1/* -*- mode: c; c-basic-offset: 8; -*- 2 * vim: noexpandtab sw=8 ts=8 sts=0: 3 * 4 * resize.c 5 * 6 * volume resize. 7 * Inspired by ext3/resize.c. 8 * 9 * Copyright (C) 2007 Oracle. All rights reserved. 10 * 11 * This program is free software; you can redistribute it and/or 12 * modify it under the terms of the GNU General Public 13 * License as published by the Free Software Foundation; either 14 * version 2 of the License, or (at your option) any later version. 15 * 16 * This program is distributed in the hope that it will be useful, 17 * but WITHOUT ANY WARRANTY; without even the implied warranty of 18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 19 * General Public License for more details. 20 * 21 * You should have received a copy of the GNU General Public 22 * License along with this program; if not, write to the 23 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 24 * Boston, MA 021110-1307, USA. 25 */ 26 27#include <linux/fs.h> 28#include <linux/types.h> 29 30#include <cluster/masklog.h> 31 32#include "ocfs2.h" 33 34#include "alloc.h" 35#include "dlmglue.h" 36#include "inode.h" 37#include "journal.h" 38#include "super.h" 39#include "sysfile.h" 40#include "uptodate.h" 41#include "ocfs2_trace.h" 42 43#include "buffer_head_io.h" 44#include "suballoc.h" 45#include "resize.h" 46 47/* 48 * Check whether there are new backup superblocks exist 49 * in the last group. If there are some, mark them or clear 50 * them in the bitmap. 51 * 52 * Return how many backups we find in the last group. 53 */ 54static u16 ocfs2_calc_new_backup_super(struct inode *inode, 55 struct ocfs2_group_desc *gd, 56 u16 cl_cpg, 57 int set) 58{ 59 int i; 60 u16 backups = 0; 61 u32 cluster; 62 u64 blkno, gd_blkno, lgd_blkno = le64_to_cpu(gd->bg_blkno); 63 64 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) { 65 blkno = ocfs2_backup_super_blkno(inode->i_sb, i); 66 cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno); 67 68 gd_blkno = ocfs2_which_cluster_group(inode, cluster); 69 if (gd_blkno < lgd_blkno) 70 continue; 71 else if (gd_blkno > lgd_blkno) 72 break; 73 74 if (set) 75 ocfs2_set_bit(cluster % cl_cpg, 76 (unsigned long *)gd->bg_bitmap); 77 else 78 ocfs2_clear_bit(cluster % cl_cpg, 79 (unsigned long *)gd->bg_bitmap); 80 backups++; 81 } 82 83 return backups; 84} 85 86static int ocfs2_update_last_group_and_inode(handle_t *handle, 87 struct inode *bm_inode, 88 struct buffer_head *bm_bh, 89 struct buffer_head *group_bh, 90 u32 first_new_cluster, 91 int new_clusters) 92{ 93 int ret = 0; 94 struct ocfs2_super *osb = OCFS2_SB(bm_inode->i_sb); 95 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bm_bh->b_data; 96 struct ocfs2_chain_list *cl = &fe->id2.i_chain; 97 struct ocfs2_chain_rec *cr; 98 struct ocfs2_group_desc *group; 99 u16 chain, num_bits, backups = 0; 100 u16 cl_bpc = le16_to_cpu(cl->cl_bpc); 101 u16 cl_cpg = le16_to_cpu(cl->cl_cpg); 102 103 trace_ocfs2_update_last_group_and_inode(new_clusters, 104 first_new_cluster); 105 106 ret = ocfs2_journal_access_gd(handle, INODE_CACHE(bm_inode), 107 group_bh, OCFS2_JOURNAL_ACCESS_WRITE); 108 if (ret < 0) { 109 mlog_errno(ret); 110 goto out; 111 } 112 113 group = (struct ocfs2_group_desc *)group_bh->b_data; 114 115 /* update the group first. */ 116 num_bits = new_clusters * cl_bpc; 117 le16_add_cpu(&group->bg_bits, num_bits); 118 le16_add_cpu(&group->bg_free_bits_count, num_bits); 119 120 /* 121 * check whether there are some new backup superblocks exist in 122 * this group and update the group bitmap accordingly. 