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 resize_bg; /* number of allocated 235 new_group_data */ 236 ext4_group_t count; /* number of groups in @groups 237 */ 238}; 239 240/* 241 * Avoiding memory allocation failures due to too many groups added each time. 242 */ 243#define MAX_RESIZE_BG 16384 244 245/* 246 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of 247 * @flexbg_size. 248 * 249 * Returns NULL on failure otherwise address of the allocated structure. 250 */ 251static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size) 252{ 253 struct ext4_new_flex_group_data *flex_gd; 254 255 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS); 256 if (flex_gd == NULL) 257 goto out3; 258 259 if (unlikely(flexbg_size > MAX_RESIZE_BG)) 260 flex_gd->resize_bg = MAX_RESIZE_BG; 261 else 262 flex_gd->resize_bg = flexbg_size; 263 264 flex_gd->groups = kmalloc_array(flex_gd->resize_bg, 265 sizeof(struct ext4_new_group_data), 266 GFP_NOFS); 267 if (flex_gd->groups == NULL) 268 goto out2; 269 270 flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16), 271 GFP_NOFS); 272 if (flex_gd->bg_flags == NULL) 273 goto out1; 274 275 return flex_gd; 276 277out1: 278 kfree(flex_gd->groups); 279out2: 280 kfree(flex_gd); 281out3: 282 return NULL; 283} 284 285static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd) 286{ 287 kfree(flex_gd->bg_flags); 288 kfree(flex_gd->groups); 289 kfree(flex_gd); 290} 291 292/* 293 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps 294 * and inode tables for a flex group. 295 * 296 * This function is used by 64bit-resize. Note that this function allocates 297 * group tables from the 1st group of groups contained by @flexgd, which may 298 * be a partial of a flex group. 299 * 300 * @sb: super block of fs to which the groups belongs 301 * 302 * Returns 0 on a successful allocation of the metadata blocks in the 303 * block group. 304 */ 305static int ext4_alloc_group_tables(struct super_block *sb, 306 struct ext4_new_flex_group_data *flex_gd, 307 unsigned int flexbg_size) 308{ 309 struct ext4_new_group_data *group_data = flex_gd->groups; 310 ext4_fsblk_t start_blk; 311 ext4_fsblk_t last_blk; 312 ext4_group_t src_group; 313 ext4_group_t bb_index = 0; 314 ext4_group_t ib_index = 0; 315 ext4_group_t it_index = 0; 316 ext4_group_t group; 317 ext4_group_t last_group; 318 unsigned overhead; 319 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0; 320 int i; 321 322 BUG_ON(flex_gd->count == 0 || group_data == NULL); 323 324 src_group = group_data[0].group; 325 last_group = src_group + flex_gd->count - 1; 326 327 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) != 328 (last_group & ~(flexbg_size - 1)))); 329next_group: 330 group = group_data[0].group; 331 if (src_group >= group_data[0].group + flex_gd->count) 332 return -ENOSPC; 333 start_blk = ext4_group_first_block_no(sb, src_group); 334 last_blk = start_blk + group_data[src_group - group].blocks_count; 335 336 overhead = ext4_group_overhead_blocks(sb, src_group); 337 338 start_blk += overhead; 339 340 /* We collect contiguous blocks as much as possible. */ 341 src_group++; 342 for (; src_group <= last_group; src_group++) { 343 overhead = ext4_group_overhead_blocks(sb, src_group); 344 if (overhead == 0) 345 last_blk += group_data[src_group - group].blocks_count; 346 else 347 break; 348 } 349 350 /* Allocate block bitmaps */ 351 for (; bb_index < flex_gd->count; bb_index++) { 352 if (start_blk >= last_blk) 353 goto next_group; 354 group_data[bb_index].block_bitmap = start_blk++; 355 group = ext4_get_group_number(sb, start_blk - 1); 356 group -= group_data[0].group; 357 group_data[group].mdata_blocks++; 358 flex_gd->bg_flags[group] &= uninit_mask; 359 } 360 361 /* Allocate inode bitmaps */ 362 for (; ib_index < flex_gd->count; ib_index++) { 363 if (start_blk >= last_blk) 364 goto next_group; 365 group_data[ib_index].inode_bitmap = start_blk++; 366 group = ext4_get_group_number(sb, start_blk - 1); 367 group -= group_data[0].group; 368 group_data[group].mdata_blocks++; 369 flex_gd->bg_flags[group] &= uninit_mask; 370 } 371 372 /* Allocate inode tables */ 373 for (; it_index < flex_gd->count; it_index++) { 374 unsigned int itb = EXT4_SB(sb)->s_itb_per_group; 375 ext4_fsblk_t next_group_start; 376 377 if (start_blk + itb > last_blk) 378 goto next_group; 379 group_data[it_index].inode_table = start_blk; 380 group = ext4_get_group_number(sb, start_blk); 381 next_group_start = ext4_group_first_block_no(sb, group + 1); 382 group -= group_data[0].group; 383 384 if (start_blk + itb > next_group_start) { 385 flex_gd->bg_flags[group + 1] &= uninit_mask; 386 overhead = start_blk + itb - next_group_start; 387 group_data[group + 1].mdata_blocks += overhead; 388 itb -= overhead; 389 } 390 391 group_data[group].mdata_blocks += itb; 392 flex_gd->bg_flags[group] &= uninit_mask; 393 start_blk += EXT4_SB(sb)->s_itb_per_group; 394 } 395 396 /* Update free clusters count to exclude metadata blocks */ 397 for (i = 0; i < flex_gd->count; i++) { 398 group_data[i].free_clusters_count -= 399 EXT4_NUM_B2C(EXT4_SB(sb), 400 group_data[i].mdata_blocks); 401 } 402 403 if (test_opt(sb, DEBUG)) { 404 int i; 405 group = group_data[0].group; 406 407 printk(KERN_DEBUG "EXT4-fs: adding a flex group with " 408 "%u groups, flexbg size is %u:\n", flex_gd->count, 409 flexbg_size); 410 411 for (i = 0; i < flex_gd->count; i++) { 412 ext4_debug( 413 "adding %s group %u: %u blocks (%u free, %u mdata blocks)\n", 414 ext4_bg_has_super(sb, group + i) ? "normal" : 415 "no-super", group + i, 416 group_data[i].blocks_count, 417 group_data[i].free_clusters_count, 418 group_data[i].mdata_blocks); 419 } 420 } 421 return 0; 422} 423 424static struct buffer_head *bclean(handle_t *handle, struct super_block *sb, 425 ext4_fsblk_t blk) 426{ 427 struct buffer_head *bh; 428 int err; 429 430 bh = sb_getblk(sb, blk); 431 if (unlikely(!bh)) 432 return ERR_PTR(-ENOMEM); 433 BUFFER_TRACE(bh, "get_write_access"); 434 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); 435 if (err) { 436 brelse(bh); 437 bh = ERR_PTR(err); 438 } else { 439 memset(bh->b_data, 0, sb->s_blocksize); 440 set_buffer_uptodate(bh); 441 } 442 443 return bh; 444} 445 446static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits) 447{ 448 return ext4_journal_ensure_credits_fn(handle, credits, 449 EXT4_MAX_TRANS_DATA, 0, 0); 450} 451 452/* 453 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used. 454 * 455 * Helper function for ext4_setup_new_group_blocks() which set . 456 * 457 * @sb: super block 458 * @handle: journal handle 459 * @flex_gd: flex group data 460 */ 461static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle, 462 struct ext4_new_flex_group_data *flex_gd, 463 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster) 464{ 465 struct ext4_sb_info *sbi = EXT4_SB(sb); 466 ext4_group_t count = last_cluster - first_cluster + 1; 467 ext4_group_t count2; 468 469 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster, 470 last_cluster); 471 for (count2 = count; count > 0; 472 count -= count2, first_cluster += count2) { 473 ext4_fsblk_t start; 474 struct buffer_head *bh; 475 ext4_group_t group; 476 int err; 477 478 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster)); 479 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group)); 480 group -= flex_gd->groups[0].group; 481 482 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start); 483 if (count2 > count) 484 count2 = count; 485 486 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) { 487 BUG_ON(flex_gd->count > 1); 488 continue; 489 } 490 491 err = ext4_resize_ensure_credits_batch(handle, 1); 492 if (err < 0) 493 return err; 494 495 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap); 496 if (unlikely(!bh)) 497 return -ENOMEM; 498 499 BUFFER_TRACE(bh, "get_write_access"); 500 err = ext4_journal_get_write_access(handle, sb, bh, 501 EXT4_JTR_NONE); 502 if (err) { 503 brelse(bh); 504 return err; 505 } 506 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", 507 first_cluster, first_cluster - start, count2); 508 mb_set_bits(bh->b_data, first_cluster - start, count2); 509 510 err = ext4_handle_dirty_metadata(handle, NULL, bh); 511 brelse(bh); 512 if (unlikely(err)) 513 return err; 514 } 515 516 return 0; 517} 518 519/* 520 * Set up the block and inode bitmaps, and the inode table for the new groups. 