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, false);
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 if (dip->i_entries == (u32)-1)
351 if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
357 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
360 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
361 (dip->i_inode.i_mode & S_ISUID) &&
362 !uid_eq(dip->i_inode.i_uid, GLOBAL_ROOT_UID)) {
363 if (S_ISDIR(inode->i_mode))
364 inode->i_mode |= S_ISUID;
365 else if (!uid_eq(dip->i_inode.i_uid, current_fsuid()))
366 inode->i_mode &= ~07111;
367 inode->i_uid = dip->i_inode.i_uid;
369 inode->i_uid = current_fsuid();
371 if (dip->i_inode.i_mode & S_ISGID) {
372 if (S_ISDIR(inode->i_mode))
373 inode->i_mode |= S_ISGID;
374 inode->i_gid = dip->i_inode.i_gid;
376 inode->i_gid = current_fsgid();
379 static int alloc_dinode(struct gfs2_inode *ip, u32 flags)
381 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
382 struct gfs2_alloc_parms ap = { .target = RES_DINODE, .aflags = flags, };
386 error = gfs2_quota_lock_check(ip);
390 error = gfs2_inplace_reserve(ip, &ap);
394 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 0);
398 error = gfs2_alloc_blocks(ip, &ip->i_no_addr, &dblocks, 1, &ip->i_generation);
399 ip->i_no_formal_ino = ip->i_generation;
400 ip->i_inode.i_ino = ip->i_no_addr;
401 ip->i_goal = ip->i_no_addr;
406 gfs2_inplace_release(ip);
408 gfs2_quota_unlock(ip);
413 static void gfs2_init_dir(struct buffer_head *dibh,
414 const struct gfs2_inode *parent)
416 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
417 struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
419 gfs2_qstr2dirent(&gfs2_qdot, GFS2_DIRENT_SIZE(gfs2_qdot.len), dent);
420 dent->de_inum = di->di_num; /* already GFS2 endian */
421 dent->de_type = cpu_to_be16(DT_DIR);
423 dent = (struct gfs2_dirent *)((char*)dent + GFS2_DIRENT_SIZE(1));
424 gfs2_qstr2dirent(&gfs2_qdotdot, dibh->b_size - GFS2_DIRENT_SIZE(1) - sizeof(struct gfs2_dinode), dent);
425 gfs2_inum_out(parent, dent);
426 dent->de_type = cpu_to_be16(DT_DIR);
431 * init_dinode - Fill in a new dinode structure
432 * @dip: The directory this inode is being created in
434 * @symname: The symlink destination (if a symlink)
435 * @bhp: The buffer head (returned to caller)
439 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
442 struct gfs2_dinode *di;
443 struct buffer_head *dibh;
445 dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
446 gfs2_trans_add_meta(ip->i_gl, dibh);
447 di = (struct gfs2_dinode *)dibh->b_data;
448 gfs2_dinode_out(ip, di);
450 di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
451 di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
455 memset(&di->__pad4, 0, sizeof(di->__pad4));
456 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
457 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
459 switch(ip->i_inode.i_mode & S_IFMT) {
461 gfs2_init_dir(dibh, dip);
464 memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
468 set_buffer_uptodate(dibh);
472 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
473 struct gfs2_inode *ip, int arq)
475 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
476 struct gfs2_alloc_parms ap = { .target = sdp->sd_max_dirres, };
480 error = gfs2_quota_lock_check(dip);
482 goto fail_quota_locks;
484 error = gfs2_inplace_reserve(dip, &ap);
486 goto fail_quota_locks;
488 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
489 dip->i_rgd->rd_length +
491 RES_STATFS + RES_QUOTA, 0);
495 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
497 goto fail_quota_locks;
500 error = gfs2_dir_add(&dip->i_inode, name, ip);
507 gfs2_inplace_release(dip);
509 gfs2_quota_unlock(dip);
513 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
516 const struct xattr *xattr;
519 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
520 err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
522 GFS2_EATYPE_SECURITY);
529 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
