1 /* 2 * Copyright (C) International Business Machines Corp., 2000-2004 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See 12 * the GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write to the Free Software 16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 17 */ 18 19 /* 20 * jfs_imap.c: inode allocation map manager 21 * 22 * Serialization: 23 * Each AG has a simple lock which is used to control the serialization of 24 * the AG level lists. This lock should be taken first whenever an AG 25 * level list will be modified or accessed. 26 * 27 * Each IAG is locked by obtaining the buffer for the IAG page. 28 * 29 * There is also a inode lock for the inode map inode. A read lock needs to 30 * be taken whenever an IAG is read from the map or the global level 31 * information is read. A write lock needs to be taken whenever the global 32 * level information is modified or an atomic operation needs to be used. 33 * 34 * If more than one IAG is read at one time, the read lock may not 35 * be given up until all of the IAG's are read. Otherwise, a deadlock 36 * may occur when trying to obtain the read lock while another thread 37 * holding the read lock is waiting on the IAG already being held. 38 * 39 * The control page of the inode map is read into memory by diMount(). 40 * Thereafter it should only be modified in memory and then it will be 41 * written out when the filesystem is unmounted by diUnmount(). 42 */ 43 44 #include <linux/fs.h> 45 #include <linux/buffer_head.h> 46 #include <linux/pagemap.h> 47 #include <linux/quotaops.h> 48 49 #include "jfs_incore.h" 50 #include "jfs_filsys.h" 51 #include "jfs_dinode.h" 52 #include "jfs_dmap.h" 53 #include "jfs_imap.h" 54 #include "jfs_metapage.h" 55 #include "jfs_superblock.h" 56 #include "jfs_debug.h" 57 58 /* 59 * imap locks 60 */ 61 /* iag free list lock */ 62 #define IAGFREE_LOCK_INIT(imap) init_MUTEX(&imap->im_freelock) 63 #define IAGFREE_LOCK(imap) down(&imap->im_freelock) 64 #define IAGFREE_UNLOCK(imap) up(&imap->im_freelock) 65 66 /* per ag iag list locks */ 67 #define AG_LOCK_INIT(imap,index) init_MUTEX(&(imap->im_aglock[index])) 68 #define AG_LOCK(imap,agno) down(&imap->im_aglock[agno]) 69 #define AG_UNLOCK(imap,agno) up(&imap->im_aglock[agno]) 70 71 /* 72 * external references 73 */ 74 extern struct address_space_operations jfs_aops; 75 76 /* 77 * forward references 78 */ 79 static int diAllocAG(struct inomap *, int, boolean_t, struct inode *); 80 static int diAllocAny(struct inomap *, int, boolean_t, struct inode *); 81 static int diAllocBit(struct inomap *, struct iag *, int); 82 static int diAllocExt(struct inomap *, int, struct inode *); 83 static int diAllocIno(struct inomap *, int, struct inode *); 84 static int diFindFree(u32, int); 85 static int diNewExt(struct inomap *, struct iag *, int); 86 static int diNewIAG(struct inomap *, int *, int, struct metapage **); 87 static void duplicateIXtree(struct super_block *, s64, int, s64 *); 88 89 static int diIAGRead(struct inomap * imap, int, struct metapage **); 90 static int copy_from_dinode(struct dinode *, struct inode *); 91 static void copy_to_dinode(struct dinode *, struct inode *); 92 93 /* 94 * debug code for double-checking inode map 95 */ 96 /* #define _JFS_DEBUG_IMAP 1 */ 97 98 #ifdef _JFS_DEBUG_IMAP 99 #define DBG_DIINIT(imap) DBGdiInit(imap) 100 #define DBG_DIALLOC(imap, ino) DBGdiAlloc(imap, ino) 101 #define DBG_DIFREE(imap, ino) DBGdiFree(imap, ino) 102 103 static void *DBGdiInit(struct inomap * imap); 104 static void DBGdiAlloc(struct inomap * imap, ino_t ino); 105 static void DBGdiFree(struct inomap * imap, ino_t ino); 106 #else 107 #define DBG_DIINIT(imap) 108 #define DBG_DIALLOC(imap, ino) 109 #define DBG_DIFREE(imap, ino) 110 #endif /* _JFS_DEBUG_IMAP */ 111 112 /* 113 * NAME: diMount() 114 * 115 * FUNCTION: initialize the incore inode map control structures for 116 * a fileset or aggregate init time. 117 * 118 * the inode map's control structure (dinomap) is 119 * brought in from disk and placed in virtual memory. 120 * 121 * PARAMETERS: 122 * ipimap - pointer to inode map inode for the aggregate or fileset. 123 * 124 * RETURN VALUES: 125 * 0 - success 126 * -ENOMEM - insufficient free virtual memory. 127 * -EIO - i/o error. 128 */ 129 int diMount(struct inode *ipimap) 130 { 131 struct inomap *imap; 132 struct metapage *mp; 133 int index; 134 struct dinomap_disk *dinom_le; 135 136 /* 137 * allocate/initialize the in-memory inode map control structure 138 */ 139 /* allocate the in-memory inode map control structure. */ 140 imap = (struct inomap *) kmalloc(sizeof(struct inomap), GFP_KERNEL); 141 if (imap == NULL) { 142 jfs_err("diMount: kmalloc returned NULL!"); 143 return -ENOMEM; 144 } 145 146 /* read the on-disk inode map control structure. */ 147 148 mp = read_metapage(ipimap, 149 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage, 150 PSIZE, 0); 151 if (mp == NULL) { 152 kfree(imap); 153 return -EIO; 154 } 155 156 /* copy the on-disk version to the in-memory version. */ 157 dinom_le = (struct dinomap_disk *) mp->data; 158 imap->im_freeiag = le32_to_cpu(dinom_le->in_freeiag); 159 imap->im_nextiag = le32_to_cpu(dinom_le->in_nextiag); 160 atomic_set(&imap->im_numinos, le32_to_cpu(dinom_le->in_numinos)); 161 atomic_set(&imap->im_numfree, le32_to_cpu(dinom_le->in_numfree)); 162 imap->im_nbperiext = le32_to_cpu(dinom_le->in_nbperiext); 163 imap->im_l2nbperiext = le32_to_cpu(dinom_le->in_l2nbperiext); 164 for (index = 0; index < MAXAG; index++) { 165 imap->im_agctl[index].inofree = 166 le32_to_cpu(dinom_le->in_agctl[index].inofree); 167 imap->im_agctl[index].extfree = 168 le32_to_cpu(dinom_le->in_agctl[index].extfree); 169 imap->im_agctl[index].numinos = 170 le32_to_cpu(dinom_le->in_agctl[index].numinos); 171 imap->im_agctl[index].numfree = 172 le32_to_cpu(dinom_le->in_agctl[index].numfree); 173 } 174 175 /* release the buffer. */ 176 release_metapage(mp); 177 178 /* 179 * allocate/initialize inode allocation map locks 180 */ 181 /* allocate and init iag free list lock */ 182 IAGFREE_LOCK_INIT(imap); 183 184 /* allocate and init ag list locks */ 185 for (index = 0; index < MAXAG; index++) { 186 AG_LOCK_INIT(imap, index); 187 } 188 189 /* bind the inode map inode and inode map control structure 190 * to each other. 191 */ 192 imap->im_ipimap = ipimap; 193 JFS_IP(ipimap)->i_imap = imap; 194 195 // DBG_DIINIT(imap); 196 197 return (0); 198 } 199 200 201 /* 202 * NAME: diUnmount() 203 * 204 * FUNCTION: write to disk the incore inode map control structures for 205 * a fileset or aggregate at unmount time. 206 * 207 * PARAMETERS: 208 * ipimap - pointer to inode map inode for the aggregate or fileset. 209 * 210 * RETURN VALUES: 211 * 0 - success 212 * -ENOMEM - insufficient free virtual memory. 213 * -EIO - i/o error. 214 */ 215 int diUnmount(struct inode *ipimap, int mounterror) 216 { 217 struct inomap *imap = JFS_IP(ipimap)->i_imap; 218 219 /* 220 * update the on-disk inode map control structure 221 */ 222 223 if (!(mounterror || isReadOnly(ipimap))) 224 diSync(ipimap); 225 226 /* 227 * Invalidate the page cache buffers 228 */ 229 truncate_inode_pages(ipimap->i_mapping, 0); 230 231 /* 232 * free in-memory control structure 233 */ 234 kfree(imap); 235 236 return (0); 237 } 238 239 240 /* 241 * diSync() 242 */ 243 int diSync(struct inode *ipimap) 244 { 245 struct dinomap_disk *dinom_le; 246 struct inomap *imp = JFS_IP(ipimap)->i_imap; 247 struct metapage *mp; 248 int index; 249 250 /* 251 * write imap global conrol page 252 */ 253 /* read the on-disk inode map control structure */ 254 mp = get_metapage(ipimap, 255 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage, 256 PSIZE, 0); 257 if (mp == NULL) { 258 jfs_err("diSync: get_metapage failed!"); 259 return -EIO; 260 } 261 262 /* copy the in-memory version to the on-disk version */ 263 dinom_le = (struct dinomap_disk *) mp->data; 264 dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag); 265 dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag); 266 dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos)); 267 dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree)); 268 dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext); 269 dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext); 270 for (index = 0; index < MAXAG; index++) { 271 dinom_le->in_agctl[index].inofree = 272 cpu_to_le32(imp->im_agctl[index].inofree); 273 dinom_le->in_agctl[index].extfree = 274 cpu_to_le32(imp->im_agctl[index].extfree); 275 dinom_le->in_agctl[index].numinos = 276 cpu_to_le32(imp->im_agctl[index].numinos); 277 dinom_le->in_agctl[index].numfree = 278 cpu_to_le32(imp->im_agctl[index].numfree); 279 } 280 281 /* write out the control structure */ 282 write_metapage(mp); 283 284 /* 285 * write out dirty pages of imap 286 */ 287 filemap_fdatawrite(ipimap->i_mapping); 288 filemap_fdatawait(ipimap->i_mapping); 289 290 diWriteSpecial(ipimap, 0); 291 292 return (0); 293 } 294 295 296 /* 297 * NAME: diRead() 298 * 299 * FUNCTION: initialize an incore inode from disk. 300 * 301 * on entry, the specifed incore inode should itself 302 * specify the disk inode number corresponding to the 303 * incore inode (i.e. i_number should be initialized). 304 * 305 * this routine handles incore inode initialization for 306 * both "special" and "regular" inodes. special inodes 307 * are those required early in the mount process and 308 * require special handling since much of the file system 309 * is not yet initialized. these "special" inodes are 310 * identified by a NULL inode map inode pointer and are 311 * actually initialized by a call to diReadSpecial(). 312 * 313 * for regular inodes, the iag describing the disk inode 314 * is read from disk to determine the inode extent address 315 * for the disk inode. with the inode extent address in 316 * hand, the page of the extent that contains the disk 317 * inode is read and the disk inode is copied to the 318 * incore inode. 319 * 320 * PARAMETERS: 321 * ip - pointer to incore inode to be initialized from disk. 322 * 323 * RETURN VALUES: 324 * 0 - success 325 * -EIO - i/o error. 326 * -ENOMEM - insufficient memory 327 * 328 */ 329 int diRead(struct inode *ip) 330 { 331 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 332 int iagno, ino, extno, rc; 333 struct inode *ipimap; 334 struct dinode *dp; 335 struct iag *iagp; 336 struct metapage *mp; 337 s64 blkno, agstart; 338 struct inomap *imap; 339 int block_offset; 340 int inodes_left; 341 uint pageno; 342 int rel_inode; 343 344 jfs_info("diRead: ino = %ld", ip->i_ino); 345 346 ipimap = sbi->ipimap; 347 JFS_IP(ip)->ipimap = ipimap; 348 349 /* determine the iag number for this inode (number) */ 350 iagno = INOTOIAG(ip->i_ino); 351 352 /* read the iag */ 353 imap = JFS_IP(ipimap)->i_imap; 354 IREAD_LOCK(ipimap); 355 rc = diIAGRead(imap, iagno, &mp); 356 IREAD_UNLOCK(ipimap); 357 if (rc) { 358 jfs_err("diRead: diIAGRead returned %d", rc); 359 return (rc); 360 } 361 362 iagp = (struct iag *) mp->data; 363 364 /* determine inode extent that holds the disk inode */ 365 ino = ip->i_ino & (INOSPERIAG - 1); 366 extno = ino >> L2INOSPEREXT; 367 368 if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) || 369 (addressPXD(&iagp->inoext[extno]) == 0)) { 370 release_metapage(mp); 371 return -ESTALE; 372 } 373 374 /* get disk block number of the page within the inode extent 375 * that holds the disk inode. 