521 * This doesn't need to be part of the main transaction, since we are only 522 * changing blocks outside the actual filesystem. We still do journaling to 523 * ensure the recovery is correct in case of a failure just after resize. 524 * If any part of this fails, we simply abort the resize. 525 * 526 * setup_new_flex_group_blocks handles a flex group as follow: 527 * 1. copy super block and GDT, and initialize group tables if necessary. 528 * In this step, we only set bits in blocks bitmaps for blocks taken by 529 * super block and GDT. 530 * 2. allocate group tables in block bitmaps, that is, set bits in block 531 * bitmap for blocks taken by group tables. 532 */ 533static int setup_new_flex_group_blocks(struct super_block *sb, 534 struct ext4_new_flex_group_data *flex_gd) 535{ 536 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group}; 537 ext4_fsblk_t start; 538 ext4_fsblk_t block; 539 struct ext4_sb_info *sbi = EXT4_SB(sb); 540 struct ext4_super_block *es = sbi->s_es; 541 struct ext4_new_group_data *group_data = flex_gd->groups; 542 __u16 *bg_flags = flex_gd->bg_flags; 543 handle_t *handle; 544 ext4_group_t group, count; 545 struct buffer_head *bh = NULL; 546 int reserved_gdb, i, j, err = 0, err2; 547 int meta_bg; 548 549 BUG_ON(!flex_gd->count || !group_data || 550 group_data[0].group != sbi->s_groups_count); 551 552 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks); 553 meta_bg = ext4_has_feature_meta_bg(sb); 554 555 /* This transaction may be extended/restarted along the way */ 556 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA); 557 if (IS_ERR(handle)) 558 return PTR_ERR(handle); 559 560 group = group_data[0].group; 561 for (i = 0; i < flex_gd->count; i++, group++) { 562 unsigned long gdblocks; 563 ext4_grpblk_t overhead; 564 565 gdblocks = ext4_bg_num_gdb(sb, group); 566 start = ext4_group_first_block_no(sb, group); 567 568 if (meta_bg == 0 && !ext4_bg_has_super(sb, group)) 569 goto handle_itb; 570 571 if (meta_bg == 1) 572 goto handle_itb; 573 574 block = start + ext4_bg_has_super(sb, group); 575 /* Copy all of the GDT blocks into the backup in this group */ 576 for (j = 0; j < gdblocks; j++, block++) { 577 struct buffer_head *gdb; 578 579 ext4_debug("update backup group %#04llx\n", block); 580 err = ext4_resize_ensure_credits_batch(handle, 1); 581 if (err < 0) 582 goto out; 583 584 gdb = sb_getblk(sb, block); 585 if (unlikely(!gdb)) { 586 err = -ENOMEM; 587 goto out; 588 } 589 590 BUFFER_TRACE(gdb, "get_write_access"); 591 err = ext4_journal_get_write_access(handle, sb, gdb, 592 EXT4_JTR_NONE); 593 if (err) { 594 brelse(gdb); 595 goto out; 596 } 597 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi, 598 s_group_desc, j)->b_data, gdb->b_size); 599 set_buffer_uptodate(gdb); 600 601 err = ext4_handle_dirty_metadata(handle, NULL, gdb); 602 if (unlikely(err)) { 603 brelse(gdb); 604 goto out; 605 } 606 brelse(gdb); 607 } 608 609 /* Zero out all of the reserved backup group descriptor 610 * table blocks 611 */ 612 if (ext4_bg_has_super(sb, group)) { 613 err = sb_issue_zeroout(sb, gdblocks + start + 1, 614 reserved_gdb, GFP_NOFS); 615 if (err) 616 goto out; 617 } 618 619handle_itb: 620 /* Initialize group tables of the grop @group */ 621 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED)) 622 goto handle_bb; 623 624 /* Zero out all of the inode table blocks */ 625 block = group_data[i].inode_table; 626 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n", 627 block, sbi->s_itb_per_group); 628 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, 629 GFP_NOFS); 630 if (err) 631 goto out; 632 633handle_bb: 634 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT) 635 goto handle_ib; 636 637 /* Initialize block bitmap of the @group */ 638 block = group_data[i].block_bitmap; 639 err = ext4_resize_ensure_credits_batch(handle, 1); 640 if (err < 0) 641 goto out; 642 643 bh = bclean(handle, sb, block); 644 if (IS_ERR(bh)) { 645 err = PTR_ERR(bh); 646 goto out; 647 } 648 overhead = ext4_group_overhead_blocks(sb, group); 649 if (overhead != 0) { 650 ext4_debug("mark backup superblock %#04llx (+0)\n", 651 start); 652 mb_set_bits(bh->b_data, 0, 653 EXT4_NUM_B2C(sbi, overhead)); 654 } 655 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count), 656 sb->s_blocksize * 8, bh->b_data); 657 err = ext4_handle_dirty_metadata(handle, NULL, bh); 658 brelse(bh); 659 if (err) 660 goto out; 661 662handle_ib: 663 if (bg_flags[i] & EXT4_BG_INODE_UNINIT) 664 continue; 665 666 /* Initialize inode bitmap of the @group */ 667 block = group_data[i].inode_bitmap; 668 err = ext4_resize_ensure_credits_batch(handle, 1); 669 if (err < 0) 670 goto out; 671 /* Mark unused entries in inode bitmap used */ 672 bh = bclean(handle, sb, block); 673 if (IS_ERR(bh)) { 674 err = PTR_ERR(bh); 675 goto out; 676 } 677 678 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), 679 sb->s_blocksize * 8, bh->b_data); 680 err = ext4_handle_dirty_metadata(handle, NULL, bh); 681 brelse(bh); 682 if (err) 683 goto out; 684 } 685 686 /* Mark group tables in block bitmap */ 687 for (j = 0; j < GROUP_TABLE_COUNT; j++) { 688 count = group_table_count[j]; 689 start = (&group_data[0].block_bitmap)[j]; 690 block = start; 691 for (i = 1; i < flex_gd->count; i++) { 692 block += group_table_count[j]; 693 if (block == (&group_data[i].block_bitmap)[j]) { 694 count += group_table_count[j]; 695 continue; 696 } 697 err = set_flexbg_block_bitmap(sb, handle, 698 flex_gd, 699 EXT4_B2C(sbi, start), 700 EXT4_B2C(sbi, 701 start + count 702 - 1)); 703 if (err) 704 goto out; 705 count = group_table_count[j]; 706 start = (&group_data[i].block_bitmap)[j]; 707 block = start; 708 } 709 710 if (count) { 711 err = set_flexbg_block_bitmap(sb, handle, 712 flex_gd, 713 EXT4_B2C(sbi, start), 714 EXT4_B2C(sbi, 715 start + count 716 - 1)); 717 if (err) 718 goto out; 719 } 720 } 721 722out: 723 err2 = ext4_journal_stop(handle); 724 if (err2 && !err) 725 err = err2; 726 727 return err; 728} 729 730/* 731 * Iterate through the groups which hold BACKUP superblock/GDT copies in an 732 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before 733 * calling this for the first time. In a sparse filesystem it will be the 734 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ... 735 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ... 736 */ 737unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three, 738 unsigned int *five, unsigned int *seven) 739{ 740 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 741 unsigned int *min = three; 742 int mult = 3; 743 unsigned int ret; 744 745 if (ext4_has_feature_sparse_super2(sb)) { 746 do { 747 if (*min > 2) 748 return UINT_MAX; 749 ret = le32_to_cpu(es->s_backup_bgs[*min - 1]); 750 *min += 1; 751 } while (!ret); 752 return ret; 753 } 754 755 if (!ext4_has_feature_sparse_super(sb)) { 756 ret = *min; 757 *min += 1; 758 return ret; 759 } 760 761 if (*five < *min) { 762 min = five; 763 mult = 5; 764 } 765 if (*seven < *min) { 766 min = seven; 767 mult = 7; 768 } 769 770 ret = *min; 771 *min *= mult; 772 773 return ret; 774} 775 776/* 777 * Check that all of the backup GDT blocks are held in the primary GDT block. 