2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2011 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/namei.h>
16 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/crc32.h>
20 #include <linux/fiemap.h>
21 #include <linux/security.h>
22 #include <asm/uaccess.h>
40 struct gfs2_skip_data {
46 static int iget_test(struct inode *inode, void *opaque)
48 struct gfs2_inode *ip = GFS2_I(inode);
49 struct gfs2_skip_data *data = opaque;
51 if (ip->i_no_addr == data->no_addr) {
52 if (data->non_block &&
53 inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
62 static int iget_set(struct inode *inode, void *opaque)
64 struct gfs2_inode *ip = GFS2_I(inode);
65 struct gfs2_skip_data *data = opaque;
69 inode->i_ino = (unsigned long)(data->no_addr);
70 ip->i_no_addr = data->no_addr;
74 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int non_block)
76 unsigned long hash = (unsigned long)no_addr;
77 struct gfs2_skip_data data;
79 data.no_addr = no_addr;
81 data.non_block = non_block;
82 return ilookup5(sb, hash, iget_test, &data);
85 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr,
88 struct gfs2_skip_data data;
89 unsigned long hash = (unsigned long)no_addr;
91 data.no_addr = no_addr;
93 data.non_block = non_block;
94 return iget5_locked(sb, hash, iget_test, iget_set, &data);
98 * gfs2_set_iop - Sets inode operations
99 * @inode: The inode with correct i_mode filled in
101 * GFS2 lookup code fills in vfs inode contents based on info obtained
102 * from directory entry inside gfs2_inode_lookup().
105 static void gfs2_set_iop(struct inode *inode)
107 struct gfs2_sbd *sdp = GFS2_SB(inode);
108 umode_t mode = inode->i_mode;
111 inode->i_op = &gfs2_file_iops;
112 if (gfs2_localflocks(sdp))
113 inode->i_fop = &gfs2_file_fops_nolock;
115 inode->i_fop = &gfs2_file_fops;
116 } else if (S_ISDIR(mode)) {
117 inode->i_op = &gfs2_dir_iops;
118 if (gfs2_localflocks(sdp))
119 inode->i_fop = &gfs2_dir_fops_nolock;
121 inode->i_fop = &gfs2_dir_fops;
122 } else if (S_ISLNK(mode)) {
123 inode->i_op = &gfs2_symlink_iops;
125 inode->i_op = &gfs2_file_iops;
126 init_special_inode(inode, inode->i_mode, inode->i_rdev);
131 * gfs2_inode_lookup - Lookup an inode
132 * @sb: The super block
133 * @no_addr: The inode number
134 * @type: The type of the inode
135 * non_block: Can we block on inodes that are being freed?
137 * Returns: A VFS inode, or an error
140 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
141 u64 no_addr, u64 no_formal_ino, int non_block)
144 struct gfs2_inode *ip;
145 struct gfs2_glock *io_gl = NULL;
148 inode = gfs2_iget(sb, no_addr, non_block);
152 return ERR_PTR(-ENOBUFS);
154 if (inode->i_state & I_NEW) {
155 struct gfs2_sbd *sdp = GFS2_SB(inode);
156 ip->i_no_formal_ino = no_formal_ino;
158 error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
161 ip->i_gl->gl_object = ip;
163 error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
167 set_bit(GIF_INVALID, &ip->i_flags);
168 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
172 ip->i_iopen_gh.gh_gl->gl_object = ip;
173 gfs2_glock_put(io_gl);
176 if (type == DT_UNKNOWN) {
177 /* Inode glock must be locked already */
178 error = gfs2_inode_refresh(GFS2_I(inode));
182 inode->i_mode = DT2IF(type);
186 unlock_new_inode(inode);
192 ip->i_iopen_gh.gh_gl->gl_object = NULL;
193 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
196 gfs2_glock_put(io_gl);
198 ip->i_gl->gl_object = NULL;
199 gfs2_glock_put(ip->i_gl);
202 return ERR_PTR(error);
205 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
206 u64 *no_formal_ino, unsigned int blktype)
208 struct super_block *sb = sdp->sd_vfs;
209 struct gfs2_holder i_gh;
210 struct inode *inode = NULL;
213 /* Must not read in block until block type is verified */
214 error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
215 LM_ST_EXCLUSIVE, GL_SKIP, &i_gh);
217 return ERR_PTR(error);
219 error = gfs2_check_blk_type(sdp, no_addr, blktype);
223 inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
227 /* Two extra checks for NFS only */
230 if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
234 if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
241 gfs2_glock_dq_uninit(&i_gh);
242 return error ? ERR_PTR(error) : inode;
249 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
253 gfs2_str2qstr(&qstr, name);
254 inode = gfs2_lookupi(dip, &qstr, 1);
255 /* gfs2_lookupi has inconsistent callers: vfs
256 * related routines expect NULL for no entry found,
257 * gfs2_lookup_simple callers expect ENOENT
258 * and do not check for NULL.
