xref: /linux/fs/gfs2/dir.c (revision bab2c80e5a6c855657482eac9e97f5f3eedb509a)
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 /*
11  * Implements Extendible Hashing as described in:
12  *   "Extendible Hashing" by Fagin, et al in
13  *     __ACM Trans. on Database Systems__, Sept 1979.
14  *
15  *
16  * Here's the layout of dirents which is essentially the same as that of ext2
17  * within a single block. The field de_name_len is the number of bytes
18  * actually required for the name (no null terminator). The field de_rec_len
19  * is the number of bytes allocated to the dirent. The offset of the next
20  * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
21  * deleted, the preceding dirent inherits its allocated space, ie
22  * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
23  * by adding de_rec_len to the current dirent, this essentially causes the
24  * deleted dirent to get jumped over when iterating through all the dirents.
25  *
26  * When deleting the first dirent in a block, there is no previous dirent so
27  * the field de_ino is set to zero to designate it as deleted. When allocating
28  * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
29  * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
30  * dirent is allocated. Otherwise it must go through all the 'used' dirents
31  * searching for one in which the amount of total space minus the amount of
32  * used space will provide enough space for the new dirent.
33  *
34  * There are two types of blocks in which dirents reside. In a stuffed dinode,
35  * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
36  * the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
37  * beginning of the leaf block. The dirents reside in leaves when
38  *
39  * dip->i_diskflags & GFS2_DIF_EXHASH is true
40  *
41  * Otherwise, the dirents are "linear", within a single stuffed dinode block.
42  *
43  * When the dirents are in leaves, the actual contents of the directory file are
44  * used as an array of 64-bit block pointers pointing to the leaf blocks. The
45  * dirents are NOT in the directory file itself. There can be more than one
46  * block pointer in the array that points to the same leaf. In fact, when a
47  * directory is first converted from linear to exhash, all of the pointers
48  * point to the same leaf.
49  *
50  * When a leaf is completely full, the size of the hash table can be
51  * doubled unless it is already at the maximum size which is hard coded into
52  * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
53  * but never before the maximum hash table size has been reached.
54  */
55 
56 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
57 
58 #include <linux/slab.h>
59 #include <linux/spinlock.h>
60 #include <linux/buffer_head.h>
61 #include <linux/sort.h>
62 #include <linux/gfs2_ondisk.h>
63 #include <linux/crc32.h>
64 #include <linux/vmalloc.h>
65 #include <linux/bio.h>
66 
67 #include "gfs2.h"
68 #include "incore.h"
69 #include "dir.h"
70 #include "glock.h"
71 #include "inode.h"
72 #include "meta_io.h"
73 #include "quota.h"
74 #include "rgrp.h"
75 #include "trans.h"
76 #include "bmap.h"
77 #include "util.h"
78 
79 #define IS_LEAF     1 /* Hashed (leaf) directory */
80 #define IS_DINODE   2 /* Linear (stuffed dinode block) directory */
81 
82 #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
83 
84 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
85 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
86 #define GFS2_HASH_INDEX_MASK 0xffffc000
87 #define GFS2_USE_HASH_FLAG 0x2000
88 
89 struct qstr gfs2_qdot __read_mostly;
90 struct qstr gfs2_qdotdot __read_mostly;
91 
92 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
93 			    const struct qstr *name, void *opaque);
94 
95 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
96 			    struct buffer_head **bhp)
97 {
98 	struct buffer_head *bh;
99 
100 	bh = gfs2_meta_new(ip->i_gl, block);
101 	gfs2_trans_add_meta(ip->i_gl, bh);
102 	gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
103 	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
104 	*bhp = bh;
105 	return 0;
106 }
107 
108 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
109 					struct buffer_head **bhp)
110 {
111 	struct buffer_head *bh;
112 	int error;
113 
114 	error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh);
115 	if (error)
116 		return error;
117 	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
118 		brelse(bh);
119 		return -EIO;
120 	}
121 	*bhp = bh;
122 	return 0;
123 }
124 
125 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
126 				  unsigned int offset, unsigned int size)
127 {
128 	struct buffer_head *dibh;
129 	int error;
130 
131 	error = gfs2_meta_inode_buffer(ip, &dibh);
132 	if (error)
133 		return error;
134 
135 	gfs2_trans_add_meta(ip->i_gl, dibh);
136 	memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
137 	if (ip->i_inode.i_size < offset + size)
138 		i_size_write(&ip->i_inode, offset + size);
139 	ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
140 	gfs2_dinode_out(ip, dibh->b_data);
141 
142 	brelse(dibh);
143 
144 	return size;
145 }
146 
147 
148 
149 /**
150  * gfs2_dir_write_data - Write directory information to the inode
151  * @ip: The GFS2 inode
152  * @buf: The buffer containing information to be written
153  * @offset: The file offset to start writing at
154  * @size: The amount of data to write
155  *
156  * Returns: The number of bytes correctly written or error code
157  */
158 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
159 			       u64 offset, unsigned int size)
160 {
161 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
162 	struct buffer_head *dibh;
163 	u64 lblock, dblock;
164 	u32 extlen = 0;
165 	unsigned int o;
166 	int copied = 0;
167 	int error = 0;
168 	int new = 0;
169 
170 	if (!size)
171 		return 0;
172 
173 	if (gfs2_is_stuffed(ip) && offset + size <= gfs2_max_stuffed_size(ip))
174 		return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
175 					      size);
176 
177 	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
178 		return -EINVAL;
179 
180 	if (gfs2_is_stuffed(ip)) {
181 		error = gfs2_unstuff_dinode(ip, NULL);
182 		if (error)
183 			return error;
184 	}
185 
186 	lblock = offset;
187 	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
188 
189 	while (copied < size) {
190 		unsigned int amount;
191 		struct buffer_head *bh;
192 
193 		amount = size - copied;
194 		if (amount > sdp->sd_sb.sb_bsize - o)
195 			amount = sdp->sd_sb.sb_bsize - o;
196 
197 		if (!extlen) {
198 			new = 1;
199 			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
200 						&dblock, &extlen);
201 			if (error)
202 				goto fail;
203 			error = -EIO;
204 			if (gfs2_assert_withdraw(sdp, dblock))
205 				goto fail;
206 		}
207 
208 		if (amount == sdp->sd_jbsize || new)
209 			error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
210 		else
211 			error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
212 
213 		if (error)
214 			goto fail;
215 
216 		gfs2_trans_add_meta(ip->i_gl, bh);
217 		memcpy(bh->b_data + o, buf, amount);
218 		brelse(bh);
219 
220 		buf += amount;
221 		copied += amount;
222 		lblock++;
223 		dblock++;
224 		extlen--;
225 
226 		o = sizeof(struct gfs2_meta_header);
227 	}
228 
229 out:
230 	error = gfs2_meta_inode_buffer(ip, &dibh);
231 	if (error)
232 		return error;
233 
234 	if (ip->i_inode.i_size < offset + copied)
235 		i_size_write(&ip->i_inode, offset + copied);
236 	ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode);
237 
238 	gfs2_trans_add_meta(ip->i_gl, dibh);
239 	gfs2_dinode_out(ip, dibh->b_data);
240 	brelse(dibh);
241 
242 	return copied;
243 fail:
244 	if (copied)
245 		goto out;
246 	return error;
247 }
248 
249 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
250 				 unsigned int size)
251 {
252 	struct buffer_head *dibh;
253 	int error;
254 
255 	error = gfs2_meta_inode_buffer(ip, &dibh);
256 	if (!error) {
257 		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
258 		brelse(dibh);
259 	}
260 
261 	return (error) ? error : size;
262 }
263 
264 
265 /**
266  * gfs2_dir_read_data - Read a data from a directory inode
267  * @ip: The GFS2 Inode
268  * @buf: The buffer to place result into
269  * @size: Amount of data to transfer
270  *
271  * Returns: The amount of data actually copied or the error
272  */
273 static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
274 			      unsigned int size)
275 {
276 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
277 	u64 lblock, dblock;
278 	u32 extlen = 0;
279 	unsigned int o;
280 	int copied = 0;
281 	int error = 0;
282 
283 	if (gfs2_is_stuffed(ip))
284 		return gfs2_dir_read_stuffed(ip, buf, size);
285 
286 	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
287 		return -EINVAL;
288 
289 	lblock = 0;
290 	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
291 
292 	while (copied < size) {
293 		unsigned int amount;
294 		struct buffer_head *bh;
295 		int new;
296 
297 		amount = size - copied;
298 		if (amount > sdp->sd_sb.sb_bsize - o)
299 			amount = sdp->sd_sb.sb_bsize - o;
300 
301 		if (!extlen) {
302 			new = 0;
303 			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
304 						&dblock, &extlen);
305 			if (error || !dblock)
306 				goto fail;
307 			BUG_ON(extlen < 1);
308 			bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
309 		} else {
310 			error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
311 			if (error)
312 				goto fail;
313 		}
314 		error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
315 		if (error) {
316 			brelse(bh);
317 			goto fail;
318 		}
319 		dblock++;
320 		extlen--;
321 		memcpy(buf, bh->b_data + o, amount);
322 		brelse(bh);
323 		buf += (amount/sizeof(__be64));
324 		copied += amount;
325 		lblock++;
326 		o = sizeof(struct gfs2_meta_header);
327 	}
328 
329 	return copied;
330 fail:
331 	return (copied) ? copied : error;
332 }
333 
334 /**
335  * gfs2_dir_get_hash_table - Get pointer to the dir hash table
336  * @ip: The inode in question
337  *
338  * Returns: The hash table or an error
339  */
340 
341 static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
342 {
343 	struct inode *inode = &ip->i_inode;
344 	int ret;
345 	u32 hsize;
346 	__be64 *hc;
347 
348 	BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
349 
350 	hc = ip->i_hash_cache;
351 	if (hc)
352 		return hc;
353 
354 	hsize = BIT(ip->i_depth);
355 	hsize *= sizeof(__be64);
356 	if (hsize != i_size_read(&ip->i_inode)) {
357 		gfs2_consist_inode(ip);
358 		return ERR_PTR(-EIO);
359 	}
360 
361 	hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
362 	if (hc == NULL)
363 		hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL);
364 
365 	if (hc == NULL)
366 		return ERR_PTR(-ENOMEM);
367 
368 	ret = gfs2_dir_read_data(ip, hc, hsize);
369 	if (ret < 0) {
370 		kvfree(hc);
371 		return ERR_PTR(ret);
372 	}
373 
374 	spin_lock(&inode->i_lock);
375 	if (likely(!ip->i_hash_cache)) {
376 		ip->i_hash_cache = hc;
377 		hc = NULL;
378 	}
379 	spin_unlock(&inode->i_lock);
380 	kvfree(hc);
381 
382 	return ip->i_hash_cache;
383 }
384 
385 /**
386  * gfs2_dir_hash_inval - Invalidate dir hash
387  * @ip: The directory inode
388  *
389  * Must be called with an exclusive glock, or during glock invalidation.
