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(-ENOMEM);
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);
473 * gfs2_trans_da_blocks - Calculate number of blocks to link inode
474 * @dip: The directory we are linking into
475 * @da: The dir add information
476 * @nr_inodes: The number of inodes involved
478 * This calculate the number of blocks we need to reserve in a
479 * transaction to link @nr_inodes into a directory. In most cases
480 * @nr_inodes will be 2 (the directory plus the inode being linked in)
481 * but in case of rename, 4 may be required.
483 * Returns: Number of blocks
486 static unsigned gfs2_trans_da_blks(const struct gfs2_inode *dip,
487 const struct gfs2_diradd *da,
490 return da->nr_blocks + gfs2_rg_blocks(dip, da->nr_blocks) +
491 (nr_inodes * RES_DINODE) + RES_QUOTA + RES_STATFS;
494 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
495 struct gfs2_inode *ip, struct gfs2_diradd *da)
497 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
498 struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
502 error = gfs2_quota_lock_check(dip);
504 goto fail_quota_locks;
506 error = gfs2_inplace_reserve(dip, &ap);
508 goto fail_quota_locks;
510 error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, da, 2), 0);
514 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
516 goto fail_quota_locks;
519 error = gfs2_dir_add(&dip->i_inode, name, ip, da);
526 gfs2_inplace_release(dip);
528 gfs2_quota_unlock(dip);
532 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
535 const struct xattr *xattr;
538 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
539 err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
541 GFS2_EATYPE_SECURITY);
548 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
549 const struct qstr *qstr)
551 return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
552 &gfs2_initxattrs, NULL);
556 * gfs2_create_inode - Create a new inode
557 * @dir: The parent directory
558 * @dentry: The new dentry
559 * @file: If non-NULL, the file which is being opened
560 * @mode: The permissions on the new inode
561 * @dev: For device nodes, this is the device number
562 * @symname: For symlinks, this is the link destination
563 * @size: The initial size of the inode (ignored for directories)
565 * Returns: 0 on success, or error code
568 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
570 umode_t mode, dev_t dev, const char *symname,
571 unsigned int size, int excl, int *opened)
573 const struct qstr *name = &dentry->d_name;
574 struct gfs2_holder ghs[2];
575 struct inode *inode = NULL;
576 struct gfs2_inode *dip = GFS2_I(dir), *ip;
577 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
578 struct gfs2_glock *io_gl;
582 struct gfs2_diradd da = { .bh = NULL, };
584 if (!name->len || name->len > GFS2_FNAMESIZE)
585 return -ENAMETOOLONG;
587 error = gfs2_rs_alloc(dip);
591 error = gfs2_rindex_update(sdp);
595 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
599 error = create_ok(dip, name, mode);
603 inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
604 error = PTR_ERR(inode);
605 if (!IS_ERR(inode)) {
606 d = d_splice_alias(inode, dentry);
612 if (S_ISREG(inode->i_mode)) {
614 error = finish_open(file, dentry, gfs2_open_common, opened);
616 error = finish_no_open(file, d);
621 gfs2_glock_dq_uninit(ghs);
623 } else if (error != -ENOENT) {
627 error = gfs2_diradd_alloc_required(dir, name, &da);
631 inode = new_inode(sdp->sd_vfs);
637 error = gfs2_rs_alloc(ip);
639 goto fail_free_inode;
641 inode->i_mode = mode;
642 set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
644 inode->i_size = size;
645 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
646 gfs2_set_inode_blocks(inode, 1);
647 munge_mode_uid_gid(dip, inode);
648 ip->i_goal = dip->i_goal;
655 switch(mode & S_IFMT) {
657 if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
658 gfs2_tune_get(sdp, gt_new_files_jdata))
659 ip->i_diskflags |= GFS2_DIF_JDATA;
660 gfs2_set_aops(inode);
663 ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
664 ip->i_diskflags |= GFS2_DIF_JDATA;
668 gfs2_set_inode_flags(inode);
670 if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
671 (dip->i_diskflags & GFS2_DIF_TOPDIR))
672 aflags |= GFS2_AF_ORLOV;
674 error = alloc_dinode(ip, aflags);
676 goto fail_free_inode;
678 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
680 goto fail_free_inode;
682 ip->i_gl->gl_object = ip;
683 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
