1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/ext4/resize.c
   4 *
   5 * Support for resizing an ext4 filesystem while it is mounted.
   6 *
   7 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
   8 *
   9 * This could probably be made into a module, because it is not often in use.
  10 */
  11
  12
  13#define EXT4FS_DEBUG
  14
  15#include <linux/errno.h>
  16#include <linux/slab.h>
  17#include <linux/jiffies.h>
  18
  19#include "ext4_jbd2.h"
  20
  21struct ext4_rcu_ptr {
  22	struct rcu_head rcu;
  23	void *ptr;
  24};
  25
  26static void ext4_rcu_ptr_callback(struct rcu_head *head)
  27{
  28	struct ext4_rcu_ptr *ptr;
  29
  30	ptr = container_of(head, struct ext4_rcu_ptr, rcu);
  31	kvfree(ptr->ptr);
  32	kfree(ptr);
  33}
  34
  35void ext4_kvfree_array_rcu(void *to_free)
  36{
  37	struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
  38
  39	if (ptr) {
  40		ptr->ptr = to_free;
  41		call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
  42		return;
  43	}
  44	synchronize_rcu();
  45	kvfree(to_free);
  46}
  47
  48int ext4_resize_begin(struct super_block *sb)
  49{
  50	struct ext4_sb_info *sbi = EXT4_SB(sb);
  51	int ret = 0;
  52
  53	if (!capable(CAP_SYS_RESOURCE))
  54		return -EPERM;
  55
  56	/*
  57	 * If the reserved GDT blocks is non-zero, the resize_inode feature
  58	 * should always be set.
  59	 */
  60	if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
  61	    !ext4_has_feature_resize_inode(sb)) {
  62		ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
  63		return -EFSCORRUPTED;
  64	}
  65
  66	/*
  67	 * If we are not using the primary superblock/GDT copy don't resize,
  68         * because the user tools have no way of handling this.  Probably a
  69         * bad time to do it anyways.
  70         */
  71	if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
  72	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
  73		ext4_warning(sb, "won't resize using backup superblock at %llu",
  74			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
  75		return -EPERM;
  76	}
  77
  78	/*
  79	 * We are not allowed to do online-resizing on a filesystem mounted
  80	 * with error, because it can destroy the filesystem easily.
  81	 */
  82	if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  83		ext4_warning(sb, "There are errors in the filesystem, "
  84			     "so online resizing is not allowed");
  85		return -EPERM;
  86	}
  87
  88	if (ext4_has_feature_sparse_super2(sb)) {
  89		ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
  90		return -EOPNOTSUPP;
  91	}
  92
  93	if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
  94				  &EXT4_SB(sb)->s_ext4_flags))
  95		ret = -EBUSY;
  96
  97	return ret;
  98}
  99
 100int ext4_resize_end(struct super_block *sb, bool update_backups)
 101{
 102	clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
 103	smp_mb__after_atomic();
 104	if (update_backups)
 105		return ext4_update_overhead(sb, true);
 106	return 0;
 107}
 108
 109static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
 110					     ext4_group_t group) {
 111	return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
 112	       EXT4_DESC_PER_BLOCK_BITS(sb);
 113}
 114
 115static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
 116					     ext4_group_t group) {
 117	group = ext4_meta_bg_first_group(sb, group);
 118	return ext4_group_first_block_no(sb, group);
 119}
 120
 121static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
 122						ext4_group_t group) {
 123	ext4_grpblk_t overhead;
 124	overhead = ext4_bg_num_gdb(sb, group);
 125	if (ext4_bg_has_super(sb, group))
 126		overhead += 1 +
 127			  le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 128	return overhead;
 129}
 130
 131#define outside(b, first, last)	((b) < (first) || (b) >= (last))
 132#define inside(b, first, last)	((b) >= (first) && (b) < (last))
 133
 134static int verify_group_input(struct super_block *sb,
 135			      struct ext4_new_group_data *input)
 136{
 137	struct ext4_sb_info *sbi = EXT4_SB(sb);
 138	struct ext4_super_block *es = sbi->s_es;
 139	ext4_fsblk_t start = ext4_blocks_count(es);
 140	ext4_fsblk_t end = start + input->blocks_count;
 141	ext4_group_t group = input->group;
 142	ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
 143	unsigned overhead;
 144	ext4_fsblk_t metaend;
 145	struct buffer_head *bh = NULL;
 146	ext4_grpblk_t free_blocks_count, offset;
 147	int err = -EINVAL;
 148
 149	if (group != sbi->s_groups_count) {
 150		ext4_warning(sb, "Cannot add at group %u (only %u groups)",
 151			     input->group, sbi->s_groups_count);
 152		return -EINVAL;
 153	}
 154
 155	overhead = ext4_group_overhead_blocks(sb, group);
 156	metaend = start + overhead;
 157	input->free_clusters_count = free_blocks_count =
 158		input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
 
 159
 160	if (test_opt(sb, DEBUG))
 161		printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 162		       "(%d free, %u reserved)\n",
 163		       ext4_bg_has_super(sb, input->group) ? "normal" :
 164		       "no-super", input->group, input->blocks_count,
 165		       free_blocks_count, input->reserved_blocks);
 166
 167	ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 168	if (offset != 0)
 169			ext4_warning(sb, "Last group not full");
 170	else if (input->reserved_blocks > input->blocks_count / 5)
 171		ext4_warning(sb, "Reserved blocks too high (%u)",
 172			     input->reserved_blocks);
 173	else if (free_blocks_count < 0)
 174		ext4_warning(sb, "Bad blocks count %u",
 175			     input->blocks_count);
 176	else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
 177		err = PTR_ERR(bh);
 178		bh = NULL;
 179		ext4_warning(sb, "Cannot read last block (%llu)",
 180			     end - 1);
 181	} else if (outside(input->block_bitmap, start, end))
 182		ext4_warning(sb, "Block bitmap not in group (block %llu)",
 183			     (unsigned long long)input->block_bitmap);
 184	else if (outside(input->inode_bitmap, start, end))
 185		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 186			     (unsigned long long)input->inode_bitmap);
 187	else if (outside(input->inode_table, start, end) ||
 188		 outside(itend - 1, start, end))
 189		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 190			     (unsigned long long)input->inode_table, itend - 1);
 191	else if (input->inode_bitmap == input->block_bitmap)
 192		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 193			     (unsigned long long)input->block_bitmap);
 194	else if (inside(input->block_bitmap, input->inode_table, itend))
 195		ext4_warning(sb, "Block bitmap (%llu) in inode table "
 196			     "(%llu-%llu)",
 197			     (unsigned long long)input->block_bitmap,
 198			     (unsigned long long)input->inode_table, itend - 1);
 199	else if (inside(input->inode_bitmap, input->inode_table, itend))
 200		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 201			     "(%llu-%llu)",
 202			     (unsigned long long)input->inode_bitmap,
 203			     (unsigned long long)input->inode_table, itend - 1);
 204	else if (inside(input->block_bitmap, start, metaend))
 205		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 206			     (unsigned long long)input->block_bitmap,
 207			     start, metaend - 1);
 208	else if (inside(input->inode_bitmap, start, metaend))
 209		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 210			     (unsigned long long)input->inode_bitmap,
 211			     start, metaend - 1);
 212	else if (inside(input->inode_table, start, metaend) ||
 213		 inside(itend - 1, start, metaend))
 214		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 215			     "(%llu-%llu)",
 216			     (unsigned long long)input->inode_table,
 217			     itend - 1, start, metaend - 1);
 218	else
 219		err = 0;
 220	brelse(bh);
 221
 222	return err;
 223}
 224
 225/*
 226 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 227 * group each time.
 228 */
 229struct ext4_new_flex_group_data {
 230	struct ext4_new_group_data *groups;	/* new_group_data for groups
 231						   in the flex group */
 232	__u16 *bg_flags;			/* block group flags of groups
 233						   in @groups */
 
 
 234	ext4_group_t count;			/* number of groups in @groups
 235						 */
 236};
 237
 238/*
 239 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 240 * @flexbg_size.
 
