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, bool is_ctl); 67 static int dbBackSplit(dmtree_t *tp, int leafno, bool is_ctl); 68 static int dbJoin(dmtree_t *tp, int leafno, int newval, bool is_ctl); 69 static void dbAdjTree(dmtree_t *tp, int leafno, int newval, bool is_ctl); 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, bool is_ctl); 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 bmp->db_l2nbperpage = le32_to_cpu(dbmp_le->dn_l2nbperpage); 182 bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag); 183 bmp->db_maxlevel = le32_to_cpu(dbmp_le->dn_maxlevel); 184 bmp->db_maxag = le32_to_cpu(dbmp_le->dn_maxag); 185 bmp->db_agpref = le32_to_cpu(dbmp_le->dn_agpref); 186 bmp->db_aglevel = le32_to_cpu(dbmp_le->dn_aglevel); 187 bmp->db_agheight = le32_to_cpu(dbmp_le->dn_agheight); 188 bmp->db_agwidth = le32_to_cpu(dbmp_le->dn_agwidth); 189 bmp->db_agstart = le32_to_cpu(dbmp_le->dn_agstart); 190 bmp->db_agl2size = le32_to_cpu(dbmp_le->dn_agl2size); 191 192 if ((bmp->db_l2nbperpage > L2PSIZE - L2MINBLOCKSIZE) || 193 (bmp->db_l2nbperpage < 0) || 194 !bmp->db_numag || (bmp->db_numag > MAXAG) || 195 (bmp->db_maxag >= MAXAG) || (bmp->db_maxag < 0) || 196 (bmp->db_agpref >= MAXAG) || (bmp->db_agpref < 0) || 197 (bmp->db_agheight < 0) || (bmp->db_agheight > (L2LPERCTL >> 1)) || 198 (bmp->db_agwidth < 1) || (bmp->db_agwidth > (LPERCTL / MAXAG)) || 199 (bmp->db_agwidth > (1 << (L2LPERCTL - (bmp->db_agheight << 1)))) || 200 (bmp->db_agstart < 0) || 201 (bmp->db_agstart > (CTLTREESIZE - 1 - bmp->db_agwidth * (MAXAG - 1))) || 202 (bmp->db_agl2size > L2MAXL2SIZE - L2MAXAG) || 203 (bmp->db_agl2size < 0) || 204 ((bmp->db_mapsize - 1) >> bmp->db_agl2size) > MAXAG) { 205 err = -EINVAL; 206 goto err_release_metapage; 207 } 208 209 for (i = 0; i < MAXAG; i++) 210 bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]); 211 bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize); 212 bmp->db_maxfreebud = dbmp_le->dn_maxfreebud; 213 214 /* release the buffer. */ 215 release_metapage(mp); 216 217 /* bind the bmap inode and the bmap descriptor to each other. */ 218 bmp->db_ipbmap = ipbmap; 219 JFS_SBI(ipbmap->i_sb)->bmap = bmp; 220 221 memset(bmp->db_active, 0, sizeof(bmp->db_active)); 222 223 /* 224 * allocate/initialize the bmap lock 225 */ 226 BMAP_LOCK_INIT(bmp); 227 228 return (0); 229 230 err_release_metapage: 231 release_metapage(mp); 232 err_kfree_bmp: 233 kfree(bmp); 234 return err; 235 } 236 237 238 /* 239 * NAME: dbUnmount() 240 * 241 * FUNCTION: terminate the block allocation map in preparation for 242 * file system unmount. 243 * 244 * the in-core bmap descriptor is written to disk and 245 * the memory for this descriptor is freed. 246 * 247 * PARAMETERS: 248 * ipbmap - pointer to in-core inode for the block map. 249 * 250 * RETURN VALUES: 251 * 0 - success 252 * -EIO - i/o error 253 */ 254 int dbUnmount(struct inode *ipbmap, int mounterror) 255 { 256 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 257 258 if (!(mounterror || isReadOnly(ipbmap))) 259 dbSync(ipbmap); 260 261 /* 262 * Invalidate the page cache buffers 263 */ 264 truncate_inode_pages(ipbmap->i_mapping, 0); 265 266 /* free the memory for the in-memory bmap. */ 267 kfree(bmp); 268 JFS_SBI(ipbmap->i_sb)->bmap = NULL; 269 270 return (0); 271 } 272 273 /* 274 * dbSync() 275 */ 276 int dbSync(struct inode *ipbmap) 277 { 278 struct dbmap_disk *dbmp_le; 279 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 280 struct metapage *mp; 281 int i; 282 283 /* 284 * write bmap global control page 285 */ 286 /* get the buffer for the on-disk bmap descriptor. */ 287 mp = read_metapage(ipbmap, 288 BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage, 289 PSIZE, 0); 290 if (mp == NULL) { 291 jfs_err("dbSync: read_metapage failed!"); 292 return -EIO; 293 } 294 /* copy the in-memory version of the bmap to the on-disk version */ 295 dbmp_le = (struct dbmap_disk *) mp->data; 296 dbmp_le->dn_mapsize = cpu_to_le64(bmp->db_mapsize); 297 dbmp_le->dn_nfree = cpu_to_le64(bmp->db_nfree); 298 dbmp_le->dn_l2nbperpage = cpu_to_le32(bmp->db_l2nbperpage); 299 dbmp_le->dn_numag = cpu_to_le32(bmp->db_numag); 300 dbmp_le->dn_maxlevel = cpu_to_le32(bmp->db_maxlevel); 301 dbmp_le->dn_maxag = cpu_to_le32(bmp->db_maxag); 302 dbmp_le->dn_agpref = cpu_to_le32(bmp->db_agpref); 303 dbmp_le->dn_aglevel = cpu_to_le32(bmp->db_aglevel); 304 dbmp_le->dn_agheight = cpu_to_le32(bmp->db_agheight); 305 dbmp_le->dn_agwidth = cpu_to_le32(bmp->db_agwidth); 306 dbmp_le->dn_agstart = cpu_to_le32(bmp->db_agstart); 307 dbmp_le->dn_agl2size = cpu_to_le32(bmp->db_agl2size); 308 for (i = 0; i < MAXAG; i++) 309 dbmp_le->dn_agfree[i] = cpu_to_le64(bmp->db_agfree[i]); 310 dbmp_le->dn_agsize = cpu_to_le64(bmp->db_agsize); 311 dbmp_le->dn_maxfreebud = bmp->db_maxfreebud; 312 313 /* write the buffer */ 314 write_metapage(mp); 315 316 /* 317 * write out dirty pages of bmap 318 */ 319 filemap_write_and_wait(ipbmap->i_mapping); 320 321 diWriteSpecial(ipbmap, 0); 322 323 return (0); 324 } 325 326 /* 327 * NAME: dbFree() 328 * 329 * FUNCTION: free the specified block range from the working block 330 * allocation map. 331 * 332 * the blocks will be free from the working map one dmap 333 * at a time. 334 * 335 * PARAMETERS: 336 * ip - pointer to in-core inode; 337 * blkno - starting block number to be freed. 338 * nblocks - number of blocks to be freed. 339 * 340 * RETURN VALUES: 341 * 0 - success 342 * -EIO - i/o error 343 */ 344 int dbFree(struct inode *ip, s64 blkno, s64 nblocks) 345 { 346 struct metapage *mp; 347 struct dmap *dp; 348 int nb, rc; 349 s64 lblkno, rem; 350 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 351 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap; 352 struct super_block *sb = ipbmap->i_sb; 353 354 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP); 355 356 /* block to be freed better be within the mapsize. */ 357 if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) { 358 IREAD_UNLOCK(ipbmap); 359 printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n", 360 (unsigned long long) blkno, 361 (unsigned long long) nblocks); 362 jfs_error(ip->i_sb, "block to be freed is outside the map\n"); 363 return -EIO; 364 } 365 366 /** 367 * TRIM the blocks, when mounted with discard option 368 */ 369 if (JFS_SBI(sb)->flag & JFS_DISCARD) 370 if (JFS_SBI(sb)->minblks_trim <= nblocks) 371 jfs_issue_discard(ipbmap, blkno, nblocks); 372 373 /* 374 * free the blocks a dmap at a time. 375 */ 376 mp = NULL; 377 for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) { 378 /* release previous dmap if any */ 379 if (mp) { 380 write_metapage(mp); 381 } 382 383 /* get the buffer for the current dmap. */ 384 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 385 mp = read_metapage(ipbmap, lblkno, PSIZE, 0); 386 if (mp == NULL) { 387 IREAD_UNLOCK(ipbmap); 388 return -EIO; 389 } 390 dp = (struct dmap *) mp->data; 391 392 /* determine the number of blocks to be freed from 393 * this dmap. 394 */ 395 nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1))); 396 397 /* free the blocks. */ 398 if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) { 399 jfs_error(ip->i_sb, "error in block map\n"); 400 release_metapage(mp); 401 IREAD_UNLOCK(ipbmap); 402 return (rc); 403 } 404 } 405 406 /* write the last buffer. */ 407 if (mp) 408 write_metapage(mp); 409 410 IREAD_UNLOCK(ipbmap); 411 412 return (0); 413 } 414 415 416 /* 417 * NAME: dbUpdatePMap() 418 * 419 * FUNCTION: update the allocation state (free or allocate) of the 420 * specified block range in the persistent block allocation map. 421 * 422 * the blocks will be updated in the persistent map one 423 * dmap at a time. 424 * 425 * PARAMETERS: 426 * ipbmap - pointer to in-core inode for the block map. 427 * free - 'true' if block range is to be freed from the persistent 428 * map; 'false' if it is to be allocated. 429 * blkno - starting block number of the range. 430 * nblocks - number of contiguous blocks in the range. 431 * tblk - transaction block; 432 * 433 * RETURN VALUES: 434 * 0 - success 435 * -EIO - i/o error 436 */ 437 int 438 dbUpdatePMap(struct inode *ipbmap, 439 int free, s64 blkno, s64 nblocks, struct tblock * tblk) 440 { 441 int nblks, dbitno, wbitno, rbits; 442 int word, nbits, nwords; 443 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 444 s64 lblkno, rem, lastlblkno; 445 u32 mask; 446 struct dmap *dp; 447 struct metapage *mp; 448 struct jfs_log *log; 449 int lsn, difft, diffp; 450 unsigned long flags; 451 452 /* the blocks better be within the mapsize. */ 453 if (blkno + nblocks > bmp->db_mapsize) { 454 printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n", 455 (unsigned long long) blkno, 456 (unsigned long long) nblocks); 457 jfs_error(ipbmap->i_sb, "blocks are outside the map\n"); 458 return -EIO; 459 } 460 461 /* compute delta of transaction lsn from log syncpt */ 462 lsn = tblk->lsn; 463 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log; 464 logdiff(difft, lsn, log); 465 466 /* 467 * update the block state a dmap at a time. 468 */ 469 mp = NULL; 470 lastlblkno = 0; 471 for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) { 472 /* get the buffer for the current dmap. */ 473 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 474 if (lblkno != lastlblkno) { 475 if (mp) { 476 write_metapage(mp); 477 } 478 479 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 480 0); 481 if (mp == NULL) 482 return -EIO; 483 metapage_wait_for_io(mp); 484 } 485 dp = (struct dmap *) mp->data; 486 487 /* determine the bit number and word within the dmap of 488 * the starting block. also determine how many blocks 489 * are to be updated within this dmap. 490 */ 491 dbitno = blkno & (BPERDMAP - 1); 492 word = dbitno >> L2DBWORD; 493 nblks = min(rem, (s64)BPERDMAP - dbitno); 494 495 /* update the bits of the dmap words. the first and last 496 * words may only have a subset of their bits updated. if 497 * this is the case, we'll work against that word (i.e. 498 * partial first and/or last) only in a single pass. a 499 * single pass will also be used to update all words that 500 * are to have all their bits updated. 501 */ 502 for (rbits = nblks; rbits > 0; 503 rbits -= nbits, dbitno += nbits) { 504 /* determine the bit number within the word and 505 * the number of bits within the word. 506 */ 507 wbitno = dbitno & (DBWORD - 1); 508 nbits = min(rbits, DBWORD - wbitno); 509 510 /* check if only part of the word is to be updated. */ 511 if (nbits < DBWORD) { 512 /* update (free or allocate) the bits 513 * in this word. 514 */ 515 mask = 516 (ONES << (DBWORD - nbits) >> wbitno); 517 if (free) 518 dp->pmap[word] &= 519 cpu_to_le32(~mask); 520 else 521 dp->pmap[word] |= 522 cpu_to_le32(mask); 523 524 word += 1; 525 } else { 526 /* one or more words are to have all 527 * their bits updated. determine how 528 * many words and how many bits. 529 */ 530 nwords = rbits >> L2DBWORD; 531 nbits = nwords << L2DBWORD; 532 533 /* update (free or allocate) the bits 534 * in these words. 535 */ 536 if (free) 537 memset(&dp->pmap[word], 0, 538 nwords * 4); 539 else 540 memset(&dp->pmap[word], (int) ONES, 541 nwords * 4); 542 543 word += nwords; 544 } 545 } 546 547 /* 548 * update dmap lsn 549 */ 550 if (lblkno == lastlblkno) 551 continue; 552 553 lastlblkno = lblkno; 554 555 LOGSYNC_LOCK(log, flags); 556 if (mp->lsn != 0) { 557 /* inherit older/smaller lsn */ 558 logdiff(diffp, mp->lsn, log); 559 if (difft < diffp) { 560 mp->lsn = lsn; 561 562 /* move bp after tblock in logsync list */ 563 list_move(&mp->synclist, &tblk->synclist); 564 } 565 566 /* inherit younger/larger clsn */ 567 logdiff(difft, tblk->clsn, log); 568 logdiff(diffp, mp->clsn, log); 569 if (difft > diffp) 570 mp->clsn = tblk->clsn; 571 } else { 572 mp->log = log; 573 mp->lsn = lsn; 574 575 /* insert bp after tblock in logsync list */ 576 log->count++; 577 list_add(&mp->synclist, &tblk->synclist); 578 579 mp->clsn = tblk->clsn; 580 } 581 LOGSYNC_UNLOCK(log, flags); 582 } 583 584 /* write the last buffer. */ 585 if (mp) { 586 write_metapage(mp); 587 } 588 589 return (0); 590 } 591 592 593 /* 594 * NAME: dbNextAG() 595 * 596 * FUNCTION: find the preferred allocation group for new allocations. 597 * 598 * Within the allocation groups, we maintain a preferred 599 * allocation group which consists of a group with at least 600 * average free space. It is the preferred group that we target 601 * new inode allocation towards. The tie-in between inode 602 * allocation and block allocation occurs as we allocate the 603 * first (data) block of an inode and specify the inode (block) 604 * as the allocation hint for this block. 605 * 606 * We try to avoid having more than one open file growing in 607 * an allocation group, as this will lead to fragmentation. 608 * This differs from the old OS/2 method of trying to keep 609 * empty ags around for large allocations. 610 * 611 * PARAMETERS: 612 * ipbmap - pointer to in-core inode for the block map. 613 * 614 * RETURN VALUES: 615 * the preferred allocation group number. 616 */ 617 int dbNextAG(struct inode *ipbmap) 618 { 619 s64 avgfree; 620 int agpref; 621 s64 hwm = 0; 622 int i; 623 int next_best = -1; 624 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 625 626 BMAP_LOCK(bmp); 627 628 /* determine the average number of free blocks within the ags. */ 629 avgfree = (u32)bmp->db_nfree / bmp->db_numag; 630 631 /* 632 * if the current preferred ag does not have an active allocator 633 * and has at least average freespace, return it 634 */ 635 agpref = bmp->db_agpref; 636 if ((atomic_read(&bmp->db_active[agpref]) == 0) && 637 (bmp->db_agfree[agpref] >= avgfree)) 638 goto unlock; 639 640 /* From the last preferred ag, find the next one with at least 641 * average free space. 642 */ 643 for (i = 0 ; i < bmp->db_numag; i++, agpref++) { 644 if (agpref >= bmp->db_numag) 645 agpref = 0; 646 647 if (atomic_read(&bmp->db_active[agpref])) 648 /* open file is currently growing in this ag */ 649 continue; 650 if (bmp->db_agfree[agpref] >= avgfree) { 651 /* Return this one */ 652 bmp->db_agpref = agpref; 653 goto unlock; 654 } else if (bmp->db_agfree[agpref] > hwm) { 655 /* Less than avg. freespace, but best so far */ 656 hwm = bmp->db_agfree[agpref]; 657 next_best = agpref; 658 } 659 } 660 661 /* 662 * If no inactive ag was found with average freespace, use the 663 * next best 664 */ 665 if (next_best != -1) 666 bmp->db_agpref = next_best; 667 /* else leave db_agpref unchanged */ 668 unlock: 669 BMAP_UNLOCK(bmp); 670 671 /* return the preferred group. 672 */ 673 return (bmp->db_agpref); 674 } 675 676 /* 677 * NAME: dbAlloc() 678 * 679 * FUNCTION: attempt to allocate a specified number of contiguous free 680 * blocks from the working allocation block map. 681 * 682 * the block allocation policy uses hints and a multi-step 683 * approach. 