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, umode_t *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);
394 error = gfs2_inplace_reserve(dip, RES_DINODE);
398 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
402 error = gfs2_alloc_blocks(dip, no_addr, &dblocks, 1, generation);
407 gfs2_inplace_release(dip);
412 static void gfs2_init_dir(struct buffer_head *dibh,
413 const struct gfs2_inode *parent)
415 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
416 struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
418 gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
419 dent->de_inum = di->di_num; /* already GFS2 endian */
420 dent->de_type = cpu_to_be16(DT_DIR);
422 dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
423 gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
424 gfs2_inum_out(parent, dent);
425 dent->de_type = cpu_to_be16(DT_DIR);
430 * init_dinode - Fill in a new dinode structure
431 * @dip: The directory this inode is being created in
432 * @gl: The glock covering the new inode
433 * @inum: The inode number
434 * @mode: The file permissions
435 * @uid: The uid of the new inode
436 * @gid: The gid of the new inode
437 * @generation: The generation number of the new inode
438 * @dev: The device number (if a device node)
439 * @symname: The symlink destination (if a symlink)
440 * @size: The inode size (ignored for directories)
441 * @bhp: The buffer head (returned to caller)
445 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
446 const struct gfs2_inum_host *inum, umode_t mode,
447 unsigned int uid, unsigned int gid,
448 const u64 *generation, dev_t dev, const char *symname,
449 unsigned size, struct buffer_head **bhp)
451 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
452 struct gfs2_dinode *di;
453 struct buffer_head *dibh;
454 struct timespec tv = CURRENT_TIME;
456 dibh = gfs2_meta_new(gl, inum->no_addr);
457 gfs2_trans_add_bh(gl, dibh, 1);
458 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
459 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
460 di = (struct gfs2_dinode *)dibh->b_data;
462 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
463 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
464 di->di_mode = cpu_to_be32(mode);
465 di->di_uid = cpu_to_be32(uid);
466 di->di_gid = cpu_to_be32(gid);
468 di->di_size = cpu_to_be64(size);
469 di->di_blocks = cpu_to_be64(1);
470 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec);
471 di->di_major = cpu_to_be32(MAJOR(dev));
472 di->di_minor = cpu_to_be32(MINOR(dev));
473 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
474 di->di_generation = cpu_to_be64(*generation);
477 di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
483 memset(&di->__pad4, 0, sizeof(di->__pad4));
485 di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
486 di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
487 di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
488 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
490 switch(mode & S_IFMT) {
492 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
493 gfs2_tune_get(sdp, gt_new_files_jdata))
494 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
497 di->di_flags |= cpu_to_be32(dip->i_diskflags &
498 GFS2_DIF_INHERIT_JDATA);
499 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
500 di->di_size = cpu_to_be64(sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode));
501 di->di_entries = cpu_to_be32(2);
502 gfs2_init_dir(dibh, dip);
505 memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, size);
509 set_buffer_uptodate(dibh);
514 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
515 umode_t mode, const struct gfs2_inum_host *inum,
516 const u64 *generation, dev_t dev, const char *symname,
517 unsigned int size, struct buffer_head **bhp)
519 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
520 unsigned int uid, gid;
523 munge_mode_uid_gid(dip, &mode, &uid, &gid);
524 if (!gfs2_qadata_get(dip))
527 error = gfs2_quota_lock(dip, uid, gid);
531 error = gfs2_quota_check(dip, uid, gid);
535 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
539 init_dinode(dip, gl, inum, mode, uid, gid, generation, dev, symname, size, bhp);
