xref: /openbmc/linux/fs/gfs2/inode.c (revision b34e08d5)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2011 Red Hat, Inc.  All rights reserved.
4  *
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.
8  */
9 
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>
15 #include <linux/mm.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>
23 
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "acl.h"
27 #include "bmap.h"
28 #include "dir.h"
29 #include "xattr.h"
30 #include "glock.h"
31 #include "inode.h"
32 #include "meta_io.h"
33 #include "quota.h"
34 #include "rgrp.h"
35 #include "trans.h"
36 #include "util.h"
37 #include "super.h"
38 #include "glops.h"
39 
40 struct gfs2_skip_data {
41 	u64 no_addr;
42 	int skipped;
43 	int non_block;
44 };
45 
46 static int iget_test(struct inode *inode, void *opaque)
47 {
48 	struct gfs2_inode *ip = GFS2_I(inode);
49 	struct gfs2_skip_data *data = opaque;
50 
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)) {
54 			data->skipped = 1;
55 			return 0;
56 		}
57 		return 1;
58 	}
59 	return 0;
60 }
61 
62 static int iget_set(struct inode *inode, void *opaque)
63 {
64 	struct gfs2_inode *ip = GFS2_I(inode);
65 	struct gfs2_skip_data *data = opaque;
66 
67 	if (data->skipped)
68 		return -ENOENT;
69 	inode->i_ino = (unsigned long)(data->no_addr);
70 	ip->i_no_addr = data->no_addr;
71 	return 0;
72 }
73 
74 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int non_block)
75 {
76 	unsigned long hash = (unsigned long)no_addr;
77 	struct gfs2_skip_data data;
78 
79 	data.no_addr = no_addr;
80 	data.skipped = 0;
81 	data.non_block = non_block;
82 	return ilookup5(sb, hash, iget_test, &data);
83 }
84 
85 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr,
86 			       int non_block)
87 {
88 	struct gfs2_skip_data data;
89 	unsigned long hash = (unsigned long)no_addr;
90 
91 	data.no_addr = no_addr;
92 	data.skipped = 0;
93 	data.non_block = non_block;
94 	return iget5_locked(sb, hash, iget_test, iget_set, &data);
95 }
96 
97 /**
98  * gfs2_set_iop - Sets inode operations
99  * @inode: The inode with correct i_mode filled in
100  *
101  * GFS2 lookup code fills in vfs inode contents based on info obtained
102  * from directory entry inside gfs2_inode_lookup().
103  */
104 
105 static void gfs2_set_iop(struct inode *inode)
106 {
107 	struct gfs2_sbd *sdp = GFS2_SB(inode);
108 	umode_t mode = inode->i_mode;
109 
110 	if (S_ISREG(mode)) {
111 		inode->i_op = &gfs2_file_iops;
112 		if (gfs2_localflocks(sdp))
113 			inode->i_fop = &gfs2_file_fops_nolock;
114 		else
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;
120 		else
121 			inode->i_fop = &gfs2_dir_fops;
122 	} else if (S_ISLNK(mode)) {
123 		inode->i_op = &gfs2_symlink_iops;
124 	} else {
125 		inode->i_op = &gfs2_file_iops;
126 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
127 	}
128 }
129 
130 /**
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?
136  *
137  * Returns: A VFS inode, or an error
138  */
139 
140 struct inode *gfs2_inode_lookup(struct super_block *sb, unsigned int type,
141 				u64 no_addr, u64 no_formal_ino, int non_block)
142 {
143 	struct inode *inode;
144 	struct gfs2_inode *ip;
145 	struct gfs2_glock *io_gl = NULL;
146 	int error;
147 
148 	inode = gfs2_iget(sb, no_addr, non_block);
149 	ip = GFS2_I(inode);
150 
151 	if (!inode)
152 		return ERR_PTR(-ENOMEM);
153 
154 	if (inode->i_state & I_NEW) {
155 		struct gfs2_sbd *sdp = GFS2_SB(inode);
156 		ip->i_no_formal_ino = no_formal_ino;
157 
158 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
159 		if (unlikely(error))
160 			goto fail;
161 		ip->i_gl->gl_object = ip;
162 
163 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
164 		if (unlikely(error))
165 			goto fail_put;
166 
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);
169 		if (unlikely(error))
170 			goto fail_iopen;
171 
172 		ip->i_iopen_gh.gh_gl->gl_object = ip;
173 		gfs2_glock_put(io_gl);
174 		io_gl = NULL;
175 
176 		if (type == DT_UNKNOWN) {
177 			/* Inode glock must be locked already */
178 			error = gfs2_inode_refresh(GFS2_I(inode));
179 			if (error)
180 				goto fail_refresh;
181 		} else {
182 			inode->i_mode = DT2IF(type);
183 		}
184 
185 		gfs2_set_iop(inode);
186 		unlock_new_inode(inode);
187 	}
188 
189 	return inode;
190 
191 fail_refresh:
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);
195 fail_iopen:
196 	if (io_gl)
197 		gfs2_glock_put(io_gl);
198 fail_put:
199 	ip->i_gl->gl_object = NULL;
200 	gfs2_glock_put(ip->i_gl);
201 fail:
202 	iget_failed(inode);
203 	return ERR_PTR(error);
204 }
205 
206 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
207 				  u64 *no_formal_ino, unsigned int blktype)
208 {
209 	struct super_block *sb = sdp->sd_vfs;
210 	struct gfs2_holder i_gh;
211 	struct inode *inode = NULL;
212 	int error;
213 
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);
217 	if (error)
218 		return ERR_PTR(error);
219 
220 	error = gfs2_check_blk_type(sdp, no_addr, blktype);
221 	if (error)
222 		goto fail;
223 
224 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0, 1);
225 	if (IS_ERR(inode))
226 		goto fail;
227 
228 	/* Two extra checks for NFS only */
229 	if (no_formal_ino) {
230 		error = -ESTALE;
231 		if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
232 			goto fail_iput;
233 
234 		error = -EIO;
235 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
236 			goto fail_iput;
237 
238 		error = 0;
239 	}
240 
241 fail:
242 	gfs2_glock_dq_uninit(&i_gh);
243 	return error ? ERR_PTR(error) : inode;
244 fail_iput:
245 	iput(inode);
246 	goto fail;
247 }
248 
249 
250 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
251 {
252 	struct qstr qstr;
253 	struct inode *inode;
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.
260 	 */
261 	if (inode == NULL)
262 		return ERR_PTR(-ENOENT);
263 	else
264 		return inode;
265 }
266 
267 
268 /**
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
274  *
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.
