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_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_REVAL_PAGECACHE))
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, false);
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 nfs_open_context *ctx = nfs_file_open_context(file);
211 struct inode *inode = file_inode(file);
212 int do_resend, status;
215 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
217 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
218 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
219 status = nfs_commit_inode(inode, FLUSH_SYNC);
221 status = file_check_and_advance_wb_err(file);
226 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
234 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
237 struct inode *inode = file_inode(file);
239 trace_nfs_fsync_enter(inode);
242 ret = file_write_and_wait_range(file, start, end);
245 ret = nfs_file_fsync_commit(file, datasync);
247 ret = pnfs_sync_inode(inode, !!datasync);
249 * If nfs_file_fsync_commit detected a server reboot, then
250 * resend all dirty pages that might have been covered by
251 * the NFS_CONTEXT_RESEND_WRITES flag
255 } while (ret == -EAGAIN);
257 trace_nfs_fsync_exit(inode, ret);
260 EXPORT_SYMBOL_GPL(nfs_file_fsync);
263 * Decide whether a read/modify/write cycle may be more efficient
264 * then a modify/write/read cycle when writing to a page in the
267 * Some pNFS layout drivers can only read/write at a certain block
268 * granularity like all block devices and therefore we must perform
269 * read/modify/write whenever a page hasn't read yet and the data
270 * to be written there is not aligned to a block boundary and/or
271 * smaller than the block size.
273 * The modify/write/read cycle may occur if a page is read before
274 * being completely filled by the writer. In this situation, the
275 * page must be completely written to stable storage on the server
276 * before it can be refilled by reading in the page from the server.
277 * This can lead to expensive, small, FILE_SYNC mode writes being
280 * It may be more efficient to read the page first if the file is
281 * open for reading in addition to writing, the page is not marked
282 * as Uptodate, it is not dirty or waiting to be committed,
283 * indicating that it was previously allocated and then modified,
284 * that there were valid bytes of data in that range of the file,
285 * and that the new data won't completely replace the old data in
286 * that range of the file.
288 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
290 unsigned int pglen = nfs_page_length(page);
291 unsigned int offset = pos & (PAGE_SIZE - 1);
292 unsigned int end = offset + len;
294 return !pglen || (end >= pglen && !offset);
297 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
298 loff_t pos, unsigned int len)
301 * Up-to-date pages, those with ongoing or full-page write
302 * don't need read/modify/write
304 if (PageUptodate(page) || PagePrivate(page) ||
305 nfs_full_page_write(page, pos, len))
308 if (pnfs_ld_read_whole_page(file->f_mapping->host))
310 /* Open for reading too? */
311 if (file->f_mode & FMODE_READ)
317 * This does the "real" work of the write. We must allocate and lock the
318 * page to be sent back to the generic routine, which then copies the
319 * data from user space.
321 * If the writer ends up delaying the write, the writer needs to
322 * increment the page use counts until he is done with the page.
324 static int nfs_write_begin(struct file *file, struct address_space *mapping,
325 loff_t pos, unsigned len, unsigned flags,
326 struct page **pagep, void **fsdata)
329 pgoff_t index = pos >> PAGE_SHIFT;
333 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
334 file, mapping->host->i_ino, len, (long long) pos);
337 page = grab_cache_page_write_begin(mapping, index, flags);
342 ret = nfs_flush_incompatible(file, page);
346 } else if (!once_thru &&
347 nfs_want_read_modify_write(file, page, pos, len)) {
349 ret = nfs_readpage(file, page);
357 static int nfs_write_end(struct file *file, struct address_space *mapping,
358 loff_t pos, unsigned len, unsigned copied,
359 struct page *page, void *fsdata)
361 unsigned offset = pos & (PAGE_SIZE - 1);
362 struct nfs_open_context *ctx = nfs_file_open_context(file);
365 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
366 file, mapping->host->i_ino, len, (long long) pos);
369 * Zero any uninitialised parts of the page, and then mark the page
370 * as up to date if it turns out that we're extending the file.