530 const struct qstr *qstr)
532 return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
533 &gfs2_initxattrs, NULL);
537 * gfs2_create_inode - Create a new inode
538 * @dir: The parent directory
539 * @dentry: The new dentry
540 * @file: If non-NULL, the file which is being opened
541 * @mode: The permissions on the new inode
542 * @dev: For device nodes, this is the device number
543 * @symname: For symlinks, this is the link destination
544 * @size: The initial size of the inode (ignored for directories)
546 * Returns: 0 on success, or error code
549 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
551 umode_t mode, dev_t dev, const char *symname,
552 unsigned int size, int excl, int *opened)
554 const struct qstr *name = &dentry->d_name;
555 struct gfs2_holder ghs[2];
556 struct inode *inode = NULL;
557 struct gfs2_inode *dip = GFS2_I(dir), *ip;
558 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
559 struct gfs2_glock *io_gl;
565 if (!name->len || name->len > GFS2_FNAMESIZE)
566 return -ENAMETOOLONG;
568 error = gfs2_rs_alloc(dip);
572 error = gfs2_rindex_update(sdp);
576 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
580 error = create_ok(dip, name, mode);
584 inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
585 error = PTR_ERR(inode);
586 if (!IS_ERR(inode)) {
587 d = d_splice_alias(inode, dentry);
590 if (S_ISREG(inode->i_mode)) {
592 error = finish_open(file, dentry, gfs2_open_common, opened);
594 error = finish_no_open(file, d);
599 gfs2_glock_dq_uninit(ghs);
601 } else if (error != -ENOENT) {
605 arq = error = gfs2_diradd_alloc_required(dir, name);
609 inode = new_inode(sdp->sd_vfs);
615 error = gfs2_rs_alloc(ip);
617 goto fail_free_inode;
619 inode->i_mode = mode;
620 set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
622 inode->i_size = size;
623 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
624 gfs2_set_inode_blocks(inode, 1);
625 munge_mode_uid_gid(dip, inode);
626 ip->i_goal = dip->i_goal;
633 switch(mode & S_IFMT) {
635 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
636 gfs2_tune_get(sdp, gt_new_files_jdata))
637 ip->i_diskflags |= GFS2_DIF_JDATA;
638 gfs2_set_aops(inode);
641 ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
642 ip->i_diskflags |= GFS2_DIF_JDATA;
646 gfs2_set_inode_flags(inode);
648 if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
649 (dip->i_diskflags & GFS2_DIF_TOPDIR))
650 aflags |= GFS2_AF_ORLOV;
652 error = alloc_dinode(ip, aflags);
654 goto fail_free_inode;
656 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
658 goto fail_free_inode;
660 ip->i_gl->gl_object = ip;
661 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
663 goto fail_free_inode;
665 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
669 init_dinode(dip, ip, symname);
672 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
676 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
680 ip->i_iopen_gh.gh_gl->gl_object = ip;
681 gfs2_glock_put(io_gl);
683 insert_inode_hash(inode);
685 error = gfs2_acl_create(dip, inode);
689 error = gfs2_security_init(dip, ip, name);
693 error = link_dinode(dip, name, ip, arq);
697 mark_inode_dirty(inode);
698 d_instantiate(dentry, inode);
700 *opened |= FILE_CREATED;
701 error = finish_open(file, dentry, gfs2_open_common, opened);
703 gfs2_glock_dq_uninit(ghs);
704 gfs2_glock_dq_uninit(ghs + 1);
708 gfs2_glock_dq_uninit(ghs + 1);
710 gfs2_glock_put(ip->i_gl);
714 gfs2_glock_dq_uninit(ghs + 1);
717 gfs2_glock_put(ip->i_gl);
718 gfs2_rs_delete(ip, NULL);
719 free_inode_nonrcu(inode);
722 gfs2_glock_dq_uninit(ghs);
723 if (inode && !IS_ERR(inode)) {
725 mark_inode_dirty(inode);
726 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
734 * gfs2_create - Create a file
735 * @dir: The directory in which to create the file
736 * @dentry: The dentry of the new file
737 * @mode: The mode of the new file
742 static int gfs2_create(struct inode *dir, struct dentry *dentry,
743 umode_t mode, bool excl)
