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/ext2_fs.h>
22 #include <linux/falloc.h>
23 #include <linux/swap.h>
24 #include <linux/crc32.h>
25 #include <linux/writeback.h>
26 #include <asm/uaccess.h>
27 #include <linux/dlm.h>
28 #include <linux/dlm_plock.h>
45 * gfs2_llseek - seek to a location in a file
48 * @origin: 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 origin)
58 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
59 struct gfs2_holder i_gh;
63 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
66 error = generic_file_llseek_unlocked(file, offset, origin);
67 gfs2_glock_dq_uninit(&i_gh);
70 error = generic_file_llseek_unlocked(file, offset, origin);
76 * gfs2_readdir - Read directory entries from a directory
77 * @file: The directory to read from
78 * @dirent: Buffer for dirents
79 * @filldir: Function used to do the copying
84 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
86 struct inode *dir = file->f_mapping->host;
87 struct gfs2_inode *dip = GFS2_I(dir);
88 struct gfs2_holder d_gh;
89 u64 offset = file->f_pos;
92 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
93 error = gfs2_glock_nq(&d_gh);
95 gfs2_holder_uninit(&d_gh);
99 error = gfs2_dir_read(dir, &offset, dirent, filldir);
101 gfs2_glock_dq_uninit(&d_gh);
103 file->f_pos = offset;
110 * @table: A table of 32 u32 flags
111 * @val: a 32 bit value to convert
113 * This function can be used to convert between fsflags values and
114 * GFS2's own flags values.
116 * Returns: the converted flags
118 static u32 fsflags_cvt(const u32 *table, u32 val)
130 static const u32 fsflags_to_gfs2[32] = {
132 [4] = GFS2_DIF_IMMUTABLE,
133 [5] = GFS2_DIF_APPENDONLY,
134 [7] = GFS2_DIF_NOATIME,
135 [12] = GFS2_DIF_EXHASH,
136 [14] = GFS2_DIF_INHERIT_JDATA,
139 static const u32 gfs2_to_fsflags[32] = {
140 [gfs2fl_Sync] = FS_SYNC_FL,
141 [gfs2fl_Immutable] = FS_IMMUTABLE_FL,
142 [gfs2fl_AppendOnly] = FS_APPEND_FL,
143 [gfs2fl_NoAtime] = FS_NOATIME_FL,
144 [gfs2fl_ExHash] = FS_INDEX_FL,
145 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
148 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
150 struct inode *inode = filp->f_path.dentry->d_inode;
151 struct gfs2_inode *ip = GFS2_I(inode);
152 struct gfs2_holder gh;
156 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
157 error = gfs2_glock_nq(&gh);
161 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
162 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
163 fsflags |= FS_JOURNAL_DATA_FL;
164 if (put_user(fsflags, ptr))
168 gfs2_holder_uninit(&gh);
172 void gfs2_set_inode_flags(struct inode *inode)
174 struct gfs2_inode *ip = GFS2_I(inode);
175 unsigned int flags = inode->i_flags;
177 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
178 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
179 inode->i_flags |= S_NOSEC;
180 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
181 flags |= S_IMMUTABLE;
182 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
184 if (ip->i_diskflags & GFS2_DIF_NOATIME)
186 if (ip->i_diskflags & GFS2_DIF_SYNC)
188 inode->i_flags = flags;
191 /* Flags that can be set by user space */
192 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
193 GFS2_DIF_IMMUTABLE| \
194 GFS2_DIF_APPENDONLY| \
198 GFS2_DIF_INHERIT_JDATA)
201 * gfs2_set_flags - set flags on an inode
203 * @flags: The flags to set
204 * @mask: Indicates which flags are valid
207 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
209 struct inode *inode = filp->f_path.dentry->d_inode;
210 struct gfs2_inode *ip = GFS2_I(inode);
211 struct gfs2_sbd *sdp = GFS2_SB(inode);
212 struct buffer_head *bh;
213 struct gfs2_holder gh;
215 u32 new_flags, flags;
217 error = mnt_want_write(filp->f_path.mnt);
221 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
226 if (!inode_owner_or_capable(inode))
230 flags = ip->i_diskflags;
231 new_flags = (flags & ~mask) | (reqflags & mask);
232 if ((new_flags ^ flags) == 0)
236 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
240 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
242 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
244 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
245 !capable(CAP_LINUX_IMMUTABLE))
247 if (!IS_IMMUTABLE(inode)) {
248 error = gfs2_permission(inode, MAY_WRITE);