279  *
280  * Returns: errno
281  */
282 
283 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
284 			   int is_root)
285 {
286 	struct super_block *sb = dir->i_sb;
287 	struct gfs2_inode *dip = GFS2_I(dir);
288 	struct gfs2_holder d_gh;
289 	int error = 0;
290 	struct inode *inode = NULL;
291 	int unlock = 0;
292 
293 	if (!name->len || name->len > GFS2_FNAMESIZE)
294 		return ERR_PTR(-ENAMETOOLONG);
295 
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)) {
299 		igrab(dir);
300 		return dir;
301 	}
302 
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);
305 		if (error)
306 			return ERR_PTR(error);
307 		unlock = 1;
308 	}
309 
310 	if (!is_root) {
311 		error = gfs2_permission(dir, MAY_EXEC);
312 		if (error)
313 			goto out;
314 	}
315 
316 	inode = gfs2_dir_search(dir, name, false);
317 	if (IS_ERR(inode))
318 		error = PTR_ERR(inode);
319 out:
320 	if (unlock)
321 		gfs2_glock_dq_uninit(&d_gh);
322 	if (error == -ENOENT)
323 		return NULL;
324 	return inode ? inode : ERR_PTR(error);
325 }
326 
327 /**
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
331  * @mode:
332  *
333  * Returns: errno
334  */
335 
336 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
337 		     umode_t mode)
338 {
339 	int error;
340 
341 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
342 	if (error)
343 		return error;
344 
345 	/*  Don't create entries in an unlinked directory  */
346 	if (!dip->i_inode.i_nlink)
347 		return -ENOENT;
348 
349 	if (dip->i_entries == (u32)-1)
350 		return -EFBIG;
351 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
352 		return -EMLINK;
353 
354 	return 0;
355 }
356 
357 static void munge_mode_uid_gid(const struct gfs2_inode *dip,
358 			       struct inode *inode)
359 {
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;
368 	} else
369 		inode->i_uid = current_fsuid();
370 
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;
375 	} else
376 		inode->i_gid = current_fsgid();
377 }
378 
379 static int alloc_dinode(struct gfs2_inode *ip, u32 flags, unsigned *dblocks)
380 {
381 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
382 	struct gfs2_alloc_parms ap = { .target = *dblocks, .aflags = flags, };
383 	int error;
384 
385 	error = gfs2_quota_lock_check(ip);
386 	if (error)
387 		goto out;
388 
389 	error = gfs2_inplace_reserve(ip, &ap);
390 	if (error)
391 		goto out_quota;
392 
393 	error = gfs2_trans_begin(sdp, (*dblocks * RES_RG_BIT) + RES_STATFS + RES_QUOTA, 0);
394 	if (error)
395 		goto out_ipreserv;
396 
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;
401 
402 	gfs2_trans_end(sdp);
403 
404 out_ipreserv:
405 	gfs2_inplace_release(ip);
406 out_quota:
407 	gfs2_quota_unlock(ip);
408 out:
409 	return error;
410 }
411 
412 static void gfs2_init_dir(struct buffer_head *dibh,
413 			  const struct gfs2_inode *parent)
414 {
415 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
416 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
417 
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);
421 
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);
426 
427 }
428 
429 /**
430  * gfs2_init_xattr - Initialise an xattr block for a new inode
431  * @ip: The inode in question
432  *
433  * This sets up an empty xattr block for a new inode, ready to
434  * take any ACLs, LSM xattrs, etc.
435  */
436 
437 static void gfs2_init_xattr(struct gfs2_inode *ip)
438 {
439 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
440 	struct buffer_head *bh;
441 	struct gfs2_ea_header *ea;
442 
443 	bh = gfs2_meta_new(ip->i_gl, ip->i_eattr);
444 	gfs2_trans_add_meta(ip->i_gl, bh);
445 	gfs2_metatype_set(bh, GFS2_METATYPE_EA, GFS2_FORMAT_EA);
446 	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
447 
448 	ea = GFS2_EA_BH2FIRST(bh);
449 	ea->ea_rec_len = cpu_to_be32(sdp->sd_jbsize);
450 	ea->ea_type = GFS2_EATYPE_UNUSED;
451 	ea->ea_flags = GFS2_EAFLAG_LAST;
452 
453 	brelse(bh);
454 }
455 
456 /**
457  * init_dinode - Fill in a new dinode structure
458  * @dip: The directory this inode is being created in
459  * @ip: The inode
460  * @symname: The symlink destination (if a symlink)
461  * @bhp: The buffer head (returned to caller)
462  *
463  */
464 
465 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
466 			const char *symname)
467 {
468 	struct gfs2_dinode *di;
469 	struct buffer_head *dibh;
470 
471 	dibh = gfs2_meta_new(ip->i_gl, ip->i_no_addr);
472 	gfs2_trans_add_meta(ip->i_gl, dibh);
473 	di = (struct gfs2_dinode *)dibh->b_data;
474 	gfs2_dinode_out(ip, di);
475 
476 	di->di_major = cpu_to_be32(MAJOR(ip->i_inode.i_rdev));
477 	di->di_minor = cpu_to_be32(MINOR(ip->i_inode.i_rdev));
478 	di->__pad1 = 0;
479 	di->__pad2 = 0;
480 	di->__pad3 = 0;
481 	memset(&di->__pad4, 0, sizeof(di->__pad4));
482 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
483 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
484 
485 	switch(ip->i_inode.i_mode & S_IFMT) {
486 	case S_IFDIR:
487 		gfs2_init_dir(dibh, dip);
488 		break;
489 	case S_IFLNK:
490 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
491 		break;
492 	}
493 
494 	set_buffer_uptodate(dibh);
495 	brelse(dibh);
496 }
497 
498 /**
499  * gfs2_trans_da_blocks - Calculate number of blocks to link inode
500  * @dip: The directory we are linking into
501  * @da: The dir add information
502  * @nr_inodes: The number of inodes involved
503  *
504  * This calculate the number of blocks we need to reserve in a
505  * transaction to link @nr_inodes into a directory. In most cases
506  * @nr_inodes will be 2 (the directory plus the inode being linked in)
507  * but in case of rename, 4 may be required.
