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);
385 error = gfs2_quota_lock_check(ip);
389 error = gfs2_inplace_reserve(ip, RES_DINODE, flags);
393 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 0);
397 error = gfs2_alloc_blocks(ip, &ip->i_no_addr, &dblocks, 1, &ip->i_generation);
398 ip->i_no_formal_ino = ip->i_generation;
399 ip->i_inode.i_ino = ip->i_no_addr;
400 ip->i_goal = ip->i_no_addr;
405 gfs2_inplace_release(ip);
407 gfs2_quota_unlock(ip);
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
433 * @symname: The symlink destination (if a symlink)
434 * @bhp: The buffer head (returned to caller)
438 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
441 struct gfs2_dinode *di;
442 struct buffer_head *dibh;
444 dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
445 gfs2_trans_add_meta(ip->i_gl, dibh);
446 di = (struct gfs2_dinode *)dibh->b_data;
447 gfs2_dinode_out(ip, di);
449 di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
450 di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
454 memset(&di->__pad4, 0, sizeof(di->__pad4));
455 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
456 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
458 switch(ip->i_inode.i_mode & S_IFMT) {
460 gfs2_init_dir(dibh, dip);
463 memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
467 set_buffer_uptodate(dibh);
471 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
472 struct gfs2_inode *ip, int arq)
474 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
478 error = gfs2_quota_lock_check(dip);
480 goto fail_quota_locks;
482 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
484 goto fail_quota_locks;
486 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
487 dip->i_rgd->rd_length +
489 RES_STATFS + RES_QUOTA, 0);
493 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
495 goto fail_quota_locks;
498 error = gfs2_dir_add(&dip->i_inode, name, ip);
505 gfs2_inplace_release(dip);
507 gfs2_quota_unlock(dip);
511 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
514 const struct xattr *xattr;
517 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
518 err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
520 GFS2_EATYPE_SECURITY);
527 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
528 const struct qstr *qstr)
530 return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
531 &gfs2_initxattrs, NULL);
535 * gfs2_create_inode - Create a new inode
536 * @dir: The parent directory
537 * @dentry: The new dentry
538 * @file: If non-NULL, the file which is being opened
539 * @mode: The permissions on the new inode
540 * @dev: For device nodes, this is the device number
541 * @symname: For symlinks, this is the link destination
542 * @size: The initial size of the inode (ignored for directories)
544 * Returns: 0 on success, or error code
547 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
549 umode_t mode, dev_t dev, const char *symname,
550 unsigned int size, int excl, int *opened)
552 const struct qstr *name = &dentry->d_name;
553 struct gfs2_holder ghs[2];
554 struct inode *inode = NULL;
555 struct gfs2_inode *dip = GFS2_I(dir), *ip;
556 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
557 struct gfs2_glock *io_gl;
563 if (!name->len || name->len > GFS2_FNAMESIZE)
564 return -ENAMETOOLONG;
566 error = gfs2_rs_alloc(dip);
570 error = gfs2_rindex_update(sdp);
574 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
578 error = create_ok(dip, name, mode);
582 inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
583 error = PTR_ERR(inode);
584 if (!IS_ERR(inode)) {
585 d = d_splice_alias(inode, dentry);
588 if (S_ISREG(inode->i_mode)) {
590 error = finish_open(file, dentry, gfs2_open_common, opened);
592 error = finish_no_open(file, d);
597 gfs2_glock_dq_uninit(ghs);
599 } else if (error != -ENOENT) {
603 arq = error = gfs2_diradd_alloc_required(dir, name);
607 inode = new_inode(sdp->sd_vfs);
613 error = gfs2_rs_alloc(ip);
615 goto fail_free_inode;
617 inode->i_mode = mode;
618 set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
620 inode->i_size = size;
621 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
622 gfs2_set_inode_blocks(inode, 1);
623 munge_mode_uid_gid(dip, inode);
624 ip->i_goal = dip->i_goal;
631 switch(mode & S_IFMT) {
633 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
634 gfs2_tune_get(sdp, gt_new_files_jdata))
635 ip->i_diskflags |= GFS2_DIF_JDATA;
636 gfs2_set_aops(inode);
639 ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
640 ip->i_diskflags |= GFS2_DIF_JDATA;
644 gfs2_set_inode_flags(inode);
646 if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
