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 18 #include "ext4_jbd2.h" 19 20 int ext4_resize_begin(struct super_block *sb) 21 { 22 struct ext4_sb_info *sbi = EXT4_SB(sb); 23 int ret = 0; 24 25 if (!capable(CAP_SYS_RESOURCE)) 26 return -EPERM; 27 28 /* 29 * If we are not using the primary superblock/GDT copy don't resize, 30 * because the user tools have no way of handling this. Probably a 31 * bad time to do it anyways. 32 */ 33 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) != 34 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) { 35 ext4_warning(sb, "won't resize using backup superblock at %llu", 36 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr); 37 return -EPERM; 38 } 39 40 /* 41 * We are not allowed to do online-resizing on a filesystem mounted 42 * with error, because it can destroy the filesystem easily. 43 */ 44 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { 45 ext4_warning(sb, "There are errors in the filesystem, " 46 "so online resizing is not allowed"); 47 return -EPERM; 48 } 49 50 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING, 51 &EXT4_SB(sb)->s_ext4_flags)) 52 ret = -EBUSY; 53 54 return ret; 55 } 56 57 void ext4_resize_end(struct super_block *sb) 58 { 59 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags); 60 smp_mb__after_atomic(); 61 } 62 63 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb, 64 ext4_group_t group) { 65 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) << 66 EXT4_DESC_PER_BLOCK_BITS(sb); 67 } 68 69 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb, 70 ext4_group_t group) { 71 group = ext4_meta_bg_first_group(sb, group); 72 return ext4_group_first_block_no(sb, group); 73 } 74 75 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb, 76 ext4_group_t group) { 77 ext4_grpblk_t overhead; 78 overhead = ext4_bg_num_gdb(sb, group); 79 if (ext4_bg_has_super(sb, group)) 80 overhead += 1 + 81 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); 82 return overhead; 83 } 84 85 #define outside(b, first, last) ((b) < (first) || (b) >= (last)) 86 #define inside(b, first, last) ((b) >= (first) && (b) < (last)) 87 88 static int verify_group_input(struct super_block *sb, 89 struct ext4_new_group_data *input) 90 { 91 struct ext4_sb_info *sbi = EXT4_SB(sb); 92 struct ext4_super_block *es = sbi->s_es; 93 ext4_fsblk_t start = ext4_blocks_count(es); 94 ext4_fsblk_t end = start + input->blocks_count; 95 ext4_group_t group = input->group; 96 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group; 97 unsigned overhead; 98 ext4_fsblk_t metaend; 99 struct buffer_head *bh = NULL; 100 ext4_grpblk_t free_blocks_count, offset; 101 int err = -EINVAL; 102 103 if (group != sbi->s_groups_count) { 104 ext4_warning(sb, "Cannot add at group %u (only %u groups)", 105 input->group, sbi->s_groups_count); 106 return -EINVAL; 107 } 108 109 overhead = ext4_group_overhead_blocks(sb, group); 110 metaend = start + overhead; 111 input->free_clusters_count = free_blocks_count = 112 input->blocks_count - 2 - overhead - sbi->s_itb_per_group; 113 114 if (test_opt(sb, DEBUG)) 115 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks " 116 "(%d free, %u reserved)\n", 117 ext4_bg_has_super(sb, input->group) ? "normal" : 118 "no-super", input->group, input->blocks_count, 119 free_blocks_count, input->reserved_blocks); 120 121 ext4_get_group_no_and_offset(sb, start, NULL, &offset); 122 if (offset != 0) 123 ext4_warning(sb, "Last group not full"); 124 else if (input->reserved_blocks > input->blocks_count / 5) 125 ext4_warning(sb, "Reserved blocks too high (%u)", 126 input->reserved_blocks); 127 else if (free_blocks_count < 0) 128 ext4_warning(sb, "Bad blocks count %u", 129 input->blocks_count); 130 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) { 131 err = PTR_ERR(bh); 132 bh = NULL; 133 ext4_warning(sb, "Cannot read last block (%llu)", 134 end - 1); 135 } else if (outside(input->block_bitmap, start, end)) 136 ext4_warning(sb, "Block bitmap not in group (block %llu)", 137 (unsigned long long)input->block_bitmap); 138 else if (outside(input->inode_bitmap, start, end)) 139 ext4_warning(sb, "Inode bitmap not in group (block %llu)", 140 (unsigned long long)input->inode_bitmap); 141 else if (outside(input->inode_table, start, end) || 142 outside(itend - 1, start, end)) 143 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)", 144 (unsigned long long)input->inode_table, itend - 1); 145 else if (input->inode_bitmap == input->block_bitmap) 146 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)", 147 (unsigned long long)input->block_bitmap); 148 else if (inside(input->block_bitmap, input->inode_table, itend)) 149 ext4_warning(sb, "Block bitmap (%llu) in inode table " 150 "(%llu-%llu)", 151 (unsigned long long)input->block_bitmap, 152 (unsigned long long)input->inode_table, itend - 1); 153 else if (inside(input->inode_bitmap, input->inode_table, itend)) 154 ext4_warning(sb, "Inode bitmap (%llu) in inode table " 155 "(%llu-%llu)", 156 (unsigned long long)input->inode_bitmap, 157 (unsigned long long)input->inode_table, itend - 1); 158 else if (inside(input->block_bitmap, start, metaend)) 159 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)", 160 (unsigned long long)input->block_bitmap, 161 start, metaend - 1); 162 else if (inside(input->inode_bitmap, start, metaend)) 163 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)", 164 (unsigned long long)input->inode_bitmap, 165 start, metaend - 1); 166 else if (inside(input->inode_table, start, metaend) || 167 inside(itend - 1, start, metaend)) 168 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table " 169 "(%llu-%llu)", 170 (unsigned long long)input->inode_table, 171 itend - 1, start, metaend - 1); 172 else 173 err = 0; 174 brelse(bh); 175 176 return err; 177 } 178 179 /* 180 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex 181 * group each time. 182 */ 183 struct ext4_new_flex_group_data { 184 struct ext4_new_group_data *groups; /* new_group_data for groups 185 in the flex group */ 186 __u16 *bg_flags; /* block group flags of groups 187 in @groups */ 188 ext4_group_t count; /* number of groups in @groups 189 */ 190 }; 191 192 /* 193 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of 194 * @flexbg_size. 195 * 196 * Returns NULL on failure otherwise address of the allocated structure. 197 */ 198 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size) 199 { 200 struct ext4_new_flex_group_data *flex_gd; 201 202 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS); 203 if (flex_gd == NULL) 204 goto out3; 205 206 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data)) 207 goto out2; 208 flex_gd->count = flexbg_size; 209 210 flex_gd->groups = kmalloc_array(flexbg_size, 211 sizeof(struct ext4_new_group_data), 212 GFP_NOFS); 213 if (flex_gd->groups == NULL) 214 goto out2; 215 216 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16), 217 GFP_NOFS); 218 if (flex_gd->bg_flags == NULL) 219 goto out1; 220 221 return flex_gd; 222 223 out1: 224 kfree(flex_gd->groups); 225 out2: 226 kfree(flex_gd); 227 out3: 228 return NULL; 229 } 230 231 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd) 232 { 233 kfree(flex_gd->bg_flags); 234 kfree(flex_gd->groups); 235 kfree(flex_gd); 236 } 237 238 /* 239 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps 240 * and inode tables for a flex group. 241 * 242 * This function is used by 64bit-resize. Note that this function allocates 243 * group tables from the 1st group of groups contained by @flexgd, which may 244 * be a partial of a flex group. 245 * 246 * @sb: super block of fs to which the groups belongs 247 * 248 * Returns 0 on a successful allocation of the metadata blocks in the 249 * block group. 