540 gfs2_quota_change(dip, +1, uid, gid);
544 gfs2_quota_unlock(dip);
546 gfs2_qadata_put(dip);
550 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
551 struct gfs2_inode *ip)
553 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
554 struct gfs2_qadata *qa;
556 struct buffer_head *dibh;
559 qa = gfs2_qadata_get(dip);
563 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
567 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
568 if (alloc_required < 0)
569 goto fail_quota_locks;
570 if (alloc_required) {
571 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
573 goto fail_quota_locks;
575 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres);
577 goto fail_quota_locks;
579 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
580 dip->i_rgd->rd_length +
582 RES_STATFS + RES_QUOTA, 0);
586 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
588 goto fail_quota_locks;
591 error = gfs2_dir_add(&dip->i_inode, name, ip);
595 error = gfs2_meta_inode_buffer(ip, &dibh);
598 set_nlink(&ip->i_inode, S_ISDIR(ip->i_inode.i_mode) ? 2 : 1);
599 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
600 gfs2_dinode_out(ip, dibh->b_data);
608 gfs2_inplace_release(dip);
611 gfs2_quota_unlock(dip);
614 gfs2_qadata_put(dip);
618 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
621 const struct xattr *xattr;
624 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
625 err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
627 GFS2_EATYPE_SECURITY);
634 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
635 const struct qstr *qstr)
637 return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
638 &gfs2_initxattrs, NULL);
642 * gfs2_create_inode - Create a new inode
643 * @dir: The parent directory
644 * @dentry: The new dentry
645 * @mode: The permissions on the new inode
646 * @dev: For device nodes, this is the device number
647 * @symname: For symlinks, this is the link destination
648 * @size: The initial size of the inode (ignored for directories)
650 * Returns: 0 on success, or error code
653 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
654 umode_t mode, dev_t dev, const char *symname,
655 unsigned int size, int excl)
657 const struct qstr *name = &dentry->d_name;
658 struct gfs2_holder ghs[2];
659 struct inode *inode = NULL;
660 struct gfs2_inode *dip = GFS2_I(dir);
661 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
662 struct gfs2_inum_host inum = { .no_addr = 0, .no_formal_ino = 0 };
665 struct buffer_head *bh = NULL;
667 if (!name->len || name->len > GFS2_FNAMESIZE)
668 return -ENAMETOOLONG;
670 error = gfs2_rs_alloc(dip);
674 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
678 error = create_ok(dip, name, mode);
679 if ((error == -EEXIST) && S_ISREG(mode) && !excl) {
680 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
681 gfs2_glock_dq_uninit(ghs);
682 d_instantiate(dentry, inode);
683 return IS_ERR(inode) ? PTR_ERR(inode) : 0;
688 error = alloc_dinode(dip, &inum.no_addr, &generation);
691 inum.no_formal_ino = generation;
693 error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
694 LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
698 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev, symname, size, &bh);
702 inode = gfs2_inode_lookup(dir->i_sb, IF2DT(mode), inum.no_addr,
703 inum.no_formal_ino, 0);
707 error = gfs2_inode_refresh(GFS2_I(inode));
711 /* the new inode needs a reservation so it can allocate xattrs. */
712 error = gfs2_rs_alloc(GFS2_I(inode));
716 error = gfs2_acl_create(dip, inode);
720 error = gfs2_security_init(dip, GFS2_I(inode), name);
724 error = link_dinode(dip, name, GFS2_I(inode));
732 /* Check if we reserved space in the rgrp. Function link_dinode may
733 not, depending on whether alloc is required. */
734 if (gfs2_mb_reserved(dip))
735 gfs2_inplace_release(dip);
736 gfs2_quota_unlock(dip);
737 gfs2_qadata_put(dip);
738 mark_inode_dirty(inode);
739 gfs2_glock_dq_uninit_m(2, ghs);
740 d_instantiate(dentry, inode);
744 gfs2_glock_dq_uninit(ghs + 1);
746 gfs2_glock_dq_uninit(ghs);
747 if (inode && !IS_ERR(inode)) {
748 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