647 (dip->i_diskflags & GFS2_DIF_TOPDIR))
648 aflags |= GFS2_AF_ORLOV;
650 error = alloc_dinode(ip, aflags);
652 goto fail_free_inode;
654 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
656 goto fail_free_inode;
658 ip->i_gl->gl_object = ip;
659 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
661 goto fail_free_inode;
663 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
667 init_dinode(dip, ip, symname);
670 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
674 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
678 ip->i_iopen_gh.gh_gl->gl_object = ip;
679 gfs2_glock_put(io_gl);
681 insert_inode_hash(inode);
683 error = gfs2_acl_create(dip, inode);
687 error = gfs2_security_init(dip, ip, name);
691 error = link_dinode(dip, name, ip, arq);
695 mark_inode_dirty(inode);
696 d_instantiate(dentry, inode);
698 error = finish_open(file, dentry, gfs2_open_common, opened);
699 gfs2_glock_dq_uninit(ghs);
700 gfs2_glock_dq_uninit(ghs + 1);
704 gfs2_glock_dq_uninit(ghs + 1);
706 gfs2_glock_put(ip->i_gl);
710 gfs2_glock_dq_uninit(ghs + 1);
713 gfs2_glock_put(ip->i_gl);
715 free_inode_nonrcu(inode);
718 gfs2_glock_dq_uninit(ghs);
719 if (inode && !IS_ERR(inode)) {
721 mark_inode_dirty(inode);
722 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
730 * gfs2_create - Create a file
731 * @dir: The directory in which to create the file
732 * @dentry: The dentry of the new file
733 * @mode: The mode of the new file
738 static int gfs2_create(struct inode *dir, struct dentry *dentry,
739 umode_t mode, bool excl)
741 return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
745 * __gfs2_lookup - Look up a filename in a directory and return its inode
746 * @dir: The directory inode
747 * @dentry: The dentry of the new inode
748 * @file: File to be opened
749 * @opened: atomic_open flags
755 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
756 struct file *file, int *opened)
760 struct gfs2_holder gh;
761 struct gfs2_glock *gl;
764 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
768 return ERR_CAST(inode);
770 gl = GFS2_I(inode)->i_gl;
771 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
774 return ERR_PTR(error);
777 d = d_splice_alias(inode, dentry);
778 if (file && S_ISREG(inode->i_mode))
779 error = finish_open(file, dentry, gfs2_open_common, opened);
781 gfs2_glock_dq_uninit(&gh);
784 return ERR_PTR(error);
789 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
792 return __gfs2_lookup(dir, dentry, NULL, NULL);
796 * gfs2_link - Link to a file
797 * @old_dentry: The inode to link
798 * @dir: Add link to this directory
799 * @dentry: The name of the link
801 * Link the inode in "old_dentry" into the directory "dir" with the
807 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
808 struct dentry *dentry)
810 struct gfs2_inode *dip = GFS2_I(dir);
811 struct gfs2_sbd *sdp = GFS2_SB(dir);
812 struct inode *inode = old_dentry->d_inode;
813 struct gfs2_inode *ip = GFS2_I(inode);
814 struct gfs2_holder ghs[2];
815 struct buffer_head *dibh;
819 if (S_ISDIR(inode->i_mode))
822 error = gfs2_rs_alloc(dip);
826 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
827 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
829 error = gfs2_glock_nq(ghs); /* parent */
833 error = gfs2_glock_nq(ghs + 1); /* child */
838 if (inode->i_nlink == 0)
841 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
845 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
856 if (!dip->i_inode.i_nlink)
859 if (dip->i_entries == (u32)-1)
862 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
865 if (!ip->i_inode.i_nlink)
868 if (ip->i_inode.i_nlink == (u32)-1)
871 alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
876 if (alloc_required) {
877 error = gfs2_quota_lock_check(dip);
881 error = gfs2_inplace_reserve(dip, sdp->sd_max_dirres, 0);
885 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
886 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
887 2 * RES_DINODE + RES_STATFS +
892 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
897 error = gfs2_meta_inode_buffer(ip, &dibh);
901 error = gfs2_dir_add(dir, &dentry->d_name, ip);
905 gfs2_trans_add_meta(ip->i_gl, dibh);
906 inc_nlink(&ip->i_inode);
907 ip->i_inode.i_ctime = CURRENT_TIME;
909 d_instantiate(dentry, inode);
910 mark_inode_dirty(inode);
918 gfs2_inplace_release(dip);
921 gfs2_quota_unlock(dip);
923 gfs2_glock_dq(ghs + 1);
927 gfs2_holder_uninit(ghs);
928 gfs2_holder_uninit(ghs + 1);
933 * gfs2_unlink_ok - check to see that a inode is still in a directory
934 * @dip: the directory
935 * @name: the name of the file
938 * Assumes that the lock on (at least) @dip is held.
