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(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 mark_inode_dirty(inode);
733 gfs2_glock_dq_uninit_m(2, ghs);
734 d_instantiate(dentry, inode);
738 gfs2_glock_dq_uninit(ghs + 1);
740 gfs2_glock_dq_uninit(ghs);
741 if (inode && !IS_ERR(inode)) {
742 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
752 * gfs2_create - Create a file
753 * @dir: The directory in which to create the file
754 * @dentry: The dentry of the new file
755 * @mode: The mode of the new file
760 static int gfs2_create(struct inode *dir, struct dentry *dentry,
761 int mode, struct nameidata *nd)
767 ret = gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0);
768 if (ret != -EEXIST || (nd && (nd->flags & LOOKUP_EXCL)))
771 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
775 return PTR_ERR(inode);
779 d_instantiate(dentry, inode);
784 * gfs2_lookup - Look up a filename in a directory and return its inode
785 * @dir: The directory inode
786 * @dentry: The dentry of the new inode
787 * @nd: passed from Linux VFS, ignored by us
789 * Called by the VFS layer. Lock dir and call gfs2_lookupi()
794 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
795 struct nameidata *nd)
797 struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
798 if (inode && !IS_ERR(inode)) {
799 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
800 struct gfs2_holder gh;
802 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
805 return ERR_PTR(error);
807 gfs2_glock_dq_uninit(&gh);
809 return d_splice_alias(inode, dentry);
813 * gfs2_link - Link to a file
814 * @old_dentry: The inode to link
815 * @dir: Add link to this directory
816 * @dentry: The name of the link
818 * Link the inode in "old_dentry" into the directory "dir" with the
824 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
825 struct dentry *dentry)
827 struct gfs2_inode *dip = GFS2_I(dir);
828 struct gfs2_sbd *sdp = GFS2_SB(dir);
829 struct inode *inode = old_dentry->d_inode;
830 struct gfs2_inode *ip = GFS2_I(inode);
831 struct gfs2_holder ghs[2];
832 struct buffer_head *dibh;
836 if (S_ISDIR(inode->i_mode))
839 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
840 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
842 error = gfs2_glock_nq(ghs); /* parent */
846 error = gfs2_glock_nq(ghs + 1); /* child */
851 if (inode->i_nlink == 0)
854 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
858 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
869 if (!dip->i_inode.i_nlink)
872 if (dip->i_entries == (u32)-1)
875 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
878 if (!ip->i_inode.i_nlink)
881 if (ip->i_inode.i_nlink == (u32)-1)
884 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
889 if (alloc_required) {
890 struct gfs2_alloc *al = gfs2_alloc_get(dip);
896 error = gfs2_quota_lock_check(dip);
900 al->al_requested = sdp->sd_max_dirres;
902 error = gfs2_inplace_reserve(dip);
906 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
908 2 * RES_DINODE + RES_STATFS +
913 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
918 error = gfs2_meta_inode_buffer(ip, &dibh);
922 error = gfs2_dir_add(dir, &dentry->d_name, ip);
926 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
927 inc_nlink(&ip->i_inode);
928 ip->i_inode.i_ctime = CURRENT_TIME;
930 d_instantiate(dentry, inode);
931 mark_inode_dirty(inode);
939 gfs2_inplace_release(dip);
942 gfs2_quota_unlock(dip);
947 gfs2_glock_dq(ghs + 1);
951 gfs2_holder_uninit(ghs);
952 gfs2_holder_uninit(ghs + 1);
957 * gfs2_unlink_ok - check to see that a inode is still in a directory
958 * @dip: the directory
959 * @name: the name of the file
962 * Assumes that the lock on (at least) @dip is held.
964 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
967 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
968 const struct gfs2_inode *ip)
972 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
975 if ((dip->i_inode.i_mode & S_ISVTX) &&
976 dip->i_inode.i_uid != current_fsuid() &&
977 ip->i_inode.i_uid != current_fsuid() && !capable(CAP_FOWNER))
980 if (IS_APPEND(&dip->i_inode))
983 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
987 error = gfs2_dir_check(&dip->i_inode, name, ip);
995 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
996 * @dip: The parent directory
997 * @name: The name of the entry in the parent directory
998 * @bh: The inode buffer for the inode to be removed
999 * @inode: The inode to be removed
1001 * Called with all the locks and in a transaction. This will only be
1002 * called for a directory after it has been checked to ensure it is empty.
