1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4 * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   5 */
   6
   7/*
   8 * Implements Extendible Hashing as described in:
   9 *   "Extendible Hashing" by Fagin, et al in
  10 *     __ACM Trans. on Database Systems__, Sept 1979.
  11 *
  12 *
  13 * Here's the layout of dirents which is essentially the same as that of ext2
  14 * within a single block. The field de_name_len is the number of bytes
  15 * actually required for the name (no null terminator). The field de_rec_len
  16 * is the number of bytes allocated to the dirent. The offset of the next
  17 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
  18 * deleted, the preceding dirent inherits its allocated space, ie
  19 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
  20 * by adding de_rec_len to the current dirent, this essentially causes the
  21 * deleted dirent to get jumped over when iterating through all the dirents.
  22 *
  23 * When deleting the first dirent in a block, there is no previous dirent so
  24 * the field de_ino is set to zero to designate it as deleted. When allocating
  25 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
  26 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
  27 * dirent is allocated. Otherwise it must go through all the 'used' dirents
  28 * searching for one in which the amount of total space minus the amount of
  29 * used space will provide enough space for the new dirent.
  30 *
  31 * There are two types of blocks in which dirents reside. In a stuffed dinode,
  32 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
  33 * the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
  34 * beginning of the leaf block. The dirents reside in leaves when
  35 *
  36 * dip->i_diskflags & GFS2_DIF_EXHASH is true
  37 *
  38 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
  39 *
  40 * When the dirents are in leaves, the actual contents of the directory file are
  41 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
  42 * dirents are NOT in the directory file itself. There can be more than one
  43 * block pointer in the array that points to the same leaf. In fact, when a
  44 * directory is first converted from linear to exhash, all of the pointers
  45 * point to the same leaf.
  46 *
  47 * When a leaf is completely full, the size of the hash table can be
  48 * doubled unless it is already at the maximum size which is hard coded into
  49 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
  50 * but never before the maximum hash table size has been reached.
  51 */
  52
  53#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  54
  55#include <linux/slab.h>
  56#include <linux/spinlock.h>
  57#include <linux/buffer_head.h>
  58#include <linux/sort.h>
  59#include <linux/gfs2_ondisk.h>
  60#include <linux/crc32.h>
  61#include <linux/vmalloc.h>
  62#include <linux/bio.h>
  63
  64#include "gfs2.h"
  65#include "incore.h"
  66#include "dir.h"
  67#include "glock.h"
  68#include "inode.h"
  69#include "meta_io.h"
  70#include "quota.h"
  71#include "rgrp.h"
  72#include "trans.h"
  73#include "bmap.h"
  74#include "util.h"
  75
  76#define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
  77
  78#define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
  79#define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
  80#define GFS2_HASH_INDEX_MASK 0xffffc000
  81#define GFS2_USE_HASH_FLAG 0x2000
  82
  83struct qstr gfs2_qdot __read_mostly;
  84struct qstr gfs2_qdotdot __read_mostly;
  85
  86typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
  87			    const struct qstr *name, void *opaque);
  88
  89int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
  90			    struct buffer_head **bhp)
  91{
  92	struct buffer_head *bh;
  93
  94	bh = gfs2_meta_new(ip->i_gl, block);
  95	gfs2_trans_add_meta(ip->i_gl, bh);
  96	gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
  97	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
  98	*bhp = bh;
  99	return 0;
 100}
 101
 102static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
 103					struct buffer_head **bhp)
 104{
 105	struct buffer_head *bh;
 106	int error;
 107
 108	error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh);
 109	if (error)
 110		return error;
 111	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
 112		brelse(bh);
 113		return -EIO;
 114	}
 115	*bhp = bh;
 116	return 0;
 117}
 118
 119static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
 120				  unsigned int offset, unsigned int size)
 121{
 122	struct buffer_head *dibh;
 123	int error;
 124
 125	error = gfs2_meta_inode_buffer(ip, &dibh);
 126	if (error)
 127		return error;
 128
 129	gfs2_trans_add_meta(ip->i_gl, dibh);
 130	memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
 131	if (ip->i_inode.i_size < offset + size)
 132		i_size_write(&ip->i_inode, offset + size);
 133	inode_set_mtime_to_ts(&ip->i_inode, inode_set_ctime_current(&ip->i_inode));
 134	gfs2_dinode_out(ip, dibh->b_data);
 135
 136	brelse(dibh);
 137
 138	return size;
 139}
 140
 141
 142
 143/**
 144 * gfs2_dir_write_data - Write directory information to the inode
 145 * @ip: The GFS2 inode
 146 * @buf: The buffer containing information to be written
 147 * @offset: The file offset to start writing at
 148 * @size: The amount of data to write
 149 *
 150 * Returns: The number of bytes correctly written or error code
 151 */
 152static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
 153			       u64 offset, unsigned int size)
 154{
 155	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 156	struct buffer_head *dibh;
 157	u64 lblock, dblock;
 158	u32 extlen = 0;
 159	unsigned int o;
 160	int copied = 0;
 161	int error = 0;
 162	bool new = false;
 163
 164	if (!size)
 165		return 0;
 166
 167	if (gfs2_is_stuffed(ip) && offset + size <= gfs2_max_stuffed_size(ip))
 168		return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
 169					      size);
 170
 171	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 172		return -EINVAL;
 173
 174	if (gfs2_is_stuffed(ip)) {
 175		error = gfs2_unstuff_dinode(ip);
 176		if (error)
 177			return error;
 178	}
 179
 180	lblock = offset;
 181	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 182
 183	while (copied < size) {
 184		unsigned int amount;
 185		struct buffer_head *bh;
 186
 187		amount = size - copied;
 188		if (amount > sdp->sd_sb.sb_bsize - o)
 189			amount = sdp->sd_sb.sb_bsize - o;
 190
 191		if (!extlen) {
 192			extlen = 1;
 193			error = gfs2_alloc_extent(&ip->i_inode, lblock, &dblock,
 194						  &extlen, &new);
 195			if (error)
 196				goto fail;
 197			error = -EIO;
 198			if (gfs2_assert_withdraw(sdp, dblock))
 199				goto fail;
 200		}
 201
 202		if (amount == sdp->sd_jbsize || new)
 203			error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
 204		else
 205			error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
 206
 207		if (error)
 208			goto fail;
 209
 210		gfs2_trans_add_meta(ip->i_gl, bh);
 211		memcpy(bh->b_data + o, buf, amount);
 212		brelse(bh);
 213
 214		buf += amount;
 215		copied += amount;
 216		lblock++;
 217		dblock++;
 218		extlen--;
 219
 220		o = sizeof(struct gfs2_meta_header);
 221	}
 222
 223out:
 224	error = gfs2_meta_inode_buffer(ip, &dibh);
 225	if (error)
 226		return error;
 227
