1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
7 #include <linux/slab.h>
8 #include <linux/spinlock.h>
9 #include <linux/compat.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
12 #include <linux/pagemap.h>
13 #include <linux/uio.h>
14 #include <linux/blkdev.h>
16 #include <linux/mount.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/falloc.h>
20 #include <linux/swap.h>
21 #include <linux/crc32.h>
22 #include <linux/writeback.h>
23 #include <linux/uaccess.h>
24 #include <linux/dlm.h>
25 #include <linux/dlm_plock.h>
26 #include <linux/delay.h>
27 #include <linux/backing-dev.h>
28 #include <linux/fileattr.h>
46 * gfs2_llseek - seek to a location in a file
49 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
51 * SEEK_END requires the glock for the file because it references the
54 * Returns: The new offset, or errno
57 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
59 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
60 struct gfs2_holder i_gh;
65 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
68 error = generic_file_llseek(file, offset, whence);
69 gfs2_glock_dq_uninit(&i_gh);
74 error = gfs2_seek_data(file, offset);
78 error = gfs2_seek_hole(file, offset);
84 * These don't reference inode->i_size and don't depend on the
85 * block mapping, so we don't need the glock.
87 error = generic_file_llseek(file, offset, whence);
97 * gfs2_readdir - Iterator for a directory
98 * @file: The directory to read from
99 * @ctx: What to feed directory entries to
104 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
106 struct inode *dir = file->f_mapping->host;
107 struct gfs2_inode *dip = GFS2_I(dir);
108 struct gfs2_holder d_gh;
111 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
115 error = gfs2_dir_read(dir, ctx, &file->f_ra);
117 gfs2_glock_dq_uninit(&d_gh);
123 * struct fsflag_gfs2flag
125 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
126 * and to GFS2_DIF_JDATA for non-directories.
131 } fsflag_gfs2flag[] = {
132 {FS_SYNC_FL, GFS2_DIF_SYNC},
133 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
134 {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
135 {FS_NOATIME_FL, GFS2_DIF_NOATIME},
136 {FS_INDEX_FL, GFS2_DIF_EXHASH},
137 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
138 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
141 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
146 if (S_ISDIR(inode->i_mode))
147 gfsflags &= ~GFS2_DIF_JDATA;
149 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
151 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
152 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
153 fsflags |= fsflag_gfs2flag[i].fsflag;
157 int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
159 struct inode *inode = d_inode(dentry);
160 struct gfs2_inode *ip = GFS2_I(inode);
161 struct gfs2_holder gh;
165 if (d_is_special(dentry))
168 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
169 error = gfs2_glock_nq(&gh);
173 fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
175 fileattr_fill_flags(fa, fsflags);
179 gfs2_holder_uninit(&gh);
183 void gfs2_set_inode_flags(struct inode *inode)
185 struct gfs2_inode *ip = GFS2_I(inode);
186 unsigned int flags = inode->i_flags;
188 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
189 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
191 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
192 flags |= S_IMMUTABLE;
193 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
195 if (ip->i_diskflags & GFS2_DIF_NOATIME)
197 if (ip->i_diskflags & GFS2_DIF_SYNC)
199 inode->i_flags = flags;
202 /* Flags that can be set by user space */
203 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
204 GFS2_DIF_IMMUTABLE| \
205 GFS2_DIF_APPENDONLY| \
209 GFS2_DIF_INHERIT_JDATA)
212 * do_gfs2_set_flags - set flags on an inode
214 * @reqflags: The flags to set
215 * @mask: Indicates which flags are valid
218 static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask)
220 struct gfs2_inode *ip = GFS2_I(inode);
221 struct gfs2_sbd *sdp = GFS2_SB(inode);
222 struct buffer_head *bh;
223 struct gfs2_holder gh;
225 u32 new_flags, flags;
227 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
232 flags = ip->i_diskflags;
233 new_flags = (flags & ~mask) | (reqflags & mask);
234 if ((new_flags ^ flags) == 0)
237 if (!IS_IMMUTABLE(inode)) {
238 error = gfs2_permission(&init_user_ns, inode, MAY_WRITE);
242 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
243 if (new_flags & GFS2_DIF_JDATA)
244 gfs2_log_flush(sdp, ip->i_gl,
245 GFS2_LOG_HEAD_FLUSH_NORMAL |
247 error = filemap_fdatawrite(inode->i_mapping);
250 error = filemap_fdatawait(inode->i_mapping);
253 if (new_flags & GFS2_DIF_JDATA)