250 */ 251 static int ext4_alloc_group_tables(struct super_block *sb, 252 struct ext4_new_flex_group_data *flex_gd, 253 int flexbg_size) 254 { 255 struct ext4_new_group_data *group_data = flex_gd->groups; 256 ext4_fsblk_t start_blk; 257 ext4_fsblk_t last_blk; 258 ext4_group_t src_group; 259 ext4_group_t bb_index = 0; 260 ext4_group_t ib_index = 0; 261 ext4_group_t it_index = 0; 262 ext4_group_t group; 263 ext4_group_t last_group; 264 unsigned overhead; 265 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0; 266 int i; 267 268 BUG_ON(flex_gd->count == 0 || group_data == NULL); 269 270 src_group = group_data[0].group; 271 last_group = src_group + flex_gd->count - 1; 272 273 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) != 274 (last_group & ~(flexbg_size - 1)))); 275 next_group: 276 group = group_data[0].group; 277 if (src_group >= group_data[0].group + flex_gd->count) 278 return -ENOSPC; 279 start_blk = ext4_group_first_block_no(sb, src_group); 280 last_blk = start_blk + group_data[src_group - group].blocks_count; 281 282 overhead = ext4_group_overhead_blocks(sb, src_group); 283 284 start_blk += overhead; 285 286 /* We collect contiguous blocks as much as possible. */ 287 src_group++; 288 for (; src_group <= last_group; src_group++) { 289 overhead = ext4_group_overhead_blocks(sb, src_group); 290 if (overhead == 0) 291 last_blk += group_data[src_group - group].blocks_count; 292 else 293 break; 294 } 295 296 /* Allocate block bitmaps */ 297 for (; bb_index < flex_gd->count; bb_index++) { 298 if (start_blk >= last_blk) 299 goto next_group; 300 group_data[bb_index].block_bitmap = start_blk++; 301 group = ext4_get_group_number(sb, start_blk - 1); 302 group -= group_data[0].group; 303 group_data[group].mdata_blocks++; 304 flex_gd->bg_flags[group] &= uninit_mask; 305 } 306 307 /* Allocate inode bitmaps */ 308 for (; ib_index < flex_gd->count; ib_index++) { 309 if (start_blk >= last_blk) 310 goto next_group; 311 group_data[ib_index].inode_bitmap = start_blk++; 312 group = ext4_get_group_number(sb, start_blk - 1); 313 group -= group_data[0].group; 314 group_data[group].mdata_blocks++; 315 flex_gd->bg_flags[group] &= uninit_mask; 316 } 317 318 /* Allocate inode tables */ 319 for (; it_index < flex_gd->count; it_index++) { 320 unsigned int itb = EXT4_SB(sb)->s_itb_per_group; 321 ext4_fsblk_t next_group_start; 322 323 if (start_blk + itb > last_blk) 324 goto next_group; 325 group_data[it_index].inode_table = start_blk; 326 group = ext4_get_group_number(sb, start_blk); 327 next_group_start = ext4_group_first_block_no(sb, group + 1); 328 group -= group_data[0].group; 329 330 if (start_blk + itb > next_group_start) { 331 flex_gd->bg_flags[group + 1] &= uninit_mask; 332 overhead = start_blk + itb - next_group_start; 333 group_data[group + 1].mdata_blocks += overhead; 334 itb -= overhead; 335 } 336 337 group_data[group].mdata_blocks += itb; 338 flex_gd->bg_flags[group] &= uninit_mask; 339 start_blk += EXT4_SB(sb)->s_itb_per_group; 340 } 341 342 /* Update free clusters count to exclude metadata blocks */ 343 for (i = 0; i < flex_gd->count; i++) { 344 group_data[i].free_clusters_count -= 345 EXT4_NUM_B2C(EXT4_SB(sb), 346 group_data[i].mdata_blocks); 347 } 348 349 if (test_opt(sb, DEBUG)) { 350 int i; 351 group = group_data[0].group; 352 353 printk(KERN_DEBUG "EXT4-fs: adding a flex group with " 354 "%d groups, flexbg size is %d:\n", flex_gd->count, 355 flexbg_size); 356 357 for (i = 0; i < flex_gd->count; i++) { 358 ext4_debug( 359 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n", 360 ext4_bg_has_super(sb, group + i) ? "normal" : 361 "no-super", group + i, 362 group_data[i].blocks_count, 363 group_data[i].free_clusters_count, 364 group_data[i].mdata_blocks); 365 } 366 } 367 return 0; 368 } 369 370 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb, 371 ext4_fsblk_t blk) 372 { 373 struct buffer_head *bh; 374 int err; 375 376 bh = sb_getblk(sb, blk); 377 if (unlikely(!bh)) 378 return ERR_PTR(-ENOMEM); 379 BUFFER_TRACE(bh, "get_write_access"); 380 if ((err = ext4_journal_get_write_access(handle, bh))) { 381 brelse(bh); 382 bh = ERR_PTR(err); 383 } else { 384 memset(bh->b_data, 0, sb->s_blocksize); 385 set_buffer_uptodate(bh); 386 } 387 388 return bh; 389 } 390 391 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits) 392 { 393 return ext4_journal_ensure_credits_fn(handle, credits, 394 EXT4_MAX_TRANS_DATA, 0, 0); 395 } 396 397 /* 398 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used. 399 * 400 * Helper function for ext4_setup_new_group_blocks() which set . 401 * 402 * @sb: super block 403 * @handle: journal handle 404 * @flex_gd: flex group data 405 */ 406 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle, 407 struct ext4_new_flex_group_data *flex_gd, 408 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster) 409 { 410 struct ext4_sb_info *sbi = EXT4_SB(sb); 411 ext4_group_t count = last_cluster - first_cluster + 1; 412 ext4_group_t count2; 413 414 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster, 415 last_cluster); 416 for (count2 = count; count > 0; 417 count -= count2, first_cluster += count2) { 418 ext4_fsblk_t start; 419 struct buffer_head *bh; 420 ext4_group_t group; 421 int err; 422 423 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster)); 424 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group)); 425 group -= flex_gd->groups[0].group; 426 427 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start); 428 if (count2 > count) 429 count2 = count; 430 431 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) { 432 BUG_ON(flex_gd->count > 1); 433 continue; 434 } 435 436 err = ext4_resize_ensure_credits_batch(handle, 1); 437 if (err < 0) 438 return err; 439 440 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap); 441 if (unlikely(!bh)) 442 return -ENOMEM; 443 444 BUFFER_TRACE(bh, "get_write_access"); 445 err = ext4_journal_get_write_access(handle, bh); 446 if (err) { 447 brelse(bh); 448 return err; 449 } 450 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", 451 first_cluster, first_cluster - start, count2); 452 ext4_set_bits(bh->b_data, first_cluster - start, count2); 453 454 err = ext4_handle_dirty_metadata(handle, NULL, bh); 455 brelse(bh); 456 if (unlikely(err)) 457 return err; 458 } 459 460 return 0; 461 } 462 463 /* 464 * Set up the block and inode bitmaps, and the inode table for the new groups. 465 * This doesn't need to be part of the main transaction, since we are only 466 * changing blocks outside the actual filesystem. We still do journaling to 467 * ensure the recovery is correct in case of a failure just after resize. 468 * If any part of this fails, we simply abort the resize. 469 * 470 * setup_new_flex_group_blocks handles a flex group as follow: 471 * 1. copy super block and GDT, and initialize group tables if necessary. 472 * In this step, we only set bits in blocks bitmaps for blocks taken by 473 * super block and GDT. 474 * 2. allocate group tables in block bitmaps, that is, set bits in block 475 * bitmap for blocks taken by group tables. 476 */ 477 static int setup_new_flex_group_blocks(struct super_block *sb, 478 struct ext4_new_flex_group_data *flex_gd) 479 { 480 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group}; 481 ext4_fsblk_t start; 482 ext4_fsblk_t block; 483 struct ext4_sb_info *sbi = EXT4_SB(sb); 484 struct ext4_super_block *es = sbi->s_es; 485 struct ext4_new_group_data *group_data = flex_gd->groups; 486 __u16 *bg_flags = flex_gd->bg_flags; 487 handle_t *handle; 488 ext4_group_t group, count; 489 struct buffer_head *bh = NULL; 490 int reserved_gdb, i, j, err = 0, err2; 491 int meta_bg; 492 493 BUG_ON(!flex_gd->count || !group_data || 494 group_data[0].group != sbi->s_groups_count); 495 496 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks); 497 meta_bg = ext4_has_feature_meta_bg(sb); 498 499 /* This transaction may be extended/restarted along the way */ 500 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA); 501 if (IS_ERR(handle)) 502 return PTR_ERR(handle); 503 504 group = group_data[0].group; 505 for (i = 0; i < flex_gd->count; i++, group++) { 506 unsigned long gdblocks; 507 ext4_grpblk_t overhead; 508 509 gdblocks = ext4_bg_num_gdb(sb, group); 510 start = ext4_group_first_block_no(sb, group); 511 512 if (meta_bg == 0 && !ext4_bg_has_super(sb, group)) 513 goto handle_itb; 514 515 if (meta_bg == 1) { 516 ext4_group_t first_group; 517 first_group = ext4_meta_bg_first_group(sb, group); 518 if (first_group != group + 1 && 519 