dir.c 49.6 KB
Newer Older
D
David Teigland 已提交
1 2
/*
 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3
 * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
D
David Teigland 已提交
4 5 6
 *
 * This copyrighted material is made available to anyone wishing to use,
 * modify, copy, or redistribute it subject to the terms and conditions
7
 * of the GNU General Public License version 2.
D
David Teigland 已提交
8 9 10
 */

/*
11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38
 * Implements Extendible Hashing as described in:
 *   "Extendible Hashing" by Fagin, et al in
 *     __ACM Trans. on Database Systems__, Sept 1979.
 *
 *
 * Here's the layout of dirents which is essentially the same as that of ext2
 * within a single block. The field de_name_len is the number of bytes
 * actually required for the name (no null terminator). The field de_rec_len
 * is the number of bytes allocated to the dirent. The offset of the next
 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
 * deleted, the preceding dirent inherits its allocated space, ie
 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
 * by adding de_rec_len to the current dirent, this essentially causes the
 * deleted dirent to get jumped over when iterating through all the dirents.
 *
 * When deleting the first dirent in a block, there is no previous dirent so
 * the field de_ino is set to zero to designate it as deleted. When allocating
 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
 * dirent is allocated. Otherwise it must go through all the 'used' dirents
 * searching for one in which the amount of total space minus the amount of
 * used space will provide enough space for the new dirent.
 *
 * There are two types of blocks in which dirents reside. In a stuffed dinode,
 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
 * the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
 * beginning of the leaf block. The dirents reside in leaves when
 *
39
 * dip->i_diskflags & GFS2_DIF_EXHASH is true
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54
 *
 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
 *
 * When the dirents are in leaves, the actual contents of the directory file are
 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
 * dirents are NOT in the directory file itself. There can be more than one
 * block pointer in the array that points to the same leaf. In fact, when a
 * directory is first converted from linear to exhash, all of the pointers
 * point to the same leaf.
 *
 * When a leaf is completely full, the size of the hash table can be
 * doubled unless it is already at the maximum size which is hard coded into
 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
 * but never before the maximum hash table size has been reached.
 */
D
David Teigland 已提交
55 56 57 58 59

#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/buffer_head.h>
#include <linux/sort.h>
60
#include <linux/gfs2_ondisk.h>
61
#include <linux/crc32.h>
62
#include <linux/vmalloc.h>
D
David Teigland 已提交
63 64

#include "gfs2.h"
65
#include "incore.h"
D
David Teigland 已提交
66 67 68 69 70 71 72
#include "dir.h"
#include "glock.h"
#include "inode.h"
#include "meta_io.h"
#include "quota.h"
#include "rgrp.h"
#include "trans.h"
73
#include "bmap.h"
74
#include "util.h"
D
David Teigland 已提交
75 76 77 78

#define IS_LEAF     1 /* Hashed (leaf) directory */
#define IS_DINODE   2 /* Linear (stuffed dinode block) directory */

79 80
#define MAX_RA_BLOCKS 32 /* max read-ahead blocks */

81 82
#define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
#define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
D
David Teigland 已提交
83

84 85 86
struct qstr gfs2_qdot __read_mostly;
struct qstr gfs2_qdotdot __read_mostly;

87 88
typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
			    const struct qstr *name, void *opaque);
D
David Teigland 已提交
89

90
int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
91
			    struct buffer_head **bhp)
92 93 94
{
	struct buffer_head *bh;

95
	bh = gfs2_meta_new(ip->i_gl, block);
96
	gfs2_trans_add_meta(ip->i_gl, bh);
97 98
	gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
99 100 101 102
	*bhp = bh;
	return 0;
}

103
static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
104 105 106 107
					struct buffer_head **bhp)
{
	struct buffer_head *bh;
	int error;
108

S
Steven Whitehouse 已提交
109
	error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
110 111
	if (error)
		return error;
112
	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
113 114 115 116 117 118
		brelse(bh);
		return -EIO;
	}
	*bhp = bh;
	return 0;
}
119 120 121 122 123 124 125 126 127 128 129

static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
				  unsigned int offset, unsigned int size)
{
	struct buffer_head *dibh;
	int error;

	error = gfs2_meta_inode_buffer(ip, &dibh);
	if (error)
		return error;

130
	gfs2_trans_add_meta(ip->i_gl, dibh);
131
	memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
S
Steven Whitehouse 已提交
132 133
	if (ip->i_inode.i_size < offset + size)
		i_size_write(&ip->i_inode, offset + size);
134
	ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
135
	gfs2_dinode_out(ip, dibh->b_data);
136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153

	brelse(dibh);

	return size;
}



/**
 * gfs2_dir_write_data - Write directory information to the inode
 * @ip: The GFS2 inode
 * @buf: The buffer containing information to be written
 * @offset: The file offset to start writing at
 * @size: The amount of data to write
 *
 * Returns: The number of bytes correctly written or error code
 */
static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
154
			       u64 offset, unsigned int size)
155
{
156
	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
157
	struct buffer_head *dibh;
158 159
	u64 lblock, dblock;
	u32 extlen = 0;
160 161 162
	unsigned int o;
	int copied = 0;
	int error = 0;
163
	int new = 0;
164 165 166 167 168 169

	if (!size)
		return 0;

	if (gfs2_is_stuffed(ip) &&
	    offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
170 171
		return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
					      size);
172 173 174 175 176

	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
		return -EINVAL;

	if (gfs2_is_stuffed(ip)) {
177
		error = gfs2_unstuff_dinode(ip, NULL);
178
		if (error)
179
			return error;
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194
	}

	lblock = offset;
	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);

	while (copied < size) {
		unsigned int amount;
		struct buffer_head *bh;

		amount = size - copied;
		if (amount > sdp->sd_sb.sb_bsize - o)
			amount = sdp->sd_sb.sb_bsize - o;

		if (!extlen) {
			new = 1;
195
			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
S
Steven Whitehouse 已提交
196
						&dblock, &extlen);
197 198 199 200 201 202 203
			if (error)
				goto fail;
			error = -EIO;
			if (gfs2_assert_withdraw(sdp, dblock))
				goto fail;
		}

204 205 206 207 208
		if (amount == sdp->sd_jbsize || new)
			error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
		else
			error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);

209 210 211
		if (error)
			goto fail;

212
		gfs2_trans_add_meta(ip->i_gl, bh);
213 214 215
		memcpy(bh->b_data + o, buf, amount);
		brelse(bh);

216
		buf += amount;
217 218 219 220 221 222 223 224 225 226 227 228 229
		copied += amount;
		lblock++;
		dblock++;
		extlen--;

		o = sizeof(struct gfs2_meta_header);
	}

out:
	error = gfs2_meta_inode_buffer(ip, &dibh);
	if (error)
		return error;

S
Steven Whitehouse 已提交
230 231
	if (ip->i_inode.i_size < offset + copied)
		i_size_write(&ip->i_inode, offset + copied);
232
	ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
233

234
	gfs2_trans_add_meta(ip->i_gl, dibh);
235
	gfs2_dinode_out(ip, dibh->b_data);
236 237 238 239 240 241 242 243 244
	brelse(dibh);

	return copied;
fail:
	if (copied)
		goto out;
	return error;
}

245 246
static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
				 unsigned int size)
247 248 249 250 251 252
{
	struct buffer_head *dibh;
	int error;

	error = gfs2_meta_inode_buffer(ip, &dibh);
	if (!error) {
253
		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268
		brelse(dibh);
	}

	return (error) ? error : size;
}


/**
 * gfs2_dir_read_data - Read a data from a directory inode
 * @ip: The GFS2 Inode
 * @buf: The buffer to place result into
 * @size: Amount of data to transfer
 *
 * Returns: The amount of data actually copied or the error
 */
269 270
static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
			      unsigned int size)
271
{
272
	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
273 274
	u64 lblock, dblock;
	u32 extlen = 0;
275 276 277 278 279
	unsigned int o;
	int copied = 0;
	int error = 0;

	if (gfs2_is_stuffed(ip))
280
		return gfs2_dir_read_stuffed(ip, buf, size);
281 282 283 284

	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
		return -EINVAL;