390  */
391 void gfs2_dir_hash_inval(struct gfs2_inode *ip)
392 {
393 	__be64 *hc;
394 
395 	spin_lock(&ip->i_inode.i_lock);
396 	hc = ip->i_hash_cache;
397 	ip->i_hash_cache = NULL;
398 	spin_unlock(&ip->i_inode.i_lock);
399 
400 	kvfree(hc);
401 }
402 
403 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
404 {
405 	return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
406 }
407 
408 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
409 				     const struct qstr *name, int ret)
410 {
411 	if (!gfs2_dirent_sentinel(dent) &&
412 	    be32_to_cpu(dent->de_hash) == name->hash &&
413 	    be16_to_cpu(dent->de_name_len) == name->len &&
414 	    memcmp(dent+1, name->name, name->len) == 0)
415 		return ret;
416 	return 0;
417 }
418 
419 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
420 			    const struct qstr *name,
421 			    void *opaque)
422 {
423 	return __gfs2_dirent_find(dent, name, 1);
424 }
425 
426 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
427 			    const struct qstr *name,
428 			    void *opaque)
429 {
430 	return __gfs2_dirent_find(dent, name, 2);
431 }
432 
433 /*
434  * name->name holds ptr to start of block.
435  * name->len holds size of block.
436  */
437 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
438 			    const struct qstr *name,
439 			    void *opaque)
440 {
441 	const char *start = name->name;
442 	const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
443 	if (name->len == (end - start))
444 		return 1;
445 	return 0;
446 }
447 
448 /* Look for the dirent that contains the offset specified in data. Once we
449  * find that dirent, there must be space available there for the new dirent */
450 static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
451 				  const struct qstr *name,
452 				  void *ptr)
453 {
454 	unsigned required = GFS2_DIRENT_SIZE(name->len);
455 	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
456 	unsigned totlen = be16_to_cpu(dent->de_rec_len);
457 
458 	if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
459 		return 0;
460 	if (gfs2_dirent_sentinel(dent))
461 		actual = 0;
462 	if (ptr < (void *)dent + actual)
463 		return -1;
464 	if ((void *)dent + totlen >= ptr + required)
465 		return 1;
466 	return -1;
467 }
468 
469 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
470 				  const struct qstr *name,
471 				  void *opaque)
472 {
473 	unsigned required = GFS2_DIRENT_SIZE(name->len);
474 	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
475 	unsigned totlen = be16_to_cpu(dent->de_rec_len);
476 
477 	if (gfs2_dirent_sentinel(dent))
478 		actual = 0;
479 	if (totlen - actual >= required)
480 		return 1;
481 	return 0;
482 }
483 
484 struct dirent_gather {
485 	const struct gfs2_dirent **pdent;
486 	unsigned offset;
487 };
488 
489 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
490 			      const struct qstr *name,
491 			      void *opaque)
492 {
493 	struct dirent_gather *g = opaque;
494 	if (!gfs2_dirent_sentinel(dent)) {
495 		g->pdent[g->offset++] = dent;
496 	}
497 	return 0;
498 }
499 
500 /*
501  * Other possible things to check:
502  * - Inode located within filesystem size (and on valid block)
503  * - Valid directory entry type
504  * Not sure how heavy-weight we want to make this... could also check
505  * hash is correct for example, but that would take a lot of extra time.
506  * For now the most important thing is to check that the various sizes
507  * are correct.
508  */
509 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
510 			     unsigned int size, unsigned int len, int first)
511 {
512 	const char *msg = "gfs2_dirent too small";
513 	if (unlikely(size < sizeof(struct gfs2_dirent)))
514 		goto error;
515 	msg = "gfs2_dirent misaligned";
516 	if (unlikely(offset & 0x7))
517 		goto error;
518 	msg = "gfs2_dirent points beyond end of block";
519 	if (unlikely(offset + size > len))
520 		goto error;
521 	msg = "zero inode number";
522 	if (unlikely(!first && gfs2_dirent_sentinel(dent)))
523 		goto error;
524 	msg = "name length is greater than space in dirent";
525 	if (!gfs2_dirent_sentinel(dent) &&
526 	    unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
527 		     size))
528 		goto error;
529 	return 0;
530 error:
531 	pr_warn("%s: %s (%s)\n",
532 		__func__, msg, first ? "first in block" : "not first in block");
533 	return -EIO;
534 }
535 
536 static int gfs2_dirent_offset(const void *buf)
537 {
538 	const struct gfs2_meta_header *h = buf;
539 	int offset;
540 
541 	BUG_ON(buf == NULL);
542 
543 	switch(be32_to_cpu(h->mh_type)) {
544 	case GFS2_METATYPE_LF:
545 		offset = sizeof(struct gfs2_leaf);
546 		break;
547 	case GFS2_METATYPE_DI:
548 		offset = sizeof(struct gfs2_dinode);
549 		break;
550 	default:
551 		goto wrong_type;
552 	}
553 	return offset;
554 wrong_type:
555 	pr_warn("%s: wrong block type %u\n", __func__, be32_to_cpu(h->mh_type));
556 	return -1;
557 }
558 
559 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
560 					    unsigned int len, gfs2_dscan_t scan,
561 					    const struct qstr *name,
562 					    void *opaque)
563 {
564 	struct gfs2_dirent *dent, *prev;
565 	unsigned offset;
566 	unsigned size;
567 	int ret = 0;
568 
569 	ret = gfs2_dirent_offset(buf);
570 	if (ret < 0)
571 		goto consist_inode;
572 
573 	offset = ret;
574 	prev = NULL;
575 	dent = buf + offset;
576 	size = be16_to_cpu(dent->de_rec_len);
577 	if (gfs2_check_dirent(dent, offset, size, len, 1))
578 		goto consist_inode;
579 	do {
580 		ret = scan(dent, name, opaque);
581 		if (ret)
582 			break;
583 		offset += size;
584 		if (offset == len)
585 			break;
586 		prev = dent;
587 		dent = buf + offset;
588 		size = be16_to_cpu(dent->de_rec_len);
589 		if (gfs2_check_dirent(dent, offset, size, len, 0))
590 			goto consist_inode;
591 	} while(1);
592 
593 	switch(ret) {
594 	case 0:
595 		return NULL;
596 	case 1:
597 		return dent;
598 	case 2:
599 		return prev ? prev : dent;
600 	default:
601 		BUG_ON(ret > 0);
602 		return ERR_PTR(ret);
603 	}
604 
605 consist_inode:
606 	gfs2_consist_inode(GFS2_I(inode));
607 	return ERR_PTR(-EIO);
608 }
609 
610 static int dirent_check_reclen(struct gfs2_inode *dip,
611 			       const struct gfs2_dirent *d, const void *end_p)
612 {
613 	const void *ptr = d;
614 	u16 rec_len = be16_to_cpu(d->de_rec_len);
615 
616 	if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
617 		goto broken;
618 	ptr += rec_len;
619 	if (ptr < end_p)
620 		return rec_len;
621 	if (ptr == end_p)
622 		return -ENOENT;
623 broken:
624 	gfs2_consist_inode(dip);
625 	return -EIO;
626 }
627 
628 /**
629  * dirent_next - Next dirent
630  * @dip: the directory
631  * @bh: The buffer
632  * @dent: Pointer to list of dirents
633  *
634  * Returns: 0 on success, error code otherwise
635  */
636 
637 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
638 		       struct gfs2_dirent **dent)
639 {
640 	struct gfs2_dirent *cur = *dent, *tmp;
641 	char *bh_end = bh->b_data + bh->b_size;
642 	int ret;
643 
644 	ret = dirent_check_reclen(dip, cur, bh_end);
645 	if (ret < 0)
646 		return ret;
647 
648 	tmp = (void *)cur + ret;
649 	ret = dirent_check_reclen(dip, tmp, bh_end);
650 	if (ret == -EIO)
651 		return ret;
652 
653         /* Only the first dent could ever have de_inum.no_addr == 0 */
654 	if (gfs2_dirent_sentinel(tmp)) {
655 		gfs2_consist_inode(dip);
656 		return -EIO;
657 	}
658 
659 	*dent = tmp;
660 	return 0;
661 }
662 
663 /**
664  * dirent_del - Delete a dirent
665  * @dip: The GFS2 inode
666  * @bh: The buffer
667  * @prev: The previous dirent
668  * @cur: The current dirent
669  *
670  */
671 
672 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
673 		       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
674 {
675 	u16 cur_rec_len, prev_rec_len;
676 
677 	if (gfs2_dirent_sentinel(cur)) {
678 		gfs2_consist_inode(dip);
679 		return;
680 	}
681 
682 	gfs2_trans_add_meta(dip->i_gl, bh);
683 
684 	/* If there is no prev entry, this is the first entry in the block.
