4 * Copyright (C) 1992 Rick Sladkey
6 * Changes Copyright (C) 1994 by Florian La Roche
7 * - Do not copy data too often around in the kernel.
8 * - In nfs_file_read the return value of kmalloc wasn't checked.
9 * - Put in a better version of read look-ahead buffering. Original idea
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
16 * nfs regular file handling functions
19 #include <linux/module.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/fcntl.h>
24 #include <linux/stat.h>
25 #include <linux/nfs_fs.h>
26 #include <linux/nfs_mount.h>
28 #include <linux/pagemap.h>
29 #include <linux/gfp.h>
30 #include <linux/swap.h>
32 #include <linux/uaccess.h>
34 #include "delegation.h"
42 #define NFSDBG_FACILITY NFSDBG_FILE
44 static const struct vm_operations_struct nfs_file_vm_ops;
46 /* Hack for future NFS swap support */
48 # define IS_SWAPFILE(inode) (0)
51 int nfs_check_flags(int flags)
53 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
58 EXPORT_SYMBOL_GPL(nfs_check_flags);
64 nfs_file_open(struct inode *inode, struct file *filp)
68 dprintk("NFS: open file(%pD2)\n", filp);
70 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
71 res = nfs_check_flags(filp->f_flags);
75 res = nfs_open(inode, filp);
80 nfs_file_release(struct inode *inode, struct file *filp)
82 dprintk("NFS: release(%pD2)\n", filp);
84 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
85 nfs_file_clear_open_context(filp);
88 EXPORT_SYMBOL_GPL(nfs_file_release);
91 * nfs_revalidate_size - Revalidate the file size
92 * @inode - pointer to inode struct
93 * @file - pointer to struct file
95 * Revalidates the file length. This is basically a wrapper around
96 * nfs_revalidate_inode() that takes into account the fact that we may
97 * have cached writes (in which case we don't care about the server's
98 * idea of what the file length is), or O_DIRECT (in which case we
99 * shouldn't trust the cache).
101 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
103 struct nfs_server *server = NFS_SERVER(inode);
105 if (filp->f_flags & O_DIRECT)
107 if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE))
111 return __nfs_revalidate_inode(server, inode);
114 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
116 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
117 filp, offset, whence);
120 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
121 * the cached file length
123 if (whence != SEEK_SET && whence != SEEK_CUR) {
124 struct inode *inode = filp->f_mapping->host;
126 int retval = nfs_revalidate_file_size(inode, filp);
128 return (loff_t)retval;
131 return generic_file_llseek(filp, offset, whence);
133 EXPORT_SYMBOL_GPL(nfs_file_llseek);
136 * Flush all dirty pages, and check for write errors.
139 nfs_file_flush(struct file *file, fl_owner_t id)
141 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 return vfs_fsync(file, 0);
154 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
156 struct inode *inode = file_inode(iocb->ki_filp);
159 if (iocb->ki_flags & IOCB_DIRECT)
160 return nfs_file_direct_read(iocb, to);
162 dprintk("NFS: read(%pD2, %zu@%lu)\n",
164 iov_iter_count(to), (unsigned long) iocb->ki_pos);
166 nfs_start_io_read(inode);
167 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
169 result = generic_file_read_iter(iocb, to);
171 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
173 nfs_end_io_read(inode);
176 EXPORT_SYMBOL_GPL(nfs_file_read);
179 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
181 struct inode *inode = file_inode(file);
184 dprintk("NFS: mmap(%pD2)\n", file);
186 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
187 * so we call that before revalidating the mapping
189 status = generic_file_mmap(file, vma);
191 vma->vm_ops = &nfs_file_vm_ops;
192 status = nfs_revalidate_mapping(inode, file->f_mapping);
196 EXPORT_SYMBOL_GPL(nfs_file_mmap);
199 * Flush any dirty pages for this process, and check for write errors.
200 * The return status from this call provides a reliable indication of
201 * whether any write errors occurred for this process.
203 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
204 * disk, but it retrieves and clears ctx->error after synching, despite
205 * the two being set at the same time in nfs_context_set_write_error().
206 * This is because the former is used to notify the _next_ call to
207 * nfs_file_write() that a write error occurred, and hence cause it to
208 * fall back to doing a synchronous write.