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1) 520 goto handle_itb; 521 } 522 523 block = start + ext4_bg_has_super(sb, group); 524 /* Copy all of the GDT blocks into the backup in this group */ 525 for (j = 0; j < gdblocks; j++, block++) { 526 struct buffer_head *gdb; 527 528 ext4_debug("update backup group %#04llx\n", block); 529 err = ext4_resize_ensure_credits_batch(handle, 1); 530 if (err < 0) 531 goto out; 532 533 gdb = sb_getblk(sb, block); 534 if (unlikely(!gdb)) { 535 err = -ENOMEM; 536 goto out; 537 } 538 539 BUFFER_TRACE(gdb, "get_write_access"); 540 err = ext4_journal_get_write_access(handle, gdb); 541 if (err) { 542 brelse(gdb); 543 goto out; 544 } 545 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data, 546 gdb->b_size); 547 set_buffer_uptodate(gdb); 548 549 err = ext4_handle_dirty_metadata(handle, NULL, gdb); 550 if (unlikely(err)) { 551 brelse(gdb); 552 goto out; 553 } 554 brelse(gdb); 555 } 556 557 /* Zero out all of the reserved backup group descriptor 558 * table blocks 559 */ 560 if (ext4_bg_has_super(sb, group)) { 561 err = sb_issue_zeroout(sb, gdblocks + start + 1, 562 reserved_gdb, GFP_NOFS); 563 if (err) 564 goto out; 565 } 566 567 handle_itb: 568 /* Initialize group tables of the grop @group */ 569 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED)) 570 goto handle_bb; 571 572 /* Zero out all of the inode table blocks */ 573 block = group_data[i].inode_table; 574 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n", 575 block, sbi->s_itb_per_group); 576 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, 577 GFP_NOFS); 578 if (err) 579 goto out; 580 581 handle_bb: 582 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT) 583 goto handle_ib; 584 585 /* Initialize block bitmap of the @group */ 586 block = group_data[i].block_bitmap; 587 err = ext4_resize_ensure_credits_batch(handle, 1); 588 if (err < 0) 589 goto out; 590 591 bh = bclean(handle, sb, block); 592 if (IS_ERR(bh)) { 593 err = PTR_ERR(bh); 594 goto out; 595 } 596 overhead = ext4_group_overhead_blocks(sb, group); 597 if (overhead != 0) { 598 ext4_debug("mark backup superblock %#04llx (+0)\n", 599 start); 600 ext4_set_bits(bh->b_data, 0, 601 EXT4_NUM_B2C(sbi, overhead)); 602 } 603 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count), 604 sb->s_blocksize * 8, bh->b_data); 605 err = ext4_handle_dirty_metadata(handle, NULL, bh); 606 brelse(bh); 607 if (err) 608 goto out; 609 610 handle_ib: 611 if (bg_flags[i] & EXT4_BG_INODE_UNINIT) 612 continue; 613 614 /* Initialize inode bitmap of the @group */ 615 block = group_data[i].inode_bitmap; 616 err = ext4_resize_ensure_credits_batch(handle, 1); 617 if (err < 0) 618 goto out; 619 /* Mark unused entries in inode bitmap used */ 620 bh = bclean(handle, sb, block); 621 if (IS_ERR(bh)) { 622 err = PTR_ERR(bh); 623 goto out; 624 } 625 626 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), 627 sb->s_blocksize * 8, bh->b_data); 628 err = ext4_handle_dirty_metadata(handle, NULL, bh); 629 brelse(bh); 630 if (err) 631 goto out; 632 } 633 634 /* Mark group tables in block bitmap */ 635 for (j = 0; j < GROUP_TABLE_COUNT; j++) { 636 count = group_table_count[j]; 637 start = (&group_data[0].block_bitmap)[j]; 638 block = start; 639 for (i = 1; i < flex_gd->count; i++) { 640 block += group_table_count[j]; 641 if (block == (&group_data[i].block_bitmap)[j]) { 642 count += group_table_count[j]; 643 continue; 644 } 645 err = set_flexbg_block_bitmap(sb, handle, 646 flex_gd, 647 EXT4_B2C(sbi, start), 648 EXT4_B2C(sbi, 649 start + count 650 - 1)); 651 if (err) 652 goto out; 653 count = group_table_count[j]; 654 start = (&group_data[i].block_bitmap)[j]; 655 block = start; 656 } 657 658 if (count) { 659 err = set_flexbg_block_bitmap(sb, handle, 660 flex_gd, 661 EXT4_B2C(sbi, start), 662 EXT4_B2C(sbi, 663 start + count 664 - 1)); 665 if (err) 666 goto out; 667 } 668 } 669 670 out: 671 err2 = ext4_journal_stop(handle); 672 if (err2 && !err) 673 err = err2; 674 675 return err; 676 } 677 678 /* 679 * Iterate through the groups which hold BACKUP superblock/GDT copies in an 680 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before 681 * calling this for the first time. In a sparse filesystem it will be the 682 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ... 683 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ... 684 */ 685 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three, 686 unsigned *five, unsigned *seven) 687 { 688 unsigned *min = three; 689 int mult = 3; 690 unsigned ret; 691 692 if (!ext4_has_feature_sparse_super(sb)) { 693 ret = *min; 694 *min += 1; 695 return ret; 696 } 697 698 if (*five < *min) { 699 min = five; 700 mult = 5; 701 } 702 if (*seven < *min) { 703 min = seven; 704 mult = 7; 705 } 706 707 ret = *min; 708 *min *= mult; 709 710 return ret; 711 } 712 713 /* 714 * Check that all of the backup GDT blocks are held in the primary GDT block. 715 * It is assumed that they are stored in group order. Returns the number of 716 * groups in current filesystem that have BACKUPS, or -ve error code. 717 */ 718 static int verify_reserved_gdb(struct super_block *sb, 719 ext4_group_t end, 720 struct buffer_head *primary) 721 { 722 const ext4_fsblk_t blk = primary->b_blocknr; 723 unsigned three = 1; 724 unsigned five = 5; 725 unsigned seven = 7; 726 unsigned grp; 727 __le32 *p = (__le32 *)primary->b_data; 728 int gdbackups = 0; 729 730 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) { 731 if (le32_to_cpu(*p++) != 732 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){ 733 ext4_warning(sb, "reserved GDT %llu" 734 " missing grp %d (%llu)", 735 blk, grp, 736 grp * 737 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) + 738 blk); 739 return -EINVAL; 740 } 741 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb)) 742 return -EFBIG; 743 } 744 745 return gdbackups; 746 } 747 748 /* 749 * Called when we need to bring a reserved group descriptor table block into 750 * use from the resize inode. The primary copy of the new GDT block currently 751 * is an indirect block (under the double indirect block in the resize inode). 752 * The new backup GDT blocks will be stored as leaf blocks in this indirect 753 * block, in group order. Even though we know all the block numbers we need, 754 * we check to ensure that the resize inode has actually reserved these blocks. 755 * 756 * Don't need to update the block bitmaps because the blocks are still in use. 757 * 758 * We get all of the error cases out of the way, so that we are sure to not 759 * fail once we start modifying the data on disk, because JBD has no rollback. 760 */ 761 static int add_new_gdb(handle_t *handle, struct inode *inode, 762 ext4_group_t group) 763 { 764 struct super_block *sb = inode->i_sb; 765 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 766 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 767 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num; 768 struct buffer_head **o_group_desc, **n_group_desc = NULL; 769 struct buffer_head *dind = NULL; 770 struct buffer_head *gdb_bh = NULL; 771 int gdbackups; 772 struct ext4_iloc iloc = { .bh = NULL }; 773 __le32 *data; 774 int err; 775 776 if (test_opt(sb, DEBUG)) 777 printk(KERN_DEBUG 778 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n", 779 gdb_num); 780 781 gdb_bh = ext4_sb_bread(sb, gdblock, 0); 782 if (IS_ERR(gdb_bh)) 783 return PTR_ERR(gdb_bh); 784 785 gdbackups = verify_reserved_gdb(sb, group, gdb_bh); 786 if (gdbackups < 0) { 787 err = gdbackups; 788 goto errout; 789 } 790 791 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; 792 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0); 793 if (IS_ERR(dind)) { 794 err = PTR_ERR(dind); 795 dind = NULL; 796 goto errout; 797 } 798 799 data = (__le32 *)dind->b_data; 800 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) { 801 ext4_warning(sb, "new group %u GDT block %llu not reserved", 802 group, gdblock); 803 err = -EINVAL; 804 goto errout; 805 } 806 807 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); 808 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); 809 if (unlikely(err)) 810 goto errout; 811 812 BUFFER_TRACE(gdb_bh, "get_write_access"); 813 err = ext4_journal_get_write_access(handle, gdb_bh); 814 if (unlikely(err)) 815 goto errout; 816 817 BUFFER_TRACE(dind, "get_write_access"); 818 err = ext4_journal_get_write_access(handle, dind); 819 if (unlikely(err)) 820 ext4_std_error(sb, err); 821 822 /* ext4_reserve_inode_write() gets a reference on the iloc */ 823 err = ext4_reserve_inode_write(handle, inode, &iloc); 824 if (unlikely(err)) 825 goto errout; 826 827 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *), 828 GFP_KERNEL); 829 if (!n_group_desc) { 830 err = -ENOMEM; 831 ext4_warning(sb, "not enough memory for %lu groups", 832 gdb_num + 1); 833 goto errout; 834 } 835 836 /* 837 * Finally, we have all of the possible failures behind us... 838 * 839 * Remove new GDT block from inode double-indirect block and clear out 840 * the new GDT block for use (which also "frees" the backup GDT blocks 841 * from the reserved inode). We don't need to change the bitmaps for 842 * these blocks, because they are marked as in-use from being in the 843 * reserved inode, and will become GDT blocks (primary and backup). 