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);
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);
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(const struct gfs2_inode *dip,
370 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
371 (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
372 if (S_ISDIR(inode->i_mode))
373 inode->i_mode |= S_ISUID;
374 else if (dip->i_inode.i_uid != current_fsuid())
375 inode->i_mode &= ~07111;
376 inode->i_uid = dip->i_inode.i_uid;
378 inode->i_uid = current_fsuid();
380 if (dip->i_inode.i_mode & S_ISGID) {
381 if (S_ISDIR(inode->i_mode))
382 inode->i_mode |= S_ISGID;
383 inode->i_gid = dip->i_inode.i_gid;
385 inode->i_gid = current_fsgid();
388 static int alloc_dinode(struct gfs2_inode *ip, u32 flags)
390 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
394 error = gfs2_inplace_reserve(ip, RES_DINODE, flags);
398 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
402 error = gfs2_alloc_blocks(ip, &ip->i_no_addr, &dblocks, 1, &ip->i_generation);
403 ip->i_no_formal_ino = ip->i_generation;
404 ip->i_inode.i_ino = ip->i_no_addr;
405 ip->i_goal = ip->i_no_addr;
410 gfs2_inplace_release(ip);
415 static void gfs2_init_dir(struct buffer_head *dibh,
416 const struct gfs2_inode *parent)
418 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
419 struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
421 gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
422 dent->de_inum = di->di_num; /* already GFS2 endian */
423 dent->de_type = cpu_to_be16(DT_DIR);
425 dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
426 gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
427 gfs2_inum_out(parent, dent);
428 dent->de_type = cpu_to_be16(DT_DIR);
433 * init_dinode - Fill in a new dinode structure
434 * @dip: The directory this inode is being created in
436 * @symname: The symlink destination (if a symlink)
437 * @bhp: The buffer head (returned to caller)
441 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
442 const char *symname, struct buffer_head **bhp)
444 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
445 struct gfs2_dinode *di;
446 struct buffer_head *dibh;
447 struct timespec tv = CURRENT_TIME;
449 dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
450 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
451 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
452 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
453 di = (struct gfs2_dinode *)dibh->b_data;
455 di->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
456 di->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
457 di->di_mode = cpu_to_be32(ip->i_inode.i_mode);
458 di->di_uid = cpu_to_be32(ip->i_inode.i_uid);
459 di->di_gid = cpu_to_be32(ip->i_inode.i_gid);
461 di->di_size = cpu_to_be64(ip->i_inode.i_size);
462 di->di_blocks = cpu_to_be64(1);
463 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec);
464 di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
465 di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
466 di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_no_addr);
467 di->di_generation = cpu_to_be64(ip->i_generation);
470 di->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) ? GFS2_FORMAT_DE : 0);
476 memset(&di->__pad4, 0, sizeof(di->__pad4));
478 di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
479 di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
480 di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
481 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
483 switch(ip->i_inode.i_mode & S_IFMT) {
485 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
486 gfs2_tune_get(sdp, gt_new_files_jdata))
487 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
490 di->di_flags |= cpu_to_be32(dip->i_diskflags &
491 GFS2_DIF_INHERIT_JDATA);
492 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
493 di->di_size = cpu_to_be64(sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode));
494 di->di_entries = cpu_to_be32(2);
495 gfs2_init_dir(dibh, dip);
498 memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
502 set_buffer_uptodate(dibh);
507 static int make_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
508 const char *symname, struct buffer_head **bhp)
510 struct inode *inode = &ip->i_inode;
511 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