508  *
509  * Returns: Number of blocks
510  */
511 
512 static unsigned gfs2_trans_da_blks(const struct gfs2_inode *dip,
513 				   const struct gfs2_diradd *da,
514 				   unsigned nr_inodes)
515 {
516 	return da->nr_blocks + gfs2_rg_blocks(dip, da->nr_blocks) +
517 	       (nr_inodes * RES_DINODE) + RES_QUOTA + RES_STATFS;
518 }
519 
520 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
521 		       struct gfs2_inode *ip, struct gfs2_diradd *da)
522 {
523 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
524 	struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
525 	int error;
526 
527 	if (da->nr_blocks) {
528 		error = gfs2_quota_lock_check(dip);
529 		if (error)
530 			goto fail_quota_locks;
531 
532 		error = gfs2_inplace_reserve(dip, &ap);
533 		if (error)
534 			goto fail_quota_locks;
535 
536 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, da, 2), 0);
537 		if (error)
538 			goto fail_ipreserv;
539 	} else {
540 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
541 		if (error)
542 			goto fail_quota_locks;
543 	}
544 
545 	error = gfs2_dir_add(&dip->i_inode, name, ip, da);
546 	if (error)
547 		goto fail_end_trans;
548 
549 fail_end_trans:
550 	gfs2_trans_end(sdp);
551 fail_ipreserv:
552 	gfs2_inplace_release(dip);
553 fail_quota_locks:
554 	gfs2_quota_unlock(dip);
555 	return error;
556 }
557 
558 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
559 		    void *fs_info)
560 {
561 	const struct xattr *xattr;
562 	int err = 0;
563 
564 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
565 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
566 				       xattr->value_len, 0,
567 				       GFS2_EATYPE_SECURITY);
568 		if (err < 0)
569 			break;
570 	}
571 	return err;
572 }
573 
574 /**
575  * gfs2_create_inode - Create a new inode
576  * @dir: The parent directory
577  * @dentry: The new dentry
578  * @file: If non-NULL, the file which is being opened
579  * @mode: The permissions on the new inode
580  * @dev: For device nodes, this is the device number
581  * @symname: For symlinks, this is the link destination
582  * @size: The initial size of the inode (ignored for directories)
583  *
584  * Returns: 0 on success, or error code
585  */
586 
587 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
588 			     struct file *file,
589 			     umode_t mode, dev_t dev, const char *symname,
590 			     unsigned int size, int excl, int *opened)
591 {
592 	const struct qstr *name = &dentry->d_name;
593 	struct posix_acl *default_acl, *acl;
594 	struct gfs2_holder ghs[2];
595 	struct inode *inode = NULL;
596 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
597 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
598 	struct gfs2_glock *io_gl;
599 	struct dentry *d;
600 	int error, free_vfs_inode = 0;
601 	u32 aflags = 0;
602 	unsigned blocks = 1;
603 	struct gfs2_diradd da = { .bh = NULL, };
604 
605 	if (!name->len || name->len > GFS2_FNAMESIZE)
606 		return -ENAMETOOLONG;
607 
608 	error = gfs2_rs_alloc(dip);
609 	if (error)
610 		return error;
611 
612 	error = gfs2_rindex_update(sdp);
613 	if (error)
614 		return error;
615 
616 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
617 	if (error)
618 		goto fail;
619 
620 	error = create_ok(dip, name, mode);
621 	if (error)
622 		goto fail_gunlock;
623 
624 	inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
625 	error = PTR_ERR(inode);
626 	if (!IS_ERR(inode)) {
627 		d = d_splice_alias(inode, dentry);
628 		error = PTR_ERR(d);
629 		if (IS_ERR(d))
630 			goto fail_gunlock;
631 		error = 0;
632 		if (file) {
633 			if (S_ISREG(inode->i_mode)) {
634 				WARN_ON(d != NULL);
635 				error = finish_open(file, dentry, gfs2_open_common, opened);
636 			} else {
637 				error = finish_no_open(file, d);
638 			}
639 		} else {
640 			dput(d);
641 		}
642 		gfs2_glock_dq_uninit(ghs);
643 		return error;
644 	} else if (error != -ENOENT) {
645 		goto fail_gunlock;
646 	}
647 
648 	error = gfs2_diradd_alloc_required(dir, name, &da);
649 	if (error < 0)
650 		goto fail_gunlock;
651 
652 	inode = new_inode(sdp->sd_vfs);
653 	error = -ENOMEM;
654 	if (!inode)
655 		goto fail_gunlock;
656 
657 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
658 	if (error)
659 		goto fail_free_vfs_inode;
660 
661 	ip = GFS2_I(inode);
662 	error = gfs2_rs_alloc(ip);
663 	if (error)
664 		goto fail_free_acls;
665 
666 	inode->i_mode = mode;
667 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
668 	inode->i_rdev = dev;
669 	inode->i_size = size;
670 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
671 	gfs2_set_inode_blocks(inode, 1);
672 	munge_mode_uid_gid(dip, inode);
673 	ip->i_goal = dip->i_goal;
674 	ip->i_diskflags = 0;
675 	ip->i_eattr = 0;
676 	ip->i_height = 0;
677 	ip->i_depth = 0;
678 	ip->i_entries = 0;
679 
680 	switch(mode & S_IFMT) {
681 	case S_IFREG:
682 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
683 		    gfs2_tune_get(sdp, gt_new_files_jdata))
684 			ip->i_diskflags |= GFS2_DIF_JDATA;
685 		gfs2_set_aops(inode);
686 		break;
687 	case S_IFDIR:
688 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
689 		ip->i_diskflags |= GFS2_DIF_JDATA;
690 		ip->i_entries = 2;
691 		break;
692 	}
693 	gfs2_set_inode_flags(inode);
694 
695 	if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
696 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
697 		aflags |= GFS2_AF_ORLOV;
698 
699 	if (default_acl || acl)
700 		blocks++;
701 
702 	error = alloc_dinode(ip, aflags, &blocks);
703 	if (error)
704 		goto fail_free_inode;
705 
706 	gfs2_set_inode_blocks(inode, blocks);
707 
708 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
709 	if (error)
710 		goto fail_free_inode;
711 
712 	ip->i_gl->gl_object = ip;
713 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
714 	if (error)
715 		goto fail_free_inode;
716 
717 	error = gfs2_trans_begin(sdp, blocks, 0);
718 	if (error)
719 		goto fail_gunlock2;
720 
721 	if (blocks > 1) {
722 		ip->i_eattr = ip->i_no_addr + 1;
723 		gfs2_init_xattr(ip);
724 	}
725 	init_dinode(dip, ip, symname);
726 	gfs2_trans_end(sdp);
727 
728 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
729 	if (error)
730 		goto fail_gunlock2;
731 
732 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
733 	if (error)
734 		goto fail_gunlock2;
735 
736 	ip->i_iopen_gh.gh_gl->gl_object = ip;
737 	gfs2_glock_put(io_gl);
738 	gfs2_set_iop(inode);
739 	insert_inode_hash(inode);
740 
741 	if (default_acl) {
742 		error = gfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
743 		posix_acl_release(default_acl);
744 	}
745 	if (acl) {
746 		if (!error)
747 			error = gfs2_set_acl(inode, acl, ACL_TYPE_ACCESS);
748 		posix_acl_release(acl);
749 	}
750 
751 	if (error)
752 		goto fail_gunlock3;
753 
