2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2011 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/namei.h>
16 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/crc32.h>
20 #include <linux/fiemap.h>
21 #include <linux/security.h>
22 #include <asm/uaccess.h>
40 struct gfs2_skip_data {
46 static int iget_test(struct inode *inode, void *opaque)
48 struct gfs2_inode *ip = GFS2_I(inode);
49 struct gfs2_skip_data *data = opaque;
51 if (ip->i_no_addr == data->no_addr) {
52 if (data->non_block &&
53 inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
62 static int iget_set(struct inode *inode, void *opaque)
64 struct gfs2_inode *ip = GFS2_I(inode);
65 struct gfs2_skip_data *data = opaque;
69 inode->i_ino = (unsigned long)(data->no_addr);
70 ip->i_no_addr = data->no_addr;
74 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int non_block)
76 unsigned long hash = (unsigned long)no_addr;
77 struct gfs2_skip_data data;
79 data.no_addr = no_addr;
81 data.non_block = non_block;
82 return ilookup5(sb, hash, iget_test, &data);
85 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr,
88 struct gfs2_skip_data data;
89 unsigned long hash = (unsigned long)no_addr;
91 data.no_addr = no_addr;
93 data.non_block = non_block;
94 return iget5_locked(sb, hash, iget_test, iget_set, &data);
98 * gfs2_set_iop - Sets inode operations
99 * @inode: The inode with correct i_mode filled in
101 * GFS2 lookup code fills in vfs inode contents based on info obtained
102 * from directory entry inside gfs2_inode_lookup().
105 static void gfs2_set_iop(struct inode *inode)
107 struct gfs2_sbd *sdp = GFS2_SB(inode);
108 umode_t mode = inode->i_mode;
111 inode->i_op = &gfs2_file_iops;
112 if (gfs2_localflocks(sdp))
113 inode->i_fop = &gfs2_file_fops_nolock;
115 inode->i_fop = &gfs2_file_fops;
116 } else if (S_ISDIR(mode)) {
117 inode->i_op = &gfs2_dir_iops;
118 if (gfs2_localflocks(sdp))
119 inode->i_fop = &gfs2_dir_fops_nolock;
121 inode->i_fop = &gfs2_dir_fops;
122 } else if (S_ISLNK(mode)) {
123 inode->i_op = &gfs2_symlink_iops;
125 inode->i_op = &gfs2_file_iops;
126 init_special_inode(inode, inode->i_mode, inode->i_rdev);
131 * gfs2_inode_lookup - Lookup an inode
132 * @sb: The super block
133 * @no_addr: The inode number
134 * @type: The type of the inode
135 * non_block: Can we block on inodes that are being freed?
137 * Returns: A VFS inode, or an error
140 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
141 u64 no_addr, u64 no_formal_ino, int non_block)
144 struct gfs2_inode *ip;
145 struct gfs2_glock *io_gl = NULL;
148 inode = gfs2_iget(sb, no_addr, non_block);
152 return ERR_PTR(-ENOBUFS);
154 if (inode->i_state & I_NEW) {
155 struct gfs2_sbd *sdp = GFS2_SB(inode);
156 ip->i_no_formal_ino = no_formal_ino;
158 error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
161 ip->i_gl->gl_object = ip;
163 error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
167 set_bit(GIF_INVALID, &ip->i_flags);
168 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
172 ip->i_iopen_gh.gh_gl->gl_object = ip;
173 gfs2_glock_put(io_gl);
176 if (type == DT_UNKNOWN) {
177 /* Inode glock must be locked already */
178 error = gfs2_inode_refresh(GFS2_I(inode));
182 inode->i_mode = DT2IF(type);
186 unlock_new_inode(inode);
192 ip->i_iopen_gh.gh_gl->gl_object = NULL;
193 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
196 gfs2_glock_put(io_gl);
198 ip->i_gl->gl_object = NULL;
199 gfs2_glock_put(ip->i_gl);
202 return ERR_PTR(error);
205 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
206 u64 *no_formal_ino, unsigned int blktype)
208 struct super_block *sb = sdp->sd_vfs;
209 struct gfs2_holder i_gh;
210 struct inode *inode = NULL;
213 /* Must not read in block until block type is verified */
214 error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
215 LM_ST_EXCLUSIVE, GL_SKIP, &i_gh);
217 return ERR_PTR(error);
219 error = gfs2_check_blk_type(sdp, no_addr, blktype);
223 inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
227 /* Two extra checks for NFS only */
230 if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
234 if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
241 gfs2_glock_dq_uninit(&i_gh);
242 return error ? ERR_PTR(error) : inode;
249 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
253 gfs2_str2qstr(&qstr, name);
254 inode = gfs2_lookupi(dip, &qstr, 1);
255 /* gfs2_lookupi has inconsistent callers: vfs
256 * related routines expect NULL for no entry found,
257 * gfs2_lookup_simple callers expect ENOENT
258 * and do not check for NULL.