514 error = gfs2_rindex_update(sdp);
518 error = gfs2_quota_lock(dip, inode->i_uid, inode->i_gid);
522 error = gfs2_quota_check(dip, inode->i_uid, inode->i_gid);
526 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
530 init_dinode(dip, ip, symname, bhp);
531 gfs2_quota_change(dip, +1, inode->i_uid, inode->i_gid);
535 gfs2_quota_unlock(dip);
539 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
540 struct gfs2_inode *ip)
542 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
544 struct buffer_head *dibh;
547 error = gfs2_rindex_update(sdp);
551 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
555 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
556 if (alloc_required < 0)
557 goto fail_quota_locks;
558 if (alloc_required) {
559 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
561 goto fail_quota_locks;
563 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
565 goto fail_quota_locks;
567 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
568 dip->i_rgd->rd_length +
570 RES_STATFS + RES_QUOTA, 0);
574 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
576 goto fail_quota_locks;
579 error = gfs2_dir_add(&dip->i_inode, name, ip);
583 error = gfs2_meta_inode_buffer(ip, &dibh);
586 set_nlink(&ip->i_inode, S_ISDIR(ip->i_inode.i_mode) ? 2 : 1);
587 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
588 gfs2_dinode_out(ip, dibh->b_data);
597 gfs2_inplace_release(dip);
600 gfs2_quota_unlock(dip);
606 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
609 const struct xattr *xattr;
612 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
613 err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
615 GFS2_EATYPE_SECURITY);
622 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
623 const struct qstr *qstr)
625 return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
626 &gfs2_initxattrs, NULL);
630 * gfs2_create_inode - Create a new inode
631 * @dir: The parent directory
632 * @dentry: The new dentry
633 * @mode: The permissions on the new inode
634 * @dev: For device nodes, this is the device number
635 * @symname: For symlinks, this is the link destination
636 * @size: The initial size of the inode (ignored for directories)
638 * Returns: 0 on success, or error code
641 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
642 umode_t mode, dev_t dev, const char *symname,
643 unsigned int size, int excl)
645 const struct qstr *name = &dentry->d_name;
646 struct gfs2_holder ghs[2];
647 struct inode *inode = NULL;
648 struct gfs2_inode *dip = GFS2_I(dir), *ip;
649 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
650 struct gfs2_glock *io_gl;
652 struct buffer_head *bh = NULL;
655 if (!name->len || name->len > GFS2_FNAMESIZE)
656 return -ENAMETOOLONG;
658 error = gfs2_rs_alloc(dip);
662 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
666 error = create_ok(dip, name, mode);
667 if ((error == -EEXIST) && S_ISREG(mode) && !excl) {
668 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
669 gfs2_glock_dq_uninit(ghs);
670 d_instantiate(dentry, inode);
671 return IS_ERR(inode) ? PTR_ERR(inode) : 0;
676 inode = new_inode(sdp->sd_vfs);
678 gfs2_glock_dq_uninit(ghs);
682 error = gfs2_rs_alloc(ip);
684 goto fail_free_inode;
686 set_bit(GIF_INVALID, &ip->i_flags);
687 inode->i_mode = mode;
689 inode->i_size = size;
690 munge_mode_uid_gid(dip, inode);
691 ip->i_goal = dip->i_goal;
693 if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
694 (dip->i_diskflags & GFS2_DIF_TOPDIR))
695 aflags |= GFS2_AF_ORLOV;
697 error = alloc_dinode(ip, aflags);
699 goto fail_free_inode;
701 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
703 goto fail_free_inode;
705 ip->i_gl->gl_object = ip;
706 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
708 goto fail_free_inode;
710 error = make_dinode(dip, ip, symname, &bh);
714 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
718 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
722 ip->i_iopen_gh.gh_gl->gl_object = ip;
723 gfs2_glock_put(io_gl);
725 insert_inode_hash(inode);
727 error = gfs2_inode_refresh(ip);
731 error = gfs2_acl_create(dip, inode);
735 error = gfs2_security_init(dip, ip, name);
739 error = link_dinode(dip, name, ip);
747 gfs2_inplace_release(dip);
748 gfs2_quota_unlock(dip);
749 mark_inode_dirty(inode);
750 gfs2_glock_dq_uninit_m(2, ghs);