684 * 685 * for allocation requests smaller than the number of blocks 686 * per dmap, we first try to allocate the new blocks 687 * immediately following the hint. if these blocks are not 688 * available, we try to allocate blocks near the hint. if 689 * no blocks near the hint are available, we next try to 690 * allocate within the same dmap as contains the hint. 691 * 692 * if no blocks are available in the dmap or the allocation 693 * request is larger than the dmap size, we try to allocate 694 * within the same allocation group as contains the hint. if 695 * this does not succeed, we finally try to allocate anywhere 696 * within the aggregate. 697 * 698 * we also try to allocate anywhere within the aggregate 699 * for allocation requests larger than the allocation group 700 * size or requests that specify no hint value. 701 * 702 * PARAMETERS: 703 * ip - pointer to in-core inode; 704 * hint - allocation hint. 705 * nblocks - number of contiguous blocks in the range. 706 * results - on successful return, set to the starting block number 707 * of the newly allocated contiguous range. 708 * 709 * RETURN VALUES: 710 * 0 - success 711 * -ENOSPC - insufficient disk resources 712 * -EIO - i/o error 713 */ 714 int dbAlloc(struct inode *ip, s64 hint, s64 nblocks, s64 * results) 715 { 716 int rc, agno; 717 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 718 struct bmap *bmp; 719 struct metapage *mp; 720 s64 lblkno, blkno; 721 struct dmap *dp; 722 int l2nb; 723 s64 mapSize; 724 int writers; 725 726 /* assert that nblocks is valid */ 727 assert(nblocks > 0); 728 729 /* get the log2 number of blocks to be allocated. 730 * if the number of blocks is not a log2 multiple, 731 * it will be rounded up to the next log2 multiple. 732 */ 733 l2nb = BLKSTOL2(nblocks); 734 735 bmp = JFS_SBI(ip->i_sb)->bmap; 736 737 mapSize = bmp->db_mapsize; 738 739 /* the hint should be within the map */ 740 if (hint >= mapSize) { 741 jfs_error(ip->i_sb, "the hint is outside the map\n"); 742 return -EIO; 743 } 744 745 /* if the number of blocks to be allocated is greater than the 746 * allocation group size, try to allocate anywhere. 747 */ 748 if (l2nb > bmp->db_agl2size) { 749 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP); 750 751 rc = dbAllocAny(bmp, nblocks, l2nb, results); 752 753 goto write_unlock; 754 } 755 756 /* 757 * If no hint, let dbNextAG recommend an allocation group 758 */ 759 if (hint == 0) 760 goto pref_ag; 761 762 /* we would like to allocate close to the hint. adjust the 763 * hint to the block following the hint since the allocators 764 * will start looking for free space starting at this point. 765 */ 766 blkno = hint + 1; 767 768 if (blkno >= bmp->db_mapsize) 769 goto pref_ag; 770 771 agno = blkno >> bmp->db_agl2size; 772 773 /* check if blkno crosses over into a new allocation group. 774 * if so, check if we should allow allocations within this 775 * allocation group. 776 */ 777 if ((blkno & (bmp->db_agsize - 1)) == 0) 778 /* check if the AG is currently being written to. 779 * if so, call dbNextAG() to find a non-busy 780 * AG with sufficient free space. 781 */ 782 if (atomic_read(&bmp->db_active[agno])) 783 goto pref_ag; 784 785 /* check if the allocation request size can be satisfied from a 786 * single dmap. if so, try to allocate from the dmap containing 787 * the hint using a tiered strategy. 788 */ 789 if (nblocks <= BPERDMAP) { 790 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP); 791 792 /* get the buffer for the dmap containing the hint. 793 */ 794 rc = -EIO; 795 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 796 mp = read_metapage(ipbmap, lblkno, PSIZE, 0); 797 if (mp == NULL) 798 goto read_unlock; 799 800 dp = (struct dmap *) mp->data; 801 802 /* first, try to satisfy the allocation request with the 803 * blocks beginning at the hint. 804 */ 805 if ((rc = dbAllocNext(bmp, dp, blkno, (int) nblocks)) 806 != -ENOSPC) { 807 if (rc == 0) { 808 *results = blkno; 809 mark_metapage_dirty(mp); 810 } 811 812 release_metapage(mp); 813 goto read_unlock; 814 } 815 816 writers = atomic_read(&bmp->db_active[agno]); 817 if ((writers > 1) || 818 ((writers == 1) && (JFS_IP(ip)->active_ag != agno))) { 819 /* 820 * Someone else is writing in this allocation 821 * group. To avoid fragmenting, try another ag 822 */ 823 release_metapage(mp); 824 IREAD_UNLOCK(ipbmap); 825 goto pref_ag; 826 } 827 828 /* next, try to satisfy the allocation request with blocks 829 * near the hint. 830 */ 831 if ((rc = 832 dbAllocNear(bmp, dp, blkno, (int) nblocks, l2nb, results)) 833 != -ENOSPC) { 834 if (rc == 0) 835 mark_metapage_dirty(mp); 836 837 release_metapage(mp); 838 goto read_unlock; 839 } 840 841 /* try to satisfy the allocation request with blocks within 842 * the same dmap as the hint. 843 */ 844 if ((rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results)) 845 != -ENOSPC) { 846 if (rc == 0) 847 mark_metapage_dirty(mp); 848 849 release_metapage(mp); 850 goto read_unlock; 851 } 852 853 release_metapage(mp); 854 IREAD_UNLOCK(ipbmap); 855 } 856 857 /* try to satisfy the allocation request with blocks within 858 * the same allocation group as the hint. 859 */ 860 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP); 861 if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) != -ENOSPC) 862 goto write_unlock; 863 864 IWRITE_UNLOCK(ipbmap); 865 866 867 pref_ag: 868 /* 869 * Let dbNextAG recommend a preferred allocation group 870 */ 871 agno = dbNextAG(ipbmap); 872 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP); 873 874 /* Try to allocate within this allocation group. if that fails, try to 875 * allocate anywhere in the map. 876 */ 877 if ((rc = dbAllocAG(bmp, agno, nblocks, l2nb, results)) == -ENOSPC) 878 rc = dbAllocAny(bmp, nblocks, l2nb, results); 879 880 write_unlock: 881 IWRITE_UNLOCK(ipbmap); 882 883 return (rc); 884 885 read_unlock: 886 IREAD_UNLOCK(ipbmap); 887 888 return (rc); 889 } 890 891 /* 892 * NAME: dbReAlloc() 893 * 894 * FUNCTION: attempt to extend a current allocation by a specified 895 * number of blocks. 896 * 897 * this routine attempts to satisfy the allocation request 898 * by first trying to extend the existing allocation in 899 * place by allocating the additional blocks as the blocks 900 * immediately following the current allocation. if these 901 * blocks are not available, this routine will attempt to 902 * allocate a new set of contiguous blocks large enough 903 * to cover the existing allocation plus the additional 904 * number of blocks required. 905 * 906 * PARAMETERS: 907 * ip - pointer to in-core inode requiring allocation. 908 * blkno - starting block of the current allocation. 909 * nblocks - number of contiguous blocks within the current 910 * allocation. 911 * addnblocks - number of blocks to add to the allocation. 912 * results - on successful return, set to the starting block number 913 * of the existing allocation if the existing allocation 914 * was extended in place or to a newly allocated contiguous 915 * range if the existing allocation could not be extended 916 * in place. 917 * 918 * RETURN VALUES: 919 * 0 - success 920 * -ENOSPC - insufficient disk resources 921 * -EIO - i/o error 922 */ 923 int 924 dbReAlloc(struct inode *ip, 925 s64 blkno, s64 nblocks, s64 addnblocks, s64 * results) 926 { 927 int rc; 928 929 /* try to extend the allocation in place. 930 */ 931 if ((rc = dbExtend(ip, blkno, nblocks, addnblocks)) == 0) { 932 *results = blkno; 933 return (0); 934 } else { 935 if (rc != -ENOSPC) 936 return (rc); 937 } 938 939 /* could not extend the allocation in place, so allocate a 940 * new set of blocks for the entire request (i.e. try to get 941 * a range of contiguous blocks large enough to cover the 942 * existing allocation plus the additional blocks.) 943 */ 944 return (dbAlloc 945 (ip, blkno + nblocks - 1, addnblocks + nblocks, results)); 946 } 947 948 949 /* 950 * NAME: dbExtend() 951 * 952 * FUNCTION: attempt to extend a current allocation by a specified 953 * number of blocks. 954 * 955 * this routine attempts to satisfy the allocation request 956 * by first trying to extend the existing allocation in 957 * place by allocating the additional blocks as the blocks 958 * immediately following the current allocation. 959 * 960 * PARAMETERS: 961 * ip - pointer to in-core inode requiring allocation. 962 * blkno - starting block of the current allocation. 963 * nblocks - number of contiguous blocks within the current 964 * allocation. 965 * addnblocks - number of blocks to add to the allocation. 966 * 967 * RETURN VALUES: 968 * 0 - success 969 * -ENOSPC - insufficient disk resources 970 * -EIO - i/o error 971 */ 972 static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks) 973 { 974 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 975 s64 lblkno, lastblkno, extblkno; 976 uint rel_block; 977 struct metapage *mp; 978 struct dmap *dp; 979 int rc; 980 struct inode *ipbmap = sbi->ipbmap; 981 struct bmap *bmp; 982 983 /* 984 * We don't want a non-aligned extent to cross a page boundary 985 */ 986 if (((rel_block = blkno & (sbi->nbperpage - 1))) && 987 (rel_block + nblocks + addnblocks > sbi->nbperpage)) 988 return -ENOSPC; 989 990 /* get the last block of the current allocation */ 991 lastblkno = blkno + nblocks - 1; 992 993 /* determine the block number of the block following 994 * the existing allocation. 995 */ 996 extblkno = lastblkno + 1; 997 998 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP); 999 1000 /* better be within the file system */ 1001 bmp = sbi->bmap; 1002 if (lastblkno < 0 || lastblkno >= bmp->db_mapsize) { 1003 IREAD_UNLOCK(ipbmap); 1004 jfs_error(ip->i_sb, "the block is outside the filesystem\n"); 1005 return -EIO; 1006 } 1007 1008 /* we'll attempt to extend the current allocation in place by 1009 * allocating the additional blocks as the blocks immediately 1010 * following the current allocation. we only try to extend the 1011 * current allocation in place if the number of additional blocks 1012 * can fit into a dmap, the last block of the current allocation 1013 * is not the last block of the file system, and the start of the 1014 * inplace extension is not on an allocation group boundary. 1015 */ 1016 if (addnblocks > BPERDMAP || extblkno >= bmp->db_mapsize || 1017 (extblkno & (bmp->db_agsize - 1)) == 0) { 1018 IREAD_UNLOCK(ipbmap); 1019 return -ENOSPC; 1020 } 1021 1022 /* get the buffer for the dmap containing the first block 1023 * of the extension. 1024 */ 1025 lblkno = BLKTODMAP(extblkno, bmp->db_l2nbperpage); 1026 mp = read_metapage(ipbmap, lblkno, PSIZE, 0); 1027 if (mp == NULL) { 1028 IREAD_UNLOCK(ipbmap); 1029 return -EIO; 1030 } 1031 1032 dp = (struct dmap *) mp->data; 1033 1034 /* try to allocate the blocks immediately following the 1035 * current allocation. 1036 */ 1037 rc = dbAllocNext(bmp, dp, extblkno, (int) addnblocks); 1038 1039 IREAD_UNLOCK(ipbmap); 1040 1041 /* were we successful ? */ 1042 if (rc == 0) 1043 write_metapage(mp); 1044 else 1045 /* we were not successful */ 1046 release_metapage(mp); 1047 1048 return (rc); 1049 } 1050 1051 1052 /* 1053 * NAME: dbAllocNext() 1054 * 1055 * FUNCTION: attempt to allocate the blocks of the specified block 1056 * range within a dmap. 1057 * 1058 * PARAMETERS: 1059 * bmp - pointer to bmap descriptor 1060 * dp - pointer to dmap. 1061 * blkno - starting block number of the range. 1062 * nblocks - number of contiguous free blocks of the range. 1063 * 1064 * RETURN VALUES: 1065 * 0 - success 1066 * -ENOSPC - insufficient disk resources 1067 * -EIO - i/o error 1068 * 1069 * serialization: IREAD_LOCK(ipbmap) held on entry/exit; 1070 */ 1071 static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno, 1072 int nblocks) 1073 { 1074 int dbitno, word, rembits, nb, nwords, wbitno, nw; 1075 int l2size; 1076 s8 *leaf; 1077 u32 mask; 1078 1079 if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) { 1080 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n"); 1081 return -EIO; 1082 } 1083 1084 /* pick up a pointer to the leaves of the dmap tree. 1085 */ 1086 leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx); 1087 1088 /* determine the bit number and word within the dmap of the 1089 * starting block. 1090 */ 1091 dbitno = blkno & (BPERDMAP - 1); 1092 word = dbitno >> L2DBWORD; 1093 1094 /* check if the specified block range is contained within 1095 * this dmap. 1096 */ 1097 if (dbitno + nblocks > BPERDMAP) 1098 return -ENOSPC; 1099 1100 /* check if the starting leaf indicates that anything 1101 * is free. 1102 */ 1103 if (leaf[word] == NOFREE) 1104 return -ENOSPC; 1105 1106 /* check the dmaps words corresponding to block range to see 1107 * if the block range is free. not all bits of the first and 1108 * last words may be contained within the block range. if this 1109 * is the case, we'll work against those words (i.e. partial first 1110 * and/or last) on an individual basis (a single pass) and examine 1111 * the actual bits to determine if they are free. a single pass 1112 * will be used for all dmap words fully contained within the 1113 * specified range. within this pass, the leaves of the dmap 1114 * tree will be examined to determine if the blocks are free. a 1115 * single leaf may describe the free space of multiple dmap 1116 * words, so we may visit only a subset of the actual leaves 1117 * corresponding to the dmap words of the block range. 1118 */ 1119 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) { 1120 /* determine the bit number within the word and 1121 * the number of bits within the word. 1122 */ 1123 wbitno = dbitno & (DBWORD - 1); 1124 nb = min(rembits, DBWORD - wbitno); 1125 1126 /* check if only part of the word is to be examined. 1127 */ 1128 if (nb < DBWORD) { 1129 /* check if the bits are free. 1130 */ 1131 mask = (ONES << (DBWORD - nb) >> wbitno); 1132 if ((mask & ~le32_to_cpu(dp->wmap[word])) != mask) 1133 return -ENOSPC; 1134 1135 word += 1; 1136 } else { 1137 /* one or more dmap words are fully contained 1138 * within the block range. determine how many 1139 * words and how many bits. 1140 */ 1141 nwords = rembits >> L2DBWORD; 1142 nb = nwords << L2DBWORD; 1143 1144 /* now examine the appropriate leaves to determine 1145 * if the blocks are free. 1146 */ 1147 while (nwords > 0) { 1148 /* does the leaf describe any free space ? 1149 */ 1150 if (leaf[word] < BUDMIN) 1151 return -ENOSPC; 1152 1153 /* determine the l2 number of bits provided 1154 * by this leaf. 1155 */ 1156 l2size = 1157 min_t(int, leaf[word], NLSTOL2BSZ(nwords)); 1158 1159 /* determine how many words were handled. 1160 */ 1161 nw = BUDSIZE(l2size, BUDMIN); 1162 1163 nwords -= nw; 1164 word += nw; 1165 } 1166 } 1167 } 1168 1169 /* allocate the blocks. 1170 */ 1171 return (dbAllocDmap(bmp, dp, blkno, nblocks)); 1172 } 1173 1174 1175 /* 1176 * NAME: dbAllocNear() 1177 * 1178 * FUNCTION: attempt to allocate a number of contiguous free blocks near 1179 * a specified block (hint) within a dmap. 1180 * 1181 * starting with the dmap leaf that covers the hint, we'll 1182 * check the next four contiguous leaves for sufficient free 1183 * space. if sufficient free space is found, we'll allocate 1184 * the desired free space. 