940 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
943 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
944 const struct gfs2_inode *ip)
948 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
951 if ((dip->i_inode.i_mode & S_ISVTX) &&
952 !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
953 !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
956 if (IS_APPEND(&dip->i_inode))
959 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
963 error = gfs2_dir_check(&dip->i_inode, name, ip);
971 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
972 * @dip: The parent directory
973 * @name: The name of the entry in the parent directory
974 * @inode: The inode to be removed
976 * Called with all the locks and in a transaction. This will only be
977 * called for a directory after it has been checked to ensure it is empty.
979 * Returns: 0 on success, or an error
982 static int gfs2_unlink_inode(struct gfs2_inode *dip,
983 const struct dentry *dentry)
985 struct inode *inode = dentry->d_inode;
986 struct gfs2_inode *ip = GFS2_I(inode);
989 error = gfs2_dir_del(dip, dentry);
994 inode->i_ctime = CURRENT_TIME;
995 if (S_ISDIR(inode->i_mode))
999 mark_inode_dirty(inode);
1000 if (inode->i_nlink == 0)
1001 gfs2_unlink_di(inode);
1007 * gfs2_unlink - Unlink an inode (this does rmdir as well)
1008 * @dir: The inode of the directory containing the inode to unlink
1009 * @dentry: The file itself
1011 * This routine uses the type of the inode as a flag to figure out
1012 * whether this is an unlink or an rmdir.
1017 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1019 struct gfs2_inode *dip = GFS2_I(dir);
1020 struct gfs2_sbd *sdp = GFS2_SB(dir);
1021 struct inode *inode = dentry->d_inode;
1022 struct gfs2_inode *ip = GFS2_I(inode);
1023 struct gfs2_holder ghs[3];
1024 struct gfs2_rgrpd *rgd;
1027 error = gfs2_rindex_update(sdp);
1033 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1034 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1036 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1040 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1043 error = gfs2_glock_nq(ghs); /* parent */
1047 error = gfs2_glock_nq(ghs + 1); /* child */
1052 if (inode->i_nlink == 0)
1055 if (S_ISDIR(inode->i_mode)) {
1057 if (ip->i_entries > 2 || inode->i_nlink > 2)
1061 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1065 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1069 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1073 error = gfs2_unlink_inode(dip, dentry);
1076 gfs2_trans_end(sdp);
1078 gfs2_glock_dq(ghs + 2);
1080 gfs2_glock_dq(ghs + 1);
1084 gfs2_holder_uninit(ghs + 2);
1086 gfs2_holder_uninit(ghs + 1);
1087 gfs2_holder_uninit(ghs);
1092 * gfs2_symlink - Create a symlink
1093 * @dir: The directory to create the symlink in
1094 * @dentry: The dentry to put the symlink in
1095 * @symname: The thing which the link points to
1100 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1101 const char *symname)
1103 struct gfs2_sbd *sdp = GFS2_SB(dir);
1106 size = strlen(symname);
1107 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1108 return -ENAMETOOLONG;
1110 return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1114 * gfs2_mkdir - Make a directory
1115 * @dir: The parent directory of the new one
1116 * @dentry: The dentry of the new directory
1117 * @mode: The mode of the new directory
1122 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1124 struct gfs2_sbd *sdp = GFS2_SB(dir);
1125 unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1126 return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1130 * gfs2_mknod - Make a special file
1131 * @dir: The directory in which the special file will reside
1132 * @dentry: The dentry of the special file
1133 * @mode: The mode of the special file
1134 * @dev: The device specification of the special file
1138 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1141 return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1145 * gfs2_atomic_open - Atomically open a file
1146 * @dir: The directory
1147 * @dentry: The proposed new entry
1148 * @file: The proposed new struct file
1149 * @flags: open flags
1151 * @opened: Flag to say whether the file has been opened or not
1153 * Returns: error code or 0 for success
1156 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1157 struct file *file, unsigned flags,
1158 umode_t mode, int *opened)
1161 bool excl = !!(flags & O_EXCL);
1163 d = __gfs2_lookup(dir, dentry, file, opened);
1168 if (dentry->d_inode) {
1169 if (!(*opened & FILE_OPENED))
1170 return finish_no_open(file, dentry);
1176 if (!(flags & O_CREAT))
1179 return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1183 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1187 * Follow @to back to the root and make sure we don't encounter @this
1188 * Assumes we already hold the rename lock.