261 return ERR_PTR(-ENOENT);
268 * gfs2_lookupi - Look up a filename in a directory and return its inode
269 * @d_gh: An initialized holder for the directory glock
270 * @name: The name of the inode to look for
271 * @is_root: If 1, ignore the caller's permissions
272 * @i_gh: An uninitialized holder for the new inode glock
274 * This can be called via the VFS filldir function when NFS is doing
275 * a readdirplus and the inode which its intending to stat isn't
276 * already in cache. In this case we must not take the directory glock
277 * again, since the readdir call will have already taken that lock.
282 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
285 struct super_block *sb = dir->i_sb;
286 struct gfs2_inode *dip = GFS2_I(dir);
287 struct gfs2_holder d_gh;
289 struct inode *inode = NULL;
292 if (!name->len || name->len > GFS2_FNAMESIZE)
293 return ERR_PTR(-ENAMETOOLONG);
295 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
296 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
297 dir == sb->s_root->d_inode)) {
302 if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
303 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
305 return ERR_PTR(error);
310 error = gfs2_permission(dir, MAY_EXEC);
315 inode = gfs2_dir_search(dir, name);
317 error = PTR_ERR(inode);
320 gfs2_glock_dq_uninit(&d_gh);
321 if (error == -ENOENT)
323 return inode ? inode : ERR_PTR(error);
327 * create_ok - OK to create a new on-disk inode here?
328 * @dip: Directory in which dinode is to be created
329 * @name: Name of new dinode
335 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
340 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
344 /* Don't create entries in an unlinked directory */
345 if (!dip->i_inode.i_nlink)
348 error = gfs2_dir_check(&dip->i_inode, name, NULL);
359 if (dip->i_entries == (u32)-1)
361 if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
367 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
368 unsigned int *uid, unsigned int *gid)
370 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
371 (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
374 else if (dip->i_inode.i_uid != current_fsuid())
376 *uid = dip->i_inode.i_uid;
378 *uid = current_fsuid();
380 if (dip->i_inode.i_mode & S_ISGID) {
383 *gid = dip->i_inode.i_gid;
385 *gid = current_fsgid();
388 static int alloc_dinode(struct gfs2_inode *dip, u64 *no_addr, u64 *generation)
390 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
394 if (gfs2_alloc_get(dip) == NULL)
397 dip->i_alloc->al_requested = RES_DINODE;
398 error = gfs2_inplace_reserve(dip);
402 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
406 error = gfs2_alloc_blocks(dip, no_addr, &dblocks, 1, generation);
411 gfs2_inplace_release(dip);
417 static void gfs2_init_dir(struct buffer_head *dibh,
418 const struct gfs2_inode *parent)
420 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
421 struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
423 gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
424 dent->de_inum = di->di_num; /* already GFS2 endian */
425 dent->de_type = cpu_to_be16(DT_DIR);
427 dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
428 gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
429 gfs2_inum_out(parent, dent);
430 dent->de_type = cpu_to_be16(DT_DIR);
435 * init_dinode - Fill in a new dinode structure
436 * @dip: The directory this inode is being created in
437 * @gl: The glock covering the new inode
438 * @inum: The inode number
439 * @mode: The file permissions
440 * @uid: The uid of the new inode
441 * @gid: The gid of the new inode
442 * @generation: The generation number of the new inode
443 * @dev: The device number (if a device node)
444 * @symname: The symlink destination (if a symlink)
445 * @size: The inode size (ignored for directories)
446 * @bhp: The buffer head (returned to caller)
450 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
451 const struct gfs2_inum_host *inum, unsigned int mode,
452 unsigned int uid, unsigned int gid,
453 const u64 *generation, dev_t dev, const char *symname,
454 unsigned size, struct buffer_head **bhp)
456 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
457 struct gfs2_dinode *di;
458 struct buffer_head *dibh;
459 struct timespec tv = CURRENT_TIME;
461 dibh = gfs2_meta_new(gl, inum->no_addr);
462 gfs2_trans_add_bh(gl, dibh, 1);
463 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
464 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
465 di = (struct gfs2_dinode *)dibh->b_data;
