dir.c

   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#include <linux/log2.h>
  64
  65#include "gfs2.h"
  66#include "incore.h"
  67#include "dir.h"
  68#include "glock.h"
  69#include "inode.h"
  70#include "meta_io.h"
  71#include "quota.h"
  72#include "rgrp.h"
  73#include "trans.h"
  74#include "bmap.h"
  75#include "util.h"
  76
  77#define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
  78
  79#define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
  80#define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
  81#define GFS2_HASH_INDEX_MASK 0xffffc000
  82#define GFS2_USE_HASH_FLAG 0x2000
  83
  84struct qstr gfs2_qdot __read_mostly;
  85struct qstr gfs2_qdotdot __read_mostly;
  86
  87typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
  88			    const struct qstr *name, void *opaque);
  89
  90int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
  91			    struct buffer_head **bhp)
  92{
  93	struct buffer_head *bh;
  94
  95	bh = gfs2_meta_new(ip->i_gl, block);
  96	gfs2_trans_add_meta(ip->i_gl, bh);
  97	gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
  98	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
  99	*bhp = bh;
 100	return 0;
 101}
 102
 103static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
 104					struct buffer_head **bhp)
 105{
 106	struct buffer_head *bh;
 107	int error;
 108
 109	error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh);
 110	if (error)
 111		return error;
 112	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
 113		brelse(bh);
 114		return -EIO;
 115	}
 116	*bhp = bh;
 117	return 0;
 118}
 119
 120static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
 121				  unsigned int offset, unsigned int size)
 122{
 123	struct buffer_head *dibh;
 124	int error;
 125
 126	error = gfs2_meta_inode_buffer(ip, &dibh);
 127	if (error)
 128		return error;
 129
 130	gfs2_trans_add_meta(ip->i_gl, dibh);
 131	memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
 132	if (ip->i_inode.i_size < offset + size)
 133		i_size_write(&ip->i_inode, offset + size);
 134	inode_set_mtime_to_ts(&ip->i_inode, inode_set_ctime_current(&ip->i_inode));
 135	gfs2_dinode_out(ip, dibh->b_data);
 136
 137	brelse(dibh);
 138
 139	return size;
 140}
 141
 142
 143
 144/**
 145 * gfs2_dir_write_data - Write directory information to the inode
 146 * @ip: The GFS2 inode
 147 * @buf: The buffer containing information to be written
 148 * @offset: The file offset to start writing at
 149 * @size: The amount of data to write
 150 *
 151 * Returns: The number of bytes correctly written or error code
 152 */
 153static 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	u64 lblock, dblock;
 159	u32 extlen = 0;
 160	unsigned int o;
 161	int copied = 0;
 162	int error = 0;
 163	bool new = false;
 164
 165	if (!size)
 166		return 0;
 167
 168	if (gfs2_is_stuffed(ip) && offset + size <= gfs2_max_stuffed_size(ip))
 169		return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
 170					      size);
 171
 172	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 173		return -EINVAL;
 174
 175	if (gfs2_is_stuffed(ip)) {
 176		error = gfs2_unstuff_dinode(ip);
 177		if (error)
 178			return error;
 179	}
 180
 181	lblock = offset;
 182	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 183
 184	while (copied < size) {
 185		unsigned int amount;
 186		struct buffer_head *bh;
 187
 188		amount = size - copied;
 189		if (amount > sdp->sd_sb.sb_bsize - o)
 190			amount = sdp->sd_sb.sb_bsize - o;
 191
 192		if (!extlen) {
 193			extlen = 1;
 194			error = gfs2_alloc_extent(&ip->i_inode, lblock, &dblock,
 195						  &extlen, &new);
 196			if (error)
 197				goto fail;
 198			error = -EIO;
 199			if (gfs2_assert_withdraw(sdp, dblock))
 200				goto fail;
 201		}
 202
 203		if (amount == sdp->sd_jbsize || new)
 204			error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
 205		else
 206			error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
 207
 208		if (error)
 209			goto fail;
 210
 211		gfs2_trans_add_meta(ip->i_gl, bh);
 212		memcpy(bh->b_data + o, buf, amount);
 213		brelse(bh);
 214
 215		buf += amount;
 216		copied += amount;
 217		lblock++;
 218		dblock++;
 219		extlen--;
 220
 221		o = sizeof(struct gfs2_meta_header);
 222	}
 223
 224out:
 225	error = gfs2_meta_inode_buffer(ip, &dibh);
 226	if (error)
 227		return error;
 228
 229	if (ip->i_inode.i_size < offset + copied)
 230		i_size_write(&ip->i_inode, offset + copied);
 231	inode_set_mtime_to_ts(&ip->i_inode, inode_set_ctime_current(&ip->i_inode));
 232
 233	gfs2_trans_add_meta(ip->i_gl, dibh);
 234	gfs2_dinode_out(ip, dibh->b_data);
 235	brelse(dibh);
 236
 237	return copied;
 238fail:
 239	if (copied)
 240		goto out;
 241	return error;
 242}
 243
 244static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
 245				 unsigned int size)
 246{
 247	struct buffer_head *dibh;
 248	int error;
 249
 250	error = gfs2_meta_inode_buffer(ip, &dibh);
 251	if (!error) {
 252		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
 253		brelse(dibh);
 254	}
 255
 256	return (error) ? error : size;
 257}
 258
 259
 260/**
 261 * gfs2_dir_read_data - Read a data from a directory inode
 262 * @ip: The GFS2 Inode
 263 * @buf: The buffer to place result into
 264 * @size: Amount of data to transfer
 265 *
 266 * Returns: The amount of data actually copied or the error
 267 */
 268static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
 269			      unsigned int size)
 270{
 271	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 272	u64 lblock, dblock;
 273	u32 extlen = 0;
 274	unsigned int o;
 275	int copied = 0;
 276	int error = 0;
 277
 278	if (gfs2_is_stuffed(ip))
 279		return gfs2_dir_read_stuffed(ip, buf, size);
 280
 281	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 282		return -EINVAL;
 283
 284	lblock = 0;
 285	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 286
 287	while (copied < size) {
 288		unsigned int amount;
 289		struct buffer_head *bh;
 290
 291		amount = size - copied;
 292		if (amount > sdp->sd_sb.sb_bsize - o)
 293			amount = sdp->sd_sb.sb_bsize - o;
 294
 295		if (!extlen) {
 296			extlen = 32;
 297			error = gfs2_get_extent(&ip->i_inode, lblock,
 298						&dblock, &extlen);
 299			if (error || !dblock)
 300				goto fail;
 301			BUG_ON(extlen < 1);
 302			bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
 303		} else {
 304			error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh);
 305			if (error)
 306				goto fail;
 307		}
 308		error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
 309		if (error) {
 310			brelse(bh);
 311			goto fail;
 312		}
 313		dblock++;
 314		extlen--;
 315		memcpy(buf, bh->b_data + o, amount);
 316		brelse(bh);
 317		buf += (amount/sizeof(__be64));
 318		copied += amount;
 319		lblock++;
 320		o = sizeof(struct gfs2_meta_header);
 321	}
 322
 323	return copied;
 324fail:
 325	return (copied) ? copied : error;
 326}
 327
 328/**
 329 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
 330 * @ip: The inode in question
 331 *
 332 * Returns: The hash table or an error
 333 */
 334
 335static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
 336{
 337	struct inode *inode = &ip->i_inode;
 338	int ret;
 339	u32 hsize;
 340	__be64 *hc;
 341
 342	BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
 343
 344	hc = ip->i_hash_cache;
 345	if (hc)
 346		return hc;
 347
 348	hsize = BIT(ip->i_depth);
 349	hsize *= sizeof(__be64);
 350	if (hsize != i_size_read(&ip->i_inode)) {
 351		gfs2_consist_inode(ip);
 352		return ERR_PTR(-EIO);
 353	}
 354
 355	hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
 356	if (hc == NULL)
 357		hc = __vmalloc(hsize, GFP_NOFS);
 358
 359	if (hc == NULL)
 360		return ERR_PTR(-ENOMEM);
 361
 362	ret = gfs2_dir_read_data(ip, hc, hsize);
 363	if (ret < 0) {
 364		kvfree(hc);
 365		return ERR_PTR(ret);
 366	}
 367
 368	spin_lock(&inode->i_lock);
 369	if (likely(!ip->i_hash_cache)) {
 370		ip->i_hash_cache = hc;
 371		hc = NULL;
 372	}
 373	spin_unlock(&inode->i_lock);
 374	kvfree(hc);
 375
 376	return ip->i_hash_cache;
 377}
 378
 379/**
 380 * gfs2_dir_hash_inval - Invalidate dir hash
 381 * @ip: The directory inode
 382 *
 383 * Must be called with an exclusive glock, or during glock invalidation.
