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);
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 inode *inode = file_inode(file);
225 struct nfs_inode *nfsi = NFS_I(inode);
226 long save_nredirtied = atomic_long_read(&nfsi->redirtied_pages);
230 trace_nfs_fsync_enter(inode);
233 ret = file_write_and_wait_range(file, start, end);
236 ret = nfs_file_fsync_commit(file, datasync);
239 ret = pnfs_sync_inode(inode, !!datasync);
242 nredirtied = atomic_long_read(&nfsi->redirtied_pages);
243 if (nredirtied == save_nredirtied)
245 save_nredirtied = nredirtied;
248 trace_nfs_fsync_exit(inode, ret);
251 EXPORT_SYMBOL_GPL(nfs_file_fsync);
254 * Decide whether a read/modify/write cycle may be more efficient
255 * then a modify/write/read cycle when writing to a page in the
258 * Some pNFS layout drivers can only read/write at a certain block
259 * granularity like all block devices and therefore we must perform
260 * read/modify/write whenever a page hasn't read yet and the data
261 * to be written there is not aligned to a block boundary and/or
262 * smaller than the block size.
264 * The modify/write/read cycle may occur if a page is read before
265 * being completely filled by the writer. In this situation, the
266 * page must be completely written to stable storage on the server
267 * before it can be refilled by reading in the page from the server.
268 * This can lead to expensive, small, FILE_SYNC mode writes being
271 * It may be more efficient to read the page first if the file is
272 * open for reading in addition to writing, the page is not marked
273 * as Uptodate, it is not dirty or waiting to be committed,
274 * indicating that it was previously allocated and then modified,
275 * that there were valid bytes of data in that range of the file,
276 * and that the new data won't completely replace the old data in
277 * that range of the file.
279 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
281 unsigned int pglen = nfs_page_length(page);
282 unsigned int offset = pos & (PAGE_SIZE - 1);
283 unsigned int end = offset + len;
285 return !pglen || (end >= pglen && !offset);
288 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
289 loff_t pos, unsigned int len)
292 * Up-to-date pages, those with ongoing or full-page write
293 * don't need read/modify/write
295 if (PageUptodate(page) || PagePrivate(page) ||
296 nfs_full_page_write(page, pos, len))
299 if (pnfs_ld_read_whole_page(file->f_mapping->host))
301 /* Open for reading too? */
302 if (file->f_mode & FMODE_READ)
308 * This does the "real" work of the write. We must allocate and lock the
309 * page to be sent back to the generic routine, which then copies the
310 * data from user space.
312 * If the writer ends up delaying the write, the writer needs to
313 * increment the page use counts until he is done with the page.
315 static int nfs_write_begin(struct file *file, struct address_space *mapping,
316 loff_t pos, unsigned len,
317 struct page **pagep, void **fsdata)
320 pgoff_t index = pos >> PAGE_SHIFT;
324 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
325 file, mapping->host->i_ino, len, (long long) pos);
328 page = grab_cache_page_write_begin(mapping, index);
333 ret = nfs_flush_incompatible(file, page);
337 } else if (!once_thru &&
338 nfs_want_read_modify_write(file, page, pos, len)) {
340 ret = nfs_read_folio(file, page_folio(page));
348 static int nfs_write_end(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned copied,
350 struct page *page, void *fsdata)
352 unsigned offset = pos & (PAGE_SIZE - 1);
353 struct nfs_open_context *ctx = nfs_file_open_context(file);
356 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
357 file, mapping->host->i_ino, len, (long long) pos);
360 * Zero any uninitialised parts of the page, and then mark the page
361 * as up to date if it turns out that we're extending the file.
363 if (!PageUptodate(page)) {
364 unsigned pglen = nfs_page_length(page);
365 unsigned end = offset + copied;
368 zero_user_segments(page, 0, offset,
370 SetPageUptodate(page);
371 } else if (end >= pglen) {
372 zero_user_segment(page, end, PAGE_SIZE);
374 SetPageUptodate(page);
376 zero_user_segment(page, pglen, PAGE_SIZE);
379 status = nfs_updatepage(file, page, offset, copied);
386 NFS_I(mapping->host)->write_io += copied;
388 if (nfs_ctx_key_to_expire(ctx, mapping->host))
389 nfs_wb_all(mapping->host);
395 * Partially or wholly invalidate a page
396 * - Release the private state associated with a page if undergoing complete
398 * - Called if either PG_private or PG_fscache is set on the page
399 * - Caller holds page lock
401 static void nfs_invalidate_folio(struct folio *folio, size_t offset,
404 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
405 folio->index, offset, length);
407 if (offset != 0 || length < folio_size(folio))
409 /* Cancel any unstarted writes on this page */
410 nfs_wb_folio_cancel(folio->mapping->host, folio);
411 folio_wait_fscache(folio);
415 * Attempt to release the private state associated with a folio
416 * - Called if either private or fscache flags are set on the folio
417 * - Caller holds folio lock
418 * - Return true (may release folio) or false (may not)
420 static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
422 dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
424 /* If the private flag is set, then the folio is not freeable */
425 if (folio_test_private(folio))
427 return nfs_fscache_release_folio(folio, gfp);
430 static void nfs_check_dirty_writeback(struct folio *folio,
431 bool *dirty, bool *writeback)
433 struct nfs_inode *nfsi;
434 struct address_space *mapping = folio->mapping;
437 * Check if an unstable folio is currently being committed and
438 * if so, have the VM treat it as if the folio is under writeback
439 * so it will not block due to folios that will shortly be freeable.
