xref: /openbmc/linux/fs/gfs2/inode.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 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/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/posix_acl.h>
16 #include <linux/sort.h>
17 #include <linux/gfs2_ondisk.h>
18 #include <linux/crc32.h>
19 #include <linux/lm_interface.h>
20 #include <linux/security.h>
21 
22 #include "gfs2.h"
23 #include "incore.h"
24 #include "acl.h"
25 #include "bmap.h"
26 #include "dir.h"
27 #include "eattr.h"
28 #include "glock.h"
29 #include "glops.h"
30 #include "inode.h"
31 #include "log.h"
32 #include "meta_io.h"
33 #include "ops_address.h"
34 #include "ops_file.h"
35 #include "ops_inode.h"
36 #include "quota.h"
37 #include "rgrp.h"
38 #include "trans.h"
39 #include "util.h"
40 
41 static int iget_test(struct inode *inode, void *opaque)
42 {
43 	struct gfs2_inode *ip = GFS2_I(inode);
44 	struct gfs2_inum_host *inum = opaque;
45 
46 	if (ip->i_num.no_addr == inum->no_addr &&
47 	    inode->i_private != NULL)
48 		return 1;
49 
50 	return 0;
51 }
52 
53 static int iget_set(struct inode *inode, void *opaque)
54 {
55 	struct gfs2_inode *ip = GFS2_I(inode);
56 	struct gfs2_inum_host *inum = opaque;
57 
58 	ip->i_num = *inum;
59 	inode->i_ino = inum->no_addr;
60 	return 0;
61 }
62 
63 struct inode *gfs2_ilookup(struct super_block *sb, struct gfs2_inum_host *inum)
64 {
65 	return ilookup5(sb, (unsigned long)inum->no_addr,
66 			iget_test, inum);
67 }
68 
69 static struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum_host *inum)
70 {
71 	return iget5_locked(sb, (unsigned long)inum->no_addr,
72 		     iget_test, iget_set, inum);
73 }
74 
75 /**
76  * gfs2_inode_lookup - Lookup an inode
77  * @sb: The super block
78  * @inum: The inode number
79  * @type: The type of the inode
80  *
81  * Returns: A VFS inode, or an error
82  */
83 
84 struct inode *gfs2_inode_lookup(struct super_block *sb, struct gfs2_inum_host *inum, unsigned int type)
85 {
86 	struct inode *inode = gfs2_iget(sb, inum);
87 	struct gfs2_inode *ip = GFS2_I(inode);
88 	struct gfs2_glock *io_gl;
89 	int error;
90 
91 	if (!inode)
92 		return ERR_PTR(-ENOBUFS);
93 
94 	if (inode->i_state & I_NEW) {
95 		struct gfs2_sbd *sdp = GFS2_SB(inode);
96 		umode_t mode = DT2IF(type);
97 		inode->i_private = ip;
98 		inode->i_mode = mode;
99 
100 		if (S_ISREG(mode)) {
101 			inode->i_op = &gfs2_file_iops;
102 			inode->i_fop = &gfs2_file_fops;
103 			inode->i_mapping->a_ops = &gfs2_file_aops;
104 		} else if (S_ISDIR(mode)) {
105 			inode->i_op = &gfs2_dir_iops;
106 			inode->i_fop = &gfs2_dir_fops;
107 		} else if (S_ISLNK(mode)) {
108 			inode->i_op = &gfs2_symlink_iops;
109 		} else {
110 			inode->i_op = &gfs2_dev_iops;
111 		}
112 
113 		error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
114 		if (unlikely(error))
115 			goto fail;
116 		ip->i_gl->gl_object = ip;
117 
118 		error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
119 		if (unlikely(error))
120 			goto fail_put;
121 
122 		set_bit(GIF_INVALID, &ip->i_flags);
123 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
124 		if (unlikely(error))
125 			goto fail_iopen;
126 
127 		gfs2_glock_put(io_gl);
128 		unlock_new_inode(inode);
129 	}
130 
131 	return inode;
132 fail_iopen:
133 	gfs2_glock_put(io_gl);
134 fail_put:
135 	ip->i_gl->gl_object = NULL;
136 	gfs2_glock_put(ip->i_gl);
137 fail:
138 	iput(inode);
139 	return ERR_PTR(error);
140 }
141 
142 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
143 {
144 	struct gfs2_dinode_host *di = &ip->i_di;
145 	const struct gfs2_dinode *str = buf;
146 
147 	if (ip->i_num.no_addr != be64_to_cpu(str->di_num.no_addr)) {
148 		if (gfs2_consist_inode(ip))
149 			gfs2_dinode_print(ip);
150 		return -EIO;
151 	}
152 	if (ip->i_num.no_formal_ino != be64_to_cpu(str->di_num.no_formal_ino))
153 		return -ESTALE;
154 
155 	ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
156 	ip->i_inode.i_rdev = 0;
157 	switch (ip->i_inode.i_mode & S_IFMT) {
158 	case S_IFBLK:
159 	case S_IFCHR:
160 		ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
161 					   be32_to_cpu(str->di_minor));
162 		break;
163 	};
164 
165 	ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
166 	ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
167 	/*
168 	 * We will need to review setting the nlink count here in the
169 	 * light of the forthcoming ro bind mount work. This is a reminder
170 	 * to do that.
