2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/pagemap.h>
15 #include <linux/uio.h>
16 #include <linux/blkdev.h>
18 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/falloc.h>
22 #include <linux/swap.h>
23 #include <linux/crc32.h>
24 #include <linux/writeback.h>
25 #include <linux/uaccess.h>
26 #include <linux/dlm.h>
27 #include <linux/dlm_plock.h>
28 #include <linux/delay.h>
29 #include <linux/backing-dev.h>
47 * gfs2_llseek - seek to a location in a file
50 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
52 * SEEK_END requires the glock for the file because it references the
55 * Returns: The new offset, or errno
58 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
60 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
61 struct gfs2_holder i_gh;
66 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
69 error = generic_file_llseek(file, offset, whence);
70 gfs2_glock_dq_uninit(&i_gh);
75 error = gfs2_seek_data(file, offset);
79 error = gfs2_seek_hole(file, offset);
85 * These don't reference inode->i_size and don't depend on the
86 * block mapping, so we don't need the glock.
88 error = generic_file_llseek(file, offset, whence);
98 * gfs2_readdir - Iterator for a directory
99 * @file: The directory to read from
100 * @ctx: What to feed directory entries to
105 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
107 struct inode *dir = file->f_mapping->host;
108 struct gfs2_inode *dip = GFS2_I(dir);
109 struct gfs2_holder d_gh;
112 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
116 error = gfs2_dir_read(dir, ctx, &file->f_ra);
118 gfs2_glock_dq_uninit(&d_gh);
126 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
127 * and to GFS2_DIF_JDATA for non-directories.
132 } fsflag_gfs2flag[] = {
133 {FS_SYNC_FL, GFS2_DIF_SYNC},
134 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
135 {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
136 {FS_NOATIME_FL, GFS2_DIF_NOATIME},
137 {FS_INDEX_FL, GFS2_DIF_EXHASH},
138 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
139 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
142 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
144 struct inode *inode = file_inode(filp);
145 struct gfs2_inode *ip = GFS2_I(inode);
146 struct gfs2_holder gh;
148 u32 gfsflags, fsflags = 0;
150 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
151 error = gfs2_glock_nq(&gh);
155 gfsflags = ip->i_diskflags;
156 if (S_ISDIR(inode->i_mode))
157 gfsflags &= ~GFS2_DIF_JDATA;
159 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
160 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
161 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
162 fsflags |= fsflag_gfs2flag[i].fsflag;
164 if (put_user(fsflags, ptr))
169 gfs2_holder_uninit(&gh);
173 void gfs2_set_inode_flags(struct inode *inode)
175 struct gfs2_inode *ip = GFS2_I(inode);
176 unsigned int flags = inode->i_flags;
178 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
179 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
181 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
182 flags |= S_IMMUTABLE;
183 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
185 if (ip->i_diskflags & GFS2_DIF_NOATIME)
187 if (ip->i_diskflags & GFS2_DIF_SYNC)
189 inode->i_flags = flags;
192 /* Flags that can be set by user space */
193 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
194 GFS2_DIF_IMMUTABLE| \
195 GFS2_DIF_APPENDONLY| \
199 GFS2_DIF_INHERIT_JDATA)
202 * do_gfs2_set_flags - set flags on an inode
203 * @filp: file pointer
204 * @reqflags: The flags to set
205 * @mask: Indicates which flags are valid
208 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
210 struct inode *inode = file_inode(filp);
211 struct gfs2_inode *ip = GFS2_I(inode);
212 struct gfs2_sbd *sdp = GFS2_SB(inode);
213 struct buffer_head *bh;
214 struct gfs2_holder gh;
216 u32 new_flags, flags;
218 error = mnt_want_write_file(filp);
