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 #include <linux/slab.h> 49 50 #include "jfs_incore.h" 51 #include "jfs_inode.h" 52 #include "jfs_filsys.h" 53 #include "jfs_dinode.h" 54 #include "jfs_dmap.h" 55 #include "jfs_imap.h" 56 #include "jfs_metapage.h" 57 #include "jfs_superblock.h" 58 #include "jfs_debug.h" 59 60 /* 61 * imap locks 62 */ 63 /* iag free list lock */ 64 #define IAGFREE_LOCK_INIT(imap) mutex_init(&imap->im_freelock) 65 #define IAGFREE_LOCK(imap) mutex_lock(&imap->im_freelock) 66 #define IAGFREE_UNLOCK(imap) mutex_unlock(&imap->im_freelock) 67 68 /* per ag iag list locks */ 69 #define AG_LOCK_INIT(imap,index) mutex_init(&(imap->im_aglock[index])) 70 #define AG_LOCK(imap,agno) mutex_lock(&imap->im_aglock[agno]) 71 #define AG_UNLOCK(imap,agno) mutex_unlock(&imap->im_aglock[agno]) 72 73 /* 74 * forward references 75 */ 76 static int diAllocAG(struct inomap *, int, bool, struct inode *); 77 static int diAllocAny(struct inomap *, int, bool, struct inode *); 78 static int diAllocBit(struct inomap *, struct iag *, int); 79 static int diAllocExt(struct inomap *, int, struct inode *); 80 static int diAllocIno(struct inomap *, int, struct inode *); 81 static int diFindFree(u32, int); 82 static int diNewExt(struct inomap *, struct iag *, int); 83 static int diNewIAG(struct inomap *, int *, int, struct metapage **); 84 static void duplicateIXtree(struct super_block *, s64, int, s64 *); 85 86 static int diIAGRead(struct inomap * imap, int, struct metapage **); 87 static int copy_from_dinode(struct dinode *, struct inode *); 88 static void copy_to_dinode(struct dinode *, struct inode *); 89 90 /* 91 * NAME: diMount() 92 * 93 * FUNCTION: initialize the incore inode map control structures for 94 * a fileset or aggregate init time. 95 * 96 * the inode map's control structure (dinomap) is 97 * brought in from disk and placed in virtual memory. 98 * 99 * PARAMETERS: 100 * ipimap - pointer to inode map inode for the aggregate or fileset. 101 * 102 * RETURN VALUES: 103 * 0 - success 104 * -ENOMEM - insufficient free virtual memory. 105 * -EIO - i/o error. 106 */ 107 int diMount(struct inode *ipimap) 108 { 109 struct inomap *imap; 110 struct metapage *mp; 111 int index; 112 struct dinomap_disk *dinom_le; 113 114 /* 115 * allocate/initialize the in-memory inode map control structure 116 */ 117 /* allocate the in-memory inode map control structure. */ 118 imap = kmalloc(sizeof(struct inomap), GFP_KERNEL); 119 if (imap == NULL) { 120 jfs_err("diMount: kmalloc returned NULL!"); 121 return -ENOMEM; 122 } 123 124 /* read the on-disk inode map control structure. */ 125 126 mp = read_metapage(ipimap, 127 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage, 128 PSIZE, 0); 129 if (mp == NULL) { 130 kfree(imap); 131 return -EIO; 132 } 133 134 /* copy the on-disk version to the in-memory version. */ 135 dinom_le = (struct dinomap_disk *) mp->data; 136 imap->im_freeiag = le32_to_cpu(dinom_le->in_freeiag); 137 imap->im_nextiag = le32_to_cpu(dinom_le->in_nextiag); 138 atomic_set(&imap->im_numinos, le32_to_cpu(dinom_le->in_numinos)); 139 atomic_set(&imap->im_numfree, le32_to_cpu(dinom_le->in_numfree)); 140 imap->im_nbperiext = le32_to_cpu(dinom_le->in_nbperiext); 141 imap->im_l2nbperiext = le32_to_cpu(dinom_le->in_l2nbperiext); 142 for (index = 0; index < MAXAG; index++) { 143 imap->im_agctl[index].inofree = 144 le32_to_cpu(dinom_le->in_agctl[index].inofree); 145 imap->im_agctl[index].extfree = 146 le32_to_cpu(dinom_le->in_agctl[index].extfree); 147 imap->im_agctl[index].numinos = 148 le32_to_cpu(dinom_le->in_agctl[index].numinos); 149 imap->im_agctl[index].numfree = 150 le32_to_cpu(dinom_le->in_agctl[index].numfree); 151 } 152 153 /* release the buffer. */ 154 release_metapage(mp); 155 156 /* 157 * allocate/initialize inode allocation map locks 158 */ 159 /* allocate and init iag free list lock */ 160 IAGFREE_LOCK_INIT(imap); 161 162 /* allocate and init ag list locks */ 163 for (index = 0; index < MAXAG; index++) { 164 AG_LOCK_INIT(imap, index); 165 } 166 167 /* bind the inode map inode and inode map control structure 168 * to each other. 169 */ 170 imap->im_ipimap = ipimap; 171 JFS_IP(ipimap)->i_imap = imap; 172 173 return (0); 174 } 175 176 177 /* 178 * NAME: diUnmount() 179 * 180 * FUNCTION: write to disk the incore inode map control structures for 181 * a fileset or aggregate at unmount time. 182 * 183 * PARAMETERS: 184 * ipimap - pointer to inode map inode for the aggregate or fileset. 185 * 186 * RETURN VALUES: 187 * 0 - success 188 * -ENOMEM - insufficient free virtual memory. 189 * -EIO - i/o error. 190 */ 191 int diUnmount(struct inode *ipimap, int mounterror) 192 { 193 struct inomap *imap = JFS_IP(ipimap)->i_imap; 194 195 /* 196 * update the on-disk inode map control structure 197 */ 198 199 if (!(mounterror || isReadOnly(ipimap))) 200 diSync(ipimap); 201 202 /* 203 * Invalidate the page cache buffers 204 */ 205 truncate_inode_pages(ipimap->i_mapping, 0); 206 207 /* 208 * free in-memory control structure 209 */ 210 kfree(imap); 211 212 return (0); 213 } 214 215 216 /* 217 * diSync() 218 */ 219 int diSync(struct inode *ipimap) 220 { 221 struct dinomap_disk *dinom_le; 222 struct inomap *imp = JFS_IP(ipimap)->i_imap; 223 struct metapage *mp; 224 int index; 225 226 /* 227 * write imap global conrol page 228 */ 229 /* read the on-disk inode map control structure */ 230 mp = get_metapage(ipimap, 231 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage, 232 PSIZE, 0); 233 if (mp == NULL) { 234 jfs_err("diSync: get_metapage failed!"); 235 return -EIO; 236 } 237 238 /* copy the in-memory version to the on-disk version */ 239 dinom_le = (struct dinomap_disk *) mp->data; 240 dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag); 241 dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag); 242 dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos)); 243 dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree)); 244 dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext); 245 dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext); 246 for (index = 0; index < MAXAG; index++) { 247 dinom_le->in_agctl[index].inofree = 248 cpu_to_le32(imp->im_agctl[index].inofree); 249 dinom_le->in_agctl[index].extfree = 250 cpu_to_le32(imp->im_agctl[index].extfree); 251 dinom_le->in_agctl[index].numinos = 252 cpu_to_le32(imp->im_agctl[index].numinos); 253 dinom_le->in_agctl[index].numfree = 254 cpu_to_le32(imp->im_agctl[index].numfree); 255 } 256 257 /* write out the control structure */ 258 write_metapage(mp); 259 260 /* 261 * write out dirty pages of imap 262 */ 263 filemap_write_and_wait(ipimap->i_mapping); 264 265 diWriteSpecial(ipimap, 0); 266 267 return (0); 268 } 269 270 271 /* 272 * NAME: diRead() 273 * 274 * FUNCTION: initialize an incore inode from disk. 275 * 276 * on entry, the specifed incore inode should itself 277 * specify the disk inode number corresponding to the 278 * incore inode (i.e. i_number should be initialized). 