376 */ 377 blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage); 378 379 /* get the ag for the iag */ 380 agstart = le64_to_cpu(iagp->agstart); 381 382 release_metapage(mp); 383 384 rel_inode = (ino & (INOSPERPAGE - 1)); 385 pageno = blkno >> sbi->l2nbperpage; 386 387 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) { 388 /* 389 * OS/2 didn't always align inode extents on page boundaries 390 */ 391 inodes_left = 392 (sbi->nbperpage - block_offset) << sbi->l2niperblk; 393 394 if (rel_inode < inodes_left) 395 rel_inode += block_offset << sbi->l2niperblk; 396 else { 397 pageno += 1; 398 rel_inode -= inodes_left; 399 } 400 } 401 402 /* read the page of disk inode */ 403 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1); 404 if (mp == 0) { 405 jfs_err("diRead: read_metapage failed"); 406 return -EIO; 407 } 408 409 /* locate the the disk inode requested */ 410 dp = (struct dinode *) mp->data; 411 dp += rel_inode; 412 413 if (ip->i_ino != le32_to_cpu(dp->di_number)) { 414 jfs_error(ip->i_sb, "diRead: i_ino != di_number"); 415 rc = -EIO; 416 } else if (le32_to_cpu(dp->di_nlink) == 0) 417 rc = -ESTALE; 418 else 419 /* copy the disk inode to the in-memory inode */ 420 rc = copy_from_dinode(dp, ip); 421 422 release_metapage(mp); 423 424 /* set the ag for the inode */ 425 JFS_IP(ip)->agno = BLKTOAG(agstart, sbi); 426 JFS_IP(ip)->active_ag = -1; 427 428 return (rc); 429 } 430 431 432 /* 433 * NAME: diReadSpecial() 434 * 435 * FUNCTION: initialize a 'special' inode from disk. 436 * 437 * this routines handles aggregate level inodes. The 438 * inode cache cannot differentiate between the 439 * aggregate inodes and the filesystem inodes, so we 440 * handle these here. We don't actually use the aggregate 441 * inode map, since these inodes are at a fixed location 442 * and in some cases the aggregate inode map isn't initialized 443 * yet. 444 * 445 * PARAMETERS: 446 * sb - filesystem superblock 447 * inum - aggregate inode number 448 * secondary - 1 if secondary aggregate inode table 449 * 450 * RETURN VALUES: 451 * new inode - success 452 * NULL - i/o error. 453 */ 454 struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary) 455 { 456 struct jfs_sb_info *sbi = JFS_SBI(sb); 457 uint address; 458 struct dinode *dp; 459 struct inode *ip; 460 struct metapage *mp; 461 462 ip = new_inode(sb); 463 if (ip == NULL) { 464 jfs_err("diReadSpecial: new_inode returned NULL!"); 465 return ip; 466 } 467 468 if (secondary) { 469 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage; 470 JFS_IP(ip)->ipimap = sbi->ipaimap2; 471 } else { 472 address = AITBL_OFF >> L2PSIZE; 473 JFS_IP(ip)->ipimap = sbi->ipaimap; 474 } 475 476 ASSERT(inum < INOSPEREXT); 477 478 ip->i_ino = inum; 479 480 address += inum >> 3; /* 8 inodes per 4K page */ 481 482 /* read the page of fixed disk inode (AIT) in raw mode */ 483 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1); 484 if (mp == NULL) { 485 ip->i_nlink = 1; /* Don't want iput() deleting it */ 486 iput(ip); 487 return (NULL); 488 } 489 490 /* get the pointer to the disk inode of interest */ 491 dp = (struct dinode *) (mp->data); 492 dp += inum % 8; /* 8 inodes per 4K page */ 493 494 /* copy on-disk inode to in-memory inode */ 495 if ((copy_from_dinode(dp, ip)) != 0) { 496 /* handle bad return by returning NULL for ip */ 497 ip->i_nlink = 1; /* Don't want iput() deleting it */ 498 iput(ip); 499 /* release the page */ 500 release_metapage(mp); 501 return (NULL); 502 503 } 504 505 ip->i_mapping->a_ops = &jfs_aops; 506 mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS); 507 508 /* Allocations to metadata inodes should not affect quotas */ 509 ip->i_flags |= S_NOQUOTA; 510 511 if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) { 512 sbi->gengen = le32_to_cpu(dp->di_gengen); 513 sbi->inostamp = le32_to_cpu(dp->di_inostamp); 514 } 515 516 /* release the page */ 517 release_metapage(mp); 518 519 return (ip); 520 } 521 522 /* 523 * NAME: diWriteSpecial() 524 * 525 * FUNCTION: Write the special inode to disk 526 * 527 * PARAMETERS: 528 * ip - special inode 529 * secondary - 1 if secondary aggregate inode table 530 * 531 * RETURN VALUES: none 532 */ 533 534 void diWriteSpecial(struct inode *ip, int secondary) 535 { 536 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 537 uint address; 538 struct dinode *dp; 539 ino_t inum = ip->i_ino; 540 struct metapage *mp; 541 542 ip->i_state &= ~I_DIRTY; 543 544 if (secondary) 545 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage; 546 else 547 address = AITBL_OFF >> L2PSIZE; 548 549 ASSERT(inum < INOSPEREXT); 550 551 address += inum >> 3; /* 8 inodes per 4K page */ 552 553 /* read the page of fixed disk inode (AIT) in raw mode */ 554 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1); 555 if (mp == NULL) { 556 jfs_err("diWriteSpecial: failed to read aggregate inode " 557 "extent!"); 558 return; 559 } 560 561 /* get the pointer to the disk inode of interest */ 562 dp = (struct dinode *) (mp->data); 563 dp += inum % 8; /* 8 inodes per 4K page */ 564 565 /* copy on-disk inode to in-memory inode */ 566 copy_to_dinode(dp, ip); 567 memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288); 568 569 if (inum == FILESYSTEM_I) 570 dp->di_gengen = cpu_to_le32(sbi->gengen); 571 572 /* write the page */ 573 write_metapage(mp); 574 } 575 576 /* 577 * NAME: diFreeSpecial() 578 * 579 * FUNCTION: Free allocated space for special inode 580 */ 581 void diFreeSpecial(struct inode *ip) 582 { 583 if (ip == NULL) { 584 jfs_err("diFreeSpecial called with NULL ip!"); 585 return; 586 } 587 filemap_fdatawrite(ip->i_mapping); 588 filemap_fdatawait(ip->i_mapping); 589 truncate_inode_pages(ip->i_mapping, 0); 590 iput(ip); 591 } 592 593 594 595 /* 596 * NAME: diWrite() 597 * 598 * FUNCTION: write the on-disk inode portion of the in-memory inode 599 * to its corresponding on-disk inode. 600 * 601 * on entry, the specifed incore inode should itself 602 * specify the disk inode number corresponding to the 603 * incore inode (i.e. i_number should be initialized). 604 * 605 * the inode contains the inode extent address for the disk 606 * inode. with the inode extent address in hand, the 607 * page of the extent that contains the disk inode is 608 * read and the disk inode portion of the incore inode 609 * is copied to the disk inode. 610 * 611 * PARAMETERS: 612 * tid - transacation id 613 * ip - pointer to incore inode to be written to the inode extent. 614 * 615 * RETURN VALUES: 616 * 0 - success 617 * -EIO - i/o error. 618 */ 619 int diWrite(tid_t tid, struct inode *ip) 620 { 621 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 622 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 623 int rc = 0; 624 s32 ino; 625 struct dinode *dp; 626 s64 blkno; 627 int block_offset; 628 int inodes_left; 629 struct metapage *mp; 630 uint pageno; 631 int rel_inode; 632 int dioffset; 633 struct inode *ipimap; 634 uint type; 635 lid_t lid; 636 struct tlock *ditlck, *tlck; 637 struct linelock *dilinelock, *ilinelock; 638 struct lv *lv; 639 int n; 640 641 ipimap = jfs_ip->ipimap; 642 643 ino = ip->i_ino & (INOSPERIAG - 1); 644 645 if (!addressPXD(&(jfs_ip->ixpxd)) || 646 (lengthPXD(&(jfs_ip->ixpxd)) != 647 JFS_IP(ipimap)->i_imap->im_nbperiext)) { 648 jfs_error(ip->i_sb, "diWrite: ixpxd invalid"); 649 return -EIO; 650 } 651 652 /* 653 * read the page of disk inode containing the specified inode: 654 */ 655 /* compute the block address of the page */ 656 blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage); 657 658 rel_inode = (ino & (INOSPERPAGE - 1)); 659 pageno = blkno >> sbi->l2nbperpage; 660 661 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) { 662 /* 663 * OS/2 didn't always align inode extents on page boundaries 664 */ 665 inodes_left = 666 (sbi->nbperpage - block_offset) << sbi->l2niperblk; 667 668 if (rel_inode < inodes_left) 669 rel_inode += block_offset << sbi->l2niperblk; 670 else { 671 pageno += 1; 672 rel_inode -= inodes_left; 673 } 674 } 675 /* read the page of disk inode */ 676 retry: 677 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1); 678 if (mp == 0) 679 return -EIO; 680 681 /* get the pointer to the disk inode */ 682 dp = (struct dinode *) mp->data; 683 dp += rel_inode; 684 685 dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE; 686 687 /* 688 * acquire transaction lock on the on-disk inode; 689 * N.B. tlock is acquired on ipimap not ip; 690 */ 691 if ((ditlck = 692 txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL) 693 goto retry; 694 dilinelock = (struct linelock *) & ditlck->lock; 695 696 /* 697 * copy btree root from in-memory inode to on-disk inode 698 * 699 * (tlock is taken from inline B+-tree root in in-memory 700 * inode when the B+-tree root is updated, which is pointed 701 * by jfs_ip->blid as well as being on tx tlock list) 702 * 703 * further processing of btree root is based on the copy 704 * in in-memory inode, where txLog() will log from, and, 705 * for xtree root, txUpdateMap() will update map and reset 706 * XAD_NEW bit; 707 */ 708 709 if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) { 710 /* 711 * This is the special xtree inside the directory for storing 712 * the directory table 713 */ 714 xtpage_t *p, *xp; 715 xad_t *xad; 716 717 jfs_ip->xtlid = 0; 718 tlck = lid_to_tlock(lid); 719 assert(tlck->type & tlckXTREE); 720 tlck->type |= tlckBTROOT; 721 tlck->mp = mp; 722 ilinelock = (struct linelock *) & tlck->lock; 723 724 /* 725 * copy xtree root from inode to dinode: 726 */ 727 p = &jfs_ip->i_xtroot; 728 xp = (xtpage_t *) &dp->di_dirtable; 729 lv = ilinelock->lv; 730 for (n = 0; n < ilinelock->index; n++, lv++) { 731 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset], 732 lv->length << L2XTSLOTSIZE); 733 } 734 735 /* reset on-disk (metadata page) xtree XAD_NEW bit */ 736 xad = &xp->xad[XTENTRYSTART]; 737 for (n = XTENTRYSTART; 738 n < le16_to_cpu(xp->header.nextindex); n++, xad++) 739 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) 740 xad->flag &= ~(XAD_NEW | XAD_EXTENDED); 741 } 742 743 if ((lid = jfs_ip->blid) == 0) 744 goto inlineData; 745 jfs_ip->blid = 0; 746 747 tlck = lid_to_tlock(lid); 748 type = tlck->type; 749 tlck->type |= tlckBTROOT; 750 tlck->mp = mp; 751 ilinelock = (struct linelock *) & tlck->lock; 752 753 /* 754 * regular file: 16 byte (XAD slot) granularity 755 */ 756 if (type & tlckXTREE) { 757 xtpage_t *p, *xp; 758 xad_t *xad; 759 760 /* 761 * copy xtree root from inode to dinode: 762 */ 763 p = &jfs_ip->i_xtroot; 764 xp = &dp->di_xtroot; 765 lv = ilinelock->lv; 766 for (n = 0; n < ilinelock->index; n++, lv++) { 767 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset], 768 lv->length << L2XTSLOTSIZE); 769 } 770 771 /* reset on-disk (metadata page) xtree XAD_NEW bit */ 772 xad = &xp->xad[XTENTRYSTART]; 773 for (n = XTENTRYSTART; 774 n < le16_to_cpu(xp->header.nextindex); n++, xad++) 775 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) 776 xad->flag &= ~(XAD_NEW | XAD_EXTENDED); 777 } 778 /* 779 * directory: 32 byte (directory entry slot) granularity 780 */ 781 else if (type & tlckDTREE) { 782 dtpage_t *p, *xp; 783 784 /* 785 * copy dtree root from inode to dinode: 786 */ 787 p = (dtpage_t *) &jfs_ip->i_dtroot; 788 xp = (dtpage_t *) & dp->di_dtroot; 789 lv = ilinelock->lv; 790 for (n = 0; n < ilinelock->index; n++, lv++) { 791 memcpy(&xp->slot[lv->offset], &p->slot[lv->offset], 792 lv->length << L2DTSLOTSIZE); 793 } 794 } else { 795 jfs_err("diWrite: UFO tlock"); 796 } 797 798 inlineData: 799 /* 800 * copy inline symlink from in-memory inode to on-disk inode 801 */ 802 if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) { 803 lv = & dilinelock->lv[dilinelock->index]; 804 lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE; 805 lv->length = 2; 806 memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE); 807 dilinelock->index++; 808 } 809 /* 810 * copy inline data from in-memory inode to on-disk inode: 811 * 128 byte slot granularity 812 */ 813 if (test_cflag(COMMIT_Inlineea, ip)) { 814 lv = & dilinelock->lv[dilinelock->index]; 815 lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE; 816 lv->length = 1; 817 memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE); 818 dilinelock->index++; 819 820 clear_cflag(COMMIT_Inlineea, ip); 821 } 822 823 /* 824 * lock/copy inode base: 128 byte slot granularity 825 */ 826 // baseDinode: 827 lv = & dilinelock->lv[dilinelock->index]; 828 lv->offset = dioffset >> L2INODESLOTSIZE; 829 copy_to_dinode(dp, ip); 830 if (test_and_clear_cflag(COMMIT_Dirtable, ip)) { 831 lv->length = 2; 832 memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96); 833 } else 834 lv->length = 1; 835 dilinelock->index++; 836 837 #ifdef _JFS_FASTDASD 838 /* 839 * We aren't logging changes to the DASD used in directory inodes, 840 * but we need to write them to disk. If we don't unmount cleanly, 841 * mount will recalculate the DASD used. 842 */ 843 if (S_ISDIR(ip->i_mode) 844 && (ip->i_ipmnt->i_mntflag & JFS_DASD_ENABLED)) 845 memcpy(&dp->di_DASD, &ip->i_DASD, sizeof(struct dasd)); 846 #endif /* _JFS_FASTDASD */ 847 848 /* release the buffer holding the updated on-disk inode. 