xref: /openbmc/linux/fs/gfs2/inode.c (revision 3c1c0ae1)
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(-ENOBUFS);
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)
380 {
381 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
382 	struct gfs2_alloc_parms ap = { .target = RES_DINODE, .aflags = flags, };
383 	int error;
384 	int dblocks = 1;
385 
386 	error = gfs2_quota_lock_check(ip);
387 	if (error)
388 		goto out;
389 
390 	error = gfs2_inplace_reserve(ip, &ap);
391 	if (error)
392 		goto out_quota;
393 
394 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 0);
395 	if (error)
396 		goto out_ipreserv;
397 
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;
402 
403 	gfs2_trans_end(sdp);
404 
405 out_ipreserv:
406 	gfs2_inplace_release(ip);
407 out_quota:
408 	gfs2_quota_unlock(ip);
409 out:
410 	return error;
411 }
412 
413 static void gfs2_init_dir(struct buffer_head *dibh,
414 			  const struct gfs2_inode *parent)
415 {
416 	struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
417 	struct gfs2_dirent *dent = (struct gfs2_dirent *)(di+1);
418 
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);
422 
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);
427 
428 }
429 
430 /**
431  * init_dinode - Fill in a new dinode structure
432  * @dip: The directory this inode is being created in
433  * @ip: The inode
434  * @symname: The symlink destination (if a symlink)
435  * @bhp: The buffer head (returned to caller)
436  *
437  */
438 
439 static void init_dinode(struct gfs2_inode *dip, struct gfs2_inode *ip,
440 			const char *symname)
441 {
442 	struct gfs2_dinode *di;
443 	struct buffer_head *dibh;
444 
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);
449 
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));
452 	di->__pad1 = 0;
453 	di->__pad2 = 0;
454 	di->__pad3 = 0;
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));
458 
459 	switch(ip->i_inode.i_mode & S_IFMT) {
460 	case S_IFDIR:
461 		gfs2_init_dir(dibh, dip);
462 		break;
463 	case S_IFLNK:
464 		memcpy(dibh->b_data + sizeof(struct gfs2_dinode), symname, ip->i_inode.i_size);
465 		break;
466 	}
467 
468 	set_buffer_uptodate(dibh);
469 	brelse(dibh);
470 }
471 
472 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
473 		       struct gfs2_inode *ip, struct gfs2_diradd *da)
474 {
475 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
476 	struct gfs2_alloc_parms ap = { .target = da->nr_blocks, };
477 	int error;
478 
479 	if (da->nr_blocks) {
480 		error = gfs2_quota_lock_check(dip);
481 		if (error)
482 			goto fail_quota_locks;
483 
484 		error = gfs2_inplace_reserve(dip, &ap);
485 		if (error)
486 			goto fail_quota_locks;
487 
488 		error = gfs2_trans_begin(sdp, da->nr_blocks +
489 					 gfs2_rg_blocks(dip, da->nr_blocks) +
490 					 2 * RES_DINODE +
491 					 RES_STATFS + RES_QUOTA, 0);
492 		if (error)
493 			goto fail_ipreserv;
494 	} else {
495 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
496 		if (error)
497 			goto fail_quota_locks;
498 	}
499 
500 	error = gfs2_dir_add(&dip->i_inode, name, ip);
501 	if (error)
502 		goto fail_end_trans;
503 
504 fail_end_trans:
505 	gfs2_trans_end(sdp);
506 fail_ipreserv:
507 	gfs2_inplace_release(dip);
508 fail_quota_locks:
509 	gfs2_quota_unlock(dip);
510 	return error;
511 }
512 
513 static int gfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
514 		    void *fs_info)
515 {
516 	const struct xattr *xattr;
517 	int err = 0;
518 
519 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
520 		err = __gfs2_xattr_set(inode, xattr->name, xattr->value,
521 				       xattr->value_len, 0,
522 				       GFS2_EATYPE_SECURITY);
523 		if (err < 0)
524 			break;
525 	}
526 	return err;
527 }
528 
529 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip,
530 			      const struct qstr *qstr)
531 {
532 	return security_inode_init_security(&ip->i_inode, &dip->i_inode, qstr,
533 					    &gfs2_initxattrs, NULL);
534 }
535 
536 /**
537  * gfs2_create_inode - Create a new inode
538  * @dir: The parent directory
539  * @dentry: The new dentry
540  * @file: If non-NULL, the file which is being opened
541  * @mode: The permissions on the new inode
542  * @dev: For device nodes, this is the device number
543  * @symname: For symlinks, this is the link destination
544  * @size: The initial size of the inode (ignored for directories)
545  *
546  * Returns: 0 on success, or error code
547  */
548 
549 static int gfs2_create_inode(struct inode *dir, struct dentry *dentry,
550 			     struct file *file,
551 			     umode_t mode, dev_t dev, const char *symname,
552 			     unsigned int size, int excl, int *opened)
553 {
554 	const struct qstr *name = &dentry->d_name;
555 	struct gfs2_holder ghs[2];
556 	struct inode *inode = NULL;
557 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
558 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
559 	struct gfs2_glock *io_gl;
560 	struct dentry *d;
561 	int error;
562 	u32 aflags = 0;
563 	struct gfs2_diradd da;
564 
565 	if (!name->len || name->len > GFS2_FNAMESIZE)
566 		return -ENAMETOOLONG;
567 
568 	error = gfs2_rs_alloc(dip);
569 	if (error)
570 		return error;
571 
572 	error = gfs2_rindex_update(sdp);
573 	if (error)
574 		return error;
575 
576 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
577 	if (error)
578 		goto fail;
579 
580 	error = create_ok(dip, name, mode);
581 	if (error)
582 		goto fail_gunlock;
583 
584 	inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
585 	error = PTR_ERR(inode);
586 	if (!IS_ERR(inode)) {
587 		d = d_splice_alias(inode, dentry);
588 		error = 0;
589 		if (file) {