685 goto fail_free_inode;
687 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
691 init_dinode(dip, ip, symname);
694 error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
698 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
702 ip->i_iopen_gh.gh_gl->gl_object = ip;
703 gfs2_glock_put(io_gl);
705 insert_inode_hash(inode);
707 error = gfs2_acl_create(dip, inode);
711 error = gfs2_security_init(dip, ip, name);
715 error = link_dinode(dip, name, ip, &da);
719 mark_inode_dirty(inode);
720 d_instantiate(dentry, inode);
722 *opened |= FILE_CREATED;
723 error = finish_open(file, dentry, gfs2_open_common, opened);
725 gfs2_glock_dq_uninit(ghs);
726 gfs2_glock_dq_uninit(ghs + 1);
730 gfs2_glock_dq_uninit(ghs + 1);
732 gfs2_glock_put(ip->i_gl);
736 gfs2_glock_dq_uninit(ghs + 1);
739 gfs2_glock_put(ip->i_gl);
740 gfs2_rs_delete(ip, NULL);
741 free_inode_nonrcu(inode);
744 gfs2_dir_no_add(&da);
745 gfs2_glock_dq_uninit(ghs);
746 if (inode && !IS_ERR(inode)) {
748 mark_inode_dirty(inode);
749 set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
757 * gfs2_create - Create a file
758 * @dir: The directory in which to create the file
759 * @dentry: The dentry of the new file
760 * @mode: The mode of the new file
765 static int gfs2_create(struct inode *dir, struct dentry *dentry,
766 umode_t mode, bool excl)
768 return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
772 * __gfs2_lookup - Look up a filename in a directory and return its inode
773 * @dir: The directory inode
774 * @dentry: The dentry of the new inode
775 * @file: File to be opened
776 * @opened: atomic_open flags
782 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
783 struct file *file, int *opened)
787 struct gfs2_holder gh;
788 struct gfs2_glock *gl;
791 inode = gfs2_lookupi(dir, &dentry->d_name, 0);
795 return ERR_CAST(inode);
797 gl = GFS2_I(inode)->i_gl;
798 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
801 return ERR_PTR(error);
804 d = d_splice_alias(inode, dentry);
807 gfs2_glock_dq_uninit(&gh);
810 if (file && S_ISREG(inode->i_mode))
811 error = finish_open(file, dentry, gfs2_open_common, opened);
813 gfs2_glock_dq_uninit(&gh);
816 return ERR_PTR(error);
821 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
824 return __gfs2_lookup(dir, dentry, NULL, NULL);
828 * gfs2_link - Link to a file
829 * @old_dentry: The inode to link
830 * @dir: Add link to this directory
831 * @dentry: The name of the link
833 * Link the inode in "old_dentry" into the directory "dir" with the
839 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
840 struct dentry *dentry)
842 struct gfs2_inode *dip = GFS2_I(dir);
843 struct gfs2_sbd *sdp = GFS2_SB(dir);
844 struct inode *inode = old_dentry->d_inode;
845 struct gfs2_inode *ip = GFS2_I(inode);
846 struct gfs2_holder ghs[2];
847 struct buffer_head *dibh;
848 struct gfs2_diradd da = { .bh = NULL, };
851 if (S_ISDIR(inode->i_mode))
854 error = gfs2_rs_alloc(dip);
858 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
859 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
861 error = gfs2_glock_nq(ghs); /* parent */
865 error = gfs2_glock_nq(ghs + 1); /* child */
870 if (inode->i_nlink == 0)
873 error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
877 error = gfs2_dir_check(dir, &dentry->d_name, NULL);
888 if (!dip->i_inode.i_nlink)
891 if (dip->i_entries == (u32)-1)
894 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
897 if (!ip->i_inode.i_nlink)
900 if (ip->i_inode.i_nlink == (u32)-1)
903 error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
908 struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
909 error = gfs2_quota_lock_check(dip);
913 error = gfs2_inplace_reserve(dip, &ap);
917 error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, &da, 2), 0);
921 error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
926 error = gfs2_meta_inode_buffer(ip, &dibh);
930 error = gfs2_dir_add(dir, &dentry->d_name, ip, &da);
934 gfs2_trans_add_meta(ip->i_gl, dibh);
935 inc_nlink(&ip->i_inode);
936 ip->i_inode.i_ctime = CURRENT_TIME;
938 d_instantiate(dentry, inode);
939 mark_inode_dirty(inode);
947 gfs2_inplace_release(dip);
950 gfs2_quota_unlock(dip);
952 gfs2_dir_no_add(&da);
953 gfs2_glock_dq(ghs + 1);
957 gfs2_holder_uninit(ghs);
958 gfs2_holder_uninit(ghs + 1);
963 * gfs2_unlink_ok - check to see that a inode is still in a directory
964 * @dip: the directory
965 * @name: the name of the file
968 * Assumes that the lock on (at least) @dip is held.
