xref: /openbmc/linux/fs/gfs2/inode.c (revision 7b9cff46)
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 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 	int arq;
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 	arq = error = gfs2_diradd_alloc_required(dir, name);
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, arq);
694 	if (error)
695 		goto fail_gunlock3;
696 
697 	mark_inode_dirty(inode);
698 	d_instantiate(dentry, inode);
699 	if (file)
700 		error = finish_open(file, dentry, gfs2_open_common, opened);
701 	gfs2_glock_dq_uninit(ghs);
702 	gfs2_glock_dq_uninit(ghs + 1);
703 	return error;
704 
705 fail_gunlock3:
706 	gfs2_glock_dq_uninit(ghs + 1);
707 	if (ip->i_gl)
708 		gfs2_glock_put(ip->i_gl);
709 	goto fail_gunlock;
710 
711 fail_gunlock2:
712 	gfs2_glock_dq_uninit(ghs + 1);
713 fail_free_inode:
714 	if (ip->i_gl)
715 		gfs2_glock_put(ip->i_gl);
716 	gfs2_rs_delete(ip, NULL);
717 	free_inode_nonrcu(inode);
718 	inode = NULL;
719 fail_gunlock:
720 	gfs2_glock_dq_uninit(ghs);
721 	if (inode && !IS_ERR(inode)) {
722 		clear_nlink(inode);
723 		mark_inode_dirty(inode);
724 		set_bit(GIF_ALLOC_FAILED, &GFS2_I(inode)->i_flags);
725 		iput(inode);
726 	}
727 fail:
728 	return error;
729 }
730 
731 /**
732  * gfs2_create - Create a file
733  * @dir: The directory in which to create the file
734  * @dentry: The dentry of the new file
735  * @mode: The mode of the new file
736  *
737  * Returns: errno
738  */
739 
740 static int gfs2_create(struct inode *dir, struct dentry *dentry,
741 		       umode_t mode, bool excl)
742 {
743 	return gfs2_create_inode(dir, dentry, NULL, S_IFREG | mode, 0, NULL, 0, excl, NULL);
744 }
745 
746 /**
747  * __gfs2_lookup - Look up a filename in a directory and return its inode
748  * @dir: The directory inode
749  * @dentry: The dentry of the new inode
750  * @file: File to be opened
751  * @opened: atomic_open flags
752  *
753  *
754  * Returns: errno
755  */
756 
757 static struct dentry *__gfs2_lookup(struct inode *dir, struct dentry *dentry,
758 				    struct file *file, int *opened)
759 {
760 	struct inode *inode;
761 	struct dentry *d;
762 	struct gfs2_holder gh;
763 	struct gfs2_glock *gl;
764 	int error;
765 
766 	inode = gfs2_lookupi(dir, &dentry->d_name, 0);
767 	if (!inode)
768 		return NULL;
769 	if (IS_ERR(inode))
770 		return ERR_CAST(inode);
771 
772 	gl = GFS2_I(inode)->i_gl;
773 	error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
774 	if (error) {
775 		iput(inode);
776 		return ERR_PTR(error);
777 	}
778 
779 	d = d_splice_alias(inode, dentry);
780 	if (file && S_ISREG(inode->i_mode))
781 		error = finish_open(file, dentry, gfs2_open_common, opened);
782 
783 	gfs2_glock_dq_uninit(&gh);
784 	if (error) {
785 		dput(d);
786 		return ERR_PTR(error);
787 	}
788 	return d;
789 }
790 
791 static struct dentry *gfs2_lookup(struct inode *dir, struct dentry *dentry,
792 				  unsigned flags)
793 {
794 	return __gfs2_lookup(dir, dentry, NULL, NULL);
795 }
796 
797 /**
798  * gfs2_link - Link to a file
799  * @old_dentry: The inode to link
800  * @dir: Add link to this directory
801  * @dentry: The name of the link
802  *
803  * Link the inode in "old_dentry" into the directory "dir" with the
804  * name in "dentry".