1185 * 1186 * PARAMETERS: 1187 * bmp - pointer to bmap descriptor 1188 * dp - pointer to dmap. 1189 * blkno - block number to allocate near. 1190 * nblocks - actual number of contiguous free blocks desired. 1191 * l2nb - log2 number of contiguous free blocks desired. 1192 * results - on successful return, set to the starting block number 1193 * of the newly allocated range. 1194 * 1195 * RETURN VALUES: 1196 * 0 - success 1197 * -ENOSPC - insufficient disk resources 1198 * -EIO - i/o error 1199 * 1200 * serialization: IREAD_LOCK(ipbmap) held on entry/exit; 1201 */ 1202 static int 1203 dbAllocNear(struct bmap * bmp, 1204 struct dmap * dp, s64 blkno, int nblocks, int l2nb, s64 * results) 1205 { 1206 int word, lword, rc; 1207 s8 *leaf; 1208 1209 if (dp->tree.leafidx != cpu_to_le32(LEAFIND)) { 1210 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmap page\n"); 1211 return -EIO; 1212 } 1213 1214 leaf = dp->tree.stree + le32_to_cpu(dp->tree.leafidx); 1215 1216 /* determine the word within the dmap that holds the hint 1217 * (i.e. blkno). also, determine the last word in the dmap 1218 * that we'll include in our examination. 1219 */ 1220 word = (blkno & (BPERDMAP - 1)) >> L2DBWORD; 1221 lword = min(word + 4, LPERDMAP); 1222 1223 /* examine the leaves for sufficient free space. 1224 */ 1225 for (; word < lword; word++) { 1226 /* does the leaf describe sufficient free space ? 1227 */ 1228 if (leaf[word] < l2nb) 1229 continue; 1230 1231 /* determine the block number within the file system 1232 * of the first block described by this dmap word. 1233 */ 1234 blkno = le64_to_cpu(dp->start) + (word << L2DBWORD); 1235 1236 /* if not all bits of the dmap word are free, get the 1237 * starting bit number within the dmap word of the required 1238 * string of free bits and adjust the block number with the 1239 * value. 1240 */ 1241 if (leaf[word] < BUDMIN) 1242 blkno += 1243 dbFindBits(le32_to_cpu(dp->wmap[word]), l2nb); 1244 1245 /* allocate the blocks. 1246 */ 1247 if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0) 1248 *results = blkno; 1249 1250 return (rc); 1251 } 1252 1253 return -ENOSPC; 1254 } 1255 1256 1257 /* 1258 * NAME: dbAllocAG() 1259 * 1260 * FUNCTION: attempt to allocate the specified number of contiguous 1261 * free blocks within the specified allocation group. 1262 * 1263 * unless the allocation group size is equal to the number 1264 * of blocks per dmap, the dmap control pages will be used to 1265 * find the required free space, if available. we start the 1266 * search at the highest dmap control page level which 1267 * distinctly describes the allocation group's free space 1268 * (i.e. the highest level at which the allocation group's 1269 * free space is not mixed in with that of any other group). 1270 * in addition, we start the search within this level at a 1271 * height of the dmapctl dmtree at which the nodes distinctly 1272 * describe the allocation group's free space. at this height, 1273 * the allocation group's free space may be represented by 1 1274 * or two sub-trees, depending on the allocation group size. 1275 * we search the top nodes of these subtrees left to right for 1276 * sufficient free space. if sufficient free space is found, 1277 * the subtree is searched to find the leftmost leaf that 1278 * has free space. once we have made it to the leaf, we 1279 * move the search to the next lower level dmap control page 1280 * corresponding to this leaf. we continue down the dmap control 1281 * pages until we find the dmap that contains or starts the 1282 * sufficient free space and we allocate at this dmap. 1283 * 1284 * if the allocation group size is equal to the dmap size, 1285 * we'll start at the dmap corresponding to the allocation 1286 * group and attempt the allocation at this level. 1287 * 1288 * the dmap control page search is also not performed if the 1289 * allocation group is completely free and we go to the first 1290 * dmap of the allocation group to do the allocation. this is 1291 * done because the allocation group may be part (not the first 1292 * part) of a larger binary buddy system, causing the dmap 1293 * control pages to indicate no free space (NOFREE) within 1294 * the allocation group. 1295 * 1296 * PARAMETERS: 1297 * bmp - pointer to bmap descriptor 1298 * agno - allocation group number. 1299 * nblocks - actual number of contiguous free blocks desired. 1300 * l2nb - log2 number of contiguous free blocks desired. 1301 * results - on successful return, set to the starting block number 1302 * of the newly allocated range. 1303 * 1304 * RETURN VALUES: 1305 * 0 - success 1306 * -ENOSPC - insufficient disk resources 1307 * -EIO - i/o error 1308 * 1309 * note: IWRITE_LOCK(ipmap) held on entry/exit; 1310 */ 1311 static int 1312 dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results) 1313 { 1314 struct metapage *mp; 1315 struct dmapctl *dcp; 1316 int rc, ti, i, k, m, n, agperlev; 1317 s64 blkno, lblkno; 1318 int budmin; 1319 1320 /* allocation request should not be for more than the 1321 * allocation group size. 1322 */ 1323 if (l2nb > bmp->db_agl2size) { 1324 jfs_error(bmp->db_ipbmap->i_sb, 1325 "allocation request is larger than the allocation group size\n"); 1326 return -EIO; 1327 } 1328 1329 /* determine the starting block number of the allocation 1330 * group. 1331 */ 1332 blkno = (s64) agno << bmp->db_agl2size; 1333 1334 /* check if the allocation group size is the minimum allocation 1335 * group size or if the allocation group is completely free. if 1336 * the allocation group size is the minimum size of BPERDMAP (i.e. 1337 * 1 dmap), there is no need to search the dmap control page (below) 1338 * that fully describes the allocation group since the allocation 1339 * group is already fully described by a dmap. in this case, we 1340 * just call dbAllocCtl() to search the dmap tree and allocate the 1341 * required space if available. 1342 * 1343 * if the allocation group is completely free, dbAllocCtl() is 1344 * also called to allocate the required space. this is done for 1345 * two reasons. first, it makes no sense searching the dmap control 1346 * pages for free space when we know that free space exists. second, 1347 * the dmap control pages may indicate that the allocation group 1348 * has no free space if the allocation group is part (not the first 1349 * part) of a larger binary buddy system. 1350 */ 1351 if (bmp->db_agsize == BPERDMAP 1352 || bmp->db_agfree[agno] == bmp->db_agsize) { 1353 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results); 1354 if ((rc == -ENOSPC) && 1355 (bmp->db_agfree[agno] == bmp->db_agsize)) { 1356 printk(KERN_ERR "blkno = %Lx, blocks = %Lx\n", 1357 (unsigned long long) blkno, 1358 (unsigned long long) nblocks); 1359 jfs_error(bmp->db_ipbmap->i_sb, 1360 "dbAllocCtl failed in free AG\n"); 1361 } 1362 return (rc); 1363 } 1364 1365 /* the buffer for the dmap control page that fully describes the 1366 * allocation group. 1367 */ 1368 lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, bmp->db_aglevel); 1369 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1370 if (mp == NULL) 1371 return -EIO; 1372 dcp = (struct dmapctl *) mp->data; 1373 budmin = dcp->budmin; 1374 1375 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) { 1376 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n"); 1377 release_metapage(mp); 1378 return -EIO; 1379 } 1380 1381 /* search the subtree(s) of the dmap control page that describes 1382 * the allocation group, looking for sufficient free space. to begin, 1383 * determine how many allocation groups are represented in a dmap 1384 * control page at the control page level (i.e. L0, L1, L2) that 1385 * fully describes an allocation group. next, determine the starting 1386 * tree index of this allocation group within the control page. 1387 */ 1388 agperlev = 1389 (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / bmp->db_agwidth; 1390 ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1)); 1391 1392 if (ti < 0 || ti >= le32_to_cpu(dcp->nleafs)) { 1393 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n"); 1394 release_metapage(mp); 1395 return -EIO; 1396 } 1397 1398 /* dmap control page trees fan-out by 4 and a single allocation 1399 * group may be described by 1 or 2 subtrees within the ag level 1400 * dmap control page, depending upon the ag size. examine the ag's 1401 * subtrees for sufficient free space, starting with the leftmost 1402 * subtree. 1403 */ 1404 for (i = 0; i < bmp->db_agwidth; i++, ti++) { 1405 /* is there sufficient free space ? 1406 */ 1407 if (l2nb > dcp->stree[ti]) 1408 continue; 1409 1410 /* sufficient free space found in a subtree. now search down 1411 * the subtree to find the leftmost leaf that describes this 1412 * free space. 1413 */ 1414 for (k = bmp->db_agheight; k > 0; k--) { 1415 for (n = 0, m = (ti << 2) + 1; n < 4; n++) { 1416 if (l2nb <= dcp->stree[m + n]) { 1417 ti = m + n; 1418 break; 1419 } 1420 } 1421 if (n == 4) { 1422 jfs_error(bmp->db_ipbmap->i_sb, 1423 "failed descending stree\n"); 1424 release_metapage(mp); 1425 return -EIO; 1426 } 1427 } 1428 1429 /* determine the block number within the file system 1430 * that corresponds to this leaf. 1431 */ 1432 if (bmp->db_aglevel == 2) 1433 blkno = 0; 1434 else if (bmp->db_aglevel == 1) 1435 blkno &= ~(MAXL1SIZE - 1); 1436 else /* bmp->db_aglevel == 0 */ 1437 blkno &= ~(MAXL0SIZE - 1); 1438 1439 blkno += 1440 ((s64) (ti - le32_to_cpu(dcp->leafidx))) << budmin; 1441 1442 /* release the buffer in preparation for going down 1443 * the next level of dmap control pages. 1444 */ 1445 release_metapage(mp); 1446 1447 /* check if we need to continue to search down the lower 1448 * level dmap control pages. we need to if the number of 1449 * blocks required is less than maximum number of blocks 1450 * described at the next lower level. 1451 */ 1452 if (l2nb < budmin) { 1453 1454 /* search the lower level dmap control pages to get 1455 * the starting block number of the dmap that 1456 * contains or starts off the free space. 1457 */ 1458 if ((rc = 1459 dbFindCtl(bmp, l2nb, bmp->db_aglevel - 1, 1460 &blkno))) { 1461 if (rc == -ENOSPC) { 1462 jfs_error(bmp->db_ipbmap->i_sb, 1463 "control page inconsistent\n"); 1464 return -EIO; 1465 } 1466 return (rc); 1467 } 1468 } 1469 1470 /* allocate the blocks. 1471 */ 1472 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results); 1473 if (rc == -ENOSPC) { 1474 jfs_error(bmp->db_ipbmap->i_sb, 1475 "unable to allocate blocks\n"); 1476 rc = -EIO; 1477 } 1478 return (rc); 1479 } 1480 1481 /* no space in the allocation group. release the buffer and 1482 * return -ENOSPC. 1483 */ 1484 release_metapage(mp); 1485 1486 return -ENOSPC; 1487 } 1488 1489 1490 /* 1491 * NAME: dbAllocAny() 1492 * 1493 * FUNCTION: attempt to allocate the specified number of contiguous 1494 * free blocks anywhere in the file system. 1495 * 1496 * dbAllocAny() attempts to find the sufficient free space by 1497 * searching down the dmap control pages, starting with the 1498 * highest level (i.e. L0, L1, L2) control page. if free space 1499 * large enough to satisfy the desired free space is found, the 1500 * desired free space is allocated. 1501 * 1502 * PARAMETERS: 1503 * bmp - pointer to bmap descriptor 1504 * nblocks - actual number of contiguous free blocks desired. 1505 * l2nb - log2 number of contiguous free blocks desired. 1506 * results - on successful return, set to the starting block number 1507 * of the newly allocated range. 1508 * 1509 * RETURN VALUES: 1510 * 0 - success 1511 * -ENOSPC - insufficient disk resources 1512 * -EIO - i/o error 1513 * 1514 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit; 1515 */ 1516 static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results) 1517 { 1518 int rc; 1519 s64 blkno = 0; 1520 1521 /* starting with the top level dmap control page, search 1522 * down the dmap control levels for sufficient free space. 1523 * if free space is found, dbFindCtl() returns the starting 1524 * block number of the dmap that contains or starts off the 1525 * range of free space. 1526 */ 1527 if ((rc = dbFindCtl(bmp, l2nb, bmp->db_maxlevel, &blkno))) 1528 return (rc); 1529 1530 /* allocate the blocks. 1531 */ 1532 rc = dbAllocCtl(bmp, nblocks, l2nb, blkno, results); 1533 if (rc == -ENOSPC) { 1534 jfs_error(bmp->db_ipbmap->i_sb, "unable to allocate blocks\n"); 1535 return -EIO; 1536 } 1537 return (rc); 1538 } 1539 1540 1541 /* 1542 * NAME: dbDiscardAG() 1543 * 1544 * FUNCTION: attempt to discard (TRIM) all free blocks of specific AG 1545 * 1546 * algorithm: 1547 * 1) allocate blocks, as large as possible and save them 1548 * while holding IWRITE_LOCK on ipbmap 1549 * 2) trim all these saved block/length values 1550 * 3) mark the blocks free again 1551 * 1552 * benefit: 1553 * - we work only on one ag at some time, minimizing how long we 1554 * need to lock ipbmap 1555 * - reading / writing the fs is possible most time, even on 1556 * trimming 1557 * 1558 * downside: 1559 * - we write two times to the dmapctl and dmap pages 1560 * - but for me, this seems the best way, better ideas? 1561 * /TR 2012 1562 * 1563 * PARAMETERS: 1564 * ip - pointer to in-core inode 1565 * agno - ag to trim 1566 * minlen - minimum value of contiguous blocks 1567 * 1568 * RETURN VALUES: 1569 * s64 - actual number of blocks trimmed 1570 */ 1571 s64 dbDiscardAG(struct inode *ip, int agno, s64 minlen) 1572 { 1573 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 1574 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap; 1575 s64 nblocks, blkno; 1576 u64 trimmed = 0; 1577 int rc, l2nb; 1578 struct super_block *sb = ipbmap->i_sb; 1579 1580 struct range2trim { 1581 u64 blkno; 1582 u64 nblocks; 1583 } *totrim, *tt; 1584 1585 /* max blkno / nblocks pairs to trim */ 1586 int count = 0, range_cnt; 1587 u64 max_ranges; 1588 1589 /* prevent others from writing new stuff here, while trimming */ 1590 IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP); 1591 1592 nblocks = bmp->db_agfree[agno]; 1593 max_ranges = nblocks; 1594 do_div(max_ranges, minlen); 1595 range_cnt = min_t(u64, max_ranges + 1, 32 * 1024); 1596 totrim = kmalloc_array(range_cnt, sizeof(struct range2trim), GFP_NOFS); 1597 if (totrim == NULL) { 1598 jfs_error(bmp->db_ipbmap->i_sb, "no memory for trim array\n"); 1599 IWRITE_UNLOCK(ipbmap); 1600 return 0; 1601 } 1602 1603 tt = totrim; 1604 while (nblocks >= minlen) { 1605 l2nb = BLKSTOL2(nblocks); 1606 1607 /* 0 = okay, -EIO = fatal, -ENOSPC -> try smaller block */ 1608 rc = dbAllocAG(bmp, agno, nblocks, l2nb, &blkno); 1609 if (rc == 0) { 1610 tt->blkno = blkno; 1611 tt->nblocks = nblocks; 1612 tt++; count++; 1613 1614 /* the whole ag is free, trim now */ 1615 if (bmp->db_agfree[agno] == 0) 1616 break; 1617 1618 /* give a hint for the next while */ 1619 nblocks = bmp->db_agfree[agno]; 1620 continue; 1621 } else if (rc == -ENOSPC) { 1622 /* search for next smaller log2 block */ 1623 l2nb = BLKSTOL2(nblocks) - 1; 1624 if (unlikely(l2nb < 0)) 1625 break; 1626 nblocks = 1LL << l2nb; 1627 } else { 1628 /* Trim any already allocated blocks */ 1629 jfs_error(bmp->db_ipbmap->i_sb, "-EIO\n"); 1630 break; 1631 } 1632 1633 /* check, if our trim array is full */ 1634 if (unlikely(count >= range_cnt - 1)) 1635 break; 1636 } 1637 IWRITE_UNLOCK(ipbmap); 1638 1639 tt->nblocks = 0; /* mark the current end */ 1640 for (tt = totrim; tt->nblocks != 0; tt++) { 1641 /* when mounted with online discard, dbFree() will 1642 * call jfs_issue_discard() itself */ 1643 if (!