467 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
468 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
469 di->di_mode = cpu_to_be32(mode);
470 di->di_uid = cpu_to_be32(uid);
471 di->di_gid = cpu_to_be32(gid);
473 di->di_size = cpu_to_be64(size);
474 di->di_blocks = cpu_to_be64(1);
475 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(tv.tv_sec);
476 di->di_major = cpu_to_be32(MAJOR(dev));
477 di->di_minor = cpu_to_be32(MINOR(dev));
478 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
479 di->di_generation = cpu_to_be64(*generation);
482 di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
488 memset(&di->__pad4, 0, sizeof(di->__pad4));
490 di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
491 di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
492 di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
493 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
495 switch(mode & S_IFMT) {
497 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
498 gfs2_tune_get(sdp, gt_new_files_jdata))
499 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
502 di->di_flags |= cpu_to_be32(dip->i_diskflags &
503 GFS2_DIF_INHERIT_JDATA);
504 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
505 di->di_size = cpu_to_be64(sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode));
506 di->di_entries = cpu_to_be32(2);
507 gfs2_init_dir(dibh, dip);
510 memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, size);
514 set_buffer_uptodate(dibh);
519 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
520 unsigned int mode, const struct gfs2_inum_host *inum,
521 const u64 *generation, dev_t dev, const char *symname,
522 unsigned int size, struct buffer_head **bhp)
524 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
525 unsigned int uid, gid;
528 munge_mode_uid_gid(dip, &mode, &uid, &gid);
529 if (!gfs2_alloc_get(dip))
532 error = gfs2_quota_lock(dip, uid, gid);
536 error = gfs2_quota_check(dip, uid, gid);
540 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
544 init_dinode(dip, gl, inum, mode, uid, gid, generation, dev, symname, size, bhp);
545 gfs2_quota_change(dip, +1, uid, gid);
549 gfs2_quota_unlock(dip);
555 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
556 struct gfs2_inode *ip)
558 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
559 struct gfs2_alloc *al;
561 struct buffer_head *dibh;
564 al = gfs2_alloc_get(dip);
568 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
572 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
573 if (alloc_required < 0)
574 goto fail_quota_locks;
575 if (alloc_required) {
576 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
578 goto fail_quota_locks;
580 al->al_requested = sdp->sd_max_dirres;
582 error = gfs2_inplace_reserve(dip);
584 goto fail_quota_locks;
586 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
587 dip->i_rgd->rd_length +
589 RES_STATFS + RES_QUOTA, 0);
593 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
595 goto fail_quota_locks;
598 error = gfs2_dir_add(&dip->i_inode, name, ip);
602 error = gfs2_meta_inode_buffer(ip, &dibh);
605 inc_nlink(&ip->i_inode);
606 if (S_ISDIR(ip->i_inode.i_mode))
607 inc_nlink(&ip->i_inode);
608 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
609 gfs2_dinode_out(ip, dibh->b_data);
617 gfs2_inplace_release(dip);
620 gfs2_quota_unlock(dip);
627 int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
630 const struct xattr *xattr;
633 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
634 err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
636 GFS2_EATYPE_SECURITY);
643 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
644 const struct qstr *qstr)
646 return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
647 &gfs2_initxattrs, NULL);
651 * gfs2_create_inode - Create a new inode
652 * @dir: The parent directory
653 * @dentry: The new dentry
654 * @mode: The permissions on the new inode
655 * @dev: For device nodes, this is the device number
656 * @symname: For symlinks, this is the link destination
657 * @size: The initial size of the inode (ignored for directories)
659 * Returns: 0 on success, or error code
662 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
663 unsigned int mode, dev_t dev, const char *symname,
664 unsigned int size, int excl)
666 const struct qstr *name = &dentry->d_name;
667 struct gfs2_holder ghs[2];
668 struct inode *inode = NULL;
669 struct gfs2_inode *dip = GFS2_I(dir);
670 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