171 	 */
172 	ip->i_inode.i_nlink = be32_to_cpu(str->di_nlink);
173 	di->di_size = be64_to_cpu(str->di_size);
174 	i_size_write(&ip->i_inode, di->di_size);
175 	di->di_blocks = be64_to_cpu(str->di_blocks);
176 	gfs2_set_inode_blocks(&ip->i_inode);
177 	ip->i_inode.i_atime.tv_sec = be64_to_cpu(str->di_atime);
178 	ip->i_inode.i_atime.tv_nsec = 0;
179 	ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
180 	ip->i_inode.i_mtime.tv_nsec = 0;
181 	ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
182 	ip->i_inode.i_ctime.tv_nsec = 0;
183 
184 	di->di_goal_meta = be64_to_cpu(str->di_goal_meta);
185 	di->di_goal_data = be64_to_cpu(str->di_goal_data);
186 	di->di_generation = be64_to_cpu(str->di_generation);
187 
188 	di->di_flags = be32_to_cpu(str->di_flags);
189 	gfs2_set_inode_flags(&ip->i_inode);
190 	di->di_height = be16_to_cpu(str->di_height);
191 
192 	di->di_depth = be16_to_cpu(str->di_depth);
193 	di->di_entries = be32_to_cpu(str->di_entries);
194 
195 	di->di_eattr = be64_to_cpu(str->di_eattr);
196 	return 0;
197 }
198 
199 /**
200  * gfs2_inode_refresh - Refresh the incore copy of the dinode
201  * @ip: The GFS2 inode
202  *
203  * Returns: errno
204  */
205 
206 int gfs2_inode_refresh(struct gfs2_inode *ip)
207 {
208 	struct buffer_head *dibh;
209 	int error;
210 
211 	error = gfs2_meta_inode_buffer(ip, &dibh);
212 	if (error)
213 		return error;
214 
215 	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) {
216 		brelse(dibh);
217 		return -EIO;
218 	}
219 
220 	error = gfs2_dinode_in(ip, dibh->b_data);
221 	brelse(dibh);
222 	clear_bit(GIF_INVALID, &ip->i_flags);
223 
224 	return error;
225 }
226 
227 int gfs2_dinode_dealloc(struct gfs2_inode *ip)
228 {
229 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
230 	struct gfs2_alloc *al;
231 	struct gfs2_rgrpd *rgd;
232 	int error;
233 
234 	if (ip->i_di.di_blocks != 1) {
235 		if (gfs2_consist_inode(ip))
236 			gfs2_dinode_print(ip);
237 		return -EIO;
238 	}
239 
240 	al = gfs2_alloc_get(ip);
241 
242 	error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
243 	if (error)
244 		goto out;
245 
246 	error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
247 	if (error)
248 		goto out_qs;
249 
250 	rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
251 	if (!rgd) {
252 		gfs2_consist_inode(ip);
253 		error = -EIO;
254 		goto out_rindex_relse;
255 	}
256 
257 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
258 				   &al->al_rgd_gh);
259 	if (error)
260 		goto out_rindex_relse;
261 
262 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1);
263 	if (error)
264 		goto out_rg_gunlock;
265 
266 	gfs2_trans_add_gl(ip->i_gl);
267 
268 	gfs2_free_di(rgd, ip);
269 
270 	gfs2_trans_end(sdp);
271 	clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
272 
273 out_rg_gunlock:
274 	gfs2_glock_dq_uninit(&al->al_rgd_gh);
275 out_rindex_relse:
276 	gfs2_glock_dq_uninit(&al->al_ri_gh);
277 out_qs:
278 	gfs2_quota_unhold(ip);
279 out:
280 	gfs2_alloc_put(ip);
281 	return error;
282 }
283 
284 /**
285  * gfs2_change_nlink - Change nlink count on inode
286  * @ip: The GFS2 inode
287  * @diff: The change in the nlink count required
288  *
289  * Returns: errno
290  */
291 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
292 {
293 	struct buffer_head *dibh;
294 	u32 nlink;
295 	int error;
296 
297 	BUG_ON(diff != 1 && diff != -1);
298 	nlink = ip->i_inode.i_nlink + diff;
299 
300 	/* If we are reducing the nlink count, but the new value ends up being
301 	   bigger than the old one, we must have underflowed. */
302 	if (diff < 0 && nlink > ip->i_inode.i_nlink) {
303 		if (gfs2_consist_inode(ip))
304 			gfs2_dinode_print(ip);
305 		return -EIO;
306 	}
307 
308 	error = gfs2_meta_inode_buffer(ip, &dibh);
309 	if (error)
310 		return error;
311 
312 	if (diff > 0)
313 		inc_nlink(&ip->i_inode);
314 	else
315 		drop_nlink(&ip->i_inode);
316 
317 	ip->i_inode.i_ctime = CURRENT_TIME_SEC;
318 
319 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
320 	gfs2_dinode_out(ip, dibh->b_data);
321 	brelse(dibh);
322 	mark_inode_dirty(&ip->i_inode);
323 
324 	if (ip->i_inode.i_nlink == 0)
325 		gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */
326 
327 	return error;
328 }
329 
330 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
331 {
332 	struct qstr qstr;