279 * 280 * this routine handles incore inode initialization for 281 * both "special" and "regular" inodes. special inodes 282 * are those required early in the mount process and 283 * require special handling since much of the file system 284 * is not yet initialized. these "special" inodes are 285 * identified by a NULL inode map inode pointer and are 286 * actually initialized by a call to diReadSpecial(). 287 * 288 * for regular inodes, the iag describing the disk inode 289 * is read from disk to determine the inode extent address 290 * for the disk inode. with the inode extent address in 291 * hand, the page of the extent that contains the disk 292 * inode is read and the disk inode is copied to the 293 * incore inode. 294 * 295 * PARAMETERS: 296 * ip - pointer to incore inode to be initialized from disk. 297 * 298 * RETURN VALUES: 299 * 0 - success 300 * -EIO - i/o error. 301 * -ENOMEM - insufficient memory 302 * 303 */ 304 int diRead(struct inode *ip) 305 { 306 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 307 int iagno, ino, extno, rc; 308 struct inode *ipimap; 309 struct dinode *dp; 310 struct iag *iagp; 311 struct metapage *mp; 312 s64 blkno, agstart; 313 struct inomap *imap; 314 int block_offset; 315 int inodes_left; 316 unsigned long pageno; 317 int rel_inode; 318 319 jfs_info("diRead: ino = %ld", ip->i_ino); 320 321 ipimap = sbi->ipimap; 322 JFS_IP(ip)->ipimap = ipimap; 323 324 /* determine the iag number for this inode (number) */ 325 iagno = INOTOIAG(ip->i_ino); 326 327 /* read the iag */ 328 imap = JFS_IP(ipimap)->i_imap; 329 IREAD_LOCK(ipimap, RDWRLOCK_IMAP); 330 rc = diIAGRead(imap, iagno, &mp); 331 IREAD_UNLOCK(ipimap); 332 if (rc) { 333 jfs_err("diRead: diIAGRead returned %d", rc); 334 return (rc); 335 } 336 337 iagp = (struct iag *) mp->data; 338 339 /* determine inode extent that holds the disk inode */ 340 ino = ip->i_ino & (INOSPERIAG - 1); 341 extno = ino >> L2INOSPEREXT; 342 343 if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) || 344 (addressPXD(&iagp->inoext[extno]) == 0)) { 345 release_metapage(mp); 346 return -ESTALE; 347 } 348 349 /* get disk block number of the page within the inode extent 350 * that holds the disk inode. 351 */ 352 blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage); 353 354 /* get the ag for the iag */ 355 agstart = le64_to_cpu(iagp->agstart); 356 357 release_metapage(mp); 358 359 rel_inode = (ino & (INOSPERPAGE - 1)); 360 pageno = blkno >> sbi->l2nbperpage; 361 362 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) { 363 /* 364 * OS/2 didn't always align inode extents on page boundaries 365 */ 366 inodes_left = 367 (sbi->nbperpage - block_offset) << sbi->l2niperblk; 368 369 if (rel_inode < inodes_left) 370 rel_inode += block_offset << sbi->l2niperblk; 371 else { 372 pageno += 1; 373 rel_inode -= inodes_left; 374 } 375 } 376 377 /* read the page of disk inode */ 378 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1); 379 if (!mp) { 380 jfs_err("diRead: read_metapage failed"); 381 return -EIO; 382 } 383 384 /* locate the disk inode requested */ 385 dp = (struct dinode *) mp->data; 386 dp += rel_inode; 387 388 if (ip->i_ino != le32_to_cpu(dp->di_number)) { 389 jfs_error(ip->i_sb, "i_ino != di_number\n"); 390 rc = -EIO; 391 } else if (le32_to_cpu(dp->di_nlink) == 0) 392 rc = -ESTALE; 393 else 394 /* copy the disk inode to the in-memory inode */ 395 rc = copy_from_dinode(dp, ip); 396 397 release_metapage(mp); 398 399 /* set the ag for the inode */ 400 JFS_IP(ip)->agstart = agstart; 401 JFS_IP(ip)->active_ag = -1; 402 403 return (rc); 404 } 405 406 407 /* 408 * NAME: diReadSpecial() 409 * 410 * FUNCTION: initialize a 'special' inode from disk. 411 * 412 * this routines handles aggregate level inodes. The 413 * inode cache cannot differentiate between the 414 * aggregate inodes and the filesystem inodes, so we 415 * handle these here. We don't actually use the aggregate 416 * inode map, since these inodes are at a fixed location 417 * and in some cases the aggregate inode map isn't initialized 418 * yet. 419 * 420 * PARAMETERS: 421 * sb - filesystem superblock 422 * inum - aggregate inode number 423 * secondary - 1 if secondary aggregate inode table 424 * 425 * RETURN VALUES: 426 * new inode - success 427 * NULL - i/o error. 428 */ 429 struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary) 430 { 431 struct jfs_sb_info *sbi = JFS_SBI(sb); 432 uint address; 433 struct dinode *dp; 434 struct inode *ip; 435 struct metapage *mp; 436 437 ip = new_inode(sb); 438 if (ip == NULL) { 439 jfs_err("diReadSpecial: new_inode returned NULL!"); 440 return ip; 441 } 442 443 if (secondary) { 444 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage; 445 JFS_IP(ip)->ipimap = sbi->ipaimap2; 446 } else { 447 address = AITBL_OFF >> L2PSIZE; 448 JFS_IP(ip)->ipimap = sbi->ipaimap; 449 } 450 451 ASSERT(inum < INOSPEREXT); 452 453 ip->i_ino = inum; 454 455 address += inum >> 3; /* 8 inodes per 4K page */ 456 457 /* read the page of fixed disk inode (AIT) in raw mode */ 458 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1); 459 if (mp == NULL) { 460 set_nlink(ip, 1); /* Don't want iput() deleting it */ 461 iput(ip); 462 return (NULL); 463 } 464 465 /* get the pointer to the disk inode of interest */ 466 dp = (struct dinode *) (mp->data); 467 dp += inum % 8; /* 8 inodes per 4K page */ 468 469 /* copy on-disk inode to in-memory inode */ 470 if ((copy_from_dinode(dp, ip)) != 0) { 471 /* handle bad return by returning NULL for ip */ 472 set_nlink(ip, 1); /* Don't want iput() deleting it */ 473 iput(ip); 474 /* release the page */ 475 release_metapage(mp); 476 return (NULL); 477 478 } 479 480 ip->i_mapping->a_ops = &jfs_metapage_aops; 481 mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS); 482 483 /* Allocations to metadata inodes should not affect quotas */ 484 ip->i_flags |= S_NOQUOTA; 485 486 if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) { 487 sbi->gengen = le32_to_cpu(dp->di_gengen); 488 sbi->inostamp = le32_to_cpu(dp->di_inostamp); 489 } 490 491 /* release the page */ 492 release_metapage(mp); 493 494 /* 495 * __mark_inode_dirty expects inodes to be hashed. Since we don't 496 * want special inodes in the fileset inode space, we make them 497 * appear hashed, but do not put on any lists. hlist_del() 498 * will work fine and require no locking. 