758 * gfs2_create - Create a file
759 * @dir: The directory in which to create the file
760 * @dentry: The dentry of the new file
761 * @mode: The mode of the new file
766 static int gfs2_create(struct inode *dir, struct dentry *dentry,
767 umode_t mode, struct nameidata *nd)
770 if (nd && (nd->flags & LOOKUP_EXCL))
772 return gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0, excl);
776 * gfs2_lookup - Look up a filename in a directory and return its inode
777 * @dir: The directory inode
778 * @dentry: The dentry of the new inode
779 * @nd: passed from Linux VFS, ignored by us
781 * Called by the VFS layer. Lock dir and call gfs2_lookupi()
786 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
787 struct nameidata *nd)
789 struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
790 if (inode && !IS_ERR(inode)) {
791 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
792 struct gfs2_holder gh;
794 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
797 return ERR_PTR(error);
799 gfs2_glock_dq_uninit(&gh);
801 return d_splice_alias(inode, dentry);
805 * gfs2_link - Link to a file
806 * @old_dentry: The inode to link
807 * @dir: Add link to this directory
808 * @dentry: The name of the link
810 * Link the inode in "old_dentry" into the directory "dir" with the
816 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
817 struct dentry *dentry)
819 struct gfs2_inode *dip = GFS2_I(dir);
820 struct gfs2_sbd *sdp = GFS2_SB(dir);
821 struct inode *inode = old_dentry->d_inode;
822 struct gfs2_inode *ip = GFS2_I(inode);
823 struct gfs2_holder ghs[2];
824 struct buffer_head *dibh;
828 if (S_ISDIR(inode->i_mode))
831 error = gfs2_rs_alloc(dip);
835 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
836 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
838 error = gfs2_glock_nq(ghs); /* parent */
842 error = gfs2_glock_nq(ghs + 1); /* child */
847 if (inode->i_nlink == 0)
850 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
854 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
865 if (!dip->i_inode.i_nlink)
868 if (dip->i_entries == (u32)-1)
871 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
874 if (!ip->i_inode.i_nlink)
877 if (ip->i_inode.i_nlink == (u32)-1)
880 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
885 if (alloc_required) {
886 struct gfs2_qadata *qa = gfs2_qadata_get(dip);
893 error = gfs2_quota_lock_check(dip);
897 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres);
901 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
902 gfs2_rg_blocks(dip) +
903 2 * RES_DINODE + RES_STATFS +
908 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
913 error = gfs2_meta_inode_buffer(ip, &dibh);
917 error = gfs2_dir_add(dir, &dentry->d_name, ip);
921 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
922 inc_nlink(&ip->i_inode);
923 ip->i_inode.i_ctime = CURRENT_TIME;
925 d_instantiate(dentry, inode);
926 mark_inode_dirty(inode);
934 gfs2_inplace_release(dip);
937 gfs2_quota_unlock(dip);
940 gfs2_qadata_put(dip);
942 gfs2_glock_dq(ghs + 1);
946 gfs2_holder_uninit(ghs);
947 gfs2_holder_uninit(ghs + 1);
952 * gfs2_unlink_ok - check to see that a inode is still in a directory
953 * @dip: the directory
954 * @name: the name of the file
957 * Assumes that the lock on (at least) @dip is held.
959 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
962 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
963 const struct gfs2_inode *ip)
967 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
970 if ((dip->i_inode.i_mode & S_ISVTX) &&
971 dip->i_inode.i_uid != current_fsuid() &&
972 ip->i_inode.i_uid != current_fsuid() && !capable(CAP_FOWNER))
975 if (IS_APPEND(&dip->i_inode))
978 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
982 error = gfs2_dir_check(&dip->i_inode, name, ip);
990 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
991 * @dip: The parent directory
992 * @name: The name of the entry in the parent directory
993 * @bh: The inode buffer for the inode to be removed
994 * @inode: The inode to be removed
996 * Called with all the locks and in a transaction. This will only be
997 * called for a directory after it has been checked to ensure it is empty.