685 	   The de_rec_len is already as big as it needs to be.  Just zero
686 	   out the inode number and return.  */
687 
688 	if (!prev) {
689 		cur->de_inum.no_addr = 0;
690 		cur->de_inum.no_formal_ino = 0;
691 		return;
692 	}
693 
694 	/*  Combine this dentry with the previous one.  */
695 
696 	prev_rec_len = be16_to_cpu(prev->de_rec_len);
697 	cur_rec_len = be16_to_cpu(cur->de_rec_len);
698 
699 	if ((char *)prev + prev_rec_len != (char *)cur)
700 		gfs2_consist_inode(dip);
701 	if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
702 		gfs2_consist_inode(dip);
703 
704 	prev_rec_len += cur_rec_len;
705 	prev->de_rec_len = cpu_to_be16(prev_rec_len);
706 }
707 
708 
709 static struct gfs2_dirent *do_init_dirent(struct inode *inode,
710 					  struct gfs2_dirent *dent,
711 					  const struct qstr *name,
712 					  struct buffer_head *bh,
713 					  unsigned offset)
714 {
715 	struct gfs2_inode *ip = GFS2_I(inode);
716 	struct gfs2_dirent *ndent;
717 	unsigned totlen;
718 
719 	totlen = be16_to_cpu(dent->de_rec_len);
720 	BUG_ON(offset + name->len > totlen);
721 	gfs2_trans_add_meta(ip->i_gl, bh);
722 	ndent = (struct gfs2_dirent *)((char *)dent + offset);
723 	dent->de_rec_len = cpu_to_be16(offset);
724 	gfs2_qstr2dirent(name, totlen - offset, ndent);
725 	return ndent;
726 }
727 
728 
729 /*
730  * Takes a dent from which to grab space as an argument. Returns the
731  * newly created dent.
732  */
733 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
734 					    struct gfs2_dirent *dent,
735 					    const struct qstr *name,
736 					    struct buffer_head *bh)
737 {
738 	unsigned offset = 0;
739 
740 	if (!gfs2_dirent_sentinel(dent))
741 		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
742 	return do_init_dirent(inode, dent, name, bh, offset);
743 }
744 
745 static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
746 						   struct buffer_head *bh,
747 						   const struct qstr *name,
748 						   void *ptr)
749 {
750 	struct gfs2_dirent *dent;
751 	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
752 				gfs2_dirent_find_offset, name, ptr);
753 	if (!dent || IS_ERR(dent))
754 		return dent;
755 	return do_init_dirent(inode, dent, name, bh,
756 			      (unsigned)(ptr - (void *)dent));
757 }
758 
759 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
760 		    struct buffer_head **bhp)
761 {
762 	int error;
763 
764 	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
765 	if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
766 		/* pr_info("block num=%llu\n", leaf_no); */
767 		error = -EIO;
768 	}
769 
770 	return error;
771 }
772 
773 /**
774  * get_leaf_nr - Get a leaf number associated with the index
775  * @dip: The GFS2 inode
776  * @index:
777  * @leaf_out:
778  *
779  * Returns: 0 on success, error code otherwise
780  */
781 
782 static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
783 		       u64 *leaf_out)
784 {
785 	__be64 *hash;
786 	int error;
787 
788 	hash = gfs2_dir_get_hash_table(dip);
789 	error = PTR_ERR_OR_ZERO(hash);
790 
791 	if (!error)
792 		*leaf_out = be64_to_cpu(*(hash + index));
793 
794 	return error;
795 }
796 
797 static int get_first_leaf(struct gfs2_inode *dip, u32 index,
798 			  struct buffer_head **bh_out)
799 {
800 	u64 leaf_no;
801 	int error;
802 
803 	error = get_leaf_nr(dip, index, &leaf_no);
804 	if (!error)
805 		error = get_leaf(dip, leaf_no, bh_out);
806 
807 	return error;
808 }
809 
810 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
811 					      const struct qstr *name,
812 					      gfs2_dscan_t scan,
813 					      struct buffer_head **pbh)
814 {
815 	struct buffer_head *bh;
816 	struct gfs2_dirent *dent;
817 	struct gfs2_inode *ip = GFS2_I(inode);
818 	int error;
819 
820 	if (ip->i_diskflags & GFS2_DIF_EXHASH) {
821 		struct gfs2_leaf *leaf;
822 		unsigned int hsize = BIT(ip->i_depth);
823 		unsigned int index;
824 		u64 ln;
825 		if (hsize * sizeof(u64) != i_size_read(inode)) {
826 			gfs2_consist_inode(ip);
827 			return ERR_PTR(-EIO);
828 		}
829 
830 		index = name->hash >> (32 - ip->i_depth);
831 		error = get_first_leaf(ip, index, &bh);
832 		if (error)
833 			return ERR_PTR(error);
834 		do {
835 			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
836 						scan, name, NULL);
837 			if (dent)
838 				goto got_dent;
839 			leaf = (struct gfs2_leaf *)bh->b_data;
840 			ln = be64_to_cpu(leaf->lf_next);
841 			brelse(bh);
842 			if (!ln)
843 				break;
844 
845 			error = get_leaf(ip, ln, &bh);
846 		} while(!error);
847 
848 		return error ? ERR_PTR(error) : NULL;
849 	}
850 
851 
852 	error = gfs2_meta_inode_buffer(ip, &bh);
853 	if (error)
854 		return ERR_PTR(error);
855 	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
856 got_dent:
857 	if (unlikely(dent == NULL || IS_ERR(dent))) {
858 		brelse(bh);
859 		bh = NULL;
860 	}
861 	*pbh = bh;
862 	return dent;
863 }
864 
865 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
866 {
867 	struct gfs2_inode *ip = GFS2_I(inode);
868 	unsigned int n = 1;
869 	u64 bn;
870 	int error;
871 	struct buffer_head *bh;
872 	struct gfs2_leaf *leaf;
873 	struct gfs2_dirent *dent;
874 	struct timespec64 tv = current_time(inode);
875 
876 	error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
877 	if (error)
878 		return NULL;
879 	bh = gfs2_meta_new(ip->i_gl, bn);
880 	if (!bh)
881 		return NULL;
882 
883 	gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
884 	gfs2_trans_add_meta(ip->i_gl, bh);
885 	gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
886 	leaf = (struct gfs2_leaf *)bh->b_data;
887 	leaf->lf_depth = cpu_to_be16(depth);
888 	leaf->lf_entries = 0;
889 	leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
890 	leaf->lf_next = 0;
891 	leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
892 	leaf->lf_dist = cpu_to_be32(1);
893 	leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
894 	leaf->lf_sec = cpu_to_be64(tv.tv_sec);
895 	memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
896 	dent = (struct gfs2_dirent *)(leaf+1);
897 	gfs2_qstr2dirent(&empty_name, bh->b_size - sizeof(struct gfs2_leaf), dent);
898 	*pbh = bh;
899 	return leaf;
900 }
901 
902 /**
903  * dir_make_exhash - Convert a stuffed directory into an ExHash directory
904  * @dip: The GFS2 inode
905  *
906  * Returns: 0 on success, error code otherwise
907  */
908 
909 static int dir_make_exhash(struct inode *inode)
910 {
911 	struct gfs2_inode *dip = GFS2_I(inode);
912 	struct gfs2_sbd *sdp = GFS2_SB(inode);
913 	struct gfs2_dirent *dent;
914 	struct qstr args;
915 	struct buffer_head *bh, *dibh;
916 	struct gfs2_leaf *leaf;
917 	int y;
918 	u32 x;
919 	__be64 *lp;
920 	u64 bn;
921 	int error;
922 
923 	error = gfs2_meta_inode_buffer(dip, &dibh);