211 nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
213 struct nfs_open_context *ctx = nfs_file_open_context(file);
214 struct inode *inode = file_inode(file);
215 int have_error, do_resend, status;
218 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
220 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
221 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
222 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
223 status = nfs_commit_inode(inode, FLUSH_SYNC);
224 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
226 ret = xchg(&ctx->error, 0);
234 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
242 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
245 struct inode *inode = file_inode(file);
247 trace_nfs_fsync_enter(inode);
250 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
253 ret = nfs_file_fsync_commit(file, start, end, datasync);
255 ret = pnfs_sync_inode(inode, !!datasync);
257 * If nfs_file_fsync_commit detected a server reboot, then
258 * resend all dirty pages that might have been covered by
259 * the NFS_CONTEXT_RESEND_WRITES flag
263 } while (ret == -EAGAIN);
265 trace_nfs_fsync_exit(inode, ret);
268 EXPORT_SYMBOL_GPL(nfs_file_fsync);
271 * Decide whether a read/modify/write cycle may be more efficient
272 * then a modify/write/read cycle when writing to a page in the
275 * The modify/write/read cycle may occur if a page is read before
276 * being completely filled by the writer. In this situation, the
277 * page must be completely written to stable storage on the server
278 * before it can be refilled by reading in the page from the server.
279 * This can lead to expensive, small, FILE_SYNC mode writes being
282 * It may be more efficient to read the page first if the file is
283 * open for reading in addition to writing, the page is not marked
284 * as Uptodate, it is not dirty or waiting to be committed,
285 * indicating that it was previously allocated and then modified,
286 * that there were valid bytes of data in that range of the file,
287 * and that the new data won't completely replace the old data in
288 * that range of the file.
290 static int nfs_want_read_modify_write(struct file *file, struct page *page,
291 loff_t pos, unsigned len)
293 unsigned int pglen = nfs_page_length(page);
294 unsigned int offset = pos & (PAGE_SIZE - 1);
295 unsigned int end = offset + len;
297 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
298 if (!PageUptodate(page))
303 if ((file->f_mode & FMODE_READ) && /* open for read? */
304 !PageUptodate(page) && /* Uptodate? */
305 !PagePrivate(page) && /* i/o request already? */
306 pglen && /* valid bytes of file? */
307 (end < pglen || offset)) /* replace all valid bytes? */
313 * This does the "real" work of the write. We must allocate and lock the
314 * page to be sent back to the generic routine, which then copies the
315 * data from user space.
317 * If the writer ends up delaying the write, the writer needs to
318 * increment the page use counts until he is done with the page.
320 static int nfs_write_begin(struct file *file, struct address_space *mapping,
321 loff_t pos, unsigned len, unsigned flags,
322 struct page **pagep, void **fsdata)
325 pgoff_t index = pos >> PAGE_SHIFT;
329 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
330 file, mapping->host->i_ino, len, (long long) pos);
333 page = grab_cache_page_write_begin(mapping, index, flags);
338 ret = nfs_flush_incompatible(file, page);
342 } else if (!once_thru &&
343 nfs_want_read_modify_write(file, page, pos, len)) {
345 ret = nfs_readpage(file, page);
353 static int nfs_write_end(struct file *file, struct address_space *mapping,
354 loff_t pos, unsigned len, unsigned copied,
355 struct page *page, void *fsdata)
357 unsigned offset = pos & (PAGE_SIZE - 1);
358 struct nfs_open_context *ctx = nfs_file_open_context(file);
361 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
362 file, mapping->host->i_ino, len, (long long) pos);
365 * Zero any uninitialised parts of the page, and then mark the page
366 * as up to date if it turns out that we're extending the file.
368 if (!PageUptodate(page)) {
369 unsigned pglen = nfs_page_length(page);
370 unsigned end = offset + copied;
373 zero_user_segments(page, 0, offset,
375 SetPageUptodate(page);
376 } else if (end >= pglen) {
377 zero_user_segment(page, end, PAGE_SIZE);
379 SetPageUptodate(page);
381 zero_user_segment(page, pglen, PAGE_SIZE);
384 status = nfs_updatepage(file, page, offset, copied);
391 NFS_I(mapping->host)->write_io += copied;
393 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
394 status = nfs_wb_all(mapping->host);
403 * Partially or wholly invalidate a page
404 * - Release the private state associated with a page if undergoing complete
406 * - Called if either PG_private or PG_fscache is set on the page
407 * - Caller holds page lock
409 static void nfs_invalidate_page(struct page *page, unsigned int offset,
412 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
413 page, offset, length);
415 if (offset != 0 || length < PAGE_SIZE)
417 /* Cancel any unstarted writes on this page */
418 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
420 nfs_fscache_invalidate_page(page, page->mapping->host);
424 * Attempt to release the private state associated with a page
425 * - Called if either PG_private or PG_fscache is set on the page
426 * - Caller holds page lock
427 * - Return true (may release page) or false (may not)
429 static int nfs_release_page(struct page *page, gfp_t gfp)
431 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
433 /* If PagePrivate() is set, then the page is not freeable */
434 if (PagePrivate(page))
436 return nfs_fscache_release_page(page, gfp);
439 static void nfs_check_dirty_writeback(struct page *page,
440 bool *dirty, bool *writeback)
442 struct nfs_inode *nfsi;
443 struct address_space *mapping = page_file_mapping(page);
445 if (!mapping || PageSwapCache(page))
449 * Check if an unstable page is currently being committed and
450 * if so, have the VM treat it as if the page is under writeback
451 * so it will not block due to pages that will shortly be freeable.