254 gfs2_ordered_del_inode(ip);
256 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
259 error = gfs2_meta_inode_buffer(ip, &bh);
262 inode->i_ctime = current_time(inode);
263 gfs2_trans_add_meta(ip->i_gl, bh);
264 ip->i_diskflags = new_flags;
265 gfs2_dinode_out(ip, bh->b_data);
267 gfs2_set_inode_flags(inode);
268 gfs2_set_aops(inode);
272 gfs2_glock_dq_uninit(&gh);
276 int gfs2_fileattr_set(struct user_namespace *mnt_userns,
277 struct dentry *dentry, struct fileattr *fa)
279 struct inode *inode = d_inode(dentry);
280 u32 fsflags = fa->flags, gfsflags = 0;
284 if (d_is_special(dentry))
287 if (fileattr_has_fsx(fa))
290 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
291 if (fsflags & fsflag_gfs2flag[i].fsflag) {
292 fsflags &= ~fsflag_gfs2flag[i].fsflag;
293 gfsflags |= fsflag_gfs2flag[i].gfsflag;
296 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
299 mask = GFS2_FLAGS_USER_SET;
300 if (S_ISDIR(inode->i_mode)) {
301 mask &= ~GFS2_DIF_JDATA;
303 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
304 if (gfsflags & GFS2_DIF_TOPDIR)
306 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
309 return do_gfs2_set_flags(inode, gfsflags, mask);
312 static int gfs2_getlabel(struct file *filp, char __user *label)
314 struct inode *inode = file_inode(filp);
315 struct gfs2_sbd *sdp = GFS2_SB(inode);
317 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
323 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
327 return gfs2_fitrim(filp, (void __user *)arg);
328 case FS_IOC_GETFSLABEL:
329 return gfs2_getlabel(filp, (char __user *)arg);
336 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
339 /* Keep this list in sync with gfs2_ioctl */
341 case FS_IOC_GETFSLABEL:
347 return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
350 #define gfs2_compat_ioctl NULL
354 * gfs2_size_hint - Give a hint to the size of a write request
355 * @filep: The struct file
356 * @offset: The file offset of the write
357 * @size: The length of the write
359 * When we are about to do a write, this function records the total
360 * write size in order to provide a suitable hint to the lower layers
361 * about how many blocks will be required.
365 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
367 struct inode *inode = file_inode(filep);
368 struct gfs2_sbd *sdp = GFS2_SB(inode);
369 struct gfs2_inode *ip = GFS2_I(inode);
370 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
371 int hint = min_t(size_t, INT_MAX, blks);
373 if (hint > atomic_read(&ip->i_sizehint))
374 atomic_set(&ip->i_sizehint, hint);
378 * gfs2_allocate_page_backing - Allocate blocks for a write fault
379 * @page: The (locked) page to allocate backing for
380 * @length: Size of the allocation
382 * We try to allocate all the blocks required for the page in one go. This
383 * might fail for various reasons, so we keep trying until all the blocks to
384 * back this page are allocated. If some of the blocks are already allocated,
387 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
389 u64 pos = page_offset(page);
392 struct iomap iomap = { };
394 if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
397 if (length < iomap.length)
398 iomap.length = length;
399 length -= iomap.length;
401 } while (length > 0);
407 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
408 * @vmf: The virtual memory fault containing the page to become writable
410 * When the page becomes writable, we need to ensure that we have
411 * blocks allocated on disk to back that page.
414 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
416 struct page *page = vmf->page;
417 struct inode *inode = file_inode(vmf->vma->vm_file);
418 struct gfs2_inode *ip = GFS2_I(inode);
419 struct gfs2_sbd *sdp = GFS2_SB(inode);
420 struct gfs2_alloc_parms ap = { .aflags = 0, };
421 u64 offset = page_offset(page);
422 unsigned int data_blocks, ind_blocks, rblocks;
423 vm_fault_t ret = VM_FAULT_LOCKED;
424 struct gfs2_holder gh;
429 sb_start_pagefault(inode->i_sb);
431 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
432 err = gfs2_glock_nq(&gh);
434 ret = block_page_mkwrite_return(err);
438 /* Check page index against inode size */
439 size = i_size_read(inode);
440 if (offset >= size) {
441 ret = VM_FAULT_SIGBUS;
445 /* Update file times before taking page lock */
446 file_update_time(vmf->vma->vm_file);
448 /* page is wholly or partially inside EOF */
449 if (size - offset < PAGE_SIZE)
450 length = size - offset;
454 gfs2_size_hint(vmf->vma->vm_file, offset, length);
456 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
457 set_bit(GIF_SW_PAGED, &ip->i_flags);
460 * iomap_writepage / iomap_writepages currently don't support inline
461 * files, so always unstuff here.
