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 #define MLOG_MASK_PREFIX ML_DISK_ALLOC 31 #include <cluster/masklog.h> 32 33 #include "ocfs2.h" 34 35 #include "alloc.h" 36 #include "dlmglue.h" 37 #include "inode.h" 38 #include "journal.h" 39 #include "super.h" 40 #include "sysfile.h" 41 #include "uptodate.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 */ 54 static u16 ocfs2_calc_new_backup_super(struct inode *inode, 55 struct ocfs2_group_desc *gd, 56 int new_clusters, 57 u32 first_new_cluster, 58 u16 cl_cpg, 59 int set) 60 { 61 int i; 62 u16 backups = 0; 63 u32 cluster; 64 u64 blkno, gd_blkno, lgd_blkno = le64_to_cpu(gd->bg_blkno); 65 66 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) { 67 blkno = ocfs2_backup_super_blkno(inode->i_sb, i); 68 cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno); 69 70 gd_blkno = ocfs2_which_cluster_group(inode, cluster); 71 if (gd_blkno < lgd_blkno) 72 continue; 73 else if (gd_blkno > lgd_blkno) 74 break; 75 76 if (set) 77 ocfs2_set_bit(cluster % cl_cpg, 78 (unsigned long *)gd->bg_bitmap); 79 else 80 ocfs2_clear_bit(cluster % cl_cpg, 81 (unsigned long *)gd->bg_bitmap); 82 backups++; 83 } 84 85 mlog_exit_void(); 86 return backups; 87 } 88 89 static int ocfs2_update_last_group_and_inode(handle_t *handle, 90 struct inode *bm_inode, 91 struct buffer_head *bm_bh, 92 struct buffer_head *group_bh, 93 u32 first_new_cluster, 94 int new_clusters) 95 { 96 int ret = 0; 97 struct ocfs2_super *osb = OCFS2_SB(bm_inode->i_sb); 98 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bm_bh->b_data; 99 struct ocfs2_chain_list *cl = &fe->id2.i_chain; 100 struct ocfs2_chain_rec *cr; 101 struct ocfs2_group_desc *group; 102 u16 chain, num_bits, backups = 0; 103 u16 cl_bpc = le16_to_cpu(cl->cl_bpc); 104 u16 cl_cpg = le16_to_cpu(cl->cl_cpg); 105 106 mlog_entry("(new_clusters=%d, first_new_cluster = %u)\n", 107 new_clusters, first_new_cluster); 108 109 ret = ocfs2_journal_access_gd(handle, INODE_CACHE(bm_inode), 110 group_bh, OCFS2_JOURNAL_ACCESS_WRITE); 111 if (ret < 0) { 112 mlog_errno(ret); 113 goto out; 114 } 115 116 group = (struct ocfs2_group_desc *)group_bh->b_data; 117 118 /* update the group first. */ 119 num_bits = new_clusters * cl_bpc; 120 le16_add_cpu(&group->bg_bits, num_bits); 121 le16_add_cpu(&group->bg_free_bits_count, num_bits); 122 123 /* 124 * check whether there are some new backup superblocks exist in 125 * this group and update the group bitmap accordingly. 126 */ 127 if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, 128 OCFS2_FEATURE_COMPAT_BACKUP_SB)) { 129 backups = ocfs2_calc_new_backup_super(bm_inode, 130 group, 131 new_clusters, 132 first_new_cluster, 133 cl_cpg, 1); 134 le16_add_cpu(&group->bg_free_bits_count, -1 * backups); 135 } 136 137 ocfs2_journal_dirty(handle, group_bh); 138 139 /* update the inode accordingly. */ 140 ret = ocfs2_journal_access_di(handle, INODE_CACHE(bm_inode), bm_bh, 141 OCFS2_JOURNAL_ACCESS_WRITE); 142 if (ret < 0) { 143 mlog_errno(ret); 144 goto out_rollback; 145 } 146 147 chain = le16_to_cpu(group->bg_chain); 148 cr = (&cl->cl_recs[chain]); 149 le32_add_cpu(&cr->c_total, num_bits); 150 le32_add_cpu(&cr->c_free, num_bits); 151 le32_add_cpu(&fe->id1.bitmap1.i_total, num_bits); 152 le32_add_cpu(&fe->i_clusters, new_clusters); 153 154 if (backups) { 