751 d_instantiate(dentry, inode);
755 gfs2_glock_dq_uninit(ghs + 1);
757 gfs2_glock_put(ip->i_gl);
761 gfs2_glock_dq_uninit(ghs + 1);
764 gfs2_glock_put(ip->i_gl);
766 free_inode_nonrcu(inode);
769 gfs2_glock_dq_uninit(ghs);
770 if (inode && !IS_ERR(inode)) {
771 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
781 * gfs2_create - Create a file
782 * @dir: The directory in which to create the file
783 * @dentry: The dentry of the new file
784 * @mode: The mode of the new file
789 static int gfs2_create(struct inode *dir, struct dentry *dentry,
790 umode_t mode, bool excl)
792 return gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0, excl);
796 * gfs2_lookup - Look up a filename in a directory and return its inode
797 * @dir: The directory inode
798 * @dentry: The dentry of the new inode
799 * @nd: passed from Linux VFS, ignored by us
801 * Called by the VFS layer. Lock dir and call gfs2_lookupi()
806 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
809 struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
810 if (inode && !IS_ERR(inode)) {
811 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
812 struct gfs2_holder gh;
814 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
817 return ERR_PTR(error);
819 gfs2_glock_dq_uninit(&gh);
821 return d_splice_alias(inode, dentry);
825 * gfs2_link - Link to a file
826 * @old_dentry: The inode to link
827 * @dir: Add link to this directory
828 * @dentry: The name of the link
830 * Link the inode in "old_dentry" into the directory "dir" with the
836 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
837 struct dentry *dentry)
839 struct gfs2_inode *dip = GFS2_I(dir);
840 struct gfs2_sbd *sdp = GFS2_SB(dir);
841 struct inode *inode = old_dentry->d_inode;
842 struct gfs2_inode *ip = GFS2_I(inode);
843 struct gfs2_holder ghs[2];
844 struct buffer_head *dibh;
848 if (S_ISDIR(inode->i_mode))
851 error = gfs2_rs_alloc(dip);
855 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
856 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
858 error = gfs2_glock_nq(ghs); /* parent */
862 error = gfs2_glock_nq(ghs + 1); /* child */
867 if (inode->i_nlink == 0)
870 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
874 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
885 if (!dip->i_inode.i_nlink)
888 if (dip->i_entries == (u32)-1)
891 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
894 if (!ip->i_inode.i_nlink)
897 if (ip->i_inode.i_nlink == (u32)-1)
900 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
905 if (alloc_required) {
906 error = gfs2_quota_lock_check(dip);
910 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
914 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
915 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
916 2 * RES_DINODE + RES_STATFS +
921 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
926 error = gfs2_meta_inode_buffer(ip, &dibh);
930 error = gfs2_dir_add(dir, &dentry->d_name, ip);
934 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
935 inc_nlink(&ip->i_inode);
936 ip->i_inode.i_ctime = CURRENT_TIME;
938 d_instantiate(dentry, inode);
939 mark_inode_dirty(inode);
947 gfs2_inplace_release(dip);
950 gfs2_quota_unlock(dip);
952 gfs2_glock_dq(ghs + 1);
956 gfs2_holder_uninit(ghs);
957 gfs2_holder_uninit(ghs + 1);
962 * gfs2_unlink_ok - check to see that a inode is still in a directory
963 * @dip: the directory
964 * @name: the name of the file
967 * Assumes that the lock on (at least) @dip is held.
969 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
972 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
973 const struct gfs2_inode *ip)
977 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
980 if ((dip->i_inode.i_mode & S_ISVTX) &&
981 dip->i_inode.i_uid != current_fsuid() &&
982 ip->i_inode.i_uid != current_fsuid() && !capable(CAP_FOWNER))
985 if (IS_APPEND(&dip->i_inode))
988 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
992 error = gfs2_dir_check(&dip->i_inode, name, ip);
1000 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1001 * @dip: The parent directory
1002 * @name: The name of the entry in the parent directory
1003 * @inode: The inode to be removed
1005 * Called with all the locks and in a transaction. This will only be
1006 * called for a directory after it has been checked to ensure it is empty.
