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) &&
372 !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
373 if (S_ISDIR(inode->i_mode))
374 inode->i_mode |= S_ISUID;
375 else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
376 inode->i_mode &= ~07111;
377 inode->i_uid = dip->i_inode.i_uid;
379 inode->i_uid = current_fsuid();
381 if (dip->i_inode.i_mode & S_ISGID) {
382 if (S_ISDIR(inode->i_mode))
383 inode->i_mode |= S_ISGID;
384 inode->i_gid = dip->i_inode.i_gid;
386 inode->i_gid = current_fsgid();
389 static int alloc_dinode(struct gfs2_inode *ip, u32 flags)
391 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
395 error = gfs2_quota_lock_check(ip);
399 error = gfs2_inplace_reserve(ip, RES_DINODE, flags);
403 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 0);
407 error = gfs2_alloc_blocks(ip, &ip->i_no_addr, &dblocks, 1, &ip->i_generation);
408 ip->i_no_formal_ino = ip->i_generation;
409 ip->i_inode.i_ino = ip->i_no_addr;
410 ip->i_goal = ip->i_no_addr;
415 gfs2_inplace_release(ip);
417 gfs2_quota_unlock(ip);
422 static void gfs2_init_dir(struct buffer_head *dibh,
423 const struct gfs2_inode *parent)
425 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
426 struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
428 gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
429 dent->de_inum = di->di_num; /* already GFS2 endian */
430 dent->de_type = cpu_to_be16(DT_DIR);
432 dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
433 gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
434 gfs2_inum_out(parent, dent);
435 dent->de_type = cpu_to_be16(DT_DIR);
440 * init_dinode - Fill in a new dinode structure
441 * @dip: The directory this inode is being created in
443 * @symname: The symlink destination (if a symlink)
444 * @bhp: The buffer head (returned to caller)
448 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
451 struct gfs2_dinode *di;
452 struct buffer_head *dibh;
454 dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
455 gfs2_trans_add_meta(ip->i_gl, dibh);
456 di = (struct gfs2_dinode *)dibh->b_data;
457 gfs2_dinode_out(ip, di);
459 di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
460 di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
464 memset(&di->__pad4, 0, sizeof(di->__pad4));
465 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
466 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
468 switch(ip->i_inode.i_mode & S_IFMT) {
470 gfs2_init_dir(dibh, dip);
473 memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
477 set_buffer_uptodate(dibh);
481 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
482 struct gfs2_inode *ip, int arq)
484 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
488 error = gfs2_quota_lock_check(dip);
490 goto fail_quota_locks;
492 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
494 goto fail_quota_locks;
496 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
497 dip->i_rgd->rd_length +
499 RES_STATFS + RES_QUOTA, 0);
503 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
505 goto fail_quota_locks;
508 error = gfs2_dir_add(&dip->i_inode, name, ip);
515 gfs2_inplace_release(dip);
517 gfs2_quota_unlock(dip);
521 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
524 const struct xattr *xattr;
527 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
528 err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
530 GFS2_EATYPE_SECURITY);
537 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
538 const struct qstr *qstr)
540 return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
541 &gfs2_initxattrs, NULL);
545 * gfs2_create_inode - Create a new inode
546 * @dir: The parent directory
547 * @dentry: The new dentry
548 * @mode: The permissions on the new inode
549 * @dev: For device nodes, this is the device number
550 * @symname: For symlinks, this is the link destination
551 * @size: The initial size of the inode (ignored for directories)
553 * Returns: 0 on success, or error code
556 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
557 umode_t mode, dev_t dev, const char *symname,
558 unsigned int size, int excl)
560 const struct qstr *name = &dentry->d_name;
561 struct gfs2_holder ghs[2];
562 struct inode *inode = NULL;
563 struct gfs2_inode *dip = GFS2_I(dir), *ip;
564 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
565 struct gfs2_glock *io_gl;
570 if (!name->len || name->len > GFS2_FNAMESIZE)
571 return -ENAMETOOLONG;
573 error = gfs2_rs_alloc(dip);
577 error = gfs2_rindex_update(sdp);
581 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
585 error = create_ok(dip, name, mode);
586 if ((error == -EEXIST) && S_ISREG(mode) && !excl) {
587 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
588 gfs2_glock_dq_uninit(ghs);
589 d_instantiate(dentry, inode);
590 return IS_ERR(inode) ? PTR_ERR(inode) : 0;
595 arq = error = gfs2_diradd_alloc_required(dir, name);
599 inode = new_inode(sdp->sd_vfs);
605 error = gfs2_rs_alloc(ip);
607 goto fail_free_inode;
609 inode->i_mode = mode;