(JFS_SBI(sb)->flag & JFS_DISCARD)) 1644 jfs_issue_discard(ip, tt->blkno, tt->nblocks); 1645 dbFree(ip, tt->blkno, tt->nblocks); 1646 trimmed += tt->nblocks; 1647 } 1648 kfree(totrim); 1649 1650 return trimmed; 1651 } 1652 1653 /* 1654 * NAME: dbFindCtl() 1655 * 1656 * FUNCTION: starting at a specified dmap control page level and block 1657 * number, search down the dmap control levels for a range of 1658 * contiguous free blocks large enough to satisfy an allocation 1659 * request for the specified number of free blocks. 1660 * 1661 * if sufficient contiguous free blocks are found, this routine 1662 * returns the starting block number within a dmap page that 1663 * contains or starts a range of contiqious free blocks that 1664 * is sufficient in size. 1665 * 1666 * PARAMETERS: 1667 * bmp - pointer to bmap descriptor 1668 * level - starting dmap control page level. 1669 * l2nb - log2 number of contiguous free blocks desired. 1670 * *blkno - on entry, starting block number for conducting the search. 1671 * on successful return, the first block within a dmap page 1672 * that contains or starts a range of contiguous free blocks. 1673 * 1674 * RETURN VALUES: 1675 * 0 - success 1676 * -ENOSPC - insufficient disk resources 1677 * -EIO - i/o error 1678 * 1679 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit; 1680 */ 1681 static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno) 1682 { 1683 int rc, leafidx, lev; 1684 s64 b, lblkno; 1685 struct dmapctl *dcp; 1686 int budmin; 1687 struct metapage *mp; 1688 1689 /* starting at the specified dmap control page level and block 1690 * number, search down the dmap control levels for the starting 1691 * block number of a dmap page that contains or starts off 1692 * sufficient free blocks. 1693 */ 1694 for (lev = level, b = *blkno; lev >= 0; lev--) { 1695 /* get the buffer of the dmap control page for the block 1696 * number and level (i.e. L0, L1, L2). 1697 */ 1698 lblkno = BLKTOCTL(b, bmp->db_l2nbperpage, lev); 1699 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1700 if (mp == NULL) 1701 return -EIO; 1702 dcp = (struct dmapctl *) mp->data; 1703 budmin = dcp->budmin; 1704 1705 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) { 1706 jfs_error(bmp->db_ipbmap->i_sb, 1707 "Corrupt dmapctl page\n"); 1708 release_metapage(mp); 1709 return -EIO; 1710 } 1711 1712 /* search the tree within the dmap control page for 1713 * sufficient free space. if sufficient free space is found, 1714 * dbFindLeaf() returns the index of the leaf at which 1715 * free space was found. 1716 */ 1717 rc = dbFindLeaf((dmtree_t *) dcp, l2nb, &leafidx, true); 1718 1719 /* release the buffer. 1720 */ 1721 release_metapage(mp); 1722 1723 /* space found ? 1724 */ 1725 if (rc) { 1726 if (lev != level) { 1727 jfs_error(bmp->db_ipbmap->i_sb, 1728 "dmap inconsistent\n"); 1729 return -EIO; 1730 } 1731 return -ENOSPC; 1732 } 1733 1734 /* adjust the block number to reflect the location within 1735 * the dmap control page (i.e. the leaf) at which free 1736 * space was found. 1737 */ 1738 b += (((s64) leafidx) << budmin); 1739 1740 /* we stop the search at this dmap control page level if 1741 * the number of blocks required is greater than or equal 1742 * to the maximum number of blocks described at the next 1743 * (lower) level. 1744 */ 1745 if (l2nb >= budmin) 1746 break; 1747 } 1748 1749 *blkno = b; 1750 return (0); 1751 } 1752 1753 1754 /* 1755 * NAME: dbAllocCtl() 1756 * 1757 * FUNCTION: attempt to allocate a specified number of contiguous 1758 * blocks starting within a specific dmap. 1759 * 1760 * this routine is called by higher level routines that search 1761 * the dmap control pages above the actual dmaps for contiguous 1762 * free space. the result of successful searches by these 1763 * routines are the starting block numbers within dmaps, with 1764 * the dmaps themselves containing the desired contiguous free 1765 * space or starting a contiguous free space of desired size 1766 * that is made up of the blocks of one or more dmaps. these 1767 * calls should not fail due to insufficent resources. 1768 * 1769 * this routine is called in some cases where it is not known 1770 * whether it will fail due to insufficient resources. more 1771 * specifically, this occurs when allocating from an allocation 1772 * group whose size is equal to the number of blocks per dmap. 1773 * in this case, the dmap control pages are not examined prior 1774 * to calling this routine (to save pathlength) and the call 1775 * might fail. 1776 * 1777 * for a request size that fits within a dmap, this routine relies 1778 * upon the dmap's dmtree to find the requested contiguous free 1779 * space. for request sizes that are larger than a dmap, the 1780 * requested free space will start at the first block of the 1781 * first dmap (i.e. blkno). 1782 * 1783 * PARAMETERS: 1784 * bmp - pointer to bmap descriptor 1785 * nblocks - actual number of contiguous free blocks to allocate. 1786 * l2nb - log2 number of contiguous free blocks to allocate. 1787 * blkno - starting block number of the dmap to start the allocation 1788 * from. 1789 * results - on successful return, set to the starting block number 1790 * of the newly allocated range. 1791 * 1792 * RETURN VALUES: 1793 * 0 - success 1794 * -ENOSPC - insufficient disk resources 1795 * -EIO - i/o error 1796 * 1797 * serialization: IWRITE_LOCK(ipbmap) held on entry/exit; 1798 */ 1799 static int 1800 dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno, s64 * results) 1801 { 1802 int rc, nb; 1803 s64 b, lblkno, n; 1804 struct metapage *mp; 1805 struct dmap *dp; 1806 1807 /* check if the allocation request is confined to a single dmap. 1808 */ 1809 if (l2nb <= L2BPERDMAP) { 1810 /* get the buffer for the dmap. 1811 */ 1812 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 1813 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1814 if (mp == NULL) 1815 return -EIO; 1816 dp = (struct dmap *) mp->data; 1817 1818 if (dp->tree.budmin < 0) { 1819 release_metapage(mp); 1820 return -EIO; 1821 } 1822 1823 /* try to allocate the blocks. 1824 */ 1825 rc = dbAllocDmapLev(bmp, dp, (int) nblocks, l2nb, results); 1826 if (rc == 0) 1827 mark_metapage_dirty(mp); 1828 1829 release_metapage(mp); 1830 1831 return (rc); 1832 } 1833 1834 /* allocation request involving multiple dmaps. it must start on 1835 * a dmap boundary. 1836 */ 1837 assert((blkno & (BPERDMAP - 1)) == 0); 1838 1839 /* allocate the blocks dmap by dmap. 1840 */ 1841 for (n = nblocks, b = blkno; n > 0; n -= nb, b += nb) { 1842 /* get the buffer for the dmap. 1843 */ 1844 lblkno = BLKTODMAP(b, bmp->db_l2nbperpage); 1845 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1846 if (mp == NULL) { 1847 rc = -EIO; 1848 goto backout; 1849 } 1850 dp = (struct dmap *) mp->data; 1851 1852 /* the dmap better be all free. 1853 */ 1854 if (dp->tree.stree[ROOT] != L2BPERDMAP) { 1855 release_metapage(mp); 1856 jfs_error(bmp->db_ipbmap->i_sb, 1857 "the dmap is not all free\n"); 1858 rc = -EIO; 1859 goto backout; 1860 } 1861 1862 /* determine how many blocks to allocate from this dmap. 1863 */ 1864 nb = min_t(s64, n, BPERDMAP); 1865 1866 /* allocate the blocks from the dmap. 1867 */ 1868 if ((rc = dbAllocDmap(bmp, dp, b, nb))) { 1869 release_metapage(mp); 1870 goto backout; 1871 } 1872 1873 /* write the buffer. 1874 */ 1875 write_metapage(mp); 1876 } 1877 1878 /* set the results (starting block number) and return. 1879 */ 1880 *results = blkno; 1881 return (0); 1882 1883 /* something failed in handling an allocation request involving 1884 * multiple dmaps. we'll try to clean up by backing out any 1885 * allocation that has already happened for this request. if 1886 * we fail in backing out the allocation, we'll mark the file 1887 * system to indicate that blocks have been leaked. 1888 */ 1889 backout: 1890 1891 /* try to backout the allocations dmap by dmap. 1892 */ 1893 for (n = nblocks - n, b = blkno; n > 0; 1894 n -= BPERDMAP, b += BPERDMAP) { 1895 /* get the buffer for this dmap. 1896 */ 1897 lblkno = BLKTODMAP(b, bmp->db_l2nbperpage); 1898 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 1899 if (mp == NULL) { 1900 /* could not back out. mark the file system 1901 * to indicate that we have leaked blocks. 1902 */ 1903 jfs_error(bmp->db_ipbmap->i_sb, 1904 "I/O Error: Block Leakage\n"); 1905 continue; 1906 } 1907 dp = (struct dmap *) mp->data; 1908 1909 /* free the blocks is this dmap. 1910 */ 1911 if (dbFreeDmap(bmp, dp, b, BPERDMAP)) { 1912 /* could not back out. mark the file system 1913 * to indicate that we have leaked blocks. 1914 */ 1915 release_metapage(mp); 1916 jfs_error(bmp->db_ipbmap->i_sb, "Block Leakage\n"); 1917 continue; 1918 } 1919 1920 /* write the buffer. 1921 */ 1922 write_metapage(mp); 1923 } 1924 1925 return (rc); 1926 } 1927 1928 1929 /* 1930 * NAME: dbAllocDmapLev() 1931 * 1932 * FUNCTION: attempt to allocate a specified number of contiguous blocks 1933 * from a specified dmap. 1934 * 1935 * this routine checks if the contiguous blocks are available. 1936 * if so, nblocks of blocks are allocated; otherwise, ENOSPC is 1937 * returned. 1938 * 1939 * PARAMETERS: 1940 * mp - pointer to bmap descriptor 1941 * dp - pointer to dmap to attempt to allocate blocks from. 1942 * l2nb - log2 number of contiguous block desired. 1943 * nblocks - actual number of contiguous block desired. 1944 * results - on successful return, set to the starting block number 1945 * of the newly allocated range. 1946 * 1947 * RETURN VALUES: 1948 * 0 - success 1949 * -ENOSPC - insufficient disk resources 1950 * -EIO - i/o error 1951 * 1952 * serialization: IREAD_LOCK(ipbmap), e.g., from dbAlloc(), or 1953 * IWRITE_LOCK(ipbmap), e.g., dbAllocCtl(), held on entry/exit; 1954 */ 1955 static int 1956 dbAllocDmapLev(struct bmap * bmp, 1957 struct dmap * dp, int nblocks, int l2nb, s64 * results) 1958 { 1959 s64 blkno; 1960 int leafidx, rc; 1961 1962 /* can't be more than a dmaps worth of blocks */ 1963 assert(l2nb <= L2BPERDMAP); 1964 1965 /* search the tree within the dmap page for sufficient 1966 * free space. if sufficient free space is found, dbFindLeaf() 1967 * returns the index of the leaf at which free space was found. 1968 */ 1969 if (dbFindLeaf((dmtree_t *) &dp->tree, l2nb, &leafidx, false)) 1970 return -ENOSPC; 1971 1972 if (leafidx < 0) 1973 return -EIO; 1974 1975 /* determine the block number within the file system corresponding 1976 * to the leaf at which free space was found. 1977 */ 1978 blkno = le64_to_cpu(dp->start) + (leafidx << L2DBWORD); 1979 1980 /* if not all bits of the dmap word are free, get the starting 1981 * bit number within the dmap word of the required string of free 1982 * bits and adjust the block number with this value. 1983 */ 1984 if (dp->tree.stree[leafidx + LEAFIND] < BUDMIN) 1985 blkno += dbFindBits(le32_to_cpu(dp->wmap[leafidx]), l2nb); 1986 1987 /* allocate the blocks */ 1988 if ((rc = dbAllocDmap(bmp, dp, blkno, nblocks)) == 0) 1989 *results = blkno; 1990 1991 return (rc); 1992 } 1993 1994 1995 /* 1996 * NAME: dbAllocDmap() 1997 * 1998 * FUNCTION: adjust the disk allocation map to reflect the allocation 1999 * of a specified block range within a dmap. 2000 * 2001 * this routine allocates the specified blocks from the dmap 2002 * through a call to dbAllocBits(). if the allocation of the 2003 * block range causes the maximum string of free blocks within 2004 * the dmap to change (i.e. the value of the root of the dmap's 2005 * dmtree), this routine will cause this change to be reflected 2006 * up through the appropriate levels of the dmap control pages 2007 * by a call to dbAdjCtl() for the L0 dmap control page that 2008 * covers this dmap. 2009 * 2010 * PARAMETERS: 2011 * bmp - pointer to bmap descriptor 2012 * dp - pointer to dmap to allocate the block range from. 2013 * blkno - starting block number of the block to be allocated. 2014 * nblocks - number of blocks to be allocated. 2015 * 2016 * RETURN VALUES: 2017 * 0 - success 2018 * -EIO - i/o error 2019 * 2020 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2021 */ 2022 static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno, 2023 int nblocks) 2024 { 2025 s8 oldroot; 2026 int rc; 2027 2028 /* save the current value of the root (i.e. maximum free string) 2029 * of the dmap tree. 2030 */ 2031 oldroot = dp->tree.stree[ROOT]; 2032 2033 /* allocate the specified (blocks) bits */ 2034 dbAllocBits(bmp, dp, blkno, nblocks); 2035 2036 /* if the root has not changed, done. */ 2037 if (dp->tree.stree[ROOT] == oldroot) 2038 return (0); 2039 2040 /* root changed. bubble the change up to the dmap control pages. 2041 * if the adjustment of the upper level control pages fails, 2042 * backout the bit allocation (thus making everything consistent). 2043 */ 2044 if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 1, 0))) 2045 dbFreeBits(bmp, dp, blkno, nblocks); 2046 2047 return (rc); 2048 } 2049 2050 2051 /* 2052 * NAME: dbFreeDmap() 2053 * 2054 * FUNCTION: adjust the disk allocation map to reflect the allocation 2055 * of a specified block range within a dmap. 2056 * 2057 * this routine frees the specified blocks from the dmap through 2058 * a call to dbFreeBits(). if the deallocation of the block range 2059 * causes the maximum string of free blocks within the dmap to 2060 * change (i.e. the value of the root of the dmap's dmtree), this 2061 * routine will cause this change to be reflected up through the 2062 * appropriate levels of the dmap control pages by a call to 2063 * dbAdjCtl() for the L0 dmap control page that covers this dmap. 2064 * 2065 * PARAMETERS: 2066 * bmp - pointer to bmap descriptor 2067 * dp - pointer to dmap to free the block range from. 2068 * blkno - starting block number of the block to be freed. 2069 * nblocks - number of blocks to be freed. 