123 */ 124 if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, 125 OCFS2_FEATURE_COMPAT_BACKUP_SB)) { 126 backups = ocfs2_calc_new_backup_super(bm_inode, 127 group, 128 cl_cpg, 1); 129 le16_add_cpu(&group->bg_free_bits_count, -1 * backups); 130 } 131 132 ocfs2_journal_dirty(handle, group_bh); 133 134 /* update the inode accordingly. */ 135 ret = ocfs2_journal_access_di(handle, INODE_CACHE(bm_inode), bm_bh, 136 OCFS2_JOURNAL_ACCESS_WRITE); 137 if (ret < 0) { 138 mlog_errno(ret); 139 goto out_rollback; 140 } 141 142 chain = le16_to_cpu(group->bg_chain); 143 cr = (&cl->cl_recs[chain]); 144 le32_add_cpu(&cr->c_total, num_bits); 145 le32_add_cpu(&cr->c_free, num_bits); 146 le32_add_cpu(&fe->id1.bitmap1.i_total, num_bits); 147 le32_add_cpu(&fe->i_clusters, new_clusters); 148 149 if (backups) { 150 le32_add_cpu(&cr->c_free, -1 * backups); 151 le32_add_cpu(&fe->id1.bitmap1.i_used, backups); 152 } 153 154 spin_lock(&OCFS2_I(bm_inode)->ip_lock); 155 OCFS2_I(bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 156 le64_add_cpu(&fe->i_size, (u64)new_clusters << osb->s_clustersize_bits); 157 spin_unlock(&OCFS2_I(bm_inode)->ip_lock); 158 i_size_write(bm_inode, le64_to_cpu(fe->i_size)); 159 160 ocfs2_journal_dirty(handle, bm_bh); 161 162out_rollback: 163 if (ret < 0) { 164 ocfs2_calc_new_backup_super(bm_inode, 165 group, 166 cl_cpg, 0); 167 le16_add_cpu(&group->bg_free_bits_count, backups); 168 le16_add_cpu(&group->bg_bits, -1 * num_bits); 169 le16_add_cpu(&group->bg_free_bits_count, -1 * num_bits); 170 } 171out: 172 if (ret) 173 mlog_errno(ret); 174 return ret; 175} 176 177static int update_backups(struct inode * inode, u32 clusters, char *data) 178{ 179 int i, ret = 0; 180 u32 cluster; 181 u64 blkno; 182 struct buffer_head *backup = NULL; 183 struct ocfs2_dinode *backup_di = NULL; 184 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 185 186 /* calculate the real backups we need to update. */ 187 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) { 188 blkno = ocfs2_backup_super_blkno(inode->i_sb, i); 189 cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno); 190 if (cluster > clusters) 191 break; 192 193 ret = ocfs2_read_blocks_sync(osb, blkno, 1, &backup); 194 if (ret < 0) { 195 mlog_errno(ret); 196 break; 197 } 198 199 memcpy(backup->b_data, data, inode->i_sb->s_blocksize); 200 201 backup_di = (struct ocfs2_dinode *)backup->b_data; 202 backup_di->i_blkno = cpu_to_le64(blkno); 203 204 ret = ocfs2_write_super_or_backup(osb, backup); 205 brelse(backup); 206 backup = NULL; 207 if (ret < 0) { 208 mlog_errno(ret); 209 break; 210 } 211 } 212 213 return ret; 214} 215 216static void ocfs2_update_super_and_backups(struct inode *inode, 217 int new_clusters) 218{ 219 int ret; 220 u32 clusters = 0; 221 struct buffer_head *super_bh = NULL; 222 struct ocfs2_dinode *super_di = NULL; 223 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 224 225 /* 226 * update the superblock last. 227 * It doesn't matter if the write failed. 