xref: /linux/fs/gfs2/inode.c (revision 394d83c17fac2b7bcf05cb99d1e945135767bb6b)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 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/security.h>
20 #include <linux/time.h>
21 
22 #include "gfs2.h"
23 #include "incore.h"
24 #include "acl.h"
25 #include "bmap.h"
26 #include "dir.h"
27 #include "xattr.h"
28 #include "glock.h"
29 #include "glops.h"
30 #include "inode.h"
31 #include "log.h"
32 #include "meta_io.h"
33 #include "quota.h"
34 #include "rgrp.h"
35 #include "trans.h"
36 #include "util.h"
37 
38 struct gfs2_inum_range_host {
39 	u64 ir_start;
40 	u64 ir_length;
41 };
42 
43 static int iget_test(struct inode *inode, void *opaque)
44 {
45 	struct gfs2_inode *ip = GFS2_I(inode);
46 	u64 *no_addr = opaque;
47 
48 	if (ip->i_no_addr == *no_addr)
49 		return 1;
50 
51 	return 0;
52 }
53 
54 static int iget_set(struct inode *inode, void *opaque)
55 {
56 	struct gfs2_inode *ip = GFS2_I(inode);
57 	u64 *no_addr = opaque;
58 
59 	inode->i_ino = (unsigned long)*no_addr;
60 	ip->i_no_addr = *no_addr;
61 	return 0;
62 }
63 
64 struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr)
65 {
66 	unsigned long hash = (unsigned long)no_addr;
67 	return ilookup5(sb, hash, iget_test, &no_addr);
68 }
69 
70 static struct inode *gfs2_iget(struct super_block *sb, u64 no_addr)
71 {
72 	unsigned long hash = (unsigned long)no_addr;
73 	return iget5_locked(sb, hash, iget_test, iget_set, &no_addr);
74 }
75 
76 /**
77  * GFS2 lookup code fills in vfs inode contents based on info obtained
78  * from directory entry inside gfs2_inode_lookup(). This has caused issues
79  * with NFS code path since its get_dentry routine doesn't have the relevant
80  * directory entry when gfs2_inode_lookup() is invoked. Part of the code
81  * segment inside gfs2_inode_lookup code needs to get moved around.
82  *
83  * Clears I_NEW as well.
84  **/
85 
86 void gfs2_set_iop(struct inode *inode)
87 {
88 	struct gfs2_sbd *sdp = GFS2_SB(inode);
89 	umode_t mode = inode->i_mode;
90 
91 	if (S_ISREG(mode)) {
92 		inode->i_op = &gfs2_file_iops;
93 		if (gfs2_localflocks(sdp))
94 			inode->i_fop = &gfs2_file_fops_nolock;
95 		else
96 			inode->i_fop = &gfs2_file_fops;
97 	} else if (S_ISDIR(mode)) {
98 		inode->i_op = &gfs2_dir_iops;
99 		if (gfs2_localflocks(sdp))
100 			inode->i_fop = &gfs2_dir_fops_nolock;
101 		else
102 			inode->i_fop = &gfs2_dir_fops;
103 	} else if (S_ISLNK(mode)) {
104 		inode->i_op = &gfs2_symlink_iops;
105 	} else {
106 		inode->i_op = &gfs2_file_iops;
107 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
108 	}
109 
110 	unlock_new_inode(inode);
111 }
112 
113 /**
114  * gfs2_inode_lookup - Lookup an inode
115  * @sb: The super block
116  * @no_addr: The inode number
117  * @type: The type of the inode
118  *
119  * Returns: A VFS inode, or an error
120  */
121 
122 struct inode *gfs2_inode_lookup(struct super_block *sb,
123 				unsigned int type,
124 				u64 no_addr,
125 				u64 no_formal_ino)
126 {
127 	struct inode *inode;
128 	struct gfs2_inode *ip;
129 	struct gfs2_glock *io_gl = NULL;
130 	int error;
131 
132 	inode = gfs2_iget(sb, no_addr);
133 	ip = GFS2_I(inode);
134 
135 	if (!inode)
136 		return ERR_PTR(-ENOBUFS);
137 
138 	if (inode->i_state & I_NEW) {
139 		struct gfs2_sbd *sdp = GFS2_SB(inode);
140 		ip->i_no_formal_ino = no_formal_ino;
141 
142 		error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
143 		if (unlikely(error))
144 			goto fail;
145 		ip->i_gl->gl_object = ip;
146 
147 		error = gfs2_glock_get(sdp, no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
148 		if (unlikely(error))
149 			goto fail_put;
150 
151 		set_bit(GIF_INVALID, &ip->i_flags);
152 		error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
153 		if (unlikely(error))
154 			goto fail_iopen;
155 		ip->i_iopen_gh.gh_gl->gl_object = ip;
156 
157 		gfs2_glock_put(io_gl);
158 		io_gl = NULL;
159 
160 		if ((type == DT_UNKNOWN) && (no_formal_ino == 0))
161 			goto gfs2_nfsbypass;
162 
163 		inode->i_mode = DT2IF(type);
164 
165 		/*
166 		 * We must read the inode in order to work out its type in
167 		 * this case. Note that this doesn't happen often as we normally
168 		 * know the type beforehand. This code path only occurs during
169 		 * unlinked inode recovery (where it is safe to do this glock,
170 		 * which is not true in the general case).
