xref: /openbmc/linux/fs/gfs2/inode.c (revision e01580bf)
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, int arq)
474 {
475 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
476 	struct gfs2_alloc_parms ap = { .target = sdp->sd_max_dirres, };
477 	int error;
478 
479 	if (arq) {
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, sdp->sd_max_dirres +
489 					 dip->i_rgd->rd_length +
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 posix_acl *default_acl, *acl;
556 	struct gfs2_holder ghs[2];
557 	struct inode *inode = NULL;
558 	struct gfs2_inode *dip = GFS2_I(dir), *ip;
559 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
560 	struct gfs2_glock *io_gl;
561 	struct dentry *d;
562 	int error;
563 	u32 aflags = 0;
564 	int arq;
565 
566 	if (!name->len || name->len > GFS2_FNAMESIZE)
567 		return -ENAMETOOLONG;
568 
569 	error = gfs2_rs_alloc(dip);
570 	if (error)
571 		return error;
572 
573 	error = gfs2_rindex_update(sdp);
574 	if (error)
575 		return error;
576 
577 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
578 	if (error)
579 		goto fail;
580 
581 	error = create_ok(dip, name, mode);
582 	if (error)
583 		goto fail_gunlock;
584 
585 	inode = gfs2_dir_search(dir, &dentry->d_name, !S_ISREG(mode) || excl);
586 	error = PTR_ERR(inode);
587 	if (!IS_ERR(inode)) {
588 		d = d_splice_alias(inode, dentry);
589 		error = 0;
590 		if (file) {
591 			if (S_ISREG(inode->i_mode)) {
592 				WARN_ON(d != NULL);
593 				error = finish_open(file, dentry, gfs2_open_common, opened);
594 			} else {
595 				error = finish_no_open(file, d);
596 			}
597 		} else {
598 			dput(d);
599 		}
600 		gfs2_glock_dq_uninit(ghs);
601 		return error;
602 	} else if (error != -ENOENT) {
603 		goto fail_gunlock;
604 	}
605 
606 	arq = error = gfs2_diradd_alloc_required(dir, name);
607 	if (error < 0)
608 		goto fail_gunlock;
609 
610 	inode = new_inode(sdp->sd_vfs);
611 	error = -ENOMEM;
612 	if (!inode)
613 		goto fail_gunlock;
614 
615 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
616 	if (error)
617 		goto fail_free_vfs_inode;
618 
619 	ip = GFS2_I(inode);
620 	error = gfs2_rs_alloc(ip);
621 	if (error)
622 		goto fail_free_acls;
623 
624 	inode->i_mode = mode;
625 	set_nlink(inode, S_ISDIR(mode) ? 2 : 1);
626 	inode->i_rdev = dev;
627 	inode->i_size = size;
628 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
629 	gfs2_set_inode_blocks(inode, 1);
630 	munge_mode_uid_gid(dip, inode);
631 	ip->i_goal = dip->i_goal;
632 	ip->i_diskflags = 0;
633 	ip->i_eattr = 0;
634 	ip->i_height = 0;
635 	ip->i_depth = 0;
636 	ip->i_entries = 0;
637 
638 	switch(mode & S_IFMT) {
639 	case S_IFREG:
640 		if ((dip->i_diskflags & GFS2_DIF_INHERIT_JDATA) ||
641 		    gfs2_tune_get(sdp, gt_new_files_jdata))
642 			ip->i_diskflags |= GFS2_DIF_JDATA;
643 		gfs2_set_aops(inode);
644 		break;
645 	case S_IFDIR:
646 		ip->i_diskflags |= (dip->i_diskflags & GFS2_DIF_INHERIT_JDATA);
647 		ip->i_diskflags |= GFS2_DIF_JDATA;
648 		ip->i_entries = 2;
649 		break;
650 	}
651 	gfs2_set_inode_flags(inode);
652 
653 	if ((GFS2_I(sdp->sd_root_dir->d_inode) == dip) ||
654 	    (dip->i_diskflags & GFS2_DIF_TOPDIR))
655 		aflags |= GFS2_AF_ORLOV;
656 
657 	error = alloc_dinode(ip, aflags);
658 	if (error)
659 		goto fail_free_inode;
660 
661 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
662 	if (error)
663 		goto fail_free_inode;
664 
665 	ip->i_gl->gl_object = ip;
666 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
667 	if (error)
668 		goto fail_free_inode;
669 
670 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
671 	if (error)
672 		goto fail_gunlock2;
673 
674 	init_dinode(dip, ip, symname);
675 	gfs2_trans_end(sdp);
676 
677 	error = gfs2_glock_get(sdp, ip->i_no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