252 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
253 if (flags & GFS2_DIF_JDATA)
254 gfs2_log_flush(sdp, ip->i_gl);
255 error = filemap_fdatawrite(inode->i_mapping);
258 error = filemap_fdatawait(inode->i_mapping);
262 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
265 error = gfs2_meta_inode_buffer(ip, &bh);
268 gfs2_trans_add_bh(ip->i_gl, bh, 1);
269 ip->i_diskflags = new_flags;
270 gfs2_dinode_out(ip, bh->b_data);
272 gfs2_set_inode_flags(inode);
273 gfs2_set_aops(inode);
277 gfs2_glock_dq_uninit(&gh);
279 mnt_drop_write(filp->f_path.mnt);
283 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
285 struct inode *inode = filp->f_path.dentry->d_inode;
286 u32 fsflags, gfsflags;
288 if (get_user(fsflags, ptr))
291 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
292 if (!S_ISDIR(inode->i_mode)) {
293 if (gfsflags & GFS2_DIF_INHERIT_JDATA)
294 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
295 return do_gfs2_set_flags(filp, gfsflags, ~0);
297 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
300 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
303 case FS_IOC_GETFLAGS:
304 return gfs2_get_flags(filp, (u32 __user *)arg);
305 case FS_IOC_SETFLAGS:
306 return gfs2_set_flags(filp, (u32 __user *)arg);
312 * gfs2_allocate_page_backing - Use bmap to allocate blocks
313 * @page: The (locked) page to allocate backing for
315 * We try to allocate all the blocks required for the page in
316 * one go. This might fail for various reasons, so we keep
317 * trying until all the blocks to back this page are allocated.
318 * If some of the blocks are already allocated, thats ok too.
321 static int gfs2_allocate_page_backing(struct page *page)
323 struct inode *inode = page->mapping->host;
324 struct buffer_head bh;
325 unsigned long size = PAGE_CACHE_SIZE;
326 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
331 gfs2_block_map(inode, lblock, &bh, 1);
332 if (!buffer_mapped(&bh))
335 lblock += (bh.b_size >> inode->i_blkbits);
341 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
342 * @vma: The virtual memory area
343 * @page: The page which is about to become writable
345 * When the page becomes writable, we need to ensure that we have
346 * blocks allocated on disk to back that page.
349 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
351 struct page *page = vmf->page;
352 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
353 struct gfs2_inode *ip = GFS2_I(inode);
354 struct gfs2_sbd *sdp = GFS2_SB(inode);
355 unsigned long last_index;
356 u64 pos = page->index << PAGE_CACHE_SHIFT;
357 unsigned int data_blocks, ind_blocks, rblocks;
358 struct gfs2_holder gh;
359 struct gfs2_alloc *al;
362 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
363 ret = gfs2_glock_nq(&gh);
367 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
368 set_bit(GIF_SW_PAGED, &ip->i_flags);
370 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE))
373 al = gfs2_alloc_get(ip);
377 ret = gfs2_quota_lock_check(ip);
380 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
381 al->al_requested = data_blocks + ind_blocks;
382 ret = gfs2_inplace_reserve(ip);
384 goto out_quota_unlock;
386 rblocks = RES_DINODE + ind_blocks;
387 if (gfs2_is_jdata(ip))
388 rblocks += data_blocks ? data_blocks : 1;
389 if (ind_blocks || data_blocks) {
390 rblocks += RES_STATFS + RES_QUOTA;
391 rblocks += gfs2_rg_blocks(al);
393 ret = gfs2_trans_begin(sdp, rblocks, 0);
399 last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
400 if (page->index > last_index)
401 goto out_unlock_page;
403 if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
404 goto out_unlock_page;
405 if (gfs2_is_stuffed(ip)) {
406 ret = gfs2_unstuff_dinode(ip, page);
408 goto out_unlock_page;
410 ret = gfs2_allocate_page_backing(page);
416 gfs2_inplace_release(ip);
418 gfs2_quota_unlock(ip);
424 gfs2_holder_uninit(&gh);
428 ret = VM_FAULT_SIGBUS;
432 static const struct vm_operations_struct gfs2_vm_ops = {
433 .fault = filemap_fault,
434 .page_mkwrite = gfs2_page_mkwrite,
439 * @file: The file to map
440 * @vma: The VMA which described the mapping
442 * There is no need to get a lock here unless we should be updating
443 * atime. We ignore any locking errors since the only consequence is
444 * a missed atime update (which will just be deferred until later).
