1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) International Business Machines Corp., 2000-2004 4 * Portions Copyright (C) Tino Reichardt, 2012 5 */ 6 7 #include <linux/fs.h> 8 #include <linux/slab.h> 9 #include "jfs_incore.h" 10 #include "jfs_superblock.h" 11 #include "jfs_dmap.h" 12 #include "jfs_imap.h" 13 #include "jfs_lock.h" 14 #include "jfs_metapage.h" 15 #include "jfs_debug.h" 16 #include "jfs_discard.h" 17 18 /* 19 * SERIALIZATION of the Block Allocation Map. 20 * 21 * the working state of the block allocation map is accessed in 22 * two directions: 23 * 24 * 1) allocation and free requests that start at the dmap 25 * level and move up through the dmap control pages (i.e. 26 * the vast majority of requests). 27 * 28 * 2) allocation requests that start at dmap control page 29 * level and work down towards the dmaps. 30 * 31 * the serialization scheme used here is as follows. 32 * 33 * requests which start at the bottom are serialized against each 34 * other through buffers and each requests holds onto its buffers 35 * as it works it way up from a single dmap to the required level 36 * of dmap control page. 37 * requests that start at the top are serialized against each other 38 * and request that start from the bottom by the multiple read/single 39 * write inode lock of the bmap inode. requests starting at the top 40 * take this lock in write mode while request starting at the bottom 41 * take the lock in read mode. a single top-down request may proceed 42 * exclusively while multiple bottoms-up requests may proceed 43 * simultaneously (under the protection of busy buffers). 44 * 45 * in addition to information found in dmaps and dmap control pages, 46 * the working state of the block allocation map also includes read/ 47 * write information maintained in the bmap descriptor (i.e. total 48 * free block count, allocation group level free block counts). 49 * a single exclusive lock (BMAP_LOCK) is used to guard this information 50 * in the face of multiple-bottoms up requests. 51 * (lock ordering: IREAD_LOCK, BMAP_LOCK); 52 * 53 * accesses to the persistent state of the block allocation map (limited 54 * to the persistent bitmaps in dmaps) is guarded by (busy) buffers. 55 */ 56 57 #define BMAP_LOCK_INIT(bmp) mutex_init(&bmp->db_bmaplock) 58 #define BMAP_LOCK(bmp) mutex_lock(&bmp->db_bmaplock) 59 #define BMAP_UNLOCK(bmp) mutex_unlock(&bmp->db_bmaplock) 60 61 /* 62 * forward references 63 */ 64 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno, 65 int nblocks); 66 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval); 67 static int dbBackSplit(dmtree_t * tp, int leafno); 68 static int dbJoin(dmtree_t * tp, int leafno, int newval); 69 static void dbAdjTree(dmtree_t * tp, int leafno, int newval); 70 static int dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc, 71 int level); 72 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results); 73 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno, 74 int nblocks); 75 static int dbAllocNear(struct bmap * bmp, struct dmap * dp, s64 blkno, 76 int nblocks, 77 int l2nb, s64 * results); 78 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno, 79 int nblocks); 80 static int dbAllocDmapLev(struct bmap * bmp, struct dmap * dp, int nblocks, 81 int l2nb, 82 s64 * results); 83 static int dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, 84 s64 * results); 85 static int dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno, 86 s64 * results); 87 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks); 88 static int dbFindBits(u32 word, int l2nb); 89 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno); 90 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx); 91 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno, 92 int nblocks); 93 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno, 94 int nblocks); 95 static int dbMaxBud(u8 * cp); 96 static int blkstol2(s64 nb); 97 98 static int cntlz(u32 value); 99 static int cnttz(u32 word); 100 101 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno, 102 int nblocks); 103 static int dbInitDmap(struct dmap * dp, s64 blkno, int nblocks); 104 static int dbInitDmapTree(struct dmap * dp); 105 static int dbInitTree(struct dmaptree * dtp); 106 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i); 107 static int dbGetL2AGSize(s64 nblocks); 108 109 /* 110 * buddy table 111 * 112 * table used for determining buddy sizes within characters of 113 * dmap bitmap words. the characters themselves serve as indexes 114 * into the table, with the table elements yielding the maximum 115 * binary buddy of free bits within the character. 116 */ 117 static const s8 budtab[256] = { 118 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 119 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 120 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 121 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 122 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 123 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 124 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 125 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 126 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 127 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 128 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 129 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 130 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 131 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 132 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 133 2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, -1 134 }; 135 136 /* 137 * NAME: dbMount() 138 * 139 * FUNCTION: initializate the block allocation map. 140 * 141 * memory is allocated for the in-core bmap descriptor and 142 * the in-core descriptor is initialized from disk. 143 * 144 * PARAMETERS: 145 * ipbmap - pointer to in-core inode for the block map. 146 * 147 * RETURN VALUES: 148 * 0 - success 149 * -ENOMEM - insufficient memory 150 * -EIO - i/o error 151 * -EINVAL - wrong bmap data 152 */ 153 int dbMount(struct inode *ipbmap) 154 { 155 struct bmap *bmp; 156 struct dbmap_disk *dbmp_le; 157 struct metapage *mp; 158 int i, err; 159 160 /* 161 * allocate/initialize the in-memory bmap descriptor 162 */ 163 /* allocate memory for the in-memory bmap descriptor */ 164 bmp = kmalloc(sizeof(struct bmap), GFP_KERNEL); 165 if (bmp == NULL) 166 return -ENOMEM; 167 168 /* read the on-disk bmap descriptor. */ 169 mp = read_metapage(ipbmap, 170 BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage, 171 PSIZE, 0); 172 if (mp == NULL) { 173 err = -EIO; 174 goto err_kfree_bmp; 175 } 176 177 /* copy the on-disk bmap descriptor to its in-memory version. */ 178 dbmp_le = (struct dbmap_disk *) mp->data; 179 bmp->db_mapsize = le64_to_cpu(dbmp_le->dn_mapsize); 180 bmp->db_nfree = le64_to_cpu(dbmp_le->dn_nfree); 181 182 bmp->db_l2nbperpage = le32_to_cpu(dbmp_le->dn_l2nbperpage); 183 if (bmp->db_l2nbperpage > L2PSIZE - L2MINBLOCKSIZE) { 184 err = -EINVAL; 185 goto err_release_metapage; 186 } 187 188 bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag); 189 if (!bmp->db_numag) { 190 err = -EINVAL; 191 goto err_release_metapage; 192 } 193 194 bmp->db_maxlevel = le32_to_cpu(dbmp_le->dn_maxlevel); 195 bmp->db_maxag = le32_to_cpu(dbmp_le->dn_maxag); 196 bmp->db_agpref = le32_to_cpu(dbmp_le->dn_agpref); 197 bmp->db_aglevel = le32_to_cpu(dbmp_le->dn_aglevel); 198 bmp->db_agheight = le32_to_cpu(dbmp_le->dn_agheight); 199 bmp->db_agwidth = le32_to_cpu(dbmp_le->dn_agwidth); 200 bmp->db_agstart = le32_to_cpu(dbmp_le->dn_agstart); 201 bmp->db_agl2size = le32_to_cpu(dbmp_le->dn_agl2size); 202 if (bmp->db_agl2size > L2MAXL2SIZE - L2MAXAG || 203 bmp->db_agl2size < 0) { 204 err = -EINVAL; 205 goto err_release_metapage; 206 } 207 208 if (((bmp->db_mapsize - 1) >> bmp->db_agl2size) > MAXAG) { 209 err = -EINVAL; 210 goto err_release_metapage; 211 } 212 213 for (i = 0; i < MAXAG; i++) 214 bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]); 215 bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize); 216 bmp->db_maxfreebud = dbmp_le->dn_maxfreebud; 217 218 /* release the buffer. */ 219 release_metapage(mp); 220 221 /* bind the bmap inode and the bmap descriptor to each other. */ 222 bmp->db_ipbmap = ipbmap; 223 JFS_SBI(ipbmap->i_sb)->bmap = bmp; 224 225 memset(bmp->db_active, 0, sizeof(bmp->db_active)); 226 227 /* 228 * allocate/initialize the bmap lock 229 */ 230 BMAP_LOCK_INIT(bmp); 231 232 return (0); 233 234 err_release_metapage: 235 release_metapage(mp); 236 err_kfree_bmp: 237 kfree(bmp); 238 return err; 239 } 240 241 242 /* 243 * NAME: dbUnmount() 244 * 245 * FUNCTION: terminate the block allocation map in preparation for 246 * file system unmount. 247 * 248 * the in-core bmap descriptor is written to disk and 249 * the memory for this descriptor is freed. 250 * 251 * PARAMETERS: 252 * ipbmap - pointer to in-core inode for the block map. 253 * 254 * RETURN VALUES: 255 * 0 - success 256 * -EIO - i/o error 257 */ 258 int dbUnmount(struct inode *ipbmap, int mounterror) 259 { 260 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 261 262 if (!(mounterror || isReadOnly(ipbmap))) 263 dbSync(ipbmap); 264 265 /* 266 * Invalidate the page cache buffers 267 */ 268 truncate_inode_pages(ipbmap->i_mapping, 0); 269 270 /* free the memory for the in-memory bmap. */ 271 kfree(bmp); 272 273 return (0); 274 } 275 276 /* 277 * dbSync() 278 */ 279 int dbSync(struct inode *ipbmap) 280 { 281 struct dbmap_disk *dbmp_le; 282 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 283 struct metapage *mp; 284 int i; 285 286 /* 287 * write bmap global control page 288 */ 289 /* get the buffer for the on-disk bmap descriptor. */ 290 mp = read_metapage(ipbmap, 291 BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage, 292 PSIZE, 0); 293 if (mp == NULL) { 294 jfs_err("dbSync: read_metapage failed!"); 295 return -EIO; 296 } 297 /* copy the in-memory version of the bmap to the on-disk version */ 298 dbmp_le = (struct dbmap_disk *) mp->data; 299 dbmp_le->dn_mapsize = cpu_to_le64(bmp->db_mapsize); 300 dbmp_le->dn_nfree = cpu_to_le64(bmp->db_nfree); 301 dbmp_le->dn_l2nbperpage = cpu_to_le32(bmp->db_l2nbperpage); 302 dbmp_le->dn_numag = cpu_to_le32(bmp->db_numag); 303 dbmp_le->dn_maxlevel = cpu_to_le32(bmp->db_maxlevel); 304 dbmp_le->dn_maxag = cpu_to_le32(bmp->db_maxag); 305 dbmp_le->dn_agpref = cpu_to_le32(bmp->db_agpref); 306 dbmp_le->dn_aglevel = cpu_to_le32(bmp->db_aglevel); 307 dbmp_le->dn_agheight = cpu_to_le32(bmp->db_agheight); 308 dbmp_le->dn_agwidth = cpu_to_le32(bmp->db_agwidth); 309 dbmp_le->dn_agstart = cpu_to_le32(bmp->db_agstart); 310 dbmp_le->dn_agl2size = cpu_to_le32(bmp->db_agl2size); 311 for (i = 0; i < MAXAG; i++) 312 dbmp_le->dn_agfree[i] = cpu_to_le64(bmp->db_agfree[i]); 313 dbmp_le->dn_agsize = cpu_to_le64(bmp->db_agsize); 314 dbmp_le->dn_maxfreebud = bmp->db_maxfreebud; 315 316 /* write the buffer */ 317 write_metapage(mp); 318 319 /* 320 * write out dirty pages of bmap 321 */ 322 filemap_write_and_wait(ipbmap->i_mapping); 323 324 diWriteSpecial(ipbmap, 0); 325 326 return (0); 327 } 328 329 /* 330 * NAME: dbFree() 331 * 332 * FUNCTION: free the specified block range from the working block 333 * allocation map. 334 * 335 * the blocks will be free from the working map one dmap 336 * at a time. 337 * 338 * PARAMETERS: 339 * ip - pointer to in-core inode; 340 * blkno - starting block number to be freed. 341 * nblocks - number of blocks to be freed. 342 * 343 * RETURN VALUES: 344 * 0 - success 345 * -EIO - i/o error 346 */ 347 int dbFree(struct inode *ip, s64 blkno, s64 nblocks) 348 { 349 struct metapage *mp; 350 struct dmap *dp; 351 int nb, rc; 352 s64 lblkno, rem; 353 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 354 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap; 355 struct super_block *sb = ipbmap->i_sb; 356 357 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP); 358 359 /* block to be freed better be within the mapsize. */ 360 if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) { 361 IREAD_UNLOCK(ipbmap); 362 printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n", 363 (unsigned long long) blkno, 364 (unsigned long long) nblocks); 365 jfs_error(ip->i_sb, "block to be freed is outside the map\n"); 366 return -EIO; 367 } 368 369 /** 370 * TRIM the blocks, when mounted with discard option 371 */ 372 if (JFS_SBI(sb)->flag & JFS_DISCARD) 373 if (JFS_SBI(sb)->minblks_trim <= nblocks) 374 jfs_issue_discard(ipbmap, blkno, nblocks); 375 376 /* 377 * free the blocks a dmap at a time. 378 */ 379 mp = NULL; 380 for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) { 381 /* release previous dmap if any */ 382 if (mp) { 383 write_metapage(mp); 384 } 385 386 /* get the buffer for the current dmap. */ 387 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 388 mp = read_metapage(ipbmap, lblkno, PSIZE, 0); 389 if (mp == NULL) { 390 IREAD_UNLOCK(ipbmap); 391 return -EIO; 392 } 393 dp = (struct dmap *) mp->data; 394 395 /* determine the number of blocks to be freed from 396 * this dmap. 397 */ 398 nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1))); 399 400 /* free the blocks. */ 401 if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) { 402 jfs_error(ip->i_sb, "error in block map\n"); 403 release_metapage(mp); 404 IREAD_UNLOCK(ipbmap); 405 return (rc); 406 } 407 } 408 409 /* write the last buffer. */ 410 if (mp) 411 write_metapage(mp); 412 413 IREAD_UNLOCK(ipbmap); 414 415 return (0); 416 } 417 418 419 /* 420 * NAME: dbUpdatePMap() 421 * 422 * FUNCTION: update the allocation state (free or allocate) of the 423 * specified block range in the persistent block allocation map. 424 * 425 * the blocks will be updated in the persistent map one 426 * dmap at a time. 427 * 428 * PARAMETERS: 429 * ipbmap - pointer to in-core inode for the block map. 430 * free - 'true' if block range is to be freed from the persistent 431 * map; 'false' if it is to be allocated. 432 * blkno - starting block number of the range. 433 * nblocks - number of contiguous blocks in the range. 434 * tblk - transaction block; 435 * 436 * RETURN VALUES: 437 * 0 - success 438 * -EIO - i/o error 439 */ 440 int 441 dbUpdatePMap(struct inode *ipbmap, 442 int free, s64 blkno, s64 nblocks, struct tblock * tblk) 443 { 444 int nblks, dbitno, wbitno, rbits; 445 int word, nbits, nwords; 446 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 447 s64 lblkno, rem, lastlblkno; 448 u32 mask; 449 struct dmap *dp; 450 struct metapage *mp; 451 struct jfs_log *log; 452 int lsn, difft, diffp; 453 unsigned long flags; 454 455 /* the blocks better be within the mapsize. */ 456 if (blkno + nblocks > bmp->db_mapsize) { 457 printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n", 458 (unsigned long long) blkno, 459 (unsigned long long) nblocks); 460 jfs_error(ipbmap->i_sb, "blocks are outside the map\n"); 461 return -EIO; 462 } 463 464 /* compute delta of transaction lsn from log syncpt */ 465 lsn = tblk->lsn; 466 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log; 467 logdiff(difft, lsn, log); 468 469 /* 470 * update the block state a dmap at a time. 471 */ 472 mp = NULL; 473 lastlblkno = 0; 474 for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) { 475 /* get the buffer for the current dmap. */ 476 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 477 if (lblkno != lastlblkno) { 478 if (mp) { 479 write_metapage(mp); 480 } 481 482 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 483 0); 484 if (mp == NULL) 485 return -EIO; 486 metapage_wait_for_io(mp); 487 } 488 dp = (struct dmap *) mp->data; 489 490 /* determine the bit number and word within the dmap of 491 * the starting block. also determine how many blocks 492 * are to be updated within this dmap. 493 */ 494 dbitno = blkno & (BPERDMAP - 1); 495 word = dbitno >> L2DBWORD; 496 nblks = min(rem, (s64)BPERDMAP - dbitno); 497 498 /* update the bits of the dmap words. the first and last 499 * words may only have a subset of their bits updated. if 500 * this is the case, we'll work against that word (i.e. 501 * partial first and/or last) only in a single pass. a 502 * single pass will also be used to update all words that 503 * are to have all their bits updated. 