778 * It is assumed that they are stored in group order. Returns the number of 779 * groups in current filesystem that have BACKUPS, or -ve error code. 780 */ 781static int verify_reserved_gdb(struct super_block *sb, 782 ext4_group_t end, 783 struct buffer_head *primary) 784{ 785 const ext4_fsblk_t blk = primary->b_blocknr; 786 unsigned three = 1; 787 unsigned five = 5; 788 unsigned seven = 7; 789 unsigned grp; 790 __le32 *p = (__le32 *)primary->b_data; 791 int gdbackups = 0; 792 793 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) { 794 if (le32_to_cpu(*p++) != 795 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){ 796 ext4_warning(sb, "reserved GDT %llu" 797 " missing grp %d (%llu)", 798 blk, grp, 799 grp * 800 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) + 801 blk); 802 return -EINVAL; 803 } 804 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb)) 805 return -EFBIG; 806 } 807 808 return gdbackups; 809} 810 811/* 812 * Called when we need to bring a reserved group descriptor table block into 813 * use from the resize inode. The primary copy of the new GDT block currently 814 * is an indirect block (under the double indirect block in the resize inode). 815 * The new backup GDT blocks will be stored as leaf blocks in this indirect 816 * block, in group order. Even though we know all the block numbers we need, 817 * we check to ensure that the resize inode has actually reserved these blocks. 818 * 819 * Don't need to update the block bitmaps because the blocks are still in use. 820 * 821 * We get all of the error cases out of the way, so that we are sure to not 822 * fail once we start modifying the data on disk, because JBD has no rollback. 823 */ 824static int add_new_gdb(handle_t *handle, struct inode *inode, 825 ext4_group_t group) 826{ 827 struct super_block *sb = inode->i_sb; 828 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 829 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 830 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num; 831 struct buffer_head **o_group_desc, **n_group_desc = NULL; 832 struct buffer_head *dind = NULL; 833 struct buffer_head *gdb_bh = NULL; 834 int gdbackups; 835 struct ext4_iloc iloc = { .bh = NULL }; 836 __le32 *data; 837 int err; 838 839 if (test_opt(sb, DEBUG)) 840 printk(KERN_DEBUG 841 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n", 842 gdb_num); 843 844 gdb_bh = ext4_sb_bread(sb, gdblock, 0); 845 if (IS_ERR(gdb_bh)) 846 return PTR_ERR(gdb_bh); 847 848 gdbackups = verify_reserved_gdb(sb, group, gdb_bh); 849 if (gdbackups < 0) { 850 err = gdbackups; 851 goto errout; 852 } 853 854 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; 855 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0); 856 if (IS_ERR(dind)) { 857 err = PTR_ERR(dind); 858 dind = NULL; 859 goto errout; 860 } 861 862 data = (__le32 *)dind->b_data; 863 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) { 864 ext4_warning(sb, "new group %u GDT block %llu not reserved", 865 group, gdblock); 866 err = -EINVAL; 867 goto errout; 868 } 869 870 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); 871 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh, 872 EXT4_JTR_NONE); 873 if (unlikely(err)) 874 goto errout; 875 876 BUFFER_TRACE(gdb_bh, "get_write_access"); 877 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE); 878 if (unlikely(err)) 879 goto errout; 880 881 BUFFER_TRACE(dind, "get_write_access"); 882 err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE); 883 if (unlikely(err)) { 884 ext4_std_error(sb, err); 885 goto errout; 886 } 887 888 /* ext4_reserve_inode_write() gets a reference on the iloc */ 889 err = ext4_reserve_inode_write(handle, inode, &iloc); 890 if (unlikely(err)) 891 goto errout; 892 893 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *), 894 GFP_KERNEL); 895 if (!n_group_desc) { 896 err = -ENOMEM; 897 ext4_warning(sb, "not enough memory for %lu groups", 898 gdb_num + 1); 899 goto errout; 900 } 901 902 /* 903 * Finally, we have all of the possible failures behind us... 904 * 905 * Remove new GDT block from inode double-indirect block and clear out 906 * the new GDT block for use (which also "frees" the backup GDT blocks 907 * from the reserved inode). We don't need to change the bitmaps for 908 * these blocks, because they are marked as in-use from being in the 909 * reserved inode, and will become GDT blocks (primary and backup). 910 */ 911 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0; 912 err = ext4_handle_dirty_metadata(handle, NULL, dind); 913 if (unlikely(err)) { 914 ext4_std_error(sb, err); 915 goto errout; 916 } 917 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 918 (9 - EXT4_SB(sb)->s_cluster_bits); 919 ext4_mark_iloc_dirty(handle, inode, &iloc); 920 memset(gdb_bh->b_data, 0, sb->s_blocksize); 921 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh); 922 if (unlikely(err)) { 923 ext4_std_error(sb, err); 924 iloc.bh = NULL; 925 goto errout; 926 } 927 brelse(dind); 928 929 rcu_read_lock(); 930 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc); 931 memcpy(n_group_desc, o_group_desc, 932 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); 933 rcu_read_unlock(); 934 n_group_desc[gdb_num] = gdb_bh; 935 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc); 936 EXT4_SB(sb)->s_gdb_count++; 937 ext4_kvfree_array_rcu(o_group_desc); 938 939 lock_buffer(EXT4_SB(sb)->s_sbh); 940 le16_add_cpu(&es->s_reserved_gdt_blocks, -1); 941 ext4_superblock_csum_set(sb); 942 unlock_buffer(EXT4_SB(sb)->s_sbh); 943 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); 944 if (err) 945 ext4_std_error(sb, err); 946 return err; 947errout: 948 kvfree(n_group_desc); 949 brelse(iloc.bh); 950 brelse(dind); 951 brelse(gdb_bh); 952 953 ext4_debug("leaving with error %d\n", err); 954 return err; 955} 956 957/* 958 * add_new_gdb_meta_bg is the sister of add_new_gdb. 959 */ 960static int add_new_gdb_meta_bg(struct super_block *sb, 961 handle_t *handle, ext4_group_t group) { 962 ext4_fsblk_t gdblock; 963 struct buffer_head *gdb_bh; 964 struct buffer_head **o_group_desc, **n_group_desc; 965 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 966 int err; 967 968 gdblock = ext4_meta_bg_first_block_no(sb, group) + 969 ext4_bg_has_super(sb, group); 970 gdb_bh = ext4_sb_bread(sb, gdblock, 0); 971 if (IS_ERR(gdb_bh)) 972 return PTR_ERR(gdb_bh); 973 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *), 974 GFP_KERNEL); 975 if (!n_group_desc) { 976 brelse(gdb_bh); 977 err = -ENOMEM; 978 ext4_warning(sb, "not enough memory for %lu groups", 979 gdb_num + 1); 980 return err; 981 } 982 983 rcu_read_lock(); 984 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc); 985 memcpy(n_group_desc, o_group_desc, 986 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); 987 rcu_read_unlock(); 988 n_group_desc[gdb_num] = gdb_bh; 989 990 BUFFER_TRACE(gdb_bh, "get_write_access"); 991 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE); 992 if (err) { 993 kvfree(n_group_desc); 994 brelse(gdb_bh); 995 return err; 996 } 997 998 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc); 999 EXT4_SB(sb)->s_gdb_count++; 1000 ext4_kvfree_array_rcu(o_group_desc); 1001 return err; 1002} 1003 1004/* 1005 * Called when we are adding a new group which has a backup copy of each of 1006 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks. 1007 * We need to add these reserved backup GDT blocks to the resize inode, so 1008 * that they are kept for future resizing and not allocated to files. 1009 * 1010 * Each reserved backup GDT block will go into a different indirect block. 1011 * The indirect blocks are actually the primary reserved GDT blocks, 1012 * so we know in advance what their block numbers are. We only get the 1013 * double-indirect block to verify it is pointing to the primary reserved 1014 * GDT blocks so we don't overwrite a data block by accident. The reserved 1015 * backup GDT blocks are stored in their reserved primary GDT block. 