805  *
806  * Returns: errno
807  */
808 
809 static int gfs2_link(struct dentry *old_dentry, struct inode *dir,
810 		     struct dentry *dentry)
811 {
812 	struct gfs2_inode *dip = GFS2_I(dir);
813 	struct gfs2_sbd *sdp = GFS2_SB(dir);
814 	struct inode *inode = old_dentry->d_inode;
815 	struct gfs2_inode *ip = GFS2_I(inode);
816 	struct gfs2_holder ghs[2];
817 	struct buffer_head *dibh;
818 	int alloc_required;
819 	int error;
820 
821 	if (S_ISDIR(inode->i_mode))
822 		return -EPERM;
823 
824 	error = gfs2_rs_alloc(dip);
825 	if (error)
826 		return error;
827 
828 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
829 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + 1);
830 
831 	error = gfs2_glock_nq(ghs); /* parent */
832 	if (error)
833 		goto out_parent;
834 
835 	error = gfs2_glock_nq(ghs + 1); /* child */
836 	if (error)
837 		goto out_child;
838 
839 	error = -ENOENT;
840 	if (inode->i_nlink == 0)
841 		goto out_gunlock;
842 
843 	error = gfs2_permission(dir, MAY_WRITE | MAY_EXEC);
844 	if (error)
845 		goto out_gunlock;
846 
847 	error = gfs2_dir_check(dir, &dentry->d_name, NULL);
848 	switch (error) {
849 	case -ENOENT:
850 		break;
851 	case 0:
852 		error = -EEXIST;
853 	default:
854 		goto out_gunlock;
855 	}
856 
857 	error = -EINVAL;
858 	if (!dip->i_inode.i_nlink)
859 		goto out_gunlock;
860 	error = -EFBIG;
861 	if (dip->i_entries == (u32)-1)
862 		goto out_gunlock;
863 	error = -EPERM;
864 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
865 		goto out_gunlock;
866 	error = -EINVAL;
867 	if (!ip->i_inode.i_nlink)
868 		goto out_gunlock;
869 	error = -EMLINK;
870 	if (ip->i_inode.i_nlink == (u32)-1)
871 		goto out_gunlock;
872 
873 	alloc_required = error = gfs2_diradd_alloc_required(dir, &dentry->d_name);
874 	if (error < 0)
875 		goto out_gunlock;
876 	error = 0;
877 
878 	if (alloc_required) {
879 		struct gfs2_alloc_parms ap = { .target = sdp->sd_max_dirres, };
880 		error = gfs2_quota_lock_check(dip);
881 		if (error)
882 			goto out_gunlock;
883 
884 		error = gfs2_inplace_reserve(dip, &ap);
885 		if (error)
886 			goto out_gunlock_q;
887 
888 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
889 					 gfs2_rg_blocks(dip, sdp->sd_max_dirres) +
890 					 2 * RES_DINODE + RES_STATFS +
891 					 RES_QUOTA, 0);
892 		if (error)
893 			goto out_ipres;
894 	} else {
895 		error = gfs2_trans_begin(sdp, 2 * RES_DINODE + RES_LEAF, 0);
896 		if (error)
897 			goto out_ipres;
898 	}
899 
900 	error = gfs2_meta_inode_buffer(ip, &dibh);
901 	if (error)
902 		goto out_end_trans;
903 
904 	error = gfs2_dir_add(dir, &dentry->d_name, ip);
905 	if (error)
906 		goto out_brelse;
907 
908 	gfs2_trans_add_meta(ip->i_gl, dibh);
909 	inc_nlink(&ip->i_inode);
910 	ip->i_inode.i_ctime = CURRENT_TIME;
911 	ihold(inode);
912 	d_instantiate(dentry, inode);
913 	mark_inode_dirty(inode);
914 
915 out_brelse:
916 	brelse(dibh);
917 out_end_trans:
918 	gfs2_trans_end(sdp);
919 out_ipres:
920 	if (alloc_required)
921 		gfs2_inplace_release(dip);
922 out_gunlock_q:
923 	if (alloc_required)
924 		gfs2_quota_unlock(dip);
925 out_gunlock:
926 	gfs2_glock_dq(ghs + 1);
927 out_child:
928 	gfs2_glock_dq(ghs);
929 out_parent:
930 	gfs2_holder_uninit(ghs);
931 	gfs2_holder_uninit(ghs + 1);
932 	return error;
933 }
934 
935 /*
936  * gfs2_unlink_ok - check to see that a inode is still in a directory
937  * @dip: the directory
938  * @name: the name of the file
939  * @ip: the inode
940  *
941  * Assumes that the lock on (at least) @dip is held.
942  *
943  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
944  */
945 
946 static int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
947 			  const struct gfs2_inode *ip)
948 {
949 	int error;
950 
951 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
952 		return -EPERM;
953 
954 	if ((dip->i_inode.i_mode & S_ISVTX) &&
955 	    !uid_eq(dip->i_inode.i_uid, current_fsuid()) &&
956 	    !uid_eq(ip->i_inode.i_uid, current_fsuid()) && !capable(CAP_FOWNER))
957 		return -EPERM;
958 
959 	if (IS_APPEND(&dip->i_inode))
960 		return -EPERM;
961 
962 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC);
963 	if (error)
964 		return error;
965 
966 	error = gfs2_dir_check(&dip->i_inode, name, ip);
967 	if (error)
968 		return error;
969 
970 	return 0;
971 }
972 
973 /**
974  * gfs2_unlink_inode - Removes an inode from its parent dir and unlinks it
975  * @dip: The parent directory
976  * @name: The name of the entry in the parent directory
977  * @inode: The inode to be removed
978  *
979  * Called with all the locks and in a transaction. This will only be
980  * called for a directory after it has been checked to ensure it is empty.
981  *
982  * Returns: 0 on success, or an error
983  */
984 
985 static int gfs2_unlink_inode(struct gfs2_inode *dip,
986 			     const struct dentry *dentry)
987 {
988 	struct inode *inode = dentry->d_inode;
989 	struct gfs2_inode *ip = GFS2_I(inode);
990 	int error;
991 
992 	error = gfs2_dir_del(dip, dentry);
993 	if (error)
994 		return error;
995 
996 	ip->i_entries = 0;
997 	inode->i_ctime = CURRENT_TIME;
998 	if (S_ISDIR(inode->i_mode))
999 		clear_nlink(inode);
1000 	else
1001 		drop_nlink(inode);
1002 	mark_inode_dirty(inode);
1003 	if (inode->i_nlink == 0)
1004 		gfs2_unlink_di(inode);
1005 	return 0;
1006 }
1007 
1008 
1009 /**
1010  * gfs2_unlink - Unlink an inode (this does rmdir as well)
1011  * @dir: The inode of the directory containing the inode to unlink
1012  * @dentry: The file itself
1013  *
1014  * This routine uses the type of the inode as a flag to figure out
1015  * whether this is an unlink or an rmdir.