610 set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
612 inode->i_size = size;
613 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
614 gfs2_set_inode_blocks(inode, 1);
615 munge_mode_uid_gid(dip, inode);
616 ip->i_goal = dip->i_goal;
623 switch(mode & S_IFMT) {
625 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
626 gfs2_tune_get(sdp, gt_new_files_jdata))
627 ip->i_diskflags |= GFS2_DIF_JDATA;
628 gfs2_set_aops(inode);
631 ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
632 ip->i_diskflags |= GFS2_DIF_JDATA;
636 gfs2_set_inode_flags(inode);
638 if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
639 (dip->i_diskflags & GFS2_DIF_TOPDIR))
640 aflags |= GFS2_AF_ORLOV;
642 error = alloc_dinode(ip, aflags);
644 goto fail_free_inode;
646 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
648 goto fail_free_inode;
650 ip->i_gl->gl_object = ip;
651 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
653 goto fail_free_inode;
655 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
659 init_dinode(dip, ip, symname);
662 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
666 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
670 ip->i_iopen_gh.gh_gl->gl_object = ip;
671 gfs2_glock_put(io_gl);
673 insert_inode_hash(inode);
675 error = gfs2_acl_create(dip, inode);
679 error = gfs2_security_init(dip, ip, name);
683 error = link_dinode(dip, name, ip, arq);
687 mark_inode_dirty(inode);
688 gfs2_glock_dq_uninit(ghs);
689 gfs2_glock_dq_uninit(ghs + 1);
690 d_instantiate(dentry, inode);
694 gfs2_glock_dq_uninit(ghs + 1);
696 gfs2_glock_put(ip->i_gl);
700 gfs2_glock_dq_uninit(ghs + 1);
703 gfs2_glock_put(ip->i_gl);
705 free_inode_nonrcu(inode);
708 gfs2_glock_dq_uninit(ghs);
709 if (inode && !IS_ERR(inode)) {
711 mark_inode_dirty(inode);
712 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
720 * gfs2_create - Create a file
721 * @dir: The directory in which to create the file
722 * @dentry: The dentry of the new file
723 * @mode: The mode of the new file
728 static int gfs2_create(struct inode *dir, struct dentry *dentry,
729 umode_t mode, bool excl)
731 return gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0, excl);
735 * gfs2_lookup - Look up a filename in a directory and return its inode
736 * @dir: The directory inode
737 * @dentry: The dentry of the new inode
738 * @nd: passed from Linux VFS, ignored by us
740 * Called by the VFS layer. Lock dir and call gfs2_lookupi()
745 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
748 struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
749 if (inode && !IS_ERR(inode)) {
750 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
751 struct gfs2_holder gh;
753 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
756 return ERR_PTR(error);
758 gfs2_glock_dq_uninit(&gh);
760 return d_splice_alias(inode, dentry);
764 * gfs2_link - Link to a file
765 * @old_dentry: The inode to link
766 * @dir: Add link to this directory
767 * @dentry: The name of the link
769 * Link the inode in "old_dentry" into the directory "dir" with the
775 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
776 struct dentry *dentry)
778 struct gfs2_inode *dip = GFS2_I(dir);
779 struct gfs2_sbd *sdp = GFS2_SB(dir);
780 struct inode *inode = old_dentry->d_inode;
781 struct gfs2_inode *ip = GFS2_I(inode);
782 struct gfs2_holder ghs[2];
783 struct buffer_head *dibh;
787 if (S_ISDIR(inode->i_mode))
790 error = gfs2_rs_alloc(dip);
794 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
795 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
797 error = gfs2_glock_nq(ghs); /* parent */
801 error = gfs2_glock_nq(ghs + 1); /* child */
806 if (inode->i_nlink == 0)
809 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
813 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
824 if (!dip->i_inode.i_nlink)
827 if (dip->i_entries == (u32)-1)
830 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
833 if (!ip->i_inode.i_nlink)
836 if (ip->i_inode.i_nlink == (u32)-1)
839 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
844 if (alloc_required) {
845 error = gfs2_quota_lock_check(dip);
849 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
853 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
854 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
855 2 * RES_DINODE + RES_STATFS +
860 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
865 error = gfs2_meta_inode_buffer(ip, &dibh);
869 error = gfs2_dir_add(dir, &dentry->d_name, ip);
873 gfs2_trans_add_meta(ip->i_gl, dibh);
874 inc_nlink(&ip->i_inode);
875 ip->i_inode.i_ctime = CURRENT_TIME;
877 d_instantiate(dentry, inode);
878 mark_inode_dirty(inode);
886 gfs2_inplace_release(dip);
889 gfs2_quota_unlock(dip);
891 gfs2_glock_dq(ghs + 1);
895 gfs2_holder_uninit(ghs);
896 gfs2_holder_uninit(ghs + 1);
901 * gfs2_unlink_ok - check to see that a inode is still in a directory
902 * @dip: the directory
903 * @name: the name of the file
906 * Assumes that the lock on (at least) @dip is held.
