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_flags |= GL_NOCACHE;
193 ip->i_iopen_gh.gh_gl->gl_object = NULL;
194 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
197 gfs2_glock_put(io_gl);
199 ip->i_gl->gl_object = NULL;
200 gfs2_glock_put(ip->i_gl);
203 return ERR_PTR(error);
206 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
207 u64 *no_formal_ino, unsigned int blktype)
209 struct super_block *sb = sdp->sd_vfs;
210 struct gfs2_holder i_gh;
211 struct inode *inode = NULL;
214 /* Must not read in block until block type is verified */
215 error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
216 LM_ST_EXCLUSIVE, GL_SKIP, &i_gh);
218 return ERR_PTR(error);
220 error = gfs2_check_blk_type(sdp, no_addr, blktype);
224 inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
228 /* Two extra checks for NFS only */
231 if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
235 if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
242 gfs2_glock_dq_uninit(&i_gh);
243 return error ? ERR_PTR(error) : inode;
250 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
254 gfs2_str2qstr(&qstr, name);
255 inode = gfs2_lookupi(dip, &qstr, 1);
256 /* gfs2_lookupi has inconsistent callers: vfs
257 * related routines expect NULL for no entry found,
258 * gfs2_lookup_simple callers expect ENOENT
259 * and do not check for NULL.
262 return ERR_PTR(-ENOENT);
269 * gfs2_lookupi - Look up a filename in a directory and return its inode
270 * @d_gh: An initialized holder for the directory glock
271 * @name: The name of the inode to look for
272 * @is_root: If 1, ignore the caller's permissions
273 * @i_gh: An uninitialized holder for the new inode glock
275 * This can be called via the VFS filldir function when NFS is doing
276 * a readdirplus and the inode which its intending to stat isn't
277 * already in cache. In this case we must not take the directory glock
278 * again, since the readdir call will have already taken that lock.
283 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
286 struct super_block *sb = dir->i_sb;
287 struct gfs2_inode *dip = GFS2_I(dir);
288 struct gfs2_holder d_gh;
290 struct inode *inode = NULL;
293 if (!name->len || name->len > GFS2_FNAMESIZE)
294 return ERR_PTR(-ENAMETOOLONG);
296 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
297 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
298 dir == sb->s_root->d_inode)) {
303 if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
304 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
306 return ERR_PTR(error);
311 error = gfs2_permission(dir, MAY_EXEC);
316 inode = gfs2_dir_search(dir, name);
318 error = PTR_ERR(inode);
321 gfs2_glock_dq_uninit(&d_gh);
322 if (error == -ENOENT)
324 return inode ? inode : ERR_PTR(error);
328 * create_ok - OK to create a new on-disk inode here?
329 * @dip: Directory in which dinode is to be created
330 * @name: Name of new dinode
336 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
341 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
345 /* Don't create entries in an unlinked directory */
346 if (!dip->i_inode.i_nlink)
349 error = gfs2_dir_check(&dip->i_inode, name, NULL);
360 if (dip->i_entries == (u32)-1)
362 if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
368 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
371 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
372 (dip->i_inode.i_mode & S_ISUID) &&
373 !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
374 if (S_ISDIR(inode->i_mode))
375 inode->i_mode |= S_ISUID;
376 else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
377 inode->i_mode &= ~07111;
378 inode->i_uid = dip->i_inode.i_uid;
380 inode->i_uid = current_fsuid();
382 if (dip->i_inode.i_mode & S_ISGID) {
383 if (S_ISDIR(inode->i_mode))
384 inode->i_mode |= S_ISGID;
385 inode->i_gid = dip->i_inode.i_gid;
387 inode->i_gid = current_fsgid();
390 static int alloc_dinode(struct gfs2_inode *ip, u32 flags)
392 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
396 error = gfs2_quota_lock_check(ip);
400 error = gfs2_inplace_reserve(ip, RES_DINODE, flags);
404 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 0);
408 error = gfs2_alloc_blocks(ip, &ip->i_no_addr, &dblocks, 1, &ip->i_generation);
409 ip->i_no_formal_ino = ip->i_generation;
410 ip->i_inode.i_ino = ip->i_no_addr;
411 ip->i_goal = ip->i_no_addr;
416 gfs2_inplace_release(ip);
418 gfs2_quota_unlock(ip);
423 static void gfs2_init_dir(struct buffer_head *dibh,