372 if (!PageUptodate(page)) {
373 unsigned pglen = nfs_page_length(page);
374 unsigned end = offset + copied;
377 zero_user_segments(page, 0, offset,
379 SetPageUptodate(page);
380 } else if (end >= pglen) {
381 zero_user_segment(page, end, PAGE_SIZE);
383 SetPageUptodate(page);
385 zero_user_segment(page, pglen, PAGE_SIZE);
388 status = nfs_updatepage(file, page, offset, copied);
395 NFS_I(mapping->host)->write_io += copied;
397 if (nfs_ctx_key_to_expire(ctx, mapping->host))
398 nfs_wb_all(mapping->host);
404 * Partially or wholly invalidate a page
405 * - Release the private state associated with a page if undergoing complete
407 * - Called if either PG_private or PG_fscache is set on the page
408 * - Caller holds page lock
410 static void nfs_invalidate_page(struct page *page, unsigned int offset,
413 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
414 page, offset, length);
416 if (offset != 0 || length < PAGE_SIZE)
418 /* Cancel any unstarted writes on this page */
419 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
421 nfs_fscache_invalidate_page(page, page->mapping->host);
425 * Attempt to release the private state associated with a page
426 * - Called if either PG_private or PG_fscache is set on the page
427 * - Caller holds page lock
428 * - Return true (may release page) or false (may not)
430 static int nfs_release_page(struct page *page, gfp_t gfp)
432 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
434 /* If PagePrivate() is set, then the page is not freeable */
435 if (PagePrivate(page))
437 return nfs_fscache_release_page(page, gfp);
440 static void nfs_check_dirty_writeback(struct page *page,
441 bool *dirty, bool *writeback)
443 struct nfs_inode *nfsi;
444 struct address_space *mapping = page_file_mapping(page);
446 if (!mapping || PageSwapCache(page))
450 * Check if an unstable page is currently being committed and
451 * if so, have the VM treat it as if the page is under writeback
452 * so it will not block due to pages that will shortly be freeable.
454 nfsi = NFS_I(mapping->host);
455 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
461 * If PagePrivate() is set, then the page is not freeable and as the
462 * inode is not being committed, it's not going to be cleaned in the
463 * near future so treat it as dirty
465 if (PagePrivate(page))
470 * Attempt to clear the private state associated with a page when an error
471 * occurs that requires the cached contents of an inode to be written back or
473 * - Called if either PG_private or fscache is set on the page
474 * - Caller holds page lock
475 * - Return 0 if successful, -error otherwise
477 static int nfs_launder_page(struct page *page)
479 struct inode *inode = page_file_mapping(page)->host;
480 struct nfs_inode *nfsi = NFS_I(inode);
482 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
483 inode->i_ino, (long long)page_offset(page));
485 nfs_fscache_wait_on_page_write(nfsi, page);
486 return nfs_wb_page(inode, page);
489 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
492 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
496 return rpc_clnt_swap_activate(clnt);
499 static void nfs_swap_deactivate(struct file *file)
501 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
503 rpc_clnt_swap_deactivate(clnt);
506 const struct address_space_operations nfs_file_aops = {
507 .readpage = nfs_readpage,
508 .readpages = nfs_readpages,
509 .set_page_dirty = __set_page_dirty_nobuffers,
510 .writepage = nfs_writepage,
511 .writepages = nfs_writepages,
512 .write_begin = nfs_write_begin,
513 .write_end = nfs_write_end,
514 .invalidatepage = nfs_invalidate_page,
515 .releasepage = nfs_release_page,
516 .direct_IO = nfs_direct_IO,
517 #ifdef CONFIG_MIGRATION
518 .migratepage = nfs_migrate_page,
520 .launder_page = nfs_launder_page,
521 .is_dirty_writeback = nfs_check_dirty_writeback,
522 .error_remove_page = generic_error_remove_page,
523 .swap_activate = nfs_swap_activate,
524 .swap_deactivate = nfs_swap_deactivate,
528 * Notification that a PTE pointing to an NFS page is about to be made
529 * writable, implying that someone is about to modify the page through a
530 * shared-writable mapping
532 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
534 struct page *page = vmf->page;
535 struct file *filp = vmf->vma->vm_file;
536 struct inode *inode = file_inode(filp);
538 vm_fault_t ret = VM_FAULT_NOPAGE;
539 struct address_space *mapping;
541 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
542 filp, filp->f_mapping->host->i_ino,
543 (long long)page_offset(page));
545 sb_start_pagefault(inode->i_sb);
547 /* make sure the cache has finished storing the page */
548 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
550 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
551 nfs_wait_bit_killable, TASK_KILLABLE);
554 mapping = page_file_mapping(page);
555 if (mapping != inode->i_mapping)
558 wait_on_page_writeback(page);
560 pagelen = nfs_page_length(page);
564 ret = VM_FAULT_LOCKED;
565 if (nfs_flush_incompatible(filp, page) == 0 &&
566 nfs_updatepage(filp, page, 0, pagelen) == 0)
569 ret = VM_FAULT_SIGBUS;
573 sb_end_pagefault(inode->i_sb);
577 static const struct vm_operations_struct nfs_file_vm_ops = {
578 .fault = filemap_fault,
579 .map_pages = filemap_map_pages,
580 .page_mkwrite = nfs_vm_page_mkwrite,
583 static int nfs_need_check_write(struct file *filp, struct inode *inode,
586 struct nfs_open_context *ctx;
588 ctx = nfs_file_open_context(filp);
589 if (nfs_error_is_fatal_on_server(error) ||
590 nfs_ctx_key_to_expire(ctx, inode))
595 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
597 struct file *file = iocb->ki_filp;
598 struct inode *inode = file_inode(file);
599 unsigned long written = 0;
604 result = nfs_key_timeout_notify(file, inode);
608 if (iocb->ki_flags & IOCB_DIRECT)
609 return nfs_file_direct_write(iocb, from, false);
611 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
612 file, iov_iter_count(from), (long long) iocb->ki_pos);
614 if (IS_SWAPFILE(inode))
617 * O_APPEND implies that we must revalidate the file length.