 384 */
 385void gfs2_dir_hash_inval(struct gfs2_inode *ip)
 386{
 387	__be64 *hc;
 388
 389	spin_lock(&ip->i_inode.i_lock);
 390	hc = ip->i_hash_cache;
 391	ip->i_hash_cache = NULL;
 392	spin_unlock(&ip->i_inode.i_lock);
 393
 394	kvfree(hc);
 395}
 396
 397static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
 398{
 399	return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
 400}
 401
 402static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
 403				     const struct qstr *name, int ret)
 404{
 405	if (!gfs2_dirent_sentinel(dent) &&
 406	    be32_to_cpu(dent->de_hash) == name->hash &&
 407	    be16_to_cpu(dent->de_name_len) == name->len &&
 408	    memcmp(dent+1, name->name, name->len) == 0)
 409		return ret;
 410	return 0;
 411}
 412
 413static int gfs2_dirent_find(const struct gfs2_dirent *dent,
 414			    const struct qstr *name,
 415			    void *opaque)
 416{
 417	return __gfs2_dirent_find(dent, name, 1);
 418}
 419
 420static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
 421			    const struct qstr *name,
 422			    void *opaque)
 423{
 424	return __gfs2_dirent_find(dent, name, 2);
 425}
 426
 427/*
 428 * name->name holds ptr to start of block.
 429 * name->len holds size of block.
 430 */
 431static int gfs2_dirent_last(const struct gfs2_dirent *dent,
 432			    const struct qstr *name,
 433			    void *opaque)
 434{
 435	const char *start = name->name;
 436	const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
 437	if (name->len == (end - start))
 438		return 1;
 439	return 0;
 440}
 441
 442/* Look for the dirent that contains the offset specified in data. Once we
 443 * find that dirent, there must be space available there for the new dirent */
 444static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
 445				  const struct qstr *name,
 446				  void *ptr)
 447{
 448	unsigned required = GFS2_DIRENT_SIZE(name->len);
 449	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 450	unsigned totlen = be16_to_cpu(dent->de_rec_len);
 451
 452	if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
 453		return 0;
 454	if (gfs2_dirent_sentinel(dent))
 455		actual = 0;
 456	if (ptr < (void *)dent + actual)
 457		return -1;
 458	if ((void *)dent + totlen >= ptr + required)
 459		return 1;
 460	return -1;
 461}
 462
 463static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
 464				  const struct qstr *name,
 465				  void *opaque)
 466{
 467	unsigned required = GFS2_DIRENT_SIZE(name->len);
 468	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 469	unsigned totlen = be16_to_cpu(dent->de_rec_len);
 470
 471	if (gfs2_dirent_sentinel(dent))
 472		actual = 0;
 473	if (totlen - actual >= required)
 474		return 1;
 475	return 0;
 476}
 477
 478struct dirent_gather {
 479	const struct gfs2_dirent **pdent;
 480	unsigned offset;
 481};
 482
 483static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
 484			      const struct qstr *name,
 485			      void *opaque)
 486{
 487	struct dirent_gather *g = opaque;
 488	if (!gfs2_dirent_sentinel(dent)) {
 489		g->pdent[g->offset++] = dent;
 490	}
 491	return 0;
 492}
 493
 494/*
 495 * Other possible things to check:
 496 * - Inode located within filesystem size (and on valid block)
 497 * - Valid directory entry type
 498 * Not sure how heavy-weight we want to make this... could also check
 499 * hash is correct for example, but that would take a lot of extra time.
 500 * For now the most important thing is to check that the various sizes
 501 * are correct.
 502 */
 503static int gfs2_check_dirent(struct gfs2_sbd *sdp,
 504			     struct gfs2_dirent *dent, unsigned int offset,
 505			     unsigned int size, unsigned int len, int first)
 506{
 507	const char *msg = "gfs2_dirent too small";
 508	if (unlikely(size < sizeof(struct gfs2_dirent)))
 509		goto error;
 510	msg = "gfs2_dirent misaligned";
 511	if (unlikely(offset & 0x7))
 512		goto error;
 513	msg = "gfs2_dirent points beyond end of block";
 514	if (unlikely(offset + size > len))
 515		goto error;
 516	msg = "zero inode number";
 517	if (unlikely(!first && gfs2_dirent_sentinel(dent)))
 518		goto error;
 519	msg = "name length is greater than space in dirent";
 520	if (!gfs2_dirent_sentinel(dent) &&
 521	    unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
 522		     size))
 523		goto error;
 524	return 0;
 525error:
 526	fs_warn(sdp, "%s: %s (%s)\n",
 527		__func__, msg, first ? "first in block" : "not first in block");
 528	return -EIO;
 529}
 530
 531static int gfs2_dirent_offset(struct gfs2_sbd *sdp, const void *buf)
 532{
 533	const struct gfs2_meta_header *h = buf;
 534	int offset;
 535
 536	BUG_ON(buf == NULL);
 537
 538	switch(be32_to_cpu(h->mh_type)) {
 539	case GFS2_METATYPE_LF:
 540		offset = sizeof(struct gfs2_leaf);
 541		break;
 542	case GFS2_METATYPE_DI:
 543		offset = sizeof(struct gfs2_dinode);
 544		break;
 545	default:
 546		goto wrong_type;
 547	}
 548	return offset;
 549wrong_type:
 550	fs_warn(sdp, "%s: wrong block type %u\n", __func__,
 551		be32_to_cpu(h->mh_type));
 552	return -1;
 553}
 554
 555static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
 556					    unsigned int len, gfs2_dscan_t scan,
 557					    const struct qstr *name,
 558					    void *opaque)
 559{
 560	struct gfs2_dirent *dent, *prev;
 561	unsigned offset;
 562	unsigned size;
 563	int ret = 0;
 564
 565	ret = gfs2_dirent_offset(GFS2_SB(inode), buf);
 566	if (ret < 0) {
 567		gfs2_consist_inode(GFS2_I(inode));
 568		return ERR_PTR(-EIO);
 569	}
 570	offset = ret;
 571	prev = NULL;
 572	dent = buf + offset;
 573	size = be16_to_cpu(dent->de_rec_len);
 574	if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size, len, 1)) {
 575		gfs2_consist_inode(GFS2_I(inode));
 576		return ERR_PTR(-EIO);
 577	}
 578	do {
 579		ret = scan(dent, name, opaque);
 580		if (ret)
 581			break;
 582		offset += size;
 583		if (offset == len)
 584			break;
 585		prev = dent;
 586		dent = buf + offset;
 587		size = be16_to_cpu(dent->de_rec_len);
 588		if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size,
 589				      len, 0)) {
 590			gfs2_consist_inode(GFS2_I(inode));
 591			return ERR_PTR(-EIO);
 592		}
 593	} while(1);
 594
 595	switch(ret) {
 596	case 0:
 597		return NULL;
 598	case 1:
 599		return dent;
 600	case 2:
 601		return prev ? prev : dent;
 602	default:
 603		BUG_ON(ret > 0);
 604		return ERR_PTR(ret);
 605	}
 606}
 607
 608static int dirent_check_reclen(struct gfs2_inode *dip,
 609			       const struct gfs2_dirent *d, const void *end_p)
 610{
 611	const void *ptr = d;
 612	u16 rec_len = be16_to_cpu(d->de_rec_len);
 613
 614	if (unlikely(rec_len < sizeof(struct gfs2_dirent))) {
 615		gfs2_consist_inode(dip);
 616		return -EIO;
 617	}
 618	ptr += rec_len;
 619	if (ptr < end_p)
 620		return rec_len;
 621	if (ptr == end_p)
 622		return -ENOENT;
 623
 624	gfs2_consist_inode(dip);
 625	return -EIO;
 626}
 627
 628/**
 629 * dirent_next - Next dirent
 630 * @dip: the directory
 631 * @bh: The buffer
 632 * @dent: Pointer to list of dirents
 633 *
 634 * Returns: 0 on success, error code otherwise
 635 */
 636
 637static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
 638		       struct gfs2_dirent **dent)
 639{
 640	struct gfs2_dirent *cur = *dent, *tmp;
 641	char *bh_end = bh->b_data + bh->b_size;
 642	int ret;
 643
 644	ret = dirent_check_reclen(dip, cur, bh_end);
 645	if (ret < 0)
 646		return ret;
 647
 648	tmp = (void *)cur + ret;
 649	ret = dirent_check_reclen(dip, tmp, bh_end);
 650	if (ret == -EIO)
 651		return ret;
 652
 653        /* Only the first dent could ever have de_inum.no_addr == 0 */
 654	if (gfs2_dirent_sentinel(tmp)) {
 655		gfs2_consist_inode(dip);
 656		return -EIO;
 657	}
 658
 659	*dent = tmp;
 660	return 0;
 661}
 662
 663/**
 664 * dirent_del - Delete a dirent
 665 * @dip: The GFS2 inode
 666 * @bh: The buffer
 667 * @prev: The previous dirent
 668 * @cur: The current dirent
 669 *
 670 */
 671
 672static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
 673		       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
 674{
 675	u16 cur_rec_len, prev_rec_len;
 676
 677	if (gfs2_dirent_sentinel(cur)) {
 678		gfs2_consist_inode(dip);
 679		return;
 680	}
 681
 682	gfs2_trans_add_meta(dip->i_gl, bh);
 683
 684	/* If there is no prev entry, this is the first entry in the block.