504 */ 505 for (rbits = nblks; rbits > 0; 506 rbits -= nbits, dbitno += nbits) { 507 /* determine the bit number within the word and 508 * the number of bits within the word. 509 */ 510 wbitno = dbitno & (DBWORD - 1); 511 nbits = min(rbits, DBWORD - wbitno); 512 513 /* check if only part of the word is to be updated. */ 514 if (nbits < DBWORD) { 515 /* update (free or allocate) the bits 516 * in this word. 517 */ 518 mask = 519 (ONES << (DBWORD - nbits) >> wbitno); 520 if (free) 521 dp->pmap[word] &= 522 cpu_to_le32(~mask); 523 else 524 dp->pmap[word] |= 525 cpu_to_le32(mask); 526 527 word += 1; 528 } else { 529 /* one or more words are to have all 530 * their bits updated. determine how 531 * many words and how many bits. 532 */ 533 nwords = rbits >> L2DBWORD; 534 nbits = nwords << L2DBWORD; 535 536 /* update (free or allocate) the bits 537 * in these words. 538 */ 539 if (free) 540 memset(&dp->pmap[word], 0, 541 nwords * 4); 542 else 543 memset(&dp->pmap[word], (int) ONES, 544 nwords * 4); 545 546 word += nwords; 547 } 548 } 549 550 /* 551 * update dmap lsn 552 */ 553 if (lblkno == lastlblkno) 554 continue; 555 556 lastlblkno = lblkno; 557 558 LOGSYNC_LOCK(log, flags); 559 if (mp->lsn != 0) { 560 /* inherit older/smaller lsn */ 561 logdiff(diffp, mp->lsn, log); 562 if (difft < diffp) { 563 mp->lsn = lsn; 564 565 /* move bp after tblock in logsync list */ 566 list_move(&mp->synclist, &tblk->synclist); 567 } 568 569 /* inherit younger/larger clsn */ 570 logdiff(difft, tblk->clsn, log); 571 logdiff(diffp, mp->clsn, log); 572 if (difft > diffp) 573 mp->clsn = tblk->clsn; 574 } else { 575 mp->log = log; 576 mp->lsn = lsn; 577 578 /* insert bp after tblock in logsync list */ 579 log->count++; 580 list_add(&mp->synclist, &tblk->synclist); 581 582 mp->clsn = tblk->clsn; 583 } 584 LOGSYNC_UNLOCK(log, flags); 585 } 586 587 /* write the last buffer. */ 588 if (mp) { 589 write_metapage(mp); 590 } 591 592 return (0); 593 } 594 595 596 /* 597 * NAME: dbNextAG() 598 * 599 * FUNCTION: find the preferred allocation group for new allocations. 600 * 601 * Within the allocation groups, we maintain a preferred 602 * allocation group which consists of a group with at least 603 * average free space. It is the preferred group that we target 604 * new inode allocation towards. The tie-in between inode 605 * allocation and block allocation occurs as we allocate the 606 * first (data) block of an inode and specify the inode (block) 607 * as the allocation hint for this block. 608 * 609 * We try to avoid having more than one open file growing in 610 * an allocation group, as this will lead to fragmentation. 611 * This differs from the old OS/2 method of trying to keep 612 * empty ags around for large allocations. 613 * 614 * PARAMETERS: 615 * ipbmap - pointer to in-core inode for the block map. 616 * 617 * RETURN VALUES: 618 * the preferred allocation group number. 619 */ 620 int dbNextAG(struct inode *ipbmap) 621 { 622 s64 avgfree; 623 int agpref; 624 s64 hwm = 0; 625 int i; 626 int next_best = -1; 627 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 628 629 BMAP_LOCK(bmp); 630 631 /* determine the average number of free blocks within the ags. */ 632 avgfree = (u32)bmp->db_nfree / bmp->db_numag; 633 634 /* 635 * if the current preferred ag does not have an active allocator 636 * and has at least average freespace, return it 637 */ 638 agpref = bmp->db_agpref; 639 if ((atomic_read(&bmp->db_active[agpref]) == 0) && 640 (bmp->db_agfree[agpref] >= avgfree)) 641 goto unlock; 642 643 /* From the last preferred ag, find the next one with at least 644 * average free space. 645 */ 646 for (i = 0 ; i < bmp->db_numag; i++, agpref++) { 647 if (agpref == bmp->db_numag) 648 agpref = 0; 649 650 if (atomic_read(&bmp->db_active[agpref])) 651 /* open file is currently growing in this ag */ 652 continue; 653 if (bmp->db_agfree[agpref] >= avgfree) { 654 /* Return this one */ 655 bmp->db_agpref = agpref; 656 goto unlock; 657 } else if (bmp->db_agfree[agpref] > hwm) { 658 /* Less than avg. freespace, but best so far */ 659 hwm = bmp->db_agfree[agpref]; 660 next_best = agpref; 661 } 662 } 663 664 /* 665 * If no inactive ag was found with average freespace, use the 666 * next best 667 */ 668 if (next_best != -1) 669 bmp->db_agpref = next_best; 670 /* else leave db_agpref unchanged */ 671 unlock: 672 BMAP_UNLOCK(bmp); 673 674 /* return the preferred group. 675 */ 676 return (bmp->db_agpref); 677 } 678 679 /* 680 * NAME: dbAlloc() 681 * 682 * FUNCTION: attempt to allocate a specified number of contiguous free 683 * blocks from the working allocation block map. 684 * 685 * the block allocation policy uses hints and a multi-step 686 * approach. 687 * 688 * for allocation requests smaller than the number of blocks 689 * per dmap, we first try to allocate the new blocks 690 * immediately following the hint. if these blocks are not 691 * available, we try to allocate blocks near the hint. if 692 * no blocks near the hint are available, we next try to 693 * allocate within the same dmap as contains the hint. 694 * 695 * if no blocks are available in the dmap or the allocation 696 * request is larger than the dmap size, we try to allocate 697 * within the same allocation group as contains the hint. if 698 * this does not succeed, we finally try to allocate anywhere 699 * within the aggregate. 700 * 701 * we also try to allocate anywhere within the aggregate 702 * for allocation requests larger than the allocation group 703 * size or requests that specify no hint value. 704 * 705 * PARAMETERS: 706 * ip - pointer to in-core inode; 707 * hint - allocation hint. 708 * nblocks - number of contiguous blocks in the range. 709 * results - on successful return, set to the starting block number 710 * of the newly allocated contiguous range. 711 * 712 * RETURN VALUES: 713 * 0 - success 714 * -ENOSPC - insufficient disk resources 715 * -EIO - i/o error 716 */ 717 int dbAlloc(struct inode *ip, s64 hint, s64 nblocks, s64 * results) 718 { 719 int rc, agno; 720 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 721 struct bmap *bmp; 722 struct metapage *mp; 723 s64 lblkno, blkno; 724 struct dmap *dp; 725 int l2nb; 726 s64 mapSize; 727 int writers; 728 729 /* assert that nblocks is valid */ 730 assert(nblocks > 0); 731 732 /* get the log2 number of blocks to be allocated. 733 * if the number of blocks is not a log2 multiple, 734 * it will be rounded up to the next log2 multiple. 735 */ 736 l2nb = BLKSTOL2(nblocks); 737 738 bmp = JFS_SBI(ip->i_sb)->bmap; 739 740 mapSize = bmp->db_mapsize; 741 742 /* the hint should be within the map */ 743 if (hint >= mapSize) { 744 jfs_error(ip->i_sb, "the hint is outside the map\n"); 745 return -EIO; 746 } 747 748 /* if the number of blocks to be allocated is greater than the 749 * allocation group size, try to allocate anywhere. 750 */ 751 if (l2nb > bmp->db_agl2size) { 752 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP); 753 754 rc = dbAllocAny(bmp, nblocks, l2nb, results); 755 756 goto write_unlock; 757 } 758 759 /* 760 * If no hint, let dbNextAG recommend an allocation group 761 */ 762 if (hint == 0) 763 goto pref_ag; 764 765 /* we would like to allocate close to the hint. adjust the 766 * hint to the block following the hint since the allocators 767 * will start looking for free space starting at this point. 768 */ 769 blkno = hint + 1; 770 771 if (blkno >= bmp->db_mapsize) 772 goto pref_ag; 773 774 agno = blkno >> bmp->db_agl2size; 775 776 /* check if blkno crosses over into a new allocation group. 777 * if so, check if we should allow allocations within this 778 * allocation group. 779 */ 780 if ((blkno & (bmp->db_agsize - 1)) == 0) 781 /* check if the AG is currently being written to. 782 * if so, call dbNextAG() to find a non-busy 783 * AG with sufficient free space. 784 */ 785 if (atomic_read(&bmp->db_active[agno])) 786 goto pref_ag; 787 788 /* check if the allocation request size can be satisfied from a 789 * single dmap. if so, try to allocate from the dmap containing 790 * the hint using a tiered strategy. 791 */ 792 if (nblocks <= BPERDMAP) { 793 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP); 794 795 /* get the buffer for the dmap containing the hint. 796 */ 797 rc = -EIO; 798 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 799 mp = read_metapage(ipbmap, lblkno, PSIZE, 0); 800 if (mp == NULL) 801 goto read_unlock; 802 803 dp = (struct dmap *) mp->data; 804 805 /* first, try to satisfy the allocation request with the 806 * blocks beginning at the hint. 807 */ 808 if ((rc = dbAllocNext(bmp, dp, blkno, (int) nblocks)) 809 != -ENOSPC) { 810 if (rc == 0) { 811 *results = blkno; 812 mark_metapage_dirty(mp); 813 } 814 815 release_metapage(mp); 816 goto read_unlock; 817 } 818 819 writers = atomic_read(&bmp->db_active[agno]); 820 if ((writers > 1) || 821 ((writers == 1) && (JFS_IP(ip)->active_ag != agno))) { 822 /* 823 * Someone else is writing in this allocation 824 * group. To avoid fragmenting, try another ag 825 */ 826 release_metapage(mp); 827 IREAD_UNLOCK(ipbmap); 828 goto pref_ag; 829 } 830 831 /* next, try to satisfy the allocation request with blocks 832 * near the hint. 833 */ 834 if ((rc = 835 dbAllocNear(bmp, dp, blkno, (int) nblocks, l2nb, results)) 836 != -ENOSPC) { 837 if (rc == 0) 838 mark_metapage_dirty(mp); 839 840 release_metapage(mp); 841 goto read_unlock; 842 } 843 844 /* try to satisfy the allocation request with blocks within 845 * the same dmap as the hint. 846 */ 847 if ((rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results)) 848 != -ENOSPC) { 849 if (rc == 0) 850 mark_metapage_dirty(mp); 851 852 release_metapage(mp); 853 goto read_unlock; 854 } 855 856 release_metapage(mp); 857 IREAD_UNLOCK(ipbmap); 858 } 859 860 /* try to satisfy the allocation request with blocks within 861 * the same allocation group as the hint. 862 */ 863 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP); 864 if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) != -ENOSPC) 865 goto write_unlock; 866 867 IWRITE_UNLOCK(ipbmap); 868 869 870 pref_ag: 871 /* 872 * Let dbNextAG recommend a preferred allocation group 873 */ 874 agno = dbNextAG(ipbmap); 875 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP); 876 877 /* Try to allocate within this allocation group. if that fails, try to 878 * allocate anywhere in the map. 879 */ 880 if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) == -ENOSPC) 881 rc = dbAllocAny(bmp, nblocks, l2nb, results); 882 883 write_unlock: 884 IWRITE_UNLOCK(ipbmap); 885 886 return (rc); 887 888 read_unlock: 889 IREAD_UNLOCK(ipbmap); 890 891 return (rc); 892 } 893 894 /* 895 * NAME: dbReAlloc() 896 * 897 * FUNCTION: attempt to extend a current allocation by a specified 898 * number of blocks. 899 * 900 * this routine attempts to satisfy the allocation request 901 * by first trying to extend the existing allocation in 902 * place by allocating the additional blocks as the blocks 903 * immediately following the current allocation. if these 904 * blocks are not available, this routine will attempt to 905 * allocate a new set of contiguous blocks large enough 906 * to cover the existing allocation plus the additional 907 * number of blocks required. 908 * 909 * PARAMETERS: 910 * ip - pointer to in-core inode requiring allocation. 911 * blkno - starting block of the current allocation. 912 * nblocks - number of contiguous blocks within the current 913 * allocation. 914 * addnblocks - number of blocks to add to the allocation. 915 * results - on successful return, set to the starting block number 916 * of the existing allocation if the existing allocation 917 * was extended in place or to a newly allocated contiguous 918 * range if the existing allocation could not be extended 919 * in place. 920 * 921 * RETURN VALUES: 922 * 0 - success 923 * -ENOSPC - insufficient disk resources 924 * -EIO - i/o error 925 */ 926 int 927 dbReAlloc(struct inode *ip, 928 s64 blkno, s64 nblocks, s64 addnblocks, s64 * results) 929 { 930 int rc; 931 932 /* try to extend the allocation in place. 933 */ 934 if ((rc = dbExtend(ip, blkno, nblocks, addnblocks)) == 0) { 935 *results = blkno; 936 return (0); 937 } else { 938 if (rc != -ENOSPC) 939 return (rc); 940 } 941 942 /* could not extend the allocation in place, so allocate a 943 * new set of blocks for the entire request (i.e. try to get 944 * a range of contiguous blocks large enough to cover the 945 * existing allocation plus the additional blocks.) 946 */ 947 return (dbAlloc 948 (ip, blkno + nblocks - 1, addnblocks + nblocks, results)); 949 } 950 951 952 /* 953 * NAME: dbExtend() 954 * 955 * FUNCTION: attempt to extend a current allocation by a specified 956 * number of blocks. 957 * 958 * this routine attempts to satisfy the allocation request 959 * by first trying to extend the existing allocation in 960 * place by allocating the additional blocks as the blocks 961 * immediately following the current allocation. 962 * 963 * PARAMETERS: 964 * ip - pointer to in-core inode requiring allocation. 965 * blkno - starting block of the current allocation. 966 * nblocks - number of contiguous blocks within the current 967 * allocation. 968 * addnblocks - number of blocks to add to the allocation. 969 * 970 * RETURN VALUES: 971 * 0 - success 972 * -ENOSPC - insufficient disk resources 973 * -EIO - i/o error 974 */ 975 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks) 976 { 977 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 978 s64 lblkno, lastblkno, extblkno; 979 uint rel_block; 980 struct metapage *mp; 981 struct dmap *dp; 982 int rc; 983 struct inode *ipbmap = sbi->ipbmap; 984 struct bmap *bmp; 985 986 /* 987 * We don't want a non-aligned extent to cross a page boundary 988 */ 989 if (((rel_block = blkno & (sbi->nbperpage - 1))) && 990 (rel_block + nblocks + addnblocks > sbi->nbperpage)) 991 return -ENOSPC; 992 993 /* get the last block of the current allocation */ 994 lastblkno = blkno + nblocks - 1; 995 996 /* determine the block number of the block following 997 * the existing allocation. 998 */ 999 extblkno = lastblkno + 1; 1000 1001 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP); 1002 1003 /* better be within the file system */ 1004 bmp = sbi->bmap; 1005 if (lastblkno < 0 || lastblkno >= bmp->db_mapsize) { 1006 IREAD_UNLOCK(ipbmap); 1007 jfs_error(ip->i_sb, "the block is outside the filesystem\n"); 1008 return -EIO; 1009 } 1010 1011 /* we'll attempt to extend the current allocation in place by 1012 * allocating the additional blocks as the blocks immediately 1013 * following the current allocation. we only try to extend the 1014 * current allocation in place if the number of additional blocks 1015 * can fit into a dmap, the last block of the current allocation 1016 * is not the last block of the file system, and the start of the 1017 * inplace extension is not on an allocation group boundary. 