619 if (iocb->ki_flags & IOCB_APPEND) {
620 result = nfs_revalidate_file_size(inode, file);
624 if (iocb->ki_pos > i_size_read(inode))
625 nfs_revalidate_mapping(inode, file->f_mapping);
627 since = filemap_sample_wb_err(file->f_mapping);
628 nfs_start_io_write(inode);
629 result = generic_write_checks(iocb, from);
631 current->backing_dev_info = inode_to_bdi(inode);
632 result = generic_perform_write(file, from, iocb->ki_pos);
633 current->backing_dev_info = NULL;
635 nfs_end_io_write(inode);
640 iocb->ki_pos += written;
641 result = generic_write_sync(iocb, written);
645 /* Return error values */
646 error = filemap_check_wb_err(file->f_mapping, since);
647 if (nfs_need_check_write(file, inode, error)) {
648 int err = nfs_wb_all(inode);
652 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
657 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
660 EXPORT_SYMBOL_GPL(nfs_file_write);
663 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
665 struct inode *inode = filp->f_mapping->host;
667 unsigned int saved_type = fl->fl_type;
669 /* Try local locking first */
670 posix_test_lock(filp, fl);
671 if (fl->fl_type != F_UNLCK) {
672 /* found a conflict */
675 fl->fl_type = saved_type;
677 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
683 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
687 fl->fl_type = F_UNLCK;
692 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
694 struct inode *inode = filp->f_mapping->host;
695 struct nfs_lock_context *l_ctx;
699 * Flush all pending writes before doing anything
704 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
705 if (!IS_ERR(l_ctx)) {
706 status = nfs_iocounter_wait(l_ctx);
707 nfs_put_lock_context(l_ctx);
708 /* NOTE: special case
709 * If we're signalled while cleaning up locks on process exit, we
710 * still need to complete the unlock.
712 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
717 * Use local locking if mounted with "-onolock" or with appropriate
721 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
723 status = locks_lock_file_wait(filp, fl);
728 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
730 struct inode *inode = filp->f_mapping->host;
734 * Flush all pending writes before doing anything
737 status = nfs_sync_mapping(filp->f_mapping);
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 * Invalidate cache to prevent missing any changes. If
754 * the file is mapped, clear the page cache as well so
755 * those mappings will be loaded.
757 * This makes locking act as a cache coherency point.
759 nfs_sync_mapping(filp->f_mapping);
760 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
761 nfs_zap_caches(inode);
762 if (mapping_mapped(filp->f_mapping))
763 nfs_revalidate_mapping(inode, filp->f_mapping);
770 * Lock a (portion of) a file
772 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
774 struct inode *inode = filp->f_mapping->host;
778 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
779 filp, fl->fl_type, fl->fl_flags,
780 (long long)fl->fl_start, (long long)fl->fl_end);
782 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
784 /* No mandatory locks over NFS */
785 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
788 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
791 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
792 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
798 ret = do_getlk(filp, cmd, fl, is_local);
799 else if (fl->fl_type == F_UNLCK)
800 ret = do_unlk(filp, cmd, fl, is_local);
802 ret = do_setlk(filp, cmd, fl, is_local);
806 EXPORT_SYMBOL_GPL(nfs_lock);
809 * Lock a (portion of) a file
811 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
813 struct inode *inode = filp->f_mapping->host;
816 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
817 filp, fl->fl_type, fl->fl_flags);
819 if (!(fl->fl_flags & FL_FLOCK))
823 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
824 * any standard. In principle we might be able to support LOCK_MAND
825 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
826 * NFS code is not set up for it.
828 if (fl->fl_type & LOCK_MAND)
831 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
834 /* We're simulating flock() locks using posix locks on the server */
835 if (fl->fl_type == F_UNLCK)
836 return do_unlk(filp, cmd, fl, is_local);
837 return do_setlk(filp, cmd, fl, is_local);
839 EXPORT_SYMBOL_GPL(nfs_flock);
841 const struct file_operations nfs_file_operations = {
842 .llseek = nfs_file_llseek,
843 .read_iter = nfs_file_read,
844 .write_iter = nfs_file_write,
845 .mmap = nfs_file_mmap,
846 .open = nfs_file_open,
847 .flush = nfs_file_flush,
848 .release = nfs_file_release,
849 .fsync = nfs_file_fsync,
852 .splice_read = generic_file_splice_read,
853 .splice_write = iter_file_splice_write,
854 .check_flags = nfs_check_flags,
855 .setlease = simple_nosetlease,
857 EXPORT_SYMBOL_GPL(nfs_file_operations);