1016 */ 1017static int reserve_backup_gdb(handle_t *handle, struct inode *inode, 1018 ext4_group_t group) 1019{ 1020 struct super_block *sb = inode->i_sb; 1021 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); 1022 int cluster_bits = EXT4_SB(sb)->s_cluster_bits; 1023 struct buffer_head **primary; 1024 struct buffer_head *dind; 1025 struct ext4_iloc iloc; 1026 ext4_fsblk_t blk; 1027 __le32 *data, *end; 1028 int gdbackups = 0; 1029 int res, i; 1030 int err; 1031 1032 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS); 1033 if (!primary) 1034 return -ENOMEM; 1035 1036 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; 1037 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0); 1038 if (IS_ERR(dind)) { 1039 err = PTR_ERR(dind); 1040 dind = NULL; 1041 goto exit_free; 1042 } 1043 1044 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count; 1045 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count % 1046 EXT4_ADDR_PER_BLOCK(sb)); 1047 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb); 1048 1049 /* Get each reserved primary GDT block and verify it holds backups */ 1050 for (res = 0; res < reserved_gdb; res++, blk++) { 1051 if (le32_to_cpu(*data) != blk) { 1052 ext4_warning(sb, "reserved block %llu" 1053 " not at offset %ld", 1054 blk, 1055 (long)(data - (__le32 *)dind->b_data)); 1056 err = -EINVAL; 1057 goto exit_bh; 1058 } 1059 primary[res] = ext4_sb_bread(sb, blk, 0); 1060 if (IS_ERR(primary[res])) { 1061 err = PTR_ERR(primary[res]); 1062 primary[res] = NULL; 1063 goto exit_bh; 1064 } 1065 gdbackups = verify_reserved_gdb(sb, group, primary[res]); 1066 if (gdbackups < 0) { 1067 brelse(primary[res]); 1068 err = gdbackups; 1069 goto exit_bh; 1070 } 1071 if (++data >= end) 1072 data = (__le32 *)dind->b_data; 1073 } 1074 1075 for (i = 0; i < reserved_gdb; i++) { 1076 BUFFER_TRACE(primary[i], "get_write_access"); 1077 if ((err = ext4_journal_get_write_access(handle, sb, primary[i], 1078 EXT4_JTR_NONE))) 1079 goto exit_bh; 1080 } 1081 1082 if ((err = ext4_reserve_inode_write(handle, inode, &iloc))) 1083 goto exit_bh; 1084 1085 /* 1086 * Finally we can add each of the reserved backup GDT blocks from 1087 * the new group to its reserved primary GDT block. 1088 */ 1089 blk = group * EXT4_BLOCKS_PER_GROUP(sb); 1090 for (i = 0; i < reserved_gdb; i++) { 1091 int err2; 1092 data = (__le32 *)primary[i]->b_data; 1093 /* printk("reserving backup %lu[%u] = %lu\n", 1094 primary[i]->b_blocknr, gdbackups, 1095 blk + primary[i]->b_blocknr); */ 1096 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr); 1097 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]); 1098 if (!err) 1099 err = err2; 1100 } 1101 1102 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits); 1103 ext4_mark_iloc_dirty(handle, inode, &iloc); 1104 1105exit_bh: 1106 while (--res >= 0) 1107 brelse(primary[res]); 1108 brelse(dind); 1109 1110exit_free: 1111 kfree(primary); 1112 1113 return err; 1114} 1115 1116static inline void ext4_set_block_group_nr(struct super_block *sb, char *data, 1117 ext4_group_t group) 1118{ 1119 struct ext4_super_block *es = (struct ext4_super_block *) data; 1120 1121 es->s_block_group_nr = cpu_to_le16(group); 1122 if (ext4_has_metadata_csum(sb)) 1123 es->s_checksum = ext4_superblock_csum(sb, es); 1124} 1125 1126/* 1127 * Update the backup copies of the ext4 metadata. These don't need to be part 1128 * of the main resize transaction, because e2fsck will re-write them if there 1129 * is a problem (basically only OOM will cause a problem). However, we 1130 * _should_ update the backups if possible, in case the primary gets trashed 1131 * for some reason and we need to run e2fsck from a backup superblock. The 1132 * important part is that the new block and inode counts are in the backup 1133 * superblocks, and the location of the new group metadata in the GDT backups. 1134 * 1135 * We do not need take the s_resize_lock for this, because these 1136 * blocks are not otherwise touched by the filesystem code when it is 1137 * mounted. We don't need to worry about last changing from 1138 * sbi->s_groups_count, because the worst that can happen is that we 1139 * do not copy the full number of backups at this time. The resize 1140 * which changed s_groups_count will backup again. 1141 */ 1142static void update_backups(struct super_block *sb, sector_t blk_off, char *data, 1143 int size, int meta_bg) 1144{ 1145 struct ext4_sb_info *sbi = EXT4_SB(sb); 1146 ext4_group_t last; 1147 const int bpg = EXT4_BLOCKS_PER_GROUP(sb); 1148 unsigned three = 1; 1149 unsigned five = 5; 1150 unsigned seven = 7; 1151 ext4_group_t group = 0; 1152 int rest = sb->s_blocksize - size; 1153 handle_t *handle; 1154 int err = 0, err2; 1155 1156 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA); 1157 if (IS_ERR(handle)) { 1158 group = 1; 1159 err = PTR_ERR(handle); 1160 goto exit_err; 1161 } 1162 1163 if (meta_bg == 0) { 1164 group = ext4_list_backups(sb, &three, &five, &seven); 1165 last = sbi->s_groups_count; 1166 } else { 1167 group = ext4_get_group_number(sb, blk_off) + 1; 1168 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2); 1169 } 1170 1171 while (group < sbi->s_groups_count) { 1172 struct buffer_head *bh; 1173 ext4_fsblk_t backup_block; 1174 int has_super = ext4_bg_has_super(sb, group); 1175 ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group); 1176 1177 /* Out of journal space, and can't get more - abort - so sad */ 1178 err = ext4_resize_ensure_credits_batch(handle, 1); 1179 if (err < 0) 1180 break; 1181 1182 if (meta_bg == 0) 1183 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off; 1184 else 1185 backup_block = first_block + has_super; 1186 1187 bh = sb_getblk(sb, backup_block); 1188 if (unlikely(!bh)) { 1189 err = -ENOMEM; 1190 break; 1191 } 1192 ext4_debug("update metadata backup %llu(+%llu)\n", 1193 backup_block, backup_block - 1194 ext4_group_first_block_no(sb, group)); 1195 BUFFER_TRACE(bh, "get_write_access"); 1196 if ((err = ext4_journal_get_write_access(handle, sb, bh, 1197 EXT4_JTR_NONE))) { 1198 brelse(bh); 1199 break; 1200 } 1201 lock_buffer(bh); 1202 memcpy(bh->b_data, data, size); 1203 if (rest) 1204 memset(bh->b_data + size, 0, rest); 1205 if (has_super && (backup_block == first_block)) 1206 ext4_set_block_group_nr(sb, bh->b_data, group); 1207 set_buffer_uptodate(bh); 1208 unlock_buffer(bh); 1209 err = ext4_handle_dirty_metadata(handle, NULL, bh); 1210 if (unlikely(err)) 1211 ext4_std_error(sb, err); 1212 brelse(bh); 1213 1214 if (meta_bg == 0) 1215 group = ext4_list_backups(sb, &three, &five, &seven); 1216 else if (group == last) 1217 break; 1218 else 1219 group = last; 1220 } 1221 if ((err2 = ext4_journal_stop(handle)) && !err) 1222 err = err2; 1223 1224 /* 1225 * Ugh! Need to have e2fsck write the backup copies. It is too 1226 * late to revert the resize, we shouldn't fail just because of 1227 * the backup copies (they are only needed in case of corruption). 1228 * 1229 * However, if we got here we have a journal problem too, so we 1230 * can't really start a transaction to mark the superblock. 