745 return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
749 * __gfs2_lookup - Look up a filename in a directory and return its inode
750 * @dir: The directory inode
751 * @dentry: The dentry of the new inode
752 * @file: File to be opened
753 * @opened: atomic_open flags
759 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
760 struct file *file, int *opened)
764 struct gfs2_holder gh;
765 struct gfs2_glock *gl;
768 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
772 return ERR_CAST(inode);
774 gl = GFS2_I(inode)->i_gl;
775 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
778 return ERR_PTR(error);
781 d = d_splice_alias(inode, dentry);
782 if (file && S_ISREG(inode->i_mode))
783 error = finish_open(file, dentry, gfs2_open_common, opened);
785 gfs2_glock_dq_uninit(&gh);
788 return ERR_PTR(error);
793 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
796 return __gfs2_lookup(dir, dentry, NULL, NULL);
800 * gfs2_link - Link to a file
801 * @old_dentry: The inode to link
802 * @dir: Add link to this directory
803 * @dentry: The name of the link
805 * Link the inode in "old_dentry" into the directory "dir" with the
811 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
812 struct dentry *dentry)
814 struct gfs2_inode *dip = GFS2_I(dir);
815 struct gfs2_sbd *sdp = GFS2_SB(dir);
816 struct inode *inode = old_dentry->d_inode;
817 struct gfs2_inode *ip = GFS2_I(inode);
818 struct gfs2_holder ghs[2];
819 struct buffer_head *dibh;
823 if (S_ISDIR(inode->i_mode))
826 error = gfs2_rs_alloc(dip);
830 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
831 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
833 error = gfs2_glock_nq(ghs); /* parent */
837 error = gfs2_glock_nq(ghs + 1); /* child */
842 if (inode->i_nlink == 0)
845 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
849 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
860 if (!dip->i_inode.i_nlink)
863 if (dip->i_entries == (u32)-1)
866 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
869 if (!ip->i_inode.i_nlink)
872 if (ip->i_inode.i_nlink == (u32)-1)
875 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
880 if (alloc_required) {
881 struct gfs2_alloc_parms ap = { .target = sdp->sd_max_dirres, };
882 error = gfs2_quota_lock_check(dip);
886 error = gfs2_inplace_reserve(dip, &ap);
890 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
891 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
892 2 * RES_DINODE + RES_STATFS +
897 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
902 error = gfs2_meta_inode_buffer(ip, &dibh);
906 error = gfs2_dir_add(dir, &dentry->d_name, ip);
910 gfs2_trans_add_meta(ip->i_gl, dibh);
911 inc_nlink(&ip->i_inode);
912 ip->i_inode.i_ctime = CURRENT_TIME;
914 d_instantiate(dentry, inode);
915 mark_inode_dirty(inode);
923 gfs2_inplace_release(dip);
926 gfs2_quota_unlock(dip);
928 gfs2_glock_dq(ghs + 1);
932 gfs2_holder_uninit(ghs);
933 gfs2_holder_uninit(ghs + 1);
938 * gfs2_unlink_ok - check to see that a inode is still in a directory
939 * @dip: the directory
940 * @name: the name of the file
943 * Assumes that the lock on (at least) @dip is held.
945 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
948 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
949 const struct gfs2_inode *ip)
953 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
956 if ((dip->i_inode.i_mode & S_ISVTX) &&
957 !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
958 !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
961 if (IS_APPEND(&dip->i_inode))
964 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
968 error = gfs2_dir_check(&dip->i_inode, name, ip);
976 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
977 * @dip: The parent directory
978 * @name: The name of the entry in the parent directory
979 * @inode: The inode to be removed
981 * Called with all the locks and in a transaction. This will only be
982 * called for a directory after it has been checked to ensure it is empty.