424 const struct gfs2_inode *parent)
426 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
427 struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
429 gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
430 dent->de_inum = di->di_num; /* already GFS2 endian */
431 dent->de_type = cpu_to_be16(DT_DIR);
433 dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
434 gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
435 gfs2_inum_out(parent, dent);
436 dent->de_type = cpu_to_be16(DT_DIR);
441 * init_dinode - Fill in a new dinode structure
442 * @dip: The directory this inode is being created in
444 * @symname: The symlink destination (if a symlink)
445 * @bhp: The buffer head (returned to caller)
449 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
452 struct gfs2_dinode *di;
453 struct buffer_head *dibh;
455 dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
456 gfs2_trans_add_meta(ip->i_gl, dibh);
457 di = (struct gfs2_dinode *)dibh->b_data;
458 gfs2_dinode_out(ip, di);
460 di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
461 di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
465 memset(&di->__pad4, 0, sizeof(di->__pad4));
466 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
467 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
469 switch(ip->i_inode.i_mode & S_IFMT) {
471 gfs2_init_dir(dibh, dip);
474 memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
478 set_buffer_uptodate(dibh);
482 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
483 struct gfs2_inode *ip, int arq)
485 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
489 error = gfs2_quota_lock_check(dip);
491 goto fail_quota_locks;
493 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
495 goto fail_quota_locks;
497 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
498 dip->i_rgd->rd_length +
500 RES_STATFS + RES_QUOTA, 0);
504 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
506 goto fail_quota_locks;
509 error = gfs2_dir_add(&dip->i_inode, name, ip);
516 gfs2_inplace_release(dip);
518 gfs2_quota_unlock(dip);
522 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
525 const struct xattr *xattr;
528 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
529 err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
531 GFS2_EATYPE_SECURITY);
538 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
539 const struct qstr *qstr)
541 return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
542 &gfs2_initxattrs, NULL);
546 * gfs2_create_inode - Create a new inode
547 * @dir: The parent directory
548 * @dentry: The new dentry
549 * @mode: The permissions on the new inode
550 * @dev: For device nodes, this is the device number
551 * @symname: For symlinks, this is the link destination
552 * @size: The initial size of the inode (ignored for directories)
554 * Returns: 0 on success, or error code
557 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
558 umode_t mode, dev_t dev, const char *symname,
559 unsigned int size, int excl)
561 const struct qstr *name = &dentry->d_name;
562 struct gfs2_holder ghs[2];
563 struct inode *inode = NULL;
564 struct gfs2_inode *dip = GFS2_I(dir), *ip;
565 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
566 struct gfs2_glock *io_gl;
571 if (!name->len || name->len > GFS2_FNAMESIZE)
572 return -ENAMETOOLONG;
574 error = gfs2_rs_alloc(dip);
578 error = gfs2_rindex_update(sdp);
582 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
586 error = create_ok(dip, name, mode);
587 if ((error == -EEXIST) && S_ISREG(mode) && !excl) {
588 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
589 gfs2_glock_dq_uninit(ghs);
590 d_instantiate(dentry, inode);
591 return IS_ERR(inode) ? PTR_ERR(inode) : 0;
596 arq = error = gfs2_diradd_alloc_required(dir, name);
600 inode = new_inode(sdp->sd_vfs);
606 error = gfs2_rs_alloc(ip);
608 goto fail_free_inode;
610 inode->i_mode = mode;
611 set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
613 inode->i_size = size;
614 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
615 gfs2_set_inode_blocks(inode, 1);
616 munge_mode_uid_gid(dip, inode);
617 ip->i_goal = dip->i_goal;
624 switch(mode & S_IFMT) {
626 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
627 gfs2_tune_get(sdp, gt_new_files_jdata))
628 ip->i_diskflags |= GFS2_DIF_JDATA;
629 gfs2_set_aops(inode);
632 ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
633 ip->i_diskflags |= GFS2_DIF_JDATA;
