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>
45 * gfs2_llseek - seek to a location in a file
48 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
50 * SEEK_END requires the glock for the file because it references the
53 * Returns: The new offset, or errno
56 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
58 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
59 struct gfs2_holder i_gh;
64 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
67 error = generic_file_llseek(file, offset, whence);
68 gfs2_glock_dq_uninit(&i_gh);
73 error = gfs2_seek_data(file, offset);
77 error = gfs2_seek_hole(file, offset);
83 * These don't reference inode->i_size and don't depend on the
84 * block mapping, so we don't need the glock.
86 error = generic_file_llseek(file, offset, whence);
96 * gfs2_readdir - Iterator for a directory
97 * @file: The directory to read from
98 * @ctx: What to feed directory entries to
103 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
105 struct inode *dir = file->f_mapping->host;
106 struct gfs2_inode *dip = GFS2_I(dir);
107 struct gfs2_holder d_gh;
110 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
114 error = gfs2_dir_read(dir, ctx, &file->f_ra);
116 gfs2_glock_dq_uninit(&d_gh);
124 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
125 * and to GFS2_DIF_JDATA for non-directories.
130 } fsflag_gfs2flag[] = {
131 {FS_SYNC_FL, GFS2_DIF_SYNC},
132 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
133 {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
134 {FS_NOATIME_FL, GFS2_DIF_NOATIME},
135 {FS_INDEX_FL, GFS2_DIF_EXHASH},
136 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
137 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
140 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
145 if (S_ISDIR(inode->i_mode))
146 gfsflags &= ~GFS2_DIF_JDATA;
148 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
150 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
151 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
152 fsflags |= fsflag_gfs2flag[i].fsflag;
156 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
158 struct inode *inode = file_inode(filp);
159 struct gfs2_inode *ip = GFS2_I(inode);
160 struct gfs2_holder gh;
164 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
165 error = gfs2_glock_nq(&gh);
169 fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
171 if (put_user(fsflags, ptr))
176 gfs2_holder_uninit(&gh);
180 void gfs2_set_inode_flags(struct inode *inode)
182 struct gfs2_inode *ip = GFS2_I(inode);
183 unsigned int flags = inode->i_flags;
185 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
186 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
188 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
189 flags |= S_IMMUTABLE;
190 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
192 if (ip->i_diskflags & GFS2_DIF_NOATIME)
194 if (ip->i_diskflags & GFS2_DIF_SYNC)
196 inode->i_flags = flags;
199 /* Flags that can be set by user space */
200 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
201 GFS2_DIF_IMMUTABLE| \
202 GFS2_DIF_APPENDONLY| \
206 GFS2_DIF_INHERIT_JDATA)
209 * do_gfs2_set_flags - set flags on an inode
210 * @filp: file pointer
211 * @reqflags: The flags to set
212 * @mask: Indicates which flags are valid
213 * @fsflags: The FS_* inode flags passed in
216 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask,
219 struct inode *inode = file_inode(filp);
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, oldflags;
227 error = mnt_want_write_file(filp);
231 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
235 oldflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
236 error = vfs_ioc_setflags_prepare(inode, oldflags, fsflags);
241 if (!inode_owner_or_capable(inode))
245 flags = ip->i_diskflags;
246 new_flags = (flags & ~mask) | (reqflags & mask);
247 if ((new_flags ^ flags) == 0)
251 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
253 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
255 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
256 !capable(CAP_LINUX_IMMUTABLE))
258 if (!IS_IMMUTABLE(inode)) {
259 error = gfs2_permission(inode, MAY_WRITE);
263 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
264 if (new_flags & GFS2_DIF_JDATA)
265 gfs2_log_flush(sdp, ip->i_gl,
266 GFS2_LOG_HEAD_FLUSH_NORMAL |
268 error = filemap_fdatawrite(inode->i_mapping);
271 error = filemap_fdatawait(inode->i_mapping);
274 if (new_flags & GFS2_DIF_JDATA)
275 gfs2_ordered_del_inode(ip);
277 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
280 error = gfs2_meta_inode_buffer(ip, &bh);
283 inode->i_ctime = current_time(inode);
284 gfs2_trans_add_meta(ip->i_gl, bh);
285 ip->i_diskflags = new_flags;
