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
10 * and implementation by Wai S Kok elekokws@ee.nus.sg.
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 <asm/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);
104 struct nfs_inode *nfsi = NFS_I(inode);
105 const unsigned long force_reval = NFS_INO_REVAL_PAGECACHE|NFS_INO_REVAL_FORCED;
106 unsigned long cache_validity = nfsi->cache_validity;
108 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) &&
109 (cache_validity & force_reval) != force_reval)
112 if (filp->f_flags & O_DIRECT)
114 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
116 if (nfs_attribute_timeout(inode))
121 return __nfs_revalidate_inode(server, inode);
124 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
126 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
127 filp, offset, whence);
130 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
131 * the cached file length
133 if (whence != SEEK_SET && whence != SEEK_CUR) {
134 struct inode *inode = filp->f_mapping->host;
136 int retval = nfs_revalidate_file_size(inode, filp);
138 return (loff_t)retval;
141 return generic_file_llseek(filp, offset, whence);
143 EXPORT_SYMBOL_GPL(nfs_file_llseek);
146 * Flush all dirty pages, and check for write errors.
149 nfs_file_flush(struct file *file, fl_owner_t id)
151 struct inode *inode = file_inode(file);
153 dprintk("NFS: flush(%pD2)\n", file);
155 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
156 if ((file->f_mode & FMODE_WRITE) == 0)
159 /* Flush writes to the server and return any errors */
160 return vfs_fsync(file, 0);
164 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
166 struct inode *inode = file_inode(iocb->ki_filp);
169 if (iocb->ki_flags & IOCB_DIRECT)
170 return nfs_file_direct_read(iocb, to);
172 dprintk("NFS: read(%pD2, %zu@%lu)\n",
174 iov_iter_count(to), (unsigned long) iocb->ki_pos);
176 nfs_start_io_read(inode);
177 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
179 result = generic_file_read_iter(iocb, to);
181 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
183 nfs_end_io_read(inode);
186 EXPORT_SYMBOL_GPL(nfs_file_read);
189 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
191 struct inode *inode = file_inode(file);
194 dprintk("NFS: mmap(%pD2)\n", file);
196 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
197 * so we call that before revalidating the mapping
199 status = generic_file_mmap(file, vma);
201 vma->vm_ops = &nfs_file_vm_ops;
202 status = nfs_revalidate_mapping(inode, file->f_mapping);
206 EXPORT_SYMBOL_GPL(nfs_file_mmap);
209 * Flush any dirty pages for this process, and check for write errors.
210 * The return status from this call provides a reliable indication of
211 * whether any write errors occurred for this process.
213 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
214 * disk, but it retrieves and clears ctx->error after synching, despite
215 * the two being set at the same time in nfs_context_set_write_error().
216 * This is because the former is used to notify the _next_ call to
217 * nfs_file_write() that a write error occurred, and hence cause it to
218 * fall back to doing a synchronous write.
221 nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
223 struct nfs_open_context *ctx = nfs_file_open_context(file);
224 struct inode *inode = file_inode(file);
225 int have_error, do_resend, status;
228 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
230 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
231 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
232 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
233 status = nfs_commit_inode(inode, FLUSH_SYNC);
234 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
236 ret = xchg(&ctx->error, 0);
244 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
252 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
255 struct inode *inode = file_inode(file);
257 trace_nfs_fsync_enter(inode);
260 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
263 ret = nfs_file_fsync_commit(file, start, end, datasync);
265 ret = pnfs_sync_inode(inode, !!datasync);
267 * If nfs_file_fsync_commit detected a server reboot, then
268 * resend all dirty pages that might have been covered by
269 * the NFS_CONTEXT_RESEND_WRITES flag
273 } while (ret == -EAGAIN);
275 trace_nfs_fsync_exit(inode, ret);
278 EXPORT_SYMBOL_GPL(nfs_file_fsync);
281 * Decide whether a read/modify/write cycle may be more efficient
282 * then a modify/write/read cycle when writing to a page in the
285 * The modify/write/read cycle may occur if a page is read before
286 * being completely filled by the writer. In this situation, the
287 * page must be completely written to stable storage on the server
288 * before it can be refilled by reading in the page from the server.