671 struct gfs2_inum_host inum = { .no_addr = 0, .no_formal_ino = 0 };
674 struct buffer_head *bh = NULL;
676 if (!name->len || name->len > GFS2_FNAMESIZE)
677 return -ENAMETOOLONG;
679 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
683 error = create_ok(dip, name, mode);
684 if ((error == -EEXIST) && S_ISREG(mode) && !excl) {
685 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
686 gfs2_glock_dq_uninit(ghs);
687 d_instantiate(dentry, inode);
688 return IS_ERR(inode) ? PTR_ERR(inode) : 0;
693 error = alloc_dinode(dip, &inum.no_addr, &generation);
696 inum.no_formal_ino = generation;
698 error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
699 LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
703 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev, symname, size, &bh);
707 inode = gfs2_inode_lookup(dir->i_sb, IF2DT(mode), inum.no_addr,
708 inum.no_formal_ino, 0);
712 error = gfs2_inode_refresh(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 gfs2_inplace_release(dip);
733 gfs2_quota_unlock(dip);
735 mark_inode_dirty(inode);
736 gfs2_glock_dq_uninit_m(2, ghs);
737 d_instantiate(dentry, inode);
741 gfs2_glock_dq_uninit(ghs + 1);
743 gfs2_glock_dq_uninit(ghs);
744 if (inode && !IS_ERR(inode)) {
745 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
755 * gfs2_create - Create a file
756 * @dir: The directory in which to create the file
757 * @dentry: The dentry of the new file
758 * @mode: The mode of the new file
763 static int gfs2_create(struct inode *dir, struct dentry *dentry,
764 int mode, struct nameidata *nd)
767 if (nd && (nd->flags & LOOKUP_EXCL))
769 return gfs2_create_inode(dir, dentry, S_IFREG | mode, 0, NULL, 0, excl);
773 * gfs2_lookup - Look up a filename in a directory and return its inode
774 * @dir: The directory inode
775 * @dentry: The dentry of the new inode
776 * @nd: passed from Linux VFS, ignored by us
778 * Called by the VFS layer. Lock dir and call gfs2_lookupi()
783 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
784 struct nameidata *nd)
786 struct inode *inode = gfs2_lookupi(dir, &dentry->d_name, 0);
787 if (inode && !IS_ERR(inode)) {
788 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
789 struct gfs2_holder gh;
791 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
794 return ERR_PTR(error);
796 gfs2_glock_dq_uninit(&gh);
798 return d_splice_alias(inode, dentry);
802 * gfs2_link - Link to a file
803 * @old_dentry: The inode to link
804 * @dir: Add link to this directory
805 * @dentry: The name of the link
807 * Link the inode in "old_dentry" into the directory "dir" with the
813 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
814 struct dentry *dentry)
816 struct gfs2_inode *dip = GFS2_I(dir);
817 struct gfs2_sbd *sdp = GFS2_SB(dir);
818 struct inode *inode = old_dentry->d_inode;
819 struct gfs2_inode *ip = GFS2_I(inode);
820 struct gfs2_holder ghs[2];
821 struct buffer_head *dibh;
825 if (S_ISDIR(inode->i_mode))
828 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
829 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
831 error = gfs2_glock_nq(ghs); /* parent */
835 error = gfs2_glock_nq(ghs + 1); /* child */
840 if (inode->i_nlink == 0)
843 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
847 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
858 if (!dip->i_inode.i_nlink)
861 if (dip->i_entries == (u32)-1)
864 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
867 if (!ip->i_inode.i_nlink)
870 if (ip->i_inode.i_nlink == (u32)-1)
873 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
878 if (alloc_required) {
879 struct gfs2_alloc *al = gfs2_alloc_get(dip);
885 error = gfs2_quota_lock_check(dip);
889 al->al_requested = sdp->sd_max_dirres;
891 error = gfs2_inplace_reserve(dip);
895 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
896 gfs2_rg_blocks(dip) +
897 2 * RES_DINODE + RES_STATFS +
902 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
907 error = gfs2_meta_inode_buffer(ip, &dibh);
911 error = gfs2_dir_add(dir, &dentry->d_name, ip);
915 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
916 inc_nlink(&ip->i_inode);
917 ip->i_inode.i_ctime = CURRENT_TIME;
919 d_instantiate(dentry, inode);
920 mark_inode_dirty(inode);
928 gfs2_inplace_release(dip);
931 gfs2_quota_unlock(dip);
936 gfs2_glock_dq(ghs + 1);
940 gfs2_holder_uninit(ghs);
941 gfs2_holder_uninit(ghs + 1);
946 * gfs2_unlink_ok - check to see that a inode is still in a directory
947 * @dip: the directory
948 * @name: the name of the file
951 * Assumes that the lock on (at least) @dip is held.