 685	   The de_rec_len is already as big as it needs to be.  Just zero
 686	   out the inode number and return.  */
 687
 688	if (!prev) {
 689		cur->de_inum.no_addr = 0;
 690		cur->de_inum.no_formal_ino = 0;
 691		return;
 692	}
 693
 694	/*  Combine this dentry with the previous one.  */
 695
 696	prev_rec_len = be16_to_cpu(prev->de_rec_len);
 697	cur_rec_len = be16_to_cpu(cur->de_rec_len);
 698
 699	if ((char *)prev + prev_rec_len != (char *)cur)
 700		gfs2_consist_inode(dip);
 701	if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
 702		gfs2_consist_inode(dip);
 703
 704	prev_rec_len += cur_rec_len;
 705	prev->de_rec_len = cpu_to_be16(prev_rec_len);
 706}
 707
 708
 709static struct gfs2_dirent *do_init_dirent(struct inode *inode,
 710					  struct gfs2_dirent *dent,
 711					  const struct qstr *name,
 712					  struct buffer_head *bh,
 713					  unsigned offset)
 714{
 715	struct gfs2_inode *ip = GFS2_I(inode);
 716	struct gfs2_dirent *ndent;
 717	unsigned totlen;
 718
 719	totlen = be16_to_cpu(dent->de_rec_len);
 720	BUG_ON(offset + name->len > totlen);
 721	gfs2_trans_add_meta(ip->i_gl, bh);
 722	ndent = (struct gfs2_dirent *)((char *)dent + offset);
 723	dent->de_rec_len = cpu_to_be16(offset);
 724	gfs2_qstr2dirent(name, totlen - offset, ndent);
 725	return ndent;
 726}
 727
 728
 729/*
 730 * Takes a dent from which to grab space as an argument. Returns the
 731 * newly created dent.
 732 */
 733static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
 734					    struct gfs2_dirent *dent,
 735					    const struct qstr *name,
 736					    struct buffer_head *bh)
 737{
 738	unsigned offset = 0;
 739
 740	if (!gfs2_dirent_sentinel(dent))
 741		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 742	return do_init_dirent(inode, dent, name, bh, offset);
 743}
 744
 745static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
 746						   struct buffer_head *bh,
 747						   const struct qstr *name,
 748						   void *ptr)
 749{
 750	struct gfs2_dirent *dent;
 751	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 752				gfs2_dirent_find_offset, name, ptr);
 753	if (IS_ERR_OR_NULL(dent))
 754		return dent;
 755	return do_init_dirent(inode, dent, name, bh,
 756			      (unsigned)(ptr - (void *)dent));
 757}
 758
 759static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
 760		    struct buffer_head **bhp)
 761{
 762	int error;
 763
 764	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp);
 765	if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
 766		/* pr_info("block num=%llu\n", leaf_no); */
 767		error = -EIO;
 768	}
 769
 770	return error;
 771}
 772
 773/**
 774 * get_leaf_nr - Get a leaf number associated with the index
 775 * @dip: The GFS2 inode
 776 * @index: hash table index of the targeted leaf
 777 * @leaf_out: Resulting leaf block number
 778 *
 779 * Returns: 0 on success, error code otherwise
 780 */
 781
 782static int get_leaf_nr(struct gfs2_inode *dip, u32 index, u64 *leaf_out)
 783{
 784	__be64 *hash;
 785	int error;
 786
 787	hash = gfs2_dir_get_hash_table(dip);
 788	error = PTR_ERR_OR_ZERO(hash);
 789
 790	if (!error)
 791		*leaf_out = be64_to_cpu(*(hash + index));
 792
 793	return error;
 794}
 795
 796static int get_first_leaf(struct gfs2_inode *dip, u32 index,
 797			  struct buffer_head **bh_out)
 798{
 799	u64 leaf_no;
 800	int error;
 801
 802	error = get_leaf_nr(dip, index, &leaf_no);
 803	if (!error)
 804		error = get_leaf(dip, leaf_no, bh_out);
 805
 806	return error;
 807}
 808
 809static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
 810					      const struct qstr *name,
 811					      gfs2_dscan_t scan,
 812					      struct buffer_head **pbh)
 813{
 814	struct buffer_head *bh;
 815	struct gfs2_dirent *dent;
 816	struct gfs2_inode *ip = GFS2_I(inode);
 817	int error;
 818
 819	if (ip->i_diskflags & GFS2_DIF_EXHASH) {
 820		struct gfs2_leaf *leaf;
 821		unsigned int hsize = BIT(ip->i_depth);
 822		unsigned int index;
 823		u64 ln;
 824		if (hsize * sizeof(u64) != i_size_read(inode)) {
 825			gfs2_consist_inode(ip);
 826			return ERR_PTR(-EIO);
 827		}
 828
 829		index = name->hash >> (32 - ip->i_depth);
 830		error = get_first_leaf(ip, index, &bh);
 831		if (error)
 832			return ERR_PTR(error);
 833		do {
 834			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 835						scan, name, NULL);
 836			if (dent)
 837				goto got_dent;
 838			leaf = (struct gfs2_leaf *)bh->b_data;
 839			ln = be64_to_cpu(leaf->lf_next);
 840			brelse(bh);
 841			if (!ln)
 842				break;
 843
 844			error = get_leaf(ip, ln, &bh);
 845		} while(!error);
 846
 847		return error ? ERR_PTR(error) : NULL;
 848	}
 849
 850
 851	error = gfs2_meta_inode_buffer(ip, &bh);
 852	if (error)
 853		return ERR_PTR(error);
 854	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
 855got_dent:
 856	if (IS_ERR_OR_NULL(dent)) {
 857		brelse(bh);
 858		bh = NULL;
 859	}
 860	*pbh = bh;
 861	return dent;
 862}
 863
 864static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
 865{
 866	struct gfs2_inode *ip = GFS2_I(inode);
 867	unsigned int n = 1;
 868	u64 bn;
 869	int error;
 870	struct buffer_head *bh;
 871	struct gfs2_leaf *leaf;
 872	struct gfs2_dirent *dent;
 873	struct timespec64 tv = current_time(inode);
 874
 875	error = gfs2_alloc_blocks(ip, &bn, &n, 0);
 876	if (error)
 877		return NULL;
 878	bh = gfs2_meta_new(ip->i_gl, bn);
 879	if (!bh)
 880		return NULL;
 881
 882	gfs2_trans_remove_revoke(GFS2_SB(inode), bn, 1);
 883	gfs2_trans_add_meta(ip->i_gl, bh);
 884	gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
 885	leaf = (struct gfs2_leaf *)bh->b_data;
 886	leaf->lf_depth = cpu_to_be16(depth);
 887	leaf->lf_entries = 0;
 888	leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
 889	leaf->lf_next = 0;
 890	leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
 891	leaf->lf_dist = cpu_to_be32(1);
 892	leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
 893	leaf->lf_sec = cpu_to_be64(tv.tv_sec);
 894	memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
 895	dent = (struct gfs2_dirent *)(leaf+1);
 896	gfs2_qstr2dirent(&empty_name, bh->b_size - sizeof(struct gfs2_leaf), dent);
 897	*pbh = bh;
 898	return leaf;
 899}
 900
 901/**
 902 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
 903 * @inode: The directory inode to be converted to exhash
 904 *
 905 * Returns: 0 on success, error code otherwise
 906 */
 907
 908static int dir_make_exhash(struct inode *inode)
 909{
 910	struct gfs2_inode *dip = GFS2_I(inode);
 911	struct gfs2_sbd *sdp = GFS2_SB(inode);
 912	struct gfs2_dirent *dent;
 913	struct qstr args;
 914	struct buffer_head *bh, *dibh;