844 */ 845 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0; 846 err = ext4_handle_dirty_metadata(handle, NULL, dind); 847 if (unlikely(err)) { 848 ext4_std_error(sb, err); 849 goto errout; 850 } 851 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 852 (9 - EXT4_SB(sb)->s_cluster_bits); 853 ext4_mark_iloc_dirty(handle, inode, &iloc); 854 memset(gdb_bh->b_data, 0, sb->s_blocksize); 855 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh); 856 if (unlikely(err)) { 857 ext4_std_error(sb, err); 858 iloc.bh = NULL; 859 goto errout; 860 } 861 brelse(dind); 862 863 o_group_desc = EXT4_SB(sb)->s_group_desc; 864 memcpy(n_group_desc, o_group_desc, 865 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); 866 n_group_desc[gdb_num] = gdb_bh; 867 EXT4_SB(sb)->s_group_desc = n_group_desc; 868 EXT4_SB(sb)->s_gdb_count++; 869 kvfree(o_group_desc); 870 871 le16_add_cpu(&es->s_reserved_gdt_blocks, -1); 872 err = ext4_handle_dirty_super(handle, sb); 873 if (err) 874 ext4_std_error(sb, err); 875 return err; 876 errout: 877 kvfree(n_group_desc); 878 brelse(iloc.bh); 879 brelse(dind); 880 brelse(gdb_bh); 881 882 ext4_debug("leaving with error %d\n", err); 883 return err; 884 } 885 886 /* 887 * add_new_gdb_meta_bg is the sister of add_new_gdb. 888 */ 889 static int add_new_gdb_meta_bg(struct super_block *sb, 890 handle_t *handle, ext4_group_t group) { 891 ext4_fsblk_t gdblock; 892 struct buffer_head *gdb_bh; 893 struct buffer_head **o_group_desc, **n_group_desc; 894 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 895 int err; 896 897 gdblock = ext4_meta_bg_first_block_no(sb, group) + 898 ext4_bg_has_super(sb, group); 899 gdb_bh = ext4_sb_bread(sb, gdblock, 0); 900 if (IS_ERR(gdb_bh)) 901 return PTR_ERR(gdb_bh); 902 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *), 903 GFP_KERNEL); 904 if (!n_group_desc) { 905 brelse(gdb_bh); 906 err = -ENOMEM; 907 ext4_warning(sb, "not enough memory for %lu groups", 908 gdb_num + 1); 909 return err; 910 } 911 912 o_group_desc = EXT4_SB(sb)->s_group_desc; 913 memcpy(n_group_desc, o_group_desc, 914 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); 915 n_group_desc[gdb_num] = gdb_bh; 916 917 BUFFER_TRACE(gdb_bh, "get_write_access"); 918 err = ext4_journal_get_write_access(handle, gdb_bh); 919 if (err) { 920 kvfree(n_group_desc); 921 brelse(gdb_bh); 922 return err; 923 } 924 925 EXT4_SB(sb)->s_group_desc = n_group_desc; 926 EXT4_SB(sb)->s_gdb_count++; 927 kvfree(o_group_desc); 928 return err; 929 } 930 931 /* 932 * Called when we are adding a new group which has a backup copy of each of 933 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks. 934 * We need to add these reserved backup GDT blocks to the resize inode, so 935 * that they are kept for future resizing and not allocated to files. 936 * 937 * Each reserved backup GDT block will go into a different indirect block. 938 * The indirect blocks are actually the primary reserved GDT blocks, 939 * so we know in advance what their block numbers are. We only get the 940 * double-indirect block to verify it is pointing to the primary reserved 941 * GDT blocks so we don't overwrite a data block by accident. The reserved 942 * backup GDT blocks are stored in their reserved primary GDT block. 943 */ 944 static int reserve_backup_gdb(handle_t *handle, struct inode *inode, 945 ext4_group_t group) 946 { 947 struct super_block *sb = inode->i_sb; 948 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); 949 int cluster_bits = EXT4_SB(sb)->s_cluster_bits; 950 struct buffer_head **primary; 951 struct buffer_head *dind; 952 struct ext4_iloc iloc; 953 ext4_fsblk_t blk; 954 __le32 *data, *end; 955 int gdbackups = 0; 956 int res, i; 957 int err; 958 959 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS); 960 if (!primary) 961 return -ENOMEM; 962 963 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; 964 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0); 965 if (IS_ERR(dind)) { 966 err = PTR_ERR(dind); 967 dind = NULL; 968 goto exit_free; 969 } 970 971 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count; 972 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count % 973 EXT4_ADDR_PER_BLOCK(sb)); 974 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb); 975 976 /* Get each reserved primary GDT block and verify it holds backups */ 977 for (res = 0; res < reserved_gdb; res++, blk++) { 978 if (le32_to_cpu(*data) != blk) { 979 ext4_warning(sb, "reserved block %llu" 980 " not at offset %ld", 981 blk, 982 (long)(data - (__le32 *)dind->b_data)); 983 err = -EINVAL; 984 goto exit_bh; 985 } 986 primary[res] = ext4_sb_bread(sb, blk, 0); 987 if (IS_ERR(primary[res])) { 988 err = PTR_ERR(primary[res]); 989 primary[res] = NULL; 990 goto exit_bh; 991 } 992 gdbackups = verify_reserved_gdb(sb, group, primary[res]); 993 if (gdbackups < 0) { 994 brelse(primary[res]); 995 err = gdbackups; 996 goto exit_bh; 997 } 998 if (++data >= end) 999 data = (__le32 *)dind->b_data; 1000 } 1001 1002 for (i = 0; i < reserved_gdb; i++) { 1003 BUFFER_TRACE(primary[i], "get_write_access"); 1004 if ((err = ext4_journal_get_write_access(handle, primary[i]))) 1005 goto exit_bh; 1006 } 1007 1008 if ((err = ext4_reserve_inode_write(handle, inode, &iloc))) 1009 goto exit_bh; 1010 1011 /* 1012 * Finally we can add each of the reserved backup GDT blocks from 1013 * the new group to its reserved primary GDT block. 1014 */ 1015 blk = group * EXT4_BLOCKS_PER_GROUP(sb); 1016 for (i = 0; i < reserved_gdb; i++) { 1017 int err2; 1018 data = (__le32 *)primary[i]->b_data; 1019 /* printk("reserving backup %lu[%u] = %lu\n", 1020 primary[i]->b_blocknr, gdbackups, 1021 blk + primary[i]->b_blocknr); */ 1022 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr); 1023 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]); 1024 if (!err) 1025 err = err2; 1026 } 1027 1028 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits); 1029 ext4_mark_iloc_dirty(handle, inode, &iloc); 1030 1031 exit_bh: 1032 while (--res >= 0) 1033 brelse(primary[res]); 1034 brelse(dind); 1035 1036 exit_free: 1037 kfree(primary); 1038 1039 return err; 1040 } 1041 1042 /* 1043 * Update the backup copies of the ext4 metadata. These don't need to be part 1044 * of the main resize transaction, because e2fsck will re-write them if there 1045 * is a problem (basically only OOM will cause a problem). However, we 1046 * _should_ update the backups if possible, in case the primary gets trashed 1047 * for some reason and we need to run e2fsck from a backup superblock. The 1048 * important part is that the new block and inode counts are in the backup 1049 * superblocks, and the location of the new group metadata in the GDT backups. 1050 * 1051 * We do not need take the s_resize_lock for this, because these 1052 * blocks are not otherwise touched by the filesystem code when it is 1053 * mounted. We don't need to worry about last changing from 1054 * sbi->s_groups_count, because the worst that can happen is that we 1055 * do not copy the full number of backups at this time. The resize 1056 * which changed s_groups_count will backup again. 