 
 
 
 
 241 *
 242 * Returns NULL on failure otherwise address of the allocated structure.
 243 */
 244static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 
 245{
 
 
 246	struct ext4_new_flex_group_data *flex_gd;
 247
 248	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 249	if (flex_gd == NULL)
 250		goto out3;
 251
 252	if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
 253		goto out2;
 254	flex_gd->count = flexbg_size;
 
 
 
 
 
 
 
 
 
 
 255
 256	flex_gd->groups = kmalloc_array(flexbg_size,
 257					sizeof(struct ext4_new_group_data),
 258					GFP_NOFS);
 259	if (flex_gd->groups == NULL)
 260		goto out2;
 261
 262	flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
 263					  GFP_NOFS);
 264	if (flex_gd->bg_flags == NULL)
 265		goto out1;
 266
 267	return flex_gd;
 268
 269out1:
 270	kfree(flex_gd->groups);
 271out2:
 272	kfree(flex_gd);
 273out3:
 274	return NULL;
 275}
 276
 277static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 278{
 279	kfree(flex_gd->bg_flags);
 280	kfree(flex_gd->groups);
 281	kfree(flex_gd);
 282}
 283
 284/*
 285 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 286 * and inode tables for a flex group.
 287 *
 288 * This function is used by 64bit-resize.  Note that this function allocates
 289 * group tables from the 1st group of groups contained by @flexgd, which may
 290 * be a partial of a flex group.
 291 *
 292 * @sb: super block of fs to which the groups belongs
 293 *
 294 * Returns 0 on a successful allocation of the metadata blocks in the
 295 * block group.
 296 */
 297static int ext4_alloc_group_tables(struct super_block *sb,
 298				struct ext4_new_flex_group_data *flex_gd,
 299				int flexbg_size)
 300{
 301	struct ext4_new_group_data *group_data = flex_gd->groups;
 302	ext4_fsblk_t start_blk;
 303	ext4_fsblk_t last_blk;
 304	ext4_group_t src_group;
 305	ext4_group_t bb_index = 0;
 306	ext4_group_t ib_index = 0;
 307	ext4_group_t it_index = 0;
 308	ext4_group_t group;
 309	ext4_group_t last_group;
 310	unsigned overhead;
 311	__u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
 312	int i;
 313
 314	BUG_ON(flex_gd->count == 0 || group_data == NULL);
 315
 316	src_group = group_data[0].group;
 317	last_group  = src_group + flex_gd->count - 1;
 318
 319	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 320	       (last_group & ~(flexbg_size - 1))));
 321next_group:
 322	group = group_data[0].group;
 323	if (src_group >= group_data[0].group + flex_gd->count)
 324		return -ENOSPC;
 325	start_blk = ext4_group_first_block_no(sb, src_group);
 326	last_blk = start_blk + group_data[src_group - group].blocks_count;
 327
 328	overhead = ext4_group_overhead_blocks(sb, src_group);
 329
 330	start_blk += overhead;
 331
 332	/* We collect contiguous blocks as much as possible. */
 333	src_group++;
 334	for (; src_group <= last_group; src_group++) {
 335		overhead = ext4_group_overhead_blocks(sb, src_group);
 336		if (overhead == 0)
 337			last_blk += group_data[src_group - group].blocks_count;
 338		else
 339			break;
 340	}
 341
 342	/* Allocate block bitmaps */
 343	for (; bb_index < flex_gd->count; bb_index++) {
 344		if (start_blk >= last_blk)
 345			goto next_group;
 346		group_data[bb_index].block_bitmap = start_blk++;
 347		group = ext4_get_group_number(sb, start_blk - 1);
 348		group -= group_data[0].group;
 349		group_data[group].mdata_blocks++;
 350		flex_gd->bg_flags[group] &= uninit_mask;
 351	}
 352
 353	/* Allocate inode bitmaps */
 354	for (; ib_index < flex_gd->count; ib_index++) {
 355		if (start_blk >= last_blk)
 356			goto next_group;
 357		group_data[ib_index].inode_bitmap = start_blk++;
 358		group = ext4_get_group_number(sb, start_blk - 1);
 359		group -= group_data[0].group;
 360		group_data[group].mdata_blocks++;
 361		flex_gd->bg_flags[group] &= uninit_mask;
 362	}
 363
 364	/* Allocate inode tables */
 365	for (; it_index < flex_gd->count; it_index++) {
 366		unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
 367		ext4_fsblk_t next_group_start;
 368
 369		if (start_blk + itb > last_blk)
 370			goto next_group;
 371		group_data[it_index].inode_table = start_blk;
 372		group = ext4_get_group_number(sb, start_blk);
 373		next_group_start = ext4_group_first_block_no(sb, group + 1);
 374		group -= group_data[0].group;
 375
 376		if (start_blk + itb > next_group_start) {
 377			flex_gd->bg_flags[group + 1] &= uninit_mask;
 378			overhead = start_blk + itb - next_group_start;
 379			group_data[group + 1].mdata_blocks += overhead;
 380			itb -= overhead;
 381		}
 382
 383		group_data[group].mdata_blocks += itb;
 384		flex_gd->bg_flags[group] &= uninit_mask;
 385		start_blk += EXT4_SB(sb)->s_itb_per_group;
 386	}
 387
 388	/* Update free clusters count to exclude metadata blocks */
 389	for (i = 0; i < flex_gd->count; i++) {
 390		group_data[i].free_clusters_count -=
 391				EXT4_NUM_B2C(EXT4_SB(sb),
 392					     group_data[i].mdata_blocks);
 393	}
 394
 395	if (test_opt(sb, DEBUG)) {
 396		int i;
 397		group = group_data[0].group;
 398
 399		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 400		       "%d groups, flexbg size is %d:\n", flex_gd->count,
 401		       flexbg_size);
 402
 403		for (i = 0; i < flex_gd->count; i++) {
 404			ext4_debug(
 405			       "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
 406			       ext4_bg_has_super(sb, group + i) ? "normal" :
 407			       "no-super", group + i,
 408			       group_data[i].blocks_count,
 409			       group_data[i].free_clusters_count,
 410			       group_data[i].mdata_blocks);
 411		}
 412	}
 413	return 0;
 414}
 415
 416static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 417				  ext4_fsblk_t blk)
 418{
 419	struct buffer_head *bh;
 420	int err;
 421
 422	bh = sb_getblk(sb, blk);
 423	if (unlikely(!bh))
 424		return ERR_PTR(-ENOMEM);
 425	BUFFER_TRACE(bh, "get_write_access");
 426	err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
 427	if (err) {
 428		brelse(bh);
 429		bh = ERR_PTR(err);
 430	} else {
 431		memset(bh->b_data, 0, sb->s_blocksize);
 432		set_buffer_uptodate(bh);
 433	}
 434
 435	return bh;
 436}
 437
 438static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
 439{
 440	return ext4_journal_ensure_credits_fn(handle, credits,
 441		EXT4_MAX_TRANS_DATA, 0, 0);
 442}
 443
 444/*
 445 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
 446 *
 447 * Helper function for ext4_setup_new_group_blocks() which set .
 448 *
 449 * @sb: super block
 450 * @handle: journal handle
 451 * @flex_gd: flex group data
 452 */
 453static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 454			struct ext4_new_flex_group_data *flex_gd,
 455			ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
 456{
 457	struct ext4_sb_info *sbi = EXT4_SB(sb);
 458	ext4_group_t count = last_cluster - first_cluster + 1;
 459	ext4_group_t count2;
 460
 461	ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
 462		   last_cluster);
 463	for (count2 = count; count > 0;
 464	     count -= count2, first_cluster += count2) {
 465		ext4_fsblk_t start;
 466		struct buffer_head *bh;
 467		ext4_group_t group;
 468		int err;
 469
 470		group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
 471		start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
 472		group -= flex_gd->groups[0].group;
 473
 474		count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
 475		if (count2 > count)
 476			count2 = count;
 477
 478		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 479			BUG_ON(flex_gd->count > 1);