754 	error = security_inode_init_security(&ip->i_inode, &dip->i_inode, name,
755 					     &gfs2_initxattrs, NULL);
756 	if (error)
757 		goto fail_gunlock3;
758 
759 	error = link_dinode(dip, name, ip, &da);
760 	if (error)
761 		goto fail_gunlock3;
762 
763 	mark_inode_dirty(inode);
764 	d_instantiate(dentry, inode);
765 	if (file) {
766 		*opened |= FILE_CREATED;
767 		error = finish_open(file, dentry, gfs2_open_common, opened);
768 	}
769 	gfs2_glock_dq_uninit(ghs);
770 	gfs2_glock_dq_uninit(ghs + 1);
771 	return error;
772 
773 fail_gunlock3:
774 	gfs2_glock_dq_uninit(ghs + 1);
775 	if (ip->i_gl)
776 		gfs2_glock_put(ip->i_gl);
777 	goto fail_gunlock;
778 
779 fail_gunlock2:
780 	gfs2_glock_dq_uninit(ghs + 1);
781 fail_free_inode:
782 	if (ip->i_gl)
783 		gfs2_glock_put(ip->i_gl);
784 	gfs2_rs_delete(ip, NULL);
785 fail_free_acls:
786 	if (default_acl)
787 		posix_acl_release(default_acl);
788 	if (acl)
789 		posix_acl_release(acl);
790 fail_free_vfs_inode:
791 	free_vfs_inode = 1;
792 fail_gunlock:
793 	gfs2_dir_no_add(&da);
794 	gfs2_glock_dq_uninit(ghs);
795 	if (inode && !IS_ERR(inode)) {
796 		clear_nlink(inode);
797 		if (!free_vfs_inode)
798 			mark_inode_dirty(inode);
799 		set_bit(free_vfs_inode ? GIF_FREE_VFS_INODE : GIF_ALLOC_FAILED,
800 			&GFS2_I(inode)->i_flags);
801 		iput(inode);
802 	}
803 fail:
804 	return error;
805 }
806 
807 /**
808  * gfs2_create - Create a file
809  * @dir: The directory in which to create the file
810  * @dentry: The dentry of the new file
811  * @mode: The mode of the new file
812  *
813  * Returns: errno
814  */
815 
816 static int gfs2_create(struct inode *dir, struct dentry *dentry,
817 		       umode_t mode, bool excl)
818 {
819 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
820 }
821 
822 /**
823  * __gfs2_lookup - Look up a filename in a directory and return its inode
824  * @dir: The directory inode
825  * @dentry: The dentry of the new inode
826  * @file: File to be opened
827  * @opened: atomic_open flags
828  *
829  *
830  * Returns: errno
831  */
832 
833 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
834 				    struct file *file, int *opened)
835 {
836 	struct inode *inode;
837 	struct dentry *d;
838 	struct gfs2_holder gh;
839 	struct gfs2_glock *gl;
840 	int error;
841 
842 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
843 	if (!inode)
844 		return NULL;
845 	if (IS_ERR(inode))
846 		return ERR_CAST(inode);
847 
848 	gl = GFS2_I(inode)->i_gl;
849 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
850 	if (error) {
851 		iput(inode);
852 		return ERR_PTR(error);
853 	}
854 
855 	d = d_splice_alias(inode, dentry);
856 	if (IS_ERR(d)) {
857 		iput(inode);
858 		gfs2_glock_dq_uninit(&gh);
859 		return d;
860 	}
861 	if (file && S_ISREG(inode->i_mode))
862 		error = finish_open(file, dentry, gfs2_open_common, opened);
863 
864 	gfs2_glock_dq_uninit(&gh);
865 	if (error) {
866 		dput(d);
867 		return ERR_PTR(error);
868 	}
869 	return d;
870 }
871 
872 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
873 				  unsigned flags)
874 {
875 	return __gfs2_lookup(dir, dentry, NULL, NULL);
876 }
877 
878 /**
879  * gfs2_link - Link to a file
880  * @old_dentry: The inode to link
881  * @dir: Add link to this directory
882  * @dentry: The name of the link
883  *
884  * Link the inode in "old_dentry" into the directory "dir" with the
885  * name in "dentry".
886  *
887  * Returns: errno
888  */
889 
890 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
891 		     struct dentry *dentry)
892 {
893 	struct gfs2_inode *dip = GFS2_I(dir);
894 	struct gfs2_sbd *sdp = GFS2_SB(dir);
895 	struct inode *inode = old_dentry->d_inode;
896 	struct gfs2_inode *ip = GFS2_I(inode);
897 	struct gfs2_holder ghs[2];
898 	struct buffer_head *dibh;
899 	struct gfs2_diradd da = { .bh = NULL, };
900 	int error;
901 
902 	if (S_ISDIR(inode->i_mode))
903 		return -EPERM;
904 
905 	error = gfs2_rs_alloc(dip);
906 	if (error)
907 		return error;
908 
909 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
910 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
911 
912 	error = gfs2_glock_nq(ghs); /* parent */
913 	if (error)
914 		goto out_parent;
915 
916 	error = gfs2_glock_nq(ghs + 1); /* child */
917 	if (error)
918 		goto out_child;
919 
920 	error = -ENOENT;
921 	if (inode->i_nlink == 0)
922 		goto out_gunlock;
923 
924 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
925 	if (error)
926 		goto out_gunlock;
927 
928 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
929 	switch (error) {
930 	case -ENOENT:
931 		break;
932 	case 0:
933 		error = -EEXIST;
934 	default:
935 		goto out_gunlock;
936 	}
937 
938 	error = -EINVAL;
939 	if (!dip->i_inode.i_nlink)
940 		goto out_gunlock;
941 	error = -EFBIG;
942 	if (dip->i_entries == (u32)-1)
943 		goto out_gunlock;
944 	error = -EPERM;
945 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
946 		goto out_gunlock;
947 	error = -EINVAL;
948 	if (!ip->i_inode.i_nlink)
949 		goto out_gunlock;
950 	error = -EMLINK;
951 	if (ip->i_inode.i_nlink == (u32)-1)
952 		goto out_gunlock;
953 
954 	error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
955 	if (error < 0)
956 		goto out_gunlock;
957 
958 	if (da.nr_blocks) {
959 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
960 		error = gfs2_quota_lock_check(dip);
961 		if (error)
962 			goto out_gunlock;
963 
964 		error = gfs2_inplace_reserve(dip, &ap);
965 		if (error)
966 			goto out_gunlock_q;
967 
968 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(dip, &da, 2), 0);
969 		if (error)
970 			goto out_ipres;
971 	} else {
972 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
973 		if (error)
974 			goto out_ipres;
975 	}
976 
977 	error = gfs2_meta_inode_buffer(ip, &dibh);
978 	if (error)
979 		goto out_end_trans;
980 
981 	error = gfs2_dir_add(dir, &dentry->d_name, ip, &da);
982 	if (error)
983 		goto out_brelse;
984 
985 	gfs2_trans_add_meta(ip->i_gl, dibh);
986 	inc_nlink(&ip->i_inode);
987 	ip->i_inode.i_ctime = CURRENT_TIME;
988 	ihold(inode);
989 	d_instantiate(dentry, inode);
990 	mark_inode_dirty(inode);
991 
992 out_brelse:
993 	brelse(dibh);
994 out_end_trans:
995 	gfs2_trans_end(sdp);
996 out_ipres:
997 	if (da.nr_blocks)
998 		gfs2_inplace_release(dip);
999 out_gunlock_q:
1000 	if (da.nr_blocks)
1001 		gfs2_quota_unlock(dip);
1002 out_gunlock:
1003 	gfs2_dir_no_add(&da);
1004 	gfs2_glock_dq(ghs + 1);
1005 out_child:
1006 	gfs2_glock_dq(ghs);
1007 out_parent:
1008 	gfs2_holder_uninit(ghs);
1009 	gfs2_holder_uninit(ghs + 1);
1010 	return error;
1011 }
1012 
1013 /*
1014  * gfs2_unlink_ok - check to see that a inode is still in a directory
1015  * @dip: the directory
1016  * @name: the name of the file
1017  * @ip: the inode
1018  *
1019  * Assumes that the lock on (at least) @dip is held.