1018 */ 1019 if (addnblocks > BPERDMAP || extblkno >= bmp->db_mapsize || 1020 (extblkno & (bmp->db_agsize - 1)) == 0) { 1021 IREAD_UNLOCK(ipbmap); 1022 return -ENOSPC; 1023 } 1024 1025 /* get the buffer for the dmap containing the first block 1026 * of the extension. 1027 */ 1028 lblkno = BLKTODMAP(extblkno, bmp->db_l2nbperpage); 1029 mp = read_metapage(ipbmap, lblkno, PSIZE, 0); 1030 if (mp == NULL) { 1031 IREAD_UNLOCK(ipbmap); 1032 return -EIO; 1033 } 1034 1035 dp = (struct dmap *) mp->data; 1036 1037 /* try to allocate the blocks immediately following the 1038 * current allocation. 1039 */ 1040 rc = dbAllocNext(bmp, dp, extblkno, (int) addnblocks); 1041 1042 IREAD_UNLOCK(ipbmap); 1043 1044 /* were we successful ? */ 1045 if (rc == 0) 1046 write_metapage(mp); 1047 else 1048 /* we were not successful */ 1049 release_metapage(mp); 1050 1051 return (rc); 1052 } 1053 1054 1055 /* 1056 * NAME: dbAllocNext() 1057 * 1058 * FUNCTION: attempt to allocate the blocks of the specified block 1059 * range within a dmap. 1060 * 1061 * PARAMETERS: 1062 * bmp - pointer to bmap descriptor 1063 * dp - pointer to dmap. 1064 * blkno - starting block number of the range. 1065 * nblocks - number of contiguous free blocks of the range. 1066 * 1067 * RETURN VALUES: 1068 * 0 - success 1069 * -ENOSPC - insufficient disk resources 1070 * -EIO - i/o error 1071 * 1072 * serialization: IREAD_LOCK(ipbmap) held on entry/exit; 1073 */ 1074 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno, 1075 int nblocks) 1076 { 1077 int dbitno, word, rembits, nb, nwords, wbitno, nw; 1078 int l2size; 1079 s8 *leaf; 1080 u32 mask; 1081 1082 if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) { 1083 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n"); 1084 return -EIO; 1085 } 1086 1087 /* pick up a pointer to the leaves of the dmap tree. 1088 */ 1089 leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx); 1090 1091 /* determine the bit number and word within the dmap of the 1092 * starting block. 1093 */ 1094 dbitno = blkno & (BPERDMAP - 1); 1095 word = dbitno >> L2DBWORD; 1096 1097 /* check if the specified block range is contained within 1098 * this dmap. 1099 */ 1100 if (dbitno + nblocks > BPERDMAP) 1101 return -ENOSPC; 1102 1103 /* check if the starting leaf indicates that anything 1104 * is free. 1105 */ 1106 if (leaf[word] == NOFREE) 1107 return -ENOSPC; 1108 1109 /* check the dmaps words corresponding to block range to see 1110 * if the block range is free. not all bits of the first and 1111 * last words may be contained within the block range. if this 1112 * is the case, we'll work against those words (i.e. partial first 1113 * and/or last) on an individual basis (a single pass) and examine 1114 * the actual bits to determine if they are free. a single pass 1115 * will be used for all dmap words fully contained within the 1116 * specified range. within this pass, the leaves of the dmap 1117 * tree will be examined to determine if the blocks are free. a 1118 * single leaf may describe the free space of multiple dmap 1119 * words, so we may visit only a subset of the actual leaves 1120 * corresponding to the dmap words of the block range. 1121 */ 1122 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) { 1123 /* determine the bit number within the word and 1124 * the number of bits within the word. 1125 */ 1126 wbitno = dbitno & (DBWORD - 1); 1127 nb = min(rembits, DBWORD - wbitno); 1128 1129 /* check if only part of the word is to be examined. 1130 */ 1131 if (nb < DBWORD) { 1132 /* check if the bits are free. 1133 */ 1134 mask = (ONES << (DBWORD - nb) >> wbitno); 1135 if ((mask & ~le32_to_cpu(dp->wmap[word])) != mask) 1136 return -ENOSPC; 1137 1138 word += 1; 1139 } else { 1140 /* one or more dmap words are fully contained 1141 * within the block range. determine how many 1142 * words and how many bits. 1143 */ 1144 nwords = rembits >> L2DBWORD; 1145 nb = nwords << L2DBWORD; 1146 1147 /* now examine the appropriate leaves to determine 1148 * if the blocks are free. 1149 */ 1150 while (nwords > 0) { 1151 /* does the leaf describe any free space ? 1152 */ 1153 if (leaf[word] < BUDMIN) 1154 return -ENOSPC; 1155 1156 /* determine the l2 number of bits provided 1157 * by this leaf. 1158 */ 1159 l2size = 1160 min_t(int, leaf[word], NLSTOL2BSZ(nwords)); 1161 1162 /* determine how many words were handled. 1163 */ 1164 nw = BUDSIZE(l2size, BUDMIN); 1165 1166 nwords -= nw; 1167 word += nw; 1168 } 1169 } 1170 } 1171 1172 /* allocate the blocks. 1173 */ 1174 return (dbAllocDmap(bmp, dp, blkno, nblocks)); 1175 } 1176 1177 1178 /* 1179 * NAME: dbAllocNear() 1180 * 1181 * FUNCTION: attempt to allocate a number of contiguous free blocks near 1182 * a specified block (hint) within a dmap. 1183 * 1184 * starting with the dmap leaf that covers the hint, we'll 1185 * check the next four contiguous leaves for sufficient free 1186 * space. if sufficient free space is found, we'll allocate 1187 * the desired free space. 1188 * 1189 * PARAMETERS: 1190 * bmp - pointer to bmap descriptor 1191 * dp - pointer to dmap. 1192 * blkno - block number to allocate near. 1193 * nblocks - actual number of contiguous free blocks desired. 1194 * l2nb - log2 number of contiguous free blocks desired. 1195 * results - on successful return, set to the starting block number 1196 * of the newly allocated range. 1197 * 1198 * RETURN VALUES: 1199 * 0 - success 1200 * -ENOSPC - insufficient disk resources 1201 * -EIO - i/o error 1202 * 1203 * serialization: IREAD_LOCK(ipbmap) held on entry/exit; 1204 */ 1205 static int 1206 dbAllocNear(struct bmap * bmp, 1207 struct dmap * dp, s64 blkno, int nblocks, int l2nb, s64 * results) 1208 { 1209 int word, lword, rc; 1210 s8 *leaf; 1211 1212 if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) { 1213 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n"); 1214 return -EIO; 1215 } 1216 1217 leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx); 1218 1219 /* determine the word within the dmap that holds the hint 1220 * (i.e. blkno). also, determine the last word in the dmap 1221 * that we'll include in our examination. 1222 */ 1223 word = (blkno & (BPERDMAP - 1)) >> L2DBWORD; 1224 lword = min(word + 4, LPERDMAP); 1225 1226 /* examine the leaves for sufficient free space. 1227 */ 1228 for (; word < lword; word++) { 1229 /* does the leaf describe sufficient free space ? 1230 */ 1231 if (leaf[word] < l2nb) 1232 continue; 1233 1234 /* determine the block number within the file system 1235 * of the first block described by this dmap word. 1236 */ 1237 blkno = le64_to_cpu(dp->start) + (word << L2DBWORD); 1238 1239 /* if not all bits of the dmap word are free, get the 1240 * starting bit number within the dmap word of the required 1241 * string of free bits and adjust the block number with the 1242 * value. 1243 */ 1244 if (leaf[word] < BUDMIN) 1245 blkno += 1246 dbFindBits(le32_to_cpu(dp->wmap[word]), l2nb); 1247 1248 /* allocate the blocks. 1249 */ 1250 if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0) 1251 *results = blkno; 1252 1253 return (rc); 1254 } 1255 1256 return -ENOSPC; 1257 } 1258 1259 1260 /* 1261 * NAME: dbAllocAG() 1262 * 1263 * FUNCTION: attempt to allocate the specified number of contiguous 1264 * free blocks within the specified allocation group. 1265 * 1266 * unless the allocation group size is equal to the number 1267 * of blocks per dmap, the dmap control pages will be used to 1268 * find the required free space, if available. we start the 1269 * search at the highest dmap control page level which 1270 * distinctly describes the allocation group's free space 1271 * (i.e. the highest level at which the allocation group's 1272 * free space is not mixed in with that of any other group). 1273 * in addition, we start the search within this level at a 1274 * height of the dmapctl dmtree at which the nodes distinctly 1275 * describe the allocation group's free space. at this height, 1276 * the allocation group's free space may be represented by 1 1277 * or two sub-trees, depending on the allocation group size. 1278 * we search the top nodes of these subtrees left to right for 1279 * sufficient free space. if sufficient free space is found, 1280 * the subtree is searched to find the leftmost leaf that 1281 * has free space. once we have made it to the leaf, we 1282 * move the search to the next lower level dmap control page 1283 * corresponding to this leaf. we continue down the dmap control 1284 * pages until we find the dmap that contains or starts the 1285 * sufficient free space and we allocate at this dmap. 1286 * 1287 * if the allocation group size is equal to the dmap size, 1288 * we'll start at the dmap corresponding to the allocation 1289 * group and attempt the allocation at this level. 1290 * 1291 * the dmap control page search is also not performed if the 1292 * allocation group is completely free and we go to the first 1293 * dmap of the allocation group to do the allocation. this is 1294 * done because the allocation group may be part (not the first 1295 * part) of a larger binary buddy system, causing the dmap 1296 * control pages to indicate no free space (NOFREE) within 1297 * the allocation group. 1298 * 1299 * PARAMETERS: 1300 * bmp - pointer to bmap descriptor 1301 * agno - allocation group number. 1302 * nblocks - actual number of contiguous free blocks desired. 1303 * l2nb - log2 number of contiguous free blocks desired. 1304 * results - on successful return, set to the starting block number 1305 * of the newly allocated range. 1306 * 1307 * RETURN VALUES: 1308 * 0 - success 1309 * -ENOSPC - insufficient disk resources 1310 * -EIO - i/o error 1311 * 1312 * note: IWRITE_LOCK(ipmap) held on entry/exit; 1313 */ 1314 static int 1315 dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results) 1316 { 1317 struct metapage *mp; 1318 struct dmapctl *dcp; 1319 int rc, ti, i, k, m, n, agperlev; 1320 s64 blkno, lblkno; 1321 int budmin; 1322 1323 /* allocation request should not be for more than the 1324 * allocation group size. 1325 */ 1326 if (l2nb > bmp->db_agl2size) { 1327 jfs_error(bmp->db_ipbmap->i_sb, 1328 "allocation request is larger than the allocation group size\n"); 1329 return -EIO; 1330 } 1331 1332 /* determine the starting block number of the allocation 1333 * group. 1334 */ 1335 blkno = (s64) agno << bmp->db_agl2size; 1336 1337 /* check if the allocation group size is the minimum allocation 1338 * group size or if the allocation group is completely free. if 1339 * the allocation group size is the minimum size of BPERDMAP (i.e. 1340 * 1 dmap), there is no need to search the dmap control page (below) 1341 * that fully describes the allocation group since the allocation 1342 * group is already fully described by a dmap. in this case, we 1343 * just call dbAllocCtl() to search the dmap tree and allocate the 1344 * required space if available. 1345 * 1346 * if the allocation group is completely free, dbAllocCtl() is 1347 * also called to allocate the required space. this is done for 1348 * two reasons. first, it makes no sense searching the dmap control 1349 * pages for free space when we know that free space exists. second, 1350 * the dmap control pages may indicate that the allocation group 1351 * has no free space if the allocation group is part (not the first 1352 * part) of a larger binary buddy system. 1353 */ 1354 if (bmp->db_agsize == BPERDMAP 1355 || bmp->db_agfree[agno] == bmp->db_agsize) { 1356 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results); 1357 if ((rc == -ENOSPC) && 1358 (bmp->db_agfree[agno] == bmp->db_agsize)) { 1359 printk(KERN_ERR "blkno = %Lx, blocks = %Lx\n", 1360 (unsigned long long) blkno, 1361 (unsigned long long) nblocks); 1362 jfs_error(bmp->db_ipbmap->i_sb, 1363 "dbAllocCtl failed in free AG\n"); 1364 } 1365 return (rc); 1366 } 1367 1368 /* the buffer for the dmap control page that fully describes the 1369 * allocation group. 1370 */ 1371 lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, bmp->db_aglevel); 1372 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1373 if (mp == NULL) 1374 return -EIO; 1375 dcp = (struct dmapctl *) mp->data; 1376 budmin = dcp->budmin; 1377 1378 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) { 1379 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n"); 1380 release_metapage(mp); 1381 return -EIO; 1382 } 1383 1384 /* search the subtree(s) of the dmap control page that describes 1385 * the allocation group, looking for sufficient free space. to begin, 1386 * determine how many allocation groups are represented in a dmap 1387 * control page at the control page level (i.e. L0, L1, L2) that 1388 * fully describes an allocation group. next, determine the starting 1389 * tree index of this allocation group within the control page. 1390 */ 1391 agperlev = 1392 (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / bmp->db_agwidth; 1393 ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1)); 1394 1395 /* dmap control page trees fan-out by 4 and a single allocation 1396 * group may be described by 1 or 2 subtrees within the ag level 1397 * dmap control page, depending upon the ag size. examine the ag's 1398 * subtrees for sufficient free space, starting with the leftmost 1399 * subtree. 1400 */ 1401 for (i = 0; i < bmp->db_agwidth; i++, ti++) { 1402 /* is there sufficient free space ? 1403 */ 1404 if (l2nb > dcp->stree[ti]) 1405 continue; 1406 1407 /* sufficient free space found in a subtree. now search down 1408 * the subtree to find the leftmost leaf that describes this 1409 * free space. 1410 */ 1411 for (k = bmp->db_agheight; k > 0; k--) { 1412 for (n = 0, m = (ti << 2) + 1; n < 4; n++) { 1413 if (l2nb <= dcp->stree[m + n]) { 1414 ti = m + n; 1415 break; 1416 } 1417 } 1418 if (n == 4) { 1419 jfs_error(bmp->db_ipbmap->i_sb, 1420 "failed descending stree\n"); 1421 release_metapage(mp); 1422 return -EIO; 1423 } 1424 } 1425 1426 /* determine the block number within the file system 1427 * that corresponds to this leaf. 1428 */ 1429 if (bmp->db_aglevel == 2) 1430 blkno = 0; 1431 else if (bmp->db_aglevel == 1) 1432 blkno &= ~(MAXL1SIZE - 1); 1433 else /* bmp->db_aglevel == 0 */ 1434 blkno &= ~(MAXL0SIZE - 1); 1435 1436 blkno += 1437 ((s64) (ti - le32_to_cpu(dcp->leafidx))) << budmin; 1438 1439 /* release the buffer in preparation for going down 1440 * the next level of dmap control pages. 1441 */ 1442 release_metapage(mp); 1443 1444 /* check if we need to continue to search down the lower 1445 * level dmap control pages. we need to if the number of 1446 * blocks required is less than maximum number of blocks 1447 * described at the next lower level. 1448 */ 1449 if (l2nb < budmin) { 1450 1451 /* search the lower level dmap control pages to get 1452 * the starting block number of the dmap that 1453 * contains or starts off the free space. 1454 */ 1455 if ((rc = 1456 dbFindCtl(bmp, l2nb, bmp->db_aglevel - 1, 1457 &blkno))) { 1458 if (rc == -ENOSPC) { 1459 jfs_error(bmp->db_ipbmap->i_sb, 1460 "control page inconsistent\n"); 1461 return -EIO; 1462 } 1463 return (rc); 1464 } 1465 } 1466 1467 /* allocate the blocks. 1468 */ 1469 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results); 1470 if (rc == -ENOSPC) { 1471 jfs_error(bmp->db_ipbmap->i_sb, 1472 "unable to allocate blocks\n"); 1473 rc = -EIO; 1474 } 1475 return (rc); 1476 } 1477 1478 /* no space in the allocation group. release the buffer and 1479 * return -ENOSPC. 