 480			continue;
 481		}
 482
 483		err = ext4_resize_ensure_credits_batch(handle, 1);
 484		if (err < 0)
 485			return err;
 486
 487		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 488		if (unlikely(!bh))
 489			return -ENOMEM;
 490
 491		BUFFER_TRACE(bh, "get_write_access");
 492		err = ext4_journal_get_write_access(handle, sb, bh,
 493						    EXT4_JTR_NONE);
 494		if (err) {
 495			brelse(bh);
 496			return err;
 497		}
 498		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
 499			   first_cluster, first_cluster - start, count2);
 500		mb_set_bits(bh->b_data, first_cluster - start, count2);
 501
 502		err = ext4_handle_dirty_metadata(handle, NULL, bh);
 503		brelse(bh);
 504		if (unlikely(err))
 505			return err;
 506	}
 507
 508	return 0;
 509}
 510
 511/*
 512 * Set up the block and inode bitmaps, and the inode table for the new groups.
 513 * This doesn't need to be part of the main transaction, since we are only
 514 * changing blocks outside the actual filesystem.  We still do journaling to
 515 * ensure the recovery is correct in case of a failure just after resize.
 516 * If any part of this fails, we simply abort the resize.
 517 *
 518 * setup_new_flex_group_blocks handles a flex group as follow:
 519 *  1. copy super block and GDT, and initialize group tables if necessary.
 520 *     In this step, we only set bits in blocks bitmaps for blocks taken by
 521 *     super block and GDT.
 522 *  2. allocate group tables in block bitmaps, that is, set bits in block
 523 *     bitmap for blocks taken by group tables.
 524 */
 525static int setup_new_flex_group_blocks(struct super_block *sb,
 526				struct ext4_new_flex_group_data *flex_gd)
 527{
 528	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 529	ext4_fsblk_t start;
 530	ext4_fsblk_t block;
 531	struct ext4_sb_info *sbi = EXT4_SB(sb);
 532	struct ext4_super_block *es = sbi->s_es;
 533	struct ext4_new_group_data *group_data = flex_gd->groups;
 534	__u16 *bg_flags = flex_gd->bg_flags;
 535	handle_t *handle;
 536	ext4_group_t group, count;
 537	struct buffer_head *bh = NULL;
 538	int reserved_gdb, i, j, err = 0, err2;
 539	int meta_bg;
 540
 541	BUG_ON(!flex_gd->count || !group_data ||
 542	       group_data[0].group != sbi->s_groups_count);
 543
 544	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 545	meta_bg = ext4_has_feature_meta_bg(sb);
 546
 547	/* This transaction may be extended/restarted along the way */
 548	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 549	if (IS_ERR(handle))
 550		return PTR_ERR(handle);
 551
 552	group = group_data[0].group;
 553	for (i = 0; i < flex_gd->count; i++, group++) {
 554		unsigned long gdblocks;
 555		ext4_grpblk_t overhead;
 556
 557		gdblocks = ext4_bg_num_gdb(sb, group);
 558		start = ext4_group_first_block_no(sb, group);
 559
 560		if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 561			goto handle_itb;
 562
 563		if (meta_bg == 1) {
 564			ext4_group_t first_group;
 565			first_group = ext4_meta_bg_first_group(sb, group);
 566			if (first_group != group + 1 &&
 567			    first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 568				goto handle_itb;
 569		}
 570
 571		block = start + ext4_bg_has_super(sb, group);
 572		/* Copy all of the GDT blocks into the backup in this group */
 573		for (j = 0; j < gdblocks; j++, block++) {
 574			struct buffer_head *gdb;
 575
 576			ext4_debug("update backup group %#04llx\n", block);
 577			err = ext4_resize_ensure_credits_batch(handle, 1);
 578			if (err < 0)
 579				goto out;
 580
 581			gdb = sb_getblk(sb, block);
 582			if (unlikely(!gdb)) {
 583				err = -ENOMEM;
 584				goto out;
 585			}
 586
 587			BUFFER_TRACE(gdb, "get_write_access");
 588			err = ext4_journal_get_write_access(handle, sb, gdb,
 589							    EXT4_JTR_NONE);
 590			if (err) {
 591				brelse(gdb);
 592				goto out;
 593			}
 594			memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
 595				s_group_desc, j)->b_data, gdb->b_size);
 596			set_buffer_uptodate(gdb);
 597
 598			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 599			if (unlikely(err)) {
 600				brelse(gdb);
 601				goto out;
 602			}
 603			brelse(gdb);
 604		}
 605
 606		/* Zero out all of the reserved backup group descriptor
 607		 * table blocks
 608		 */
 609		if (ext4_bg_has_super(sb, group)) {
 610			err = sb_issue_zeroout(sb, gdblocks + start + 1,
 611					reserved_gdb, GFP_NOFS);
 612			if (err)
 613				goto out;
 614		}
 615
 616handle_itb:
 617		/* Initialize group tables of the grop @group */
 618		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 619			goto handle_bb;
 620
 621		/* Zero out all of the inode table blocks */
 622		block = group_data[i].inode_table;
 623		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 624			   block, sbi->s_itb_per_group);
 625		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 626				       GFP_NOFS);
 627		if (err)
 628			goto out;
 629
 630handle_bb:
 631		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 632			goto handle_ib;
 633
 634		/* Initialize block bitmap of the @group */
 635		block = group_data[i].block_bitmap;
 636		err = ext4_resize_ensure_credits_batch(handle, 1);
 637		if (err < 0)
 638			goto out;
 639
 640		bh = bclean(handle, sb, block);
 641		if (IS_ERR(bh)) {
 642			err = PTR_ERR(bh);
 643			goto out;
 644		}
 645		overhead = ext4_group_overhead_blocks(sb, group);
 646		if (overhead != 0) {
 647			ext4_debug("mark backup superblock %#04llx (+0)\n",
 648				   start);
 649			mb_set_bits(bh->b_data, 0,
 650				      EXT4_NUM_B2C(sbi, overhead));
 651		}
 652		ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
 653				     sb->s_blocksize * 8, bh->b_data);
 654		err = ext4_handle_dirty_metadata(handle, NULL, bh);
 655		brelse(bh);
 656		if (err)
 657			goto out;
 658
 659handle_ib:
 660		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 661			continue;
 662
 663		/* Initialize inode bitmap of the @group */
 664		block = group_data[i].inode_bitmap;
 665		err = ext4_resize_ensure_credits_batch(handle, 1);
 666		if (err < 0)
 667			goto out;
 668		/* Mark unused entries in inode bitmap used */
 669		bh = bclean(handle, sb, block);
 670		if (IS_ERR(bh)) {
 671			err = PTR_ERR(bh);
 672			goto out;
 673		}
 674
 675		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 676				     sb->s_blocksize * 8, bh->b_data);
 677		err = ext4_handle_dirty_metadata(handle, NULL, bh);
 678		brelse(bh);
 679		if (err)
 680			goto out;
 681	}
 682
 683	/* Mark group tables in block bitmap */
 684	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 685		count = group_table_count[j];
 686		start = (&group_data[0].block_bitmap)[j];
 687		block = start;
 688		for (i = 1; i < flex_gd->count; i++) {
 689			block += group_table_count[j];
 690			if (block == (&group_data[i].block_bitmap)[j]) {
 691				count += group_table_count[j];
 692				continue;
 693			}
 694			err = set_flexbg_block_bitmap(sb, handle,
 695						      flex_gd,
 696						      EXT4_B2C(sbi, start),
 697						      EXT4_B2C(sbi,
 698							       start + count
 699							       - 1));
 700			if (err)
 701				goto out;
 702			count = group_table_count[j];
 703			start = (&group_data[i].block_bitmap)[j];
 704			block = start;
 705		}
 706
 707		if (count) {
 708			err = set_flexbg_block_bitmap(sb, handle,
 709						      flex_gd,
 710						      EXT4_B2C(sbi, start),
 711						      EXT4_B2C(sbi,
 712							       start + count
 713							       - 1));
 714			if (err)
 715				goto out;
 716		}
 717	}
 718
 719out:
 720	err2 = ext4_journal_stop(handle);
 721	if (err2 && !err)
 722		err = err2;
 723
 724	return err;
 725}
 726
 727/*
 728 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 729 * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 730 * calling this for the first time.  In a sparse filesystem it will be the
 731 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 732 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 733 */
 734unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three,
 735			       unsigned int *five, unsigned int *seven)
 736{
 737	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 738	unsigned int *min = three;
 739	int mult = 3;
 740	unsigned int ret;
 741
 742	if (ext4_has_feature_sparse_super2(sb)) {
 743		do {
 744			if (*min > 2)
 745				return UINT_MAX;
 746			ret = le32_to_cpu(es->s_backup_bgs[*min - 1]);
 747			*min += 1;
 748		} while (!ret);
 749		return ret;
 750	}
 751
 752	if (!ext4_has_feature_sparse_super(sb)) {
 753		ret = *min;
 754		*min += 1;
 755		return ret;
 756	}
 757
 758	if (*five < *min) {
 759		min = five;
 760		mult = 5;
 761	}
 762	if (*seven < *min) {
 763		min = seven;
 764		mult = 7;
 765	}
 766
 767	ret = *min;
 768	*min *= mult;
 769
 770	return ret;
 771}
 772
 773/*
 774 * Check that all of the backup GDT blocks are held in the primary GDT block.
 775 * It is assumed that they are stored in group order.  Returns the number of
 776 * groups in current filesystem that have BACKUPS, or -ve error code.
 777 */
 778static int verify_reserved_gdb(struct super_block *sb,
 779			       ext4_group_t end,
 780			       struct buffer_head *primary)
 781{
 782	const ext4_fsblk_t blk = primary->b_blocknr;
 783	unsigned three = 1;
 784	unsigned five = 5;
 785	unsigned seven = 7;
 786	unsigned grp;
 787	__le32 *p = (__le32 *)primary->b_data;
 788	int gdbackups = 0;
 789
 790	while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 791		if (le32_to_cpu(*p++) !=
 792		    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 793			ext4_warning(sb, "reserved GDT %llu"
 794				     " missing grp %d (%llu)",
 795				     blk, grp,
 796				     grp *
 797				     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 798				     blk);
 799			return -EINVAL;
 800		}
 801		if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 802			return -EFBIG;
 803	}
 804
 805	return gdbackups;
 806}
 807
 808/*
 809 * Called when we need to bring a reserved group descriptor table block into
 810 * use from the resize inode.  The primary copy of the new GDT block currently
 811 * is an indirect block (under the double indirect block in the resize inode).
 812 * The new backup GDT blocks will be stored as leaf blocks in this indirect
 813 * block, in group order.  Even though we know all the block numbers we need,
 814 * we check to ensure that the resize inode has actually reserved these blocks.
 815 *
 816 * Don't need to update the block bitmaps because the blocks are still in use.
 817 *
 818 * We get all of the error cases out of the way, so that we are sure to not
 819 * fail once we start modifying the data on disk, because JBD has no rollback.
 820 */
 821static int add_new_gdb(handle_t *handle, struct inode *inode,
 822		       ext4_group_t group)
 823{
 824	struct super_block *sb = inode->i_sb;
 825	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 826	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 827	ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 828	struct buffer_head **o_group_desc, **n_group_desc = NULL;
 829	struct buffer_head *dind = NULL;
 830	struct buffer_head *gdb_bh = NULL;
 831	int gdbackups;
 832	struct ext4_iloc iloc = { .bh = NULL };
 833	__le32 *data;
 834	int err;
 835
 836	if (test_opt(sb, DEBUG))
 837		printk(KERN_DEBUG
 838		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 839		       gdb_num);
 840
 841	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 842	if (IS_ERR(gdb_bh))
 843		return PTR_ERR(gdb_bh);
 844
 845	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 846	if (gdbackups < 0) {
 847		err = gdbackups;
 848		goto errout;
 849	}
 850
 851	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 852	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
 853	if (IS_ERR(dind)) {
 854		err = PTR_ERR(dind);
 855		dind = NULL;
 856		goto errout;
 857	}
 858
 859	data = (__le32 *)dind->b_data;
 860	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 861		ext4_warning(sb, "new group %u GDT block %llu not reserved",
 862			     group, gdblock);
 863		err = -EINVAL;
 864		goto errout;
 865	}
 866
 867	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 868	err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
 869					    EXT4_JTR_NONE);
 870	if (unlikely(err))
 871		goto errout;
 872
 873	BUFFER_TRACE(gdb_bh, "get_write_access");
 874	err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
 875	if (unlikely(err))
 876		goto errout;
 877
 878	BUFFER_TRACE(dind, "get_write_access");
 879	err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
 880	if (unlikely(err)) {
 881		ext4_std_error(sb, err);
 882		goto errout;
 883	}
 884
 885	/* ext4_reserve_inode_write() gets a reference on the iloc */
 886	err = ext4_reserve_inode_write(handle, inode, &iloc);
 887	if (unlikely(err))
 888		goto errout;
 889
 890	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 891				GFP_KERNEL);
 892	if (!n_group_desc) {
 893		err = -ENOMEM;
 894		ext4_warning(sb, "not enough memory for %lu groups",
 895			     gdb_num + 1);
 896		goto errout;
 897	}
 898
 899	/*
 900	 * Finally, we have all of the possible failures behind us...
 901	 *
 902	 * Remove new GDT block from inode double-indirect block and clear out
 903	 * the new GDT block for use (which also "frees" the backup GDT blocks
 904	 * from the reserved inode).  We don't need to change the bitmaps for
 905	 * these blocks, because they are marked as in-use from being in the
 906	 * reserved inode, and will become GDT blocks (primary and backup).
 907	 */
 908	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 909	err = ext4_handle_dirty_metadata(handle, NULL, dind);
 910	if (unlikely(err)) {
 911		ext4_std_error(sb, err);
 912		goto errout;
 913	}
 914	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
 915			   (9 - EXT4_SB(sb)->s_cluster_bits);
 916	ext4_mark_iloc_dirty(handle, inode, &iloc);
 917	memset(gdb_bh->b_data, 0, sb->s_blocksize);
 918	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 919	if (unlikely(err)) {
 920		ext4_std_error(sb, err);
 921		iloc.bh = NULL;
 922		goto errout;
 923	}
 924	brelse(dind);
 925
 926	rcu_read_lock();
 927	o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
 928	memcpy(n_group_desc, o_group_desc,
 929	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 930	rcu_read_unlock();
 931	n_group_desc[gdb_num] = gdb_bh;
 932	rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
 933	EXT4_SB(sb)->s_gdb_count++;
 934	ext4_kvfree_array_rcu(o_group_desc);
 935
 936	lock_buffer(EXT4_SB(sb)->s_sbh);
 937	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 938	ext4_superblock_csum_set(sb);
 939	unlock_buffer(EXT4_SB(sb)->s_sbh);
 940	err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 941	if (err)
 942		ext4_std_error(sb, err);
 943	return err;
 944errout:
 945	kvfree(n_group_desc);
 946	brelse(iloc.bh);
 947	brelse(dind);
 948	brelse(gdb_bh);
 949
 950	ext4_debug("leaving with error %d\n", err);
 951	return err;
 952}
 953
 954/*
 955 * add_new_gdb_meta_bg is the sister of add_new_gdb.
 