953 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
956 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
957 const struct gfs2_inode *ip)
961 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
964 if ((dip->i_inode.i_mode & S_ISVTX) &&
965 dip->i_inode.i_uid != current_fsuid() &&
966 ip->i_inode.i_uid != current_fsuid() && !capable(CAP_FOWNER))
969 if (IS_APPEND(&dip->i_inode))
972 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
976 error = gfs2_dir_check(&dip->i_inode, name, ip);
984 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
985 * @dip: The parent directory
986 * @name: The name of the entry in the parent directory
987 * @bh: The inode buffer for the inode to be removed
988 * @inode: The inode to be removed
990 * Called with all the locks and in a transaction. This will only be
991 * called for a directory after it has been checked to ensure it is empty.
993 * Returns: 0 on success, or an error
996 static int gfs2_unlink_inode(struct gfs2_inode *dip,
997 const struct dentry *dentry,
998 struct buffer_head *bh)
1000 struct inode *inode = dentry->d_inode;
1001 struct gfs2_inode *ip = GFS2_I(inode);
1004 error = gfs2_dir_del(dip, dentry);
1009 inode->i_ctime = CURRENT_TIME;
1010 if (S_ISDIR(inode->i_mode))
1014 mark_inode_dirty(inode);
1015 if (inode->i_nlink == 0)
1016 gfs2_unlink_di(inode);
1022 * gfs2_unlink - Unlink an inode (this does rmdir as well)
1023 * @dir: The inode of the directory containing the inode to unlink
1024 * @dentry: The file itself
1026 * This routine uses the type of the inode as a flag to figure out
1027 * whether this is an unlink or an rmdir.
1032 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1034 struct gfs2_inode *dip = GFS2_I(dir);
1035 struct gfs2_sbd *sdp = GFS2_SB(dir);
1036 struct inode *inode = dentry->d_inode;
1037 struct gfs2_inode *ip = GFS2_I(inode);
1038 struct buffer_head *bh;
1039 struct gfs2_holder ghs[3];
1040 struct gfs2_rgrpd *rgd;
1043 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1044 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1046 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1049 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1052 error = gfs2_glock_nq(ghs); /* parent */
1056 error = gfs2_glock_nq(ghs + 1); /* child */
1061 if (inode->i_nlink == 0)
1064 if (S_ISDIR(inode->i_mode)) {
1066 if (ip->i_entries > 2 || inode->i_nlink > 2)
1070 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1074 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1078 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1082 error = gfs2_meta_inode_buffer(ip, &bh);
1086 error = gfs2_unlink_inode(dip, dentry, bh);
1090 gfs2_trans_end(sdp);
1092 gfs2_glock_dq(ghs + 2);
1094 gfs2_glock_dq(ghs + 1);
1098 gfs2_holder_uninit(ghs + 2);
1100 gfs2_holder_uninit(ghs + 1);
1101 gfs2_holder_uninit(ghs);
1106 * gfs2_symlink - Create a symlink
1107 * @dir: The directory to create the symlink in
1108 * @dentry: The dentry to put the symlink in
1109 * @symname: The thing which the link points to
1114 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1115 const char *symname)
1117 struct gfs2_sbd *sdp = GFS2_SB(dir);
1120 size = strlen(symname);
1121 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1122 return -ENAMETOOLONG;
1124 return gfs2_create_inode(dir, dentry, S_IFLNK | S_IRWXUGO, 0, symname, size, 0);
1128 * gfs2_mkdir - Make a directory
1129 * @dir: The parent directory of the new one
1130 * @dentry: The dentry of the new directory
1131 * @mode: The mode of the new directory
1136 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1138 return gfs2_create_inode(dir, dentry, S_IFDIR | mode, 0, NULL, 0, 0);
1142 * gfs2_mknod - Make a special file
1143 * @dir: The directory in which the special file will reside
1144 * @dentry: The dentry of the special file
1145 * @mode: The mode of the special file
1146 * @dev: The device specification of the special file
1150 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, int mode,
1153 return gfs2_create_inode(dir, dentry, mode, dev, NULL, 0, 0);
1157 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1161 * Follow @to back to the root and make sure we don't encounter @this
1162 * Assumes we already hold the rename lock.
