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/crc32.h>
23 #include <linux/writeback.h>
24 #include <asm/uaccess.h>
25 #include <linux/dlm.h>
26 #include <linux/dlm_plock.h>
43 * gfs2_llseek - seek to a location in a file
46 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
48 * SEEK_END requires the glock for the file because it references the
51 * Returns: The new offset, or errno
54 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
56 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
57 struct gfs2_holder i_gh;
61 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
64 error = generic_file_llseek_unlocked(file, offset, origin);
65 gfs2_glock_dq_uninit(&i_gh);
68 error = generic_file_llseek_unlocked(file, offset, origin);
74 * gfs2_readdir - Read directory entries from a directory
75 * @file: The directory to read from
76 * @dirent: Buffer for dirents
77 * @filldir: Function used to do the copying
82 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
84 struct inode *dir = file->f_mapping->host;
85 struct gfs2_inode *dip = GFS2_I(dir);
86 struct gfs2_holder d_gh;
87 u64 offset = file->f_pos;
90 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
91 error = gfs2_glock_nq(&d_gh);
93 gfs2_holder_uninit(&d_gh);
97 error = gfs2_dir_read(dir, &offset, dirent, filldir);
99 gfs2_glock_dq_uninit(&d_gh);
101 file->f_pos = offset;
108 * @table: A table of 32 u32 flags
109 * @val: a 32 bit value to convert
111 * This function can be used to convert between fsflags values and
112 * GFS2's own flags values.
114 * Returns: the converted flags
116 static u32 fsflags_cvt(const u32 *table, u32 val)
128 static const u32 fsflags_to_gfs2[32] = {
130 [4] = GFS2_DIF_IMMUTABLE,
131 [5] = GFS2_DIF_APPENDONLY,
132 [7] = GFS2_DIF_NOATIME,
133 [12] = GFS2_DIF_EXHASH,
134 [14] = GFS2_DIF_INHERIT_JDATA,
137 static const u32 gfs2_to_fsflags[32] = {
138 [gfs2fl_Sync] = FS_SYNC_FL,
139 [gfs2fl_Immutable] = FS_IMMUTABLE_FL,
140 [gfs2fl_AppendOnly] = FS_APPEND_FL,
141 [gfs2fl_NoAtime] = FS_NOATIME_FL,
142 [gfs2fl_ExHash] = FS_INDEX_FL,
143 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
146 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
148 struct inode *inode = filp->f_path.dentry->d_inode;
149 struct gfs2_inode *ip = GFS2_I(inode);
150 struct gfs2_holder gh;
154 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
155 error = gfs2_glock_nq(&gh);
159 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
160 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
161 fsflags |= FS_JOURNAL_DATA_FL;
162 if (put_user(fsflags, ptr))
166 gfs2_holder_uninit(&gh);
170 void gfs2_set_inode_flags(struct inode *inode)
172 struct gfs2_inode *ip = GFS2_I(inode);
173 unsigned int flags = inode->i_flags;
175 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
176 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
177 flags |= S_IMMUTABLE;
178 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
180 if (ip->i_diskflags & GFS2_DIF_NOATIME)
182 if (ip->i_diskflags & GFS2_DIF_SYNC)
184 inode->i_flags = flags;
187 /* Flags that can be set by user space */
188 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
189 GFS2_DIF_IMMUTABLE| \
190 GFS2_DIF_APPENDONLY| \
194 GFS2_DIF_INHERIT_JDATA)
197 * gfs2_set_flags - set flags on an inode
199 * @flags: The flags to set
200 * @mask: Indicates which flags are valid
203 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
205 struct inode *inode = filp->f_path.dentry->d_inode;
206 struct gfs2_inode *ip = GFS2_I(inode);
207 struct gfs2_sbd *sdp = GFS2_SB(inode);
208 struct buffer_head *bh;
209 struct gfs2_holder gh;
211 u32 new_flags, flags;
213 error = mnt_want_write(filp->f_path.mnt);
217 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
222 if (!is_owner_or_cap(inode))
226 flags = ip->i_diskflags;
227 new_flags = (flags & ~mask) | (reqflags & mask);
228 if ((new_flags ^ flags) == 0)
232 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
236 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
238 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
240 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
241 !capable(CAP_LINUX_IMMUTABLE))
243 if (!IS_IMMUTABLE(inode)) {
244 error = gfs2_permission(inode, MAY_WRITE);