1008 * Returns: 0 on success, or an error
1011 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1012 const struct dentry *dentry)
1014 struct inode *inode = dentry->d_inode;
1015 struct gfs2_inode *ip = GFS2_I(inode);
1018 error = gfs2_dir_del(dip, dentry);
1023 inode->i_ctime = CURRENT_TIME;
1024 if (S_ISDIR(inode->i_mode))
1028 mark_inode_dirty(inode);
1029 if (inode->i_nlink == 0)
1030 gfs2_unlink_di(inode);
1036 * gfs2_unlink - Unlink an inode (this does rmdir as well)
1037 * @dir: The inode of the directory containing the inode to unlink
1038 * @dentry: The file itself
1040 * This routine uses the type of the inode as a flag to figure out
1041 * whether this is an unlink or an rmdir.
1046 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1048 struct gfs2_inode *dip = GFS2_I(dir);
1049 struct gfs2_sbd *sdp = GFS2_SB(dir);
1050 struct inode *inode = dentry->d_inode;
1051 struct gfs2_inode *ip = GFS2_I(inode);
1052 struct gfs2_holder ghs[3];
1053 struct gfs2_rgrpd *rgd;
1056 error = gfs2_rindex_update(sdp);
1062 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1063 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1065 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1069 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1072 error = gfs2_glock_nq(ghs); /* parent */
1076 error = gfs2_glock_nq(ghs + 1); /* child */
1081 if (inode->i_nlink == 0)
1084 if (S_ISDIR(inode->i_mode)) {
1086 if (ip->i_entries > 2 || inode->i_nlink > 2)
1090 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1094 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1098 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1102 error = gfs2_unlink_inode(dip, dentry);
1105 gfs2_trans_end(sdp);
1107 gfs2_glock_dq(ghs + 2);
1109 gfs2_glock_dq(ghs + 1);
1113 gfs2_holder_uninit(ghs + 2);
1115 gfs2_holder_uninit(ghs + 1);
1116 gfs2_holder_uninit(ghs);
1121 * gfs2_symlink - Create a symlink
1122 * @dir: The directory to create the symlink in
1123 * @dentry: The dentry to put the symlink in
1124 * @symname: The thing which the link points to
1129 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1130 const char *symname)
1132 struct gfs2_sbd *sdp = GFS2_SB(dir);
1135 size = strlen(symname);
1136 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1137 return -ENAMETOOLONG;
1139 return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1143 * gfs2_mkdir - Make a directory
1144 * @dir: The parent directory of the new one
1145 * @dentry: The dentry of the new directory
1146 * @mode: The mode of the new directory
1151 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1153 return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0, 0);
1157 * gfs2_mknod - Make a special file
1158 * @dir: The directory in which the special file will reside
1159 * @dentry: The dentry of the special file
1160 * @mode: The mode of the special file
1161 * @dev: The device specification of the special file
1165 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1168 return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0, 0);
1172 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1176 * Follow @to back to the root and make sure we don't encounter @this
1177 * Assumes we already hold the rename lock.