 915	struct gfs2_leaf *leaf;
 916	u32 x;
 917	__be64 *lp;
 918	u64 bn;
 919	int error;
 920
 921	error = gfs2_meta_inode_buffer(dip, &dibh);
 922	if (error)
 923		return error;
 924
 925	/*  Turn over a new leaf  */
 926
 927	leaf = new_leaf(inode, &bh, 0);
 928	if (!leaf)
 929		return -ENOSPC;
 930	bn = bh->b_blocknr;
 931
 932	gfs2_assert(sdp, dip->i_entries < BIT(16));
 933	leaf->lf_entries = cpu_to_be16(dip->i_entries);
 934
 935	/*  Copy dirents  */
 936
 937	gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
 938			     sizeof(struct gfs2_dinode));
 939
 940	/*  Find last entry  */
 941
 942	x = 0;
 943	args.len = bh->b_size - sizeof(struct gfs2_dinode) +
 944		   sizeof(struct gfs2_leaf);
 945	args.name = bh->b_data;
 946	dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
 947				gfs2_dirent_last, &args, NULL);
 948	if (!dent) {
 949		brelse(bh);
 950		brelse(dibh);
 951		return -EIO;
 952	}
 953	if (IS_ERR(dent)) {
 954		brelse(bh);
 955		brelse(dibh);
 956		return PTR_ERR(dent);
 957	}
 958
 959	/*  Adjust the last dirent's record length
 960	   (Remember that dent still points to the last entry.)  */
 961
 962	dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
 963		sizeof(struct gfs2_dinode) -
 964		sizeof(struct gfs2_leaf));
 965
 966	brelse(bh);
 967
 968	/*  We're done with the new leaf block, now setup the new
 969	    hash table.  */
 970
 971	gfs2_trans_add_meta(dip->i_gl, dibh);
 972	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 973
 974	lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
 975
 976	for (x = sdp->sd_hash_ptrs; x--; lp++)
 977		*lp = cpu_to_be64(bn);
 978
 979	i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
 980	gfs2_add_inode_blocks(&dip->i_inode, 1);
 981	dip->i_diskflags |= GFS2_DIF_EXHASH;
 982	dip->i_depth = ilog2(sdp->sd_hash_ptrs);
 983
 984	gfs2_dinode_out(dip, dibh->b_data);
 985
 986	brelse(dibh);
 987
 988	return 0;
 989}
 990
 991/**
 992 * dir_split_leaf - Split a leaf block into two
 993 * @inode: The directory inode to be split
 994 * @name: name of the dirent we're trying to insert
 995 *
 996 * Returns: 0 on success, error code on failure
 997 */
 998
 999static int dir_split_leaf(struct inode *inode, const struct qstr *name)
1000{
1001	struct gfs2_inode *dip = GFS2_I(inode);
1002	struct buffer_head *nbh, *obh, *dibh;
1003	struct gfs2_leaf *nleaf, *oleaf;
1004	struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
1005	u32 start, len, half_len, divider;
1006	u64 bn, leaf_no;
1007	__be64 *lp;
1008	u32 index;
1009	int x;
1010	int error;
1011
1012	index = name->hash >> (32 - dip->i_depth);
1013	error = get_leaf_nr(dip, index, &leaf_no);
1014	if (error)
1015		return error;
1016
1017	/*  Get the old leaf block  */
1018	error = get_leaf(dip, leaf_no, &obh);
1019	if (error)
1020		return error;
1021
1022	oleaf = (struct gfs2_leaf *)obh->b_data;
1023	if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
1024		brelse(obh);
1025		return 1; /* can't split */
1026	}
1027
1028	gfs2_trans_add_meta(dip->i_gl, obh);
1029
1030	nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
1031	if (!nleaf) {
1032		brelse(obh);
1033		return -ENOSPC;
1034	}
1035	bn = nbh->b_blocknr;
1036
1037	/*  Compute the start and len of leaf pointers in the hash table.  */
1038	len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth));
1039	half_len = len >> 1;
1040	if (!half_len) {
1041		fs_warn(GFS2_SB(inode), "i_depth %u lf_depth %u index %u\n",
1042			dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
1043		gfs2_consist_inode(dip);
1044		error = -EIO;
1045		goto fail_brelse;
1046	}
1047
1048	start = (index & ~(len - 1));
1049
1050	/* Change the pointers.
1051	   Don't bother distinguishing stuffed from non-stuffed.
1052	   This code is complicated enough already. */
1053	lp = kmalloc_array(half_len, sizeof(__be64), GFP_NOFS);
1054	if (!lp) {
1055		error = -ENOMEM;
1056		goto fail_brelse;
1057	}
1058
1059	/*  Change the pointers  */
1060	for (x = 0; x < half_len; x++)
1061		lp[x] = cpu_to_be64(bn);
1062
1063	gfs2_dir_hash_inval(dip);
1064
1065	error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1066				    half_len * sizeof(u64));
1067	if (error != half_len * sizeof(u64)) {
1068		if (error >= 0)
1069			error = -EIO;
1070		goto fail_lpfree;
1071	}
1072
1073	kfree(lp);
1074
1075	/*  Compute the divider  */
1076	divider = (start + half_len) << (32 - dip->i_depth);
1077
1078	/*  Copy the entries  */
1079	dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1080
1081	do {
1082		next = dent;
1083		if (dirent_next(dip, obh, &next))
1084			next = NULL;
1085
1086		if (!gfs2_dirent_sentinel(dent) &&
1087		    be32_to_cpu(dent->de_hash) < divider) {
1088			struct qstr str;
1089			void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
1090			str.name = (char*)(dent+1);
1091			str.len = be16_to_cpu(dent->de_name_len);
1092			str.hash = be32_to_cpu(dent->de_hash);
1093			new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
1094			if (IS_ERR(new)) {
1095				error = PTR_ERR(new);
1096				break;
1097			}
1098
1099			new->de_inum = dent->de_inum; /* No endian worries */
1100			new->de_type = dent->de_type; /* No endian worries */
1101			be16_add_cpu(&nleaf->lf_entries, 1);
1102
1103			dirent_del(dip, obh, prev, dent);
1104
1105			if (!oleaf->lf_entries)
1106				gfs2_consist_inode(dip);
1107			be16_add_cpu(&oleaf->lf_entries, -1);
1108
1109			if (!prev)
1110				prev = dent;
1111		} else {
1112			prev = dent;
1113		}
1114		dent = next;
1115	} while (dent);
1116
1117	oleaf->lf_depth = nleaf->lf_depth;
1118
1119	error = gfs2_meta_inode_buffer(dip, &dibh);
1120	if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1121		gfs2_trans_add_meta(dip->i_gl, dibh);
1122		gfs2_add_inode_blocks(&dip->i_inode, 1);
1123		gfs2_dinode_out(dip, dibh->b_data);
1124		brelse(dibh);
1125	}
1126
1127	brelse(obh);
1128	brelse(nbh);
1129
1130	return error;
1131
1132fail_lpfree:
1133	kfree(lp);
1134
1135fail_brelse:
1136	brelse(obh);
1137	brelse(nbh);
1138	return error;
1139}
1140
1141/**
1142 * dir_double_exhash - Double size of ExHash table
1143 * @dip: The GFS2 dinode
1144 *
1145 * Returns: 0 on success, error code on failure
1146 */
1147
1148static int dir_double_exhash(struct gfs2_inode *dip)
1149{
1150	struct buffer_head *dibh;
1151	u32 hsize;
1152	u32 hsize_bytes;
1153	__be64 *hc;
1154	__be64 *hc2, *h;
1155	int x;
1156	int error = 0;
1157
1158	hsize = BIT(dip->i_depth);
1159	hsize_bytes = hsize * sizeof(__be64);
1160
1161	hc = gfs2_dir_get_hash_table(dip);
1162	if (IS_ERR(hc))
1163		return PTR_ERR(hc);
1164
1165	hc2 = kmalloc_array(hsize_bytes, 2, GFP_NOFS | __GFP_NOWARN);