286 gfs2_dinode_out(ip, bh->b_data);
288 gfs2_set_inode_flags(inode);
289 gfs2_set_aops(inode);
293 gfs2_glock_dq_uninit(&gh);
295 mnt_drop_write_file(filp);
299 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
301 struct inode *inode = file_inode(filp);
302 u32 fsflags, gfsflags = 0;
306 if (get_user(fsflags, ptr))
309 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
310 if (fsflags & fsflag_gfs2flag[i].fsflag) {
311 fsflags &= ~fsflag_gfs2flag[i].fsflag;
312 gfsflags |= fsflag_gfs2flag[i].gfsflag;
315 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
318 mask = GFS2_FLAGS_USER_SET;
319 if (S_ISDIR(inode->i_mode)) {
320 mask &= ~GFS2_DIF_JDATA;
322 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
323 if (gfsflags & GFS2_DIF_TOPDIR)
325 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
328 return do_gfs2_set_flags(filp, gfsflags, mask, fsflags);
331 static int gfs2_getlabel(struct file *filp, char __user *label)
333 struct inode *inode = file_inode(filp);
334 struct gfs2_sbd *sdp = GFS2_SB(inode);
336 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
342 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
345 case FS_IOC_GETFLAGS:
346 return gfs2_get_flags(filp, (u32 __user *)arg);
347 case FS_IOC_SETFLAGS:
348 return gfs2_set_flags(filp, (u32 __user *)arg);
350 return gfs2_fitrim(filp, (void __user *)arg);
351 case FS_IOC_GETFSLABEL:
352 return gfs2_getlabel(filp, (char __user *)arg);
359 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
362 /* These are just misnamed, they actually get/put from/to user an int */
363 case FS_IOC32_GETFLAGS:
364 cmd = FS_IOC_GETFLAGS;
366 case FS_IOC32_SETFLAGS:
367 cmd = FS_IOC_SETFLAGS;
369 /* Keep this list in sync with gfs2_ioctl */
371 case FS_IOC_GETFSLABEL:
377 return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
380 #define gfs2_compat_ioctl NULL
384 * gfs2_size_hint - Give a hint to the size of a write request
385 * @filep: The struct file
386 * @offset: The file offset of the write
387 * @size: The length of the write
389 * When we are about to do a write, this function records the total
390 * write size in order to provide a suitable hint to the lower layers
391 * about how many blocks will be required.
395 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
397 struct inode *inode = file_inode(filep);
398 struct gfs2_sbd *sdp = GFS2_SB(inode);
399 struct gfs2_inode *ip = GFS2_I(inode);
400 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
401 int hint = min_t(size_t, INT_MAX, blks);
403 if (hint > atomic_read(&ip->i_sizehint))
404 atomic_set(&ip->i_sizehint, hint);
408 * gfs2_allocate_page_backing - Allocate blocks for a write fault
409 * @page: The (locked) page to allocate backing for
410 * @length: Size of the allocation
412 * We try to allocate all the blocks required for the page in one go. This
413 * might fail for various reasons, so we keep trying until all the blocks to
414 * back this page are allocated. If some of the blocks are already allocated,
417 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
419 u64 pos = page_offset(page);
422 struct iomap iomap = { };
424 if (gfs2_iomap_get_alloc(page->mapping->host, pos, length, &iomap))
427 if (length < iomap.length)
428 iomap.length = length;
429 length -= iomap.length;
431 } while (length > 0);
437 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
438 * @vma: The virtual memory area
439 * @vmf: The virtual memory fault containing the page to become writable
441 * When the page becomes writable, we need to ensure that we have
442 * blocks allocated on disk to back that page.
445 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
447 struct page *page = vmf->page;
448 struct inode *inode = file_inode(vmf->vma->vm_file);
449 struct gfs2_inode *ip = GFS2_I(inode);
450 struct gfs2_sbd *sdp = GFS2_SB(inode);
451 struct gfs2_alloc_parms ap = { .aflags = 0, };
452 u64 offset = page_offset(page);
453 unsigned int data_blocks, ind_blocks, rblocks;
454 struct gfs2_holder gh;
459 sb_start_pagefault(inode->i_sb);
461 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
462 ret = gfs2_glock_nq(&gh);
466 /* Check page index against inode size */
467 size = i_size_read(inode);
468 if (offset >= size) {
473 /* Update file times before taking page lock */
474 file_update_time(vmf->vma->vm_file);
476 /* page is wholly or partially inside EOF */
477 if (offset > size - PAGE_SIZE)
478 length = offset_in_page(size);
482 gfs2_size_hint(vmf->vma->vm_file, offset, length);
484 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
485 set_bit(GIF_SW_PAGED, &ip->i_flags);
488 * iomap_writepage / iomap_writepages currently don't support inline
489 * files, so always unstuff here.