849 * the buffer will be later written by commit processing. 850 */ 851 write_metapage(mp); 852 853 return (rc); 854 } 855 856 857 /* 858 * NAME: diFree(ip) 859 * 860 * FUNCTION: free a specified inode from the inode working map 861 * for a fileset or aggregate. 862 * 863 * if the inode to be freed represents the first (only) 864 * free inode within the iag, the iag will be placed on 865 * the ag free inode list. 866 * 867 * freeing the inode will cause the inode extent to be 868 * freed if the inode is the only allocated inode within 869 * the extent. in this case all the disk resource backing 870 * up the inode extent will be freed. in addition, the iag 871 * will be placed on the ag extent free list if the extent 872 * is the first free extent in the iag. if freeing the 873 * extent also means that no free inodes will exist for 874 * the iag, the iag will also be removed from the ag free 875 * inode list. 876 * 877 * the iag describing the inode will be freed if the extent 878 * is to be freed and it is the only backed extent within 879 * the iag. in this case, the iag will be removed from the 880 * ag free extent list and ag free inode list and placed on 881 * the inode map's free iag list. 882 * 883 * a careful update approach is used to provide consistency 884 * in the face of updates to multiple buffers. under this 885 * approach, all required buffers are obtained before making 886 * any updates and are held until all updates are complete. 887 * 888 * PARAMETERS: 889 * ip - inode to be freed. 890 * 891 * RETURN VALUES: 892 * 0 - success 893 * -EIO - i/o error. 894 */ 895 int diFree(struct inode *ip) 896 { 897 int rc; 898 ino_t inum = ip->i_ino; 899 struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp; 900 struct metapage *mp, *amp, *bmp, *cmp, *dmp; 901 int iagno, ino, extno, bitno, sword, agno; 902 int back, fwd; 903 u32 bitmap, mask; 904 struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap; 905 struct inomap *imap = JFS_IP(ipimap)->i_imap; 906 pxd_t freepxd; 907 tid_t tid; 908 struct inode *iplist[3]; 909 struct tlock *tlck; 910 struct pxd_lock *pxdlock; 911 912 /* 913 * This is just to suppress compiler warnings. The same logic that 914 * references these variables is used to initialize them. 915 */ 916 aiagp = biagp = ciagp = diagp = NULL; 917 918 /* get the iag number containing the inode. 919 */ 920 iagno = INOTOIAG(inum); 921 922 /* make sure that the iag is contained within 923 * the map. 924 */ 925 if (iagno >= imap->im_nextiag) { 926 dump_mem("imap", imap, 32); 927 jfs_error(ip->i_sb, 928 "diFree: inum = %d, iagno = %d, nextiag = %d", 929 (uint) inum, iagno, imap->im_nextiag); 930 return -EIO; 931 } 932 933 /* get the allocation group for this ino. 934 */ 935 agno = JFS_IP(ip)->agno; 936 937 /* Lock the AG specific inode map information 938 */ 939 AG_LOCK(imap, agno); 940 941 /* Obtain read lock in imap inode. Don't release it until we have 942 * read all of the IAG's that we are going to. 943 */ 944 IREAD_LOCK(ipimap); 945 946 /* read the iag. 947 */ 948 if ((rc = diIAGRead(imap, iagno, &mp))) { 949 IREAD_UNLOCK(ipimap); 950 AG_UNLOCK(imap, agno); 951 return (rc); 952 } 953 iagp = (struct iag *) mp->data; 954 955 /* get the inode number and extent number of the inode within 956 * the iag and the inode number within the extent. 957 */ 958 ino = inum & (INOSPERIAG - 1); 959 extno = ino >> L2INOSPEREXT; 960 bitno = ino & (INOSPEREXT - 1); 961 mask = HIGHORDER >> bitno; 962 963 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { 964 jfs_error(ip->i_sb, 965 "diFree: wmap shows inode already free"); 966 } 967 968 if (!addressPXD(&iagp->inoext[extno])) { 969 release_metapage(mp); 970 IREAD_UNLOCK(ipimap); 971 AG_UNLOCK(imap, agno); 972 jfs_error(ip->i_sb, "diFree: invalid inoext"); 973 return -EIO; 974 } 975 976 /* compute the bitmap for the extent reflecting the freed inode. 977 */ 978 bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask; 979 980 if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) { 981 release_metapage(mp); 982 IREAD_UNLOCK(ipimap); 983 AG_UNLOCK(imap, agno); 984 jfs_error(ip->i_sb, "diFree: numfree > numinos"); 985 return -EIO; 986 } 987 /* 988 * inode extent still has some inodes or below low water mark: 989 * keep the inode extent; 990 */ 991 if (bitmap || 992 imap->im_agctl[agno].numfree < 96 || 993 (imap->im_agctl[agno].numfree < 288 && 994 (((imap->im_agctl[agno].numfree * 100) / 995 imap->im_agctl[agno].numinos) <= 25))) { 996 /* if the iag currently has no free inodes (i.e., 997 * the inode being freed is the first free inode of iag), 998 * insert the iag at head of the inode free list for the ag. 999 */ 1000 if (iagp->nfreeinos == 0) { 1001 /* check if there are any iags on the ag inode 1002 * free list. if so, read the first one so that 1003 * we can link the current iag onto the list at 1004 * the head. 1005 */ 1006 if ((fwd = imap->im_agctl[agno].inofree) >= 0) { 1007 /* read the iag that currently is the head 1008 * of the list. 1009 */ 1010 if ((rc = diIAGRead(imap, fwd, &))) { 1011 IREAD_UNLOCK(ipimap); 1012 AG_UNLOCK(imap, agno); 1013 release_metapage(mp); 1014 return (rc); 1015 } 1016 aiagp = (struct iag *) amp->data; 1017 1018 /* make current head point back to the iag. 1019 */ 1020 aiagp->inofreeback = cpu_to_le32(iagno); 1021 1022 write_metapage(amp); 1023 } 1024 1025 /* iag points forward to current head and iag 1026 * becomes the new head of the list. 1027 */ 1028 iagp->inofreefwd = 1029 cpu_to_le32(imap->im_agctl[agno].inofree); 1030 iagp->inofreeback = cpu_to_le32(-1); 1031 imap->im_agctl[agno].inofree = iagno; 1032 } 1033 IREAD_UNLOCK(ipimap); 1034 1035 /* update the free inode summary map for the extent if 1036 * freeing the inode means the extent will now have free 1037 * inodes (i.e., the inode being freed is the first free 1038 * inode of extent), 1039 */ 1040 if (iagp->wmap[extno] == cpu_to_le32(ONES)) { 1041 sword = extno >> L2EXTSPERSUM; 1042 bitno = extno & (EXTSPERSUM - 1); 1043 iagp->inosmap[sword] &= 1044 cpu_to_le32(~(HIGHORDER >> bitno)); 1045 } 1046 1047 /* update the bitmap. 1048 */ 1049 iagp->wmap[extno] = cpu_to_le32(bitmap); 1050 DBG_DIFREE(imap, inum); 1051 1052 /* update the free inode counts at the iag, ag and 1053 * map level. 1054 */ 1055 iagp->nfreeinos = 1056 cpu_to_le32(le32_to_cpu(iagp->nfreeinos) + 1); 1057 imap->im_agctl[agno].numfree += 1; 1058 atomic_inc(&imap->im_numfree); 1059 1060 /* release the AG inode map lock 1061 */ 1062 AG_UNLOCK(imap, agno); 1063 1064 /* write the iag */ 1065 write_metapage(mp); 1066 1067 return (0); 1068 } 1069 1070 1071 /* 1072 * inode extent has become free and above low water mark: 1073 * free the inode extent; 1074 */ 1075 1076 /* 1077 * prepare to update iag list(s) (careful update step 1) 1078 */ 1079 amp = bmp = cmp = dmp = NULL; 1080 fwd = back = -1; 1081 1082 /* check if the iag currently has no free extents. if so, 1083 * it will be placed on the head of the ag extent free list. 1084 */ 1085 if (iagp->nfreeexts == 0) { 1086 /* check if the ag extent free list has any iags. 1087 * if so, read the iag at the head of the list now. 1088 * this (head) iag will be updated later to reflect 1089 * the addition of the current iag at the head of 1090 * the list. 1091 */ 1092 if ((fwd = imap->im_agctl[agno].extfree) >= 0) { 1093 if ((rc = diIAGRead(imap, fwd, &))) 1094 goto error_out; 1095 aiagp = (struct iag *) amp->data; 1096 } 1097 } else { 1098 /* iag has free extents. check if the addition of a free 1099 * extent will cause all extents to be free within this 1100 * iag. if so, the iag will be removed from the ag extent 1101 * free list and placed on the inode map's free iag list. 1102 */ 1103 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) { 1104 /* in preparation for removing the iag from the 1105 * ag extent free list, read the iags preceeding 1106 * and following the iag on the ag extent free 1107 * list. 1108 */ 1109 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) { 1110 if ((rc = diIAGRead(imap, fwd, &))) 1111 goto error_out; 1112 aiagp = (struct iag *) amp->data; 1113 } 1114 1115 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) { 1116 if ((rc = diIAGRead(imap, back, &bmp))) 1117 goto error_out; 1118 biagp = (struct iag *) bmp->data; 1119 } 1120 } 1121 } 1122 1123 /* remove the iag from the ag inode free list if freeing 1124 * this extent cause the iag to have no free inodes. 1125 */ 1126 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) { 1127 int inofreeback = le32_to_cpu(iagp->inofreeback); 1128 int inofreefwd = le32_to_cpu(iagp->inofreefwd); 1129 1130 /* in preparation for removing the iag from the 1131 * ag inode free list, read the iags preceeding 1132 * and following the iag on the ag inode free 1133 * list. before reading these iags, we must make 1134 * sure that we already don't have them in hand 1135 * from up above, since re-reading an iag (buffer) 1136 * we are currently holding would cause a deadlock. 1137 */ 1138 if (inofreefwd >= 0) { 1139 1140 if (inofreefwd == fwd) 1141 ciagp = (struct iag *) amp->data; 1142 else if (inofreefwd == back) 1143 ciagp = (struct iag *) bmp->data; 1144 else { 1145 if ((rc = 1146 diIAGRead(imap, inofreefwd, &cmp))) 1147 goto error_out; 1148 ciagp = (struct iag *) cmp->data; 1149 } 1150 assert(ciagp != NULL); 1151 } 1152 1153 if (inofreeback >= 0) { 1154 if (inofreeback == fwd) 1155 diagp = (struct iag *) amp->data; 1156 else if (inofreeback == back) 1157 diagp = (struct iag *) bmp->data; 1158 else { 1159 if ((rc = 1160 diIAGRead(imap, inofreeback, &dmp))) 1161 goto error_out; 1162 diagp = (struct iag *) dmp->data; 1163 } 1164 assert(diagp != NULL); 1165 } 1166 } 1167 1168 IREAD_UNLOCK(ipimap); 1169 1170 /* 1171 * invalidate any page of the inode extent freed from buffer cache; 1172 */ 1173 freepxd = iagp->inoext[extno]; 1174 invalidate_pxd_metapages(ip, freepxd); 1175 1176 /* 1177 * update iag list(s) (careful update step 2) 1178 */ 1179 /* add the iag to the ag extent free list if this is the 1180 * first free extent for the iag. 1181 */ 1182 if (iagp->nfreeexts == 0) { 1183 if (fwd >= 0) 1184 aiagp->extfreeback = cpu_to_le32(iagno); 1185 1186 iagp->extfreefwd = 1187 cpu_to_le32(imap->im_agctl[agno].extfree); 1188 iagp->extfreeback = cpu_to_le32(-1); 1189 imap->im_agctl[agno].extfree = iagno; 1190 } else { 1191 /* remove the iag from the ag extent list if all extents 1192 * are now free and place it on the inode map iag free list. 1193 */ 1194 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) { 1195 if (fwd >= 0) 1196 aiagp->extfreeback = iagp->extfreeback; 1197 1198 if (back >= 0) 1199 biagp->extfreefwd = iagp->extfreefwd; 1200 else 1201 imap->im_agctl[agno].extfree = 1202 le32_to_cpu(iagp->extfreefwd); 1203 1204 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); 1205 1206 IAGFREE_LOCK(imap); 1207 iagp->iagfree = cpu_to_le32(imap->im_freeiag); 1208 imap->im_freeiag = iagno; 1209 IAGFREE_UNLOCK(imap); 1210 } 1211 } 1212 1213 /* remove the iag from the ag inode free list if freeing 1214 * this extent causes the iag to have no free inodes. 1215 */ 1216 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) { 1217 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) 1218 ciagp->inofreeback = iagp->inofreeback; 1219 1220 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) 1221 diagp->inofreefwd = iagp->inofreefwd; 1222 else 1223 imap->im_agctl[agno].inofree = 1224 le32_to_cpu(iagp->inofreefwd); 1225 1226 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); 1227 } 1228 1229 /* update the inode extent address and working map 1230 * to reflect the free extent. 1231 * the permanent map should have been updated already 1232 * for the inode being freed. 1233 */ 1234 if (iagp->pmap[extno] != 0) { 1235 jfs_error(ip->i_sb, "diFree: the pmap does not show inode free"); 1236 } 1237 iagp->wmap[extno] = 0; 1238 DBG_DIFREE(imap, inum); 1239 PXDlength(&iagp->inoext[extno], 0); 1240 PXDaddress(&iagp->inoext[extno], 0); 1241 1242 /* update the free extent and free inode summary maps 1243 * to reflect the freed extent. 1244 * the inode summary map is marked to indicate no inodes 1245 * available for the freed extent. 