1016  *
1017  * Returns: errno
1018  */
1019 
1020 static int gfs2_unlink(struct inode *dir, struct dentry *dentry)
1021 {
1022 	struct gfs2_inode *dip = GFS2_I(dir);
1023 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1024 	struct inode *inode = dentry->d_inode;
1025 	struct gfs2_inode *ip = GFS2_I(inode);
1026 	struct gfs2_holder ghs[3];
1027 	struct gfs2_rgrpd *rgd;
1028 	int error;
1029 
1030 	error = gfs2_rindex_update(sdp);
1031 	if (error)
1032 		return error;
1033 
1034 	error = -EROFS;
1035 
1036 	gfs2_holder_init(dip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1037 	gfs2_holder_init(ip->i_gl,  LM_ST_EXCLUSIVE, 0, ghs + 1);
1038 
1039 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1040 	if (!rgd)
1041 		goto out_inodes;
1042 
1043 	gfs2_holder_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + 2);
1044 
1045 
1046 	error = gfs2_glock_nq(ghs); /* parent */
1047 	if (error)
1048 		goto out_parent;
1049 
1050 	error = gfs2_glock_nq(ghs + 1); /* child */
1051 	if (error)
1052 		goto out_child;
1053 
1054 	error = -ENOENT;
1055 	if (inode->i_nlink == 0)
1056 		goto out_rgrp;
1057 
1058 	if (S_ISDIR(inode->i_mode)) {
1059 		error = -ENOTEMPTY;
1060 		if (ip->i_entries > 2 || inode->i_nlink > 2)
1061 			goto out_rgrp;
1062 	}
1063 
1064 	error = gfs2_glock_nq(ghs + 2); /* rgrp */
1065 	if (error)
1066 		goto out_rgrp;
1067 
1068 	error = gfs2_unlink_ok(dip, &dentry->d_name, ip);
1069 	if (error)
1070 		goto out_gunlock;
1071 
1072 	error = gfs2_trans_begin(sdp, 2*RES_DINODE + 3*RES_LEAF + RES_RG_BIT, 0);
1073 	if (error)
1074 		goto out_end_trans;
1075 
1076 	error = gfs2_unlink_inode(dip, dentry);
1077 
1078 out_end_trans:
1079 	gfs2_trans_end(sdp);
1080 out_gunlock:
1081 	gfs2_glock_dq(ghs + 2);
1082 out_rgrp:
1083 	gfs2_glock_dq(ghs + 1);
1084 out_child:
1085 	gfs2_glock_dq(ghs);
1086 out_parent:
1087 	gfs2_holder_uninit(ghs + 2);
1088 out_inodes:
1089 	gfs2_holder_uninit(ghs + 1);
1090 	gfs2_holder_uninit(ghs);
1091 	return error;
1092 }
1093 
1094 /**
1095  * gfs2_symlink - Create a symlink
1096  * @dir: The directory to create the symlink in
1097  * @dentry: The dentry to put the symlink in
1098  * @symname: The thing which the link points to
1099  *
1100  * Returns: errno
1101  */
1102 
1103 static int gfs2_symlink(struct inode *dir, struct dentry *dentry,
1104 			const char *symname)
1105 {
1106 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1107 	unsigned int size;
1108 
1109 	size = strlen(symname);
1110 	if (size > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode) - 1)
1111 		return -ENAMETOOLONG;
1112 
1113 	return gfs2_create_inode(dir, dentry, NULL, S_IFLNK | S_IRWXUGO, 0, symname, size, 0, NULL);
1114 }
1115 
1116 /**
1117  * gfs2_mkdir - Make a directory
1118  * @dir: The parent directory of the new one
1119  * @dentry: The dentry of the new directory
1120  * @mode: The mode of the new directory
1121  *
1122  * Returns: errno
1123  */
1124 
1125 static int gfs2_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1126 {
1127 	struct gfs2_sbd *sdp = GFS2_SB(dir);
1128 	unsigned dsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode);
1129 	return gfs2_create_inode(dir, dentry, NULL, S_IFDIR | mode, 0, NULL, dsize, 0, NULL);
1130 }
1131 
1132 /**
1133  * gfs2_mknod - Make a special file
1134  * @dir: The directory in which the special file will reside
1135  * @dentry: The dentry of the special file
1136  * @mode: The mode of the special file
1137  * @dev: The device specification of the special file
1138  *
1139  */
1140 
1141 static int gfs2_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1142 		      dev_t dev)
1143 {
1144 	return gfs2_create_inode(dir, dentry, NULL, mode, dev, NULL, 0, 0, NULL);
1145 }
1146 
1147 /**
1148  * gfs2_atomic_open - Atomically open a file
1149  * @dir: The directory
1150  * @dentry: The proposed new entry
1151  * @file: The proposed new struct file
1152  * @flags: open flags
1153  * @mode: File mode
1154  * @opened: Flag to say whether the file has been opened or not
1155  *
1156  * Returns: error code or 0 for success
1157  */
1158 
1159 static int gfs2_atomic_open(struct inode *dir, struct dentry *dentry,
1160                             struct file *file, unsigned flags,
1161                             umode_t mode, int *opened)
1162 {
1163 	struct dentry *d;
1164 	bool excl = !!(flags & O_EXCL);
1165 
1166 	d = __gfs2_lookup(dir, dentry, file, opened);
1167 	if (IS_ERR(d))
1168 		return PTR_ERR(d);
1169 	if (d != NULL)
1170 		dentry = d;
1171 	if (dentry->d_inode) {
1172 		if (!(*opened & FILE_OPENED))
1173 			return finish_no_open(file, dentry);
1174 		dput(d);
1175 		return 0;
1176 	}
1177 
1178 	BUG_ON(d != NULL);
1179 	if (!(flags & O_CREAT))
1180 		return -ENOENT;
1181 
1182 	return gfs2_create_inode(dir, dentry, file, S_IFREG | mode, 0, NULL, 0, excl, opened);
1183 }
1184 
1185 /*
1186  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1187  * @this: move this
1188  * @to: to here
1189  *
1190  * Follow @to back to the root and make sure we don't encounter @this
1191  * Assumes we already hold the rename lock.