261 return ERR_PTR(-ENOENT);
268 * gfs2_lookupi - Look up a filename in a directory and return its inode
269 * @d_gh: An initialized holder for the directory glock
270 * @name: The name of the inode to look for
271 * @is_root: If 1, ignore the caller's permissions
272 * @i_gh: An uninitialized holder for the new inode glock
274 * This can be called via the VFS filldir function when NFS is doing
275 * a readdirplus and the inode which its intending to stat isn't
276 * already in cache. In this case we must not take the directory glock
277 * again, since the readdir call will have already taken that lock.
282 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
285 struct super_block *sb = dir->i_sb;
286 struct gfs2_inode *dip = GFS2_I(dir);
287 struct gfs2_holder d_gh;
289 struct inode *inode = NULL;
292 if (!name->len || name->len > GFS2_FNAMESIZE)
293 return ERR_PTR(-ENAMETOOLONG);
295 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
296 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
297 dir == sb->s_root->d_inode)) {
302 if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
303 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
305 return ERR_PTR(error);
310 error = gfs2_permission(dir, MAY_EXEC, 0);
315 inode = gfs2_dir_search(dir, name);
317 error = PTR_ERR(inode);
320 gfs2_glock_dq_uninit(&d_gh);
321 if (error == -ENOENT)
323 return inode ? inode : ERR_PTR(error);
327 * create_ok - OK to create a new on-disk inode here?
328 * @dip: Directory in which dinode is to be created
329 * @name: Name of new dinode
335 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
340 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, 0);
344 /* Don't create entries in an unlinked directory */
345 if (!dip->i_inode.i_nlink)
348 error = gfs2_dir_check(&dip->i_inode, name, NULL);
359 if (dip->i_entries == (u32)-1)
361 if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
367 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
368 unsigned int *uid, unsigned int *gid)
370 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
371 (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
374 else if (dip->i_inode.i_uid != current_fsuid())
376 *uid = dip->i_inode.i_uid;
378 *uid = current_fsuid();
380 if (dip->i_inode.i_mode & S_ISGID) {
383 *gid = dip->i_inode.i_gid;
385 *gid = current_fsgid();
388 static int alloc_dinode(struct gfs2_inode *dip, u64 *no_addr, u64 *generation)
390 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
393 if (gfs2_alloc_get(dip) == NULL)
396 dip->i_alloc->al_requested = RES_DINODE;
397 error = gfs2_inplace_reserve(dip);
401 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
405 error = gfs2_alloc_di(dip, no_addr, generation);
410 gfs2_inplace_release(dip);
416 static void gfs2_init_dir(struct buffer_head *dibh,
417 const struct gfs2_inode *parent)
419 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
420 struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
422 gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
423 dent->de_inum = di->di_num; /* already GFS2 endian */
424 dent->de_type = cpu_to_be16(DT_DIR);
426 dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
427 gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
428 gfs2_inum_out(parent, dent);
429 dent->de_type = cpu_to_be16(DT_DIR);
434 * init_dinode - Fill in a new dinode structure
435 * @dip: The directory this inode is being created in
436 * @gl: The glock covering the new inode
437 * @inum: The inode number
438 * @mode: The file permissions
439 * @uid: The uid of the new inode
440 * @gid: The gid of the new inode
441 * @generation: The generation number of the new inode
442 * @dev: The device number (if a device node)
443 * @symname: The symlink destination (if a symlink)
444 * @size: The inode size (ignored for directories)
445 * @bhp: The buffer head (returned to caller)
449 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
450 const struct gfs2_inum_host *inum, unsigned int mode,
451 unsigned int uid, unsigned int gid,
452 const u64 *generation, dev_t dev, const char *symname,
453 unsigned size, struct buffer_head **bhp)
455 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
456 struct gfs2_dinode *di;
457 struct buffer_head *dibh;
458 struct timespec tv = CURRENT_TIME;
460 dibh = gfs2_meta_new(gl, inum->no_addr);
461 gfs2_trans_add_bh(gl, dibh, 1);
462 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
463 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
464 di = (struct gfs2_dinode *)dibh->b_data;