1246 */ 1247 sword = extno >> L2EXTSPERSUM; 1248 bitno = extno & (EXTSPERSUM - 1); 1249 mask = HIGHORDER >> bitno; 1250 iagp->inosmap[sword] |= cpu_to_le32(mask); 1251 iagp->extsmap[sword] &= cpu_to_le32(~mask); 1252 1253 /* update the number of free inodes and number of free extents 1254 * for the iag. 1255 */ 1256 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) - 1257 (INOSPEREXT - 1)); 1258 iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) + 1); 1259 1260 /* update the number of free inodes and backed inodes 1261 * at the ag and inode map level. 1262 */ 1263 imap->im_agctl[agno].numfree -= (INOSPEREXT - 1); 1264 imap->im_agctl[agno].numinos -= INOSPEREXT; 1265 atomic_sub(INOSPEREXT - 1, &imap->im_numfree); 1266 atomic_sub(INOSPEREXT, &imap->im_numinos); 1267 1268 if (amp) 1269 write_metapage(amp); 1270 if (bmp) 1271 write_metapage(bmp); 1272 if (cmp) 1273 write_metapage(cmp); 1274 if (dmp) 1275 write_metapage(dmp); 1276 1277 /* 1278 * start transaction to update block allocation map 1279 * for the inode extent freed; 1280 * 1281 * N.B. AG_LOCK is released and iag will be released below, and 1282 * other thread may allocate inode from/reusing the ixad freed 1283 * BUT with new/different backing inode extent from the extent 1284 * to be freed by the transaction; 1285 */ 1286 tid = txBegin(ipimap->i_sb, COMMIT_FORCE); 1287 down(&JFS_IP(ipimap)->commit_sem); 1288 1289 /* acquire tlock of the iag page of the freed ixad 1290 * to force the page NOHOMEOK (even though no data is 1291 * logged from the iag page) until NOREDOPAGE|FREEXTENT log 1292 * for the free of the extent is committed; 1293 * write FREEXTENT|NOREDOPAGE log record 1294 * N.B. linelock is overlaid as freed extent descriptor; 1295 */ 1296 tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE); 1297 pxdlock = (struct pxd_lock *) & tlck->lock; 1298 pxdlock->flag = mlckFREEPXD; 1299 pxdlock->pxd = freepxd; 1300 pxdlock->index = 1; 1301 1302 write_metapage(mp); 1303 1304 iplist[0] = ipimap; 1305 1306 /* 1307 * logredo needs the IAG number and IAG extent index in order 1308 * to ensure that the IMap is consistent. The least disruptive 1309 * way to pass these values through to the transaction manager 1310 * is in the iplist array. 1311 * 1312 * It's not pretty, but it works. 1313 */ 1314 iplist[1] = (struct inode *) (size_t)iagno; 1315 iplist[2] = (struct inode *) (size_t)extno; 1316 1317 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE); 1318 1319 txEnd(tid); 1320 up(&JFS_IP(ipimap)->commit_sem); 1321 1322 /* unlock the AG inode map information */ 1323 AG_UNLOCK(imap, agno); 1324 1325 return (0); 1326 1327 error_out: 1328 IREAD_UNLOCK(ipimap); 1329 1330 if (amp) 1331 release_metapage(amp); 1332 if (bmp) 1333 release_metapage(bmp); 1334 if (cmp) 1335 release_metapage(cmp); 1336 if (dmp) 1337 release_metapage(dmp); 1338 1339 AG_UNLOCK(imap, agno); 1340 1341 release_metapage(mp); 1342 1343 return (rc); 1344 } 1345 1346 /* 1347 * There are several places in the diAlloc* routines where we initialize 1348 * the inode. 1349 */ 1350 static inline void 1351 diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp) 1352 { 1353 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 1354 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 1355 1356 ip->i_ino = (iagno << L2INOSPERIAG) + ino; 1357 DBG_DIALLOC(JFS_IP(ipimap)->i_imap, ip->i_ino); 1358 jfs_ip->ixpxd = iagp->inoext[extno]; 1359 jfs_ip->agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi); 1360 jfs_ip->active_ag = -1; 1361 } 1362 1363 1364 /* 1365 * NAME: diAlloc(pip,dir,ip) 1366 * 1367 * FUNCTION: allocate a disk inode from the inode working map 1368 * for a fileset or aggregate. 1369 * 1370 * PARAMETERS: 1371 * pip - pointer to incore inode for the parent inode. 1372 * dir - TRUE if the new disk inode is for a directory. 1373 * ip - pointer to a new inode 1374 * 1375 * RETURN VALUES: 1376 * 0 - success. 1377 * -ENOSPC - insufficient disk resources. 1378 * -EIO - i/o error. 1379 */ 1380 int diAlloc(struct inode *pip, boolean_t dir, struct inode *ip) 1381 { 1382 int rc, ino, iagno, addext, extno, bitno, sword; 1383 int nwords, rem, i, agno; 1384 u32 mask, inosmap, extsmap; 1385 struct inode *ipimap; 1386 struct metapage *mp; 1387 ino_t inum; 1388 struct iag *iagp; 1389 struct inomap *imap; 1390 1391 /* get the pointers to the inode map inode and the 1392 * corresponding imap control structure. 1393 */ 1394 ipimap = JFS_SBI(pip->i_sb)->ipimap; 1395 imap = JFS_IP(ipimap)->i_imap; 1396 JFS_IP(ip)->ipimap = ipimap; 1397 JFS_IP(ip)->fileset = FILESYSTEM_I; 1398 1399 /* for a directory, the allocation policy is to start 1400 * at the ag level using the preferred ag. 1401 */ 1402 if (dir == TRUE) { 1403 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap); 1404 AG_LOCK(imap, agno); 1405 goto tryag; 1406 } 1407 1408 /* for files, the policy starts off by trying to allocate from 1409 * the same iag containing the parent disk inode: 1410 * try to allocate the new disk inode close to the parent disk 1411 * inode, using parent disk inode number + 1 as the allocation 1412 * hint. (we use a left-to-right policy to attempt to avoid 1413 * moving backward on the disk.) compute the hint within the 1414 * file system and the iag. 1415 */ 1416 1417 /* get the ag number of this iag */ 1418 agno = JFS_IP(pip)->agno; 1419 1420 if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) { 1421 /* 1422 * There is an open file actively growing. We want to 1423 * allocate new inodes from a different ag to avoid 1424 * fragmentation problems. 1425 */ 1426 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap); 1427 AG_LOCK(imap, agno); 1428 goto tryag; 1429 } 1430 1431 inum = pip->i_ino + 1; 1432 ino = inum & (INOSPERIAG - 1); 1433 1434 /* back off the the hint if it is outside of the iag */ 1435 if (ino == 0) 1436 inum = pip->i_ino; 1437 1438 /* lock the AG inode map information */ 1439 AG_LOCK(imap, agno); 1440 1441 /* Get read lock on imap inode */ 1442 IREAD_LOCK(ipimap); 1443 1444 /* get the iag number and read the iag */ 1445 iagno = INOTOIAG(inum); 1446 if ((rc = diIAGRead(imap, iagno, &mp))) { 1447 IREAD_UNLOCK(ipimap); 1448 AG_UNLOCK(imap, agno); 1449 return (rc); 1450 } 1451 iagp = (struct iag *) mp->data; 1452 1453 /* determine if new inode extent is allowed to be added to the iag. 1454 * new inode extent can be added to the iag if the ag 1455 * has less than 32 free disk inodes and the iag has free extents. 1456 */ 1457 addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts); 1458 1459 /* 1460 * try to allocate from the IAG 1461 */ 1462 /* check if the inode may be allocated from the iag 1463 * (i.e. the inode has free inodes or new extent can be added). 1464 */ 1465 if (iagp->nfreeinos || addext) { 1466 /* determine the extent number of the hint. 1467 */ 1468 extno = ino >> L2INOSPEREXT; 1469 1470 /* check if the extent containing the hint has backed 1471 * inodes. if so, try to allocate within this extent. 1472 */ 1473 if (addressPXD(&iagp->inoext[extno])) { 1474 bitno = ino & (INOSPEREXT - 1); 1475 if ((bitno = 1476 diFindFree(le32_to_cpu(iagp->wmap[extno]), 1477 bitno)) 1478 < INOSPEREXT) { 1479 ino = (extno << L2INOSPEREXT) + bitno; 1480 1481 /* a free inode (bit) was found within this 1482 * extent, so allocate it. 1483 */ 1484 rc = diAllocBit(imap, iagp, ino); 1485 IREAD_UNLOCK(ipimap); 1486 if (rc) { 1487 assert(rc == -EIO); 1488 } else { 1489 /* set the results of the allocation 1490 * and write the iag. 1491 */ 1492 diInitInode(ip, iagno, ino, extno, 1493 iagp); 1494 mark_metapage_dirty(mp); 1495 } 1496 release_metapage(mp); 1497 1498 /* free the AG lock and return. 1499 */ 1500 AG_UNLOCK(imap, agno); 1501 return (rc); 1502 } 1503 1504 if (!addext) 1505 extno = 1506 (extno == 1507 EXTSPERIAG - 1) ? 0 : extno + 1; 1508 } 1509 1510 /* 1511 * no free inodes within the extent containing the hint. 1512 * 1513 * try to allocate from the backed extents following 1514 * hint or, if appropriate (i.e. addext is true), allocate 1515 * an extent of free inodes at or following the extent 1516 * containing the hint. 1517 * 1518 * the free inode and free extent summary maps are used 1519 * here, so determine the starting summary map position 1520 * and the number of words we'll have to examine. again, 1521 * the approach is to allocate following the hint, so we 1522 * might have to initially ignore prior bits of the summary 1523 * map that represent extents prior to the extent containing 1524 * the hint and later revisit these bits. 1525 */ 1526 bitno = extno & (EXTSPERSUM - 1); 1527 nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1; 1528 sword = extno >> L2EXTSPERSUM; 1529 1530 /* mask any prior bits for the starting words of the 1531 * summary map. 1532 */ 1533 mask = ONES << (EXTSPERSUM - bitno); 1534 inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask; 1535 extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask; 1536 1537 /* scan the free inode and free extent summary maps for 1538 * free resources. 1539 */ 1540 for (i = 0; i < nwords; i++) { 1541 /* check if this word of the free inode summary 1542 * map describes an extent with free inodes. 1543 */ 1544 if (~inosmap) { 1545 /* an extent with free inodes has been 1546 * found. determine the extent number 1547 * and the inode number within the extent. 1548 */ 1549 rem = diFindFree(inosmap, 0); 1550 extno = (sword << L2EXTSPERSUM) + rem; 1551 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 1552 0); 1553 if (rem >= INOSPEREXT) { 1554 IREAD_UNLOCK(ipimap); 1555 release_metapage(mp); 1556 AG_UNLOCK(imap, agno); 1557 jfs_error(ip->i_sb, 1558 "diAlloc: can't find free bit " 1559 "in wmap"); 1560 return EIO; 1561 } 1562 1563 /* determine the inode number within the 1564 * iag and allocate the inode from the 1565 * map. 1566 */ 1567 ino = (extno << L2INOSPEREXT) + rem; 1568 rc = diAllocBit(imap, iagp, ino); 1569 IREAD_UNLOCK(ipimap); 1570 if (rc) 1571 assert(rc == -EIO); 1572 else { 1573 /* set the results of the allocation 1574 * and write the iag. 1575 */ 1576 diInitInode(ip, iagno, ino, extno, 1577 iagp); 1578 mark_metapage_dirty(mp); 1579 } 1580 release_metapage(mp); 1581 1582 /* free the AG lock and return. 1583 */ 1584 AG_UNLOCK(imap, agno); 1585 return (rc); 1586 1587 } 1588 1589 /* check if we may allocate an extent of free 1590 * inodes and whether this word of the free 1591 * extents summary map describes a free extent. 1592 */ 1593 if (addext && ~extsmap) { 1594 /* a free extent has been found. determine 1595 * the extent number. 1596 */ 1597 rem = diFindFree(extsmap, 0); 1598 extno = (sword << L2EXTSPERSUM) + rem; 1599 1600 /* allocate an extent of free inodes. 1601 */ 1602 if ((rc = diNewExt(imap, iagp, extno))) { 1603 /* if there is no disk space for a 1604 * new extent, try to allocate the 1605 * disk inode from somewhere else. 1606 */ 1607 if (rc == -ENOSPC) 1608 break; 1609 1610 assert(rc == -EIO); 1611 } else { 1612 /* set the results of the allocation 1613 * and write the iag. 1614 */ 1615 diInitInode(ip, iagno, 1616 extno << L2INOSPEREXT, 1617 extno, iagp); 1618 mark_metapage_dirty(mp); 1619 } 1620 release_metapage(mp); 1621 /* free the imap inode & the AG lock & return. 1622 */ 1623 IREAD_UNLOCK(ipimap); 1624 AG_UNLOCK(imap, agno); 1625 return (rc); 1626 } 1627 1628 /* move on to the next set of summary map words. 1629 */ 1630 sword = (sword == SMAPSZ - 1) ? 0 : sword + 1; 1631 inosmap = le32_to_cpu(iagp->inosmap[sword]); 1632 extsmap = le32_to_cpu(iagp->extsmap[sword]); 1633 } 1634 } 1635 /* unlock imap inode */ 1636 IREAD_UNLOCK(ipimap); 1637 1638 /* nothing doing in this iag, so release it. */ 1639 release_metapage(mp); 1640 1641 tryag: 1642 /* 1643 * try to allocate anywhere within the same AG as the parent inode. 1644 */ 1645 rc = diAllocAG(imap, agno, dir, ip); 1646 1647 AG_UNLOCK(imap, agno); 1648 1649 if (rc != -ENOSPC) 1650 return (rc); 1651 1652 /* 1653 * try to allocate in any AG. 1654 */ 1655 return (diAllocAny(imap, agno, dir, ip)); 1656 } 1657 1658 1659 /* 1660 * NAME: diAllocAG(imap,agno,dir,ip) 1661 * 1662 * FUNCTION: allocate a disk inode from the allocation group. 