908 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
911 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
912 const struct gfs2_inode *ip)
916 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
919 if ((dip->i_inode.i_mode & S_ISVTX) &&
920 !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
921 !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
924 if (IS_APPEND(&dip->i_inode))
927 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
931 error = gfs2_dir_check(&dip->i_inode, name, ip);
939 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
940 * @dip: The parent directory
941 * @name: The name of the entry in the parent directory
942 * @inode: The inode to be removed
944 * Called with all the locks and in a transaction. This will only be
945 * called for a directory after it has been checked to ensure it is empty.
947 * Returns: 0 on success, or an error
950 static int gfs2_unlink_inode(struct gfs2_inode *dip,
951 const struct dentry *dentry)
953 struct inode *inode = dentry->d_inode;
954 struct gfs2_inode *ip = GFS2_I(inode);
957 error = gfs2_dir_del(dip, dentry);
962 inode->i_ctime = CURRENT_TIME;
963 if (S_ISDIR(inode->i_mode))
967 mark_inode_dirty(inode);
968 if (inode->i_nlink == 0)
969 gfs2_unlink_di(inode);
975 * gfs2_unlink - Unlink an inode (this does rmdir as well)
976 * @dir: The inode of the directory containing the inode to unlink
977 * @dentry: The file itself
979 * This routine uses the type of the inode as a flag to figure out
980 * whether this is an unlink or an rmdir.
985 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
987 struct gfs2_inode *dip = GFS2_I(dir);
988 struct gfs2_sbd *sdp = GFS2_SB(dir);
989 struct inode *inode = dentry->d_inode;
990 struct gfs2_inode *ip = GFS2_I(inode);
991 struct gfs2_holder ghs[3];
992 struct gfs2_rgrpd *rgd;
995 error = gfs2_rindex_update(sdp);
1001 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1002 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1004 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1008 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1011 error = gfs2_glock_nq(ghs); /* parent */
1015 error = gfs2_glock_nq(ghs + 1); /* child */
1020 if (inode->i_nlink == 0)
1023 if (S_ISDIR(inode->i_mode)) {
1025 if (ip->i_entries > 2 || inode->i_nlink > 2)
1029 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1033 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1037 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1041 error = gfs2_unlink_inode(dip, dentry);
1044 gfs2_trans_end(sdp);
1046 gfs2_glock_dq(ghs + 2);
1048 gfs2_glock_dq(ghs + 1);
1052 gfs2_holder_uninit(ghs + 2);
1054 gfs2_holder_uninit(ghs + 1);
1055 gfs2_holder_uninit(ghs);
1060 * gfs2_symlink - Create a symlink
1061 * @dir: The directory to create the symlink in
1062 * @dentry: The dentry to put the symlink in
1063 * @symname: The thing which the link points to
1068 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1069 const char *symname)
1071 struct gfs2_sbd *sdp = GFS2_SB(dir);
1074 size = strlen(symname);
1075 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1076 return -ENAMETOOLONG;
1078 return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1082 * gfs2_mkdir - Make a directory
1083 * @dir: The parent directory of the new one
1084 * @dentry: The dentry of the new directory
1085 * @mode: The mode of the new directory
1090 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1092 struct gfs2_sbd *sdp = GFS2_SB(dir);
1093 unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1094 return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, dsize, 0);
1098 * gfs2_mknod - Make a special file
1099 * @dir: The directory in which the special file will reside
1100 * @dentry: The dentry of the special file
1101 * @mode: The mode of the special file
1102 * @dev: The device specification of the special file
1106 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1109 return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0, 0);
1113 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1117 * Follow @to back to the root and make sure we don't encounter @this
1118 * Assumes we already hold the rename lock.