289 * This can lead to expensive, small, FILE_SYNC mode writes being
292 * It may be more efficient to read the page first if the file is
293 * open for reading in addition to writing, the page is not marked
294 * as Uptodate, it is not dirty or waiting to be committed,
295 * indicating that it was previously allocated and then modified,
296 * that there were valid bytes of data in that range of the file,
297 * and that the new data won't completely replace the old data in
298 * that range of the file.
300 static int nfs_want_read_modify_write(struct file *file, struct page *page,
301 loff_t pos, unsigned len)
303 unsigned int pglen = nfs_page_length(page);
304 unsigned int offset = pos & (PAGE_SIZE - 1);
305 unsigned int end = offset + len;
307 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
308 if (!PageUptodate(page))
313 if ((file->f_mode & FMODE_READ) && /* open for read? */
314 !PageUptodate(page) && /* Uptodate? */
315 !PagePrivate(page) && /* i/o request already? */
316 pglen && /* valid bytes of file? */
317 (end < pglen || offset)) /* replace all valid bytes? */
323 * This does the "real" work of the write. We must allocate and lock the
324 * page to be sent back to the generic routine, which then copies the
325 * data from user space.
327 * If the writer ends up delaying the write, the writer needs to
328 * increment the page use counts until he is done with the page.
330 static int nfs_write_begin(struct file *file, struct address_space *mapping,
331 loff_t pos, unsigned len, unsigned flags,
332 struct page **pagep, void **fsdata)
335 pgoff_t index = pos >> PAGE_SHIFT;
339 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
340 file, mapping->host->i_ino, len, (long long) pos);
343 page = grab_cache_page_write_begin(mapping, index, flags);
348 ret = nfs_flush_incompatible(file, page);
352 } else if (!once_thru &&
353 nfs_want_read_modify_write(file, page, pos, len)) {
355 ret = nfs_readpage(file, page);
363 static int nfs_write_end(struct file *file, struct address_space *mapping,
364 loff_t pos, unsigned len, unsigned copied,
365 struct page *page, void *fsdata)
367 unsigned offset = pos & (PAGE_SIZE - 1);
368 struct nfs_open_context *ctx = nfs_file_open_context(file);
371 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
372 file, mapping->host->i_ino, len, (long long) pos);
375 * Zero any uninitialised parts of the page, and then mark the page
376 * as up to date if it turns out that we're extending the file.
378 if (!PageUptodate(page)) {
379 unsigned pglen = nfs_page_length(page);
380 unsigned end = offset + len;
383 zero_user_segments(page, 0, offset,
385 SetPageUptodate(page);
386 } else if (end >= pglen) {
387 zero_user_segment(page, end, PAGE_SIZE);
389 SetPageUptodate(page);
391 zero_user_segment(page, pglen, PAGE_SIZE);
394 status = nfs_updatepage(file, page, offset, copied);
401 NFS_I(mapping->host)->write_io += copied;
403 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
404 status = nfs_wb_all(mapping->host);
413 * Partially or wholly invalidate a page
414 * - Release the private state associated with a page if undergoing complete
416 * - Called if either PG_private or PG_fscache is set on the page
417 * - Caller holds page lock
419 static void nfs_invalidate_page(struct page *page, unsigned int offset,
422 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
423 page, offset, length);
425 if (offset != 0 || length < PAGE_SIZE)
427 /* Cancel any unstarted writes on this page */
428 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
430 nfs_fscache_invalidate_page(page, page->mapping->host);
434 * Attempt to release the private state associated with a page
435 * - Called if either PG_private or PG_fscache is set on the page
436 * - Caller holds page lock
437 * - Return true (may release page) or false (may not)
439 static int nfs_release_page(struct page *page, gfp_t gfp)
441 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
443 /* If PagePrivate() is set, then the page is not freeable */
444 if (PagePrivate(page))
446 return nfs_fscache_release_page(page, gfp);
449 static void nfs_check_dirty_writeback(struct page *page,
450 bool *dirty, bool *writeback)
452 struct nfs_inode *nfsi;
453 struct address_space *mapping = page_file_mapping(page);
455 if (!mapping || PageSwapCache(page))
459 * Check if an unstable page is currently being committed and
460 * if so, have the VM treat it as if the page is under writeback
461 * so it will not block due to pages that will shortly be freeable.