1192  *
1193  * Returns: errno
1194  */
1195 
1196 static int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1197 {
1198 	struct inode *dir = &to->i_inode;
1199 	struct super_block *sb = dir->i_sb;
1200 	struct inode *tmp;
1201 	int error = 0;
1202 
1203 	igrab(dir);
1204 
1205 	for (;;) {
1206 		if (dir == &this->i_inode) {
1207 			error = -EINVAL;
1208 			break;
1209 		}
1210 		if (dir == sb->s_root->d_inode) {
1211 			error = 0;
1212 			break;
1213 		}
1214 
1215 		tmp = gfs2_lookupi(dir, &gfs2_qdotdot, 1);
1216 		if (IS_ERR(tmp)) {
1217 			error = PTR_ERR(tmp);
1218 			break;
1219 		}
1220 
1221 		iput(dir);
1222 		dir = tmp;
1223 	}
1224 
1225 	iput(dir);
1226 
1227 	return error;
1228 }
1229 
1230 /**
1231  * gfs2_rename - Rename a file
1232  * @odir: Parent directory of old file name
1233  * @odentry: The old dentry of the file
1234  * @ndir: Parent directory of new file name
1235  * @ndentry: The new dentry of the file
1236  *
1237  * Returns: errno
1238  */
1239 
1240 static int gfs2_rename(struct inode *odir, struct dentry *odentry,
1241 		       struct inode *ndir, struct dentry *ndentry)
1242 {
1243 	struct gfs2_inode *odip = GFS2_I(odir);
1244 	struct gfs2_inode *ndip = GFS2_I(ndir);
1245 	struct gfs2_inode *ip = GFS2_I(odentry->d_inode);
1246 	struct gfs2_inode *nip = NULL;
1247 	struct gfs2_sbd *sdp = GFS2_SB(odir);
1248 	struct gfs2_holder ghs[5], r_gh = { .gh_gl = NULL, };
1249 	struct gfs2_rgrpd *nrgd;
1250 	unsigned int num_gh;
1251 	int dir_rename = 0;
1252 	int alloc_required = 0;
1253 	unsigned int x;
1254 	int error;
1255 
1256 	if (ndentry->d_inode) {
1257 		nip = GFS2_I(ndentry->d_inode);
1258 		if (ip == nip)
1259 			return 0;
1260 	}
1261 
1262 	error = gfs2_rindex_update(sdp);
1263 	if (error)
1264 		return error;
1265 
1266 	error = gfs2_rs_alloc(ndip);
1267 	if (error)
1268 		return error;
1269 
1270 	if (odip != ndip) {
1271 		error = gfs2_glock_nq_init(sdp->sd_rename_gl, LM_ST_EXCLUSIVE,
1272 					   0, &r_gh);
1273 		if (error)
1274 			goto out;
1275 
1276 		if (S_ISDIR(ip->i_inode.i_mode)) {
1277 			dir_rename = 1;
1278 			/* don't move a dirctory into it's subdir */
1279 			error = gfs2_ok_to_move(ip, ndip);
1280 			if (error)
1281 				goto out_gunlock_r;
1282 		}
1283 	}
1284 
1285 	num_gh = 1;
1286 	gfs2_holder_init(odip->i_gl, LM_ST_EXCLUSIVE, 0, ghs);
1287 	if (odip != ndip) {
1288 		gfs2_holder_init(ndip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1289 		num_gh++;
1290 	}
1291 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1292 	num_gh++;
1293 
1294 	if (nip) {
1295 		gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh);
1296 		num_gh++;
1297 		/* grab the resource lock for unlink flag twiddling
1298 		 * this is the case of the target file already existing
1299 		 * so we unlink before doing the rename
1300 		 */
1301 		nrgd = gfs2_blk2rgrpd(sdp, nip->i_no_addr, 1);
1302 		if (nrgd)
1303 			gfs2_holder_init(nrgd->rd_gl, LM_ST_EXCLUSIVE, 0, ghs + num_gh++);
1304 	}
1305 
1306 	for (x = 0; x < num_gh; x++) {
1307 		error = gfs2_glock_nq(ghs + x);
1308 		if (error)
1309 			goto out_gunlock;
1310 	}
1311 
1312 	error = -ENOENT;
1313 	if (ip->i_inode.i_nlink == 0)
1314 		goto out_gunlock;
1315 
1316 	/* Check out the old directory */
1317 
1318 	error = gfs2_unlink_ok(odip, &odentry->d_name, ip);
1319 	if (error)
1320 		goto out_gunlock;
1321 
1322 	/* Check out the new directory */
1323 
1324 	if (nip) {
1325 		error = gfs2_unlink_ok(ndip, &ndentry->d_name, nip);
1326 		if (error)
1327 			goto out_gunlock;
1328 
1329 		if (nip->i_inode.i_nlink == 0) {