590 			if (S_ISREG(inode->i_mode)) {
591 				WARN_ON(d != NULL);
592 				error = finish_open(file, dentry, gfs2_open_common, opened);
593 			} else {
594 				error = finish_no_open(file, d);
595 			}
596 		} else {
597 			dput(d);
598 		}
599 		gfs2_glock_dq_uninit(ghs);
600 		return error;
601 	} else if (error != -ENOENT) {
602 		goto fail_gunlock;
603 	}
604 
605 	error = gfs2_diradd_alloc_required(dir, name, &da);
606 	if (error < 0)
607 		goto fail_gunlock;
608 
609 	inode = new_inode(sdp->sd_vfs);
610 	error = -ENOMEM;
611 	if (!inode)
612 		goto fail_gunlock;
613 
614 	ip = GFS2_I(inode);
615 	error = gfs2_rs_alloc(ip);
616 	if (error)
617 		goto fail_free_inode;
618 
619 	inode->i_mode = mode;
620 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
621 	inode->i_rdev = dev;
622 	inode->i_size = size;
623 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
624 	gfs2_set_inode_blocks(inode, 1);
625 	munge_mode_uid_gid(dip, inode);
626 	ip->i_goal = dip->i_goal;
627 	ip->i_diskflags = 0;
628 	ip->i_eattr = 0;
629 	ip->i_height = 0;
630 	ip->i_depth = 0;
631 	ip->i_entries = 0;
632 
633 	switch(mode & S_IFMT) {
634 	case S_IFREG:
635 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
636 		    gfs2_tune_get(sdp, gt_new_files_jdata))
637 			ip->i_diskflags |= GFS2_DIF_JDATA;
638 		gfs2_set_aops(inode);
639 		break;
640 	case S_IFDIR:
641 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
642 		ip->i_diskflags |= GFS2_DIF_JDATA;
643 		ip->i_entries = 2;
644 		break;
645 	}
646 	gfs2_set_inode_flags(inode);
647 
648 	if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
649 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
650 		aflags |= GFS2_AF_ORLOV;
651 
652 	error = alloc_dinode(ip, aflags);
653 	if (error)
654 		goto fail_free_inode;
655 
656 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
657 	if (error)
658 		goto fail_free_inode;
659 
660 	ip->i_gl->gl_object = ip;
661 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
662 	if (error)
663 		goto fail_free_inode;
664 
665 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
666 	if (error)
667 		goto fail_gunlock2;
668 
669 	init_dinode(dip, ip, symname);
670 	gfs2_trans_end(sdp);
671 
672 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
673 	if (error)
674 		goto fail_gunlock2;
675 
676 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
677 	if (error)
678 		goto fail_gunlock2;
679 
680 	ip->i_iopen_gh.gh_gl->gl_object = ip;
681 	gfs2_glock_put(io_gl);
682 	gfs2_set_iop(inode);
683 	insert_inode_hash(inode);
684 
685 	error = gfs2_acl_create(dip, inode);
686 	if (error)
687 		goto fail_gunlock3;
688 
689 	error = gfs2_security_init(dip, ip, name);
690 	if (error)
691 		goto fail_gunlock3;
692 
693 	error = link_dinode(dip, name, ip, &da);
694 	if (error)
695 		goto fail_gunlock3;
696 
697 	mark_inode_dirty(inode);
698 	d_instantiate(dentry, inode);
699 	if (file) {
700 		*opened |= FILE_CREATED;
701 		error = finish_open(file, dentry, gfs2_open_common, opened);
702 	}
703 	gfs2_glock_dq_uninit(ghs);
704 	gfs2_glock_dq_uninit(ghs + 1);
705 	return error;
706 
707 fail_gunlock3:
708 	gfs2_glock_dq_uninit(ghs + 1);
709 	if (ip->i_gl)
710 		gfs2_glock_put(ip->i_gl);
711 	goto fail_gunlock;
712 
713 fail_gunlock2:
714 	gfs2_glock_dq_uninit(ghs + 1);
715 fail_free_inode:
716 	if (ip->i_gl)
717 		gfs2_glock_put(ip->i_gl);
718 	gfs2_rs_delete(ip, NULL);
719 	free_inode_nonrcu(inode);
720 	inode = NULL;
721 fail_gunlock:
722 	gfs2_glock_dq_uninit(ghs);
723 	if (inode && !IS_ERR(inode)) {
724 		clear_nlink(inode);
725 		mark_inode_dirty(inode);
726 		set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
727 		iput(inode);
728 	}
729 fail:
730 	return error;
731 }
732 
733 /**
734  * gfs2_create - Create a file
735  * @dir: The directory in which to create the file
736  * @dentry: The dentry of the new file
737  * @mode: The mode of the new file
738  *
739  * Returns: errno
740  */
741 
742 static int gfs2_create(struct inode *dir, struct dentry *dentry,
743 		       umode_t mode, bool excl)
744 {
745 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
746 }
747 
748 /**
749  * __gfs2_lookup - Look up a filename in a directory and return its inode
750  * @dir: The directory inode
751  * @dentry: The dentry of the new inode
752  * @file: File to be opened
753  * @opened: atomic_open flags
754  *
755  *
756  * Returns: errno
757  */
758 
759 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
760 				    struct file *file, int *opened)
761 {
762 	struct inode *inode;
763 	struct dentry *d;
764 	struct gfs2_holder gh;
765 	struct gfs2_glock *gl;
766 	int error;
767 
768 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
769 	if (!inode)
770 		return NULL;
771 	if (IS_ERR(inode))
772 		return ERR_CAST(inode);
773 
774 	gl = GFS2_I(inode)->i_gl;
775 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
776 	if (error) {
777 		iput(inode);
778 		return ERR_PTR(error);
779 	}
780 
781 	d = d_splice_alias(inode, dentry);
782 	if (file && S_ISREG(inode->i_mode))
783 		error = finish_open(file, dentry, gfs2_open_common, opened);
784 
785 	gfs2_glock_dq_uninit(&gh);
786 	if (error) {
787 		dput(d);
788 		return ERR_PTR(error);
789 	}
790 	return d;
791 }
792 
793 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
794 				  unsigned flags)
795 {
796 	return __gfs2_lookup(dir, dentry, NULL, NULL);
797 }
798 
799 /**
800  * gfs2_link - Link to a file
801  * @old_dentry: The inode to link
802  * @dir: Add link to this directory
803  * @dentry: The name of the link
804  *
805  * Link the inode in "old_dentry" into the directory "dir" with the
806  * name in "dentry".