 228	if (ip->i_inode.i_size < offset + copied)
 229		i_size_write(&ip->i_inode, offset + copied);
 230	inode_set_mtime_to_ts(&ip->i_inode, inode_set_ctime_current(&ip->i_inode));
 231
 232	gfs2_trans_add_meta(ip->i_gl, dibh);
 233	gfs2_dinode_out(ip, dibh->b_data);
 234	brelse(dibh);
 235
 236	return copied;
 237fail:
 238	if (copied)
 239		goto out;
 240	return error;
 241}
 242
 243static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
 244				 unsigned int size)
 245{
 246	struct buffer_head *dibh;
 247	int error;
 248
 249	error = gfs2_meta_inode_buffer(ip, &dibh);
 250	if (!error) {
 251		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
 252		brelse(dibh);
 253	}
 254
 255	return (error) ? error : size;
 256}
 257
 258
 259/**
 260 * gfs2_dir_read_data - Read a data from a directory inode
 261 * @ip: The GFS2 Inode
 262 * @buf: The buffer to place result into
 263 * @size: Amount of data to transfer
 264 *
 265 * Returns: The amount of data actually copied or the error
 266 */
 267static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
 268			      unsigned int size)
 269{
 270	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 271	u64 lblock, dblock;
 272	u32 extlen = 0;
 273	unsigned int o;
 274	int copied = 0;
 275	int error = 0;
 276
 277	if (gfs2_is_stuffed(ip))
 278		return gfs2_dir_read_stuffed(ip, buf, size);
 279
 280	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 281		return -EINVAL;
 282
 283	lblock = 0;
 284	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 285
 286	while (copied < size) {
 287		unsigned int amount;
 288		struct buffer_head *bh;
 289
 290		amount = size - copied;
 291		if (amount > sdp->sd_sb.sb_bsize - o)
 292			amount = sdp->sd_sb.sb_bsize - o;
 293
 294		if (!extlen) {
 295			extlen = 32;
 296			error = gfs2_get_extent(&ip->i_inode, lblock,
 297						&dblock, &extlen);
 298			if (error || !dblock)
 299				goto fail;
 300			BUG_ON(extlen < 1);
 301			bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
 302		} else {
 303			error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
 304			if (error)
 305				goto fail;
 306		}
 307		error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
 308		if (error) {
 309			brelse(bh);
 310			goto fail;
 311		}
 312		dblock++;
 313		extlen--;
 314		memcpy(buf, bh->b_data + o, amount);
 315		brelse(bh);
 316		buf += (amount/sizeof(__be64));
 317		copied += amount;
 318		lblock++;
 319		o = sizeof(struct gfs2_meta_header);
 320	}
 321
 322	return copied;
 323fail:
 324	return (copied) ? copied : error;
 325}
 326
 327/**
 328 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
 329 * @ip: The inode in question
 330 *
 331 * Returns: The hash table or an error
 332 */
 333
 334static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
 335{
 336	struct inode *inode = &ip->i_inode;
 337	int ret;
 338	u32 hsize;
 339	__be64 *hc;
 340
 341	BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
 342
 343	hc = ip->i_hash_cache;
 344	if (hc)
 345		return hc;
 346
 347	hsize = BIT(ip->i_depth);
 348	hsize *= sizeof(__be64);
 349	if (hsize != i_size_read(&ip->i_inode)) {
 350		gfs2_consist_inode(ip);
 351		return ERR_PTR(-EIO);
 352	}
 353
 354	hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
 355	if (hc == NULL)
 356		hc = __vmalloc(hsize, GFP_NOFS);
 357
 358	if (hc == NULL)
 359		return ERR_PTR(-ENOMEM);
 360
 361	ret = gfs2_dir_read_data(ip, hc, hsize);
 362	if (ret < 0) {
 363		kvfree(hc);
 364		return ERR_PTR(ret);
 365	}
 366
 367	spin_lock(&inode->i_lock);
 368	if (likely(!ip->i_hash_cache)) {
 369		ip->i_hash_cache = hc;
 370		hc = NULL;
 371	}
 372	spin_unlock(&inode->i_lock);
 373	kvfree(hc);
 374
 375	return ip->i_hash_cache;
 376}
 377
 378/**
 379 * gfs2_dir_hash_inval - Invalidate dir hash
 380 * @ip: The directory inode
 381 *
 382 * Must be called with an exclusive glock, or during glock invalidation.
 383 */
 384void gfs2_dir_hash_inval(struct gfs2_inode *ip)
 385{
 386	__be64 *hc;
 387
 388	spin_lock(&ip->i_inode.i_lock);
 389	hc = ip->i_hash_cache;
 390	ip->i_hash_cache = NULL;
 391	spin_unlock(&ip->i_inode.i_lock);
 392
 393	kvfree(hc);
 394}
 395
 396static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
 397{
 398	return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
 399}
 400
 401static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
 402				     const struct qstr *name, int ret)
 403{
 404	if (!gfs2_dirent_sentinel(dent) &&
 405	    be32_to_cpu(dent->de_hash) == name->hash &&
 406	    be16_to_cpu(dent->de_name_len) == name->len &&
 407	    memcmp(dent+1, name->name, name->len) == 0)
 408		return ret;
 409	return 0;
 410}
 411
 412static int gfs2_dirent_find(const struct gfs2_dirent *dent,
 413			    const struct qstr *name,
 414			    void *opaque)
 415{
 416	return __gfs2_dirent_find(dent, name, 1);
 417}
 418
 419static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
 420			    const struct qstr *name,
 421			    void *opaque)
 422{
 423	return __gfs2_dirent_find(dent, name, 2);
 424}
 425
 426/*
 427 * name->name holds ptr to start of block.
 428 * name->len holds size of block.
 429 */
 430static int gfs2_dirent_last(const struct gfs2_dirent *dent,
 431			    const struct qstr *name,
 432			    void *opaque)
 433{
 434	const char *start = name->name;
 435	const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
 436	if (name->len == (end - start))
 437		return 1;
 438	return 0;
 439}
 440
 441/* Look for the dirent that contains the offset specified in data. Once we
 442 * find that dirent, there must be space available there for the new dirent */
 443static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
 444				  const struct qstr *name,
 445				  void *ptr)
 446{
 447	unsigned required = GFS2_DIRENT_SIZE(name->len);
 448	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 449	unsigned totlen = be16_to_cpu(dent->de_rec_len);
 450
 451	if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
 452		return 0;
 453	if (gfs2_dirent_sentinel(dent))
 454		actual = 0;
 455	if (ptr < (void *)dent + actual)
 456		return -1;
 457	if ((void *)dent + totlen >= ptr + required)
 458		return 1;
 459	return -1;
 460}
 461
 462static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
 463				  const struct qstr *name,
 464				  void *opaque)
 465{
 466	unsigned required = GFS2_DIRENT_SIZE(name->len);
 467	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 468	unsigned totlen = be16_to_cpu(dent->de_rec_len);
 469
 470	if (gfs2_dirent_sentinel(dent))
 471		actual = 0;
 472	if (totlen - actual >= required)
 473		return 1;
 474	return 0;
 475}
 476
 477struct dirent_gather {
 478	const struct gfs2_dirent **pdent;
 479	unsigned offset;
 480};
 481
 482static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
 483			      const struct qstr *name,
 484			      void *opaque)
 485{
 486	struct dirent_gather *g = opaque;
 487	if (!gfs2_dirent_sentinel(dent)) {
 488		g->pdent[g->offset++] = dent;
 489	}
 490	return 0;
 491}
 492
 493/*
 494 * Other possible things to check:
 495 * - Inode located within filesystem size (and on valid block)
 496 * - Valid directory entry type
 497 * Not sure how heavy-weight we want to make this... could also check
 498 * hash is correct for example, but that would take a lot of extra time.
 499 * For now the most important thing is to check that the various sizes
 500 * are correct.