499 */ 500 hlist_add_fake(&ip->i_hash); 501 502 return (ip); 503 } 504 505 /* 506 * NAME: diWriteSpecial() 507 * 508 * FUNCTION: Write the special inode to disk 509 * 510 * PARAMETERS: 511 * ip - special inode 512 * secondary - 1 if secondary aggregate inode table 513 * 514 * RETURN VALUES: none 515 */ 516 517 void diWriteSpecial(struct inode *ip, int secondary) 518 { 519 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 520 uint address; 521 struct dinode *dp; 522 ino_t inum = ip->i_ino; 523 struct metapage *mp; 524 525 if (secondary) 526 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage; 527 else 528 address = AITBL_OFF >> L2PSIZE; 529 530 ASSERT(inum < INOSPEREXT); 531 532 address += inum >> 3; /* 8 inodes per 4K page */ 533 534 /* read the page of fixed disk inode (AIT) in raw mode */ 535 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1); 536 if (mp == NULL) { 537 jfs_err("diWriteSpecial: failed to read aggregate inode extent!"); 538 return; 539 } 540 541 /* get the pointer to the disk inode of interest */ 542 dp = (struct dinode *) (mp->data); 543 dp += inum % 8; /* 8 inodes per 4K page */ 544 545 /* copy on-disk inode to in-memory inode */ 546 copy_to_dinode(dp, ip); 547 memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288); 548 549 if (inum == FILESYSTEM_I) 550 dp->di_gengen = cpu_to_le32(sbi->gengen); 551 552 /* write the page */ 553 write_metapage(mp); 554 } 555 556 /* 557 * NAME: diFreeSpecial() 558 * 559 * FUNCTION: Free allocated space for special inode 560 */ 561 void diFreeSpecial(struct inode *ip) 562 { 563 if (ip == NULL) { 564 jfs_err("diFreeSpecial called with NULL ip!"); 565 return; 566 } 567 filemap_write_and_wait(ip->i_mapping); 568 truncate_inode_pages(ip->i_mapping, 0); 569 iput(ip); 570 } 571 572 573 574 /* 575 * NAME: diWrite() 576 * 577 * FUNCTION: write the on-disk inode portion of the in-memory inode 578 * to its corresponding on-disk inode. 579 * 580 * on entry, the specifed incore inode should itself 581 * specify the disk inode number corresponding to the 582 * incore inode (i.e. i_number should be initialized). 583 * 584 * the inode contains the inode extent address for the disk 585 * inode. with the inode extent address in hand, the 586 * page of the extent that contains the disk inode is 587 * read and the disk inode portion of the incore inode 588 * is copied to the disk inode. 589 * 590 * PARAMETERS: 591 * tid - transacation id 592 * ip - pointer to incore inode to be written to the inode extent. 593 * 594 * RETURN VALUES: 595 * 0 - success 596 * -EIO - i/o error. 597 */ 598 int diWrite(tid_t tid, struct inode *ip) 599 { 600 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 601 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 602 int rc = 0; 603 s32 ino; 604 struct dinode *dp; 605 s64 blkno; 606 int block_offset; 607 int inodes_left; 608 struct metapage *mp; 609 unsigned long pageno; 610 int rel_inode; 611 int dioffset; 612 struct inode *ipimap; 613 uint type; 614 lid_t lid; 615 struct tlock *ditlck, *tlck; 616 struct linelock *dilinelock, *ilinelock; 617 struct lv *lv; 618 int n; 619 620 ipimap = jfs_ip->ipimap; 621 622 ino = ip->i_ino & (INOSPERIAG - 1); 623 624 if (!addressPXD(&(jfs_ip->ixpxd)) || 625 (lengthPXD(&(jfs_ip->ixpxd)) != 626 JFS_IP(ipimap)->i_imap->im_nbperiext)) { 627 jfs_error(ip->i_sb, "ixpxd invalid\n"); 628 return -EIO; 629 } 630 631 /* 632 * read the page of disk inode containing the specified inode: 633 */ 634 /* compute the block address of the page */ 635 blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage); 636 637 rel_inode = (ino & (INOSPERPAGE - 1)); 638 pageno = blkno >> sbi->l2nbperpage; 639 640 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) { 641 /* 642 * OS/2 didn't always align inode extents on page boundaries 643 */ 644 inodes_left = 645 (sbi->nbperpage - block_offset) << sbi->l2niperblk; 646 647 if (rel_inode < inodes_left) 648 rel_inode += block_offset << sbi->l2niperblk; 649 else { 650 pageno += 1; 651 rel_inode -= inodes_left; 652 } 653 } 654 /* read the page of disk inode */ 655 retry: 656 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1); 657 if (!mp) 658 return -EIO; 659 660 /* get the pointer to the disk inode */ 661 dp = (struct dinode *) mp->data; 662 dp += rel_inode; 663 664 dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE; 665 666 /* 667 * acquire transaction lock on the on-disk inode; 668 * N.B. tlock is acquired on ipimap not ip; 669 */ 670 if ((ditlck = 671 txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL) 672 goto retry; 673 dilinelock = (struct linelock *) & ditlck->lock; 674 675 /* 676 * copy btree root from in-memory inode to on-disk inode 677 * 678 * (tlock is taken from inline B+-tree root in in-memory 679 * inode when the B+-tree root is updated, which is pointed 680 * by jfs_ip->blid as well as being on tx tlock list) 681 * 682 * further processing of btree root is based on the copy 683 * in in-memory inode, where txLog() will log from, and, 684 * for xtree root, txUpdateMap() will update map and reset 685 * XAD_NEW bit; 686 */ 687 688 if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) { 689 /* 690 * This is the special xtree inside the directory for storing 691 * the directory table 692 */ 693 xtpage_t *p, *xp; 694 xad_t *xad; 695 696 jfs_ip->xtlid = 0; 697 tlck = lid_to_tlock(lid); 698 assert(tlck->type & tlckXTREE); 699 tlck->type |= tlckBTROOT; 700 tlck->mp = mp; 701 ilinelock = (struct linelock *) & tlck->lock; 702 703 /* 704 * copy xtree root from inode to dinode: 705 */ 706 p = &jfs_ip->i_xtroot; 707 xp = (xtpage_t *) &dp->di_dirtable; 708 lv = ilinelock->lv; 709 for (n = 0; n < ilinelock->index; n++, lv++) { 710 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset], 711 lv->length << L2XTSLOTSIZE); 712 } 713 714 /* reset on-disk (metadata page) xtree XAD_NEW bit */ 715 xad = &xp->xad[XTENTRYSTART]; 716 for (n = XTENTRYSTART; 717 n < le16_to_cpu(xp->header.nextindex); n++, xad++) 718 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) 719 xad->flag &= ~(XAD_NEW | XAD_EXTENDED); 720 } 721 722 if ((lid = jfs_ip->blid) == 0) 723 goto inlineData; 724 jfs_ip->blid = 0; 725 726 tlck = lid_to_tlock(lid); 727 type = tlck->type; 728 tlck->type |= tlckBTROOT; 729 tlck->mp = mp; 730 ilinelock = (struct linelock *) & tlck->lock; 731 732 /* 733 * regular file: 16 byte (XAD slot) granularity 734 */ 735 if (type & tlckXTREE) { 736 xtpage_t *p, *xp; 737 xad_t *xad; 738 739 /* 740 * copy xtree root from inode to dinode: 741 */ 742 p = &jfs_ip->i_xtroot; 743 xp = &dp->di_xtroot; 744 lv = ilinelock->lv; 745 for (n = 0; n < ilinelock->index; n++, lv++) { 746 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset], 747 lv->length << L2XTSLOTSIZE); 748 } 749 750 /* reset on-disk (metadata page) xtree XAD_NEW bit */ 751 xad = &xp->xad[XTENTRYSTART]; 752 for (n = XTENTRYSTART; 753 n < le16_to_cpu(xp->header.nextindex); n++, xad++) 754 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) 755 xad->flag &= ~(XAD_NEW | XAD_EXTENDED); 756 } 757 /* 758 * directory: 32 byte (directory entry slot) granularity 759 */ 760 else if (type & tlckDTREE) { 761 dtpage_t *p, *xp; 762 763 /* 764 * copy dtree root from inode to dinode: 765 */ 766 p = (dtpage_t *) &jfs_ip->i_dtroot; 767 xp = (dtpage_t *) & dp->di_dtroot; 768 lv = ilinelock->lv; 769 for (n = 0; n < ilinelock->index; n++, lv++) { 770 memcpy(&xp->slot[lv->offset], &p->slot[lv->offset], 771 lv->length << L2DTSLOTSIZE); 772 } 773 } else { 774 jfs_err("diWrite: UFO tlock"); 775 } 776 777 inlineData: 778 /* 779 * copy inline symlink from in-memory inode to on-disk inode 780 */ 781 if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) { 782 lv = & dilinelock->lv[dilinelock->index]; 783 lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE; 784 lv->length = 2; 785 memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE); 786 dilinelock->index++; 787 } 788 /* 789 * copy inline data from in-memory inode to on-disk inode: 790 * 128 byte slot granularity 791 */ 792 if (test_cflag(COMMIT_Inlineea, ip)) { 793 lv = & dilinelock->lv[dilinelock->index]; 794 lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE; 795 lv->length = 1; 796 memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE); 797 dilinelock->index++; 798 799 clear_cflag(COMMIT_Inlineea, ip); 800 } 801 802 /* 803 * lock/copy inode base: 128 byte slot granularity 804 */ 805 lv = & dilinelock->lv[dilinelock->index]; 806 lv->offset = dioffset >> L2INODESLOTSIZE; 807 copy_to_dinode(dp, ip); 808 if (test_and_clear_cflag(COMMIT_Dirtable, ip)) { 809 lv->length = 2; 810 memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96); 811 } else 812 lv->length = 1; 813 dilinelock->index++; 814 815 /* release the buffer holding the updated on-disk inode. 