807  *
808  * Returns: errno
809  */
810 
811 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
812 		     struct dentry *dentry)
813 {
814 	struct gfs2_inode *dip = GFS2_I(dir);
815 	struct gfs2_sbd *sdp = GFS2_SB(dir);
816 	struct inode *inode = old_dentry->d_inode;
817 	struct gfs2_inode *ip = GFS2_I(inode);
818 	struct gfs2_holder ghs[2];
819 	struct buffer_head *dibh;
820 	struct gfs2_diradd da;
821 	int error;
822 
823 	if (S_ISDIR(inode->i_mode))
824 		return -EPERM;
825 
826 	error = gfs2_rs_alloc(dip);
827 	if (error)
828 		return error;
829 
830 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
831 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
832 
833 	error = gfs2_glock_nq(ghs); /* parent */
834 	if (error)
835 		goto out_parent;
836 
837 	error = gfs2_glock_nq(ghs + 1); /* child */
838 	if (error)
839 		goto out_child;
840 
841 	error = -ENOENT;
842 	if (inode->i_nlink == 0)
843 		goto out_gunlock;
844 
845 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
846 	if (error)
847 		goto out_gunlock;
848 
849 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
850 	switch (error) {
851 	case -ENOENT:
852 		break;
853 	case 0:
854 		error = -EEXIST;
855 	default:
856 		goto out_gunlock;
857 	}
858 
859 	error = -EINVAL;
860 	if (!dip->i_inode.i_nlink)
861 		goto out_gunlock;
862 	error = -EFBIG;
863 	if (dip->i_entries == (u32)-1)
864 		goto out_gunlock;
865 	error = -EPERM;
866 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
867 		goto out_gunlock;
868 	error = -EINVAL;
869 	if (!ip->i_inode.i_nlink)
870 		goto out_gunlock;
871 	error = -EMLINK;
872 	if (ip->i_inode.i_nlink == (u32)-1)
873 		goto out_gunlock;
874 
875 	error = gfs2_diradd_alloc_required(dir, &dentry->d_name, &da);
876 	if (error < 0)
877 		goto out_gunlock;
878 
879 	if (da.nr_blocks) {
880 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
881 		error = gfs2_quota_lock_check(dip);
882 		if (error)
883 			goto out_gunlock;
884 
885 		error = gfs2_inplace_reserve(dip, &ap);
886 		if (error)
887 			goto out_gunlock_q;
888 
889 		error = gfs2_trans_begin(sdp, da.nr_blocks +
890 					 gfs2_rg_blocks(dip, da.nr_blocks) +
891 					 2 * RES_DINODE + RES_STATFS +
892 					 RES_QUOTA, 0);
893 		if (error)
894 			goto out_ipres;
895 	} else {
896 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
897 		if (error)
898 			goto out_ipres;
899 	}
900 
901 	error = gfs2_meta_inode_buffer(ip, &dibh);
902 	if (error)
903 		goto out_end_trans;
904 
905 	error = gfs2_dir_add(dir, &dentry->d_name, ip);
906 	if (error)
907 		goto out_brelse;
908 
909 	gfs2_trans_add_meta(ip->i_gl, dibh);
910 	inc_nlink(&ip->i_inode);
911 	ip->i_inode.i_ctime = CURRENT_TIME;
912 	ihold(inode);
913 	d_instantiate(dentry, inode);
914 	mark_inode_dirty(inode);
915 
916 out_brelse:
917 	brelse(dibh);
918 out_end_trans:
919 	gfs2_trans_end(sdp);
920 out_ipres:
921 	if (da.nr_blocks)
922 		gfs2_inplace_release(dip);
923 out_gunlock_q:
924 	if (da.nr_blocks)
925 		gfs2_quota_unlock(dip);
926 out_gunlock:
927 	gfs2_glock_dq(ghs + 1);
928 out_child:
929 	gfs2_glock_dq(ghs);
930 out_parent:
931 	gfs2_holder_uninit(ghs);
932 	gfs2_holder_uninit(ghs + 1);
933 	return error;
934 }
935 
936 /*
937  * gfs2_unlink_ok - check to see that a inode is still in a directory
938  * @dip: the directory
939  * @name: the name of the file
940  * @ip: the inode
941  *
942  * Assumes that the lock on (at least) @dip is held.