466 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
467 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
468 di->di_mode = cpu_to_be32(mode);
469 di->di_uid = cpu_to_be32(uid);
470 di->di_gid = cpu_to_be32(gid);
472 di->di_size = cpu_to_be64(size);
473 di->di_blocks = cpu_to_be64(1);
474 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec);
475 di->di_major = cpu_to_be32(MAJOR(dev));
476 di->di_minor = cpu_to_be32(MINOR(dev));
477 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
478 di->di_generation = cpu_to_be64(*generation);
481 di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
487 memset(&di->__pad4, 0, sizeof(di->__pad4));
489 di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
490 di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
491 di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
492 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
494 switch(mode & S_IFMT) {
496 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
497 gfs2_tune_get(sdp, gt_new_files_jdata))
498 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
501 di->di_flags |= cpu_to_be32(dip->i_diskflags &
502 GFS2_DIF_INHERIT_JDATA);
503 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
504 di->di_size = cpu_to_be64(sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode));
505 di->di_entries = cpu_to_be32(2);
506 gfs2_init_dir(dibh, dip);
509 memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, size);
513 set_buffer_uptodate(dibh);
518 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
519 unsigned int mode, const struct gfs2_inum_host *inum,
520 const u64 *generation, dev_t dev, const char *symname,
521 unsigned int size, struct buffer_head **bhp)
523 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
524 unsigned int uid, gid;
527 munge_mode_uid_gid(dip, &mode, &uid, &gid);
528 if (!gfs2_alloc_get(dip))
531 error = gfs2_quota_lock(dip, uid, gid);
535 error = gfs2_quota_check(dip, uid, gid);
539 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
543 init_dinode(dip, gl, inum, mode, uid, gid, generation, dev, symname, size, bhp);
544 gfs2_quota_change(dip, +1, uid, gid);
548 gfs2_quota_unlock(dip);
554 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
555 struct gfs2_inode *ip)
557 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
558 struct gfs2_alloc *al;
560 struct buffer_head *dibh;
563 al = gfs2_alloc_get(dip);
567 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
571 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
572 if (alloc_required < 0)
573 goto fail_quota_locks;
574 if (alloc_required) {
575 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
577 goto fail_quota_locks;
579 al->al_requested = sdp->sd_max_dirres;
581 error = gfs2_inplace_reserve(dip);
583 goto fail_quota_locks;
585 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
586 al->al_rgd->rd_length +
588 RES_STATFS + RES_QUOTA, 0);
592 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
594 goto fail_quota_locks;
597 error = gfs2_dir_add(&dip->i_inode, name, ip);
601 error = gfs2_meta_inode_buffer(ip, &dibh);
604 inc_nlink(&ip->i_inode);
605 if (S_ISDIR(ip->i_inode.i_mode))
606 inc_nlink(&ip->i_inode);
607 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
608 gfs2_dinode_out(ip, dibh->b_data);
616 if (dip->i_alloc->al_rgd)
617 gfs2_inplace_release(dip);
620 gfs2_quota_unlock(dip);
627 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
628 const struct qstr *qstr)
635 err = security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
636 &name, &value, &len);
639 if (err == -EOPNOTSUPP)
644 err = __gfs2_xattr_set(&ip->i_inode, name, value, len, 0,
645 GFS2_EATYPE_SECURITY);
653 * gfs2_create_inode - Create a new inode
654 * @dir: The parent directory
655 * @dentry: The new dentry
656 * @mode: The permissions on the new inode
657 * @dev: For device nodes, this is the device number
658 * @symname: For symlinks, this is the link destination
659 * @size: The initial size of the inode (ignored for directories)
661 * Returns: 0 on success, or error code
664 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
665 unsigned int mode, dev_t dev, const char *symname,
668 const struct qstr *name = &dentry->d_name;
669 struct gfs2_holder ghs[2];
670 struct inode *inode = NULL;
671 struct gfs2_inode *dip = GFS2_I(dir);
672 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
673 struct gfs2_inum_host inum = { .no_addr = 0, .no_formal_ino = 0 };
676 struct buffer_head *bh = NULL;
678 if (!name->len || name->len > GFS2_FNAMESIZE)