155 le32_add_cpu(&cr->c_free, -1 * backups); 156 le32_add_cpu(&fe->id1.bitmap1.i_used, backups); 157 } 158 159 spin_lock(&OCFS2_I(bm_inode)->ip_lock); 160 OCFS2_I(bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 161 le64_add_cpu(&fe->i_size, new_clusters << osb->s_clustersize_bits); 162 spin_unlock(&OCFS2_I(bm_inode)->ip_lock); 163 i_size_write(bm_inode, le64_to_cpu(fe->i_size)); 164 165 ocfs2_journal_dirty(handle, bm_bh); 166 167 out_rollback: 168 if (ret < 0) { 169 ocfs2_calc_new_backup_super(bm_inode, 170 group, 171 new_clusters, 172 first_new_cluster, 173 cl_cpg, 0); 174 le16_add_cpu(&group->bg_free_bits_count, backups); 175 le16_add_cpu(&group->bg_bits, -1 * num_bits); 176 le16_add_cpu(&group->bg_free_bits_count, -1 * num_bits); 177 } 178 out: 179 mlog_exit(ret); 180 return ret; 181 } 182 183 static int update_backups(struct inode * inode, u32 clusters, char *data) 184 { 185 int i, ret = 0; 186 u32 cluster; 187 u64 blkno; 188 struct buffer_head *backup = NULL; 189 struct ocfs2_dinode *backup_di = NULL; 190 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 191 192 /* calculate the real backups we need to update. */ 193 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) { 194 blkno = ocfs2_backup_super_blkno(inode->i_sb, i); 195 cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno); 196 if (cluster > clusters) 197 break; 198 199 ret = ocfs2_read_blocks_sync(osb, blkno, 1, &backup); 200 if (ret < 0) { 201 mlog_errno(ret); 202 break; 203 } 204 205 memcpy(backup->b_data, data, inode->i_sb->s_blocksize); 206 207 backup_di = (struct ocfs2_dinode *)backup->b_data; 208 backup_di->i_blkno = cpu_to_le64(blkno); 209 210 ret = ocfs2_write_super_or_backup(osb, backup); 211 brelse(backup); 212 backup = NULL; 213 if (ret < 0) { 214 mlog_errno(ret); 215 break; 216 } 217 } 218 219 return ret; 220 } 221 222 static void ocfs2_update_super_and_backups(struct inode *inode, 223 int new_clusters) 224 { 225 int ret; 226 u32 clusters = 0; 227 struct buffer_head *super_bh = NULL; 228 struct ocfs2_dinode *super_di = NULL; 229 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 230 231 /* 232 * update the superblock last. 233 * It doesn't matter if the write failed. 234 */ 235 ret = ocfs2_read_blocks_sync(osb, OCFS2_SUPER_BLOCK_BLKNO, 1, 236 &super_bh); 237 if (ret < 0) { 238 mlog_errno(ret); 239 goto out; 240 } 241 242 super_di = (struct ocfs2_dinode *)super_bh->b_data; 243 le32_add_cpu(&super_di->i_clusters, new_clusters); 244 clusters = le32_to_cpu(super_di->i_clusters); 245 246 ret = ocfs2_write_super_or_backup(osb, super_bh); 247 if (ret < 0) { 248 mlog_errno(ret); 249 goto out; 250 } 251 252 if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_COMPAT_BACKUP_SB)) 253 ret = update_backups(inode, clusters, super_bh->b_data); 254 255 out: 256 brelse(super_bh); 257 if (ret) 258 printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s" 259 " during fs resize. This condition is not fatal," 260 " but fsck.ocfs2 should be run to fix it\n", 261 osb->dev_str); 262 return; 263 } 264 265 /* 266 * Extend the filesystem to the new number of clusters specified. This entry 267 * point is only used to extend the current filesystem to the end of the last 268 * existing group. 