943  *
944  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
945  */
946 
947 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
948 			  const struct gfs2_inode *ip)
949 {
950 	int error;
951 
952 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
953 		return -EPERM;
954 
955 	if ((dip->i_inode.i_mode & S_ISVTX) &&
956 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
957 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
958 		return -EPERM;
959 
960 	if (IS_APPEND(&dip->i_inode))
961 		return -EPERM;
962 
963 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
964 	if (error)
965 		return error;
966 
967 	error = gfs2_dir_check(&dip->i_inode, name, ip);
968 	if (error)
969 		return error;
970 
971 	return 0;
972 }
973 
974 /**
975  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
976  * @dip: The parent directory
977  * @name: The name of the entry in the parent directory
978  * @inode: The inode to be removed
979  *
980  * Called with all the locks and in a transaction. This will only be
981  * called for a directory after it has been checked to ensure it is empty.
982  *
983  * Returns: 0 on success, or an error
984  */
985 
986 static int gfs2_unlink_inode(struct gfs2_inode *dip,
987 			     const struct dentry *dentry)
988 {
989 	struct inode *inode = dentry->d_inode;
990 	struct gfs2_inode *ip = GFS2_I(inode);
991 	int error;
992 
993 	error = gfs2_dir_del(dip, dentry);
994 	if (error)
995 		return error;
996 
997 	ip->i_entries = 0;
998 	inode->i_ctime = CURRENT_TIME;
999 	if (S_ISDIR(inode->i_mode))
1000 		clear_nlink(inode);
1001 	else
1002 		drop_nlink(inode);
1003 	mark_inode_dirty(inode);
1004 	if (inode->i_nlink == 0)
1005 		gfs2_unlink_di(inode);
1006 	return 0;
1007 }
1008 
1009 
1010 /**
1011  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1012  * @dir: The inode of the directory containing the inode to unlink
1013  * @dentry: The file itself
1014  *
1015  * This routine uses the type of the inode as a flag to figure out
1016  * whether this is an unlink or an rmdir.
1017  *
1018  * Returns: errno
1019  */
1020 
1021 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1022 {
1023 	struct gfs2_inode *dip = GFS2_I(dir);
1024 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1025 	struct inode *inode = dentry->d_inode;
1026 	struct gfs2_inode *ip = GFS2_I(inode);
1027 	struct gfs2_holder ghs[3];
1028 	struct gfs2_rgrpd *rgd;
1029 	int error;
1030 
1031 	error = gfs2_rindex_update(sdp);
1032 	if (error)
1033 		return error;
1034 
1035 	error = -EROFS;
1036 
1037 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1038 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1039 
1040 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1041 	if (!rgd)
1042 		goto out_inodes;
1043 
1044 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1045 
1046 
1047 	error = gfs2_glock_nq(ghs); /* parent */
1048 	if (error)
1049 		goto out_parent;
1050 
1051 	error = gfs2_glock_nq(ghs + 1); /* child */
1052 	if (error)
1053 		goto out_child;
1054 
1055 	error = -ENOENT;
1056 	if (inode->i_nlink == 0)
1057 		goto out_rgrp;
1058 
1059 	if (S_ISDIR(inode->i_mode)) {
1060 		error = -ENOTEMPTY;
1061 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1062 			goto out_rgrp;
1063 	}
1064 
1065 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1066 	if (error)
1067 		goto out_rgrp;
1068 
1069 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1070 	if (error)
1071 		goto out_gunlock;
1072 
1073 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1074 	if (error)
1075 		goto out_end_trans;
1076 
1077 	error = gfs2_unlink_inode(dip, dentry);
1078 
1079 out_end_trans:
1080 	gfs2_trans_end(sdp);
1081 out_gunlock:
1082 	gfs2_glock_dq(ghs + 2);
1083 out_rgrp:
1084 	gfs2_glock_dq(ghs + 1);
1085 out_child:
1086 	gfs2_glock_dq(ghs);
1087 out_parent:
1088 	gfs2_holder_uninit(ghs + 2);
1089 out_inodes:
1090 	gfs2_holder_uninit(ghs + 1);
1091 	gfs2_holder_uninit(ghs);
1092 	return error;
1093 }
1094 
1095 /**
1096  * gfs2_symlink - Create a symlink
1097  * @dir: The directory to create the symlink in
1098  * @dentry: The dentry to put the symlink in
1099  * @symname: The thing which the link points to
1100  *
1101  * Returns: errno
1102  */
1103 
1104 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1105 			const char *symname)
1106 {
1107 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1108 	unsigned int size;
1109 
1110 	size = strlen(symname);
1111 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1112 		return -ENAMETOOLONG;
1113 
1114 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1115 }
1116 
1117 /**
1118  * gfs2_mkdir - Make a directory
1119  * @dir: The parent directory of the new one
1120  * @dentry: The dentry of the new directory
1121  * @mode: The mode of the new directory
1122  *
1123  * Returns: errno
1124  */
1125 
1126 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1127 {
1128 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1129 	unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1130 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1131 }
1132 
1133 /**
1134  * gfs2_mknod - Make a special file
1135  * @dir: The directory in which the special file will reside
1136  * @dentry: The dentry of the special file
1137  * @mode: The mode of the special file
1138  * @dev: The device specification of the special file
1139  *
1140  */
1141 
1142 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1143 		      dev_t dev)
1144 {
1145 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1146 }
1147 
1148 /**
1149  * gfs2_atomic_open - Atomically open a file
1150  * @dir: The directory
1151  * @dentry: The proposed new entry
1152  * @file: The proposed new struct file
1153  * @flags: open flags
1154  * @mode: File mode
1155  * @opened: Flag to say whether the file has been opened or not
1156  *
1157  * Returns: error code or 0 for success
1158  */
1159 
1160 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1161                             struct file *file, unsigned flags,
1162                             umode_t mode, int *opened)
1163 {
1164 	struct dentry *d;
1165 	bool excl = !!(flags & O_EXCL);
1166 
1167 	d = __gfs2_lookup(dir, dentry, file, opened);
1168 	if (IS_ERR(d))
1169 		return PTR_ERR(d);
1170 	if (d != NULL)
1171 		dentry = d;
1172 	if (dentry->d_inode) {
1173 		if (!(*opened & FILE_OPENED)) {
1174 			if (d == NULL)
1175 				dget(dentry);
1176 			return finish_no_open(file, dentry);
1177 		}
1178 		dput(d);
1179 		return 0;
1180 	}
1181 
1182 	BUG_ON(d != NULL);
1183 	if (!(flags & O_CREAT))
1184 		return -ENOENT;
1185 
1186 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1187 }
1188 
1189 /*
1190  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1191  * @this: move this
1192  * @to: to here
1193  *
1194  * Follow @to back to the root and make sure we don't encounter @this
1195  * Assumes we already hold the rename lock.