999 * Returns: 0 on success, or an error
1002 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1003 const struct dentry *dentry,
1004 struct buffer_head *bh)
1006 struct inode *inode = dentry->d_inode;
1007 struct gfs2_inode *ip = GFS2_I(inode);
1010 error = gfs2_dir_del(dip, dentry);
1015 inode->i_ctime = CURRENT_TIME;
1016 if (S_ISDIR(inode->i_mode))
1020 mark_inode_dirty(inode);
1021 if (inode->i_nlink == 0)
1022 gfs2_unlink_di(inode);
1028 * gfs2_unlink - Unlink an inode (this does rmdir as well)
1029 * @dir: The inode of the directory containing the inode to unlink
1030 * @dentry: The file itself
1032 * This routine uses the type of the inode as a flag to figure out
1033 * whether this is an unlink or an rmdir.
1038 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1040 struct gfs2_inode *dip = GFS2_I(dir);
1041 struct gfs2_sbd *sdp = GFS2_SB(dir);
1042 struct inode *inode = dentry->d_inode;
1043 struct gfs2_inode *ip = GFS2_I(inode);
1044 struct buffer_head *bh;
1045 struct gfs2_holder ghs[3];
1046 struct gfs2_rgrpd *rgd;
1049 error = gfs2_rindex_update(sdp);
1055 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1056 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1058 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1062 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1065 error = gfs2_glock_nq(ghs); /* parent */
1069 error = gfs2_glock_nq(ghs + 1); /* child */
1074 if (inode->i_nlink == 0)
1077 if (S_ISDIR(inode->i_mode)) {
1079 if (ip->i_entries > 2 || inode->i_nlink > 2)
1083 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1087 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1091 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1095 error = gfs2_meta_inode_buffer(ip, &bh);
1099 error = gfs2_unlink_inode(dip, dentry, bh);
1103 gfs2_trans_end(sdp);
1105 gfs2_glock_dq(ghs + 2);
1107 gfs2_glock_dq(ghs + 1);
1111 gfs2_holder_uninit(ghs + 2);
1113 gfs2_holder_uninit(ghs + 1);
1114 gfs2_holder_uninit(ghs);
1119 * gfs2_symlink - Create a symlink
1120 * @dir: The directory to create the symlink in
1121 * @dentry: The dentry to put the symlink in
1122 * @symname: The thing which the link points to
1127 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1128 const char *symname)
1130 struct gfs2_sbd *sdp = GFS2_SB(dir);
1133 size = strlen(symname);
1134 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1135 return -ENAMETOOLONG;
1137 return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1141 * gfs2_mkdir - Make a directory
1142 * @dir: The parent directory of the new one
1143 * @dentry: The dentry of the new directory
1144 * @mode: The mode of the new directory
1149 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1151 return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0, 0);
1155 * gfs2_mknod - Make a special file
1156 * @dir: The directory in which the special file will reside
1157 * @dentry: The dentry of the special file
1158 * @mode: The mode of the special file
1159 * @dev: The device specification of the special file
1163 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1166 return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0, 0);
1170 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1174 * Follow @to back to the root and make sure we don't encounter @this
1175 * Assumes we already hold the rename lock.