 501 */
 502static int gfs2_check_dirent(struct gfs2_sbd *sdp,
 503			     struct gfs2_dirent *dent, unsigned int offset,
 504			     unsigned int size, unsigned int len, int first)
 505{
 506	const char *msg = "gfs2_dirent too small";
 507	if (unlikely(size < sizeof(struct gfs2_dirent)))
 508		goto error;
 509	msg = "gfs2_dirent misaligned";
 510	if (unlikely(offset & 0x7))
 511		goto error;
 512	msg = "gfs2_dirent points beyond end of block";
 513	if (unlikely(offset + size > len))
 514		goto error;
 515	msg = "zero inode number";
 516	if (unlikely(!first && gfs2_dirent_sentinel(dent)))
 517		goto error;
 518	msg = "name length is greater than space in dirent";
 519	if (!gfs2_dirent_sentinel(dent) &&
 520	    unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
 521		     size))
 522		goto error;
 523	return 0;
 524error:
 525	fs_warn(sdp, "%s: %s (%s)\n",
 526		__func__, msg, first ? "first in block" : "not first in block");
 527	return -EIO;
 528}
 529
 530static int gfs2_dirent_offset(struct gfs2_sbd *sdp, const void *buf)
 531{
 532	const struct gfs2_meta_header *h = buf;
 533	int offset;
 534
 535	BUG_ON(buf == NULL);
 536
 537	switch(be32_to_cpu(h->mh_type)) {
 538	case GFS2_METATYPE_LF:
 539		offset = sizeof(struct gfs2_leaf);
 540		break;
 541	case GFS2_METATYPE_DI:
 542		offset = sizeof(struct gfs2_dinode);
 543		break;
 544	default:
 545		goto wrong_type;
 546	}
 547	return offset;
 548wrong_type:
 549	fs_warn(sdp, "%s: wrong block type %u\n", __func__,
 550		be32_to_cpu(h->mh_type));
 551	return -1;
 552}
 553
 554static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
 555					    unsigned int len, gfs2_dscan_t scan,
 556					    const struct qstr *name,
 557					    void *opaque)
 558{
 559	struct gfs2_dirent *dent, *prev;
 560	unsigned offset;
 561	unsigned size;
 562	int ret = 0;
 563
 564	ret = gfs2_dirent_offset(GFS2_SB(inode), buf);
 565	if (ret < 0)
 566		goto consist_inode;
 567
 568	offset = ret;
 569	prev = NULL;
 570	dent = buf + offset;
 571	size = be16_to_cpu(dent->de_rec_len);
 572	if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size, len, 1))
 573		goto consist_inode;
 574	do {
 575		ret = scan(dent, name, opaque);
 576		if (ret)
 577			break;
 578		offset += size;
 579		if (offset == len)
 580			break;
 581		prev = dent;
 582		dent = buf + offset;
 583		size = be16_to_cpu(dent->de_rec_len);
 584		if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size,
 585				      len, 0))
 586			goto consist_inode;
 587	} while(1);
 588
 589	switch(ret) {
 590	case 0:
 591		return NULL;
 592	case 1:
 593		return dent;
 594	case 2:
 595		return prev ? prev : dent;
 596	default:
 597		BUG_ON(ret > 0);
 598		return ERR_PTR(ret);
 599	}
 600
 601consist_inode:
 602	gfs2_consist_inode(GFS2_I(inode));
 603	return ERR_PTR(-EIO);
 604}
 605
 606static int dirent_check_reclen(struct gfs2_inode *dip,
 607			       const struct gfs2_dirent *d, const void *end_p)
 608{
 609	const void *ptr = d;
 610	u16 rec_len = be16_to_cpu(d->de_rec_len);
 611
 612	if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
 613		goto broken;
 614	ptr += rec_len;
 615	if (ptr < end_p)
 616		return rec_len;
 617	if (ptr == end_p)
 618		return -ENOENT;
 619broken:
 620	gfs2_consist_inode(dip);
 621	return -EIO;
 622}
 623
 624/**
 625 * dirent_next - Next dirent
 626 * @dip: the directory
 627 * @bh: The buffer
 628 * @dent: Pointer to list of dirents
 629 *
 630 * Returns: 0 on success, error code otherwise
 631 */
 632
 633static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
 634		       struct gfs2_dirent **dent)
 635{
 636	struct gfs2_dirent *cur = *dent, *tmp;
 637	char *bh_end = bh->b_data + bh->b_size;
 638	int ret;
 639
 640	ret = dirent_check_reclen(dip, cur, bh_end);
 641	if (ret < 0)
 642		return ret;
 643
 644	tmp = (void *)cur + ret;
 645	ret = dirent_check_reclen(dip, tmp, bh_end);
 646	if (ret == -EIO)
 647		return ret;
 648
 649        /* Only the first dent could ever have de_inum.no_addr == 0 */
 650	if (gfs2_dirent_sentinel(tmp)) {
 651		gfs2_consist_inode(dip);
 652		return -EIO;
 653	}
 654
 655	*dent = tmp;
 656	return 0;
 657}
 658
 659/**
 660 * dirent_del - Delete a dirent
 661 * @dip: The GFS2 inode
 662 * @bh: The buffer
 663 * @prev: The previous dirent
 664 * @cur: The current dirent
 665 *
 666 */
 667
 668static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
 669		       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
 670{
 671	u16 cur_rec_len, prev_rec_len;
 672
 673	if (gfs2_dirent_sentinel(cur)) {
 674		gfs2_consist_inode(dip);
 675		return;
 676	}
 677
 678	gfs2_trans_add_meta(dip->i_gl, bh);
 679
 680	/* If there is no prev entry, this is the first entry in the block.
 681	   The de_rec_len is already as big as it needs to be.  Just zero
 682	   out the inode number and return.  */
 683
 684	if (!prev) {
 685		cur->de_inum.no_addr = 0;
 686		cur->de_inum.no_formal_ino = 0;
 687		return;
 688	}
 689
 690	/*  Combine this dentry with the previous one.  */
 691
 692	prev_rec_len = be16_to_cpu(prev->de_rec_len);
 693	cur_rec_len = be16_to_cpu(cur->de_rec_len);
 694
 695	if ((char *)prev + prev_rec_len != (char *)cur)
 696		gfs2_consist_inode(dip);
 697	if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
 698		gfs2_consist_inode(dip);
 699
 700	prev_rec_len += cur_rec_len;
 701	prev->de_rec_len = cpu_to_be16(prev_rec_len);
 702}
 703
 704
 705static struct gfs2_dirent *do_init_dirent(struct inode *inode,
 706					  struct gfs2_dirent *dent,
 707					  const struct qstr *name,
 708					  struct buffer_head *bh,
 709					  unsigned offset)
 710{
 711	struct gfs2_inode *ip = GFS2_I(inode);
 712	struct gfs2_dirent *ndent;
 713	unsigned totlen;
 714
 715	totlen = be16_to_cpu(dent->de_rec_len);
 716	BUG_ON(offset + name->len > totlen);
 717	gfs2_trans_add_meta(ip->i_gl, bh);
 718	ndent = (struct gfs2_dirent *)((char *)dent + offset);
 719	dent->de_rec_len = cpu_to_be16(offset);
 720	gfs2_qstr2dirent(name, totlen - offset, ndent);
 721	return ndent;
 722}
 723
 724
 725/*
 726 * Takes a dent from which to grab space as an argument. Returns the
 727 * newly created dent.
 728 */
 729static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
 730					    struct gfs2_dirent *dent,
 731					    const struct qstr *name,
 732					    struct buffer_head *bh)
 733{
 734	unsigned offset = 0;
 735
 736	if (!gfs2_dirent_sentinel(dent))
 737		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 738	return do_init_dirent(inode, dent, name, bh, offset);
 739}
 740
 741static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
 742						   struct buffer_head *bh,
 743						   const struct qstr *name,
 744						   void *ptr)
 745{
 746	struct gfs2_dirent *dent;
 747	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 748				gfs2_dirent_find_offset, name, ptr);
 749	if (IS_ERR_OR_NULL(dent))
 750		return dent;
 751	return do_init_dirent(inode, dent, name, bh,
 752			      (unsigned)(ptr - (void *)dent));
 753}
 754
 755static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
 756		    struct buffer_head **bhp)
 757{
 758	int error;
 759
 760	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
 761	if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
 762		/* pr_info("block num=%llu\n", leaf_no); */
 763		error = -EIO;
 764	}
 765
 766	return error;
 767}
 768
 769/**
 770 * get_leaf_nr - Get a leaf number associated with the index
 771 * @dip: The GFS2 inode
 772 * @index: hash table index of the targeted leaf
 773 * @leaf_out: Resulting leaf block number