984 * Returns: 0 on success, or an error
987 static int gfs2_unlink_inode(struct gfs2_inode *dip,
988 const struct dentry *dentry)
990 struct inode *inode = dentry->d_inode;
991 struct gfs2_inode *ip = GFS2_I(inode);
994 error = gfs2_dir_del(dip, dentry);
999 inode->i_ctime = CURRENT_TIME;
1000 if (S_ISDIR(inode->i_mode))
1004 mark_inode_dirty(inode);
1005 if (inode->i_nlink == 0)
1006 gfs2_unlink_di(inode);
1012 * gfs2_unlink - Unlink an inode (this does rmdir as well)
1013 * @dir: The inode of the directory containing the inode to unlink
1014 * @dentry: The file itself
1016 * This routine uses the type of the inode as a flag to figure out
1017 * whether this is an unlink or an rmdir.
1022 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1024 struct gfs2_inode *dip = GFS2_I(dir);
1025 struct gfs2_sbd *sdp = GFS2_SB(dir);
1026 struct inode *inode = dentry->d_inode;
1027 struct gfs2_inode *ip = GFS2_I(inode);
1028 struct gfs2_holder ghs[3];
1029 struct gfs2_rgrpd *rgd;
1032 error = gfs2_rindex_update(sdp);
1038 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1039 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1041 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1045 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1048 error = gfs2_glock_nq(ghs); /* parent */
1052 error = gfs2_glock_nq(ghs + 1); /* child */
1057 if (inode->i_nlink == 0)
1060 if (S_ISDIR(inode->i_mode)) {
1062 if (ip->i_entries > 2 || inode->i_nlink > 2)
1066 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1070 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1074 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1078 error = gfs2_unlink_inode(dip, dentry);
1081 gfs2_trans_end(sdp);
1083 gfs2_glock_dq(ghs + 2);
1085 gfs2_glock_dq(ghs + 1);
1089 gfs2_holder_uninit(ghs + 2);
1091 gfs2_holder_uninit(ghs + 1);
1092 gfs2_holder_uninit(ghs);
1097 * gfs2_symlink - Create a symlink
1098 * @dir: The directory to create the symlink in
1099 * @dentry: The dentry to put the symlink in
1100 * @symname: The thing which the link points to
1105 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1106 const char *symname)
1108 struct gfs2_sbd *sdp = GFS2_SB(dir);
1111 size = strlen(symname);
1112 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1113 return -ENAMETOOLONG;
1115 return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1119 * gfs2_mkdir - Make a directory
1120 * @dir: The parent directory of the new one
1121 * @dentry: The dentry of the new directory
1122 * @mode: The mode of the new directory
1127 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1129 struct gfs2_sbd *sdp = GFS2_SB(dir);
1130 unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1131 return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1135 * gfs2_mknod - Make a special file
1136 * @dir: The directory in which the special file will reside
1137 * @dentry: The dentry of the special file
1138 * @mode: The mode of the special file
1139 * @dev: The device specification of the special file
1143 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1146 return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1150 * gfs2_atomic_open - Atomically open a file
1151 * @dir: The directory
1152 * @dentry: The proposed new entry
1153 * @file: The proposed new struct file
1154 * @flags: open flags
1156 * @opened: Flag to say whether the file has been opened or not
1158 * Returns: error code or 0 for success
1161 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1162 struct file *file, unsigned flags,
1163 umode_t mode, int *opened)
1166 bool excl = !!(flags & O_EXCL);
1168 d = __gfs2_lookup(dir, dentry, file, opened);
1173 if (dentry->d_inode) {
1174 if (!(*opened & FILE_OPENED))
1175 return finish_no_open(file, dentry);
1181 if (!(flags & O_CREAT))
1184 return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1188 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1192 * Follow @to back to the root and make sure we don't encounter @this
1193 * Assumes we already hold the rename lock.