492 if (!gfs2_is_stuffed(ip) &&
493 !gfs2_write_alloc_required(ip, offset, length)) {
495 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
502 ret = gfs2_rindex_update(sdp);
506 gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
507 ap.target = data_blocks + ind_blocks;
508 ret = gfs2_quota_lock_check(ip, &ap);
511 ret = gfs2_inplace_reserve(ip, &ap);
513 goto out_quota_unlock;
515 rblocks = RES_DINODE + ind_blocks;
516 if (gfs2_is_jdata(ip))
517 rblocks += data_blocks ? data_blocks : 1;
518 if (ind_blocks || data_blocks) {
519 rblocks += RES_STATFS + RES_QUOTA;
520 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
522 ret = gfs2_trans_begin(sdp, rblocks, 0);
528 /* If truncated, we must retry the operation, we may have raced
529 * with the glock demotion code.
531 if (!PageUptodate(page) || page->mapping != inode->i_mapping)
534 /* Unstuff, if required, and allocate backing blocks for page */
536 if (gfs2_is_stuffed(ip))
537 ret = gfs2_unstuff_dinode(ip, page);
539 ret = gfs2_allocate_page_backing(page, length);
546 gfs2_inplace_release(ip);
548 gfs2_quota_unlock(ip);
552 gfs2_holder_uninit(&gh);
554 set_page_dirty(page);
555 wait_for_stable_page(page);
557 sb_end_pagefault(inode->i_sb);
558 return block_page_mkwrite_return(ret);
561 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
563 struct inode *inode = file_inode(vmf->vma->vm_file);
564 struct gfs2_inode *ip = GFS2_I(inode);
565 struct gfs2_holder gh;
569 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
570 err = gfs2_glock_nq(&gh);
572 ret = block_page_mkwrite_return(err);
575 ret = filemap_fault(vmf);
578 gfs2_holder_uninit(&gh);
582 static const struct vm_operations_struct gfs2_vm_ops = {
584 .map_pages = filemap_map_pages,
585 .page_mkwrite = gfs2_page_mkwrite,
590 * @file: The file to map
591 * @vma: The VMA which described the mapping
593 * There is no need to get a lock here unless we should be updating
594 * atime. We ignore any locking errors since the only consequence is
595 * a missed atime update (which will just be deferred until later).
600 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
602 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
604 if (!(file->f_flags & O_NOATIME) &&
605 !IS_NOATIME(&ip->i_inode)) {
606 struct gfs2_holder i_gh;
609 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
613 /* grab lock to update inode */
614 gfs2_glock_dq_uninit(&i_gh);
617 vma->vm_ops = &gfs2_vm_ops;
623 * gfs2_open_common - This is common to open and atomic_open
624 * @inode: The inode being opened
625 * @file: The file being opened
627 * This maybe called under a glock or not depending upon how it has
628 * been called. We must always be called under a glock for regular
629 * files, however. For other file types, it does not matter whether
630 * we hold the glock or not.
632 * Returns: Error code or 0 for success
635 int gfs2_open_common(struct inode *inode, struct file *file)
637 struct gfs2_file *fp;
640 if (S_ISREG(inode->i_mode)) {
641 ret = generic_file_open(inode, file);
646 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
650 mutex_init(&fp->f_fl_mutex);
652 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
653 file->private_data = fp;
654 if (file->f_mode & FMODE_WRITE) {
655 ret = gfs2_qa_get(GFS2_I(inode));
662 kfree(file->private_data);
663 file->private_data = NULL;
668 * gfs2_open - open a file
669 * @inode: the inode to open
670 * @file: the struct file for this opening
672 * After atomic_open, this function is only used for opening files
673 * which are already cached. We must still get the glock for regular
674 * files to ensure that we have the file size uptodate for the large
675 * file check which is in the common code. That is only an issue for
676 * regular files though.