285
	lblock = 0;
286 287 288 289 290 291 292 293 294 295 296 297 298
	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);

	while (copied < size) {
		unsigned int amount;
		struct buffer_head *bh;
		int new;

		amount = size - copied;
		if (amount > sdp->sd_sb.sb_bsize - o)
			amount = sdp->sd_sb.sb_bsize - o;

		if (!extlen) {
			new = 0;
299
			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
S
Steven Whitehouse 已提交
300
						&dblock, &extlen);
S
Steven Whitehouse 已提交
301
			if (error || !dblock)
302
				goto fail;
S
Steven Whitehouse 已提交
303 304
			BUG_ON(extlen < 1);
			bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
305
		} else {
S
Steven Whitehouse 已提交
306
			error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
307 308
			if (error)
				goto fail;
S
Steven Whitehouse 已提交
309 310 311 312 313 314 315 316
		}
		error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
		if (error) {
			brelse(bh);
			goto fail;
		}
		dblock++;
		extlen--;
317 318
		memcpy(buf, bh->b_data + o, amount);
		brelse(bh);
319
		buf += (amount/sizeof(__be64));
320 321 322 323 324 325 326 327 328 329
		copied += amount;
		lblock++;
		o = sizeof(struct gfs2_meta_header);
	}

	return copied;
fail:
	return (copied) ? copied : error;
}

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
/**
 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
 * @ip: The inode in question
 *
 * Returns: The hash table or an error
 */

static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
{
	struct inode *inode = &ip->i_inode;
	int ret;
	u32 hsize;
	__be64 *hc;

	BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));

	hc = ip->i_hash_cache;
	if (hc)
		return hc;

	hsize = 1 << ip->i_depth;
	hsize *= sizeof(__be64);
	if (hsize != i_size_read(&ip->i_inode)) {
		gfs2_consist_inode(ip);
		return ERR_PTR(-EIO);
	}

357 358 359 360
	hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
	if (hc == NULL)
		hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL);

361 362 363
	if (hc == NULL)
		return ERR_PTR(-ENOMEM);

364
	ret = gfs2_dir_read_data(ip, hc, hsize);
365
	if (ret < 0) {
366 367 368 369
		if (is_vmalloc_addr(hc))
			vfree(hc);
		else
			kfree(hc);
370 371 372 373
		return ERR_PTR(ret);
	}

	spin_lock(&inode->i_lock);
374 375 376 377 378 379
	if (ip->i_hash_cache) {
		if (is_vmalloc_addr(hc))
			vfree(hc);
		else
			kfree(hc);
	} else {
380
		ip->i_hash_cache = hc;
381
	}
382 383 384 385 386 387 388 389 390 391 392 393 394 395 396
	spin_unlock(&inode->i_lock);

	return ip->i_hash_cache;
}

/**
 * gfs2_dir_hash_inval - Invalidate dir hash
 * @ip: The directory inode
 *
 * Must be called with an exclusive glock, or during glock invalidation.
 */
void gfs2_dir_hash_inval(struct gfs2_inode *ip)
{
	__be64 *hc = ip->i_hash_cache;
	ip->i_hash_cache = NULL;
397 398 399 400
	if (is_vmalloc_addr(hc))
		vfree(hc);
	else
		kfree(hc);
401 402
}

403 404 405 406 407
static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
{
	return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
}

408 409 410
static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
				     const struct qstr *name, int ret)
{
411
	if (!gfs2_dirent_sentinel(dent) &&
412 413
	    be32_to_cpu(dent->de_hash) == name->hash &&
	    be16_to_cpu(dent->de_name_len) == name->len &&
414
	    memcmp(dent+1, name->name, name->len) == 0)
415 416 417 418 419
		return ret;
	return 0;
}

static int gfs2_dirent_find(const struct gfs2_dirent *dent,
420 421
			    const struct qstr *name,
			    void *opaque)
422 423 424 425 426
{
	return __gfs2_dirent_find(dent, name, 1);
}

static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
427 428
			    const struct qstr *name,
			    void *opaque)
429 430 431 432 433 434 435
{
	return __gfs2_dirent_find(dent, name, 2);
}

/*
 * name->name holds ptr to start of block.
 * name->len holds size of block.
D
David Teigland 已提交
436
 */
437
static int gfs2_dirent_last(const struct gfs2_dirent *dent,
438 439
			    const struct qstr *name,
			    void *opaque)
440 441 442 443 444 445 446
{
	const char *start = name->name;
	const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
	if (name->len == (end - start))
		return 1;
	return 0;
}
D
David Teigland 已提交
447

448
static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
449 450
				  const struct qstr *name,
				  void *opaque)
D
David Teigland 已提交
451
{
452 453 454 455
	unsigned required = GFS2_DIRENT_SIZE(name->len);
	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
	unsigned totlen = be16_to_cpu(dent->de_rec_len);

456
	if (gfs2_dirent_sentinel(dent))
B
Bob Peterson 已提交
457
		actual = 0;
J
Jan Engelhardt 已提交
458
	if (totlen - actual >= required)
459 460 461 462
		return 1;
	return 0;
}

463 464 465 466 467 468 469 470 471 472
struct dirent_gather {
	const struct gfs2_dirent **pdent;
	unsigned offset;
};

static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
			      const struct qstr *name,
			      void *opaque)
{
	struct dirent_gather *g = opaque;
473
	if (!gfs2_dirent_sentinel(dent)) {
474 475 476 477 478
		g->pdent[g->offset++] = dent;
	}
	return 0;
}

479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
/*
 * Other possible things to check:
 * - Inode located within filesystem size (and on valid block)
 * - Valid directory entry type
 * Not sure how heavy-weight we want to make this... could also check
 * hash is correct for example, but that would take a lot of extra time.
 * For now the most important thing is to check that the various sizes
 * are correct.
 */
static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
			     unsigned int size, unsigned int len, int first)
{
	const char *msg = "gfs2_dirent too small";
	if (unlikely(size < sizeof(struct gfs2_dirent)))
		goto error;
	msg = "gfs2_dirent misaligned";
	if (unlikely(offset & 0x7))
		goto error;
	msg = "gfs2_dirent points beyond end of block";
	if (unlikely(offset + size > len))
		goto error;
	msg = "zero inode number";
501
	if (unlikely(!first && gfs2_dirent_sentinel(dent)))
502 503
		goto error;
	msg = "name length is greater than space in dirent";
504
	if (!gfs2_dirent_sentinel(dent) &&
505 506 507 508 509 510 511 512
	    unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
		     size))
		goto error;
	return 0;
error:
	printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
	       first ? "first in block" : "not first in block");
	return -EIO;
D
David Teigland 已提交
513 514
}

515
static int gfs2_dirent_offset(const void *buf)
516
{
517 518
	const struct gfs2_meta_header *h = buf;
	int offset;
519 520 521

	BUG_ON(buf == NULL);

522
	switch(be32_to_cpu(h->mh_type)) {
523 524 525 526 527 528 529 530 531
	case GFS2_METATYPE_LF:
		offset = sizeof(struct gfs2_leaf);
		break;
	case GFS2_METATYPE_DI:
		offset = sizeof(struct gfs2_dinode);
		break;
	default:
		goto wrong_type;
	}
532 533 534
	return offset;
wrong_type:
	printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
535
	       be32_to_cpu(h->mh_type));
536 537 538
	return -1;
}

539
static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
540 541 542 543 544 545 546 547
					    unsigned int len, gfs2_dscan_t scan,
					    const struct qstr *name,
					    void *opaque)
{
	struct gfs2_dirent *dent, *prev;
	unsigned offset;
	unsigned size;
	int ret = 0;
548

549 550 551 552 553
	ret = gfs2_dirent_offset(buf);
	if (ret < 0)
		goto consist_inode;

	offset = ret;
554
	prev = NULL;
555
	dent = buf + offset;
556 557 558 559
	size = be16_to_cpu(dent->de_rec_len);
	if (gfs2_check_dirent(dent, offset, size, len, 1))
		goto consist_inode;
	do {
560
		ret = scan(dent, name, opaque);
561 562 563 564 565 566
		if (ret)
			break;
		offset += size;
		if (offset == len)
			break;
		prev = dent;
567
		dent = buf + offset;
568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585
		size = be16_to_cpu(dent->de_rec_len);
		if (gfs2_check_dirent(dent, offset, size, len, 0))
			goto consist_inode;
	} while(1);

	switch(ret) {
	case 0:
		return NULL;
	case 1:
		return dent;
	case 2:
		return prev ? prev : dent;
	default:
		BUG_ON(ret > 0);
		return ERR_PTR(ret);
	}

consist_inode:
586
	gfs2_consist_inode(GFS2_I(inode));
587 588 589
	return ERR_PTR(-EIO);
}

590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607
static int dirent_check_reclen(struct gfs2_inode *dip,
			       const struct gfs2_dirent *d, const void *end_p)
{
	const void *ptr = d;
	u16 rec_len = be16_to_cpu(d->de_rec_len);

	if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
		goto broken;
	ptr += rec_len;
	if (ptr < end_p)
		return rec_len;
	if (ptr == end_p)
		return -ENOENT;
broken:
	gfs2_consist_inode(dip);
	return -EIO;
}

D
David Teigland 已提交
608 609 610 611 612 613 614 615 616 617 618 619
/**
 * dirent_next - Next dirent
 * @dip: the directory
 * @bh: The buffer
 * @dent: Pointer to list of dirents
 *
 * Returns: 0 on success, error code otherwise
 */

static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
		       struct gfs2_dirent **dent)
{
620 621 622
	struct gfs2_dirent *cur = *dent, *tmp;
	char *bh_end = bh->b_data + bh->b_size;
	int ret;
D
David Teigland 已提交
623