924 	if (error)
925 		return error;
926 
927 	/*  Turn over a new leaf  */
928 
929 	leaf = new_leaf(inode, &bh, 0);
930 	if (!leaf)
931 		return -ENOSPC;
932 	bn = bh->b_blocknr;
933 
934 	gfs2_assert(sdp, dip->i_entries < BIT(16));
935 	leaf->lf_entries = cpu_to_be16(dip->i_entries);
936 
937 	/*  Copy dirents  */
938 
939 	gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
940 			     sizeof(struct gfs2_dinode));
941 
942 	/*  Find last entry  */
943 
944 	x = 0;
945 	args.len = bh->b_size - sizeof(struct gfs2_dinode) +
946 		   sizeof(struct gfs2_leaf);
947 	args.name = bh->b_data;
948 	dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
949 				gfs2_dirent_last, &args, NULL);
950 	if (!dent) {
951 		brelse(bh);
952 		brelse(dibh);
953 		return -EIO;
954 	}
955 	if (IS_ERR(dent)) {
956 		brelse(bh);
957 		brelse(dibh);
958 		return PTR_ERR(dent);
959 	}
960 
961 	/*  Adjust the last dirent's record length
962 	   (Remember that dent still points to the last entry.)  */
963 
964 	dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
965 		sizeof(struct gfs2_dinode) -
966 		sizeof(struct gfs2_leaf));
967 
968 	brelse(bh);
969 
970 	/*  We're done with the new leaf block, now setup the new
971 	    hash table.  */
972 
973 	gfs2_trans_add_meta(dip->i_gl, dibh);
974 	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
975 
976 	lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
977 
978 	for (x = sdp->sd_hash_ptrs; x--; lp++)
979 		*lp = cpu_to_be64(bn);
980 
981 	i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
982 	gfs2_add_inode_blocks(&dip->i_inode, 1);
983 	dip->i_diskflags |= GFS2_DIF_EXHASH;
984 
985 	for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
986 	dip->i_depth = y;
987 
988 	gfs2_dinode_out(dip, dibh->b_data);
989 
990 	brelse(dibh);
991 
992 	return 0;
993 }
994 
995 /**
996  * dir_split_leaf - Split a leaf block into two
997  * @dip: The GFS2 inode
998  * @index:
999  * @leaf_no:
1000  *
1001  * Returns: 0 on success, error code on failure
1002  */
1003 
1004 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
1005 {
1006 	struct gfs2_inode *dip = GFS2_I(inode);
1007 	struct buffer_head *nbh, *obh, *dibh;
1008 	struct gfs2_leaf *nleaf, *oleaf;
1009 	struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
1010 	u32 start, len, half_len, divider;
1011 	u64 bn, leaf_no;
1012 	__be64 *lp;
1013 	u32 index;
1014 	int x, moved = 0;
1015 	int error;
1016 
1017 	index = name->hash >> (32 - dip->i_depth);
1018 	error = get_leaf_nr(dip, index, &leaf_no);
1019 	if (error)
1020 		return error;
1021 
1022 	/*  Get the old leaf block  */
1023 	error = get_leaf(dip, leaf_no, &obh);
1024 	if (error)
1025 		return error;
1026 
1027 	oleaf = (struct gfs2_leaf *)obh->b_data;
1028 	if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
1029 		brelse(obh);
1030 		return 1; /* can't split */
1031 	}
1032 
1033 	gfs2_trans_add_meta(dip->i_gl, obh);
1034 
1035 	nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
1036 	if (!nleaf) {
1037 		brelse(obh);
1038 		return -ENOSPC;
1039 	}
1040 	bn = nbh->b_blocknr;
1041 
1042 	/*  Compute the start and len of leaf pointers in the hash table.  */
1043 	len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth));
1044 	half_len = len >> 1;
1045 	if (!half_len) {
1046 		pr_warn("i_depth %u lf_depth %u index %u\n",
1047 			dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
1048 		gfs2_consist_inode(dip);
1049 		error = -EIO;
1050 		goto fail_brelse;
1051 	}
1052 
1053 	start = (index & ~(len - 1));
1054 
1055 	/* Change the pointers.
1056 	   Don't bother distinguishing stuffed from non-stuffed.
1057 	   This code is complicated enough already. */
1058 	lp = kmalloc_array(half_len, sizeof(__be64), GFP_NOFS);
1059 	if (!lp) {
1060 		error = -ENOMEM;
1061 		goto fail_brelse;
1062 	}
1063 
1064 	/*  Change the pointers  */
1065 	for (x = 0; x < half_len; x++)
1066 		lp[x] = cpu_to_be64(bn);
1067 
1068 	gfs2_dir_hash_inval(dip);
1069 
1070 	error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1071 				    half_len * sizeof(u64));
1072 	if (error != half_len * sizeof(u64)) {
1073 		if (error >= 0)
1074 			error = -EIO;
1075 		goto fail_lpfree;
1076 	}
1077 
1078 	kfree(lp);
1079 
1080 	/*  Compute the divider  */
1081 	divider = (start + half_len) << (32 - dip->i_depth);
1082 
1083 	/*  Copy the entries  */
1084 	dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1085 
1086 	do {
1087 		next = dent;
1088 		if (dirent_next(dip, obh, &next))
1089 			next = NULL;
1090 
1091 		if (!gfs2_dirent_sentinel(dent) &&
1092 		    be32_to_cpu(dent->de_hash) < divider) {
1093 			struct qstr str;
1094 			void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
1095 			str.name = (char*)(dent+1);
1096 			str.len = be16_to_cpu(dent->de_name_len);
1097 			str.hash = be32_to_cpu(dent->de_hash);
1098 			new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
1099 			if (IS_ERR(new)) {
1100 				error = PTR_ERR(new);
1101 				break;
1102 			}
1103 
1104 			new->de_inum = dent->de_inum; /* No endian worries */
1105 			new->de_type = dent->de_type; /* No endian worries */
1106 			be16_add_cpu(&nleaf->lf_entries, 1);
1107 
1108 			dirent_del(dip, obh, prev, dent);
1109 
1110 			if (!oleaf->lf_entries)
1111 				gfs2_consist_inode(dip);
1112 			be16_add_cpu(&oleaf->lf_entries, -1);
1113 
1114 			if (!prev)
1115 				prev = dent;
1116 
1117 			moved = 1;
1118 		} else {
1119 			prev = dent;
1120 		}
1121 		dent = next;
1122 	} while (dent);
1123 
1124 	oleaf->lf_depth = nleaf->lf_depth;
1125 
1126 	error = gfs2_meta_inode_buffer(dip, &dibh);
1127 	if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1128 		gfs2_trans_add_meta(dip->i_gl, dibh);
1129 		gfs2_add_inode_blocks(&dip->i_inode, 1);
1130 		gfs2_dinode_out(dip, dibh->b_data);
1131 		brelse(dibh);
1132 	}
1133 
1134 	brelse(obh);
1135 	brelse(nbh);
1136 
1137 	return error;
1138 
1139 fail_lpfree:
1140 	kfree(lp);
1141 
1142 fail_brelse:
1143 	brelse(obh);
1144 	brelse(nbh);
1145 	return error;
1146 }
1147 
1148 /**
1149  * dir_double_exhash - Double size of ExHash table
1150  * @dip: The GFS2 dinode
1151  *
1152  * Returns: 0 on success, error code on failure
1153  */
1154 
1155 static int dir_double_exhash(struct gfs2_inode *dip)
1156 {
1157 	struct buffer_head *dibh;
1158 	u32 hsize;
1159 	u32 hsize_bytes;
1160 	__be64 *hc;
1161 	__be64 *hc2, *h;
1162 	int x;
1163 	int error = 0;
1164 
1165 	hsize = BIT(dip->i_depth);
1166 	hsize_bytes = hsize * sizeof(__be64);
1167 
1168 	hc = gfs2_dir_get_hash_table(dip);
1169 	if (IS_ERR(hc))
1170 		return PTR_ERR(hc);
1171 
1172 	hc2 = kmalloc_array(hsize_bytes, 2, GFP_NOFS | __GFP_NOWARN);
1173 	if (hc2 == NULL)
1174 		hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL);