1180 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1182 struct inode *dir = &to->i_inode;
1183 struct super_block *sb = dir->i_sb;
1190 if (dir == &this->i_inode) {
1194 if (dir == sb->s_root->d_inode) {
1199 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1201 error = PTR_ERR(tmp);
1215 * gfs2_rename - Rename a file
1216 * @odir: Parent directory of old file name
1217 * @odentry: The old dentry of the file
1218 * @ndir: Parent directory of new file name
1219 * @ndentry: The new dentry of the file
1224 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1225 struct inode *ndir, struct dentry *ndentry)
1227 struct gfs2_inode *odip = GFS2_I(odir);
1228 struct gfs2_inode *ndip = GFS2_I(ndir);
1229 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1230 struct gfs2_inode *nip = NULL;
1231 struct gfs2_sbd *sdp = GFS2_SB(odir);
1232 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1233 struct gfs2_rgrpd *nrgd;
1234 unsigned int num_gh;
1236 int alloc_required = 0;
1240 if (ndentry->d_inode) {
1241 nip = GFS2_I(ndentry->d_inode);
1246 error = gfs2_rindex_update(sdp);
1250 error = gfs2_rs_alloc(ndip);
1255 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1260 if (S_ISDIR(ip->i_inode.i_mode)) {
1262 /* don't move a dirctory into it's subdir */
1263 error = gfs2_ok_to_move(ip, ndip);
1270 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1272 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1275 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1279 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1281 /* grab the resource lock for unlink flag twiddling
1282 * this is the case of the target file already existing
1283 * so we unlink before doing the rename
1285 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1287 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1290 for (x = 0; x < num_gh; x++) {
1291 error = gfs2_glock_nq(ghs + x);
1297 if (ip->i_inode.i_nlink == 0)
1300 /* Check out the old directory */
1302 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1306 /* Check out the new directory */
1309 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1313 if (nip->i_inode.i_nlink == 0) {
1318 if (S_ISDIR(nip->i_inode.i_mode)) {
1319 if (nip->i_entries < 2) {
1320 gfs2_consist_inode(nip);
1324 if (nip->i_entries > 2) {
1330 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1334 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1346 if (!ndip->i_inode.i_nlink) {
1350 if (ndip->i_entries == (u32)-1) {
1354 if (S_ISDIR(ip->i_inode.i_mode) &&
1355 ndip->i_inode.i_nlink == (u32)-1) {
1362 /* Check out the dir to be renamed */
1365 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1371 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1372 error = alloc_required;
1376 if (alloc_required) {
1377 struct gfs2_qadata *qa = gfs2_qadata_get(ndip);
1384 error = gfs2_quota_lock_check(ndip);
1388 error = gfs2_inplace_reserve(ndip, sdp->sd_max_dirres);
1392 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1393 gfs2_rg_blocks(ndip) +
1394 4 * RES_DINODE + 4 * RES_LEAF +
1395 RES_STATFS + RES_QUOTA + 4, 0);
1399 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1400 5 * RES_LEAF + 4, 0);
1405 /* Remove the target file, if it exists */
1408 struct buffer_head *bh;
1409 error = gfs2_meta_inode_buffer(nip, &bh);
1412 error = gfs2_unlink_inode(ndip, ndentry, bh);
1417 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1421 struct buffer_head *dibh;
1422 error = gfs2_meta_inode_buffer(ip, &dibh);
1425 ip->i_inode.i_ctime = CURRENT_TIME;
1426 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1427 gfs2_dinode_out(ip, dibh->b_data);
1431 error = gfs2_dir_del(odip, odentry);
1435 error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1440 gfs2_trans_end(sdp);
1443 gfs2_inplace_release(ndip);
1446 gfs2_quota_unlock(ndip);
1449 gfs2_qadata_put(ndip);
1452 gfs2_glock_dq(ghs + x);
1453 gfs2_holder_uninit(ghs + x);
1457 gfs2_glock_dq_uninit(&r_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_qadata_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);
1642 gfs2_qadata_put(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_rs_alloc(ip);
1668 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1673 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1676 error = inode_change_ok(inode, attr);
1680 if (attr->ia_valid & ATTR_SIZE)
1681 error = gfs2_setattr_size(inode, attr->ia_size);
1682 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1683 error = setattr_chown(inode, attr);
1684 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1685 error = gfs2_acl_chmod(ip, attr);
1687 error = gfs2_setattr_simple(inode, attr);
1691 mark_inode_dirty(inode);
1692 gfs2_glock_dq_uninit(&i_gh);
1697 * gfs2_getattr - Read out an inode's attributes
1698 * @mnt: The vfsmount the inode is being accessed from
1699 * @dentry: The dentry to stat
1700 * @stat: The inode's stats
1702 * This may be called from the VFS directly, or from within GFS2 with the
1703 * inode locked, so we look to see if the glock is already locked and only
1704 * lock the glock if its not already been done. Note that its the NFS
1705 * readdirplus operation which causes this to be called (from filldir)
1706 * with the glock already held.