637 gfs2_set_inode_flags(inode);
639 if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
640 (dip->i_diskflags & GFS2_DIF_TOPDIR))
641 aflags |= GFS2_AF_ORLOV;
643 error = alloc_dinode(ip, aflags);
645 goto fail_free_inode;
647 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
649 goto fail_free_inode;
651 ip->i_gl->gl_object = ip;
652 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
654 goto fail_free_inode;
656 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
660 init_dinode(dip, ip, symname);
663 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
667 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
671 ip->i_iopen_gh.gh_gl->gl_object = ip;
672 gfs2_glock_put(io_gl);
674 insert_inode_hash(inode);
676 error = gfs2_acl_create(dip, inode);
680 error = gfs2_security_init(dip, ip, name);
684 error = link_dinode(dip, name, ip, arq);
688 mark_inode_dirty(inode);
689 gfs2_glock_dq_uninit(ghs);
690 gfs2_glock_dq_uninit(ghs + 1);
691 d_instantiate(dentry, inode);
695 gfs2_glock_dq_uninit(ghs + 1);
697 gfs2_glock_put(ip->i_gl);
701 gfs2_glock_dq_uninit(ghs + 1);
704 gfs2_glock_put(ip->i_gl);
706 free_inode_nonrcu(inode);
709 gfs2_glock_dq_uninit(ghs);
710 if (inode && !IS_ERR(inode)) {
712 mark_inode_dirty(inode);
713 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
721 * gfs2_create - Create a file
722 * @dir: The directory in which to create the file
723 * @dentry: The dentry of the new file
724 * @mode: The mode of the new file
729 static int gfs2_create(struct inode *dir, struct dentry *dentry,
730 umode_t mode, bool excl)
732 return gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0, excl);
736 * gfs2_lookup - Look up a filename in a directory and return its inode
737 * @dir: The directory inode
738 * @dentry: The dentry of the new inode
739 * @nd: passed from Linux VFS, ignored by us
741 * Called by the VFS layer. Lock dir and call gfs2_lookupi()
746 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
749 struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
750 if (inode && !IS_ERR(inode)) {
751 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
752 struct gfs2_holder gh;
754 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
757 return ERR_PTR(error);
759 gfs2_glock_dq_uninit(&gh);
761 return d_splice_alias(inode, dentry);
765 * gfs2_link - Link to a file
766 * @old_dentry: The inode to link
767 * @dir: Add link to this directory
768 * @dentry: The name of the link
770 * Link the inode in "old_dentry" into the directory "dir" with the
776 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
777 struct dentry *dentry)
779 struct gfs2_inode *dip = GFS2_I(dir);
780 struct gfs2_sbd *sdp = GFS2_SB(dir);
781 struct inode *inode = old_dentry->d_inode;
782 struct gfs2_inode *ip = GFS2_I(inode);
783 struct gfs2_holder ghs[2];
784 struct buffer_head *dibh;
788 if (S_ISDIR(inode->i_mode))
791 error = gfs2_rs_alloc(dip);
795 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
796 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
798 error = gfs2_glock_nq(ghs); /* parent */
802 error = gfs2_glock_nq(ghs + 1); /* child */
807 if (inode->i_nlink == 0)
810 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
814 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
825 if (!dip->i_inode.i_nlink)
828 if (dip->i_entries == (u32)-1)
831 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
834 if (!ip->i_inode.i_nlink)
837 if (ip->i_inode.i_nlink == (u32)-1)
840 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
845 if (alloc_required) {
846 error = gfs2_quota_lock_check(dip);
850 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
854 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
855 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
856 2 * RES_DINODE + RES_STATFS +
861 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
866 error = gfs2_meta_inode_buffer(ip, &dibh);
870 error = gfs2_dir_add(dir, &dentry->d_name, ip);
874 gfs2_trans_add_meta(ip->i_gl, dibh);
875 inc_nlink(&ip->i_inode);
876 ip->i_inode.i_ctime = CURRENT_TIME;
878 d_instantiate(dentry, inode);
879 mark_inode_dirty(inode);
887 gfs2_inplace_release(dip);
890 gfs2_quota_unlock(dip);
892 gfs2_glock_dq(ghs + 1);
896 gfs2_holder_uninit(ghs);
897 gfs2_holder_uninit(ghs + 1);
902 * gfs2_unlink_ok - check to see that a inode is still in a directory
903 * @dip: the directory
904 * @name: the name of the file
907 * Assumes that the lock on (at least) @dip is held.