1166	if (hc2 == NULL)
1167		hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS);
1168
1169	if (!hc2)
1170		return -ENOMEM;
1171
1172	h = hc2;
1173	error = gfs2_meta_inode_buffer(dip, &dibh);
1174	if (error)
1175		goto out_kfree;
1176
1177	for (x = 0; x < hsize; x++) {
1178		*h++ = *hc;
1179		*h++ = *hc;
1180		hc++;
1181	}
1182
1183	error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1184	if (error != (hsize_bytes * 2))
1185		goto fail;
1186
1187	gfs2_dir_hash_inval(dip);
1188	dip->i_hash_cache = hc2;
1189	dip->i_depth++;
1190	gfs2_dinode_out(dip, dibh->b_data);
1191	brelse(dibh);
1192	return 0;
1193
1194fail:
1195	/* Replace original hash table & size */
1196	gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1197	i_size_write(&dip->i_inode, hsize_bytes);
1198	gfs2_dinode_out(dip, dibh->b_data);
1199	brelse(dibh);
1200out_kfree:
1201	kvfree(hc2);
1202	return error;
1203}
1204
1205/**
1206 * compare_dents - compare directory entries by hash value
1207 * @a: first dent
1208 * @b: second dent
1209 *
1210 * When comparing the hash entries of @a to @b:
1211 *   gt: returns 1
1212 *   lt: returns -1
1213 *   eq: returns 0
1214 */
1215
1216static int compare_dents(const void *a, const void *b)
1217{
1218	const struct gfs2_dirent *dent_a, *dent_b;
1219	u32 hash_a, hash_b;
1220	int ret = 0;
1221
1222	dent_a = *(const struct gfs2_dirent **)a;
1223	hash_a = dent_a->de_cookie;
1224
1225	dent_b = *(const struct gfs2_dirent **)b;
1226	hash_b = dent_b->de_cookie;
1227
1228	if (hash_a > hash_b)
1229		ret = 1;
1230	else if (hash_a < hash_b)
1231		ret = -1;
1232	else {
1233		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1234		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1235
1236		if (len_a > len_b)
1237			ret = 1;
1238		else if (len_a < len_b)
1239			ret = -1;
1240		else
1241			ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1242	}
1243
1244	return ret;
1245}
1246
1247/**
1248 * do_filldir_main - read out directory entries
1249 * @dip: The GFS2 inode
1250 * @ctx: what to feed the entries to
1251 * @darr: an array of struct gfs2_dirent pointers to read
1252 * @entries: the number of entries in darr
1253 * @sort_start: index of the directory array to start our sort
1254 * @copied: pointer to int that's non-zero if a entry has been copied out
1255 *
1256 * Jump through some hoops to make sure that if there are hash collsions,
1257 * they are read out at the beginning of a buffer.  We want to minimize
1258 * the possibility that they will fall into different readdir buffers or
1259 * that someone will want to seek to that location.
1260 *
1261 * Returns: errno, >0 if the actor tells you to stop
1262 */
1263
1264static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
1265			   struct gfs2_dirent **darr, u32 entries,
1266			   u32 sort_start, int *copied)
1267{
1268	const struct gfs2_dirent *dent, *dent_next;
1269	u64 off, off_next;
1270	unsigned int x, y;
1271	int run = 0;
1272
1273	if (sort_start < entries)
1274		sort(&darr[sort_start], entries - sort_start,
1275		     sizeof(struct gfs2_dirent *), compare_dents, NULL);
1276
1277	dent_next = darr[0];
1278	off_next = dent_next->de_cookie;
1279
1280	for (x = 0, y = 1; x < entries; x++, y++) {
1281		dent = dent_next;
1282		off = off_next;
1283
1284		if (y < entries) {
1285			dent_next = darr[y];
1286			off_next = dent_next->de_cookie;
1287
1288			if (off < ctx->pos)
1289				continue;
1290			ctx->pos = off;
1291
1292			if (off_next == off) {
1293				if (*copied && !run)
1294					return 1;
1295				run = 1;
1296			} else
1297				run = 0;
1298		} else {
1299			if (off < ctx->pos)
1300				continue;
1301			ctx->pos = off;
1302		}
1303
1304		if (!dir_emit(ctx, (const char *)(dent + 1),
1305				be16_to_cpu(dent->de_name_len),
1306				be64_to_cpu(dent->de_inum.no_addr),
1307				be16_to_cpu(dent->de_type)))
1308			return 1;
1309
1310		*copied = 1;
1311	}
1312
1313	/* Increment the ctx->pos by one, so the next time we come into the
1314	   do_filldir fxn, we get the next entry instead of the last one in the
1315	   current leaf */
1316
1317	ctx->pos++;
1318
1319	return 0;
1320}
1321
1322static void *gfs2_alloc_sort_buffer(unsigned size)
1323{
1324	void *ptr = NULL;
1325
1326	if (size < KMALLOC_MAX_SIZE)
1327		ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1328	if (!ptr)
1329		ptr = __vmalloc(size, GFP_NOFS);
1330	return ptr;
1331}
1332
1333
1334static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
1335			    unsigned leaf_nr, struct gfs2_dirent **darr,
1336			    unsigned entries)
1337{
1338	int sort_id = -1;
1339	int i;
1340	
1341	for (i = 0; i < entries; i++) {
1342		unsigned offset;
1343
1344		darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
1345		darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
1346
1347		if (!sdp->sd_args.ar_loccookie)
1348			continue;
1349		offset = (char *)(darr[i]) -
1350			(bh->b_data + gfs2_dirent_offset(sdp, bh->b_data));
1351		offset /= GFS2_MIN_DIRENT_SIZE;
1352		offset += leaf_nr * sdp->sd_max_dents_per_leaf;
1353		if (offset >= GFS2_USE_HASH_FLAG ||
1354		    leaf_nr >= GFS2_USE_HASH_FLAG) {
1355			darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
1356			if (sort_id < 0)
1357				sort_id = i;
1358			continue;
1359		}
1360		darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
1361		darr[i]->de_cookie |= offset;
1362	}
1363	return sort_id;
1364}	
1365
1366
1367static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
1368			      int *copied, unsigned *depth,
1369			      u64 leaf_no)
1370{
1371	struct gfs2_inode *ip = GFS2_I(inode);
1372	struct gfs2_sbd *sdp = GFS2_SB(inode);
1373	struct buffer_head *bh;
1374	struct gfs2_leaf *lf;
1375	unsigned entries = 0, entries2 = 0;
1376	unsigned leaves = 0, leaf = 0, offset, sort_offset;
1377	struct gfs2_dirent **darr, *dent;
1378	struct dirent_gather g;
1379	struct buffer_head **larr;
1380	int error, i, need_sort = 0, sort_id;
1381	u64 lfn = leaf_no;
1382
1383	do {
1384		error = get_leaf(ip, lfn, &bh);
1385		if (error)
1386			goto out;
1387		lf = (struct gfs2_leaf *)bh->b_data;
1388		if (leaves == 0)
1389			*depth = be16_to_cpu(lf->lf_depth);
1390		entries += be16_to_cpu(lf->lf_entries);
1391		leaves++;
1392		lfn = be64_to_cpu(lf->lf_next);
1393		brelse(bh);
1394	} while(lfn);
1395
1396	if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
1397		need_sort = 1;
1398		sort_offset = 0;
1399	}
1400
1401	if (!entries)
1402		return 0;
1403
1404	error = -ENOMEM;
1405	/*
1406	 * The extra 99 entries are not normally used, but are a buffer
1407	 * zone in case the number of entries in the leaf is corrupt.