1182 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1184 struct inode *dir = &to->i_inode;
1185 struct super_block *sb = dir->i_sb;
1192 if (dir == &this->i_inode) {
1196 if (dir == sb->s_root->d_inode) {
1201 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1203 error = PTR_ERR(tmp);
1217 * gfs2_rename - Rename a file
1218 * @odir: Parent directory of old file name
1219 * @odentry: The old dentry of the file
1220 * @ndir: Parent directory of new file name
1221 * @ndentry: The new dentry of the file
1226 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1227 struct inode *ndir, struct dentry *ndentry)
1229 struct gfs2_inode *odip = GFS2_I(odir);
1230 struct gfs2_inode *ndip = GFS2_I(ndir);
1231 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1232 struct gfs2_inode *nip = NULL;
1233 struct gfs2_sbd *sdp = GFS2_SB(odir);
1234 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1235 struct gfs2_rgrpd *nrgd;
1236 unsigned int num_gh;
1238 int alloc_required = 0;
1242 if (ndentry->d_inode) {
1243 nip = GFS2_I(ndentry->d_inode);
1248 error = gfs2_rindex_update(sdp);
1252 error = gfs2_rs_alloc(ndip);
1257 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1262 if (S_ISDIR(ip->i_inode.i_mode)) {
1264 /* don't move a dirctory into it's subdir */
1265 error = gfs2_ok_to_move(ip, ndip);
1272 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1274 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1277 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1281 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1283 /* grab the resource lock for unlink flag twiddling
1284 * this is the case of the target file already existing
1285 * so we unlink before doing the rename
1287 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1289 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1292 for (x = 0; x < num_gh; x++) {
1293 error = gfs2_glock_nq(ghs + x);
1299 if (ip->i_inode.i_nlink == 0)
1302 /* Check out the old directory */
1304 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1308 /* Check out the new directory */
1311 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1315 if (nip->i_inode.i_nlink == 0) {
1320 if (S_ISDIR(nip->i_inode.i_mode)) {
1321 if (nip->i_entries < 2) {
1322 gfs2_consist_inode(nip);
1326 if (nip->i_entries > 2) {
1332 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1336 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1348 if (!ndip->i_inode.i_nlink) {
1352 if (ndip->i_entries == (u32)-1) {
1356 if (S_ISDIR(ip->i_inode.i_mode) &&
1357 ndip->i_inode.i_nlink == (u32)-1) {
1364 /* Check out the dir to be renamed */
1367 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1373 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1374 error = alloc_required;
1378 if (alloc_required) {
1379 error = gfs2_quota_lock_check(ndip);
1383 error = gfs2_inplace_reserve(ndip, sdp->sd_max_dirres, 0);
1387 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1388 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1389 4 * RES_DINODE + 4 * RES_LEAF +
1390 RES_STATFS + RES_QUOTA + 4, 0);
1394 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1395 5 * RES_LEAF + 4, 0);
1400 /* Remove the target file, if it exists */
1403 error = gfs2_unlink_inode(ndip, ndentry);
1406 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1410 struct buffer_head *dibh;
1411 error = gfs2_meta_inode_buffer(ip, &dibh);
1414 ip->i_inode.i_ctime = CURRENT_TIME;
1415 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1416 gfs2_dinode_out(ip, dibh->b_data);
1420 error = gfs2_dir_del(odip, odentry);
1424 error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1429 gfs2_trans_end(sdp);
1432 gfs2_inplace_release(ndip);
1435 gfs2_quota_unlock(ndip);
1438 gfs2_glock_dq(ghs + x);
1439 gfs2_holder_uninit(ghs + x);
1443 gfs2_glock_dq_uninit(&r_gh);
1449 * gfs2_follow_link - Follow a symbolic link
1450 * @dentry: The dentry of the link
1451 * @nd: Data that we pass to vfs_follow_link()
1453 * This can handle symlinks of any size.
1455 * Returns: 0 on success or error code
1458 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1460 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1461 struct gfs2_holder i_gh;
1462 struct buffer_head *dibh;
1467 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1468 error = gfs2_glock_nq(&i_gh);
1470 gfs2_holder_uninit(&i_gh);
1471 nd_set_link(nd, ERR_PTR(error));
1475 size = (unsigned int)i_size_read(&ip->i_inode);
1477 gfs2_consist_inode(ip);
1478 buf = ERR_PTR(-EIO);
1482 error = gfs2_meta_inode_buffer(ip, &dibh);
1484 buf = ERR_PTR(error);
1488 buf = kzalloc(size + 1, GFP_NOFS);
1490 buf = ERR_PTR(-ENOMEM);
1492 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1495 gfs2_glock_dq_uninit(&i_gh);
1496 nd_set_link(nd, buf);
1500 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1502 char *s = nd_get_link(nd);
1510 * @mask: The mask to be tested
1511 * @flags: Indicates whether this is an RCU path walk or not
1513 * This may be called from the VFS directly, or from within GFS2 with the
1514 * inode locked, so we look to see if the glock is already locked and only
1515 * lock the glock if its not already been done.