909 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
912 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
913 const struct gfs2_inode *ip)
917 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
920 if ((dip->i_inode.i_mode & S_ISVTX) &&
921 !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
922 !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
925 if (IS_APPEND(&dip->i_inode))
928 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
932 error = gfs2_dir_check(&dip->i_inode, name, ip);
940 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
941 * @dip: The parent directory
942 * @name: The name of the entry in the parent directory
943 * @inode: The inode to be removed
945 * Called with all the locks and in a transaction. This will only be
946 * called for a directory after it has been checked to ensure it is empty.
948 * Returns: 0 on success, or an error
951 static int gfs2_unlink_inode(struct gfs2_inode *dip,
952 const struct dentry *dentry)
954 struct inode *inode = dentry->d_inode;
955 struct gfs2_inode *ip = GFS2_I(inode);
958 error = gfs2_dir_del(dip, dentry);
963 inode->i_ctime = CURRENT_TIME;
964 if (S_ISDIR(inode->i_mode))
968 mark_inode_dirty(inode);
969 if (inode->i_nlink == 0)
970 gfs2_unlink_di(inode);
976 * gfs2_unlink - Unlink an inode (this does rmdir as well)
977 * @dir: The inode of the directory containing the inode to unlink
978 * @dentry: The file itself
980 * This routine uses the type of the inode as a flag to figure out
981 * whether this is an unlink or an rmdir.
986 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
988 struct gfs2_inode *dip = GFS2_I(dir);
989 struct gfs2_sbd *sdp = GFS2_SB(dir);
990 struct inode *inode = dentry->d_inode;
991 struct gfs2_inode *ip = GFS2_I(inode);
992 struct gfs2_holder ghs[3];
993 struct gfs2_rgrpd *rgd;
996 error = gfs2_rindex_update(sdp);
1002 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1003 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1005 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1009 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1012 error = gfs2_glock_nq(ghs); /* parent */
1016 error = gfs2_glock_nq(ghs + 1); /* child */
1021 if (inode->i_nlink == 0)
1024 if (S_ISDIR(inode->i_mode)) {
1026 if (ip->i_entries > 2 || inode->i_nlink > 2)
1030 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1034 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1038 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1042 error = gfs2_unlink_inode(dip, dentry);
1045 gfs2_trans_end(sdp);
1047 gfs2_glock_dq(ghs + 2);
1049 gfs2_glock_dq(ghs + 1);
1053 gfs2_holder_uninit(ghs + 2);
1055 gfs2_holder_uninit(ghs + 1);
1056 gfs2_holder_uninit(ghs);
1061 * gfs2_symlink - Create a symlink
1062 * @dir: The directory to create the symlink in
1063 * @dentry: The dentry to put the symlink in
1064 * @symname: The thing which the link points to
1069 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1070 const char *symname)
1072 struct gfs2_sbd *sdp = GFS2_SB(dir);
1075 size = strlen(symname);
1076 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1077 return -ENAMETOOLONG;
1079 return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1083 * gfs2_mkdir - Make a directory
1084 * @dir: The parent directory of the new one
1085 * @dentry: The dentry of the new directory
1086 * @mode: The mode of the new directory
1091 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1093 struct gfs2_sbd *sdp = GFS2_SB(dir);
1094 unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1095 return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, dsize, 0);
1099 * gfs2_mknod - Make a special file
1100 * @dir: The directory in which the special file will reside
1101 * @dentry: The dentry of the special file
1102 * @mode: The mode of the special file
1103 * @dev: The device specification of the special file
1107 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1110 return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0, 0);
1114 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1118 * Follow @to back to the root and make sure we don't encounter @this
1119 * Assumes we already hold the rename lock.