269 */ 270 int ocfs2_group_extend(struct inode * inode, int new_clusters) 271 { 272 int ret; 273 handle_t *handle; 274 struct buffer_head *main_bm_bh = NULL; 275 struct buffer_head *group_bh = NULL; 276 struct inode *main_bm_inode = NULL; 277 struct ocfs2_dinode *fe = NULL; 278 struct ocfs2_group_desc *group = NULL; 279 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 280 u16 cl_bpc; 281 u32 first_new_cluster; 282 u64 lgd_blkno; 283 284 mlog_entry_void(); 285 286 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) 287 return -EROFS; 288 289 if (new_clusters < 0) 290 return -EINVAL; 291 else if (new_clusters == 0) 292 return 0; 293 294 main_bm_inode = ocfs2_get_system_file_inode(osb, 295 GLOBAL_BITMAP_SYSTEM_INODE, 296 OCFS2_INVALID_SLOT); 297 if (!main_bm_inode) { 298 ret = -EINVAL; 299 mlog_errno(ret); 300 goto out; 301 } 302 303 mutex_lock(&main_bm_inode->i_mutex); 304 305 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); 306 if (ret < 0) { 307 mlog_errno(ret); 308 goto out_mutex; 309 } 310 311 fe = (struct ocfs2_dinode *)main_bm_bh->b_data; 312 313 /* main_bm_bh is validated by inode read inside ocfs2_inode_lock(), 314 * so any corruption is a code bug. */ 315 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 316 317 if (le16_to_cpu(fe->id2.i_chain.cl_cpg) != 318 ocfs2_group_bitmap_size(osb->sb, 0, 319 osb->s_feature_incompat) * 8) { 320 mlog(ML_ERROR, "The disk is too old and small. " 321 "Force to do offline resize."); 322 ret = -EINVAL; 323 goto out_unlock; 324 } 325 326 first_new_cluster = le32_to_cpu(fe->i_clusters); 327 lgd_blkno = ocfs2_which_cluster_group(main_bm_inode, 328 first_new_cluster - 1); 329 330 ret = ocfs2_read_group_descriptor(main_bm_inode, fe, lgd_blkno, 331 &group_bh); 332 if (ret < 0) { 333 mlog_errno(ret); 334 goto out_unlock; 335 } 336 group = (struct ocfs2_group_desc *)group_bh->b_data; 337 338 cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc); 339 if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters > 340 le16_to_cpu(fe->id2.i_chain.cl_cpg)) { 341 ret = -EINVAL; 342 goto out_unlock; 343 } 344 345 mlog(0, "extend the last group at %llu, new clusters = %d\n", 346 (unsigned long long)le64_to_cpu(group->bg_blkno), new_clusters); 347 348 handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS); 349 if (IS_ERR(handle)) { 350 mlog_errno(PTR_ERR(handle)); 351 ret = -EINVAL; 352 goto out_unlock; 353 } 354 355 /* update the last group descriptor and inode. */ 356 ret = ocfs2_update_last_group_and_inode(handle, main_bm_inode, 357 main_bm_bh, group_bh, 358 first_new_cluster, 359 new_clusters); 360 if (ret) { 361 mlog_errno(ret); 362 goto out_commit; 363 } 364 365 ocfs2_update_super_and_backups(main_bm_inode, new_clusters); 366 367 out_commit: 368 ocfs2_commit_trans(osb, handle); 369 out_unlock: 370 brelse(group_bh); 371 brelse(main_bm_bh); 372 373 ocfs2_inode_unlock(main_bm_inode, 1); 374 375 out_mutex: 376 mutex_unlock(&main_bm_inode->i_mutex); 377 iput(main_bm_inode); 378 379 out: 380 mlog_exit_void(); 381 return ret; 382 } 383 384 static int ocfs2_check_new_group(struct inode *inode, 385 struct ocfs2_dinode *di, 386 struct ocfs2_new_group_input *input, 387 struct buffer_head *group_bh) 388 { 389 int ret; 390 struct ocfs2_group_desc *gd = 391 (struct ocfs2_group_desc *)group_bh->b_data; 392 u16 cl_bpc = le16_to_cpu(di->id2.i_chain.cl_bpc); 393 394 ret = ocfs2_check_group_descriptor(inode->i_sb, di, group_bh); 395 if (ret) 396 goto out; 397 398 ret = -EINVAL; 399 if (le16_to_cpu(gd->bg_chain) != input->chain) 400 mlog(ML_ERROR, "Group descriptor # %llu has bad chain %u " 401 "while input has %u set.