1231 * Chicken out and just set the flag on the hope it will be written 1232 * to disk, and if not - we will simply wait until next fsck. 1233 */ 1234exit_err: 1235 if (err) { 1236 ext4_warning(sb, "can't update backup for group %u (err %d), " 1237 "forcing fsck on next reboot", group, err); 1238 sbi->s_mount_state &= ~EXT4_VALID_FS; 1239 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 1240 mark_buffer_dirty(sbi->s_sbh); 1241 } 1242} 1243 1244/* 1245 * ext4_add_new_descs() adds @count group descriptor of groups 1246 * starting at @group 1247 * 1248 * @handle: journal handle 1249 * @sb: super block 1250 * @group: the group no. of the first group desc to be added 1251 * @resize_inode: the resize inode 1252 * @count: number of group descriptors to be added 1253 */ 1254static int ext4_add_new_descs(handle_t *handle, struct super_block *sb, 1255 ext4_group_t group, struct inode *resize_inode, 1256 ext4_group_t count) 1257{ 1258 struct ext4_sb_info *sbi = EXT4_SB(sb); 1259 struct ext4_super_block *es = sbi->s_es; 1260 struct buffer_head *gdb_bh; 1261 int i, gdb_off, gdb_num, err = 0; 1262 int meta_bg; 1263 1264 meta_bg = ext4_has_feature_meta_bg(sb); 1265 for (i = 0; i < count; i++, group++) { 1266 int reserved_gdb = ext4_bg_has_super(sb, group) ? 1267 le16_to_cpu(es->s_reserved_gdt_blocks) : 0; 1268 1269 gdb_off = group % EXT4_DESC_PER_BLOCK(sb); 1270 gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 1271 1272 /* 1273 * We will only either add reserved group blocks to a backup group 1274 * or remove reserved blocks for the first group in a new group block. 1275 * Doing both would be mean more complex code, and sane people don't 1276 * use non-sparse filesystems anymore. This is already checked above. 1277 */ 1278 if (gdb_off) { 1279 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, 1280 gdb_num); 1281 BUFFER_TRACE(gdb_bh, "get_write_access"); 1282 err = ext4_journal_get_write_access(handle, sb, gdb_bh, 1283 EXT4_JTR_NONE); 1284 1285 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group)) 1286 err = reserve_backup_gdb(handle, resize_inode, group); 1287 } else if (meta_bg != 0) { 1288 err = add_new_gdb_meta_bg(sb, handle, group); 1289 } else { 1290 err = add_new_gdb(handle, resize_inode, group); 1291 } 1292 if (err) 1293 break; 1294 } 1295 return err; 1296} 1297 1298static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block) 1299{ 1300 struct buffer_head *bh = sb_getblk(sb, block); 1301 if (unlikely(!bh)) 1302 return NULL; 1303 if (!bh_uptodate_or_lock(bh)) { 1304 if (ext4_read_bh(bh, 0, NULL) < 0) { 1305 brelse(bh); 1306 return NULL; 1307 } 1308 } 1309 1310 return bh; 1311} 1312 1313static int ext4_set_bitmap_checksums(struct super_block *sb, 1314 struct ext4_group_desc *gdp, 1315 struct ext4_new_group_data *group_data) 1316{ 1317 struct buffer_head *bh; 1318 1319 if (!ext4_has_metadata_csum(sb)) 1320 return 0; 1321 1322 bh = ext4_get_bitmap(sb, group_data->inode_bitmap); 1323 if (!bh) 1324 return -EIO; 1325 ext4_inode_bitmap_csum_set(sb, gdp, bh, 1326 EXT4_INODES_PER_GROUP(sb) / 8); 1327 brelse(bh); 1328 1329 bh = ext4_get_bitmap(sb, group_data->block_bitmap); 1330 if (!bh) 1331 return -EIO; 1332 ext4_block_bitmap_csum_set(sb, gdp, bh); 1333 brelse(bh); 1334 1335 return 0; 1336} 1337 1338/* 1339 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg 1340 */ 1341static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb, 1342 struct ext4_new_flex_group_data *flex_gd) 1343{ 1344 struct ext4_new_group_data *group_data = flex_gd->groups; 1345 struct ext4_group_desc *gdp; 1346 struct ext4_sb_info *sbi = EXT4_SB(sb); 1347 struct buffer_head *gdb_bh; 1348 ext4_group_t group; 1349 __u16 *bg_flags = flex_gd->bg_flags; 1350 int i, gdb_off, gdb_num, err = 0; 1351 1352 1353 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) { 1354 group = group_data->group; 1355 1356 gdb_off = group % EXT4_DESC_PER_BLOCK(sb); 1357 gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 1358 1359 /* 1360 * get_write_access() has been called on gdb_bh by ext4_add_new_desc(). 1361 */ 1362 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num); 1363 /* Update group descriptor block for new group */ 1364 gdp = (struct ext4_group_desc *)(gdb_bh->b_data + 1365 gdb_off * EXT4_DESC_SIZE(sb)); 1366 1367 memset(gdp, 0, EXT4_DESC_SIZE(sb)); 1368 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap); 1369 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap); 1370 err = ext4_set_bitmap_checksums(sb, gdp, group_data); 1371 if (err) { 1372 ext4_std_error(sb, err); 1373 break; 1374 } 1375 1376 ext4_inode_table_set(sb, gdp, group_data->inode_table); 1377 ext4_free_group_clusters_set(sb, gdp, 1378 group_data->free_clusters_count); 1379 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb)); 1380 if (ext4_has_group_desc_csum(sb)) 1381 ext4_itable_unused_set(sb, gdp, 1382 EXT4_INODES_PER_GROUP(sb)); 1383 gdp->bg_flags = cpu_to_le16(*bg_flags); 1384 ext4_group_desc_csum_set(sb, group, gdp); 1385 1386 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh); 1387 if (unlikely(err)) { 1388 ext4_std_error(sb, err); 1389 break; 1390 } 1391 1392 /* 1393 * We can allocate memory for mb_alloc based on the new group 1394 * descriptor 1395 */ 1396 err = ext4_mb_add_groupinfo(sb, group, gdp); 1397 if (err) 1398 break; 1399 } 1400 return err; 1401} 1402 1403static void ext4_add_overhead(struct super_block *sb, 1404 const ext4_fsblk_t overhead) 1405{ 1406 struct ext4_sb_info *sbi = EXT4_SB(sb); 1407 struct ext4_super_block *es = sbi->s_es; 1408 1409 sbi->s_overhead += overhead; 1410 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead); 1411 smp_wmb(); 1412} 1413 1414/* 1415 * ext4_update_super() updates the super block so that the newly added 1416 * groups can be seen by the filesystem. 1417 * 1418 * @sb: super block 1419 * @flex_gd: new added groups 1420 */ 1421static void ext4_update_super(struct super_block *sb, 1422 struct ext4_new_flex_group_data *flex_gd) 1423{ 1424 ext4_fsblk_t blocks_count = 0; 1425 ext4_fsblk_t free_blocks = 0; 1426 ext4_fsblk_t reserved_blocks = 0; 1427 struct ext4_new_group_data *group_data = flex_gd->groups; 1428 struct ext4_sb_info *sbi = EXT4_SB(sb); 1429 struct ext4_super_block *es = sbi->s_es; 1430 int i; 1431 1432 BUG_ON(flex_gd->count == 0 || group_data == NULL); 1433 /* 1434 * Make the new blocks and inodes valid next. We do this before 1435 * increasing the group count so that once the group is enabled, 1436 * all of its blocks and inodes are already valid. 1437 * 1438 * We always allocate group-by-group, then block-by-block or 1439 * inode-by-inode within a group, so enabling these 1440 * blocks/inodes before the group is live won't actually let us 1441 * allocate the new space yet. 1442 */ 1443 for (i = 0; i < flex_gd->count; i++) { 1444 blocks_count += group_data[i].blocks_count; 1445 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count); 1446 } 1447 1448 reserved_blocks = ext4_r_blocks_count(es) * 100; 1449 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es)); 1450 reserved_blocks *= blocks_count; 1451 do_div(reserved_blocks, 100); 1452 1453 lock_buffer(sbi->s_sbh); 1454 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count); 1455 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks); 1456 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) * 1457 flex_gd->count); 1458 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) * 1459 flex_gd->count); 1460 1461 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es)); 1462 /* 1463 * We need to protect s_groups_count against other CPUs seeing 1464 * inconsistent state in the superblock. 1465 * 1466 * The precise rules we use are: 1467 * 1468 * * Writers must perform a smp_wmb() after updating all 1469 * dependent data and before modifying the groups count 1470 * 1471 * * Readers must perform an smp_rmb() after reading the groups 1472 * count and before reading any dependent data. 1473 * 1474 * NB. These rules can be relaxed when checking the group count 1475 * while freeing data, as we can only allocate from a block 1476 * group after serialising against the group count, and we can 1477 * only then free after serialising in turn against that 1478 * allocation. 