1124 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1126 struct inode *dir = &to->i_inode;
1127 struct super_block *sb = dir->i_sb;
1134 if (dir == &this->i_inode) {
1138 if (dir == sb->s_root->d_inode) {
1143 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1145 error = PTR_ERR(tmp);
1159 * gfs2_rename - Rename a file
1160 * @odir: Parent directory of old file name
1161 * @odentry: The old dentry of the file
1162 * @ndir: Parent directory of new file name
1163 * @ndentry: The new dentry of the file
1168 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1169 struct inode *ndir, struct dentry *ndentry)
1171 struct gfs2_inode *odip = GFS2_I(odir);
1172 struct gfs2_inode *ndip = GFS2_I(ndir);
1173 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1174 struct gfs2_inode *nip = NULL;
1175 struct gfs2_sbd *sdp = GFS2_SB(odir);
1176 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1177 struct gfs2_rgrpd *nrgd;
1178 unsigned int num_gh;
1180 int alloc_required = 0;
1184 if (ndentry->d_inode) {
1185 nip = GFS2_I(ndentry->d_inode);
1190 error = gfs2_rindex_update(sdp);
1194 error = gfs2_rs_alloc(ndip);
1199 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1204 if (S_ISDIR(ip->i_inode.i_mode)) {
1206 /* don't move a dirctory into it's subdir */
1207 error = gfs2_ok_to_move(ip, ndip);
1214 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1216 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1219 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1223 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1225 /* grab the resource lock for unlink flag twiddling
1226 * this is the case of the target file already existing
1227 * so we unlink before doing the rename
1229 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1231 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1234 for (x = 0; x < num_gh; x++) {
1235 error = gfs2_glock_nq(ghs + x);
1241 if (ip->i_inode.i_nlink == 0)
1244 /* Check out the old directory */
1246 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1250 /* Check out the new directory */
1253 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1257 if (nip->i_inode.i_nlink == 0) {
1262 if (S_ISDIR(nip->i_inode.i_mode)) {
1263 if (nip->i_entries < 2) {
1264 gfs2_consist_inode(nip);
1268 if (nip->i_entries > 2) {
1274 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1278 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1290 if (!ndip->i_inode.i_nlink) {
1294 if (ndip->i_entries == (u32)-1) {
1298 if (S_ISDIR(ip->i_inode.i_mode) &&
1299 ndip->i_inode.i_nlink == (u32)-1) {
1306 /* Check out the dir to be renamed */
1309 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1315 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1316 error = alloc_required;
1320 if (alloc_required) {
1321 error = gfs2_quota_lock_check(ndip);
1325 error = gfs2_inplace_reserve(ndip, sdp->sd_max_dirres, 0);
1329 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1330 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1331 4 * RES_DINODE + 4 * RES_LEAF +
1332 RES_STATFS + RES_QUOTA + 4, 0);
1336 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1337 5 * RES_LEAF + 4, 0);
1342 /* Remove the target file, if it exists */
1345 error = gfs2_unlink_inode(ndip, ndentry);
1348 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1352 struct buffer_head *dibh;
1353 error = gfs2_meta_inode_buffer(ip, &dibh);
1356 ip->i_inode.i_ctime = CURRENT_TIME;
1357 gfs2_trans_add_meta(ip->i_gl, dibh);
1358 gfs2_dinode_out(ip, dibh->b_data);
1362 error = gfs2_dir_del(odip, odentry);
1366 error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1371 gfs2_trans_end(sdp);
1374 gfs2_inplace_release(ndip);
1377 gfs2_quota_unlock(ndip);
1380 gfs2_glock_dq(ghs + x);
1381 gfs2_holder_uninit(ghs + x);
1385 gfs2_glock_dq_uninit(&r_gh);
1391 * gfs2_follow_link - Follow a symbolic link
1392 * @dentry: The dentry of the link
1393 * @nd: Data that we pass to vfs_follow_link()
1395 * This can handle symlinks of any size.