441 nfsi = NFS_I(mapping->host);
442 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
448 * If the private flag is set, then the folio is not freeable
449 * and as the inode is not being committed, it's not going to
450 * be cleaned in the near future so treat it as dirty
452 if (folio_test_private(folio))
457 * Attempt to clear the private state associated with a page when an error
458 * occurs that requires the cached contents of an inode to be written back or
460 * - Called if either PG_private or fscache is set on the page
461 * - Caller holds page lock
462 * - Return 0 if successful, -error otherwise
464 static int nfs_launder_folio(struct folio *folio)
466 struct inode *inode = folio->mapping->host;
468 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
469 inode->i_ino, folio_pos(folio));
471 folio_wait_fscache(folio);
472 return nfs_wb_page(inode, &folio->page);
475 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
478 unsigned long blocks;
481 struct inode *inode = file_inode(file);
482 struct rpc_clnt *clnt = NFS_CLIENT(inode);
483 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
485 spin_lock(&inode->i_lock);
486 blocks = inode->i_blocks;
487 isize = inode->i_size;
488 spin_unlock(&inode->i_lock);
489 if (blocks*512 < isize) {
490 pr_warn("swap activate: swapfile has holes\n");
494 ret = rpc_clnt_swap_activate(clnt);
497 ret = add_swap_extent(sis, 0, sis->max, 0);
499 rpc_clnt_swap_deactivate(clnt);
505 if (cl->rpc_ops->enable_swap)
506 cl->rpc_ops->enable_swap(inode);
508 sis->flags |= SWP_FS_OPS;
512 static void nfs_swap_deactivate(struct file *file)
514 struct inode *inode = file_inode(file);
515 struct rpc_clnt *clnt = NFS_CLIENT(inode);
516 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
518 rpc_clnt_swap_deactivate(clnt);
519 if (cl->rpc_ops->disable_swap)
520 cl->rpc_ops->disable_swap(file_inode(file));
523 const struct address_space_operations nfs_file_aops = {
524 .read_folio = nfs_read_folio,
525 .readahead = nfs_readahead,
526 .dirty_folio = filemap_dirty_folio,
527 .writepage = nfs_writepage,
528 .writepages = nfs_writepages,
529 .write_begin = nfs_write_begin,
530 .write_end = nfs_write_end,
531 .invalidate_folio = nfs_invalidate_folio,
532 .release_folio = nfs_release_folio,
533 .migrate_folio = nfs_migrate_folio,
534 .launder_folio = nfs_launder_folio,
535 .is_dirty_writeback = nfs_check_dirty_writeback,
536 .error_remove_page = generic_error_remove_page,
537 .swap_activate = nfs_swap_activate,
538 .swap_deactivate = nfs_swap_deactivate,
539 .swap_rw = nfs_swap_rw,
543 * Notification that a PTE pointing to an NFS page is about to be made
544 * writable, implying that someone is about to modify the page through a
545 * shared-writable mapping
547 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
549 struct page *page = vmf->page;
550 struct file *filp = vmf->vma->vm_file;
551 struct inode *inode = file_inode(filp);
553 vm_fault_t ret = VM_FAULT_NOPAGE;
554 struct address_space *mapping;
556 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
557 filp, filp->f_mapping->host->i_ino,
558 (long long)page_offset(page));
560 sb_start_pagefault(inode->i_sb);
562 /* make sure the cache has finished storing the page */
563 if (PageFsCache(page) &&
564 wait_on_page_fscache_killable(vmf->page) < 0) {
565 ret = VM_FAULT_RETRY;
569 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
570 nfs_wait_bit_killable,
571 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
574 mapping = page_file_mapping(page);
575 if (mapping != inode->i_mapping)
578 wait_on_page_writeback(page);
580 pagelen = nfs_page_length(page);
584 ret = VM_FAULT_LOCKED;
585 if (nfs_flush_incompatible(filp, page) == 0 &&
586 nfs_updatepage(filp, page, 0, pagelen) == 0)
589 ret = VM_FAULT_SIGBUS;
593 sb_end_pagefault(inode->i_sb);
597 static const struct vm_operations_struct nfs_file_vm_ops = {
598 .fault = filemap_fault,
599 .map_pages = filemap_map_pages,
600 .page_mkwrite = nfs_vm_page_mkwrite,
603 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
605 struct file *file = iocb->ki_filp;
606 struct inode *inode = file_inode(file);
607 unsigned int mntflags = NFS_SERVER(inode)->flags;
608 ssize_t result, written;
612 result = nfs_key_timeout_notify(file, inode);
616 if (iocb->ki_flags & IOCB_DIRECT)
617 return nfs_file_direct_write(iocb, from, false);
619 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
620 file, iov_iter_count(from), (long long) iocb->ki_pos);
622 if (IS_SWAPFILE(inode))
625 * O_APPEND implies that we must revalidate the file length.