970 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
973 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
974 const struct gfs2_inode *ip)
978 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
981 if ((dip->i_inode.i_mode & S_ISVTX) &&
982 !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
983 !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
986 if (IS_APPEND(&dip->i_inode))
989 error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
993 error = gfs2_dir_check(&dip->i_inode, name, ip);
1001 * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1002 * @dip: The parent directory
1003 * @name: The name of the entry in the parent directory
1004 * @inode: The inode to be removed
1006 * Called with all the locks and in a transaction. This will only be
1007 * called for a directory after it has been checked to ensure it is empty.
1009 * Returns: 0 on success, or an error
1012 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1013 const struct dentry *dentry)
1015 struct inode *inode = dentry->d_inode;
1016 struct gfs2_inode *ip = GFS2_I(inode);
1019 error = gfs2_dir_del(dip, dentry);
1024 inode->i_ctime = CURRENT_TIME;
1025 if (S_ISDIR(inode->i_mode))
1029 mark_inode_dirty(inode);
1030 if (inode->i_nlink == 0)
1031 gfs2_unlink_di(inode);
1037 * gfs2_unlink - Unlink an inode (this does rmdir as well)
1038 * @dir: The inode of the directory containing the inode to unlink
1039 * @dentry: The file itself
1041 * This routine uses the type of the inode as a flag to figure out
1042 * whether this is an unlink or an rmdir.
1047 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1049 struct gfs2_inode *dip = GFS2_I(dir);
1050 struct gfs2_sbd *sdp = GFS2_SB(dir);
1051 struct inode *inode = dentry->d_inode;
1052 struct gfs2_inode *ip = GFS2_I(inode);
1053 struct gfs2_holder ghs[3];
1054 struct gfs2_rgrpd *rgd;
1057 error = gfs2_rindex_update(sdp);
1063 gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1064 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
1066 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1070 gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1073 error = gfs2_glock_nq(ghs); /* parent */
1077 error = gfs2_glock_nq(ghs + 1); /* child */
1082 if (inode->i_nlink == 0)
1085 if (S_ISDIR(inode->i_mode)) {
1087 if (ip->i_entries > 2 || inode->i_nlink > 2)
1091 error = gfs2_glock_nq(ghs + 2); /* rgrp */
1095 error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1099 error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1103 error = gfs2_unlink_inode(dip, dentry);
1106 gfs2_trans_end(sdp);
1108 gfs2_glock_dq(ghs + 2);
1110 gfs2_glock_dq(ghs + 1);
1114 gfs2_holder_uninit(ghs + 2);
1116 gfs2_holder_uninit(ghs + 1);
1117 gfs2_holder_uninit(ghs);
1122 * gfs2_symlink - Create a symlink
1123 * @dir: The directory to create the symlink in
1124 * @dentry: The dentry to put the symlink in
1125 * @symname: The thing which the link points to
1130 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1131 const char *symname)
1133 struct gfs2_sbd *sdp = GFS2_SB(dir);
1136 size = strlen(symname);
1137 if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1138 return -ENAMETOOLONG;
1140 return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1144 * gfs2_mkdir - Make a directory
1145 * @dir: The parent directory of the new one
1146 * @dentry: The dentry of the new directory
1147 * @mode: The mode of the new directory
1152 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1154 struct gfs2_sbd *sdp = GFS2_SB(dir);
1155 unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1156 return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1160 * gfs2_mknod - Make a special file
1161 * @dir: The directory in which the special file will reside
1162 * @dentry: The dentry of the special file
1163 * @mode: The mode of the special file
1164 * @dev: The device specification of the special file
1168 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1171 return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1175 * gfs2_atomic_open - Atomically open a file
1176 * @dir: The directory
1177 * @dentry: The proposed new entry
1178 * @file: The proposed new struct file
1179 * @flags: open flags
1181 * @opened: Flag to say whether the file has been opened or not
1183 * Returns: error code or 0 for success
1186 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1187 struct file *file, unsigned flags,
1188 umode_t mode, int *opened)
1191 bool excl = !!(flags & O_EXCL);
1193 d = __gfs2_lookup(dir, dentry, file, opened);
1198 if (dentry->d_inode) {
1199 if (!(*opened & FILE_OPENED)) {
1202 return finish_no_open(file, dentry);
1209 if (!(flags & O_CREAT))
1212 return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1216 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1220 * Follow @to back to the root and make sure we don't encounter @this
1221 * Assumes we already hold the rename lock.