624 625 626
	ret = dirent_check_reclen(dip, cur, bh_end);
	if (ret < 0)
		return ret;
627

628 629 630 631
	tmp = (void *)cur + ret;
	ret = dirent_check_reclen(dip, tmp, bh_end);
	if (ret == -EIO)
		return ret;
632

D
David Teigland 已提交
633
        /* Only the first dent could ever have de_inum.no_addr == 0 */
634
	if (gfs2_dirent_sentinel(tmp)) {
D
David Teigland 已提交
635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654
		gfs2_consist_inode(dip);
		return -EIO;
	}

	*dent = tmp;
	return 0;
}

/**
 * dirent_del - Delete a dirent
 * @dip: The GFS2 inode
 * @bh: The buffer
 * @prev: The previous dirent
 * @cur: The current dirent
 *
 */

static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
		       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
{
655
	u16 cur_rec_len, prev_rec_len;
D
David Teigland 已提交
656

657
	if (gfs2_dirent_sentinel(cur)) {
D
David Teigland 已提交
658 659 660 661
		gfs2_consist_inode(dip);
		return;
	}

662
	gfs2_trans_add_meta(dip->i_gl, bh);
D
David Teigland 已提交
663 664 665 666 667 668

	/* If there is no prev entry, this is the first entry in the block.
	   The de_rec_len is already as big as it needs to be.  Just zero
	   out the inode number and return.  */

	if (!prev) {
669 670
		cur->de_inum.no_addr = 0;
		cur->de_inum.no_formal_ino = 0;
D
David Teigland 已提交
671 672 673 674 675
		return;
	}

	/*  Combine this dentry with the previous one.  */

676 677
	prev_rec_len = be16_to_cpu(prev->de_rec_len);
	cur_rec_len = be16_to_cpu(cur->de_rec_len);
D
David Teigland 已提交
678 679 680 681 682 683 684

	if ((char *)prev + prev_rec_len != (char *)cur)
		gfs2_consist_inode(dip);
	if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
		gfs2_consist_inode(dip);

	prev_rec_len += cur_rec_len;
685
	prev->de_rec_len = cpu_to_be16(prev_rec_len);
D
David Teigland 已提交
686 687
}

688 689 690
/*
 * Takes a dent from which to grab space as an argument. Returns the
 * newly created dent.
D
David Teigland 已提交
691
 */
692 693 694 695
static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
					    struct gfs2_dirent *dent,
					    const struct qstr *name,
					    struct buffer_head *bh)
D
David Teigland 已提交
696
{
697
	struct gfs2_inode *ip = GFS2_I(inode);
698 699 700
	struct gfs2_dirent *ndent;
	unsigned offset = 0, totlen;

701
	if (!gfs2_dirent_sentinel(dent))
702 703 704
		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
	totlen = be16_to_cpu(dent->de_rec_len);
	BUG_ON(offset + name->len > totlen);
705
	gfs2_trans_add_meta(ip->i_gl, bh);
706 707 708 709
	ndent = (struct gfs2_dirent *)((char *)dent + offset);
	dent->de_rec_len = cpu_to_be16(offset);
	gfs2_qstr2dirent(name, totlen - offset, ndent);
	return ndent;
D
David Teigland 已提交
710 711
}

712 713 714
static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
					     struct buffer_head *bh,
					     const struct qstr *name)
D
David Teigland 已提交
715 716
{
	struct gfs2_dirent *dent;
717
	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
718
				gfs2_dirent_find_space, name, NULL);
719 720 721
	if (!dent || IS_ERR(dent))
		return dent;
	return gfs2_init_dirent(inode, dent, name, bh);
D
David Teigland 已提交
722 723
}

724
static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
D
David Teigland 已提交
725 726 727 728
		    struct buffer_head **bhp)
{
	int error;

S
Steven Whitehouse 已提交
729
	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
730 731
	if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
		/* printk(KERN_INFO "block num=%llu\n", leaf_no); */
D
David Teigland 已提交
732
		error = -EIO;
733
	}
D
David Teigland 已提交
734 735 736 737 738 739 740 741 742 743 744 745 746

	return error;
}

/**
 * get_leaf_nr - Get a leaf number associated with the index
 * @dip: The GFS2 inode
 * @index:
 * @leaf_out:
 *
 * Returns: 0 on success, error code otherwise
 */

747 748
static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
		       u64 *leaf_out)
D
David Teigland 已提交
749
{
750
	__be64 *hash;
D
David Teigland 已提交
751

752 753 754 755
	hash = gfs2_dir_get_hash_table(dip);
	if (IS_ERR(hash))
		return PTR_ERR(hash);
	*leaf_out = be64_to_cpu(*(hash + index));
D
David Teigland 已提交
756 757 758
	return 0;
}

759
static int get_first_leaf(struct gfs2_inode *dip, u32 index,
D
David Teigland 已提交
760 761
			  struct buffer_head **bh_out)
{
762
	u64 leaf_no;
D
David Teigland 已提交
763 764 765 766 767 768 769 770 771
	int error;

	error = get_leaf_nr(dip, index, &leaf_no);
	if (!error)
		error = get_leaf(dip, leaf_no, bh_out);

	return error;
}

772 773 774 775
static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
					      const struct qstr *name,
					      gfs2_dscan_t scan,
					      struct buffer_head **pbh)
D
David Teigland 已提交
776
{
777 778
	struct buffer_head *bh;
	struct gfs2_dirent *dent;
779
	struct gfs2_inode *ip = GFS2_I(inode);
D
David Teigland 已提交
780 781
	int error;

782
	if (ip->i_diskflags & GFS2_DIF_EXHASH) {
783
		struct gfs2_leaf *leaf;
784
		unsigned hsize = 1 << ip->i_depth;
785 786
		unsigned index;
		u64 ln;
S
Steven Whitehouse 已提交
787
		if (hsize * sizeof(u64) != i_size_read(inode)) {
788 789 790
			gfs2_consist_inode(ip);
			return ERR_PTR(-EIO);
		}
791

792
		index = name->hash >> (32 - ip->i_depth);
793 794 795 796 797
		error = get_first_leaf(ip, index, &bh);
		if (error)
			return ERR_PTR(error);
		do {
			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
798
						scan, name, NULL);
799 800 801 802
			if (dent)
				goto got_dent;
			leaf = (struct gfs2_leaf *)bh->b_data;
			ln = be64_to_cpu(leaf->lf_next);
803
			brelse(bh);
804 805
			if (!ln)
				break;
806

807 808
			error = get_leaf(ip, ln, &bh);
		} while(!error);
D
David Teigland 已提交
809

810
		return error ? ERR_PTR(error) : NULL;
D
David Teigland 已提交
811 812
	}

813

814 815 816
	error = gfs2_meta_inode_buffer(ip, &bh);
	if (error)
		return ERR_PTR(error);
817
	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
818
got_dent:
819
	if (unlikely(dent == NULL || IS_ERR(dent))) {
820 821 822
		brelse(bh);
		bh = NULL;
	}
823 824 825
	*pbh = bh;
	return dent;
}
D
David Teigland 已提交
826

827 828
static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
{
829
	struct gfs2_inode *ip = GFS2_I(inode);
830
	unsigned int n = 1;
831 832 833
	u64 bn;
	int error;
	struct buffer_head *bh;
834 835
	struct gfs2_leaf *leaf;
	struct gfs2_dirent *dent;
836
	struct qstr name = { .name = "" };
837
	struct timespec tv = CURRENT_TIME;
838

839
	error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
840 841 842
	if (error)
		return NULL;
	bh = gfs2_meta_new(ip->i_gl, bn);
843 844
	if (!bh)
		return NULL;
845

846
	gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
847
	gfs2_trans_add_meta(ip->i_gl, bh);
848 849 850
	gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
	leaf = (struct gfs2_leaf *)bh->b_data;
	leaf->lf_depth = cpu_to_be16(depth);
851
	leaf->lf_entries = 0;
852
	leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
853
	leaf->lf_next = 0;
854 855 856 857 858
	leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
	leaf->lf_dist = cpu_to_be32(1);
	leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
	leaf->lf_sec = cpu_to_be64(tv.tv_sec);
	memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
859
	dent = (struct gfs2_dirent *)(leaf+1);
860
	gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
861 862
	*pbh = bh;
	return leaf;
D
David Teigland 已提交
863 864 865 866 867 868 869 870 871
}

/**
 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
 * @dip: The GFS2 inode
 *
 * Returns: 0 on success, error code otherwise
 */