333 	struct inode *inode;
334 	gfs2_str2qstr(&qstr, name);
335 	inode = gfs2_lookupi(dip, &qstr, 1, NULL);
336 	/* gfs2_lookupi has inconsistent callers: vfs
337 	 * related routines expect NULL for no entry found,
338 	 * gfs2_lookup_simple callers expect ENOENT
339 	 * and do not check for NULL.
340 	 */
341 	if (inode == NULL)
342 		return ERR_PTR(-ENOENT);
343 	else
344 		return inode;
345 }
346 
347 
348 /**
349  * gfs2_lookupi - Look up a filename in a directory and return its inode
350  * @d_gh: An initialized holder for the directory glock
351  * @name: The name of the inode to look for
352  * @is_root: If 1, ignore the caller's permissions
353  * @i_gh: An uninitialized holder for the new inode glock
354  *
355  * This can be called via the VFS filldir function when NFS is doing
356  * a readdirplus and the inode which its intending to stat isn't
357  * already in cache. In this case we must not take the directory glock
358  * again, since the readdir call will have already taken that lock.
359  *
360  * Returns: errno
361  */
362 
363 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
364 			   int is_root, struct nameidata *nd)
365 {
366 	struct super_block *sb = dir->i_sb;
367 	struct gfs2_inode *dip = GFS2_I(dir);
368 	struct gfs2_holder d_gh;
369 	struct gfs2_inum_host inum;
370 	unsigned int type;
371 	int error;
372 	struct inode *inode = NULL;
373 	int unlock = 0;
374 
375 	if (!name->len || name->len > GFS2_FNAMESIZE)
376 		return ERR_PTR(-ENAMETOOLONG);
377 
378 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
379 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
380 	     dir == sb->s_root->d_inode)) {
381 		igrab(dir);
382 		return dir;
383 	}
384 
385 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == 0) {
386 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
387 		if (error)
388 			return ERR_PTR(error);
389 		unlock = 1;
390 	}
391 
392 	if (!is_root) {
393 		error = permission(dir, MAY_EXEC, NULL);
394 		if (error)
395 			goto out;
396 	}
397 
398 	error = gfs2_dir_search(dir, name, &inum, &type);
399 	if (error)
400 		goto out;
401 
402 	inode = gfs2_inode_lookup(sb, &inum, type);
403 
404 out:
405 	if (unlock)
406 		gfs2_glock_dq_uninit(&d_gh);
407 	if (error == -ENOENT)
408 		return NULL;
409 	return inode ? inode : ERR_PTR(error);
410 }
411 
412 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino)
413 {
414 	struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode);
415 	struct buffer_head *bh;
416 	struct gfs2_inum_range_host ir;
417 	int error;
418 
419 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
420 	if (error)
421 		return error;
422 	mutex_lock(&sdp->sd_inum_mutex);
423 
424 	error = gfs2_meta_inode_buffer(ip, &bh);
425 	if (error) {
426 		mutex_unlock(&sdp->sd_inum_mutex);
427 		gfs2_trans_end(sdp);
428 		return error;
429 	}
430 
431 	gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
432 
433 	if (ir.ir_length) {
434 		*formal_ino = ir.ir_start++;
435 		ir.ir_length--;
436 		gfs2_trans_add_bh(ip->i_gl, bh, 1);
437 		gfs2_inum_range_out(&ir,
438 				    bh->b_data + sizeof(struct gfs2_dinode));
439 		brelse(bh);
440 		mutex_unlock(&sdp->sd_inum_mutex);
441 		gfs2_trans_end(sdp);
442 		return 0;
443 	}
444 
445 	brelse(bh);
446 
447 	mutex_unlock(&sdp->sd_inum_mutex);
448 	gfs2_trans_end(sdp);
449 
450 	return 1;
451 }
452 
453 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino)
454 {
455 	struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode);
456 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode);
457 	struct gfs2_holder gh;
458 	struct buffer_head *bh;
459 	struct gfs2_inum_range_host ir;
460 	int error;
461 
462 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
463 	if (error)
464 		return error;
465 
466 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
467 	if (error)
468 		goto out;
469 	mutex_lock(&sdp->sd_inum_mutex);
470 
471 	error = gfs2_meta_inode_buffer(ip, &bh);
472 	if (error)
473 		goto out_end_trans;
474 
475 	gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
476 
477 	if (!ir.ir_length) {
478 		struct buffer_head *m_bh;
479 		u64 x, y;
480 		__be64 z;
481 
482 		error = gfs2_meta_inode_buffer(m_ip, &m_bh);
483 		if (error)
484 			goto out_brelse;
485 
486 		z = *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode));