\n", 402 (unsigned long long)le64_to_cpu(gd->bg_blkno), 403 le16_to_cpu(gd->bg_chain), input->chain); 404 else if (le16_to_cpu(gd->bg_bits) != input->clusters * cl_bpc) 405 mlog(ML_ERROR, "Group descriptor # %llu has bit count %u but " 406 "input has %u clusters set\n", 407 (unsigned long long)le64_to_cpu(gd->bg_blkno), 408 le16_to_cpu(gd->bg_bits), input->clusters); 409 else if (le16_to_cpu(gd->bg_free_bits_count) != input->frees * cl_bpc) 410 mlog(ML_ERROR, "Group descriptor # %llu has free bit count %u " 411 "but it should have %u set\n", 412 (unsigned long long)le64_to_cpu(gd->bg_blkno), 413 le16_to_cpu(gd->bg_bits), 414 input->frees * cl_bpc); 415 else 416 ret = 0; 417 418 out: 419 return ret; 420 } 421 422 static int ocfs2_verify_group_and_input(struct inode *inode, 423 struct ocfs2_dinode *di, 424 struct ocfs2_new_group_input *input, 425 struct buffer_head *group_bh) 426 { 427 u16 cl_count = le16_to_cpu(di->id2.i_chain.cl_count); 428 u16 cl_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg); 429 u16 next_free = le16_to_cpu(di->id2.i_chain.cl_next_free_rec); 430 u32 cluster = ocfs2_blocks_to_clusters(inode->i_sb, input->group); 431 u32 total_clusters = le32_to_cpu(di->i_clusters); 432 int ret = -EINVAL; 433 434 if (cluster < total_clusters) 435 mlog(ML_ERROR, "add a group which is in the current volume.\n"); 436 else if (input->chain >= cl_count) 437 mlog(ML_ERROR, "input chain exceeds the limit.\n"); 438 else if (next_free != cl_count && next_free != input->chain) 439 mlog(ML_ERROR, 440 "the add group should be in chain %u\n", next_free); 441 else if (total_clusters + input->clusters < total_clusters) 442 mlog(ML_ERROR, "add group's clusters overflow.\n"); 443 else if (input->clusters > cl_cpg) 444 mlog(ML_ERROR, "the cluster exceeds the maximum of a group\n"); 445 else if (input->frees > input->clusters) 446 mlog(ML_ERROR, "the free cluster exceeds the total clusters\n"); 447 else if (total_clusters % cl_cpg != 0) 448 mlog(ML_ERROR, 449 "the last group isn't full. Use group extend first.\n"); 450 else if (input->group != ocfs2_which_cluster_group(inode, cluster)) 451 mlog(ML_ERROR, "group blkno is invalid\n"); 452 else if ((ret = ocfs2_check_new_group(inode, di, input, group_bh))) 453 mlog(ML_ERROR, "group descriptor check failed.\n"); 454 else 455 ret = 0; 456 457 return ret; 458 } 459 460 /* Add a new group descriptor to global_bitmap. */ 461 int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input) 462 { 463 int ret; 464 handle_t *handle; 465 struct buffer_head *main_bm_bh = NULL; 466 struct inode *main_bm_inode = NULL; 467 struct ocfs2_dinode *fe = NULL; 468 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 469 struct buffer_head *group_bh = NULL; 470 struct ocfs2_group_desc *group = NULL; 471 struct ocfs2_chain_list *cl; 472 struct ocfs2_chain_rec *cr; 473 u16 cl_bpc; 474 475 mlog_entry_void(); 476 477 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) 478 return -EROFS; 479 480 main_bm_inode = ocfs2_get_system_file_inode(osb, 481 GLOBAL_BITMAP_SYSTEM_INODE, 482 