171 		 */
172 		if (type == DT_UNKNOWN) {
173 			struct gfs2_holder gh;
174 			error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
175 			if (unlikely(error))
176 				goto fail_glock;
177 			/* Inode is now uptodate */
178 			gfs2_glock_dq_uninit(&gh);
179 		}
180 
181 		gfs2_set_iop(inode);
182 	}
183 
184 gfs2_nfsbypass:
185 	return inode;
186 fail_glock:
187 	gfs2_glock_dq(&ip->i_iopen_gh);
188 fail_iopen:
189 	if (io_gl)
190 		gfs2_glock_put(io_gl);
191 fail_put:
192 	if (inode->i_state & I_NEW)
193 		ip->i_gl->gl_object = NULL;
194 	gfs2_glock_put(ip->i_gl);
195 fail:
196 	if (inode->i_state & I_NEW)
197 		iget_failed(inode);
198 	else
199 		iput(inode);
200 	return ERR_PTR(error);
201 }
202 
203 struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr,
204 				  u64 *no_formal_ino, unsigned int blktype)
205 {
206 	struct super_block *sb = sdp->sd_vfs;
207 	struct gfs2_holder i_gh;
208 	struct inode *inode;
209 	int error;
210 
211 	error = gfs2_glock_nq_num(sdp, no_addr, &gfs2_inode_glops,
212 				  LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
213 	if (error)
214 		return ERR_PTR(error);
215 
216 	error = gfs2_check_blk_type(sdp, no_addr, blktype);
217 	if (error)
218 		goto fail;
219 
220 	inode = gfs2_inode_lookup(sb, DT_UNKNOWN, no_addr, 0);
221 	if (IS_ERR(inode))
222 		goto fail;
223 
224 	error = gfs2_inode_refresh(GFS2_I(inode));
225 	if (error)
226 		goto fail_iput;
227 
228 	/* Pick up the works we bypass in gfs2_inode_lookup */
229 	if (inode->i_state & I_NEW)
230 		gfs2_set_iop(inode);
231 
232 	/* Two extra checks for NFS only */
233 	if (no_formal_ino) {
234 		error = -ESTALE;
235 		if (GFS2_I(inode)->i_no_formal_ino != *no_formal_ino)
236 			goto fail_iput;
237 
238 		error = -EIO;
239 		if (GFS2_I(inode)->i_diskflags & GFS2_DIF_SYSTEM)
240 			goto fail_iput;
241 
242 		error = 0;
243 	}
244 
245 fail:
246 	gfs2_glock_dq_uninit(&i_gh);
247 	return error ? ERR_PTR(error) : inode;
248 fail_iput:
249 	iput(inode);
250 	goto fail;
251 }
252 
253 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
254 {
255 	const struct gfs2_dinode *str = buf;
256 	struct timespec atime;
257 	u16 height, depth;
258 
259 	if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
260 		goto corrupt;
261 	ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
262 	ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
263 	ip->i_inode.i_rdev = 0;
264 	switch (ip->i_inode.i_mode & S_IFMT) {
265 	case S_IFBLK:
266 	case S_IFCHR:
267 		ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
268 					   be32_to_cpu(str->di_minor));
269 		break;
270 	};
271 
272 	ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
273 	ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
274 	/*
275 	 * We will need to review setting the nlink count here in the
276 	 * light of the forthcoming ro bind mount work. This is a reminder
277 	 * to do that.