453 nfsi = NFS_I(mapping->host);
454 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
460 * If PagePrivate() is set, then the page is not freeable and as the
461 * inode is not being committed, it's not going to be cleaned in the
462 * near future so treat it as dirty
464 if (PagePrivate(page))
469 * Attempt to clear the private state associated with a page when an error
470 * occurs that requires the cached contents of an inode to be written back or
472 * - Called if either PG_private or fscache is set on the page
473 * - Caller holds page lock
474 * - Return 0 if successful, -error otherwise
476 static int nfs_launder_page(struct page *page)
478 struct inode *inode = page_file_mapping(page)->host;
479 struct nfs_inode *nfsi = NFS_I(inode);
481 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
482 inode->i_ino, (long long)page_offset(page));
484 nfs_fscache_wait_on_page_write(nfsi, page);
485 return nfs_wb_launder_page(inode, page);
488 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
491 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
495 return rpc_clnt_swap_activate(clnt);
498 static void nfs_swap_deactivate(struct file *file)
500 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
502 rpc_clnt_swap_deactivate(clnt);
505 const struct address_space_operations nfs_file_aops = {
506 .readpage = nfs_readpage,
507 .readpages = nfs_readpages,
508 .set_page_dirty = __set_page_dirty_nobuffers,
509 .writepage = nfs_writepage,
510 .writepages = nfs_writepages,
511 .write_begin = nfs_write_begin,
512 .write_end = nfs_write_end,
513 .invalidatepage = nfs_invalidate_page,
514 .releasepage = nfs_release_page,
515 .direct_IO = nfs_direct_IO,
516 #ifdef CONFIG_MIGRATION
517 .migratepage = nfs_migrate_page,
519 .launder_page = nfs_launder_page,
520 .is_dirty_writeback = nfs_check_dirty_writeback,
521 .error_remove_page = generic_error_remove_page,
522 .swap_activate = nfs_swap_activate,
523 .swap_deactivate = nfs_swap_deactivate,
527 * Notification that a PTE pointing to an NFS page is about to be made
528 * writable, implying that someone is about to modify the page through a
529 * shared-writable mapping
531 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
533 struct page *page = vmf->page;
534 struct file *filp = vma->vm_file;
535 struct inode *inode = file_inode(filp);
537 int ret = VM_FAULT_NOPAGE;
538 struct address_space *mapping;
540 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
541 filp, filp->f_mapping->host->i_ino,
542 (long long)page_offset(page));
544 sb_start_pagefault(inode->i_sb);
546 /* make sure the cache has finished storing the page */
547 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
549 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
550 nfs_wait_bit_killable, TASK_KILLABLE);
553 mapping = page_file_mapping(page);
554 if (mapping != inode->i_mapping)
557 wait_on_page_writeback(page);
559 pagelen = nfs_page_length(page);
563 ret = VM_FAULT_LOCKED;
564 if (nfs_flush_incompatible(filp, page) == 0 &&
565 nfs_updatepage(filp, page, 0, pagelen) == 0)
568 ret = VM_FAULT_SIGBUS;
572 sb_end_pagefault(inode->i_sb);
576 static const struct vm_operations_struct nfs_file_vm_ops = {
577 .fault = filemap_fault,
578 .map_pages = filemap_map_pages,
579 .page_mkwrite = nfs_vm_page_mkwrite,
582 static int nfs_need_check_write(struct file *filp, struct inode *inode)
584 struct nfs_open_context *ctx;
586 ctx = nfs_file_open_context(filp);
587 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
588 nfs_ctx_key_to_expire(ctx, inode))
593 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
595 struct file *file = iocb->ki_filp;
596 struct inode *inode = file_inode(file);
597 unsigned long written = 0;
600 result = nfs_key_timeout_notify(file, inode);
604 if (iocb->ki_flags & IOCB_DIRECT)
605 return nfs_file_direct_write(iocb, from);
607 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
608 file, iov_iter_count(from), (long long) iocb->ki_pos);
610 if (IS_SWAPFILE(inode))
613 * O_APPEND implies that we must revalidate the file length.