872
static int dir_make_exhash(struct inode *inode)
D
David Teigland 已提交
873
{
874 875
	struct gfs2_inode *dip = GFS2_I(inode);
	struct gfs2_sbd *sdp = GFS2_SB(inode);
D
David Teigland 已提交
876
	struct gfs2_dirent *dent;
877
	struct qstr args;
D
David Teigland 已提交
878 879 880
	struct buffer_head *bh, *dibh;
	struct gfs2_leaf *leaf;
	int y;
881
	u32 x;
A
Al Viro 已提交
882 883
	__be64 *lp;
	u64 bn;
D
David Teigland 已提交
884 885 886 887 888 889 890 891
	int error;

	error = gfs2_meta_inode_buffer(dip, &dibh);
	if (error)
		return error;

	/*  Turn over a new leaf  */

892 893 894 895
	leaf = new_leaf(inode, &bh, 0);
	if (!leaf)
		return -ENOSPC;
	bn = bh->b_blocknr;
D
David Teigland 已提交
896

897 898
	gfs2_assert(sdp, dip->i_entries < (1 << 16));
	leaf->lf_entries = cpu_to_be16(dip->i_entries);
D
David Teigland 已提交
899 900 901 902 903 904 905 906 907

	/*  Copy dirents  */

	gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
			     sizeof(struct gfs2_dinode));

	/*  Find last entry  */

	x = 0;
908 909 910
	args.len = bh->b_size - sizeof(struct gfs2_dinode) +
		   sizeof(struct gfs2_leaf);
	args.name = bh->b_data;
911
	dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
912
				gfs2_dirent_last, &args, NULL);
913 914 915 916 917 918 919 920 921
	if (!dent) {
		brelse(bh);
		brelse(dibh);
		return -EIO;
	}
	if (IS_ERR(dent)) {
		brelse(bh);
		brelse(dibh);
		return PTR_ERR(dent);
D
David Teigland 已提交
922 923 924 925 926
	}

	/*  Adjust the last dirent's record length
	   (Remember that dent still points to the last entry.)  */

927
	dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
D
David Teigland 已提交
928
		sizeof(struct gfs2_dinode) -
929
		sizeof(struct gfs2_leaf));
D
David Teigland 已提交
930 931 932 933 934 935

	brelse(bh);

	/*  We're done with the new leaf block, now setup the new
	    hash table.  */

936
	gfs2_trans_add_meta(dip->i_gl, dibh);
D
David Teigland 已提交
937 938
	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));

A
Al Viro 已提交
939
	lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
D
David Teigland 已提交
940 941 942 943

	for (x = sdp->sd_hash_ptrs; x--; lp++)
		*lp = cpu_to_be64(bn);

S
Steven Whitehouse 已提交
944
	i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
945
	gfs2_add_inode_blocks(&dip->i_inode, 1);
946
	dip->i_diskflags |= GFS2_DIF_EXHASH;
D
David Teigland 已提交
947 948

	for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
949
	dip->i_depth = y;
D
David Teigland 已提交
950

951
	gfs2_dinode_out(dip, dibh->b_data);
D
David Teigland 已提交
952 953 954 955 956 957 958 959 960 961 962 963 964 965 966

	brelse(dibh);

	return 0;
}

/**
 * dir_split_leaf - Split a leaf block into two
 * @dip: The GFS2 inode
 * @index:
 * @leaf_no:
 *
 * Returns: 0 on success, error code on failure
 */

967
static int dir_split_leaf(struct inode *inode, const struct qstr *name)
D
David Teigland 已提交
968
{
969
	struct gfs2_inode *dip = GFS2_I(inode);
D
David Teigland 已提交
970 971
	struct buffer_head *nbh, *obh, *dibh;
	struct gfs2_leaf *nleaf, *oleaf;
972
	struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
973
	u32 start, len, half_len, divider;
A
Al Viro 已提交
974 975
	u64 bn, leaf_no;
	__be64 *lp;
976
	u32 index;
D
David Teigland 已提交
977 978 979
	int x, moved = 0;
	int error;

980
	index = name->hash >> (32 - dip->i_depth);
981 982 983
	error = get_leaf_nr(dip, index, &leaf_no);
	if (error)
		return error;
D
David Teigland 已提交
984 985 986 987

	/*  Get the old leaf block  */
	error = get_leaf(dip, leaf_no, &obh);
	if (error)
988
		return error;
D
David Teigland 已提交
989 990

	oleaf = (struct gfs2_leaf *)obh->b_data;
991
	if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
992 993 994 995
		brelse(obh);
		return 1; /* can't split */
	}

996
	gfs2_trans_add_meta(dip->i_gl, obh);
D
David Teigland 已提交
997

998 999 1000 1001 1002 1003
	nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
	if (!nleaf) {
		brelse(obh);
		return -ENOSPC;
	}
	bn = nbh->b_blocknr;
D
David Teigland 已提交
1004

1005
	/*  Compute the start and len of leaf pointers in the hash table.  */
1006
	len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth));
D
David Teigland 已提交
1007 1008
	half_len = len >> 1;
	if (!half_len) {
1009
		printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
D
David Teigland 已提交
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019
		gfs2_consist_inode(dip);
		error = -EIO;
		goto fail_brelse;
	}

	start = (index & ~(len - 1));

	/* Change the pointers.
	   Don't bother distinguishing stuffed from non-stuffed.
	   This code is complicated enough already. */
1020 1021 1022 1023 1024 1025
	lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS);
	if (!lp) {
		error = -ENOMEM;
		goto fail_brelse;
	}

D
David Teigland 已提交
1026 1027 1028 1029
	/*  Change the pointers  */
	for (x = 0; x < half_len; x++)
		lp[x] = cpu_to_be64(bn);

1030 1031
	gfs2_dir_hash_inval(dip);

1032 1033 1034
	error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
				    half_len * sizeof(u64));
	if (error != half_len * sizeof(u64)) {
D
David Teigland 已提交
1035 1036 1037 1038 1039 1040 1041 1042
		if (error >= 0)
			error = -EIO;
		goto fail_lpfree;
	}

	kfree(lp);

	/*  Compute the divider  */
1043
	divider = (start + half_len) << (32 - dip->i_depth);
D
David Teigland 已提交
1044 1045

	/*  Copy the entries  */
1046
	dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
D
David Teigland 已提交
1047 1048 1049 1050 1051 1052

	do {
		next = dent;
		if (dirent_next(dip, obh, &next))
			next = NULL;

1053
		if (!gfs2_dirent_sentinel(dent) &&
D
David Teigland 已提交
1054
		    be32_to_cpu(dent->de_hash) < divider) {
1055 1056 1057 1058
			struct qstr str;
			str.name = (char*)(dent+1);
			str.len = be16_to_cpu(dent->de_name_len);
			str.hash = be32_to_cpu(dent->de_hash);
1059
			new = gfs2_dirent_alloc(inode, nbh, &str);
1060 1061 1062 1063
			if (IS_ERR(new)) {
				error = PTR_ERR(new);
				break;
			}
D
David Teigland 已提交
1064 1065 1066

			new->de_inum = dent->de_inum; /* No endian worries */
			new->de_type = dent->de_type; /* No endian worries */
M
Marcin Slusarz 已提交
1067
			be16_add_cpu(&nleaf->lf_entries, 1);
D
David Teigland 已提交
1068 1069 1070 1071 1072

			dirent_del(dip, obh, prev, dent);

			if (!oleaf->lf_entries)
				gfs2_consist_inode(dip);
M
Marcin Slusarz 已提交
1073
			be16_add_cpu(&oleaf->lf_entries, -1);
D
David Teigland 已提交
1074 1075 1076 1077 1078

			if (!prev)
				prev = dent;

			moved = 1;
1079
		} else {
D
David Teigland 已提交
1080
			prev = dent;
1081
		}
D
David Teigland 已提交
1082
		dent = next;
1083
	} while (dent);
D
David Teigland 已提交
1084

1085
	oleaf->lf_depth = nleaf->lf_depth;
D
David Teigland 已提交
1086 1087

	error = gfs2_meta_inode_buffer(dip, &dibh);
1088
	if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1089
		gfs2_trans_add_meta(dip->i_gl, dibh);
1090
		gfs2_add_inode_blocks(&dip->i_inode, 1);
1091
		gfs2_dinode_out(dip, dibh->b_data);
D
David Teigland 已提交
1092 1093 1094 1095 1096 1097 1098 1099
		brelse(dibh);
	}

	brelse(obh);
	brelse(nbh);

	return error;

1100
fail_lpfree:
D
David Teigland 已提交
1101 1102
	kfree(lp);

1103
fail_brelse:
D
David Teigland 已提交
1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118
	brelse(obh);
	brelse(nbh);
	return error;
}

/**
 * dir_double_exhash - Double size of ExHash table
 * @dip: The GFS2 dinode
 *
 * Returns: 0 on success, error code on failure
 */

static int dir_double_exhash(struct gfs2_inode *dip)
{
	struct buffer_head *dibh;
1119
	u32 hsize;
1120 1121 1122
	u32 hsize_bytes;
	__be64 *hc;
	__be64 *hc2, *h;
D
David Teigland 已提交
1123 1124 1125
	int x;
	int error = 0;