1167 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1169 struct inode *dir = &to->i_inode;
1170 struct super_block *sb = dir->i_sb;
1177 if (dir == &this->i_inode) {
1181 if (dir == sb->s_root->d_inode) {
1186 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1188 error = PTR_ERR(tmp);
1202 * gfs2_rename - Rename a file
1203 * @odir: Parent directory of old file name
1204 * @odentry: The old dentry of the file
1205 * @ndir: Parent directory of new file name
1206 * @ndentry: The new dentry of the file
1211 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1212 struct inode *ndir, struct dentry *ndentry)
1214 struct gfs2_inode *odip = GFS2_I(odir);
1215 struct gfs2_inode *ndip = GFS2_I(ndir);
1216 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1217 struct gfs2_inode *nip = NULL;
1218 struct gfs2_sbd *sdp = GFS2_SB(odir);
1219 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1220 struct gfs2_rgrpd *nrgd;
1221 unsigned int num_gh;
1223 int alloc_required = 0;
1227 if (ndentry->d_inode) {
1228 nip = GFS2_I(ndentry->d_inode);
1234 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1239 if (S_ISDIR(ip->i_inode.i_mode)) {
1241 /* don't move a dirctory into it's subdir */
1242 error = gfs2_ok_to_move(ip, ndip);
1249 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1251 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1254 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1258 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1260 /* grab the resource lock for unlink flag twiddling
1261 * this is the case of the target file already existing
1262 * so we unlink before doing the rename
1264 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr);
1266 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1269 for (x = 0; x < num_gh; x++) {
1270 error = gfs2_glock_nq(ghs + x);
1276 if (ip->i_inode.i_nlink == 0)
1279 /* Check out the old directory */
1281 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1285 /* Check out the new directory */
1288 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1292 if (nip->i_inode.i_nlink == 0) {
1297 if (S_ISDIR(nip->i_inode.i_mode)) {
1298 if (nip->i_entries < 2) {
1299 gfs2_consist_inode(nip);
1303 if (nip->i_entries > 2) {
1309 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1313 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1325 if (!ndip->i_inode.i_nlink) {
1329 if (ndip->i_entries == (u32)-1) {
1333 if (S_ISDIR(ip->i_inode.i_mode) &&
1334 ndip->i_inode.i_nlink == (u32)-1) {
1341 /* Check out the dir to be renamed */
1344 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1350 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1351 error = alloc_required;
1356 if (alloc_required) {
1357 struct gfs2_alloc *al = gfs2_alloc_get(ndip);
1363 error = gfs2_quota_lock_check(ndip);
1367 al->al_requested = sdp->sd_max_dirres;
1369 error = gfs2_inplace_reserve(ndip);
1373 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1374 gfs2_rg_blocks(ndip) +
1375 4 * RES_DINODE + 4 * RES_LEAF +
1376 RES_STATFS + RES_QUOTA + 4, 0);
1380 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1381 5 * RES_LEAF + 4, 0);
1386 /* Remove the target file, if it exists */
1389 struct buffer_head *bh;
1390 error = gfs2_meta_inode_buffer(nip, &bh);
1393 error = gfs2_unlink_inode(ndip, ndentry, bh);
1398 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1402 struct buffer_head *dibh;
1403 error = gfs2_meta_inode_buffer(ip, &dibh);
1406 ip->i_inode.i_ctime = CURRENT_TIME;
1407 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1408 gfs2_dinode_out(ip, dibh->b_data);
1412 error = gfs2_dir_del(odip, odentry);
1416 error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1421 gfs2_trans_end(sdp);
1424 gfs2_inplace_release(ndip);
1427 gfs2_quota_unlock(ndip);
1430 gfs2_alloc_put(ndip);
1433 gfs2_glock_dq(ghs + x);
1434 gfs2_holder_uninit(ghs + x);
1438 gfs2_glock_dq_uninit(&r_gh);
1444 * gfs2_follow_link - Follow a symbolic link
1445 * @dentry: The dentry of the link
1446 * @nd: Data that we pass to vfs_follow_link()
1448 * This can handle symlinks of any size.