1123 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1125 struct inode *dir = &to->i_inode;
1126 struct super_block *sb = dir->i_sb;
1133 if (dir == &this->i_inode) {
1137 if (dir == sb->s_root->d_inode) {
1142 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1144 error = PTR_ERR(tmp);
1158 * gfs2_rename - Rename a file
1159 * @odir: Parent directory of old file name
1160 * @odentry: The old dentry of the file
1161 * @ndir: Parent directory of new file name
1162 * @ndentry: The new dentry of the file
1167 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1168 struct inode *ndir, struct dentry *ndentry)
1170 struct gfs2_inode *odip = GFS2_I(odir);
1171 struct gfs2_inode *ndip = GFS2_I(ndir);
1172 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1173 struct gfs2_inode *nip = NULL;
1174 struct gfs2_sbd *sdp = GFS2_SB(odir);
1175 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1176 struct gfs2_rgrpd *nrgd;
1177 unsigned int num_gh;
1179 int alloc_required = 0;
1183 if (ndentry->d_inode) {
1184 nip = GFS2_I(ndentry->d_inode);
1189 error = gfs2_rindex_update(sdp);
1193 error = gfs2_rs_alloc(ndip);
1198 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1203 if (S_ISDIR(ip->i_inode.i_mode)) {
1205 /* don't move a dirctory into it's subdir */
1206 error = gfs2_ok_to_move(ip, ndip);
1213 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1215 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1218 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1222 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1224 /* grab the resource lock for unlink flag twiddling
1225 * this is the case of the target file already existing
1226 * so we unlink before doing the rename
1228 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1230 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1233 for (x = 0; x < num_gh; x++) {
1234 error = gfs2_glock_nq(ghs + x);
1240 if (ip->i_inode.i_nlink == 0)
1243 /* Check out the old directory */
1245 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1249 /* Check out the new directory */
1252 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1256 if (nip->i_inode.i_nlink == 0) {
1261 if (S_ISDIR(nip->i_inode.i_mode)) {
1262 if (nip->i_entries < 2) {
1263 gfs2_consist_inode(nip);
1267 if (nip->i_entries > 2) {
1273 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1277 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1289 if (!ndip->i_inode.i_nlink) {
1293 if (ndip->i_entries == (u32)-1) {
1297 if (S_ISDIR(ip->i_inode.i_mode) &&
1298 ndip->i_inode.i_nlink == (u32)-1) {
1305 /* Check out the dir to be renamed */
1308 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1314 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1315 error = alloc_required;
1319 if (alloc_required) {
1320 error = gfs2_quota_lock_check(ndip);
1324 error = gfs2_inplace_reserve(ndip, sdp->sd_max_dirres, 0);
1328 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1329 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1330 4 * RES_DINODE + 4 * RES_LEAF +
1331 RES_STATFS + RES_QUOTA + 4, 0);
1335 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1336 5 * RES_LEAF + 4, 0);
1341 /* Remove the target file, if it exists */
1344 error = gfs2_unlink_inode(ndip, ndentry);
1347 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1351 struct buffer_head *dibh;
1352 error = gfs2_meta_inode_buffer(ip, &dibh);
1355 ip->i_inode.i_ctime = CURRENT_TIME;
1356 gfs2_trans_add_meta(ip->i_gl, dibh);
1357 gfs2_dinode_out(ip, dibh->b_data);
1361 error = gfs2_dir_del(odip, odentry);
1365 error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1370 gfs2_trans_end(sdp);
1373 gfs2_inplace_release(ndip);
1376 gfs2_quota_unlock(ndip);
1379 gfs2_glock_dq(ghs + x);
1380 gfs2_holder_uninit(ghs + x);
1384 gfs2_glock_dq_uninit(&r_gh);
1390 * gfs2_follow_link - Follow a symbolic link
1391 * @dentry: The dentry of the link
1392 * @nd: Data that we pass to vfs_follow_link()
1394 * This can handle symlinks of any size.