222 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
227 if (!inode_owner_or_capable(inode))
231 flags = ip->i_diskflags;
232 new_flags = (flags & ~mask) | (reqflags & mask);
233 if ((new_flags ^ flags) == 0)
237 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
239 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
241 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
242 !capable(CAP_LINUX_IMMUTABLE))
244 if (!IS_IMMUTABLE(inode)) {
245 error = gfs2_permission(inode, MAY_WRITE);
249 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
250 if (new_flags & GFS2_DIF_JDATA)
251 gfs2_log_flush(sdp, ip->i_gl,
252 GFS2_LOG_HEAD_FLUSH_NORMAL |
254 error = filemap_fdatawrite(inode->i_mapping);
257 error = filemap_fdatawait(inode->i_mapping);
260 if (new_flags & GFS2_DIF_JDATA)
261 gfs2_ordered_del_inode(ip);
263 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
266 error = gfs2_meta_inode_buffer(ip, &bh);
269 inode->i_ctime = current_time(inode);
270 gfs2_trans_add_meta(ip->i_gl, bh);
271 ip->i_diskflags = new_flags;
272 gfs2_dinode_out(ip, bh->b_data);
274 gfs2_set_inode_flags(inode);
275 gfs2_set_aops(inode);
279 gfs2_glock_dq_uninit(&gh);
281 mnt_drop_write_file(filp);
285 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
287 struct inode *inode = file_inode(filp);
288 u32 fsflags, gfsflags = 0;
292 if (get_user(fsflags, ptr))
295 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
296 if (fsflags & fsflag_gfs2flag[i].fsflag) {
297 fsflags &= ~fsflag_gfs2flag[i].fsflag;
298 gfsflags |= fsflag_gfs2flag[i].gfsflag;
301 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
304 mask = GFS2_FLAGS_USER_SET;
305 if (S_ISDIR(inode->i_mode)) {
306 mask &= ~GFS2_DIF_JDATA;
308 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
309 if (gfsflags & GFS2_DIF_TOPDIR)
311 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
314 return do_gfs2_set_flags(filp, gfsflags, mask);
317 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
320 case FS_IOC_GETFLAGS:
321 return gfs2_get_flags(filp, (u32 __user *)arg);
322 case FS_IOC_SETFLAGS:
323 return gfs2_set_flags(filp, (u32 __user *)arg);
325 return gfs2_fitrim(filp, (void __user *)arg);
331 * gfs2_size_hint - Give a hint to the size of a write request
332 * @filep: The struct file
333 * @offset: The file offset of the write
334 * @size: The length of the write
336 * When we are about to do a write, this function records the total
337 * write size in order to provide a suitable hint to the lower layers
338 * about how many blocks will be required.
342 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
344 struct inode *inode = file_inode(filep);
345 struct gfs2_sbd *sdp = GFS2_SB(inode);
346 struct gfs2_inode *ip = GFS2_I(inode);
347 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
348 int hint = min_t(size_t, INT_MAX, blks);
350 if (hint > atomic_read(&ip->i_res.rs_sizehint))
351 atomic_set(&ip->i_res.rs_sizehint, hint);
355 * gfs2_allocate_page_backing - Use bmap to allocate blocks
356 * @page: The (locked) page to allocate backing for
358 * We try to allocate all the blocks required for the page in
359 * one go. This might fail for various reasons, so we keep
360 * trying until all the blocks to back this page are allocated.
361 * If some of the blocks are already allocated, thats ok too.
364 static int gfs2_allocate_page_backing(struct page *page)
366 struct inode *inode = page->mapping->host;
367 struct buffer_head bh;
368 unsigned long size = PAGE_SIZE;
369 u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits);
374 gfs2_block_map(inode, lblock, &bh, 1);
375 if (!buffer_mapped(&bh))
378 lblock += (bh.b_size >> inode->i_blkbits);
384 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
385 * @vma: The virtual memory area
386 * @vmf: The virtual memory fault containing the page to become writable
388 * When the page becomes writable, we need to ensure that we have
389 * blocks allocated on disk to back that page.