681 static int gfs2_open(struct inode *inode, struct file *file)
683 struct gfs2_inode *ip = GFS2_I(inode);
684 struct gfs2_holder i_gh;
686 bool need_unlock = false;
688 if (S_ISREG(ip->i_inode.i_mode)) {
689 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
696 error = gfs2_open_common(inode, file);
699 gfs2_glock_dq_uninit(&i_gh);
705 * gfs2_release - called to close a struct file
706 * @inode: the inode the struct file belongs to
707 * @file: the struct file being closed
712 static int gfs2_release(struct inode *inode, struct file *file)
714 struct gfs2_inode *ip = GFS2_I(inode);
716 kfree(file->private_data);
717 file->private_data = NULL;
719 if (file->f_mode & FMODE_WRITE) {
720 gfs2_rs_delete(ip, &inode->i_writecount);
727 * gfs2_fsync - sync the dirty data for a file (across the cluster)
728 * @file: the file that points to the dentry
729 * @start: the start position in the file to sync
730 * @end: the end position in the file to sync
731 * @datasync: set if we can ignore timestamp changes
733 * We split the data flushing here so that we don't wait for the data
734 * until after we've also sent the metadata to disk. Note that for
735 * data=ordered, we will write & wait for the data at the log flush
736 * stage anyway, so this is unlikely to make much of a difference
737 * except in the data=writeback case.
739 * If the fdatawrite fails due to any reason except -EIO, we will
740 * continue the remainder of the fsync, although we'll still report
741 * the error at the end. This is to match filemap_write_and_wait_range()
747 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
750 struct address_space *mapping = file->f_mapping;
751 struct inode *inode = mapping->host;
752 int sync_state = inode->i_state & I_DIRTY_ALL;
753 struct gfs2_inode *ip = GFS2_I(inode);
754 int ret = 0, ret1 = 0;
756 if (mapping->nrpages) {
757 ret1 = filemap_fdatawrite_range(mapping, start, end);
762 if (!gfs2_is_jdata(ip))
763 sync_state &= ~I_DIRTY_PAGES;
765 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
768 ret = sync_inode_metadata(inode, 1);
771 if (gfs2_is_jdata(ip))
772 ret = file_write_and_wait(file);
775 gfs2_ail_flush(ip->i_gl, 1);
778 if (mapping->nrpages)
779 ret = file_fdatawait_range(file, start, end);
781 return ret ? ret : ret1;
784 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
785 struct gfs2_holder *gh)
787 struct file *file = iocb->ki_filp;
788 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
789 size_t count = iov_iter_count(to);
793 return 0; /* skip atime */
795 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
796 ret = gfs2_glock_nq(gh);
800 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
801 is_sync_kiocb(iocb));
805 gfs2_holder_uninit(gh);
809 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
810 struct gfs2_holder *gh)
812 struct file *file = iocb->ki_filp;
813 struct inode *inode = file->f_mapping->host;
814 struct gfs2_inode *ip = GFS2_I(inode);
815 size_t len = iov_iter_count(from);
816 loff_t offset = iocb->ki_pos;
820 * Deferred lock, even if its a write, since we do no allocation on
821 * this path. All we need to change is the atime, and this lock mode
822 * ensures that other nodes have flushed their buffered read caches
823 * (i.e. their page cache entries for this inode). We do not,
824 * unfortunately, have the option of only flushing a range like the
827 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
828 ret = gfs2_glock_nq(gh);
832 /* Silently fall back to buffered I/O when writing beyond EOF */
833 if (offset + len > i_size_read(&ip->i_inode))
836 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
837 is_sync_kiocb(iocb));
843 gfs2_holder_uninit(gh);
847 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
849 struct gfs2_inode *ip;
850 struct gfs2_holder gh;
854 if (iocb->ki_flags & IOCB_DIRECT) {
855 ret = gfs2_file_direct_read(iocb, to, &gh);
856 if (likely(ret != -ENOTBLK))
858 iocb->ki_flags &= ~IOCB_DIRECT;
860 iocb->ki_flags |= IOCB_NOIO;
861 ret = generic_file_read_iter(iocb, to);
862 iocb->ki_flags &= ~IOCB_NOIO;
864 if (!iov_iter_count(to))
870 if (iocb->ki_flags & IOCB_NOWAIT)
873 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
874 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
875 ret = gfs2_glock_nq(&gh);
878 ret = generic_file_read_iter(iocb, to);
883 gfs2_holder_uninit(&gh);
884 return written ? written : ret;
888 * gfs2_file_write_iter - Perform a write to a file
889 * @iocb: The io context
890 * @from: The data to write
892 * We have to do a lock/unlock here to refresh the inode size for
893 * O_APPEND writes, otherwise we can land up writing at the wrong
894 * offset. There is still a race, but provided the app is using its
895 * own file locking, this will make O_APPEND work as expected.