1057 */ 1058 static void update_backups(struct super_block *sb, sector_t blk_off, char *data, 1059 int size, int meta_bg) 1060 { 1061 struct ext4_sb_info *sbi = EXT4_SB(sb); 1062 ext4_group_t last; 1063 const int bpg = EXT4_BLOCKS_PER_GROUP(sb); 1064 unsigned three = 1; 1065 unsigned five = 5; 1066 unsigned seven = 7; 1067 ext4_group_t group = 0; 1068 int rest = sb->s_blocksize - size; 1069 handle_t *handle; 1070 int err = 0, err2; 1071 1072 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA); 1073 if (IS_ERR(handle)) { 1074 group = 1; 1075 err = PTR_ERR(handle); 1076 goto exit_err; 1077 } 1078 1079 if (meta_bg == 0) { 1080 group = ext4_list_backups(sb, &three, &five, &seven); 1081 last = sbi->s_groups_count; 1082 } else { 1083 group = ext4_get_group_number(sb, blk_off) + 1; 1084 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2); 1085 } 1086 1087 while (group < sbi->s_groups_count) { 1088 struct buffer_head *bh; 1089 ext4_fsblk_t backup_block; 1090 1091 /* Out of journal space, and can't get more - abort - so sad */ 1092 err = ext4_resize_ensure_credits_batch(handle, 1); 1093 if (err < 0) 1094 break; 1095 1096 if (meta_bg == 0) 1097 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off; 1098 else 1099 backup_block = (ext4_group_first_block_no(sb, group) + 1100 ext4_bg_has_super(sb, group)); 1101 1102 bh = sb_getblk(sb, backup_block); 1103 if (unlikely(!bh)) { 1104 err = -ENOMEM; 1105 break; 1106 } 1107 ext4_debug("update metadata backup %llu(+%llu)\n", 1108 backup_block, backup_block - 1109 ext4_group_first_block_no(sb, group)); 1110 BUFFER_TRACE(bh, "get_write_access"); 1111 if ((err = ext4_journal_get_write_access(handle, bh))) { 1112 brelse(bh); 1113 break; 1114 } 1115 lock_buffer(bh); 1116 memcpy(bh->b_data, data, size); 1117 if (rest) 1118 memset(bh->b_data + size, 0, rest); 1119 set_buffer_uptodate(bh); 1120 unlock_buffer(bh); 1121 err = ext4_handle_dirty_metadata(handle, NULL, bh); 1122 if (unlikely(err)) 1123 ext4_std_error(sb, err); 1124 brelse(bh); 1125 1126 if (meta_bg == 0) 1127 group = ext4_list_backups(sb, &three, &five, &seven); 1128 else if (group == last) 1129 break; 1130 else 1131 group = last; 1132 } 1133 if ((err2 = ext4_journal_stop(handle)) && !err) 1134 err = err2; 1135 1136 /* 1137 * Ugh! Need to have e2fsck write the backup copies. It is too 1138 * late to revert the resize, we shouldn't fail just because of 1139 * the backup copies (they are only needed in case of corruption). 1140 * 1141 * However, if we got here we have a journal problem too, so we 1142 * can't really start a transaction to mark the superblock. 1143 * Chicken out and just set the flag on the hope it will be written 1144 * to disk, and if not - we will simply wait until next fsck. 1145 */ 1146 exit_err: 1147 if (err) { 1148 ext4_warning(sb, "can't update backup for group %u (err %d), " 1149 "forcing fsck on next reboot", group, err); 1150 sbi->s_mount_state &= ~EXT4_VALID_FS; 1151 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS); 1152 mark_buffer_dirty(sbi->s_sbh); 1153 } 1154 } 1155 1156 /* 1157 * ext4_add_new_descs() adds @count group descriptor of groups 1158 * starting at @group 1159 * 1160 * @handle: journal handle 1161 * @sb: super block 1162 * @group: the group no. of the first group desc to be added 1163 * @resize_inode: the resize inode 1164 * @count: number of group descriptors to be added 1165 */ 1166 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb, 1167 ext4_group_t group, struct inode *resize_inode, 1168 ext4_group_t count) 1169 { 1170 struct ext4_sb_info *sbi = EXT4_SB(sb); 1171 struct ext4_super_block *es = sbi->s_es; 1172 struct buffer_head *gdb_bh; 1173 int i, gdb_off, gdb_num, err = 0; 1174 int meta_bg; 1175 1176 meta_bg = ext4_has_feature_meta_bg(sb); 1177 for (i = 0; i < count; i++, group++) { 1178 int reserved_gdb = ext4_bg_has_super(sb, group) ? 1179 le16_to_cpu(es->s_reserved_gdt_blocks) : 0; 1180 1181 gdb_off = group % EXT4_DESC_PER_BLOCK(sb); 1182 gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 1183 1184 /* 1185 * We will only either add reserved group blocks to a backup group 1186 * or remove reserved blocks for the first group in a new group block. 1187 * Doing both would be mean more complex code, and sane people don't 1188 * use non-sparse filesystems anymore. This is already checked above. 1189 */ 1190 if (gdb_off) { 1191 gdb_bh = sbi->s_group_desc[gdb_num]; 1192 BUFFER_TRACE(gdb_bh, "get_write_access"); 1193 err = ext4_journal_get_write_access(handle, gdb_bh); 1194 1195 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group)) 1196 err = reserve_backup_gdb(handle, resize_inode, group); 1197 } else if (meta_bg != 0) { 1198 err = add_new_gdb_meta_bg(sb, handle, group); 1199 } else { 1200 err = add_new_gdb(handle, resize_inode, group); 1201 } 1202 if (err) 1203 break; 1204 } 1205 return err; 1206 } 1207 1208 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block) 1209 { 1210 struct buffer_head *bh = sb_getblk(sb, block); 1211 if (unlikely(!bh)) 1212 return NULL; 1213 if (!bh_uptodate_or_lock(bh)) { 1214 if (bh_submit_read(bh) < 0) { 1215 brelse(bh); 1216 return NULL; 1217 } 1218 } 1219 1220 return bh; 1221 } 1222 1223 static int ext4_set_bitmap_checksums(struct super_block *sb, 1224 ext4_group_t group, 1225 struct ext4_group_desc *gdp, 1226 struct ext4_new_group_data *group_data) 1227 { 1228 struct buffer_head *bh; 1229 1230 if (!ext4_has_metadata_csum(sb)) 1231 return 0; 1232 1233 bh = ext4_get_bitmap(sb, group_data->inode_bitmap); 1234 if (!bh) 1235 return -EIO; 1236 ext4_inode_bitmap_csum_set(sb, group, gdp, bh, 1237 EXT4_INODES_PER_GROUP(sb) / 8); 1238 brelse(bh); 1239 1240 bh = ext4_get_bitmap(sb, group_data->block_bitmap); 1241 if (!bh) 1242 return -EIO; 1243 ext4_block_bitmap_csum_set(sb, group, gdp, bh); 1244 brelse(bh); 1245 1246 return 0; 1247 } 1248 1249 /* 1250 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg 1251 */ 1252 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb, 1253 struct ext4_new_flex_group_data *flex_gd) 1254 { 1255 struct ext4_new_group_data *group_data = flex_gd->groups; 1256 struct ext4_group_desc *gdp; 1257 struct ext4_sb_info *sbi = EXT4_SB(sb); 1258 struct buffer_head *gdb_bh; 1259 ext4_group_t group; 1260 __u16 *bg_flags = flex_gd->bg_flags; 1261 int i, gdb_off, gdb_num, err = 0; 1262 1263 1264 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) { 1265 group = group_data->group; 1266 1267 gdb_off = group % EXT4_DESC_PER_BLOCK(sb); 1268 gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 1269 1270 /* 1271 * get_write_access() has been called on gdb_bh by ext4_add_new_desc(). 1272 */ 1273 gdb_bh = sbi->s_group_desc[gdb_num]; 1274 /* Update group descriptor block for new group */ 1275 gdp = (struct ext4_group_desc *)(gdb_bh->b_data + 1276 gdb_off * EXT4_DESC_SIZE(sb)); 1277 1278 memset(gdp, 0, EXT4_DESC_SIZE(sb)); 1279 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap); 1280 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap); 1281 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data); 1282 if (err) { 1283 ext4_std_error(sb, err); 1284 break; 1285 } 1286 1287 ext4_inode_table_set(sb, gdp, group_data->inode_table); 1288 ext4_free_group_clusters_set(sb, gdp, 1289 group_data->free_clusters_count); 1290 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb)); 1291 if (ext4_has_group_desc_csum(sb)) 1292 ext4_itable_unused_set(sb, gdp, 1293 EXT4_INODES_PER_GROUP(sb)); 1294 gdp->bg_flags = cpu_to_le16(*bg_flags); 1295 ext4_group_desc_csum_set(sb, group, gdp); 1296 1297 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh); 1298 if (unlikely(err)) { 1299 ext4_std_error(sb, err); 1300 break; 1301 } 1302 1303 /* 1304 * We can allocate memory for mb_alloc based on the new group 1305 * descriptor 1306 */ 1307 err = ext4_mb_add_groupinfo(sb, group, gdp); 1308 if (err) 1309 break; 1310 } 1311 return err; 1312 } 1313 1314 /* 1315 * ext4_update_super() updates the super block so that the newly added 1316 * groups can be seen by the filesystem. 