1226 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1228 struct inode *dir = &to->i_inode;
1229 struct super_block *sb = dir->i_sb;
1236 if (dir == &this->i_inode) {
1240 if (dir == sb->s_root->d_inode) {
1245 tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1247 error = PTR_ERR(tmp);
1261 * gfs2_rename - Rename a file
1262 * @odir: Parent directory of old file name
1263 * @odentry: The old dentry of the file
1264 * @ndir: Parent directory of new file name
1265 * @ndentry: The new dentry of the file
1270 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1271 struct inode *ndir, struct dentry *ndentry)
1273 struct gfs2_inode *odip = GFS2_I(odir);
1274 struct gfs2_inode *ndip = GFS2_I(ndir);
1275 struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1276 struct gfs2_inode *nip = NULL;
1277 struct gfs2_sbd *sdp = GFS2_SB(odir);
1278 struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1279 struct gfs2_rgrpd *nrgd;
1280 unsigned int num_gh;
1282 struct gfs2_diradd da = { .nr_blocks = 0, };
1286 if (ndentry->d_inode) {
1287 nip = GFS2_I(ndentry->d_inode);
1292 error = gfs2_rindex_update(sdp);
1296 error = gfs2_rs_alloc(ndip);
1301 error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1306 if (S_ISDIR(ip->i_inode.i_mode)) {
1308 /* don't move a dirctory into it's subdir */
1309 error = gfs2_ok_to_move(ip, ndip);
1316 gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1318 gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1321 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1325 gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1327 /* grab the resource lock for unlink flag twiddling
1328 * this is the case of the target file already existing
1329 * so we unlink before doing the rename
1331 nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1333 gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1336 for (x = 0; x < num_gh; x++) {
1337 error = gfs2_glock_nq(ghs + x);
1343 if (ip->i_inode.i_nlink == 0)
1346 /* Check out the old directory */
1348 error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1352 /* Check out the new directory */
1355 error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1359 if (nip->i_inode.i_nlink == 0) {
1364 if (S_ISDIR(nip->i_inode.i_mode)) {
1365 if (nip->i_entries < 2) {
1366 gfs2_consist_inode(nip);
1370 if (nip->i_entries > 2) {
1376 error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1380 error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1392 if (!ndip->i_inode.i_nlink) {
1396 if (ndip->i_entries == (u32)-1) {
1400 if (S_ISDIR(ip->i_inode.i_mode) &&
1401 ndip->i_inode.i_nlink == (u32)-1) {
1408 /* Check out the dir to be renamed */
1411 error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1417 error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1423 struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1424 error = gfs2_quota_lock_check(ndip);
1428 error = gfs2_inplace_reserve(ndip, &ap);
1432 error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(ndip, &da, 4) +
1433 4 * RES_LEAF + 4, 0);
1437 error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1438 5 * RES_LEAF + 4, 0);
1443 /* Remove the target file, if it exists */
1446 error = gfs2_unlink_inode(ndip, ndentry);
1449 error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1453 struct buffer_head *dibh;
1454 error = gfs2_meta_inode_buffer(ip, &dibh);
1457 ip->i_inode.i_ctime = CURRENT_TIME;
1458 gfs2_trans_add_meta(ip->i_gl, dibh);
1459 gfs2_dinode_out(ip, dibh->b_data);
1463 error = gfs2_dir_del(odip, odentry);
1467 error = gfs2_dir_add(ndir, &ndentry->d_name, ip, &da);
1472 gfs2_trans_end(sdp);
1475 gfs2_inplace_release(ndip);
1478 gfs2_quota_unlock(ndip);
1480 gfs2_dir_no_add(&da);
1482 gfs2_glock_dq(ghs + x);
1483 gfs2_holder_uninit(ghs + x);
1487 gfs2_glock_dq_uninit(&r_gh);
1493 * gfs2_follow_link - Follow a symbolic link
1494 * @dentry: The dentry of the link
1495 * @nd: Data that we pass to vfs_follow_link()
1497 * This can handle symlinks of any size.