1479 */ 1480 smp_wmb(); 1481 1482 /* Update the global fs size fields */ 1483 sbi->s_groups_count += flex_gd->count; 1484 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 1485 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 1486 1487 /* Update the reserved block counts only once the new group is 1488 * active. */ 1489 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) + 1490 reserved_blocks); 1491 1492 /* Update the free space counts */ 1493 percpu_counter_add(&sbi->s_freeclusters_counter, 1494 EXT4_NUM_B2C(sbi, free_blocks)); 1495 percpu_counter_add(&sbi->s_freeinodes_counter, 1496 EXT4_INODES_PER_GROUP(sb) * flex_gd->count); 1497 1498 ext4_debug("free blocks count %llu", 1499 percpu_counter_read(&sbi->s_freeclusters_counter)); 1500 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) { 1501 ext4_group_t flex_group; 1502 struct flex_groups *fg; 1503 1504 flex_group = ext4_flex_group(sbi, group_data[0].group); 1505 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); 1506 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks), 1507 &fg->free_clusters); 1508 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count, 1509 &fg->free_inodes); 1510 } 1511 1512 /* 1513 * Update the fs overhead information. 1514 * 1515 * For bigalloc, if the superblock already has a properly calculated 1516 * overhead, update it with a value based on numbers already computed 1517 * above for the newly allocated capacity. 1518 */ 1519 if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0)) 1520 ext4_add_overhead(sb, 1521 EXT4_NUM_B2C(sbi, blocks_count - free_blocks)); 1522 else 1523 ext4_calculate_overhead(sb); 1524 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead); 1525 1526 ext4_superblock_csum_set(sb); 1527 unlock_buffer(sbi->s_sbh); 1528 if (test_opt(sb, DEBUG)) 1529 printk(KERN_DEBUG "EXT4-fs: added group %u:" 1530 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count, 1531 blocks_count, free_blocks, reserved_blocks); 1532} 1533 1534/* Add a flex group to an fs. Ensure we handle all possible error conditions 1535 * _before_ we start modifying the filesystem, because we cannot abort the 1536 * transaction and not have it write the data to disk. 1537 */ 1538static int ext4_flex_group_add(struct super_block *sb, 1539 struct inode *resize_inode, 1540 struct ext4_new_flex_group_data *flex_gd) 1541{ 1542 struct ext4_sb_info *sbi = EXT4_SB(sb); 1543 struct ext4_super_block *es = sbi->s_es; 1544 ext4_fsblk_t o_blocks_count; 1545 ext4_grpblk_t last; 1546 ext4_group_t group; 1547 handle_t *handle; 1548 unsigned reserved_gdb; 1549 int err = 0, err2 = 0, credit; 1550 1551 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags); 1552 1553 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks); 1554 o_blocks_count = ext4_blocks_count(es); 1555 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); 1556 BUG_ON(last); 1557 1558 err = setup_new_flex_group_blocks(sb, flex_gd); 1559 if (err) 1560 goto exit; 1561 /* 1562 * We will always be modifying at least the superblock and GDT 1563 * blocks. If we are adding a group past the last current GDT block, 1564 * we will also modify the inode and the dindirect block. If we 1565 * are adding a group with superblock/GDT backups we will also 1566 * modify each of the reserved GDT dindirect blocks. 1567 */ 1568 credit = 3; /* sb, resize inode, resize inode dindirect */ 1569 /* GDT blocks */ 1570 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb)); 1571 credit += reserved_gdb; /* Reserved GDT dindirect blocks */ 1572 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit); 1573 if (IS_ERR(handle)) { 1574 err = PTR_ERR(handle); 1575 goto exit; 1576 } 1577 1578 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 1579 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh, 1580 EXT4_JTR_NONE); 1581 if (err) 1582 goto exit_journal; 1583 1584 group = flex_gd->groups[0].group; 1585 BUG_ON(group != sbi->s_groups_count); 1586 err = ext4_add_new_descs(handle, sb, group, 1587 resize_inode, flex_gd->count); 1588 if (err) 1589 goto exit_journal; 1590 1591 err = ext4_setup_new_descs(handle, sb, flex_gd); 1592 if (err) 1593 goto exit_journal; 1594 1595 ext4_update_super(sb, flex_gd); 1596 1597 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); 1598 1599exit_journal: 1600 err2 = ext4_journal_stop(handle); 1601 if (!err) 1602 err = err2; 1603 1604 if (!err) { 1605 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 1606 int gdb_num_end = ((group + flex_gd->count - 1) / 1607 EXT4_DESC_PER_BLOCK(sb)); 1608 int meta_bg = ext4_has_feature_meta_bg(sb) && 1609 gdb_num >= le32_to_cpu(es->s_first_meta_bg); 1610 sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr - 1611 ext4_group_first_block_no(sb, 0); 1612 sector_t old_gdb = 0; 1613 1614 update_backups(sb, ext4_group_first_block_no(sb, 0), 1615 (char *)es, sizeof(struct ext4_super_block), 0); 1616 for (; gdb_num <= gdb_num_end; gdb_num++) { 1617 struct buffer_head *gdb_bh; 1618 1619 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, 1620 gdb_num); 1621 if (old_gdb == gdb_bh->b_blocknr) 1622 continue; 1623 update_backups(sb, gdb_bh->b_blocknr - padding_blocks, 1624 gdb_bh->b_data, gdb_bh->b_size, meta_bg); 1625 old_gdb = gdb_bh->b_blocknr; 1626 } 1627 } 1628exit: 1629 return err; 1630} 1631 1632static int ext4_setup_next_flex_gd(struct super_block *sb, 1633 struct ext4_new_flex_group_data *flex_gd, 1634 ext4_fsblk_t n_blocks_count) 1635{ 1636 struct ext4_sb_info *sbi = EXT4_SB(sb); 1637 struct ext4_super_block *es = sbi->s_es; 1638 struct ext4_new_group_data *group_data = flex_gd->groups; 1639 ext4_fsblk_t o_blocks_count; 1640 ext4_group_t n_group; 1641 ext4_group_t group; 1642 ext4_group_t last_group; 1643 ext4_grpblk_t last; 1644 ext4_grpblk_t clusters_per_group; 1645 unsigned long i; 1646 1647 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb); 1648 1649 o_blocks_count = ext4_blocks_count(es); 1650 1651 if (o_blocks_count == n_blocks_count) 1652 return 0; 1653 1654 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); 1655 BUG_ON(last); 1656 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last); 1657 1658 last_group = group | (flex_gd->resize_bg - 1); 1659 if (last_group > n_group) 1660 last_group = n_group; 1661 1662 flex_gd->count = last_group - group + 1; 1663 1664 for (i = 0; i < flex_gd->count; i++) { 1665 int overhead; 1666 1667 group_data[i].group = group + i; 1668 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb); 1669 overhead = ext4_group_overhead_blocks(sb, group + i); 1670 group_data[i].mdata_blocks = overhead; 1671 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb); 1672 if (ext4_has_group_desc_csum(sb)) { 1673 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT | 1674 EXT4_BG_INODE_UNINIT; 1675 if (!test_opt(sb, INIT_INODE_TABLE)) 1676 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED; 1677 } else 1678 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED; 1679 } 1680 1681 if (last_group == n_group && ext4_has_group_desc_csum(sb)) 1682 /* We need to initialize block bitmap of last group. */ 1683 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT; 1684 1685 if ((last_group == n_group) && (last != clusters_per_group - 1)) { 1686 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1); 1687 group_data[i - 1].free_clusters_count -= clusters_per_group - 1688 last - 1; 1689 } 1690 1691 return 1; 1692} 1693 1694/* Add group descriptor data to an existing or new group descriptor block. 1695 * Ensure we handle all possible error conditions _before_ we start modifying 1696 * the filesystem, because we cannot abort the transaction and not have it 1697 * write the data to disk. 1698 * 1699 * If we are on a GDT block boundary, we need to get the reserved GDT block. 1700 * Otherwise, we may need to add backup GDT blocks for a sparse group. 