816 * the buffer will be later written by commit processing. 817 */ 818 write_metapage(mp); 819 820 return (rc); 821 } 822 823 824 /* 825 * NAME: diFree(ip) 826 * 827 * FUNCTION: free a specified inode from the inode working map 828 * for a fileset or aggregate. 829 * 830 * if the inode to be freed represents the first (only) 831 * free inode within the iag, the iag will be placed on 832 * the ag free inode list. 833 * 834 * freeing the inode will cause the inode extent to be 835 * freed if the inode is the only allocated inode within 836 * the extent. in this case all the disk resource backing 837 * up the inode extent will be freed. in addition, the iag 838 * will be placed on the ag extent free list if the extent 839 * is the first free extent in the iag. if freeing the 840 * extent also means that no free inodes will exist for 841 * the iag, the iag will also be removed from the ag free 842 * inode list. 843 * 844 * the iag describing the inode will be freed if the extent 845 * is to be freed and it is the only backed extent within 846 * the iag. in this case, the iag will be removed from the 847 * ag free extent list and ag free inode list and placed on 848 * the inode map's free iag list. 849 * 850 * a careful update approach is used to provide consistency 851 * in the face of updates to multiple buffers. under this 852 * approach, all required buffers are obtained before making 853 * any updates and are held until all updates are complete. 854 * 855 * PARAMETERS: 856 * ip - inode to be freed. 857 * 858 * RETURN VALUES: 859 * 0 - success 860 * -EIO - i/o error. 861 */ 862 int diFree(struct inode *ip) 863 { 864 int rc; 865 ino_t inum = ip->i_ino; 866 struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp; 867 struct metapage *mp, *amp, *bmp, *cmp, *dmp; 868 int iagno, ino, extno, bitno, sword, agno; 869 int back, fwd; 870 u32 bitmap, mask; 871 struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap; 872 struct inomap *imap = JFS_IP(ipimap)->i_imap; 873 pxd_t freepxd; 874 tid_t tid; 875 struct inode *iplist[3]; 876 struct tlock *tlck; 877 struct pxd_lock *pxdlock; 878 879 /* 880 * This is just to suppress compiler warnings. The same logic that 881 * references these variables is used to initialize them. 882 */ 883 aiagp = biagp = ciagp = diagp = NULL; 884 885 /* get the iag number containing the inode. 886 */ 887 iagno = INOTOIAG(inum); 888 889 /* make sure that the iag is contained within 890 * the map. 891 */ 892 if (iagno >= imap->im_nextiag) { 893 print_hex_dump(KERN_ERR, "imap: ", DUMP_PREFIX_ADDRESS, 16, 4, 894 imap, 32, 0); 895 jfs_error(ip->i_sb, "inum = %d, iagno = %d, nextiag = %d\n", 896 (uint) inum, iagno, imap->im_nextiag); 897 return -EIO; 898 } 899 900 /* get the allocation group for this ino. 901 */ 902 agno = BLKTOAG(JFS_IP(ip)->agstart, JFS_SBI(ip->i_sb)); 903 904 /* Lock the AG specific inode map information 905 */ 906 AG_LOCK(imap, agno); 907 908 /* Obtain read lock in imap inode. Don't release it until we have 909 * read all of the IAG's that we are going to. 910 */ 911 IREAD_LOCK(ipimap, RDWRLOCK_IMAP); 912 913 /* read the iag. 914 */ 915 if ((rc = diIAGRead(imap, iagno, &mp))) { 916 IREAD_UNLOCK(ipimap); 917 AG_UNLOCK(imap, agno); 918 return (rc); 919 } 920 iagp = (struct iag *) mp->data; 921 922 /* get the inode number and extent number of the inode within 923 * the iag and the inode number within the extent. 924 */ 925 ino = inum & (INOSPERIAG - 1); 926 extno = ino >> L2INOSPEREXT; 927 bitno = ino & (INOSPEREXT - 1); 928 mask = HIGHORDER >> bitno; 929 930 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { 931 jfs_error(ip->i_sb, "wmap shows inode already free\n"); 932 } 933 934 if (!addressPXD(&iagp->inoext[extno])) { 935 release_metapage(mp); 936 IREAD_UNLOCK(ipimap); 937 AG_UNLOCK(imap, agno); 938 jfs_error(ip->i_sb, "invalid inoext\n"); 939 return -EIO; 940 } 941 942 /* compute the bitmap for the extent reflecting the freed inode. 943 */ 944 bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask; 945 946 if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) { 947 release_metapage(mp); 948 IREAD_UNLOCK(ipimap); 949 AG_UNLOCK(imap, agno); 950 jfs_error(ip->i_sb, "numfree > numinos\n"); 951 return -EIO; 952 } 953 /* 954 * inode extent still has some inodes or below low water mark: 955 * keep the inode extent; 956 */ 957 if (bitmap || 958 imap->im_agctl[agno].numfree < 96 || 959 (imap->im_agctl[agno].numfree < 288 && 960 (((imap->im_agctl[agno].numfree * 100) / 961 imap->im_agctl[agno].numinos) <= 25))) { 962 /* if the iag currently has no free inodes (i.e., 963 * the inode being freed is the first free inode of iag), 964 * insert the iag at head of the inode free list for the ag. 965 */ 966 if (iagp->nfreeinos == 0) { 967 /* check if there are any iags on the ag inode 968 * free list. if so, read the first one so that 969 * we can link the current iag onto the list at 970 * the head. 971 */ 972 if ((fwd = imap->im_agctl[agno].inofree) >= 0) { 973 /* read the iag that currently is the head 974 * of the list. 975 */ 976 if ((rc = diIAGRead(imap, fwd, &))) { 977 IREAD_UNLOCK(ipimap); 978 AG_UNLOCK(imap, agno); 979 release_metapage(mp); 980 return (rc); 981 } 982 aiagp = (struct iag *) amp->data; 983 984 /* make current head point back to the iag. 985 */ 986 aiagp->inofreeback = cpu_to_le32(iagno); 987 988 write_metapage(amp); 989 } 990 991 /* iag points forward to current head and iag 992 * becomes the new head of the list. 993 */ 994 iagp->inofreefwd = 995 cpu_to_le32(imap->im_agctl[agno].inofree); 996 iagp->inofreeback = cpu_to_le32(-1); 997 imap->im_agctl[agno].inofree = iagno; 998 } 999 IREAD_UNLOCK(ipimap); 1000 1001 /* update the free inode summary map for the extent if 1002 * freeing the inode means the extent will now have free 1003 * inodes (i.e., the inode being freed is the first free 1004 * inode of extent), 1005 */ 1006 if (iagp->wmap[extno] == cpu_to_le32(ONES)) { 1007 sword = extno >> L2EXTSPERSUM; 1008 bitno = extno & (EXTSPERSUM - 1); 1009 iagp->inosmap[sword] &= 1010 cpu_to_le32(~(HIGHORDER >> bitno)); 1011 } 1012 1013 /* update the bitmap. 1014 */ 1015 iagp->wmap[extno] = cpu_to_le32(bitmap); 1016 1017 /* update the free inode counts at the iag, ag and 1018 * map level. 