899 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
901 struct file *file = iocb->ki_filp;
902 struct inode *inode = file_inode(file);
903 struct gfs2_inode *ip = GFS2_I(inode);
904 struct gfs2_holder gh;
907 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
909 if (iocb->ki_flags & IOCB_APPEND) {
910 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
913 gfs2_glock_dq_uninit(&gh);
917 ret = generic_write_checks(iocb, from);
921 ret = file_remove_privs(file);
925 ret = file_update_time(file);
929 if (iocb->ki_flags & IOCB_DIRECT) {
930 struct address_space *mapping = file->f_mapping;
931 ssize_t buffered, ret2;
933 ret = gfs2_file_direct_write(iocb, from, &gh);
934 if (ret < 0 || !iov_iter_count(from))
937 iocb->ki_flags |= IOCB_DSYNC;
938 current->backing_dev_info = inode_to_bdi(inode);
939 buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
940 current->backing_dev_info = NULL;
941 if (unlikely(buffered <= 0))
945 * We need to ensure that the page cache pages are written to
946 * disk and invalidated to preserve the expected O_DIRECT
947 * semantics. If the writeback or invalidate fails, only report
948 * the direct I/O range as we don't know if the buffered pages
951 iocb->ki_pos += buffered;
952 ret2 = generic_write_sync(iocb, buffered);
953 invalidate_mapping_pages(mapping,
954 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
955 (iocb->ki_pos - 1) >> PAGE_SHIFT);
956 if (!ret || ret2 > 0)
959 current->backing_dev_info = inode_to_bdi(inode);
960 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
961 current->backing_dev_info = NULL;
962 if (likely(ret > 0)) {
964 ret = generic_write_sync(iocb, ret);
973 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
976 struct super_block *sb = inode->i_sb;
977 struct gfs2_inode *ip = GFS2_I(inode);
978 loff_t end = offset + len;
979 struct buffer_head *dibh;
982 error = gfs2_meta_inode_buffer(ip, &dibh);
986 gfs2_trans_add_meta(ip->i_gl, dibh);
988 if (gfs2_is_stuffed(ip)) {
989 error = gfs2_unstuff_dinode(ip, NULL);
994 while (offset < end) {
995 struct iomap iomap = { };
997 error = gfs2_iomap_get_alloc(inode, offset, end - offset,
1001 offset = iomap.offset + iomap.length;
1002 if (!(iomap.flags & IOMAP_F_NEW))
1004 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1005 iomap.length >> inode->i_blkbits,
1008 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1018 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1019 * blocks, determine how many bytes can be written.
1020 * @ip: The inode in question.
1021 * @len: Max cap of bytes. What we return in *len must be <= this.
1022 * @data_blocks: Compute and return the number of data blocks needed
1023 * @ind_blocks: Compute and return the number of indirect blocks needed
1024 * @max_blocks: The total blocks available to work with.