1004 * Returns: 0 on success, or an error
1007 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1008 const struct dentry *dentry,
1009 struct buffer_head *bh)
1011 struct inode *inode = dentry->d_inode;
1012 struct gfs2_inode *ip = GFS2_I(inode);
1015 error = gfs2_dir_del(dip, dentry);
1020 inode->i_ctime = CURRENT_TIME;
1021 if (S_ISDIR(inode->i_mode))
1025 mark_inode_dirty(inode);
1026 if (inode->i_nlink == 0)
1027 gfs2_unlink_di(inode);
1033 * gfs2_unlink - Unlink an inode (this does rmdir as well)
1034 * @dir: The inode of the directory containing the inode to unlink
1035 * @dentry: The file itself
1037 * This routine uses the type of the inode as a flag to figure out
1038 * whether this is an unlink or an rmdir.
1043 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1045 struct gfs2_inode *dip = GFS2_I(dir);
1046 struct gfs2_sbd *sdp = GFS2_SB(dir);
1047 struct inode *inode = dentry->d_inode;
1048 struct gfs2_inode *ip = GFS2_I(inode);
1049 struct buffer_head *bh;
1050 struct gfs2_holder ghs[3];
1051 struct gfs2_rgrpd *rgd;
1052 struct gfs2_holder ri_gh;
1055 error = gfs2_rindex_hold(sdp, &ri_gh);
1059 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1060 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1062 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1063 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1066 error = gfs2_glock_nq(ghs); /* parent */
1070 error = gfs2_glock_nq(ghs + 1); /* child */
1075 if (inode->i_nlink == 0)
1078 if (S_ISDIR(inode->i_mode)) {
1080 if (ip->i_entries > 2 || inode->i_nlink > 2)
1084 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1088 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1092 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1096 error = gfs2_meta_inode_buffer(ip, &bh);
1100 error = gfs2_unlink_inode(dip, dentry, bh);
1104 gfs2_trans_end(sdp);
1106 gfs2_glock_dq(ghs + 2);
1108 gfs2_holder_uninit(ghs + 2);
1109 gfs2_glock_dq(ghs + 1);
1111 gfs2_holder_uninit(ghs + 1);
1114 gfs2_holder_uninit(ghs);
1115 gfs2_glock_dq_uninit(&ri_gh);
1120 * gfs2_symlink - Create a symlink
1121 * @dir: The directory to create the symlink in
1122 * @dentry: The dentry to put the symlink in
1123 * @symname: The thing which the link points to
1128 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1129 const char *symname)
1131 struct gfs2_sbd *sdp = GFS2_SB(dir);
1134 size = strlen(symname);
1135 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1136 return -ENAMETOOLONG;
1138 return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size);
1142 * gfs2_mkdir - Make a directory
1143 * @dir: The parent directory of the new one
1144 * @dentry: The dentry of the new directory
1145 * @mode: The mode of the new directory
1150 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1152 return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0);
1156 * gfs2_mknod - Make a special file
1157 * @dir: The directory in which the special file will reside
1158 * @dentry: The dentry of the special file
1159 * @mode: The mode of the special file
1160 * @dev: The device specification of the special file
1164 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, int mode,
1167 return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0);
1171 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1175 * Follow @to back to the root and make sure we don't encounter @this
1176 * Assumes we already hold the rename lock.