1126
	hsize = 1 << dip->i_depth;
1127
	hsize_bytes = hsize * sizeof(__be64);
D
David Teigland 已提交
1128

1129 1130 1131
	hc = gfs2_dir_get_hash_table(dip);
	if (IS_ERR(hc))
		return PTR_ERR(hc);
D
David Teigland 已提交
1132

1133
	hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN);
1134 1135 1136
	if (hc2 == NULL)
		hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL);

1137
	if (!hc2)
1138
		return -ENOMEM;
D
David Teigland 已提交
1139

1140
	h = hc2;
1141 1142 1143
	error = gfs2_meta_inode_buffer(dip, &dibh);
	if (error)
		goto out_kfree;
D
David Teigland 已提交
1144

1145 1146 1147 1148
	for (x = 0; x < hsize; x++) {
		*h++ = *hc;
		*h++ = *hc;
		hc++;
D
David Teigland 已提交
1149 1150
	}

1151 1152 1153
	error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
	if (error != (hsize_bytes * 2))
		goto fail;
D
David Teigland 已提交
1154

1155 1156 1157 1158 1159 1160
	gfs2_dir_hash_inval(dip);
	dip->i_hash_cache = hc2;
	dip->i_depth++;
	gfs2_dinode_out(dip, dibh->b_data);
	brelse(dibh);
	return 0;
D
David Teigland 已提交
1161

1162
fail:
1163 1164 1165 1166 1167 1168
	/* Replace original hash table & size */
	gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
	i_size_write(&dip->i_inode, hsize_bytes);
	gfs2_dinode_out(dip, dibh->b_data);
	brelse(dibh);
out_kfree:
1169 1170 1171 1172
	if (is_vmalloc_addr(hc2))
		vfree(hc2);
	else
		kfree(hc2);
D
David Teigland 已提交
1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188
	return error;
}

/**
 * compare_dents - compare directory entries by hash value
 * @a: first dent
 * @b: second dent
 *
 * When comparing the hash entries of @a to @b:
 *   gt: returns 1
 *   lt: returns -1
 *   eq: returns 0
 */

static int compare_dents(const void *a, const void *b)
{
1189
	const struct gfs2_dirent *dent_a, *dent_b;
1190
	u32 hash_a, hash_b;
D
David Teigland 已提交
1191 1192
	int ret = 0;

1193
	dent_a = *(const struct gfs2_dirent **)a;
1194
	hash_a = be32_to_cpu(dent_a->de_hash);
D
David Teigland 已提交
1195

1196
	dent_b = *(const struct gfs2_dirent **)b;
1197
	hash_b = be32_to_cpu(dent_b->de_hash);
D
David Teigland 已提交
1198 1199 1200 1201 1202 1203

	if (hash_a > hash_b)
		ret = 1;
	else if (hash_a < hash_b)
		ret = -1;
	else {
1204 1205
		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
D
David Teigland 已提交
1206 1207 1208 1209 1210 1211

		if (len_a > len_b)
			ret = 1;
		else if (len_a < len_b)
			ret = -1;
		else
1212
			ret = memcmp(dent_a + 1, dent_b + 1, len_a);
D
David Teigland 已提交
1213 1214 1215 1216 1217 1218 1219 1220
	}

	return ret;
}

/**
 * do_filldir_main - read out directory entries
 * @dip: The GFS2 inode
A
Al Viro 已提交
1221
 * @ctx: what to feed the entries to
D
David Teigland 已提交
1222 1223 1224 1225 1226 1227 1228 1229 1230
 * @darr: an array of struct gfs2_dirent pointers to read
 * @entries: the number of entries in darr
 * @copied: pointer to int that's non-zero if a entry has been copied out
 *
 * Jump through some hoops to make sure that if there are hash collsions,
 * they are read out at the beginning of a buffer.  We want to minimize
 * the possibility that they will fall into different readdir buffers or
 * that someone will want to seek to that location.
 *
A
Al Viro 已提交
1231
 * Returns: errno, >0 if the actor tells you to stop
D
David Teigland 已提交
1232 1233
 */

A
Al Viro 已提交
1234
static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1235
			   const struct gfs2_dirent **darr, u32 entries,
D
David Teigland 已提交
1236 1237
			   int *copied)
{
1238
	const struct gfs2_dirent *dent, *dent_next;
1239
	u64 off, off_next;
D
David Teigland 已提交
1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257
	unsigned int x, y;
	int run = 0;

	sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);

	dent_next = darr[0];
	off_next = be32_to_cpu(dent_next->de_hash);
	off_next = gfs2_disk_hash2offset(off_next);

	for (x = 0, y = 1; x < entries; x++, y++) {
		dent = dent_next;
		off = off_next;

		if (y < entries) {
			dent_next = darr[y];
			off_next = be32_to_cpu(dent_next->de_hash);
			off_next = gfs2_disk_hash2offset(off_next);

A
Al Viro 已提交
1258
			if (off < ctx->pos)
D
David Teigland 已提交
1259
				continue;
A
Al Viro 已提交
1260
			ctx->pos = off;
D
David Teigland 已提交
1261 1262 1263 1264 1265 1266 1267 1268

			if (off_next == off) {
				if (*copied && !run)
					return 1;
				run = 1;
			} else
				run = 0;
		} else {
A
Al Viro 已提交
1269
			if (off < ctx->pos)
D
David Teigland 已提交
1270
				continue;
A
Al Viro 已提交
1271
			ctx->pos = off;
D
David Teigland 已提交
1272 1273
		}

A
Al Viro 已提交
1274
		if (!dir_emit(ctx, (const char *)(dent + 1),
1275
				be16_to_cpu(dent->de_name_len),
A
Al Viro 已提交
1276 1277
				be64_to_cpu(dent->de_inum.no_addr),
				be16_to_cpu(dent->de_type)))
D
David Teigland 已提交
1278 1279 1280 1281 1282
			return 1;

		*copied = 1;
	}

A
Al Viro 已提交
1283
	/* Increment the ctx->pos by one, so the next time we come into the
D
David Teigland 已提交
1284 1285 1286
	   do_filldir fxn, we get the next entry instead of the last one in the
	   current leaf */

A
Al Viro 已提交
1287
	ctx->pos++;
D
David Teigland 已提交
1288 1289 1290 1291

	return 0;
}

1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310
static void *gfs2_alloc_sort_buffer(unsigned size)
{
	void *ptr = NULL;

	if (size < KMALLOC_MAX_SIZE)
		ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
	if (!ptr)
		ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
	return ptr;
}

static void gfs2_free_sort_buffer(void *ptr)
{
	if (is_vmalloc_addr(ptr))
		vfree(ptr);
	else
		kfree(ptr);
}

A
Al Viro 已提交
1311 1312
static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
			      int *copied, unsigned *depth,
1313
			      u64 leaf_no)
D
David Teigland 已提交
1314
{
1315
	struct gfs2_inode *ip = GFS2_I(inode);
1316
	struct gfs2_sbd *sdp = GFS2_SB(inode);
1317 1318
	struct buffer_head *bh;
	struct gfs2_leaf *lf;
1319
	unsigned entries = 0, entries2 = 0;
1320 1321 1322 1323 1324 1325 1326
	unsigned leaves = 0;
	const struct gfs2_dirent **darr, *dent;
	struct dirent_gather g;
	struct buffer_head **larr;
	int leaf = 0;
	int error, i;
	u64 lfn = leaf_no;
D
David Teigland 已提交
1327 1328

	do {
1329
		error = get_leaf(ip, lfn, &bh);
D
David Teigland 已提交
1330
		if (error)
1331 1332 1333 1334 1335 1336 1337 1338 1339
			goto out;
		lf = (struct gfs2_leaf *)bh->b_data;
		if (leaves == 0)
			*depth = be16_to_cpu(lf->lf_depth);
		entries += be16_to_cpu(lf->lf_entries);
		leaves++;
		lfn = be64_to_cpu(lf->lf_next);
		brelse(bh);
	} while(lfn);
D
David Teigland 已提交
1340 1341 1342 1343

	if (!entries)
		return 0;

1344
	error = -ENOMEM;
1345 1346 1347 1348 1349 1350
	/*
	 * The extra 99 entries are not normally used, but are a buffer
	 * zone in case the number of entries in the leaf is corrupt.
	 * 99 is the maximum number of entries that can fit in a single
	 * leaf block.
	 */
1351
	larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1352 1353 1354 1355 1356 1357
	if (!larr)
		goto out;
	darr = (const struct gfs2_dirent **)(larr + leaves);
	g.pdent = darr;
	g.offset = 0;
	lfn = leaf_no;
D
David Teigland 已提交
1358