1330 			error = -EAGAIN;
1331 			goto out_gunlock;
1332 		}
1333 
1334 		if (S_ISDIR(nip->i_inode.i_mode)) {
1335 			if (nip->i_entries < 2) {
1336 				gfs2_consist_inode(nip);
1337 				error = -EIO;
1338 				goto out_gunlock;
1339 			}
1340 			if (nip->i_entries > 2) {
1341 				error = -ENOTEMPTY;
1342 				goto out_gunlock;
1343 			}
1344 		}
1345 	} else {
1346 		error = gfs2_permission(ndir, MAY_WRITE | MAY_EXEC);
1347 		if (error)
1348 			goto out_gunlock;
1349 
1350 		error = gfs2_dir_check(ndir, &ndentry->d_name, NULL);
1351 		switch (error) {
1352 		case -ENOENT:
1353 			error = 0;
1354 			break;
1355 		case 0:
1356 			error = -EEXIST;
1357 		default:
1358 			goto out_gunlock;
1359 		};
1360 
1361 		if (odip != ndip) {
1362 			if (!ndip->i_inode.i_nlink) {
1363 				error = -ENOENT;
1364 				goto out_gunlock;
1365 			}
1366 			if (ndip->i_entries == (u32)-1) {
1367 				error = -EFBIG;
1368 				goto out_gunlock;
1369 			}
1370 			if (S_ISDIR(ip->i_inode.i_mode) &&
1371 			    ndip->i_inode.i_nlink == (u32)-1) {
1372 				error = -EMLINK;
1373 				goto out_gunlock;
1374 			}
1375 		}
1376 	}
1377 
1378 	/* Check out the dir to be renamed */
1379 
1380 	if (dir_rename) {
1381 		error = gfs2_permission(odentry->d_inode, MAY_WRITE);
1382 		if (error)
1383 			goto out_gunlock;
1384 	}
1385 
1386 	if (nip == NULL)
1387 		alloc_required = gfs2_diradd_alloc_required(ndir, &ndentry->d_name);
1388 	error = alloc_required;
1389 	if (error < 0)
1390 		goto out_gunlock;
1391 
1392 	if (alloc_required) {
1393 		struct gfs2_alloc_parms ap = { .target = sdp->sd_max_dirres, };
1394 		error = gfs2_quota_lock_check(ndip);
1395 		if (error)
1396 			goto out_gunlock;
1397 
1398 		error = gfs2_inplace_reserve(ndip, &ap);
1399 		if (error)
1400 			goto out_gunlock_q;
1401 
1402 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
1403 					 gfs2_rg_blocks(ndip, sdp->sd_max_dirres) +
1404 					 4 * RES_DINODE + 4 * RES_LEAF +
1405 					 RES_STATFS + RES_QUOTA + 4, 0);
1406 		if (error)
1407 			goto out_ipreserv;
1408 	} else {
1409 		error = gfs2_trans_begin(sdp, 4 * RES_DINODE +
1410 					 5 * RES_LEAF + 4, 0);
1411 		if (error)
1412 			goto out_gunlock;
1413 	}
1414 
1415 	/* Remove the target file, if it exists */
1416 
1417 	if (nip)
1418 		error = gfs2_unlink_inode(ndip, ndentry);
1419 
1420 	if (dir_rename) {
1421 		error = gfs2_dir_mvino(ip, &gfs2_qdotdot, ndip, DT_DIR);
1422 		if (error)
1423 			goto out_end_trans;
1424 	} else {
1425 		struct buffer_head *dibh;
1426 		error = gfs2_meta_inode_buffer(ip, &dibh);
1427 		if (error)
1428 			goto out_end_trans;
1429 		ip->i_inode.i_ctime = CURRENT_TIME;
1430 		gfs2_trans_add_meta(ip->i_gl, dibh);
1431 		gfs2_dinode_out(ip, dibh->b_data);
1432 		brelse(dibh);
1433 	}
1434 
1435 	error = gfs2_dir_del(odip, odentry);
1436 	if (error)
1437 		goto out_end_trans;
1438 
1439 	error = gfs2_dir_add(ndir, &ndentry->d_name, ip);
1440 	if (error)
1441 		goto out_end_trans;
1442 
1443 out_end_trans:
1444 	gfs2_trans_end(sdp);
1445 out_ipreserv:
1446 	if (alloc_required)
1447 		gfs2_inplace_release(ndip);
1448 out_gunlock_q:
1449 	if (alloc_required)
1450 		gfs2_quota_unlock(ndip);
1451 out_gunlock:
1452 	while (x--) {
1453 		gfs2_glock_dq(ghs + x);
1454 		gfs2_holder_uninit(ghs + x);
1455 	}
1456 out_gunlock_r:
1457 	if (r_gh.gh_gl)
1458 		gfs2_glock_dq_uninit(&r_gh);
1459 out:
1460 	return error;
1461 }
1462 
1463 /**
1464  * gfs2_follow_link - Follow a symbolic link
1465  * @dentry: The dentry of the link
1466  * @nd: Data that we pass to vfs_follow_link()
1467  *
1468  * This can handle symlinks of any size.