 774 *
 775 * Returns: 0 on success, error code otherwise
 776 */
 777
 778static int get_leaf_nr(struct gfs2_inode *dip, u32 index, u64 *leaf_out)
 779{
 780	__be64 *hash;
 781	int error;
 782
 783	hash = gfs2_dir_get_hash_table(dip);
 784	error = PTR_ERR_OR_ZERO(hash);
 785
 786	if (!error)
 787		*leaf_out = be64_to_cpu(*(hash + index));
 788
 789	return error;
 790}
 791
 792static int get_first_leaf(struct gfs2_inode *dip, u32 index,
 793			  struct buffer_head **bh_out)
 794{
 795	u64 leaf_no;
 796	int error;
 797
 798	error = get_leaf_nr(dip, index, &leaf_no);
 799	if (!error)
 800		error = get_leaf(dip, leaf_no, bh_out);
 801
 802	return error;
 803}
 804
 805static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
 806					      const struct qstr *name,
 807					      gfs2_dscan_t scan,
 808					      struct buffer_head **pbh)
 809{
 810	struct buffer_head *bh;
 811	struct gfs2_dirent *dent;
 812	struct gfs2_inode *ip = GFS2_I(inode);
 813	int error;
 814
 815	if (ip->i_diskflags & GFS2_DIF_EXHASH) {
 816		struct gfs2_leaf *leaf;
 817		unsigned int hsize = BIT(ip->i_depth);
 818		unsigned int index;
 819		u64 ln;
 820		if (hsize * sizeof(u64) != i_size_read(inode)) {
 821			gfs2_consist_inode(ip);
 822			return ERR_PTR(-EIO);
 823		}
 824
 825		index = name->hash >> (32 - ip->i_depth);
 826		error = get_first_leaf(ip, index, &bh);
 827		if (error)
 828			return ERR_PTR(error);
 829		do {
 830			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 831						scan, name, NULL);
 832			if (dent)
 833				goto got_dent;
 834			leaf = (struct gfs2_leaf *)bh->b_data;
 835			ln = be64_to_cpu(leaf->lf_next);
 836			brelse(bh);
 837			if (!ln)
 838				break;
 839
 840			error = get_leaf(ip, ln, &bh);
 841		} while(!error);
 842
 843		return error ? ERR_PTR(error) : NULL;
 844	}
 845
 846
 847	error = gfs2_meta_inode_buffer(ip, &bh);
 848	if (error)
 849		return ERR_PTR(error);
 850	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
 851got_dent:
 852	if (IS_ERR_OR_NULL(dent)) {
 853		brelse(bh);
 854		bh = NULL;
 855	}
 856	*pbh = bh;
 857	return dent;
 858}
 859
 860static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
 861{
 862	struct gfs2_inode *ip = GFS2_I(inode);
 863	unsigned int n = 1;
 864	u64 bn;
 865	int error;
 866	struct buffer_head *bh;
 867	struct gfs2_leaf *leaf;
 868	struct gfs2_dirent *dent;
 869	struct timespec64 tv = current_time(inode);
 870
 871	error = gfs2_alloc_blocks(ip, &bn, &n, 0);
 872	if (error)
 873		return NULL;
 874	bh = gfs2_meta_new(ip->i_gl, bn);
 875	if (!bh)
 876		return NULL;
 877
 878	gfs2_trans_remove_revoke(GFS2_SB(inode), bn, 1);
 879	gfs2_trans_add_meta(ip->i_gl, bh);
 880	gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
 881	leaf = (struct gfs2_leaf *)bh->b_data;
 882	leaf->lf_depth = cpu_to_be16(depth);
 883	leaf->lf_entries = 0;
 884	leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
 885	leaf->lf_next = 0;
 886	leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
 887	leaf->lf_dist = cpu_to_be32(1);
 888	leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
 889	leaf->lf_sec = cpu_to_be64(tv.tv_sec);
 890	memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
 891	dent = (struct gfs2_dirent *)(leaf+1);
 892	gfs2_qstr2dirent(&empty_name, bh->b_size - sizeof(struct gfs2_leaf), dent);
 893	*pbh = bh;
 894	return leaf;
 895}
 896
 897/**
 898 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
 899 * @inode: The directory inode to be converted to exhash
 900 *
 901 * Returns: 0 on success, error code otherwise
 902 */
 903
 904static int dir_make_exhash(struct inode *inode)
 905{
 906	struct gfs2_inode *dip = GFS2_I(inode);
 907	struct gfs2_sbd *sdp = GFS2_SB(inode);
 908	struct gfs2_dirent *dent;
 909	struct qstr args;
 910	struct buffer_head *bh, *dibh;
 911	struct gfs2_leaf *leaf;
 912	int y;
 913	u32 x;
 914	__be64 *lp;
 915	u64 bn;
 916	int error;
 917
 918	error = gfs2_meta_inode_buffer(dip, &dibh);
 919	if (error)
 920		return error;
 921
 922	/*  Turn over a new leaf  */
 923
 924	leaf = new_leaf(inode, &bh, 0);
 925	if (!leaf)
 926		return -ENOSPC;
 927	bn = bh->b_blocknr;
 928
 929	gfs2_assert(sdp, dip->i_entries < BIT(16));
 930	leaf->lf_entries = cpu_to_be16(dip->i_entries);
 931
 932	/*  Copy dirents  */
 933
 934	gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
 935			     sizeof(struct gfs2_dinode));
 936
 937	/*  Find last entry  */
 938
 939	x = 0;
 940	args.len = bh->b_size - sizeof(struct gfs2_dinode) +
 941		   sizeof(struct gfs2_leaf);
 942	args.name = bh->b_data;
 943	dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
 944				gfs2_dirent_last, &args, NULL);
 945	if (!dent) {
 946		brelse(bh);
 947		brelse(dibh);
 948		return -EIO;
 949	}
 950	if (IS_ERR(dent)) {
 951		brelse(bh);
 952		brelse(dibh);
 953		return PTR_ERR(dent);
 954	}
 955
 956	/*  Adjust the last dirent's record length
 957	   (Remember that dent still points to the last entry.)  */
 958
 959	dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
 960		sizeof(struct gfs2_dinode) -
 961		sizeof(struct gfs2_leaf));
 962
 963	brelse(bh);
 964
 965	/*  We're done with the new leaf block, now setup the new
 966	    hash table.  */
 967
 968	gfs2_trans_add_meta(dip->i_gl, dibh);
 969	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 970
 971	lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
 972
 973	for (x = sdp->sd_hash_ptrs; x--; lp++)
 974		*lp = cpu_to_be64(bn);
 975
 976	i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
 977	gfs2_add_inode_blocks(&dip->i_inode, 1);
 978	dip->i_diskflags |= GFS2_DIF_EXHASH;
 979
 980	for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
 981	dip->i_depth = y;
 982
 983	gfs2_dinode_out(dip, dibh->b_data);
 984
 985	brelse(dibh);
 986
 987	return 0;
 988}
 989
 990/**
 991 * dir_split_leaf - Split a leaf block into two
 992 * @inode: The directory inode to be split
 993 * @name: name of the dirent we're trying to insert
 994 *
 995 * Returns: 0 on success, error code on failure
 996 */
 997
 998static int dir_split_leaf(struct inode *inode, const struct qstr *name)
 999{
1000	struct gfs2_inode *dip = GFS2_I(inode);
1001	struct buffer_head *nbh, *obh, *dibh;
1002	struct gfs2_leaf *nleaf, *oleaf;
1003	struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
1004	u32 start, len, half_len, divider;
1005	u64 bn, leaf_no;
1006	__be64 *lp;
1007	u32 index;
1008	int x;
1009	int error;
1010
1011	index = name->hash >> (32 - dip->i_depth);
1012	error = get_leaf_nr(dip, index, &leaf_no);
1013	if (error)
1014		return error;
1015
1016	/*  Get the old leaf block  */
1017	error = get_leaf(dip, leaf_no, &obh);
1018	if (error)
1019		return error;
1020
1021	oleaf = (struct gfs2_leaf *)obh->b_data;
1022	if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
1023		brelse(obh);
1024		return 1; /* can't split */
1025	}
1026
1027	gfs2_trans_add_meta(dip->i_gl, obh);
1028
1029	nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
1030	if (!nleaf) {
1031		brelse(obh);
1032		return -ENOSPC;
1033	}
1034	bn = nbh->b_blocknr;
1035
1036	/*  Compute the start and len of leaf pointers in the hash table.  */
1037	len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth));
1038	half_len = len >> 1;
1039	if (!half_len) {
1040		fs_warn(GFS2_SB(inode), "i_depth %u lf_depth %u index %u\n",
1041			dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
1042		gfs2_consist_inode(dip);
1043		error = -EIO;
1044		goto fail_brelse;
1045	}
1046
1047	start = (index & ~(len - 1));
1048
1049	/* Change the pointers.