392 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
394 struct page *page = vmf->page;
395 struct inode *inode = file_inode(vmf->vma->vm_file);
396 struct gfs2_inode *ip = GFS2_I(inode);
397 struct gfs2_sbd *sdp = GFS2_SB(inode);
398 struct gfs2_alloc_parms ap = { .aflags = 0, };
399 unsigned long last_index;
400 u64 pos = page->index << PAGE_SHIFT;
401 unsigned int data_blocks, ind_blocks, rblocks;
402 struct gfs2_holder gh;
406 sb_start_pagefault(inode->i_sb);
408 ret = gfs2_rsqa_alloc(ip);
412 gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE);
414 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
415 ret = gfs2_glock_nq(&gh);
419 /* Update file times before taking page lock */
420 file_update_time(vmf->vma->vm_file);
422 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
423 set_bit(GIF_SW_PAGED, &ip->i_flags);
425 if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) {
427 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
434 ret = gfs2_rindex_update(sdp);
438 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
439 ap.target = data_blocks + ind_blocks;
440 ret = gfs2_quota_lock_check(ip, &ap);
443 ret = gfs2_inplace_reserve(ip, &ap);
445 goto out_quota_unlock;
447 rblocks = RES_DINODE + ind_blocks;
448 if (gfs2_is_jdata(ip))
449 rblocks += data_blocks ? data_blocks : 1;
450 if (ind_blocks || data_blocks) {
451 rblocks += RES_STATFS + RES_QUOTA;
452 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
454 ret = gfs2_trans_begin(sdp, rblocks, 0);
460 size = i_size_read(inode);
461 last_index = (size - 1) >> PAGE_SHIFT;
462 /* Check page index against inode size */
463 if (size == 0 || (page->index > last_index))
467 /* If truncated, we must retry the operation, we may have raced
468 * with the glock demotion code.
470 if (!PageUptodate(page) || page->mapping != inode->i_mapping)
473 /* Unstuff, if required, and allocate backing blocks for page */
475 if (gfs2_is_stuffed(ip))
476 ret = gfs2_unstuff_dinode(ip, page);
478 ret = gfs2_allocate_page_backing(page);
485 gfs2_inplace_release(ip);
487 gfs2_quota_unlock(ip);
491 gfs2_holder_uninit(&gh);
493 set_page_dirty(page);
494 wait_for_stable_page(page);
497 sb_end_pagefault(inode->i_sb);
498 return block_page_mkwrite_return(ret);
501 static const struct vm_operations_struct gfs2_vm_ops = {
502 .fault = filemap_fault,
503 .map_pages = filemap_map_pages,
504 .page_mkwrite = gfs2_page_mkwrite,
509 * @file: The file to map
510 * @vma: The VMA which described the mapping
512 * There is no need to get a lock here unless we should be updating
513 * atime. We ignore any locking errors since the only consequence is
514 * a missed atime update (which will just be deferred until later).
519 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
521 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
523 if (!(file->f_flags & O_NOATIME) &&
524 !IS_NOATIME(&ip->i_inode)) {
525 struct gfs2_holder i_gh;
528 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
532 /* grab lock to update inode */
533 gfs2_glock_dq_uninit(&i_gh);
536 vma->vm_ops = &gfs2_vm_ops;
542 * gfs2_open_common - This is common to open and atomic_open
543 * @inode: The inode being opened
544 * @file: The file being opened
546 * This maybe called under a glock or not depending upon how it has
547 * been called. We must always be called under a glock for regular
548 * files, however. For other file types, it does not matter whether
549 * we hold the glock or not.
551 * Returns: Error code or 0 for success
554 int gfs2_open_common(struct inode *inode, struct file *file)
556 struct gfs2_file *fp;
559 if (S_ISREG(inode->i_mode)) {
560 ret = generic_file_open(inode, file);
565 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
569 mutex_init(&fp->f_fl_mutex);
571 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
572 file->private_data = fp;
577 * gfs2_open - open a file
578 * @inode: the inode to open
579 * @file: the struct file for this opening
581 * After atomic_open, this function is only used for opening files
582 * which are already cached. We must still get the glock for regular
583 * files to ensure that we have the file size uptodate for the large
584 * file check which is in the common code. That is only an issue for
585 * regular files though.