464 if (!gfs2_is_stuffed(ip) &&
465 !gfs2_write_alloc_required(ip, offset, length)) {
467 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
468 ret = VM_FAULT_NOPAGE;
474 err = gfs2_rindex_update(sdp);
476 ret = block_page_mkwrite_return(err);
480 gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
481 ap.target = data_blocks + ind_blocks;
482 err = gfs2_quota_lock_check(ip, &ap);
484 ret = block_page_mkwrite_return(err);
487 err = gfs2_inplace_reserve(ip, &ap);
489 ret = block_page_mkwrite_return(err);
490 goto out_quota_unlock;
493 rblocks = RES_DINODE + ind_blocks;
494 if (gfs2_is_jdata(ip))
495 rblocks += data_blocks ? data_blocks : 1;
496 if (ind_blocks || data_blocks) {
497 rblocks += RES_STATFS + RES_QUOTA;
498 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
500 err = gfs2_trans_begin(sdp, rblocks, 0);
502 ret = block_page_mkwrite_return(err);
506 /* Unstuff, if required, and allocate backing blocks for page */
507 if (gfs2_is_stuffed(ip)) {
508 err = gfs2_unstuff_dinode(ip);
510 ret = block_page_mkwrite_return(err);
516 /* If truncated, we must retry the operation, we may have raced
517 * with the glock demotion code.
519 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
520 ret = VM_FAULT_NOPAGE;
521 goto out_page_locked;
524 err = gfs2_allocate_page_backing(page, length);
526 ret = block_page_mkwrite_return(err);
529 if (ret != VM_FAULT_LOCKED)
534 gfs2_inplace_release(ip);
536 gfs2_quota_unlock(ip);
540 gfs2_holder_uninit(&gh);
541 if (ret == VM_FAULT_LOCKED) {
542 set_page_dirty(page);
543 wait_for_stable_page(page);
545 sb_end_pagefault(inode->i_sb);
549 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
551 struct inode *inode = file_inode(vmf->vma->vm_file);
552 struct gfs2_inode *ip = GFS2_I(inode);
553 struct gfs2_holder gh;
557 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
558 err = gfs2_glock_nq(&gh);
560 ret = block_page_mkwrite_return(err);
563 ret = filemap_fault(vmf);
566 gfs2_holder_uninit(&gh);
570 static const struct vm_operations_struct gfs2_vm_ops = {
572 .map_pages = filemap_map_pages,
573 .page_mkwrite = gfs2_page_mkwrite,
578 * @file: The file to map
579 * @vma: The VMA which described the mapping
581 * There is no need to get a lock here unless we should be updating
582 * atime. We ignore any locking errors since the only consequence is
583 * a missed atime update (which will just be deferred until later).
588 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
590 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
592 if (!(file->f_flags & O_NOATIME) &&
593 !IS_NOATIME(&ip->i_inode)) {
594 struct gfs2_holder i_gh;
597 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
601 /* grab lock to update inode */
602 gfs2_glock_dq_uninit(&i_gh);
605 vma->vm_ops = &gfs2_vm_ops;
611 * gfs2_open_common - This is common to open and atomic_open
612 * @inode: The inode being opened
613 * @file: The file being opened
615 * This maybe called under a glock or not depending upon how it has
616 * been called. We must always be called under a glock for regular
617 * files, however. For other file types, it does not matter whether
618 * we hold the glock or not.
620 * Returns: Error code or 0 for success
623 int gfs2_open_common(struct inode *inode, struct file *file)
625 struct gfs2_file *fp;
628 if (S_ISREG(inode->i_mode)) {
629 ret = generic_file_open(inode, file);
634 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
638 mutex_init(&fp->f_fl_mutex);
640 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
641 file->private_data = fp;
642 if (file->f_mode & FMODE_WRITE) {
643 ret = gfs2_qa_get(GFS2_I(inode));
650 kfree(file->private_data);
651 file->private_data = NULL;
656 * gfs2_open - open a file
657 * @inode: the inode to open
658 * @file: the struct file for this opening
660 * After atomic_open, this function is only used for opening files
661 * which are already cached. We must still get the glock for regular
662 * files to ensure that we have the file size uptodate for the large
663 * file check which is in the common code. That is only an issue for
664 * regular files though.