278 	 */
279 	ip->i_inode.i_nlink = be32_to_cpu(str->di_nlink);
280 	i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
281 	gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
282 	atime.tv_sec = be64_to_cpu(str->di_atime);
283 	atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
284 	if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
285 		ip->i_inode.i_atime = atime;
286 	ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
287 	ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
288 	ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
289 	ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
290 
291 	ip->i_goal = be64_to_cpu(str->di_goal_meta);
292 	ip->i_generation = be64_to_cpu(str->di_generation);
293 
294 	ip->i_diskflags = be32_to_cpu(str->di_flags);
295 	gfs2_set_inode_flags(&ip->i_inode);
296 	height = be16_to_cpu(str->di_height);
297 	if (unlikely(height > GFS2_MAX_META_HEIGHT))
298 		goto corrupt;
299 	ip->i_height = (u8)height;
300 
301 	depth = be16_to_cpu(str->di_depth);
302 	if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
303 		goto corrupt;
304 	ip->i_depth = (u8)depth;
305 	ip->i_entries = be32_to_cpu(str->di_entries);
306 
307 	ip->i_eattr = be64_to_cpu(str->di_eattr);
308 	if (S_ISREG(ip->i_inode.i_mode))
309 		gfs2_set_aops(&ip->i_inode);
310 
311 	return 0;
312 corrupt:
313 	if (gfs2_consist_inode(ip))
314 		gfs2_dinode_print(ip);
315 	return -EIO;
316 }
317 
318 /**
319  * gfs2_inode_refresh - Refresh the incore copy of the dinode
320  * @ip: The GFS2 inode
321  *
322  * Returns: errno
323  */
324 
325 int gfs2_inode_refresh(struct gfs2_inode *ip)
326 {
327 	struct buffer_head *dibh;
328 	int error;
329 
330 	error = gfs2_meta_inode_buffer(ip, &dibh);
331 	if (error)
332 		return error;
333 
334 	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) {
335 		brelse(dibh);
336 		return -EIO;
337 	}
338 
339 	error = gfs2_dinode_in(ip, dibh->b_data);
340 	brelse(dibh);
341 	clear_bit(GIF_INVALID, &ip->i_flags);
342 
343 	return error;
344 }
345 
346 int gfs2_dinode_dealloc(struct gfs2_inode *ip)
347 {
348 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
349 	struct gfs2_alloc *al;
350 	struct gfs2_rgrpd *rgd;
351 	int error;
352 
353 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
354 		if (gfs2_consist_inode(ip))
355 			gfs2_dinode_print(ip);
356 		return -EIO;
357 	}
358 
359 	al = gfs2_alloc_get(ip);
360 	if (!al)
361 		return -ENOMEM;
362 
363 	error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
364 	if (error)
365 		goto out;
366 
367 	error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
368 	if (error)
369 		goto out_qs;
370 
371 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
372 	if (!rgd) {
373 		gfs2_consist_inode(ip);
374 		error = -EIO;
375 		goto out_rindex_relse;
376 	}
377 
378 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
379 				   &al->al_rgd_gh);
380 	if (error)
381 		goto out_rindex_relse;
382 
383 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1);
384 	if (error)
385 		goto out_rg_gunlock;
386 
387 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
388 	set_bit(GLF_LFLUSH, &ip->i_gl->gl_flags);
389 
390 	gfs2_free_di(rgd, ip);
391 
392 	gfs2_trans_end(sdp);
393 
394 out_rg_gunlock:
395 	gfs2_glock_dq_uninit(&al->al_rgd_gh);
396 out_rindex_relse:
397 	gfs2_glock_dq_uninit(&al->al_ri_gh);
398 out_qs:
399 	gfs2_quota_unhold(ip);
400 out:
401 	gfs2_alloc_put(ip);
402 	return error;
403 }
404 
405 /**
406  * gfs2_change_nlink - Change nlink count on inode
407  * @ip: The GFS2 inode
408  * @diff: The change in the nlink count required
409  *
410  * Returns: errno
411  */
412 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
413 {
414 	struct buffer_head *dibh;
415 	u32 nlink;
416 	int error;
417 
418 	BUG_ON(diff != 1 && diff != -1);
419 	nlink = ip->i_inode.i_nlink + diff;
420 
421 	/* If we are reducing the nlink count, but the new value ends up being
422 	   bigger than the old one, we must have underflowed. */
423 	if (diff < 0 && nlink > ip->i_inode.i_nlink) {
424 		if (gfs2_consist_inode(ip))
425 			gfs2_dinode_print(ip);
426 		return -EIO;
427 	}
428 
429 	error = gfs2_meta_inode_buffer(ip, &dibh);
430 	if (error)
431 		return error;
432 
433 	if (diff > 0)
434 		inc_nlink(&ip->i_inode);
435 	else
436 		drop_nlink(&ip->i_inode);
437 
438 	ip->i_inode.i_ctime = CURRENT_TIME;
439 
440 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
441 	gfs2_dinode_out(ip, dibh->b_data);
442 	brelse(dibh);
443 	mark_inode_dirty(&ip->i_inode);
444 
445 	if (ip->i_inode.i_nlink == 0)
446 		gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */
447 
448 	return error;
449 }
450 
451 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
452 {
453 	struct qstr qstr;
454 	struct inode *inode;
455 	gfs2_str2qstr(&qstr, name);
456 	inode = gfs2_lookupi(dip, &qstr, 1);
457 	/* gfs2_lookupi has inconsistent callers: vfs
458 	 * related routines expect NULL for no entry found,
459 	 * gfs2_lookup_simple callers expect ENOENT
460 	 * and do not check for NULL.