678 	if (error)
679 		goto fail_gunlock2;
680 
681 	error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
682 	if (error)
683 		goto fail_gunlock2;
684 
685 	ip->i_iopen_gh.gh_gl->gl_object = ip;
686 	gfs2_glock_put(io_gl);
687 	gfs2_set_iop(inode);
688 	insert_inode_hash(inode);
689 
690 	if (default_acl) {
691 		error = gfs2_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
692 		posix_acl_release(default_acl);
693 	}
694 	if (acl) {
695 		if (!error)
696 			error = gfs2_set_acl(inode, acl, ACL_TYPE_ACCESS);
697 		posix_acl_release(acl);
698 	}
699 
700 	if (error)
701 		goto fail_gunlock3;
702 
703 	error = gfs2_security_init(dip, ip, name);
704 	if (error)
705 		goto fail_gunlock3;
706 
707 	error = link_dinode(dip, name, ip, arq);
708 	if (error)
709 		goto fail_gunlock3;
710 
711 	mark_inode_dirty(inode);
712 	d_instantiate(dentry, inode);
713 	if (file) {
714 		*opened |= FILE_CREATED;
715 		error = finish_open(file, dentry, gfs2_open_common, opened);
716 	}
717 	gfs2_glock_dq_uninit(ghs);
718 	gfs2_glock_dq_uninit(ghs + 1);
719 	return error;
720 
721 fail_gunlock3:
722 	gfs2_glock_dq_uninit(ghs + 1);
723 	if (ip->i_gl)
724 		gfs2_glock_put(ip->i_gl);
725 	goto fail_gunlock;
726 
727 fail_gunlock2:
728 	gfs2_glock_dq_uninit(ghs + 1);
729 fail_free_inode:
730 	if (ip->i_gl)
731 		gfs2_glock_put(ip->i_gl);
732 	gfs2_rs_delete(ip, NULL);
733 fail_free_acls:
734 	if (default_acl)
735 		posix_acl_release(default_acl);
736 	if (acl)
737 		posix_acl_release(acl);
738 fail_free_vfs_inode:
739 	free_inode_nonrcu(inode);
740 	inode = NULL;
741 fail_gunlock:
742 	gfs2_glock_dq_uninit(ghs);
743 	if (inode && !IS_ERR(inode)) {
744 		clear_nlink(inode);
745 		mark_inode_dirty(inode);
746 		set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
747 		iput(inode);
748 	}
749 fail:
750 	return error;
751 }
752 
753 /**
754  * gfs2_create - Create a file
755  * @dir: The directory in which to create the file
756  * @dentry: The dentry of the new file
757  * @mode: The mode of the new file
758  *
759  * Returns: errno
760  */
761 
762 static int gfs2_create(struct inode *dir, struct dentry *dentry,
763 		       umode_t mode, bool excl)
764 {
765 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
766 }
767 
768 /**
769  * __gfs2_lookup - Look up a filename in a directory and return its inode
770  * @dir: The directory inode
771  * @dentry: The dentry of the new inode
772  * @file: File to be opened
773  * @opened: atomic_open flags
774  *
775  *
776  * Returns: errno
777  */
778 
779 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
780 				    struct file *file, int *opened)
781 {
782 	struct inode *inode;
783 	struct dentry *d;
784 	struct gfs2_holder gh;
785 	struct gfs2_glock *gl;
786 	int error;
787 
788 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
789 	if (!inode)
790 		return NULL;
791 	if (IS_ERR(inode))
792 		return ERR_CAST(inode);
793 
794 	gl = GFS2_I(inode)->i_gl;
795 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
796 	if (error) {
797 		iput(inode);
798 		return ERR_PTR(error);
799 	}
800 
801 	d = d_splice_alias(inode, dentry);
802 	if (file && S_ISREG(inode->i_mode))
803 		error = finish_open(file, dentry, gfs2_open_common, opened);
804 
805 	gfs2_glock_dq_uninit(&gh);
806 	if (error) {
807 		dput(d);
808 		return ERR_PTR(error);
809 	}
810 	return d;
811 }
812 
813 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
814 				  unsigned flags)
815 {
816 	return __gfs2_lookup(dir, dentry, NULL, NULL);
817 }
818 
819 /**
820  * gfs2_link - Link to a file
821  * @old_dentry: The inode to link
822  * @dir: Add link to this directory
823  * @dentry: The name of the link
824  *
825  * Link the inode in "old_dentry" into the directory "dir" with the
826  * name in "dentry".