1408	 * 99 is the maximum number of entries that can fit in a single
1409	 * leaf block.
1410	 */
1411	larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1412	if (!larr)
1413		goto out;
1414	darr = (struct gfs2_dirent **)(larr + leaves);
1415	g.pdent = (const struct gfs2_dirent **)darr;
1416	g.offset = 0;
1417	lfn = leaf_no;
1418
1419	do {
1420		error = get_leaf(ip, lfn, &bh);
1421		if (error)
1422			goto out_free;
1423		lf = (struct gfs2_leaf *)bh->b_data;
1424		lfn = be64_to_cpu(lf->lf_next);
1425		if (lf->lf_entries) {
1426			offset = g.offset;
1427			entries2 += be16_to_cpu(lf->lf_entries);
1428			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1429						gfs2_dirent_gather, NULL, &g);
1430			error = PTR_ERR(dent);
1431			if (IS_ERR(dent))
1432				goto out_free;
1433			if (entries2 != g.offset) {
1434				fs_warn(sdp, "Number of entries corrupt in dir "
1435						"leaf %llu, entries2 (%u) != "
1436						"g.offset (%u)\n",
1437					(unsigned long long)bh->b_blocknr,
1438					entries2, g.offset);
1439				gfs2_consist_inode(ip);
1440				error = -EIO;
1441				goto out_free;
1442			}
1443			error = 0;
1444			sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
1445						   be16_to_cpu(lf->lf_entries));
1446			if (!need_sort && sort_id >= 0) {
1447				need_sort = 1;
1448				sort_offset = offset + sort_id;
1449			}
1450			larr[leaf++] = bh;
1451		} else {
1452			larr[leaf++] = NULL;
1453			brelse(bh);
1454		}
1455	} while(lfn);
1456
1457	BUG_ON(entries2 != entries);
1458	error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
1459				sort_offset : entries, copied);
1460out_free:
1461	for(i = 0; i < leaf; i++)
1462		brelse(larr[i]);
1463	kvfree(larr);
1464out:
1465	return error;
1466}
1467
1468/**
1469 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1470 * @inode: the directory inode
1471 * @hsize: hash table size
1472 * @index: index into the hash table
1473 * @f_ra: read-ahead parameters
1474 *
1475 * Note: we can't calculate each index like dir_e_read can because we don't
1476 * have the leaf, and therefore we don't have the depth, and therefore we
1477 * don't have the length. So we have to just read enough ahead to make up
1478 * for the loss of information.
1479 */
1480static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1481			       struct file_ra_state *f_ra)
1482{
1483	struct gfs2_inode *ip = GFS2_I(inode);
1484	struct gfs2_glock *gl = ip->i_gl;
1485	struct buffer_head *bh;
1486	u64 blocknr = 0, last;
1487	unsigned count;
1488
1489	/* First check if we've already read-ahead for the whole range. */
1490	if (index + MAX_RA_BLOCKS < f_ra->start)
1491		return;
1492
1493	f_ra->start = max((pgoff_t)index, f_ra->start);
1494	for (count = 0; count < MAX_RA_BLOCKS; count++) {
1495		if (f_ra->start >= hsize) /* if exceeded the hash table */
1496			break;
1497
1498		last = blocknr;
1499		blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1500		f_ra->start++;
1501		if (blocknr == last)
1502			continue;
1503
1504		bh = gfs2_getbuf(gl, blocknr, 1);
1505		if (trylock_buffer(bh)) {
1506			if (buffer_uptodate(bh)) {
1507				unlock_buffer(bh);
1508				brelse(bh);
1509				continue;
1510			}
1511			bh->b_end_io = end_buffer_read_sync;
1512			submit_bh(REQ_OP_READ | REQ_RAHEAD | REQ_META |
1513				  REQ_PRIO, bh);
1514			continue;
1515		}
1516		brelse(bh);
1517	}
1518}
1519
1520/**
1521 * dir_e_read - Reads the entries from a directory into a filldir buffer
1522 * @inode: the directory inode
1523 * @ctx: actor to feed the entries to
1524 * @f_ra: read-ahead parameters
1525 *
1526 * Returns: errno
1527 */
1528
1529static int dir_e_read(struct inode *inode, struct dir_context *ctx,
1530		      struct file_ra_state *f_ra)
1531{
1532	struct gfs2_inode *dip = GFS2_I(inode);
1533	u32 hsize, len = 0;
1534	u32 hash, index;
1535	__be64 *lp;
1536	int copied = 0;
1537	int error = 0;
1538	unsigned depth = 0;
1539
1540	hsize = BIT(dip->i_depth);
1541	hash = gfs2_dir_offset2hash(ctx->pos);
1542	index = hash >> (32 - dip->i_depth);
1543
1544	if (dip->i_hash_cache == NULL)
1545		f_ra->start = 0;
1546	lp = gfs2_dir_get_hash_table(dip);
1547	if (IS_ERR(lp))
1548		return PTR_ERR(lp);
1549
1550	gfs2_dir_readahead(inode, hsize, index, f_ra);
1551
1552	while (index < hsize) {
1553		error = gfs2_dir_read_leaf(inode, ctx,
1554					   &copied, &depth,
1555					   be64_to_cpu(lp[index]));
1556		if (error)
1557			break;
1558
1559		len = BIT(dip->i_depth - depth);
1560		index = (index & ~(len - 1)) + len;
1561	}
1562
1563	if (error > 0)
1564		error = 0;
1565	return error;
1566}
1567
1568int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
1569		  struct file_ra_state *f_ra)
1570{
1571	struct gfs2_inode *dip = GFS2_I(inode);
1572	struct gfs2_sbd *sdp = GFS2_SB(inode);
1573	struct dirent_gather g;
1574	struct gfs2_dirent **darr, *dent;
1575	struct buffer_head *dibh;
1576	int copied = 0;
1577	int error;
1578
1579	if (!dip->i_entries)
1580		return 0;
1581
1582	if (dip->i_diskflags & GFS2_DIF_EXHASH)
1583		return dir_e_read(inode, ctx, f_ra);
1584
1585	if (!gfs2_is_stuffed(dip)) {
1586		gfs2_consist_inode(dip);
1587		return -EIO;
1588	}
1589
1590	error = gfs2_meta_inode_buffer(dip, &dibh);
1591	if (error)
1592		return error;
1593
1594	error = -ENOMEM;
1595	/* 96 is max number of dirents which can be stuffed into an inode */
1596	darr = kmalloc_array(96, sizeof(struct gfs2_dirent *), GFP_NOFS);
1597	if (darr) {
1598		g.pdent = (const struct gfs2_dirent **)darr;
1599		g.offset = 0;
1600		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1601					gfs2_dirent_gather, NULL, &g);
1602		if (IS_ERR(dent)) {
1603			error = PTR_ERR(dent);
1604			goto out;
1605		}
1606		if (dip->i_entries != g.offset) {
1607			fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1608				"ip->i_entries (%u) != g.offset (%u)\n",
1609				(unsigned long long)dip->i_no_addr,
1610				dip->i_entries,
1611				g.offset);
1612			gfs2_consist_inode(dip);
1613			error = -EIO;
1614			goto out;
1615		}
1616		gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
1617		error = do_filldir_main(dip, ctx, darr,
1618					dip->i_entries, 0, &copied);
1619out:
1620		kfree(darr);
1621	}
1622
1623	if (error > 0)
1624		error = 0;
1625
1626	brelse(dibh);
1627
1628	return error;
1629}
1630
1631/**
1632 * gfs2_dir_search - Search a directory
1633 * @dir: The GFS2 directory inode
1634 * @name: The name we are looking up
1635 * @fail_on_exist: Fail if the name exists rather than looking it up
1636 *
1637 * This routine searches a directory for a file or another directory.
1638 * Assumes a glock is held on dip.