1193 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1195 struct inode *dir = &to->i_inode;
1196 struct super_block *sb = dir->i_sb;
1203 if (dir == &this->i_inode) {
1207 if (dir == sb->s_root->d_inode) {
1212 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1214 error = PTR_ERR(tmp);
1228 * gfs2_rename - Rename a file
1229 * @odir: Parent directory of old file name
1230 * @odentry: The old dentry of the file
1231 * @ndir: Parent directory of new file name
1232 * @ndentry: The new dentry of the file
1237 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1238 struct inode *ndir, struct dentry *ndentry)
1240 struct gfs2_inode *odip = GFS2_I(odir);
1241 struct gfs2_inode *ndip = GFS2_I(ndir);
1242 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1243 struct gfs2_inode *nip = NULL;
1244 struct gfs2_sbd *sdp = GFS2_SB(odir);
1245 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1246 struct gfs2_rgrpd *nrgd;
1247 unsigned int num_gh;
1249 int alloc_required = 0;
1253 if (ndentry->d_inode) {
1254 nip = GFS2_I(ndentry->d_inode);
1259 error = gfs2_rindex_update(sdp);
1263 error = gfs2_rs_alloc(ndip);
1268 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1273 if (S_ISDIR(ip->i_inode.i_mode)) {
1275 /* don't move a dirctory into it's subdir */
1276 error = gfs2_ok_to_move(ip, ndip);
1283 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1285 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1288 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1292 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1294 /* grab the resource lock for unlink flag twiddling
1295 * this is the case of the target file already existing
1296 * so we unlink before doing the rename
1298 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1300 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1303 for (x = 0; x < num_gh; x++) {
1304 error = gfs2_glock_nq(ghs + x);
1310 if (ip->i_inode.i_nlink == 0)
1313 /* Check out the old directory */
1315 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1319 /* Check out the new directory */
1322 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1326 if (nip->i_inode.i_nlink == 0) {
1331 if (S_ISDIR(nip->i_inode.i_mode)) {
1332 if (nip->i_entries < 2) {
1333 gfs2_consist_inode(nip);
1337 if (nip->i_entries > 2) {
1343 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1347 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1359 if (!ndip->i_inode.i_nlink) {
1363 if (ndip->i_entries == (u32)-1) {
1367 if (S_ISDIR(ip->i_inode.i_mode) &&
1368 ndip->i_inode.i_nlink == (u32)-1) {
1375 /* Check out the dir to be renamed */
1378 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1384 alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1385 error = alloc_required;
1389 if (alloc_required) {
1390 error = gfs2_quota_lock_check(ndip);
1394 error = gfs2_inplace_reserve(ndip, sdp->sd_max_dirres, 0);
1398 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1399 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1400 4 * RES_DINODE + 4 * RES_LEAF +
1401 RES_STATFS + RES_QUOTA + 4, 0);
1405 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1406 5 * RES_LEAF + 4, 0);
1411 /* Remove the target file, if it exists */
1414 error = gfs2_unlink_inode(ndip, ndentry);
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_meta(ip->i_gl, dibh);
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_glock_dq(ghs + x);
1450 gfs2_holder_uninit(ghs + x);
1454 gfs2_glock_dq_uninit(&r_gh);
1460 * gfs2_follow_link - Follow a symbolic link
1461 * @dentry: The dentry of the link
1462 * @nd: Data that we pass to vfs_follow_link()
1464 * This can handle symlinks of any size.