1198 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1200 struct inode *dir = &to->i_inode;
1201 struct super_block *sb = dir->i_sb;
1208 if (dir == &this->i_inode) {
1212 if (dir == sb->s_root->d_inode) {
1217 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1219 error = PTR_ERR(tmp);
1233 * gfs2_rename - Rename a file
1234 * @odir: Parent directory of old file name
1235 * @odentry: The old dentry of the file
1236 * @ndir: Parent directory of new file name
1237 * @ndentry: The new dentry of the file
1242 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1243 struct inode *ndir, struct dentry *ndentry)
1245 struct gfs2_inode *odip = GFS2_I(odir);
1246 struct gfs2_inode *ndip = GFS2_I(ndir);
1247 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1248 struct gfs2_inode *nip = NULL;
1249 struct gfs2_sbd *sdp = GFS2_SB(odir);
1250 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1251 struct gfs2_rgrpd *nrgd;
1252 unsigned int num_gh;
1254 int alloc_required = 0;
1258 if (ndentry->d_inode) {
1259 nip = GFS2_I(ndentry->d_inode);
1264 error = gfs2_rindex_update(sdp);
1268 error = gfs2_rs_alloc(ndip);
1273 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1278 if (S_ISDIR(ip->i_inode.i_mode)) {
1280 /* don't move a dirctory into it's subdir */
1281 error = gfs2_ok_to_move(ip, ndip);
1288 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1290 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1293 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1297 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1299 /* grab the resource lock for unlink flag twiddling
1300 * this is the case of the target file already existing
1301 * so we unlink before doing the rename
1303 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1305 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1308 for (x = 0; x < num_gh; x++) {
1309 error = gfs2_glock_nq(ghs + x);
1315 if (ip->i_inode.i_nlink == 0)
1318 /* Check out the old directory */
1320 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1324 /* Check out the new directory */
1327 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1331 if (nip->i_inode.i_nlink == 0) {
1336 if (S_ISDIR(nip->i_inode.i_mode)) {
1337 if (nip->i_entries < 2) {
1338 gfs2_consist_inode(nip);
1342 if (nip->i_entries > 2) {
1348 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1352 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1364 if (!ndip->i_inode.i_nlink) {
1368 if (ndip->i_entries == (u32)-1) {
1372 if (S_ISDIR(ip->i_inode.i_mode) &&
1373 ndip->i_inode.i_nlink == (u32)-1) {
1380 /* Check out the dir to be renamed */
1383 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1389 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1390 error = alloc_required;
1394 if (alloc_required) {
1395 struct gfs2_alloc_parms ap = { .target = sdp->sd_max_dirres, };
1396 error = gfs2_quota_lock_check(ndip);
1400 error = gfs2_inplace_reserve(ndip, &ap);
1404 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1405 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1406 4 * RES_DINODE + 4 * RES_LEAF +
1407 RES_STATFS + RES_QUOTA + 4, 0);
1411 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1412 5 * RES_LEAF + 4, 0);
1417 /* Remove the target file, if it exists */
1420 error = gfs2_unlink_inode(ndip, ndentry);
1423 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1427 struct buffer_head *dibh;
1428 error = gfs2_meta_inode_buffer(ip, &dibh);
1431 ip->i_inode.i_ctime = CURRENT_TIME;
1432 gfs2_trans_add_meta(ip->i_gl, dibh);
1433 gfs2_dinode_out(ip, dibh->b_data);
1437 error = gfs2_dir_del(odip, odentry);
1441 error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1446 gfs2_trans_end(sdp);
1449 gfs2_inplace_release(ndip);
1452 gfs2_quota_unlock(ndip);
1455 gfs2_glock_dq(ghs + x);
1456 gfs2_holder_uninit(ghs + x);
1460 gfs2_glock_dq_uninit(&r_gh);
1466 * gfs2_follow_link - Follow a symbolic link
1467 * @dentry: The dentry of the link
1468 * @nd: Data that we pass to vfs_follow_link()
1470 * This can handle symlinks of any size.