1639 *
1640 * Returns: errno
1641 */
1642
1643struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
1644			      bool fail_on_exist)
1645{
1646	struct buffer_head *bh;
1647	struct gfs2_dirent *dent;
1648	u64 addr, formal_ino;
1649	u16 dtype;
1650
1651	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1652	if (dent) {
1653		struct inode *inode;
1654		u16 rahead;
1655
1656		if (IS_ERR(dent))
1657			return ERR_CAST(dent);
1658		dtype = be16_to_cpu(dent->de_type);
1659		rahead = be16_to_cpu(dent->de_rahead);
1660		addr = be64_to_cpu(dent->de_inum.no_addr);
1661		formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
1662		brelse(bh);
1663		if (fail_on_exist)
1664			return ERR_PTR(-EEXIST);
1665		inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino,
1666					  GFS2_BLKST_FREE /* ignore */);
1667		if (!IS_ERR(inode))
1668			GFS2_I(inode)->i_rahead = rahead;
1669		return inode;
1670	}
1671	return ERR_PTR(-ENOENT);
1672}
1673
1674int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1675		   const struct gfs2_inode *ip)
1676{
1677	struct buffer_head *bh;
1678	struct gfs2_dirent *dent;
1679	int ret = -ENOENT;
1680
1681	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1682	if (dent) {
1683		if (IS_ERR(dent))
1684			return PTR_ERR(dent);
1685		if (ip) {
1686			if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1687				goto out;
1688			if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1689			    ip->i_no_formal_ino)
1690				goto out;
1691			if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1692			    be16_to_cpu(dent->de_type))) {
1693				gfs2_consist_inode(GFS2_I(dir));
1694				ret = -EIO;
1695				goto out;
1696			}
1697		}
1698		ret = 0;
1699out:
1700		brelse(bh);
1701	}
1702	return ret;
1703}
1704
1705/**
1706 * dir_new_leaf - Add a new leaf onto hash chain
1707 * @inode: The directory
1708 * @name: The name we are adding
1709 *
1710 * This adds a new dir leaf onto an existing leaf when there is not
1711 * enough space to add a new dir entry. This is a last resort after
1712 * we've expanded the hash table to max size and also split existing
1713 * leaf blocks, so it will only occur for very large directories.
1714 *
1715 * The dist parameter is set to 1 for leaf blocks directly attached
1716 * to the hash table, 2 for one layer of indirection, 3 for two layers
1717 * etc. We are thus able to tell the difference between an old leaf
1718 * with dist set to zero (i.e. "don't know") and a new one where we
1719 * set this information for debug/fsck purposes.
1720 *
1721 * Returns: 0 on success, or -ve on error
1722 */
1723
1724static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1725{
1726	struct buffer_head *bh, *obh;
1727	struct gfs2_inode *ip = GFS2_I(inode);
1728	struct gfs2_leaf *leaf, *oleaf;
1729	u32 dist = 1;
1730	int error;
1731	u32 index;
1732	u64 bn;
1733
1734	index = name->hash >> (32 - ip->i_depth);
1735	error = get_first_leaf(ip, index, &obh);
1736	if (error)
1737		return error;
1738	do {
1739		dist++;
1740		oleaf = (struct gfs2_leaf *)obh->b_data;
1741		bn = be64_to_cpu(oleaf->lf_next);
1742		if (!bn)
1743			break;
1744		brelse(obh);
1745		error = get_leaf(ip, bn, &obh);
1746		if (error)
1747			return error;
1748	} while(1);
1749
1750	gfs2_trans_add_meta(ip->i_gl, obh);
1751
1752	leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1753	if (!leaf) {
1754		brelse(obh);
1755		return -ENOSPC;
1756	}
1757	leaf->lf_dist = cpu_to_be32(dist);
1758	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1759	brelse(bh);
1760	brelse(obh);
1761
1762	error = gfs2_meta_inode_buffer(ip, &bh);
1763	if (error)
1764		return error;
1765	gfs2_trans_add_meta(ip->i_gl, bh);
1766	gfs2_add_inode_blocks(&ip->i_inode, 1);
1767	gfs2_dinode_out(ip, bh->b_data);
1768	brelse(bh);
1769	return 0;
1770}
1771
1772static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip)
1773{
1774	u64 where = ip->i_no_addr + 1;
1775	if (ip->i_eattr == where)
1776		return 1;
1777	return 0;
1778}
1779
1780/**
1781 * gfs2_dir_add - Add new filename into directory
1782 * @inode: The directory inode
1783 * @name: The new name
1784 * @nip: The GFS2 inode to be linked in to the directory
1785 * @da: The directory addition info
1786 *
1787 * If the call to gfs2_diradd_alloc_required resulted in there being
1788 * no need to allocate any new directory blocks, then it will contain
1789 * a pointer to the directory entry and the bh in which it resides. We
1790 * can use that without having to repeat the search. If there was no
1791 * free space, then we must now create more space.
1792 *
1793 * Returns: 0 on success, error code on failure
1794 */
1795
1796int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1797		 const struct gfs2_inode *nip, struct gfs2_diradd *da)
1798{
1799	struct gfs2_inode *ip = GFS2_I(inode);
1800	struct buffer_head *bh = da->bh;
1801	struct gfs2_dirent *dent = da->dent;
1802	struct timespec64 tv;
1803	struct gfs2_leaf *leaf;
1804	int error;
1805
1806	while(1) {
1807		if (da->bh == NULL) {
1808			dent = gfs2_dirent_search(inode, name,
1809						  gfs2_dirent_find_space, &bh);
1810		}
1811		if (dent) {
1812			if (IS_ERR(dent))
1813				return PTR_ERR(dent);
1814			dent = gfs2_init_dirent(inode, dent, name, bh);
1815			gfs2_inum_out(nip, dent);
1816			dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1817			dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip));
1818			tv = inode_set_ctime_current(&ip->i_inode);
1819			if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1820				leaf = (struct gfs2_leaf *)bh->b_data;
1821				be16_add_cpu(&leaf->lf_entries, 1);
1822				leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1823				leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1824			}
1825			da->dent = NULL;
1826			da->bh = NULL;
1827			brelse(bh);
1828			ip->i_entries++;
1829			inode_set_mtime_to_ts(&ip->i_inode, tv);
1830			if (S_ISDIR(nip->i_inode.i_mode))
1831				inc_nlink(&ip->i_inode);
1832			mark_inode_dirty(inode);
1833			error = 0;
1834			break;
1835		}
1836		if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1837			error = dir_make_exhash(inode);
1838			if (error)
1839				break;
1840			continue;
1841		}
1842		error = dir_split_leaf(inode, name);
1843		if (error == 0)
1844			continue;
1845		if (error < 0)
1846			break;
1847		if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1848			error = dir_double_exhash(ip);
1849			if (error)
1850				break;
1851			error = dir_split_leaf(inode, name);
1852			if (error < 0)
1853				break;
1854			if (error == 0)
1855				continue;
1856		}
1857		error = dir_new_leaf(inode, name);
1858		if (!error)
1859			continue;
1860		error = -ENOSPC;
1861		break;
1862	}
1863	return error;
1864}
1865
1866
1867/**
1868 * gfs2_dir_del - Delete a directory entry
1869 * @dip: The GFS2 inode
1870 * @dentry: The directory entry we want to delete
1871 *
1872 * Returns: 0 on success, error code on failure
1873 */
1874
1875int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
1876{
1877	const struct qstr *name = &dentry->d_name;
1878	struct gfs2_dirent *dent, *prev = NULL;
1879	struct buffer_head *bh;
1880	struct timespec64 tv;
1881
1882	/* Returns _either_ the entry (if its first in block) or the
1883	   previous entry otherwise */
1884	dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1885	if (!dent) {
1886		gfs2_consist_inode(dip);
1887		return -EIO;
1888	}
1889	if (IS_ERR(dent)) {
1890		gfs2_consist_inode(dip);
1891		return PTR_ERR(dent);
1892	}
1893	/* If not first in block, adjust pointers accordingly */
1894	if (gfs2_dirent_find(dent, name, NULL) == 0) {
1895		prev = dent;
1896		dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1897	}
1898
1899	dirent_del(dip, bh, prev, dent);
1900	tv = inode_set_ctime_current(&dip->i_inode);
1901	if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1902		struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1903		u16 entries = be16_to_cpu(leaf->lf_entries);
1904		if (!entries)
1905			gfs2_consist_inode(dip);
1906		leaf->lf_entries = cpu_to_be16(--entries);
1907		leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
1908		leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1909	}
1910	brelse(bh);
1911
1912	if (!dip->i_entries)
1913		gfs2_consist_inode(dip);
1914	dip->i_entries--;
1915	inode_set_mtime_to_ts(&dip->i_inode, tv);
1916	if (d_is_dir(dentry))
1917		drop_nlink(&dip->i_inode);
1918	mark_inode_dirty(&dip->i_inode);
1919
1920	return 0;
1921}
1922
1923/**
1924 * gfs2_dir_mvino - Change inode number of directory entry
1925 * @dip: The GFS2 directory inode
1926 * @filename: the filename to be moved
1927 * @nip: the new GFS2 inode
1928 * @new_type: the de_type of the new dirent
1929 *
1930 * This routine changes the inode number of a directory entry.  It's used
1931 * by rename to change ".." when a directory is moved.