1396 * Returns: 0 on success or error code
1399 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1401 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1402 struct gfs2_holder i_gh;
1403 struct buffer_head *dibh;
1408 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1409 error = gfs2_glock_nq(&i_gh);
1411 gfs2_holder_uninit(&i_gh);
1412 nd_set_link(nd, ERR_PTR(error));
1416 size = (unsigned int)i_size_read(&ip->i_inode);
1418 gfs2_consist_inode(ip);
1419 buf = ERR_PTR(-EIO);
1423 error = gfs2_meta_inode_buffer(ip, &dibh);
1425 buf = ERR_PTR(error);
1429 buf = kzalloc(size + 1, GFP_NOFS);
1431 buf = ERR_PTR(-ENOMEM);
1433 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1436 gfs2_glock_dq_uninit(&i_gh);
1437 nd_set_link(nd, buf);
1441 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1443 char *s = nd_get_link(nd);
1451 * @mask: The mask to be tested
1452 * @flags: Indicates whether this is an RCU path walk or not
1454 * This may be called from the VFS directly, or from within GFS2 with the
1455 * inode locked, so we look to see if the glock is already locked and only
1456 * lock the glock if its not already been done.
1461 int gfs2_permission(struct inode *inode, int mask)
1463 struct gfs2_inode *ip;
1464 struct gfs2_holder i_gh;
1470 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1471 if (mask & MAY_NOT_BLOCK)
1473 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1479 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1482 error = generic_permission(inode, mask);
1484 gfs2_glock_dq_uninit(&i_gh);
1489 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1491 setattr_copy(inode, attr);
1492 mark_inode_dirty(inode);
1497 * gfs2_setattr_simple -
1504 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1508 if (current->journal_info)
1509 return __gfs2_setattr_simple(inode, attr);
1511 error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1515 error = __gfs2_setattr_simple(inode, attr);
1516 gfs2_trans_end(GFS2_SB(inode));
1520 static int setattr_chown(struct inode *inode, struct iattr *attr)
1522 struct gfs2_inode *ip = GFS2_I(inode);
1523 struct gfs2_sbd *sdp = GFS2_SB(inode);
1528 ouid = inode->i_uid;
1529 ogid = inode->i_gid;
1530 nuid = attr->ia_uid;
1531 ngid = attr->ia_gid;
1533 if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1534 ouid = nuid = NO_UID_QUOTA_CHANGE;
1535 if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1536 ogid = ngid = NO_GID_QUOTA_CHANGE;
1538 error = gfs2_quota_lock(ip, nuid, ngid);
1542 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1543 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1544 error = gfs2_quota_check(ip, nuid, ngid);
1549 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1553 error = gfs2_setattr_simple(inode, attr);
1557 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1558 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1559 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1560 gfs2_quota_change(ip, -blocks, ouid, ogid);
1561 gfs2_quota_change(ip, blocks, nuid, ngid);
1565 gfs2_trans_end(sdp);
1567 gfs2_quota_unlock(ip);
1572 * gfs2_setattr - Change attributes on an inode
1573 * @dentry: The dentry which is changing
1574 * @attr: The structure describing the change
1576 * The VFS layer wants to change one or more of an inodes attributes. Write
1577 * that change out to disk.
1582 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1584 struct inode *inode = dentry->d_inode;
1585 struct gfs2_inode *ip = GFS2_I(inode);
1586 struct gfs2_holder i_gh;
1589 error = gfs2_rs_alloc(ip);
1593 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1598 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1601 error = inode_change_ok(inode, attr);
1605 if (attr->ia_valid & ATTR_SIZE)
1606 error = gfs2_setattr_size(inode, attr->ia_size);
1607 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1608 error = setattr_chown(inode, attr);
1609 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1610 error = gfs2_acl_chmod(ip, attr);
1612 error = gfs2_setattr_simple(inode, attr);
1616 mark_inode_dirty(inode);
1617 gfs2_glock_dq_uninit(&i_gh);
1622 * gfs2_getattr - Read out an inode's attributes
1623 * @mnt: The vfsmount the inode is being accessed from
1624 * @dentry: The dentry to stat
1625 * @stat: The inode's stats
1627 * This may be called from the VFS directly, or from within GFS2 with the
1628 * inode locked, so we look to see if the glock is already locked and only
1629 * lock the glock if its not already been done. Note that its the NFS
1630 * readdirplus operation which causes this to be called (from filldir)
1631 * with the glock already held.