679 return -ENAMETOOLONG;
681 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
685 error = create_ok(dip, name, mode);
689 error = alloc_dinode(dip, &inum.no_addr, &generation);
692 inum.no_formal_ino = generation;
694 error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
695 LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
699 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev, symname, size, &bh);
703 inode = gfs2_inode_lookup(dir->i_sb, IF2DT(mode), inum.no_addr,
704 inum.no_formal_ino, 0);
708 error = gfs2_inode_refresh(GFS2_I(inode));
712 error = gfs2_acl_create(dip, inode);
716 error = gfs2_security_init(dip, GFS2_I(inode), name);
720 error = link_dinode(dip, name, GFS2_I(inode));
728 if (dip->i_alloc->al_rgd)
729 gfs2_inplace_release(dip);
730 gfs2_quota_unlock(dip);
732 gfs2_glock_dq_uninit_m(2, ghs);
733 mark_inode_dirty(inode);
734 d_instantiate(dentry, inode);
738 gfs2_glock_dq_uninit(ghs + 1);
739 if (inode && !IS_ERR(inode))
742 gfs2_glock_dq_uninit(ghs);
750 * gfs2_create - Create a file
751 * @dir: The directory in which to create the file
752 * @dentry: The dentry of the new file
753 * @mode: The mode of the new file
758 static int gfs2_create(struct inode *dir, struct dentry *dentry,
759 int mode, struct nameidata *nd)
765 ret = gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0);
766 if (ret != -EEXIST || (nd && (nd->flags & LOOKUP_EXCL)))
769 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
773 return PTR_ERR(inode);
777 d_instantiate(dentry, inode);
782 * gfs2_lookup - Look up a filename in a directory and return its inode
783 * @dir: The directory inode
784 * @dentry: The dentry of the new inode
785 * @nd: passed from Linux VFS, ignored by us
787 * Called by the VFS layer. Lock dir and call gfs2_lookupi()
792 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
793 struct nameidata *nd)
795 struct inode *inode = NULL;
797 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
798 if (inode && IS_ERR(inode))
799 return ERR_CAST(inode);
802 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
803 struct gfs2_holder gh;
805 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
808 return ERR_PTR(error);
810 gfs2_glock_dq_uninit(&gh);
811 return d_splice_alias(inode, dentry);
813 d_add(dentry, inode);
819 * gfs2_link - Link to a file
820 * @old_dentry: The inode to link
821 * @dir: Add link to this directory
822 * @dentry: The name of the link
824 * Link the inode in "old_dentry" into the directory "dir" with the
830 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
831 struct dentry *dentry)
833 struct gfs2_inode *dip = GFS2_I(dir);
834 struct gfs2_sbd *sdp = GFS2_SB(dir);
835 struct inode *inode = old_dentry->d_inode;
836 struct gfs2_inode *ip = GFS2_I(inode);
837 struct gfs2_holder ghs[2];
838 struct buffer_head *dibh;
842 if (S_ISDIR(inode->i_mode))
845 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
846 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
848 error = gfs2_glock_nq(ghs); /* parent */
852 error = gfs2_glock_nq(ghs + 1); /* child */
857 if (inode->i_nlink == 0)
860 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC, 0);
864 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
875 if (!dip->i_inode.i_nlink)
878 if (dip->i_entries == (u32)-1)
881 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
884 if (!ip->i_inode.i_nlink)
887 if (ip->i_inode.i_nlink == (u32)-1)
890 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
895 if (alloc_required) {
896 struct gfs2_alloc *al = gfs2_alloc_get(dip);
902 error = gfs2_quota_lock_check(dip);
906 al->al_requested = sdp->sd_max_dirres;
908 error = gfs2_inplace_reserve(dip);
912 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
914 2 * RES_DINODE + RES_STATFS +
919 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
924 error = gfs2_meta_inode_buffer(ip, &dibh);
928 error = gfs2_dir_add(dir, &dentry->d_name, ip);
932 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
933 inc_nlink(&ip->i_inode);
934 ip->i_inode.i_ctime = CURRENT_TIME;
935 gfs2_dinode_out(ip, dibh->b_data);
936 mark_inode_dirty(&ip->i_inode);
944 gfs2_inplace_release(dip);
947 gfs2_quota_unlock(dip);
952 gfs2_glock_dq(ghs + 1);
956 gfs2_holder_uninit(ghs);
957 gfs2_holder_uninit(ghs + 1);
960 d_instantiate(dentry, inode);
961 mark_inode_dirty(inode);
967 * gfs2_unlink_ok - check to see that a inode is still in a directory
968 * @dip: the directory
969 * @name: the name of the file
972 * Assumes that the lock on (at least) @dip is held.