1520 int gfs2_permission(struct inode *inode, int mask)
1522 struct gfs2_inode *ip;
1523 struct gfs2_holder i_gh;
1529 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1530 if (mask & MAY_NOT_BLOCK)
1532 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1538 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1541 error = generic_permission(inode, mask);
1543 gfs2_glock_dq_uninit(&i_gh);
1548 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1550 setattr_copy(inode, attr);
1551 mark_inode_dirty(inode);
1556 * gfs2_setattr_simple -
1563 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1567 if (current->journal_info)
1568 return __gfs2_setattr_simple(inode, attr);
1570 error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1574 error = __gfs2_setattr_simple(inode, attr);
1575 gfs2_trans_end(GFS2_SB(inode));
1579 static int setattr_chown(struct inode *inode, struct iattr *attr)
1581 struct gfs2_inode *ip = GFS2_I(inode);
1582 struct gfs2_sbd *sdp = GFS2_SB(inode);
1583 u32 ouid, ogid, nuid, ngid;
1586 ouid = inode->i_uid;
1587 ogid = inode->i_gid;
1588 nuid = attr->ia_uid;
1589 ngid = attr->ia_gid;
1591 if (!(attr->ia_valid & ATTR_UID) || ouid == nuid)
1592 ouid = nuid = NO_QUOTA_CHANGE;
1593 if (!(attr->ia_valid & ATTR_GID) || ogid == ngid)
1594 ogid = ngid = NO_QUOTA_CHANGE;
1596 error = gfs2_quota_lock(ip, nuid, ngid);
1600 if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1601 error = gfs2_quota_check(ip, nuid, ngid);
1606 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1610 error = gfs2_setattr_simple(inode, attr);
1614 if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1615 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1616 gfs2_quota_change(ip, -blocks, ouid, ogid);
1617 gfs2_quota_change(ip, blocks, nuid, ngid);
1621 gfs2_trans_end(sdp);
1623 gfs2_quota_unlock(ip);
1628 * gfs2_setattr - Change attributes on an inode
1629 * @dentry: The dentry which is changing
1630 * @attr: The structure describing the change
1632 * The VFS layer wants to change one or more of an inodes attributes. Write
1633 * that change out to disk.
1638 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1640 struct inode *inode = dentry->d_inode;
1641 struct gfs2_inode *ip = GFS2_I(inode);
1642 struct gfs2_holder i_gh;
1645 error = gfs2_rs_alloc(ip);
1649 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1654 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1657 error = inode_change_ok(inode, attr);
1661 if (attr->ia_valid & ATTR_SIZE)
1662 error = gfs2_setattr_size(inode, attr->ia_size);
1663 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1664 error = setattr_chown(inode, attr);
1665 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1666 error = gfs2_acl_chmod(ip, attr);
1668 error = gfs2_setattr_simple(inode, attr);
1672 mark_inode_dirty(inode);
1673 gfs2_glock_dq_uninit(&i_gh);
1678 * gfs2_getattr - Read out an inode's attributes
1679 * @mnt: The vfsmount the inode is being accessed from
1680 * @dentry: The dentry to stat
1681 * @stat: The inode's stats
1683 * This may be called from the VFS directly, or from within GFS2 with the
1684 * inode locked, so we look to see if the glock is already locked and only
1685 * lock the glock if its not already been done. Note that its the NFS
1686 * readdirplus operation which causes this to be called (from filldir)
1687 * with the glock already held.