449 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
451 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
453 if (!(file->f_flags & O_NOATIME) &&
454 !IS_NOATIME(&ip->i_inode)) {
455 struct gfs2_holder i_gh;
458 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
459 error = gfs2_glock_nq(&i_gh);
462 gfs2_glock_dq(&i_gh);
464 gfs2_holder_uninit(&i_gh);
468 vma->vm_ops = &gfs2_vm_ops;
469 vma->vm_flags |= VM_CAN_NONLINEAR;
475 * gfs2_open - open a file
476 * @inode: the inode to open
477 * @file: the struct file for this opening
482 static int gfs2_open(struct inode *inode, struct file *file)
484 struct gfs2_inode *ip = GFS2_I(inode);
485 struct gfs2_holder i_gh;
486 struct gfs2_file *fp;
489 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
493 mutex_init(&fp->f_fl_mutex);
495 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
496 file->private_data = fp;
498 if (S_ISREG(ip->i_inode.i_mode)) {
499 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
504 if (!(file->f_flags & O_LARGEFILE) &&
505 i_size_read(inode) > MAX_NON_LFS) {
510 gfs2_glock_dq_uninit(&i_gh);
516 gfs2_glock_dq_uninit(&i_gh);
518 file->private_data = NULL;
524 * gfs2_close - called to close a struct file
525 * @inode: the inode the struct file belongs to
526 * @file: the struct file being closed
531 static int gfs2_close(struct inode *inode, struct file *file)
533 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
534 struct gfs2_file *fp;
536 fp = file->private_data;
537 file->private_data = NULL;
539 if (gfs2_assert_warn(sdp, fp))
548 * gfs2_fsync - sync the dirty data for a file (across the cluster)
549 * @file: the file that points to the dentry
550 * @start: the start position in the file to sync
551 * @end: the end position in the file to sync
552 * @datasync: set if we can ignore timestamp changes
554 * The VFS will flush data for us. We only need to worry
555 * about metadata here.
560 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
563 struct inode *inode = file->f_mapping->host;
564 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
565 struct gfs2_inode *ip = GFS2_I(inode);
568 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
571 mutex_lock(&inode->i_mutex);
574 sync_state &= ~I_DIRTY_SYNC;
577 ret = sync_inode_metadata(inode, 1);
579 mutex_unlock(&inode->i_mutex);
582 gfs2_ail_flush(ip->i_gl);
585 mutex_unlock(&inode->i_mutex);
590 * gfs2_file_aio_write - Perform a write to a file
591 * @iocb: The io context
592 * @iov: The data to write
593 * @nr_segs: Number of @iov segments
594 * @pos: The file position
596 * We have to do a lock/unlock here to refresh the inode size for
597 * O_APPEND writes, otherwise we can land up writing at the wrong
598 * offset. There is still a race, but provided the app is using its
599 * own file locking, this will make O_APPEND work as expected.
603 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
604 unsigned long nr_segs, loff_t pos)
606 struct file *file = iocb->ki_filp;
608 if (file->f_flags & O_APPEND) {
609 struct dentry *dentry = file->f_dentry;
610 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
611 struct gfs2_holder gh;
614 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
617 gfs2_glock_dq_uninit(&gh);
620 return generic_file_aio_write(iocb, iov, nr_segs, pos);
623 static int empty_write_end(struct page *page, unsigned from,
624 unsigned to, int mode)
626 struct inode *inode = page->mapping->host;
627 struct gfs2_inode *ip = GFS2_I(inode);
628 struct buffer_head *bh;
629 unsigned offset, blksize = 1 << inode->i_blkbits;
630 pgoff_t end_index = i_size_read(inode) >> PAGE_CACHE_SHIFT;
632 zero_user(page, from, to-from);
633 mark_page_accessed(page);
635 if (page->index < end_index || !(mode & FALLOC_FL_KEEP_SIZE)) {
636 if (!gfs2_is_writeback(ip))
637 gfs2_page_add_databufs(ip, page, from, to);
639 block_commit_write(page, from, to);
644 bh = page_buffers(page);
645 while (offset < to) {
646 if (offset >= from) {
647 set_buffer_uptodate(bh);
648 mark_buffer_dirty(bh);
649 clear_buffer_new(bh);
650 write_dirty_buffer(bh, WRITE);
653 bh = bh->b_this_page;
657 bh = page_buffers(page);
658 while (offset < to) {
659 if (offset >= from) {
661 if (!buffer_uptodate(bh))
665 bh = bh->b_this_page;
670 static int needs_empty_write(sector_t block, struct inode *inode)
673 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
675 bh_map.b_size = 1 << inode->i_blkbits;
676 error = gfs2_block_map(inode, block, &bh_map, 0);
679 return !buffer_mapped(&bh_map);