1932 * Assumes a glock is held on dvp.
1933 *
1934 * Returns: errno
1935 */
1936
1937int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1938		   const struct gfs2_inode *nip, unsigned int new_type)
1939{
1940	struct buffer_head *bh;
1941	struct gfs2_dirent *dent;
1942
1943	dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1944	if (!dent) {
1945		gfs2_consist_inode(dip);
1946		return -EIO;
1947	}
1948	if (IS_ERR(dent))
1949		return PTR_ERR(dent);
1950
1951	gfs2_trans_add_meta(dip->i_gl, bh);
1952	gfs2_inum_out(nip, dent);
1953	dent->de_type = cpu_to_be16(new_type);
1954	brelse(bh);
1955
1956	inode_set_mtime_to_ts(&dip->i_inode, inode_set_ctime_current(&dip->i_inode));
1957	mark_inode_dirty_sync(&dip->i_inode);
1958	return 0;
1959}
1960
1961/**
1962 * leaf_dealloc - Deallocate a directory leaf
1963 * @dip: the directory
1964 * @index: the hash table offset in the directory
1965 * @len: the number of pointers to this leaf
1966 * @leaf_no: the leaf number
1967 * @leaf_bh: buffer_head for the starting leaf
1968 * @last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1969 *
1970 * Returns: errno
1971 */
1972
1973static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1974			u64 leaf_no, struct buffer_head *leaf_bh,
1975			int last_dealloc)
1976{
1977	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1978	struct gfs2_leaf *tmp_leaf;
1979	struct gfs2_rgrp_list rlist;
1980	struct buffer_head *bh, *dibh;
1981	u64 blk, nblk;
1982	unsigned int rg_blocks = 0, l_blocks = 0;
1983	char *ht;
1984	unsigned int x, size = len * sizeof(u64);
1985	int error;
1986
1987	error = gfs2_rindex_update(sdp);
1988	if (error)
1989		return error;
1990
1991	memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1992
1993	ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
1994	if (ht == NULL)
1995		ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO);
1996	if (!ht)
1997		return -ENOMEM;
1998
1999	error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
2000	if (error)
2001		goto out;
2002
2003	/*  Count the number of leaves  */
2004	bh = leaf_bh;
2005
2006	for (blk = leaf_no; blk; blk = nblk) {
2007		if (blk != leaf_no) {
2008			error = get_leaf(dip, blk, &bh);
2009			if (error)
2010				goto out_rlist;
2011		}
2012		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2013		nblk = be64_to_cpu(tmp_leaf->lf_next);
2014		if (blk != leaf_no)
2015			brelse(bh);
2016
2017		gfs2_rlist_add(dip, &rlist, blk);
2018		l_blocks++;
2019	}
2020
2021	gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, LM_FLAG_NODE_SCOPE);
2022
2023	for (x = 0; x < rlist.rl_rgrps; x++) {
2024		struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl);
2025
2026		rg_blocks += rgd->rd_length;
2027	}
2028
2029	error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
2030	if (error)
2031		goto out_rlist;
2032
2033	error = gfs2_trans_begin(sdp,
2034			rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
2035			RES_DINODE + RES_STATFS + RES_QUOTA, RES_DINODE +
2036				 l_blocks);
2037	if (error)
2038		goto out_rg_gunlock;
2039
2040	bh = leaf_bh;
2041
2042	for (blk = leaf_no; blk; blk = nblk) {
2043		struct gfs2_rgrpd *rgd;
2044
2045		if (blk != leaf_no) {
2046			error = get_leaf(dip, blk, &bh);
2047			if (error)
2048				goto out_end_trans;
2049		}
2050		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
2051		nblk = be64_to_cpu(tmp_leaf->lf_next);
2052		if (blk != leaf_no)
2053			brelse(bh);
2054
2055		rgd = gfs2_blk2rgrpd(sdp, blk, true);
2056		gfs2_free_meta(dip, rgd, blk, 1);
2057		gfs2_add_inode_blocks(&dip->i_inode, -1);
2058	}
2059
2060	error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
2061	if (error != size) {
2062		if (error >= 0)
2063			error = -EIO;
2064		goto out_end_trans;
2065	}
2066
2067	error = gfs2_meta_inode_buffer(dip, &dibh);
2068	if (error)
2069		goto out_end_trans;
2070
2071	gfs2_trans_add_meta(dip->i_gl, dibh);
2072	/* On the last dealloc, make this a regular file in case we crash.
2073	   (We don't want to free these blocks a second time.)  */
2074	if (last_dealloc)
2075		dip->i_inode.i_mode = S_IFREG;
2076	gfs2_dinode_out(dip, dibh->b_data);
2077	brelse(dibh);
2078
2079out_end_trans:
2080	gfs2_trans_end(sdp);
2081out_rg_gunlock:
2082	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
2083out_rlist:
2084	gfs2_rlist_free(&rlist);
2085	gfs2_quota_unhold(dip);
2086out:
2087	kvfree(ht);
2088	return error;
2089}
2090
2091/**
2092 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2093 * @dip: the directory
2094 *
2095 * Dealloc all on-disk directory leaves to FREEMETA state
2096 * Change on-disk inode type to "regular file"
2097 *
2098 * Returns: errno
2099 */
2100
2101int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
2102{
2103	struct buffer_head *bh;
2104	struct gfs2_leaf *leaf;
2105	u32 hsize, len;
2106	u32 index = 0, next_index;
2107	__be64 *lp;
2108	u64 leaf_no;
2109	int error = 0, last;
2110
2111	hsize = BIT(dip->i_depth);
2112
2113	lp = gfs2_dir_get_hash_table(dip);
2114	if (IS_ERR(lp))
2115		return PTR_ERR(lp);
2116
2117	while (index < hsize) {
2118		leaf_no = be64_to_cpu(lp[index]);
2119		if (leaf_no) {
2120			error = get_leaf(dip, leaf_no, &bh);
2121			if (error)
2122				goto out;
2123			leaf = (struct gfs2_leaf *)bh->b_data;
2124			len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth));
2125
2126			next_index = (index & ~(len - 1)) + len;
2127			last = ((next_index >= hsize) ? 1 : 0);
2128			error = leaf_dealloc(dip, index, len, leaf_no, bh,
2129					     last);
2130			brelse(bh);
2131			if (error)
2132				goto out;
2133			index = next_index;
2134		} else
2135			index++;
2136	}
2137
2138	if (index != hsize) {
2139		gfs2_consist_inode(dip);
2140		error = -EIO;
2141	}
2142
2143out:
2144
2145	return error;
2146}
2147
2148/**
2149 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2150 * @inode: the directory inode being written to
2151 * @name: the filename that's going to be added
2152 * @da: The structure to return dir alloc info
2153 *
2154 * Returns: 0 if ok, -ve on error
2155 */
2156
2157int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
2158			       struct gfs2_diradd *da)
2159{
2160	struct gfs2_inode *ip = GFS2_I(inode);
2161	struct gfs2_sbd *sdp = GFS2_SB(inode);
2162	const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
2163	struct gfs2_dirent *dent;
2164	struct buffer_head *bh;
2165
2166	da->nr_blocks = 0;
2167	da->bh = NULL;
2168	da->dent = NULL;
2169
2170	dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2171	if (!dent) {
2172		da->nr_blocks = sdp->sd_max_dirres;
2173		if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
2174		    (GFS2_DIRENT_SIZE(name->len) < extra))
2175			da->nr_blocks = 1;
2176		return 0;
2177	}
2178	if (IS_ERR(dent))
2179		return PTR_ERR(dent);
2180
2181	if (da->save_loc) {
2182		da->bh = bh;
2183		da->dent = dent;
2184	} else {
2185		brelse(bh);
2186	}
2187	return 0;
2188}
2189