1711 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1714 struct inode *inode = dentry->d_inode;
1715 struct gfs2_inode *ip = GFS2_I(inode);
1716 struct gfs2_holder gh;
1720 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1721 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1727 generic_fillattr(inode, stat);
1729 gfs2_glock_dq_uninit(&gh);
1734 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1735 const void *data, size_t size, int flags)
1737 struct inode *inode = dentry->d_inode;
1738 struct gfs2_inode *ip = GFS2_I(inode);
1739 struct gfs2_holder gh;
1742 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1743 ret = gfs2_glock_nq(&gh);
1745 ret = generic_setxattr(dentry, name, data, size, flags);
1748 gfs2_holder_uninit(&gh);
1752 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1753 void *data, size_t size)
1755 struct inode *inode = dentry->d_inode;
1756 struct gfs2_inode *ip = GFS2_I(inode);
1757 struct gfs2_holder gh;
1760 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1761 ret = gfs2_glock_nq(&gh);
1763 ret = generic_getxattr(dentry, name, data, size);
1766 gfs2_holder_uninit(&gh);
1770 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1772 struct inode *inode = dentry->d_inode;
1773 struct gfs2_inode *ip = GFS2_I(inode);
1774 struct gfs2_holder gh;
1777 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1778 ret = gfs2_glock_nq(&gh);
1780 ret = generic_removexattr(dentry, name);
1783 gfs2_holder_uninit(&gh);
1787 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1790 struct gfs2_inode *ip = GFS2_I(inode);
1791 struct gfs2_holder gh;
1794 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1798 mutex_lock(&inode->i_mutex);
1800 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1804 if (gfs2_is_stuffed(ip)) {
1805 u64 phys = ip->i_no_addr << inode->i_blkbits;
1806 u64 size = i_size_read(inode);
1807 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1808 FIEMAP_EXTENT_DATA_INLINE;
1809 phys += sizeof(struct gfs2_dinode);
1811 if (start + len > size)
1814 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1819 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1823 gfs2_glock_dq_uninit(&gh);
1825 mutex_unlock(&inode->i_mutex);
1829 const struct inode_operations gfs2_file_iops = {
1830 .permission = gfs2_permission,
1831 .setattr = gfs2_setattr,
1832 .getattr = gfs2_getattr,
1833 .setxattr = gfs2_setxattr,
1834 .getxattr = gfs2_getxattr,
1835 .listxattr = gfs2_listxattr,
1836 .removexattr = gfs2_removexattr,
1837 .fiemap = gfs2_fiemap,
1838 .get_acl = gfs2_get_acl,
1841 const struct inode_operations gfs2_dir_iops = {
1842 .create = gfs2_create,
1843 .lookup = gfs2_lookup,
1845 .unlink = gfs2_unlink,
1846 .symlink = gfs2_symlink,
1847 .mkdir = gfs2_mkdir,
1848 .rmdir = gfs2_unlink,
1849 .mknod = gfs2_mknod,
1850 .rename = gfs2_rename,
1851 .permission = gfs2_permission,
1852 .setattr = gfs2_setattr,
1853 .getattr = gfs2_getattr,
1854 .setxattr = gfs2_setxattr,
1855 .getxattr = gfs2_getxattr,
1856 .listxattr = gfs2_listxattr,
1857 .removexattr = gfs2_removexattr,
1858 .fiemap = gfs2_fiemap,
1859 .get_acl = gfs2_get_acl,
1862 const struct inode_operations gfs2_symlink_iops = {
1863 .readlink = generic_readlink,
1864 .follow_link = gfs2_follow_link,
1865 .put_link = gfs2_put_link,
1866 .permission = gfs2_permission,
1867 .setattr = gfs2_setattr,
1868 .getattr = gfs2_getattr,
1869 .setxattr = gfs2_setxattr,
1870 .getxattr = gfs2_getxattr,
1871 .listxattr = gfs2_listxattr,
1872 .removexattr = gfs2_removexattr,
1873 .fiemap = gfs2_fiemap,
1874 .get_acl = gfs2_get_acl,