1480 */ 1481 release_metapage(mp); 1482 1483 return -ENOSPC; 1484 } 1485 1486 1487 /* 1488 * NAME: dbAllocAny() 1489 * 1490 * FUNCTION: attempt to allocate the specified number of contiguous 1491 * free blocks anywhere in the file system. 1492 * 1493 * dbAllocAny() attempts to find the sufficient free space by 1494 * searching down the dmap control pages, starting with the 1495 * highest level (i.e. L0, L1, L2) control page. if free space 1496 * large enough to satisfy the desired free space is found, the 1497 * desired free space is allocated. 1498 * 1499 * PARAMETERS: 1500 * bmp - pointer to bmap descriptor 1501 * nblocks - actual number of contiguous free blocks desired. 1502 * l2nb - log2 number of contiguous free blocks desired. 1503 * results - on successful return, set to the starting block number 1504 * of the newly allocated range. 1505 * 1506 * RETURN VALUES: 1507 * 0 - success 1508 * -ENOSPC - insufficient disk resources 1509 * -EIO - i/o error 1510 * 1511 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit; 1512 */ 1513 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results) 1514 { 1515 int rc; 1516 s64 blkno = 0; 1517 1518 /* starting with the top level dmap control page, search 1519 * down the dmap control levels for sufficient free space. 1520 * if free space is found, dbFindCtl() returns the starting 1521 * block number of the dmap that contains or starts off the 1522 * range of free space. 1523 */ 1524 if ((rc = dbFindCtl(bmp, l2nb, bmp->db_maxlevel, &blkno))) 1525 return (rc); 1526 1527 /* allocate the blocks. 1528 */ 1529 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results); 1530 if (rc == -ENOSPC) { 1531 jfs_error(bmp->db_ipbmap->i_sb, "unable to allocate blocks\n"); 1532 return -EIO; 1533 } 1534 return (rc); 1535 } 1536 1537 1538 /* 1539 * NAME: dbDiscardAG() 1540 * 1541 * FUNCTION: attempt to discard (TRIM) all free blocks of specific AG 1542 * 1543 * algorithm: 1544 * 1) allocate blocks, as large as possible and save them 1545 * while holding IWRITE_LOCK on ipbmap 1546 * 2) trim all these saved block/length values 1547 * 3) mark the blocks free again 1548 * 1549 * benefit: 1550 * - we work only on one ag at some time, minimizing how long we 1551 * need to lock ipbmap 1552 * - reading / writing the fs is possible most time, even on 1553 * trimming 1554 * 1555 * downside: 1556 * - we write two times to the dmapctl and dmap pages 1557 * - but for me, this seems the best way, better ideas? 1558 * /TR 2012 1559 * 1560 * PARAMETERS: 1561 * ip - pointer to in-core inode 1562 * agno - ag to trim 1563 * minlen - minimum value of contiguous blocks 1564 * 1565 * RETURN VALUES: 1566 * s64 - actual number of blocks trimmed 1567 */ 1568 s64 dbDiscardAG(struct inode *ip, int agno, s64 minlen) 1569 { 1570 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 1571 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap; 1572 s64 nblocks, blkno; 1573 u64 trimmed = 0; 1574 int rc, l2nb; 1575 struct super_block *sb = ipbmap->i_sb; 1576 1577 struct range2trim { 1578 u64 blkno; 1579 u64 nblocks; 1580 } *totrim, *tt; 1581 1582 /* max blkno / nblocks pairs to trim */ 1583 int count = 0, range_cnt; 1584 u64 max_ranges; 1585 1586 /* prevent others from writing new stuff here, while trimming */ 1587 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP); 1588 1589 nblocks = bmp->db_agfree[agno]; 1590 max_ranges = nblocks; 1591 do_div(max_ranges, minlen); 1592 range_cnt = min_t(u64, max_ranges + 1, 32 * 1024); 1593 totrim = kmalloc_array(range_cnt, sizeof(struct range2trim), GFP_NOFS); 1594 if (totrim == NULL) { 1595 jfs_error(bmp->db_ipbmap->i_sb, "no memory for trim array\n"); 1596 IWRITE_UNLOCK(ipbmap); 1597 return 0; 1598 } 1599 1600 tt = totrim; 1601 while (nblocks >= minlen) { 1602 l2nb = BLKSTOL2(nblocks); 1603 1604 /* 0 = okay, -EIO = fatal, -ENOSPC -> try smaller block */ 1605 rc = dbAllocAG(bmp, agno, nblocks, l2nb, &blkno); 1606 if (rc == 0) { 1607 tt->blkno = blkno; 1608 tt->nblocks = nblocks; 1609 tt++; count++; 1610 1611 /* the whole ag is free, trim now */ 1612 if (bmp->db_agfree[agno] == 0) 1613 break; 1614 1615 /* give a hint for the next while */ 1616 nblocks = bmp->db_agfree[agno]; 1617 continue; 1618 } else if (rc == -ENOSPC) { 1619 /* search for next smaller log2 block */ 1620 l2nb = BLKSTOL2(nblocks) - 1; 1621 nblocks = 1LL << l2nb; 1622 } else { 1623 /* Trim any already allocated blocks */ 1624 jfs_error(bmp->db_ipbmap->i_sb, "-EIO\n"); 1625 break; 1626 } 1627 1628 /* check, if our trim array is full */ 1629 if (unlikely(count >= range_cnt - 1)) 1630 break; 1631 } 1632 IWRITE_UNLOCK(ipbmap); 1633 1634 tt->nblocks = 0; /* mark the current end */ 1635 for (tt = totrim; tt->nblocks != 0; tt++) { 1636 /* when mounted with online discard, dbFree() will 1637 * call jfs_issue_discard() itself */ 1638 if (!(JFS_SBI(sb)->flag & JFS_DISCARD)) 1639 jfs_issue_discard(ip, tt->blkno, tt->nblocks); 1640 dbFree(ip, tt->blkno, tt->nblocks); 1641 trimmed += tt->nblocks; 1642 } 1643 kfree(totrim); 1644 1645 return trimmed; 1646 } 1647 1648 /* 1649 * NAME: dbFindCtl() 1650 * 1651 * FUNCTION: starting at a specified dmap control page level and block 1652 * number, search down the dmap control levels for a range of 1653 * contiguous free blocks large enough to satisfy an allocation 1654 * request for the specified number of free blocks. 1655 * 1656 * if sufficient contiguous free blocks are found, this routine 1657 * returns the starting block number within a dmap page that 1658 * contains or starts a range of contiqious free blocks that 1659 * is sufficient in size. 1660 * 1661 * PARAMETERS: 1662 * bmp - pointer to bmap descriptor 1663 * level - starting dmap control page level. 1664 * l2nb - log2 number of contiguous free blocks desired. 1665 * *blkno - on entry, starting block number for conducting the search. 1666 * on successful return, the first block within a dmap page 1667 * that contains or starts a range of contiguous free blocks. 1668 * 1669 * RETURN VALUES: 1670 * 0 - success 1671 * -ENOSPC - insufficient disk resources 1672 * -EIO - i/o error 1673 * 1674 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit; 1675 */ 1676 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno) 1677 { 1678 int rc, leafidx, lev; 1679 s64 b, lblkno; 1680 struct dmapctl *dcp; 1681 int budmin; 1682 struct metapage *mp; 1683 1684 /* starting at the specified dmap control page level and block 1685 * number, search down the dmap control levels for the starting 1686 * block number of a dmap page that contains or starts off 1687 * sufficient free blocks. 1688 */ 1689 for (lev = level, b = *blkno; lev >= 0; lev--) { 1690 /* get the buffer of the dmap control page for the block 1691 * number and level (i.e. L0, L1, L2). 1692 */ 1693 lblkno = BLKTOCTL(b, bmp->db_l2nbperpage, lev); 1694 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1695 if (mp == NULL) 1696 return -EIO; 1697 dcp = (struct dmapctl *) mp->data; 1698 budmin = dcp->budmin; 1699 1700 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) { 1701 jfs_error(bmp->db_ipbmap->i_sb, 1702 "Corrupt dmapctl page\n"); 1703 release_metapage(mp); 1704 return -EIO; 1705 } 1706 1707 /* search the tree within the dmap control page for 1708 * sufficient free space. if sufficient free space is found, 1709 * dbFindLeaf() returns the index of the leaf at which 1710 * free space was found. 1711 */ 1712 rc = dbFindLeaf((dmtree_t *) dcp, l2nb, &leafidx); 1713 1714 /* release the buffer. 1715 */ 1716 release_metapage(mp); 1717 1718 /* space found ? 1719 */ 1720 if (rc) { 1721 if (lev != level) { 1722 jfs_error(bmp->db_ipbmap->i_sb, 1723 "dmap inconsistent\n"); 1724 return -EIO; 1725 } 1726 return -ENOSPC; 1727 } 1728 1729 /* adjust the block number to reflect the location within 1730 * the dmap control page (i.e. the leaf) at which free 1731 * space was found. 1732 */ 1733 b += (((s64) leafidx) << budmin); 1734 1735 /* we stop the search at this dmap control page level if 1736 * the number of blocks required is greater than or equal 1737 * to the maximum number of blocks described at the next 1738 * (lower) level. 1739 */ 1740 if (l2nb >= budmin) 1741 break; 1742 } 1743 1744 *blkno = b; 1745 return (0); 1746 } 1747 1748 1749 /* 1750 * NAME: dbAllocCtl() 1751 * 1752 * FUNCTION: attempt to allocate a specified number of contiguous 1753 * blocks starting within a specific dmap. 1754 * 1755 * this routine is called by higher level routines that search 1756 * the dmap control pages above the actual dmaps for contiguous 1757 * free space. the result of successful searches by these 1758 * routines are the starting block numbers within dmaps, with 1759 * the dmaps themselves containing the desired contiguous free 1760 * space or starting a contiguous free space of desired size 1761 * that is made up of the blocks of one or more dmaps. these 1762 * calls should not fail due to insufficent resources. 1763 * 1764 * this routine is called in some cases where it is not known 1765 * whether it will fail due to insufficient resources. more 1766 * specifically, this occurs when allocating from an allocation 1767 * group whose size is equal to the number of blocks per dmap. 1768 * in this case, the dmap control pages are not examined prior 1769 * to calling this routine (to save pathlength) and the call 1770 * might fail. 1771 * 1772 * for a request size that fits within a dmap, this routine relies 1773 * upon the dmap's dmtree to find the requested contiguous free 1774 * space. for request sizes that are larger than a dmap, the 1775 * requested free space will start at the first block of the 1776 * first dmap (i.e. blkno). 1777 * 1778 * PARAMETERS: 1779 * bmp - pointer to bmap descriptor 1780 * nblocks - actual number of contiguous free blocks to allocate. 1781 * l2nb - log2 number of contiguous free blocks to allocate. 1782 * blkno - starting block number of the dmap to start the allocation 1783 * from. 1784 * results - on successful return, set to the starting block number 1785 * of the newly allocated range. 1786 * 1787 * RETURN VALUES: 1788 * 0 - success 1789 * -ENOSPC - insufficient disk resources 1790 * -EIO - i/o error 1791 * 1792 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit; 1793 */ 1794 static int 1795 dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno, s64 * results) 1796 { 1797 int rc, nb; 1798 s64 b, lblkno, n; 1799 struct metapage *mp; 1800 struct dmap *dp; 1801 1802 /* check if the allocation request is confined to a single dmap. 1803 */ 1804 if (l2nb <= L2BPERDMAP) { 1805 /* get the buffer for the dmap. 1806 */ 1807 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 1808 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1809 if (mp == NULL) 1810 return -EIO; 1811 dp = (struct dmap *) mp->data; 1812 1813 /* try to allocate the blocks. 1814 */ 1815 rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results); 1816 if (rc == 0) 1817 mark_metapage_dirty(mp); 1818 1819 release_metapage(mp); 1820 1821 return (rc); 1822 } 1823 1824 /* allocation request involving multiple dmaps. it must start on 1825 * a dmap boundary. 1826 */ 1827 assert((blkno & (BPERDMAP - 1)) == 0); 1828 1829 /* allocate the blocks dmap by dmap. 1830 */ 1831 for (n = nblocks, b = blkno; n > 0; n -= nb, b += nb) { 1832 /* get the buffer for the dmap. 1833 */ 1834 lblkno = BLKTODMAP(b, bmp->db_l2nbperpage); 1835 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1836 if (mp == NULL) { 1837 rc = -EIO; 1838 goto backout; 1839 } 1840 dp = (struct dmap *) mp->data; 1841 1842 /* the dmap better be all free. 1843 */ 1844 if (dp->tree.stree[ROOT] != L2BPERDMAP) { 1845 release_metapage(mp); 1846 jfs_error(bmp->db_ipbmap->i_sb, 1847 "the dmap is not all free\n"); 1848 rc = -EIO; 1849 goto backout; 1850 } 1851 1852 /* determine how many blocks to allocate from this dmap. 1853 */ 1854 nb = min_t(s64, n, BPERDMAP); 1855 1856 /* allocate the blocks from the dmap. 1857 */ 1858 if ((rc = dbAllocDmap(bmp, dp, b, nb))) { 1859 release_metapage(mp); 1860 goto backout; 1861 } 1862 1863 /* write the buffer. 1864 */ 1865 write_metapage(mp); 1866 } 1867 1868 /* set the results (starting block number) and return. 1869 */ 1870 *results = blkno; 1871 return (0); 1872 1873 /* something failed in handling an allocation request involving 1874 * multiple dmaps. we'll try to clean up by backing out any 1875 * allocation that has already happened for this request. if 1876 * we fail in backing out the allocation, we'll mark the file 1877 * system to indicate that blocks have been leaked. 1878 */ 1879 backout: 1880 1881 /* try to backout the allocations dmap by dmap. 1882 */ 1883 for (n = nblocks - n, b = blkno; n > 0; 1884 n -= BPERDMAP, b += BPERDMAP) { 1885 /* get the buffer for this dmap. 1886 */ 1887 lblkno = BLKTODMAP(b, bmp->db_l2nbperpage); 1888 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1889 if (mp == NULL) { 1890 /* could not back out. mark the file system 1891 * to indicate that we have leaked blocks. 1892 */ 1893 jfs_error(bmp->db_ipbmap->i_sb, 1894 "I/O Error: Block Leakage\n"); 1895 continue; 1896 } 1897 dp = (struct dmap *) mp->data; 1898 1899 /* free the blocks is this dmap. 1900 */ 1901 if (dbFreeDmap(bmp, dp, b, BPERDMAP)) { 1902 /* could not back out. mark the file system 1903 * to indicate that we have leaked blocks. 1904 */ 1905 release_metapage(mp); 1906 jfs_error(bmp->db_ipbmap->i_sb, "Block Leakage\n"); 1907 continue; 1908 } 1909 1910 /* write the buffer. 1911 */ 1912 write_metapage(mp); 1913 } 1914 1915 return (rc); 1916 } 1917 1918 1919 /* 1920 * NAME: dbAllocDmapLev() 1921 * 1922 * FUNCTION: attempt to allocate a specified number of contiguous blocks 1923 * from a specified dmap. 1924 * 1925 * this routine checks if the contiguous blocks are available. 1926 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is 1927 * returned. 1928 * 1929 * PARAMETERS: 1930 * mp - pointer to bmap descriptor 1931 * dp - pointer to dmap to attempt to allocate blocks from. 1932 * l2nb - log2 number of contiguous block desired. 1933 * nblocks - actual number of contiguous block desired. 1934 * results - on successful return, set to the starting block number 1935 * of the newly allocated range. 1936 * 1937 * RETURN VALUES: 1938 * 0 - success 1939 * -ENOSPC - insufficient disk resources 1940 * -EIO - i/o error 1941 * 1942 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or 1943 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit; 1944 */ 1945 static int 1946 dbAllocDmapLev(struct bmap * bmp, 1947 struct dmap * dp, int nblocks, int l2nb, s64 * results) 1948 { 1949 s64 blkno; 1950 int leafidx, rc; 1951 1952 /* can't be more than a dmaps worth of blocks */ 1953 assert(l2nb <= L2BPERDMAP); 1954 1955 /* search the tree within the dmap page for sufficient 1956 * free space. if sufficient free space is found, dbFindLeaf() 1957 * returns the index of the leaf at which free space was found. 1958 */ 1959 if (dbFindLeaf((dmtree_t *) & dp->tree, l2nb, &leafidx)) 1960 return -ENOSPC; 1961 1962 if (leafidx < 0) 1963 return -EIO; 1964 1965 /* determine the block number within the file system corresponding 1966 * to the leaf at which free space was found. 1967 */ 1968 blkno = le64_to_cpu(dp->start) + (leafidx << L2DBWORD); 1969 1970 /* if not all bits of the dmap word are free, get the starting 1971 * bit number within the dmap word of the required string of free 1972 * bits and adjust the block number with this value. 