1466 * Returns: 0 on success or error code
1469 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1471 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1472 struct gfs2_holder i_gh;
1473 struct buffer_head *dibh;
1478 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1479 error = gfs2_glock_nq(&i_gh);
1481 gfs2_holder_uninit(&i_gh);
1482 nd_set_link(nd, ERR_PTR(error));
1486 size = (unsigned int)i_size_read(&ip->i_inode);
1488 gfs2_consist_inode(ip);
1489 buf = ERR_PTR(-EIO);
1493 error = gfs2_meta_inode_buffer(ip, &dibh);
1495 buf = ERR_PTR(error);
1499 buf = kzalloc(size + 1, GFP_NOFS);
1501 buf = ERR_PTR(-ENOMEM);
1503 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1506 gfs2_glock_dq_uninit(&i_gh);
1507 nd_set_link(nd, buf);
1511 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1513 char *s = nd_get_link(nd);
1521 * @mask: The mask to be tested
1522 * @flags: Indicates whether this is an RCU path walk or not
1524 * This may be called from the VFS directly, or from within GFS2 with the
1525 * inode locked, so we look to see if the glock is already locked and only
1526 * lock the glock if its not already been done.
1531 int gfs2_permission(struct inode *inode, int mask)
1533 struct gfs2_inode *ip;
1534 struct gfs2_holder i_gh;
1540 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1541 if (mask & MAY_NOT_BLOCK)
1543 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1549 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1552 error = generic_permission(inode, mask);
1554 gfs2_glock_dq_uninit(&i_gh);
1559 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1561 setattr_copy(inode, attr);
1562 mark_inode_dirty(inode);
1567 * gfs2_setattr_simple -
1574 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1578 if (current->journal_info)
1579 return __gfs2_setattr_simple(inode, attr);
1581 error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1585 error = __gfs2_setattr_simple(inode, attr);
1586 gfs2_trans_end(GFS2_SB(inode));
1590 static int setattr_chown(struct inode *inode, struct iattr *attr)
1592 struct gfs2_inode *ip = GFS2_I(inode);
1593 struct gfs2_sbd *sdp = GFS2_SB(inode);
1598 ouid = inode->i_uid;
1599 ogid = inode->i_gid;
1600 nuid = attr->ia_uid;
1601 ngid = attr->ia_gid;
1603 if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1604 ouid = nuid = NO_UID_QUOTA_CHANGE;
1605 if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1606 ogid = ngid = NO_GID_QUOTA_CHANGE;
1608 error = gfs2_quota_lock(ip, nuid, ngid);
1612 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1613 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1614 error = gfs2_quota_check(ip, nuid, ngid);
1619 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1623 error = gfs2_setattr_simple(inode, attr);
1627 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1628 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1629 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1630 gfs2_quota_change(ip, -blocks, ouid, ogid);
1631 gfs2_quota_change(ip, blocks, nuid, ngid);
1635 gfs2_trans_end(sdp);
1637 gfs2_quota_unlock(ip);
1642 * gfs2_setattr - Change attributes on an inode
1643 * @dentry: The dentry which is changing
1644 * @attr: The structure describing the change
1646 * The VFS layer wants to change one or more of an inodes attributes. Write
1647 * that change out to disk.
1652 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1654 struct inode *inode = dentry->d_inode;
1655 struct gfs2_inode *ip = GFS2_I(inode);
1656 struct gfs2_holder i_gh;
1659 error = gfs2_rs_alloc(ip);
1663 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1668 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1671 error = inode_change_ok(inode, attr);
1675 if (attr->ia_valid & ATTR_SIZE)
1676 error = gfs2_setattr_size(inode, attr->ia_size);
1677 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1678 error = setattr_chown(inode, attr);
1679 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1680 error = gfs2_acl_chmod(ip, attr);
1682 error = gfs2_setattr_simple(inode, attr);
1686 mark_inode_dirty(inode);
1687 gfs2_glock_dq_uninit(&i_gh);
1692 * gfs2_getattr - Read out an inode's attributes
1693 * @mnt: The vfsmount the inode is being accessed from
1694 * @dentry: The dentry to stat
1695 * @stat: The inode's stats
1697 * This may be called from the VFS directly, or from within GFS2 with the
1698 * inode locked, so we look to see if the glock is already locked and only
1699 * lock the glock if its not already been done. Note that its the NFS
1700 * readdirplus operation which causes this to be called (from filldir)
1701 * with the glock already held.