669 static int gfs2_open(struct inode *inode, struct file *file)
671 struct gfs2_inode *ip = GFS2_I(inode);
672 struct gfs2_holder i_gh;
674 bool need_unlock = false;
676 if (S_ISREG(ip->i_inode.i_mode)) {
677 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
684 error = gfs2_open_common(inode, file);
687 gfs2_glock_dq_uninit(&i_gh);
693 * gfs2_release - called to close a struct file
694 * @inode: the inode the struct file belongs to
695 * @file: the struct file being closed
700 static int gfs2_release(struct inode *inode, struct file *file)
702 struct gfs2_inode *ip = GFS2_I(inode);
704 kfree(file->private_data);
705 file->private_data = NULL;
707 if (file->f_mode & FMODE_WRITE) {
708 if (gfs2_rs_active(&ip->i_res))
709 gfs2_rs_delete(ip, &inode->i_writecount);
716 * gfs2_fsync - sync the dirty data for a file (across the cluster)
717 * @file: the file that points to the dentry
718 * @start: the start position in the file to sync
719 * @end: the end position in the file to sync
720 * @datasync: set if we can ignore timestamp changes
722 * We split the data flushing here so that we don't wait for the data
723 * until after we've also sent the metadata to disk. Note that for
724 * data=ordered, we will write & wait for the data at the log flush
725 * stage anyway, so this is unlikely to make much of a difference
726 * except in the data=writeback case.
728 * If the fdatawrite fails due to any reason except -EIO, we will
729 * continue the remainder of the fsync, although we'll still report
730 * the error at the end. This is to match filemap_write_and_wait_range()
736 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
739 struct address_space *mapping = file->f_mapping;
740 struct inode *inode = mapping->host;
741 int sync_state = inode->i_state & I_DIRTY;
742 struct gfs2_inode *ip = GFS2_I(inode);
743 int ret = 0, ret1 = 0;
745 if (mapping->nrpages) {
746 ret1 = filemap_fdatawrite_range(mapping, start, end);
751 if (!gfs2_is_jdata(ip))
752 sync_state &= ~I_DIRTY_PAGES;
754 sync_state &= ~I_DIRTY_SYNC;
757 ret = sync_inode_metadata(inode, 1);
760 if (gfs2_is_jdata(ip))
761 ret = file_write_and_wait(file);
764 gfs2_ail_flush(ip->i_gl, 1);
767 if (mapping->nrpages)
768 ret = file_fdatawait_range(file, start, end);
770 return ret ? ret : ret1;
773 static inline bool should_fault_in_pages(ssize_t ret, struct iov_iter *i,
777 size_t count = iov_iter_count(i);
783 if (ret <= 0 && ret != -EFAULT)
785 if (!iter_is_iovec(i))
788 if (*prev_count != count || !*window_size) {
789 int pages, nr_dirtied;
791 pages = min_t(int, BIO_MAX_VECS, DIV_ROUND_UP(count, PAGE_SIZE));
792 nr_dirtied = max(current->nr_dirtied_pause -
793 current->nr_dirtied, 1);
794 pages = min(pages, nr_dirtied);
798 p = i->iov[0].iov_base + i->iov_offset;
799 *window_size = (size_t)PAGE_SIZE * pages - offset_in_page(p);
803 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
804 struct gfs2_holder *gh)
806 struct file *file = iocb->ki_filp;
807 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
808 size_t prev_count = 0, window_size = 0;
813 * In this function, we disable page faults when we're holding the
814 * inode glock while doing I/O. If a page fault occurs, we indicate
815 * that the inode glock may be dropped, fault in the pages manually,
818 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
819 * physical as well as manual page faults, and we need to disable both
822 * For direct I/O, gfs2 takes the inode glock in deferred mode. This
823 * locking mode is compatible with other deferred holders, so multiple
824 * processes and nodes can do direct I/O to a file at the same time.
825 * There's no guarantee that reads or writes will be atomic. Any
826 * coordination among readers and writers needs to happen externally.