1663 * 1664 * this routine first determines if a new extent of free 1665 * inodes should be added for the allocation group, with 1666 * the current request satisfied from this extent. if this 1667 * is the case, an attempt will be made to do just that. if 1668 * this attempt fails or it has been determined that a new 1669 * extent should not be added, an attempt is made to satisfy 1670 * the request by allocating an existing (backed) free inode 1671 * from the allocation group. 1672 * 1673 * PRE CONDITION: Already have the AG lock for this AG. 1674 * 1675 * PARAMETERS: 1676 * imap - pointer to inode map control structure. 1677 * agno - allocation group to allocate from. 1678 * dir - TRUE if the new disk inode is for a directory. 1679 * ip - pointer to the new inode to be filled in on successful return 1680 * with the disk inode number allocated, its extent address 1681 * and the start of the ag. 1682 * 1683 * RETURN VALUES: 1684 * 0 - success. 1685 * -ENOSPC - insufficient disk resources. 1686 * -EIO - i/o error. 1687 */ 1688 static int 1689 diAllocAG(struct inomap * imap, int agno, boolean_t dir, struct inode *ip) 1690 { 1691 int rc, addext, numfree, numinos; 1692 1693 /* get the number of free and the number of backed disk 1694 * inodes currently within the ag. 1695 */ 1696 numfree = imap->im_agctl[agno].numfree; 1697 numinos = imap->im_agctl[agno].numinos; 1698 1699 if (numfree > numinos) { 1700 jfs_error(ip->i_sb, "diAllocAG: numfree > numinos"); 1701 return -EIO; 1702 } 1703 1704 /* determine if we should allocate a new extent of free inodes 1705 * within the ag: for directory inodes, add a new extent 1706 * if there are a small number of free inodes or number of free 1707 * inodes is a small percentage of the number of backed inodes. 1708 */ 1709 if (dir == TRUE) 1710 addext = (numfree < 64 || 1711 (numfree < 256 1712 && ((numfree * 100) / numinos) <= 20)); 1713 else 1714 addext = (numfree == 0); 1715 1716 /* 1717 * try to allocate a new extent of free inodes. 1718 */ 1719 if (addext) { 1720 /* if free space is not avaliable for this new extent, try 1721 * below to allocate a free and existing (already backed) 1722 * inode from the ag. 1723 */ 1724 if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC) 1725 return (rc); 1726 } 1727 1728 /* 1729 * try to allocate an existing free inode from the ag. 1730 */ 1731 return (diAllocIno(imap, agno, ip)); 1732 } 1733 1734 1735 /* 1736 * NAME: diAllocAny(imap,agno,dir,iap) 1737 * 1738 * FUNCTION: allocate a disk inode from any other allocation group. 1739 * 1740 * this routine is called when an allocation attempt within 1741 * the primary allocation group has failed. if attempts to 1742 * allocate an inode from any allocation group other than the 1743 * specified primary group. 1744 * 1745 * PARAMETERS: 1746 * imap - pointer to inode map control structure. 1747 * agno - primary allocation group (to avoid). 1748 * dir - TRUE if the new disk inode is for a directory. 1749 * ip - pointer to a new inode to be filled in on successful return 1750 * with the disk inode number allocated, its extent address 1751 * and the start of the ag. 1752 * 1753 * RETURN VALUES: 1754 * 0 - success. 1755 * -ENOSPC - insufficient disk resources. 1756 * -EIO - i/o error. 1757 */ 1758 static int 1759 diAllocAny(struct inomap * imap, int agno, boolean_t dir, struct inode *ip) 1760 { 1761 int ag, rc; 1762 int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag; 1763 1764 1765 /* try to allocate from the ags following agno up to 1766 * the maximum ag number. 1767 */ 1768 for (ag = agno + 1; ag <= maxag; ag++) { 1769 AG_LOCK(imap, ag); 1770 1771 rc = diAllocAG(imap, ag, dir, ip); 1772 1773 AG_UNLOCK(imap, ag); 1774 1775 if (rc != -ENOSPC) 1776 return (rc); 1777 } 1778 1779 /* try to allocate from the ags in front of agno. 1780 */ 1781 for (ag = 0; ag < agno; ag++) { 1782 AG_LOCK(imap, ag); 1783 1784 rc = diAllocAG(imap, ag, dir, ip); 1785 1786 AG_UNLOCK(imap, ag); 1787 1788 if (rc != -ENOSPC) 1789 return (rc); 1790 } 1791 1792 /* no free disk inodes. 1793 */ 1794 return -ENOSPC; 1795 } 1796 1797 1798 /* 1799 * NAME: diAllocIno(imap,agno,ip) 1800 * 1801 * FUNCTION: allocate a disk inode from the allocation group's free 1802 * inode list, returning an error if this free list is 1803 * empty (i.e. no iags on the list). 1804 * 1805 * allocation occurs from the first iag on the list using 1806 * the iag's free inode summary map to find the leftmost 1807 * free inode in the iag. 1808 * 1809 * PRE CONDITION: Already have AG lock for this AG. 1810 * 1811 * PARAMETERS: 1812 * imap - pointer to inode map control structure. 1813 * agno - allocation group. 1814 * ip - pointer to new inode to be filled in on successful return 1815 * with the disk inode number allocated, its extent address 1816 * and the start of the ag. 1817 * 1818 * RETURN VALUES: 1819 * 0 - success. 1820 * -ENOSPC - insufficient disk resources. 1821 * -EIO - i/o error. 1822 */ 1823 static int diAllocIno(struct inomap * imap, int agno, struct inode *ip) 1824 { 1825 int iagno, ino, rc, rem, extno, sword; 1826 struct metapage *mp; 1827 struct iag *iagp; 1828 1829 /* check if there are iags on the ag's free inode list. 1830 */ 1831 if ((iagno = imap->im_agctl[agno].inofree) < 0) 1832 return -ENOSPC; 1833 1834 /* obtain read lock on imap inode */ 1835 IREAD_LOCK(imap->im_ipimap); 1836 1837 /* read the iag at the head of the list. 1838 */ 1839 if ((rc = diIAGRead(imap, iagno, &mp))) { 1840 IREAD_UNLOCK(imap->im_ipimap); 1841 return (rc); 1842 } 1843 iagp = (struct iag *) mp->data; 1844 1845 /* better be free inodes in this iag if it is on the 1846 * list. 1847 */ 1848 if (!iagp->nfreeinos) { 1849 IREAD_UNLOCK(imap->im_ipimap); 1850 release_metapage(mp); 1851 jfs_error(ip->i_sb, 1852 "diAllocIno: nfreeinos = 0, but iag on freelist"); 1853 return -EIO; 1854 } 1855 1856 /* scan the free inode summary map to find an extent 1857 * with free inodes. 1858 */ 1859 for (sword = 0;; sword++) { 1860 if (sword >= SMAPSZ) { 1861 IREAD_UNLOCK(imap->im_ipimap); 1862 release_metapage(mp); 1863 jfs_error(ip->i_sb, 1864 "diAllocIno: free inode not found in summary map"); 1865 return -EIO; 1866 } 1867 1868 if (~iagp->inosmap[sword]) 1869 break; 1870 } 1871 1872 /* found a extent with free inodes. determine 1873 * the extent number. 1874 */ 1875 rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0); 1876 if (rem >= EXTSPERSUM) { 1877 IREAD_UNLOCK(imap->im_ipimap); 1878 release_metapage(mp); 1879 jfs_error(ip->i_sb, "diAllocIno: no free extent found"); 1880 return -EIO; 1881 } 1882 extno = (sword << L2EXTSPERSUM) + rem; 1883 1884 /* find the first free inode in the extent. 1885 */ 1886 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0); 1887 if (rem >= INOSPEREXT) { 1888 IREAD_UNLOCK(imap->im_ipimap); 1889 release_metapage(mp); 1890 jfs_error(ip->i_sb, "diAllocIno: free inode not found"); 1891 return -EIO; 1892 } 1893 1894 /* compute the inode number within the iag. 1895 */ 1896 ino = (extno << L2INOSPEREXT) + rem; 1897 1898 /* allocate the inode. 1899 */ 1900 rc = diAllocBit(imap, iagp, ino); 1901 IREAD_UNLOCK(imap->im_ipimap); 1902 if (rc) { 1903 release_metapage(mp); 1904 return (rc); 1905 } 1906 1907 /* set the results of the allocation and write the iag. 1908 */ 1909 diInitInode(ip, iagno, ino, extno, iagp); 1910 write_metapage(mp); 1911 1912 return (0); 1913 } 1914 1915 1916 /* 1917 * NAME: diAllocExt(imap,agno,ip) 1918 * 1919 * FUNCTION: add a new extent of free inodes to an iag, allocating 1920 * an inode from this extent to satisfy the current allocation 1921 * request. 1922 * 1923 * this routine first tries to find an existing iag with free 1924 * extents through the ag free extent list. if list is not 1925 * empty, the head of the list will be selected as the home 1926 * of the new extent of free inodes. otherwise (the list is 1927 * empty), a new iag will be allocated for the ag to contain 1928 * the extent. 1929 * 1930 * once an iag has been selected, the free extent summary map 1931 * is used to locate a free extent within the iag and diNewExt() 1932 * is called to initialize the extent, with initialization 1933 * including the allocation of the first inode of the extent 1934 * for the purpose of satisfying this request. 1935 * 1936 * PARAMETERS: 1937 * imap - pointer to inode map control structure. 1938 * agno - allocation group number. 1939 * ip - pointer to new inode to be filled in on successful return 1940 * with the disk inode number allocated, its extent address 1941 * and the start of the ag. 1942 * 1943 * RETURN VALUES: 1944 * 0 - success. 1945 * -ENOSPC - insufficient disk resources. 1946 * -EIO - i/o error. 1947 */ 1948 static int diAllocExt(struct inomap * imap, int agno, struct inode *ip) 1949 { 1950 int rem, iagno, sword, extno, rc; 1951 struct metapage *mp; 1952 struct iag *iagp; 1953 1954 /* check if the ag has any iags with free extents. if not, 1955 * allocate a new iag for the ag. 1956 */ 1957 if ((iagno = imap->im_agctl[agno].extfree) < 0) { 1958 /* If successful, diNewIAG will obtain the read lock on the 1959 * imap inode. 1960 */ 1961 if ((rc = diNewIAG(imap, &iagno, agno, &mp))) { 1962 return (rc); 1963 } 1964 iagp = (struct iag *) mp->data; 1965 1966 /* set the ag number if this a brand new iag 1967 */ 1968 iagp->agstart = 1969 cpu_to_le64(AGTOBLK(agno, imap->im_ipimap)); 1970 } else { 1971 /* read the iag. 1972 */ 1973 IREAD_LOCK(imap->im_ipimap); 1974 if ((rc = diIAGRead(imap, iagno, &mp))) { 1975 IREAD_UNLOCK(imap->im_ipimap); 1976 jfs_error(ip->i_sb, "diAllocExt: error reading iag"); 1977 return rc; 1978 } 1979 iagp = (struct iag *) mp->data; 1980 } 1981 1982 /* using the free extent summary map, find a free extent. 1983 */ 1984 for (sword = 0;; sword++) { 1985 if (sword >= SMAPSZ) { 1986 release_metapage(mp); 1987 IREAD_UNLOCK(imap->im_ipimap); 1988 jfs_error(ip->i_sb, 1989 "diAllocExt: free ext summary map not found"); 1990 return -EIO; 1991 } 1992 if (~iagp->extsmap[sword]) 1993 break; 1994 } 1995 1996 /* determine the extent number of the free extent. 1997 */ 1998 rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0); 1999 if (rem >= EXTSPERSUM) { 2000 release_metapage(mp); 2001 IREAD_UNLOCK(imap->im_ipimap); 2002 jfs_error(ip->i_sb, "diAllocExt: free extent not found"); 2003 return -EIO; 2004 } 2005 extno = (sword << L2EXTSPERSUM) + rem; 2006 2007 /* initialize the new extent. 2008 */ 2009 rc = diNewExt(imap, iagp, extno); 2010 IREAD_UNLOCK(imap->im_ipimap); 2011 if (rc) { 2012 /* something bad happened. if a new iag was allocated, 2013 * place it back on the inode map's iag free list, and 2014 * clear the ag number information. 2015 */ 2016 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { 2017 IAGFREE_LOCK(imap); 2018 iagp->iagfree = cpu_to_le32(imap->im_freeiag); 2019 imap->im_freeiag = iagno; 2020 IAGFREE_UNLOCK(imap); 2021 } 2022 write_metapage(mp); 2023 return (rc); 2024 } 2025 2026 /* set the results of the allocation and write the iag. 2027 */ 2028 diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp); 2029 2030 write_metapage(mp); 2031 2032 return (0); 2033 } 2034 2035 2036 /* 2037 * NAME: diAllocBit(imap,iagp,ino) 2038 * 2039 * FUNCTION: allocate a backed inode from an iag. 2040 * 2041 * this routine performs the mechanics of allocating a 2042 * specified inode from a backed extent. 2043 * 2044 * if the inode to be allocated represents the last free 2045 * inode within the iag, the iag will be removed from the 2046 * ag free inode list. 2047 * 2048 * a careful update approach is used to provide consistency 2049 * in the face of updates to multiple buffers. under this 2050 * approach, all required buffers are obtained before making 2051 * any updates and are held all are updates are complete. 2052 * 2053 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on 2054 * this AG. Must have read lock on imap inode. 2055 * 2056 * PARAMETERS: 2057 * imap - pointer to inode map control structure. 2058 * iagp - pointer to iag. 2059 * ino - inode number to be allocated within the iag. 2060 * 2061 * RETURN VALUES: 2062 * 0 - success. 2063 * -ENOSPC - insufficient disk resources. 2064 * -EIO - i/o error. 