615 if (iocb->ki_flags & IOCB_APPEND) {
616 result = nfs_revalidate_file_size(inode, file);
621 nfs_start_io_write(inode);
622 result = generic_write_checks(iocb, from);
624 current->backing_dev_info = inode_to_bdi(inode);
625 result = generic_perform_write(file, from, iocb->ki_pos);
626 current->backing_dev_info = NULL;
628 nfs_end_io_write(inode);
632 result = generic_write_sync(iocb, result);
636 iocb->ki_pos += written;
638 /* Return error values */
639 if (nfs_need_check_write(file, inode)) {
640 int err = vfs_fsync(file, 0);
644 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
649 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
652 EXPORT_SYMBOL_GPL(nfs_file_write);
655 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
657 struct inode *inode = filp->f_mapping->host;
659 unsigned int saved_type = fl->fl_type;
661 /* Try local locking first */
662 posix_test_lock(filp, fl);
663 if (fl->fl_type != F_UNLCK) {
664 /* found a conflict */
667 fl->fl_type = saved_type;
669 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
675 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
679 fl->fl_type = F_UNLCK;
684 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
686 struct inode *inode = filp->f_mapping->host;
687 struct nfs_lock_context *l_ctx;
691 * Flush all pending writes before doing anything
696 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
697 if (!IS_ERR(l_ctx)) {
698 status = nfs_iocounter_wait(l_ctx);
699 nfs_put_lock_context(l_ctx);
704 /* NOTE: special case
705 * If we're signalled while cleaning up locks on process exit, we
706 * still need to complete the unlock.
709 * Use local locking if mounted with "-onolock" or with appropriate
713 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
715 status = locks_lock_file_wait(filp, fl);
720 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
722 struct inode *inode = filp->f_mapping->host;
726 * Flush all pending writes before doing anything
729 status = nfs_sync_mapping(filp->f_mapping);
734 * Use local locking if mounted with "-onolock" or with appropriate
738 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
740 status = locks_lock_file_wait(filp, fl);
745 * Revalidate the cache if the server has time stamps granular
746 * enough to detect subsecond changes. Otherwise, clear the
747 * cache to prevent missing any changes.
749 * This makes locking act as a cache coherency point.
751 nfs_sync_mapping(filp->f_mapping);
752 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
753 nfs_zap_mapping(inode, filp->f_mapping);
759 * Lock a (portion of) a file
761 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
763 struct inode *inode = filp->f_mapping->host;
767 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
768 filp, fl->fl_type, fl->fl_flags,
769 (long long)fl->fl_start, (long long)fl->fl_end);
771 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
773 /* No mandatory locks over NFS */
774 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
777 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
780 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
781 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
787 ret = do_getlk(filp, cmd, fl, is_local);
788 else if (fl->fl_type == F_UNLCK)
789 ret = do_unlk(filp, cmd, fl, is_local);
791 ret = do_setlk(filp, cmd, fl, is_local);
795 EXPORT_SYMBOL_GPL(nfs_lock);
798 * Lock a (portion of) a file
800 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
802 struct inode *inode = filp->f_mapping->host;
805 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
806 filp, fl->fl_type, fl->fl_flags);
808 if (!(fl->fl_flags & FL_FLOCK))
812 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
813 * any standard. In principle we might be able to support LOCK_MAND
814 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
815 * NFS code is not set up for it.
817 if (fl->fl_type & LOCK_MAND)
820 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
823 /* We're simulating flock() locks using posix locks on the server */
824 if (fl->fl_type == F_UNLCK)
825 return do_unlk(filp, cmd, fl, is_local);
826 return do_setlk(filp, cmd, fl, is_local);
828 EXPORT_SYMBOL_GPL(nfs_flock);
830 const struct file_operations nfs_file_operations = {
831 .llseek = nfs_file_llseek,
832 .read_iter = nfs_file_read,
833 .write_iter = nfs_file_write,
834 .mmap = nfs_file_mmap,
835 .open = nfs_file_open,
836 .flush = nfs_file_flush,
837 .release = nfs_file_release,
838 .fsync = nfs_file_fsync,
841 .splice_read = generic_file_splice_read,
842 .splice_write = iter_file_splice_write,
843 .check_flags = nfs_check_flags,
844 .setlease = simple_nosetlease,
846 EXPORT_SYMBOL_GPL(nfs_file_operations);