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