1499 * Returns: 0 on success or error code
1502 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1504 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1505 struct gfs2_holder i_gh;
1506 struct buffer_head *dibh;
1511 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1512 error = gfs2_glock_nq(&i_gh);
1514 gfs2_holder_uninit(&i_gh);
1515 nd_set_link(nd, ERR_PTR(error));
1519 size = (unsigned int)i_size_read(&ip->i_inode);
1521 gfs2_consist_inode(ip);
1522 buf = ERR_PTR(-EIO);
1526 error = gfs2_meta_inode_buffer(ip, &dibh);
1528 buf = ERR_PTR(error);
1532 buf = kzalloc(size + 1, GFP_NOFS);
1534 buf = ERR_PTR(-ENOMEM);
1536 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1539 gfs2_glock_dq_uninit(&i_gh);
1540 nd_set_link(nd, buf);
1547 * @mask: The mask to be tested
1548 * @flags: Indicates whether this is an RCU path walk or not
1550 * This may be called from the VFS directly, or from within GFS2 with the
1551 * inode locked, so we look to see if the glock is already locked and only
1552 * lock the glock if its not already been done.
1557 int gfs2_permission(struct inode *inode, int mask)
1559 struct gfs2_inode *ip;
1560 struct gfs2_holder i_gh;
1566 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1567 if (mask & MAY_NOT_BLOCK)
1569 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1575 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1578 error = generic_permission(inode, mask);
1580 gfs2_glock_dq_uninit(&i_gh);
1585 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1587 setattr_copy(inode, attr);
1588 mark_inode_dirty(inode);
1593 * gfs2_setattr_simple -
1600 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1604 if (current->journal_info)
1605 return __gfs2_setattr_simple(inode, attr);
1607 error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1611 error = __gfs2_setattr_simple(inode, attr);
1612 gfs2_trans_end(GFS2_SB(inode));
1616 static int setattr_chown(struct inode *inode, struct iattr *attr)
1618 struct gfs2_inode *ip = GFS2_I(inode);
1619 struct gfs2_sbd *sdp = GFS2_SB(inode);
1624 ouid = inode->i_uid;
1625 ogid = inode->i_gid;
1626 nuid = attr->ia_uid;
1627 ngid = attr->ia_gid;
1629 if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1630 ouid = nuid = NO_UID_QUOTA_CHANGE;
1631 if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1632 ogid = ngid = NO_GID_QUOTA_CHANGE;
1634 error = get_write_access(inode);
1638 error = gfs2_rs_alloc(ip);
1642 error = gfs2_rindex_update(sdp);
1646 error = gfs2_quota_lock(ip, nuid, ngid);
1650 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1651 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1652 error = gfs2_quota_check(ip, nuid, ngid);
1657 error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1661 error = gfs2_setattr_simple(inode, attr);
1665 if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1666 !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1667 u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1668 gfs2_quota_change(ip, -blocks, ouid, ogid);
1669 gfs2_quota_change(ip, blocks, nuid, ngid);
1673 gfs2_trans_end(sdp);
1675 gfs2_quota_unlock(ip);
1677 put_write_access(inode);
1682 * gfs2_setattr - Change attributes on an inode
1683 * @dentry: The dentry which is changing
1684 * @attr: The structure describing the change
1686 * The VFS layer wants to change one or more of an inodes attributes. Write
1687 * that change out to disk.
1692 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1694 struct inode *inode = dentry->d_inode;
1695 struct gfs2_inode *ip = GFS2_I(inode);
1696 struct gfs2_holder i_gh;
1699 error = gfs2_rs_alloc(ip);
1703 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1708 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1711 error = inode_change_ok(inode, attr);
1715 if (attr->ia_valid & ATTR_SIZE)
1716 error = gfs2_setattr_size(inode, attr->ia_size);
1717 else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1718 error = setattr_chown(inode, attr);
1719 else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1720 error = gfs2_acl_chmod(ip, attr);
1722 error = gfs2_setattr_simple(inode, attr);
1726 mark_inode_dirty(inode);
1727 gfs2_glock_dq_uninit(&i_gh);
1732 * gfs2_getattr - Read out an inode's attributes
1733 * @mnt: The vfsmount the inode is being accessed from
1734 * @dentry: The dentry to stat
1735 * @stat: The inode's stats
1737 * This may be called from the VFS directly, or from within GFS2 with the
1738 * inode locked, so we look to see if the glock is already locked and only
1739 * lock the glock if its not already been done. Note that its the NFS
1740 * readdirplus operation which causes this to be called (from filldir)
1741 * with the glock already held.