1181 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1183 struct inode *dir = &to->i_inode;
1184 struct super_block *sb = dir->i_sb;
1191 if (dir == &this->i_inode) {
1195 if (dir == sb->s_root->d_inode) {
1200 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1202 error = PTR_ERR(tmp);
1216 * gfs2_rename - Rename a file
1217 * @odir: Parent directory of old file name
1218 * @odentry: The old dentry of the file
1219 * @ndir: Parent directory of new file name
1220 * @ndentry: The new dentry of the file
1225 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1226 struct inode *ndir, struct dentry *ndentry)
1228 struct gfs2_inode *odip = GFS2_I(odir);
1229 struct gfs2_inode *ndip = GFS2_I(ndir);
1230 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1231 struct gfs2_inode *nip = NULL;
1232 struct gfs2_sbd *sdp = GFS2_SB(odir);
1233 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, }, ri_gh;
1234 struct gfs2_rgrpd *nrgd;
1235 unsigned int num_gh;
1237 int alloc_required = 0;
1241 if (ndentry->d_inode) {
1242 nip = GFS2_I(ndentry->d_inode);
1247 error = gfs2_rindex_hold(sdp, &ri_gh);
1252 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1257 if (S_ISDIR(ip->i_inode.i_mode)) {
1259 /* don't move a dirctory into it's subdir */
1260 error = gfs2_ok_to_move(ip, ndip);
1267 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1269 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1272 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1276 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1278 /* grab the resource lock for unlink flag twiddling
1279 * this is the case of the target file already existing
1280 * so we unlink before doing the rename
1282 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr);
1284 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1287 for (x = 0; x < num_gh; x++) {
1288 error = gfs2_glock_nq(ghs + x);
1294 if (ip->i_inode.i_nlink == 0)
1297 /* Check out the old directory */
1299 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1303 /* Check out the new directory */
1306 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1310 if (nip->i_inode.i_nlink == 0) {
1315 if (S_ISDIR(nip->i_inode.i_mode)) {
1316 if (nip->i_entries < 2) {
1317 gfs2_consist_inode(nip);
1321 if (nip->i_entries > 2) {
1327 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1331 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1343 if (!ndip->i_inode.i_nlink) {
1347 if (ndip->i_entries == (u32)-1) {
1351 if (S_ISDIR(ip->i_inode.i_mode) &&
1352 ndip->i_inode.i_nlink == (u32)-1) {
1359 /* Check out the dir to be renamed */
1362 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1368 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1369 error = alloc_required;
1374 if (alloc_required) {
1375 struct gfs2_alloc *al = gfs2_alloc_get(ndip);
1381 error = gfs2_quota_lock_check(ndip);
1385 al->al_requested = sdp->sd_max_dirres;
1387 error = gfs2_inplace_reserve_ri(ndip);
1391 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1392 gfs2_rg_blocks(al) +
1393 4 * RES_DINODE + 4 * RES_LEAF +
1394 RES_STATFS + RES_QUOTA + 4, 0);
1398 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1399 5 * RES_LEAF + 4, 0);
1404 /* Remove the target file, if it exists */
1407 struct buffer_head *bh;
1408 error = gfs2_meta_inode_buffer(nip, &bh);
1411 error = gfs2_unlink_inode(ndip, ndentry, bh);
1416 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1420 struct buffer_head *dibh;
1421 error = gfs2_meta_inode_buffer(ip, &dibh);
1424 ip->i_inode.i_ctime = CURRENT_TIME;
1425 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1426 gfs2_dinode_out(ip, dibh->b_data);
1430 error = gfs2_dir_del(odip, odentry);
1434 error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1439 gfs2_trans_end(sdp);
1442 gfs2_inplace_release(ndip);
1445 gfs2_quota_unlock(ndip);
1448 gfs2_alloc_put(ndip);
1451 gfs2_glock_dq(ghs + x);
1452 gfs2_holder_uninit(ghs + x);
1456 gfs2_glock_dq_uninit(&r_gh);
1458 gfs2_glock_dq_uninit(&ri_gh);
1463 * gfs2_follow_link - Follow a symbolic link
1464 * @dentry: The dentry of the link
1465 * @nd: Data that we pass to vfs_follow_link()
1467 * This can handle symlinks of any size.