2065 */ 2066 static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino) 2067 { 2068 int extno, bitno, agno, sword, rc; 2069 struct metapage *amp = NULL, *bmp = NULL; 2070 struct iag *aiagp = NULL, *biagp = NULL; 2071 u32 mask; 2072 2073 /* check if this is the last free inode within the iag. 2074 * if so, it will have to be removed from the ag free 2075 * inode list, so get the iags preceeding and following 2076 * it on the list. 2077 */ 2078 if (iagp->nfreeinos == cpu_to_le32(1)) { 2079 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) { 2080 if ((rc = 2081 diIAGRead(imap, le32_to_cpu(iagp->inofreefwd), 2082 &))) 2083 return (rc); 2084 aiagp = (struct iag *) amp->data; 2085 } 2086 2087 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) { 2088 if ((rc = 2089 diIAGRead(imap, 2090 le32_to_cpu(iagp->inofreeback), 2091 &bmp))) { 2092 if (amp) 2093 release_metapage(amp); 2094 return (rc); 2095 } 2096 biagp = (struct iag *) bmp->data; 2097 } 2098 } 2099 2100 /* get the ag number, extent number, inode number within 2101 * the extent. 2102 */ 2103 agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb)); 2104 extno = ino >> L2INOSPEREXT; 2105 bitno = ino & (INOSPEREXT - 1); 2106 2107 /* compute the mask for setting the map. 2108 */ 2109 mask = HIGHORDER >> bitno; 2110 2111 /* the inode should be free and backed. 2112 */ 2113 if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) || 2114 ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) || 2115 (addressPXD(&iagp->inoext[extno]) == 0)) { 2116 if (amp) 2117 release_metapage(amp); 2118 if (bmp) 2119 release_metapage(bmp); 2120 2121 jfs_error(imap->im_ipimap->i_sb, 2122 "diAllocBit: iag inconsistent"); 2123 return -EIO; 2124 } 2125 2126 /* mark the inode as allocated in the working map. 2127 */ 2128 iagp->wmap[extno] |= cpu_to_le32(mask); 2129 2130 /* check if all inodes within the extent are now 2131 * allocated. if so, update the free inode summary 2132 * map to reflect this. 2133 */ 2134 if (iagp->wmap[extno] == cpu_to_le32(ONES)) { 2135 sword = extno >> L2EXTSPERSUM; 2136 bitno = extno & (EXTSPERSUM - 1); 2137 iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno); 2138 } 2139 2140 /* if this was the last free inode in the iag, remove the 2141 * iag from the ag free inode list. 2142 */ 2143 if (iagp->nfreeinos == cpu_to_le32(1)) { 2144 if (amp) { 2145 aiagp->inofreeback = iagp->inofreeback; 2146 write_metapage(amp); 2147 } 2148 2149 if (bmp) { 2150 biagp->inofreefwd = iagp->inofreefwd; 2151 write_metapage(bmp); 2152 } else { 2153 imap->im_agctl[agno].inofree = 2154 le32_to_cpu(iagp->inofreefwd); 2155 } 2156 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); 2157 } 2158 2159 /* update the free inode count at the iag, ag, inode 2160 * map levels. 2161 */ 2162 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) - 1); 2163 imap->im_agctl[agno].numfree -= 1; 2164 atomic_dec(&imap->im_numfree); 2165 2166 return (0); 2167 } 2168 2169 2170 /* 2171 * NAME: diNewExt(imap,iagp,extno) 2172 * 2173 * FUNCTION: initialize a new extent of inodes for an iag, allocating 2174 * the first inode of the extent for use for the current 2175 * allocation request. 2176 * 2177 * disk resources are allocated for the new extent of inodes 2178 * and the inodes themselves are initialized to reflect their 2179 * existence within the extent (i.e. their inode numbers and 2180 * inode extent addresses are set) and their initial state 2181 * (mode and link count are set to zero). 2182 * 2183 * if the iag is new, it is not yet on an ag extent free list 2184 * but will now be placed on this list. 2185 * 2186 * if the allocation of the new extent causes the iag to 2187 * have no free extent, the iag will be removed from the 2188 * ag extent free list. 2189 * 2190 * if the iag has no free backed inodes, it will be placed 2191 * on the ag free inode list, since the addition of the new 2192 * extent will now cause it to have free inodes. 2193 * 2194 * a careful update approach is used to provide consistency 2195 * (i.e. list consistency) in the face of updates to multiple 2196 * buffers. under this approach, all required buffers are 2197 * obtained before making any updates and are held until all 2198 * updates are complete. 2199 * 2200 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on 2201 * this AG. Must have read lock on imap inode. 2202 * 2203 * PARAMETERS: 2204 * imap - pointer to inode map control structure. 2205 * iagp - pointer to iag. 2206 * extno - extent number. 2207 * 2208 * RETURN VALUES: 2209 * 0 - success. 2210 * -ENOSPC - insufficient disk resources. 2211 * -EIO - i/o error. 2212 */ 2213 static int diNewExt(struct inomap * imap, struct iag * iagp, int extno) 2214 { 2215 int agno, iagno, fwd, back, freei = 0, sword, rc; 2216 struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL; 2217 struct metapage *amp, *bmp, *cmp, *dmp; 2218 struct inode *ipimap; 2219 s64 blkno, hint; 2220 int i, j; 2221 u32 mask; 2222 ino_t ino; 2223 struct dinode *dp; 2224 struct jfs_sb_info *sbi; 2225 2226 /* better have free extents. 2227 */ 2228 if (!iagp->nfreeexts) { 2229 jfs_error(imap->im_ipimap->i_sb, "diNewExt: no free extents"); 2230 return -EIO; 2231 } 2232 2233 /* get the inode map inode. 2234 */ 2235 ipimap = imap->im_ipimap; 2236 sbi = JFS_SBI(ipimap->i_sb); 2237 2238 amp = bmp = cmp = NULL; 2239 2240 /* get the ag and iag numbers for this iag. 2241 */ 2242 agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi); 2243 iagno = le32_to_cpu(iagp->iagnum); 2244 2245 /* check if this is the last free extent within the 2246 * iag. if so, the iag must be removed from the ag 2247 * free extent list, so get the iags preceeding and 2248 * following the iag on this list. 2249 */ 2250 if (iagp->nfreeexts == cpu_to_le32(1)) { 2251 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) { 2252 if ((rc = diIAGRead(imap, fwd, &))) 2253 return (rc); 2254 aiagp = (struct iag *) amp->data; 2255 } 2256 2257 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) { 2258 if ((rc = diIAGRead(imap, back, &bmp))) 2259 goto error_out; 2260 biagp = (struct iag *) bmp->data; 2261 } 2262 } else { 2263 /* the iag has free extents. if all extents are free 2264 * (as is the case for a newly allocated iag), the iag 2265 * must be added to the ag free extent list, so get 2266 * the iag at the head of the list in preparation for 2267 * adding this iag to this list. 2268 */ 2269 fwd = back = -1; 2270 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { 2271 if ((fwd = imap->im_agctl[agno].extfree) >= 0) { 2272 if ((rc = diIAGRead(imap, fwd, &))) 2273 goto error_out; 2274 aiagp = (struct iag *) amp->data; 2275 } 2276 } 2277 } 2278 2279 /* check if the iag has no free inodes. if so, the iag 2280 * will have to be added to the ag free inode list, so get 2281 * the iag at the head of the list in preparation for 2282 * adding this iag to this list. in doing this, we must 2283 * check if we already have the iag at the head of 2284 * the list in hand. 2285 */ 2286 if (iagp->nfreeinos == 0) { 2287 freei = imap->im_agctl[agno].inofree; 2288 2289 if (freei >= 0) { 2290 if (freei == fwd) { 2291 ciagp = aiagp; 2292 } else if (freei == back) { 2293 ciagp = biagp; 2294 } else { 2295 if ((rc = diIAGRead(imap, freei, &cmp))) 2296 goto error_out; 2297 ciagp = (struct iag *) cmp->data; 2298 } 2299 if (ciagp == NULL) { 2300 jfs_error(imap->im_ipimap->i_sb, 2301 "diNewExt: ciagp == NULL"); 2302 rc = -EIO; 2303 goto error_out; 2304 } 2305 } 2306 } 2307 2308 /* allocate disk space for the inode extent. 2309 */ 2310 if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0)) 2311 hint = ((s64) agno << sbi->bmap->db_agl2size) - 1; 2312 else 2313 hint = addressPXD(&iagp->inoext[extno - 1]) + 2314 lengthPXD(&iagp->inoext[extno - 1]) - 1; 2315 2316 if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno))) 2317 goto error_out; 2318 2319 /* compute the inode number of the first inode within the 2320 * extent. 2321 */ 2322 ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT); 2323 2324 /* initialize the inodes within the newly allocated extent a 2325 * page at a time. 2326 */ 2327 for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) { 2328 /* get a buffer for this page of disk inodes. 2329 */ 2330 dmp = get_metapage(ipimap, blkno + i, PSIZE, 1); 2331 if (dmp == NULL) { 2332 rc = -EIO; 2333 goto error_out; 2334 } 2335 dp = (struct dinode *) dmp->data; 2336 2337 /* initialize the inode number, mode, link count and 2338 * inode extent address. 2339 */ 2340 for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) { 2341 dp->di_inostamp = cpu_to_le32(sbi->inostamp); 2342 dp->di_number = cpu_to_le32(ino); 2343 dp->di_fileset = cpu_to_le32(FILESYSTEM_I); 2344 dp->di_mode = 0; 2345 dp->di_nlink = 0; 2346 PXDaddress(&(dp->di_ixpxd), blkno); 2347 PXDlength(&(dp->di_ixpxd), imap->im_nbperiext); 2348 } 2349 write_metapage(dmp); 2350 } 2351 2352 /* if this is the last free extent within the iag, remove the 2353 * iag from the ag free extent list. 2354 */ 2355 if (iagp->nfreeexts == cpu_to_le32(1)) { 2356 if (fwd >= 0) 2357 aiagp->extfreeback = iagp->extfreeback; 2358 2359 if (back >= 0) 2360 biagp->extfreefwd = iagp->extfreefwd; 2361 else 2362 imap->im_agctl[agno].extfree = 2363 le32_to_cpu(iagp->extfreefwd); 2364 2365 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); 2366 } else { 2367 /* if the iag has all free extents (newly allocated iag), 2368 * add the iag to the ag free extent list. 2369 */ 2370 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { 2371 if (fwd >= 0) 2372 aiagp->extfreeback = cpu_to_le32(iagno); 2373 2374 iagp->extfreefwd = cpu_to_le32(fwd); 2375 iagp->extfreeback = cpu_to_le32(-1); 2376 imap->im_agctl[agno].extfree = iagno; 2377 } 2378 } 2379 2380 /* if the iag has no free inodes, add the iag to the 2381 * ag free inode list. 2382 */ 2383 if (iagp->nfreeinos == 0) { 2384 if (freei >= 0) 2385 ciagp->inofreeback = cpu_to_le32(iagno); 2386 2387 iagp->inofreefwd = 2388 cpu_to_le32(imap->im_agctl[agno].inofree); 2389 iagp->inofreeback = cpu_to_le32(-1); 2390 imap->im_agctl[agno].inofree = iagno; 2391 } 2392 2393 /* initialize the extent descriptor of the extent. */ 2394 PXDlength(&iagp->inoext[extno], imap->im_nbperiext); 2395 PXDaddress(&iagp->inoext[extno], blkno); 2396 2397 /* initialize the working and persistent map of the extent. 2398 * the working map will be initialized such that 2399 * it indicates the first inode of the extent is allocated. 2400 */ 2401 iagp->wmap[extno] = cpu_to_le32(HIGHORDER); 2402 iagp->pmap[extno] = 0; 2403 2404 /* update the free inode and free extent summary maps 2405 * for the extent to indicate the extent has free inodes 2406 * and no longer represents a free extent. 2407 */ 2408 sword = extno >> L2EXTSPERSUM; 2409 mask = HIGHORDER >> (extno & (EXTSPERSUM - 1)); 2410 iagp->extsmap[sword] |= cpu_to_le32(mask); 2411 iagp->inosmap[sword] &= cpu_to_le32(~mask); 2412 2413 /* update the free inode and free extent counts for the 2414 * iag. 2415 */ 2416 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) + 2417 (INOSPEREXT - 1)); 2418 iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) - 1); 2419 2420 /* update the free and backed inode counts for the ag. 2421 */ 2422 imap->im_agctl[agno].numfree += (INOSPEREXT - 1); 2423 imap->im_agctl[agno].numinos += INOSPEREXT; 2424 2425 /* update the free and backed inode counts for the inode map. 2426 */ 2427 atomic_add(INOSPEREXT - 1, &imap->im_numfree); 2428 atomic_add(INOSPEREXT, &imap->im_numinos); 2429 2430 /* write the iags. 2431 */ 2432 if (amp) 2433 write_metapage(amp); 2434 if (bmp) 2435 write_metapage(bmp); 2436 if (cmp) 2437 write_metapage(cmp); 2438 2439 return (0); 2440 2441 error_out: 2442 2443 /* release the iags. 2444 */ 2445 if (amp) 2446 release_metapage(amp); 2447 if (bmp) 2448 release_metapage(bmp); 2449 if (cmp) 2450 release_metapage(cmp); 2451 2452 return (rc); 2453 } 2454 2455 2456 /* 2457 * NAME: diNewIAG(imap,iagnop,agno) 2458 * 2459 * FUNCTION: allocate a new iag for an allocation group. 2460 * 2461 * first tries to allocate the iag from the inode map 2462 * iagfree list: 2463 * if the list has free iags, the head of the list is removed 2464 * and returned to satisfy the request. 2465 * if the inode map's iag free list is empty, the inode map 2466 * is extended to hold a new iag. this new iag is initialized 2467 * and returned to satisfy the request. 