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 /* Hack for future NFS swap support */
49 # define IS_SWAPFILE(inode) (0)
52 int nfs_check_flags(int flags)
54 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
59 EXPORT_SYMBOL_GPL(nfs_check_flags);
65 nfs_file_open(struct inode *inode, struct file *filp)
69 dprintk("NFS: open file(%pD2)\n", filp);
71 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
72 res = nfs_check_flags(filp->f_flags);
76 res = nfs_open(inode, filp);
81 nfs_file_release(struct inode *inode, struct file *filp)
83 dprintk("NFS: release(%pD2)\n", filp);
85 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
86 nfs_file_clear_open_context(filp);
89 EXPORT_SYMBOL_GPL(nfs_file_release);
92 * nfs_revalidate_file_size - Revalidate the file size
93 * @inode: pointer to inode struct
94 * @filp: pointer to struct file
96 * Revalidates the file length. This is basically a wrapper around
97 * nfs_revalidate_inode() that takes into account the fact that we may
98 * have cached writes (in which case we don't care about the server's
99 * idea of what the file length is), or O_DIRECT (in which case we
100 * shouldn't trust the cache).
102 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
104 struct nfs_server *server = NFS_SERVER(inode);
106 if (filp->f_flags & O_DIRECT)
108 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE))
112 return __nfs_revalidate_inode(server, inode);
115 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
117 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
118 filp, offset, whence);
121 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
122 * the cached file length
124 if (whence != SEEK_SET && whence != SEEK_CUR) {
125 struct inode *inode = filp->f_mapping->host;
127 int retval = nfs_revalidate_file_size(inode, filp);
129 return (loff_t)retval;
132 return generic_file_llseek(filp, offset, whence);
134 EXPORT_SYMBOL_GPL(nfs_file_llseek);
137 * Flush all dirty pages, and check for write errors.
140 nfs_file_flush(struct file *file, fl_owner_t id)
142 struct inode *inode = file_inode(file);
145 dprintk("NFS: flush(%pD2)\n", file);
147 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
148 if ((file->f_mode & FMODE_WRITE) == 0)
151 /* Flush writes to the server and return any errors */
152 since = filemap_sample_wb_err(file->f_mapping);
154 return filemap_check_wb_err(file->f_mapping, since);
158 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
160 struct inode *inode = file_inode(iocb->ki_filp);
163 if (iocb->ki_flags & IOCB_DIRECT)
164 return nfs_file_direct_read(iocb, to);
166 dprintk("NFS: read(%pD2, %zu@%lu)\n",
168 iov_iter_count(to), (unsigned long) iocb->ki_pos);
170 nfs_start_io_read(inode);
171 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
173 result = generic_file_read_iter(iocb, to);
175 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
177 nfs_end_io_read(inode);
180 EXPORT_SYMBOL_GPL(nfs_file_read);
183 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
185 struct inode *inode = file_inode(file);
188 dprintk("NFS: mmap(%pD2)\n", file);
190 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
191 * so we call that before revalidating the mapping
193 status = generic_file_mmap(file, vma);
195 vma->vm_ops = &nfs_file_vm_ops;
196 status = nfs_revalidate_mapping(inode, file->f_mapping);
200 EXPORT_SYMBOL_GPL(nfs_file_mmap);
203 * Flush any dirty pages for this process, and check for write errors.
204 * The return status from this call provides a reliable indication of
205 * whether any write errors occurred for this process.