1359 1360
	do {
		error = get_leaf(ip, lfn, &bh);
D
David Teigland 已提交
1361
		if (error)
1362
			goto out_free;
1363 1364 1365
		lf = (struct gfs2_leaf *)bh->b_data;
		lfn = be64_to_cpu(lf->lf_next);
		if (lf->lf_entries) {
1366
			entries2 += be16_to_cpu(lf->lf_entries);
1367 1368 1369
			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
						gfs2_dirent_gather, NULL, &g);
			error = PTR_ERR(dent);
1370
			if (IS_ERR(dent))
1371
				goto out_free;
1372
			if (entries2 != g.offset) {
1373 1374 1375 1376 1377
				fs_warn(sdp, "Number of entries corrupt in dir "
						"leaf %llu, entries2 (%u) != "
						"g.offset (%u)\n",
					(unsigned long long)bh->b_blocknr,
					entries2, g.offset);
1378 1379
					
				error = -EIO;
1380
				goto out_free;
1381 1382 1383
			}
			error = 0;
			larr[leaf++] = bh;
D
David Teigland 已提交
1384
		} else {
1385
			brelse(bh);
D
David Teigland 已提交
1386
		}
1387
	} while(lfn);
D
David Teigland 已提交
1388

1389
	BUG_ON(entries2 != entries);
A
Al Viro 已提交
1390
	error = do_filldir_main(ip, ctx, darr, entries, copied);
1391
out_free:
1392 1393
	for(i = 0; i < leaf; i++)
		brelse(larr[i]);
1394
	gfs2_free_sort_buffer(larr);
1395
out:
D
David Teigland 已提交
1396 1397 1398
	return error;
}

1399 1400
/**
 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1401 1402 1403 1404
 *
 * Note: we can't calculate each index like dir_e_read can because we don't
 * have the leaf, and therefore we don't have the depth, and therefore we
 * don't have the length. So we have to just read enough ahead to make up
1405 1406
 * for the loss of information.
 */
1407 1408 1409 1410 1411 1412 1413 1414 1415 1416
static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
			       struct file_ra_state *f_ra)
{
	struct gfs2_inode *ip = GFS2_I(inode);
	struct gfs2_glock *gl = ip->i_gl;
	struct buffer_head *bh;
	u64 blocknr = 0, last;
	unsigned count;

	/* First check if we've already read-ahead for the whole range. */
1417
	if (index + MAX_RA_BLOCKS < f_ra->start)
1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444
		return;

	f_ra->start = max((pgoff_t)index, f_ra->start);
	for (count = 0; count < MAX_RA_BLOCKS; count++) {
		if (f_ra->start >= hsize) /* if exceeded the hash table */
			break;

		last = blocknr;
		blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
		f_ra->start++;
		if (blocknr == last)
			continue;

		bh = gfs2_getbuf(gl, blocknr, 1);
		if (trylock_buffer(bh)) {
			if (buffer_uptodate(bh)) {
				unlock_buffer(bh);
				brelse(bh);
				continue;
			}
			bh->b_end_io = end_buffer_read_sync;
			submit_bh(READA | REQ_META, bh);
			continue;
		}
		brelse(bh);
	}
}
1445

D
David Teigland 已提交
1446
/**
1447 1448
 * dir_e_read - Reads the entries from a directory into a filldir buffer
 * @dip: dinode pointer
A
Al Viro 已提交
1449
 * @ctx: actor to feed the entries to
D
David Teigland 已提交
1450
 *
1451
 * Returns: errno
D
David Teigland 已提交
1452 1453
 */

A
Al Viro 已提交
1454 1455
static int dir_e_read(struct inode *inode, struct dir_context *ctx,
		      struct file_ra_state *f_ra)
D
David Teigland 已提交
1456
{
1457
	struct gfs2_inode *dip = GFS2_I(inode);
1458 1459
	u32 hsize, len = 0;
	u32 hash, index;
A
Al Viro 已提交
1460
	__be64 *lp;
1461 1462
	int copied = 0;
	int error = 0;
1463
	unsigned depth = 0;
D
David Teigland 已提交
1464

1465
	hsize = 1 << dip->i_depth;
A
Al Viro 已提交
1466
	hash = gfs2_dir_offset2hash(ctx->pos);
1467
	index = hash >> (32 - dip->i_depth);
D
David Teigland 已提交
1468

1469
	if (dip->i_hash_cache == NULL)
1470
		f_ra->start = 0;
1471 1472 1473
	lp = gfs2_dir_get_hash_table(dip);
	if (IS_ERR(lp))
		return PTR_ERR(lp);
D
David Teigland 已提交
1474

1475 1476
	gfs2_dir_readahead(inode, hsize, index, f_ra);

D
David Teigland 已提交
1477
	while (index < hsize) {
A
Al Viro 已提交
1478
		error = gfs2_dir_read_leaf(inode, ctx,
1479
					   &copied, &depth,
1480
					   be64_to_cpu(lp[index]));
D
David Teigland 已提交
1481
		if (error)
1482
			break;
D
David Teigland 已提交
1483

1484
		len = 1 << (dip->i_depth - depth);
D
David Teigland 已提交
1485 1486 1487
		index = (index & ~(len - 1)) + len;
	}

1488 1489
	if (error > 0)
		error = 0;
D
David Teigland 已提交
1490 1491 1492
	return error;
}

A
Al Viro 已提交
1493 1494
int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
		  struct file_ra_state *f_ra)
D
David Teigland 已提交
1495
{
1496
	struct gfs2_inode *dip = GFS2_I(inode);
1497
	struct gfs2_sbd *sdp = GFS2_SB(inode);
1498 1499
	struct dirent_gather g;
	const struct gfs2_dirent **darr, *dent;
D
David Teigland 已提交
1500 1501 1502 1503
	struct buffer_head *dibh;
	int copied = 0;
	int error;

1504
	if (!dip->i_entries)
1505 1506
		return 0;

1507
	if (dip->i_diskflags & GFS2_DIF_EXHASH)
A
Al Viro 已提交
1508
		return dir_e_read(inode, ctx, f_ra);
1509

D
David Teigland 已提交
1510 1511 1512 1513 1514 1515 1516 1517 1518
	if (!gfs2_is_stuffed(dip)) {
		gfs2_consist_inode(dip);
		return -EIO;
	}

	error = gfs2_meta_inode_buffer(dip, &dibh);
	if (error)
		return error;

1519
	error = -ENOMEM;
1520
	/* 96 is max number of dirents which can be stuffed into an inode */
J
Josef Bacik 已提交
1521
	darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
1522 1523 1524 1525 1526 1527 1528 1529 1530
	if (darr) {
		g.pdent = darr;
		g.offset = 0;
		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
					gfs2_dirent_gather, NULL, &g);
		if (IS_ERR(dent)) {
			error = PTR_ERR(dent);
			goto out;
		}
1531
		if (dip->i_entries != g.offset) {
1532
			fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1533
				"ip->i_entries (%u) != g.offset (%u)\n",
1534
				(unsigned long long)dip->i_no_addr,
1535
				dip->i_entries,
1536 1537 1538 1539
				g.offset);
			error = -EIO;
			goto out;
		}
A
Al Viro 已提交
1540
		error = do_filldir_main(dip, ctx, darr,
1541
					dip->i_entries, &copied);
1542 1543 1544 1545
out:
		kfree(darr);
	}

D
David Teigland 已提交
1546 1547 1548 1549 1550 1551 1552 1553 1554 1555
	if (error > 0)
		error = 0;

	brelse(dibh);

	return error;
}

/**
 * gfs2_dir_search - Search a directory
1556 1557 1558
 * @dip: The GFS2 dir inode
 * @name: The name we are looking up
 * @fail_on_exist: Fail if the name exists rather than looking it up
D
David Teigland 已提交
1559 1560 1561 1562 1563 1564 1565
 *
 * This routine searches a directory for a file or another directory.
 * Assumes a glock is held on dip.
 *
 * Returns: errno
 */

1566 1567
struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
			      bool fail_on_exist)
D
David Teigland 已提交
1568
{
1569 1570
	struct buffer_head *bh;
	struct gfs2_dirent *dent;
1571 1572
	u64 addr, formal_ino;
	u16 dtype;
1573 1574 1575 1576

	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
	if (dent) {
		if (IS_ERR(dent))
1577
			return ERR_CAST(dent);
1578 1579 1580
		dtype = be16_to_cpu(dent->de_type);
		addr = be64_to_cpu(dent->de_inum.no_addr);
		formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
1581
		brelse(bh);
1582 1583 1584
		if (fail_on_exist)
			return ERR_PTR(-EEXIST);
		return gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, 0);
1585 1586 1587 1588 1589 1590 1591 1592 1593 1594
	}
	return ERR_PTR(-ENOENT);
}

int gfs2_dir_check(struct inode *dir, const struct qstr *name,
		   const struct gfs2_inode *ip)
{
	struct buffer_head *bh;
	struct gfs2_dirent *dent;
	int ret = -ENOENT;
1595 1596 1597 1598 1599