1469  *
1470  * Returns: 0 on success or error code
1471  */
1472 
1473 static void *gfs2_follow_link(struct dentry *dentry, struct nameidata *nd)
1474 {
1475 	struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
1476 	struct gfs2_holder i_gh;
1477 	struct buffer_head *dibh;
1478 	unsigned int size;
1479 	char *buf;
1480 	int error;
1481 
1482 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
1483 	error = gfs2_glock_nq(&i_gh);
1484 	if (error) {
1485 		gfs2_holder_uninit(&i_gh);
1486 		nd_set_link(nd, ERR_PTR(error));
1487 		return NULL;
1488 	}
1489 
1490 	size = (unsigned int)i_size_read(&ip->i_inode);
1491 	if (size == 0) {
1492 		gfs2_consist_inode(ip);
1493 		buf = ERR_PTR(-EIO);
1494 		goto out;
1495 	}
1496 
1497 	error = gfs2_meta_inode_buffer(ip, &dibh);
1498 	if (error) {
1499 		buf = ERR_PTR(error);
1500 		goto out;
1501 	}
1502 
1503 	buf = kzalloc(size + 1, GFP_NOFS);
1504 	if (!buf)
1505 		buf = ERR_PTR(-ENOMEM);
1506 	else
1507 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
1508 	brelse(dibh);
1509 out:
1510 	gfs2_glock_dq_uninit(&i_gh);
1511 	nd_set_link(nd, buf);
1512 	return NULL;
1513 }
1514 
1515 static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
1516 {
1517 	char *s = nd_get_link(nd);
1518 	if (!IS_ERR(s))
1519 		kfree(s);
1520 }
1521 
1522 /**
1523  * gfs2_permission -
1524  * @inode: The inode
1525  * @mask: The mask to be tested
1526  * @flags: Indicates whether this is an RCU path walk or not
1527  *
1528  * This may be called from the VFS directly, or from within GFS2 with the
1529  * inode locked, so we look to see if the glock is already locked and only
1530  * lock the glock if its not already been done.
1531  *
1532  * Returns: errno
1533  */
1534 
1535 int gfs2_permission(struct inode *inode, int mask)
1536 {
1537 	struct gfs2_inode *ip;
1538 	struct gfs2_holder i_gh;
1539 	int error;
1540 	int unlock = 0;
1541 
1542 
1543 	ip = GFS2_I(inode);
1544 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1545 		if (mask & MAY_NOT_BLOCK)
1546 			return -ECHILD;
1547 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
1548 		if (error)
1549 			return error;
1550 		unlock = 1;
1551 	}
1552 
1553 	if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
1554 		error = -EACCES;
1555 	else
1556 		error = generic_permission(inode, mask);
1557 	if (unlock)
1558 		gfs2_glock_dq_uninit(&i_gh);
1559 
1560 	return error;
1561 }
1562 
1563 static int __gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1564 {
1565 	setattr_copy(inode, attr);
1566 	mark_inode_dirty(inode);
1567 	return 0;
1568 }
1569 
1570 /**
1571  * gfs2_setattr_simple -
1572  * @ip:
1573  * @attr:
1574  *
1575  * Returns: errno
1576  */
1577 
1578 int gfs2_setattr_simple(struct inode *inode, struct iattr *attr)
1579 {
1580 	int error;
1581 
1582 	if (current->journal_info)
1583 		return __gfs2_setattr_simple(inode, attr);
1584 
1585 	error = gfs2_trans_begin(GFS2_SB(inode), RES_DINODE, 0);
1586 	if (error)
1587 		return error;
1588 
1589 	error = __gfs2_setattr_simple(inode, attr);
1590 	gfs2_trans_end(GFS2_SB(inode));
1591 	return error;
1592 }
1593 
1594 static int setattr_chown(struct inode *inode, struct iattr *attr)
1595 {
1596 	struct gfs2_inode *ip = GFS2_I(inode);
1597 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1598 	kuid_t ouid, nuid;