487 		x = y = be64_to_cpu(z);
488 		ir.ir_start = x;
489 		ir.ir_length = GFS2_INUM_QUANTUM;
490 		x += GFS2_INUM_QUANTUM;
491 		if (x < y)
492 			gfs2_consist_inode(m_ip);
493 		z = cpu_to_be64(x);
494 		gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
495 		*(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = z;
496 
497 		brelse(m_bh);
498 	}
499 
500 	*formal_ino = ir.ir_start++;
501 	ir.ir_length--;
502 
503 	gfs2_trans_add_bh(ip->i_gl, bh, 1);
504 	gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
505 
506 out_brelse:
507 	brelse(bh);
508 out_end_trans:
509 	mutex_unlock(&sdp->sd_inum_mutex);
510 	gfs2_trans_end(sdp);
511 out:
512 	gfs2_glock_dq_uninit(&gh);
513 	return error;
514 }
515 
516 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum)
517 {
518 	int error;
519 
520 	error = pick_formal_ino_1(sdp, inum);
521 	if (error <= 0)
522 		return error;
523 
524 	error = pick_formal_ino_2(sdp, inum);
525 
526 	return error;
527 }
528 
529 /**
530  * create_ok - OK to create a new on-disk inode here?
531  * @dip:  Directory in which dinode is to be created
532  * @name:  Name of new dinode
533  * @mode:
534  *
535  * Returns: errno
536  */
537 
538 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
539 		     unsigned int mode)
540 {
541 	int error;
542 
543 	error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL);
544 	if (error)
545 		return error;
546 
547 	/*  Don't create entries in an unlinked directory  */
548 	if (!dip->i_inode.i_nlink)
549 		return -EPERM;
550 
551 	error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL);
552 	switch (error) {
553 	case -ENOENT:
554 		error = 0;
555 		break;
556 	case 0:
557 		return -EEXIST;
558 	default:
559 		return error;
560 	}
561 
562 	if (dip->i_di.di_entries == (u32)-1)
563 		return -EFBIG;
564 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
565 		return -EMLINK;
566 
567 	return 0;
568 }
569 
570 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
571 			       unsigned int *uid, unsigned int *gid)
572 {
573 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
574 	    (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
575 		if (S_ISDIR(*mode))
576 			*mode |= S_ISUID;
577 		else if (dip->i_inode.i_uid != current->fsuid)
578 			*mode &= ~07111;
579 		*uid = dip->i_inode.i_uid;
580 	} else
581 		*uid = current->fsuid;
582 
583 	if (dip->i_inode.i_mode & S_ISGID) {
584 		if (S_ISDIR(*mode))
585 			*mode |= S_ISGID;
586 		*gid = dip->i_inode.i_gid;
587 	} else
588 		*gid = current->fsgid;
589 }
590 
591 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum_host *inum,
592 			u64 *generation)
593 {
594 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
595 	int error;
596 
597 	gfs2_alloc_get(dip);
598 
599 	dip->i_alloc.al_requested = RES_DINODE;
600 	error = gfs2_inplace_reserve(dip);
601 	if (error)
602 		goto out;
603 
604 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
605 	if (error)
606 		goto out_ipreserv;
607 
608 	inum->no_addr = gfs2_alloc_di(dip, generation);
609 
610 	gfs2_trans_end(sdp);
611 
612 out_ipreserv:
613 	gfs2_inplace_release(dip);
614 out:
615 	gfs2_alloc_put(dip);
616 	return error;
617 }
618 
619 /**
620  * init_dinode - Fill in a new dinode structure
621  * @dip: the directory this inode is being created in
622  * @gl: The glock covering the new inode
623  * @inum: the inode number
624  * @mode: the file permissions
625  * @uid:
626  * @gid:
627  *
628  */
629 
630 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
631 			const struct gfs2_inum_host *inum, unsigned int mode,
632 			unsigned int uid, unsigned int gid,
633 			const u64 *generation, dev_t dev)
634 {
635 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
636 	struct gfs2_dinode *di;
637 	struct buffer_head *dibh;
638 
639 	dibh = gfs2_meta_new(gl, inum->no_addr);
640 	gfs2_trans_add_bh(gl, dibh, 1);
641 	gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
642 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
643 	di = (struct gfs2_dinode *)dibh->b_data;
644 
645 	di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
646 	di->di_num.no_addr = cpu_to_be64(inum->no_addr);
647 	di->di_mode = cpu_to_be32(mode);
648 	di->di_uid = cpu_to_be32(uid);
649 	di->di_gid = cpu_to_be32(gid);
650 	di->di_nlink = 0;