 
 
 
 
 
 956 */
 957static int add_new_gdb_meta_bg(struct super_block *sb,
 958			       handle_t *handle, ext4_group_t group) {
 959	ext4_fsblk_t gdblock;
 960	struct buffer_head *gdb_bh;
 961	struct buffer_head **o_group_desc, **n_group_desc;
 962	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 963	int err;
 964
 965	gdblock = ext4_meta_bg_first_block_no(sb, group) +
 966		   ext4_bg_has_super(sb, group);
 967	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 968	if (IS_ERR(gdb_bh))
 969		return PTR_ERR(gdb_bh);
 970	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 971				GFP_KERNEL);
 972	if (!n_group_desc) {
 973		brelse(gdb_bh);
 974		err = -ENOMEM;
 975		ext4_warning(sb, "not enough memory for %lu groups",
 976			     gdb_num + 1);
 977		return err;
 978	}
 979
 980	rcu_read_lock();
 981	o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
 982	memcpy(n_group_desc, o_group_desc,
 983	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 984	rcu_read_unlock();
 985	n_group_desc[gdb_num] = gdb_bh;
 986
 987	BUFFER_TRACE(gdb_bh, "get_write_access");
 988	err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
 989	if (err) {
 990		kvfree(n_group_desc);
 991		brelse(gdb_bh);
 992		return err;
 993	}
 994
 995	rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
 996	EXT4_SB(sb)->s_gdb_count++;
 997	ext4_kvfree_array_rcu(o_group_desc);
 998	return err;
 999}
1000
1001/*
1002 * Called when we are adding a new group which has a backup copy of each of
1003 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
1004 * We need to add these reserved backup GDT blocks to the resize inode, so
1005 * that they are kept for future resizing and not allocated to files.
1006 *
1007 * Each reserved backup GDT block will go into a different indirect block.
1008 * The indirect blocks are actually the primary reserved GDT blocks,
1009 * so we know in advance what their block numbers are.  We only get the
1010 * double-indirect block to verify it is pointing to the primary reserved
1011 * GDT blocks so we don't overwrite a data block by accident.  The reserved
1012 * backup GDT blocks are stored in their reserved primary GDT block.
1013 */
1014static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1015			      ext4_group_t group)
1016{
1017	struct super_block *sb = inode->i_sb;
1018	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1019	int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1020	struct buffer_head **primary;
1021	struct buffer_head *dind;
1022	struct ext4_iloc iloc;
1023	ext4_fsblk_t blk;
1024	__le32 *data, *end;
1025	int gdbackups = 0;
1026	int res, i;
1027	int err;
1028
1029	primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1030	if (!primary)
1031		return -ENOMEM;
1032
1033	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1034	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1035	if (IS_ERR(dind)) {
1036		err = PTR_ERR(dind);
1037		dind = NULL;
1038		goto exit_free;
1039	}
1040
1041	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1042	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1043					 EXT4_ADDR_PER_BLOCK(sb));
1044	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1045
1046	/* Get each reserved primary GDT block and verify it holds backups */
1047	for (res = 0; res < reserved_gdb; res++, blk++) {
1048		if (le32_to_cpu(*data) != blk) {
1049			ext4_warning(sb, "reserved block %llu"
1050				     " not at offset %ld",
1051				     blk,
1052				     (long)(data - (__le32 *)dind->b_data));
1053			err = -EINVAL;
1054			goto exit_bh;
1055		}
1056		primary[res] = ext4_sb_bread(sb, blk, 0);
1057		if (IS_ERR(primary[res])) {
1058			err = PTR_ERR(primary[res]);
1059			primary[res] = NULL;
1060			goto exit_bh;
1061		}
1062		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1063		if (gdbackups < 0) {
1064			brelse(primary[res]);
1065			err = gdbackups;
1066			goto exit_bh;
1067		}
1068		if (++data >= end)
1069			data = (__le32 *)dind->b_data;
1070	}
1071
1072	for (i = 0; i < reserved_gdb; i++) {
1073		BUFFER_TRACE(primary[i], "get_write_access");
1074		if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1075							 EXT4_JTR_NONE)))
1076			goto exit_bh;
1077	}
1078
1079	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1080		goto exit_bh;
1081
1082	/*
1083	 * Finally we can add each of the reserved backup GDT blocks from
1084	 * the new group to its reserved primary GDT block.
1085	 */
1086	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1087	for (i = 0; i < reserved_gdb; i++) {
1088		int err2;
1089		data = (__le32 *)primary[i]->b_data;
1090		/* printk("reserving backup %lu[%u] = %lu\n",
1091		       primary[i]->b_blocknr, gdbackups,
1092		       blk + primary[i]->b_blocknr); */
1093		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1094		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1095		if (!err)
1096			err = err2;
1097	}
1098
1099	inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1100	ext4_mark_iloc_dirty(handle, inode, &iloc);
1101
1102exit_bh:
1103	while (--res >= 0)
1104		brelse(primary[res]);
1105	brelse(dind);
1106
1107exit_free:
1108	kfree(primary);
1109
1110	return err;
1111}
1112
1113static inline void ext4_set_block_group_nr(struct super_block *sb, char *data,
1114					   ext4_group_t group)
1115{
1116	struct ext4_super_block *es = (struct ext4_super_block *) data;
1117
1118	es->s_block_group_nr = cpu_to_le16(group);
1119	if (ext4_has_metadata_csum(sb))
1120		es->s_checksum = ext4_superblock_csum(sb, es);
1121}
1122
1123/*
1124 * Update the backup copies of the ext4 metadata.  These don't need to be part
1125 * of the main resize transaction, because e2fsck will re-write them if there
1126 * is a problem (basically only OOM will cause a problem).  However, we
1127 * _should_ update the backups if possible, in case the primary gets trashed
1128 * for some reason and we need to run e2fsck from a backup superblock.  The
1129 * important part is that the new block and inode counts are in the backup
1130 * superblocks, and the location of the new group metadata in the GDT backups.
1131 *
1132 * We do not need take the s_resize_lock for this, because these
1133 * blocks are not otherwise touched by the filesystem code when it is
1134 * mounted.  We don't need to worry about last changing from
1135 * sbi->s_groups_count, because the worst that can happen is that we
1136 * do not copy the full number of backups at this time.  The resize
1137 * which changed s_groups_count will backup again.
1138 */
1139static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1140			   int size, int meta_bg)
1141{
1142	struct ext4_sb_info *sbi = EXT4_SB(sb);
1143	ext4_group_t last;
1144	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1145	unsigned three = 1;
1146	unsigned five = 5;
1147	unsigned seven = 7;
1148	ext4_group_t group = 0;
1149	int rest = sb->s_blocksize - size;
1150	handle_t *handle;
1151	int err = 0, err2;
1152
1153	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1154	if (IS_ERR(handle)) {
1155		group = 1;
1156		err = PTR_ERR(handle);
1157		goto exit_err;
1158	}
1159
1160	if (meta_bg == 0) {
1161		group = ext4_list_backups(sb, &three, &five, &seven);
1162		last = sbi->s_groups_count;
1163	} else {
1164		group = ext4_get_group_number(sb, blk_off) + 1;
1165		last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1166	}
1167
1168	while (group < sbi->s_groups_count) {
1169		struct buffer_head *bh;
1170		ext4_fsblk_t backup_block;
1171		int has_super = ext4_bg_has_super(sb, group);
1172		ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group);
1173
1174		/* Out of journal space, and can't get more - abort - so sad */
1175		err = ext4_resize_ensure_credits_batch(handle, 1);
1176		if (err < 0)
1177			break;
1178
1179		if (meta_bg == 0)
1180			backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1181		else
1182			backup_block = first_block + has_super;
1183
1184		bh = sb_getblk(sb, backup_block);
1185		if (unlikely(!bh)) {
1186			err = -ENOMEM;
1187			break;
1188		}
1189		ext4_debug("update metadata backup %llu(+%llu)\n",
1190			   backup_block, backup_block -
1191			   ext4_group_first_block_no(sb, group));
1192		BUFFER_TRACE(bh, "get_write_access");
1193		if ((err = ext4_journal_get_write_access(handle, sb, bh,
1194							 EXT4_JTR_NONE)))
 