590 static int gfs2_open(struct inode *inode, struct file *file)
592 struct gfs2_inode *ip = GFS2_I(inode);
593 struct gfs2_holder i_gh;
595 bool need_unlock = false;
597 if (S_ISREG(ip->i_inode.i_mode)) {
598 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
605 error = gfs2_open_common(inode, file);
608 gfs2_glock_dq_uninit(&i_gh);
614 * gfs2_release - called to close a struct file
615 * @inode: the inode the struct file belongs to
616 * @file: the struct file being closed
621 static int gfs2_release(struct inode *inode, struct file *file)
623 struct gfs2_inode *ip = GFS2_I(inode);
625 kfree(file->private_data);
626 file->private_data = NULL;
628 if (!(file->f_mode & FMODE_WRITE))
631 gfs2_rsqa_delete(ip, &inode->i_writecount);
636 * gfs2_fsync - sync the dirty data for a file (across the cluster)
637 * @file: the file that points to the dentry
638 * @start: the start position in the file to sync
639 * @end: the end position in the file to sync
640 * @datasync: set if we can ignore timestamp changes
642 * We split the data flushing here so that we don't wait for the data
643 * until after we've also sent the metadata to disk. Note that for
644 * data=ordered, we will write & wait for the data at the log flush
645 * stage anyway, so this is unlikely to make much of a difference
646 * except in the data=writeback case.
648 * If the fdatawrite fails due to any reason except -EIO, we will
649 * continue the remainder of the fsync, although we'll still report
650 * the error at the end. This is to match filemap_write_and_wait_range()
656 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
659 struct address_space *mapping = file->f_mapping;
660 struct inode *inode = mapping->host;
661 int sync_state = inode->i_state & I_DIRTY_ALL;
662 struct gfs2_inode *ip = GFS2_I(inode);
663 int ret = 0, ret1 = 0;
665 if (mapping->nrpages) {
666 ret1 = filemap_fdatawrite_range(mapping, start, end);
671 if (!gfs2_is_jdata(ip))
672 sync_state &= ~I_DIRTY_PAGES;
674 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
677 ret = sync_inode_metadata(inode, 1);
680 if (gfs2_is_jdata(ip))
681 ret = file_write_and_wait(file);
684 gfs2_ail_flush(ip->i_gl, 1);
687 if (mapping->nrpages)
688 ret = file_fdatawait_range(file, start, end);
690 return ret ? ret : ret1;
693 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to)
695 struct file *file = iocb->ki_filp;
696 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
697 size_t count = iov_iter_count(to);
698 struct gfs2_holder gh;
702 return 0; /* skip atime */
704 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
705 ret = gfs2_glock_nq(&gh);
709 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL);
713 gfs2_holder_uninit(&gh);
717 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
719 struct file *file = iocb->ki_filp;
720 struct inode *inode = file->f_mapping->host;
721 struct gfs2_inode *ip = GFS2_I(inode);
722 size_t len = iov_iter_count(from);
723 loff_t offset = iocb->ki_pos;
724 struct gfs2_holder gh;
728 * Deferred lock, even if its a write, since we do no allocation on
729 * this path. All we need to change is the atime, and this lock mode
730 * ensures that other nodes have flushed their buffered read caches
731 * (i.e. their page cache entries for this inode). We do not,
732 * unfortunately, have the option of only flushing a range like the
735 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
736 ret = gfs2_glock_nq(&gh);
740 /* Silently fall back to buffered I/O when writing beyond EOF */
741 if (offset + len > i_size_read(&ip->i_inode))
744 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL);
749 gfs2_holder_uninit(&gh);
753 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
757 if (iocb->ki_flags & IOCB_DIRECT) {
758 ret = gfs2_file_direct_read(iocb, to);
759 if (likely(ret != -ENOTBLK))
761 iocb->ki_flags &= ~IOCB_DIRECT;
763 return generic_file_read_iter(iocb, to);
767 * gfs2_file_write_iter - Perform a write to a file
768 * @iocb: The io context
769 * @from: The data to write
771 * We have to do a lock/unlock here to refresh the inode size for
772 * O_APPEND writes, otherwise we can land up writing at the wrong
773 * offset. There is still a race, but provided the app is using its
774 * own file locking, this will make O_APPEND work as expected.