1026 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1028 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1029 unsigned int *data_blocks, unsigned int *ind_blocks,
1030 unsigned int max_blocks)
1033 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1034 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1036 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1037 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1041 *data_blocks = max_data;
1042 *ind_blocks = max_blocks - max_data;
1043 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1046 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1050 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1052 struct inode *inode = file_inode(file);
1053 struct gfs2_sbd *sdp = GFS2_SB(inode);
1054 struct gfs2_inode *ip = GFS2_I(inode);
1055 struct gfs2_alloc_parms ap = { .aflags = 0, };
1056 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1057 loff_t bytes, max_bytes, max_blks;
1059 const loff_t pos = offset;
1060 const loff_t count = len;
1061 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1062 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1063 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1065 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1067 offset &= bsize_mask;
1069 len = next - offset;
1070 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1073 bytes &= bsize_mask;
1075 bytes = sdp->sd_sb.sb_bsize;
1077 gfs2_size_hint(file, offset, len);
1079 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1080 ap.min_target = data_blocks + ind_blocks;
1085 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1091 /* We need to determine how many bytes we can actually
1092 * fallocate without exceeding quota or going over the
1093 * end of the fs. We start off optimistically by assuming
1094 * we can write max_bytes */
1095 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1097 /* Since max_bytes is most likely a theoretical max, we
1098 * calculate a more realistic 'bytes' to serve as a good
1099 * starting point for the number of bytes we may be able
1101 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1102 ap.target = data_blocks + ind_blocks;
1104 error = gfs2_quota_lock_check(ip, &ap);
1107 /* ap.allowed tells us how many blocks quota will allow
1108 * us to write. Check if this reduces max_blks */
1109 max_blks = UINT_MAX;
1111 max_blks = ap.allowed;
1113 error = gfs2_inplace_reserve(ip, &ap);
1117 /* check if the selected rgrp limits our max_blks further */
1118 if (ap.allowed && ap.allowed < max_blks)
1119 max_blks = ap.allowed;
1121 /* Almost done. Calculate bytes that can be written using
1122 * max_blks. We also recompute max_bytes, data_blocks and
1124 calc_max_reserv(ip, &max_bytes, &data_blocks,
1125 &ind_blocks, max_blks);
1127 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1128 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1129 if (gfs2_is_jdata(ip))
1130 rblocks += data_blocks ? data_blocks : 1;
1132 error = gfs2_trans_begin(sdp, rblocks,
1133 PAGE_SIZE >> inode->i_blkbits);
1135 goto out_trans_fail;
1137 error = fallocate_chunk(inode, offset, max_bytes, mode);
1138 gfs2_trans_end(sdp);
1141 goto out_trans_fail;
1144 offset += max_bytes;
1145 gfs2_inplace_release(ip);
1146 gfs2_quota_unlock(ip);
1149 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1150 i_size_write(inode, pos + count);
1151 file_update_time(file);
1152 mark_inode_dirty(inode);
1154 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1155 return vfs_fsync_range(file, pos, pos + count - 1,
1156 (file->f_flags & __O_SYNC) ? 0 : 1);
1160 gfs2_inplace_release(ip);
1162 gfs2_quota_unlock(ip);
1166 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1168 struct inode *inode = file_inode(file);
1169 struct gfs2_sbd *sdp = GFS2_SB(inode);
1170 struct gfs2_inode *ip = GFS2_I(inode);
1171 struct gfs2_holder gh;
1174 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1176 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1177 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1182 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1183 ret = gfs2_glock_nq(&gh);
1187 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1188 (offset + len) > inode->i_size) {
1189 ret = inode_newsize_ok(inode, offset + len);
1194 ret = get_write_access(inode);
1198 if (mode & FALLOC_FL_PUNCH_HOLE) {
1199 ret = __gfs2_punch_hole(file, offset, len);
1201 ret = __gfs2_fallocate(file, mode, offset, len);
1203 gfs2_rs_deltree(&ip->i_res);
1206 put_write_access(inode);
1210 gfs2_holder_uninit(&gh);
1211 inode_unlock(inode);
1215 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1216 struct file *out, loff_t *ppos,
1217 size_t len, unsigned int flags)