1195			break;
 
1196		lock_buffer(bh);
1197		memcpy(bh->b_data, data, size);
1198		if (rest)
1199			memset(bh->b_data + size, 0, rest);
1200		if (has_super && (backup_block == first_block))
1201			ext4_set_block_group_nr(sb, bh->b_data, group);
1202		set_buffer_uptodate(bh);
1203		unlock_buffer(bh);
1204		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1205		if (unlikely(err))
1206			ext4_std_error(sb, err);
1207		brelse(bh);
1208
1209		if (meta_bg == 0)
1210			group = ext4_list_backups(sb, &three, &five, &seven);
1211		else if (group == last)
1212			break;
1213		else
1214			group = last;
1215	}
1216	if ((err2 = ext4_journal_stop(handle)) && !err)
1217		err = err2;
1218
1219	/*
1220	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1221	 * late to revert the resize, we shouldn't fail just because of
1222	 * the backup copies (they are only needed in case of corruption).
1223	 *
1224	 * However, if we got here we have a journal problem too, so we
1225	 * can't really start a transaction to mark the superblock.
1226	 * Chicken out and just set the flag on the hope it will be written
1227	 * to disk, and if not - we will simply wait until next fsck.
1228	 */
1229exit_err:
1230	if (err) {
1231		ext4_warning(sb, "can't update backup for group %u (err %d), "
1232			     "forcing fsck on next reboot", group, err);
1233		sbi->s_mount_state &= ~EXT4_VALID_FS;
1234		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1235		mark_buffer_dirty(sbi->s_sbh);
1236	}
1237}
1238
1239/*
1240 * ext4_add_new_descs() adds @count group descriptor of groups
1241 * starting at @group
1242 *
1243 * @handle: journal handle
1244 * @sb: super block
1245 * @group: the group no. of the first group desc to be added
1246 * @resize_inode: the resize inode
1247 * @count: number of group descriptors to be added
1248 */
1249static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1250			      ext4_group_t group, struct inode *resize_inode,
1251			      ext4_group_t count)
1252{
1253	struct ext4_sb_info *sbi = EXT4_SB(sb);
1254	struct ext4_super_block *es = sbi->s_es;
1255	struct buffer_head *gdb_bh;
1256	int i, gdb_off, gdb_num, err = 0;
1257	int meta_bg;
1258
1259	meta_bg = ext4_has_feature_meta_bg(sb);
1260	for (i = 0; i < count; i++, group++) {
1261		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1262			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1263
1264		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1265		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1266
1267		/*
1268		 * We will only either add reserved group blocks to a backup group
1269		 * or remove reserved blocks for the first group in a new group block.
1270		 * Doing both would be mean more complex code, and sane people don't
1271		 * use non-sparse filesystems anymore.  This is already checked above.
1272		 */
1273		if (gdb_off) {
1274			gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1275						     gdb_num);
1276			BUFFER_TRACE(gdb_bh, "get_write_access");
1277			err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1278							    EXT4_JTR_NONE);
1279
1280			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1281				err = reserve_backup_gdb(handle, resize_inode, group);
1282		} else if (meta_bg != 0) {
1283			err = add_new_gdb_meta_bg(sb, handle, group);
1284		} else {
1285			err = add_new_gdb(handle, resize_inode, group);
1286		}
1287		if (err)
1288			break;
1289	}
1290	return err;
1291}
1292
1293static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1294{
1295	struct buffer_head *bh = sb_getblk(sb, block);
1296	if (unlikely(!bh))
1297		return NULL;
1298	if (!bh_uptodate_or_lock(bh)) {
1299		if (ext4_read_bh(bh, 0, NULL) < 0) {
1300			brelse(bh);
1301			return NULL;
1302		}
1303	}
1304
1305	return bh;
1306}
1307
1308static int ext4_set_bitmap_checksums(struct super_block *sb,
1309				     ext4_group_t group,
1310				     struct ext4_group_desc *gdp,
1311				     struct ext4_new_group_data *group_data)
1312{
1313	struct buffer_head *bh;
1314
1315	if (!ext4_has_metadata_csum(sb))
1316		return 0;
1317
1318	bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1319	if (!bh)
1320		return -EIO;
1321	ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1322				   EXT4_INODES_PER_GROUP(sb) / 8);
1323	brelse(bh);
1324
1325	bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1326	if (!bh)
1327		return -EIO;
1328	ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1329	brelse(bh);
1330
1331	return 0;
1332}
1333
1334/*
1335 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1336 */
1337static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1338				struct ext4_new_flex_group_data *flex_gd)
1339{
1340	struct ext4_new_group_data	*group_data = flex_gd->groups;
1341	struct ext4_group_desc		*gdp;
1342	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1343	struct buffer_head		*gdb_bh;
1344	ext4_group_t			group;
1345	__u16				*bg_flags = flex_gd->bg_flags;
1346	int				i, gdb_off, gdb_num, err = 0;
1347
1348
1349	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1350		group = group_data->group;
1351
1352		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1353		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1354
1355		/*
1356		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1357		 */
1358		gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1359		/* Update group descriptor block for new group */
1360		gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1361						 gdb_off * EXT4_DESC_SIZE(sb));
1362
1363		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1364		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1365		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1366		err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1367		if (err) {
1368			ext4_std_error(sb, err);
1369			break;
1370		}
1371
1372		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1373		ext4_free_group_clusters_set(sb, gdp,
1374					     group_data->free_clusters_count);
1375		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1376		if (ext4_has_group_desc_csum(sb))
1377			ext4_itable_unused_set(sb, gdp,
1378					       EXT4_INODES_PER_GROUP(sb));
1379		gdp->bg_flags = cpu_to_le16(*bg_flags);
1380		ext4_group_desc_csum_set(sb, group, gdp);
1381
1382		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1383		if (unlikely(err)) {
1384			ext4_std_error(sb, err);
1385			break;
1386		}
1387
1388		/*
1389		 * We can allocate memory for mb_alloc based on the new group
1390		 * descriptor
1391		 */
1392		err = ext4_mb_add_groupinfo(sb, group, gdp);
1393		if (err)
1394			break;
1395	}
1396	return err;
1397}
1398
1399static void ext4_add_overhead(struct super_block *sb,
1400                              const ext4_fsblk_t overhead)
1401{
1402       struct ext4_sb_info *sbi = EXT4_SB(sb);
1403       struct ext4_super_block *es = sbi->s_es;
1404
1405       sbi->s_overhead += overhead;
1406       es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1407       smp_wmb();
1408}
1409
1410/*
1411 * ext4_update_super() updates the super block so that the newly added
1412 * groups can be seen by the filesystem.
1413 *
1414 * @sb: super block
1415 * @flex_gd: new added groups
1416 */
1417static void ext4_update_super(struct super_block *sb,
1418			     struct ext4_new_flex_group_data *flex_gd)
1419{
1420	ext4_fsblk_t blocks_count = 0;
1421	ext4_fsblk_t free_blocks = 0;
1422	ext4_fsblk_t reserved_blocks = 0;
1423	struct ext4_new_group_data *group_data = flex_gd->groups;
1424	struct ext4_sb_info *sbi = EXT4_SB(sb);
1425	struct ext4_super_block *es = sbi->s_es;
1426	int i;
1427
1428	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1429	/*
1430	 * Make the new blocks and inodes valid next.  We do this before
1431	 * increasing the group count so that once the group is enabled,
1432	 * all of its blocks and inodes are already valid.
1433	 *
1434	 * We always allocate group-by-group, then block-by-block or
1435	 * inode-by-inode within a group, so enabling these
1436	 * blocks/inodes before the group is live won't actually let us
1437	 * allocate the new space yet.
1438	 */
1439	for (i = 0; i < flex_gd->count; i++) {
1440		blocks_count += group_data[i].blocks_count;
1441		free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1442	}
1443
1444	reserved_blocks = ext4_r_blocks_count(es) * 100;
1445	reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1446	reserved_blocks *= blocks_count;
1447	do_div(reserved_blocks, 100);
1448
1449	lock_buffer(sbi->s_sbh);
1450	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1451	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1452	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1453		     flex_gd->count);
1454	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1455		     flex_gd->count);
1456
1457	ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1458	/*
1459	 * We need to protect s_groups_count against other CPUs seeing
1460	 * inconsistent state in the superblock.
1461	 *
1462	 * The precise rules we use are:
1463	 *
1464	 * * Writers must perform a smp_wmb() after updating all
1465	 *   dependent data and before modifying the groups count
1466	 *
1467	 * * Readers must perform an smp_rmb() after reading the groups
1468	 *   count and before reading any dependent data.
1469	 *
1470	 * NB. These rules can be relaxed when checking the group count
1471	 * while freeing data, as we can only allocate from a block
1472	 * group after serialising against the group count, and we can
1473	 * only then free after serialising in turn against that
1474	 * allocation.
1475	 */
1476	smp_wmb();
1477
1478	/* Update the global fs size fields */
1479	sbi->s_groups_count += flex_gd->count;
1480	sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1481			(EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1482
1483	/* Update the reserved block counts only once the new group is
1484	 * active. */
1485	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1486				reserved_blocks);
1487
1488	/* Update the free space counts */
1489	percpu_counter_add(&sbi->s_freeclusters_counter,
1490			   EXT4_NUM_B2C(sbi, free_blocks));
1491	percpu_counter_add(&sbi->s_freeinodes_counter,
1492			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1493
1494	ext4_debug("free blocks count %llu",
1495		   percpu_counter_read(&sbi->s_freeclusters_counter));
1496	if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1497		ext4_group_t flex_group;
1498		struct flex_groups *fg;
1499
1500		flex_group = ext4_flex_group(sbi, group_data[0].group);
1501		fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1502		atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1503			     &fg->free_clusters);
1504		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1505			   &fg->free_inodes);
1506	}
1507
1508	/*
1509	 * Update the fs overhead information.
1510	 *
1511	 * For bigalloc, if the superblock already has a properly calculated
1512	 * overhead, update it with a value based on numbers already computed
1513	 * above for the newly allocated capacity.
1514	 */
1515	if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0))
1516		ext4_add_overhead(sb,
1517			EXT4_NUM_B2C(sbi, blocks_count - free_blocks));
1518	else
1519		ext4_calculate_overhead(sb);
1520	es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1521
1522	ext4_superblock_csum_set(sb);
1523	unlock_buffer(sbi->s_sbh);
1524	if (test_opt(sb, DEBUG))
1525		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1526		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1527		       blocks_count, free_blocks, reserved_blocks);
1528}
1529
1530/* Add a flex group to an fs. Ensure we handle all possible error conditions
1531 * _before_ we start modifying the filesystem, because we cannot abort the
1532 * transaction and not have it write the data to disk.
1533 */
1534static int ext4_flex_group_add(struct super_block *sb,
1535			       struct inode *resize_inode,
1536			       struct ext4_new_flex_group_data *flex_gd)
1537{
1538	struct ext4_sb_info *sbi = EXT4_SB(sb);
1539	struct ext4_super_block *es = sbi->s_es;
1540	ext4_fsblk_t o_blocks_count;
1541	ext4_grpblk_t last;
1542	ext4_group_t group;
1543	handle_t *handle;
1544	unsigned reserved_gdb;
1545	int err = 0, err2 = 0, credit;
1546
1547	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1548
1549	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1550	o_blocks_count = ext4_blocks_count(es);
1551	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1552	BUG_ON(last);
1553
1554	err = setup_new_flex_group_blocks(sb, flex_gd);
1555	if (err)
1556		goto exit;
1557	/*
1558	 * We will always be modifying at least the superblock and  GDT
1559	 * blocks.  If we are adding a group past the last current GDT block,
1560	 * we will also modify the inode and the dindirect block.  If we
1561	 * are adding a group with superblock/GDT backups  we will also
1562	 * modify each of the reserved GDT dindirect blocks.
1563	 */
1564	credit = 3;	/* sb, resize inode, resize inode dindirect */
1565	/* GDT blocks */
1566	credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1567	credit += reserved_gdb;	/* Reserved GDT dindirect blocks */
1568	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1569	if (IS_ERR(handle)) {
1570		err = PTR_ERR(handle);
1571		goto exit;
1572	}
1573
1574	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1575	err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1576					    EXT4_JTR_NONE);
1577	if (err)
1578		goto exit_journal;
1579
1580	group = flex_gd->groups[0].group;
1581	BUG_ON(group != sbi->s_groups_count);
1582	err = ext4_add_new_descs(handle, sb, group,
1583				resize_inode, flex_gd->count);
1584	if (err)
1585		goto exit_journal;
1586
1587	err = ext4_setup_new_descs(handle, sb, flex_gd);
1588	if (err)
1589		goto exit_journal;
1590
1591	ext4_update_super(sb, flex_gd);
1592
1593	err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1594
1595exit_journal:
1596	err2 = ext4_journal_stop(handle);
1597	if (!err)
1598		err = err2;
1599
1600	if (!err) {
1601		int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1602		int gdb_num_end = ((group + flex_gd->count - 1) /
1603				   EXT4_DESC_PER_BLOCK(sb));
1604		int meta_bg = ext4_has_feature_meta_bg(sb);
1605		sector_t old_gdb = 0;
 