1450 * Returns: 0 on success or error code
1453 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1455 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1456 struct gfs2_holder i_gh;
1457 struct buffer_head *dibh;
1462 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1463 error = gfs2_glock_nq(&i_gh);
1465 gfs2_holder_uninit(&i_gh);
1466 nd_set_link(nd, ERR_PTR(error));
1470 size = (unsigned int)i_size_read(&ip->i_inode);
1472 gfs2_consist_inode(ip);
1473 buf = ERR_PTR(-EIO);
1477 error = gfs2_meta_inode_buffer(ip, &dibh);
1479 buf = ERR_PTR(error);
1483 buf = kzalloc(size + 1, GFP_NOFS);
1485 buf = ERR_PTR(-ENOMEM);
1487 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1490 gfs2_glock_dq_uninit(&i_gh);
1491 nd_set_link(nd, buf);
1495 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1497 char *s = nd_get_link(nd);
1505 * @mask: The mask to be tested
1506 * @flags: Indicates whether this is an RCU path walk or not
1508 * This may be called from the VFS directly, or from within GFS2 with the
1509 * inode locked, so we look to see if the glock is already locked and only
1510 * lock the glock if its not already been done.
1515 int gfs2_permission(struct inode *inode, int mask)
1517 struct gfs2_inode *ip;
1518 struct gfs2_holder i_gh;
1524 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1525 if (mask & MAY_NOT_BLOCK)
1527 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1533 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1536 error = generic_permission(inode, mask);
1538 gfs2_glock_dq_uninit(&i_gh);
1543 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1545 setattr_copy(inode, attr);
1546 mark_inode_dirty(inode);
1551 * gfs2_setattr_simple -
1558 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1562 if (current->journal_info)
1563 return __gfs2_setattr_simple(inode, attr);
1565 error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1569 error = __gfs2_setattr_simple(inode, attr);
1570 gfs2_trans_end(GFS2_SB(inode));
1574 static int setattr_chown(struct inode *inode, struct iattr *attr)
1576 struct gfs2_inode *ip = GFS2_I(inode);
1577 struct gfs2_sbd *sdp = GFS2_SB(inode);
1578 u32 ouid, ogid, nuid, ngid;
1581 ouid = inode->i_uid;
1582 ogid = inode->i_gid;
1583 nuid = attr->ia_uid;
1584 ngid = attr->ia_gid;
1586 if (!(attr->ia_valid & ATTR_UID) || ouid == nuid)
1587 ouid = nuid = NO_QUOTA_CHANGE;
1588 if (!(attr->ia_valid & ATTR_GID) || ogid == ngid)
1589 ogid = ngid = NO_QUOTA_CHANGE;
1591 if (!gfs2_alloc_get(ip))
1594 error = gfs2_quota_lock(ip, nuid, ngid);
1598 if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1599 error = gfs2_quota_check(ip, nuid, ngid);
1604 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1608 error = gfs2_setattr_simple(inode, attr);
1612 if (ouid != NO_QUOTA_CHANGE || ogid != NO_QUOTA_CHANGE) {
1613 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1614 gfs2_quota_change(ip, -blocks, ouid, ogid);
1615 gfs2_quota_change(ip, blocks, nuid, ngid);
1619 gfs2_trans_end(sdp);
1621 gfs2_quota_unlock(ip);
1628 * gfs2_setattr - Change attributes on an inode
1629 * @dentry: The dentry which is changing
1630 * @attr: The structure describing the change
1632 * The VFS layer wants to change one or more of an inodes attributes. Write
1633 * that change out to disk.
1638 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1640 struct inode *inode = dentry->d_inode;
1641 struct gfs2_inode *ip = GFS2_I(inode);
1642 struct gfs2_holder i_gh;
1645 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1650 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1653 error = inode_change_ok(inode, attr);
1657 if (attr->ia_valid & ATTR_SIZE)
1658 error = gfs2_setattr_size(inode, attr->ia_size);
1659 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1660 error = setattr_chown(inode, attr);
1661 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1662 error = gfs2_acl_chmod(ip, attr);
1664 error = gfs2_setattr_simple(inode, attr);
1668 mark_inode_dirty(inode);
1669 gfs2_glock_dq_uninit(&i_gh);
1674 * gfs2_getattr - Read out an inode's attributes
1675 * @mnt: The vfsmount the inode is being accessed from
1676 * @dentry: The dentry to stat
1677 * @stat: The inode's stats
1679 * This may be called from the VFS directly, or from within GFS2 with the
1680 * inode locked, so we look to see if the glock is already locked and only
1681 * lock the glock if its not already been done. Note that its the NFS
1682 * readdirplus operation which causes this to be called (from filldir)
1683 * with the glock already held.