682 static int write_empty_blocks(struct page *page, unsigned from, unsigned to,
685 struct inode *inode = page->mapping->host;
686 unsigned start, end, next, blksize;
687 sector_t block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
690 blksize = 1 << inode->i_blkbits;
692 while (next < from) {
699 ret = needs_empty_write(block, inode);
700 if (unlikely(ret < 0))
704 ret = __block_write_begin(page, start, end - start,
708 ret = empty_write_end(page, start, end, mode);
721 ret = __block_write_begin(page, start, end - start, gfs2_block_map);
724 ret = empty_write_end(page, start, end, mode);
732 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
735 struct gfs2_inode *ip = GFS2_I(inode);
736 struct buffer_head *dibh;
738 u64 start = offset >> PAGE_CACHE_SHIFT;
739 unsigned int start_offset = offset & ~PAGE_CACHE_MASK;
740 u64 end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
743 unsigned int end_offset = (offset + len) & ~PAGE_CACHE_MASK;
744 unsigned int from, to;
747 end_offset = PAGE_CACHE_SIZE;
749 error = gfs2_meta_inode_buffer(ip, &dibh);
753 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
755 if (gfs2_is_stuffed(ip)) {
756 error = gfs2_unstuff_dinode(ip, NULL);
762 offset = start << PAGE_CACHE_SHIFT;
764 to = PAGE_CACHE_SIZE;
765 while (curr <= end) {
766 page = grab_cache_page_write_begin(inode->i_mapping, curr,
768 if (unlikely(!page)) {
775 error = write_empty_blocks(page, from, to, mode);
776 if (!error && offset + to > inode->i_size &&
777 !(mode & FALLOC_FL_KEEP_SIZE)) {
778 i_size_write(inode, offset + to);
781 page_cache_release(page);
785 offset += PAGE_CACHE_SIZE;
789 gfs2_dinode_out(ip, dibh->b_data);
790 mark_inode_dirty(inode);
798 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len,
799 unsigned int *data_blocks, unsigned int *ind_blocks)
801 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
802 unsigned int max_blocks = ip->i_alloc->al_rgd->rd_free_clone;
803 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
805 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
806 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
809 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
810 so it might end up with fewer data blocks */
811 if (max_data <= *data_blocks)
813 *data_blocks = max_data;
814 *ind_blocks = max_blocks - max_data;
815 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
818 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
822 static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
825 struct inode *inode = file->f_path.dentry->d_inode;
826 struct gfs2_sbd *sdp = GFS2_SB(inode);
827 struct gfs2_inode *ip = GFS2_I(inode);
828 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
829 loff_t bytes, max_bytes;
830 struct gfs2_alloc *al;
832 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
833 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
834 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
836 /* We only support the FALLOC_FL_KEEP_SIZE mode */
837 if (mode & ~FALLOC_FL_KEEP_SIZE)
840 offset &= bsize_mask;
843 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
848 bytes = sdp->sd_sb.sb_bsize;
850 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
851 error = gfs2_glock_nq(&ip->i_gh);
855 if (!gfs2_write_alloc_required(ip, offset, len))
861 al = gfs2_alloc_get(ip);
867 error = gfs2_quota_lock_check(ip);
872 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
874 al->al_requested = data_blocks + ind_blocks;
875 error = gfs2_inplace_reserve(ip);
877 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) {
881 bytes = sdp->sd_sb.sb_bsize;
887 calc_max_reserv(ip, len, &max_bytes, &data_blocks, &ind_blocks);
888 al->al_requested = data_blocks + ind_blocks;
890 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
891 RES_RG_HDR + gfs2_rg_blocks(al);
892 if (gfs2_is_jdata(ip))
893 rblocks += data_blocks ? data_blocks : 1;
895 error = gfs2_trans_begin(sdp, rblocks,
896 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
900 error = fallocate_chunk(inode, offset, max_bytes, mode);
908 gfs2_inplace_release(ip);
909 gfs2_quota_unlock(ip);
915 gfs2_inplace_release(ip);
917 gfs2_quota_unlock(ip);
921 gfs2_glock_dq(&ip->i_gh);
923 gfs2_holder_uninit(&ip->i_gh);
927 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
930 * gfs2_setlease - acquire/release a file lease
931 * @file: the file pointer
935 * We don't currently have a way to enforce a lease across the whole
936 * cluster; until we do, disable leases (by just returning -EINVAL),
937 * unless the administrator has requested purely local locking.