1196  *
1197  * Returns: errno
1198  */
1199 
1200 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1201 {
1202 	struct inode *dir = &to->i_inode;
1203 	struct super_block *sb = dir->i_sb;
1204 	struct inode *tmp;
1205 	int error = 0;
1206 
1207 	igrab(dir);
1208 
1209 	for (;;) {
1210 		if (dir == &this->i_inode) {
1211 			error = -EINVAL;
1212 			break;
1213 		}
1214 		if (dir == sb->s_root->d_inode) {
1215 			error = 0;
1216 			break;
1217 		}
1218 
1219 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1220 		if (IS_ERR(tmp)) {
1221 			error = PTR_ERR(tmp);
1222 			break;
1223 		}
1224 
1225 		iput(dir);
1226 		dir = tmp;
1227 	}
1228 
1229 	iput(dir);
1230 
1231 	return error;
1232 }
1233 
1234 /**
1235  * gfs2_rename - Rename a file
1236  * @odir: Parent directory of old file name
1237  * @odentry: The old dentry of the file
1238  * @ndir: Parent directory of new file name
1239  * @ndentry: The new dentry of the file
1240  *
1241  * Returns: errno
1242  */
1243 
1244 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1245 		       struct inode *ndir, struct dentry *ndentry)
1246 {
1247 	struct gfs2_inode *odip = GFS2_I(odir);
1248 	struct gfs2_inode *ndip = GFS2_I(ndir);
1249 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1250 	struct gfs2_inode *nip = NULL;
1251 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1252 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1253 	struct gfs2_rgrpd *nrgd;
1254 	unsigned int num_gh;
1255 	int dir_rename = 0;
1256 	struct gfs2_diradd da = { .nr_blocks = 0, };
1257 	unsigned int x;
1258 	int error;
1259 
1260 	if (ndentry->d_inode) {
1261 		nip = GFS2_I(ndentry->d_inode);
1262 		if (ip == nip)
1263 			return 0;
1264 	}
1265 
1266 	error = gfs2_rindex_update(sdp);
1267 	if (error)
1268 		return error;
1269 
1270 	error = gfs2_rs_alloc(ndip);
1271 	if (error)
1272 		return error;
1273 
1274 	if (odip != ndip) {
1275 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1276 					   0, &r_gh);
1277 		if (error)
1278 			goto out;
1279 
1280 		if (S_ISDIR(ip->i_inode.i_mode)) {
1281 			dir_rename = 1;
1282 			/* don't move a dirctory into it's subdir */
1283 			error = gfs2_ok_to_move(ip, ndip);
1284 			if (error)
1285 				goto out_gunlock_r;
1286 		}
1287 	}
1288 
1289 	num_gh = 1;
1290 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1291 	if (odip != ndip) {
1292 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1293 		num_gh++;
1294 	}
1295 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1296 	num_gh++;
1297 
1298 	if (nip) {
1299 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1300 		num_gh++;
1301 		/* grab the resource lock for unlink flag twiddling
1302 		 * this is the case of the target file already existing
1303 		 * so we unlink before doing the rename
1304 		 */
1305 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1306 		if (nrgd)
1307 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1308 	}
1309 
1310 	for (x = 0; x < num_gh; x++) {
1311 		error = gfs2_glock_nq(ghs + x);
1312 		if (error)
1313 			goto out_gunlock;
1314 	}
1315 
1316 	error = -ENOENT;
1317 	if (ip->i_inode.i_nlink == 0)
1318 		goto out_gunlock;
1319 
1320 	/* Check out the old directory */
1321 
1322 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1323 	if (error)
1324 		goto out_gunlock;
1325 
1326 	/* Check out the new directory */
1327 
1328 	if (nip) {
1329 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1330 		if (error)
1331 			goto out_gunlock;
1332 
1333 		if (nip->i_inode.i_nlink == 0) {
1334 			error = -EAGAIN;
1335 			goto out_gunlock;
1336 		}
1337 
1338 		if (S_ISDIR(nip->i_inode.i_mode)) {
1339 			if (nip->i_entries < 2) {
1340 				gfs2_consist_inode(nip);
1341 				error = -EIO;
1342 				goto out_gunlock;
1343 			}
1344 			if (nip->i_entries > 2) {
1345 				error = -ENOTEMPTY;
1346 				goto out_gunlock;
1347 			}
1348 		}
1349 	} else {
1350 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1351 		if (error)
1352 			goto out_gunlock;
1353 
1354 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1355 		switch (error) {
1356 		case -ENOENT:
1357 			error = 0;
1358 			break;
1359 		case 0:
1360 			error = -EEXIST;
1361 		default:
1362 			goto out_gunlock;
1363 		};
1364 
1365 		if (odip != ndip) {
1366 			if (!ndip->i_inode.i_nlink) {