2468 * 2469 * PARAMETERS: 2470 * imap - pointer to inode map control structure. 2471 * iagnop - pointer to an iag number set with the number of the 2472 * newly allocated iag upon successful return. 2473 * agno - allocation group number. 2474 * bpp - Buffer pointer to be filled in with new IAG's buffer 2475 * 2476 * RETURN VALUES: 2477 * 0 - success. 2478 * -ENOSPC - insufficient disk resources. 2479 * -EIO - i/o error. 2480 * 2481 * serialization: 2482 * AG lock held on entry/exit; 2483 * write lock on the map is held inside; 2484 * read lock on the map is held on successful completion; 2485 * 2486 * note: new iag transaction: 2487 * . synchronously write iag; 2488 * . write log of xtree and inode of imap; 2489 * . commit; 2490 * . synchronous write of xtree (right to left, bottom to top); 2491 * . at start of logredo(): init in-memory imap with one additional iag page; 2492 * . at end of logredo(): re-read imap inode to determine 2493 * new imap size; 2494 */ 2495 static int 2496 diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp) 2497 { 2498 int rc; 2499 int iagno, i, xlen; 2500 struct inode *ipimap; 2501 struct super_block *sb; 2502 struct jfs_sb_info *sbi; 2503 struct metapage *mp; 2504 struct iag *iagp; 2505 s64 xaddr = 0; 2506 s64 blkno; 2507 tid_t tid; 2508 #ifdef _STILL_TO_PORT 2509 xad_t xad; 2510 #endif /* _STILL_TO_PORT */ 2511 struct inode *iplist[1]; 2512 2513 /* pick up pointers to the inode map and mount inodes */ 2514 ipimap = imap->im_ipimap; 2515 sb = ipimap->i_sb; 2516 sbi = JFS_SBI(sb); 2517 2518 /* acquire the free iag lock */ 2519 IAGFREE_LOCK(imap); 2520 2521 /* if there are any iags on the inode map free iag list, 2522 * allocate the iag from the head of the list. 2523 */ 2524 if (imap->im_freeiag >= 0) { 2525 /* pick up the iag number at the head of the list */ 2526 iagno = imap->im_freeiag; 2527 2528 /* determine the logical block number of the iag */ 2529 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage); 2530 } else { 2531 /* no free iags. the inode map will have to be extented 2532 * to include a new iag. 2533 */ 2534 2535 /* acquire inode map lock */ 2536 IWRITE_LOCK(ipimap); 2537 2538 if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) { 2539 IWRITE_UNLOCK(ipimap); 2540 IAGFREE_UNLOCK(imap); 2541 jfs_error(imap->im_ipimap->i_sb, 2542 "diNewIAG: ipimap->i_size is wrong"); 2543 return -EIO; 2544 } 2545 2546 2547 /* get the next avaliable iag number */ 2548 iagno = imap->im_nextiag; 2549 2550 /* make sure that we have not exceeded the maximum inode 2551 * number limit. 2552 */ 2553 if (iagno > (MAXIAGS - 1)) { 2554 /* release the inode map lock */ 2555 IWRITE_UNLOCK(ipimap); 2556 2557 rc = -ENOSPC; 2558 goto out; 2559 } 2560 2561 /* 2562 * synchronously append new iag page. 2563 */ 2564 /* determine the logical address of iag page to append */ 2565 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage); 2566 2567 /* Allocate extent for new iag page */ 2568 xlen = sbi->nbperpage; 2569 if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) { 2570 /* release the inode map lock */ 2571 IWRITE_UNLOCK(ipimap); 2572 2573 goto out; 2574 } 2575 2576 /* assign a buffer for the page */ 2577 mp = get_metapage(ipimap, xaddr, PSIZE, 1); 2578 if (!mp) { 2579 /* Free the blocks allocated for the iag since it was 2580 * not successfully added to the inode map 2581 */ 2582 dbFree(ipimap, xaddr, (s64) xlen); 2583 2584 /* release the inode map lock */ 2585 IWRITE_UNLOCK(ipimap); 2586 2587 rc = -EIO; 2588 goto out; 2589 } 2590 iagp = (struct iag *) mp->data; 2591 2592 /* init the iag */ 2593 memset(iagp, 0, sizeof(struct iag)); 2594 iagp->iagnum = cpu_to_le32(iagno); 2595 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); 2596 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); 2597 iagp->iagfree = cpu_to_le32(-1); 2598 iagp->nfreeinos = 0; 2599 iagp->nfreeexts = cpu_to_le32(EXTSPERIAG); 2600 2601 /* initialize the free inode summary map (free extent 2602 * summary map initialization handled by bzero). 2603 */ 2604 for (i = 0; i < SMAPSZ; i++) 2605 iagp->inosmap[i] = cpu_to_le32(ONES); 2606 2607 /* 2608 * Invalidate the page after writing and syncing it. 2609 * After it's initialized, we access it in a different 2610 * address space 2611 */ 2612 set_bit(META_discard, &mp->flag); 2613 flush_metapage(mp); 2614 2615 /* 2616 * start tyransaction of update of the inode map 2617 * addressing structure pointing to the new iag page; 2618 */ 2619 tid = txBegin(sb, COMMIT_FORCE); 2620 down(&JFS_IP(ipimap)->commit_sem); 2621 2622 /* update the inode map addressing structure to point to it */ 2623 if ((rc = 2624 xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) { 2625 txEnd(tid); 2626 up(&JFS_IP(ipimap)->commit_sem); 2627 /* Free the blocks allocated for the iag since it was 2628 * not successfully added to the inode map 2629 */ 2630 dbFree(ipimap, xaddr, (s64) xlen); 2631 2632 /* release the inode map lock */ 2633 IWRITE_UNLOCK(ipimap); 2634 2635 goto out; 2636 } 2637 2638 /* update the inode map's inode to reflect the extension */ 2639 ipimap->i_size += PSIZE; 2640 inode_add_bytes(ipimap, PSIZE); 2641 2642 /* 2643 * txCommit(COMMIT_FORCE) will synchronously write address 2644 * index pages and inode after commit in careful update order 2645 * of address index pages (right to left, bottom up); 2646 */ 2647 iplist[0] = ipimap; 2648 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE); 2649 2650 txEnd(tid); 2651 up(&JFS_IP(ipimap)->commit_sem); 2652 2653 duplicateIXtree(sb, blkno, xlen, &xaddr); 2654 2655 /* update the next avaliable iag number */ 2656 imap->im_nextiag += 1; 2657 2658 /* Add the iag to the iag free list so we don't lose the iag 2659 * if a failure happens now. 2660 */ 2661 imap->im_freeiag = iagno; 2662 2663 /* Until we have logredo working, we want the imap inode & 2664 * control page to be up to date. 2665 */ 2666 diSync(ipimap); 2667 2668 /* release the inode map lock */ 2669 IWRITE_UNLOCK(ipimap); 2670 } 2671 2672 /* obtain read lock on map */ 2673 IREAD_LOCK(ipimap); 2674 2675 /* read the iag */ 2676 if ((rc = diIAGRead(imap, iagno, &mp))) { 2677 IREAD_UNLOCK(ipimap); 2678 rc = -EIO; 2679 goto out; 2680 } 2681 iagp = (struct iag *) mp->data; 2682 2683 /* remove the iag from the iag free list */ 2684 imap->im_freeiag = le32_to_cpu(iagp->iagfree); 2685 iagp->iagfree = cpu_to_le32(-1); 2686 2687 /* set the return iag number and buffer pointer */ 2688 *iagnop = iagno; 2689 *mpp = mp; 2690 2691 out: 2692 /* release the iag free lock */ 2693 IAGFREE_UNLOCK(imap); 2694 2695 return (rc); 2696 } 2697 2698 /* 2699 * NAME: diIAGRead() 2700 * 2701 * FUNCTION: get the buffer for the specified iag within a fileset 2702 * or aggregate inode map. 2703 * 2704 * PARAMETERS: 2705 * imap - pointer to inode map control structure. 2706 * iagno - iag number. 2707 * bpp - point to buffer pointer to be filled in on successful 2708 * exit. 2709 * 2710 * SERIALIZATION: 2711 * must have read lock on imap inode 2712 * (When called by diExtendFS, the filesystem is quiesced, therefore 2713 * the read lock is unnecessary.) 2714 * 2715 * RETURN VALUES: 2716 * 0 - success. 2717 * -EIO - i/o error. 2718 */ 2719 static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp) 2720 { 2721 struct inode *ipimap = imap->im_ipimap; 2722 s64 blkno; 2723 2724 /* compute the logical block number of the iag. */ 2725 blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage); 2726 2727 /* read the iag. */ 2728 *mpp = read_metapage(ipimap, blkno, PSIZE, 0); 2729 if (*mpp == NULL) { 2730 return -EIO; 2731 } 2732 2733 return (0); 2734 } 2735 2736 /* 2737 * NAME: diFindFree() 2738 * 2739 * FUNCTION: find the first free bit in a word starting at 2740 * the specified bit position. 2741 * 2742 * PARAMETERS: 2743 * word - word to be examined. 2744 * start - starting bit position. 2745 * 2746 * RETURN VALUES: 2747 * bit position of first free bit in the word or 32 if 2748 * no free bits were found. 2749 */ 2750 static int diFindFree(u32 word, int start) 2751 { 2752 int bitno; 2753 assert(start < 32); 2754 /* scan the word for the first free bit. */ 2755 for (word <<= start, bitno = start; bitno < 32; 2756 bitno++, word <<= 1) { 2757 if ((word & HIGHORDER) == 0) 2758 break; 2759 } 2760 return (bitno); 2761 } 2762 2763 /* 2764 * NAME: diUpdatePMap() 2765 * 2766 * FUNCTION: Update the persistent map in an IAG for the allocation or 2767 * freeing of the specified inode. 2768 * 2769 * PRE CONDITIONS: Working map has already been updated for allocate. 2770 * 2771 * PARAMETERS: 2772 * ipimap - Incore inode map inode 2773 * inum - Number of inode to mark in permanent map 2774 * is_free - If TRUE indicates inode should be marked freed, otherwise 2775 * indicates inode should be marked allocated. 2776 * 2777 * RETURN VALUES: 2778 * 0 for success 2779 */ 2780 int 2781 diUpdatePMap(struct inode *ipimap, 2782 unsigned long inum, boolean_t is_free, struct tblock * tblk) 2783 { 2784 int rc; 2785 struct iag *iagp; 2786 struct metapage *mp; 2787 int iagno, ino, extno, bitno; 2788 struct inomap *imap; 2789 u32 mask; 2790 struct jfs_log *log; 2791 int lsn, difft, diffp; 2792 2793 imap = JFS_IP(ipimap)->i_imap; 2794 /* get the iag number containing the inode */ 2795 iagno = INOTOIAG(inum); 2796 /* make sure that the iag is contained within the map */ 2797 if (iagno >= imap->im_nextiag) { 2798 jfs_error(ipimap->i_sb, 2799 "diUpdatePMap: the iag is outside the map"); 2800 return -EIO; 2801 } 2802 /* read the iag */ 2803 IREAD_LOCK(ipimap); 2804 rc = diIAGRead(imap, iagno, &mp); 2805 IREAD_UNLOCK(ipimap); 2806 if (rc) 2807 return (rc); 2808 iagp = (struct iag *) mp->data; 2809 /* get the inode number and extent number of the inode within 2810 * the iag and the inode number within the extent. 2811 */ 2812 ino = inum & (INOSPERIAG - 1); 2813 extno = ino >> L2INOSPEREXT; 2814 bitno = ino & (INOSPEREXT - 1); 2815 mask = HIGHORDER >> bitno; 2816 /* 2817 * mark the inode free in persistent map: 2818 */ 2819 if (is_free == TRUE) { 2820 /* The inode should have been allocated both in working 2821 * map and in persistent map; 2822 * the inode will be freed from working map at the release 2823 * of last reference release; 2824 */ 2825 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { 2826 jfs_error(ipimap->i_sb, 2827 "diUpdatePMap: inode %ld not marked as " 2828 "allocated in wmap!", inum); 2829 } 2830 if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) { 2831 jfs_error(ipimap->i_sb, 2832 "diUpdatePMap: inode %ld not marked as " 2833 "allocated in pmap!", inum); 2834 } 2835 /* update the bitmap for the extent of the freed inode */ 2836 iagp->pmap[extno] &= cpu_to_le32(~mask); 2837 } 2838 /* 2839 * mark the inode allocated in persistent map: 2840 */ 2841 else { 2842 /* The inode should be already allocated in the working map 2843 * and should be free in persistent map; 2844 */ 2845 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { 2846 release_metapage(mp); 2847 jfs_error(ipimap->i_sb, 2848 "diUpdatePMap: the inode is not allocated in " 2849 "the working map"); 2850 return -EIO; 2851 } 2852 if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) { 2853 release_metapage(mp); 2854 jfs_error(ipimap->i_sb, 2855 "diUpdatePMap: the inode is not free in the " 2856 "persistent map"); 2857 return -EIO; 2858 } 2859 /* update the bitmap for the extent of the allocated inode */ 2860 iagp->pmap[extno] |= cpu_to_le32(mask); 2861 } 2862 /* 2863 * update iag lsn 2864 */ 2865 lsn = tblk->lsn; 2866 log = JFS_SBI(tblk->sb)->log; 2867 if (mp->lsn != 0) { 2868 /* inherit older/smaller lsn */ 2869 logdiff(difft, lsn, log); 2870 logdiff(diffp, mp->lsn, log); 2871 if (difft < diffp) { 2872 mp->lsn = lsn; 2873 /* move mp after tblock in logsync list */ 2874 LOGSYNC_LOCK(log); 2875 list_move(&mp->synclist, &tblk->synclist); 2876 LOGSYNC_UNLOCK(log); 2877 } 2878 /* inherit younger/larger clsn */ 2879 LOGSYNC_LOCK(log); 2880 assert(mp->clsn); 2881 logdiff(difft, tblk->clsn, log); 2882 logdiff(diffp, mp->clsn, log); 2883 if (difft > diffp) 2884 mp->clsn = tblk->clsn; 2885 LOGSYNC_UNLOCK(log); 2886 } else { 2887 mp->log = log; 2888 mp->lsn = lsn; 2889 /* insert mp after tblock in logsync list */ 2890 LOGSYNC_LOCK(log); 2891 log->count++; 2892 list_add(&mp->synclist, &tblk->synclist); 2893 mp->clsn = tblk->clsn; 2894 LOGSYNC_UNLOCK(log); 2895 } 2896 write_metapage(mp); 2897 return (0); 2898 } 2899 2900 /* 2901 * diExtendFS() 2902 * 2903 * function: update imap for extendfs(); 2904 * 2905 * note: AG size has been increased s.t. each k old contiguous AGs are 2906 * coalesced into a new AG; 2907 */ 2908 int diExtendFS(struct inode *ipimap, struct inode *ipbmap) 2909 { 2910 int rc, rcx = 0; 2911 struct inomap *imap = JFS_IP(ipimap)->i_imap; 2912 struct iag *iagp = NULL, *hiagp = NULL; 2913 struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap; 2914 struct metapage *bp, *hbp; 2915 int i, n, head; 2916 int numinos, xnuminos = 0, xnumfree = 0; 2917 s64 agstart; 2918 2919 jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d", 2920 imap->im_nextiag, atomic_read(&imap->im_numinos), 2921 atomic_read(&imap->im_numfree)); 2922 2923 /* 2924 * reconstruct imap 2925 * 2926 * coalesce contiguous k (newAGSize/oldAGSize) AGs; 2927 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn; 2928 * note: new AG size = old AG size * (2**x). 