974 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
977 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
978 const struct gfs2_inode *ip)
982 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
985 if ((dip->i_inode.i_mode & S_ISVTX) &&
986 dip->i_inode.i_uid != current_fsuid() &&
987 ip->i_inode.i_uid != current_fsuid() && !capable(CAP_FOWNER))
990 if (IS_APPEND(&dip->i_inode))
993 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, 0);
997 error = gfs2_dir_check(&dip->i_inode, name, ip);
1005 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1006 * @dip: The parent directory
1007 * @name: The name of the entry in the parent directory
1008 * @bh: The inode buffer for the inode to be removed
1009 * @inode: The inode to be removed
1011 * Called with all the locks and in a transaction. This will only be
1012 * called for a directory after it has been checked to ensure it is empty.
1014 * Returns: 0 on success, or an error
1017 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1018 const struct dentry *dentry,
1019 struct buffer_head *bh)
1021 struct inode *inode = dentry->d_inode;
1022 struct gfs2_inode *ip = GFS2_I(inode);
1025 error = gfs2_dir_del(dip, dentry);
1030 inode->i_ctime = CURRENT_TIME;
1031 if (S_ISDIR(inode->i_mode))
1035 gfs2_trans_add_bh(ip->i_gl, bh, 1);
1036 gfs2_dinode_out(ip, bh->b_data);
1037 mark_inode_dirty(inode);
1038 if (inode->i_nlink == 0)
1039 gfs2_unlink_di(inode);
1045 * gfs2_unlink - Unlink an inode (this does rmdir as well)
1046 * @dir: The inode of the directory containing the inode to unlink
1047 * @dentry: The file itself
1049 * This routine uses the type of the inode as a flag to figure out
1050 * whether this is an unlink or an rmdir.
1055 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1057 struct gfs2_inode *dip = GFS2_I(dir);
1058 struct gfs2_sbd *sdp = GFS2_SB(dir);
1059 struct inode *inode = dentry->d_inode;
1060 struct gfs2_inode *ip = GFS2_I(inode);
1061 struct buffer_head *bh;
1062 struct gfs2_holder ghs[3];
1063 struct gfs2_rgrpd *rgd;
1064 struct gfs2_holder ri_gh;
1067 error = gfs2_rindex_hold(sdp, &ri_gh);
1071 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1072 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1074 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1075 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1078 error = gfs2_glock_nq(ghs); /* parent */
1082 error = gfs2_glock_nq(ghs + 1); /* child */
1087 if (inode->i_nlink == 0)
1090 if (S_ISDIR(inode->i_mode)) {
1092 if (ip->i_entries > 2 || inode->i_nlink > 2)
1096 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1100 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1104 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1108 error = gfs2_meta_inode_buffer(ip, &bh);
1112 error = gfs2_unlink_inode(dip, dentry, bh);
1116 gfs2_trans_end(sdp);
1118 gfs2_glock_dq(ghs + 2);
1120 gfs2_holder_uninit(ghs + 2);
1121 gfs2_glock_dq(ghs + 1);
1123 gfs2_holder_uninit(ghs + 1);
1126 gfs2_holder_uninit(ghs);
1127 gfs2_glock_dq_uninit(&ri_gh);
1132 * gfs2_symlink - Create a symlink
1133 * @dir: The directory to create the symlink in
1134 * @dentry: The dentry to put the symlink in
1135 * @symname: The thing which the link points to
1140 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1141 const char *symname)
1143 struct gfs2_sbd *sdp = GFS2_SB(dir);
1146 size = strlen(symname);
1147 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1148 return -ENAMETOOLONG;
1150 return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size);
1154 * gfs2_mkdir - Make a directory
1155 * @dir: The parent directory of the new one
1156 * @dentry: The dentry of the new directory
1157 * @mode: The mode of the new directory
1162 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1164 return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0);
1168 * gfs2_mknod - Make a special file
1169 * @dir: The directory in which the special file will reside
1170 * @dentry: The dentry of the special file
1171 * @mode: The mode of the special file
1172 * @dev: The device specification of the special file
1176 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, int mode,
1179 return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0);
1183 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1187 * Follow @to back to the root and make sure we don't encounter @this
1188 * Assumes we already hold the rename lock.