1175 
1176 	if (!hc2)
1177 		return -ENOMEM;
1178 
1179 	h = hc2;
1180 	error = gfs2_meta_inode_buffer(dip, &dibh);
1181 	if (error)
1182 		goto out_kfree;
1183 
1184 	for (x = 0; x < hsize; x++) {
1185 		*h++ = *hc;
1186 		*h++ = *hc;
1187 		hc++;
1188 	}
1189 
1190 	error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1191 	if (error != (hsize_bytes * 2))
1192 		goto fail;
1193 
1194 	gfs2_dir_hash_inval(dip);
1195 	dip->i_hash_cache = hc2;
1196 	dip->i_depth++;
1197 	gfs2_dinode_out(dip, dibh->b_data);
1198 	brelse(dibh);
1199 	return 0;
1200 
1201 fail:
1202 	/* Replace original hash table & size */
1203 	gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1204 	i_size_write(&dip->i_inode, hsize_bytes);
1205 	gfs2_dinode_out(dip, dibh->b_data);
1206 	brelse(dibh);
1207 out_kfree:
1208 	kvfree(hc2);
1209 	return error;
1210 }
1211 
1212 /**
1213  * compare_dents - compare directory entries by hash value
1214  * @a: first dent
1215  * @b: second dent
1216  *
1217  * When comparing the hash entries of @a to @b:
1218  *   gt: returns 1
1219  *   lt: returns -1
1220  *   eq: returns 0
1221  */
1222 
1223 static int compare_dents(const void *a, const void *b)
1224 {
1225 	const struct gfs2_dirent *dent_a, *dent_b;
1226 	u32 hash_a, hash_b;
1227 	int ret = 0;
1228 
1229 	dent_a = *(const struct gfs2_dirent **)a;
1230 	hash_a = dent_a->de_cookie;
1231 
1232 	dent_b = *(const struct gfs2_dirent **)b;
1233 	hash_b = dent_b->de_cookie;
1234 
1235 	if (hash_a > hash_b)
1236 		ret = 1;
1237 	else if (hash_a < hash_b)
1238 		ret = -1;
1239 	else {
1240 		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1241 		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1242 
1243 		if (len_a > len_b)
1244 			ret = 1;
1245 		else if (len_a < len_b)
1246 			ret = -1;
1247 		else
1248 			ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1249 	}
1250 
1251 	return ret;
1252 }
1253 
1254 /**
1255  * do_filldir_main - read out directory entries
1256  * @dip: The GFS2 inode
1257  * @ctx: what to feed the entries to
1258  * @darr: an array of struct gfs2_dirent pointers to read
1259  * @entries: the number of entries in darr
1260  * @copied: pointer to int that's non-zero if a entry has been copied out
1261  *
1262  * Jump through some hoops to make sure that if there are hash collsions,
1263  * they are read out at the beginning of a buffer.  We want to minimize
1264  * the possibility that they will fall into different readdir buffers or
1265  * that someone will want to seek to that location.
1266  *
1267  * Returns: errno, >0 if the actor tells you to stop
1268  */
1269 
1270 static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1271 			   struct gfs2_dirent **darr, u32 entries,
1272 			   u32 sort_start, int *copied)
1273 {
1274 	const struct gfs2_dirent *dent, *dent_next;
1275 	u64 off, off_next;
1276 	unsigned int x, y;
1277 	int run = 0;
1278 
1279 	if (sort_start < entries)
1280 		sort(&darr[sort_start], entries - sort_start,
1281 		     sizeof(struct gfs2_dirent *), compare_dents, NULL);
1282 
1283 	dent_next = darr[0];
1284 	off_next = dent_next->de_cookie;
1285 
1286 	for (x = 0, y = 1; x < entries; x++, y++) {
1287 		dent = dent_next;
1288 		off = off_next;
1289 
1290 		if (y < entries) {
1291 			dent_next = darr[y];
1292 			off_next = dent_next->de_cookie;
1293 
1294 			if (off < ctx->pos)
1295 				continue;
1296 			ctx->pos = off;
1297 
1298 			if (off_next == off) {
1299 				if (*copied && !run)
1300 					return 1;
1301 				run = 1;
1302 			} else
1303 				run = 0;
1304 		} else {
1305 			if (off < ctx->pos)
1306 				continue;
1307 			ctx->pos = off;
1308 		}
1309 
1310 		if (!dir_emit(ctx, (const char *)(dent + 1),
1311 				be16_to_cpu(dent->de_name_len),
1312 				be64_to_cpu(dent->de_inum.no_addr),
1313 				be16_to_cpu(dent->de_type)))
1314 			return 1;
1315 
1316 		*copied = 1;
1317 	}
1318 
1319 	/* Increment the ctx->pos by one, so the next time we come into the
1320 	   do_filldir fxn, we get the next entry instead of the last one in the
1321 	   current leaf */
1322 
1323 	ctx->pos++;
1324 
1325 	return 0;
1326 }
1327 
1328 static void *gfs2_alloc_sort_buffer(unsigned size)
1329 {
1330 	void *ptr = NULL;
1331 
1332 	if (size < KMALLOC_MAX_SIZE)
1333 		ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1334 	if (!ptr)
1335 		ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
1336 	return ptr;
1337 }
1338 
1339 
1340 static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
1341 			    unsigned leaf_nr, struct gfs2_dirent **darr,
1342 			    unsigned entries)
1343 {
1344 	int sort_id = -1;
1345 	int i;
1346 
1347 	for (i = 0; i < entries; i++) {
1348 		unsigned offset;
1349 
1350 		darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
1351 		darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
1352 
1353 		if (!sdp->sd_args.ar_loccookie)
1354 			continue;
1355 		offset = (char *)(darr[i]) -
1356 			 (bh->b_data + gfs2_dirent_offset(bh->b_data));
1357 		offset /= GFS2_MIN_DIRENT_SIZE;
1358 		offset += leaf_nr * sdp->sd_max_dents_per_leaf;
1359 		if (offset >= GFS2_USE_HASH_FLAG ||
1360 		    leaf_nr >= GFS2_USE_HASH_FLAG) {
1361 			darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
1362 			if (sort_id < 0)
1363 				sort_id = i;
1364 			continue;
1365 		}
1366 		darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
1367 		darr[i]->de_cookie |= offset;
1368 	}
1369 	return sort_id;
1370 }
1371 
1372 
1373 static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
1374 			      int *copied, unsigned *depth,
1375 			      u64 leaf_no)
1376 {
1377 	struct gfs2_inode *ip = GFS2_I(inode);
1378 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1379 	struct buffer_head *bh;
1380 	struct gfs2_leaf *lf;
1381 	unsigned entries = 0, entries2 = 0;
1382 	unsigned leaves = 0, leaf = 0, offset, sort_offset;
1383 	struct gfs2_dirent **darr, *dent;
1384 	struct dirent_gather g;
1385 	struct buffer_head **larr;
1386 	int error, i, need_sort = 0, sort_id;
1387 	u64 lfn = leaf_no;
1388 
1389 	do {
1390 		error = get_leaf(ip, lfn, &bh);
1391 		if (error)
1392 			goto out;
1393 		lf = (struct gfs2_leaf *)bh->b_data;
1394 		if (leaves == 0)
1395 			*depth = be16_to_cpu(lf->lf_depth);
1396 		entries += be16_to_cpu(lf->lf_entries);
1397 		leaves++;
1398 		lfn = be64_to_cpu(lf->lf_next);
1399 		brelse(bh);
1400 	} while(lfn);
1401 
1402 	if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
1403 		need_sort = 1;
1404 		sort_offset = 0;
1405 	}
1406 
1407 	if (!entries)
1408 		return 0;
1409 
1410 	error = -ENOMEM;
1411 	/*
1412 	 * The extra 99 entries are not normally used, but are a buffer
1413 	 * zone in case the number of entries in the leaf is corrupt.