827  *
828  * Returns: errno
829  */
830 
831 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
832 		     struct dentry *dentry)
833 {
834 	struct gfs2_inode *dip = GFS2_I(dir);
835 	struct gfs2_sbd *sdp = GFS2_SB(dir);
836 	struct inode *inode = old_dentry->d_inode;
837 	struct gfs2_inode *ip = GFS2_I(inode);
838 	struct gfs2_holder ghs[2];
839 	struct buffer_head *dibh;
840 	int alloc_required;
841 	int error;
842 
843 	if (S_ISDIR(inode->i_mode))
844 		return -EPERM;
845 
846 	error = gfs2_rs_alloc(dip);
847 	if (error)
848 		return error;
849 
850 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
851 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
852 
853 	error = gfs2_glock_nq(ghs); /* parent */
854 	if (error)
855 		goto out_parent;
856 
857 	error = gfs2_glock_nq(ghs + 1); /* child */
858 	if (error)
859 		goto out_child;
860 
861 	error = -ENOENT;
862 	if (inode->i_nlink == 0)
863 		goto out_gunlock;
864 
865 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
866 	if (error)
867 		goto out_gunlock;
868 
869 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
870 	switch (error) {
871 	case -ENOENT:
872 		break;
873 	case 0:
874 		error = -EEXIST;
875 	default:
876 		goto out_gunlock;
877 	}
878 
879 	error = -EINVAL;
880 	if (!dip->i_inode.i_nlink)
881 		goto out_gunlock;
882 	error = -EFBIG;
883 	if (dip->i_entries == (u32)-1)
884 		goto out_gunlock;
885 	error = -EPERM;
886 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
887 		goto out_gunlock;
888 	error = -EINVAL;
889 	if (!ip->i_inode.i_nlink)
890 		goto out_gunlock;
891 	error = -EMLINK;
892 	if (ip->i_inode.i_nlink == (u32)-1)
893 		goto out_gunlock;
894 
895 	alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
896 	if (error < 0)
897 		goto out_gunlock;
898 	error = 0;
899 
900 	if (alloc_required) {
901 		struct gfs2_alloc_parms ap = { .target = sdp->sd_max_dirres, };
902 		error = gfs2_quota_lock_check(dip);
903 		if (error)
904 			goto out_gunlock;
905 
906 		error = gfs2_inplace_reserve(dip, &ap);
907 		if (error)
908 			goto out_gunlock_q;
909 
910 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
911 					 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
912 					 2 * RES_DINODE + RES_STATFS +
913 					 RES_QUOTA, 0);
914 		if (error)
915 			goto out_ipres;
916 	} else {
917 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
918 		if (error)
919 			goto out_ipres;
920 	}
921 
922 	error = gfs2_meta_inode_buffer(ip, &dibh);
923 	if (error)
924 		goto out_end_trans;
925 
926 	error = gfs2_dir_add(dir, &dentry->d_name, ip);
927 	if (error)
928 		goto out_brelse;
929 
930 	gfs2_trans_add_meta(ip->i_gl, dibh);
931 	inc_nlink(&ip->i_inode);
932 	ip->i_inode.i_ctime = CURRENT_TIME;
933 	ihold(inode);
934 	d_instantiate(dentry, inode);
935 	mark_inode_dirty(inode);
936 
937 out_brelse:
938 	brelse(dibh);
939 out_end_trans:
940 	gfs2_trans_end(sdp);
941 out_ipres:
942 	if (alloc_required)
943 		gfs2_inplace_release(dip);
944 out_gunlock_q:
945 	if (alloc_required)
946 		gfs2_quota_unlock(dip);
947 out_gunlock:
948 	gfs2_glock_dq(ghs + 1);
949 out_child:
950 	gfs2_glock_dq(ghs);
951 out_parent:
952 	gfs2_holder_uninit(ghs);
953 	gfs2_holder_uninit(ghs + 1);
954 	return error;
955 }
956 
957 /*
958  * gfs2_unlink_ok - check to see that a inode is still in a directory
959  * @dip: the directory
960  * @name: the name of the file
961  * @ip: the inode
962  *
963  * Assumes that the lock on (at least) @dip is held.
964  *
965  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
966  */
967 
968 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
969 			  const struct gfs2_inode *ip)
970 {
971 	int error;
972 
973 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
974 		return -EPERM;
975 
976 	if ((dip->i_inode.i_mode & S_ISVTX) &&
977 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
978 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
979 		return -EPERM;
980 
981 	if (IS_APPEND(&dip->i_inode))
982 		return -EPERM;
983 
984 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
985 	if (error)
986 		return error;
987 
988 	error = gfs2_dir_check(&dip->i_inode, name, ip);
989 	if (error)
990 		return error;
991 
992 	return 0;
993 }
994 
995 /**
996  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
997  * @dip: The parent directory
998  * @name: The name of the entry in the parent directory
999  * @inode: The inode to be removed
1000  *
1001  * Called with all the locks and in a transaction. This will only be
1002  * called for a directory after it has been checked to ensure it is empty.
1003  *
1004  * Returns: 0 on success, or an error
1005  */
1006 
1007 static int gfs2_unlink_inode(struct gfs2_inode *dip,
1008 			     const struct dentry *dentry)
1009 {
1010 	struct inode *inode = dentry->d_inode;
1011 	struct gfs2_inode *ip = GFS2_I(inode);
1012 	int error;
1013 
1014 	error = gfs2_dir_del(dip, dentry);
1015 	if (error)
1016 		return error;
1017 
1018 	ip->i_entries = 0;
1019 	inode->i_ctime = CURRENT_TIME;
1020 	if (S_ISDIR(inode->i_mode))
1021 		clear_nlink(inode);
1022 	else
1023 		drop_nlink(inode);
1024 	mark_inode_dirty(inode);
1025 	if (inode->i_nlink == 0)
1026 		gfs2_unlink_di(inode);
1027 	return 0;
1028 }
1029 
1030 
1031 /**
1032  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1033  * @dir: The inode of the directory containing the inode to unlink
1034  * @dentry: The file itself
1035  *
1036  * This routine uses the type of the inode as a flag to figure out
1037  * whether this is an unlink or an rmdir.