2070 * 2071 * RETURN VALUES: 2072 * 0 - success 2073 * -EIO - i/o error 2074 * 2075 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2076 */ 2077 static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno, 2078 int nblocks) 2079 { 2080 s8 oldroot; 2081 int rc = 0, word; 2082 2083 /* save the current value of the root (i.e. maximum free string) 2084 * of the dmap tree. 2085 */ 2086 oldroot = dp->tree.stree[ROOT]; 2087 2088 /* free the specified (blocks) bits */ 2089 rc = dbFreeBits(bmp, dp, blkno, nblocks); 2090 2091 /* if error or the root has not changed, done. */ 2092 if (rc || (dp->tree.stree[ROOT] == oldroot)) 2093 return (rc); 2094 2095 /* root changed. bubble the change up to the dmap control pages. 2096 * if the adjustment of the upper level control pages fails, 2097 * backout the deallocation. 2098 */ 2099 if ((rc = dbAdjCtl(bmp, blkno, dp->tree.stree[ROOT], 0, 0))) { 2100 word = (blkno & (BPERDMAP - 1)) >> L2DBWORD; 2101 2102 /* as part of backing out the deallocation, we will have 2103 * to back split the dmap tree if the deallocation caused 2104 * the freed blocks to become part of a larger binary buddy 2105 * system. 2106 */ 2107 if (dp->tree.stree[word] == NOFREE) 2108 dbBackSplit((dmtree_t *)&dp->tree, word, false); 2109 2110 dbAllocBits(bmp, dp, blkno, nblocks); 2111 } 2112 2113 return (rc); 2114 } 2115 2116 2117 /* 2118 * NAME: dbAllocBits() 2119 * 2120 * FUNCTION: allocate a specified block range from a dmap. 2121 * 2122 * this routine updates the dmap to reflect the working 2123 * state allocation of the specified block range. it directly 2124 * updates the bits of the working map and causes the adjustment 2125 * of the binary buddy system described by the dmap's dmtree 2126 * leaves to reflect the bits allocated. it also causes the 2127 * dmap's dmtree, as a whole, to reflect the allocated range. 2128 * 2129 * PARAMETERS: 2130 * bmp - pointer to bmap descriptor 2131 * dp - pointer to dmap to allocate bits from. 2132 * blkno - starting block number of the bits to be allocated. 2133 * nblocks - number of bits to be allocated. 2134 * 2135 * RETURN VALUES: none 2136 * 2137 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2138 */ 2139 static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno, 2140 int nblocks) 2141 { 2142 int dbitno, word, rembits, nb, nwords, wbitno, nw, agno; 2143 dmtree_t *tp = (dmtree_t *) & dp->tree; 2144 int size; 2145 s8 *leaf; 2146 2147 /* pick up a pointer to the leaves of the dmap tree */ 2148 leaf = dp->tree.stree + LEAFIND; 2149 2150 /* determine the bit number and word within the dmap of the 2151 * starting block. 2152 */ 2153 dbitno = blkno & (BPERDMAP - 1); 2154 word = dbitno >> L2DBWORD; 2155 2156 /* block range better be within the dmap */ 2157 assert(dbitno + nblocks <= BPERDMAP); 2158 2159 /* allocate the bits of the dmap's words corresponding to the block 2160 * range. not all bits of the first and last words may be contained 2161 * within the block range. if this is the case, we'll work against 2162 * those words (i.e. partial first and/or last) on an individual basis 2163 * (a single pass), allocating the bits of interest by hand and 2164 * updating the leaf corresponding to the dmap word. a single pass 2165 * will be used for all dmap words fully contained within the 2166 * specified range. within this pass, the bits of all fully contained 2167 * dmap words will be marked as free in a single shot and the leaves 2168 * will be updated. a single leaf may describe the free space of 2169 * multiple dmap words, so we may update only a subset of the actual 2170 * leaves corresponding to the dmap words of the block range. 2171 */ 2172 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) { 2173 /* determine the bit number within the word and 2174 * the number of bits within the word. 2175 */ 2176 wbitno = dbitno & (DBWORD - 1); 2177 nb = min(rembits, DBWORD - wbitno); 2178 2179 /* check if only part of a word is to be allocated. 2180 */ 2181 if (nb < DBWORD) { 2182 /* allocate (set to 1) the appropriate bits within 2183 * this dmap word. 2184 */ 2185 dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb) 2186 >> wbitno); 2187 2188 /* update the leaf for this dmap word. in addition 2189 * to setting the leaf value to the binary buddy max 2190 * of the updated dmap word, dbSplit() will split 2191 * the binary system of the leaves if need be. 2192 */ 2193 dbSplit(tp, word, BUDMIN, 2194 dbMaxBud((u8 *)&dp->wmap[word]), false); 2195 2196 word += 1; 2197 } else { 2198 /* one or more dmap words are fully contained 2199 * within the block range. determine how many 2200 * words and allocate (set to 1) the bits of these 2201 * words. 2202 */ 2203 nwords = rembits >> L2DBWORD; 2204 memset(&dp->wmap[word], (int) ONES, nwords * 4); 2205 2206 /* determine how many bits. 2207 */ 2208 nb = nwords << L2DBWORD; 2209 2210 /* now update the appropriate leaves to reflect 2211 * the allocated words. 2212 */ 2213 for (; nwords > 0; nwords -= nw) { 2214 if (leaf[word] < BUDMIN) { 2215 jfs_error(bmp->db_ipbmap->i_sb, 2216 "leaf page corrupt\n"); 2217 break; 2218 } 2219 2220 /* determine what the leaf value should be 2221 * updated to as the minimum of the l2 number 2222 * of bits being allocated and the l2 number 2223 * of bits currently described by this leaf. 2224 */ 2225 size = min_t(int, leaf[word], 2226 NLSTOL2BSZ(nwords)); 2227 2228 /* update the leaf to reflect the allocation. 2229 * in addition to setting the leaf value to 2230 * NOFREE, dbSplit() will split the binary 2231 * system of the leaves to reflect the current 2232 * allocation (size). 2233 */ 2234 dbSplit(tp, word, size, NOFREE, false); 2235 2236 /* get the number of dmap words handled */ 2237 nw = BUDSIZE(size, BUDMIN); 2238 word += nw; 2239 } 2240 } 2241 } 2242 2243 /* update the free count for this dmap */ 2244 le32_add_cpu(&dp->nfree, -nblocks); 2245 2246 BMAP_LOCK(bmp); 2247 2248 /* if this allocation group is completely free, 2249 * update the maximum allocation group number if this allocation 2250 * group is the new max. 2251 */ 2252 agno = blkno >> bmp->db_agl2size; 2253 if (agno > bmp->db_maxag) 2254 bmp->db_maxag = agno; 2255 2256 /* update the free count for the allocation group and map */ 2257 bmp->db_agfree[agno] -= nblocks; 2258 bmp->db_nfree -= nblocks; 2259 2260 BMAP_UNLOCK(bmp); 2261 } 2262 2263 2264 /* 2265 * NAME: dbFreeBits() 2266 * 2267 * FUNCTION: free a specified block range from a dmap. 2268 * 2269 * this routine updates the dmap to reflect the working 2270 * state allocation of the specified block range. it directly 2271 * updates the bits of the working map and causes the adjustment 2272 * of the binary buddy system described by the dmap's dmtree 2273 * leaves to reflect the bits freed. it also causes the dmap's 2274 * dmtree, as a whole, to reflect the deallocated range. 2275 * 2276 * PARAMETERS: 2277 * bmp - pointer to bmap descriptor 2278 * dp - pointer to dmap to free bits from. 2279 * blkno - starting block number of the bits to be freed. 2280 * nblocks - number of bits to be freed. 2281 * 2282 * RETURN VALUES: 0 for success 2283 * 2284 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2285 */ 2286 static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno, 2287 int nblocks) 2288 { 2289 int dbitno, word, rembits, nb, nwords, wbitno, nw, agno; 2290 dmtree_t *tp = (dmtree_t *) & dp->tree; 2291 int rc = 0; 2292 int size; 2293 2294 /* determine the bit number and word within the dmap of the 2295 * starting block. 2296 */ 2297 dbitno = blkno & (BPERDMAP - 1); 2298 word = dbitno >> L2DBWORD; 2299 2300 /* block range better be within the dmap. 2301 */ 2302 assert(dbitno + nblocks <= BPERDMAP); 2303 2304 /* free the bits of the dmaps words corresponding to the block range. 2305 * not all bits of the first and last words may be contained within 2306 * the block range. if this is the case, we'll work against those 2307 * words (i.e. partial first and/or last) on an individual basis 2308 * (a single pass), freeing the bits of interest by hand and updating 2309 * the leaf corresponding to the dmap word. a single pass will be used 2310 * for all dmap words fully contained within the specified range. 2311 * within this pass, the bits of all fully contained dmap words will 2312 * be marked as free in a single shot and the leaves will be updated. a 2313 * single leaf may describe the free space of multiple dmap words, 2314 * so we may update only a subset of the actual leaves corresponding 2315 * to the dmap words of the block range. 2316 * 2317 * dbJoin() is used to update leaf values and will join the binary 2318 * buddy system of the leaves if the new leaf values indicate this 2319 * should be done. 2320 */ 2321 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) { 2322 /* determine the bit number within the word and 2323 * the number of bits within the word. 2324 */ 2325 wbitno = dbitno & (DBWORD - 1); 2326 nb = min(rembits, DBWORD - wbitno); 2327 2328 /* check if only part of a word is to be freed. 2329 */ 2330 if (nb < DBWORD) { 2331 /* free (zero) the appropriate bits within this 2332 * dmap word. 2333 */ 2334 dp->wmap[word] &= 2335 cpu_to_le32(~(ONES << (DBWORD - nb) 2336 >> wbitno)); 2337 2338 /* update the leaf for this dmap word. 2339 */ 2340 rc = dbJoin(tp, word, 2341 dbMaxBud((u8 *)&dp->wmap[word]), false); 2342 if (rc) 2343 return rc; 2344 2345 word += 1; 2346 } else { 2347 /* one or more dmap words are fully contained 2348 * within the block range. determine how many 2349 * words and free (zero) the bits of these words. 2350 */ 2351 nwords = rembits >> L2DBWORD; 2352 memset(&dp->wmap[word], 0, nwords * 4); 2353 2354 /* determine how many bits. 2355 */ 2356 nb = nwords << L2DBWORD; 2357 2358 /* now update the appropriate leaves to reflect 2359 * the freed words. 2360 */ 2361 for (; nwords > 0; nwords -= nw) { 2362 /* determine what the leaf value should be 2363 * updated to as the minimum of the l2 number 2364 * of bits being freed and the l2 (max) number 2365 * of bits that can be described by this leaf. 2366 */ 2367 size = 2368 min(LITOL2BSZ 2369 (word, L2LPERDMAP, BUDMIN), 2370 NLSTOL2BSZ(nwords)); 2371 2372 /* update the leaf. 2373 */ 2374 rc = dbJoin(tp, word, size, false); 2375 if (rc) 2376 return rc; 2377 2378 /* get the number of dmap words handled. 2379 */ 2380 nw = BUDSIZE(size, BUDMIN); 2381 word += nw; 2382 } 2383 } 2384 } 2385 2386 /* update the free count for this dmap. 2387 */ 2388 le32_add_cpu(&dp->nfree, nblocks); 2389 2390 BMAP_LOCK(bmp); 2391 2392 /* update the free count for the allocation group and 2393 * map. 2394 */ 2395 agno = blkno >> bmp->db_agl2size; 2396 bmp->db_nfree += nblocks; 2397 bmp->db_agfree[agno] += nblocks; 2398 2399 /* check if this allocation group is not completely free and 2400 * if it is currently the maximum (rightmost) allocation group. 2401 * if so, establish the new maximum allocation group number by 2402 * searching left for the first allocation group with allocation. 2403 */ 2404 if ((bmp->db_agfree[agno] == bmp->db_agsize && agno == bmp->db_maxag) || 2405 (agno == bmp->db_numag - 1 && 2406 bmp->db_agfree[agno] == (bmp-> db_mapsize & (BPERDMAP - 1)))) { 2407 while (bmp->db_maxag > 0) { 2408 bmp->db_maxag -= 1; 2409 if (bmp->db_agfree[bmp->db_maxag] != 2410 bmp->db_agsize) 2411 break; 2412 } 2413 2414 /* re-establish the allocation group preference if the 2415 * current preference is right of the maximum allocation 2416 * group. 2417 */ 2418 if (bmp->db_agpref > bmp->db_maxag) 2419 bmp->db_agpref = bmp->db_maxag; 2420 } 2421 2422 BMAP_UNLOCK(bmp); 2423 2424 return 0; 2425 } 2426 2427 2428 /* 2429 * NAME: dbAdjCtl() 2430 * 2431 * FUNCTION: adjust a dmap control page at a specified level to reflect 2432 * the change in a lower level dmap or dmap control page's 2433 * maximum string of free blocks (i.e. a change in the root 2434 * of the lower level object's dmtree) due to the allocation 2435 * or deallocation of a range of blocks with a single dmap. 2436 * 2437 * on entry, this routine is provided with the new value of 2438 * the lower level dmap or dmap control page root and the 2439 * starting block number of the block range whose allocation 2440 * or deallocation resulted in the root change. this range 2441 * is respresented by a single leaf of the current dmapctl 2442 * and the leaf will be updated with this value, possibly 2443 * causing a binary buddy system within the leaves to be 2444 * split or joined. the update may also cause the dmapctl's 2445 * dmtree to be updated. 2446 * 2447 * if the adjustment of the dmap control page, itself, causes its 2448 * root to change, this change will be bubbled up to the next dmap 2449 * control level by a recursive call to this routine, specifying 2450 * the new root value and the next dmap control page level to 2451 * be adjusted. 2452 * PARAMETERS: 2453 * bmp - pointer to bmap descriptor 2454 * blkno - the first block of a block range within a dmap. it is 2455 * the allocation or deallocation of this block range that 2456 * requires the dmap control page to be adjusted. 2457 * newval - the new value of the lower level dmap or dmap control 2458 * page root. 2459 * alloc - 'true' if adjustment is due to an allocation. 2460 * level - current level of dmap control page (i.e. L0, L1, L2) to 2461 * be adjusted. 2462 * 2463 * RETURN VALUES: 2464 * 0 - success 2465 * -EIO - i/o error 2466 * 2467 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2468 */ 2469 static int 2470 dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc, int level) 2471 { 2472 struct metapage *mp; 2473 s8 oldroot; 2474 int oldval; 2475 s64 lblkno; 2476 struct dmapctl *dcp; 2477 int rc, leafno, ti; 2478 2479 /* get the buffer for the dmap control page for the specified 2480 * block number and control page level. 2481 */ 2482 lblkno = BLKTOCTL(blkno, bmp->db_l2nbperpage, level); 2483 mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE, 0); 2484 if (mp == NULL) 2485 return -EIO; 2486 dcp = (struct dmapctl *) mp->data; 2487 2488 if (dcp->leafidx != cpu_to_le32(CTLLEAFIND)) { 2489 jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page\n"); 2490 release_metapage(mp); 2491 return -EIO; 2492 } 2493 2494 /* determine the leaf number corresponding to the block and 2495 * the index within the dmap control tree. 2496 */ 2497 leafno = BLKTOCTLLEAF(blkno, dcp->budmin); 2498 ti = leafno + le32_to_cpu(dcp->leafidx); 2499 2500 /* save the current leaf value and the current root level (i.e. 2501 * maximum l2 free string described by this dmapctl). 2502 */ 2503 oldval = dcp->stree[ti]; 2504 oldroot = dcp->stree[ROOT]; 2505 2506 /* check if this is a control page update for an allocation. 2507 * if so, update the leaf to reflect the new leaf value using 2508 * dbSplit(); otherwise (deallocation), use dbJoin() to update 2509 * the leaf with the new value. in addition to updating the 2510 * leaf, dbSplit() will also split the binary buddy system of 2511 * the leaves, if required, and bubble new values within the 2512 * dmapctl tree, if required. similarly, dbJoin() will join 2513 * the binary buddy system of leaves and bubble new values up 2514 * the dmapctl tree as required by the new leaf value. 