829 if (!iov_iter_count(to))
830 return 0; /* skip atime */
832 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
834 ret = gfs2_glock_nq(gh);
840 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
841 IOMAP_DIO_PARTIAL, written);
847 if (should_fault_in_pages(ret, to, &prev_count, &window_size)) {
850 gfs2_holder_allow_demote(gh);
851 leftover = fault_in_iov_iter_writeable(to, window_size);
852 gfs2_holder_disallow_demote(gh);
853 if (leftover != window_size) {
854 if (!gfs2_holder_queued(gh))
856 goto retry_under_glock;
859 if (gfs2_holder_queued(gh))
862 gfs2_holder_uninit(gh);
868 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
869 struct gfs2_holder *gh)
871 struct file *file = iocb->ki_filp;
872 struct inode *inode = file->f_mapping->host;
873 struct gfs2_inode *ip = GFS2_I(inode);
874 size_t prev_count = 0, window_size = 0;
879 * In this function, we disable page faults when we're holding the
880 * inode glock while doing I/O. If a page fault occurs, we indicate
881 * that the inode glock may be dropped, fault in the pages manually,
884 * For writes, iomap_dio_rw only triggers manual page faults, so we
885 * don't need to disable physical ones.
889 * Deferred lock, even if its a write, since we do no allocation on
890 * this path. All we need to change is the atime, and this lock mode
891 * ensures that other nodes have flushed their buffered read caches
892 * (i.e. their page cache entries for this inode). We do not,
893 * unfortunately, have the option of only flushing a range like the
896 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
898 ret = gfs2_glock_nq(gh);
902 /* Silently fall back to buffered I/O when writing beyond EOF */
903 if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
906 from->nofault = true;
907 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
908 IOMAP_DIO_PARTIAL, read);
909 from->nofault = false;
916 if (should_fault_in_pages(ret, from, &prev_count, &window_size)) {
919 gfs2_holder_allow_demote(gh);
920 leftover = fault_in_iov_iter_readable(from, window_size);
921 gfs2_holder_disallow_demote(gh);
922 if (leftover != window_size) {
923 if (!gfs2_holder_queued(gh))
925 goto retry_under_glock;
929 if (gfs2_holder_queued(gh))
932 gfs2_holder_uninit(gh);
938 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
940 struct gfs2_inode *ip;
941 struct gfs2_holder gh;
942 size_t prev_count = 0, window_size = 0;
947 * In this function, we disable page faults when we're holding the
948 * inode glock while doing I/O. If a page fault occurs, we indicate
949 * that the inode glock may be dropped, fault in the pages manually,
953 if (iocb->ki_flags & IOCB_DIRECT) {
954 ret = gfs2_file_direct_read(iocb, to, &gh);
955 if (likely(ret != -ENOTBLK))
957 iocb->ki_flags &= ~IOCB_DIRECT;
959 iocb->ki_flags |= IOCB_NOIO;
960 ret = generic_file_read_iter(iocb, to);
961 iocb->ki_flags &= ~IOCB_NOIO;
963 if (!iov_iter_count(to))
969 if (iocb->ki_flags & IOCB_NOWAIT)
972 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
973 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
975 ret = gfs2_glock_nq(&gh);
980 ret = generic_file_read_iter(iocb, to);
985 if (should_fault_in_pages(ret, to, &prev_count, &window_size)) {
988 gfs2_holder_allow_demote(&gh);
989 leftover = fault_in_iov_iter_writeable(to, window_size);
990 gfs2_holder_disallow_demote(&gh);
991 if (leftover != window_size) {
992 if (!gfs2_holder_queued(&gh)) {
997 goto retry_under_glock;
1000 if (gfs2_holder_queued(&gh))
1003 gfs2_holder_uninit(&gh);
1004 return written ? written : ret;
1007 static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
1008 struct iov_iter *from,
1009 struct gfs2_holder *gh)
1011 struct file *file = iocb->ki_filp;
1012 struct inode *inode = file_inode(file);
1013 struct gfs2_inode *ip = GFS2_I(inode);
1014 struct gfs2_sbd *sdp = GFS2_SB(inode);
1015 struct gfs2_holder *statfs_gh = NULL;
1016 size_t prev_count = 0, window_size = 0;
1017 size_t orig_count = iov_iter_count(from);
1022 * In this function, we disable page faults when we're holding the
1023 * inode glock while doing I/O. If a page fault occurs, we indicate
1024 * that the inode glock may be dropped, fault in the pages manually,
1028 if (inode == sdp->sd_rindex) {
1029 statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1034 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1036 ret = gfs2_glock_nq(gh);
1040 if (inode == sdp->sd_rindex) {
1041 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1043 ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1044 GL_NOCACHE, statfs_gh);
1049 current->backing_dev_info = inode_to_bdi(inode);
1050 pagefault_disable();
1051 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
1053 current->backing_dev_info = NULL;
1055 iocb->ki_pos += ret;
1059 if (inode == sdp->sd_rindex)
1060 gfs2_glock_dq_uninit(statfs_gh);
1062 from->count = orig_count - read;
1063 if (should_fault_in_pages(ret, from, &prev_count, &window_size)) {
1066 gfs2_holder_allow_demote(gh);
1067 leftover = fault_in_iov_iter_readable(from, window_size);
1068 gfs2_holder_disallow_demote(gh);
1069 if (leftover != window_size) {
1070 from->count = min(from->count, window_size - leftover);
1071 if (!gfs2_holder_queued(gh)) {
1076 goto retry_under_glock;
1080 if (gfs2_holder_queued(gh))
1083 gfs2_holder_uninit(gh);
1086 return read ? read : ret;
1090 * gfs2_file_write_iter - Perform a write to a file
1091 * @iocb: The io context
1092 * @from: The data to write
1094 * We have to do a lock/unlock here to refresh the inode size for
1095 * O_APPEND writes, otherwise we can land up writing at the wrong
1096 * offset. There is still a race, but provided the app is using its
1097 * own file locking, this will make O_APPEND work as expected.