1038  *
1039  * Returns: errno
1040  */
1041 
1042 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1043 {
1044 	struct gfs2_inode *dip = GFS2_I(dir);
1045 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1046 	struct inode *inode = dentry->d_inode;
1047 	struct gfs2_inode *ip = GFS2_I(inode);
1048 	struct gfs2_holder ghs[3];
1049 	struct gfs2_rgrpd *rgd;
1050 	int error;
1051 
1052 	error = gfs2_rindex_update(sdp);
1053 	if (error)
1054 		return error;
1055 
1056 	error = -EROFS;
1057 
1058 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1059 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1060 
1061 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1062 	if (!rgd)
1063 		goto out_inodes;
1064 
1065 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1066 
1067 
1068 	error = gfs2_glock_nq(ghs); /* parent */
1069 	if (error)
1070 		goto out_parent;
1071 
1072 	error = gfs2_glock_nq(ghs + 1); /* child */
1073 	if (error)
1074 		goto out_child;
1075 
1076 	error = -ENOENT;
1077 	if (inode->i_nlink == 0)
1078 		goto out_rgrp;
1079 
1080 	if (S_ISDIR(inode->i_mode)) {
1081 		error = -ENOTEMPTY;
1082 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1083 			goto out_rgrp;
1084 	}
1085 
1086 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1087 	if (error)
1088 		goto out_rgrp;
1089 
1090 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1091 	if (error)
1092 		goto out_gunlock;
1093 
1094 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1095 	if (error)
1096 		goto out_end_trans;
1097 
1098 	error = gfs2_unlink_inode(dip, dentry);
1099 
1100 out_end_trans:
1101 	gfs2_trans_end(sdp);
1102 out_gunlock:
1103 	gfs2_glock_dq(ghs + 2);
1104 out_rgrp:
1105 	gfs2_glock_dq(ghs + 1);
1106 out_child:
1107 	gfs2_glock_dq(ghs);
1108 out_parent:
1109 	gfs2_holder_uninit(ghs + 2);
1110 out_inodes:
1111 	gfs2_holder_uninit(ghs + 1);
1112 	gfs2_holder_uninit(ghs);
1113 	return error;
1114 }
1115 
1116 /**
1117  * gfs2_symlink - Create a symlink
1118  * @dir: The directory to create the symlink in
1119  * @dentry: The dentry to put the symlink in
1120  * @symname: The thing which the link points to
1121  *
1122  * Returns: errno
1123  */
1124 
1125 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1126 			const char *symname)
1127 {
1128 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1129 	unsigned int size;
1130 
1131 	size = strlen(symname);
1132 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1133 		return -ENAMETOOLONG;
1134 
1135 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1136 }
1137 
1138 /**
1139  * gfs2_mkdir - Make a directory
1140  * @dir: The parent directory of the new one
1141  * @dentry: The dentry of the new directory
1142  * @mode: The mode of the new directory
1143  *
1144  * Returns: errno
1145  */
1146 
1147 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1148 {
1149 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1150 	unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1151 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1152 }
1153 
1154 /**
1155  * gfs2_mknod - Make a special file
1156  * @dir: The directory in which the special file will reside
1157  * @dentry: The dentry of the special file
1158  * @mode: The mode of the special file
1159  * @dev: The device specification of the special file
1160  *
1161  */
1162 
1163 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1164 		      dev_t dev)
1165 {
1166 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1167 }
1168 
1169 /**
1170  * gfs2_atomic_open - Atomically open a file
1171  * @dir: The directory
1172  * @dentry: The proposed new entry
1173  * @file: The proposed new struct file
1174  * @flags: open flags
1175  * @mode: File mode
1176  * @opened: Flag to say whether the file has been opened or not
1177  *
1178  * Returns: error code or 0 for success
1179  */
1180 
1181 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1182                             struct file *file, unsigned flags,
1183                             umode_t mode, int *opened)
1184 {
1185 	struct dentry *d;
1186 	bool excl = !!(flags & O_EXCL);
1187 
1188 	d = __gfs2_lookup(dir, dentry, file, opened);
1189 	if (IS_ERR(d))
1190 		return PTR_ERR(d);
1191 	if (d != NULL)
1192 		dentry = d;
1193 	if (dentry->d_inode) {
1194 		if (!(*opened & FILE_OPENED)) {
1195 			if (d == NULL)
1196 				dget(dentry);
1197 			return finish_no_open(file, dentry);
1198 		}
1199 		dput(d);
1200 		return 0;
1201 	}
1202 
1203 	BUG_ON(d != NULL);
1204 	if (!(flags & O_CREAT))
1205 		return -ENOENT;
1206 
1207 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1208 }
1209 
1210 /*
1211  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1212  * @this: move this
1213  * @to: to here
1214  *
1215  * Follow @to back to the root and make sure we don't encounter @this
1216  * Assumes we already hold the rename lock.