	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
	if (dent) {
		if (IS_ERR(dent))
			return PTR_ERR(dent);
1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614
		if (ip) {
			if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
				goto out;
			if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
			    ip->i_no_formal_ino)
				goto out;
			if (unlikely(IF2DT(ip->i_inode.i_mode) !=
			    be16_to_cpu(dent->de_type))) {
				gfs2_consist_inode(GFS2_I(dir));
				ret = -EIO;
				goto out;
			}
		}
		ret = 0;
out:
1615 1616
		brelse(bh);
	}
1617
	return ret;
1618 1619
}

1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638
/**
 * dir_new_leaf - Add a new leaf onto hash chain
 * @inode: The directory
 * @name: The name we are adding
 *
 * This adds a new dir leaf onto an existing leaf when there is not
 * enough space to add a new dir entry. This is a last resort after
 * we've expanded the hash table to max size and also split existing
 * leaf blocks, so it will only occur for very large directories.
 *
 * The dist parameter is set to 1 for leaf blocks directly attached
 * to the hash table, 2 for one layer of indirection, 3 for two layers
 * etc. We are thus able to tell the difference between an old leaf
 * with dist set to zero (i.e. "don't know") and a new one where we
 * set this information for debug/fsck purposes.
 *
 * Returns: 0 on success, or -ve on error
 */

1639 1640 1641
static int dir_new_leaf(struct inode *inode, const struct qstr *name)
{
	struct buffer_head *bh, *obh;
1642
	struct gfs2_inode *ip = GFS2_I(inode);
1643
	struct gfs2_leaf *leaf, *oleaf;
1644
	u32 dist = 1;
D
David Teigland 已提交
1645
	int error;
1646 1647
	u32 index;
	u64 bn;
D
David Teigland 已提交
1648

1649
	index = name->hash >> (32 - ip->i_depth);
1650 1651 1652 1653
	error = get_first_leaf(ip, index, &obh);
	if (error)
		return error;
	do {
1654
		dist++;
1655 1656 1657 1658 1659 1660 1661 1662 1663
		oleaf = (struct gfs2_leaf *)obh->b_data;
		bn = be64_to_cpu(oleaf->lf_next);
		if (!bn)
			break;
		brelse(obh);
		error = get_leaf(ip, bn, &obh);
		if (error)
			return error;
	} while(1);
D
David Teigland 已提交
1664

1665
	gfs2_trans_add_meta(ip->i_gl, obh);
1666 1667 1668 1669 1670 1671

	leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
	if (!leaf) {
		brelse(obh);
		return -ENOSPC;
	}
1672
	leaf->lf_dist = cpu_to_be32(dist);
1673
	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1674 1675 1676 1677 1678 1679
	brelse(bh);
	brelse(obh);

	error = gfs2_meta_inode_buffer(ip, &bh);
	if (error)
		return error;
1680
	gfs2_trans_add_meta(ip->i_gl, bh);
1681
	gfs2_add_inode_blocks(&ip->i_inode, 1);
1682
	gfs2_dinode_out(ip, bh->b_data);
1683 1684
	brelse(bh);
	return 0;
D
David Teigland 已提交
1685 1686
}

1687 1688 1689 1690 1691 1692 1693 1694
static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip)
{
	u64 where = ip->i_no_addr + 1;
	if (ip->i_eattr == where)
		return 1;
	return 0;
}

D
David Teigland 已提交
1695 1696
/**
 * gfs2_dir_add - Add new filename into directory
1697 1698 1699 1700 1701 1702 1703 1704 1705 1706
 * @inode: The directory inode
 * @name: The new name
 * @nip: The GFS2 inode to be linked in to the directory
 * @da: The directory addition info
 *
 * If the call to gfs2_diradd_alloc_required resulted in there being
 * no need to allocate any new directory blocks, then it will contain
 * a pointer to the directory entry and the bh in which it resides. We
 * can use that without having to repeat the search. If there was no
 * free space, then we must now create more space.
D
David Teigland 已提交
1707 1708 1709 1710
 *
 * Returns: 0 on success, error code on failure
 */

1711
int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1712
		 const struct gfs2_inode *nip, struct gfs2_diradd *da)
D
David Teigland 已提交
1713
{
1714
	struct gfs2_inode *ip = GFS2_I(inode);
1715 1716
	struct buffer_head *bh = da->bh;
	struct gfs2_dirent *dent = da->dent;
1717
	struct timespec tv;
1718
	struct gfs2_leaf *leaf;
D
David Teigland 已提交
1719 1720
	int error;

1721
	while(1) {
1722 1723 1724 1725
		if (da->bh == NULL) {
			dent = gfs2_dirent_search(inode, name,
						  gfs2_dirent_find_space, &bh);
		}
1726 1727 1728 1729
		if (dent) {
			if (IS_ERR(dent))
				return PTR_ERR(dent);
			dent = gfs2_init_dirent(inode, dent, name, bh);
1730
			gfs2_inum_out(nip, dent);
1731
			dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1732
			dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip));
1733
			tv = CURRENT_TIME;
1734
			if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1735
				leaf = (struct gfs2_leaf *)bh->b_data;
M
Marcin Slusarz 已提交
1736
				be16_add_cpu(&leaf->lf_entries, 1);
1737 1738
				leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
				leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1739
			}
1740 1741
			da->dent = NULL;
			da->bh = NULL;
1742
			brelse(bh);
1743
			ip->i_entries++;
1744
			ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv;
1745 1746
			if (S_ISDIR(nip->i_inode.i_mode))
				inc_nlink(&ip->i_inode);
1747
			mark_inode_dirty(inode);
1748 1749 1750
			error = 0;
			break;
		}
1751
		if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1752 1753 1754 1755 1756 1757 1758 1759
			error = dir_make_exhash(inode);
			if (error)
				break;
			continue;
		}
		error = dir_split_leaf(inode, name);
		if (error == 0)
			continue;
1760
		if (error < 0)
1761
			break;
1762
		if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1763 1764 1765 1766
			error = dir_double_exhash(ip);
			if (error)
				break;
			error = dir_split_leaf(inode, name);
1767
			if (error < 0)
1768
				break;
1769 1770
			if (error == 0)
				continue;
1771 1772 1773 1774 1775 1776 1777
		}
		error = dir_new_leaf(inode, name);
		if (!error)
			continue;
		error = -ENOSPC;
		break;
	}
D
David Teigland 已提交
1778 1779 1780
	return error;
}

1781

D
David Teigland 已提交
1782 1783 1784 1785 1786 1787 1788 1789
/**
 * gfs2_dir_del - Delete a directory entry
 * @dip: The GFS2 inode
 * @filename: The filename
 *
 * Returns: 0 on success, error code on failure
 */

1790
int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
D
David Teigland 已提交
1791
{
1792
	const struct qstr *name = &dentry->d_name;
1793 1794
	struct gfs2_dirent *dent, *prev = NULL;
	struct buffer_head *bh;
1795
	struct timespec tv = CURRENT_TIME;
D
David Teigland 已提交
1796

1797 1798
	/* Returns _either_ the entry (if its first in block) or the
	   previous entry otherwise */
1799
	dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1800 1801 1802 1803 1804 1805 1806 1807 1808
	if (!dent) {
		gfs2_consist_inode(dip);
		return -EIO;
	}
	if (IS_ERR(dent)) {
		gfs2_consist_inode(dip);
		return PTR_ERR(dent);
	}
	/* If not first in block, adjust pointers accordingly */
1809
	if (gfs2_dirent_find(dent, name, NULL) == 0) {
1810 1811 1812 1813 1814
		prev = dent;
		dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
	}

	dirent_del(dip, bh, prev, dent);
1815
	if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1816 1817 1818 1819 1820
		struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
		u16 entries = be16_to_cpu(leaf->lf_entries);
		if (!entries)
			gfs2_consist_inode(dip);
		leaf->lf_entries = cpu_to_be16(--entries);
1821 1822
		leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
		leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1823
	}
1824
	brelse(bh);
1825

1826
	if (!dip->i_entries)
1827
		gfs2_consist_inode(dip);
1828
	dip->i_entries--;
1829
	dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv;
1830 1831
	if (S_ISDIR(dentry->d_inode->i_mode))
		drop_nlink(&dip->i_inode);
1832
	mark_inode_dirty(&dip->i_inode);
D
David Teigland 已提交
1833

S
Steven Whitehouse 已提交
1834
	return 0;
D
David Teigland 已提交
1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849
}

/**
 * gfs2_dir_mvino - Change inode number of directory entry
 * @dip: The GFS2 inode
 * @filename:
 * @new_inode:
 *
 * This routine changes the inode number of a directory entry.  It's used
 * by rename to change ".." when a directory is moved.
 * Assumes a glock is held on dvp.
 *
 * Returns: errno
 */