208 nfs_file_fsync_commit(struct file *file, int datasync)
210 struct inode *inode = file_inode(file);
213 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
215 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
216 ret = nfs_commit_inode(inode, FLUSH_SYNC);
219 return file_check_and_advance_wb_err(file);
223 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
225 struct nfs_open_context *ctx = nfs_file_open_context(file);
226 struct inode *inode = file_inode(file);
229 trace_nfs_fsync_enter(inode);
232 ret = file_write_and_wait_range(file, start, end);
235 ret = nfs_file_fsync_commit(file, datasync);
238 ret = pnfs_sync_inode(inode, !!datasync);
241 if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
244 * If nfs_file_fsync_commit detected a server reboot, then
245 * resend all dirty pages that might have been covered by
246 * the NFS_CONTEXT_RESEND_WRITES flag
252 trace_nfs_fsync_exit(inode, ret);
255 EXPORT_SYMBOL_GPL(nfs_file_fsync);
258 * Decide whether a read/modify/write cycle may be more efficient
259 * then a modify/write/read cycle when writing to a page in the
262 * Some pNFS layout drivers can only read/write at a certain block
263 * granularity like all block devices and therefore we must perform
264 * read/modify/write whenever a page hasn't read yet and the data
265 * to be written there is not aligned to a block boundary and/or
266 * smaller than the block size.
268 * The modify/write/read cycle may occur if a page is read before
269 * being completely filled by the writer. In this situation, the
270 * page must be completely written to stable storage on the server
271 * before it can be refilled by reading in the page from the server.
272 * This can lead to expensive, small, FILE_SYNC mode writes being
275 * It may be more efficient to read the page first if the file is
276 * open for reading in addition to writing, the page is not marked
277 * as Uptodate, it is not dirty or waiting to be committed,
278 * indicating that it was previously allocated and then modified,
279 * that there were valid bytes of data in that range of the file,
280 * and that the new data won't completely replace the old data in
281 * that range of the file.
283 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
285 unsigned int pglen = nfs_page_length(page);
286 unsigned int offset = pos & (PAGE_SIZE - 1);
287 unsigned int end = offset + len;
289 return !pglen || (end >= pglen && !offset);
292 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
293 loff_t pos, unsigned int len)
296 * Up-to-date pages, those with ongoing or full-page write
297 * don't need read/modify/write
299 if (PageUptodate(page) || PagePrivate(page) ||
300 nfs_full_page_write(page, pos, len))
303 if (pnfs_ld_read_whole_page(file->f_mapping->host))
305 /* Open for reading too? */
306 if (file->f_mode & FMODE_READ)
312 * This does the "real" work of the write. We must allocate and lock the
313 * page to be sent back to the generic routine, which then copies the
314 * data from user space.
316 * If the writer ends up delaying the write, the writer needs to
317 * increment the page use counts until he is done with the page.
319 static int nfs_write_begin(struct file *file, struct address_space *mapping,
320 loff_t pos, unsigned len, unsigned flags,
321 struct page **pagep, void **fsdata)
324 pgoff_t index = pos >> PAGE_SHIFT;
328 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
329 file, mapping->host->i_ino, len, (long long) pos);
332 page = grab_cache_page_write_begin(mapping, index, flags);
337 ret = nfs_flush_incompatible(file, page);
341 } else if (!once_thru &&
342 nfs_want_read_modify_write(file, page, pos, len)) {
344 ret = nfs_readpage(file, page);
352 static int nfs_write_end(struct file *file, struct address_space *mapping,
353 loff_t pos, unsigned len, unsigned copied,
354 struct page *page, void *fsdata)
356 unsigned offset = pos & (PAGE_SIZE - 1);
357 struct nfs_open_context *ctx = nfs_file_open_context(file);
360 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
361 file, mapping->host->i_ino, len, (long long) pos);
364 * Zero any uninitialised parts of the page, and then mark the page
365 * as up to date if it turns out that we're extending the file.