651 	di->di_size = 0;
652 	di->di_blocks = cpu_to_be64(1);
653 	di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
654 	di->di_major = cpu_to_be32(MAJOR(dev));
655 	di->di_minor = cpu_to_be32(MINOR(dev));
656 	di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
657 	di->di_generation = cpu_to_be64(*generation);
658 	di->di_flags = 0;
659 
660 	if (S_ISREG(mode)) {
661 		if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
662 		    gfs2_tune_get(sdp, gt_new_files_jdata))
663 			di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
664 		if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
665 		    gfs2_tune_get(sdp, gt_new_files_directio))
666 			di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
667 	} else if (S_ISDIR(mode)) {
668 		di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
669 					    GFS2_DIF_INHERIT_DIRECTIO);
670 		di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
671 					    GFS2_DIF_INHERIT_JDATA);
672 	}
673 
674 	di->__pad1 = 0;
675 	di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
676 	di->di_height = 0;
677 	di->__pad2 = 0;
678 	di->__pad3 = 0;
679 	di->di_depth = 0;
680 	di->di_entries = 0;
681 	memset(&di->__pad4, 0, sizeof(di->__pad4));
682 	di->di_eattr = 0;
683 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
684 
685 	brelse(dibh);
686 }
687 
688 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
689 		       unsigned int mode, const struct gfs2_inum_host *inum,
690 		       const u64 *generation, dev_t dev)
691 {
692 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
693 	unsigned int uid, gid;
694 	int error;
695 
696 	munge_mode_uid_gid(dip, &mode, &uid, &gid);
697 	gfs2_alloc_get(dip);
698 
699 	error = gfs2_quota_lock(dip, uid, gid);
700 	if (error)
701 		goto out;
702 
703 	error = gfs2_quota_check(dip, uid, gid);
704 	if (error)
705 		goto out_quota;
706 
707 	error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
708 	if (error)
709 		goto out_quota;
710 
711 	init_dinode(dip, gl, inum, mode, uid, gid, generation, dev);
712 	gfs2_quota_change(dip, +1, uid, gid);
713 	gfs2_trans_end(sdp);
714 
715 out_quota:
716 	gfs2_quota_unlock(dip);
717 out:
718 	gfs2_alloc_put(dip);
719 	return error;
720 }
721 
722 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
723 		       struct gfs2_inode *ip)
724 {
725 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
726 	struct gfs2_alloc *al;
727 	int alloc_required;
728 	struct buffer_head *dibh;
729 	int error;
730 
731 	al = gfs2_alloc_get(dip);
732 
733 	error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
734 	if (error)
735 		goto fail;
736 
737 	error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
738 	if (alloc_required < 0)
739 		goto fail;
740 	if (alloc_required) {
741 		error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
742 		if (error)
743 			goto fail_quota_locks;
744 
745 		al->al_requested = sdp->sd_max_dirres;
746 
747 		error = gfs2_inplace_reserve(dip);
748 		if (error)
749 			goto fail_quota_locks;
750 
751 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
752 					 al->al_rgd->rd_ri.ri_length +
753 					 2 * RES_DINODE +
754 					 RES_STATFS + RES_QUOTA, 0);
755 		if (error)
756 			goto fail_ipreserv;
757 	} else {
758 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
759 		if (error)
760 			goto fail_quota_locks;
761 	}
762 
763 	error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_inode.i_mode));
764 	if (error)
765 		goto fail_end_trans;
766 
767 	error = gfs2_meta_inode_buffer(ip, &dibh);
768 	if (error)
769 		goto fail_end_trans;
770 	ip->i_inode.i_nlink = 1;
771 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
772 	gfs2_dinode_out(ip, dibh->b_data);
773 	brelse(dibh);
774 	return 0;
775 
776 fail_end_trans:
777 	gfs2_trans_end(sdp);
778 
779 fail_ipreserv:
780 	if (dip->i_alloc.al_rgd)
781 		gfs2_inplace_release(dip);
782 
783 fail_quota_locks:
784 	gfs2_quota_unlock(dip);
785 
786 fail:
787 	gfs2_alloc_put(dip);
788 	return error;
789 }
790 
791 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip)
792 {
793 	int err;
794 	size_t len;
795 	void *value;
796 	char *name;
797 	struct gfs2_ea_request er;
798 
799 	err = security_inode_init_security(&ip->i_inode, &dip->i_inode,
800 					   &name, &value, &len);
801 
802 	if (err) {