1973 */ 1974 if (dp->tree.stree[leafidx + LEAFIND] < BUDMIN) 1975 blkno += dbFindBits(le32_to_cpu(dp->wmap[leafidx]), l2nb); 1976 1977 /* allocate the blocks */ 1978 if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0) 1979 *results = blkno; 1980 1981 return (rc); 1982 } 1983 1984 1985 /* 1986 * NAME: dbAllocDmap() 1987 * 1988 * FUNCTION: adjust the disk allocation map to reflect the allocation 1989 * of a specified block range within a dmap. 1990 * 1991 * this routine allocates the specified blocks from the dmap 1992 * through a call to dbAllocBits(). if the allocation of the 1993 * block range causes the maximum string of free blocks within 1994 * the dmap to change (i.e. the value of the root of the dmap's 1995 * dmtree), this routine will cause this change to be reflected 1996 * up through the appropriate levels of the dmap control pages 1997 * by a call to dbAdjCtl() for the L0 dmap control page that 1998 * covers this dmap. 1999 * 2000 * PARAMETERS: 2001 * bmp - pointer to bmap descriptor 2002 * dp - pointer to dmap to allocate the block range from. 2003 * blkno - starting block number of the block to be allocated. 2004 * nblocks - number of blocks to be allocated. 2005 * 2006 * RETURN VALUES: 2007 * 0 - success 2008 * -EIO - i/o error 2009 * 2010 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2011 */ 2012 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno, 2013 int nblocks) 2014 { 2015 s8 oldroot; 2016 int rc; 2017 2018 /* save the current value of the root (i.e. maximum free string) 2019 * of the dmap tree. 2020 */ 2021 oldroot = dp->tree.stree[ROOT]; 2022 2023 /* allocate the specified (blocks) bits */ 2024 dbAllocBits(bmp, dp, blkno, nblocks); 2025 2026 /* if the root has not changed, done. */ 2027 if (dp->tree.stree[ROOT] == oldroot) 2028 return (0); 2029 2030 /* root changed. bubble the change up to the dmap control pages. 2031 * if the adjustment of the upper level control pages fails, 2032 * backout the bit allocation (thus making everything consistent). 2033 */ 2034 if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 1, 0))) 2035 dbFreeBits(bmp, dp, blkno, nblocks); 2036 2037 return (rc); 2038 } 2039 2040 2041 /* 2042 * NAME: dbFreeDmap() 2043 * 2044 * FUNCTION: adjust the disk allocation map to reflect the allocation 2045 * of a specified block range within a dmap. 2046 * 2047 * this routine frees the specified blocks from the dmap through 2048 * a call to dbFreeBits(). if the deallocation of the block range 2049 * causes the maximum string of free blocks within the dmap to 2050 * change (i.e. the value of the root of the dmap's dmtree), this 2051 * routine will cause this change to be reflected up through the 2052 * appropriate levels of the dmap control pages by a call to 2053 * dbAdjCtl() for the L0 dmap control page that covers this dmap. 2054 * 2055 * PARAMETERS: 2056 * bmp - pointer to bmap descriptor 2057 * dp - pointer to dmap to free the block range from. 2058 * blkno - starting block number of the block to be freed. 2059 * nblocks - number of blocks to be freed. 2060 * 2061 * RETURN VALUES: 2062 * 0 - success 2063 * -EIO - i/o error 2064 * 2065 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2066 */ 2067 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno, 2068 int nblocks) 2069 { 2070 s8 oldroot; 2071 int rc = 0, word; 2072 2073 /* save the current value of the root (i.e. maximum free string) 2074 * of the dmap tree. 2075 */ 2076 oldroot = dp->tree.stree[ROOT]; 2077 2078 /* free the specified (blocks) bits */ 2079 rc = dbFreeBits(bmp, dp, blkno, nblocks); 2080 2081 /* if error or the root has not changed, done. */ 2082 if (rc || (dp->tree.stree[ROOT] == oldroot)) 2083 return (rc); 2084 2085 /* root changed. bubble the change up to the dmap control pages. 2086 * if the adjustment of the upper level control pages fails, 2087 * backout the deallocation. 2088 */ 2089 if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 0, 0))) { 2090 word = (blkno & (BPERDMAP - 1)) >> L2DBWORD; 2091 2092 /* as part of backing out the deallocation, we will have 2093 * to back split the dmap tree if the deallocation caused 2094 * the freed blocks to become part of a larger binary buddy 2095 * system. 2096 */ 2097 if (dp->tree.stree[word] == NOFREE) 2098 dbBackSplit((dmtree_t *) & dp->tree, word); 2099 2100 dbAllocBits(bmp, dp, blkno, nblocks); 2101 } 2102 2103 return (rc); 2104 } 2105 2106 2107 /* 2108 * NAME: dbAllocBits() 2109 * 2110 * FUNCTION: allocate a specified block range from a dmap. 2111 * 2112 * this routine updates the dmap to reflect the working 2113 * state allocation of the specified block range. it directly 2114 * updates the bits of the working map and causes the adjustment 2115 * of the binary buddy system described by the dmap's dmtree 2116 * leaves to reflect the bits allocated. it also causes the 2117 * dmap's dmtree, as a whole, to reflect the allocated range. 2118 * 2119 * PARAMETERS: 2120 * bmp - pointer to bmap descriptor 2121 * dp - pointer to dmap to allocate bits from. 2122 * blkno - starting block number of the bits to be allocated. 2123 * nblocks - number of bits to be allocated. 2124 * 2125 * RETURN VALUES: none 2126 * 2127 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2128 */ 2129 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno, 2130 int nblocks) 2131 { 2132 int dbitno, word, rembits, nb, nwords, wbitno, nw, agno; 2133 dmtree_t *tp = (dmtree_t *) & dp->tree; 2134 int size; 2135 s8 *leaf; 2136 2137 /* pick up a pointer to the leaves of the dmap tree */ 2138 leaf = dp->tree.stree + LEAFIND; 2139 2140 /* determine the bit number and word within the dmap of the 2141 * starting block. 2142 */ 2143 dbitno = blkno & (BPERDMAP - 1); 2144 word = dbitno >> L2DBWORD; 2145 2146 /* block range better be within the dmap */ 2147 assert(dbitno + nblocks <= BPERDMAP); 2148 2149 /* allocate the bits of the dmap's words corresponding to the block 2150 * range. not all bits of the first and last words may be contained 2151 * within the block range. if this is the case, we'll work against 2152 * those words (i.e. partial first and/or last) on an individual basis 2153 * (a single pass), allocating the bits of interest by hand and 2154 * updating the leaf corresponding to the dmap word. a single pass 2155 * will be used for all dmap words fully contained within the 2156 * specified range. within this pass, the bits of all fully contained 2157 * dmap words will be marked as free in a single shot and the leaves 2158 * will be updated. a single leaf may describe the free space of 2159 * multiple dmap words, so we may update only a subset of the actual 2160 * leaves corresponding to the dmap words of the block range. 2161 */ 2162 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) { 2163 /* determine the bit number within the word and 2164 * the number of bits within the word. 2165 */ 2166 wbitno = dbitno & (DBWORD - 1); 2167 nb = min(rembits, DBWORD - wbitno); 2168 2169 /* check if only part of a word is to be allocated. 2170 */ 2171 if (nb < DBWORD) { 2172 /* allocate (set to 1) the appropriate bits within 2173 * this dmap word. 2174 */ 2175 dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb) 2176 >> wbitno); 2177 2178 /* update the leaf for this dmap word. in addition 2179 * to setting the leaf value to the binary buddy max 2180 * of the updated dmap word, dbSplit() will split 2181 * the binary system of the leaves if need be. 2182 */ 2183 dbSplit(tp, word, BUDMIN, 2184 dbMaxBud((u8 *) & dp->wmap[word])); 2185 2186 word += 1; 2187 } else { 2188 /* one or more dmap words are fully contained 2189 * within the block range. determine how many 2190 * words and allocate (set to 1) the bits of these 2191 * words. 2192 */ 2193 nwords = rembits >> L2DBWORD; 2194 memset(&dp->wmap[word], (int) ONES, nwords * 4); 2195 2196 /* determine how many bits. 2197 */ 2198 nb = nwords << L2DBWORD; 2199 2200 /* now update the appropriate leaves to reflect 2201 * the allocated words. 2202 */ 2203 for (; nwords > 0; nwords -= nw) { 2204 if (leaf[word] < BUDMIN) { 2205 jfs_error(bmp->db_ipbmap->i_sb, 2206 "leaf page corrupt\n"); 2207 break; 2208 } 2209 2210 /* determine what the leaf value should be 2211 * updated to as the minimum of the l2 number 2212 * of bits being allocated and the l2 number 2213 * of bits currently described by this leaf. 2214 */ 2215 size = min_t(int, leaf[word], 2216 NLSTOL2BSZ(nwords)); 2217 2218 /* update the leaf to reflect the allocation. 2219 * in addition to setting the leaf value to 2220 * NOFREE, dbSplit() will split the binary 2221 * system of the leaves to reflect the current 2222 * allocation (size). 2223 */ 2224 dbSplit(tp, word, size, NOFREE); 2225 2226 /* get the number of dmap words handled */ 2227 nw = BUDSIZE(size, BUDMIN); 2228 word += nw; 2229 } 2230 } 2231 } 2232 2233 /* update the free count for this dmap */ 2234 le32_add_cpu(&dp->nfree, -nblocks); 2235 2236 BMAP_LOCK(bmp); 2237 2238 /* if this allocation group is completely free, 2239 * update the maximum allocation group number if this allocation 2240 * group is the new max. 2241 */ 2242 agno = blkno >> bmp->db_agl2size; 2243 if (agno > bmp->db_maxag) 2244 bmp->db_maxag = agno; 2245 2246 /* update the free count for the allocation group and map */ 2247 bmp->db_agfree[agno] -= nblocks; 2248 bmp->db_nfree -= nblocks; 2249 2250 BMAP_UNLOCK(bmp); 2251 } 2252 2253 2254 /* 2255 * NAME: dbFreeBits() 2256 * 2257 * FUNCTION: free a specified block range from a dmap. 2258 * 2259 * this routine updates the dmap to reflect the working 2260 * state allocation of the specified block range. it directly 2261 * updates the bits of the working map and causes the adjustment 2262 * of the binary buddy system described by the dmap's dmtree 2263 * leaves to reflect the bits freed. it also causes the dmap's 2264 * dmtree, as a whole, to reflect the deallocated range. 2265 * 2266 * PARAMETERS: 2267 * bmp - pointer to bmap descriptor 2268 * dp - pointer to dmap to free bits from. 2269 * blkno - starting block number of the bits to be freed. 2270 * nblocks - number of bits to be freed. 2271 * 2272 * RETURN VALUES: 0 for success 2273 * 2274 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2275 */ 2276 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno, 2277 int nblocks) 2278 { 2279 int dbitno, word, rembits, nb, nwords, wbitno, nw, agno; 2280 dmtree_t *tp = (dmtree_t *) & dp->tree; 2281 int rc = 0; 2282 int size; 2283 2284 /* determine the bit number and word within the dmap of the 2285 * starting block. 2286 */ 2287 dbitno = blkno & (BPERDMAP - 1); 2288 word = dbitno >> L2DBWORD; 2289 2290 /* block range better be within the dmap. 2291 */ 2292 assert(dbitno + nblocks <= BPERDMAP); 2293 2294 /* free the bits of the dmaps words corresponding to the block range. 2295 * not all bits of the first and last words may be contained within 2296 * the block range. if this is the case, we'll work against those 2297 * words (i.e. partial first and/or last) on an individual basis 2298 * (a single pass), freeing the bits of interest by hand and updating 2299 * the leaf corresponding to the dmap word. a single pass will be used 2300 * for all dmap words fully contained within the specified range. 2301 * within this pass, the bits of all fully contained dmap words will 2302 * be marked as free in a single shot and the leaves will be updated. a 2303 * single leaf may describe the free space of multiple dmap words, 2304 * so we may update only a subset of the actual leaves corresponding 2305 * to the dmap words of the block range. 2306 * 2307 * dbJoin() is used to update leaf values and will join the binary 2308 * buddy system of the leaves if the new leaf values indicate this 2309 * should be done. 2310 */ 2311 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) { 2312 /* determine the bit number within the word and 2313 * the number of bits within the word. 2314 */ 2315 wbitno = dbitno & (DBWORD - 1); 2316 nb = min(rembits, DBWORD - wbitno); 2317 2318 /* check if only part of a word is to be freed. 2319 */ 2320 if (nb < DBWORD) { 2321 /* free (zero) the appropriate bits within this 2322 * dmap word. 2323 */ 2324 dp->wmap[word] &= 2325 cpu_to_le32(~(ONES << (DBWORD - nb) 2326 >> wbitno)); 2327 2328 /* update the leaf for this dmap word. 2329 */ 2330 rc = dbJoin(tp, word, 2331 dbMaxBud((u8 *) & dp->wmap[word])); 2332 if (rc) 2333 return rc; 2334 2335 word += 1; 2336 } else { 2337 /* one or more dmap words are fully contained 2338 * within the block range. determine how many 2339 * words and free (zero) the bits of these words. 2340 */ 2341 nwords = rembits >> L2DBWORD; 2342 memset(&dp->wmap[word], 0, nwords * 4); 2343 2344 /* determine how many bits. 2345 */ 2346 nb = nwords << L2DBWORD; 2347 2348 /* now update the appropriate leaves to reflect 2349 * the freed words. 2350 */ 2351 for (; nwords > 0; nwords -= nw) { 2352 /* determine what the leaf value should be 2353 * updated to as the minimum of the l2 number 2354 * of bits being freed and the l2 (max) number 2355 * of bits that can be described by this leaf. 2356 */ 2357 size = 2358 min(LITOL2BSZ 2359 (word, L2LPERDMAP, BUDMIN), 2360 NLSTOL2BSZ(nwords)); 2361 2362 /* update the leaf. 2363 */ 2364 rc = dbJoin(tp, word, size); 2365 if (rc) 2366 return rc; 2367 2368 /* get the number of dmap words handled. 2369 */ 2370 nw = BUDSIZE(size, BUDMIN); 2371 word += nw; 2372 } 2373 } 2374 } 2375 2376 /* update the free count for this dmap. 2377 */ 2378 le32_add_cpu(&dp->nfree, nblocks); 2379 2380 BMAP_LOCK(bmp); 2381 2382 /* update the free count for the allocation group and 2383 * map. 2384 */ 2385 agno = blkno >> bmp->db_agl2size; 2386 bmp->db_nfree += nblocks; 2387 bmp->db_agfree[agno] += nblocks; 2388 2389 /* check if this allocation group is not completely free and 2390 * if it is currently the maximum (rightmost) allocation group. 2391 * if so, establish the new maximum allocation group number by 2392 * searching left for the first allocation group with allocation. 2393 */ 2394 if ((bmp->db_agfree[agno] == bmp->db_agsize && agno == bmp->db_maxag) || 2395 (agno == bmp->db_numag - 1 && 2396 bmp->db_agfree[agno] == (bmp-> db_mapsize & (BPERDMAP - 1)))) { 2397 while (bmp->db_maxag > 0) { 2398 bmp->db_maxag -= 1; 2399 if (bmp->db_agfree[bmp->db_maxag] != 2400 bmp->db_agsize) 2401 break; 2402 } 2403 2404 /* re-establish the allocation group preference if the 2405 * current preference is right of the maximum allocation 2406 * group. 2407 */ 2408 if (bmp->db_agpref > bmp->db_maxag) 2409 bmp->db_agpref = bmp->db_maxag; 2410 } 2411 2412 BMAP_UNLOCK(bmp); 2413 2414 return 0; 2415 } 2416 2417 2418 /* 2419 * NAME: dbAdjCtl() 2420 * 2421 * FUNCTION: adjust a dmap control page at a specified level to reflect 2422 * the change in a lower level dmap or dmap control page's 2423 * maximum string of free blocks (i.e. a change in the root 2424 * of the lower level object's dmtree) due to the allocation 2425 * or deallocation of a range of blocks with a single dmap. 2426 * 2427 * on entry, this routine is provided with the new value of 2428 * the lower level dmap or dmap control page root and the 2429 * starting block number of the block range whose allocation 2430 * or deallocation resulted in the root change. this range 2431 * is respresented by a single leaf of the current dmapctl 2432 * and the leaf will be updated with this value, possibly 2433 * causing a binary buddy system within the leaves to be 2434 * split or joined. the update may also cause the dmapctl's 2435 * dmtree to be updated. 