1101 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1103 struct file *file = iocb->ki_filp;
1104 struct inode *inode = file_inode(file);
1105 struct gfs2_inode *ip = GFS2_I(inode);
1106 struct gfs2_holder gh;
1109 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
1111 if (iocb->ki_flags & IOCB_APPEND) {
1112 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1115 gfs2_glock_dq_uninit(&gh);
1119 ret = generic_write_checks(iocb, from);
1123 ret = file_remove_privs(file);
1127 ret = file_update_time(file);
1131 if (iocb->ki_flags & IOCB_DIRECT) {
1132 struct address_space *mapping = file->f_mapping;
1133 ssize_t buffered, ret2;
1135 ret = gfs2_file_direct_write(iocb, from, &gh);
1136 if (ret < 0 || !iov_iter_count(from))
1139 iocb->ki_flags |= IOCB_DSYNC;
1140 buffered = gfs2_file_buffered_write(iocb, from, &gh);
1141 if (unlikely(buffered <= 0)) {
1148 * We need to ensure that the page cache pages are written to
1149 * disk and invalidated to preserve the expected O_DIRECT
1150 * semantics. If the writeback or invalidate fails, only report
1151 * the direct I/O range as we don't know if the buffered pages
1154 ret2 = generic_write_sync(iocb, buffered);
1155 invalidate_mapping_pages(mapping,
1156 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
1157 (iocb->ki_pos - 1) >> PAGE_SHIFT);
1158 if (!ret || ret2 > 0)
1161 ret = gfs2_file_buffered_write(iocb, from, &gh);
1162 if (likely(ret > 0))
1163 ret = generic_write_sync(iocb, ret);
1167 inode_unlock(inode);
1171 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1174 struct super_block *sb = inode->i_sb;
1175 struct gfs2_inode *ip = GFS2_I(inode);
1176 loff_t end = offset + len;
1177 struct buffer_head *dibh;
1180 error = gfs2_meta_inode_buffer(ip, &dibh);
1181 if (unlikely(error))
1184 gfs2_trans_add_meta(ip->i_gl, dibh);
1186 if (gfs2_is_stuffed(ip)) {
1187 error = gfs2_unstuff_dinode(ip);
1188 if (unlikely(error))
1192 while (offset < end) {
1193 struct iomap iomap = { };
1195 error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
1198 offset = iomap.offset + iomap.length;
1199 if (!(iomap.flags & IOMAP_F_NEW))
1201 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1202 iomap.length >> inode->i_blkbits,
1205 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1215 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1216 * blocks, determine how many bytes can be written.
1217 * @ip: The inode in question.
1218 * @len: Max cap of bytes. What we return in *len must be <= this.
1219 * @data_blocks: Compute and return the number of data blocks needed
1220 * @ind_blocks: Compute and return the number of indirect blocks needed
1221 * @max_blocks: The total blocks available to work with.