1020  *
1021  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1022  */
1023 
1024 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
1025 			  const struct gfs2_inode *ip)
1026 {
1027 	int error;
1028 
1029 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
1030 		return -EPERM;
1031 
1032 	if ((dip->i_inode.i_mode & S_ISVTX) &&
1033 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
1034 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
1035 		return -EPERM;
1036 
1037 	if (IS_APPEND(&dip->i_inode))
1038 		return -EPERM;
1039 
1040 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
1041 	if (error)
1042 		return error;
1043 
1044 	error = gfs2_dir_check(&dip->i_inode, name, ip);
1045 	if (error)
1046 		return error;
1047 
1048 	return 0;
1049 }
1050 
1051 /**
1052  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
1053  * @dip: The parent directory
1054  * @name: The name of the entry in the parent directory
1055  * @inode: The inode to be removed
1056  *
1057  * Called with all the locks and in a transaction. This will only be
1058  * called for a directory after it has been checked to ensure it is empty.
1059  *
1060  * Returns: 0 on success, or an error
1061  */
1062 
1063 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1064 			     const struct dentry *dentry)
1065 {
1066 	struct inode *inode = dentry->d_inode;
1067 	struct gfs2_inode *ip = GFS2_I(inode);
1068 	int error;
1069 
1070 	error = gfs2_dir_del(dip, dentry);
1071 	if (error)
1072 		return error;
1073 
1074 	ip->i_entries = 0;
1075 	inode->i_ctime = CURRENT_TIME;
1076 	if (S_ISDIR(inode->i_mode))
1077 		clear_nlink(inode);
1078 	else
1079 		drop_nlink(inode);
1080 	mark_inode_dirty(inode);
1081 	if (inode->i_nlink == 0)
1082 		gfs2_unlink_di(inode);
1083 	return 0;
1084 }
1085 
1086 
1087 /**
1088  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1089  * @dir: The inode of the directory containing the inode to unlink
1090  * @dentry: The file itself
1091  *
1092  * This routine uses the type of the inode as a flag to figure out
1093  * whether this is an unlink or an rmdir.
1094  *
1095  * Returns: errno
1096  */
1097 
1098 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1099 {
1100 	struct gfs2_inode *dip = GFS2_I(dir);
1101 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1102 	struct inode *inode = dentry->d_inode;
1103 	struct gfs2_inode *ip = GFS2_I(inode);
1104 	struct gfs2_holder ghs[3];
1105 	struct gfs2_rgrpd *rgd;
1106 	int error;
1107 
1108 	error = gfs2_rindex_update(sdp);
1109 	if (error)
1110 		return error;
1111 
1112 	error = -EROFS;
1113 
1114 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1115 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1116 
1117 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1118 	if (!rgd)
1119 		goto out_inodes;
1120 
1121 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1122 
1123 
1124 	error = gfs2_glock_nq(ghs); /* parent */
1125 	if (error)
1126 		goto out_parent;
1127 
1128 	error = gfs2_glock_nq(ghs + 1); /* child */
1129 	if (error)
1130 		goto out_child;
1131 
1132 	error = -ENOENT;
1133 	if (inode->i_nlink == 0)
1134 		goto out_rgrp;
1135 
1136 	if (S_ISDIR(inode->i_mode)) {
1137 		error = -ENOTEMPTY;
1138 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1139 			goto out_rgrp;
1140 	}
1141 
1142 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1143 	if (error)
1144 		goto out_rgrp;
1145 
1146 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1147 	if (error)
1148 		goto out_gunlock;
1149 
1150 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1151 	if (error)
1152 		goto out_end_trans;
1153 
1154 	error = gfs2_unlink_inode(dip, dentry);
1155 
1156 out_end_trans:
1157 	gfs2_trans_end(sdp);
1158 out_gunlock:
1159 	gfs2_glock_dq(ghs + 2);
1160 out_rgrp:
1161 	gfs2_glock_dq(ghs + 1);
1162 out_child:
1163 	gfs2_glock_dq(ghs);
1164 out_parent:
1165 	gfs2_holder_uninit(ghs + 2);
1166 out_inodes:
1167 	gfs2_holder_uninit(ghs + 1);
1168 	gfs2_holder_uninit(ghs);
1169 	return error;
1170 }
1171 
1172 /**
1173  * gfs2_symlink - Create a symlink
1174  * @dir: The directory to create the symlink in
1175  * @dentry: The dentry to put the symlink in
1176  * @symname: The thing which the link points to
1177  *
1178  * Returns: errno
1179  */
1180 
1181 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1182 			const char *symname)
1183 {
1184 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1185 	unsigned int size;
1186 
1187 	size = strlen(symname);
1188 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1189 		return -ENAMETOOLONG;
1190 
1191 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1192 }
1193 
1194 /**
1195  * gfs2_mkdir - Make a directory
1196  * @dir: The parent directory of the new one
1197  * @dentry: The dentry of the new directory
1198  * @mode: The mode of the new directory
1199  *
1200  * Returns: errno
1201  */
1202 
1203 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1204 {
1205 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1206 	unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1207 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1208 }
1209 
1210 /**
1211  * gfs2_mknod - Make a special file
1212  * @dir: The directory in which the special file will reside
1213  * @dentry: The dentry of the special file
1214  * @mode: The mode of the special file
1215  * @dev: The device specification of the special file
1216  *
1217  */
1218 
1219 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1220 		      dev_t dev)
1221 {
1222 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1223 }
1224 
1225 /**
1226  * gfs2_atomic_open - Atomically open a file
1227  * @dir: The directory
1228  * @dentry: The proposed new entry
1229  * @file: The proposed new struct file
1230  * @flags: open flags
1231  * @mode: File mode
1232  * @opened: Flag to say whether the file has been opened or not
1233  *
1234  * Returns: error code or 0 for success
1235  */
1236 
1237 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1238                             struct file *file, unsigned flags,
1239                             umode_t mode, int *opened)
1240 {
1241 	struct dentry *d;
1242 	bool excl = !!(flags & O_EXCL);
1243 
1244 	d = __gfs2_lookup(dir, dentry, file, opened);
1245 	if (IS_ERR(d))
1246 		return PTR_ERR(d);
1247 	if (d != NULL)
1248 		dentry = d;
1249 	if (dentry->d_inode) {
1250 		if (!(*opened & FILE_OPENED)) {
1251 			if (d == NULL)
1252 				dget(dentry);
1253 			return finish_no_open(file, dentry);
1254 		}
1255 		dput(d);
1256 		return 0;
1257 	}
1258 
1259 	BUG_ON(d != NULL);
1260 	if (!(flags & O_CREAT))
1261 		return -ENOENT;
1262 
1263 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1264 }
1265 
1266 /*
1267  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1268  * @this: move this
1269  * @to: to here
1270  *
1271  * Follow @to back to the root and make sure we don't encounter @this
1272  * Assumes we already hold the rename lock.