1469 * Returns: 0 on success or error code
1472 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1474 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1475 struct gfs2_holder i_gh;
1476 struct buffer_head *dibh;
1481 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1482 error = gfs2_glock_nq(&i_gh);
1484 gfs2_holder_uninit(&i_gh);
1485 nd_set_link(nd, ERR_PTR(error));
1489 size = (unsigned int)i_size_read(&ip->i_inode);
1491 gfs2_consist_inode(ip);
1492 buf = ERR_PTR(-EIO);
1496 error = gfs2_meta_inode_buffer(ip, &dibh);
1498 buf = ERR_PTR(error);
1502 buf = kzalloc(size + 1, GFP_NOFS);
1504 buf = ERR_PTR(-ENOMEM);
1506 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1509 gfs2_glock_dq_uninit(&i_gh);
1510 nd_set_link(nd, buf);
1514 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1516 char *s = nd_get_link(nd);
1524 * @mask: The mask to be tested
1525 * @flags: Indicates whether this is an RCU path walk or not
1527 * This may be called from the VFS directly, or from within GFS2 with the
1528 * inode locked, so we look to see if the glock is already locked and only
1529 * lock the glock if its not already been done.
1534 int gfs2_permission(struct inode *inode, int mask)
1536 struct gfs2_inode *ip;
1537 struct gfs2_holder i_gh;
1543 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1544 if (mask & MAY_NOT_BLOCK)
1546 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1552 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1555 error = generic_permission(inode, mask);
1557 gfs2_glock_dq_uninit(&i_gh);
1562 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1564 setattr_copy(inode, attr);
1565 mark_inode_dirty(inode);
1570 * gfs2_setattr_simple -
1577 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1581 if (current->journal_info)
1582 return __gfs2_setattr_simple(inode, attr);
1584 error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1588 error = __gfs2_setattr_simple(inode, attr);
1589 gfs2_trans_end(GFS2_SB(inode));
1593 static int setattr_chown(struct inode *inode, struct iattr *attr)
1595 struct gfs2_inode *ip = GFS2_I(inode);
1596 struct gfs2_sbd *sdp = GFS2_SB(inode);
1597 u32 ouid, ogid, nuid, ngid;
1600 ouid = inode->i_uid;
1601 ogid = inode->i_gid;
1602 nuid = attr->ia_uid;
1603 ngid = attr->ia_gid;
1605 if (!(attr->ia_valid & ATTR_UID) || ouid == nuid)
1606 ouid = nuid = NO_QUOTA_CHANGE;
1607 if (!(attr->ia_valid & ATTR_GID) || ogid == ngid)
1608 ogid = ngid = NO_QUOTA_CHANGE;
1610 if (!gfs2_alloc_get(ip))
1613 error = gfs2_quota_lock(ip, nuid, ngid);
1617 if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1618 error = gfs2_quota_check(ip, nuid, ngid);
1623 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1627 error = gfs2_setattr_simple(inode, attr);
1631 if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1632 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1633 gfs2_quota_change(ip, -blocks, ouid, ogid);
1634 gfs2_quota_change(ip, blocks, nuid, ngid);
1638 gfs2_trans_end(sdp);
1640 gfs2_quota_unlock(ip);
1647 * gfs2_setattr - Change attributes on an inode
1648 * @dentry: The dentry which is changing
1649 * @attr: The structure describing the change
1651 * The VFS layer wants to change one or more of an inodes attributes. Write
1652 * that change out to disk.
1657 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1659 struct inode *inode = dentry->d_inode;
1660 struct gfs2_inode *ip = GFS2_I(inode);
1661 struct gfs2_holder i_gh;
1664 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1669 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1672 error = inode_change_ok(inode, attr);
1676 if (attr->ia_valid & ATTR_SIZE)
1677 error = gfs2_setattr_size(inode, attr->ia_size);
1678 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1679 error = setattr_chown(inode, attr);
1680 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1681 error = gfs2_acl_chmod(ip, attr);
1683 error = gfs2_setattr_simple(inode, attr);
1687 mark_inode_dirty(inode);
1688 gfs2_glock_dq_uninit(&i_gh);
1693 * gfs2_getattr - Read out an inode's attributes
1694 * @mnt: The vfsmount the inode is being accessed from
1695 * @dentry: The dentry to stat
1696 * @stat: The inode's stats
1698 * This may be called from the VFS directly, or from within GFS2 with the
1699 * inode locked, so we look to see if the glock is already locked and only
1700 * lock the glock if its not already been done. Note that its the NFS
1701 * readdirplus operation which causes this to be called (from filldir)
1702 * with the glock already held.