2436 * 2437 * if the adjustment of the dmap control page, itself, causes its 2438 * root to change, this change will be bubbled up to the next dmap 2439 * control level by a recursive call to this routine, specifying 2440 * the new root value and the next dmap control page level to 2441 * be adjusted. 2442 * PARAMETERS: 2443 * bmp - pointer to bmap descriptor 2444 * blkno - the first block of a block range within a dmap. it is 2445 * the allocation or deallocation of this block range that 2446 * requires the dmap control page to be adjusted. 2447 * newval - the new value of the lower level dmap or dmap control 2448 * page root. 2449 * alloc - 'true' if adjustment is due to an allocation. 2450 * level - current level of dmap control page (i.e. L0, L1, L2) to 2451 * be adjusted. 2452 * 2453 * RETURN VALUES: 2454 * 0 - success 2455 * -EIO - i/o error 2456 * 2457 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2458 */ 2459 static int 2460 dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc, int level) 2461 { 2462 struct metapage *mp; 2463 s8 oldroot; 2464 int oldval; 2465 s64 lblkno; 2466 struct dmapctl *dcp; 2467 int rc, leafno, ti; 2468 2469 /* get the buffer for the dmap control page for the specified 2470 * block number and control page level. 2471 */ 2472 lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, level); 2473 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 2474 if (mp == NULL) 2475 return -EIO; 2476 dcp = (struct dmapctl *) mp->data; 2477 2478 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) { 2479 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n"); 2480 release_metapage(mp); 2481 return -EIO; 2482 } 2483 2484 /* determine the leaf number corresponding to the block and 2485 * the index within the dmap control tree. 2486 */ 2487 leafno = BLKTOCTLLEAF(blkno, dcp->budmin); 2488 ti = leafno + le32_to_cpu(dcp->leafidx); 2489 2490 /* save the current leaf value and the current root level (i.e. 2491 * maximum l2 free string described by this dmapctl). 2492 */ 2493 oldval = dcp->stree[ti]; 2494 oldroot = dcp->stree[ROOT]; 2495 2496 /* check if this is a control page update for an allocation. 2497 * if so, update the leaf to reflect the new leaf value using 2498 * dbSplit(); otherwise (deallocation), use dbJoin() to update 2499 * the leaf with the new value. in addition to updating the 2500 * leaf, dbSplit() will also split the binary buddy system of 2501 * the leaves, if required, and bubble new values within the 2502 * dmapctl tree, if required. similarly, dbJoin() will join 2503 * the binary buddy system of leaves and bubble new values up 2504 * the dmapctl tree as required by the new leaf value. 2505 */ 2506 if (alloc) { 2507 /* check if we are in the middle of a binary buddy 2508 * system. this happens when we are performing the 2509 * first allocation out of an allocation group that 2510 * is part (not the first part) of a larger binary 2511 * buddy system. if we are in the middle, back split 2512 * the system prior to calling dbSplit() which assumes 2513 * that it is at the front of a binary buddy system. 2514 */ 2515 if (oldval == NOFREE) { 2516 rc = dbBackSplit((dmtree_t *) dcp, leafno); 2517 if (rc) { 2518 release_metapage(mp); 2519 return rc; 2520 } 2521 oldval = dcp->stree[ti]; 2522 } 2523 dbSplit((dmtree_t *) dcp, leafno, dcp->budmin, newval); 2524 } else { 2525 rc = dbJoin((dmtree_t *) dcp, leafno, newval); 2526 if (rc) { 2527 release_metapage(mp); 2528 return rc; 2529 } 2530 } 2531 2532 /* check if the root of the current dmap control page changed due 2533 * to the update and if the current dmap control page is not at 2534 * the current top level (i.e. L0, L1, L2) of the map. if so (i.e. 2535 * root changed and this is not the top level), call this routine 2536 * again (recursion) for the next higher level of the mapping to 2537 * reflect the change in root for the current dmap control page. 2538 */ 2539 if (dcp->stree[ROOT] != oldroot) { 2540 /* are we below the top level of the map. if so, 2541 * bubble the root up to the next higher level. 2542 */ 2543 if (level < bmp->db_maxlevel) { 2544 /* bubble up the new root of this dmap control page to 2545 * the next level. 2546 */ 2547 if ((rc = 2548 dbAdjCtl(bmp, blkno, dcp->stree[ROOT], alloc, 2549 level + 1))) { 2550 /* something went wrong in bubbling up the new 2551 * root value, so backout the changes to the 2552 * current dmap control page. 2553 */ 2554 if (alloc) { 2555 dbJoin((dmtree_t *) dcp, leafno, 2556 oldval); 2557 } else { 2558 /* the dbJoin() above might have 2559 * caused a larger binary buddy system 2560 * to form and we may now be in the 2561 * middle of it. if this is the case, 2562 * back split the buddies. 2563 */ 2564 if (dcp->stree[ti] == NOFREE) 2565 dbBackSplit((dmtree_t *) 2566 dcp, leafno); 2567 dbSplit((dmtree_t *) dcp, leafno, 2568 dcp->budmin, oldval); 2569 } 2570 2571 /* release the buffer and return the error. 2572 */ 2573 release_metapage(mp); 2574 return (rc); 2575 } 2576 } else { 2577 /* we're at the top level of the map. update 2578 * the bmap control page to reflect the size 2579 * of the maximum free buddy system. 2580 */ 2581 assert(level == bmp->db_maxlevel); 2582 if (bmp->db_maxfreebud != oldroot) { 2583 jfs_error(bmp->db_ipbmap->i_sb, 2584 "the maximum free buddy is not the old root\n"); 2585 } 2586 bmp->db_maxfreebud = dcp->stree[ROOT]; 2587 } 2588 } 2589 2590 /* write the buffer. 2591 */ 2592 write_metapage(mp); 2593 2594 return (0); 2595 } 2596 2597 2598 /* 2599 * NAME: dbSplit() 2600 * 2601 * FUNCTION: update the leaf of a dmtree with a new value, splitting 2602 * the leaf from the binary buddy system of the dmtree's 2603 * leaves, as required. 2604 * 2605 * PARAMETERS: 2606 * tp - pointer to the tree containing the leaf. 2607 * leafno - the number of the leaf to be updated. 2608 * splitsz - the size the binary buddy system starting at the leaf 2609 * must be split to, specified as the log2 number of blocks. 2610 * newval - the new value for the leaf. 2611 * 2612 * RETURN VALUES: none 2613 * 2614 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2615 */ 2616 static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval) 2617 { 2618 int budsz; 2619 int cursz; 2620 s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx); 2621 2622 /* check if the leaf needs to be split. 2623 */ 2624 if (leaf[leafno] > tp->dmt_budmin) { 2625 /* the split occurs by cutting the buddy system in half 2626 * at the specified leaf until we reach the specified 2627 * size. pick up the starting split size (current size 2628 * - 1 in l2) and the corresponding buddy size. 2629 */ 2630 cursz = leaf[leafno] - 1; 2631 budsz = BUDSIZE(cursz, tp->dmt_budmin); 2632 2633 /* split until we reach the specified size. 2634 */ 2635 while (cursz >= splitsz) { 2636 /* update the buddy's leaf with its new value. 2637 */ 2638 dbAdjTree(tp, leafno ^ budsz, cursz); 2639 2640 /* on to the next size and buddy. 2641 */ 2642 cursz -= 1; 2643 budsz >>= 1; 2644 } 2645 } 2646 2647 /* adjust the dmap tree to reflect the specified leaf's new 2648 * value. 2649 */ 2650 dbAdjTree(tp, leafno, newval); 2651 } 2652 2653 2654 /* 2655 * NAME: dbBackSplit() 2656 * 2657 * FUNCTION: back split the binary buddy system of dmtree leaves 2658 * that hold a specified leaf until the specified leaf 2659 * starts its own binary buddy system. 2660 * 2661 * the allocators typically perform allocations at the start 2662 * of binary buddy systems and dbSplit() is used to accomplish 2663 * any required splits. in some cases, however, allocation 2664 * may occur in the middle of a binary system and requires a 2665 * back split, with the split proceeding out from the middle of 2666 * the system (less efficient) rather than the start of the 2667 * system (more efficient). the cases in which a back split 2668 * is required are rare and are limited to the first allocation 2669 * within an allocation group which is a part (not first part) 2670 * of a larger binary buddy system and a few exception cases 2671 * in which a previous join operation must be backed out. 2672 * 2673 * PARAMETERS: 2674 * tp - pointer to the tree containing the leaf. 2675 * leafno - the number of the leaf to be updated. 2676 * 2677 * RETURN VALUES: none 2678 * 2679 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2680 */ 2681 static int dbBackSplit(dmtree_t * tp, int leafno) 2682 { 2683 int budsz, bud, w, bsz, size; 2684 int cursz; 2685 s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx); 2686 2687 /* leaf should be part (not first part) of a binary 2688 * buddy system. 2689 */ 2690 assert(leaf[leafno] == NOFREE); 2691 2692 /* the back split is accomplished by iteratively finding the leaf 2693 * that starts the buddy system that contains the specified leaf and 2694 * splitting that system in two. this iteration continues until 2695 * the specified leaf becomes the start of a buddy system. 2696 * 2697 * determine maximum possible l2 size for the specified leaf. 2698 */ 2699 size = 2700 LITOL2BSZ(leafno, le32_to_cpu(tp->dmt_l2nleafs), 2701 tp->dmt_budmin); 2702 2703 /* determine the number of leaves covered by this size. this 2704 * is the buddy size that we will start with as we search for 2705 * the buddy system that contains the specified leaf. 2706 */ 2707 budsz = BUDSIZE(size, tp->dmt_budmin); 2708 2709 /* back split. 2710 */ 2711 while (leaf[leafno] == NOFREE) { 2712 /* find the leftmost buddy leaf. 2713 */ 2714 for (w = leafno, bsz = budsz;; bsz <<= 1, 2715 w = (w < bud) ? w : bud) { 2716 if (bsz >= le32_to_cpu(tp->dmt_nleafs)) { 2717 jfs_err("JFS: block map error in dbBackSplit"); 2718 return -EIO; 2719 } 2720 2721 /* determine the buddy. 2722 */ 2723 bud = w ^ bsz; 2724 2725 /* check if this buddy is the start of the system. 2726 */ 2727 if (leaf[bud] != NOFREE) { 2728 /* split the leaf at the start of the 2729 * system in two. 2730 */ 2731 cursz = leaf[bud] - 1; 2732 dbSplit(tp, bud, cursz, cursz); 2733 break; 2734 } 2735 } 2736 } 2737 2738 if (leaf[leafno] != size) { 2739 jfs_err("JFS: wrong leaf value in dbBackSplit"); 2740 return -EIO; 2741 } 2742 return 0; 2743 } 2744 2745 2746 /* 2747 * NAME: dbJoin() 2748 * 2749 * FUNCTION: update the leaf of a dmtree with a new value, joining 2750 * the leaf with other leaves of the dmtree into a multi-leaf 2751 * binary buddy system, as required. 2752 * 2753 * PARAMETERS: 2754 * tp - pointer to the tree containing the leaf. 2755 * leafno - the number of the leaf to be updated. 2756 * newval - the new value for the leaf. 2757 * 2758 * RETURN VALUES: none 2759 */ 2760 static int dbJoin(dmtree_t * tp, int leafno, int newval) 2761 { 2762 int budsz, buddy; 2763 s8 *leaf; 2764 2765 /* can the new leaf value require a join with other leaves ? 2766 */ 2767 if (newval >= tp->dmt_budmin) { 2768 /* pickup a pointer to the leaves of the tree. 2769 */ 2770 leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx); 2771 2772 /* try to join the specified leaf into a large binary 2773 * buddy system. the join proceeds by attempting to join 2774 * the specified leafno with its buddy (leaf) at new value. 2775 * if the join occurs, we attempt to join the left leaf 2776 * of the joined buddies with its buddy at new value + 1. 2777 * we continue to join until we find a buddy that cannot be 2778 * joined (does not have a value equal to the size of the 2779 * last join) or until all leaves have been joined into a 2780 * single system. 2781 * 2782 * get the buddy size (number of words covered) of 2783 * the new value. 2784 */ 2785 budsz = BUDSIZE(newval, tp->dmt_budmin); 2786 2787 /* try to join. 2788 */ 2789 while (budsz < le32_to_cpu(tp->dmt_nleafs)) { 2790 /* get the buddy leaf. 2791 */ 2792 buddy = leafno ^ budsz; 2793 2794 /* if the leaf's new value is greater than its 2795 * buddy's value, we join no more. 2796 */ 2797 if (newval > leaf[buddy]) 2798 break; 2799 2800 /* It shouldn't be less */ 2801 if (newval < leaf[buddy]) 2802 return -EIO; 2803 2804 /* check which (leafno or buddy) is the left buddy. 2805 * the left buddy gets to claim the blocks resulting 2806 * from the join while the right gets to claim none. 2807 * the left buddy is also eligible to participate in 2808 * a join at the next higher level while the right 2809 * is not. 2810 * 2811 */ 2812 if (leafno < buddy) { 2813 /* leafno is the left buddy. 2814 */ 2815 dbAdjTree(tp, buddy, NOFREE); 2816 } else { 2817 /* buddy is the left buddy and becomes 2818 * leafno. 2819 */ 2820 dbAdjTree(tp, leafno, NOFREE); 2821 leafno = buddy; 2822 } 2823 2824 /* on to try the next join. 2825 */ 2826 newval += 1; 2827 budsz <<= 1; 2828 } 2829 } 2830 2831 /* update the leaf value. 2832 */ 2833 dbAdjTree(tp, leafno, newval); 2834 2835 return 0; 2836 } 2837 2838 2839 /* 2840 * NAME: dbAdjTree() 2841 * 2842 * FUNCTION: update a leaf of a dmtree with a new value, adjusting 2843 * the dmtree, as required, to reflect the new leaf value. 2844 * the combination of any buddies must already be done before 2845 * this is called. 2846 * 2847 * PARAMETERS: 2848 * tp - pointer to the tree to be adjusted. 2849 * leafno - the number of the leaf to be updated. 2850 * newval - the new value for the leaf. 2851 * 2852 * RETURN VALUES: none 2853 */ 2854 static void dbAdjTree(dmtree_t * tp, int leafno, int newval) 2855 { 2856 int lp, pp, k; 2857 int max; 2858 2859 /* pick up the index of the leaf for this leafno. 2860 */ 2861 lp = leafno + le32_to_cpu(tp->dmt_leafidx); 2862 2863 /* is the current value the same as the old value ? if so, 2864 * there is nothing to do. 2865 */ 2866 if (tp->dmt_stree[lp] == newval) 2867 return; 2868 2869 /* set the new value. 2870 */ 2871 tp->dmt_stree[lp] = newval; 2872 2873 /* bubble the new value up the tree as required. 2874 */ 2875 for (k = 0; k < le32_to_cpu(tp->dmt_height); k++) { 2876 /* get the index of the first leaf of the 4 leaf 2877 * group containing the specified leaf (leafno). 2878 */ 2879 lp = ((lp - 1) & ~0x03) + 1; 2880 2881 /* get the index of the parent of this 4 leaf group. 2882 */ 2883 pp = (lp - 1) >> 2; 2884 2885 /* determine the maximum of the 4 leaves. 2886 */ 2887 max = TREEMAX(&tp->dmt_stree[lp]); 2888 2889 /* if the maximum of the 4 is the same as the 2890 * parent's value, we're done. 2891 */ 2892 if (tp->dmt_stree[pp] == max) 2893 break; 2894 2895 /* parent gets new value. 2896 */ 2897 tp->dmt_stree[pp] = max; 2898 2899 /* parent becomes leaf for next go-round. 2900 */ 2901 lp = pp; 2902 } 2903 } 2904 2905 2906 /* 2907 * NAME: dbFindLeaf() 2908 * 2909 * FUNCTION: search a dmtree_t for sufficient free blocks, returning 2910 * the index of a leaf describing the free blocks if 2911 * sufficient free blocks are found. 2912 * 2913 * the search starts at the top of the dmtree_t tree and 2914 * proceeds down the tree to the leftmost leaf with sufficient 2915 * free space. 2916 * 2917 * PARAMETERS: 2918 * tp - pointer to the tree to be searched. 2919 * l2nb - log2 number of free blocks to search for. 2920 * leafidx - return pointer to be set to the index of the leaf 2921 * describing at least l2nb free blocks if sufficient 2922 * free blocks are found. 2923 * 2924 * RETURN VALUES: 2925 * 0 - success 2926 * -ENOSPC - insufficient free blocks. 2927 */ 2928 static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx) 2929 { 2930 int ti, n = 0, k, x = 0; 2931 2932 /* first check the root of the tree to see if there is 2933 * sufficient free space. 2934 */ 2935 if (l2nb > tp->dmt_stree[ROOT]) 2936 return -ENOSPC; 2937 2938 /* sufficient free space available. now search down the tree 2939 * starting at the next level for the leftmost leaf that 2940 * describes sufficient free space. 