463 nfsi = NFS_I(mapping->host);
464 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
470 * If PagePrivate() is set, then the page is not freeable and as the
471 * inode is not being committed, it's not going to be cleaned in the
472 * near future so treat it as dirty
474 if (PagePrivate(page))
479 * Attempt to clear the private state associated with a page when an error
480 * occurs that requires the cached contents of an inode to be written back or
482 * - Called if either PG_private or fscache is set on the page
483 * - Caller holds page lock
484 * - Return 0 if successful, -error otherwise
486 static int nfs_launder_page(struct page *page)
488 struct inode *inode = page_file_mapping(page)->host;
489 struct nfs_inode *nfsi = NFS_I(inode);
491 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
492 inode->i_ino, (long long)page_offset(page));
494 nfs_fscache_wait_on_page_write(nfsi, page);
495 return nfs_wb_launder_page(inode, page);
498 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
501 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
505 return rpc_clnt_swap_activate(clnt);
508 static void nfs_swap_deactivate(struct file *file)
510 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
512 rpc_clnt_swap_deactivate(clnt);
515 const struct address_space_operations nfs_file_aops = {
516 .readpage = nfs_readpage,
517 .readpages = nfs_readpages,
518 .set_page_dirty = __set_page_dirty_nobuffers,
519 .writepage = nfs_writepage,
520 .writepages = nfs_writepages,
521 .write_begin = nfs_write_begin,
522 .write_end = nfs_write_end,
523 .invalidatepage = nfs_invalidate_page,
524 .releasepage = nfs_release_page,
525 .direct_IO = nfs_direct_IO,
526 #ifdef CONFIG_MIGRATION
527 .migratepage = nfs_migrate_page,
529 .launder_page = nfs_launder_page,
530 .is_dirty_writeback = nfs_check_dirty_writeback,
531 .error_remove_page = generic_error_remove_page,
532 .swap_activate = nfs_swap_activate,
533 .swap_deactivate = nfs_swap_deactivate,
537 * Notification that a PTE pointing to an NFS page is about to be made
538 * writable, implying that someone is about to modify the page through a
539 * shared-writable mapping
541 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
543 struct page *page = vmf->page;
544 struct file *filp = vma->vm_file;
545 struct inode *inode = file_inode(filp);
547 int ret = VM_FAULT_NOPAGE;
548 struct address_space *mapping;
550 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
551 filp, filp->f_mapping->host->i_ino,
552 (long long)page_offset(page));
554 sb_start_pagefault(inode->i_sb);
556 /* make sure the cache has finished storing the page */
557 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
559 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
560 nfs_wait_bit_killable, TASK_KILLABLE);
563 mapping = page_file_mapping(page);
564 if (mapping != inode->i_mapping)
567 wait_on_page_writeback(page);
569 pagelen = nfs_page_length(page);
573 ret = VM_FAULT_LOCKED;
574 if (nfs_flush_incompatible(filp, page) == 0 &&
575 nfs_updatepage(filp, page, 0, pagelen) == 0)
578 ret = VM_FAULT_SIGBUS;
582 sb_end_pagefault(inode->i_sb);
586 static const struct vm_operations_struct nfs_file_vm_ops = {
587 .fault = filemap_fault,
588 .map_pages = filemap_map_pages,
589 .page_mkwrite = nfs_vm_page_mkwrite,
592 static int nfs_need_check_write(struct file *filp, struct inode *inode)
594 struct nfs_open_context *ctx;
596 ctx = nfs_file_open_context(filp);
597 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
598 nfs_ctx_key_to_expire(ctx, inode))
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 long written = 0;
610 result = nfs_key_timeout_notify(file, inode);
614 if (iocb->ki_flags & IOCB_DIRECT)
615 return nfs_file_direct_write(iocb, from);
617 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
618 file, iov_iter_count(from), (long long) iocb->ki_pos);
620 if (IS_SWAPFILE(inode))
623 * O_APPEND implies that we must revalidate the file length.