2515 */ 2516 if (alloc) { 2517 /* check if we are in the middle of a binary buddy 2518 * system. this happens when we are performing the 2519 * first allocation out of an allocation group that 2520 * is part (not the first part) of a larger binary 2521 * buddy system. if we are in the middle, back split 2522 * the system prior to calling dbSplit() which assumes 2523 * that it is at the front of a binary buddy system. 2524 */ 2525 if (oldval == NOFREE) { 2526 rc = dbBackSplit((dmtree_t *)dcp, leafno, true); 2527 if (rc) { 2528 release_metapage(mp); 2529 return rc; 2530 } 2531 oldval = dcp->stree[ti]; 2532 } 2533 dbSplit((dmtree_t *) dcp, leafno, dcp->budmin, newval, true); 2534 } else { 2535 rc = dbJoin((dmtree_t *) dcp, leafno, newval, true); 2536 if (rc) { 2537 release_metapage(mp); 2538 return rc; 2539 } 2540 } 2541 2542 /* check if the root of the current dmap control page changed due 2543 * to the update and if the current dmap control page is not at 2544 * the current top level (i.e. L0, L1, L2) of the map. if so (i.e. 2545 * root changed and this is not the top level), call this routine 2546 * again (recursion) for the next higher level of the mapping to 2547 * reflect the change in root for the current dmap control page. 2548 */ 2549 if (dcp->stree[ROOT] != oldroot) { 2550 /* are we below the top level of the map. if so, 2551 * bubble the root up to the next higher level. 2552 */ 2553 if (level < bmp->db_maxlevel) { 2554 /* bubble up the new root of this dmap control page to 2555 * the next level. 2556 */ 2557 if ((rc = 2558 dbAdjCtl(bmp, blkno, dcp->stree[ROOT], alloc, 2559 level + 1))) { 2560 /* something went wrong in bubbling up the new 2561 * root value, so backout the changes to the 2562 * current dmap control page. 2563 */ 2564 if (alloc) { 2565 dbJoin((dmtree_t *) dcp, leafno, 2566 oldval, true); 2567 } else { 2568 /* the dbJoin() above might have 2569 * caused a larger binary buddy system 2570 * to form and we may now be in the 2571 * middle of it. if this is the case, 2572 * back split the buddies. 2573 */ 2574 if (dcp->stree[ti] == NOFREE) 2575 dbBackSplit((dmtree_t *) 2576 dcp, leafno, true); 2577 dbSplit((dmtree_t *) dcp, leafno, 2578 dcp->budmin, oldval, true); 2579 } 2580 2581 /* release the buffer and return the error. 2582 */ 2583 release_metapage(mp); 2584 return (rc); 2585 } 2586 } else { 2587 /* we're at the top level of the map. update 2588 * the bmap control page to reflect the size 2589 * of the maximum free buddy system. 2590 */ 2591 assert(level == bmp->db_maxlevel); 2592 if (bmp->db_maxfreebud != oldroot) { 2593 jfs_error(bmp->db_ipbmap->i_sb, 2594 "the maximum free buddy is not the old root\n"); 2595 } 2596 bmp->db_maxfreebud = dcp->stree[ROOT]; 2597 } 2598 } 2599 2600 /* write the buffer. 2601 */ 2602 write_metapage(mp); 2603 2604 return (0); 2605 } 2606 2607 2608 /* 2609 * NAME: dbSplit() 2610 * 2611 * FUNCTION: update the leaf of a dmtree with a new value, splitting 2612 * the leaf from the binary buddy system of the dmtree's 2613 * leaves, as required. 2614 * 2615 * PARAMETERS: 2616 * tp - pointer to the tree containing the leaf. 2617 * leafno - the number of the leaf to be updated. 2618 * splitsz - the size the binary buddy system starting at the leaf 2619 * must be split to, specified as the log2 number of blocks. 2620 * newval - the new value for the leaf. 2621 * 2622 * RETURN VALUES: none 2623 * 2624 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2625 */ 2626 static void dbSplit(dmtree_t *tp, int leafno, int splitsz, int newval, bool is_ctl) 2627 { 2628 int budsz; 2629 int cursz; 2630 s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx); 2631 2632 /* check if the leaf needs to be split. 2633 */ 2634 if (leaf[leafno] > tp->dmt_budmin) { 2635 /* the split occurs by cutting the buddy system in half 2636 * at the specified leaf until we reach the specified 2637 * size. pick up the starting split size (current size 2638 * - 1 in l2) and the corresponding buddy size. 2639 */ 2640 cursz = leaf[leafno] - 1; 2641 budsz = BUDSIZE(cursz, tp->dmt_budmin); 2642 2643 /* split until we reach the specified size. 2644 */ 2645 while (cursz >= splitsz) { 2646 /* update the buddy's leaf with its new value. 2647 */ 2648 dbAdjTree(tp, leafno ^ budsz, cursz, is_ctl); 2649 2650 /* on to the next size and buddy. 2651 */ 2652 cursz -= 1; 2653 budsz >>= 1; 2654 } 2655 } 2656 2657 /* adjust the dmap tree to reflect the specified leaf's new 2658 * value. 2659 */ 2660 dbAdjTree(tp, leafno, newval, is_ctl); 2661 } 2662 2663 2664 /* 2665 * NAME: dbBackSplit() 2666 * 2667 * FUNCTION: back split the binary buddy system of dmtree leaves 2668 * that hold a specified leaf until the specified leaf 2669 * starts its own binary buddy system. 2670 * 2671 * the allocators typically perform allocations at the start 2672 * of binary buddy systems and dbSplit() is used to accomplish 2673 * any required splits. in some cases, however, allocation 2674 * may occur in the middle of a binary system and requires a 2675 * back split, with the split proceeding out from the middle of 2676 * the system (less efficient) rather than the start of the 2677 * system (more efficient). the cases in which a back split 2678 * is required are rare and are limited to the first allocation 2679 * within an allocation group which is a part (not first part) 2680 * of a larger binary buddy system and a few exception cases 2681 * in which a previous join operation must be backed out. 2682 * 2683 * PARAMETERS: 2684 * tp - pointer to the tree containing the leaf. 2685 * leafno - the number of the leaf to be updated. 2686 * 2687 * RETURN VALUES: none 2688 * 2689 * serialization: IREAD_LOCK(ipbmap) or IWRITE_LOCK(ipbmap) held on entry/exit; 2690 */ 2691 static int dbBackSplit(dmtree_t *tp, int leafno, bool is_ctl) 2692 { 2693 int budsz, bud, w, bsz, size; 2694 int cursz; 2695 s8 *leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx); 2696 2697 /* leaf should be part (not first part) of a binary 2698 * buddy system. 2699 */ 2700 assert(leaf[leafno] == NOFREE); 2701 2702 /* the back split is accomplished by iteratively finding the leaf 2703 * that starts the buddy system that contains the specified leaf and 2704 * splitting that system in two. this iteration continues until 2705 * the specified leaf becomes the start of a buddy system. 2706 * 2707 * determine maximum possible l2 size for the specified leaf. 2708 */ 2709 size = 2710 LITOL2BSZ(leafno, le32_to_cpu(tp->dmt_l2nleafs), 2711 tp->dmt_budmin); 2712 2713 /* determine the number of leaves covered by this size. this 2714 * is the buddy size that we will start with as we search for 2715 * the buddy system that contains the specified leaf. 2716 */ 2717 budsz = BUDSIZE(size, tp->dmt_budmin); 2718 2719 /* back split. 2720 */ 2721 while (leaf[leafno] == NOFREE) { 2722 /* find the leftmost buddy leaf. 2723 */ 2724 for (w = leafno, bsz = budsz;; bsz <<= 1, 2725 w = (w < bud) ? w : bud) { 2726 if (bsz >= le32_to_cpu(tp->dmt_nleafs)) { 2727 jfs_err("JFS: block map error in dbBackSplit"); 2728 return -EIO; 2729 } 2730 2731 /* determine the buddy. 2732 */ 2733 bud = w ^ bsz; 2734 2735 /* check if this buddy is the start of the system. 2736 */ 2737 if (leaf[bud] != NOFREE) { 2738 /* split the leaf at the start of the 2739 * system in two. 2740 */ 2741 cursz = leaf[bud] - 1; 2742 dbSplit(tp, bud, cursz, cursz, is_ctl); 2743 break; 2744 } 2745 } 2746 } 2747 2748 if (leaf[leafno] != size) { 2749 jfs_err("JFS: wrong leaf value in dbBackSplit"); 2750 return -EIO; 2751 } 2752 return 0; 2753 } 2754 2755 2756 /* 2757 * NAME: dbJoin() 2758 * 2759 * FUNCTION: update the leaf of a dmtree with a new value, joining 2760 * the leaf with other leaves of the dmtree into a multi-leaf 2761 * binary buddy system, as required. 2762 * 2763 * PARAMETERS: 2764 * tp - pointer to the tree containing the leaf. 2765 * leafno - the number of the leaf to be updated. 2766 * newval - the new value for the leaf. 2767 * 2768 * RETURN VALUES: none 2769 */ 2770 static int dbJoin(dmtree_t *tp, int leafno, int newval, bool is_ctl) 2771 { 2772 int budsz, buddy; 2773 s8 *leaf; 2774 2775 /* can the new leaf value require a join with other leaves ? 2776 */ 2777 if (newval >= tp->dmt_budmin) { 2778 /* pickup a pointer to the leaves of the tree. 2779 */ 2780 leaf = tp->dmt_stree + le32_to_cpu(tp->dmt_leafidx); 2781 2782 /* try to join the specified leaf into a large binary 2783 * buddy system. the join proceeds by attempting to join 2784 * the specified leafno with its buddy (leaf) at new value. 2785 * if the join occurs, we attempt to join the left leaf 2786 * of the joined buddies with its buddy at new value + 1. 2787 * we continue to join until we find a buddy that cannot be 2788 * joined (does not have a value equal to the size of the 2789 * last join) or until all leaves have been joined into a 2790 * single system. 2791 * 2792 * get the buddy size (number of words covered) of 2793 * the new value. 2794 */ 2795 budsz = BUDSIZE(newval, tp->dmt_budmin); 2796 2797 /* try to join. 2798 */ 2799 while (budsz < le32_to_cpu(tp->dmt_nleafs)) { 2800 /* get the buddy leaf. 2801 */ 2802 buddy = leafno ^ budsz; 2803 2804 /* if the leaf's new value is greater than its 2805 * buddy's value, we join no more. 2806 */ 2807 if (newval > leaf[buddy]) 2808 break; 2809 2810 /* It shouldn't be less */ 2811 if (newval < leaf[buddy]) 2812 return -EIO; 2813 2814 /* check which (leafno or buddy) is the left buddy. 2815 * the left buddy gets to claim the blocks resulting 2816 * from the join while the right gets to claim none. 2817 * the left buddy is also eligible to participate in 2818 * a join at the next higher level while the right 2819 * is not. 2820 * 2821 */ 2822 if (leafno < buddy) { 2823 /* leafno is the left buddy. 2824 */ 2825 dbAdjTree(tp, buddy, NOFREE, is_ctl); 2826 } else { 2827 /* buddy is the left buddy and becomes 2828 * leafno. 2829 */ 2830 dbAdjTree(tp, leafno, NOFREE, is_ctl); 2831 leafno = buddy; 2832 } 2833 2834 /* on to try the next join. 2835 */ 2836 newval += 1; 2837 budsz <<= 1; 2838 } 2839 } 2840 2841 /* update the leaf value. 2842 */ 2843 dbAdjTree(tp, leafno, newval, is_ctl); 2844 2845 return 0; 2846 } 2847 2848 2849 /* 2850 * NAME: dbAdjTree() 2851 * 2852 * FUNCTION: update a leaf of a dmtree with a new value, adjusting 2853 * the dmtree, as required, to reflect the new leaf value. 2854 * the combination of any buddies must already be done before 2855 * this is called. 2856 * 2857 * PARAMETERS: 2858 * tp - pointer to the tree to be adjusted. 2859 * leafno - the number of the leaf to be updated. 2860 * newval - the new value for the leaf. 2861 * 2862 * RETURN VALUES: none 2863 */ 2864 static void dbAdjTree(dmtree_t *tp, int leafno, int newval, bool is_ctl) 2865 { 2866 int lp, pp, k; 2867 int max, size; 2868 2869 size = is_ctl ? CTLTREESIZE : TREESIZE; 2870 2871 /* pick up the index of the leaf for this leafno. 2872 */ 2873 lp = leafno + le32_to_cpu(tp->dmt_leafidx); 2874 2875 if (WARN_ON_ONCE(lp >= size || lp < 0)) 2876 return; 2877 2878 /* is the current value the same as the old value ? if so, 2879 * there is nothing to do. 2880 */ 2881 if (tp->dmt_stree[lp] == newval) 2882 return; 2883 2884 /* set the new value. 2885 */ 2886 tp->dmt_stree[lp] = newval; 2887 2888 /* bubble the new value up the tree as required. 2889 */ 2890 for (k = 0; k < le32_to_cpu(tp->dmt_height); k++) { 2891 if (lp == 0) 2892 break; 2893 2894 /* get the index of the first leaf of the 4 leaf 2895 * group containing the specified leaf (leafno). 2896 */ 2897 lp = ((lp - 1) & ~0x03) + 1; 2898 2899 /* get the index of the parent of this 4 leaf group. 2900 */ 2901 pp = (lp - 1) >> 2; 2902 2903 /* determine the maximum of the 4 leaves. 2904 */ 2905 max = TREEMAX(&tp->dmt_stree[lp]); 2906 2907 /* if the maximum of the 4 is the same as the 2908 * parent's value, we're done. 2909 */ 2910 if (tp->dmt_stree[pp] == max) 2911 break; 2912 2913 /* parent gets new value. 2914 */ 2915 tp->dmt_stree[pp] = max; 2916 2917 /* parent becomes leaf for next go-round. 2918 */ 2919 lp = pp; 2920 } 2921 } 2922 2923 2924 /* 2925 * NAME: dbFindLeaf() 2926 * 2927 * FUNCTION: search a dmtree_t for sufficient free blocks, returning 2928 * the index of a leaf describing the free blocks if 2929 * sufficient free blocks are found. 2930 * 2931 * the search starts at the top of the dmtree_t tree and 2932 * proceeds down the tree to the leftmost leaf with sufficient 2933 * free space. 2934 * 2935 * PARAMETERS: 2936 * tp - pointer to the tree to be searched. 2937 * l2nb - log2 number of free blocks to search for. 2938 * leafidx - return pointer to be set to the index of the leaf 2939 * describing at least l2nb free blocks if sufficient 2940 * free blocks are found. 2941 * is_ctl - determines if the tree is of type ctl 2942 * 2943 * RETURN VALUES: 2944 * 0 - success 2945 * -ENOSPC - insufficient free blocks. 2946 */ 2947 static int dbFindLeaf(dmtree_t *tp, int l2nb, int *leafidx, bool is_ctl) 2948 { 2949 int ti, n = 0, k, x = 0; 2950 int max_size, max_idx; 2951 2952 max_size = is_ctl ? CTLTREESIZE : TREESIZE; 2953 max_idx = is_ctl ? LPERCTL : LPERDMAP; 2954 2955 /* first check the root of the tree to see if there is 2956 * sufficient free space. 2957 */ 2958 if (l2nb > tp->dmt_stree[ROOT]) 2959 return -ENOSPC; 2960 2961 /* sufficient free space available. now search down the tree 2962 * starting at the next level for the leftmost leaf that 2963 * describes sufficient free space. 2964 */ 2965 for (k = le32_to_cpu(tp->dmt_height), ti = 1; 2966 k > 0; k--, ti = ((ti + n) << 2) + 1) { 2967 /* search the four nodes at this level, starting from 2968 * the left. 2969 */ 2970 for (x = ti, n = 0; n < 4; n++) { 2971 /* sufficient free space found. move to the next 2972 * level (or quit if this is the last level). 2973 */ 2974 if (x + n > max_size) 2975 return -ENOSPC; 2976 if (l2nb <= tp->dmt_stree[x + n]) 2977 break; 2978 } 2979 2980 /* better have found something since the higher 2981 * levels of the tree said it was here. 2982 */ 2983 assert(n < 4); 2984 } 2985 if (le32_to_cpu(tp->dmt_leafidx) >= max_idx) 2986 return -ENOSPC; 2987 2988 /* set the return to the leftmost leaf describing sufficient 2989 * free space. 2990 */ 2991 *leafidx = x + n - le32_to_cpu(tp->dmt_leafidx); 2992 2993 return (0); 2994 } 2995 2996 2997 /* 2998 * NAME: dbFindBits() 2999 * 3000 * FUNCTION: find a specified number of binary buddy free bits within a 3001 * dmap bitmap word value. 3002 * 3003 * this routine searches the bitmap value for (1 << l2nb) free 3004 * bits at (1 << l2nb) alignments within the value. 3005 * 3006 * PARAMETERS: 3007 * word - dmap bitmap word value. 3008 * l2nb - number of free bits specified as a log2 number. 