1217  *
1218  * Returns: errno
1219  */
1220 
1221 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1222 {
1223 	struct inode *dir = &to->i_inode;
1224 	struct super_block *sb = dir->i_sb;
1225 	struct inode *tmp;
1226 	int error = 0;
1227 
1228 	igrab(dir);
1229 
1230 	for (;;) {
1231 		if (dir == &this->i_inode) {
1232 			error = -EINVAL;
1233 			break;
1234 		}
1235 		if (dir == sb->s_root->d_inode) {
1236 			error = 0;
1237 			break;
1238 		}
1239 
1240 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1241 		if (IS_ERR(tmp)) {
1242 			error = PTR_ERR(tmp);
1243 			break;
1244 		}
1245 
1246 		iput(dir);
1247 		dir = tmp;
1248 	}
1249 
1250 	iput(dir);
1251 
1252 	return error;
1253 }
1254 
1255 /**
1256  * gfs2_rename - Rename a file
1257  * @odir: Parent directory of old file name
1258  * @odentry: The old dentry of the file
1259  * @ndir: Parent directory of new file name
1260  * @ndentry: The new dentry of the file
1261  *
1262  * Returns: errno
1263  */
1264 
1265 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1266 		       struct inode *ndir, struct dentry *ndentry)
1267 {
1268 	struct gfs2_inode *odip = GFS2_I(odir);
1269 	struct gfs2_inode *ndip = GFS2_I(ndir);
1270 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1271 	struct gfs2_inode *nip = NULL;
1272 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1273 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1274 	struct gfs2_rgrpd *nrgd;
1275 	unsigned int num_gh;
1276 	int dir_rename = 0;
1277 	int alloc_required = 0;
1278 	unsigned int x;
1279 	int error;
1280 
1281 	if (ndentry->d_inode) {
1282 		nip = GFS2_I(ndentry->d_inode);
1283 		if (ip == nip)
1284 			return 0;
1285 	}
1286 
1287 	error = gfs2_rindex_update(sdp);
1288 	if (error)
1289 		return error;
1290 
1291 	error = gfs2_rs_alloc(ndip);
1292 	if (error)
1293 		return error;
1294 
1295 	if (odip != ndip) {
1296 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1297 					   0, &r_gh);
1298 		if (error)
1299 			goto out;
1300 
1301 		if (S_ISDIR(ip->i_inode.i_mode)) {
1302 			dir_rename = 1;
1303 			/* don't move a dirctory into it's subdir */
1304 			error = gfs2_ok_to_move(ip, ndip);
1305 			if (error)
1306 				goto out_gunlock_r;
1307 		}
1308 	}
1309 
1310 	num_gh = 1;
1311 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1312 	if (odip != ndip) {
1313 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1314 		num_gh++;
1315 	}
1316 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1317 	num_gh++;
1318 
1319 	if (nip) {
1320 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1321 		num_gh++;
1322 		/* grab the resource lock for unlink flag twiddling
1323 		 * this is the case of the target file already existing
1324 		 * so we unlink before doing the rename
1325 		 */
1326 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1327 		if (nrgd)
1328 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1329 	}
1330 
1331 	for (x = 0; x < num_gh; x++) {
1332 		error = gfs2_glock_nq(ghs + x);
1333 		if (error)
1334 			goto out_gunlock;
1335 	}
1336 
1337 	error = -ENOENT;
1338 	if (ip->i_inode.i_nlink == 0)
1339 		goto out_gunlock;
1340 
1341 	/* Check out the old directory */
1342 
1343 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1344 	if (error)
1345 		goto out_gunlock;
1346 
1347 	/* Check out the new directory */
1348 
1349 	if (nip) {
1350 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1351 		if (error)
1352 			goto out_gunlock;
1353 
1354 		if (nip->i_inode.i_nlink == 0) {
1355 			error = -EAGAIN;
1356 			goto out_gunlock;
1357 		}
1358 
1359 		if (S_ISDIR(nip->i_inode.i_mode)) {
1360 			if (nip->i_entries < 2) {
1361 				gfs2_consist_inode(nip);
1362 				error = -EIO;
1363 				goto out_gunlock;
1364 			}
1365 			if (nip->i_entries > 2) {
1366 				error = -ENOTEMPTY;
1367 				goto out_gunlock;
1368 			}
1369 		}
1370 	} else {
1371 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1372 		if (error)
1373 			goto out_gunlock;
1374 
1375 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1376 		switch (error) {
1377 		case -ENOENT:
1378 			error = 0;
1379 			break;
1380 		case 0:
1381 			error = -EEXIST;
1382 		default:
1383 			goto out_gunlock;
1384 		};
1385 
1386 		if (odip != ndip) {
1387 			if (!ndip->i_inode.i_nlink) {
1388 				error = -ENOENT;