1050	   Don't bother distinguishing stuffed from non-stuffed.
1051	   This code is complicated enough already. */
1052	lp = kmalloc_array(half_len, sizeof(__be64), GFP_NOFS);
1053	if (!lp) {
1054		error = -ENOMEM;
1055		goto fail_brelse;
1056	}
1057
1058	/*  Change the pointers  */
1059	for (x = 0; x < half_len; x++)
1060		lp[x] = cpu_to_be64(bn);
1061
1062	gfs2_dir_hash_inval(dip);
1063
1064	error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1065				    half_len * sizeof(u64));
1066	if (error != half_len * sizeof(u64)) {
1067		if (error >= 0)
1068			error = -EIO;
1069		goto fail_lpfree;
1070	}
1071
1072	kfree(lp);
1073
1074	/*  Compute the divider  */
1075	divider = (start + half_len) << (32 - dip->i_depth);
1076
1077	/*  Copy the entries  */
1078	dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1079
1080	do {
1081		next = dent;
1082		if (dirent_next(dip, obh, &next))
1083			next = NULL;
1084
1085		if (!gfs2_dirent_sentinel(dent) &&
1086		    be32_to_cpu(dent->de_hash) < divider) {
1087			struct qstr str;
1088			void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
1089			str.name = (char*)(dent+1);
1090			str.len = be16_to_cpu(dent->de_name_len);
1091			str.hash = be32_to_cpu(dent->de_hash);
1092			new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
1093			if (IS_ERR(new)) {
1094				error = PTR_ERR(new);
1095				break;
1096			}
1097
1098			new->de_inum = dent->de_inum; /* No endian worries */
1099			new->de_type = dent->de_type; /* No endian worries */
1100			be16_add_cpu(&nleaf->lf_entries, 1);
1101
1102			dirent_del(dip, obh, prev, dent);
1103
1104			if (!oleaf->lf_entries)
1105				gfs2_consist_inode(dip);
1106			be16_add_cpu(&oleaf->lf_entries, -1);
1107
1108			if (!prev)
1109				prev = dent;
1110		} else {
1111			prev = dent;
1112		}
1113		dent = next;
1114	} while (dent);
1115
1116	oleaf->lf_depth = nleaf->lf_depth;
1117
1118	error = gfs2_meta_inode_buffer(dip, &dibh);
1119	if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1120		gfs2_trans_add_meta(dip->i_gl, dibh);
1121		gfs2_add_inode_blocks(&dip->i_inode, 1);
1122		gfs2_dinode_out(dip, dibh->b_data);
1123		brelse(dibh);
1124	}
1125
1126	brelse(obh);
1127	brelse(nbh);
1128
1129	return error;
1130
1131fail_lpfree:
1132	kfree(lp);
1133
1134fail_brelse:
1135	brelse(obh);
1136	brelse(nbh);
1137	return error;
1138}
1139
1140/**
1141 * dir_double_exhash - Double size of ExHash table
1142 * @dip: The GFS2 dinode
1143 *
1144 * Returns: 0 on success, error code on failure
1145 */
1146
1147static int dir_double_exhash(struct gfs2_inode *dip)
1148{
1149	struct buffer_head *dibh;
1150	u32 hsize;
1151	u32 hsize_bytes;
1152	__be64 *hc;
1153	__be64 *hc2, *h;
1154	int x;
1155	int error = 0;
1156
1157	hsize = BIT(dip->i_depth);
1158	hsize_bytes = hsize * sizeof(__be64);
1159
1160	hc = gfs2_dir_get_hash_table(dip);
1161	if (IS_ERR(hc))
1162		return PTR_ERR(hc);
1163
1164	hc2 = kmalloc_array(hsize_bytes, 2, GFP_NOFS | __GFP_NOWARN);
1165	if (hc2 == NULL)
1166		hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS);
1167
1168	if (!hc2)
1169		return -ENOMEM;
1170
1171	h = hc2;
1172	error = gfs2_meta_inode_buffer(dip, &dibh);
1173	if (error)
1174		goto out_kfree;
1175
1176	for (x = 0; x < hsize; x++) {
1177		*h++ = *hc;
1178		*h++ = *hc;
1179		hc++;
1180	}
1181
1182	error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1183	if (error != (hsize_bytes * 2))
1184		goto fail;
1185
1186	gfs2_dir_hash_inval(dip);
1187	dip->i_hash_cache = hc2;
1188	dip->i_depth++;
1189	gfs2_dinode_out(dip, dibh->b_data);
1190	brelse(dibh);
1191	return 0;
1192
1193fail:
1194	/* Replace original hash table & size */
1195	gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1196	i_size_write(&dip->i_inode, hsize_bytes);
1197	gfs2_dinode_out(dip, dibh->b_data);
1198	brelse(dibh);
1199out_kfree:
1200	kvfree(hc2);
1201	return error;
1202}
1203
1204/**
1205 * compare_dents - compare directory entries by hash value
1206 * @a: first dent
1207 * @b: second dent
1208 *
1209 * When comparing the hash entries of @a to @b:
1210 *   gt: returns 1
1211 *   lt: returns -1
1212 *   eq: returns 0
1213 */
1214
1215static int compare_dents(const void *a, const void *b)
1216{
1217	const struct gfs2_dirent *dent_a, *dent_b;
1218	u32 hash_a, hash_b;
1219	int ret = 0;
1220
1221	dent_a = *(const struct gfs2_dirent **)a;
1222	hash_a = dent_a->de_cookie;
1223
1224	dent_b = *(const struct gfs2_dirent **)b;
1225	hash_b = dent_b->de_cookie;
1226
1227	if (hash_a > hash_b)
1228		ret = 1;
1229	else if (hash_a < hash_b)
1230		ret = -1;
1231	else {
1232		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1233		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1234
1235		if (len_a > len_b)
1236			ret = 1;
1237		else if (len_a < len_b)
1238			ret = -1;
1239		else
1240			ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1241	}
1242
1243	return ret;
1244}
1245
1246/**
1247 * do_filldir_main - read out directory entries
1248 * @dip: The GFS2 inode
1249 * @ctx: what to feed the entries to
1250 * @darr: an array of struct gfs2_dirent pointers to read
1251 * @entries: the number of entries in darr
1252 * @sort_start: index of the directory array to start our sort
1253 * @copied: pointer to int that's non-zero if a entry has been copied out
1254 *
1255 * Jump through some hoops to make sure that if there are hash collsions,
1256 * they are read out at the beginning of a buffer.  We want to minimize
1257 * the possibility that they will fall into different readdir buffers or
1258 * that someone will want to seek to that location.
1259 *
1260 * Returns: errno, >0 if the actor tells you to stop
1261 */
1262
1263static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1264			   struct gfs2_dirent **darr, u32 entries,
1265			   u32 sort_start, int *copied)
1266{
1267	const struct gfs2_dirent *dent, *dent_next;
1268	u64 off, off_next;
1269	unsigned int x, y;
1270	int run = 0;
1271
1272	if (sort_start < entries)
1273		sort(&darr[sort_start], entries - sort_start,
1274		     sizeof(struct gfs2_dirent *), compare_dents, NULL);
1275
1276	dent_next = darr[0];
1277	off_next = dent_next->de_cookie;
1278
1279	for (x = 0, y = 1; x < entries; x++, y++) {
1280		dent = dent_next;
1281		off = off_next;
1282
1283		if (y < entries) {
1284			dent_next = darr[y];
1285			off_next = dent_next->de_cookie;
1286
1287			if (off < ctx->pos)
1288				continue;
1289			ctx->pos = off;
1290
1291			if (off_next == off) {
1292				if (*copied && !run)
1293					return 1;
1294				run = 1;
1295			} else
1296				run = 0;
1297		} else {
1298			if (off < ctx->pos)
1299				continue;
1300			ctx->pos = off;
1301		}
1302
1303		if (!dir_emit(ctx, (const char *)(dent + 1),
1304				be16_to_cpu(dent->de_name_len),
1305				be64_to_cpu(dent->de_inum.no_addr),
1306				be16_to_cpu(dent->de_type)))
1307			return 1;
1308
1309		*copied = 1;
1310	}
1311
1312	/* Increment the ctx->pos by one, so the next time we come into the
1313	   do_filldir fxn, we get the next entry instead of the last one in the
1314	   current leaf */
1315
1316	ctx->pos++;
1317
1318	return 0;
1319}
1320
1321static void *gfs2_alloc_sort_buffer(unsigned size)
1322{
1323	void *ptr = NULL;
1324
1325	if (size < KMALLOC_MAX_SIZE)
1326		ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1327	if (!ptr)
1328		ptr = __vmalloc(size, GFP_NOFS);
1329	return ptr;
1330}
1331
1332
1333static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
1334			    unsigned leaf_nr, struct gfs2_dirent **darr,
1335			    unsigned entries)
1336{
1337	int sort_id = -1;
1338	int i;
1339	
1340	for (i = 0; i < entries; i++) {
1341		unsigned offset;
1342
1343		darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
1344		darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
1345
1346		if (!sdp->sd_args.ar_loccookie)
1347			continue;
1348		offset = (char *)(darr[i]) -
1349			(bh->b_data + gfs2_dirent_offset(sdp, bh->b_data));
1350		offset /= GFS2_MIN_DIRENT_SIZE;
1351		offset += leaf_nr * sdp->sd_max_dents_per_leaf;
1352		if (offset >= GFS2_USE_HASH_FLAG ||
1353		    leaf_nr >= GFS2_USE_HASH_FLAG) {
1354			darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
1355			if (sort_id < 0)
1356				sort_id = i;
1357			continue;
1358		}
1359		darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
1360		darr[i]->de_cookie |= offset;
1361	}
1362	return sort_id;
1363}	
1364
1365
1366static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
1367			      int *copied, unsigned *depth,
1368			      u64 leaf_no)
1369{
1370	struct gfs2_inode *ip = GFS2_I(inode);
1371	struct gfs2_sbd *sdp = GFS2_SB(inode);
1372	struct buffer_head *bh;
1373	struct gfs2_leaf *lf;
1374	unsigned entries = 0, entries2 = 0;
1375	unsigned leaves = 0, leaf = 0, offset, sort_offset;
1376	struct gfs2_dirent **darr, *dent;
1377	struct dirent_gather g;
1378	struct buffer_head **larr;
1379	int error, i, need_sort = 0, sort_id;
1380	u64 lfn = leaf_no;
1381
1382	do {
1383		error = get_leaf(ip, lfn, &bh);
1384		if (error)
1385			goto out;
1386		lf = (struct gfs2_leaf *)bh->b_data;
1387		if (leaves == 0)
1388			*depth = be16_to_cpu(lf->lf_depth);
1389		entries += be16_to_cpu(lf->lf_entries);
1390		leaves++;
1391		lfn = be64_to_cpu(lf->lf_next);
1392		brelse(bh);
1393	} while(lfn);
1394
1395	if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
1396		need_sort = 1;
1397		sort_offset = 0;
1398	}
1399
1400	if (!entries)
1401		return 0;
1402
1403	error = -ENOMEM;
1404	/*
1405	 * The extra 99 entries are not normally used, but are a buffer
1406	 * zone in case the number of entries in the leaf is corrupt.