1367 				error = -ENOENT;
1368 				goto out_gunlock;
1369 			}
1370 			if (ndip->i_entries == (u32)-1) {
1371 				error = -EFBIG;
1372 				goto out_gunlock;
1373 			}
1374 			if (S_ISDIR(ip->i_inode.i_mode) &&
1375 			    ndip->i_inode.i_nlink == (u32)-1) {
1376 				error = -EMLINK;
1377 				goto out_gunlock;
1378 			}
1379 		}
1380 	}
1381 
1382 	/* Check out the dir to be renamed */
1383 
1384 	if (dir_rename) {
1385 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1386 		if (error)
1387 			goto out_gunlock;
1388 	}
1389 
1390 	if (nip == NULL) {
1391 		error = gfs2_diradd_alloc_required(ndir, &ndentry->d_name, &da);
1392 		if (error)
1393 			goto out_gunlock;
1394 	}
1395 
1396 	if (da.nr_blocks) {
1397 		struct gfs2_alloc_parms ap = { .target = da.nr_blocks, };
1398 		error = gfs2_quota_lock_check(ndip);
1399 		if (error)
1400 			goto out_gunlock;
1401 
1402 		error = gfs2_inplace_reserve(ndip, &ap);
1403 		if (error)
1404 			goto out_gunlock_q;
1405 
1406 		error = gfs2_trans_begin(sdp, da.nr_blocks +
1407 					 gfs2_rg_blocks(ndip, da.nr_blocks) +
1408 					 4 * RES_DINODE + 4 * RES_LEAF +
1409 					 RES_STATFS + RES_QUOTA + 4, 0);
1410 		if (error)
1411 			goto out_ipreserv;
1412 	} else {
1413 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1414 					 5 * RES_LEAF + 4, 0);
1415 		if (error)
1416 			goto out_gunlock;
1417 	}
1418 
1419 	/* Remove the target file, if it exists */
1420 
1421 	if (nip)
1422 		error = gfs2_unlink_inode(ndip, ndentry);
1423 
1424 	if (dir_rename) {
1425 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1426 		if (error)
1427 			goto out_end_trans;
1428 	} else {
1429 		struct buffer_head *dibh;
1430 		error = gfs2_meta_inode_buffer(ip, &dibh);
1431 		if (error)
1432 			goto out_end_trans;
1433 		ip->i_inode.i_ctime = CURRENT_TIME;
1434 		gfs2_trans_add_meta(ip->i_gl, dibh);
1435 		gfs2_dinode_out(ip, dibh->b_data);
1436 		brelse(dibh);
1437 	}
1438 
1439 	error = gfs2_dir_del(odip, odentry);
1440 	if (error)
1441 		goto out_end_trans;
1442 
1443 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1444 	if (error)
1445 		goto out_end_trans;
1446 
1447 out_end_trans:
1448 	gfs2_trans_end(sdp);
1449 out_ipreserv:
1450 	if (da.nr_blocks)
1451 		gfs2_inplace_release(ndip);
1452 out_gunlock_q:
1453 	if (da.nr_blocks)
1454 		gfs2_quota_unlock(ndip);
1455 out_gunlock:
1456 	while (x--) {
1457 		gfs2_glock_dq(ghs + x);
1458 		gfs2_holder_uninit(ghs + x);
1459 	}
1460 out_gunlock_r:
1461 	if (r_gh.gh_gl)
1462 		gfs2_glock_dq_uninit(&r_gh);
1463 out:
1464 	return error;
1465 }
1466 
1467 /**
1468  * gfs2_follow_link - Follow a symbolic link
1469  * @dentry: The dentry of the link
1470  * @nd: Data that we pass to vfs_follow_link()
1471  *
1472  * This can handle symlinks of any size.
1473  *
1474  * Returns: 0 on success or error code
1475  */
1476 
1477 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1478 {
1479 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1480 	struct gfs2_holder i_gh;
1481 	struct buffer_head *dibh;
1482 	unsigned int size;
1483 	char *buf;
1484 	int error;
1485 
1486 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1487 	error = gfs2_glock_nq(&i_gh);
1488 	if (error) {
1489 		gfs2_holder_uninit(&i_gh);
1490 		nd_set_link(nd, ERR_PTR(error));
1491 		return NULL;
1492 	}
1493 
1494 	size = (unsigned int)i_size_read(&ip->i_inode);
1495 	if (size == 0) {
1496 		gfs2_consist_inode(ip);
1497 		buf = ERR_PTR(-EIO);
1498 		goto out;
1499 	}
1500 
1501 	error = gfs2_meta_inode_buffer(ip, &dibh);
1502 	if (error) {
1503 		buf = ERR_PTR(error);
1504 		goto out;
1505 	}
1506 
1507 	buf = kzalloc(size + 1, GFP_NOFS);
1508 	if (!buf)
1509 		buf = ERR_PTR(-ENOMEM);
1510 	else
1511 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1512 	brelse(dibh);
1513 out:
1514 	gfs2_glock_dq_uninit(&i_gh);
1515 	nd_set_link(nd, buf);
1516 	return NULL;
1517 }
1518 
1519 /**
1520  * gfs2_permission -
1521  * @inode: The inode
1522  * @mask: The mask to be tested
1523  * @flags: Indicates whether this is an RCU path walk or not
1524  *
1525  * This may be called from the VFS directly, or from within GFS2 with the
1526  * inode locked, so we look to see if the glock is already locked and only
1527  * lock the glock if its not already been done.