1599 	kgid_t ogid, ngid;
1600 	int error;
1601 
1602 	ouid = inode->i_uid;
1603 	ogid = inode->i_gid;
1604 	nuid = attr->ia_uid;
1605 	ngid = attr->ia_gid;
1606 
1607 	if (!(attr->ia_valid & ATTR_UID) || uid_eq(ouid, nuid))
1608 		ouid = nuid = NO_UID_QUOTA_CHANGE;
1609 	if (!(attr->ia_valid & ATTR_GID) || gid_eq(ogid, ngid))
1610 		ogid = ngid = NO_GID_QUOTA_CHANGE;
1611 
1612 	error = gfs2_quota_lock(ip, nuid, ngid);
1613 	if (error)
1614 		return error;
1615 
1616 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1617 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1618 		error = gfs2_quota_check(ip, nuid, ngid);
1619 		if (error)
1620 			goto out_gunlock_q;
1621 	}
1622 
1623 	error = gfs2_trans_begin(sdp, RES_DINODE + 2 * RES_QUOTA, 0);
1624 	if (error)
1625 		goto out_gunlock_q;
1626 
1627 	error = gfs2_setattr_simple(inode, attr);
1628 	if (error)
1629 		goto out_end_trans;
1630 
1631 	if (!uid_eq(ouid, NO_UID_QUOTA_CHANGE) ||
1632 	    !gid_eq(ogid, NO_GID_QUOTA_CHANGE)) {
1633 		u64 blocks = gfs2_get_inode_blocks(&ip->i_inode);
1634 		gfs2_quota_change(ip, -blocks, ouid, ogid);
1635 		gfs2_quota_change(ip, blocks, nuid, ngid);
1636 	}
1637 
1638 out_end_trans:
1639 	gfs2_trans_end(sdp);
1640 out_gunlock_q:
1641 	gfs2_quota_unlock(ip);
1642 	return error;
1643 }
1644 
1645 /**
1646  * gfs2_setattr - Change attributes on an inode
1647  * @dentry: The dentry which is changing
1648  * @attr: The structure describing the change
1649  *
1650  * The VFS layer wants to change one or more of an inodes attributes.  Write
1651  * that change out to disk.
1652  *
1653  * Returns: errno
1654  */
1655 
1656 static int gfs2_setattr(struct dentry *dentry, struct iattr *attr)
1657 {
1658 	struct inode *inode = dentry->d_inode;
1659 	struct gfs2_inode *ip = GFS2_I(inode);
1660 	struct gfs2_holder i_gh;
1661 	int error;
1662 
1663 	error = gfs2_rs_alloc(ip);
1664 	if (error)
1665 		return error;
1666 
1667 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1668 	if (error)
1669 		return error;
1670 
1671 	error = -EPERM;
1672 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
1673 		goto out;
1674 
1675 	error = inode_change_ok(inode, attr);
1676 	if (error)
1677 		goto out;
1678 
1679 	if (attr->ia_valid & ATTR_SIZE)
1680 		error = gfs2_setattr_size(inode, attr->ia_size);
1681 	else if (attr->ia_valid & (ATTR_UID | ATTR_GID))
1682 		error = setattr_chown(inode, attr);
1683 	else if ((attr->ia_valid & ATTR_MODE) && IS_POSIXACL(inode))
1684 		error = gfs2_acl_chmod(ip, attr);
1685 	else
1686 		error = gfs2_setattr_simple(inode, attr);
1687 
1688 out:
1689 	if (!error)
1690 		mark_inode_dirty(inode);
1691 	gfs2_glock_dq_uninit(&i_gh);
1692 	return error;
1693 }
1694 
1695 /**
1696  * gfs2_getattr - Read out an inode's attributes
1697  * @mnt: The vfsmount the inode is being accessed from
1698  * @dentry: The dentry to stat
1699  * @stat: The inode's stats
1700  *
1701  * This may be called from the VFS directly, or from within GFS2 with the
1702  * inode locked, so we look to see if the glock is already locked and only
1703  * lock the glock if its not already been done. Note that its the NFS
1704  * readdirplus operation which causes this to be called (from filldir)
1705  * with the glock already held.