228 */ 229 ret = ocfs2_read_blocks_sync(osb, OCFS2_SUPER_BLOCK_BLKNO, 1, 230 &super_bh); 231 if (ret < 0) { 232 mlog_errno(ret); 233 goto out; 234 } 235 236 super_di = (struct ocfs2_dinode *)super_bh->b_data; 237 le32_add_cpu(&super_di->i_clusters, new_clusters); 238 clusters = le32_to_cpu(super_di->i_clusters); 239 240 ret = ocfs2_write_super_or_backup(osb, super_bh); 241 if (ret < 0) { 242 mlog_errno(ret); 243 goto out; 244 } 245 246 if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_COMPAT_BACKUP_SB)) 247 ret = update_backups(inode, clusters, super_bh->b_data); 248 249out: 250 brelse(super_bh); 251 if (ret) 252 printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s" 253 " during fs resize. This condition is not fatal," 254 " but fsck.ocfs2 should be run to fix it\n", 255 osb->dev_str); 256 return; 257} 258 259/* 260 * Extend the filesystem to the new number of clusters specified. This entry 261 * point is only used to extend the current filesystem to the end of the last 262 * existing group. 263 */ 264int ocfs2_group_extend(struct inode * inode, int new_clusters) 265{ 266 int ret; 267 handle_t *handle; 268 struct buffer_head *main_bm_bh = NULL; 269 struct buffer_head *group_bh = NULL; 270 struct inode *main_bm_inode = NULL; 271 struct ocfs2_dinode *fe = NULL; 272 struct ocfs2_group_desc *group = NULL; 273 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 274 u16 cl_bpc; 275 u32 first_new_cluster; 276 u64 lgd_blkno; 277 278 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) 279 return -EROFS; 280 281 if (new_clusters < 0) 282 return -EINVAL; 283 else if (new_clusters == 0) 284 return 0; 285 286 main_bm_inode = ocfs2_get_system_file_inode(osb, 287 GLOBAL_BITMAP_SYSTEM_INODE, 288 OCFS2_INVALID_SLOT); 289 if (!main_bm_inode) { 290 ret = -EINVAL; 291 mlog_errno(ret); 292 goto out; 293 } 294 295 mutex_lock(&main_bm_inode->i_mutex); 296 297 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); 298 if (ret < 0) { 299 mlog_errno(ret); 300 goto out_mutex; 301 } 302 303 fe = (struct ocfs2_dinode *)main_bm_bh->b_data; 304 305 /* main_bm_bh is validated by inode read inside ocfs2_inode_lock(), 306 * so any corruption is a code bug. */ 307 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 308 309 if (le16_to_cpu(fe->id2.i_chain.cl_cpg) != 310 ocfs2_group_bitmap_size(osb->sb, 0, 311 osb->s_feature_incompat) * 8) { 312 mlog(ML_ERROR, "The disk is too old and small. " 313 "Force to do offline resize."); 314 ret = -EINVAL; 315 goto out_unlock; 316 } 317 318 first_new_cluster = le32_to_cpu(fe->i_clusters); 319 lgd_blkno = ocfs2_which_cluster_group(main_bm_inode, 320 first_new_cluster - 1); 321 322 ret = ocfs2_read_group_descriptor(main_bm_inode, fe, lgd_blkno, 323 &group_bh); 324 if (ret < 0) { 325 mlog_errno(ret); 326 goto out_unlock; 327 } 328 group = (struct ocfs2_group_desc *)group_bh->b_data; 329 330 cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc); 331 if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters > 332 le16_to_cpu(fe->id2.i_chain.cl_cpg)) { 333 ret = -EINVAL; 334 goto out_unlock; 335 } 336 337 338 trace_ocfs2_group_extend( 339 (unsigned long long)le64_to_cpu(group->bg_blkno), new_clusters); 340 341 handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS); 342 if (IS_ERR(handle)) { 343 mlog_errno(PTR_ERR(handle)); 344 ret = -EINVAL; 345 goto out_unlock; 346 } 347 348 /* update the last group descriptor and inode. */ 349 ret = ocfs2_update_last_group_and_inode(handle, main_bm_inode, 350 main_bm_bh, group_bh, 351 first_new_cluster, 352 new_clusters); 353 if (ret) { 354 mlog_errno(ret); 355 goto out_commit; 356 } 357 358 ocfs2_update_super_and_backups(main_bm_inode, new_clusters); 359 360out_commit: 361 ocfs2_commit_trans(osb, handle); 362out_unlock: 363 brelse(group_bh); 364 brelse(main_bm_bh); 365 366 ocfs2_inode_unlock(main_bm_inode, 1); 367 368out_mutex: 369 mutex_unlock(&main_bm_inode->i_mutex); 370 iput(main_bm_inode); 371 372out: 373 return ret; 374} 375 376static