367 if (!PageUptodate(page)) {
368 unsigned pglen = nfs_page_length(page);
369 unsigned end = offset + copied;
372 zero_user_segments(page, 0, offset,
374 SetPageUptodate(page);
375 } else if (end >= pglen) {
376 zero_user_segment(page, end, PAGE_SIZE);
378 SetPageUptodate(page);
380 zero_user_segment(page, pglen, PAGE_SIZE);
383 status = nfs_updatepage(file, page, offset, copied);
390 NFS_I(mapping->host)->write_io += copied;
392 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
393 status = nfs_wb_all(mapping->host);
402 * Partially or wholly invalidate a page
403 * - Release the private state associated with a page if undergoing complete
405 * - Called if either PG_private or PG_fscache is set on the page
406 * - Caller holds page lock
408 static void nfs_invalidate_page(struct page *page, unsigned int offset,
411 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
412 page, offset, length);
414 if (offset != 0 || length < PAGE_SIZE)
416 /* Cancel any unstarted writes on this page */
417 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
419 nfs_fscache_invalidate_page(page, page->mapping->host);
423 * Attempt to release the private state associated with a page
424 * - Called if either PG_private or PG_fscache is set on the page
425 * - Caller holds page lock
426 * - Return true (may release page) or false (may not)
428 static int nfs_release_page(struct page *page, gfp_t gfp)
430 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
432 /* If PagePrivate() is set, then the page is not freeable */
433 if (PagePrivate(page))
435 return nfs_fscache_release_page(page, gfp);
438 static void nfs_check_dirty_writeback(struct page *page,
439 bool *dirty, bool *writeback)
441 struct nfs_inode *nfsi;
442 struct address_space *mapping = page_file_mapping(page);
444 if (!mapping || PageSwapCache(page))
448 * Check if an unstable page is currently being committed and
449 * if so, have the VM treat it as if the page is under writeback
450 * so it will not block due to pages that will shortly be freeable.
452 nfsi = NFS_I(mapping->host);
453 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
459 * If PagePrivate() is set, then the page is not freeable and as the
460 * inode is not being committed, it's not going to be cleaned in the
461 * near future so treat it as dirty
463 if (PagePrivate(page))
468 * Attempt to clear the private state associated with a page when an error
469 * occurs that requires the cached contents of an inode to be written back or
471 * - Called if either PG_private or fscache is set on the page
472 * - Caller holds page lock
473 * - Return 0 if successful, -error otherwise
475 static int nfs_launder_page(struct page *page)
477 struct inode *inode = page_file_mapping(page)->host;
478 struct nfs_inode *nfsi = NFS_I(inode);
480 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
481 inode->i_ino, (long long)page_offset(page));
483 nfs_fscache_wait_on_page_write(nfsi, page);
484 return nfs_wb_page(inode, page);
487 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
490 unsigned long blocks;
492 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
493 struct inode *inode = file->f_mapping->host;
495 spin_lock(&inode->i_lock);
496 blocks = inode->i_blocks;
497 isize = inode->i_size;
498 spin_unlock(&inode->i_lock);
499 if (blocks*512 < isize) {
500 pr_warn("swap activate: swapfile has holes\n");
506 return rpc_clnt_swap_activate(clnt);
509 static void nfs_swap_deactivate(struct file *file)
511 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
513 rpc_clnt_swap_deactivate(clnt);
516 const struct address_space_operations nfs_file_aops = {
517 .readpage = nfs_readpage,
518 .readpages = nfs_readpages,
519 .set_page_dirty = __set_page_dirty_nobuffers,
520 .writepage = nfs_writepage,
521 .writepages = nfs_writepages,
522 .write_begin = nfs_write_begin,
523 .write_end = nfs_write_end,
524 .invalidatepage = nfs_invalidate_page,
525 .releasepage = nfs_release_page,
526 .direct_IO = nfs_direct_IO,
527 #ifdef CONFIG_MIGRATION
528 .migratepage = nfs_migrate_page,
530 .launder_page = nfs_launder_page,
531 .is_dirty_writeback = nfs_check_dirty_writeback,
532 .error_remove_page = generic_error_remove_page,
533 .swap_activate = nfs_swap_activate,
534 .swap_deactivate = nfs_swap_deactivate,
538 * Notification that a PTE pointing to an NFS page is about to be made
539 * writable, implying that someone is about to modify the page through a
540 * shared-writable mapping
542 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
544 struct page *page = vmf->page;
545 struct file *filp = vmf->vma->vm_file;
546 struct inode *inode = file_inode(filp);