1317 * 1318 * @sb: super block 1319 * @flex_gd: new added groups 1320 */ 1321 static void ext4_update_super(struct super_block *sb, 1322 struct ext4_new_flex_group_data *flex_gd) 1323 { 1324 ext4_fsblk_t blocks_count = 0; 1325 ext4_fsblk_t free_blocks = 0; 1326 ext4_fsblk_t reserved_blocks = 0; 1327 struct ext4_new_group_data *group_data = flex_gd->groups; 1328 struct ext4_sb_info *sbi = EXT4_SB(sb); 1329 struct ext4_super_block *es = sbi->s_es; 1330 int i; 1331 1332 BUG_ON(flex_gd->count == 0 || group_data == NULL); 1333 /* 1334 * Make the new blocks and inodes valid next. We do this before 1335 * increasing the group count so that once the group is enabled, 1336 * all of its blocks and inodes are already valid. 1337 * 1338 * We always allocate group-by-group, then block-by-block or 1339 * inode-by-inode within a group, so enabling these 1340 * blocks/inodes before the group is live won't actually let us 1341 * allocate the new space yet. 1342 */ 1343 for (i = 0; i < flex_gd->count; i++) { 1344 blocks_count += group_data[i].blocks_count; 1345 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count); 1346 } 1347 1348 reserved_blocks = ext4_r_blocks_count(es) * 100; 1349 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es)); 1350 reserved_blocks *= blocks_count; 1351 do_div(reserved_blocks, 100); 1352 1353 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count); 1354 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks); 1355 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) * 1356 flex_gd->count); 1357 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) * 1358 flex_gd->count); 1359 1360 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es)); 1361 /* 1362 * We need to protect s_groups_count against other CPUs seeing 1363 * inconsistent state in the superblock. 1364 * 1365 * The precise rules we use are: 1366 * 1367 * * Writers must perform a smp_wmb() after updating all 1368 * dependent data and before modifying the groups count 1369 * 1370 * * Readers must perform an smp_rmb() after reading the groups 1371 * count and before reading any dependent data. 1372 * 1373 * NB. These rules can be relaxed when checking the group count 1374 * while freeing data, as we can only allocate from a block 1375 * group after serialising against the group count, and we can 1376 * only then free after serialising in turn against that 1377 * allocation. 1378 */ 1379 smp_wmb(); 1380 1381 /* Update the global fs size fields */ 1382 sbi->s_groups_count += flex_gd->count; 1383 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, 1384 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 1385 1386 /* Update the reserved block counts only once the new group is 1387 * active. */ 1388 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) + 1389 reserved_blocks); 1390 1391 /* Update the free space counts */ 1392 percpu_counter_add(&sbi->s_freeclusters_counter, 1393 EXT4_NUM_B2C(sbi, free_blocks)); 1394 percpu_counter_add(&sbi->s_freeinodes_counter, 1395 EXT4_INODES_PER_GROUP(sb) * flex_gd->count); 1396 1397 ext4_debug("free blocks count %llu", 1398 percpu_counter_read(&sbi->s_freeclusters_counter)); 1399 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) { 1400 ext4_group_t flex_group; 1401 flex_group = ext4_flex_group(sbi, group_data[0].group); 1402 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks), 1403 &sbi->s_flex_groups[flex_group].free_clusters); 1404 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count, 1405 &sbi->s_flex_groups[flex_group].free_inodes); 1406 } 1407 1408 /* 1409 * Update the fs overhead information 1410 */ 1411 ext4_calculate_overhead(sb); 1412 1413 if (test_opt(sb, DEBUG)) 1414 printk(KERN_DEBUG "EXT4-fs: added group %u:" 1415 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count, 1416 blocks_count, free_blocks, reserved_blocks); 1417 } 1418 1419 /* Add a flex group to an fs. Ensure we handle all possible error conditions 1420 * _before_ we start modifying the filesystem, because we cannot abort the 1421 * transaction and not have it write the data to disk. 1422 */ 1423 static int ext4_flex_group_add(struct super_block *sb, 1424 struct inode *resize_inode, 1425 struct ext4_new_flex_group_data *flex_gd) 1426 { 1427 struct ext4_sb_info *sbi = EXT4_SB(sb); 1428 struct ext4_super_block *es = sbi->s_es; 1429 ext4_fsblk_t o_blocks_count; 1430 ext4_grpblk_t last; 1431 ext4_group_t group; 1432 handle_t *handle; 1433 unsigned reserved_gdb; 1434 int err = 0, err2 = 0, credit; 1435 1436 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags); 1437 1438 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks); 1439 o_blocks_count = ext4_blocks_count(es); 1440 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); 1441 BUG_ON(last); 1442 1443 err = setup_new_flex_group_blocks(sb, flex_gd); 1444 if (err) 1445 goto exit; 1446 /* 1447 * We will always be modifying at least the superblock and GDT 1448 * blocks. If we are adding a group past the last current GDT block, 1449 * we will also modify the inode and the dindirect block. If we 1450 * are adding a group with superblock/GDT backups we will also 1451 * modify each of the reserved GDT dindirect blocks. 1452 */ 1453 credit = 3; /* sb, resize inode, resize inode dindirect */ 1454 /* GDT blocks */ 1455 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb)); 1456 credit += reserved_gdb; /* Reserved GDT dindirect blocks */ 1457 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit); 1458 if (IS_ERR(handle)) { 1459 err = PTR_ERR(handle); 1460 goto exit; 1461 } 1462 1463 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 1464 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 1465 if (err) 1466 goto exit_journal; 1467 1468 group = flex_gd->groups[0].group; 1469 BUG_ON(group != sbi->s_groups_count); 1470 err = ext4_add_new_descs(handle, sb, group, 1471 resize_inode, flex_gd->count); 1472 if (err) 1473 goto exit_journal; 1474 1475 err = ext4_setup_new_descs(handle, sb, flex_gd); 1476 if (err) 1477 goto exit_journal; 1478 1479 ext4_update_super(sb, flex_gd); 1480 1481 err = ext4_handle_dirty_super(handle, sb); 1482 1483 exit_journal: 1484 err2 = ext4_journal_stop(handle); 1485 if (!err) 1486 err = err2; 1487 1488 if (!err) { 1489 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb); 1490 int gdb_num_end = ((group + flex_gd->count - 1) / 1491 EXT4_DESC_PER_BLOCK(sb)); 1492 int meta_bg = ext4_has_feature_meta_bg(sb); 1493 sector_t old_gdb = 0; 1494 1495 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es, 1496 sizeof(struct ext4_super_block), 0); 1497 for (; gdb_num <= gdb_num_end; gdb_num++) { 1498 struct buffer_head *gdb_bh; 1499 1500 gdb_bh = sbi->s_group_desc[gdb_num]; 1501 if (old_gdb == gdb_bh->b_blocknr) 1502 continue; 1503 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data, 1504 gdb_bh->b_size, meta_bg); 1505 old_gdb = gdb_bh->b_blocknr; 1506 } 1507 } 1508 exit: 1509 return err; 1510 } 1511 1512 static int ext4_setup_next_flex_gd(struct super_block *sb, 1513 struct ext4_new_flex_group_data *flex_gd, 1514 ext4_fsblk_t n_blocks_count, 1515 unsigned long flexbg_size) 1516 { 1517 struct ext4_sb_info *sbi = EXT4_SB(sb); 1518 struct ext4_super_block *es = sbi->s_es; 1519 struct ext4_new_group_data *group_data = flex_gd->groups; 1520 ext4_fsblk_t o_blocks_count; 1521 ext4_group_t n_group; 1522 ext4_group_t group; 1523 ext4_group_t last_group; 1524 ext4_grpblk_t last; 1525 ext4_grpblk_t clusters_per_group; 1526 unsigned long i; 1527 1528 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb); 1529 1530 o_blocks_count = ext4_blocks_count(es); 1531 1532 if (o_blocks_count == n_blocks_count) 1533 return 0; 1534 1535 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); 1536 BUG_ON(last); 1537 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last); 1538 1539 last_group = group | (flexbg_size - 1); 1540 if (last_group > n_group) 1541 last_group = n_group; 1542 1543 flex_gd->count = last_group - group + 1; 1544 1545 for (i = 0; i < flex_gd->count; i++) { 1546 int overhead; 1547 1548 group_data[i].group = group + i; 1549 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb); 1550 overhead = ext4_group_overhead_blocks(sb, group + i); 1551 group_data[i].mdata_blocks = overhead; 1552 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb); 1553 if (ext4_has_group_desc_csum(sb)) { 1554 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT | 1555 EXT4_BG_INODE_UNINIT; 1556 if (!test_opt(sb, INIT_INODE_TABLE)) 1557 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED; 1558 } else 1559 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED; 1560 } 1561 1562 if (last_group == n_group && ext4_has_group_desc_csum(sb)) 1563 /* We need to initialize block bitmap of last group. */ 1564 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT; 1565 1566 if ((last_group == n_group) && (last != clusters_per_group - 1)) { 1567 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1); 1568 group_data[i - 1].free_clusters_count -= clusters_per_group - 1569 last - 1; 1570 } 1571 1572 return 1; 1573 } 1574 1575 /* Add group descriptor data to an existing or new group descriptor block. 