1636 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1639 struct inode *inode = dentry->d_inode;
1640 struct gfs2_inode *ip = GFS2_I(inode);
1641 struct gfs2_holder gh;
1645 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1646 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1652 generic_fillattr(inode, stat);
1654 gfs2_glock_dq_uninit(&gh);
1659 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1660 const void *data, size_t size, int flags)
1662 struct inode *inode = dentry->d_inode;
1663 struct gfs2_inode *ip = GFS2_I(inode);
1664 struct gfs2_holder gh;
1667 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1668 ret = gfs2_glock_nq(&gh);
1670 ret = gfs2_rs_alloc(ip);
1672 ret = generic_setxattr(dentry, name, data, size, flags);
1675 gfs2_holder_uninit(&gh);
1679 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1680 void *data, size_t size)
1682 struct inode *inode = dentry->d_inode;
1683 struct gfs2_inode *ip = GFS2_I(inode);
1684 struct gfs2_holder gh;
1687 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1688 ret = gfs2_glock_nq(&gh);
1690 ret = generic_getxattr(dentry, name, data, size);
1693 gfs2_holder_uninit(&gh);
1697 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1699 struct inode *inode = dentry->d_inode;
1700 struct gfs2_inode *ip = GFS2_I(inode);
1701 struct gfs2_holder gh;
1704 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1705 ret = gfs2_glock_nq(&gh);
1707 ret = gfs2_rs_alloc(ip);
1709 ret = generic_removexattr(dentry, name);
1712 gfs2_holder_uninit(&gh);
1716 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1719 struct gfs2_inode *ip = GFS2_I(inode);
1720 struct gfs2_holder gh;
1723 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1727 mutex_lock(&inode->i_mutex);
1729 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1733 if (gfs2_is_stuffed(ip)) {
1734 u64 phys = ip->i_no_addr << inode->i_blkbits;
1735 u64 size = i_size_read(inode);
1736 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1737 FIEMAP_EXTENT_DATA_INLINE;
1738 phys += sizeof(struct gfs2_dinode);
1740 if (start + len > size)
1743 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1748 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1752 gfs2_glock_dq_uninit(&gh);
1754 mutex_unlock(&inode->i_mutex);
1758 const struct inode_operations gfs2_file_iops = {
1759 .permission = gfs2_permission,
1760 .setattr = gfs2_setattr,
1761 .getattr = gfs2_getattr,
1762 .setxattr = gfs2_setxattr,
1763 .getxattr = gfs2_getxattr,
1764 .listxattr = gfs2_listxattr,
1765 .removexattr = gfs2_removexattr,
1766 .fiemap = gfs2_fiemap,
1767 .get_acl = gfs2_get_acl,
1770 const struct inode_operations gfs2_dir_iops = {
1771 .create = gfs2_create,
1772 .lookup = gfs2_lookup,
1774 .unlink = gfs2_unlink,
1775 .symlink = gfs2_symlink,
1776 .mkdir = gfs2_mkdir,
1777 .rmdir = gfs2_unlink,
1778 .mknod = gfs2_mknod,
1779 .rename = gfs2_rename,
1780 .permission = gfs2_permission,
1781 .setattr = gfs2_setattr,
1782 .getattr = gfs2_getattr,
1783 .setxattr = gfs2_setxattr,
1784 .getxattr = gfs2_getxattr,
1785 .listxattr = gfs2_listxattr,
1786 .removexattr = gfs2_removexattr,
1787 .fiemap = gfs2_fiemap,
1788 .get_acl = gfs2_get_acl,
1791 const struct inode_operations gfs2_symlink_iops = {
1792 .readlink = generic_readlink,
1793 .follow_link = gfs2_follow_link,
1794 .put_link = gfs2_put_link,
1795 .permission = gfs2_permission,
1796 .setattr = gfs2_setattr,
1797 .getattr = gfs2_getattr,
1798 .setxattr = gfs2_setxattr,
1799 .getxattr = gfs2_getxattr,
1800 .listxattr = gfs2_listxattr,
1801 .removexattr = gfs2_removexattr,
1802 .fiemap = gfs2_fiemap,
1803 .get_acl = gfs2_get_acl,