248 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
249 if (flags & GFS2_DIF_JDATA)
250 gfs2_log_flush(sdp, ip->i_gl);
251 error = filemap_fdatawrite(inode->i_mapping);
254 error = filemap_fdatawait(inode->i_mapping);
258 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
261 error = gfs2_meta_inode_buffer(ip, &bh);
264 gfs2_trans_add_bh(ip->i_gl, bh, 1);
265 ip->i_diskflags = new_flags;
266 gfs2_dinode_out(ip, bh->b_data);
268 gfs2_set_inode_flags(inode);
269 gfs2_set_aops(inode);
273 gfs2_glock_dq_uninit(&gh);
275 mnt_drop_write(filp->f_path.mnt);
279 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
281 struct inode *inode = filp->f_path.dentry->d_inode;
282 u32 fsflags, gfsflags;
284 if (get_user(fsflags, ptr))
287 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
288 if (!S_ISDIR(inode->i_mode)) {
289 if (gfsflags & GFS2_DIF_INHERIT_JDATA)
290 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
291 return do_gfs2_set_flags(filp, gfsflags, ~0);
293 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
296 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
299 case FS_IOC_GETFLAGS:
300 return gfs2_get_flags(filp, (u32 __user *)arg);
301 case FS_IOC_SETFLAGS:
302 return gfs2_set_flags(filp, (u32 __user *)arg);
308 * gfs2_allocate_page_backing - Use bmap to allocate blocks
309 * @page: The (locked) page to allocate backing for
311 * We try to allocate all the blocks required for the page in
312 * one go. This might fail for various reasons, so we keep
313 * trying until all the blocks to back this page are allocated.
314 * If some of the blocks are already allocated, thats ok too.
317 static int gfs2_allocate_page_backing(struct page *page)
319 struct inode *inode = page->mapping->host;
320 struct buffer_head bh;
321 unsigned long size = PAGE_CACHE_SIZE;
322 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
327 gfs2_block_map(inode, lblock, &bh, 1);
328 if (!buffer_mapped(&bh))
331 lblock += (bh.b_size >> inode->i_blkbits);
337 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
338 * @vma: The virtual memory area
339 * @page: The page which is about to become writable
341 * When the page becomes writable, we need to ensure that we have
342 * blocks allocated on disk to back that page.
345 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
347 struct page *page = vmf->page;
348 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
349 struct gfs2_inode *ip = GFS2_I(inode);
350 struct gfs2_sbd *sdp = GFS2_SB(inode);
351 unsigned long last_index;
352 u64 pos = page->index << PAGE_CACHE_SHIFT;
353 unsigned int data_blocks, ind_blocks, rblocks;
354 struct gfs2_holder gh;
355 struct gfs2_alloc *al;
358 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
359 ret = gfs2_glock_nq(&gh);
363 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
364 set_bit(GIF_SW_PAGED, &ip->i_flags);
366 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE))
369 al = gfs2_alloc_get(ip);
373 ret = gfs2_quota_lock_check(ip);
376 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
377 al->al_requested = data_blocks + ind_blocks;
378 ret = gfs2_inplace_reserve(ip);
380 goto out_quota_unlock;
382 rblocks = RES_DINODE + ind_blocks;
383 if (gfs2_is_jdata(ip))
384 rblocks += data_blocks ? data_blocks : 1;
385 if (ind_blocks || data_blocks)
386 rblocks += RES_STATFS + RES_QUOTA;
387 ret = gfs2_trans_begin(sdp, rblocks, 0);
393 last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
394 if (page->index > last_index)
395 goto out_unlock_page;
397 if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
398 goto out_unlock_page;
399 if (gfs2_is_stuffed(ip)) {
400 ret = gfs2_unstuff_dinode(ip, page);
402 goto out_unlock_page;
404 ret = gfs2_allocate_page_backing(page);
410 gfs2_inplace_release(ip);
412 gfs2_quota_unlock(ip);
418 gfs2_holder_uninit(&gh);
422 ret = VM_FAULT_SIGBUS;
426 static const struct vm_operations_struct gfs2_vm_ops = {
427 .fault = filemap_fault,
428 .page_mkwrite = gfs2_page_mkwrite,
433 * @file: The file to map
434 * @vma: The VMA which described the mapping
436 * There is no need to get a lock here unless we should be updating
437 * atime. We ignore any locking errors since the only consequence is
438 * a missed atime update (which will just be deferred until later).