461 	 */
462 	if (inode == NULL)
463 		return ERR_PTR(-ENOENT);
464 	else
465 		return inode;
466 }
467 
468 
469 /**
470  * gfs2_lookupi - Look up a filename in a directory and return its inode
471  * @d_gh: An initialized holder for the directory glock
472  * @name: The name of the inode to look for
473  * @is_root: If 1, ignore the caller's permissions
474  * @i_gh: An uninitialized holder for the new inode glock
475  *
476  * This can be called via the VFS filldir function when NFS is doing
477  * a readdirplus and the inode which its intending to stat isn't
478  * already in cache. In this case we must not take the directory glock
479  * again, since the readdir call will have already taken that lock.
480  *
481  * Returns: errno
482  */
483 
484 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
485 			   int is_root)
486 {
487 	struct super_block *sb = dir->i_sb;
488 	struct gfs2_inode *dip = GFS2_I(dir);
489 	struct gfs2_holder d_gh;
490 	int error = 0;
491 	struct inode *inode = NULL;
492 	int unlock = 0;
493 
494 	if (!name->len || name->len > GFS2_FNAMESIZE)
495 		return ERR_PTR(-ENAMETOOLONG);
496 
497 	if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
498 	    (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
499 	     dir == sb->s_root->d_inode)) {
500 		igrab(dir);
501 		return dir;
502 	}
503 
504 	if (gfs2_glock_is_locked_by_me(dip->i_gl) == NULL) {
505 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
506 		if (error)
507 			return ERR_PTR(error);
508 		unlock = 1;
509 	}
510 
511 	if (!is_root) {
512 		error = gfs2_permission(dir, MAY_EXEC, 0);
513 		if (error)
514 			goto out;
515 	}
516 
517 	inode = gfs2_dir_search(dir, name);
518 	if (IS_ERR(inode))
519 		error = PTR_ERR(inode);
520 out:
521 	if (unlock)
522 		gfs2_glock_dq_uninit(&d_gh);
523 	if (error == -ENOENT)
524 		return NULL;
525 	return inode ? inode : ERR_PTR(error);
526 }
527 
528 /**
529  * create_ok - OK to create a new on-disk inode here?