1528  *
1529  * Returns: errno
1530  */
1531 
1532 int gfs2_permission(struct inode *inode, int mask)
1533 {
1534 	struct gfs2_inode *ip;
1535 	struct gfs2_holder i_gh;
1536 	int error;
1537 	int unlock = 0;
1538 
1539 
1540 	ip = GFS2_I(inode);
1541 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1542 		if (mask & MAY_NOT_BLOCK)
1543 			return -ECHILD;
1544 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1545 		if (error)
1546 			return error;
1547 		unlock = 1;
1548 	}
1549 
1550 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1551 		error = -EACCES;
1552 	else
1553 		error = generic_permission(inode, mask);
1554 	if (unlock)
1555 		gfs2_glock_dq_uninit(&i_gh);
1556 
1557 	return error;
1558 }
1559 
1560 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1561 {
1562 	setattr_copy(inode, attr);
1563 	mark_inode_dirty(inode);
1564 	return 0;
1565 }
1566 
1567 /**
1568  * gfs2_setattr_simple -
1569  * @ip:
1570  * @attr:
1571  *
1572  * Returns: errno
1573  */
1574 
1575 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1576 {
1577 	int error;
1578 
1579 	if (current->journal_info)
1580 		return __gfs2_setattr_simple(inode, attr);
1581 
1582 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1583 	if (error)
1584 		return error;
1585 
1586 	error = __gfs2_setattr_simple(inode, attr);
1587 	gfs2_trans_end(GFS2_SB(inode));
1588 	return error;
1589 }
1590 
1591 static int setattr_chown(struct inode *inode, struct iattr *attr)
1592 {
1593 	struct gfs2_inode *ip = GFS2_I(inode);
1594 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1595 	kuid_t ouid, nuid;
1596 	kgid_t ogid, ngid;
1597 	int error;
1598 
1599 	ouid = inode->i_uid;
1600 	ogid = inode->i_gid;
1601 	nuid = attr->ia_uid;
1602 	ngid = attr->ia_gid;
1603 
1604 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1605 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1606 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1607 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1608 
1609 	error = gfs2_quota_lock(ip, nuid, ngid);
1610 	if (error)
1611 		return error;
1612 
1613 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1614 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1615 		error = gfs2_quota_check(ip, nuid, ngid);
1616 		if (error)
1617 			goto out_gunlock_q;
1618 	}
1619 
1620 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1621 	if (error)
1622 		goto out_gunlock_q;
1623 
1624 	error = gfs2_setattr_simple(inode, attr);
1625 	if (error)
1626 		goto out_end_trans;
1627 
1628 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1629 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1630 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1631 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1632 		gfs2_quota_change(ip, blocks, nuid, ngid);
1633 	}
1634 
1635 out_end_trans:
1636 	gfs2_trans_end(sdp);
1637 out_gunlock_q:
1638 	gfs2_quota_unlock(ip);
1639 	return error;
1640 }
1641 
1642 /**
1643  * gfs2_setattr - Change attributes on an inode
1644  * @dentry: The dentry which is changing
1645  * @attr: The structure describing the change
1646  *
1647  * The VFS layer wants to change one or more of an inodes attributes.  Write
1648  * that change out to disk.
1649  *
1650  * Returns: errno
1651  */
1652 
1653 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1654 {
1655 	struct inode *inode = dentry->d_inode;
1656 	struct gfs2_inode *ip = GFS2_I(inode);
1657 	struct gfs2_holder i_gh;
1658 	int error;
1659 
1660 	error = gfs2_rs_alloc(ip);
1661 	if (error)
1662 		return error;
1663 
1664 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1665 	if (error)
1666 		return error;
1667 
1668 	error = -EPERM;
1669 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1670 		goto out;
1671 
1672 	error = inode_change_ok(inode, attr);
1673 	if (error)
1674 		goto out;
1675 
1676 	if (attr->ia_valid & ATTR_SIZE)
1677 		error = gfs2_setattr_size(inode, attr->ia_size);
1678 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1679 		error = setattr_chown(inode, attr);
1680 	else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1681 		error = gfs2_acl_chmod(ip, attr);
1682 	else
1683 		error = gfs2_setattr_simple(inode, attr);
1684 
1685 out:
1686 	if (!error)
1687 		mark_inode_dirty(inode);
1688 	gfs2_glock_dq_uninit(&i_gh);
1689 	return error;
1690 }
1691 
1692 /**
1693  * gfs2_getattr - Read out an inode's attributes
1694  * @mnt: The vfsmount the inode is being accessed from
1695  * @dentry: The dentry to stat
1696  * @stat: The inode's stats
1697  *
1698  * This may be called from the VFS directly, or from within GFS2 with the
1699  * inode locked, so we look to see if the glock is already locked and only
1700  * lock the glock if its not already been done. Note that its the NFS
1701  * readdirplus operation which causes this to be called (from filldir)
1702  * with the glock already held.