1019 */ 1020 le32_add_cpu(&iagp->nfreeinos, 1); 1021 imap->im_agctl[agno].numfree += 1; 1022 atomic_inc(&imap->im_numfree); 1023 1024 /* release the AG inode map lock 1025 */ 1026 AG_UNLOCK(imap, agno); 1027 1028 /* write the iag */ 1029 write_metapage(mp); 1030 1031 return (0); 1032 } 1033 1034 1035 /* 1036 * inode extent has become free and above low water mark: 1037 * free the inode extent; 1038 */ 1039 1040 /* 1041 * prepare to update iag list(s) (careful update step 1) 1042 */ 1043 amp = bmp = cmp = dmp = NULL; 1044 fwd = back = -1; 1045 1046 /* check if the iag currently has no free extents. if so, 1047 * it will be placed on the head of the ag extent free list. 1048 */ 1049 if (iagp->nfreeexts == 0) { 1050 /* check if the ag extent free list has any iags. 1051 * if so, read the iag at the head of the list now. 1052 * this (head) iag will be updated later to reflect 1053 * the addition of the current iag at the head of 1054 * the list. 1055 */ 1056 if ((fwd = imap->im_agctl[agno].extfree) >= 0) { 1057 if ((rc = diIAGRead(imap, fwd, &))) 1058 goto error_out; 1059 aiagp = (struct iag *) amp->data; 1060 } 1061 } else { 1062 /* iag has free extents. check if the addition of a free 1063 * extent will cause all extents to be free within this 1064 * iag. if so, the iag will be removed from the ag extent 1065 * free list and placed on the inode map's free iag list. 1066 */ 1067 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) { 1068 /* in preparation for removing the iag from the 1069 * ag extent free list, read the iags preceding 1070 * and following the iag on the ag extent free 1071 * list. 1072 */ 1073 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) { 1074 if ((rc = diIAGRead(imap, fwd, &))) 1075 goto error_out; 1076 aiagp = (struct iag *) amp->data; 1077 } 1078 1079 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) { 1080 if ((rc = diIAGRead(imap, back, &bmp))) 1081 goto error_out; 1082 biagp = (struct iag *) bmp->data; 1083 } 1084 } 1085 } 1086 1087 /* remove the iag from the ag inode free list if freeing 1088 * this extent cause the iag to have no free inodes. 1089 */ 1090 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) { 1091 int inofreeback = le32_to_cpu(iagp->inofreeback); 1092 int inofreefwd = le32_to_cpu(iagp->inofreefwd); 1093 1094 /* in preparation for removing the iag from the 1095 * ag inode free list, read the iags preceding 1096 * and following the iag on the ag inode free 1097 * list. before reading these iags, we must make 1098 * sure that we already don't have them in hand 1099 * from up above, since re-reading an iag (buffer) 1100 * we are currently holding would cause a deadlock. 1101 */ 1102 if (inofreefwd >= 0) { 1103 1104 if (inofreefwd == fwd) 1105 ciagp = (struct iag *) amp->data; 1106 else if (inofreefwd == back) 1107 ciagp = (struct iag *) bmp->data; 1108 else { 1109 if ((rc = 1110 diIAGRead(imap, inofreefwd, &cmp))) 1111 goto error_out; 1112 ciagp = (struct iag *) cmp->data; 1113 } 1114 assert(ciagp != NULL); 1115 } 1116 1117 if (inofreeback >= 0) { 1118 if (inofreeback == fwd) 1119 diagp = (struct iag *) amp->data; 1120 else if (inofreeback == back) 1121 diagp = (struct iag *) bmp->data; 1122 else { 1123 if ((rc = 1124 diIAGRead(imap, inofreeback, &dmp))) 1125 goto error_out; 1126 diagp = (struct iag *) dmp->data; 1127 } 1128 assert(diagp != NULL); 1129 } 1130 } 1131 1132 IREAD_UNLOCK(ipimap); 1133 1134 /* 1135 * invalidate any page of the inode extent freed from buffer cache; 1136 */ 1137 freepxd = iagp->inoext[extno]; 1138 invalidate_pxd_metapages(ip, freepxd); 1139 1140 /* 1141 * update iag list(s) (careful update step 2) 1142 */ 1143 /* add the iag to the ag extent free list if this is the 1144 * first free extent for the iag. 1145 */ 1146 if (iagp->nfreeexts == 0) { 1147 if (fwd >= 0) 1148 aiagp->extfreeback = cpu_to_le32(iagno); 1149 1150 iagp->extfreefwd = 1151 cpu_to_le32(imap->im_agctl[agno].extfree); 1152 iagp->extfreeback = cpu_to_le32(-1); 1153 imap->im_agctl[agno].extfree = iagno; 1154 } else { 1155 /* remove the iag from the ag extent list if all extents 1156 * are now free and place it on the inode map iag free list. 1157 */ 1158 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) { 1159 if (fwd >= 0) 1160 aiagp->extfreeback = iagp->extfreeback; 1161 1162 if (back >= 0) 1163 biagp->extfreefwd = iagp->extfreefwd; 1164 else 1165 imap->im_agctl[agno].extfree = 1166 le32_to_cpu(iagp->extfreefwd); 1167 1168 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); 1169 1170 IAGFREE_LOCK(imap); 1171 iagp->iagfree = cpu_to_le32(imap->im_freeiag); 1172 imap->im_freeiag = iagno; 1173 IAGFREE_UNLOCK(imap); 1174 } 1175 } 1176 1177 /* remove the iag from the ag inode free list if freeing 1178 * this extent causes the iag to have no free inodes. 1179 */ 1180 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) { 1181 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) 1182 ciagp->inofreeback = iagp->inofreeback; 1183 1184 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) 1185 diagp->inofreefwd = iagp->inofreefwd; 1186 else 1187 imap->im_agctl[agno].inofree = 1188 le32_to_cpu(iagp->inofreefwd); 1189 1190 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); 1191 } 1192 1193 /* update the inode extent address and working map 1194 * to reflect the free extent. 1195 * the permanent map should have been updated already 1196 * for the inode being freed. 1197 */ 1198 if (iagp->pmap[extno] != 0) { 1199 jfs_error(ip->i_sb, "the pmap does not show inode free\n"); 1200 } 1201 iagp->wmap[extno] = 0; 1202 PXDlength(&iagp->inoext[extno], 0); 1203 PXDaddress(&iagp->inoext[extno], 0); 1204 1205 /* update the free extent and free inode summary maps 1206 * to reflect the freed extent. 1207 * the inode summary map is marked to indicate no inodes 1208 * available for the freed extent. 1209 */ 1210 sword = extno >> L2EXTSPERSUM; 1211 bitno = extno & (EXTSPERSUM - 1); 1212 mask = HIGHORDER >> bitno; 1213 iagp->inosmap[sword] |= cpu_to_le32(mask); 1214 iagp->extsmap[sword] &= cpu_to_le32(~mask); 1215 1216 /* update the number of free inodes and number of free extents 1217 * for the iag. 1218 */ 1219 le32_add_cpu(&iagp->nfreeinos, -(INOSPEREXT - 1)); 1220 le32_add_cpu(&iagp->nfreeexts, 1); 1221 1222 /* update the number of free inodes and backed inodes 1223 * at the ag and inode map level. 