1850
int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1851
		   const struct gfs2_inode *nip, unsigned int new_type)
D
David Teigland 已提交
1852
{
1853 1854
	struct buffer_head *bh;
	struct gfs2_dirent *dent;
D
David Teigland 已提交
1855 1856
	int error;

1857
	dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1858 1859 1860 1861 1862 1863
	if (!dent) {
		gfs2_consist_inode(dip);
		return -EIO;
	}
	if (IS_ERR(dent))
		return PTR_ERR(dent);
D
David Teigland 已提交
1864

1865
	gfs2_trans_add_meta(dip->i_gl, bh);
1866
	gfs2_inum_out(nip, dent);
1867 1868
	dent->de_type = cpu_to_be16(new_type);

1869
	if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1870 1871 1872 1873
		brelse(bh);
		error = gfs2_meta_inode_buffer(dip, &bh);
		if (error)
			return error;
1874
		gfs2_trans_add_meta(dip->i_gl, bh);
1875 1876
	}

1877
	dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1878
	gfs2_dinode_out(dip, bh->b_data);
1879 1880
	brelse(bh);
	return 0;
D
David Teigland 已提交
1881 1882 1883 1884 1885 1886 1887 1888
}

/**
 * leaf_dealloc - Deallocate a directory leaf
 * @dip: the directory
 * @index: the hash table offset in the directory
 * @len: the number of pointers to this leaf
 * @leaf_no: the leaf number
1889
 * @leaf_bh: buffer_head for the starting leaf
1890
 * last_dealloc: 1 if this is the final dealloc for the leaf, else 0
D
David Teigland 已提交
1891 1892 1893 1894
 *
 * Returns: errno
 */

1895
static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1896 1897
			u64 leaf_no, struct buffer_head *leaf_bh,
			int last_dealloc)
D
David Teigland 已提交
1898
{
1899
	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1900
	struct gfs2_leaf *tmp_leaf;
D
David Teigland 已提交
1901 1902
	struct gfs2_rgrp_list rlist;
	struct buffer_head *bh, *dibh;
1903
	u64 blk, nblk;
D
David Teigland 已提交
1904 1905
	unsigned int rg_blocks = 0, l_blocks = 0;
	char *ht;
1906
	unsigned int x, size = len * sizeof(u64);
D
David Teigland 已提交
1907 1908
	int error;

1909 1910 1911 1912
	error = gfs2_rindex_update(sdp);
	if (error)
		return error;

D
David Teigland 已提交
1913 1914
	memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));

1915
	ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
1916 1917
	if (ht == NULL)
		ht = vzalloc(size);
D
David Teigland 已提交
1918 1919 1920
	if (!ht)
		return -ENOMEM;

1921
	error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
D
David Teigland 已提交
1922
	if (error)
1923
		goto out;
D
David Teigland 已提交
1924 1925

	/*  Count the number of leaves  */
1926
	bh = leaf_bh;
D
David Teigland 已提交
1927

1928
	for (blk = leaf_no; blk; blk = nblk) {
1929 1930 1931 1932 1933
		if (blk != leaf_no) {
			error = get_leaf(dip, blk, &bh);
			if (error)
				goto out_rlist;
		}
1934 1935
		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
		nblk = be64_to_cpu(tmp_leaf->lf_next);
1936 1937
		if (blk != leaf_no)
			brelse(bh);
D
David Teigland 已提交
1938

1939
		gfs2_rlist_add(dip, &rlist, blk);
D
David Teigland 已提交
1940 1941 1942
		l_blocks++;
	}

1943
	gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
D
David Teigland 已提交
1944 1945 1946

	for (x = 0; x < rlist.rl_rgrps; x++) {
		struct gfs2_rgrpd *rgd;
1947
		rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1948
		rg_blocks += rgd->rd_length;
D
David Teigland 已提交
1949 1950 1951 1952 1953 1954 1955
	}

	error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
	if (error)
		goto out_rlist;

	error = gfs2_trans_begin(sdp,
1956
			rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
D
David Teigland 已提交
1957 1958 1959 1960
			RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
	if (error)
		goto out_rg_gunlock;

1961 1962
	bh = leaf_bh;

1963
	for (blk = leaf_no; blk; blk = nblk) {
1964 1965 1966 1967 1968
		if (blk != leaf_no) {
			error = get_leaf(dip, blk, &bh);
			if (error)
				goto out_end_trans;
		}
1969 1970
		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
		nblk = be64_to_cpu(tmp_leaf->lf_next);
1971 1972
		if (blk != leaf_no)
			brelse(bh);
D
David Teigland 已提交
1973 1974

		gfs2_free_meta(dip, blk, 1);
1975
		gfs2_add_inode_blocks(&dip->i_inode, -1);
D
David Teigland 已提交
1976 1977
	}

1978
	error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
D
David Teigland 已提交
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988
	if (error != size) {
		if (error >= 0)
			error = -EIO;
		goto out_end_trans;
	}

	error = gfs2_meta_inode_buffer(dip, &dibh);
	if (error)
		goto out_end_trans;

1989
	gfs2_trans_add_meta(dip->i_gl, dibh);
1990 1991 1992 1993
	/* On the last dealloc, make this a regular file in case we crash.
	   (We don't want to free these blocks a second time.)  */
	if (last_dealloc)
		dip->i_inode.i_mode = S_IFREG;
1994
	gfs2_dinode_out(dip, dibh->b_data);
D
David Teigland 已提交
1995 1996
	brelse(dibh);

1997
out_end_trans:
D
David Teigland 已提交
1998
	gfs2_trans_end(sdp);
1999
out_rg_gunlock:
D
David Teigland 已提交
2000
	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2001
out_rlist:
D
David Teigland 已提交
2002 2003
	gfs2_rlist_free(&rlist);
	gfs2_quota_unhold(dip);
2004
out:
2005 2006 2007 2008
	if (is_vmalloc_addr(ht))
		vfree(ht);
	else
		kfree(ht);
D
David Teigland 已提交
2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
	return error;
}

/**
 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
 * @dip: the directory
 *
 * Dealloc all on-disk directory leaves to FREEMETA state
 * Change on-disk inode type to "regular file"
 *
 * Returns: errno
 */

int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
{
2024 2025 2026 2027 2028 2029 2030 2031 2032 2033
	struct buffer_head *bh;
	struct gfs2_leaf *leaf;
	u32 hsize, len;
	u32 index = 0, next_index;
	__be64 *lp;
	u64 leaf_no;
	int error = 0, last;

	hsize = 1 << dip->i_depth;

2034 2035 2036
	lp = gfs2_dir_get_hash_table(dip);
	if (IS_ERR(lp))
		return PTR_ERR(lp);
2037 2038

	while (index < hsize) {
2039
		leaf_no = be64_to_cpu(lp[index]);
2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066
		if (leaf_no) {
			error = get_leaf(dip, leaf_no, &bh);
			if (error)
				goto out;
			leaf = (struct gfs2_leaf *)bh->b_data;
			len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth));

			next_index = (index & ~(len - 1)) + len;
			last = ((next_index >= hsize) ? 1 : 0);
			error = leaf_dealloc(dip, index, len, leaf_no, bh,
					     last);
			brelse(bh);
			if (error)
				goto out;
			index = next_index;
		} else
			index++;
	}

	if (index != hsize) {
		gfs2_consist_inode(dip);
		error = -EIO;
	}

out:

	return error;
D
David Teigland 已提交
2067 2068 2069 2070 2071 2072
}

/**
 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
 * @ip: the file being written to
 * @filname: the filename that's going to be added
2073
 * @da: The structure to return dir alloc info
D
David Teigland 已提交
2074
 *
2075
 * Returns: 0 if ok, -ve on error
D
David Teigland 已提交
2076 2077
 */

2078 2079
int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
			       struct gfs2_diradd *da)
D
David Teigland 已提交
2080
{
2081
	struct gfs2_inode *ip = GFS2_I(inode);
2082
	struct gfs2_sbd *sdp = GFS2_SB(inode);
2083
	const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2084 2085
	struct gfs2_dirent *dent;
	struct buffer_head *bh;
D
David Teigland 已提交
2086

2087
	da->nr_blocks = 0;
2088 2089
	da->bh = NULL;
	da->dent = NULL;
2090

2091
	dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2092
	if (!dent) {
2093
		da->nr_blocks = sdp->sd_max_dirres;
2094 2095 2096
		if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
		    (GFS2_DIRENT_SIZE(name->len) < extra))
			da->nr_blocks = 1;
2097
		return 0;
2098
	}
2099 2100
	if (IS_ERR(dent))
		return PTR_ERR(dent);
2101 2102
	da->bh = bh;
	da->dent = dent;
2103
	return 0;
D
David Teigland 已提交
2104 2105
}