803 		if (err == -EOPNOTSUPP)
804 			return 0;
805 		return err;
806 	}
807 
808 	memset(&er, 0, sizeof(struct gfs2_ea_request));
809 
810 	er.er_type = GFS2_EATYPE_SECURITY;
811 	er.er_name = name;
812 	er.er_data = value;
813 	er.er_name_len = strlen(name);
814 	er.er_data_len = len;
815 
816 	err = gfs2_ea_set_i(ip, &er);
817 
818 	kfree(value);
819 	kfree(name);
820 
821 	return err;
822 }
823 
824 /**
825  * gfs2_createi - Create a new inode
826  * @ghs: An array of two holders
827  * @name: The name of the new file
828  * @mode: the permissions on the new inode
829  *
830  * @ghs[0] is an initialized holder for the directory
831  * @ghs[1] is the holder for the inode lock
832  *
833  * If the return value is not NULL, the glocks on both the directory and the new
834  * file are held.  A transaction has been started and an inplace reservation
835  * is held, as well.
836  *
837  * Returns: An inode
838  */
839 
840 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name,
841 			   unsigned int mode, dev_t dev)
842 {
843 	struct inode *inode;
844 	struct gfs2_inode *dip = ghs->gh_gl->gl_object;
845 	struct inode *dir = &dip->i_inode;
846 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
847 	struct gfs2_inum_host inum;
848 	int error;
849 	u64 generation;
850 
851 	if (!name->len || name->len > GFS2_FNAMESIZE)
852 		return ERR_PTR(-ENAMETOOLONG);
853 
854 	gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
855 	error = gfs2_glock_nq(ghs);
856 	if (error)
857 		goto fail;
858 
859 	error = create_ok(dip, name, mode);
860 	if (error)
861 		goto fail_gunlock;
862 
863 	error = pick_formal_ino(sdp, &inum.no_formal_ino);
864 	if (error)
865 		goto fail_gunlock;
866 
867 	error = alloc_dinode(dip, &inum, &generation);
868 	if (error)
869 		goto fail_gunlock;
870 
871 	error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
872 				  LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
873 	if (error)
874 		goto fail_gunlock;
875 
876 	error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev);
877 	if (error)
878 		goto fail_gunlock2;
879 
880 	inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode));
881 	if (IS_ERR(inode))
882 		goto fail_gunlock2;
883 
884 	error = gfs2_inode_refresh(GFS2_I(inode));
885 	if (error)
886 		goto fail_iput;
887 
888 	error = gfs2_acl_create(dip, GFS2_I(inode));
889 	if (error)
890 		goto fail_iput;
891 
892 	error = gfs2_security_init(dip, GFS2_I(inode));
893 	if (error)
894 		goto fail_iput;
895 
896 	error = link_dinode(dip, name, GFS2_I(inode));
897 	if (error)
898 		goto fail_iput;
899 
900 	if (!inode)
901 		return ERR_PTR(-ENOMEM);
902 	return inode;
903 
904 fail_iput:
905 	iput(inode);
906 fail_gunlock2:
907 	gfs2_glock_dq_uninit(ghs + 1);
908 fail_gunlock:
909 	gfs2_glock_dq(ghs);
910 fail:
911 	return ERR_PTR(error);
912 }
913 
914 /**
915  * gfs2_rmdiri - Remove a directory
916  * @dip: The parent directory of the directory to be removed
917  * @name: The name of the directory to be removed
918  * @ip: The GFS2 inode of the directory to be removed
919  *
920  * Assumes Glocks on dip and ip are held
921  *
922  * Returns: errno
923  */
924 
925 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name,
926 		struct gfs2_inode *ip)
927 {
928 	struct qstr dotname;
929 	int error;
930 
931 	if (ip->i_di.di_entries != 2) {
932 		if (gfs2_consist_inode(ip))
933 			gfs2_dinode_print(ip);
934 		return -EIO;
935 	}
936 
937 	error = gfs2_dir_del(dip, name);
938 	if (error)
939 		return error;
940 
941 	error = gfs2_change_nlink(dip, -1);
942 	if (error)
943 		return error;
944 
945 	gfs2_str2qstr(&dotname, ".");
946 	error = gfs2_dir_del(ip, &dotname);
947 	if (error)
948 		return error;
949 
950 	gfs2_str2qstr(&dotname, "..");
951 	error = gfs2_dir_del(ip, &dotname);
952 	if (error)
953 		return error;
954 
955 	/* It looks odd, but it really should be done twice */
956 	error = gfs2_change_nlink(ip, -1);
957 	if (error)
958 		return error;
959 
960 	error = gfs2_change_nlink(ip, -1);
961 	if (error)
962 		return error;
963 
964 	return error;
965 }
966 
967 /*
968  * gfs2_unlink_ok - check to see that a inode is still in a directory
969  * @dip: the directory
970  * @name: the name of the file
971  * @ip: the inode
972  *
973  * Assumes that the lock on (at least) @dip is held.