1273  *
1274  * Returns: errno
1275  */
1276 
1277 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1278 {
1279 	struct inode *dir = &to->i_inode;
1280 	struct super_block *sb = dir->i_sb;
1281 	struct inode *tmp;
1282 	int error = 0;
1283 
1284 	igrab(dir);
1285 
1286 	for (;;) {
1287 		if (dir == &this->i_inode) {
1288 			error = -EINVAL;
1289 			break;
1290 		}
1291 		if (dir == sb->s_root->d_inode) {
1292 			error = 0;
1293 			break;
1294 		}
1295 
1296 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1297 		if (!tmp) {
1298 			error = -ENOENT;
1299 			break;
1300 		}
1301 		if (IS_ERR(tmp)) {
1302 			error = PTR_ERR(tmp);
1303 			break;
1304 		}
1305 
1306 		iput(dir);
1307 		dir = tmp;
1308 	}
1309 
1310 	iput(dir);
1311 
1312 	return error;
1313 }
1314 
1315 /**
1316  * gfs2_rename - Rename a file
1317  * @odir: Parent directory of old file name
1318  * @odentry: The old dentry of the file
1319  * @ndir: Parent directory of new file name
1320  * @ndentry: The new dentry of the file
1321  *
1322  * Returns: errno
1323  */
1324 
1325 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1326 		       struct inode *ndir, struct dentry *ndentry)
1327 {
1328 	struct gfs2_inode *odip = GFS2_I(odir);
1329 	struct gfs2_inode *ndip = GFS2_I(ndir);
1330 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1331 	struct gfs2_inode *nip = NULL;
1332 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1333 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1334 	struct gfs2_rgrpd *nrgd;
1335 	unsigned int num_gh;
1336 	int dir_rename = 0;
1337 	struct gfs2_diradd da = { .nr_blocks = 0, };
1338 	unsigned int x;
1339 	int error;
1340 
1341 	if (ndentry->d_inode) {
1342 		nip = GFS2_I(ndentry->d_inode);
1343 		if (ip == nip)
1344 			return 0;
1345 	}
1346 
1347 	error = gfs2_rindex_update(sdp);
1348 	if (error)
1349 		return error;
1350 
1351 	error = gfs2_rs_alloc(ndip);
1352 	if (error)
1353 		return error;
1354 
1355 	if (odip != ndip) {
1356 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1357 					   0, &r_gh);
1358 		if (error)
1359 			goto out;
1360 
1361 		if (S_ISDIR(ip->i_inode.i_mode)) {
1362 			dir_rename = 1;
1363 			/* don't move a dirctory into it's subdir */
1364 			error = gfs2_ok_to_move(ip, ndip);
1365 			if (error)
1366 				goto out_gunlock_r;
1367 		}
1368 	}
1369 
1370 	num_gh = 1;
1371 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1372 	if (odip != ndip) {
1373 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1374 		num_gh++;
1375 	}
1376 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1377 	num_gh++;
1378 
1379 	if (nip) {
1380 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1381 		num_gh++;
1382 		/* grab the resource lock for unlink flag twiddling
1383 		 * this is the case of the target file already existing
1384 		 * so we unlink before doing the rename
1385 		 */
1386 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1387 		if (nrgd)
1388 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1389 	}
1390 
1391 	for (x = 0; x < num_gh; x++) {
1392 		error = gfs2_glock_nq(ghs + x);
1393 		if (error)
1394 			goto out_gunlock;
1395 	}
1396 
1397 	error = -ENOENT;
1398 	if (ip->i_inode.i_nlink == 0)
1399 		goto out_gunlock;
1400 
1401 	/* Check out the old directory */
1402 
1403 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1404 	if (error)
1405 		goto out_gunlock;
1406 
1407 	/* Check out the new directory */
1408 
1409 	if (nip) {
1410 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1411 		if (error)
1412 			goto out_gunlock;
1413 
1414 		if (nip->i_inode.i_nlink == 0) {
1415 			error = -EAGAIN;
1416 			goto out_gunlock;
1417 		}
1418 
1419 		if (S_ISDIR(nip->i_inode.i_mode)) {
1420 			if (nip->i_entries < 2) {
1421 				gfs2_consist_inode(nip);
1422 				error = -EIO;
1423 				goto out_gunlock;
1424 			}
1425 			if (nip->i_entries > 2) {
1426 				error = -ENOTEMPTY;
1427 				goto out_gunlock;
1428 			}
1429 		}
1430 	} else {
1431 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1432 		if (error)
1433 			goto out_gunlock;
1434 
1435 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1436 		switch (error) {
1437 		case -ENOENT:
1438 			error = 0;
1439 			break;
1440 		case 0:
1441 			error = -EEXIST;
1442 		default:
1443 			goto out_gunlock;
1444 		};
1445 
1446 		if (odip != ndip) {
1447 			if (!ndip->i_inode.i_nlink) {
1448 				error = -ENOENT;
1449 				goto out_gunlock;
1450 			}
1451 			if (ndip->i_entries == (u32)-1) {
1452 				error = -EFBIG;
1453 				goto out_gunlock;
1454 			}
1455 			if (S_ISDIR(ip->i_inode.i_mode) &&
1456 			    ndip->i_inode.i_nlink == (u32)-1) {
1457 				error = -EMLINK;
1458 				goto out_gunlock;
1459 			}
1460 		}
1461 	}
1462 
1463 	/* Check out the dir to be renamed */
1464 
1465 	if (dir_rename) {
1466 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1467 		if (error)
1468 			goto out_gunlock;
1469 	}
1470 
1471 	if (nip == NULL) {
1472 		error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1473 		if (error)
1474 			goto out_gunlock;
1475 	}
1476 
1477 	if (da.nr_blocks) {
1478 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1479 		error = gfs2_quota_lock_check(ndip);
1480 		if (error)
1481 			goto out_gunlock;
1482 
1483 		error = gfs2_inplace_reserve(ndip, &ap);
1484 		if (error)
1485 			goto out_gunlock_q;
1486 
1487 		error = gfs2_trans_begin(sdp, gfs2_trans_da_blks(ndip, &da, 4) +
1488 					 4 * RES_LEAF + 4, 0);
1489 		if (error)
1490 			goto out_ipreserv;
1491 	} else {
1492 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1493 					 5 * RES_LEAF + 4, 0);
1494 		if (error)
1495 			goto out_gunlock;
1496 	}
1497 
1498 	/* Remove the target file, if it exists */
1499 
1500 	if (nip)
1501 		error = gfs2_unlink_inode(ndip, ndentry);
1502 
1503 	if (dir_rename) {
1504 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1505 		if (error)
1506 			goto out_end_trans;
1507 	} else {
1508 		struct buffer_head *dibh;
1509 		error = gfs2_meta_inode_buffer(ip, &dibh);
1510 		if (error)
1511 			goto out_end_trans;
1512 		ip->i_inode.i_ctime = CURRENT_TIME;
1513 		gfs2_trans_add_meta(ip->i_gl, dibh);
1514 		gfs2_dinode_out(ip, dibh->b_data);
1515 		brelse(dibh);
1516 	}
1517 
1518 	error = gfs2_dir_del(odip, odentry);
1519 	if (error)
1520 		goto out_end_trans;
1521 
1522 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip, &da);
1523 	if (error)
1524 		goto out_end_trans;
1525 
1526 out_end_trans:
1527 	gfs2_trans_end(sdp);
1528 out_ipreserv:
1529 	if (da.nr_blocks)
1530 		gfs2_inplace_release(ndip);
1531 out_gunlock_q:
1532 	if (da.nr_blocks)
1533 		gfs2_quota_unlock(ndip);
1534 out_gunlock:
1535 	gfs2_dir_no_add(&da);
1536 	while (x--) {
1537 		gfs2_glock_dq(ghs + x);
1538 		gfs2_holder_uninit(ghs + x);
1539 	}
1540 out_gunlock_r:
1541 	if (r_gh.gh_gl)
1542 		gfs2_glock_dq_uninit(&r_gh);
1543 out:
1544 	return error;
1545 }
1546 
1547 /**
1548  * gfs2_follow_link - Follow a symbolic link
1549  * @dentry: The dentry of the link
1550  * @nd: Data that we pass to vfs_follow_link()
1551  *
1552  * This can handle symlinks of any size.