1221 gfs2_size_hint(out, *ppos, len);
1223 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1227 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1230 * gfs2_lock - acquire/release a posix lock on a file
1231 * @file: the file pointer
1232 * @cmd: either modify or retrieve lock state, possibly wait
1233 * @fl: type and range of lock
1238 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1240 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1241 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1242 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1244 if (!(fl->fl_flags & FL_POSIX))
1246 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1249 if (cmd == F_CANCELLK) {
1252 fl->fl_type = F_UNLCK;
1254 if (unlikely(gfs2_withdrawn(sdp))) {
1255 if (fl->fl_type == F_UNLCK)
1256 locks_lock_file_wait(file, fl);
1260 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1261 else if (fl->fl_type == F_UNLCK)
1262 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1264 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1267 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1269 struct gfs2_file *fp = file->private_data;
1270 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1271 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1272 struct gfs2_glock *gl;
1278 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1279 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1281 mutex_lock(&fp->f_fl_mutex);
1283 if (gfs2_holder_initialized(fl_gh)) {
1284 struct file_lock request;
1285 if (fl_gh->gh_state == state)
1287 locks_init_lock(&request);
1288 request.fl_type = F_UNLCK;
1289 request.fl_flags = FL_FLOCK;
1290 locks_lock_file_wait(file, &request);
1291 gfs2_glock_dq(fl_gh);
1292 gfs2_holder_reinit(state, flags, fl_gh);
1294 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1295 &gfs2_flock_glops, CREATE, &gl);
1298 gfs2_holder_init(gl, state, flags, fl_gh);
1301 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1302 error = gfs2_glock_nq(fl_gh);
1303 if (error != GLR_TRYFAILED)
1305 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1306 fl_gh->gh_error = 0;
1310 gfs2_holder_uninit(fl_gh);
1311 if (error == GLR_TRYFAILED)
1314 error = locks_lock_file_wait(file, fl);
1315 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1319 mutex_unlock(&fp->f_fl_mutex);
1323 static void do_unflock(struct file *file, struct file_lock *fl)
1325 struct gfs2_file *fp = file->private_data;
1326 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1328 mutex_lock(&fp->f_fl_mutex);
1329 locks_lock_file_wait(file, fl);
1330 if (gfs2_holder_initialized(fl_gh)) {
1331 gfs2_glock_dq(fl_gh);
1332 gfs2_holder_uninit(fl_gh);
1334 mutex_unlock(&fp->f_fl_mutex);
1338 * gfs2_flock - acquire/release a flock lock on a file
1339 * @file: the file pointer
1340 * @cmd: either modify or retrieve lock state, possibly wait
1341 * @fl: type and range of lock
1346 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1348 if (!(fl->fl_flags & FL_FLOCK))
1350 if (fl->fl_type & LOCK_MAND)
1353 if (fl->fl_type == F_UNLCK) {
1354 do_unflock(file, fl);
1357 return do_flock(file, cmd, fl);
1361 const struct file_operations gfs2_file_fops = {
1362 .llseek = gfs2_llseek,
1363 .read_iter = gfs2_file_read_iter,
1364 .write_iter = gfs2_file_write_iter,
1365 .iopoll = iomap_dio_iopoll,
1366 .unlocked_ioctl = gfs2_ioctl,
1367 .compat_ioctl = gfs2_compat_ioctl,
1370 .release = gfs2_release,
1371 .fsync = gfs2_fsync,
1373 .flock = gfs2_flock,
1374 .splice_read = generic_file_splice_read,
1375 .splice_write = gfs2_file_splice_write,
1376 .setlease = simple_nosetlease,
1377 .fallocate = gfs2_fallocate,
1380 const struct file_operations gfs2_dir_fops = {
1381 .iterate_shared = gfs2_readdir,
1382 .unlocked_ioctl = gfs2_ioctl,
1383 .compat_ioctl = gfs2_compat_ioctl,
1385 .release = gfs2_release,
1386 .fsync = gfs2_fsync,
1388 .flock = gfs2_flock,
1389 .llseek = default_llseek,
1392 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1394 const struct file_operations gfs2_file_fops_nolock = {
1395 .llseek = gfs2_llseek,
1396 .read_iter = gfs2_file_read_iter,
1397 .write_iter = gfs2_file_write_iter,
1398 .iopoll = iomap_dio_iopoll,
1399 .unlocked_ioctl = gfs2_ioctl,
1400 .compat_ioctl = gfs2_compat_ioctl,
1403 .release = gfs2_release,
1404 .fsync = gfs2_fsync,
1405 .splice_read = generic_file_splice_read,
1406 .splice_write = gfs2_file_splice_write,
1407 .setlease = generic_setlease,
1408 .fallocate = gfs2_fallocate,
1411 const struct file_operations gfs2_dir_fops_nolock = {
1412 .iterate_shared = gfs2_readdir,
1413 .unlocked_ioctl = gfs2_ioctl,
1414 .compat_ioctl = gfs2_compat_ioctl,
1416 .release = gfs2_release,
1417 .fsync = gfs2_fsync,
1418 .llseek = default_llseek,