974  *
975  * Returns: 0 if the parent/child relationship is correct, errno if it isn't
976  */
977 
978 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
979 		   struct gfs2_inode *ip)
980 {
981 	struct gfs2_inum_host inum;
982 	unsigned int type;
983 	int error;
984 
985 	if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
986 		return -EPERM;
987 
988 	if ((dip->i_inode.i_mode & S_ISVTX) &&
989 	    dip->i_inode.i_uid != current->fsuid &&
990 	    ip->i_inode.i_uid != current->fsuid && !capable(CAP_FOWNER))
991 		return -EPERM;
992 
993 	if (IS_APPEND(&dip->i_inode))
994 		return -EPERM;
995 
996 	error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL);
997 	if (error)
998 		return error;
999 
1000 	error = gfs2_dir_search(&dip->i_inode, name, &inum, &type);
1001 	if (error)
1002 		return error;
1003 
1004 	if (!gfs2_inum_equal(&inum, &ip->i_num))
1005 		return -ENOENT;
1006 
1007 	if (IF2DT(ip->i_inode.i_mode) != type) {
1008 		gfs2_consist_inode(dip);
1009 		return -EIO;
1010 	}
1011 
1012 	return 0;
1013 }
1014 
1015 /*
1016  * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1017  * @this: move this
1018  * @to: to here
1019  *
1020  * Follow @to back to the root and make sure we don't encounter @this
1021  * Assumes we already hold the rename lock.
1022  *
1023  * Returns: errno
1024  */
1025 
1026 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1027 {
1028 	struct inode *dir = &to->i_inode;
1029 	struct super_block *sb = dir->i_sb;
1030 	struct inode *tmp;
1031 	struct qstr dotdot;
1032 	int error = 0;
1033 
1034 	gfs2_str2qstr(&dotdot, "..");
1035 
1036 	igrab(dir);
1037 
1038 	for (;;) {
1039 		if (dir == &this->i_inode) {
1040 			error = -EINVAL;
1041 			break;
1042 		}
1043 		if (dir == sb->s_root->d_inode) {
1044 			error = 0;
1045 			break;
1046 		}
1047 
1048 		tmp = gfs2_lookupi(dir, &dotdot, 1, NULL);
1049 		if (IS_ERR(tmp)) {
1050 			error = PTR_ERR(tmp);
1051 			break;
1052 		}
1053 
1054 		iput(dir);
1055 		dir = tmp;
1056 	}
1057 
1058 	iput(dir);
1059 
1060 	return error;
1061 }
1062 
1063 /**
1064  * gfs2_readlinki - return the contents of a symlink
1065  * @ip: the symlink's inode
1066  * @buf: a pointer to the buffer to be filled
1067  * @len: a pointer to the length of @buf
1068  *
1069  * If @buf is too small, a piece of memory is kmalloc()ed and needs
1070  * to be freed by the caller.