1553  *
1554  * Returns: 0 on success or error code
1555  */
1556 
1557 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1558 {
1559 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1560 	struct gfs2_holder i_gh;
1561 	struct buffer_head *dibh;
1562 	unsigned int size;
1563 	char *buf;
1564 	int error;
1565 
1566 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1567 	error = gfs2_glock_nq(&i_gh);
1568 	if (error) {
1569 		gfs2_holder_uninit(&i_gh);
1570 		nd_set_link(nd, ERR_PTR(error));
1571 		return NULL;
1572 	}
1573 
1574 	size = (unsigned int)i_size_read(&ip->i_inode);
1575 	if (size == 0) {
1576 		gfs2_consist_inode(ip);
1577 		buf = ERR_PTR(-EIO);
1578 		goto out;
1579 	}
1580 
1581 	error = gfs2_meta_inode_buffer(ip, &dibh);
1582 	if (error) {
1583 		buf = ERR_PTR(error);
1584 		goto out;
1585 	}
1586 
1587 	buf = kzalloc(size + 1, GFP_NOFS);
1588 	if (!buf)
1589 		buf = ERR_PTR(-ENOMEM);
1590 	else
1591 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1592 	brelse(dibh);
1593 out:
1594 	gfs2_glock_dq_uninit(&i_gh);
1595 	nd_set_link(nd, buf);
1596 	return NULL;
1597 }
1598 
1599 /**
1600  * gfs2_permission -
1601  * @inode: The inode
1602  * @mask: The mask to be tested
1603  * @flags: Indicates whether this is an RCU path walk or not
1604  *
1605  * This may be called from the VFS directly, or from within GFS2 with the
1606  * inode locked, so we look to see if the glock is already locked and only
1607  * lock the glock if its not already been done.
1608  *
1609  * Returns: errno
1610  */
1611 
1612 int gfs2_permission(struct inode *inode, int mask)
1613 {
1614 	struct gfs2_inode *ip;
1615 	struct gfs2_holder i_gh;
1616 	int error;
1617 	int unlock = 0;
1618 
1619 
1620 	ip = GFS2_I(inode);
1621 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1622 		if (mask & MAY_NOT_BLOCK)
1623 			return -ECHILD;
1624 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1625 		if (error)
1626 			return error;
1627 		unlock = 1;
1628 	}
1629 
1630 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1631 		error = -EACCES;
1632 	else
1633 		error = generic_permission(inode, mask);
1634 	if (unlock)
1635 		gfs2_glock_dq_uninit(&i_gh);
1636 
1637 	return error;
1638 }
1639 
1640 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1641 {
1642 	setattr_copy(inode, attr);
1643 	mark_inode_dirty(inode);
1644 	return 0;
1645 }
1646 
1647 /**
1648  * gfs2_setattr_simple -
1649  * @ip:
1650  * @attr:
1651  *
1652  * Returns: errno
1653  */
1654 
1655 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1656 {
1657 	int error;
1658 
1659 	if (current->journal_info)
1660 		return __gfs2_setattr_simple(inode, attr);
1661 
1662 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1663 	if (error)
1664 		return error;
1665 
1666 	error = __gfs2_setattr_simple(inode, attr);
1667 	gfs2_trans_end(GFS2_SB(inode));
1668 	return error;
1669 }
1670 
1671 static int setattr_chown(struct inode *inode, struct iattr *attr)
1672 {
1673 	struct gfs2_inode *ip = GFS2_I(inode);
1674 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1675 	kuid_t ouid, nuid;
1676 	kgid_t ogid, ngid;
1677 	int error;
1678 
1679 	ouid = inode->i_uid;
1680 	ogid = inode->i_gid;
1681 	nuid = attr->ia_uid;
1682 	ngid = attr->ia_gid;
1683 
1684 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1685 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1686 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1687 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1688 
1689 	error = get_write_access(inode);
1690 	if (error)
1691 		return error;
1692 
1693 	error = gfs2_rs_alloc(ip);
1694 	if (error)
1695 		goto out;
1696 
1697 	error = gfs2_rindex_update(sdp);
1698 	if (error)
1699 		goto out;
1700 
1701 	error = gfs2_quota_lock(ip, nuid, ngid);
1702 	if (error)
1703 		goto out;
1704 
1705 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1706 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1707 		error = gfs2_quota_check(ip, nuid, ngid);
1708 		if (error)
1709 			goto out_gunlock_q;
1710 	}
1711 
1712 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1713 	if (error)
1714 		goto out_gunlock_q;
1715 
1716 	error = gfs2_setattr_simple(inode, attr);
1717 	if (error)
1718 		goto out_end_trans;
1719 
1720 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1721 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1722 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1723 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1724 		gfs2_quota_change(ip, blocks, nuid, ngid);
1725 	}
1726 
1727 out_end_trans:
1728 	gfs2_trans_end(sdp);
1729 out_gunlock_q:
1730 	gfs2_quota_unlock(ip);
1731 out:
1732 	put_write_access(inode);
1733 	return error;
1734 }
1735 
1736 /**
1737  * gfs2_setattr - Change attributes on an inode
1738  * @dentry: The dentry which is changing
1739  * @attr: The structure describing the change
1740  *
1741  * The VFS layer wants to change one or more of an inodes attributes.  Write
1742  * that change out to disk.