2929 */ 2930 2931 /* init per AG control information im_agctl[] */ 2932 for (i = 0; i < MAXAG; i++) { 2933 imap->im_agctl[i].inofree = -1; 2934 imap->im_agctl[i].extfree = -1; 2935 imap->im_agctl[i].numinos = 0; /* number of backed inodes */ 2936 imap->im_agctl[i].numfree = 0; /* number of free backed inodes */ 2937 } 2938 2939 /* 2940 * process each iag page of the map. 2941 * 2942 * rebuild AG Free Inode List, AG Free Inode Extent List; 2943 */ 2944 for (i = 0; i < imap->im_nextiag; i++) { 2945 if ((rc = diIAGRead(imap, i, &bp))) { 2946 rcx = rc; 2947 continue; 2948 } 2949 iagp = (struct iag *) bp->data; 2950 if (le32_to_cpu(iagp->iagnum) != i) { 2951 release_metapage(bp); 2952 jfs_error(ipimap->i_sb, 2953 "diExtendFs: unexpected value of iagnum"); 2954 return -EIO; 2955 } 2956 2957 /* leave free iag in the free iag list */ 2958 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { 2959 release_metapage(bp); 2960 continue; 2961 } 2962 2963 /* agstart that computes to the same ag is treated as same; */ 2964 agstart = le64_to_cpu(iagp->agstart); 2965 /* iagp->agstart = agstart & ~(mp->db_agsize - 1); */ 2966 n = agstart >> mp->db_agl2size; 2967 2968 /* compute backed inodes */ 2969 numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts)) 2970 << L2INOSPEREXT; 2971 if (numinos > 0) { 2972 /* merge AG backed inodes */ 2973 imap->im_agctl[n].numinos += numinos; 2974 xnuminos += numinos; 2975 } 2976 2977 /* if any backed free inodes, insert at AG free inode list */ 2978 if ((int) le32_to_cpu(iagp->nfreeinos) > 0) { 2979 if ((head = imap->im_agctl[n].inofree) == -1) { 2980 iagp->inofreefwd = cpu_to_le32(-1); 2981 iagp->inofreeback = cpu_to_le32(-1); 2982 } else { 2983 if ((rc = diIAGRead(imap, head, &hbp))) { 2984 rcx = rc; 2985 goto nextiag; 2986 } 2987 hiagp = (struct iag *) hbp->data; 2988 hiagp->inofreeback = iagp->iagnum; 2989 iagp->inofreefwd = cpu_to_le32(head); 2990 iagp->inofreeback = cpu_to_le32(-1); 2991 write_metapage(hbp); 2992 } 2993 2994 imap->im_agctl[n].inofree = 2995 le32_to_cpu(iagp->iagnum); 2996 2997 /* merge AG backed free inodes */ 2998 imap->im_agctl[n].numfree += 2999 le32_to_cpu(iagp->nfreeinos); 3000 xnumfree += le32_to_cpu(iagp->nfreeinos); 3001 } 3002 3003 /* if any free extents, insert at AG free extent list */ 3004 if (le32_to_cpu(iagp->nfreeexts) > 0) { 3005 if ((head = imap->im_agctl[n].extfree) == -1) { 3006 iagp->extfreefwd = cpu_to_le32(-1); 3007 iagp->extfreeback = cpu_to_le32(-1); 3008 } else { 3009 if ((rc = diIAGRead(imap, head, &hbp))) { 3010 rcx = rc; 3011 goto nextiag; 3012 } 3013 hiagp = (struct iag *) hbp->data; 3014 hiagp->extfreeback = iagp->iagnum; 3015 iagp->extfreefwd = cpu_to_le32(head); 3016 iagp->extfreeback = cpu_to_le32(-1); 3017 write_metapage(hbp); 3018 } 3019 3020 imap->im_agctl[n].extfree = 3021 le32_to_cpu(iagp->iagnum); 3022 } 3023 3024 nextiag: 3025 write_metapage(bp); 3026 } 3027 3028 if (xnuminos != atomic_read(&imap->im_numinos) || 3029 xnumfree != atomic_read(&imap->im_numfree)) { 3030 jfs_error(ipimap->i_sb, 3031 "diExtendFs: numinos or numfree incorrect"); 3032 return -EIO; 3033 } 3034 3035 return rcx; 3036 } 3037 3038 3039 /* 3040 * duplicateIXtree() 3041 * 3042 * serialization: IWRITE_LOCK held on entry/exit 3043 * 3044 * note: shadow page with regular inode (rel.2); 3045 */ 3046 static void duplicateIXtree(struct super_block *sb, s64 blkno, 3047 int xlen, s64 *xaddr) 3048 { 3049 struct jfs_superblock *j_sb; 3050 struct buffer_head *bh; 3051 struct inode *ip; 3052 tid_t tid; 3053 3054 /* if AIT2 ipmap2 is bad, do not try to update it */ 3055 if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT) /* s_flag */ 3056 return; 3057 ip = diReadSpecial(sb, FILESYSTEM_I, 1); 3058 if (ip == NULL) { 3059 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT; 3060 if (readSuper(sb, &bh)) 3061 return; 3062 j_sb = (struct jfs_superblock *)bh->b_data; 3063 j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT); 3064 3065 mark_buffer_dirty(bh); 3066 sync_dirty_buffer(bh); 3067 brelse(bh); 3068 return; 3069 } 3070 3071 /* start transaction */ 3072 tid = txBegin(sb, COMMIT_FORCE); 3073 /* update the inode map addressing structure to point to it */ 3074 if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) { 3075 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT; 3076 txAbort(tid, 1); 3077 goto cleanup; 3078 3079 } 3080 /* update the inode map's inode to reflect the extension */ 3081 ip->i_size += PSIZE; 3082 inode_add_bytes(ip, PSIZE); 3083 txCommit(tid, 1, &ip, COMMIT_FORCE); 3084 cleanup: 3085 txEnd(tid); 3086 diFreeSpecial(ip); 3087 } 3088 3089 /* 3090 * NAME: copy_from_dinode() 3091 * 3092 * FUNCTION: Copies inode info from disk inode to in-memory inode 3093 * 3094 * RETURN VALUES: 3095 * 0 - success 3096 * -ENOMEM - insufficient memory 3097 */ 3098 static int copy_from_dinode(struct dinode * dip, struct inode *ip) 3099 { 3100 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 3101 3102 jfs_ip->fileset = le32_to_cpu(dip->di_fileset); 3103 jfs_ip->mode2 = le32_to_cpu(dip->di_mode); 3104 3105 ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff; 3106 ip->i_nlink = le32_to_cpu(dip->di_nlink); 3107 ip->i_uid = le32_to_cpu(dip->di_uid); 3108 ip->i_gid = le32_to_cpu(dip->di_gid); 3109 ip->i_size = le64_to_cpu(dip->di_size); 3110 ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec); 3111 ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec); 3112 ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec); 3113 ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec); 3114 ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec); 3115 ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec); 3116 ip->i_blksize = ip->i_sb->s_blocksize; 3117 ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks)); 3118 ip->i_generation = le32_to_cpu(dip->di_gen); 3119 3120 jfs_ip->ixpxd = dip->di_ixpxd; /* in-memory pxd's are little-endian */ 3121 jfs_ip->acl = dip->di_acl; /* as are dxd's */ 3122 jfs_ip->ea = dip->di_ea; 3123 jfs_ip->next_index = le32_to_cpu(dip->di_next_index); 3124 jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec); 3125 jfs_ip->acltype = le32_to_cpu(dip->di_acltype); 3126 3127 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) { 3128 jfs_ip->dev = le32_to_cpu(dip->di_rdev); 3129 ip->i_rdev = new_decode_dev(jfs_ip->dev); 3130 } 3131 3132 if (S_ISDIR(ip->i_mode)) { 3133 memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384); 3134 } else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) { 3135 memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288); 3136 } else 3137 memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128); 3138 3139 /* Zero the in-memory-only stuff */ 3140 jfs_ip->cflag = 0; 3141 jfs_ip->btindex = 0; 3142 jfs_ip->btorder = 0; 3143 jfs_ip->bxflag = 0; 3144 jfs_ip->blid = 0; 3145 jfs_ip->atlhead = 0; 3146 jfs_ip->atltail = 0; 3147 jfs_ip->xtlid = 0; 3148 return (0); 3149 } 3150 3151 /* 3152 * NAME: copy_to_dinode() 3153 * 3154 * FUNCTION: Copies inode info from in-memory inode to disk inode 3155 */ 3156 static void copy_to_dinode(struct dinode * dip, struct inode *ip) 3157 { 3158 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 3159 3160 dip->di_fileset = cpu_to_le32(jfs_ip->fileset); 3161 dip->di_inostamp = cpu_to_le32(JFS_SBI(ip->i_sb)->inostamp); 3162 dip->di_number = cpu_to_le32(ip->i_ino); 3163 dip->di_gen = cpu_to_le32(ip->i_generation); 3164 dip->di_size = cpu_to_le64(ip->i_size); 3165 dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks)); 3166 dip->di_nlink = cpu_to_le32(ip->i_nlink); 3167 dip->di_uid = cpu_to_le32(ip->i_uid); 3168 dip->di_gid = cpu_to_le32(ip->i_gid); 3169 /* 3170 * mode2 is only needed for storing the higher order bits. 3171 * Trust i_mode for the lower order ones 3172 */ 3173 dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) | ip->i_mode); 3174 dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec); 3175 dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec); 3176 dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec); 3177 dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec); 3178 dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec); 3179 dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec); 3180 dip->di_ixpxd = jfs_ip->ixpxd; /* in-memory pxd's are little-endian */ 3181 dip->di_acl = jfs_ip->acl; /* as are dxd's */ 3182 dip->di_ea = jfs_ip->ea; 3183 dip->di_next_index = cpu_to_le32(jfs_ip->next_index); 3184 dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime); 3185 dip->di_otime.tv_nsec = 0; 3186 dip->di_acltype = cpu_to_le32(jfs_ip->acltype); 3187 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) 3188 dip->di_rdev = cpu_to_le32(jfs_ip->dev); 3189 } 3190 3191 #ifdef _JFS_DEBUG_IMAP 3192 /* 3193 * DBGdiInit() 3194 */ 3195 static void *DBGdiInit(struct inomap * imap) 3196 { 3197 u32 *dimap; 3198 int size; 3199 size = 64 * 1024; 3200 if ((dimap = (u32 *) xmalloc(size, L2PSIZE, kernel_heap)) == NULL) 3201 assert(0); 3202 bzero((void *) dimap, size); 3203 imap->im_DBGdimap = dimap; 3204 } 3205 3206 /* 3207 * DBGdiAlloc() 3208 */ 3209 static void DBGdiAlloc(struct inomap * imap, ino_t ino) 3210 { 3211 u32 *dimap = imap->im_DBGdimap; 3212 int w, b; 3213 u32 m; 3214 w = ino >> 5; 3215 b = ino & 31; 3216 m = 0x80000000 >> b; 3217 assert(w < 64 * 256); 3218 if (dimap[w] & m) { 3219 printk("DEBUG diAlloc: duplicate alloc ino:0x%x\n", ino); 3220 } 3221 dimap[w] |= m; 3222 } 3223 3224 /* 3225 * DBGdiFree() 3226 */ 3227 static void DBGdiFree(struct inomap * imap, ino_t ino) 3228 { 3229 u32 *dimap = imap->im_DBGdimap; 3230 int w, b; 3231 u32 m; 3232 w = ino >> 5; 3233 b = ino & 31; 3234 m = 0x80000000 >> b; 3235 assert(w < 64 * 256); 3236 if ((dimap[w] & m) == 0) { 3237 printk("DEBUG diFree: duplicate free ino:0x%x\n", ino); 3238 } 3239 dimap[w] &= ~m; 3240 } 3241 3242 static void dump_cp(struct inomap * ipimap, char *function, int line) 3243 { 3244 printk("\n* ********* *\nControl Page %s %d\n", function, line); 3245 printk("FreeIAG %d\tNextIAG %d\n", ipimap->im_freeiag, 3246 ipimap->im_nextiag); 3247 printk("NumInos %d\tNumFree %d\n", 3248 atomic_read(&ipimap->im_numinos), 3249 atomic_read(&ipimap->im_numfree)); 3250 printk("AG InoFree %d\tAG ExtFree %d\n", 3251 ipimap->im_agctl[0].inofree, ipimap->im_agctl[0].extfree); 3252 printk("AG NumInos %d\tAG NumFree %d\n", 3253 ipimap->im_agctl[0].numinos, ipimap->im_agctl[0].numfree); 3254 } 3255 3256 static void dump_iag(struct iag * iag, char *function, int line) 3257 { 3258 printk("\n* ********* *\nIAG %s %d\n", function, line); 3259 printk("IagNum %d\tIAG Free %d\n", le32_to_cpu(iag->iagnum), 3260 le32_to_cpu(iag->iagfree)); 3261 printk("InoFreeFwd %d\tInoFreeBack %d\n", 3262 le32_to_cpu(iag->inofreefwd), 3263 le32_to_cpu(iag->inofreeback)); 3264 printk("ExtFreeFwd %d\tExtFreeBack %d\n", 3265 le32_to_cpu(iag->extfreefwd), 3266 le32_to_cpu(iag->extfreeback)); 3267 printk("NFreeInos %d\tNFreeExts %d\n", le32_to_cpu(iag->nfreeinos), 3268 le32_to_cpu(iag->nfreeexts)); 3269 } 3270 #endif /* _JFS_DEBUG_IMAP */ 3271