1224 */ 1225 imap->im_agctl[agno].numfree -= (INOSPEREXT - 1); 1226 imap->im_agctl[agno].numinos -= INOSPEREXT; 1227 atomic_sub(INOSPEREXT - 1, &imap->im_numfree); 1228 atomic_sub(INOSPEREXT, &imap->im_numinos); 1229 1230 if (amp) 1231 write_metapage(amp); 1232 if (bmp) 1233 write_metapage(bmp); 1234 if (cmp) 1235 write_metapage(cmp); 1236 if (dmp) 1237 write_metapage(dmp); 1238 1239 /* 1240 * start transaction to update block allocation map 1241 * for the inode extent freed; 1242 * 1243 * N.B. AG_LOCK is released and iag will be released below, and 1244 * other thread may allocate inode from/reusing the ixad freed 1245 * BUT with new/different backing inode extent from the extent 1246 * to be freed by the transaction; 1247 */ 1248 tid = txBegin(ipimap->i_sb, COMMIT_FORCE); 1249 mutex_lock(&JFS_IP(ipimap)->commit_mutex); 1250 1251 /* acquire tlock of the iag page of the freed ixad 1252 * to force the page NOHOMEOK (even though no data is 1253 * logged from the iag page) until NOREDOPAGE|FREEXTENT log 1254 * for the free of the extent is committed; 1255 * write FREEXTENT|NOREDOPAGE log record 1256 * N.B. linelock is overlaid as freed extent descriptor; 1257 */ 1258 tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE); 1259 pxdlock = (struct pxd_lock *) & tlck->lock; 1260 pxdlock->flag = mlckFREEPXD; 1261 pxdlock->pxd = freepxd; 1262 pxdlock->index = 1; 1263 1264 write_metapage(mp); 1265 1266 iplist[0] = ipimap; 1267 1268 /* 1269 * logredo needs the IAG number and IAG extent index in order 1270 * to ensure that the IMap is consistent. The least disruptive 1271 * way to pass these values through to the transaction manager 1272 * is in the iplist array. 1273 * 1274 * It's not pretty, but it works. 1275 */ 1276 iplist[1] = (struct inode *) (size_t)iagno; 1277 iplist[2] = (struct inode *) (size_t)extno; 1278 1279 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE); 1280 1281 txEnd(tid); 1282 mutex_unlock(&JFS_IP(ipimap)->commit_mutex); 1283 1284 /* unlock the AG inode map information */ 1285 AG_UNLOCK(imap, agno); 1286 1287 return (0); 1288 1289 error_out: 1290 IREAD_UNLOCK(ipimap); 1291 1292 if (amp) 1293 release_metapage(amp); 1294 if (bmp) 1295 release_metapage(bmp); 1296 if (cmp) 1297 release_metapage(cmp); 1298 if (dmp) 1299 release_metapage(dmp); 1300 1301 AG_UNLOCK(imap, agno); 1302 1303 release_metapage(mp); 1304 1305 return (rc); 1306 } 1307 1308 /* 1309 * There are several places in the diAlloc* routines where we initialize 1310 * the inode. 1311 */ 1312 static inline void 1313 diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp) 1314 { 1315 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 1316 1317 ip->i_ino = (iagno << L2INOSPERIAG) + ino; 1318 jfs_ip->ixpxd = iagp->inoext[extno]; 1319 jfs_ip->agstart = le64_to_cpu(iagp->agstart); 1320 jfs_ip->active_ag = -1; 1321 } 1322 1323 1324 /* 1325 * NAME: diAlloc(pip,dir,ip) 1326 * 1327 * FUNCTION: allocate a disk inode from the inode working map 1328 * for a fileset or aggregate. 1329 * 1330 * PARAMETERS: 1331 * pip - pointer to incore inode for the parent inode. 1332 * dir - 'true' if the new disk inode is for a directory. 1333 * ip - pointer to a new inode 1334 * 1335 * RETURN VALUES: 1336 * 0 - success. 1337 * -ENOSPC - insufficient disk resources. 1338 * -EIO - i/o error. 1339 */ 1340 int diAlloc(struct inode *pip, bool dir, struct inode *ip) 1341 { 1342 int rc, ino, iagno, addext, extno, bitno, sword; 1343 int nwords, rem, i, agno; 1344 u32 mask, inosmap, extsmap; 1345 struct inode *ipimap; 1346 struct metapage *mp; 1347 ino_t inum; 1348 struct iag *iagp; 1349 struct inomap *imap; 1350 1351 /* get the pointers to the inode map inode and the 1352 * corresponding imap control structure. 1353 */ 1354 ipimap = JFS_SBI(pip->i_sb)->ipimap; 1355 imap = JFS_IP(ipimap)->i_imap; 1356 JFS_IP(ip)->ipimap = ipimap; 1357 JFS_IP(ip)->fileset = FILESYSTEM_I; 1358 1359 /* for a directory, the allocation policy is to start 1360 * at the ag level using the preferred ag. 1361 */ 1362 if (dir) { 1363 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap); 1364 AG_LOCK(imap, agno); 1365 goto tryag; 1366 } 1367 1368 /* for files, the policy starts off by trying to allocate from 1369 * the same iag containing the parent disk inode: 1370 * try to allocate the new disk inode close to the parent disk 1371 * inode, using parent disk inode number + 1 as the allocation 1372 * hint. (we use a left-to-right policy to attempt to avoid 1373 * moving backward on the disk.) compute the hint within the 1374 * file system and the iag. 1375 */ 1376 1377 /* get the ag number of this iag */ 1378 agno = BLKTOAG(JFS_IP(pip)->agstart, JFS_SBI(pip->i_sb)); 1379 1380 if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) { 1381 /* 1382 * There is an open file actively growing. We want to 1383 * allocate new inodes from a different ag to avoid 1384 * fragmentation problems. 1385 */ 1386 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap); 1387 AG_LOCK(imap, agno); 1388 goto tryag; 1389 } 1390 1391 inum = pip->i_ino + 1; 1392 ino = inum & (INOSPERIAG - 1); 1393 1394 /* back off the hint if it is outside of the iag */ 1395 if (ino == 0) 1396 inum = pip->i_ino; 1397 1398 /* lock the AG inode map information */ 1399 AG_LOCK(imap, agno); 1400 1401 /* Get read lock on imap inode */ 1402 IREAD_LOCK(ipimap, RDWRLOCK_IMAP); 1403 1404 /* get the iag number and read the iag */ 1405 iagno = INOTOIAG(inum); 1406 if ((rc = diIAGRead(imap, iagno, &mp))) { 1407 IREAD_UNLOCK(ipimap); 1408 AG_UNLOCK(imap, agno); 1409 return (rc); 1410 } 1411 iagp = (struct iag *) mp->data; 1412 1413 /* determine if new inode extent is allowed to be added to the iag. 1414 * new inode extent can be added to the iag if the ag 1415 * has less than 32 free disk inodes and the iag has free extents. 1416 */ 1417 addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts); 1418 1419 /* 1420 * try to allocate from the IAG 1421 */ 1422 /* check if the inode may be allocated from the iag 1423 * (i.e. the inode has free inodes or new extent can be added). 1424 */ 1425 if (iagp->nfreeinos || addext) { 1426 /* determine the extent number of the hint. 1427 */ 1428 extno = ino >> L2INOSPEREXT; 1429 1430 /* check if the extent containing the hint has backed 1431 * inodes. if so, try to allocate within this extent. 1432 */ 1433 if (addressPXD(&iagp->inoext[extno])) { 1434 bitno = ino & (INOSPEREXT - 1); 1435 if ((bitno = 1436 diFindFree(le32_to_cpu(iagp->wmap[extno]), 1437 bitno)) 1438 < INOSPEREXT) { 1439 ino = (extno << L2INOSPEREXT) + bitno; 1440 1441 /* a free inode (bit) was found within this 1442 * extent, so allocate it. 1443 */ 1444 rc = diAllocBit(imap, iagp, ino); 1445 IREAD_UNLOCK(ipimap); 1446 if (rc) { 1447 assert(rc == -EIO); 1448 } else { 1449 /* set the results of the allocation 1450 * and write the iag. 1451 */ 1452 diInitInode(ip, iagno, ino, extno, 1453 iagp); 1454 mark_metapage_dirty(mp); 1455 } 1456 release_metapage(mp); 1457 1458 /* free the AG lock and return. 1459 */ 1460 AG_UNLOCK(imap, agno); 1461 return (rc); 1462 } 1463 1464 if (!addext) 1465 extno = 1466 (extno == 1467 EXTSPERIAG - 1) ? 0 : extno + 1; 1468 } 1469 1470 /* 1471 * no free inodes within the extent containing the hint. 1472 * 1473 * try to allocate from the backed extents following 1474 * hint or, if appropriate (i.e. addext is true), allocate 1475 * an extent of free inodes at or following the extent 1476 * containing the hint. 1477 * 1478 * the free inode and free extent summary maps are used 1479 * here, so determine the starting summary map position 1480 * and the number of words we'll have to examine. again, 1481 * the approach is to allocate following the hint, so we 1482 * might have to initially ignore prior bits of the summary 1483 * map that represent extents prior to the extent containing 1484 * the hint and later revisit these bits. 1485 */ 1486 bitno = extno & (EXTSPERSUM - 1); 1487 nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1; 1488 sword = extno >> L2EXTSPERSUM; 1489 1490 /* mask any prior bits for the starting words of the 1491 * summary map. 1492 */ 1493 mask = (bitno == 0) ? 