1397 * Returns: 0 on success or error code
1400 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1402 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1403 struct gfs2_holder i_gh;
1404 struct buffer_head *dibh;
1409 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1410 error = gfs2_glock_nq(&i_gh);
1412 gfs2_holder_uninit(&i_gh);
1413 nd_set_link(nd, ERR_PTR(error));
1417 size = (unsigned int)i_size_read(&ip->i_inode);
1419 gfs2_consist_inode(ip);
1420 buf = ERR_PTR(-EIO);
1424 error = gfs2_meta_inode_buffer(ip, &dibh);
1426 buf = ERR_PTR(error);
1430 buf = kzalloc(size + 1, GFP_NOFS);
1432 buf = ERR_PTR(-ENOMEM);
1434 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1437 gfs2_glock_dq_uninit(&i_gh);
1438 nd_set_link(nd, buf);
1442 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1444 char *s = nd_get_link(nd);
1452 * @mask: The mask to be tested
1453 * @flags: Indicates whether this is an RCU path walk or not
1455 * This may be called from the VFS directly, or from within GFS2 with the
1456 * inode locked, so we look to see if the glock is already locked and only
1457 * lock the glock if its not already been done.
1462 int gfs2_permission(struct inode *inode, int mask)
1464 struct gfs2_inode *ip;
1465 struct gfs2_holder i_gh;
1471 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1472 if (mask & MAY_NOT_BLOCK)
1474 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1480 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1483 error = generic_permission(inode, mask);
1485 gfs2_glock_dq_uninit(&i_gh);
1490 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1492 setattr_copy(inode, attr);
1493 mark_inode_dirty(inode);
1498 * gfs2_setattr_simple -
1505 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1509 if (current->journal_info)
1510 return __gfs2_setattr_simple(inode, attr);
1512 error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1516 error = __gfs2_setattr_simple(inode, attr);
1517 gfs2_trans_end(GFS2_SB(inode));
1521 static int setattr_chown(struct inode *inode, struct iattr *attr)
1523 struct gfs2_inode *ip = GFS2_I(inode);
1524 struct gfs2_sbd *sdp = GFS2_SB(inode);
1529 ouid = inode->i_uid;
1530 ogid = inode->i_gid;
1531 nuid = attr->ia_uid;
1532 ngid = attr->ia_gid;
1534 if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1535 ouid = nuid = NO_UID_QUOTA_CHANGE;
1536 if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1537 ogid = ngid = NO_GID_QUOTA_CHANGE;
1539 error = gfs2_quota_lock(ip, nuid, ngid);
1543 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1544 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1545 error = gfs2_quota_check(ip, nuid, ngid);
1550 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1554 error = gfs2_setattr_simple(inode, attr);
1558 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1559 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1560 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1561 gfs2_quota_change(ip, -blocks, ouid, ogid);
1562 gfs2_quota_change(ip, blocks, nuid, ngid);
1566 gfs2_trans_end(sdp);
1568 gfs2_quota_unlock(ip);
1573 * gfs2_setattr - Change attributes on an inode
1574 * @dentry: The dentry which is changing
1575 * @attr: The structure describing the change
1577 * The VFS layer wants to change one or more of an inodes attributes. Write
1578 * that change out to disk.
1583 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1585 struct inode *inode = dentry->d_inode;
1586 struct gfs2_inode *ip = GFS2_I(inode);
1587 struct gfs2_holder i_gh;
1590 error = gfs2_rs_alloc(ip);
1594 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1599 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1602 error = inode_change_ok(inode, attr);
1606 if (attr->ia_valid & ATTR_SIZE)
1607 error = gfs2_setattr_size(inode, attr->ia_size);
1608 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1609 error = setattr_chown(inode, attr);
1610 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1611 error = gfs2_acl_chmod(ip, attr);
1613 error = gfs2_setattr_simple(inode, attr);
1617 mark_inode_dirty(inode);
1618 gfs2_glock_dq_uninit(&i_gh);
1623 * gfs2_getattr - Read out an inode's attributes
1624 * @mnt: The vfsmount the inode is being accessed from
1625 * @dentry: The dentry to stat
1626 * @stat: The inode's stats
1628 * This may be called from the VFS directly, or from within GFS2 with the
1629 * inode locked, so we look to see if the glock is already locked and only
1630 * lock the glock if its not already been done. Note that its the NFS
1631 * readdirplus operation which causes this to be called (from filldir)
1632 * with the glock already held.