443 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
445 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
447 if (!(file->f_flags & O_NOATIME)) {
448 struct gfs2_holder i_gh;
451 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
452 error = gfs2_glock_nq(&i_gh);
455 gfs2_glock_dq_uninit(&i_gh);
457 vma->vm_ops = &gfs2_vm_ops;
458 vma->vm_flags |= VM_CAN_NONLINEAR;
464 * gfs2_open - open a file
465 * @inode: the inode to open
466 * @file: the struct file for this opening
471 static int gfs2_open(struct inode *inode, struct file *file)
473 struct gfs2_inode *ip = GFS2_I(inode);
474 struct gfs2_holder i_gh;
475 struct gfs2_file *fp;
478 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
482 mutex_init(&fp->f_fl_mutex);
484 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
485 file->private_data = fp;
487 if (S_ISREG(ip->i_inode.i_mode)) {
488 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
493 if (!(file->f_flags & O_LARGEFILE) &&
494 ip->i_disksize > MAX_NON_LFS) {
499 gfs2_glock_dq_uninit(&i_gh);
505 gfs2_glock_dq_uninit(&i_gh);
507 file->private_data = NULL;
513 * gfs2_close - called to close a struct file
514 * @inode: the inode the struct file belongs to
515 * @file: the struct file being closed
520 static int gfs2_close(struct inode *inode, struct file *file)
522 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
523 struct gfs2_file *fp;
525 fp = file->private_data;
526 file->private_data = NULL;
528 if (gfs2_assert_warn(sdp, fp))
537 * gfs2_fsync - sync the dirty data for a file (across the cluster)
538 * @file: the file that points to the dentry (we ignore this)
539 * @dentry: the dentry that points to the inode to sync
541 * The VFS will flush "normal" data for us. We only need to worry
542 * about metadata here. For journaled data, we just do a log flush
543 * as we can't avoid it. Otherwise we can just bale out if datasync
544 * is set. For stuffed inodes we must flush the log in order to
545 * ensure that all data is on disk.
547 * The call to write_inode_now() is there to write back metadata and
548 * the inode itself. It does also try and write the data, but thats
549 * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
555 static int gfs2_fsync(struct file *file, int datasync)
557 struct inode *inode = file->f_mapping->host;
558 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
561 if (gfs2_is_jdata(GFS2_I(inode))) {
562 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
566 if (sync_state != 0) {
568 ret = write_inode_now(inode, 0);
570 if (gfs2_is_stuffed(GFS2_I(inode)))
571 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
578 * gfs2_file_aio_write - Perform a write to a file
579 * @iocb: The io context
580 * @iov: The data to write
581 * @nr_segs: Number of @iov segments
582 * @pos: The file position
584 * We have to do a lock/unlock here to refresh the inode size for
585 * O_APPEND writes, otherwise we can land up writing at the wrong
586 * offset. There is still a race, but provided the app is using its
587 * own file locking, this will make O_APPEND work as expected.