1071  *
1072  * Returns: errno
1073  */
1074 
1075 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1076 {
1077 	struct gfs2_holder i_gh;
1078 	struct buffer_head *dibh;
1079 	unsigned int x;
1080 	int error;
1081 
1082 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1083 	error = gfs2_glock_nq_atime(&i_gh);
1084 	if (error) {
1085 		gfs2_holder_uninit(&i_gh);
1086 		return error;
1087 	}
1088 
1089 	if (!ip->i_di.di_size) {
1090 		gfs2_consist_inode(ip);
1091 		error = -EIO;
1092 		goto out;
1093 	}
1094 
1095 	error = gfs2_meta_inode_buffer(ip, &dibh);
1096 	if (error)
1097 		goto out;
1098 
1099 	x = ip->i_di.di_size + 1;
1100 	if (x > *len) {
1101 		*buf = kmalloc(x, GFP_KERNEL);
1102 		if (!*buf) {
1103 			error = -ENOMEM;
1104 			goto out_brelse;
1105 		}
1106 	}
1107 
1108 	memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1109 	*len = x;
1110 
1111 out_brelse:
1112 	brelse(dibh);
1113 out:
1114 	gfs2_glock_dq_uninit(&i_gh);
1115 	return error;
1116 }
1117 
1118 /**
1119  * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1120  *       conditionally update the inode's atime
1121  * @gh: the holder to acquire
1122  *
1123  * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1124  * Update if the difference between the current time and the inode's current
1125  * atime is greater than an interval specified at mount.
1126  *
1127  * Returns: errno
1128  */
1129 
1130 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1131 {
1132 	struct gfs2_glock *gl = gh->gh_gl;
1133 	struct gfs2_sbd *sdp = gl->gl_sbd;
1134 	struct gfs2_inode *ip = gl->gl_object;
1135 	s64 curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1136 	unsigned int state;
1137 	int flags;
1138 	int error;
1139 
1140 	if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1141 	    gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1142 	    gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1143 		return -EINVAL;
1144 
1145 	state = gh->gh_state;
1146 	flags = gh->gh_flags;
1147 
1148 	error = gfs2_glock_nq(gh);
1149 	if (error)
1150 		return error;
1151 
1152 	if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1153 	    (sdp->sd_vfs->s_flags & MS_RDONLY))
1154 		return 0;
1155 
1156 	curtime = get_seconds();
1157 	if (curtime - ip->i_inode.i_atime.tv_sec >= quantum) {
1158 		gfs2_glock_dq(gh);
1159 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY,
1160 				   gh);
1161 		error = gfs2_glock_nq(gh);
1162 		if (error)
1163 			return error;
1164 
1165 		/* Verify that atime hasn't been updated while we were
1166 		   trying to get exclusive lock. */
1167 
1168 		curtime = get_seconds();
1169 		if (curtime - ip->i_inode.i_atime.tv_sec >= quantum) {
1170 			struct buffer_head *dibh;
1171 			struct gfs2_dinode *di;
1172 
1173 			error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1174 			if (error == -EROFS)
1175 				return 0;
1176 			if (error)
1177 				goto fail;
1178 
1179 			error = gfs2_meta_inode_buffer(ip, &dibh);
1180 			if (error)
1181 				goto fail_end_trans;
1182 
1183 			ip->i_inode.i_atime.tv_sec = curtime;
1184 
1185 			gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1186 			di = (struct gfs2_dinode *)dibh->b_data;
1187 			di->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
1188 			brelse(dibh);
1189 
1190 			gfs2_trans_end(sdp);
1191 		}
1192 
1193 		/* If someone else has asked for the glock,
1194 		   unlock and let them have it. Then reacquire
1195 		   in the original state. */
1196 		if (gfs2_glock_is_blocking(gl)) {
1197 			gfs2_glock_dq(gh);
1198 			gfs2_holder_reinit(state, flags, gh);
1199 			return gfs2_glock_nq(gh);
1200 		}
1201 	}
1202 
1203 	return 0;
1204 
1205 fail_end_trans:
1206 	gfs2_trans_end(sdp);
1207 fail:
1208 	gfs2_glock_dq(gh);
1209 	return error;
1210 }
1211 
1212 static int
1213 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1214 {
1215 	struct buffer_head *dibh;
1216 	int error;
1217 
1218 	error = gfs2_meta_inode_buffer(ip, &dibh);
1219 	if (!error) {
1220 		error = inode_setattr(&ip->i_inode, attr);
1221 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1222 		gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1223 		gfs2_dinode_out(ip, dibh->b_data);
1224 		brelse(dibh);
1225 	}
1226 	return error;
1227 }
1228 
1229 /**
1230  * gfs2_setattr_simple -
1231  * @ip:
1232  * @attr:
1233  *
1234  * Called with a reference on the vnode.
1235  *
1236  * Returns: errno
1237  */
1238 
1239 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1240 {
1241 	int error;
1242 
1243 	if (current->journal_info)
1244 		return __gfs2_setattr_simple(ip, attr);
1245 
1246 	error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0);
1247 	if (error)
1248 		return error;
1249 
1250 	error = __gfs2_setattr_simple(ip, attr);
1251 	gfs2_trans_end(GFS2_SB(&ip->i_inode));
1252 	return error;
1253 }
1254 
1255