1414 	 * 99 is the maximum number of entries that can fit in a single
1415 	 * leaf block.
1416 	 */
1417 	larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1418 	if (!larr)
1419 		goto out;
1420 	darr = (struct gfs2_dirent **)(larr + leaves);
1421 	g.pdent = (const struct gfs2_dirent **)darr;
1422 	g.offset = 0;
1423 	lfn = leaf_no;
1424 
1425 	do {
1426 		error = get_leaf(ip, lfn, &bh);
1427 		if (error)
1428 			goto out_free;
1429 		lf = (struct gfs2_leaf *)bh->b_data;
1430 		lfn = be64_to_cpu(lf->lf_next);
1431 		if (lf->lf_entries) {
1432 			offset = g.offset;
1433 			entries2 += be16_to_cpu(lf->lf_entries);
1434 			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1435 						gfs2_dirent_gather, NULL, &g);
1436 			error = PTR_ERR(dent);
1437 			if (IS_ERR(dent))
1438 				goto out_free;
1439 			if (entries2 != g.offset) {
1440 				fs_warn(sdp, "Number of entries corrupt in dir "
1441 						"leaf %llu, entries2 (%u) != "
1442 						"g.offset (%u)\n",
1443 					(unsigned long long)bh->b_blocknr,
1444 					entries2, g.offset);
1445 				gfs2_consist_inode(ip);
1446 				error = -EIO;
1447 				goto out_free;
1448 			}
1449 			error = 0;
1450 			sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
1451 						   be16_to_cpu(lf->lf_entries));
1452 			if (!need_sort && sort_id >= 0) {
1453 				need_sort = 1;
1454 				sort_offset = offset + sort_id;
1455 			}
1456 			larr[leaf++] = bh;
1457 		} else {
1458 			larr[leaf++] = NULL;
1459 			brelse(bh);
1460 		}
1461 	} while(lfn);
1462 
1463 	BUG_ON(entries2 != entries);
1464 	error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
1465 				sort_offset : entries, copied);
1466 out_free:
1467 	for(i = 0; i < leaf; i++)
1468 		if (larr[i])
1469 			brelse(larr[i]);
1470 	kvfree(larr);
1471 out:
1472 	return error;
1473 }
1474 
1475 /**
1476  * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1477  *
1478  * Note: we can't calculate each index like dir_e_read can because we don't
1479  * have the leaf, and therefore we don't have the depth, and therefore we
1480  * don't have the length. So we have to just read enough ahead to make up
1481  * for the loss of information.
1482  */
1483 static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1484 			       struct file_ra_state *f_ra)
1485 {
1486 	struct gfs2_inode *ip = GFS2_I(inode);
1487 	struct gfs2_glock *gl = ip->i_gl;
1488 	struct buffer_head *bh;
1489 	u64 blocknr = 0, last;
1490 	unsigned count;
1491 
1492 	/* First check if we've already read-ahead for the whole range. */
1493 	if (index + MAX_RA_BLOCKS < f_ra->start)
1494 		return;
1495 
1496 	f_ra->start = max((pgoff_t)index, f_ra->start);
1497 	for (count = 0; count < MAX_RA_BLOCKS; count++) {
1498 		if (f_ra->start >= hsize) /* if exceeded the hash table */
1499 			break;
1500 
1501 		last = blocknr;
1502 		blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1503 		f_ra->start++;
1504 		if (blocknr == last)
1505 			continue;
1506 
1507 		bh = gfs2_getbuf(gl, blocknr, 1);
1508 		if (trylock_buffer(bh)) {
1509 			if (buffer_uptodate(bh)) {
1510 				unlock_buffer(bh);
1511 				brelse(bh);
1512 				continue;
1513 			}
1514 			bh->b_end_io = end_buffer_read_sync;
1515 			submit_bh(REQ_OP_READ,
1516 				  REQ_RAHEAD | REQ_META | REQ_PRIO,
1517 				  bh);
1518 			continue;
1519 		}
1520 		brelse(bh);
1521 	}
1522 }
1523 
1524 /**
1525  * dir_e_read - Reads the entries from a directory into a filldir buffer
1526  * @dip: dinode pointer
1527  * @ctx: actor to feed the entries to
1528  *
1529  * Returns: errno
1530  */
1531 
1532 static int dir_e_read(struct inode *inode, struct dir_context *ctx,
1533 		      struct file_ra_state *f_ra)
1534 {
1535 	struct gfs2_inode *dip = GFS2_I(inode);
1536 	u32 hsize, len = 0;
1537 	u32 hash, index;
1538 	__be64 *lp;
1539 	int copied = 0;
1540 	int error = 0;
1541 	unsigned depth = 0;
1542 
1543 	hsize = BIT(dip->i_depth);
1544 	hash = gfs2_dir_offset2hash(ctx->pos);
1545 	index = hash >> (32 - dip->i_depth);
1546 
1547 	if (dip->i_hash_cache == NULL)
1548 		f_ra->start = 0;
1549 	lp = gfs2_dir_get_hash_table(dip);
1550 	if (IS_ERR(lp))
1551 		return PTR_ERR(lp);
1552 
1553 	gfs2_dir_readahead(inode, hsize, index, f_ra);
1554 
1555 	while (index < hsize) {
1556 		error = gfs2_dir_read_leaf(inode, ctx,
1557 					   &copied, &depth,
1558 					   be64_to_cpu(lp[index]));
1559 		if (error)
1560 			break;
1561 
1562 		len = BIT(dip->i_depth - depth);
1563 		index = (index & ~(len - 1)) + len;
1564 	}
1565 
1566 	if (error > 0)
1567 		error = 0;
1568 	return error;
1569 }
1570 
1571 int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
1572 		  struct file_ra_state *f_ra)
1573 {
1574 	struct gfs2_inode *dip = GFS2_I(inode);
1575 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1576 	struct dirent_gather g;
1577 	struct gfs2_dirent **darr, *dent;
1578 	struct buffer_head *dibh;
1579 	int copied = 0;
1580 	int error;
1581 
1582 	if (!dip->i_entries)
1583 		return 0;
1584 
1585 	if (dip->i_diskflags & GFS2_DIF_EXHASH)
1586 		return dir_e_read(inode, ctx, f_ra);
1587 
1588 	if (!gfs2_is_stuffed(dip)) {
1589 		gfs2_consist_inode(dip);
1590 		return -EIO;
1591 	}
1592 
1593 	error = gfs2_meta_inode_buffer(dip, &dibh);
1594 	if (error)
1595 		return error;
1596 
1597 	error = -ENOMEM;
1598 	/* 96 is max number of dirents which can be stuffed into an inode */
1599 	darr = kmalloc_array(96, sizeof(struct gfs2_dirent *), GFP_NOFS);
1600 	if (darr) {
1601 		g.pdent = (const struct gfs2_dirent **)darr;
1602 		g.offset = 0;
1603 		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1604 					gfs2_dirent_gather, NULL, &g);
1605 		if (IS_ERR(dent)) {
1606 			error = PTR_ERR(dent);
1607 			goto out;
1608 		}
1609 		if (dip->i_entries != g.offset) {
1610 			fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1611 				"ip->i_entries (%u) != g.offset (%u)\n",
1612 				(unsigned long long)dip->i_no_addr,
1613 				dip->i_entries,
1614 				g.offset);
1615 			gfs2_consist_inode(dip);
1616 			error = -EIO;
1617 			goto out;
1618 		}
1619 		gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
1620 		error = do_filldir_main(dip, ctx, darr,
1621 					dip->i_entries, 0, &copied);
1622 out:
1623 		kfree(darr);
1624 	}
1625 
1626 	if (error > 0)
1627 		error = 0;
1628 
1629 	brelse(dibh);
1630 
1631 	return error;
1632 }
1633 
1634 /**
1635  * gfs2_dir_search - Search a directory
1636  * @dip: The GFS2 dir inode
1637  * @name: The name we are looking up
1638  * @fail_on_exist: Fail if the name exists rather than looking it up
1639  *
1640  * This routine searches a directory for a file or another directory.
1641  * Assumes a glock is held on dip.