0 : (ONES << (EXTSPERSUM - bitno)); 1494 inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask; 1495 extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask; 1496 1497 /* scan the free inode and free extent summary maps for 1498 * free resources. 1499 */ 1500 for (i = 0; i < nwords; i++) { 1501 /* check if this word of the free inode summary 1502 * map describes an extent with free inodes. 1503 */ 1504 if (~inosmap) { 1505 /* an extent with free inodes has been 1506 * found. determine the extent number 1507 * and the inode number within the extent. 1508 */ 1509 rem = diFindFree(inosmap, 0); 1510 extno = (sword << L2EXTSPERSUM) + rem; 1511 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 1512 0); 1513 if (rem >= INOSPEREXT) { 1514 IREAD_UNLOCK(ipimap); 1515 release_metapage(mp); 1516 AG_UNLOCK(imap, agno); 1517 jfs_error(ip->i_sb, 1518 "can't find free bit in wmap\n"); 1519 return -EIO; 1520 } 1521 1522 /* determine the inode number within the 1523 * iag and allocate the inode from the 1524 * map. 1525 */ 1526 ino = (extno << L2INOSPEREXT) + rem; 1527 rc = diAllocBit(imap, iagp, ino); 1528 IREAD_UNLOCK(ipimap); 1529 if (rc) 1530 assert(rc == -EIO); 1531 else { 1532 /* set the results of the allocation 1533 * and write the iag. 1534 */ 1535 diInitInode(ip, iagno, ino, extno, 1536 iagp); 1537 mark_metapage_dirty(mp); 1538 } 1539 release_metapage(mp); 1540 1541 /* free the AG lock and return. 1542 */ 1543 AG_UNLOCK(imap, agno); 1544 return (rc); 1545 1546 } 1547 1548 /* check if we may allocate an extent of free 1549 * inodes and whether this word of the free 1550 * extents summary map describes a free extent. 1551 */ 1552 if (addext && ~extsmap) { 1553 /* a free extent has been found. determine 1554 * the extent number. 1555 */ 1556 rem = diFindFree(extsmap, 0); 1557 extno = (sword << L2EXTSPERSUM) + rem; 1558 1559 /* allocate an extent of free inodes. 1560 */ 1561 if ((rc = diNewExt(imap, iagp, extno))) { 1562 /* if there is no disk space for a 1563 * new extent, try to allocate the 1564 * disk inode from somewhere else. 1565 */ 1566 if (rc == -ENOSPC) 1567 break; 1568 1569 assert(rc == -EIO); 1570 } else { 1571 /* set the results of the allocation 1572 * and write the iag. 1573 */ 1574 diInitInode(ip, iagno, 1575 extno << L2INOSPEREXT, 1576 extno, iagp); 1577 mark_metapage_dirty(mp); 1578 } 1579 release_metapage(mp); 1580 /* free the imap inode & the AG lock & return. 1581 */ 1582 IREAD_UNLOCK(ipimap); 1583 AG_UNLOCK(imap, agno); 1584 return (rc); 1585 } 1586 1587 /* move on to the next set of summary map words. 1588 */ 1589 sword = (sword == SMAPSZ - 1) ? 0 : sword + 1; 1590 inosmap = le32_to_cpu(iagp->inosmap[sword]); 1591 extsmap = le32_to_cpu(iagp->extsmap[sword]); 1592 } 1593 } 1594 /* unlock imap inode */ 1595 IREAD_UNLOCK(ipimap); 1596 1597 /* nothing doing in this iag, so release it. */ 1598 release_metapage(mp); 1599 1600 tryag: 1601 /* 1602 * try to allocate anywhere within the same AG as the parent inode. 1603 */ 1604 rc = diAllocAG(imap, agno, dir, ip); 1605 1606 AG_UNLOCK(imap, agno); 1607 1608 if (rc != -ENOSPC) 1609 return (rc); 1610 1611 /* 1612 * try to allocate in any AG. 1613 */ 1614 return (diAllocAny(imap, agno, dir, ip)); 1615 } 1616 1617 1618 /* 1619 * NAME: diAllocAG(imap,agno,dir,ip) 1620 * 1621 * FUNCTION: allocate a disk inode from the allocation group. 1622 * 1623 * this routine first determines if a new extent of free 1624 * inodes should be added for the allocation group, with 1625 * the current request satisfied from this extent. if this 1626 * is the case, an attempt will be made to do just that. if 1627 * this attempt fails or it has been determined that a new 1628 * extent should not be added, an attempt is made to satisfy 1629 * the request by allocating an existing (backed) free inode 1630 * from the allocation group. 1631 * 1632 * PRE CONDITION: Already have the AG lock for this AG. 1633 * 1634 * PARAMETERS: 1635 * imap - pointer to inode map control structure. 1636 * agno - allocation group to allocate from. 1637 * dir - 'true' if the new disk inode is for a directory. 1638 * ip - pointer to the new inode to be filled in on successful return 1639 * with the disk inode number allocated, its extent address 1640 * and the start of the ag. 1641 * 1642 * RETURN VALUES: 1643 * 0 - success. 1644 * -ENOSPC - insufficient disk resources. 1645 * -EIO - i/o error. 1646 */ 1647 static int 1648 diAllocAG(struct inomap * imap, int agno, bool dir, struct inode *ip) 1649 { 1650 int rc, addext, numfree, numinos; 1651 1652 /* get the number of free and the number of backed disk 1653 * inodes currently within the ag. 1654 */ 1655 numfree = imap->im_agctl[agno].numfree; 1656 numinos = imap->im_agctl[agno].numinos; 1657 1658 if (numfree > numinos) { 1659 jfs_error(ip->i_sb, "numfree > numinos\n"); 1660 return -EIO; 1661 } 1662 1663 /* determine if we should allocate a new extent of free inodes 1664 * within the ag: for directory inodes, add a new extent 1665 * if there are a small number of free inodes or number of free 1666 * inodes is a small percentage of the number of backed inodes. 1667 */ 1668 if (dir) 1669 addext = (numfree < 64 || 1670 (numfree < 256 1671 && ((numfree * 100) / numinos) <= 20)); 1672 else 1673 addext = (numfree == 0); 1674 1675 /* 1676 * try to allocate a new extent of free inodes. 1677 */ 1678 if (addext) { 1679 /* if free space is not available for this new extent, try 1680 * below to allocate a free and existing (already backed) 1681 * inode from the ag. 1682 */ 1683 if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC) 1684 return (rc); 1685 } 1686 1687 /* 1688 * try to allocate an existing free inode from the ag. 1689 */ 1690 return (diAllocIno(imap, agno, ip)); 1691 } 1692 1693 1694 /* 1695 * NAME: diAllocAny(imap,agno,dir,iap) 1696 * 1697 * FUNCTION: allocate a disk inode from any other allocation group. 1698 * 1699 * this routine is called when an allocation attempt within 1700 * the primary allocation group has failed. if attempts to 1701 * allocate an inode from any allocation group other than the 1702 * specified primary group. 1703 * 1704 * PARAMETERS: 1705 * imap - pointer to inode map control structure. 1706 * agno - primary allocation group (to avoid). 1707 * dir - 'true' if the new disk inode is for a directory. 1708 * ip - pointer to a new inode to be filled in on successful return 1709 * with the disk inode number allocated, its extent address 1710 * and the start of the ag. 1711 * 1712 * RETURN VALUES: 1713 * 0 - success. 1714 * -ENOSPC - insufficient disk resources. 1715 * -EIO - i/o error. 1716 */ 1717 static int 1718 diAllocAny(struct inomap * imap, int agno, bool dir, struct inode *ip) 1719 { 1720 int ag, rc; 1721 int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag; 1722 1723 1724 /* try to allocate from the ags following agno up to 1725 * the maximum ag number. 