1389 				goto out_gunlock;
1390 			}
1391 			if (ndip->i_entries == (u32)-1) {
1392 				error = -EFBIG;
1393 				goto out_gunlock;
1394 			}
1395 			if (S_ISDIR(ip->i_inode.i_mode) &&
1396 			    ndip->i_inode.i_nlink == (u32)-1) {
1397 				error = -EMLINK;
1398 				goto out_gunlock;
1399 			}
1400 		}
1401 	}
1402 
1403 	/* Check out the dir to be renamed */
1404 
1405 	if (dir_rename) {
1406 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1407 		if (error)
1408 			goto out_gunlock;
1409 	}
1410 
1411 	if (nip == NULL)
1412 		alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1413 	error = alloc_required;
1414 	if (error < 0)
1415 		goto out_gunlock;
1416 
1417 	if (alloc_required) {
1418 		struct gfs2_alloc_parms ap = { .target = sdp->sd_max_dirres, };
1419 		error = gfs2_quota_lock_check(ndip);
1420 		if (error)
1421 			goto out_gunlock;
1422 
1423 		error = gfs2_inplace_reserve(ndip, &ap);
1424 		if (error)
1425 			goto out_gunlock_q;
1426 
1427 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1428 					 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1429 					 4 * RES_DINODE + 4 * RES_LEAF +
1430 					 RES_STATFS + RES_QUOTA + 4, 0);
1431 		if (error)
1432 			goto out_ipreserv;
1433 	} else {
1434 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1435 					 5 * RES_LEAF + 4, 0);
1436 		if (error)
1437 			goto out_gunlock;
1438 	}
1439 
1440 	/* Remove the target file, if it exists */
1441 
1442 	if (nip)
1443 		error = gfs2_unlink_inode(ndip, ndentry);
1444 
1445 	if (dir_rename) {
1446 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1447 		if (error)
1448 			goto out_end_trans;
1449 	} else {
1450 		struct buffer_head *dibh;
1451 		error = gfs2_meta_inode_buffer(ip, &dibh);
1452 		if (error)
1453 			goto out_end_trans;
1454 		ip->i_inode.i_ctime = CURRENT_TIME;
1455 		gfs2_trans_add_meta(ip->i_gl, dibh);
1456 		gfs2_dinode_out(ip, dibh->b_data);
1457 		brelse(dibh);
1458 	}
1459 
1460 	error = gfs2_dir_del(odip, odentry);
1461 	if (error)
1462 		goto out_end_trans;
1463 
1464 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1465 	if (error)
1466 		goto out_end_trans;
1467 
1468 out_end_trans:
1469 	gfs2_trans_end(sdp);
1470 out_ipreserv:
1471 	if (alloc_required)
1472 		gfs2_inplace_release(ndip);
1473 out_gunlock_q:
1474 	if (alloc_required)
1475 		gfs2_quota_unlock(ndip);
1476 out_gunlock:
1477 	while (x--) {
1478 		gfs2_glock_dq(ghs + x);
1479 		gfs2_holder_uninit(ghs + x);
1480 	}
1481 out_gunlock_r:
1482 	if (r_gh.gh_gl)
1483 		gfs2_glock_dq_uninit(&r_gh);
1484 out:
1485 	return error;
1486 }
1487 
1488 /**
1489  * gfs2_follow_link - Follow a symbolic link
1490  * @dentry: The dentry of the link
1491  * @nd: Data that we pass to vfs_follow_link()
1492  *
1493  * This can handle symlinks of any size.
1494  *
1495  * Returns: 0 on success or error code
1496  */
1497 
1498 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1499 {
1500 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1501 	struct gfs2_holder i_gh;
1502 	struct buffer_head *dibh;
1503 	unsigned int size;
1504 	char *buf;
1505 	int error;
1506 
1507 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1508 	error = gfs2_glock_nq(&i_gh);
1509 	if (error) {
1510 		gfs2_holder_uninit(&i_gh);
1511 		nd_set_link(nd, ERR_PTR(error));
1512 		return NULL;
1513 	}
1514 
1515 	size = (unsigned int)i_size_read(&ip->i_inode);
1516 	if (size == 0) {
1517 		gfs2_consist_inode(ip);
1518 		buf = ERR_PTR(-EIO);
1519 		goto out;
1520 	}
1521 
1522 	error = gfs2_meta_inode_buffer(ip, &dibh);
1523 	if (error) {
1524 		buf = ERR_PTR(error);
1525 		goto out;
1526 	}
1527 
1528 	buf = kzalloc(size + 1, GFP_NOFS);
1529 	if (!buf)
1530 		buf = ERR_PTR(-ENOMEM);
1531 	else
1532 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1533 	brelse(dibh);
1534 out:
1535 	gfs2_glock_dq_uninit(&i_gh);
1536 	nd_set_link(nd, buf);
1537 	return NULL;
1538 }
1539 
1540 /**
1541  * gfs2_permission -
1542  * @inode: The inode
1543  * @mask: The mask to be tested
1544  * @flags: Indicates whether this is an RCU path walk or not
1545  *
1546  * This may be called from the VFS directly, or from within GFS2 with the
1547  * inode locked, so we look to see if the glock is already locked and only
1548  * lock the glock if its not already been done.