1688 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1691 struct inode *inode = dentry->d_inode;
1692 struct gfs2_inode *ip = GFS2_I(inode);
1693 struct gfs2_holder gh;
1697 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1698 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1704 generic_fillattr(inode, stat);
1706 gfs2_glock_dq_uninit(&gh);
1711 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1712 const void *data, size_t size, int flags)
1714 struct inode *inode = dentry->d_inode;
1715 struct gfs2_inode *ip = GFS2_I(inode);
1716 struct gfs2_holder gh;
1719 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1720 ret = gfs2_glock_nq(&gh);
1722 ret = generic_setxattr(dentry, name, data, size, flags);
1725 gfs2_holder_uninit(&gh);
1729 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1730 void *data, size_t size)
1732 struct inode *inode = dentry->d_inode;
1733 struct gfs2_inode *ip = GFS2_I(inode);
1734 struct gfs2_holder gh;
1737 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1738 ret = gfs2_glock_nq(&gh);
1740 ret = generic_getxattr(dentry, name, data, size);
1743 gfs2_holder_uninit(&gh);
1747 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1749 struct inode *inode = dentry->d_inode;
1750 struct gfs2_inode *ip = GFS2_I(inode);
1751 struct gfs2_holder gh;
1754 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1755 ret = gfs2_glock_nq(&gh);
1757 ret = generic_removexattr(dentry, name);
1760 gfs2_holder_uninit(&gh);
1764 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1767 struct gfs2_inode *ip = GFS2_I(inode);
1768 struct gfs2_holder gh;
1771 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1775 mutex_lock(&inode->i_mutex);
1777 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1781 if (gfs2_is_stuffed(ip)) {
1782 u64 phys = ip->i_no_addr << inode->i_blkbits;
1783 u64 size = i_size_read(inode);
1784 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1785 FIEMAP_EXTENT_DATA_INLINE;
1786 phys += sizeof(struct gfs2_dinode);
1788 if (start + len > size)
1791 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1796 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1800 gfs2_glock_dq_uninit(&gh);
1802 mutex_unlock(&inode->i_mutex);
1806 const struct inode_operations gfs2_file_iops = {
1807 .permission = gfs2_permission,
1808 .setattr = gfs2_setattr,
1809 .getattr = gfs2_getattr,
1810 .setxattr = gfs2_setxattr,
1811 .getxattr = gfs2_getxattr,
1812 .listxattr = gfs2_listxattr,
1813 .removexattr = gfs2_removexattr,
1814 .fiemap = gfs2_fiemap,
1815 .get_acl = gfs2_get_acl,
1818 const struct inode_operations gfs2_dir_iops = {
1819 .create = gfs2_create,
1820 .lookup = gfs2_lookup,
1822 .unlink = gfs2_unlink,
1823 .symlink = gfs2_symlink,
1824 .mkdir = gfs2_mkdir,
1825 .rmdir = gfs2_unlink,
1826 .mknod = gfs2_mknod,
1827 .rename = gfs2_rename,
1828 .permission = gfs2_permission,
1829 .setattr = gfs2_setattr,
1830 .getattr = gfs2_getattr,
1831 .setxattr = gfs2_setxattr,
1832 .getxattr = gfs2_getxattr,
1833 .listxattr = gfs2_listxattr,
1834 .removexattr = gfs2_removexattr,
1835 .fiemap = gfs2_fiemap,
1836 .get_acl = gfs2_get_acl,
1839 const struct inode_operations gfs2_symlink_iops = {
1840 .readlink = generic_readlink,
1841 .follow_link = gfs2_follow_link,
1842 .put_link = gfs2_put_link,
1843 .permission = gfs2_permission,
1844 .setattr = gfs2_setattr,
1845 .getattr = gfs2_getattr,
1846 .setxattr = gfs2_setxattr,
1847 .getxattr = gfs2_getxattr,
1848 .listxattr = gfs2_listxattr,
1849 .removexattr = gfs2_removexattr,
1850 .fiemap = gfs2_fiemap,
1851 .get_acl = gfs2_get_acl,