1223 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1225 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1226 unsigned int *data_blocks, unsigned int *ind_blocks,
1227 unsigned int max_blocks)
1230 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1231 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1233 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1234 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1238 *data_blocks = max_data;
1239 *ind_blocks = max_blocks - max_data;
1240 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1243 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1247 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1249 struct inode *inode = file_inode(file);
1250 struct gfs2_sbd *sdp = GFS2_SB(inode);
1251 struct gfs2_inode *ip = GFS2_I(inode);
1252 struct gfs2_alloc_parms ap = { .aflags = 0, };
1253 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1254 loff_t bytes, max_bytes, max_blks;
1256 const loff_t pos = offset;
1257 const loff_t count = len;
1258 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1259 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1260 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1262 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1264 offset &= bsize_mask;
1266 len = next - offset;
1267 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1270 bytes &= bsize_mask;
1272 bytes = sdp->sd_sb.sb_bsize;
1274 gfs2_size_hint(file, offset, len);
1276 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1277 ap.min_target = data_blocks + ind_blocks;
1282 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1288 /* We need to determine how many bytes we can actually
1289 * fallocate without exceeding quota or going over the
1290 * end of the fs. We start off optimistically by assuming
1291 * we can write max_bytes */
1292 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1294 /* Since max_bytes is most likely a theoretical max, we
1295 * calculate a more realistic 'bytes' to serve as a good
1296 * starting point for the number of bytes we may be able
1298 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1299 ap.target = data_blocks + ind_blocks;
1301 error = gfs2_quota_lock_check(ip, &ap);
1304 /* ap.allowed tells us how many blocks quota will allow
1305 * us to write. Check if this reduces max_blks */
1306 max_blks = UINT_MAX;
1308 max_blks = ap.allowed;
1310 error = gfs2_inplace_reserve(ip, &ap);
1314 /* check if the selected rgrp limits our max_blks further */
1315 if (ip->i_res.rs_reserved < max_blks)
1316 max_blks = ip->i_res.rs_reserved;
1318 /* Almost done. Calculate bytes that can be written using
1319 * max_blks. We also recompute max_bytes, data_blocks and
1321 calc_max_reserv(ip, &max_bytes, &data_blocks,
1322 &ind_blocks, max_blks);
1324 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1325 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1326 if (gfs2_is_jdata(ip))
1327 rblocks += data_blocks ? data_blocks : 1;
1329 error = gfs2_trans_begin(sdp, rblocks,
1330 PAGE_SIZE >> inode->i_blkbits);
1332 goto out_trans_fail;
1334 error = fallocate_chunk(inode, offset, max_bytes, mode);
1335 gfs2_trans_end(sdp);
1338 goto out_trans_fail;
1341 offset += max_bytes;
1342 gfs2_inplace_release(ip);
1343 gfs2_quota_unlock(ip);
1346 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1347 i_size_write(inode, pos + count);
1348 file_update_time(file);
1349 mark_inode_dirty(inode);
1351 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1352 return vfs_fsync_range(file, pos, pos + count - 1,
1353 (file->f_flags & __O_SYNC) ? 0 : 1);
1357 gfs2_inplace_release(ip);
1359 gfs2_quota_unlock(ip);
1363 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1365 struct inode *inode = file_inode(file);
1366 struct gfs2_sbd *sdp = GFS2_SB(inode);
1367 struct gfs2_inode *ip = GFS2_I(inode);
1368 struct gfs2_holder gh;
1371 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1373 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1374 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1379 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1380 ret = gfs2_glock_nq(&gh);
1384 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1385 (offset + len) > inode->i_size) {
1386 ret = inode_newsize_ok(inode, offset + len);
1391 ret = get_write_access(inode);
1395 if (mode & FALLOC_FL_PUNCH_HOLE) {
1396 ret = __gfs2_punch_hole(file, offset, len);
1398 ret = __gfs2_fallocate(file, mode, offset, len);
1400 gfs2_rs_deltree(&ip->i_res);
1403 put_write_access(inode);
1407 gfs2_holder_uninit(&gh);
1408 inode_unlock(inode);
1412 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1413 struct file *out, loff_t *ppos,