OCFS2_INVALID_SLOT); 483 if (!main_bm_inode) { 484 ret = -EINVAL; 485 mlog_errno(ret); 486 goto out; 487 } 488 489 mutex_lock(&main_bm_inode->i_mutex); 490 491 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); 492 if (ret < 0) { 493 mlog_errno(ret); 494 goto out_mutex; 495 } 496 497 fe = (struct ocfs2_dinode *)main_bm_bh->b_data; 498 499 if (le16_to_cpu(fe->id2.i_chain.cl_cpg) != 500 ocfs2_group_bitmap_size(osb->sb, 0, 501 osb->s_feature_incompat) * 8) { 502 mlog(ML_ERROR, "The disk is too old and small." 503 " Force to do offline resize."); 504 ret = -EINVAL; 505 goto out_unlock; 506 } 507 508 ret = ocfs2_read_blocks_sync(osb, input->group, 1, &group_bh); 509 if (ret < 0) { 510 mlog(ML_ERROR, "Can't read the group descriptor # %llu " 511 "from the device.", (unsigned long long)input->group); 512 goto out_unlock; 513 } 514 515 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), group_bh); 516 517 ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh); 518 if (ret) { 519 mlog_errno(ret); 520 goto out_unlock; 521 } 522 523 mlog(0, "Add a new group %llu in chain = %u, length = %u\n", 524 (unsigned long long)input->group, input->chain, input->clusters); 525 526 handle = ocfs2_start_trans(osb, OCFS2_GROUP_ADD_CREDITS); 527 if (IS_ERR(handle)) { 528 mlog_errno(PTR_ERR(handle)); 529 ret = -EINVAL; 530 goto out_unlock; 531 } 532 533 cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc); 534 cl = &fe->id2.i_chain; 535 cr = &cl->cl_recs[input->chain]; 536 537 ret = ocfs2_journal_access_gd(handle, INODE_CACHE(main_bm_inode), 538 group_bh, OCFS2_JOURNAL_ACCESS_WRITE); 539 if (ret < 0) { 540 mlog_errno(ret); 541 goto out_commit; 542 } 543 544 group = (struct ocfs2_group_desc *)group_bh->b_data; 545 group->bg_next_group = cr->c_blkno; 546 ocfs2_journal_dirty(handle, group_bh); 547 548 ret = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode), 549 main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE); 550 if (ret < 0) { 551 mlog_errno(ret); 552 goto out_commit; 553 } 554 555 if (input->chain == le16_to_cpu(cl->cl_next_free_rec)) { 556 le16_add_cpu(&cl->cl_next_free_rec, 1); 557 memset(cr, 0, sizeof(struct ocfs2_chain_rec)); 558 } 559 560 cr->c_blkno = cpu_to_le64(input->group); 561 le32_add_cpu(&cr->c_total, input->clusters * cl_bpc); 562 le32_add_cpu(&cr->c_free, input->frees * cl_bpc); 563 564 le32_add_cpu(&fe->id1.bitmap1.i_total, input->clusters *cl_bpc); 565 le32_add_cpu(&fe->id1.bitmap1.i_used, 566 (input->clusters - input->frees) * cl_bpc); 567 le32_add_cpu(&fe->i_clusters, input->clusters); 568 569 ocfs2_journal_dirty(handle, main_bm_bh); 570 571 spin_lock(&OCFS2_I(main_bm_inode)->ip_lock); 572 OCFS2_I(main_bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 573 le64_add_cpu(&fe->i_size, input->clusters << osb->s_clustersize_bits); 574 spin_unlock(&OCFS2_I(main_bm_inode)->ip_lock); 575 i_size_write(main_bm_inode, le64_to_cpu(fe->i_size)); 576 577 ocfs2_update_super_and_backups(main_bm_inode, input->clusters); 578 579 out_commit: 580 ocfs2_commit_trans(osb, handle); 581 out_unlock: 582 brelse(group_bh); 583 brelse(main_bm_bh); 584 585 ocfs2_inode_unlock(main_bm_inode, 1); 586 587 out_mutex: 588 mutex_unlock(&main_bm_inode->i_mutex); 589 iput(main_bm_inode); 590 591 out: 592 mlog_exit_void(); 593 return ret; 594 } 595