1576 * Ensure we handle all possible error conditions _before_ we start modifying 1577 * the filesystem, because we cannot abort the transaction and not have it 1578 * write the data to disk. 1579 * 1580 * If we are on a GDT block boundary, we need to get the reserved GDT block. 1581 * Otherwise, we may need to add backup GDT blocks for a sparse group. 1582 * 1583 * We only need to hold the superblock lock while we are actually adding 1584 * in the new group's counts to the superblock. Prior to that we have 1585 * not really "added" the group at all. We re-check that we are still 1586 * adding in the last group in case things have changed since verifying. 1587 */ 1588 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) 1589 { 1590 struct ext4_new_flex_group_data flex_gd; 1591 struct ext4_sb_info *sbi = EXT4_SB(sb); 1592 struct ext4_super_block *es = sbi->s_es; 1593 int reserved_gdb = ext4_bg_has_super(sb, input->group) ? 1594 le16_to_cpu(es->s_reserved_gdt_blocks) : 0; 1595 struct inode *inode = NULL; 1596 int gdb_off; 1597 int err; 1598 __u16 bg_flags = 0; 1599 1600 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb); 1601 1602 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) { 1603 ext4_warning(sb, "Can't resize non-sparse filesystem further"); 1604 return -EPERM; 1605 } 1606 1607 if (ext4_blocks_count(es) + input->blocks_count < 1608 ext4_blocks_count(es)) { 1609 ext4_warning(sb, "blocks_count overflow"); 1610 return -EINVAL; 1611 } 1612 1613 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) < 1614 le32_to_cpu(es->s_inodes_count)) { 1615 ext4_warning(sb, "inodes_count overflow"); 1616 return -EINVAL; 1617 } 1618 1619 if (reserved_gdb || gdb_off == 0) { 1620 if (!ext4_has_feature_resize_inode(sb) || 1621 !le16_to_cpu(es->s_reserved_gdt_blocks)) { 1622 ext4_warning(sb, 1623 "No reserved GDT blocks, can't resize"); 1624 return -EPERM; 1625 } 1626 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL); 1627 if (IS_ERR(inode)) { 1628 ext4_warning(sb, "Error opening resize inode"); 1629 return PTR_ERR(inode); 1630 } 1631 } 1632 1633 1634 err = verify_group_input(sb, input); 1635 if (err) 1636 goto out; 1637 1638 err = ext4_alloc_flex_bg_array(sb, input->group + 1); 1639 if (err) 1640 goto out; 1641 1642 err = ext4_mb_alloc_groupinfo(sb, input->group + 1); 1643 if (err) 1644 goto out; 1645 1646 flex_gd.count = 1; 1647 flex_gd.groups = input; 1648 flex_gd.bg_flags = &bg_flags; 1649 err = ext4_flex_group_add(sb, inode, &flex_gd); 1650 out: 1651 iput(inode); 1652 return err; 1653 } /* ext4_group_add */ 1654 1655 /* 1656 * extend a group without checking assuming that checking has been done. 1657 */ 1658 static int ext4_group_extend_no_check(struct super_block *sb, 1659 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add) 1660 { 1661 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 1662 handle_t *handle; 1663 int err = 0, err2; 1664 1665 /* We will update the superblock, one block bitmap, and 1666 * one group descriptor via ext4_group_add_blocks(). 1667 */ 1668 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3); 1669 if (IS_ERR(handle)) { 1670 err = PTR_ERR(handle); 1671 ext4_warning(sb, "error %d on journal start", err); 1672 return err; 1673 } 1674 1675 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); 1676 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); 1677 if (err) { 1678 ext4_warning(sb, "error %d on journal write access", err); 1679 goto errout; 1680 } 1681 1682 ext4_blocks_count_set(es, o_blocks_count + add); 1683 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add); 1684 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count, 1685 o_blocks_count + add); 1686 /* We add the blocks to the bitmap and set the group need init bit */ 1687 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add); 1688 if (err) 1689 goto errout; 1690 ext4_handle_dirty_super(handle, sb); 1691 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count, 1692 o_blocks_count + add); 1693 errout: 1694 err2 = ext4_journal_stop(handle); 1695 if (err2 && !err) 1696 err = err2; 1697 1698 if (!err) { 1699 if (test_opt(sb, DEBUG)) 1700 printk(KERN_DEBUG "EXT4-fs: extended group to %llu " 1701 "blocks\n", ext4_blocks_count(es)); 1702 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, 1703 (char *)es, sizeof(struct ext4_super_block), 0); 1704 } 1705 return err; 1706 } 1707 1708 /* 1709 * Extend the filesystem to the new number of blocks specified. This entry 1710 * point is only used to extend the current filesystem to the end of the last 1711 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>" 1712 * for emergencies (because it has no dependencies on reserved blocks). 1713 * 1714 * If we _really_ wanted, we could use default values to call ext4_group_add() 1715 * allow the "remount" trick to work for arbitrary resizing, assuming enough 1716 * GDT blocks are reserved to grow to the desired size. 1717 */ 1718 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, 1719 ext4_fsblk_t n_blocks_count) 1720 { 1721 ext4_fsblk_t o_blocks_count; 1722 ext4_grpblk_t last; 1723 ext4_grpblk_t add; 1724 struct buffer_head *bh; 1725 int err; 1726 ext4_group_t group; 1727 1728 o_blocks_count = ext4_blocks_count(es); 1729 1730 if (test_opt(sb, DEBUG)) 1731 ext4_msg(sb, KERN_DEBUG, 1732 "extending last group from %llu to %llu blocks", 1733 o_blocks_count, n_blocks_count); 1734 1735 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count) 1736 return 0; 1737 1738 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) { 1739 ext4_msg(sb, KERN_ERR, 1740 "filesystem too large to resize to %llu blocks safely", 1741 n_blocks_count); 1742 return -EINVAL; 1743 } 1744 1745 if (n_blocks_count < o_blocks_count) { 1746 ext4_warning(sb, "can't shrink FS - resize aborted"); 1747 return -EINVAL; 1748 } 1749 1750 /* Handle the remaining blocks in the last group only. */ 1751 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); 1752 1753 if (last == 0) { 1754 ext4_warning(sb, "need to use ext2online to resize further"); 1755 return -EPERM; 1756 } 1757 1758 add = EXT4_BLOCKS_PER_GROUP(sb) - last; 1759 1760 if (o_blocks_count + add < o_blocks_count) { 1761 ext4_warning(sb, "blocks_count overflow"); 1762 return -EINVAL; 1763 } 1764 1765 if (o_blocks_count + add > n_blocks_count) 1766 add = n_blocks_count - o_blocks_count; 1767 1768 if (o_blocks_count + add < n_blocks_count) 1769 ext4_warning(sb, "will only finish group (%llu blocks, %u new)", 1770 o_blocks_count + add, add); 1771 1772 /* See if the device is actually as big as what was requested */ 1773 bh = sb_bread(sb, o_blocks_count + add - 1); 1774 if (!bh) { 1775 ext4_warning(sb, "can't read last block, resize aborted"); 1776 return -ENOSPC; 1777 } 1778 brelse(bh); 1779 1780 err = ext4_group_extend_no_check(sb, o_blocks_count, add); 1781 return err; 1782 } /* ext4_group_extend */ 1783 1784 1785 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups) 1786 { 1787 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb); 1788 } 1789 1790 /* 1791 * Release the resize inode and drop the resize_inode feature