3009 * 3010 * RETURN VALUES: 3011 * starting bit number of free bits. 3012 */ 3013 static int dbFindBits(u32 word, int l2nb) 3014 { 3015 int bitno, nb; 3016 u32 mask; 3017 3018 /* get the number of bits. 3019 */ 3020 nb = 1 << l2nb; 3021 assert(nb <= DBWORD); 3022 3023 /* complement the word so we can use a mask (i.e. 0s represent 3024 * free bits) and compute the mask. 3025 */ 3026 word = ~word; 3027 mask = ONES << (DBWORD - nb); 3028 3029 /* scan the word for nb free bits at nb alignments. 3030 */ 3031 for (bitno = 0; mask != 0; bitno += nb, mask = (mask >> nb)) { 3032 if ((mask & word) == mask) 3033 break; 3034 } 3035 3036 ASSERT(bitno < 32); 3037 3038 /* return the bit number. 3039 */ 3040 return (bitno); 3041 } 3042 3043 3044 /* 3045 * NAME: dbMaxBud(u8 *cp) 3046 * 3047 * FUNCTION: determine the largest binary buddy string of free 3048 * bits within 32-bits of the map. 3049 * 3050 * PARAMETERS: 3051 * cp - pointer to the 32-bit value. 3052 * 3053 * RETURN VALUES: 3054 * largest binary buddy of free bits within a dmap word. 3055 */ 3056 static int dbMaxBud(u8 * cp) 3057 { 3058 signed char tmp1, tmp2; 3059 3060 /* check if the wmap word is all free. if so, the 3061 * free buddy size is BUDMIN. 3062 */ 3063 if (*((uint *) cp) == 0) 3064 return (BUDMIN); 3065 3066 /* check if the wmap word is half free. if so, the 3067 * free buddy size is BUDMIN-1. 3068 */ 3069 if (*((u16 *) cp) == 0 || *((u16 *) cp + 1) == 0) 3070 return (BUDMIN - 1); 3071 3072 /* not all free or half free. determine the free buddy 3073 * size thru table lookup using quarters of the wmap word. 3074 */ 3075 tmp1 = max(budtab[cp[2]], budtab[cp[3]]); 3076 tmp2 = max(budtab[cp[0]], budtab[cp[1]]); 3077 return (max(tmp1, tmp2)); 3078 } 3079 3080 3081 /* 3082 * NAME: cnttz(uint word) 3083 * 3084 * FUNCTION: determine the number of trailing zeros within a 32-bit 3085 * value. 3086 * 3087 * PARAMETERS: 3088 * value - 32-bit value to be examined. 3089 * 3090 * RETURN VALUES: 3091 * count of trailing zeros 3092 */ 3093 static int cnttz(u32 word) 3094 { 3095 int n; 3096 3097 for (n = 0; n < 32; n++, word >>= 1) { 3098 if (word & 0x01) 3099 break; 3100 } 3101 3102 return (n); 3103 } 3104 3105 3106 /* 3107 * NAME: cntlz(u32 value) 3108 * 3109 * FUNCTION: determine the number of leading zeros within a 32-bit 3110 * value. 3111 * 3112 * PARAMETERS: 3113 * value - 32-bit value to be examined. 3114 * 3115 * RETURN VALUES: 3116 * count of leading zeros 3117 */ 3118 static int cntlz(u32 value) 3119 { 3120 int n; 3121 3122 for (n = 0; n < 32; n++, value <<= 1) { 3123 if (value & HIGHORDER) 3124 break; 3125 } 3126 return (n); 3127 } 3128 3129 3130 /* 3131 * NAME: blkstol2(s64 nb) 3132 * 3133 * FUNCTION: convert a block count to its log2 value. if the block 3134 * count is not a l2 multiple, it is rounded up to the next 3135 * larger l2 multiple. 3136 * 3137 * PARAMETERS: 3138 * nb - number of blocks 3139 * 3140 * RETURN VALUES: 3141 * log2 number of blocks 3142 */ 3143 static int blkstol2(s64 nb) 3144 { 3145 int l2nb; 3146 s64 mask; /* meant to be signed */ 3147 3148 mask = (s64) 1 << (64 - 1); 3149 3150 /* count the leading bits. 3151 */ 3152 for (l2nb = 0; l2nb < 64; l2nb++, mask >>= 1) { 3153 /* leading bit found. 3154 */ 3155 if (nb & mask) { 3156 /* determine the l2 value. 3157 */ 3158 l2nb = (64 - 1) - l2nb; 3159 3160 /* check if we need to round up. 3161 */ 3162 if (~mask & nb) 3163 l2nb++; 3164 3165 return (l2nb); 3166 } 3167 } 3168 assert(0); 3169 return 0; /* fix compiler warning */ 3170 } 3171 3172 3173 /* 3174 * NAME: dbAllocBottomUp() 3175 * 3176 * FUNCTION: alloc the specified block range from the working block 3177 * allocation map. 3178 * 3179 * the blocks will be alloc from the working map one dmap 3180 * at a time. 3181 * 3182 * PARAMETERS: 3183 * ip - pointer to in-core inode; 3184 * blkno - starting block number to be freed. 3185 * nblocks - number of blocks to be freed. 3186 * 3187 * RETURN VALUES: 3188 * 0 - success 3189 * -EIO - i/o error 3190 */ 3191 int dbAllocBottomUp(struct inode *ip, s64 blkno, s64 nblocks) 3192 { 3193 struct metapage *mp; 3194 struct dmap *dp; 3195 int nb, rc; 3196 s64 lblkno, rem; 3197 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 3198 struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap; 3199 3200 IREAD_LOCK(ipbmap, RDWRLOCK_DMAP); 3201 3202 /* block to be allocated better be within the mapsize. */ 3203 ASSERT(nblocks <= bmp->db_mapsize - blkno); 3204 3205 /* 3206 * allocate the blocks a dmap at a time. 3207 */ 3208 mp = NULL; 3209 for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) { 3210 /* release previous dmap if any */ 3211 if (mp) { 3212 write_metapage(mp); 3213 } 3214 3215 /* get the buffer for the current dmap. */ 3216 lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage); 3217 mp = read_metapage(ipbmap, lblkno, PSIZE, 0); 3218 if (mp == NULL) { 3219 IREAD_UNLOCK(ipbmap); 3220 return -EIO; 3221 } 3222 dp = (struct dmap *) mp->data; 3223 3224 /* determine the number of blocks to be allocated from 3225 * this dmap. 3226 */ 3227 nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1))); 3228 3229 /* allocate the blocks. */ 3230 if ((rc = dbAllocDmapBU(bmp, dp, blkno, nb))) { 3231 release_metapage(mp); 3232 IREAD_UNLOCK(ipbmap); 3233 return (rc); 3234 } 3235 } 3236 3237 /* write the last buffer. */ 3238 write_metapage(mp); 3239 3240 IREAD_UNLOCK(ipbmap); 3241 3242 return (0); 3243 } 3244 3245 3246 static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno, 3247 int nblocks) 3248 { 3249 int rc; 3250 int dbitno, word, rembits, nb, nwords, wbitno, agno; 3251 s8 oldroot; 3252 struct dmaptree *tp = (struct dmaptree *) & dp->tree; 3253 3254 /* save the current value of the root (i.e. maximum free string) 3255 * of the dmap tree. 3256 */ 3257 oldroot = tp->stree[ROOT]; 3258 3259 /* determine the bit number and word within the dmap of the 3260 * starting block. 3261 */ 3262 dbitno = blkno & (BPERDMAP - 1); 3263 word = dbitno >> L2DBWORD; 3264 3265 /* block range better be within the dmap */ 3266 assert(dbitno + nblocks <= BPERDMAP); 3267 3268 /* allocate the bits of the dmap's words corresponding to the block 3269 * range. not all bits of the first and last words may be contained 3270 * within the block range. if this is the case, we'll work against 3271 * those words (i.e. partial first and/or last) on an individual basis 3272 * (a single pass), allocating the bits of interest by hand and 3273 * updating the leaf corresponding to the dmap word. a single pass 3274 * will be used for all dmap words fully contained within the 3275 * specified range. within this pass, the bits of all fully contained 3276 * dmap words will be marked as free in a single shot and the leaves 3277 * will be updated. a single leaf may describe the free space of 3278 * multiple dmap words, so we may update only a subset of the actual 3279 * leaves corresponding to the dmap words of the block range. 3280 */ 3281 for (rembits = nblocks; rembits > 0; rembits -= nb, dbitno += nb) { 3282 /* determine the bit number within the word and 3283 * the number of bits within the word. 3284 */ 3285 wbitno = dbitno & (DBWORD - 1); 3286 nb = min(rembits, DBWORD - wbitno); 3287 3288 /* check if only part of a word is to be allocated. 3289 */ 3290 if (nb < DBWORD) { 3291 /* allocate (set to 1) the appropriate bits within 3292 * this dmap word. 3293 */ 3294 dp->wmap[word] |= cpu_to_le32(ONES << (DBWORD - nb) 3295 >> wbitno); 3296 3297 word++; 3298 } else { 3299 /* one or more dmap words are fully contained 3300 * within the block range. determine how many 3301 * words and allocate (set to 1) the bits of these 3302 * words. 3303 */ 3304 nwords = rembits >> L2DBWORD; 3305 memset(&dp->wmap[word], (int) ONES, nwords * 4); 3306 3307 /* determine how many bits */ 3308 nb = nwords << L2DBWORD; 3309 word += nwords; 3310 } 3311 } 3312 3313 /* update the free count for this dmap */ 3314 le32_add_cpu(&dp->nfree, -nblocks); 3315 3316 /* reconstruct summary tree */ 3317 dbInitDmapTree(dp); 3318 3319 BMAP_LOCK(bmp); 3320 3321 /* if this allocation group is completely free, 3322 * update the highest active allocation group number 3323 * if this allocation group is the new max. 3324 */ 3325 agno = blkno >> bmp->db_agl2size; 3326 if (agno > bmp->db_maxag) 3327 bmp->db_maxag = agno; 3328 3329 /* update the free count for the allocation group and map */ 3330 bmp->db_agfree[agno] -= nblocks; 3331 bmp->db_nfree -= nblocks; 3332 3333 BMAP_UNLOCK(bmp); 3334 3335 /* if the root has not changed, done. */ 3336 if (tp->stree[ROOT] == oldroot) 3337 return (0); 3338 3339 /* root changed. bubble the change up to the dmap control pages. 3340 * if the adjustment of the upper level control pages fails, 3341 * backout the bit allocation (thus making everything consistent). 3342 */ 3343 if ((rc = dbAdjCtl(bmp, blkno, tp->stree[ROOT], 1, 0))) 3344 dbFreeBits(bmp, dp, blkno, nblocks); 3345 3346 return (rc); 3347 } 3348 3349 3350 /* 3351 * NAME: dbExtendFS() 3352 * 3353 * FUNCTION: extend bmap from blkno for nblocks; 3354 * dbExtendFS() updates bmap ready for dbAllocBottomUp(); 3355 * 3356 * L2 3357 * | 3358 * L1---------------------------------L1 3359 * | | 3360 * L0---------L0---------L0 L0---------L0---------L0 3361 * | | | | | | 3362 * d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,...,dn d0,.,dm; 3363 * L2L1L0d0,...,dnL0d0,...,dnL0d0,...,dnL1L0d0,...,dnL0d0,...,dnL0d0,..dm 3364 * 3365 * <---old---><----------------------------extend-----------------------> 3366 */ 3367 int dbExtendFS(struct inode *ipbmap, s64 blkno, s64 nblocks) 3368 { 3369 struct jfs_sb_info *sbi = JFS_SBI(ipbmap->i_sb); 3370 int nbperpage = sbi->nbperpage; 3371 int i, i0 = true, j, j0 = true, k, n; 3372 s64 newsize; 3373 s64 p; 3374 struct metapage *mp, *l2mp, *l1mp = NULL, *l0mp = NULL; 3375 struct dmapctl *l2dcp, *l1dcp, *l0dcp; 3376 struct dmap *dp; 3377 s8 *l0leaf, *l1leaf, *l2leaf; 3378 struct bmap *bmp = sbi->bmap; 3379 int agno, l2agsize, oldl2agsize; 3380 s64 ag_rem; 3381 3382 newsize = blkno + nblocks; 3383 3384 jfs_info("dbExtendFS: blkno:%Ld nblocks:%Ld newsize:%Ld", 3385 (long long) blkno, (long long) nblocks, (long long) newsize); 3386 3387 /* 3388 * initialize bmap control page. 3389 * 3390 * all the data in bmap control page should exclude 3391 * the mkfs hidden dmap page. 3392 */ 3393 3394 /* update mapsize */ 3395 bmp->db_mapsize = newsize; 3396 bmp->db_maxlevel = BMAPSZTOLEV(bmp->db_mapsize); 3397 3398 /* compute new AG size */ 3399 l2agsize = dbGetL2AGSize(newsize); 3400 oldl2agsize = bmp->db_agl2size; 3401 3402 bmp->db_agl2size = l2agsize; 3403 bmp->db_agsize = (s64)1 << l2agsize; 3404 3405 /* compute new number of AG */ 3406 agno = bmp->db_numag; 3407 bmp->db_numag = newsize >> l2agsize; 3408 bmp->db_numag += ((u32) newsize % (u32) bmp->db_agsize) ? 1 : 0; 3409 3410 /* 3411 * reconfigure db_agfree[] 3412 * from old AG configuration to new AG configuration; 3413 * 3414 * coalesce contiguous k (newAGSize/oldAGSize) AGs; 3415 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn; 3416 * note: new AG size = old AG size * (2**x). 3417 */ 3418 if (l2agsize == oldl2agsize) 3419 goto extend; 3420 k = 1 << (l2agsize - oldl2agsize); 3421 ag_rem = bmp->db_agfree[0]; /* save agfree[0] */ 3422 for (i = 0, n = 0; i < agno; n++) { 3423 bmp->db_agfree[n] = 0; /* init collection point */ 3424 3425 /* coalesce contiguous k AGs; */ 3426 for (j = 0; j < k && i < agno; j++, i++) { 3427 /* merge AGi to AGn */ 3428 bmp->db_agfree[n] += bmp->db_agfree[i]; 3429 } 3430 } 3431 bmp->db_agfree[0] += ag_rem; /* restore agfree[0] */ 3432 3433 for (; n < MAXAG; n++) 3434 bmp->db_agfree[n] = 0; 3435 3436 /* 3437 * update highest active ag number 3438 */ 3439 3440 bmp->db_maxag = bmp->db_maxag / k; 3441 3442 /* 3443 * extend bmap 3444 * 3445 * update bit maps and corresponding level control pages; 3446 * global control page db_nfree, db_agfree[agno], db_maxfreebud; 3447 */ 3448 extend: 3449 /* get L2 page */ 3450 p = BMAPBLKNO + nbperpage; /* L2 page */ 3451 l2mp = read_metapage(ipbmap, p, PSIZE, 0); 3452 if (!l2mp) { 3453 jfs_error(ipbmap->i_sb, "L2 page could not be read\n"); 3454 return -EIO; 3455 } 3456 l2dcp = (struct dmapctl *) l2mp->data; 3457 3458 /* compute start L1 */ 3459 k = blkno >> L2MAXL1SIZE; 3460 l2leaf = l2dcp->stree + CTLLEAFIND + k; 3461 p = BLKTOL1(blkno, sbi->l2nbperpage); /* L1 page */ 3462 3463 /* 3464 * extend each L1 in L2 3465 */ 3466 for (; k < LPERCTL; k++, p += nbperpage) { 3467 /* get L1 page */ 3468 if (j0) { 3469 /* read in L1 page: (blkno & (MAXL1SIZE - 1)) */ 3470 l1mp = read_metapage(ipbmap, p, PSIZE, 0); 3471 if (l1mp == NULL) 3472 goto errout; 3473 l1dcp = (struct dmapctl *) l1mp->data; 3474 3475 /* compute start L0 */ 3476 j = (blkno & (MAXL1SIZE - 1)) >> L2MAXL0SIZE; 3477 l1leaf = l1dcp->stree + CTLLEAFIND + j; 3478 p = BLKTOL0(blkno, sbi->l2nbperpage); 3479 j0 = false; 3480 } else { 3481 /* assign/init L1 page */ 3482 l1mp = get_metapage(ipbmap, p, PSIZE, 0); 3483 if (l1mp == NULL) 3484 goto errout; 3485 3486 l1dcp = (struct dmapctl *) l1mp->data; 3487 3488 /* compute start L0 */ 3489 j = 0; 3490 l1leaf = l1dcp->stree + CTLLEAFIND; 3491 p += nbperpage; /* 1st L0 of L1.k */ 3492 } 3493 3494 /* 3495 * extend each L0 in L1 3496 */ 3497 for (; j < LPERCTL; j++) { 3498 /* get L0 page */ 3499 if (i0) { 3500 /* read in L0 page: (blkno & (MAXL0SIZE - 1)) */ 3501 3502 l0mp = read_metapage(ipbmap, p, PSIZE, 0); 3503 if (l0mp == NULL) 3504 goto errout; 3505 l0dcp = (struct dmapctl *) l0mp->data; 3506 3507 /* compute start dmap */ 3508 i = (blkno & (MAXL0SIZE - 1)) >> 3509 L2BPERDMAP; 3510 l0leaf = l0dcp->stree + CTLLEAFIND + i; 3511 p = BLKTODMAP(blkno, 3512 sbi->l2nbperpage); 3513 i0 = false; 3514 } else { 3515 /* assign/init L0 page */ 3516 l0mp = get_metapage(ipbmap, p, PSIZE, 0); 3517 if (l0mp == NULL) 3518 goto errout; 3519 3520 l0dcp = (struct dmapctl *) l0mp->data; 3521 3522 /* compute start dmap */ 3523 i = 0; 3524 l0leaf = l0dcp->stree + CTLLEAFIND; 3525 p += nbperpage; /* 1st dmap of L0.j */ 3526 } 3527 3528 /* 3529 * extend each dmap in L0 3530 */ 3531 for (; i < LPERCTL; i++) { 3532 /* 3533 * reconstruct the dmap page, and 3534 * initialize corresponding parent L0 leaf 3535 */ 3536 if ((n = blkno & (BPERDMAP - 1))) { 3537 /* read in dmap page: */ 3538 mp = read_metapage(ipbmap, p, 3539 PSIZE, 0); 3540 if (mp == NULL) 3541 goto errout; 3542 n = min(nblocks, (s64)BPERDMAP - n); 3543 } else { 3544 /* assign/init dmap page */ 3545 mp = read_metapage(ipbmap, p, 3546 PSIZE, 0); 3547 if (mp == NULL) 3548 goto errout; 3549 3550 n = min_t(s64, nblocks, BPERDMAP); 3551 } 3552 3553 dp = (struct dmap *) mp->data; 3554 *l0leaf = dbInitDmap(dp, blkno, n); 3555 3556 bmp->db_nfree += n; 3557 agno = le64_to_cpu(dp->start) >> l2agsize; 3558 bmp->db_agfree[agno] += n; 3559 3560 write_metapage(mp); 3561 3562 l0leaf++; 3563 p += nbperpage; 3564 3565 blkno += n; 3566 nblocks -= n; 3567 if (nblocks == 0) 3568 break; 3569 } /* for each dmap in a L0 */ 3570 3571 /* 3572 * build current L0 page from its leaves, and 3573 * initialize corresponding parent L1 leaf 3574 */ 3575 *l1leaf = dbInitDmapCtl(l0dcp, 0, ++i); 3576 write_metapage(l0mp); 3577 l0mp = NULL; 3578 3579 if (nblocks) 3580 l1leaf++; /* continue for next L0 */ 3581 else { 3582 /* more than 1 L0 ? */ 3583 if (j > 0) 3584 break; /* build L1 page */ 3585 else { 3586 /* summarize in global bmap page */ 3587 bmp->db_maxfreebud = *l1leaf; 3588 release_metapage(l1mp); 3589 release_metapage(l2mp); 3590 goto finalize; 3591 } 3592 } 3593 } /* for each L0 in a L1 */ 3594 3595 /* 3596 * build current L1 page from its leaves, and 3597 * initialize corresponding parent L2 leaf 3598 */ 3599 *l2leaf = dbInitDmapCtl(l1dcp, 1, ++j); 3600 write_metapage(l1mp); 3601 l1mp = NULL; 3602 3603 if (nblocks) 3604 l2leaf++; /* continue for next L1 */ 3605 else { 3606 /* more than 1 L1 ? */ 3607 if (k > 0) 3608 break; /* build L2 page */ 3609 else { 3610 /* summarize in global bmap page */ 3611 bmp->db_maxfreebud = *l2leaf; 3612 release_metapage(l2mp); 3613 goto finalize; 3614 } 3615 } 3616 } /* for each L1 in a L2 */ 3617 3618 jfs_error(ipbmap->i_sb, "function has not returned as expected\n"); 3619 errout: 3620 if (l0mp) 3621 release_metapage(l0mp); 3622 if (l1mp) 3623 release_metapage(l1mp); 3624 release_metapage(l2mp); 3625 return -EIO; 3626 3627 /* 3628 * finalize bmap control page 3629 */ 3630 finalize: 3631 3632 return 0; 3633 } 3634 3635 3636 /* 3637 * dbFinalizeBmap() 3638 */ 3639 void dbFinalizeBmap(struct inode *ipbmap) 3640 { 3641 struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap; 3642 int actags, inactags, l2nl; 3643 s64 ag_rem, actfree, inactfree, avgfree; 3644 int i, n; 3645 3646 /* 3647 * finalize bmap control page 3648 */ 3649 //finalize: 3650 /* 3651 * compute db_agpref: preferred ag to allocate from 3652 * (the leftmost ag with average free space in it); 3653 */ 3654 //agpref: 3655 /* get the number of active ags and inactive ags */ 3656 actags = bmp->db_maxag + 1; 3657 inactags = bmp->db_numag - actags; 3658 ag_rem = bmp->db_mapsize & (bmp->db_agsize - 1); /* ??? */ 3659 3660 /* determine how many blocks are in the inactive allocation 3661 * groups. in doing this, we must account for the fact that 3662 * the rightmost group might be a partial group (i.e. file 3663 * system size is not a multiple of the group size). 3664 */ 3665 inactfree = (inactags && ag_rem) ? 3666 (((s64)inactags - 1) << bmp->db_agl2size) + ag_rem 3667 : ((s64)inactags << bmp->db_agl2size); 3668 3669 /* determine how many free blocks are in the active 3670 * allocation groups plus the average number of free blocks 3671 * within the active ags. 3672 */ 3673 actfree = bmp->db_nfree - inactfree; 3674 avgfree = (u32) actfree / (u32) actags; 3675 3676 /* if the preferred allocation group has not average free space. 3677 * re-establish the preferred group as the leftmost 3678 * group with average free space. 3679 */ 3680 if (bmp->db_agfree[bmp->db_agpref] < avgfree) { 3681 for (bmp->db_agpref = 0; bmp->db_agpref < actags; 3682 bmp->db_agpref++) { 3683 if (bmp->db_agfree[bmp->db_agpref] >= avgfree) 3684 break; 3685 } 3686 if (bmp->db_agpref >= bmp->db_numag) { 3687 jfs_error(ipbmap->i_sb, 3688 "cannot find ag with average freespace\n"); 3689 } 3690 } 3691 3692 /* 3693 * compute db_aglevel, db_agheight, db_width, db_agstart: 3694 * an ag is covered in aglevel dmapctl summary tree, 3695 * at agheight level height (from leaf) with agwidth number of nodes 3696 * each, which starts at agstart index node of the smmary tree node 3697 * array; 3698 */ 3699 bmp->db_aglevel = BMAPSZTOLEV(bmp->db_agsize); 3700 l2nl = 3701 bmp->db_agl2size - (L2BPERDMAP + bmp->db_aglevel * L2LPERCTL); 3702 bmp->db_agheight = l2nl >> 1; 3703 bmp->db_agwidth = 1 << (l2nl - (bmp->db_agheight << 1)); 3704 for (i = 5 - bmp->db_agheight, bmp->db_agstart = 0, n = 1; i > 0; 3705 i--) { 3706 bmp->db_agstart += n; 3707 n <<= 2; 3708 } 3709 3710 } 3711 3712 3713 /* 3714 * NAME: dbInitDmap()/ujfs_idmap_page() 3715 * 3716 * FUNCTION: initialize working/persistent bitmap of the dmap page 3717 * for the specified number of blocks: 3718 * 3719 * at entry, the bitmaps had been initialized as free (ZEROS); 3720 * The number of blocks will only account for the actually 3721 * existing blocks. Blocks which don't actually exist in 3722 * the aggregate will be marked as allocated (ONES); 3723 * 3724 * PARAMETERS: 3725 * dp - pointer to page of map 3726 * nblocks - number of blocks this page 3727 * 3728 * RETURNS: NONE 3729 */ 3730 static int dbInitDmap(struct dmap * dp, s64 Blkno, int nblocks) 3731 { 3732 int blkno, w, b, r, nw, nb, i; 3733 3734 /* starting block number within the dmap */ 3735 blkno = Blkno & (BPERDMAP - 1); 3736 3737 if (blkno == 0) { 3738 dp->nblocks = dp->nfree = cpu_to_le32(nblocks); 3739 dp->start = cpu_to_le64(Blkno); 3740 3741 if (nblocks == BPERDMAP) { 3742 memset(&dp->wmap[0], 0, LPERDMAP * 4); 3743 memset(&dp->pmap[0], 0, LPERDMAP * 4); 3744 goto initTree; 3745 } 3746 } else { 3747 le32_add_cpu(&dp->nblocks, nblocks); 3748 le32_add_cpu(&dp->nfree, nblocks); 3749 } 3750 3751 /* word number containing start block number */ 3752 w = blkno >> L2DBWORD; 3753 3754 /* 3755 * free the bits corresponding to the block range (ZEROS): 3756 * note: not all bits of the first and last words may be contained 3757 * within the block range. 3758 */ 3759 for (r = nblocks; r > 0; r -= nb, blkno += nb) { 3760 /* number of bits preceding range to be freed in the word */ 3761 b = blkno & (DBWORD - 1); 3762 /* number of bits to free in the word */ 3763 nb = min(r, DBWORD - b); 3764 3765 /* is partial word to be freed ? */ 3766 if (nb < DBWORD) { 3767 /* free (set to 0) from the bitmap word */ 3768 dp->wmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb) 3769 >> b)); 3770 dp->pmap[w] &= cpu_to_le32(~(ONES << (DBWORD - nb) 3771 >> b)); 3772 3773 /* skip the word freed */ 3774 w++; 3775 } else { 3776 /* free (set to 0) contiguous bitmap words */ 3777 nw = r >> L2DBWORD; 3778 memset(&dp->wmap[w], 0, nw * 4); 3779 memset(&dp->pmap[w], 0, nw * 4); 3780 3781 /* skip the words freed */ 3782 nb = nw << L2DBWORD; 3783 w += nw; 3784 } 3785 } 3786 3787 /* 3788 * mark bits following the range to be freed (non-existing 3789 * blocks) as allocated (ONES) 3790 */ 3791 3792 if (blkno == BPERDMAP) 3793 goto initTree; 3794 3795 /* the first word beyond the end of existing blocks */ 3796 w = blkno >> L2DBWORD; 3797 3798 /* does nblocks fall on a 32-bit boundary ? */ 3799 b = blkno & (DBWORD - 1); 3800 if (b) { 3801 /* mark a partial word allocated */ 3802 dp->wmap[w] = dp->pmap[w] = cpu_to_le32(ONES >> b); 3803 w++; 3804 } 3805 3806 /* set the rest of the words in the page to allocated (ONES) */ 3807 for (i = w; i < LPERDMAP; i++) 3808 dp->pmap[i] = dp->wmap[i] = cpu_to_le32(ONES); 3809 3810 /* 3811 * init tree 3812 */ 3813 initTree: 3814 return (dbInitDmapTree(dp)); 3815 } 3816 3817 3818 /* 3819 * NAME: dbInitDmapTree()/ujfs_complete_dmap() 3820 * 3821 * FUNCTION: initialize summary tree of the specified dmap: 3822 * 3823 * at entry, bitmap of the dmap has been initialized; 3824 * 3825 * PARAMETERS: 3826 * dp - dmap to complete 3827 * blkno - starting block number for this dmap 3828 * treemax - will be filled in with max free for this dmap 3829 * 3830 * RETURNS: max free string at the root of the tree 3831 */ 3832 static int dbInitDmapTree(struct dmap * dp) 3833 { 3834 struct dmaptree *tp; 3835 s8 *cp; 3836 int i; 3837 3838 /* init fixed info of tree */ 3839 tp = &dp->tree; 3840 tp->nleafs = cpu_to_le32(LPERDMAP); 3841 tp->l2nleafs = cpu_to_le32(L2LPERDMAP); 3842 tp->leafidx = cpu_to_le32(LEAFIND); 3843 tp->height = cpu_to_le32(4); 3844 tp->budmin = BUDMIN; 3845 3846 /* init each leaf from corresponding wmap word: 3847 * note: leaf is set to NOFREE(-1) if all blocks of corresponding 3848 * bitmap word are allocated. 3849 */ 3850 cp = tp->stree + le32_to_cpu(tp->leafidx); 3851 for (i = 0; i < LPERDMAP; i++) 3852 *cp++ = dbMaxBud((u8 *) & dp->wmap[i]); 3853 3854 /* build the dmap's binary buddy summary tree */ 3855 return (dbInitTree(tp)); 3856 } 3857 3858 3859 /* 3860 * NAME: dbInitTree()/ujfs_adjtree() 3861 * 3862 * FUNCTION: initialize binary buddy summary tree of a dmap or dmapctl. 3863 * 3864 * at entry, the leaves of the tree has been initialized 3865 * from corresponding bitmap word or root of summary tree 3866 * of the child control page; 3867 * configure binary buddy system at the leaf level, then 3868 * bubble up the values of the leaf nodes up the tree. 3869 * 3870 * PARAMETERS: 3871 * cp - Pointer to the root of the tree 3872 * l2leaves- Number of leaf nodes as a power of 2 3873 * l2min - Number of blocks that can be covered by a leaf 3874 * as a power of 2 3875 * 3876 * RETURNS: max free string at the root of the tree 3877 */ 3878 static int dbInitTree(struct dmaptree * dtp) 3879 { 3880 int l2max, l2free, bsize, nextb, i; 3881 int child, parent, nparent; 3882 s8 *tp, *cp, *cp1; 3883 3884 tp = dtp->stree; 3885 3886 /* Determine the maximum free string possible for the leaves */ 3887 l2max = le32_to_cpu(dtp->l2nleafs) + dtp->budmin; 3888 3889 /* 3890 * configure the leaf level into binary buddy system 3891 * 3892 * Try to combine buddies starting with a buddy size of 1 3893 * (i.e. two leaves). At a buddy size of 1 two buddy leaves 3894 * can be combined if both buddies have a maximum free of l2min; 3895 * the combination will result in the left-most buddy leaf having 3896 * a maximum free of l2min+1. 3897 * After processing all buddies for a given size, process buddies 3898 * at the next higher buddy size (i.e. current size * 2) and 3899 * the next maximum free (current free + 1). 3900 * This continues until the maximum possible buddy combination 3901 * yields maximum free. 3902 */ 3903 for (l2free = dtp->budmin, bsize = 1; l2free < l2max; 3904 l2free++, bsize = nextb) { 3905 /* get next buddy size == current buddy pair size */ 3906 nextb = bsize << 1; 3907 3908 /* scan each adjacent buddy pair at current buddy size */ 3909 for (i = 0, cp = tp + le32_to_cpu(dtp->leafidx); 3910 i < le32_to_cpu(dtp->nleafs); 3911 i += nextb, cp += nextb) { 3912 /* coalesce if both adjacent buddies are max free */ 3913 if (*cp == l2free && *(cp + bsize) == l2free) { 3914 *cp = l2free + 1; /* left take right */ 3915 *(cp + bsize) = -1; /* right give left */ 3916 } 3917 } 3918 } 3919 3920 /* 3921 * bubble summary information of leaves up the tree. 3922 * 3923 * Starting at the leaf node level, the four nodes described by 3924 * the higher level parent node are compared for a maximum free and 3925 * this maximum becomes the value of the parent node. 3926 * when all lower level nodes are processed in this fashion then 3927 * move up to the next level (parent becomes a lower level node) and 3928 * continue the process for that level. 3929 */ 3930 for (child = le32_to_cpu(dtp->leafidx), 3931 nparent = le32_to_cpu(dtp->nleafs) >> 2; 3932 nparent > 0; nparent >>= 2, child = parent) { 3933 /* get index of 1st node of parent level */ 3934 parent = (child - 1) >> 2; 3935 3936 /* set the value of the parent node as the maximum 3937 * of the four nodes of the current level. 3938 */ 3939 for (i = 0, cp = tp + child, cp1 = tp + parent; 3940 i < nparent; i++, cp += 4, cp1++) 3941 *cp1 = TREEMAX(cp); 3942 } 3943 3944 return (*tp); 3945 } 3946 3947 3948 /* 3949 * dbInitDmapCtl() 3950 * 3951 * function: initialize dmapctl page 3952 */ 3953 static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i) 3954 { /* start leaf index not covered by range */ 3955 s8 *cp; 3956 3957 dcp->nleafs = cpu_to_le32(LPERCTL); 3958 dcp->l2nleafs = cpu_to_le32(L2LPERCTL); 3959 dcp->leafidx = cpu_to_le32(CTLLEAFIND); 3960 dcp->height = cpu_to_le32(5); 3961 dcp->budmin = L2BPERDMAP + L2LPERCTL * level; 3962 3963 /* 3964 * initialize the leaves of current level that were not covered 3965 * by the specified input block range (i.e. the leaves have no 3966 * low level dmapctl or dmap). 3967 */ 3968 cp = &dcp->stree[CTLLEAFIND + i]; 3969 for (; i < LPERCTL; i++) 3970 *cp++ = NOFREE; 3971 3972 /* build the dmap's binary buddy summary tree */ 3973 return (dbInitTree((struct dmaptree *) dcp)); 3974 } 3975 3976 3977 /* 3978 * NAME: dbGetL2AGSize()/ujfs_getagl2size() 3979 * 3980 * FUNCTION: Determine log2(allocation group size) from aggregate size 3981 * 3982 * PARAMETERS: 3983 * nblocks - Number of blocks in aggregate 3984 * 3985 * RETURNS: log2(allocation group size) in aggregate blocks 3986 */ 3987 static int dbGetL2AGSize(s64 nblocks) 3988 { 3989 s64 sz; 3990 s64 m; 3991 int l2sz; 3992 3993 if (nblocks < BPERDMAP * MAXAG) 3994 return (L2BPERDMAP); 3995 3996 /* round up aggregate size to power of 2 */ 3997 m = ((u64) 1 << (64 - 1)); 3998 for (l2sz = 64; l2sz >= 0; l2sz--, m >>= 1) { 3999 if (m & nblocks) 4000 break; 4001 } 4002 4003 sz = (s64) 1 << l2sz; 4004 if (sz < nblocks) 4005 l2sz += 1; 4006 4007 /* agsize = roundupSize/max_number_of_ag */ 4008 return (l2sz - L2MAXAG); 4009 } 4010 4011 4012 /* 4013 * NAME: dbMapFileSizeToMapSize() 4014 * 4015 * FUNCTION: compute number of blocks the block allocation map file 4016 * can cover from the map file size; 4017 * 4018 * RETURNS: Number of blocks which can be covered by this block map file; 4019 */ 4020 4021 /* 4022 * maximum number of map pages at each level including control pages 4023 */ 4024 #define MAXL0PAGES (1 + LPERCTL) 4025 #define MAXL1PAGES (1 + LPERCTL * MAXL0PAGES) 4026 4027 /* 4028 * convert number of map pages to the zero origin top dmapctl level 4029 */ 4030 #define BMAPPGTOLEV(npages) \ 4031 (((npages) <= 3 + MAXL0PAGES) ? 0 : \ 4032 ((npages) <= 2 + MAXL1PAGES) ? 1 : 2) 4033 4034 s64 dbMapFileSizeToMapSize(struct inode * ipbmap) 4035 { 4036 struct super_block *sb = ipbmap->i_sb; 4037 s64 nblocks; 4038 s64 npages, ndmaps; 4039 int level, i; 4040 int complete, factor; 4041 4042 nblocks = ipbmap->i_size >> JFS_SBI(sb)->l2bsize; 4043 npages = nblocks >> JFS_SBI(sb)->l2nbperpage; 4044 level = BMAPPGTOLEV(npages); 4045 4046 /* At each level, accumulate the number of dmap pages covered by 4047 * the number of full child levels below it; 4048 * repeat for the last incomplete child level. 4049 */ 4050 ndmaps = 0; 4051 npages--; /* skip the first global control page */ 4052 /* skip higher level control pages above top level covered by map */ 4053 npages -= (2 - level); 4054 npages--; /* skip top level's control page */ 4055 for (i = level; i >= 0; i--) { 4056 factor = 4057 (i == 2) ? MAXL1PAGES : ((i == 1) ? MAXL0PAGES : 1); 4058 complete = (u32) npages / factor; 4059 ndmaps += complete * ((i == 2) ? LPERCTL * LPERCTL : 4060 ((i == 1) ? LPERCTL : 1)); 4061 4062 /* pages in last/incomplete child */ 4063 npages = (u32) npages % factor; 4064 /* skip incomplete child's level control page */ 4065 npages--; 4066 } 4067 4068 /* convert the number of dmaps into the number of blocks 4069 * which can be covered by the dmaps; 4070 */ 4071 nblocks = ndmaps << L2BPERDMAP; 4072 4073 return (nblocks); 4074 } 4075