1701 * 1702 * We only need to hold the superblock lock while we are actually adding 1703 * in the new group's counts to the superblock. Prior to that we have 1704 * not really "added" the group at all. We re-check that we are still 1705 * adding in the last group in case things have changed since verifying. 1706 */ 1707int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) 1708{ 1709 struct ext4_new_flex_group_data flex_gd; 1710 struct ext4_sb_info *sbi = EXT4_SB(sb); 1711 struct ext4_super_block *es = sbi->s_es; 1712 int reserved_gdb = ext4_bg_has_super(sb, input->group) ? 1713 le16_to_cpu(es->s_reserved_gdt_blocks) : 0; 1714 struct inode *inode = NULL; 1715 int gdb_off; 1716 int err; 1717 __u16 bg_flags = 0; 1718 1719 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb); 1720 1721 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) { 1722 ext4_warning(sb, "Can't resize non-sparse filesystem further"); 1723 return -EPERM; 1724 } 1725 1726 if (ext4_blocks_count(es) + input->blocks_count < 1727 ext4_blocks_count(es)) { 1728 ext4_warning(sb, "blocks_count overflow"); 1729 return -EINVAL; 1730 } 1731 1732 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) < 1733 le32_to_cpu(es->s_inodes_count)) { 1734 ext4_warning(sb, "inodes_count overflow"); 1735 return -EINVAL; 1736 } 1737 1738 if (reserved_gdb || gdb_off == 0) { 1739 if (!ext4_has_feature_resize_inode(sb) || 1740 !le16_to_cpu(es->s_reserved_gdt_blocks)) { 1741 ext4_warning(sb, 1742 "No reserved GDT blocks, can't resize"); 1743 return -EPERM; 1744 } 1745 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL); 1746 if (IS_ERR(inode)) { 1747 ext4_warning(sb, "Error opening resize inode"); 1748 return PTR_ERR(inode); 1749 } 1750 } 1751 1752 1753 err = verify_group_input(sb, input); 1754 if (err) 1755 goto out; 1756 1757 err = ext4_alloc_flex_bg_array(sb, input->group + 1); 1758 if (err) 1759 goto out; 1760 1761 err = ext4_mb_alloc_groupinfo(sb, input->group + 1); 1762 if (err) 1763 goto out; 1764 1765 flex_gd.count = 1; 1766 flex_gd.groups = input; 1767 flex_gd.bg_flags = &bg_flags; 1768 err = ext4_flex_group_add(sb, inode, &flex_gd); 1769out: 1770 iput(inode); 1771 return err; 1772} /* ext4_group_add */ 1773 1774/* 1775 * extend a group without checking assuming that checking has been done. 1776 */ 1777static int ext4_group_extend_no_check(struct super_block *sb, 1778 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add) 1779{ 1780 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1781 handle_t *handle; 1782 int err = 0, err2; 1783 1784 /* We will update the superblock, one block bitmap, and 1785 * one group descriptor via ext4_group_add_blocks(). 1786 */ 1787 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3); 1788 if (IS_ERR(handle)) { 1789 err = PTR_ERR(handle); 1790 ext4_warning(sb, "error %d on journal start", err); 1791 return err; 1792 } 1793 1794 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); 1795 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh, 1796 EXT4_JTR_NONE); 1797 if (err) { 1798 ext4_warning(sb, "error %d on journal write access", err); 1799 goto errout; 1800 } 1801 1802 lock_buffer(EXT4_SB(sb)->s_sbh); 1803 ext4_blocks_count_set(es, o_blocks_count + add); 1804 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add); 1805 ext4_superblock_csum_set(sb); 1806 unlock_buffer(EXT4_SB(sb)->s_sbh); 1807 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count, 1808 o_blocks_count + add); 1809 /* We add the blocks to the bitmap and set the group need init bit */ 1810 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add); 1811 if (err) 1812 goto errout; 1813 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); 1814 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count, 1815 o_blocks_count + add); 1816errout: 1817 err2 = ext4_journal_stop(handle); 1818 if (err2 && !err) 1819 err = err2; 1820 1821 if (!err) { 1822 if (test_opt(sb, DEBUG)) 1823 printk(KERN_DEBUG "EXT4-fs: extended group to %llu " 1824 "blocks\n", ext4_blocks_count(es)); 1825 update_backups(sb, ext4_group_first_block_no(sb, 0), 1826 (char *)es, sizeof(struct ext4_super_block), 0); 1827 } 1828 return err; 1829} 1830 1831/* 1832 * Extend the filesystem to the new number of blocks specified. This entry 1833 * point is only used to extend the current filesystem to the end of the last 1834 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>" 1835 * for emergencies (because it has no dependencies on reserved blocks). 1836 * 1837 * If we _really_ wanted, we could use default values to call ext4_group_add() 1838 * allow the "remount" trick to work for arbitrary resizing, assuming enough 1839 * GDT blocks are reserved to grow to the desired size. 1840 */ 1841int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, 1842 ext4_fsblk_t n_blocks_count) 1843{ 1844 ext4_fsblk_t o_blocks_count; 1845 ext4_grpblk_t last; 1846 ext4_grpblk_t add; 1847 struct buffer_head *bh; 1848 ext4_group_t group; 1849 1850 o_blocks_count = ext4_blocks_count(es); 1851 1852 if (test_opt(sb, DEBUG)) 1853 ext4_msg(sb, KERN_DEBUG, 1854 "extending last group from %llu to %llu blocks", 1855 o_blocks_count, n_blocks_count); 1856 1857 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count) 1858 return 0; 1859 1860 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) { 1861 ext4_msg(sb, KERN_ERR, 1862 "filesystem too large to resize to %llu blocks safely", 1863 n_blocks_count); 1864 return -EINVAL; 1865 } 1866 1867 if (n_blocks_count < o_blocks_count) { 1868 ext4_warning(sb, "can't shrink FS - resize aborted"); 1869 return -EINVAL; 1870 } 1871 1872 /* Handle the remaining blocks in the last group only. */ 1873 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); 1874 1875 if (last == 0) { 1876 ext4_warning(sb, "need to use ext2online to resize further"); 1877 return -EPERM; 1878 } 1879 1880 add = EXT4_BLOCKS_PER_GROUP(sb) - last; 1881 1882 if (o_blocks_count + add < o_blocks_count) { 1883 ext4_warning(sb, "blocks_count overflow"); 1884 return -EINVAL; 1885 } 1886 1887 if (o_blocks_count + add > n_blocks_count) 1888 add = n_blocks_count - o_blocks_count; 1889 1890 if (o_blocks_count + add < n_blocks_count) 1891 ext4_warning(sb, "will only finish group (%llu blocks, %u new)", 1892 o_blocks_count + add, add); 1893 1894 /* See if the device is actually as big as what was requested */ 1895 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0); 1896 if (IS_ERR(bh)) { 1897 ext4_warning(sb, "can't read last block, resize aborted"); 1898 return -ENOSPC; 1899 } 1900 brelse(bh); 1901 1902 return ext4_group_extend_no_check(sb, o_blocks_count, add); 1903} /* ext4_group_extend */ 1904 1905 1906static int num_desc_blocks(struct super_block *sb, ext4_group_t groups) 1907{ 1908 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb); 1909} 1910 1911/* 1912 * Release the resize inode and drop the resize_inode feature if there 1913 * are no more reserved gdt blocks, and then convert the file system 1914 * to enable meta_bg 1915 */ 1916static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode) 1917{ 1918 handle_t *handle; 1919 struct ext4_sb_info *sbi = EXT4_SB(sb); 1920 struct ext4_super_block *es = sbi->s_es; 1921 struct ext4_inode_info *ei = EXT4_I(inode); 1922 ext4_fsblk_t nr; 1923 int i, ret, err = 0; 1924 int credits = 1; 1925 1926 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg"); 1927 if (inode) { 1928 if (es->s_reserved_gdt_blocks) { 1929 ext4_error(sb, "Unexpected non-zero " 1930 "s_reserved_gdt_blocks"); 1931 return -EPERM; 1932 } 1933 1934 /* Do a quick sanity check of the resize inode */ 1935 if (inode->i_blocks != 1 << (inode->i_blkbits - 1936 (9 - sbi->s_cluster_bits))) 1937 goto invalid_resize_inode; 1938 for (i = 0; i < EXT4_N_BLOCKS; i++) { 1939 if (i == EXT4_DIND_BLOCK) { 1940 if (ei->i_data[i]) 1941 