530  * @dip:  Directory in which dinode is to be created
531  * @name:  Name of new dinode
532  * @mode:
533  *
534  * Returns: errno
535  */
536 
537 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
538 		     unsigned int mode)
539 {
540 	int error;
541 
542 	error = gfs2_permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, 0);
543 	if (error)
544 		return error;
545 
546 	/*  Don't create entries in an unlinked directory  */
547 	if (!dip->i_inode.i_nlink)
548 		return -EPERM;
549 
550 	error = gfs2_dir_check(&dip->i_inode, name, NULL);
551 	switch (error) {
552 	case -ENOENT:
553 		error = 0;
554 		break;
555 	case 0:
556 		return -EEXIST;
557 	default:
558 		return error;
559 	}
560 
561 	if (dip->i_entries == (u32)-1)
562 		return -EFBIG;
563 	if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
564 		return -EMLINK;
565 
566 	return 0;
567 }
568 
569 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
570 			       unsigned int *uid, unsigned int *gid)
571 {
572 	if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
573 	    (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
574 		if (S_ISDIR(*mode))
575 			*mode |= S_ISUID;
576 		else if (dip->i_inode.i_uid != current_fsuid())
577 			*mode &= ~07111;
578 		*uid = dip->i_inode.i_uid;
579 	} else
580 		*uid = current_fsuid();
581 
582 	if (dip->i_inode.i_mode & S_ISGID) {
583 		if (S_ISDIR(*mode))
584 			*mode |= S_ISGID;
585 		*gid = dip->i_inode.i_gid;
586 	} else
587 		*gid = current_fsgid();
588 }
589 
590 static int alloc_dinode(struct gfs2_inode *dip, u64 *no_addr, u64 *generation)
591 {
592 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
593 	int error;
594 
595 	if (gfs2_alloc_get(dip) == NULL)
596 		return -ENOMEM;
597 
598 	dip->i_alloc->al_requested = RES_DINODE;
599 	error = gfs2_inplace_reserve(dip);
600 	if (error)
601 		goto out;
602 
603 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
604 	if (error)
605 		goto out_ipreserv;
606 
607 	error = gfs2_alloc_di(dip, no_addr, generation);
608 
609 	gfs2_trans_end(sdp);
610 
611 out_ipreserv:
612 	gfs2_inplace_release(dip);
613 out:
614 	gfs2_alloc_put(dip);
615 	return error;
616 }
617 
618 /**
619  * init_dinode - Fill in a new dinode structure
620  * @dip: the directory this inode is being created in
621  * @gl: The glock covering the new inode
622  * @inum: the inode number
623  * @mode: the file permissions
624  * @uid:
625  * @gid:
626  *
627  */
628 
629 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
630 			const struct gfs2_inum_host *inum, unsigned int mode,
631 			unsigned int uid, unsigned int gid,
632 			const u64 *generation, dev_t dev, struct buffer_head **bhp)
633 {
634 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
635 	struct gfs2_dinode *di;
636 	struct buffer_head *dibh;
637 	struct timespec tv = CURRENT_TIME;
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(tv.tv_sec);
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_diskflags & GFS2_DIF_INHERIT_JDATA) ||
662 		    gfs2_tune_get(sdp, gt_new_files_jdata))
663 			di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
664 	} else if (S_ISDIR(mode)) {
665 		di->di_flags |= cpu_to_be32(dip->i_diskflags &
666 					    GFS2_DIF_INHERIT_JDATA);
667 	}
668 
669 	di->__pad1 = 0;
670 	di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
671 	di->di_height = 0;
672 	di->__pad2 = 0;
673 	di->__pad3 = 0;
674 	di->di_depth = 0;
675 	di->di_entries = 0;
676 	memset(&di->__pad4, 0, sizeof(di->__pad4));
677 	di->di_eattr = 0;
678 	di->di_atime_nsec = cpu_to_be32(tv.tv_nsec);
679 	di->di_mtime_nsec = cpu_to_be32(tv.tv_nsec);
680 	di->di_ctime_nsec = cpu_to_be32(tv.tv_nsec);
681 	memset(&di->di_reserved, 0, sizeof(di->di_reserved));
682 
683 	set_buffer_uptodate(dibh);
684 
685 	*bhp = 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, struct buffer_head **bhp)
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 	if (!gfs2_alloc_get(dip))
698 		return -ENOMEM;
699 
700 	error = gfs2_quota_lock(dip, uid, gid);
701 	if (error)
702 		goto out;
703 
704 	error = gfs2_quota_check(dip, uid, gid);
705 	if (error)
706 		goto out_quota;
707 
708 	error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
709 	if (error)
710 		goto out_quota;
711 
712 	init_dinode(dip, gl, inum, mode, uid, gid, generation, dev, bhp);