1414 size_t len, unsigned int flags)
1418 gfs2_size_hint(out, *ppos, len);
1420 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1424 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1427 * gfs2_lock - acquire/release a posix lock on a file
1428 * @file: the file pointer
1429 * @cmd: either modify or retrieve lock state, possibly wait
1430 * @fl: type and range of lock
1435 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1437 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1438 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1439 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1441 if (!(fl->fl_flags & FL_POSIX))
1443 if (cmd == F_CANCELLK) {
1446 fl->fl_type = F_UNLCK;
1448 if (unlikely(gfs2_withdrawn(sdp))) {
1449 if (fl->fl_type == F_UNLCK)
1450 locks_lock_file_wait(file, fl);
1454 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1455 else if (fl->fl_type == F_UNLCK)
1456 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1458 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1461 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1463 struct gfs2_file *fp = file->private_data;
1464 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1465 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1466 struct gfs2_glock *gl;
1472 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1473 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1475 mutex_lock(&fp->f_fl_mutex);
1477 if (gfs2_holder_initialized(fl_gh)) {
1478 struct file_lock request;
1479 if (fl_gh->gh_state == state)
1481 locks_init_lock(&request);
1482 request.fl_type = F_UNLCK;
1483 request.fl_flags = FL_FLOCK;
1484 locks_lock_file_wait(file, &request);
1485 gfs2_glock_dq(fl_gh);
1486 gfs2_holder_reinit(state, flags, fl_gh);
1488 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1489 &gfs2_flock_glops, CREATE, &gl);
1492 gfs2_holder_init(gl, state, flags, fl_gh);
1495 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1496 error = gfs2_glock_nq(fl_gh);
1497 if (error != GLR_TRYFAILED)
1499 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1500 fl_gh->gh_error = 0;
1504 gfs2_holder_uninit(fl_gh);
1505 if (error == GLR_TRYFAILED)
1508 error = locks_lock_file_wait(file, fl);
1509 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1513 mutex_unlock(&fp->f_fl_mutex);
1517 static void do_unflock(struct file *file, struct file_lock *fl)
1519 struct gfs2_file *fp = file->private_data;
1520 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1522 mutex_lock(&fp->f_fl_mutex);
1523 locks_lock_file_wait(file, fl);
1524 if (gfs2_holder_initialized(fl_gh)) {
1525 gfs2_glock_dq(fl_gh);
1526 gfs2_holder_uninit(fl_gh);
1528 mutex_unlock(&fp->f_fl_mutex);
1532 * gfs2_flock - acquire/release a flock lock on a file
1533 * @file: the file pointer
1534 * @cmd: either modify or retrieve lock state, possibly wait
1535 * @fl: type and range of lock
1540 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1542 if (!(fl->fl_flags & FL_FLOCK))
1545 if (fl->fl_type == F_UNLCK) {
1546 do_unflock(file, fl);
1549 return do_flock(file, cmd, fl);
1553 const struct file_operations gfs2_file_fops = {
1554 .llseek = gfs2_llseek,
1555 .read_iter = gfs2_file_read_iter,
1556 .write_iter = gfs2_file_write_iter,
1557 .iopoll = iocb_bio_iopoll,
1558 .unlocked_ioctl = gfs2_ioctl,
1559 .compat_ioctl = gfs2_compat_ioctl,
1562 .release = gfs2_release,
1563 .fsync = gfs2_fsync,
1565 .flock = gfs2_flock,
1566 .splice_read = generic_file_splice_read,
1567 .splice_write = gfs2_file_splice_write,
1568 .setlease = simple_nosetlease,
1569 .fallocate = gfs2_fallocate,
1572 const struct file_operations gfs2_dir_fops = {
1573 .iterate_shared = gfs2_readdir,
1574 .unlocked_ioctl = gfs2_ioctl,
1575 .compat_ioctl = gfs2_compat_ioctl,
1577 .release = gfs2_release,
1578 .fsync = gfs2_fsync,
1580 .flock = gfs2_flock,
1581 .llseek = default_llseek,
1584 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1586 const struct file_operations gfs2_file_fops_nolock = {
1587 .llseek = gfs2_llseek,
1588 .read_iter = gfs2_file_read_iter,
1589 .write_iter = gfs2_file_write_iter,
1590 .iopoll = iocb_bio_iopoll,
1591 .unlocked_ioctl = gfs2_ioctl,
1592 .compat_ioctl = gfs2_compat_ioctl,
1595 .release = gfs2_release,
1596 .fsync = gfs2_fsync,
1597 .splice_read = generic_file_splice_read,
1598 .splice_write = gfs2_file_splice_write,
1599 .setlease = generic_setlease,
1600 .fallocate = gfs2_fallocate,
1603 const struct file_operations gfs2_dir_fops_nolock = {
1604 .iterate_shared = gfs2_readdir,
1605 .unlocked_ioctl = gfs2_ioctl,
1606 .compat_ioctl = gfs2_compat_ioctl,
1608 .release = gfs2_release,
1609 .fsync = gfs2_fsync,
1610 .llseek = default_llseek,