778 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
780 struct file *file = iocb->ki_filp;
781 struct inode *inode = file_inode(file);
782 struct gfs2_inode *ip = GFS2_I(inode);
783 ssize_t written = 0, ret;
785 ret = gfs2_rsqa_alloc(ip);
789 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
791 if (iocb->ki_flags & IOCB_APPEND) {
792 struct gfs2_holder gh;
794 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
797 gfs2_glock_dq_uninit(&gh);
801 ret = generic_write_checks(iocb, from);
805 /* We can write back this queue in page reclaim */
806 current->backing_dev_info = inode_to_bdi(inode);
808 ret = file_remove_privs(file);
812 ret = file_update_time(file);
816 if (iocb->ki_flags & IOCB_DIRECT) {
817 struct address_space *mapping = file->f_mapping;
821 written = gfs2_file_direct_write(iocb, from);
822 if (written < 0 || !iov_iter_count(from))
825 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
826 if (unlikely(ret < 0))
831 * We need to ensure that the page cache pages are written to
832 * disk and invalidated to preserve the expected O_DIRECT
836 endbyte = pos + buffered - 1;
837 ret = filemap_write_and_wait_range(mapping, pos, endbyte);
839 iocb->ki_pos += buffered;
841 invalidate_mapping_pages(mapping,
843 endbyte >> PAGE_SHIFT);
846 * We don't know how much we wrote, so just return
847 * the number of bytes which were direct-written
851 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
857 current->backing_dev_info = NULL;
860 if (likely(ret > 0)) {
861 /* Handle various SYNC-type writes */
862 ret = generic_write_sync(iocb, ret);
864 return written ? written : ret;
867 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
870 struct super_block *sb = inode->i_sb;
871 struct gfs2_inode *ip = GFS2_I(inode);
872 loff_t end = offset + len;
873 struct buffer_head *dibh;
876 error = gfs2_meta_inode_buffer(ip, &dibh);
880 gfs2_trans_add_meta(ip->i_gl, dibh);
882 if (gfs2_is_stuffed(ip)) {
883 error = gfs2_unstuff_dinode(ip, NULL);
888 while (offset < end) {
889 struct iomap iomap = { };
891 error = gfs2_iomap_get_alloc(inode, offset, end - offset,
895 offset = iomap.offset + iomap.length;
896 if (!(iomap.flags & IOMAP_F_NEW))
898 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
899 iomap.length >> inode->i_blkbits,
902 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
911 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
912 * blocks, determine how many bytes can be written.
913 * @ip: The inode in question.
914 * @len: Max cap of bytes. What we return in *len must be <= this.
915 * @data_blocks: Compute and return the number of data blocks needed
916 * @ind_blocks: Compute and return the number of indirect blocks needed
917 * @max_blocks: The total blocks available to work with.