713 	gfs2_quota_change(dip, +1, uid, gid);
714 	gfs2_trans_end(sdp);
715 
716 out_quota:
717 	gfs2_quota_unlock(dip);
718 out:
719 	gfs2_alloc_put(dip);
720 	return error;
721 }
722 
723 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
724 		       struct gfs2_inode *ip)
725 {
726 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
727 	struct gfs2_alloc *al;
728 	int alloc_required;
729 	struct buffer_head *dibh;
730 	int error;
731 
732 	al = gfs2_alloc_get(dip);
733 	if (!al)
734 		return -ENOMEM;
735 
736 	error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
737 	if (error)
738 		goto fail;
739 
740 	error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
741 	if (alloc_required < 0)
742 		goto fail_quota_locks;
743 	if (alloc_required) {
744 		error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
745 		if (error)
746 			goto fail_quota_locks;
747 
748 		al->al_requested = sdp->sd_max_dirres;
749 
750 		error = gfs2_inplace_reserve(dip);
751 		if (error)
752 			goto fail_quota_locks;
753 
754 		error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
755 					 al->al_rgd->rd_length +
756 					 2 * RES_DINODE +
757 					 RES_STATFS + RES_QUOTA, 0);
758 		if (error)
759 			goto fail_ipreserv;
760 	} else {
761 		error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
762 		if (error)
763 			goto fail_quota_locks;
764 	}
765 
766 	error = gfs2_dir_add(&dip->i_inode, name, ip, IF2DT(ip->i_inode.i_mode));
767 	if (error)
768 		goto fail_end_trans;
769 
770 	error = gfs2_meta_inode_buffer(ip, &dibh);
771 	if (error)
772 		goto fail_end_trans;
773 	ip->i_inode.i_nlink = 1;
774 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
775 	gfs2_dinode_out(ip, dibh->b_data);
776 	brelse(dibh);
777 	return 0;
778 
779 fail_end_trans:
780 	gfs2_trans_end(sdp);
781 
782 fail_ipreserv:
783 	if (dip->i_alloc->al_rgd)
784 		gfs2_inplace_release(dip);
785 
786 fail_quota_locks:
787 	gfs2_quota_unlock(dip);
788 
789 fail:
790 	gfs2_alloc_put(dip);
791 	return error;
792 }
793 
794 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip)
795 {
796 	int err;
797 	size_t len;
798 	void *value;
799 	char *name;
800 
801 	err = security_inode_init_security(&ip->i_inode, &dip->i_inode,
802 					   &name, &value, &len);
803 
804 	if (err) {
805 		if (err == -EOPNOTSUPP)
806 			return 0;
807 		return err;
808 	}
809 
810 	err = __gfs2_xattr_set(&ip->i_inode, name, value, len, 0,
811 			       GFS2_EATYPE_SECURITY);
812 	kfree(value);
813 	kfree(name);
814 
815 	return err;
816 }
817 
818 /**
819  * gfs2_createi - Create a new inode
820  * @ghs: An array of two holders
821  * @name: The name of the new file
822  * @mode: the permissions on the new inode
823  *
824  * @ghs[0] is an initialized holder for the directory
825  * @ghs[1] is the holder for the inode lock
826  *
827  * If the return value is not NULL, the glocks on both the directory and the new
828  * file are held.  A transaction has been started and an inplace reservation
829  * is held, as well.
830  *
831  * Returns: An inode
832  */
833 
834 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name,
835 			   unsigned int mode, dev_t dev)
836 {
837 	struct inode *inode = NULL;
838 	struct gfs2_inode *dip = ghs->gh_gl->gl_object;
839 	struct inode *dir = &dip->i_inode;
840 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
841 	struct gfs2_inum_host inum = { .no_addr = 0, .no_formal_ino = 0 };
842 	int error;
843 	u64 generation;
844 	struct buffer_head *bh = NULL;
845 
846 	if (!name->len || name->len > GFS2_FNAMESIZE)
847 		return ERR_PTR(-ENAMETOOLONG);
848 
849 	gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
850 	error = gfs2_glock_nq(ghs);
851 	if (error)
852 		goto fail;
853 
854 	error = create_ok(dip, name, mode);
855 	if (error)
856 		goto fail_gunlock;
857 
858 	error = alloc_dinode(dip, &inum.no_addr, &generation);
859 	if (error)
860 		goto fail_gunlock;
861 	inum.no_formal_ino = generation;
862 
863 	error = gfs2_glock_nq_num(sdp, inum.no_addr, &gfs2_inode_glops,
864 				  LM_ST_EXCLUSIVE, GL_SKIP, ghs + 1);
865 	if (error)
866 		goto fail_gunlock;
867 
868 	error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev, &bh);
869 	if (error)
870 		goto fail_gunlock2;
871 
872 	inode = gfs2_inode_lookup(dir->i_sb, IF2DT(mode), inum.no_addr,
873 				  inum.no_formal_ino);