continue; 1942 else 1943 goto invalid_resize_inode; 1944 } 1945 if (ei->i_data[i]) 1946 goto invalid_resize_inode; 1947 } 1948 credits += 3; /* block bitmap, bg descriptor, resize inode */ 1949 } 1950 1951 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits); 1952 if (IS_ERR(handle)) 1953 return PTR_ERR(handle); 1954 1955 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 1956 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh, 1957 EXT4_JTR_NONE); 1958 if (err) 1959 goto errout; 1960 1961 lock_buffer(sbi->s_sbh); 1962 ext4_clear_feature_resize_inode(sb); 1963 ext4_set_feature_meta_bg(sb); 1964 sbi->s_es->s_first_meta_bg = 1965 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count)); 1966 ext4_superblock_csum_set(sb); 1967 unlock_buffer(sbi->s_sbh); 1968 1969 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); 1970 if (err) { 1971 ext4_std_error(sb, err); 1972 goto errout; 1973 } 1974 1975 if (inode) { 1976 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]); 1977 ext4_free_blocks(handle, inode, NULL, nr, 1, 1978 EXT4_FREE_BLOCKS_METADATA | 1979 EXT4_FREE_BLOCKS_FORGET); 1980 ei->i_data[EXT4_DIND_BLOCK] = 0; 1981 inode->i_blocks = 0; 1982 1983 err = ext4_mark_inode_dirty(handle, inode); 1984 if (err) 1985 ext4_std_error(sb, err); 1986 } 1987 1988errout: 1989 ret = ext4_journal_stop(handle); 1990 return err ? err : ret; 1991 1992invalid_resize_inode: 1993 ext4_error(sb, "corrupted/inconsistent resize inode"); 1994 return -EINVAL; 1995} 1996 1997/* 1998 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count 1999 * 2000 * @sb: super block of the fs to be resized 2001 * @n_blocks_count: the number of blocks resides in the resized fs 2002 */ 2003int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count) 2004{ 2005 struct ext4_new_flex_group_data *flex_gd = NULL; 2006 struct ext4_sb_info *sbi = EXT4_SB(sb); 2007 struct ext4_super_block *es = sbi->s_es; 2008 struct buffer_head *bh; 2009 struct inode *resize_inode = NULL; 2010 ext4_grpblk_t add, offset; 2011 unsigned long n_desc_blocks; 2012 unsigned long o_desc_blocks; 2013 ext4_group_t o_group; 2014 ext4_group_t n_group; 2015 ext4_fsblk_t o_blocks_count; 2016 ext4_fsblk_t n_blocks_count_retry = 0; 2017 unsigned long last_update_time = 0; 2018 int err = 0; 2019 int meta_bg; 2020 unsigned int flexbg_size = ext4_flex_bg_size(sbi); 2021 2022 /* See if the device is actually as big as what was requested */ 2023 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0); 2024 if (IS_ERR(bh)) { 2025 ext4_warning(sb, "can't read last block, resize aborted"); 2026 return -ENOSPC; 2027 } 2028 brelse(bh); 2029 2030 /* 2031 * For bigalloc, trim the requested size to the nearest cluster 2032 * boundary to avoid creating an unusable filesystem. We do this 2033 * silently, instead of returning an error, to avoid breaking 2034 * callers that blindly resize the filesystem to the full size of 2035 * the underlying block device. 2036 */ 2037 if (ext4_has_feature_bigalloc(sb)) 2038 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1); 2039 2040retry: 2041 o_blocks_count = ext4_blocks_count(es); 2042 2043 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu " 2044 "to %llu blocks", o_blocks_count, n_blocks_count); 2045 2046 if (n_blocks_count < o_blocks_count) { 2047 /* On-line shrinking not supported */ 2048 ext4_warning(sb, "can't shrink FS - resize aborted"); 2049 return -EINVAL; 2050 } 2051 2052 if (n_blocks_count == o_blocks_count) 2053 /* Nothing need to do */ 2054 return 0; 2055 2056 n_group = ext4_get_group_number(sb, n_blocks_count - 1); 2057 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) { 2058 ext4_warning(sb, "resize would cause inodes_count overflow"); 2059 return -EINVAL; 2060 } 2061 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset); 2062 2063 n_desc_blocks = num_desc_blocks(sb, n_group + 1); 2064 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count); 2065 2066 meta_bg = ext4_has_feature_meta_bg(sb); 2067 2068 if (ext4_has_feature_resize_inode(sb)) { 2069 if (meta_bg) { 2070 ext4_error(sb, "resize_inode and meta_bg enabled " 2071 "simultaneously"); 2072 return -EINVAL; 2073 } 2074 if (n_desc_blocks > o_desc_blocks + 2075 le16_to_cpu(es->s_reserved_gdt_blocks)) { 2076 n_blocks_count_retry = n_blocks_count; 2077 n_desc_blocks = o_desc_blocks + 2078 le16_to_cpu(es->s_reserved_gdt_blocks); 2079 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb); 2080 n_blocks_count = (ext4_fsblk_t)n_group * 2081 EXT4_BLOCKS_PER_GROUP(sb) + 2082 le32_to_cpu(es->s_first_data_block); 2083 n_group--; /* set to last group number */ 2084 } 2085 2086 if (!resize_inode) 2087 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO, 2088 EXT4_IGET_SPECIAL); 2089 if (IS_ERR(resize_inode)) { 2090 ext4_warning(sb, "Error opening resize inode"); 2091 return PTR_ERR(resize_inode); 2092 } 2093 } 2094 2095 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) { 2096 err = ext4_convert_meta_bg(sb, resize_inode); 2097 if (err) 2098 goto out; 2099 if (resize_inode) { 2100 iput(resize_inode); 2101 resize_inode = NULL; 2102 } 2103 if (n_blocks_count_retry) { 2104 n_blocks_count = n_blocks_count_retry; 2105 n_blocks_count_retry = 0; 2106 goto retry; 2107 } 2108 } 2109 2110 /* 2111 * Make sure the last group has enough space so that it's 2112 * guaranteed to have enough space for all metadata blocks 2113 * that it might need to hold. (We might not need to store 2114 * the inode table blocks in the last block group, but there 2115 * will be cases where this might be needed.) 2116 */ 2117 if ((ext4_group_first_block_no(sb, n_group) + 2118 ext4_group_overhead_blocks(sb, n_group) + 2 + 2119 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) { 2120 n_blocks_count = ext4_group_first_block_no(sb, n_group); 2121 n_group--; 2122 n_blocks_count_retry = 0; 2123 if (resize_inode) { 2124 iput(resize_inode); 2125 resize_inode = NULL; 2126 } 2127 goto retry; 2128 } 2129 2130 /* extend the last group */ 2131 if (n_group == o_group) 2132 add = n_blocks_count - o_blocks_count; 2133 else 2134 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1)); 2135 if (add > 0) { 2136 err = ext4_group_extend_no_check(sb, o_blocks_count, add); 2137 if (err) 2138 goto out; 2139 } 2140 2141 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0) 2142 goto out; 2143 2144 err = ext4_alloc_flex_bg_array(sb, n_group + 1); 2145 if (err) 2146 goto out; 2147 2148 err = ext4_mb_alloc_groupinfo(sb, n_group + 1); 2149 if (err) 2150 goto out; 2151 2152 flex_gd = alloc_flex_gd(flexbg_size); 2153 if (flex_gd == NULL) { 2154 err = -ENOMEM; 2155 goto out; 2156 } 2157 2158 /* Add flex groups. Note that a regular group is a 2159 * flex group with 1 group. 2160 */ 2161 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) { 2162 if (time_is_before_jiffies(last_update_time + HZ * 10)) { 2163 if (last_update_time) 2164 ext4_msg(sb, KERN_INFO, 2165 "resized to %llu blocks", 2166 ext4_blocks_count(es)); 2167 last_update_time = jiffies; 2168 } 2169 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0) 2170 break; 2171 err = ext4_flex_group_add(sb, resize_inode, flex_gd); 2172 if (unlikely(err)) 2173 break; 2174 } 2175 2176 if (!err && n_blocks_count_retry) { 2177 n_blocks_count = n_blocks_count_retry; 2178 n_blocks_count_retry = 0; 2179 free_flex_gd(flex_gd); 2180 flex_gd = NULL; 2181 if (resize_inode) { 2182 iput(resize_inode); 2183 resize_inode = NULL; 2184 } 2185 goto retry; 2186 } 2187 2188out: 2189 if (flex_gd) 2190 free_flex_gd(flex_gd); 2191 if (resize_inode != NULL) 2192 iput(resize_inode); 2193 if (err) 2194 ext4_warning(sb, "error (%d) occurred during " 2195 "file system resize", err); 2196 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", 2197 ext4_blocks_count(es)); 2198 return err; 2199} 2200