1407	 * 99 is the maximum number of entries that can fit in a single
1408	 * leaf block.
1409	 */
1410	larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1411	if (!larr)
1412		goto out;
1413	darr = (struct gfs2_dirent **)(larr + leaves);
1414	g.pdent = (const struct gfs2_dirent **)darr;
1415	g.offset = 0;
1416	lfn = leaf_no;
1417
1418	do {
1419		error = get_leaf(ip, lfn, &bh);
1420		if (error)
1421			goto out_free;
1422		lf = (struct gfs2_leaf *)bh->b_data;
1423		lfn = be64_to_cpu(lf->lf_next);
1424		if (lf->lf_entries) {
1425			offset = g.offset;
1426			entries2 += be16_to_cpu(lf->lf_entries);
1427			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1428						gfs2_dirent_gather, NULL, &g);
1429			error = PTR_ERR(dent);
1430			if (IS_ERR(dent))
1431				goto out_free;
1432			if (entries2 != g.offset) {
1433				fs_warn(sdp, "Number of entries corrupt in dir "
1434						"leaf %llu, entries2 (%u) != "
1435						"g.offset (%u)\n",
1436					(unsigned long long)bh->b_blocknr,
1437					entries2, g.offset);
1438				gfs2_consist_inode(ip);
1439				error = -EIO;
1440				goto out_free;
1441			}
1442			error = 0;
1443			sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
1444						   be16_to_cpu(lf->lf_entries));
1445			if (!need_sort && sort_id >= 0) {
1446				need_sort = 1;
1447				sort_offset = offset + sort_id;
1448			}
1449			larr[leaf++] = bh;
1450		} else {
1451			larr[leaf++] = NULL;
1452			brelse(bh);
1453		}
1454	} while(lfn);
1455
1456	BUG_ON(entries2 != entries);
1457	error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
1458				sort_offset : entries, copied);
1459out_free:
1460	for(i = 0; i < leaf; i++)
1461		brelse(larr[i]);
1462	kvfree(larr);
1463out:
1464	return error;
1465}
1466
1467/**
1468 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1469 * @inode: the directory inode
1470 * @hsize: hash table size
1471 * @index: index into the hash table
1472 * @f_ra: read-ahead parameters
1473 *
1474 * Note: we can't calculate each index like dir_e_read can because we don't
1475 * have the leaf, and therefore we don't have the depth, and therefore we
1476 * don't have the length. So we have to just read enough ahead to make up
1477 * for the loss of information.
1478 */
1479static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1480			       struct file_ra_state *f_ra)
1481{
1482	struct gfs2_inode *ip = GFS2_I(inode);
1483	struct gfs2_glock *gl = ip->i_gl;
1484	struct buffer_head *bh;
1485	u64 blocknr = 0, last;
1486	unsigned count;
1487
1488	/* First check if we've already read-ahead for the whole range. */
1489	if (index + MAX_RA_BLOCKS < f_ra->start)
1490		return;
1491
1492	f_ra->start = max((pgoff_t)index, f_ra->start);
1493	for (count = 0; count < MAX_RA_BLOCKS; count++) {
1494		if (f_ra->start >= hsize) /* if exceeded the hash table */
1495			break;
1496
1497		last = blocknr;
1498		blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1499		f_ra->start++;
1500		if (blocknr == last)
1501			continue;
1502
1503		bh = gfs2_getbuf(gl, blocknr, 1);
1504		if (trylock_buffer(bh)) {
1505			if (buffer_uptodate(bh)) {
1506				unlock_buffer(bh);
1507				brelse(bh);
1508				continue;
1509			}
1510			bh->b_end_io = end_buffer_read_sync;
1511			submit_bh(REQ_OP_READ | REQ_RAHEAD | REQ_META |
1512				  REQ_PRIO, bh);
1513			continue;
1514		}
1515		brelse(bh);
1516	}
1517}
1518
1519/**
1520 * dir_e_read - Reads the entries from a directory into a filldir buffer
1521 * @inode: the directory inode
1522 * @ctx: actor to feed the entries to
1523 * @f_ra: read-ahead parameters
1524 *
1525 * Returns: errno
1526 */
1527
1528static int dir_e_read(struct inode *inode, struct dir_context *ctx,
1529		      struct file_ra_state *f_ra)
1530{
1531	struct gfs2_inode *dip = GFS2_I(inode);
1532	u32 hsize, len = 0;
1533	u32 hash, index;
1534	__be64 *lp;
1535	int copied = 0;
1536	int error = 0;
1537	unsigned depth = 0;
1538
1539	hsize = BIT(dip->i_depth);
1540	hash = gfs2_dir_offset2hash(ctx->pos);
1541	index = hash >> (32 - dip->i_depth);
1542
1543	if (dip->i_hash_cache == NULL)
1544		f_ra->start = 0;
1545	lp = gfs2_dir_get_hash_table(dip);
1546	if (IS_ERR(lp))
1547		return PTR_ERR(lp);
1548
1549	gfs2_dir_readahead(inode, hsize, index, f_ra);
1550
1551	while (index < hsize) {
1552		error = gfs2_dir_read_leaf(inode, ctx,
1553					   &copied, &depth,
1554					   be64_to_cpu(lp[index]));
1555		if (error)
1556			break;
1557
1558		len = BIT(dip->i_depth - depth);
1559		index = (index & ~(len - 1)) + len;
1560	}
1561
1562	if (error > 0)
1563		error = 0;
1564	return error;
1565}
1566
1567int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
1568		  struct file_ra_state *f_ra)
1569{
1570	struct gfs2_inode *dip = GFS2_I(inode);
1571	struct gfs2_sbd *sdp = GFS2_SB(inode);
1572	struct dirent_gather g;
1573	struct gfs2_dirent **darr, *dent;
1574	struct buffer_head *dibh;
1575	int copied = 0;
1576	int error;
1577
1578	if (!dip->i_entries)
1579		return 0;
1580
1581	if (dip->i_diskflags & GFS2_DIF_EXHASH)
1582		return dir_e_read(inode, ctx, f_ra);
1583
1584	if (!gfs2_is_stuffed(dip)) {
1585		gfs2_consist_inode(dip);
1586		return -EIO;
1587	}
1588
1589	error = gfs2_meta_inode_buffer(dip, &dibh);
1590	if (error)
1591		return error;
1592
1593	error = -ENOMEM;
1594	/* 96 is max number of dirents which can be stuffed into an inode */
1595	darr = kmalloc_array(96, sizeof(struct gfs2_dirent *), GFP_NOFS);
1596	if (darr) {
1597		g.pdent = (const struct gfs2_dirent **)darr;
1598		g.offset = 0;
1599		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1600					gfs2_dirent_gather, NULL, &g);
1601		if (IS_ERR(dent)) {
1602			error = PTR_ERR(dent);
1603			goto out;
1604		}
1605		if (dip->i_entries != g.offset) {
1606			fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1607				"ip->i_entries (%u) != g.offset (%u)\n",
1608				(unsigned long long)dip->i_no_addr,
1609				dip->i_entries,
1610				g.offset);
1611			gfs2_consist_inode(dip);
1612			error = -EIO;
1613			goto out;
1614		}
1615		gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
1616		error = do_filldir_main(dip, ctx, darr,
1617					dip->i_entries, 0, &copied);
1618out:
1619		kfree(darr);
1620	}
1621
1622	if (error > 0)
1623		error = 0;
1624
1625	brelse(dibh);
1626
1627	return error;
1628}
1629
1630/**
1631 * gfs2_dir_search - Search a directory
1632 * @dir: The GFS2 directory inode
1633 * @name: The name we are looking up
1634 * @fail_on_exist: Fail if the name exists rather than looking it up
1635 *
1636 * This routine searches a directory for a file or another directory.