1746 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1749 struct inode *inode = dentry->d_inode;
1750 struct gfs2_inode *ip = GFS2_I(inode);
1751 struct gfs2_holder gh;
1755 if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1756 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1762 generic_fillattr(inode, stat);
1764 gfs2_glock_dq_uninit(&gh);
1769 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1770 const void *data, size_t size, int flags)
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_setxattr(dentry, name, data, size, flags);
1785 gfs2_holder_uninit(&gh);
1789 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1790 void *data, size_t size)
1792 struct inode *inode = dentry->d_inode;
1793 struct gfs2_inode *ip = GFS2_I(inode);
1794 struct gfs2_holder gh;
1797 /* For selinux during lookup */
1798 if (gfs2_glock_is_locked_by_me(ip->i_gl))
1799 return generic_getxattr(dentry, name, data, size);
1801 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1802 ret = gfs2_glock_nq(&gh);
1804 ret = generic_getxattr(dentry, name, data, size);
1807 gfs2_holder_uninit(&gh);
1811 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1813 struct inode *inode = dentry->d_inode;
1814 struct gfs2_inode *ip = GFS2_I(inode);
1815 struct gfs2_holder gh;
1818 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1819 ret = gfs2_glock_nq(&gh);
1821 ret = gfs2_rs_alloc(ip);
1823 ret = generic_removexattr(dentry, name);
1826 gfs2_holder_uninit(&gh);
1830 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1833 struct gfs2_inode *ip = GFS2_I(inode);
1834 struct gfs2_holder gh;
1837 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1841 mutex_lock(&inode->i_mutex);
1843 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1847 if (gfs2_is_stuffed(ip)) {
1848 u64 phys = ip->i_no_addr << inode->i_blkbits;
1849 u64 size = i_size_read(inode);
1850 u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1851 FIEMAP_EXTENT_DATA_INLINE;
1852 phys += sizeof(struct gfs2_dinode);
1854 if (start + len > size)
1857 ret = fiemap_fill_next_extent(fieinfo, start, phys,
1862 ret = __generic_block_fiemap(inode, fieinfo, start, len,
1866 gfs2_glock_dq_uninit(&gh);
1868 mutex_unlock(&inode->i_mutex);
1872 const struct inode_operations gfs2_file_iops = {
1873 .permission = gfs2_permission,
1874 .setattr = gfs2_setattr,
1875 .getattr = gfs2_getattr,
1876 .setxattr = gfs2_setxattr,
1877 .getxattr = gfs2_getxattr,
1878 .listxattr = gfs2_listxattr,
1879 .removexattr = gfs2_removexattr,
1880 .fiemap = gfs2_fiemap,
1881 .get_acl = gfs2_get_acl,
1884 const struct inode_operations gfs2_dir_iops = {
1885 .create = gfs2_create,
1886 .lookup = gfs2_lookup,
1888 .unlink = gfs2_unlink,
1889 .symlink = gfs2_symlink,
1890 .mkdir = gfs2_mkdir,
1891 .rmdir = gfs2_unlink,
1892 .mknod = gfs2_mknod,
1893 .rename = gfs2_rename,
1894 .permission = gfs2_permission,
1895 .setattr = gfs2_setattr,
1896 .getattr = gfs2_getattr,
1897 .setxattr = gfs2_setxattr,
1898 .getxattr = gfs2_getxattr,
1899 .listxattr = gfs2_listxattr,
1900 .removexattr = gfs2_removexattr,
1901 .fiemap = gfs2_fiemap,
1902 .get_acl = gfs2_get_acl,
1903 .atomic_open = gfs2_atomic_open,
1906 const struct inode_operations gfs2_symlink_iops = {
1907 .readlink = generic_readlink,
1908 .follow_link = gfs2_follow_link,
1909 .put_link = kfree_put_link,
1910 .permission = gfs2_permission,
1911 .setattr = gfs2_setattr,
1912 .getattr = gfs2_getattr,
1913 .setxattr = gfs2_setxattr,
1914 .getxattr = gfs2_getxattr,
1915 .listxattr = gfs2_listxattr,
1916 .removexattr = gfs2_removexattr,
1917 .fiemap = gfs2_fiemap,
1918 .get_acl = gfs2_get_acl,