591 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
592 unsigned long nr_segs, loff_t pos)
594 struct file *file = iocb->ki_filp;
596 if (file->f_flags & O_APPEND) {
597 struct dentry *dentry = file->f_dentry;
598 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
599 struct gfs2_holder gh;
602 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
605 gfs2_glock_dq_uninit(&gh);
608 return generic_file_aio_write(iocb, iov, nr_segs, pos);
611 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
614 * gfs2_setlease - acquire/release a file lease
615 * @file: the file pointer
619 * We don't currently have a way to enforce a lease across the whole
620 * cluster; until we do, disable leases (by just returning -EINVAL),
621 * unless the administrator has requested purely local locking.
626 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
632 * gfs2_lock - acquire/release a posix lock on a file
633 * @file: the file pointer
634 * @cmd: either modify or retrieve lock state, possibly wait
635 * @fl: type and range of lock
640 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
642 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
643 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
644 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
646 if (!(fl->fl_flags & FL_POSIX))
648 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
651 if (cmd == F_CANCELLK) {
654 fl->fl_type = F_UNLCK;
656 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
659 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
660 else if (fl->fl_type == F_UNLCK)
661 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
663 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
666 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
668 struct gfs2_file *fp = file->private_data;
669 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
670 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
671 struct gfs2_glock *gl;
676 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
677 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
679 mutex_lock(&fp->f_fl_mutex);
683 if (fl_gh->gh_state == state)
685 flock_lock_file_wait(file,
686 &(struct file_lock){.fl_type = F_UNLCK});
687 gfs2_glock_dq_wait(fl_gh);
688 gfs2_holder_reinit(state, flags, fl_gh);
690 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
691 &gfs2_flock_glops, CREATE, &gl);
694 gfs2_holder_init(gl, state, flags, fl_gh);
697 error = gfs2_glock_nq(fl_gh);
699 gfs2_holder_uninit(fl_gh);
700 if (error == GLR_TRYFAILED)
703 error = flock_lock_file_wait(file, fl);
704 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
708 mutex_unlock(&fp->f_fl_mutex);
712 static void do_unflock(struct file *file, struct file_lock *fl)
714 struct gfs2_file *fp = file->private_data;
715 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
717 mutex_lock(&fp->f_fl_mutex);
718 flock_lock_file_wait(file, fl);
720 gfs2_glock_dq_uninit(fl_gh);
721 mutex_unlock(&fp->f_fl_mutex);
725 * gfs2_flock - acquire/release a flock lock on a file
726 * @file: the file pointer
727 * @cmd: either modify or retrieve lock state, possibly wait
728 * @fl: type and range of lock
733 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
735 if (!(fl->fl_flags & FL_FLOCK))
737 if (fl->fl_type & LOCK_MAND)
740 if (fl->fl_type == F_UNLCK) {
741 do_unflock(file, fl);
744 return do_flock(file, cmd, fl);
748 const struct file_operations gfs2_file_fops = {
749 .llseek = gfs2_llseek,
750 .read = do_sync_read,
751 .aio_read = generic_file_aio_read,
752 .write = do_sync_write,
753 .aio_write = gfs2_file_aio_write,
754 .unlocked_ioctl = gfs2_ioctl,
757 .release = gfs2_close,
761 .splice_read = generic_file_splice_read,
762 .splice_write = generic_file_splice_write,
763 .setlease = gfs2_setlease,
766 const struct file_operations gfs2_dir_fops = {
767 .readdir = gfs2_readdir,
768 .unlocked_ioctl = gfs2_ioctl,
770 .release = gfs2_close,
776 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
778 const struct file_operations gfs2_file_fops_nolock = {
779 .llseek = gfs2_llseek,
780 .read = do_sync_read,
781 .aio_read = generic_file_aio_read,
782 .write = do_sync_write,
783 .aio_write = gfs2_file_aio_write,
784 .unlocked_ioctl = gfs2_ioctl,
787 .release = gfs2_close,
789 .splice_read = generic_file_splice_read,
790 .splice_write = generic_file_splice_write,
791 .setlease = generic_setlease,
794 const struct file_operations gfs2_dir_fops_nolock = {
795 .readdir = gfs2_readdir,
796 .unlocked_ioctl = gfs2_ioctl,
798 .release = gfs2_close,