1637 * Assumes a glock is held on dip.
1638 *
1639 * Returns: errno
1640 */
1641
1642struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
1643			      bool fail_on_exist)
1644{
1645	struct buffer_head *bh;
1646	struct gfs2_dirent *dent;
1647	u64 addr, formal_ino;
1648	u16 dtype;
1649
1650	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1651	if (dent) {
1652		struct inode *inode;
1653		u16 rahead;
1654
1655		if (IS_ERR(dent))
1656			return ERR_CAST(dent);
1657		dtype = be16_to_cpu(dent->de_type);
1658		rahead = be16_to_cpu(dent->de_rahead);
1659		addr = be64_to_cpu(dent->de_inum.no_addr);
1660		formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
1661		brelse(bh);
1662		if (fail_on_exist)
1663			return ERR_PTR(-EEXIST);
1664		inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino,
1665					  GFS2_BLKST_FREE /* ignore */);
1666		if (!IS_ERR(inode))
1667			GFS2_I(inode)->i_rahead = rahead;
1668		return inode;
1669	}
1670	return ERR_PTR(-ENOENT);
1671}
1672
1673int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1674		   const struct gfs2_inode *ip)
1675{
1676	struct buffer_head *bh;
1677	struct gfs2_dirent *dent;
1678	int ret = -ENOENT;
1679
1680	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1681	if (dent) {
1682		if (IS_ERR(dent))
1683			return PTR_ERR(dent);
1684		if (ip) {
1685			if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1686				goto out;
1687			if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1688			    ip->i_no_formal_ino)
1689				goto out;
1690			if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1691			    be16_to_cpu(dent->de_type))) {
1692				gfs2_consist_inode(GFS2_I(dir));
1693				ret = -EIO;
1694				goto out;
1695			}
1696		}
1697		ret = 0;
1698out:
1699		brelse(bh);
1700	}
1701	return ret;
1702}
1703
1704/**
1705 * dir_new_leaf - Add a new leaf onto hash chain
1706 * @inode: The directory
1707 * @name: The name we are adding
1708 *
1709 * This adds a new dir leaf onto an existing leaf when there is not
1710 * enough space to add a new dir entry. This is a last resort after
1711 * we've expanded the hash table to max size and also split existing
1712 * leaf blocks, so it will only occur for very large directories.
1713 *
1714 * The dist parameter is set to 1 for leaf blocks directly attached
1715 * to the hash table, 2 for one layer of indirection, 3 for two layers
1716 * etc. We are thus able to tell the difference between an old leaf
1717 * with dist set to zero (i.e. "don't know") and a new one where we
1718 * set this information for debug/fsck purposes.
1719 *
1720 * Returns: 0 on success, or -ve on error
1721 */
1722
1723static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1724{
1725	struct buffer_head *bh, *obh;
1726	struct gfs2_inode *ip = GFS2_I(inode);
1727	struct gfs2_leaf *leaf, *oleaf;
1728	u32 dist = 1;
1729	int error;
1730	u32 index;
1731	u64 bn;
1732
1733	index = name->hash >> (32 - ip->i_depth);
1734	error = get_first_leaf(ip, index, &obh);
1735	if (error)
1736		return error;
1737	do {
1738		dist++;
1739		oleaf = (struct gfs2_leaf *)obh->b_data;
1740		bn = be64_to_cpu(oleaf->lf_next);
1741		if (!bn)
1742			break;
1743		brelse(obh);
1744		error = get_leaf(ip, bn, &obh);
1745		if (error)
1746			return error;
1747	} while(1);
1748
1749	gfs2_trans_add_meta(ip->i_gl, obh);
1750
1751	leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1752	if (!leaf) {
1753		brelse(obh);
1754		return -ENOSPC;
1755	}
1756	leaf->lf_dist = cpu_to_be32(dist);
1757	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1758	brelse(bh);
1759	brelse(obh);
1760
1761	error = gfs2_meta_inode_buffer(ip, &bh);
1762	if (error)
1763		return error;
1764	gfs2_trans_add_meta(ip->i_gl, bh);
1765	gfs2_add_inode_blocks(&ip->i_inode, 1);
1766	gfs2_dinode_out(ip, bh->b_data);
1767	brelse(bh);
1768	return 0;
1769}
1770
1771static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip)
1772{
1773	u64 where = ip->i_no_addr + 1;
1774	if (ip->i_eattr == where)
1775		return 1;
1776	return 0;
1777}
1778
1779/**
1780 * gfs2_dir_add - Add new filename into directory
1781 * @inode: The directory inode
1782 * @name: The new name
1783 * @nip: The GFS2 inode to be linked in to the directory
1784 * @da: The directory addition info
1785 *
1786 * If the call to gfs2_diradd_alloc_required resulted in there being
1787 * no need to allocate any new directory blocks, then it will contain
1788 * a pointer to the directory entry and the bh in which it resides. We
1789 * can use that without having to repeat the search. If there was no
1790 * free space, then we must now create more space.
1791 *
1792 * Returns: 0 on success, error code on failure
1793 */
1794
1795int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1796		 const struct gfs2_inode *nip, struct gfs2_diradd *da)
1797{
1798	struct gfs2_inode *ip = GFS2_I(inode);
1799	struct buffer_head *bh = da->bh;
1800	struct gfs2_dirent *dent = da->dent;
1801	struct timespec64 tv;
1802	struct gfs2_leaf *leaf;
1803	int error;
1804
1805	while(1) {
1806		if (da->bh == NULL) {
1807			dent = gfs2_dirent_search(inode, name,
1808						  gfs2_dirent_find_space, &bh);
1809		}
1810		if (dent) {
1811			if (IS_ERR(dent))
1812				return PTR_ERR(dent);
1813			dent = gfs2_init_dirent(inode, dent, name, bh);
1814			gfs2_inum_out(nip, dent);
1815			dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1816			dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip));
1817			tv = inode_set_ctime_current(&ip->i_inode);
1818			if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1819				leaf = (struct gfs2_leaf *)bh->b_data;
1820				be16_add_cpu(&leaf->lf_entries, 1);
1821				leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1822				leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1823			}
1824			da->dent = NULL;
1825			da->bh = NULL;
1826			brelse(bh);
1827			ip->i_entries++;
1828			inode_set_mtime_to_ts(&ip->i_inode, tv);
1829			if (S_ISDIR(nip->i_inode.i_mode))
1830				inc_nlink(&ip->i_inode);
1831			mark_inode_dirty(inode);
1832			error = 0;
1833			break;
1834		}
1835		if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1836			error = dir_make_exhash(inode);
1837			if (error)
1838				break;
1839			continue;
1840		}
1841		error = dir_split_leaf(inode, name);
1842		if (error == 0)
1843			continue;
1844		if (error < 0)
1845			break;
1846		if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1847			error = dir_double_exhash(ip);
1848			if (error)
1849				break;
1850			error = dir_split_leaf(inode, name);
1851			if (error < 0)
1852				break;
1853			if (error == 0)
1854				continue;
1855		}
1856		error = dir_new_leaf(inode, name);
1857		if (!error)
1858			continue;
1859		error = -ENOSPC;
1860		break;
1861	}
1862	return error;
1863}
1864
1865
1866/**
1867 * gfs2_dir_del - Delete a directory entry
1868 * @dip: The GFS2 inode
1869 * @dentry: The directory entry we want to delete
1870 *
1871 * Returns: 0 on success, error code on failure
1872 */
1873
1874int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
1875{
1876	const struct qstr *name = &dentry->d_name;
1877	struct gfs2_dirent *dent, *prev = NULL;
1878	struct buffer_head *bh;
1879	struct timespec64 tv;
1880
1881	/* Returns _either_ the entry (if its first in block) or the
1882	   previous entry otherwise */
1883	dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1884	if (!dent) {
1885		gfs2_consist_inode(dip);
1886		return -EIO;
1887	}
1888	if (IS_ERR(dent)) {
1889		gfs2_consist_inode(dip);
1890		return PTR_ERR(dent);
1891	}
1892	/* If not first in block, adjust pointers accordingly */
1893	if (gfs2_dirent_find(dent, name, NULL) == 0) {
1894		prev = dent;
1895		dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1896	}
1897
1898	dirent_del(dip, bh, prev, dent);
1899	tv = inode_set_ctime_current(&dip->i_inode);
1900	if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1901		struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1902		u16 entries = be16_to_cpu(leaf->lf_entries);
1903		if (!entries)
1904			gfs2_consist_inode(dip);
1905		leaf->lf_entries = cpu_to_be16(--entries);
1906		leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1907		leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1908	}
1909	brelse(bh);
1910
1911	if (!dip->i_entries)
1912		gfs2_consist_inode(dip);
1913	dip->i_entries--;
1914	inode_set_mtime_to_ts(&dip->i_inode, tv);
1915	if (d_is_dir(dentry))
1916		drop_nlink(&dip->i_inode);
1917	mark_inode_dirty(&dip->i_inode);
1918
1919	return 0;
1920}
1921
1922/**
1923 * gfs2_dir_mvino - Change inode number of directory entry
1924 * @dip: The GFS2 directory inode
1925 * @filename: the filename to be moved
1926 * @nip: the new GFS2 inode
1927 * @new_type: the de_type of the new dirent
1928 *
1929 * This routine changes the inode number of a directory entry.  It's used
1930 * by rename to change ".." when a directory is moved.