1549  *
1550  * Returns: errno
1551  */
1552 
1553 int gfs2_permission(struct inode *inode, int mask)
1554 {
1555 	struct gfs2_inode *ip;
1556 	struct gfs2_holder i_gh;
1557 	int error;
1558 	int unlock = 0;
1559 
1560 
1561 	ip = GFS2_I(inode);
1562 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1563 		if (mask & MAY_NOT_BLOCK)
1564 			return -ECHILD;
1565 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1566 		if (error)
1567 			return error;
1568 		unlock = 1;
1569 	}
1570 
1571 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1572 		error = -EACCES;
1573 	else
1574 		error = generic_permission(inode, mask);
1575 	if (unlock)
1576 		gfs2_glock_dq_uninit(&i_gh);
1577 
1578 	return error;
1579 }
1580 
1581 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1582 {
1583 	setattr_copy(inode, attr);
1584 	mark_inode_dirty(inode);
1585 	return 0;
1586 }
1587 
1588 /**
1589  * gfs2_setattr_simple -
1590  * @ip:
1591  * @attr:
1592  *
1593  * Returns: errno
1594  */
1595 
1596 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1597 {
1598 	int error;
1599 
1600 	if (current->journal_info)
1601 		return __gfs2_setattr_simple(inode, attr);
1602 
1603 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1604 	if (error)
1605 		return error;
1606 
1607 	error = __gfs2_setattr_simple(inode, attr);
1608 	gfs2_trans_end(GFS2_SB(inode));
1609 	return error;
1610 }
1611 
1612 static int setattr_chown(struct inode *inode, struct iattr *attr)
1613 {
1614 	struct gfs2_inode *ip = GFS2_I(inode);
1615 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1616 	kuid_t ouid, nuid;
1617 	kgid_t ogid, ngid;
1618 	int error;
1619 
1620 	ouid = inode->i_uid;
1621 	ogid = inode->i_gid;
1622 	nuid = attr->ia_uid;
1623 	ngid = attr->ia_gid;
1624 
1625 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1626 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1627 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1628 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1629 
1630 	error = gfs2_quota_lock(ip, nuid, ngid);
1631 	if (error)
1632 		return error;
1633 
1634 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1635 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1636 		error = gfs2_quota_check(ip, nuid, ngid);
1637 		if (error)
1638 			goto out_gunlock_q;
1639 	}
1640 
1641 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1642 	if (error)
1643 		goto out_gunlock_q;
1644 
1645 	error = gfs2_setattr_simple(inode, attr);
1646 	if (error)
1647 		goto out_end_trans;
1648 
1649 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1650 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1651 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1652 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1653 		gfs2_quota_change(ip, blocks, nuid, ngid);
1654 	}
1655 
1656 out_end_trans:
1657 	gfs2_trans_end(sdp);
1658 out_gunlock_q:
1659 	gfs2_quota_unlock(ip);
1660 	return error;
1661 }
1662 
1663 /**
1664  * gfs2_setattr - Change attributes on an inode
1665  * @dentry: The dentry which is changing
1666  * @attr: The structure describing the change
1667  *
1668  * The VFS layer wants to change one or more of an inodes attributes.  Write
1669  * that change out to disk.
1670  *
1671  * Returns: errno
1672  */
1673 
1674 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1675 {
1676 	struct inode *inode = dentry->d_inode;
1677 	struct gfs2_inode *ip = GFS2_I(inode);
1678 	struct gfs2_holder i_gh;
1679 	int error;
1680 
1681 	error = gfs2_rs_alloc(ip);
1682 	if (error)
1683 		return error;
1684 
1685 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1686 	if (error)
1687 		return error;
1688 
1689 	error = -EPERM;
1690 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1691 		goto out;
1692 
1693 	error = inode_change_ok(inode, attr);
1694 	if (error)
1695 		goto out;
1696 
1697 	if (attr->ia_valid & ATTR_SIZE)
1698 		error = gfs2_setattr_size(inode, attr->ia_size);
1699 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1700 		error = setattr_chown(inode, attr);
1701 	else {
1702 		error = gfs2_setattr_simple(inode, attr);
1703 		if (!error && attr->ia_valid & ATTR_MODE)
1704 			error = posix_acl_chmod(inode, inode->i_mode);
1705 	}
1706 
1707 out:
1708 	if (!error)
1709 		mark_inode_dirty(inode);
1710 	gfs2_glock_dq_uninit(&i_gh);
1711 	return error;
1712 }
1713 
1714 /**
1715  * gfs2_getattr - Read out an inode's attributes
1716  * @mnt: The vfsmount the inode is being accessed from
1717  * @dentry: The dentry to stat
1718  * @stat: The inode's stats
1719  *
1720  * This may be called from the VFS directly, or from within GFS2 with the
1721  * inode locked, so we look to see if the glock is already locked and only
1722  * lock the glock if its not already been done. Note that its the NFS
1723  * readdirplus operation which causes this to be called (from filldir)
1724  * with the glock already held.