1193 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1195 struct inode *dir = &to->i_inode;
1196 struct super_block *sb = dir->i_sb;
1203 if (dir == &this->i_inode) {
1207 if (dir == sb->s_root->d_inode) {
1212 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1214 error = PTR_ERR(tmp);
1228 * gfs2_rename - Rename a file
1229 * @odir: Parent directory of old file name
1230 * @odentry: The old dentry of the file
1231 * @ndir: Parent directory of new file name
1232 * @ndentry: The new dentry of the file
1237 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1238 struct inode *ndir, struct dentry *ndentry)
1240 struct gfs2_inode *odip = GFS2_I(odir);
1241 struct gfs2_inode *ndip = GFS2_I(ndir);
1242 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1243 struct gfs2_inode *nip = NULL;
1244 struct gfs2_sbd *sdp = GFS2_SB(odir);
1245 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, }, ri_gh;
1246 struct gfs2_rgrpd *nrgd;
1247 unsigned int num_gh;
1249 int alloc_required = 0;
1253 if (ndentry->d_inode) {
1254 nip = GFS2_I(ndentry->d_inode);
1259 error = gfs2_rindex_hold(sdp, &ri_gh);
1264 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1269 if (S_ISDIR(ip->i_inode.i_mode)) {
1271 /* don't move a dirctory into it's subdir */
1272 error = gfs2_ok_to_move(ip, ndip);
1279 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1281 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1284 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1288 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1290 /* grab the resource lock for unlink flag twiddling
1291 * this is the case of the target file already existing
1292 * so we unlink before doing the rename
1294 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr);
1296 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1299 for (x = 0; x < num_gh; x++) {
1300 error = gfs2_glock_nq(ghs + x);
1306 if (ip->i_inode.i_nlink == 0)
1309 /* Check out the old directory */
1311 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1315 /* Check out the new directory */
1318 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1322 if (nip->i_inode.i_nlink == 0) {
1327 if (S_ISDIR(nip->i_inode.i_mode)) {
1328 if (nip->i_entries < 2) {
1329 gfs2_consist_inode(nip);
1333 if (nip->i_entries > 2) {
1339 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC, 0);
1343 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1355 if (!ndip->i_inode.i_nlink) {
1359 if (ndip->i_entries == (u32)-1) {
1363 if (S_ISDIR(ip->i_inode.i_mode) &&
1364 ndip->i_inode.i_nlink == (u32)-1) {
1371 /* Check out the dir to be renamed */
1374 error = gfs2_permission(odentry->d_inode, MAY_WRITE, 0);
1380 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1381 error = alloc_required;
1386 if (alloc_required) {
1387 struct gfs2_alloc *al = gfs2_alloc_get(ndip);
1393 error = gfs2_quota_lock_check(ndip);
1397 al->al_requested = sdp->sd_max_dirres;
1399 error = gfs2_inplace_reserve_ri(ndip);
1403 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1404 gfs2_rg_blocks(al) +
1405 4 * RES_DINODE + 4 * RES_LEAF +
1406 RES_STATFS + RES_QUOTA + 4, 0);
1410 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1411 5 * RES_LEAF + 4, 0);
1416 /* Remove the target file, if it exists */
1419 struct buffer_head *bh;
1420 error = gfs2_meta_inode_buffer(nip, &bh);
1423 error = gfs2_unlink_inode(ndip, ndentry, bh);
1428 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1432 struct buffer_head *dibh;
1433 error = gfs2_meta_inode_buffer(ip, &dibh);
1436 ip->i_inode.i_ctime = CURRENT_TIME;
1437 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1438 gfs2_dinode_out(ip, dibh->b_data);
1442 error = gfs2_dir_del(odip, odentry);
1446 error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1451 gfs2_trans_end(sdp);
1454 gfs2_inplace_release(ndip);
1457 gfs2_quota_unlock(ndip);
1460 gfs2_alloc_put(ndip);
1463 gfs2_glock_dq(ghs + x);
1464 gfs2_holder_uninit(ghs + x);
1468 gfs2_glock_dq_uninit(&r_gh);
1470 gfs2_glock_dq_uninit(&ri_gh);
1475 * gfs2_follow_link - Follow a symbolic link
1476 * @dentry: The dentry of the link
1477 * @nd: Data that we pass to vfs_follow_link()
1479 * This can handle symlinks of any size.