1703  *
1704  * Returns: errno
1705  */
1706 
1707 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1708 			struct kstat *stat)
1709 {
1710 	struct inode *inode = dentry->d_inode;
1711 	struct gfs2_inode *ip = GFS2_I(inode);
1712 	struct gfs2_holder gh;
1713 	int error;
1714 	int unlock = 0;
1715 
1716 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1717 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1718 		if (error)
1719 			return error;
1720 		unlock = 1;
1721 	}
1722 
1723 	generic_fillattr(inode, stat);
1724 	if (unlock)
1725 		gfs2_glock_dq_uninit(&gh);
1726 
1727 	return 0;
1728 }
1729 
1730 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1731 			 const void *data, size_t size, int flags)
1732 {
1733 	struct inode *inode = dentry->d_inode;
1734 	struct gfs2_inode *ip = GFS2_I(inode);
1735 	struct gfs2_holder gh;
1736 	int ret;
1737 
1738 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1739 	ret = gfs2_glock_nq(&gh);
1740 	if (ret == 0) {
1741 		ret = gfs2_rs_alloc(ip);
1742 		if (ret == 0)
1743 			ret = generic_setxattr(dentry, name, data, size, flags);
1744 		gfs2_glock_dq(&gh);
1745 	}
1746 	gfs2_holder_uninit(&gh);
1747 	return ret;
1748 }
1749 
1750 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1751 			     void *data, size_t size)
1752 {
1753 	struct inode *inode = dentry->d_inode;
1754 	struct gfs2_inode *ip = GFS2_I(inode);
1755 	struct gfs2_holder gh;
1756 	int ret;
1757 
1758 	/* For selinux during lookup */
1759 	if (gfs2_glock_is_locked_by_me(ip->i_gl))
1760 		return generic_getxattr(dentry, name, data, size);
1761 
1762 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1763 	ret = gfs2_glock_nq(&gh);
1764 	if (ret == 0) {
1765 		ret = generic_getxattr(dentry, name, data, size);
1766 		gfs2_glock_dq(&gh);
1767 	}
1768 	gfs2_holder_uninit(&gh);
1769 	return ret;
1770 }
1771 
1772 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1773 {
1774 	struct inode *inode = dentry->d_inode;
1775 	struct gfs2_inode *ip = GFS2_I(inode);
1776 	struct gfs2_holder gh;
1777 	int ret;
1778 
1779 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1780 	ret = gfs2_glock_nq(&gh);
1781 	if (ret == 0) {
1782 		ret = gfs2_rs_alloc(ip);
1783 		if (ret == 0)
1784 			ret = generic_removexattr(dentry, name);
1785 		gfs2_glock_dq(&gh);
1786 	}
1787 	gfs2_holder_uninit(&gh);
1788 	return ret;
1789 }
1790 
1791 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1792 		       u64 start, u64 len)
1793 {
1794 	struct gfs2_inode *ip = GFS2_I(inode);
1795 	struct gfs2_holder gh;
1796 	int ret;
1797 
1798 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1799 	if (ret)
1800 		return ret;
1801 
1802 	mutex_lock(&inode->i_mutex);
1803 
1804 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1805 	if (ret)
1806 		goto out;
1807 
1808 	if (gfs2_is_stuffed(ip)) {
1809 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1810 		u64 size = i_size_read(inode);
1811 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1812 			    FIEMAP_EXTENT_DATA_INLINE;
1813 		phys += sizeof(struct gfs2_dinode);
1814 		phys += start;
1815 		if (start + len > size)
1816 			len = size - start;
1817 		if (start < size)
1818 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1819 						      len, flags);
1820 		if (ret == 1)
1821 			ret = 0;
1822 	} else {
1823 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1824 					     gfs2_block_map);
1825 	}
1826 
1827 	gfs2_glock_dq_uninit(&gh);
1828 out:
1829 	mutex_unlock(&inode->i_mutex);
1830 	return ret;
1831 }
1832 
1833 const struct inode_operations gfs2_file_iops = {
1834 	.permission = gfs2_permission,
1835 	.setattr = gfs2_setattr,
1836 	.getattr = gfs2_getattr,
1837 	.setxattr = gfs2_setxattr,
1838 	.getxattr = gfs2_getxattr,
1839 	.listxattr = gfs2_listxattr,
1840 	.removexattr = gfs2_removexattr,
1841 	.fiemap = gfs2_fiemap,
1842 	.get_acl = gfs2_get_acl,
1843 };
1844 
1845 const struct inode_operations gfs2_dir_iops = {
1846 	.create = gfs2_create,
1847 	.lookup = gfs2_lookup,
1848 	.link = gfs2_link,
1849 	.unlink = gfs2_unlink,
1850 	.symlink = gfs2_symlink,
1851 	.mkdir = gfs2_mkdir,
1852 	.rmdir = gfs2_unlink,
1853 	.mknod = gfs2_mknod,
1854 	.rename = gfs2_rename,
1855 	.permission = gfs2_permission,
1856 	.setattr = gfs2_setattr,
1857 	.getattr = gfs2_getattr,
1858 	.setxattr = gfs2_setxattr,
1859 	.getxattr = gfs2_getxattr,
1860 	.listxattr = gfs2_listxattr,
1861 	.removexattr = gfs2_removexattr,
1862 	.fiemap = gfs2_fiemap,
1863 	.get_acl = gfs2_get_acl,
1864 	.atomic_open = gfs2_atomic_open,
1865 };
1866 
1867 const struct inode_operations gfs2_symlink_iops = {
1868 	.readlink = generic_readlink,
1869 	.follow_link = gfs2_follow_link,
1870 	.put_link = kfree_put_link,
1871 	.permission = gfs2_permission,
1872 	.setattr = gfs2_setattr,
1873 	.getattr = gfs2_getattr,
1874 	.setxattr = gfs2_setxattr,
1875 	.getxattr = gfs2_getxattr,
1876 	.listxattr = gfs2_listxattr,
1877 	.removexattr = gfs2_removexattr,
1878 	.fiemap = gfs2_fiemap,
1879 	.get_acl = gfs2_get_acl,
1880 };
1881 
1882