1725  *
1726  * Returns: errno
1727  */
1728 
1729 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1730 			struct kstat *stat)
1731 {
1732 	struct inode *inode = dentry->d_inode;
1733 	struct gfs2_inode *ip = GFS2_I(inode);
1734 	struct gfs2_holder gh;
1735 	int error;
1736 	int unlock = 0;
1737 
1738 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1739 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1740 		if (error)
1741 			return error;
1742 		unlock = 1;
1743 	}
1744 
1745 	generic_fillattr(inode, stat);
1746 	if (unlock)
1747 		gfs2_glock_dq_uninit(&gh);
1748 
1749 	return 0;
1750 }
1751 
1752 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1753 			 const void *data, size_t size, int flags)
1754 {
1755 	struct inode *inode = dentry->d_inode;
1756 	struct gfs2_inode *ip = GFS2_I(inode);
1757 	struct gfs2_holder gh;
1758 	int ret;
1759 
1760 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1761 	ret = gfs2_glock_nq(&gh);
1762 	if (ret == 0) {
1763 		ret = gfs2_rs_alloc(ip);
1764 		if (ret == 0)
1765 			ret = generic_setxattr(dentry, name, data, size, flags);
1766 		gfs2_glock_dq(&gh);
1767 	}
1768 	gfs2_holder_uninit(&gh);
1769 	return ret;
1770 }
1771 
1772 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1773 			     void *data, size_t size)
1774 {
1775 	struct inode *inode = dentry->d_inode;
1776 	struct gfs2_inode *ip = GFS2_I(inode);
1777 	struct gfs2_holder gh;
1778 	int ret;
1779 
1780 	/* For selinux during lookup */
1781 	if (gfs2_glock_is_locked_by_me(ip->i_gl))
1782 		return generic_getxattr(dentry, name, data, size);
1783 
1784 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1785 	ret = gfs2_glock_nq(&gh);
1786 	if (ret == 0) {
1787 		ret = generic_getxattr(dentry, name, data, size);
1788 		gfs2_glock_dq(&gh);
1789 	}
1790 	gfs2_holder_uninit(&gh);
1791 	return ret;
1792 }
1793 
1794 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1795 {
1796 	struct inode *inode = dentry->d_inode;
1797 	struct gfs2_inode *ip = GFS2_I(inode);
1798 	struct gfs2_holder gh;
1799 	int ret;
1800 
1801 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1802 	ret = gfs2_glock_nq(&gh);
1803 	if (ret == 0) {
1804 		ret = gfs2_rs_alloc(ip);
1805 		if (ret == 0)
1806 			ret = generic_removexattr(dentry, name);
1807 		gfs2_glock_dq(&gh);
1808 	}
1809 	gfs2_holder_uninit(&gh);
1810 	return ret;
1811 }
1812 
1813 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1814 		       u64 start, u64 len)
1815 {
1816 	struct gfs2_inode *ip = GFS2_I(inode);
1817 	struct gfs2_holder gh;
1818 	int ret;
1819 
1820 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1821 	if (ret)
1822 		return ret;
1823 
1824 	mutex_lock(&inode->i_mutex);
1825 
1826 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1827 	if (ret)
1828 		goto out;
1829 
1830 	if (gfs2_is_stuffed(ip)) {
1831 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1832 		u64 size = i_size_read(inode);
1833 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1834 			    FIEMAP_EXTENT_DATA_INLINE;
1835 		phys += sizeof(struct gfs2_dinode);
1836 		phys += start;
1837 		if (start + len > size)
1838 			len = size - start;
1839 		if (start < size)
1840 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1841 						      len, flags);
1842 		if (ret == 1)
1843 			ret = 0;
1844 	} else {
1845 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1846 					     gfs2_block_map);
1847 	}
1848 
1849 	gfs2_glock_dq_uninit(&gh);
1850 out:
1851 	mutex_unlock(&inode->i_mutex);
1852 	return ret;
1853 }
1854 
1855 const struct inode_operations gfs2_file_iops = {
1856 	.permission = gfs2_permission,
1857 	.setattr = gfs2_setattr,
1858 	.getattr = gfs2_getattr,
1859 	.setxattr = gfs2_setxattr,
1860 	.getxattr = gfs2_getxattr,
1861 	.listxattr = gfs2_listxattr,
1862 	.removexattr = gfs2_removexattr,
1863 	.fiemap = gfs2_fiemap,
1864 	.get_acl = gfs2_get_acl,
1865 	.set_acl = gfs2_set_acl,
1866 };
1867 
1868 const struct inode_operations gfs2_dir_iops = {
1869 	.create = gfs2_create,
1870 	.lookup = gfs2_lookup,
1871 	.link = gfs2_link,
1872 	.unlink = gfs2_unlink,
1873 	.symlink = gfs2_symlink,
1874 	.mkdir = gfs2_mkdir,
1875 	.rmdir = gfs2_unlink,
1876 	.mknod = gfs2_mknod,
1877 	.rename = gfs2_rename,
1878 	.permission = gfs2_permission,
1879 	.setattr = gfs2_setattr,
1880 	.getattr = gfs2_getattr,
1881 	.setxattr = gfs2_setxattr,
1882 	.getxattr = gfs2_getxattr,
1883 	.listxattr = gfs2_listxattr,
1884 	.removexattr = gfs2_removexattr,
1885 	.fiemap = gfs2_fiemap,
1886 	.get_acl = gfs2_get_acl,
1887 	.set_acl = gfs2_set_acl,
1888 	.atomic_open = gfs2_atomic_open,
1889 };
1890 
1891 const struct inode_operations gfs2_symlink_iops = {
1892 	.readlink = generic_readlink,
1893 	.follow_link = gfs2_follow_link,
1894 	.put_link = kfree_put_link,
1895 	.permission = gfs2_permission,
1896 	.setattr = gfs2_setattr,
1897 	.getattr = gfs2_getattr,
1898 	.setxattr = gfs2_setxattr,
1899 	.getxattr = gfs2_getxattr,
1900 	.listxattr = gfs2_listxattr,
1901 	.removexattr = gfs2_removexattr,
1902 	.fiemap = gfs2_fiemap,
1903 };
1904 
1905