919 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
921 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
922 unsigned int *data_blocks, unsigned int *ind_blocks,
923 unsigned int max_blocks)
926 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
927 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
929 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
930 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
934 *data_blocks = max_data;
935 *ind_blocks = max_blocks - max_data;
936 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
939 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
943 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
945 struct inode *inode = file_inode(file);
946 struct gfs2_sbd *sdp = GFS2_SB(inode);
947 struct gfs2_inode *ip = GFS2_I(inode);
948 struct gfs2_alloc_parms ap = { .aflags = 0, };
949 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
950 loff_t bytes, max_bytes, max_blks;
952 const loff_t pos = offset;
953 const loff_t count = len;
954 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
955 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
956 loff_t max_chunk_size = UINT_MAX & bsize_mask;
958 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
960 offset &= bsize_mask;
963 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
968 bytes = sdp->sd_sb.sb_bsize;
970 gfs2_size_hint(file, offset, len);
972 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
973 ap.min_target = data_blocks + ind_blocks;
978 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
984 /* We need to determine how many bytes we can actually
985 * fallocate without exceeding quota or going over the
986 * end of the fs. We start off optimistically by assuming
987 * we can write max_bytes */
988 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
990 /* Since max_bytes is most likely a theoretical max, we
991 * calculate a more realistic 'bytes' to serve as a good
992 * starting point for the number of bytes we may be able
994 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
995 ap.target = data_blocks + ind_blocks;
997 error = gfs2_quota_lock_check(ip, &ap);
1000 /* ap.allowed tells us how many blocks quota will allow
1001 * us to write. Check if this reduces max_blks */
1002 max_blks = UINT_MAX;
1004 max_blks = ap.allowed;
1006 error = gfs2_inplace_reserve(ip, &ap);
1010 /* check if the selected rgrp limits our max_blks further */
1011 if (ap.allowed && ap.allowed < max_blks)
1012 max_blks = ap.allowed;
1014 /* Almost done. Calculate bytes that can be written using
1015 * max_blks. We also recompute max_bytes, data_blocks and
1017 calc_max_reserv(ip, &max_bytes, &data_blocks,
1018 &ind_blocks, max_blks);
1020 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1021 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1022 if (gfs2_is_jdata(ip))
1023 rblocks += data_blocks ? data_blocks : 1;
1025 error = gfs2_trans_begin(sdp, rblocks,
1026 PAGE_SIZE/sdp->sd_sb.sb_bsize);
1028 goto out_trans_fail;
1030 error = fallocate_chunk(inode, offset, max_bytes, mode);
1031 gfs2_trans_end(sdp);
1034 goto out_trans_fail;
1037 offset += max_bytes;
1038 gfs2_inplace_release(ip);
1039 gfs2_quota_unlock(ip);
1042 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) {
1043 i_size_write(inode, pos + count);
1044 file_update_time(file);
1045 mark_inode_dirty(inode);
1048 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1049 return vfs_fsync_range(file, pos, pos + count - 1,
1050 (file->f_flags & __O_SYNC) ? 0 : 1);
1054 gfs2_inplace_release(ip);
1056 gfs2_quota_unlock(ip);
1060 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1062 struct inode *inode = file_inode(file);
1063 struct gfs2_sbd *sdp = GFS2_SB(inode);
1064 struct gfs2_inode *ip = GFS2_I(inode);
1065 struct gfs2_holder gh;
1068 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1070 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1071 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1076 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1077 ret = gfs2_glock_nq(&gh);
1081 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1082 (offset + len) > inode->i_size) {
1083 ret = inode_newsize_ok(inode, offset + len);
1088 ret = get_write_access(inode);
1092 if (mode & FALLOC_FL_PUNCH_HOLE) {
1093 ret = __gfs2_punch_hole(file, offset, len);
1095 ret = gfs2_rsqa_alloc(ip);
1099 ret = __gfs2_fallocate(file, mode, offset, len);
1102 gfs2_rs_deltree(&ip->i_res);
1106 put_write_access(inode);
1110 gfs2_holder_uninit(&gh);
1111 inode_unlock(inode);
1115 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1116 struct file *out, loff_t *ppos,