 
1606
1607		update_backups(sb, ext4_group_first_block_no(sb, 0),
1608			       (char *)es, sizeof(struct ext4_super_block), 0);
1609		for (; gdb_num <= gdb_num_end; gdb_num++) {
1610			struct buffer_head *gdb_bh;
1611
1612			gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1613						     gdb_num);
1614			if (old_gdb == gdb_bh->b_blocknr)
1615				continue;
1616			update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1617				       gdb_bh->b_size, meta_bg);
1618			old_gdb = gdb_bh->b_blocknr;
1619		}
1620	}
1621exit:
1622	return err;
1623}
1624
1625static int ext4_setup_next_flex_gd(struct super_block *sb,
1626				    struct ext4_new_flex_group_data *flex_gd,
1627				    ext4_fsblk_t n_blocks_count,
1628				    unsigned long flexbg_size)
1629{
1630	struct ext4_sb_info *sbi = EXT4_SB(sb);
1631	struct ext4_super_block *es = sbi->s_es;
1632	struct ext4_new_group_data *group_data = flex_gd->groups;
1633	ext4_fsblk_t o_blocks_count;
1634	ext4_group_t n_group;
1635	ext4_group_t group;
1636	ext4_group_t last_group;
1637	ext4_grpblk_t last;
1638	ext4_grpblk_t clusters_per_group;
1639	unsigned long i;
1640
1641	clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1642
1643	o_blocks_count = ext4_blocks_count(es);
1644
1645	if (o_blocks_count == n_blocks_count)
1646		return 0;
1647
1648	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1649	BUG_ON(last);
1650	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1651
1652	last_group = group | (flexbg_size - 1);
1653	if (last_group > n_group)
1654		last_group = n_group;
1655
1656	flex_gd->count = last_group - group + 1;
1657
1658	for (i = 0; i < flex_gd->count; i++) {
1659		int overhead;
1660
1661		group_data[i].group = group + i;
1662		group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1663		overhead = ext4_group_overhead_blocks(sb, group + i);
1664		group_data[i].mdata_blocks = overhead;
1665		group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1666		if (ext4_has_group_desc_csum(sb)) {
1667			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1668					       EXT4_BG_INODE_UNINIT;
1669			if (!test_opt(sb, INIT_INODE_TABLE))
1670				flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1671		} else
1672			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1673	}
1674
1675	if (last_group == n_group && ext4_has_group_desc_csum(sb))
1676		/* We need to initialize block bitmap of last group. */
1677		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1678
1679	if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1680		group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1681		group_data[i - 1].free_clusters_count -= clusters_per_group -
1682						       last - 1;
1683	}
1684
1685	return 1;
1686}
1687
1688/* Add group descriptor data to an existing or new group descriptor block.
1689 * Ensure we handle all possible error conditions _before_ we start modifying
1690 * the filesystem, because we cannot abort the transaction and not have it
1691 * write the data to disk.
1692 *
1693 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1694 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1695 *
1696 * We only need to hold the superblock lock while we are actually adding
1697 * in the new group's counts to the superblock.  Prior to that we have
1698 * not really "added" the group at all.  We re-check that we are still
1699 * adding in the last group in case things have changed since verifying.
1700 */
1701int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1702{
1703	struct ext4_new_flex_group_data flex_gd;
1704	struct ext4_sb_info *sbi = EXT4_SB(sb);
1705	struct ext4_super_block *es = sbi->s_es;
1706	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1707		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1708	struct inode *inode = NULL;
1709	int gdb_off;
1710	int err;
1711	__u16 bg_flags = 0;
1712
1713	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1714
1715	if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1716		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1717		return -EPERM;
1718	}
1719
1720	if (ext4_blocks_count(es) + input->blocks_count <
1721	    ext4_blocks_count(es)) {
1722		ext4_warning(sb, "blocks_count overflow");
1723		return -EINVAL;
1724	}
1725
1726	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1727	    le32_to_cpu(es->s_inodes_count)) {
1728		ext4_warning(sb, "inodes_count overflow");
1729		return -EINVAL;
1730	}
1731
1732	if (reserved_gdb || gdb_off == 0) {
1733		if (!ext4_has_feature_resize_inode(sb) ||
1734		    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1735			ext4_warning(sb,
1736				     "No reserved GDT blocks, can't resize");
1737			return -EPERM;
1738		}
1739		inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1740		if (IS_ERR(inode)) {
1741			ext4_warning(sb, "Error opening resize inode");
1742			return PTR_ERR(inode);
1743		}
1744	}
1745
1746
1747	err = verify_group_input(sb, input);
1748	if (err)
1749		goto out;
1750
1751	err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1752	if (err)
1753		goto out;
1754
1755	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1756	if (err)
1757		goto out;
1758
1759	flex_gd.count = 1;
1760	flex_gd.groups = input;
1761	flex_gd.bg_flags = &bg_flags;
1762	err = ext4_flex_group_add(sb, inode, &flex_gd);
1763out:
1764	iput(inode);
1765	return err;
1766} /* ext4_group_add */
1767
1768/*
1769 * extend a group without checking assuming that checking has been done.
1770 */
1771static int ext4_group_extend_no_check(struct super_block *sb,
1772				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1773{
1774	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1775	handle_t *handle;
1776	int err = 0, err2;
1777
1778	/* We will update the superblock, one block bitmap, and
1779	 * one group descriptor via ext4_group_add_blocks().
1780	 */
1781	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1782	if (IS_ERR(handle)) {
1783		err = PTR_ERR(handle);
1784		ext4_warning(sb, "error %d on journal start", err);
1785		return err;
1786	}
1787
1788	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1789	err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1790					    EXT4_JTR_NONE);
1791	if (err) {
1792		ext4_warning(sb, "error %d on journal write access", err);
1793		goto errout;
1794	}
1795
1796	lock_buffer(EXT4_SB(sb)->s_sbh);
1797	ext4_blocks_count_set(es, o_blocks_count + add);
1798	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1799	ext4_superblock_csum_set(sb);
1800	unlock_buffer(EXT4_SB(sb)->s_sbh);
1801	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1802		   o_blocks_count + add);
1803	/* We add the blocks to the bitmap and set the group need init bit */
1804	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1805	if (err)
1806		goto errout;
1807	ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1808	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1809		   o_blocks_count + add);
1810errout:
1811	err2 = ext4_journal_stop(handle);
1812	if (err2 && !err)
1813		err = err2;
1814
1815	if (!err) {
1816		if (test_opt(sb, DEBUG))
1817			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1818			       "blocks\n", ext4_blocks_count(es));
1819		update_backups(sb, ext4_group_first_block_no(sb, 0),
1820			       (char *)es, sizeof(struct ext4_super_block), 0);
1821	}
1822	return err;
1823}
1824
1825/*
1826 * Extend the filesystem to the new number of blocks specified.  This entry
1827 * point is only used to extend the current filesystem to the end of the last
1828 * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1829 * for emergencies (because it has no dependencies on reserved blocks).
1830 *
1831 * If we _really_ wanted, we could use default values to call ext4_group_add()
1832 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1833 * GDT blocks are reserved to grow to the desired size.
1834 */
1835int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1836		      ext4_fsblk_t n_blocks_count)
1837{
1838	ext4_fsblk_t o_blocks_count;
1839	ext4_grpblk_t last;
1840	ext4_grpblk_t add;
1841	struct buffer_head *bh;
1842	ext4_group_t group;
1843
1844	o_blocks_count = ext4_blocks_count(es);
1845
1846	if (test_opt(sb, DEBUG))
1847		ext4_msg(sb, KERN_DEBUG,
1848			 "extending last group from %llu to %llu blocks",
1849			 o_blocks_count, n_blocks_count);
1850
1851	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1852		return 0;
1853
1854	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1855		ext4_msg(sb, KERN_ERR,
1856			 "filesystem too large to resize to %llu blocks safely",
1857			 n_blocks_count);
1858		return -EINVAL;
1859	}
1860
1861	if (n_blocks_count < o_blocks_count) {
1862		ext4_warning(sb, "can't shrink FS - resize aborted");
1863		return -EINVAL;
1864	}
1865
1866	/* Handle the remaining blocks in the last group only. */
1867	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1868
1869	if (last == 0) {
1870		ext4_warning(sb, "need to use ext2online to resize further");
1871		return -EPERM;
1872	}
1873
1874	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1875
1876	if (o_blocks_count + add < o_blocks_count) {
1877		ext4_warning(sb, "blocks_count overflow");
1878		return -EINVAL;
1879	}
1880
1881	if (o_blocks_count + add > n_blocks_count)
1882		add = n_blocks_count - o_blocks_count;
1883
1884	if (o_blocks_count + add < n_blocks_count)
1885		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1886			     o_blocks_count + add, add);
1887
1888	/* See if the device is actually as big as what was requested */
1889	bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1890	if (IS_ERR(bh)) {
1891		ext4_warning(sb, "can't read last block, resize aborted");
1892		return -ENOSPC;
1893	}
1894	brelse(bh);
1895
1896	return ext4_group_extend_no_check(sb, o_blocks_count, add);
1897} /* ext4_group_extend */
1898
1899
1900static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1901{
1902	return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1903}
1904
1905/*
1906 * Release the resize inode and drop the resize_inode feature if there
1907 * are no more reserved gdt blocks, and then convert the file system
1908 * to enable meta_bg
1909 */
1910static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1911{
1912	handle_t *handle;
1913	struct ext4_sb_info *sbi = EXT4_SB(sb);
1914	struct ext4_super_block *es = sbi->s_es;
1915	struct ext4_inode_info *ei = EXT4_I(inode);
1916	ext4_fsblk_t nr;
1917	int i, ret, err = 0;
1918	int credits = 1;
1919
1920	ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1921	if (inode) {
1922		if (es->s_reserved_gdt_blocks) {
1923			ext4_error(sb, "Unexpected non-zero "
1924				   "s_reserved_gdt_blocks");
1925			return -EPERM;
1926		}
1927
1928		/* Do a quick sanity check of the resize inode */
1929		if (inode->i_blocks != 1 << (inode->i_blkbits -
1930					     (9 - sbi->s_cluster_bits)))
1931			goto invalid_resize_inode;
1932		for (i = 0; i < EXT4_N_BLOCKS; i++) {
1933			if (i == EXT4_DIND_BLOCK) {
1934				if (ei->i_data[i])
1935					continue;
1936				else
1937					goto invalid_resize_inode;
1938			}
1939			if (ei->i_data[i])
1940				goto invalid_resize_inode;
1941		}
1942		credits += 3;	/* block bitmap, bg descriptor, resize inode */
1943	}
1944
1945	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1946	if (IS_ERR(handle))
1947		return PTR_ERR(handle);
1948
1949	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1950	err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1951					    EXT4_JTR_NONE);
1952	if (err)
1953		goto errout;
1954
1955	lock_buffer(sbi->s_sbh);
1956	ext4_clear_feature_resize_inode(sb);
1957	ext4_set_feature_meta_bg(sb);
1958	sbi->s_es->s_first_meta_bg =
1959		cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1960	ext4_superblock_csum_set(sb);
1961	unlock_buffer(sbi->s_sbh);
1962
1963	err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1964	if (err) {
1965		ext4_std_error(sb, err);
1966		goto errout;
1967	}
1968
1969	if (inode) {
1970		nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1971		ext4_free_blocks(handle, inode, NULL, nr, 1,
1972				 EXT4_FREE_BLOCKS_METADATA |
1973				 EXT4_FREE_BLOCKS_FORGET);
1974		ei->i_data[EXT4_DIND_BLOCK] = 0;
1975		inode->i_blocks = 0;
1976
1977		err = ext4_mark_inode_dirty(handle, inode);
1978		if (err)
1979			ext4_std_error(sb, err);
1980	}
1981
1982errout:
1983	ret = ext4_journal_stop(handle);
1984	if (!err)
1985		err = ret;
1986	return ret;
1987
1988invalid_resize_inode:
1989	ext4_error(sb, "corrupted/inconsistent resize inode");
1990	return -EINVAL;
1991}
1992
1993/*
1994 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1995 *
1996 * @sb: super block of the fs to be resized
1997 * @n_blocks_count: the number of blocks resides in the resized fs
1998 */
1999int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
2000{
2001	struct ext4_new_flex_group_data *flex_gd = NULL;
2002	struct ext4_sb_info *sbi = EXT4_SB(sb);
2003	struct ext4_super_block *es = sbi->s_es;
2004	struct buffer_head *bh;
2005	struct inode *resize_inode = NULL;
2006	ext4_grpblk_t add, offset;
2007	unsigned long n_desc_blocks;
2008	unsigned long o_desc_blocks;
2009	ext4_group_t o_group;
2010	ext4_group_t n_group;
2011	ext4_fsblk_t o_blocks_count;
2012	ext4_fsblk_t n_blocks_count_retry = 0;
2013	unsigned long last_update_time = 0;
2014	int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
2015	int meta_bg;
 