int ocfs2_check_new_group(struct inode *inode, 377 struct ocfs2_dinode *di, 378 struct ocfs2_new_group_input *input, 379 struct buffer_head *group_bh) 380{ 381 int ret; 382 struct ocfs2_group_desc *gd = 383 (struct ocfs2_group_desc *)group_bh->b_data; 384 u16 cl_bpc = le16_to_cpu(di->id2.i_chain.cl_bpc); 385 386 ret = ocfs2_check_group_descriptor(inode->i_sb, di, group_bh); 387 if (ret) 388 goto out; 389 390 ret = -EINVAL; 391 if (le16_to_cpu(gd->bg_chain) != input->chain) 392 mlog(ML_ERROR, "Group descriptor # %llu has bad chain %u " 393 "while input has %u set.\n", 394 (unsigned long long)le64_to_cpu(gd->bg_blkno), 395 le16_to_cpu(gd->bg_chain), input->chain); 396 else if (le16_to_cpu(gd->bg_bits) != input->clusters * cl_bpc) 397 mlog(ML_ERROR, "Group descriptor # %llu has bit count %u but " 398 "input has %u clusters set\n", 399 (unsigned long long)le64_to_cpu(gd->bg_blkno), 400 le16_to_cpu(gd->bg_bits), input->clusters); 401 else if (le16_to_cpu(gd->bg_free_bits_count) != input->frees * cl_bpc) 402 mlog(ML_ERROR, "Group descriptor # %llu has free bit count %u " 403 "but it should have %u set\n", 404 (unsigned long long)le64_to_cpu(gd->bg_blkno), 405 le16_to_cpu(gd->bg_bits), 406 input->frees * cl_bpc); 407 else 408 ret = 0; 409 410out: 411 return ret; 412} 413 414static int ocfs2_verify_group_and_input(struct inode *inode, 415 struct ocfs2_dinode *di, 416 struct ocfs2_new_group_input *input, 417 struct buffer_head *group_bh) 418{ 419 u16 cl_count = le16_to_cpu(di->id2.i_chain.cl_count); 420 u16 cl_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg); 421 u16 next_free = le16_to_cpu(di->id2.i_chain.cl_next_free_rec); 422 u32 cluster = ocfs2_blocks_to_clusters(inode->i_sb, input->group); 423 u32 total_clusters = le32_to_cpu(di->i_clusters); 424 int ret = -EINVAL; 425 426 if (cluster < total_clusters) 427 mlog(ML_ERROR, "add a group which is in the current volume.\n"); 428 else if (input->chain >= cl_count) 429 mlog(ML_ERROR, "input chain exceeds the limit.\n"); 430 else if (next_free != cl_count && next_free != input->chain) 431 mlog(ML_ERROR, 432 "the add group should be in chain %u\n", next_free); 433 else if (total_clusters + input->clusters < total_clusters) 434 mlog(ML_ERROR, "add group's clusters overflow.\n"); 435 else if (input->clusters > cl_cpg) 436 mlog(ML_ERROR, "the cluster exceeds the maximum of a group\n"); 437 else if (input->frees > input->clusters) 438 mlog(ML_ERROR, "the free cluster exceeds the total clusters\n"); 439 else if (total_clusters % cl_cpg != 0) 440 mlog(ML_ERROR, 441 "the last group isn't full. Use group extend first.\n"); 442 else if (input->group != ocfs2_which_cluster_group(inode, cluster)) 443 mlog(ML_ERROR, "group blkno is invalid\n"); 444 else if ((ret = ocfs2_check_new_group(inode, di, input, group_bh))) 445 mlog(ML_ERROR, "group descriptor check failed.\n"); 446 else 447 ret = 0; 448 449 return ret; 450} 451 452/* Add a new group descriptor to global_bitmap. */ 453int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input) 454{ 455 int ret; 456 handle_t *handle; 457 struct buffer_head *main_bm_bh = NULL; 458 struct inode *main_bm_inode = NULL; 459 struct ocfs2_dinode *fe = NULL; 460 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 461 struct buffer_head *group_bh = NULL; 462 struct ocfs2_group_desc *group = NULL; 463 struct ocfs2_chain_list *cl; 464 struct ocfs2_chain_rec *cr; 465 u16 cl_bpc; 466 u64 bg_ptr; 467 468 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) 469 