1481 * Returns: 0 on success or error code
1484 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1486 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1487 struct gfs2_holder i_gh;
1488 struct buffer_head *dibh;
1493 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1494 error = gfs2_glock_nq(&i_gh);
1496 gfs2_holder_uninit(&i_gh);
1497 nd_set_link(nd, ERR_PTR(error));
1501 size = (unsigned int)i_size_read(&ip->i_inode);
1503 gfs2_consist_inode(ip);
1504 buf = ERR_PTR(-EIO);
1508 error = gfs2_meta_inode_buffer(ip, &dibh);
1510 buf = ERR_PTR(error);
1514 buf = kzalloc(size + 1, GFP_NOFS);
1516 buf = ERR_PTR(-ENOMEM);
1518 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1521 gfs2_glock_dq_uninit(&i_gh);
1522 nd_set_link(nd, buf);
1526 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1528 char *s = nd_get_link(nd);
1536 * @mask: The mask to be tested
1537 * @flags: Indicates whether this is an RCU path walk or not
1539 * This may be called from the VFS directly, or from within GFS2 with the
1540 * inode locked, so we look to see if the glock is already locked and only
1541 * lock the glock if its not already been done.
1546 int gfs2_permission(struct inode *inode, int mask, unsigned int flags)
1548 struct gfs2_inode *ip;
1549 struct gfs2_holder i_gh;
1555 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1556 if (flags & IPERM_FLAG_RCU)
1558 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1564 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1567 error = generic_permission(inode, mask, flags, gfs2_check_acl);
1569 gfs2_glock_dq_uninit(&i_gh);
1574 static int __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1576 struct inode *inode = &ip->i_inode;
1577 struct buffer_head *dibh;
1580 error = gfs2_meta_inode_buffer(ip, &dibh);
1584 setattr_copy(inode, attr);
1585 mark_inode_dirty(inode);
1586 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1587 gfs2_dinode_out(ip, dibh->b_data);
1593 * gfs2_setattr_simple -
1600 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1604 if (current->journal_info)
1605 return __gfs2_setattr_simple(ip, attr);
1607 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0);
1611 error = __gfs2_setattr_simple(ip, attr);
1612 gfs2_trans_end(GFS2_SB(&ip->i_inode));
1616 static int setattr_chown(struct inode *inode, struct iattr *attr)
1618 struct gfs2_inode *ip = GFS2_I(inode);
1619 struct gfs2_sbd *sdp = GFS2_SB(inode);
1620 u32 ouid, ogid, nuid, ngid;
1623 ouid = inode->i_uid;
1624 ogid = inode->i_gid;
1625 nuid = attr->ia_uid;
1626 ngid = attr->ia_gid;
1628 if (!(attr->ia_valid & ATTR_UID) || ouid == nuid)
1629 ouid = nuid = NO_QUOTA_CHANGE;
1630 if (!(attr->ia_valid & ATTR_GID) || ogid == ngid)
1631 ogid = ngid = NO_QUOTA_CHANGE;
1633 if (!gfs2_alloc_get(ip))
1636 error = gfs2_quota_lock(ip, nuid, ngid);
1640 if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1641 error = gfs2_quota_check(ip, nuid, ngid);
1646 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1650 error = gfs2_setattr_simple(ip, attr);
1654 if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1655 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1656 gfs2_quota_change(ip, -blocks, ouid, ogid);
1657 gfs2_quota_change(ip, blocks, nuid, ngid);
1661 gfs2_trans_end(sdp);
1663 gfs2_quota_unlock(ip);
1670 * gfs2_setattr - Change attributes on an inode
1671 * @dentry: The dentry which is changing
1672 * @attr: The structure describing the change
1674 * The VFS layer wants to change one or more of an inodes attributes. Write
1675 * that change out to disk.
1680 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1682 struct inode *inode = dentry->d_inode;
1683 struct gfs2_inode *ip = GFS2_I(inode);
1684 struct gfs2_holder i_gh;
1687 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1692 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1695 error = inode_change_ok(inode, attr);
1699 if (attr->ia_valid & ATTR_SIZE)
1700 error = gfs2_setattr_size(inode, attr->ia_size);
1701 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1702 error = setattr_chown(inode, attr);
1703 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1704 error = gfs2_acl_chmod(ip, attr);
1706 error = gfs2_setattr_simple(ip, attr);
1709 gfs2_glock_dq_uninit(&i_gh);
1711 mark_inode_dirty(inode);
1716 * gfs2_getattr - Read out an inode's attributes
1717 * @mnt: The vfsmount the inode is being accessed from
1718 * @dentry: The dentry to stat
1719 * @stat: The inode's stats
1721 * This may be called from the VFS directly, or from within GFS2 with the
1722 * inode locked, so we look to see if the glock is already locked and only
1723 * lock the glock if its not already been done. Note that its the NFS
1724 * readdirplus operation which causes this to be called (from filldir)
1725 * with the glock already held.