548 vm_fault_t ret = VM_FAULT_NOPAGE;
549 struct address_space *mapping;
551 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
552 filp, filp->f_mapping->host->i_ino,
553 (long long)page_offset(page));
555 sb_start_pagefault(inode->i_sb);
557 /* make sure the cache has finished storing the page */
558 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
560 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
561 nfs_wait_bit_killable, TASK_KILLABLE);
564 mapping = page_file_mapping(page);
565 if (mapping != inode->i_mapping)
568 wait_on_page_writeback(page);
570 pagelen = nfs_page_length(page);
574 ret = VM_FAULT_LOCKED;
575 if (nfs_flush_incompatible(filp, page) == 0 &&
576 nfs_updatepage(filp, page, 0, pagelen) == 0)
579 ret = VM_FAULT_SIGBUS;
583 sb_end_pagefault(inode->i_sb);
587 static const struct vm_operations_struct nfs_file_vm_ops = {
588 .fault = filemap_fault,
589 .map_pages = filemap_map_pages,
590 .page_mkwrite = nfs_vm_page_mkwrite,
593 static int nfs_need_check_write(struct file *filp, struct inode *inode,
596 struct nfs_open_context *ctx;
598 ctx = nfs_file_open_context(filp);
599 if (nfs_error_is_fatal_on_server(error) ||
600 nfs_ctx_key_to_expire(ctx, inode))
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);
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(file, from, iocb->ki_pos);
643 current->backing_dev_info = NULL;
645 nfs_end_io_write(inode);
650 iocb->ki_pos += written;
652 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
653 result = filemap_fdatawrite_range(file->f_mapping,
654 iocb->ki_pos - written,
659 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
660 result = filemap_fdatawait_range(file->f_mapping,
661 iocb->ki_pos - written,
666 result = generic_write_sync(iocb, written);
670 /* Return error values */
671 error = filemap_check_wb_err(file->f_mapping, since);
672 if (nfs_need_check_write(file, inode, error)) {
673 int err = nfs_wb_all(inode);
677 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
682 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
685 EXPORT_SYMBOL_GPL(nfs_file_write);
688 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
690 struct inode *inode = filp->f_mapping->host;
692 unsigned int saved_type = fl->fl_type;
694 /* Try local locking first */
695 posix_test_lock(filp, fl);
696 if (fl->fl_type != F_UNLCK) {
697 /* found a conflict */
700 fl->fl_type = saved_type;
702 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
708 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
712 fl->fl_type = F_UNLCK;
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);
733 /* NOTE: special case
734 * If we're signalled while cleaning up locks on process exit, we
735 * still need to complete the unlock.
737 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
742 * Use local locking if mounted with "-onolock" or with appropriate
746 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
748 status = locks_lock_file_wait(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 = locks_lock_file_wait(filp, fl);
778 * Invalidate cache to prevent missing any changes. If
779 * the file is mapped, clear the page cache as well so
780 * those mappings will be loaded.
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_caches(inode);
787 if (mapping_mapped(filp->f_mapping))
788 nfs_revalidate_mapping(inode, filp->f_mapping);
795 * Lock a (portion of) a file
797 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
799 struct inode *inode = filp->f_mapping->host;
803 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
804 filp, fl->fl_type, fl->fl_flags,
805 (long long)fl->fl_start, (long long)fl->fl_end);
807 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
809 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
812 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
813 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
819 ret = do_getlk(filp, cmd, fl, is_local);
820 else if (fl->fl_type == F_UNLCK)
821 ret = do_unlk(filp, cmd, fl, is_local);
823 ret = do_setlk(filp, cmd, fl, is_local);
827 EXPORT_SYMBOL_GPL(nfs_lock);
830 * Lock a (portion of) a file
832 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
834 struct inode *inode = filp->f_mapping->host;
837 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
838 filp, fl->fl_type, fl->fl_flags);
840 if (!(fl->fl_flags & FL_FLOCK))
844 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
845 * any standard. In principle we might be able to support LOCK_MAND
846 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
847 * NFS code is not set up for it.
849 if (fl->fl_type & LOCK_MAND)
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);