return -EROFS; 470 471 main_bm_inode = ocfs2_get_system_file_inode(osb, 472 GLOBAL_BITMAP_SYSTEM_INODE, 473 OCFS2_INVALID_SLOT); 474 if (!main_bm_inode) { 475 ret = -EINVAL; 476 mlog_errno(ret); 477 goto out; 478 } 479 480 mutex_lock(&main_bm_inode->i_mutex); 481 482 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); 483 if (ret < 0) { 484 mlog_errno(ret); 485 goto out_mutex; 486 } 487 488 fe = (struct ocfs2_dinode *)main_bm_bh->b_data; 489 490 if (le16_to_cpu(fe->id2.i_chain.cl_cpg) != 491 ocfs2_group_bitmap_size(osb->sb, 0, 492 osb->s_feature_incompat) * 8) { 493 mlog(ML_ERROR, "The disk is too old and small." 494 " Force to do offline resize."); 495 ret = -EINVAL; 496 goto out_unlock; 497 } 498 499 ret = ocfs2_read_blocks_sync(osb, input->group, 1, &group_bh); 500 if (ret < 0) { 501 mlog(ML_ERROR, "Can't read the group descriptor # %llu " 502 "from the device.", (unsigned long long)input->group); 503 goto out_unlock; 504 } 505 506 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), group_bh); 507 508 ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh); 509 if (ret) { 510 mlog_errno(ret); 511 goto out_free_group_bh; 512 } 513 514 trace_ocfs2_group_add((unsigned long long)input->group, 515 input->chain, input->clusters, input->frees); 516 517 handle = ocfs2_start_trans(osb, OCFS2_GROUP_ADD_CREDITS); 518 if (IS_ERR(handle)) { 519 mlog_errno(PTR_ERR(handle)); 520 ret = -EINVAL; 521 goto out_free_group_bh; 522 } 523 524 cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc); 525 cl = &fe->id2.i_chain; 526 cr = &cl->cl_recs[input->chain]; 527 528 ret = ocfs2_journal_access_gd(handle, INODE_CACHE(main_bm_inode), 529 group_bh, OCFS2_JOURNAL_ACCESS_WRITE); 530 if (ret < 0) { 531 mlog_errno(ret); 532 goto out_commit; 533 } 534 535 group = (struct ocfs2_group_desc *)group_bh->b_data; 536 bg_ptr = le64_to_cpu(group->bg_next_group); 537 group->bg_next_group = cr->c_blkno; 538 ocfs2_journal_dirty(handle, group_bh); 539 540 ret = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode), 541 main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE); 542 if (ret < 0) { 543 group->bg_next_group = cpu_to_le64(bg_ptr); 544 mlog_errno(ret); 545 goto out_commit; 546 } 547 548 if (input->chain == le16_to_cpu(cl->cl_next_free_rec)) { 549 le16_add_cpu(&cl->cl_next_free_rec, 1); 550 memset(cr, 0, sizeof(struct ocfs2_chain_rec)); 551 } 552 553 cr->c_blkno = cpu_to_le64(input->group); 554 le32_add_cpu(&cr->c_total, input->clusters * cl_bpc); 555 le32_add_cpu(&cr->c_free, input->frees * cl_bpc); 556 557 le32_add_cpu(&fe->id1.bitmap1.i_total, input->clusters *cl_bpc); 558 le32_add_cpu(&fe->id1.bitmap1.i_used, 559 (input->clusters - input->frees) * cl_bpc); 560 le32_add_cpu(&fe->i_clusters, input->clusters); 561 562 ocfs2_journal_dirty(handle, main_bm_bh); 563 564 spin_lock(&OCFS2_I(main_bm_inode)->ip_lock); 565 OCFS2_I(main_bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 566 le64_add_cpu(&fe->i_size, (u64)input->clusters << osb->s_clustersize_bits); 567 spin_unlock(&OCFS2_I(main_bm_inode)->ip_lock); 568 i_size_write(main_bm_inode, le64_to_cpu(fe->i_size)); 569 570 ocfs2_update_super_and_backups(main_bm_inode, input->clusters); 571 572out_commit: 573 ocfs2_commit_trans(osb, handle); 574 575out_free_group_bh: 576 brelse(group_bh); 577 578out_unlock: 579 brelse(main_bm_bh); 580 581 ocfs2_inode_unlock(main_bm_inode, 1); 582 583out_mutex: 584 mutex_unlock(&main_bm_inode->i_mutex); 585 iput(main_bm_inode); 586 587out: 588 return ret; 589} 590