2941 */ 2942 for (k = le32_to_cpu(tp->dmt_height), ti = 1; 2943 k > 0; k--, ti = ((ti + n) << 2) + 1) { 2944 /* search the four nodes at this level, starting from 2945 * the left. 2946 */ 2947 for (x = ti, n = 0; n < 4; n++) { 2948 /* sufficient free space found. move to the next 2949 * level (or quit if this is the last level). 2950 */ 2951 if (l2nb <= tp->dmt_stree[x + n]) 2952 break; 2953 } 2954 2955 /* better have found something since the higher 2956 * levels of the tree said it was here. 2957 */ 2958 assert(n < 4); 2959 } 2960 2961 /* set the return to the leftmost leaf describing sufficient 2962 * free space. 2963 */ 2964 *leafidx = x + n - le32_to_cpu(tp->dmt_leafidx); 2965 2966 return (0); 2967 } 2968 2969 2970 /* 2971 * NAME: dbFindBits() 2972 * 2973 * FUNCTION: find a specified number of binary buddy free bits within a 2974 * dmap bitmap word value. 2975 * 2976 * this routine searches the bitmap value for (1 << l2nb) free 2977 * bits at (1 << l2nb) alignments within the value. 2978 * 2979 * PARAMETERS: 2980 * word - dmap bitmap word value. 2981 * l2nb - number of free bits specified as a log2 number. 2982 * 2983 * RETURN VALUES: 2984 * starting bit number of free bits. 2985 */ 2986 static int dbFindBits(u32 word, int l2nb) 2987 { 2988 int bitno, nb; 2989 u32 mask; 2990 2991 /* get the number of bits. 2992 */ 2993 nb = 1 << l2nb; 2994 assert(nb <= DBWORD); 2995 2996 /* complement the word so we can use a mask (i.e. 0s represent 2997 * free bits) and compute the mask. 2998 */ 2999 word = ~word; 3000 mask = ONES << (DBWORD - nb); 3001 3002 /* scan the word for nb free bits at nb alignments. 3003 */ 3004 for (bitno = 0; mask != 0; bitno += nb, mask >>= nb) { 3005 if ((mask & word) == mask) 3006 break; 3007 } 3008 3009 ASSERT(bitno < 32); 3010 3011 /* return the bit number. 3012 */ 3013 return (bitno); 3014 } 3015 3016 3017 /* 3018 * NAME: dbMaxBud(u8 *cp) 3019 * 3020 * FUNCTION: determine the largest binary buddy string of free 3021 * bits within 32-bits of the map. 3022 * 3023 * PARAMETERS: 3024 * cp - pointer to the 32-bit value. 3025 * 3026 * RETURN VALUES: 3027 * largest binary buddy of free bits within a dmap word. 3028 */ 3029 static int dbMaxBud(u8 * cp) 3030 { 3031 signed char tmp1, tmp2; 3032 3033 /* check if the wmap word is all free. if so, the 3034 * free buddy size is BUDMIN. 3035 */ 3036 if (*((uint *) cp) == 0) 3037 return (BUDMIN); 3038 3039 /* check if the wmap word is half free. if so, the 3040 * free buddy size is BUDMIN-1. 3041 */ 3042 if (*((u16 *) cp) == 0 || *((u16 *) cp + 1) == 0) 3043 return (BUDMIN - 1); 3044 3045 /* not all free or half free. determine the free buddy 3046 * size thru table lookup using quarters of the wmap word. 3047 */ 3048 tmp1 = max(budtab[cp[2]], budtab[cp[3]]); 3049 tmp2 = max(budtab[cp[0]], budtab[cp[1]]); 3050 return (max(tmp1, tmp2)); 3051 } 3052 3053 3054 /* 3055 * NAME: cnttz(uint word) 3056 * 3057 * FUNCTION: determine the number of trailing zeros within a 32-bit 3058 * value. 3059 * 3060 * PARAMETERS: 3061 * value - 32-bit value to be examined. 3062 * 3063 * RETURN VALUES: 3064 * count of trailing zeros 3065 */ 3066 static int cnttz(u32 word) 3067 { 3068 int n; 3069 3070 for (n = 0; n < 32; n++, word >>= 1) { 3071 if (word & 0x01) 3072 break; 3073 } 3074 3075 return (n); 3076 } 3077 3078 3079 /* 3080 * NAME: cntlz(u32 value) 3081 * 3082 * FUNCTION: determine the number of leading zeros within a 32-bit 3083 * value. 3084 * 3085 * PARAMETERS: 3086 * value - 32-bit value to be examined. 3087 * 3088 * RETURN VALUES: 3089 * count of leading zeros 3090 */ 3091 static int cntlz(u32 value) 3092 { 3093 int n; 3094 3095 for (n = 0; n < 32; n++, value <<= 1) { 3096 if (value & HIGHORDER) 3097 break; 3098 } 3099 return (n); 3100 } 3101 3102 3103 /* 3104 * NAME: blkstol2(s64 nb) 3105 * 3106 * FUNCTION: convert a block count to its log2 value. if the block 3107 * count is not a l2 multiple, it is rounded up to the next 3108 * larger l2 multiple. 3109 * 3110 * PARAMETERS: 3111 * nb - number of blocks 3112 * 3113 * RETURN VALUES: 3114 * log2 number of blocks 3115 */ 3116 static int blkstol2(s64 nb) 3117 { 3118 int l2nb; 3119 s64 mask; /* meant to be signed */ 3120 3121 mask = (s64) 1 << (64 - 1); 3122 3123 /* count the leading bits. 3124 */ 3125 for (l2nb = 0; l2nb < 64; l2nb++, mask >>= 1) { 3126 /* leading bit found. 3127 */ 3128 if (nb & mask) { 3129 /* determine the l2 value. 3130 */ 3131 l2nb = (64 - 1) - l2nb; 3132 3133 /* check if we need to round up. 3134 */ 3135 if (~mask & nb) 3136 l2nb++; 3137 3138 return (l2nb); 3139 } 3140 } 3141 assert(0); 3142 return 0; /* fix compiler warning */ 3143 } 3144 3145 3146 /* 3147 * NAME: dbAllocBottomUp() 3148 * 3149 * FUNCTION: alloc the specified block range from the working block 3150 * allocation map. 3151 * 3152 * the blocks will be alloc from the working map one dmap 3153 * at a time. 3154 * 3155 * PARAMETERS: 3156 * ip - pointer to in-core inode; 3157 * blkno - starting block number to be freed. 3158 * nblocks - number of blocks to be freed. 3159 * 3160 * RETURN VALUES: 3161 * 0 - success 3162 * -EIO - i/o error 3163 */ 3164 int dbAllocBottomUp(struct inode *ip, s64 blkno, s64 nblocks) 3165 { 3166 struct metapage *mp; 3167 struct dmap *dp; 3168 int nb, rc; 3169 s64 lblkno, rem; 3170 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 3171 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap; 3172 3173 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP); 3174 3175 /* block to be allocated better be within the mapsize. */ 3176 ASSERT(nblocks <= bmp->db_mapsize - blkno); 3177 3178 /* 3179 * allocate the blocks a dmap at a time. 3180 */ 3181 mp = NULL; 3182 for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) { 3183 /* release previous dmap if any */ 3184 if (mp) { 3185 write_metapage(mp); 3186 } 3187 3188 /* get the buffer for the current dmap. */ 3189 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 3190 mp = read_metapage(ipbmap, lblkno, PSIZE, 0); 3191 if (mp == NULL) { 3192 IREAD_UNLOCK(ipbmap); 3193 return -EIO; 3194 } 3195 dp = (struct dmap *) mp->data; 3196 3197 /* determine the number of blocks to be allocated from 3198 * this dmap. 3199 */ 3200 nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1))); 3201 3202 /* allocate the blocks. */ 3203 if ((rc = dbAllocDmapBU(bmp, dp, blkno, nb))) { 3204 release_metapage(mp); 3205 IREAD_UNLOCK(ipbmap); 3206 return (rc); 3207 } 3208 } 3209 3210 /* write the last buffer. */ 3211 write_metapage(mp); 3212 3213 IREAD_UNLOCK(ipbmap); 3214 3215 return (0); 3216 } 3217 3218 3219 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno, 3220 int nblocks) 3221 { 3222 int rc; 3223 int dbitno, word, rembits, nb, nwords, wbitno, agno; 3224 s8 oldroot; 3225 struct dmaptree *tp = (struct dmaptree *) & dp->tree; 3226 3227 /* save the current value of the root (i.e. maximum free string) 3228 * of the dmap tree. 3229 */ 3230 oldroot = tp->stree[ROOT]; 3231 3232 /* determine the bit number and word within the dmap of the 3233 * starting block. 3234 */ 3235 dbitno = blkno & (BPERDMAP - 1); 3236 word = dbitno >> L2DBWORD; 3237 3238 /* block range better be within the dmap */ 3239 assert(dbitno + nblocks <= BPERDMAP); 3240 3241 /* allocate the bits of the dmap's words corresponding to the block 3242 * range. not all bits of the first and last words may be contained 3243 * within the block range. if this is the case, we'll work against 3244 * those words (i.e. partial first and/or last) on an individual basis 3245 * (a single pass), allocating the bits of interest by hand and 3246 * updating the leaf corresponding to the dmap word. a single pass 3247 * will be used for all dmap words fully contained within the 3248 * specified range. within this pass, the bits of all fully contained 3249 * dmap words will be marked as free in a single shot and the leaves 3250 * will be updated. a single leaf may describe the free space of 3251 * multiple dmap words, so we may update only a subset of the actual 3252 * leaves corresponding to the dmap words of the block range. 3253 */ 3254 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) { 3255 /* determine the bit number within the word and 3256 * the number of bits within the word. 3257 */ 3258 wbitno = dbitno & (DBWORD - 1); 3259 nb = min(rembits, DBWORD - wbitno); 3260 3261 /* check if only part of a word is to be allocated. 3262 */ 3263 if (nb < DBWORD) { 3264 /* allocate (set to 1) the appropriate bits within 3265 * this dmap word. 3266 */ 3267 dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb) 3268 >> wbitno); 3269 3270 word++; 3271 } else { 3272 /* one or more dmap words are fully contained 3273 * within the block range. determine how many 3274 * words and allocate (set to 1) the bits of these 3275 * words. 3276 */ 3277 nwords = rembits >> L2DBWORD; 3278 memset(&dp->wmap[word], (int) ONES, nwords * 4); 3279 3280 /* determine how many bits */ 3281 nb = nwords << L2DBWORD; 3282 word += nwords; 3283 } 3284 } 3285 3286 /* update the free count for this dmap */ 3287 le32_add_cpu(&dp->nfree, -nblocks); 3288 3289 /* reconstruct summary tree */ 3290 dbInitDmapTree(dp); 3291 3292 BMAP_LOCK(bmp); 3293 3294 /* if this allocation group is completely free, 3295 * update the highest active allocation group number 3296 * if this allocation group is the new max. 3297 */ 3298 agno = blkno >> bmp->db_agl2size; 3299 if (agno > bmp->db_maxag) 3300 bmp->db_maxag = agno; 3301 3302 /* update the free count for the allocation group and map */ 3303 bmp->db_agfree[agno] -= nblocks; 3304 bmp->db_nfree -= nblocks; 3305 3306 BMAP_UNLOCK(bmp); 3307 3308 /* if the root has not changed, done. */ 3309 if (tp->stree[ROOT] == oldroot) 3310 return (0); 3311 3312 /* root changed. bubble the change up to the dmap control pages. 3313 * if the adjustment of the upper level control pages fails, 3314 * backout the bit allocation (thus making everything consistent). 3315 */ 3316 if ((rc = dbAdjCtl(bmp, blkno, tp->stree[ROOT], 1, 0))) 3317 dbFreeBits(bmp, dp, blkno, nblocks); 3318 3319 return (rc); 3320 } 3321 3322 3323 /* 3324 * NAME: dbExtendFS() 3325 * 3326 * FUNCTION: extend bmap from blkno for nblocks; 3327 * dbExtendFS() updates bmap ready for dbAllocBottomUp(); 3328 * 3329 * L2 3330 * | 3331 * L1---------------------------------L1 3332 * | | 3333 * L0---------L0---------L0 L0---------L0---------L0 3334 * | | | | | | 3335 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm; 3336 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm 3337 * 3338 * <---old---><----------------------------extend-----------------------> 3339 */ 3340 int dbExtendFS(struct inode *ipbmap, s64 blkno, s64 nblocks) 3341 { 3342 struct jfs_sb_info *sbi = JFS_SBI(ipbmap->i_sb); 3343 int nbperpage = sbi->nbperpage; 3344 int i, i0 = true, j, j0 = true, k, n; 3345 s64 newsize; 3346 s64 p; 3347 struct metapage *mp, *l2mp, *l1mp = NULL, *l0mp = NULL; 3348 struct dmapctl *l2dcp, *l1dcp, *l0dcp; 3349 struct dmap *dp; 3350 s8 *l0leaf, *l1leaf, *l2leaf; 3351 struct bmap *bmp = sbi->bmap; 3352 int agno, l2agsize, oldl2agsize; 3353 s64 ag_rem; 3354 3355 newsize = blkno + nblocks; 3356 3357 jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld", 3358 (long long) blkno, (long long) nblocks, (long long) newsize); 3359 3360 /* 3361 * initialize bmap control page. 3362 * 3363 * all the data in bmap control page should exclude 3364 * the mkfs hidden dmap page. 3365 */ 3366 3367 /* update mapsize */ 3368 bmp->db_mapsize = newsize; 3369 bmp->db_maxlevel = BMAPSZTOLEV(bmp->db_mapsize); 3370 3371 /* compute new AG size */ 3372 l2agsize = dbGetL2AGSize(newsize); 3373 oldl2agsize = bmp->db_agl2size; 3374 3375 bmp->db_agl2size = l2agsize; 3376 bmp->db_agsize = 1 << l2agsize; 3377 3378 /* compute new number of AG */ 3379 agno = bmp->db_numag; 3380 bmp->db_numag = newsize >> l2agsize; 3381 bmp->db_numag += ((u32) newsize % (u32) bmp->db_agsize) ? 1 : 0; 3382 3383 /* 3384 * reconfigure db_agfree[] 3385 * from old AG configuration to new AG configuration; 3386 * 3387 * coalesce contiguous k (newAGSize/oldAGSize) AGs; 3388 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn; 3389 * note: new AG size = old AG size * (2**x). 3390 */ 3391 if (l2agsize == oldl2agsize) 3392 goto extend; 3393 k = 1 << (l2agsize - oldl2agsize); 3394 ag_rem = bmp->db_agfree[0]; /* save agfree[0] */ 3395 for (i = 0, n = 0; i < agno; n++) { 3396 bmp->db_agfree[n] = 0; /* init collection point */ 3397 3398 /* coalesce contiguous k AGs; */ 3399 for (j = 0; j < k && i < agno; j++, i++) { 3400 /* merge AGi to AGn */ 3401 bmp->db_agfree[n] += bmp->db_agfree[i]; 3402 } 3403 } 3404 bmp->db_agfree[0] += ag_rem; /* restore agfree[0] */ 3405 3406 for (; n < MAXAG; n++) 3407 bmp->db_agfree[n] = 0; 3408 3409 /* 3410 * update highest active ag number 3411 */ 3412 3413 bmp->db_maxag = bmp->db_maxag / k; 3414 3415 /* 3416 * extend bmap 3417 * 3418 * update bit maps and corresponding level control pages; 3419 * global control page db_nfree, db_agfree[agno], db_maxfreebud; 3420 */ 3421 extend: 3422 /* get L2 page */ 3423 p = BMAPBLKNO + nbperpage; /* L2 page */ 3424 l2mp = read_metapage(ipbmap, p, PSIZE, 0); 3425 if (!l2mp) { 3426 jfs_error(ipbmap->i_sb, "L2 page could not be read\n"); 3427 return -EIO; 3428 } 3429 l2dcp = (struct dmapctl *) l2mp->data; 3430 3431 /* compute start L1 */ 3432 k = blkno >> L2MAXL1SIZE; 3433 l2leaf = l2dcp->stree + CTLLEAFIND + k; 3434 p = BLKTOL1(blkno, sbi->l2nbperpage); /* L1 page */ 3435 3436 /* 3437 * extend each L1 in L2 3438 */ 3439 for (; k < LPERCTL; k++, p += nbperpage) { 3440 /* get L1 page */ 3441 if (j0) { 3442 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */ 3443 l1mp = read_metapage(ipbmap, p, PSIZE, 0); 3444 if (l1mp == NULL) 3445 goto errout; 3446 l1dcp = (struct dmapctl *) l1mp->data; 3447 3448 /* compute start L0 */ 3449 j = (blkno & (MAXL1SIZE - 1)) >> L2MAXL0SIZE; 3450 l1leaf = l1dcp->stree + CTLLEAFIND + j; 3451 p = BLKTOL0(blkno, sbi->l2nbperpage); 3452 j0 = false; 3453 } else { 3454 /* assign/init L1 page */ 3455 l1mp = get_metapage(ipbmap, p, PSIZE, 0); 3456 if (l1mp == NULL) 3457 goto errout; 3458 3459 l1dcp = (struct dmapctl *) l1mp->data; 3460 3461 /* compute start L0 */ 3462 j = 0; 3463 l1leaf = l1dcp->stree + CTLLEAFIND; 3464 p += nbperpage; /* 1st L0 of L1.k */ 3465 } 3466 3467 /* 3468 * extend each L0 in L1 3469 */ 3470 for (; j < LPERCTL; j++) { 3471 /* get L0 page */ 3472 if (i0) { 3473 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */ 3474 3475 l0mp = read_metapage(ipbmap, p, PSIZE, 0); 3476 if (l0mp == NULL) 3477 goto errout; 3478 l0dcp = (struct dmapctl *) l0mp->data; 3479 3480 /* compute start dmap */ 3481 i = (blkno & (MAXL0SIZE - 1)) >> 3482 L2BPERDMAP; 3483 l0leaf = l0dcp->stree + CTLLEAFIND + i; 3484 p = BLKTODMAP(blkno, 3485 sbi->l2nbperpage); 3486 i0 = false; 3487 } else { 3488 /* assign/init L0 page */ 3489 l0mp = get_metapage(ipbmap, p, PSIZE, 0); 3490 if (l0mp == NULL) 3491 goto errout; 3492 3493 l0dcp = (struct dmapctl *) l0mp->data; 3494 3495 /* compute start dmap */ 3496 i = 0; 3497 l0leaf = l0dcp->stree + CTLLEAFIND; 3498 p += nbperpage; /* 1st dmap of L0.j */ 3499 } 3500 3501 /* 3502 * extend each dmap in L0 3503 */ 3504 for (; i < LPERCTL; i++) { 3505 /* 3506 * reconstruct the dmap page, and 3507 * initialize corresponding parent L0 leaf 3508 */ 3509 if ((n = blkno & (BPERDMAP - 1))) { 3510 /* read in dmap page: */ 3511 mp = read_metapage(ipbmap, p, 3512 PSIZE, 0); 3513 if (mp == NULL) 3514 goto errout; 3515 n = min(nblocks, (s64)BPERDMAP - n); 3516 } else { 3517 /* assign/init dmap page */ 3518 mp = read_metapage(ipbmap, p, 3519 PSIZE, 0); 3520 if (mp == NULL) 3521 goto errout; 3522 3523 n = min_t(s64, nblocks, BPERDMAP); 3524 } 3525 3526 dp = (struct dmap *) mp->data; 3527 *l0leaf = dbInitDmap(dp, blkno, n); 3528 3529 bmp->db_nfree += n; 3530 agno = le64_to_cpu(dp->start) >> l2agsize; 3531 bmp->db_agfree[agno] += n; 3532 3533 write_metapage(mp); 3534 3535 l0leaf++; 3536 p += nbperpage; 3537 3538 blkno += n; 3539 nblocks -= n; 3540 if (nblocks == 0) 3541 break; 3542 } /* for each dmap in a L0 */ 3543 3544 /* 3545 * build current L0 page from its leaves, and 3546 * initialize corresponding parent L1 leaf 3547 */ 3548 *l1leaf = dbInitDmapCtl(l0dcp, 0, ++i); 3549 write_metapage(l0mp); 3550 l0mp = NULL; 3551 3552 if (nblocks) 3553 l1leaf++; /* continue for next L0 */ 3554 else { 3555 /* more than 1 L0 ? */ 3556 if (j > 0) 3557 break; /* build L1 page */ 3558 else { 3559 /* summarize in global bmap page */ 3560 bmp->db_maxfreebud = *l1leaf; 3561 release_metapage(l1mp); 3562 release_metapage(l2mp); 3563 goto finalize; 3564 } 3565 } 3566 } /* for each L0 in a L1 */ 3567 3568 /* 3569 * build current L1 page from its leaves, and 3570 * initialize corresponding parent L2 leaf 3571 */ 3572 *l2leaf = dbInitDmapCtl(l1dcp, 1, ++j); 3573 write_metapage(l1mp); 3574 l1mp = NULL; 3575 3576 if (nblocks) 3577 l2leaf++; /* continue for next L1 */ 3578 else { 3579 /* more than 1 L1 ? */ 3580 if (k > 0) 3581 break; /* build L2 page */ 3582 else { 3583 /* summarize in global bmap page */ 3584 bmp->db_maxfreebud = *l2leaf; 3585 release_metapage(l2mp); 3586 goto finalize; 3587 } 3588 } 3589 } /* for each L1 in a L2 */ 3590 3591 jfs_error(ipbmap->i_sb, "function has not returned as expected\n"); 3592 errout: 3593 if (l0mp) 3594 release_metapage(l0mp); 3595 if (l1mp) 3596 release_metapage(l1mp); 3597 release_metapage(l2mp); 3598 return -EIO; 3599 3600 /* 3601 * finalize bmap control page 3602 */ 3603 finalize: 3604 3605 return 0; 3606 } 3607 3608 3609 /* 3610 * dbFinalizeBmap() 3611 */ 3612 void dbFinalizeBmap(struct inode *ipbmap) 3613 { 3614 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 3615 int actags, inactags, l2nl; 3616 s64 ag_rem, actfree, inactfree, avgfree; 3617 int i, n; 3618 3619 /* 3620 * finalize bmap control page 3621 */ 3622 //finalize: 3623 /* 3624 * compute db_agpref: preferred ag to allocate from 3625 * (the leftmost ag with average free space in it); 3626 */ 3627 //agpref: 3628 /* get the number of active ags and inactive ags */ 3629 actags = bmp->db_maxag + 1; 3630 inactags = bmp->db_numag - actags; 3631 ag_rem = bmp->db_mapsize & (bmp->db_agsize - 1); /* ??? */ 3632 3633 /* determine how many blocks are in the inactive allocation 3634 * groups. in doing this, we must account for the fact that 3635 * the rightmost group might be a partial group (i.e. file 3636 * system size is not a multiple of the group size). 3637 */ 3638 inactfree = (inactags && ag_rem) ? 3639 ((inactags - 1) << bmp->db_agl2size) + ag_rem 3640 : inactags << bmp->db_agl2size; 3641 3642 /* determine how many free blocks are in the active 3643 * allocation groups plus the average number of free blocks 3644 * within the active ags. 3645 */ 3646 actfree = bmp->db_nfree - inactfree; 3647 avgfree = (u32) actfree / (u32) actags; 3648 3649 /* if the preferred allocation group has not average free space. 3650 * re-establish the preferred group as the leftmost 3651 * group with average free space. 3652 */ 3653 if (bmp->db_agfree[bmp->db_agpref] < avgfree) { 3654 for (bmp->db_agpref = 0; bmp->db_agpref < actags; 3655 bmp->db_agpref++) { 3656 if (bmp->db_agfree[bmp->db_agpref] >= avgfree) 3657 break; 3658 } 3659 if (bmp->db_agpref >= bmp->db_numag) { 3660 jfs_error(ipbmap->i_sb, 3661 "cannot find ag with average freespace\n"); 3662 } 3663 } 3664 3665 /* 3666 * compute db_aglevel, db_agheight, db_width, db_agstart: 3667 * an ag is covered in aglevel dmapctl summary tree, 3668 * at agheight level height (from leaf) with agwidth number of nodes 3669 * each, which starts at agstart index node of the smmary tree node 3670 * array; 3671 */ 3672 bmp->db_aglevel = BMAPSZTOLEV(bmp->db_agsize); 3673 l2nl = 3674 bmp->db_agl2size - (L2BPERDMAP + bmp->db_aglevel * L2LPERCTL); 3675 bmp->db_agheight = l2nl >> 1; 3676 bmp->db_agwidth = 1 << (l2nl - (bmp->db_agheight << 1)); 3677 for (i = 5 - bmp->db_agheight, bmp->db_agstart = 0, n = 1; i > 0; 3678 i--) { 3679 bmp->db_agstart += n; 3680 n <<= 2; 3681 } 3682 3683 } 3684 3685 3686 /* 3687 * NAME: dbInitDmap()/ujfs_idmap_page() 3688 * 3689 * FUNCTION: initialize working/persistent bitmap of the dmap page 3690 * for the specified number of blocks: 3691 * 3692 * at entry, the bitmaps had been initialized as free (ZEROS); 3693 * The number of blocks will only account for the actually 3694 * existing blocks. Blocks which don't actually exist in 3695 * the aggregate will be marked as allocated (ONES); 3696 * 3697 * PARAMETERS: 3698 * dp - pointer to page of map 3699 * nblocks - number of blocks this page 3700 * 3701 * RETURNS: NONE 3702 */ 3703 static int dbInitDmap(struct dmap * dp, s64 Blkno, int nblocks) 3704 { 3705 int blkno, w, b, r, nw, nb, i; 3706 3707 /* starting block number within the dmap */ 3708 blkno = Blkno & (BPERDMAP - 1); 3709 3710 if (blkno == 0) { 3711 dp->nblocks = dp->nfree = cpu_to_le32(nblocks); 3712 dp->start = cpu_to_le64(Blkno); 3713 3714 if (nblocks == BPERDMAP) { 3715 memset(&dp->wmap[0], 0, LPERDMAP * 4); 3716 memset(&dp->pmap[0], 0, LPERDMAP * 4); 3717 goto initTree; 3718 } 3719 } else { 3720 le32_add_cpu(&dp->nblocks, nblocks); 3721 le32_add_cpu(&dp->nfree, nblocks); 3722 } 3723 3724 /* word number containing start block number */ 3725 w = blkno >> L2DBWORD; 3726 3727 /* 3728 * free the bits corresponding to the block range (ZEROS): 3729 * note: not all bits of the first and last words may be contained 3730 * within the block range. 3731 */ 3732 for (r = nblocks; r > 0; r -= nb, blkno += nb) { 3733 /* number of bits preceding range to be freed in the word */ 3734 b = blkno & (DBWORD - 1); 3735 /* number of bits to free in the word */ 3736 nb = min(r, DBWORD - b); 3737 3738 /* is partial word to be freed ? */ 3739 if (nb < DBWORD) { 3740 /* free (set to 0) from the bitmap word */ 3741 dp->wmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb) 3742 >> b)); 3743 dp->pmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb) 3744 >> b)); 3745 3746 /* skip the word freed */ 3747 w++; 3748 } else { 3749 /* free (set to 0) contiguous bitmap words */ 3750 nw = r >> L2DBWORD; 3751 memset(&dp->wmap[w], 0, nw * 4); 3752 memset(&dp->pmap[w], 0, nw * 4); 3753 3754 /* skip the words freed */ 3755 nb = nw << L2DBWORD; 3756 w += nw; 3757 } 3758 } 3759 3760 /* 3761 * mark bits following the range to be freed (non-existing 3762 * blocks) as allocated (ONES) 3763 */ 3764 3765 if (blkno == BPERDMAP) 3766 goto initTree; 3767 3768 /* the first word beyond the end of existing blocks */ 3769 w = blkno >> L2DBWORD; 3770 3771 /* does nblocks fall on a 32-bit boundary ? */ 3772 b = blkno & (DBWORD - 1); 3773 if (b) { 3774 /* mark a partial word allocated */ 3775 dp->wmap[w] = dp->pmap[w] = cpu_to_le32(ONES >> b); 3776 w++; 3777 } 3778 3779 /* set the rest of the words in the page to allocated (ONES) */ 3780 for (i = w; i < LPERDMAP; i++) 3781 dp->pmap[i] = dp->wmap[i] = cpu_to_le32(ONES); 3782 3783 /* 3784 * init tree 3785 */ 3786 initTree: 3787 return (dbInitDmapTree(dp)); 3788 } 3789 3790 3791 /* 3792 * NAME: dbInitDmapTree()/ujfs_complete_dmap() 3793 * 3794 * FUNCTION: initialize summary tree of the specified dmap: 3795 * 3796 * at entry, bitmap of the dmap has been initialized; 3797 * 3798 * PARAMETERS: 3799 * dp - dmap to complete 3800 * blkno - starting block number for this dmap 3801 * treemax - will be filled in with max free for this dmap 3802 * 3803 * RETURNS: max free string at the root of the tree 3804 */ 3805 static int dbInitDmapTree(struct dmap * dp) 3806 { 3807 struct dmaptree *tp; 3808 s8 *cp; 3809 int i; 3810 3811 /* init fixed info of tree */ 3812 tp = &dp->tree; 3813 tp->nleafs = cpu_to_le32(LPERDMAP); 3814 tp->l2nleafs = cpu_to_le32(L2LPERDMAP); 3815 tp->leafidx = cpu_to_le32(LEAFIND); 3816 tp->height = cpu_to_le32(4); 3817 tp->budmin = BUDMIN; 3818 3819 /* init each leaf from corresponding wmap word: 3820 * note: leaf is set to NOFREE(-1) if all blocks of corresponding 3821 * bitmap word are allocated. 3822 */ 3823 cp = tp->stree + le32_to_cpu(tp->leafidx); 3824 for (i = 0; i < LPERDMAP; i++) 3825 *cp++ = dbMaxBud((u8 *) & dp->wmap[i]); 3826 3827 /* build the dmap's binary buddy summary tree */ 3828 return (dbInitTree(tp)); 3829 } 3830 3831 3832 /* 3833 * NAME: dbInitTree()/ujfs_adjtree() 3834 * 3835 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl. 3836 * 3837 * at entry, the leaves of the tree has been initialized 3838 * from corresponding bitmap word or root of summary tree 3839 * of the child control page; 3840 * configure binary buddy system at the leaf level, then 3841 * bubble up the values of the leaf nodes up the tree. 3842 * 3843 * PARAMETERS: 3844 * cp - Pointer to the root of the tree 3845 * l2leaves- Number of leaf nodes as a power of 2 3846 * l2min - Number of blocks that can be covered by a leaf 3847 * as a power of 2 3848 * 3849 * RETURNS: max free string at the root of the tree 3850 */ 3851 static int dbInitTree(struct dmaptree * dtp) 3852 { 3853 int l2max, l2free, bsize, nextb, i; 3854 int child, parent, nparent; 3855 s8 *tp, *cp, *cp1; 3856 3857 tp = dtp->stree; 3858 3859 /* Determine the maximum free string possible for the leaves */ 3860 l2max = le32_to_cpu(dtp->l2nleafs) + dtp->budmin; 3861 3862 /* 3863 * configure the leaf level into binary buddy system 3864 * 3865 * Try to combine buddies starting with a buddy size of 1 3866 * (i.e. two leaves). At a buddy size of 1 two buddy leaves 3867 * can be combined if both buddies have a maximum free of l2min; 3868 * the combination will result in the left-most buddy leaf having 3869 * a maximum free of l2min+1. 3870 * After processing all buddies for a given size, process buddies 3871 * at the next higher buddy size (i.e. current size * 2) and 3872 * the next maximum free (current free + 1). 3873 * This continues until the maximum possible buddy combination 3874 * yields maximum free. 3875 */ 3876 for (l2free = dtp->budmin, bsize = 1; l2free < l2max; 3877 l2free++, bsize = nextb) { 3878 /* get next buddy size == current buddy pair size */ 3879 nextb = bsize << 1; 3880 3881 /* scan each adjacent buddy pair at current buddy size */ 3882 for (i = 0, cp = tp + le32_to_cpu(dtp->leafidx); 3883 i < le32_to_cpu(dtp->nleafs); 3884 i += nextb, cp += nextb) { 3885 /* coalesce if both adjacent buddies are max free */ 3886 if (*cp == l2free && *(cp + bsize) == l2free) { 3887 *cp = l2free + 1; /* left take right */ 3888 *(cp + bsize) = -1; /* right give left */ 3889 } 3890 } 3891 } 3892 3893 /* 3894 * bubble summary information of leaves up the tree. 3895 * 3896 * Starting at the leaf node level, the four nodes described by 3897 * the higher level parent node are compared for a maximum free and 3898 * this maximum becomes the value of the parent node. 3899 * when all lower level nodes are processed in this fashion then 3900 * move up to the next level (parent becomes a lower level node) and 3901 * continue the process for that level. 3902 */ 3903 for (child = le32_to_cpu(dtp->leafidx), 3904 nparent = le32_to_cpu(dtp->nleafs) >> 2; 3905 nparent > 0; nparent >>= 2, child = parent) { 3906 /* get index of 1st node of parent level */ 3907 parent = (child - 1) >> 2; 3908 3909 /* set the value of the parent node as the maximum 3910 * of the four nodes of the current level. 3911 */ 3912 for (i = 0, cp = tp + child, cp1 = tp + parent; 3913 i < nparent; i++, cp += 4, cp1++) 3914 *cp1 = TREEMAX(cp); 3915 } 3916 3917 return (*tp); 3918 } 3919 3920 3921 /* 3922 * dbInitDmapCtl() 3923 * 3924 * function: initialize dmapctl page 3925 */ 3926 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i) 3927 { /* start leaf index not covered by range */ 3928 s8 *cp; 3929 3930 dcp->nleafs = cpu_to_le32(LPERCTL); 3931 dcp->l2nleafs = cpu_to_le32(L2LPERCTL); 3932 dcp->leafidx = cpu_to_le32(CTLLEAFIND); 3933 dcp->height = cpu_to_le32(5); 3934 dcp->budmin = L2BPERDMAP + L2LPERCTL * level; 3935 3936 /* 3937 * initialize the leaves of current level that were not covered 3938 * by the specified input block range (i.e. the leaves have no 3939 * low level dmapctl or dmap). 3940 */ 3941 cp = &dcp->stree[CTLLEAFIND + i]; 3942 for (; i < LPERCTL; i++) 3943 *cp++ = NOFREE; 3944 3945 /* build the dmap's binary buddy summary tree */ 3946 return (dbInitTree((struct dmaptree *) dcp)); 3947 } 3948 3949 3950 /* 3951 * NAME: dbGetL2AGSize()/ujfs_getagl2size() 3952 * 3953 * FUNCTION: Determine log2(allocation group size) from aggregate size 3954 * 3955 * PARAMETERS: 3956 * nblocks - Number of blocks in aggregate 3957 * 3958 * RETURNS: log2(allocation group size) in aggregate blocks 3959 */ 3960 static int dbGetL2AGSize(s64 nblocks) 3961 { 3962 s64 sz; 3963 s64 m; 3964 int l2sz; 3965 3966 if (nblocks < BPERDMAP * MAXAG) 3967 return (L2BPERDMAP); 3968 3969 /* round up aggregate size to power of 2 */ 3970 m = ((u64) 1 << (64 - 1)); 3971 for (l2sz = 64; l2sz >= 0; l2sz--, m >>= 1) { 3972 if (m & nblocks) 3973 break; 3974 } 3975 3976 sz = (s64) 1 << l2sz; 3977 if (sz < nblocks) 3978 l2sz += 1; 3979 3980 /* agsize = roundupSize/max_number_of_ag */ 3981 return (l2sz - L2MAXAG); 3982 } 3983 3984 3985 /* 3986 * NAME: dbMapFileSizeToMapSize() 3987 * 3988 * FUNCTION: compute number of blocks the block allocation map file 3989 * can cover from the map file size; 3990 * 3991 * RETURNS: Number of blocks which can be covered by this block map file; 3992 */ 3993 3994 /* 3995 * maximum number of map pages at each level including control pages 3996 */ 3997 #define MAXL0PAGES (1 + LPERCTL) 3998 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES) 3999 4000 /* 4001 * convert number of map pages to the zero origin top dmapctl level 4002 */ 4003 #define BMAPPGTOLEV(npages) \ 4004 (((npages) <= 3 + MAXL0PAGES) ? 0 : \ 4005 ((npages) <= 2 + MAXL1PAGES) ? 1 : 2) 4006 4007 s64 dbMapFileSizeToMapSize(struct inode * ipbmap) 4008 { 4009 struct super_block *sb = ipbmap->i_sb; 4010 s64 nblocks; 4011 s64 npages, ndmaps; 4012 int level, i; 4013 int complete, factor; 4014 4015 nblocks = ipbmap->i_size >> JFS_SBI(sb)->l2bsize; 4016 npages = nblocks >> JFS_SBI(sb)->l2nbperpage; 4017 level = BMAPPGTOLEV(npages); 4018 4019 /* At each level, accumulate the number of dmap pages covered by 4020 * the number of full child levels below it; 4021 * repeat for the last incomplete child level. 4022 */ 4023 ndmaps = 0; 4024 npages--; /* skip the first global control page */ 4025 /* skip higher level control pages above top level covered by map */ 4026 npages -= (2 - level); 4027 npages--; /* skip top level's control page */ 4028 for (i = level; i >= 0; i--) { 4029 factor = 4030 (i == 2) ? MAXL1PAGES : ((i == 1) ? MAXL0PAGES : 1); 4031 complete = (u32) npages / factor; 4032 ndmaps += complete * ((i == 2) ? LPERCTL * LPERCTL : 4033 ((i == 1) ? LPERCTL : 1)); 4034 4035 /* pages in last/incomplete child */ 4036 npages = (u32) npages % factor; 4037 /* skip incomplete child's level control page */ 4038 npages--; 4039 } 4040 4041 /* convert the number of dmaps into the number of blocks 4042 * which can be covered by the dmaps; 4043 */ 4044 nblocks = ndmaps << L2BPERDMAP; 4045 4046 return (nblocks); 4047 } 4048