1472 * Returns: 0 on success or error code
1475 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1477 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1478 struct gfs2_holder i_gh;
1479 struct buffer_head *dibh;
1484 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1485 error = gfs2_glock_nq(&i_gh);
1487 gfs2_holder_uninit(&i_gh);
1488 nd_set_link(nd, ERR_PTR(error));
1492 size = (unsigned int)i_size_read(&ip->i_inode);
1494 gfs2_consist_inode(ip);
1495 buf = ERR_PTR(-EIO);
1499 error = gfs2_meta_inode_buffer(ip, &dibh);
1501 buf = ERR_PTR(error);
1505 buf = kzalloc(size + 1, GFP_NOFS);
1507 buf = ERR_PTR(-ENOMEM);
1509 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1512 gfs2_glock_dq_uninit(&i_gh);
1513 nd_set_link(nd, buf);
1520 * @mask: The mask to be tested
1521 * @flags: Indicates whether this is an RCU path walk or not
1523 * This may be called from the VFS directly, or from within GFS2 with the
1524 * inode locked, so we look to see if the glock is already locked and only
1525 * lock the glock if its not already been done.
1530 int gfs2_permission(struct inode *inode, int mask)
1532 struct gfs2_inode *ip;
1533 struct gfs2_holder i_gh;
1539 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1540 if (mask & MAY_NOT_BLOCK)
1542 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1548 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1551 error = generic_permission(inode, mask);
1553 gfs2_glock_dq_uninit(&i_gh);
1558 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1560 setattr_copy(inode, attr);
1561 mark_inode_dirty(inode);
1566 * gfs2_setattr_simple -
1573 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1577 if (current->journal_info)
1578 return __gfs2_setattr_simple(inode, attr);
1580 error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1584 error = __gfs2_setattr_simple(inode, attr);
1585 gfs2_trans_end(GFS2_SB(inode));
1589 static int setattr_chown(struct inode *inode, struct iattr *attr)
1591 struct gfs2_inode *ip = GFS2_I(inode);
1592 struct gfs2_sbd *sdp = GFS2_SB(inode);
1597 ouid = inode->i_uid;
1598 ogid = inode->i_gid;
1599 nuid = attr->ia_uid;
1600 ngid = attr->ia_gid;
1602 if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1603 ouid = nuid = NO_UID_QUOTA_CHANGE;
1604 if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1605 ogid = ngid = NO_GID_QUOTA_CHANGE;
1607 error = gfs2_quota_lock(ip, nuid, ngid);
1611 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1612 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1613 error = gfs2_quota_check(ip, nuid, ngid);
1618 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1622 error = gfs2_setattr_simple(inode, attr);
1626 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1627 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1628 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1629 gfs2_quota_change(ip, -blocks, ouid, ogid);
1630 gfs2_quota_change(ip, blocks, nuid, ngid);
1634 gfs2_trans_end(sdp);
1636 gfs2_quota_unlock(ip);
1641 * gfs2_setattr - Change attributes on an inode
1642 * @dentry: The dentry which is changing
1643 * @attr: The structure describing the change
1645 * The VFS layer wants to change one or more of an inodes attributes. Write
1646 * that change out to disk.
1651 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1653 struct inode *inode = dentry->d_inode;
1654 struct gfs2_inode *ip = GFS2_I(inode);
1655 struct gfs2_holder i_gh;
1658 error = gfs2_rs_alloc(ip);
1662 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1667 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1670 error = inode_change_ok(inode, attr);
1674 if (attr->ia_valid & ATTR_SIZE)
1675 error = gfs2_setattr_size(inode, attr->ia_size);
1676 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1677 error = setattr_chown(inode, attr);
1678 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1679 error = gfs2_acl_chmod(ip, attr);
1681 error = gfs2_setattr_simple(inode, attr);
1685 mark_inode_dirty(inode);
1686 gfs2_glock_dq_uninit(&i_gh);
1691 * gfs2_getattr - Read out an inode's attributes
1692 * @mnt: The vfsmount the inode is being accessed from
1693 * @dentry: The dentry to stat
1694 * @stat: The inode's stats
1696 * This may be called from the VFS directly, or from within GFS2 with the
1697 * inode locked, so we look to see if the glock is already locked and only
1698 * lock the glock if its not already been done. Note that its the NFS
1699 * readdirplus operation which causes this to be called (from filldir)
1700 * with the glock already held.