1730 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1733 struct inode *inode = dentry->d_inode;
1734 struct gfs2_inode *ip = GFS2_I(inode);
1735 struct gfs2_holder gh;
1739 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1740 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1746 generic_fillattr(inode, stat);
1748 gfs2_glock_dq_uninit(&gh);
1753 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1754 const void *data, size_t size, int flags)
1756 struct inode *inode = dentry->d_inode;
1757 struct gfs2_inode *ip = GFS2_I(inode);
1758 struct gfs2_holder gh;
1761 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1762 ret = gfs2_glock_nq(&gh);
1764 ret = generic_setxattr(dentry, name, data, size, flags);
1767 gfs2_holder_uninit(&gh);
1771 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1772 void *data, size_t size)
1774 struct inode *inode = dentry->d_inode;
1775 struct gfs2_inode *ip = GFS2_I(inode);
1776 struct gfs2_holder gh;
1779 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1780 ret = gfs2_glock_nq(&gh);
1782 ret = generic_getxattr(dentry, name, data, size);
1785 gfs2_holder_uninit(&gh);
1789 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1791 struct inode *inode = dentry->d_inode;
1792 struct gfs2_inode *ip = GFS2_I(inode);
1793 struct gfs2_holder gh;
1796 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1797 ret = gfs2_glock_nq(&gh);
1799 ret = generic_removexattr(dentry, name);
1802 gfs2_holder_uninit(&gh);
1806 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1809 struct gfs2_inode *ip = GFS2_I(inode);
1810 struct gfs2_holder gh;
1813 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1817 mutex_lock(&inode->i_mutex);
1819 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1823 if (gfs2_is_stuffed(ip)) {
1824 u64 phys = ip->i_no_addr << inode->i_blkbits;
1825 u64 size = i_size_read(inode);
1826 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1827 FIEMAP_EXTENT_DATA_INLINE;
1828 phys += sizeof(struct gfs2_dinode);
1830 if (start + len > size)
1833 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1838 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1842 gfs2_glock_dq_uninit(&gh);
1844 mutex_unlock(&inode->i_mutex);
1848 const struct inode_operations gfs2_file_iops = {
1849 .permission = gfs2_permission,
1850 .setattr = gfs2_setattr,
1851 .getattr = gfs2_getattr,
1852 .setxattr = gfs2_setxattr,
1853 .getxattr = gfs2_getxattr,
1854 .listxattr = gfs2_listxattr,
1855 .removexattr = gfs2_removexattr,
1856 .fiemap = gfs2_fiemap,
1859 const struct inode_operations gfs2_dir_iops = {
1860 .create = gfs2_create,
1861 .lookup = gfs2_lookup,
1863 .unlink = gfs2_unlink,
1864 .symlink = gfs2_symlink,
1865 .mkdir = gfs2_mkdir,
1866 .rmdir = gfs2_unlink,
1867 .mknod = gfs2_mknod,
1868 .rename = gfs2_rename,
1869 .permission = gfs2_permission,
1870 .setattr = gfs2_setattr,
1871 .getattr = gfs2_getattr,
1872 .setxattr = gfs2_setxattr,
1873 .getxattr = gfs2_getxattr,
1874 .listxattr = gfs2_listxattr,
1875 .removexattr = gfs2_removexattr,
1876 .fiemap = gfs2_fiemap,
1879 const struct inode_operations gfs2_symlink_iops = {
1880 .readlink = generic_readlink,
1881 .follow_link = gfs2_follow_link,
1882 .put_link = gfs2_put_link,
1883 .permission = gfs2_permission,
1884 .setattr = gfs2_setattr,
1885 .getattr = gfs2_getattr,
1886 .setxattr = gfs2_setxattr,
1887 .getxattr = gfs2_getxattr,
1888 .listxattr = gfs2_listxattr,
1889 .removexattr = gfs2_removexattr,
1890 .fiemap = gfs2_fiemap,