if there 1792 * are no more reserved gdt blocks, and then convert the file system 1793 * to enable meta_bg 1794 */ 1795 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode) 1796 { 1797 handle_t *handle; 1798 struct ext4_sb_info *sbi = EXT4_SB(sb); 1799 struct ext4_super_block *es = sbi->s_es; 1800 struct ext4_inode_info *ei = EXT4_I(inode); 1801 ext4_fsblk_t nr; 1802 int i, ret, err = 0; 1803 int credits = 1; 1804 1805 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg"); 1806 if (inode) { 1807 if (es->s_reserved_gdt_blocks) { 1808 ext4_error(sb, "Unexpected non-zero " 1809 "s_reserved_gdt_blocks"); 1810 return -EPERM; 1811 } 1812 1813 /* Do a quick sanity check of the resize inode */ 1814 if (inode->i_blocks != 1 << (inode->i_blkbits - 1815 (9 - sbi->s_cluster_bits))) 1816 goto invalid_resize_inode; 1817 for (i = 0; i < EXT4_N_BLOCKS; i++) { 1818 if (i == EXT4_DIND_BLOCK) { 1819 if (ei->i_data[i]) 1820 continue; 1821 else 1822 goto invalid_resize_inode; 1823 } 1824 if (ei->i_data[i]) 1825 goto invalid_resize_inode; 1826 } 1827 credits += 3; /* block bitmap, bg descriptor, resize inode */ 1828 } 1829 1830 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits); 1831 if (IS_ERR(handle)) 1832 return PTR_ERR(handle); 1833 1834 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 1835 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 1836 if (err) 1837 goto errout; 1838 1839 ext4_clear_feature_resize_inode(sb); 1840 ext4_set_feature_meta_bg(sb); 1841 sbi->s_es->s_first_meta_bg = 1842 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count)); 1843 1844 err = ext4_handle_dirty_super(handle, sb); 1845 if (err) { 1846 ext4_std_error(sb, err); 1847 goto errout; 1848 } 1849 1850 if (inode) { 1851 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]); 1852 ext4_free_blocks(handle, inode, NULL, nr, 1, 1853 EXT4_FREE_BLOCKS_METADATA | 1854 EXT4_FREE_BLOCKS_FORGET); 1855 ei->i_data[EXT4_DIND_BLOCK] = 0; 1856 inode->i_blocks = 0; 1857 1858 err = ext4_mark_inode_dirty(handle, inode); 1859 if (err) 1860 ext4_std_error(sb, err); 1861 } 1862 1863 errout: 1864 ret = ext4_journal_stop(handle); 1865 if (!err) 1866 err = ret; 1867 return ret; 1868 1869 invalid_resize_inode: 1870 ext4_error(sb, "corrupted/inconsistent resize inode"); 1871 return -EINVAL; 1872 } 1873 1874 /* 1875 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count 1876 * 1877 * @sb: super block of the fs to be resized 1878 * @n_blocks_count: the number of blocks resides in the resized fs 1879 */ 1880 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count) 1881 { 1882 struct ext4_new_flex_group_data *flex_gd = NULL; 1883 struct ext4_sb_info *sbi = EXT4_SB(sb); 1884 struct ext4_super_block *es = sbi->s_es; 1885 struct buffer_head *bh; 1886 struct inode *resize_inode = NULL; 1887 ext4_grpblk_t add, offset; 1888 unsigned long n_desc_blocks; 1889 unsigned long o_desc_blocks; 1890 ext4_group_t o_group; 1891 ext4_group_t n_group; 1892 ext4_fsblk_t o_blocks_count; 1893 ext4_fsblk_t n_blocks_count_retry = 0; 1894 unsigned long last_update_time = 0; 1895 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex; 1896 int meta_bg; 1897 1898 /* See if the device is actually as big as what was requested */ 1899 bh = sb_bread(sb, n_blocks_count - 1); 1900 if (!bh) { 1901 ext4_warning(sb, "can't read last block, resize aborted"); 1902 return -ENOSPC; 1903 } 1904 brelse(bh); 1905 1906 retry: 1907 o_blocks_count = ext4_blocks_count(es); 1908 1909 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu " 1910 "to %llu blocks", o_blocks_count, n_blocks_count); 1911 1912 if (n_blocks_count < o_blocks_count) { 1913 /* On-line shrinking not supported */ 1914 ext4_warning(sb, "can't shrink FS - resize aborted"); 1915 return -EINVAL; 1916 } 1917 1918 if (n_blocks_count == o_blocks_count) 1919 /* Nothing need to do */ 1920 return 0; 1921 1922 n_group = ext4_get_group_number(sb, n_blocks_count - 1); 1923 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) { 1924 ext4_warning(sb, "resize would cause inodes_count overflow"); 1925 return -EINVAL; 1926 } 1927 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset); 1928 1929 n_desc_blocks = num_desc_blocks(sb, n_group + 1); 1930 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count); 1931 1932 meta_bg = ext4_has_feature_meta_bg(sb); 1933 1934 if (ext4_has_feature_resize_inode(sb)) { 1935 if (meta_bg) { 1936 ext4_error(sb, "resize_inode and meta_bg enabled " 1937 "simultaneously"); 1938 return -EINVAL; 1939 } 1940 if (n_desc_blocks > o_desc_blocks + 1941 le16_to_cpu(es->s_reserved_gdt_blocks)) { 1942 n_blocks_count_retry = n_blocks_count; 1943 n_desc_blocks = o_desc_blocks + 1944 le16_to_cpu(es->s_reserved_gdt_blocks); 1945 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb); 1946 n_blocks_count = (ext4_fsblk_t)n_group * 1947 EXT4_BLOCKS_PER_GROUP(sb) + 1948 le32_to_cpu(es->s_first_data_block); 1949 n_group--; /* set to last group number */ 1950 } 1951 1952 if (!resize_inode) 1953 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO, 1954 EXT4_IGET_SPECIAL); 1955 if (IS_ERR(resize_inode)) { 1956 ext4_warning(sb, "Error opening resize inode"); 1957 return PTR_ERR(resize_inode); 1958 } 1959 } 1960 1961 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) { 1962 err = ext4_convert_meta_bg(sb, resize_inode); 1963 if (err) 1964 goto out; 1965 if (resize_inode) { 1966 iput(resize_inode); 1967 resize_inode = NULL; 1968 } 1969 if (n_blocks_count_retry) { 1970 n_blocks_count = n_blocks_count_retry; 1971 n_blocks_count_retry = 0; 1972 goto retry; 1973 } 1974 } 1975 1976 /* 1977 * Make sure the last group has enough space so that it's 1978 * guaranteed to have enough space for all metadata blocks 1979 * that it might need to hold. (We might not need to store 1980 * the inode table blocks in the last block group, but there 1981 * will be cases where this might be needed.) 1982 */ 1983 if ((ext4_group_first_block_no(sb, n_group) + 1984 ext4_group_overhead_blocks(sb, n_group) + 2 + 1985 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) { 1986 n_blocks_count = ext4_group_first_block_no(sb, n_group); 1987 n_group--; 1988 n_blocks_count_retry = 0; 1989 if (resize_inode) { 1990 iput(resize_inode); 1991 resize_inode = NULL; 1992 } 1993 goto retry; 1994 } 1995 1996 /* extend the last group */ 1997 if (n_group == o_group) 1998 add = n_blocks_count - o_blocks_count; 1999 else 2000 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1)); 2001 if (add > 0) { 2002 err = ext4_group_extend_no_check(sb, o_blocks_count, add); 2003 if (err) 2004 goto out; 2005 } 2006 2007 if (ext4_blocks_count(es) == n_blocks_count) 2008 goto out; 2009 2010 err = ext4_alloc_flex_bg_array(sb, n_group + 1); 2011 if (err) 2012 goto out; 2013 2014 err = ext4_mb_alloc_groupinfo(sb, n_group + 1); 2015 if (err) 2016 goto out; 2017 2018 flex_gd = alloc_flex_gd(flexbg_size); 2019 if (flex_gd == NULL) { 2020 err = -ENOMEM; 2021 goto out; 2022 } 2023 2024 /* Add flex groups. Note that a regular group is a 2025 * flex group with 1 group. 2026 */ 2027 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count, 2028 flexbg_size)) { 2029 if (jiffies - last_update_time > HZ * 10) { 2030 if (last_update_time) 2031 ext4_msg(sb, KERN_INFO, 2032 "resized to %llu blocks", 2033 ext4_blocks_count(es)); 2034 last_update_time = jiffies; 2035 } 2036 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0) 2037 break; 2038 err = ext4_flex_group_add(sb, resize_inode, flex_gd); 2039 if (unlikely(err)) 2040 break; 2041 } 2042 2043 if (!err && n_blocks_count_retry) { 2044 n_blocks_count = n_blocks_count_retry; 2045 n_blocks_count_retry = 0; 2046 free_flex_gd(flex_gd); 2047 flex_gd = NULL; 2048 if (resize_inode) { 2049 iput(resize_inode); 2050 resize_inode = NULL; 2051 } 2052 goto retry; 2053 } 2054 2055 out: 2056 if (flex_gd) 2057 free_flex_gd(flex_gd); 2058 if (resize_inode != NULL) 2059 iput(resize_inode); 2060 if (err) 2061 ext4_warning(sb, "error (%d) occurred during " 2062 "file system resize", err); 2063 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", 2064 ext4_blocks_count(es)); 2065 return err; 2066 } 2067