625 if (iocb->ki_flags & IOCB_APPEND) {
626 result = nfs_revalidate_file_size(inode, file);
631 nfs_start_io_write(inode);
632 result = generic_write_checks(iocb, from);
634 current->backing_dev_info = inode_to_bdi(inode);
635 result = generic_perform_write(file, from, iocb->ki_pos);
636 current->backing_dev_info = NULL;
638 nfs_end_io_write(inode);
642 result = generic_write_sync(iocb, result);
646 iocb->ki_pos += written;
648 /* Return error values */
649 if (nfs_need_check_write(file, inode)) {
650 int err = vfs_fsync(file, 0);
654 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
659 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
662 EXPORT_SYMBOL_GPL(nfs_file_write);
665 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
667 struct inode *inode = filp->f_mapping->host;
669 unsigned int saved_type = fl->fl_type;
671 /* Try local locking first */
672 posix_test_lock(filp, fl);
673 if (fl->fl_type != F_UNLCK) {
674 /* found a conflict */
677 fl->fl_type = saved_type;
679 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
685 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
689 fl->fl_type = F_UNLCK;
694 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
696 struct inode *inode = filp->f_mapping->host;
697 struct nfs_lock_context *l_ctx;
701 * Flush all pending writes before doing anything
706 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
707 if (!IS_ERR(l_ctx)) {
708 status = nfs_iocounter_wait(l_ctx);
709 nfs_put_lock_context(l_ctx);
714 /* NOTE: special case
715 * If we're signalled while cleaning up locks on process exit, we
716 * still need to complete the unlock.
719 * Use local locking if mounted with "-onolock" or with appropriate
723 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
725 status = locks_lock_file_wait(filp, fl);
730 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
732 struct inode *inode = filp->f_mapping->host;
736 * Flush all pending writes before doing anything
739 status = nfs_sync_mapping(filp->f_mapping);
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 * Revalidate the cache if the server has time stamps granular
756 * enough to detect subsecond changes. Otherwise, clear the
757 * cache to prevent missing any changes.
759 * This makes locking act as a cache coherency point.
761 nfs_sync_mapping(filp->f_mapping);
762 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
763 nfs_zap_mapping(inode, filp->f_mapping);
769 * Lock a (portion of) a file
771 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
773 struct inode *inode = filp->f_mapping->host;
777 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
778 filp, fl->fl_type, fl->fl_flags,
779 (long long)fl->fl_start, (long long)fl->fl_end);
781 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
783 /* No mandatory locks over NFS */
784 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
787 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
790 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
791 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
797 ret = do_getlk(filp, cmd, fl, is_local);
798 else if (fl->fl_type == F_UNLCK)
799 ret = do_unlk(filp, cmd, fl, is_local);
801 ret = do_setlk(filp, cmd, fl, is_local);
805 EXPORT_SYMBOL_GPL(nfs_lock);
808 * Lock a (portion of) a file
810 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
812 struct inode *inode = filp->f_mapping->host;
815 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
816 filp, fl->fl_type, fl->fl_flags);
818 if (!(fl->fl_flags & FL_FLOCK))
822 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
823 * any standard. In principle we might be able to support LOCK_MAND
824 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
825 * NFS code is not set up for it.
827 if (fl->fl_type & LOCK_MAND)
830 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
833 /* We're simulating flock() locks using posix locks on the server */
834 if (fl->fl_type == F_UNLCK)
835 return do_unlk(filp, cmd, fl, is_local);
836 return do_setlk(filp, cmd, fl, is_local);
838 EXPORT_SYMBOL_GPL(nfs_flock);
840 const struct file_operations nfs_file_operations = {
841 .llseek = nfs_file_llseek,
842 .read_iter = nfs_file_read,
843 .write_iter = nfs_file_write,
844 .mmap = nfs_file_mmap,
845 .open = nfs_file_open,
846 .flush = nfs_file_flush,
847 .release = nfs_file_release,
848 .fsync = nfs_file_fsync,
851 .splice_read = generic_file_splice_read,
852 .splice_write = iter_file_splice_write,
853 .check_flags = nfs_check_flags,
854 .setlease = simple_nosetlease,
856 EXPORT_SYMBOL_GPL(nfs_file_operations);