1931 * Assumes a glock is held on dvp.
1932 *
1933 * Returns: errno
1934 */
1935
1936int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1937		   const struct gfs2_inode *nip, unsigned int new_type)
1938{
1939	struct buffer_head *bh;
1940	struct gfs2_dirent *dent;
1941
1942	dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1943	if (!dent) {
1944		gfs2_consist_inode(dip);
1945		return -EIO;
1946	}
1947	if (IS_ERR(dent))
1948		return PTR_ERR(dent);
1949
1950	gfs2_trans_add_meta(dip->i_gl, bh);
1951	gfs2_inum_out(nip, dent);
1952	dent->de_type = cpu_to_be16(new_type);
1953	brelse(bh);
1954
1955	inode_set_mtime_to_ts(&dip->i_inode, inode_set_ctime_current(&dip->i_inode));
1956	mark_inode_dirty_sync(&dip->i_inode);
1957	return 0;
1958}
1959
1960/**
1961 * leaf_dealloc - Deallocate a directory leaf
1962 * @dip: the directory
1963 * @index: the hash table offset in the directory
1964 * @len: the number of pointers to this leaf
1965 * @leaf_no: the leaf number
1966 * @leaf_bh: buffer_head for the starting leaf
1967 * @last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1968 *
1969 * Returns: errno
1970 */
1971
1972static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1973			u64 leaf_no, struct buffer_head *leaf_bh,
1974			int last_dealloc)
1975{
1976	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1977	struct gfs2_leaf *tmp_leaf;
1978	struct gfs2_rgrp_list rlist;
1979	struct buffer_head *bh, *dibh;
1980	u64 blk, nblk;
1981	unsigned int rg_blocks = 0, l_blocks = 0;
1982	char *ht;
1983	unsigned int x, size = len * sizeof(u64);
1984	int error;
1985
1986	error = gfs2_rindex_update(sdp);
1987	if (error)
1988		return error;
1989
1990	memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1991
1992	ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
1993	if (ht == NULL)
1994		ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO);
1995	if (!ht)
1996		return -ENOMEM;
1997
1998	error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1999	if (error)
2000		goto out;
2001
2002	/*  Count the number of leaves  */
2003	bh = leaf_bh;
2004
2005	for (blk = leaf_no; blk; blk = nblk) {
2006		if (blk != leaf_no) {
2007			error = get_leaf(dip, blk, &bh);
2008			if (error)
2009				goto out_rlist;
2010		}
2011		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2012		nblk = be64_to_cpu(tmp_leaf->lf_next);
2013		if (blk != leaf_no)
2014			brelse(bh);
2015
2016		gfs2_rlist_add(dip, &rlist, blk);
2017		l_blocks++;
2018	}
2019
2020	gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, LM_FLAG_NODE_SCOPE);
2021
2022	for (x = 0; x < rlist.rl_rgrps; x++) {
2023		struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl);
2024
2025		rg_blocks += rgd->rd_length;
2026	}
2027
2028	error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
2029	if (error)
2030		goto out_rlist;
2031
2032	error = gfs2_trans_begin(sdp,
2033			rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
2034			RES_DINODE + RES_STATFS + RES_QUOTA, RES_DINODE +
2035				 l_blocks);
2036	if (error)
2037		goto out_rg_gunlock;
2038
2039	bh = leaf_bh;
2040
2041	for (blk = leaf_no; blk; blk = nblk) {
2042		struct gfs2_rgrpd *rgd;
2043
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		rgd = gfs2_blk2rgrpd(sdp, blk, true);
2055		gfs2_free_meta(dip, rgd, blk, 1);
2056		gfs2_add_inode_blocks(&dip->i_inode, -1);
2057	}
2058
2059	error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
2060	if (error != size) {
2061		if (error >= 0)
2062			error = -EIO;
2063		goto out_end_trans;
2064	}
2065
2066	error = gfs2_meta_inode_buffer(dip, &dibh);
2067	if (error)
2068		goto out_end_trans;
2069
2070	gfs2_trans_add_meta(dip->i_gl, dibh);
2071	/* On the last dealloc, make this a regular file in case we crash.
2072	   (We don't want to free these blocks a second time.)  */
2073	if (last_dealloc)
2074		dip->i_inode.i_mode = S_IFREG;
2075	gfs2_dinode_out(dip, dibh->b_data);
2076	brelse(dibh);
2077
2078out_end_trans:
2079	gfs2_trans_end(sdp);
2080out_rg_gunlock:
2081	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2082out_rlist:
2083	gfs2_rlist_free(&rlist);
2084	gfs2_quota_unhold(dip);
2085out:
2086	kvfree(ht);
2087	return error;
2088}
2089
2090/**
2091 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2092 * @dip: the directory
2093 *
2094 * Dealloc all on-disk directory leaves to FREEMETA state
2095 * Change on-disk inode type to "regular file"
2096 *
2097 * Returns: errno
2098 */
2099
2100int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
2101{
2102	struct buffer_head *bh;
2103	struct gfs2_leaf *leaf;
2104	u32 hsize, len;
2105	u32 index = 0, next_index;
2106	__be64 *lp;
2107	u64 leaf_no;
2108	int error = 0, last;
2109
2110	hsize = BIT(dip->i_depth);
2111
2112	lp = gfs2_dir_get_hash_table(dip);
2113	if (IS_ERR(lp))
2114		return PTR_ERR(lp);
2115
2116	while (index < hsize) {
2117		leaf_no = be64_to_cpu(lp[index]);
2118		if (leaf_no) {
2119			error = get_leaf(dip, leaf_no, &bh);
2120			if (error)
2121				goto out;
2122			leaf = (struct gfs2_leaf *)bh->b_data;
2123			len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth));
2124
2125			next_index = (index & ~(len - 1)) + len;
2126			last = ((next_index >= hsize) ? 1 : 0);
2127			error = leaf_dealloc(dip, index, len, leaf_no, bh,
2128					     last);
2129			brelse(bh);
2130			if (error)
2131				goto out;
2132			index = next_index;
2133		} else
2134			index++;
2135	}
2136
2137	if (index != hsize) {
2138		gfs2_consist_inode(dip);
2139		error = -EIO;
2140	}
2141
2142out:
2143
2144	return error;
2145}
2146
2147/**
2148 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2149 * @inode: the directory inode being written to
2150 * @name: the filename that's going to be added
2151 * @da: The structure to return dir alloc info
2152 *
2153 * Returns: 0 if ok, -ve on error
2154 */
2155
2156int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
2157			       struct gfs2_diradd *da)
2158{
2159	struct gfs2_inode *ip = GFS2_I(inode);
2160	struct gfs2_sbd *sdp = GFS2_SB(inode);
2161	const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2162	struct gfs2_dirent *dent;
2163	struct buffer_head *bh;
2164
2165	da->nr_blocks = 0;
2166	da->bh = NULL;
2167	da->dent = NULL;
2168
2169	dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2170	if (!dent) {
2171		da->nr_blocks = sdp->sd_max_dirres;
2172		if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
2173		    (GFS2_DIRENT_SIZE(name->len) < extra))
2174			da->nr_blocks = 1;
2175		return 0;
2176	}
2177	if (IS_ERR(dent))
2178		return PTR_ERR(dent);
2179
2180	if (da->save_loc) {
2181		da->bh = bh;
2182		da->dent = dent;
2183	} else {
2184		brelse(bh);
2185	}
2186	return 0;
2187}
2188