2016
2017	/* See if the device is actually as big as what was requested */
2018	bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
2019	if (IS_ERR(bh)) {
2020		ext4_warning(sb, "can't read last block, resize aborted");
2021		return -ENOSPC;
2022	}
2023	brelse(bh);
2024
2025	/*
2026	 * For bigalloc, trim the requested size to the nearest cluster
2027	 * boundary to avoid creating an unusable filesystem. We do this
2028	 * silently, instead of returning an error, to avoid breaking
2029	 * callers that blindly resize the filesystem to the full size of
2030	 * the underlying block device.
2031	 */
2032	if (ext4_has_feature_bigalloc(sb))
2033		n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
2034
2035retry:
2036	o_blocks_count = ext4_blocks_count(es);
2037
2038	ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
2039		 "to %llu blocks", o_blocks_count, n_blocks_count);
2040
2041	if (n_blocks_count < o_blocks_count) {
2042		/* On-line shrinking not supported */
2043		ext4_warning(sb, "can't shrink FS - resize aborted");
2044		return -EINVAL;
2045	}
2046
2047	if (n_blocks_count == o_blocks_count)
2048		/* Nothing need to do */
2049		return 0;
2050
2051	n_group = ext4_get_group_number(sb, n_blocks_count - 1);
2052	if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2053		ext4_warning(sb, "resize would cause inodes_count overflow");
2054		return -EINVAL;
2055	}
2056	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2057
2058	n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2059	o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2060
2061	meta_bg = ext4_has_feature_meta_bg(sb);
2062
2063	if (ext4_has_feature_resize_inode(sb)) {
2064		if (meta_bg) {
2065			ext4_error(sb, "resize_inode and meta_bg enabled "
2066				   "simultaneously");
2067			return -EINVAL;
2068		}
2069		if (n_desc_blocks > o_desc_blocks +
2070		    le16_to_cpu(es->s_reserved_gdt_blocks)) {
2071			n_blocks_count_retry = n_blocks_count;
2072			n_desc_blocks = o_desc_blocks +
2073				le16_to_cpu(es->s_reserved_gdt_blocks);
2074			n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2075			n_blocks_count = (ext4_fsblk_t)n_group *
2076				EXT4_BLOCKS_PER_GROUP(sb) +
2077				le32_to_cpu(es->s_first_data_block);
2078			n_group--; /* set to last group number */
2079		}
2080
2081		if (!resize_inode)
2082			resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2083						 EXT4_IGET_SPECIAL);
2084		if (IS_ERR(resize_inode)) {
2085			ext4_warning(sb, "Error opening resize inode");
2086			return PTR_ERR(resize_inode);
2087		}
2088	}
2089
2090	if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2091		err = ext4_convert_meta_bg(sb, resize_inode);
2092		if (err)
2093			goto out;
2094		if (resize_inode) {
2095			iput(resize_inode);
2096			resize_inode = NULL;
2097		}
2098		if (n_blocks_count_retry) {
2099			n_blocks_count = n_blocks_count_retry;
2100			n_blocks_count_retry = 0;
2101			goto retry;
2102		}
2103	}
2104
2105	/*
2106	 * Make sure the last group has enough space so that it's
2107	 * guaranteed to have enough space for all metadata blocks
2108	 * that it might need to hold.  (We might not need to store
2109	 * the inode table blocks in the last block group, but there
2110	 * will be cases where this might be needed.)
2111	 */
2112	if ((ext4_group_first_block_no(sb, n_group) +
2113	     ext4_group_overhead_blocks(sb, n_group) + 2 +
2114	     sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2115		n_blocks_count = ext4_group_first_block_no(sb, n_group);
2116		n_group--;
2117		n_blocks_count_retry = 0;
2118		if (resize_inode) {
2119			iput(resize_inode);
2120			resize_inode = NULL;
2121		}
2122		goto retry;
2123	}
2124
2125	/* extend the last group */
2126	if (n_group == o_group)
2127		add = n_blocks_count - o_blocks_count;
2128	else
2129		add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2130	if (add > 0) {
2131		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2132		if (err)
2133			goto out;
2134	}
2135
2136	if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2137		goto out;
2138
2139	err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2140	if (err)
2141		goto out;
2142
2143	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2144	if (err)
2145		goto out;
2146
2147	flex_gd = alloc_flex_gd(flexbg_size);
2148	if (flex_gd == NULL) {
2149		err = -ENOMEM;
2150		goto out;
2151	}
2152
2153	/* Add flex groups. Note that a regular group is a
2154	 * flex group with 1 group.
2155	 */
2156	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2157					      flexbg_size)) {
2158		if (time_is_before_jiffies(last_update_time + HZ * 10)) {
2159			if (last_update_time)
2160				ext4_msg(sb, KERN_INFO,
2161					 "resized to %llu blocks",
2162					 ext4_blocks_count(es));
2163			last_update_time = jiffies;
2164		}
2165		if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2166			break;
2167		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2168		if (unlikely(err))
2169			break;
2170	}
2171
2172	if (!err && n_blocks_count_retry) {
2173		n_blocks_count = n_blocks_count_retry;
2174		n_blocks_count_retry = 0;
2175		free_flex_gd(flex_gd);
2176		flex_gd = NULL;
2177		if (resize_inode) {
2178			iput(resize_inode);
2179			resize_inode = NULL;
2180		}
2181		goto retry;
2182	}
2183
2184out:
2185	if (flex_gd)
2186		free_flex_gd(flex_gd);
2187	if (resize_inode != NULL)
2188		iput(resize_inode);
2189	if (err)
2190		ext4_warning(sb, "error (%d) occurred during "
2191			     "file system resize", err);
2192	ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2193		 ext4_blocks_count(es));
2194	return err;
2195}