874 	if (IS_ERR(inode))
875 		goto fail_gunlock2;
876 
877 	error = gfs2_inode_refresh(GFS2_I(inode));
878 	if (error)
879 		goto fail_gunlock2;
880 
881 	error = gfs2_acl_create(dip, inode);
882 	if (error)
883 		goto fail_gunlock2;
884 
885 	error = gfs2_security_init(dip, GFS2_I(inode));
886 	if (error)
887 		goto fail_gunlock2;
888 
889 	error = link_dinode(dip, name, GFS2_I(inode));
890 	if (error)
891 		goto fail_gunlock2;
892 
893 	if (bh)
894 		brelse(bh);
895 	return inode;
896 
897 fail_gunlock2:
898 	gfs2_glock_dq_uninit(ghs + 1);
899 	if (inode && !IS_ERR(inode))
900 		iput(inode);
901 fail_gunlock:
902 	gfs2_glock_dq(ghs);
903 fail:
904 	if (bh)
905 		brelse(bh);
906 	return ERR_PTR(error);
907 }
908 
909 static int __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
910 {
911 	struct inode *inode = &ip->i_inode;
912 	struct buffer_head *dibh;
913 	int error;
914 
915 	error = gfs2_meta_inode_buffer(ip, &dibh);
916 	if (error)
917 		return error;
918 
919 	setattr_copy(inode, attr);
920 	mark_inode_dirty(inode);
921 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
922 	gfs2_dinode_out(ip, dibh->b_data);
923 	brelse(dibh);
924 	return 0;
925 }
926 
927 /**
928  * gfs2_setattr_simple -
929  * @ip:
930  * @attr:
931  *
932  * Called with a reference on the vnode.
933  *
934  * Returns: errno
935  */
936 
937 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
938 {
939 	int error;
940 
941 	if (current->journal_info)
942 		return __gfs2_setattr_simple(ip, attr);
943 
944 	error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0);
945 	if (error)
946 		return error;
947 
948 	error = __gfs2_setattr_simple(ip, attr);
949 	gfs2_trans_end(GFS2_SB(&ip->i_inode));
950 	return error;
951 }
952 
953 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
954 {
955 	struct gfs2_dinode *str = buf;
956 
957 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
958 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
959 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
960 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
961 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
962 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
963 	str->di_uid = cpu_to_be32(ip->i_inode.i_uid);
964 	str->di_gid = cpu_to_be32(ip->i_inode.i_gid);
965 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
966 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
967 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
968 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
969 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
970 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
971 
972 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
973 	str->di_goal_data = cpu_to_be64(ip->i_goal);
974 	str->di_generation = cpu_to_be64(ip->i_generation);
975 
976 	str->di_flags = cpu_to_be32(ip->i_diskflags);
977 	str->di_height = cpu_to_be16(ip->i_height);
978 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
979 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
980 					     GFS2_FORMAT_DE : 0);
981 	str->di_depth = cpu_to_be16(ip->i_depth);
982 	str->di_entries = cpu_to_be32(ip->i_entries);
983 
984 	str->di_eattr = cpu_to_be64(ip->i_eattr);
985 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
986 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
987 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
988 }
989 
990 void gfs2_dinode_print(const struct gfs2_inode *ip)
991 {
992 	printk(KERN_INFO "  no_formal_ino = %llu\n",
993 	       (unsigned long long)ip->i_no_formal_ino);
994 	printk(KERN_INFO "  no_addr = %llu\n",
995 	       (unsigned long long)ip->i_no_addr);
996 	printk(KERN_INFO "  i_size = %llu\n",
997 	       (unsigned long long)i_size_read(&ip->i_inode));
998 	printk(KERN_INFO "  blocks = %llu\n",
999 	       (unsigned long long)gfs2_get_inode_blocks(&ip->i_inode));
1000 	printk(KERN_INFO "  i_goal = %llu\n",
1001 	       (unsigned long long)ip->i_goal);
1002 	printk(KERN_INFO "  i_diskflags = 0x%.8X\n", ip->i_diskflags);
1003 	printk(KERN_INFO "  i_height = %u\n", ip->i_height);
1004 	printk(KERN_INFO "  i_depth = %u\n", ip->i_depth);
1005 	printk(KERN_INFO "  i_entries = %u\n", ip->i_entries);
1006 	printk(KERN_INFO "  i_eattr = %llu\n",
1007 	       (unsigned long long)ip->i_eattr);
1008 }
1009 
1010