1642  *
1643  * Returns: errno
1644  */
1645 
1646 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
1647 			      bool fail_on_exist)
1648 {
1649 	struct buffer_head *bh;
1650 	struct gfs2_dirent *dent;
1651 	u64 addr, formal_ino;
1652 	u16 dtype;
1653 
1654 	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1655 	if (dent) {
1656 		struct inode *inode;
1657 		u16 rahead;
1658 
1659 		if (IS_ERR(dent))
1660 			return ERR_CAST(dent);
1661 		dtype = be16_to_cpu(dent->de_type);
1662 		rahead = be16_to_cpu(dent->de_rahead);
1663 		addr = be64_to_cpu(dent->de_inum.no_addr);
1664 		formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
1665 		brelse(bh);
1666 		if (fail_on_exist)
1667 			return ERR_PTR(-EEXIST);
1668 		inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino,
1669 					  GFS2_BLKST_FREE /* ignore */);
1670 		if (!IS_ERR(inode))
1671 			GFS2_I(inode)->i_rahead = rahead;
1672 		return inode;
1673 	}
1674 	return ERR_PTR(-ENOENT);
1675 }
1676 
1677 int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1678 		   const struct gfs2_inode *ip)
1679 {
1680 	struct buffer_head *bh;
1681 	struct gfs2_dirent *dent;
1682 	int ret = -ENOENT;
1683 
1684 	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1685 	if (dent) {
1686 		if (IS_ERR(dent))
1687 			return PTR_ERR(dent);
1688 		if (ip) {
1689 			if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1690 				goto out;
1691 			if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1692 			    ip->i_no_formal_ino)
1693 				goto out;
1694 			if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1695 			    be16_to_cpu(dent->de_type))) {
1696 				gfs2_consist_inode(GFS2_I(dir));
1697 				ret = -EIO;
1698 				goto out;
1699 			}
1700 		}
1701 		ret = 0;
1702 out:
1703 		brelse(bh);
1704 	}
1705 	return ret;
1706 }
1707 
1708 /**
1709  * dir_new_leaf - Add a new leaf onto hash chain
1710  * @inode: The directory
1711  * @name: The name we are adding
1712  *
1713  * This adds a new dir leaf onto an existing leaf when there is not
1714  * enough space to add a new dir entry. This is a last resort after
1715  * we've expanded the hash table to max size and also split existing
1716  * leaf blocks, so it will only occur for very large directories.
1717  *
1718  * The dist parameter is set to 1 for leaf blocks directly attached
1719  * to the hash table, 2 for one layer of indirection, 3 for two layers
1720  * etc. We are thus able to tell the difference between an old leaf
1721  * with dist set to zero (i.e. "don't know") and a new one where we
1722  * set this information for debug/fsck purposes.
1723  *
1724  * Returns: 0 on success, or -ve on error
1725  */
1726 
1727 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1728 {
1729 	struct buffer_head *bh, *obh;
1730 	struct gfs2_inode *ip = GFS2_I(inode);
1731 	struct gfs2_leaf *leaf, *oleaf;
1732 	u32 dist = 1;
1733 	int error;
1734 	u32 index;
1735 	u64 bn;
1736 
1737 	index = name->hash >> (32 - ip->i_depth);
1738 	error = get_first_leaf(ip, index, &obh);
1739 	if (error)
1740 		return error;
1741 	do {
1742 		dist++;
1743 		oleaf = (struct gfs2_leaf *)obh->b_data;
1744 		bn = be64_to_cpu(oleaf->lf_next);
1745 		if (!bn)
1746 			break;
1747 		brelse(obh);
1748 		error = get_leaf(ip, bn, &obh);
1749 		if (error)
1750 			return error;
1751 	} while(1);
1752 
1753 	gfs2_trans_add_meta(ip->i_gl, obh);
1754 
1755 	leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1756 	if (!leaf) {
1757 		brelse(obh);
1758 		return -ENOSPC;
1759 	}
1760 	leaf->lf_dist = cpu_to_be32(dist);
1761 	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1762 	brelse(bh);
1763 	brelse(obh);
1764 
1765 	error = gfs2_meta_inode_buffer(ip, &bh);
1766 	if (error)
1767 		return error;
1768 	gfs2_trans_add_meta(ip->i_gl, bh);
1769 	gfs2_add_inode_blocks(&ip->i_inode, 1);
1770 	gfs2_dinode_out(ip, bh->b_data);
1771 	brelse(bh);
1772 	return 0;
1773 }
1774 
1775 static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip)
1776 {
1777 	u64 where = ip->i_no_addr + 1;
1778 	if (ip->i_eattr == where)
1779 		return 1;
1780 	return 0;
1781 }
1782 
1783 /**
1784  * gfs2_dir_add - Add new filename into directory
1785  * @inode: The directory inode
1786  * @name: The new name
1787  * @nip: The GFS2 inode to be linked in to the directory
1788  * @da: The directory addition info
1789  *
1790  * If the call to gfs2_diradd_alloc_required resulted in there being
1791  * no need to allocate any new directory blocks, then it will contain
1792  * a pointer to the directory entry and the bh in which it resides. We
1793  * can use that without having to repeat the search. If there was no
1794  * free space, then we must now create more space.
1795  *
1796  * Returns: 0 on success, error code on failure
1797  */
1798 
1799 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1800 		 const struct gfs2_inode *nip, struct gfs2_diradd *da)
1801 {
1802 	struct gfs2_inode *ip = GFS2_I(inode);
1803 	struct buffer_head *bh = da->bh;
1804 	struct gfs2_dirent *dent = da->dent;
1805 	struct timespec64 tv;
1806 	struct gfs2_leaf *leaf;
1807 	int error;
1808 
1809 	while(1) {
1810 		if (da->bh == NULL) {
1811 			dent = gfs2_dirent_search(inode, name,
1812 						  gfs2_dirent_find_space, &bh);
1813 		}
1814 		if (dent) {
1815 			if (IS_ERR(dent))
1816 				return PTR_ERR(dent);
1817 			dent = gfs2_init_dirent(inode, dent, name, bh);
1818 			gfs2_inum_out(nip, dent);
1819 			dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1820 			dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip));
1821 			tv = current_time(&ip->i_inode);
1822 			if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1823 				leaf = (struct gfs2_leaf *)bh->b_data;
1824 				be16_add_cpu(&leaf->lf_entries, 1);
1825 				leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1826 				leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1827 			}
1828 			da->dent = NULL;
1829 			da->bh = NULL;
1830 			brelse(bh);
1831 			ip->i_entries++;
1832 			ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv;
1833 			if (S_ISDIR(nip->i_inode.i_mode))
1834 				inc_nlink(&ip->i_inode);
1835 			mark_inode_dirty(inode);
1836 			error = 0;
1837 			break;
1838 		}
1839 		if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1840 			error = dir_make_exhash(inode);
1841 			if (error)
1842 				break;
1843 			continue;
1844 		}
1845 		error = dir_split_leaf(inode, name);
1846 		if (error == 0)
1847 			continue;
1848 		if (error < 0)
1849 			break;
1850 		if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1851 			error = dir_double_exhash(ip);
1852 			if (error)
1853 				break;
1854 			error = dir_split_leaf(inode, name);
1855 			if (error < 0)
1856 				break;
1857 			if (error == 0)
1858 				continue;
1859 		}
1860 		error = dir_new_leaf(inode, name);
1861 		if (!error)
1862 			continue;
1863 		error = -ENOSPC;
1864 		break;
1865 	}
1866 	return error;
1867 }
1868 
1869 
1870 /**
1871  * gfs2_dir_del - Delete a directory entry
1872  * @dip: The GFS2 inode
1873  * @filename: The filename
1874  *
1875  * Returns: 0 on success, error code on failure
1876  */
1877 
1878 int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
1879 {
1880 	const struct qstr *name = &dentry->d_name;
1881 	struct gfs2_dirent *dent, *prev = NULL;
1882 	struct buffer_head *bh;
1883 	struct timespec64 tv = current_time(&dip->i_inode);
1884 
1885 	/* Returns _either_ the entry (if its first in block) or the
1886 	   previous entry otherwise */
1887 	dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1888 	if (!dent) {
1889 		gfs2_consist_inode(dip);
1890 		return -EIO;
1891 	}
1892 	if (IS_ERR(dent)) {
1893 		gfs2_consist_inode(dip);
1894 		return PTR_ERR(dent);
1895 	}
1896 	/* If not first in block, adjust pointers accordingly */
1897 	if (gfs2_dirent_find(dent, name, NULL) == 0) {
1898 		prev = dent;
1899 		dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1900 	}
1901 
1902 	dirent_del(dip, bh, prev, dent);
1903 	if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1904 		struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1905 		u16 entries = be16_to_cpu(leaf->lf_entries);
1906 		if (!entries)
1907 			gfs2_consist_inode(dip);
1908 		leaf->lf_entries = cpu_to_be16(--entries);
1909 		leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1910 		leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1911 	}
1912 	brelse(bh);
1913 
1914 	if (!dip->i_entries)
1915 		gfs2_consist_inode(dip);
1916 	dip->i_entries--;
1917 	dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv;
1918 	if (d_is_dir(dentry))
1919 		drop_nlink(&dip->i_inode);
1920 	mark_inode_dirty(&dip->i_inode);
1921 
1922 	return 0;
1923 }
1924 
1925 /**
1926  * gfs2_dir_mvino - Change inode number of directory entry
1927  * @dip: The GFS2 inode
1928  * @filename:
1929  * @new_inode:
1930  *
1931  * This routine changes the inode number of a directory entry.  It's used
1932  * by rename to change ".." when a directory is moved.