1707 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1710 struct inode *inode = dentry->d_inode;
1711 struct gfs2_inode *ip = GFS2_I(inode);
1712 struct gfs2_holder gh;
1716 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1717 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1723 generic_fillattr(inode, stat);
1725 gfs2_glock_dq_uninit(&gh);
1730 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1731 const void *data, size_t size, int flags)
1733 struct inode *inode = dentry->d_inode;
1734 struct gfs2_inode *ip = GFS2_I(inode);
1735 struct gfs2_holder gh;
1738 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1739 ret = gfs2_glock_nq(&gh);
1741 ret = generic_setxattr(dentry, name, data, size, flags);
1744 gfs2_holder_uninit(&gh);
1748 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1749 void *data, size_t size)
1751 struct inode *inode = dentry->d_inode;
1752 struct gfs2_inode *ip = GFS2_I(inode);
1753 struct gfs2_holder gh;
1756 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1757 ret = gfs2_glock_nq(&gh);
1759 ret = generic_getxattr(dentry, name, data, size);
1762 gfs2_holder_uninit(&gh);
1766 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1768 struct inode *inode = dentry->d_inode;
1769 struct gfs2_inode *ip = GFS2_I(inode);
1770 struct gfs2_holder gh;
1773 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1774 ret = gfs2_glock_nq(&gh);
1776 ret = generic_removexattr(dentry, name);
1779 gfs2_holder_uninit(&gh);
1783 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1786 struct gfs2_inode *ip = GFS2_I(inode);
1787 struct gfs2_holder gh;
1790 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1794 mutex_lock(&inode->i_mutex);
1796 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1800 if (gfs2_is_stuffed(ip)) {
1801 u64 phys = ip->i_no_addr << inode->i_blkbits;
1802 u64 size = i_size_read(inode);
1803 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1804 FIEMAP_EXTENT_DATA_INLINE;
1805 phys += sizeof(struct gfs2_dinode);
1807 if (start + len > size)
1810 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1815 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1819 gfs2_glock_dq_uninit(&gh);
1821 mutex_unlock(&inode->i_mutex);
1825 const struct inode_operations gfs2_file_iops = {
1826 .permission = gfs2_permission,
1827 .setattr = gfs2_setattr,
1828 .getattr = gfs2_getattr,
1829 .setxattr = gfs2_setxattr,
1830 .getxattr = gfs2_getxattr,
1831 .listxattr = gfs2_listxattr,
1832 .removexattr = gfs2_removexattr,
1833 .fiemap = gfs2_fiemap,
1834 .get_acl = gfs2_get_acl,
1837 const struct inode_operations gfs2_dir_iops = {
1838 .create = gfs2_create,
1839 .lookup = gfs2_lookup,
1841 .unlink = gfs2_unlink,
1842 .symlink = gfs2_symlink,
1843 .mkdir = gfs2_mkdir,
1844 .rmdir = gfs2_unlink,
1845 .mknod = gfs2_mknod,
1846 .rename = gfs2_rename,
1847 .permission = gfs2_permission,
1848 .setattr = gfs2_setattr,
1849 .getattr = gfs2_getattr,
1850 .setxattr = gfs2_setxattr,
1851 .getxattr = gfs2_getxattr,
1852 .listxattr = gfs2_listxattr,
1853 .removexattr = gfs2_removexattr,
1854 .fiemap = gfs2_fiemap,
1855 .get_acl = gfs2_get_acl,
1858 const struct inode_operations gfs2_symlink_iops = {
1859 .readlink = generic_readlink,
1860 .follow_link = gfs2_follow_link,
1861 .put_link = gfs2_put_link,
1862 .permission = gfs2_permission,
1863 .setattr = gfs2_setattr,
1864 .getattr = gfs2_getattr,
1865 .setxattr = gfs2_setxattr,
1866 .getxattr = gfs2_getxattr,
1867 .listxattr = gfs2_listxattr,
1868 .removexattr = gfs2_removexattr,
1869 .fiemap = gfs2_fiemap,
1870 .get_acl = gfs2_get_acl,