1637 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1640 struct inode *inode = dentry->d_inode;
1641 struct gfs2_inode *ip = GFS2_I(inode);
1642 struct gfs2_holder gh;
1646 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1647 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1653 generic_fillattr(inode, stat);
1655 gfs2_glock_dq_uninit(&gh);
1660 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1661 const void *data, size_t size, int flags)
1663 struct inode *inode = dentry->d_inode;
1664 struct gfs2_inode *ip = GFS2_I(inode);
1665 struct gfs2_holder gh;
1668 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1669 ret = gfs2_glock_nq(&gh);
1671 ret = gfs2_rs_alloc(ip);
1673 ret = generic_setxattr(dentry, name, data, size, flags);
1676 gfs2_holder_uninit(&gh);
1680 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1681 void *data, size_t size)
1683 struct inode *inode = dentry->d_inode;
1684 struct gfs2_inode *ip = GFS2_I(inode);
1685 struct gfs2_holder gh;
1688 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1689 ret = gfs2_glock_nq(&gh);
1691 ret = generic_getxattr(dentry, name, data, size);
1694 gfs2_holder_uninit(&gh);
1698 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1700 struct inode *inode = dentry->d_inode;
1701 struct gfs2_inode *ip = GFS2_I(inode);
1702 struct gfs2_holder gh;
1705 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1706 ret = gfs2_glock_nq(&gh);
1708 ret = gfs2_rs_alloc(ip);
1710 ret = generic_removexattr(dentry, name);
1713 gfs2_holder_uninit(&gh);
1717 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1720 struct gfs2_inode *ip = GFS2_I(inode);
1721 struct gfs2_holder gh;
1724 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1728 mutex_lock(&inode->i_mutex);
1730 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1734 if (gfs2_is_stuffed(ip)) {
1735 u64 phys = ip->i_no_addr << inode->i_blkbits;
1736 u64 size = i_size_read(inode);
1737 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1738 FIEMAP_EXTENT_DATA_INLINE;
1739 phys += sizeof(struct gfs2_dinode);
1741 if (start + len > size)
1744 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1749 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1753 gfs2_glock_dq_uninit(&gh);
1755 mutex_unlock(&inode->i_mutex);
1759 const struct inode_operations gfs2_file_iops = {
1760 .permission = gfs2_permission,
1761 .setattr = gfs2_setattr,
1762 .getattr = gfs2_getattr,
1763 .setxattr = gfs2_setxattr,
1764 .getxattr = gfs2_getxattr,
1765 .listxattr = gfs2_listxattr,
1766 .removexattr = gfs2_removexattr,
1767 .fiemap = gfs2_fiemap,
1768 .get_acl = gfs2_get_acl,
1771 const struct inode_operations gfs2_dir_iops = {
1772 .create = gfs2_create,
1773 .lookup = gfs2_lookup,
1775 .unlink = gfs2_unlink,
1776 .symlink = gfs2_symlink,
1777 .mkdir = gfs2_mkdir,
1778 .rmdir = gfs2_unlink,
1779 .mknod = gfs2_mknod,
1780 .rename = gfs2_rename,
1781 .permission = gfs2_permission,
1782 .setattr = gfs2_setattr,
1783 .getattr = gfs2_getattr,
1784 .setxattr = gfs2_setxattr,
1785 .getxattr = gfs2_getxattr,
1786 .listxattr = gfs2_listxattr,
1787 .removexattr = gfs2_removexattr,
1788 .fiemap = gfs2_fiemap,
1789 .get_acl = gfs2_get_acl,
1792 const struct inode_operations gfs2_symlink_iops = {
1793 .readlink = generic_readlink,
1794 .follow_link = gfs2_follow_link,
1795 .put_link = gfs2_put_link,
1796 .permission = gfs2_permission,
1797 .setattr = gfs2_setattr,
1798 .getattr = gfs2_getattr,
1799 .setxattr = gfs2_setxattr,
1800 .getxattr = gfs2_getxattr,
1801 .listxattr = gfs2_listxattr,
1802 .removexattr = gfs2_removexattr,
1803 .fiemap = gfs2_fiemap,
1804 .get_acl = gfs2_get_acl,