939 * Locking: called under lock_flocks
944 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
950 * gfs2_lock - acquire/release a posix lock on a file
951 * @file: the file pointer
952 * @cmd: either modify or retrieve lock state, possibly wait
953 * @fl: type and range of lock
958 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
960 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
961 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
962 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
964 if (!(fl->fl_flags & FL_POSIX))
966 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
969 if (cmd == F_CANCELLK) {
972 fl->fl_type = F_UNLCK;
974 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
977 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
978 else if (fl->fl_type == F_UNLCK)
979 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
981 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
984 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
986 struct gfs2_file *fp = file->private_data;
987 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
988 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
989 struct gfs2_glock *gl;
994 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
995 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
997 mutex_lock(&fp->f_fl_mutex);
1001 if (fl_gh->gh_state == state)
1003 flock_lock_file_wait(file,
1004 &(struct file_lock){.fl_type = F_UNLCK});
1005 gfs2_glock_dq_wait(fl_gh);
1006 gfs2_holder_reinit(state, flags, fl_gh);
1008 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1009 &gfs2_flock_glops, CREATE, &gl);
1012 gfs2_holder_init(gl, state, flags, fl_gh);
1015 error = gfs2_glock_nq(fl_gh);
1017 gfs2_holder_uninit(fl_gh);
1018 if (error == GLR_TRYFAILED)
1021 error = flock_lock_file_wait(file, fl);
1022 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1026 mutex_unlock(&fp->f_fl_mutex);
1030 static void do_unflock(struct file *file, struct file_lock *fl)
1032 struct gfs2_file *fp = file->private_data;
1033 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1035 mutex_lock(&fp->f_fl_mutex);
1036 flock_lock_file_wait(file, fl);
1038 gfs2_glock_dq_wait(fl_gh);
1039 gfs2_holder_uninit(fl_gh);
1041 mutex_unlock(&fp->f_fl_mutex);
1045 * gfs2_flock - acquire/release a flock lock on a file
1046 * @file: the file pointer
1047 * @cmd: either modify or retrieve lock state, possibly wait
1048 * @fl: type and range of lock
1053 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1055 if (!(fl->fl_flags & FL_FLOCK))
1057 if (fl->fl_type & LOCK_MAND)
1060 if (fl->fl_type == F_UNLCK) {
1061 do_unflock(file, fl);
1064 return do_flock(file, cmd, fl);
1068 const struct file_operations gfs2_file_fops = {
1069 .llseek = gfs2_llseek,
1070 .read = do_sync_read,
1071 .aio_read = generic_file_aio_read,
1072 .write = do_sync_write,
1073 .aio_write = gfs2_file_aio_write,
1074 .unlocked_ioctl = gfs2_ioctl,
1077 .release = gfs2_close,
1078 .fsync = gfs2_fsync,
1080 .flock = gfs2_flock,
1081 .splice_read = generic_file_splice_read,
1082 .splice_write = generic_file_splice_write,
1083 .setlease = gfs2_setlease,
1084 .fallocate = gfs2_fallocate,
1087 const struct file_operations gfs2_dir_fops = {
1088 .readdir = gfs2_readdir,
1089 .unlocked_ioctl = gfs2_ioctl,
1091 .release = gfs2_close,
1092 .fsync = gfs2_fsync,
1094 .flock = gfs2_flock,
1095 .llseek = default_llseek,
1098 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1100 const struct file_operations gfs2_file_fops_nolock = {
1101 .llseek = gfs2_llseek,
1102 .read = do_sync_read,
1103 .aio_read = generic_file_aio_read,
1104 .write = do_sync_write,
1105 .aio_write = gfs2_file_aio_write,
1106 .unlocked_ioctl = gfs2_ioctl,
1109 .release = gfs2_close,
1110 .fsync = gfs2_fsync,
1111 .splice_read = generic_file_splice_read,
1112 .splice_write = generic_file_splice_write,
1113 .setlease = generic_setlease,
1114 .fallocate = gfs2_fallocate,
1117 const struct file_operations gfs2_dir_fops_nolock = {
1118 .readdir = gfs2_readdir,
1119 .unlocked_ioctl = gfs2_ioctl,
1121 .release = gfs2_close,
1122 .fsync = gfs2_fsync,
1123 .llseek = default_llseek,