1743  *
1744  * Returns: errno
1745  */
1746 
1747 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1748 {
1749 	struct inode *inode = dentry->d_inode;
1750 	struct gfs2_inode *ip = GFS2_I(inode);
1751 	struct gfs2_holder i_gh;
1752 	int error;
1753 
1754 	error = gfs2_rs_alloc(ip);
1755 	if (error)
1756 		return error;
1757 
1758 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1759 	if (error)
1760 		return error;
1761 
1762 	error = -EPERM;
1763 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1764 		goto out;
1765 
1766 	error = inode_change_ok(inode, attr);
1767 	if (error)
1768 		goto out;
1769 
1770 	if (attr->ia_valid & ATTR_SIZE)
1771 		error = gfs2_setattr_size(inode, attr->ia_size);
1772 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1773 		error = setattr_chown(inode, attr);
1774 	else {
1775 		error = gfs2_setattr_simple(inode, attr);
1776 		if (!error && attr->ia_valid & ATTR_MODE)
1777 			error = posix_acl_chmod(inode, inode->i_mode);
1778 	}
1779 
1780 out:
1781 	if (!error)
1782 		mark_inode_dirty(inode);
1783 	gfs2_glock_dq_uninit(&i_gh);
1784 	return error;
1785 }
1786 
1787 /**
1788  * gfs2_getattr - Read out an inode's attributes
1789  * @mnt: The vfsmount the inode is being accessed from
1790  * @dentry: The dentry to stat
1791  * @stat: The inode's stats
1792  *
1793  * This may be called from the VFS directly, or from within GFS2 with the
1794  * inode locked, so we look to see if the glock is already locked and only
1795  * lock the glock if its not already been done. Note that its the NFS
1796  * readdirplus operation which causes this to be called (from filldir)
1797  * with the glock already held.
1798  *
1799  * Returns: errno
1800  */
1801 
1802 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1803 			struct kstat *stat)
1804 {
1805 	struct inode *inode = dentry->d_inode;
1806 	struct gfs2_inode *ip = GFS2_I(inode);
1807 	struct gfs2_holder gh;
1808 	int error;
1809 	int unlock = 0;
1810 
1811 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1812 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1813 		if (error)
1814 			return error;
1815 		unlock = 1;
1816 	}
1817 
1818 	generic_fillattr(inode, stat);
1819 	if (unlock)
1820 		gfs2_glock_dq_uninit(&gh);
1821 
1822 	return 0;
1823 }
1824 
1825 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1826 			 const void *data, size_t size, int flags)
1827 {
1828 	struct inode *inode = dentry->d_inode;
1829 	struct gfs2_inode *ip = GFS2_I(inode);
1830 	struct gfs2_holder gh;
1831 	int ret;
1832 
1833 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1834 	ret = gfs2_glock_nq(&gh);
1835 	if (ret == 0) {
1836 		ret = gfs2_rs_alloc(ip);
1837 		if (ret == 0)
1838 			ret = generic_setxattr(dentry, name, data, size, flags);
1839 		gfs2_glock_dq(&gh);
1840 	}
1841 	gfs2_holder_uninit(&gh);
1842 	return ret;
1843 }
1844 
1845 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1846 			     void *data, size_t size)
1847 {
1848 	struct inode *inode = dentry->d_inode;
1849 	struct gfs2_inode *ip = GFS2_I(inode);
1850 	struct gfs2_holder gh;
1851 	int ret;
1852 
1853 	/* For selinux during lookup */
1854 	if (gfs2_glock_is_locked_by_me(ip->i_gl))
1855 		return generic_getxattr(dentry, name, data, size);
1856 
1857 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1858 	ret = gfs2_glock_nq(&gh);
1859 	if (ret == 0) {
1860 		ret = generic_getxattr(dentry, name, data, size);
1861 		gfs2_glock_dq(&gh);
1862 	}
1863 	gfs2_holder_uninit(&gh);
1864 	return ret;
1865 }
1866 
1867 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1868 {
1869 	struct inode *inode = dentry->d_inode;
1870 	struct gfs2_inode *ip = GFS2_I(inode);
1871 	struct gfs2_holder gh;
1872 	int ret;
1873 
1874 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1875 	ret = gfs2_glock_nq(&gh);
1876 	if (ret == 0) {
1877 		ret = gfs2_rs_alloc(ip);
1878 		if (ret == 0)
1879 			ret = generic_removexattr(dentry, name);
1880 		gfs2_glock_dq(&gh);
1881 	}
1882 	gfs2_holder_uninit(&gh);
1883 	return ret;
1884 }
1885 
1886 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1887 		       u64 start, u64 len)
1888 {
1889 	struct gfs2_inode *ip = GFS2_I(inode);
1890 	struct gfs2_holder gh;
1891 	int ret;
1892 
1893 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1894 	if (ret)
1895 		return ret;
1896 
1897 	mutex_lock(&inode->i_mutex);
1898 
1899 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1900 	if (ret)
1901 		goto out;
1902 
1903 	if (gfs2_is_stuffed(ip)) {
1904 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1905 		u64 size = i_size_read(inode);
1906 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1907 			    FIEMAP_EXTENT_DATA_INLINE;
1908 		phys += sizeof(struct gfs2_dinode);
1909 		phys += start;
1910 		if (start + len > size)
1911 			len = size - start;
1912 		if (start < size)
1913 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1914 						      len, flags);
1915 		if (ret == 1)
1916 			ret = 0;
1917 	} else {
1918 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1919 					     gfs2_block_map);
1920 	}
1921 
1922 	gfs2_glock_dq_uninit(&gh);
1923 out:
1924 	mutex_unlock(&inode->i_mutex);
1925 	return ret;
1926 }
1927 
1928 const struct inode_operations gfs2_file_iops = {
1929 	.permission = gfs2_permission,
1930 	.setattr = gfs2_setattr,
1931 	.getattr = gfs2_getattr,
1932 	.setxattr = gfs2_setxattr,
1933 	.getxattr = gfs2_getxattr,
1934 	.listxattr = gfs2_listxattr,
1935 	.removexattr = gfs2_removexattr,
1936 	.fiemap = gfs2_fiemap,
1937 	.get_acl = gfs2_get_acl,
1938 	.set_acl = gfs2_set_acl,
1939 };
1940 
1941 const struct inode_operations gfs2_dir_iops = {
1942 	.create = gfs2_create,
1943 	.lookup = gfs2_lookup,
1944 	.link = gfs2_link,
1945 	.unlink = gfs2_unlink,
1946 	.symlink = gfs2_symlink,
1947 	.mkdir = gfs2_mkdir,
1948 	.rmdir = gfs2_unlink,
1949 	.mknod = gfs2_mknod,
1950 	.rename = gfs2_rename,
1951 	.permission = gfs2_permission,
1952 	.setattr = gfs2_setattr,
1953 	.getattr = gfs2_getattr,
1954 	.setxattr = gfs2_setxattr,
1955 	.getxattr = gfs2_getxattr,
1956 	.listxattr = gfs2_listxattr,
1957 	.removexattr = gfs2_removexattr,
1958 	.fiemap = gfs2_fiemap,
1959 	.get_acl = gfs2_get_acl,
1960 	.set_acl = gfs2_set_acl,
1961 	.atomic_open = gfs2_atomic_open,
1962 };
1963 
1964 const struct inode_operations gfs2_symlink_iops = {
1965 	.readlink = generic_readlink,
1966 	.follow_link = gfs2_follow_link,
1967 	.put_link = kfree_put_link,
1968 	.permission = gfs2_permission,
1969 	.setattr = gfs2_setattr,
1970 	.getattr = gfs2_getattr,
1971 	.setxattr = gfs2_setxattr,
1972 	.getxattr = gfs2_getxattr,
1973 	.listxattr = gfs2_listxattr,
1974 	.removexattr = gfs2_removexattr,
1975 	.fiemap = gfs2_fiemap,
1976 };
1977 
1978