1692 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1695 struct inode *inode = dentry->d_inode;
1696 struct gfs2_inode *ip = GFS2_I(inode);
1697 struct gfs2_holder gh;
1701 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1702 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1708 generic_fillattr(inode, stat);
1710 gfs2_glock_dq_uninit(&gh);
1715 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1716 const void *data, size_t size, int flags)
1718 struct inode *inode = dentry->d_inode;
1719 struct gfs2_inode *ip = GFS2_I(inode);
1720 struct gfs2_holder gh;
1723 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1724 ret = gfs2_glock_nq(&gh);
1726 ret = gfs2_rs_alloc(ip);
1728 ret = generic_setxattr(dentry, name, data, size, flags);
1731 gfs2_holder_uninit(&gh);
1735 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1736 void *data, size_t size)
1738 struct inode *inode = dentry->d_inode;
1739 struct gfs2_inode *ip = GFS2_I(inode);
1740 struct gfs2_holder gh;
1743 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1744 ret = gfs2_glock_nq(&gh);
1746 ret = generic_getxattr(dentry, name, data, size);
1749 gfs2_holder_uninit(&gh);
1753 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1755 struct inode *inode = dentry->d_inode;
1756 struct gfs2_inode *ip = GFS2_I(inode);
1757 struct gfs2_holder gh;
1760 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1761 ret = gfs2_glock_nq(&gh);
1763 ret = gfs2_rs_alloc(ip);
1765 ret = generic_removexattr(dentry, name);
1768 gfs2_holder_uninit(&gh);
1772 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1775 struct gfs2_inode *ip = GFS2_I(inode);
1776 struct gfs2_holder gh;
1779 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1783 mutex_lock(&inode->i_mutex);
1785 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1789 if (gfs2_is_stuffed(ip)) {
1790 u64 phys = ip->i_no_addr << inode->i_blkbits;
1791 u64 size = i_size_read(inode);
1792 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1793 FIEMAP_EXTENT_DATA_INLINE;
1794 phys += sizeof(struct gfs2_dinode);
1796 if (start + len > size)
1799 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1804 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1808 gfs2_glock_dq_uninit(&gh);
1810 mutex_unlock(&inode->i_mutex);
1814 const struct inode_operations gfs2_file_iops = {
1815 .permission = gfs2_permission,
1816 .setattr = gfs2_setattr,
1817 .getattr = gfs2_getattr,
1818 .setxattr = gfs2_setxattr,
1819 .getxattr = gfs2_getxattr,
1820 .listxattr = gfs2_listxattr,
1821 .removexattr = gfs2_removexattr,
1822 .fiemap = gfs2_fiemap,
1823 .get_acl = gfs2_get_acl,
1826 const struct inode_operations gfs2_dir_iops = {
1827 .create = gfs2_create,
1828 .lookup = gfs2_lookup,
1830 .unlink = gfs2_unlink,
1831 .symlink = gfs2_symlink,
1832 .mkdir = gfs2_mkdir,
1833 .rmdir = gfs2_unlink,
1834 .mknod = gfs2_mknod,
1835 .rename = gfs2_rename,
1836 .permission = gfs2_permission,
1837 .setattr = gfs2_setattr,
1838 .getattr = gfs2_getattr,
1839 .setxattr = gfs2_setxattr,
1840 .getxattr = gfs2_getxattr,
1841 .listxattr = gfs2_listxattr,
1842 .removexattr = gfs2_removexattr,
1843 .fiemap = gfs2_fiemap,
1844 .get_acl = gfs2_get_acl,
1847 const struct inode_operations gfs2_symlink_iops = {
1848 .readlink = generic_readlink,
1849 .follow_link = gfs2_follow_link,
1850 .put_link = gfs2_put_link,
1851 .permission = gfs2_permission,
1852 .setattr = gfs2_setattr,
1853 .getattr = gfs2_getattr,
1854 .setxattr = gfs2_setxattr,
1855 .getxattr = gfs2_getxattr,
1856 .listxattr = gfs2_listxattr,
1857 .removexattr = gfs2_removexattr,
1858 .fiemap = gfs2_fiemap,
1859 .get_acl = gfs2_get_acl,