1706 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1709 struct inode *inode = dentry->d_inode;
1710 struct gfs2_inode *ip = GFS2_I(inode);
1711 struct gfs2_holder gh;
1715 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1716 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1722 generic_fillattr(inode, stat);
1724 gfs2_glock_dq_uninit(&gh);
1729 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1730 const void *data, size_t size, int flags)
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_EXCLUSIVE, 0, &gh);
1738 ret = gfs2_glock_nq(&gh);
1740 ret = gfs2_rs_alloc(ip);
1742 ret = generic_setxattr(dentry, name, data, size, flags);
1745 gfs2_holder_uninit(&gh);
1749 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1750 void *data, size_t size)
1752 struct inode *inode = dentry->d_inode;
1753 struct gfs2_inode *ip = GFS2_I(inode);
1754 struct gfs2_holder gh;
1757 /* For selinux during lookup */
1758 if (gfs2_glock_is_locked_by_me(ip->i_gl))
1759 return generic_getxattr(dentry, name, data, size);
1761 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1762 ret = gfs2_glock_nq(&gh);
1764 ret = generic_getxattr(dentry, name, data, size);
1767 gfs2_holder_uninit(&gh);
1771 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1773 struct inode *inode = dentry->d_inode;
1774 struct gfs2_inode *ip = GFS2_I(inode);
1775 struct gfs2_holder gh;
1778 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1779 ret = gfs2_glock_nq(&gh);
1781 ret = gfs2_rs_alloc(ip);
1783 ret = generic_removexattr(dentry, name);
1786 gfs2_holder_uninit(&gh);
1790 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1793 struct gfs2_inode *ip = GFS2_I(inode);
1794 struct gfs2_holder gh;
1797 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1801 mutex_lock(&inode->i_mutex);
1803 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1807 if (gfs2_is_stuffed(ip)) {
1808 u64 phys = ip->i_no_addr << inode->i_blkbits;
1809 u64 size = i_size_read(inode);
1810 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1811 FIEMAP_EXTENT_DATA_INLINE;
1812 phys += sizeof(struct gfs2_dinode);
1814 if (start + len > size)
1817 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1822 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1826 gfs2_glock_dq_uninit(&gh);
1828 mutex_unlock(&inode->i_mutex);
1832 const struct inode_operations gfs2_file_iops = {
1833 .permission = gfs2_permission,
1834 .setattr = gfs2_setattr,
1835 .getattr = gfs2_getattr,
1836 .setxattr = gfs2_setxattr,
1837 .getxattr = gfs2_getxattr,
1838 .listxattr = gfs2_listxattr,
1839 .removexattr = gfs2_removexattr,
1840 .fiemap = gfs2_fiemap,
1841 .get_acl = gfs2_get_acl,
1844 const struct inode_operations gfs2_dir_iops = {
1845 .create = gfs2_create,
1846 .lookup = gfs2_lookup,
1848 .unlink = gfs2_unlink,
1849 .symlink = gfs2_symlink,
1850 .mkdir = gfs2_mkdir,
1851 .rmdir = gfs2_unlink,
1852 .mknod = gfs2_mknod,
1853 .rename = gfs2_rename,
1854 .permission = gfs2_permission,
1855 .setattr = gfs2_setattr,
1856 .getattr = gfs2_getattr,
1857 .setxattr = gfs2_setxattr,
1858 .getxattr = gfs2_getxattr,
1859 .listxattr = gfs2_listxattr,
1860 .removexattr = gfs2_removexattr,
1861 .fiemap = gfs2_fiemap,
1862 .get_acl = gfs2_get_acl,
1863 .atomic_open = gfs2_atomic_open,
1866 const struct inode_operations gfs2_symlink_iops = {
1867 .readlink = generic_readlink,
1868 .follow_link = gfs2_follow_link,
1869 .put_link = gfs2_put_link,
1870 .permission = gfs2_permission,
1871 .setattr = gfs2_setattr,
1872 .getattr = gfs2_getattr,
1873 .setxattr = gfs2_setxattr,
1874 .getxattr = gfs2_getxattr,
1875 .listxattr = gfs2_listxattr,
1876 .removexattr = gfs2_removexattr,
1877 .fiemap = gfs2_fiemap,
1878 .get_acl = gfs2_get_acl,