1705 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1708 struct inode *inode = dentry->d_inode;
1709 struct gfs2_inode *ip = GFS2_I(inode);
1710 struct gfs2_holder gh;
1714 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1715 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1721 generic_fillattr(inode, stat);
1723 gfs2_glock_dq_uninit(&gh);
1728 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1729 const void *data, size_t size, int flags)
1731 struct inode *inode = dentry->d_inode;
1732 struct gfs2_inode *ip = GFS2_I(inode);
1733 struct gfs2_holder gh;
1736 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1737 ret = gfs2_glock_nq(&gh);
1739 ret = gfs2_rs_alloc(ip);
1741 ret = generic_setxattr(dentry, name, data, size, flags);
1744 gfs2_holder_uninit(&gh);
1748 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1749 void *data, size_t size)
1751 struct inode *inode = dentry->d_inode;
1752 struct gfs2_inode *ip = GFS2_I(inode);
1753 struct gfs2_holder gh;
1756 /* For selinux during lookup */
1757 if (gfs2_glock_is_locked_by_me(ip->i_gl))
1758 return generic_getxattr(dentry, name, data, size);
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 = gfs2_rs_alloc(ip);
1782 ret = generic_removexattr(dentry, name);
1785 gfs2_holder_uninit(&gh);
1789 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1792 struct gfs2_inode *ip = GFS2_I(inode);
1793 struct gfs2_holder gh;
1796 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1800 mutex_lock(&inode->i_mutex);
1802 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1806 if (gfs2_is_stuffed(ip)) {
1807 u64 phys = ip->i_no_addr << inode->i_blkbits;
1808 u64 size = i_size_read(inode);
1809 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1810 FIEMAP_EXTENT_DATA_INLINE;
1811 phys += sizeof(struct gfs2_dinode);
1813 if (start + len > size)
1816 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1821 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1825 gfs2_glock_dq_uninit(&gh);
1827 mutex_unlock(&inode->i_mutex);
1831 const struct inode_operations gfs2_file_iops = {
1832 .permission = gfs2_permission,
1833 .setattr = gfs2_setattr,
1834 .getattr = gfs2_getattr,
1835 .setxattr = gfs2_setxattr,
1836 .getxattr = gfs2_getxattr,
1837 .listxattr = gfs2_listxattr,
1838 .removexattr = gfs2_removexattr,
1839 .fiemap = gfs2_fiemap,
1840 .get_acl = gfs2_get_acl,
1843 const struct inode_operations gfs2_dir_iops = {
1844 .create = gfs2_create,
1845 .lookup = gfs2_lookup,
1847 .unlink = gfs2_unlink,
1848 .symlink = gfs2_symlink,
1849 .mkdir = gfs2_mkdir,
1850 .rmdir = gfs2_unlink,
1851 .mknod = gfs2_mknod,
1852 .rename = gfs2_rename,
1853 .permission = gfs2_permission,
1854 .setattr = gfs2_setattr,
1855 .getattr = gfs2_getattr,
1856 .setxattr = gfs2_setxattr,
1857 .getxattr = gfs2_getxattr,
1858 .listxattr = gfs2_listxattr,
1859 .removexattr = gfs2_removexattr,
1860 .fiemap = gfs2_fiemap,
1861 .get_acl = gfs2_get_acl,
1862 .atomic_open = gfs2_atomic_open,
1865 const struct inode_operations gfs2_symlink_iops = {
1866 .readlink = generic_readlink,
1867 .follow_link = gfs2_follow_link,
1868 .put_link = kfree_put_link,
1869 .permission = gfs2_permission,
1870 .setattr = gfs2_setattr,
1871 .getattr = gfs2_getattr,
1872 .setxattr = gfs2_setxattr,
1873 .getxattr = gfs2_getxattr,
1874 .listxattr = gfs2_listxattr,
1875 .removexattr = gfs2_removexattr,
1876 .fiemap = gfs2_fiemap,
1877 .get_acl = gfs2_get_acl,