627 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
628 result = nfs_revalidate_file_size(inode, file);
633 nfs_clear_invalid_mapping(file->f_mapping);
635 since = filemap_sample_wb_err(file->f_mapping);
636 nfs_start_io_write(inode);
637 result = generic_write_checks(iocb, from);
639 current->backing_dev_info = inode_to_bdi(inode);
640 result = generic_perform_write(iocb, from);
641 current->backing_dev_info = NULL;
643 nfs_end_io_write(inode);
648 iocb->ki_pos += written;
649 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
651 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
652 result = filemap_fdatawrite_range(file->f_mapping,
653 iocb->ki_pos - written,
658 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
659 filemap_fdatawait_range(file->f_mapping,
660 iocb->ki_pos - written,
663 result = generic_write_sync(iocb, written);
668 /* Return error values */
669 error = filemap_check_wb_err(file->f_mapping, since);
677 error = file_check_and_advance_wb_err(file);
684 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
687 EXPORT_SYMBOL_GPL(nfs_file_write);
690 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
692 struct inode *inode = filp->f_mapping->host;
694 unsigned int saved_type = fl->fl_type;
696 /* Try local locking first */
697 posix_test_lock(filp, fl);
698 if (fl->fl_type != F_UNLCK) {
699 /* found a conflict */
702 fl->fl_type = saved_type;
704 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
710 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
714 fl->fl_type = F_UNLCK;
719 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
721 struct inode *inode = filp->f_mapping->host;
722 struct nfs_lock_context *l_ctx;
726 * Flush all pending writes before doing anything
731 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
732 if (!IS_ERR(l_ctx)) {
733 status = nfs_iocounter_wait(l_ctx);
734 nfs_put_lock_context(l_ctx);
735 /* NOTE: special case
736 * If we're signalled while cleaning up locks on process exit, we
737 * still need to complete the unlock.
739 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
744 * Use local locking if mounted with "-onolock" or with appropriate
748 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
750 status = locks_lock_file_wait(filp, fl);
755 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
757 struct inode *inode = filp->f_mapping->host;
761 * Flush all pending writes before doing anything
764 status = nfs_sync_mapping(filp->f_mapping);
769 * Use local locking if mounted with "-onolock" or with appropriate
773 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
775 status = locks_lock_file_wait(filp, fl);
780 * Invalidate cache to prevent missing any changes. If
781 * the file is mapped, clear the page cache as well so
782 * those mappings will be loaded.
784 * This makes locking act as a cache coherency point.
786 nfs_sync_mapping(filp->f_mapping);
787 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
788 nfs_zap_caches(inode);
789 if (mapping_mapped(filp->f_mapping))
790 nfs_revalidate_mapping(inode, filp->f_mapping);
797 * Lock a (portion of) a file
799 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
801 struct inode *inode = filp->f_mapping->host;
805 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
806 filp, fl->fl_type, fl->fl_flags,
807 (long long)fl->fl_start, (long long)fl->fl_end);
809 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
811 if (fl->fl_flags & FL_RECLAIM)
814 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
817 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
818 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
824 ret = do_getlk(filp, cmd, fl, is_local);
825 else if (fl->fl_type == F_UNLCK)
826 ret = do_unlk(filp, cmd, fl, is_local);
828 ret = do_setlk(filp, cmd, fl, is_local);
832 EXPORT_SYMBOL_GPL(nfs_lock);
835 * Lock a (portion of) a file
837 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
839 struct inode *inode = filp->f_mapping->host;
842 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
843 filp, fl->fl_type, fl->fl_flags);
845 if (!(fl->fl_flags & FL_FLOCK))
848 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
851 /* We're simulating flock() locks using posix locks on the server */
852 if (fl->fl_type == F_UNLCK)
853 return do_unlk(filp, cmd, fl, is_local);
854 return do_setlk(filp, cmd, fl, is_local);
856 EXPORT_SYMBOL_GPL(nfs_flock);
858 const struct file_operations nfs_file_operations = {
859 .llseek = nfs_file_llseek,
860 .read_iter = nfs_file_read,
861 .write_iter = nfs_file_write,
862 .mmap = nfs_file_mmap,
863 .open = nfs_file_open,
864 .flush = nfs_file_flush,
865 .release = nfs_file_release,
866 .fsync = nfs_file_fsync,
869 .splice_read = generic_file_splice_read,
870 .splice_write = iter_file_splice_write,
871 .check_flags = nfs_check_flags,
872 .setlease = simple_nosetlease,
874 EXPORT_SYMBOL_GPL(nfs_file_operations);