1117 size_t len, unsigned int flags)
1120 struct gfs2_inode *ip = GFS2_I(out->f_mapping->host);
1122 error = gfs2_rsqa_alloc(ip);
1124 return (ssize_t)error;
1126 gfs2_size_hint(out, *ppos, len);
1128 return iter_file_splice_write(pipe, out, ppos, len, flags);
1131 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1134 * gfs2_lock - acquire/release a posix lock on a file
1135 * @file: the file pointer
1136 * @cmd: either modify or retrieve lock state, possibly wait
1137 * @fl: type and range of lock
1142 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1144 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1145 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1146 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1148 if (!(fl->fl_flags & FL_POSIX))
1150 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1153 if (cmd == F_CANCELLK) {
1156 fl->fl_type = F_UNLCK;
1158 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1159 if (fl->fl_type == F_UNLCK)
1160 locks_lock_file_wait(file, fl);
1164 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1165 else if (fl->fl_type == F_UNLCK)
1166 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1168 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1171 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1173 struct gfs2_file *fp = file->private_data;
1174 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1175 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1176 struct gfs2_glock *gl;
1182 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1183 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1185 mutex_lock(&fp->f_fl_mutex);
1187 if (gfs2_holder_initialized(fl_gh)) {
1188 if (fl_gh->gh_state == state)
1190 locks_lock_file_wait(file,
1191 &(struct file_lock) {
1193 .fl_flags = FL_FLOCK
1195 gfs2_glock_dq(fl_gh);
1196 gfs2_holder_reinit(state, flags, fl_gh);
1198 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1199 &gfs2_flock_glops, CREATE, &gl);
1202 gfs2_holder_init(gl, state, flags, fl_gh);
1205 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1206 error = gfs2_glock_nq(fl_gh);
1207 if (error != GLR_TRYFAILED)
1209 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1210 fl_gh->gh_error = 0;
1214 gfs2_holder_uninit(fl_gh);
1215 if (error == GLR_TRYFAILED)
1218 error = locks_lock_file_wait(file, fl);
1219 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1223 mutex_unlock(&fp->f_fl_mutex);
1227 static void do_unflock(struct file *file, struct file_lock *fl)
1229 struct gfs2_file *fp = file->private_data;
1230 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1232 mutex_lock(&fp->f_fl_mutex);
1233 locks_lock_file_wait(file, fl);
1234 if (gfs2_holder_initialized(fl_gh)) {
1235 gfs2_glock_dq(fl_gh);
1236 gfs2_holder_uninit(fl_gh);
1238 mutex_unlock(&fp->f_fl_mutex);
1242 * gfs2_flock - acquire/release a flock lock on a file
1243 * @file: the file pointer
1244 * @cmd: either modify or retrieve lock state, possibly wait
1245 * @fl: type and range of lock
1250 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1252 if (!(fl->fl_flags & FL_FLOCK))
1254 if (fl->fl_type & LOCK_MAND)
1257 if (fl->fl_type == F_UNLCK) {
1258 do_unflock(file, fl);
1261 return do_flock(file, cmd, fl);
1265 const struct file_operations gfs2_file_fops = {
1266 .llseek = gfs2_llseek,
1267 .read_iter = gfs2_file_read_iter,
1268 .write_iter = gfs2_file_write_iter,
1269 .unlocked_ioctl = gfs2_ioctl,
1272 .release = gfs2_release,
1273 .fsync = gfs2_fsync,
1275 .flock = gfs2_flock,
1276 .splice_read = generic_file_splice_read,
1277 .splice_write = gfs2_file_splice_write,
1278 .setlease = simple_nosetlease,
1279 .fallocate = gfs2_fallocate,
1282 const struct file_operations gfs2_dir_fops = {
1283 .iterate_shared = gfs2_readdir,
1284 .unlocked_ioctl = gfs2_ioctl,
1286 .release = gfs2_release,
1287 .fsync = gfs2_fsync,
1289 .flock = gfs2_flock,
1290 .llseek = default_llseek,
1293 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1295 const struct file_operations gfs2_file_fops_nolock = {
1296 .llseek = gfs2_llseek,
1297 .read_iter = gfs2_file_read_iter,
1298 .write_iter = gfs2_file_write_iter,
1299 .unlocked_ioctl = gfs2_ioctl,
1302 .release = gfs2_release,
1303 .fsync = gfs2_fsync,
1304 .splice_read = generic_file_splice_read,
1305 .splice_write = gfs2_file_splice_write,
1306 .setlease = generic_setlease,
1307 .fallocate = gfs2_fallocate,
1310 const struct file_operations gfs2_dir_fops_nolock = {
1311 .iterate_shared = gfs2_readdir,
1312 .unlocked_ioctl = gfs2_ioctl,
1314 .release = gfs2_release,
1315 .fsync = gfs2_fsync,
1316 .llseek = default_llseek,