1933  * Assumes a glock is held on dvp.
1934  *
1935  * Returns: errno
1936  */
1937 
1938 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1939 		   const struct gfs2_inode *nip, unsigned int new_type)
1940 {
1941 	struct buffer_head *bh;
1942 	struct gfs2_dirent *dent;
1943 
1944 	dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1945 	if (!dent) {
1946 		gfs2_consist_inode(dip);
1947 		return -EIO;
1948 	}
1949 	if (IS_ERR(dent))
1950 		return PTR_ERR(dent);
1951 
1952 	gfs2_trans_add_meta(dip->i_gl, bh);
1953 	gfs2_inum_out(nip, dent);
1954 	dent->de_type = cpu_to_be16(new_type);
1955 	brelse(bh);
1956 
1957 	dip->i_inode.i_mtime = dip->i_inode.i_ctime = current_time(&dip->i_inode);
1958 	mark_inode_dirty_sync(&dip->i_inode);
1959 	return 0;
1960 }
1961 
1962 /**
1963  * leaf_dealloc - Deallocate a directory leaf
1964  * @dip: the directory
1965  * @index: the hash table offset in the directory
1966  * @len: the number of pointers to this leaf
1967  * @leaf_no: the leaf number
1968  * @leaf_bh: buffer_head for the starting leaf
1969  * last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1970  *
1971  * Returns: errno
1972  */
1973 
1974 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1975 			u64 leaf_no, struct buffer_head *leaf_bh,
1976 			int last_dealloc)
1977 {
1978 	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1979 	struct gfs2_leaf *tmp_leaf;
1980 	struct gfs2_rgrp_list rlist;
1981 	struct buffer_head *bh, *dibh;
1982 	u64 blk, nblk;
1983 	unsigned int rg_blocks = 0, l_blocks = 0;
1984 	char *ht;
1985 	unsigned int x, size = len * sizeof(u64);
1986 	int error;
1987 
1988 	error = gfs2_rindex_update(sdp);
1989 	if (error)
1990 		return error;
1991 
1992 	memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1993 
1994 	ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
1995 	if (ht == NULL)
1996 		ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO,
1997 			       PAGE_KERNEL);
1998 	if (!ht)
1999 		return -ENOMEM;
2000 
2001 	error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
2002 	if (error)
2003 		goto out;
2004 
2005 	/*  Count the number of leaves  */
2006 	bh = leaf_bh;
2007 
2008 	for (blk = leaf_no; blk; blk = nblk) {
2009 		if (blk != leaf_no) {
2010 			error = get_leaf(dip, blk, &bh);
2011 			if (error)
2012 				goto out_rlist;
2013 		}
2014 		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2015 		nblk = be64_to_cpu(tmp_leaf->lf_next);
2016 		if (blk != leaf_no)
2017 			brelse(bh);
2018 
2019 		gfs2_rlist_add(dip, &rlist, blk);
2020 		l_blocks++;
2021 	}
2022 
2023 	gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
2024 
2025 	for (x = 0; x < rlist.rl_rgrps; x++) {
2026 		struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl);
2027 
2028 		rg_blocks += rgd->rd_length;
2029 	}
2030 
2031 	error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
2032 	if (error)
2033 		goto out_rlist;
2034 
2035 	error = gfs2_trans_begin(sdp,
2036 			rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
2037 			RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
2038 	if (error)
2039 		goto out_rg_gunlock;
2040 
2041 	bh = leaf_bh;
2042 
2043 	for (blk = leaf_no; blk; blk = nblk) {
2044 		if (blk != leaf_no) {
2045 			error = get_leaf(dip, blk, &bh);
2046 			if (error)
2047 				goto out_end_trans;
2048 		}
2049 		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2050 		nblk = be64_to_cpu(tmp_leaf->lf_next);
2051 		if (blk != leaf_no)
2052 			brelse(bh);
2053 
2054 		gfs2_free_meta(dip, blk, 1);
2055 		gfs2_add_inode_blocks(&dip->i_inode, -1);
2056 	}
2057 
2058 	error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
2059 	if (error != size) {
2060 		if (error >= 0)
2061 			error = -EIO;
2062 		goto out_end_trans;
2063 	}
2064 
2065 	error = gfs2_meta_inode_buffer(dip, &dibh);
2066 	if (error)
2067 		goto out_end_trans;
2068 
2069 	gfs2_trans_add_meta(dip->i_gl, dibh);
2070 	/* On the last dealloc, make this a regular file in case we crash.
2071 	   (We don't want to free these blocks a second time.)  */
2072 	if (last_dealloc)
2073 		dip->i_inode.i_mode = S_IFREG;
2074 	gfs2_dinode_out(dip, dibh->b_data);
2075 	brelse(dibh);
2076 
2077 out_end_trans:
2078 	gfs2_trans_end(sdp);
2079 out_rg_gunlock:
2080 	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2081 out_rlist:
2082 	gfs2_rlist_free(&rlist);
2083 	gfs2_quota_unhold(dip);
2084 out:
2085 	kvfree(ht);
2086 	return error;
2087 }
2088 
2089 /**
2090  * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2091  * @dip: the directory
2092  *
2093  * Dealloc all on-disk directory leaves to FREEMETA state
2094  * Change on-disk inode type to "regular file"
2095  *
2096  * Returns: errno
2097  */
2098 
2099 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
2100 {
2101 	struct buffer_head *bh;
2102 	struct gfs2_leaf *leaf;
2103 	u32 hsize, len;
2104 	u32 index = 0, next_index;
2105 	__be64 *lp;
2106 	u64 leaf_no;
2107 	int error = 0, last;
2108 
2109 	hsize = BIT(dip->i_depth);
2110 
2111 	lp = gfs2_dir_get_hash_table(dip);
2112 	if (IS_ERR(lp))
2113 		return PTR_ERR(lp);
2114 
2115 	while (index < hsize) {
2116 		leaf_no = be64_to_cpu(lp[index]);
2117 		if (leaf_no) {
2118 			error = get_leaf(dip, leaf_no, &bh);
2119 			if (error)
2120 				goto out;
2121 			leaf = (struct gfs2_leaf *)bh->b_data;
2122 			len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth));
2123 
2124 			next_index = (index & ~(len - 1)) + len;
2125 			last = ((next_index >= hsize) ? 1 : 0);
2126 			error = leaf_dealloc(dip, index, len, leaf_no, bh,
2127 					     last);
2128 			brelse(bh);
2129 			if (error)
2130 				goto out;
2131 			index = next_index;
2132 		} else
2133 			index++;
2134 	}
2135 
2136 	if (index != hsize) {
2137 		gfs2_consist_inode(dip);
2138 		error = -EIO;
2139 	}
2140 
2141 out:
2142 
2143 	return error;
2144 }
2145 
2146 /**
2147  * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2148  * @ip: the file being written to
2149  * @filname: the filename that's going to be added
2150  * @da: The structure to return dir alloc info
2151  *
2152  * Returns: 0 if ok, -ve on error
2153  */
2154 
2155 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
2156 			       struct gfs2_diradd *da)
2157 {
2158 	struct gfs2_inode *ip = GFS2_I(inode);
2159 	struct gfs2_sbd *sdp = GFS2_SB(inode);
2160 	const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2161 	struct gfs2_dirent *dent;
2162 	struct buffer_head *bh;
2163 
2164 	da->nr_blocks = 0;
2165 	da->bh = NULL;
2166 	da->dent = NULL;
2167 
2168 	dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2169 	if (!dent) {
2170 		da->nr_blocks = sdp->sd_max_dirres;
2171 		if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
2172 		    (GFS2_DIRENT_SIZE(name->len) < extra))
2173 			da->nr_blocks = 1;
2174 		return 0;
2175 	}
2176 	if (IS_ERR(dent))
2177 		return PTR_ERR(dent);
2178 
2179 	if (da->save_loc) {
2180 		da->bh = bh;
2181 		da->dent = dent;
2182 	} else {
2183 		brelse(bh);
2184 	}
2185 	return 0;
2186 }
2187 
2188