is_local);
858 return do_setlk(filp, cmd, fl, is_local);
860 EXPORT_SYMBOL_GPL(nfs_flock);
862 const struct file_operations nfs_file_operations = {
863 .llseek = nfs_file_llseek,
864 .read_iter = nfs_file_read,
865 .write_iter = nfs_file_write,
866 .mmap = nfs_file_mmap,
867 .open = nfs_file_open,
868 .flush = nfs_file_flush,
869 .release = nfs_file_release,
870 .fsync = nfs_file_fsync,
873 .splice_read = generic_file_splice_read,
874 .splice_write = iter_file_splice_write,
875 .check_flags = nfs_check_flags,
876 .setlease = simple_nosetlease,
878 EXPORT_SYMBOL_GPL(nfs_file_operations);