1706  *
1707  * Returns: errno
1708  */
1709 
1710 static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
1711 			struct kstat *stat)
1712 {
1713 	struct inode *inode = dentry->d_inode;
1714 	struct gfs2_inode *ip = GFS2_I(inode);
1715 	struct gfs2_holder gh;
1716 	int error;
1717 	int unlock = 0;
1718 
1719 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
1720 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1721 		if (error)
1722 			return error;
1723 		unlock = 1;
1724 	}
1725 
1726 	generic_fillattr(inode, stat);
1727 	if (unlock)
1728 		gfs2_glock_dq_uninit(&gh);
1729 
1730 	return 0;
1731 }
1732 
1733 static int gfs2_setxattr(struct dentry *dentry, const char *name,
1734 			 const void *data, size_t size, int flags)
1735 {
1736 	struct inode *inode = dentry->d_inode;
1737 	struct gfs2_inode *ip = GFS2_I(inode);
1738 	struct gfs2_holder gh;
1739 	int ret;
1740 
1741 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1742 	ret = gfs2_glock_nq(&gh);
1743 	if (ret == 0) {
1744 		ret = gfs2_rs_alloc(ip);
1745 		if (ret == 0)
1746 			ret = generic_setxattr(dentry, name, data, size, flags);
1747 		gfs2_glock_dq(&gh);
1748 	}
1749 	gfs2_holder_uninit(&gh);
1750 	return ret;
1751 }
1752 
1753 static ssize_t gfs2_getxattr(struct dentry *dentry, const char *name,
1754 			     void *data, size_t size)
1755 {
1756 	struct inode *inode = dentry->d_inode;
1757 	struct gfs2_inode *ip = GFS2_I(inode);
1758 	struct gfs2_holder gh;
1759 	int ret;
1760 
1761 	/* For selinux during lookup */
1762 	if (gfs2_glock_is_locked_by_me(ip->i_gl))
1763 		return generic_getxattr(dentry, name, data, size);
1764 
1765 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &gh);
1766 	ret = gfs2_glock_nq(&gh);
1767 	if (ret == 0) {
1768 		ret = generic_getxattr(dentry, name, data, size);
1769 		gfs2_glock_dq(&gh);
1770 	}
1771 	gfs2_holder_uninit(&gh);
1772 	return ret;
1773 }
1774 
1775 static int gfs2_removexattr(struct dentry *dentry, const char *name)
1776 {
1777 	struct inode *inode = dentry->d_inode;
1778 	struct gfs2_inode *ip = GFS2_I(inode);
1779 	struct gfs2_holder gh;
1780 	int ret;
1781 
1782 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1783 	ret = gfs2_glock_nq(&gh);
1784 	if (ret == 0) {
1785 		ret = gfs2_rs_alloc(ip);
1786 		if (ret == 0)
1787 			ret = generic_removexattr(dentry, name);
1788 		gfs2_glock_dq(&gh);
1789 	}
1790 	gfs2_holder_uninit(&gh);
1791 	return ret;
1792 }
1793 
1794 static int gfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1795 		       u64 start, u64 len)
1796 {
1797 	struct gfs2_inode *ip = GFS2_I(inode);
1798 	struct gfs2_holder gh;
1799 	int ret;
1800 
1801 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1802 	if (ret)
1803 		return ret;
1804 
1805 	mutex_lock(&inode->i_mutex);
1806 
1807 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1808 	if (ret)
1809 		goto out;
1810 
1811 	if (gfs2_is_stuffed(ip)) {
1812 		u64 phys = ip->i_no_addr << inode->i_blkbits;
1813 		u64 size = i_size_read(inode);
1814 		u32 flags = FIEMAP_EXTENT_LAST|FIEMAP_EXTENT_NOT_ALIGNED|
1815 			    FIEMAP_EXTENT_DATA_INLINE;
1816 		phys += sizeof(struct gfs2_dinode);
1817 		phys += start;
1818 		if (start + len > size)
1819 			len = size - start;
1820 		if (start < size)
1821 			ret = fiemap_fill_next_extent(fieinfo, start, phys,
1822 						      len, flags);
1823 		if (ret == 1)
1824 			ret = 0;
1825 	} else {
1826 		ret = __generic_block_fiemap(inode, fieinfo, start, len,
1827 					     gfs2_block_map);
1828 	}
1829 
1830 	gfs2_glock_dq_uninit(&gh);
1831 out:
1832 	mutex_unlock(&inode->i_mutex);
1833 	return ret;
1834 }
1835 
1836 const struct inode_operations gfs2_file_iops = {
1837 	.permission = gfs2_permission,
1838 	.setattr = gfs2_setattr,
1839 	.getattr = gfs2_getattr,
1840 	.setxattr = gfs2_setxattr,
1841 	.getxattr = gfs2_getxattr,
1842 	.listxattr = gfs2_listxattr,
1843 	.removexattr = gfs2_removexattr,
1844 	.fiemap = gfs2_fiemap,
1845 	.get_acl = gfs2_get_acl,
1846 };
1847 
1848 const struct inode_operations gfs2_dir_iops = {
1849 	.create = gfs2_create,
1850 	.lookup = gfs2_lookup,
1851 	.link = gfs2_link,
1852 	.unlink = gfs2_unlink,
1853 	.symlink = gfs2_symlink,
1854 	.mkdir = gfs2_mkdir,
1855 	.rmdir = gfs2_unlink,
1856 	.mknod = gfs2_mknod,
1857 	.rename = gfs2_rename,
1858 	.permission = gfs2_permission,
1859 	.setattr = gfs2_setattr,
1860 	.getattr = gfs2_getattr,
1861 	.setxattr = gfs2_setxattr,
1862 	.getxattr = gfs2_getxattr,
1863 	.listxattr = gfs2_listxattr,
1864 	.removexattr = gfs2_removexattr,
1865 	.fiemap = gfs2_fiemap,
1866 	.get_acl = gfs2_get_acl,
1867 	.atomic_open = gfs2_atomic_open,
1868 };
1869 
1870 const struct inode_operations gfs2_symlink_iops = {
1871 	.readlink = generic_readlink,
1872 	.follow_link = gfs2_follow_link,
1873 	.put_link = gfs2_put_link,
1874 	.permission = gfs2_permission,
1875 	.setattr = gfs2_setattr,
1876 	.getattr = gfs2_getattr,
1877 	.setxattr = gfs2_setxattr,
1878 	.getxattr = gfs2_getxattr,
1879 	.listxattr = gfs2_listxattr,
1880 	.removexattr = gfs2_removexattr,
1881 	.fiemap = gfs2_fiemap,
1882 	.get_acl = gfs2_get_acl,
1883 };
1884 
1885