1726 */ 1727 for (ag = agno + 1; ag <= maxag; ag++) { 1728 AG_LOCK(imap, ag); 1729 1730 rc = diAllocAG(imap, ag, dir, ip); 1731 1732 AG_UNLOCK(imap, ag); 1733 1734 if (rc != -ENOSPC) 1735 return (rc); 1736 } 1737 1738 /* try to allocate from the ags in front of agno. 1739 */ 1740 for (ag = 0; ag < agno; ag++) { 1741 AG_LOCK(imap, ag); 1742 1743 rc = diAllocAG(imap, ag, dir, ip); 1744 1745 AG_UNLOCK(imap, ag); 1746 1747 if (rc != -ENOSPC) 1748 return (rc); 1749 } 1750 1751 /* no free disk inodes. 1752 */ 1753 return -ENOSPC; 1754 } 1755 1756 1757 /* 1758 * NAME: diAllocIno(imap,agno,ip) 1759 * 1760 * FUNCTION: allocate a disk inode from the allocation group's free 1761 * inode list, returning an error if this free list is 1762 * empty (i.e. no iags on the list). 1763 * 1764 * allocation occurs from the first iag on the list using 1765 * the iag's free inode summary map to find the leftmost 1766 * free inode in the iag. 1767 * 1768 * PRE CONDITION: Already have AG lock for this AG. 1769 * 1770 * PARAMETERS: 1771 * imap - pointer to inode map control structure. 1772 * agno - allocation group. 1773 * ip - pointer to new inode to be filled in on successful return 1774 * with the disk inode number allocated, its extent address 1775 * and the start of the ag. 1776 * 1777 * RETURN VALUES: 1778 * 0 - success. 1779 * -ENOSPC - insufficient disk resources. 1780 * -EIO - i/o error. 1781 */ 1782 static int diAllocIno(struct inomap * imap, int agno, struct inode *ip) 1783 { 1784 int iagno, ino, rc, rem, extno, sword; 1785 struct metapage *mp; 1786 struct iag *iagp; 1787 1788 /* check if there are iags on the ag's free inode list. 1789 */ 1790 if ((iagno = imap->im_agctl[agno].inofree) < 0) 1791 return -ENOSPC; 1792 1793 /* obtain read lock on imap inode */ 1794 IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP); 1795 1796 /* read the iag at the head of the list. 1797 */ 1798 if ((rc = diIAGRead(imap, iagno, &mp))) { 1799 IREAD_UNLOCK(imap->im_ipimap); 1800 return (rc); 1801 } 1802 iagp = (struct iag *) mp->data; 1803 1804 /* better be free inodes in this iag if it is on the 1805 * list. 1806 */ 1807 if (!iagp->nfreeinos) { 1808 IREAD_UNLOCK(imap->im_ipimap); 1809 release_metapage(mp); 1810 jfs_error(ip->i_sb, "nfreeinos = 0, but iag on freelist\n"); 1811 return -EIO; 1812 } 1813 1814 /* scan the free inode summary map to find an extent 1815 * with free inodes. 1816 */ 1817 for (sword = 0;; sword++) { 1818 if (sword >= SMAPSZ) { 1819 IREAD_UNLOCK(imap->im_ipimap); 1820 release_metapage(mp); 1821 jfs_error(ip->i_sb, 1822 "free inode not found in summary map\n"); 1823 return -EIO; 1824 } 1825 1826 if (~iagp->inosmap[sword]) 1827 break; 1828 } 1829 1830 /* found a extent with free inodes. determine 1831 * the extent number. 1832 */ 1833 rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0); 1834 if (rem >= EXTSPERSUM) { 1835 IREAD_UNLOCK(imap->im_ipimap); 1836 release_metapage(mp); 1837 jfs_error(ip->i_sb, "no free extent found\n"); 1838 return -EIO; 1839 } 1840 extno = (sword << L2EXTSPERSUM) + rem; 1841 1842 /* find the first free inode in the extent. 1843 */ 1844 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0); 1845 if (rem >= INOSPEREXT) { 1846 IREAD_UNLOCK(imap->im_ipimap); 1847 release_metapage(mp); 1848 jfs_error(ip->i_sb, "free inode not found\n"); 1849 return -EIO; 1850 } 1851 1852 /* compute the inode number within the iag. 1853 */ 1854 ino = (extno << L2INOSPEREXT) + rem; 1855 1856 /* allocate the inode. 1857 */ 1858 rc = diAllocBit(imap, iagp, ino); 1859 IREAD_UNLOCK(imap->im_ipimap); 1860 if (rc) { 1861 release_metapage(mp); 1862 return (rc); 1863 } 1864 1865 /* set the results of the allocation and write the iag. 1866 */ 1867 diInitInode(ip, iagno, ino, extno, iagp); 1868 write_metapage(mp); 1869 1870 return (0); 1871 } 1872 1873 1874 /* 1875 * NAME: diAllocExt(imap,agno,ip) 1876 * 1877 * FUNCTION: add a new extent of free inodes to an iag, allocating 1878 * an inode from this extent to satisfy the current allocation 1879 * request. 1880 * 1881 * this routine first tries to find an existing iag with free 1882 * extents through the ag free extent list. if list is not 1883 * empty, the head of the list will be selected as the home 1884 * of the new extent of free inodes. otherwise (the list is 1885 * empty), a new iag will be allocated for the ag to contain 1886 * the extent. 1887 * 1888 * once an iag has been selected, the free extent summary map 1889 * is used to locate a free extent within the iag and diNewExt() 1890 * is called to initialize the extent, with initialization 1891 * including the allocation of the first inode of the extent 1892 * for the purpose of satisfying this request. 1893 * 1894 * PARAMETERS: 1895 * imap - pointer to inode map control structure. 1896 * agno - allocation group number. 1897 * ip - pointer to new inode to be filled in on successful return 1898 * with the disk inode number allocated, its extent address 1899 * and the start of the ag. 1900 * 1901 * RETURN VALUES: 1902 * 0 - success. 1903 * -ENOSPC - insufficient disk resources. 1904 * -EIO - i/o error. 1905 */ 1906 static int diAllocExt(struct inomap * imap, int agno, struct inode *ip) 1907 { 1908 int rem, iagno, sword, extno, rc; 1909 struct metapage *mp; 1910 struct iag *iagp; 1911 1912 /* check if the ag has any iags with free extents. if not, 1913 * allocate a new iag for the ag. 1914 */ 1915 if ((iagno = imap->im_agctl[agno].extfree) < 0) { 1916 /* If successful, diNewIAG will obtain the read lock on the 1917 * imap inode. 1918 */ 1919 if ((rc = diNewIAG(imap, &iagno, agno, &mp))) { 1920 return (rc); 1921 } 1922 iagp = (struct iag *) mp->data; 1923 1924 /* set the ag number if this a brand new iag 1925 */ 1926 iagp->agstart = 1927 cpu_to_le64(AGTOBLK(agno, imap->im_ipimap)); 1928 } else { 1929 /* read the iag. 1930 */ 1931 IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP); 1932 if ((rc = diIAGRead(imap, iagno, &mp))) { 1933 IREAD_UNLOCK(imap->im_ipimap); 1934 jfs_error(ip->i_sb, "error reading iag\n"); 1935 return rc; 1936 } 1937 iagp = (struct iag *) mp->data; 1938 } 1939 1940 /* using the free extent summary map, find a free extent. 1941 */ 1942 for (sword = 0;; sword++) { 1943 if (sword >= SMAPSZ) { 1944 release_metapage(mp); 1945 IREAD_UNLOCK(imap->im_ipimap); 1946 jfs_error(ip->i_sb, "free ext summary map not found\n"); 1947 return -EIO; 1948 } 1949 if (~iagp->extsmap[sword]) 1950 break; 1951 } 1952 1953 /* determine the extent number of the free extent. 1954 */ 1955 rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0); 1956 if (rem >= EXTSPERSUM) { 1957 release_metapage(mp); 1958 IREAD_UNLOCK(imap->im_ipimap); 1959 jfs_error(ip->i_sb, "free extent not found\n"); 1960 return -EIO; 1961 } 1962 extno = (sword << L2EXTSPERSUM) + rem; 1963 1964 /* initialize the new extent. 1965 */ 1966 rc = diNewExt(imap, iagp, extno); 1967 IREAD_UNLOCK(imap->im_ipimap); 1968 if (rc) { 1969 /* something bad happened. if a new iag was allocated, 1970 * place it back on the inode map's iag free list, and 1971 * clear the ag number information. 1972 */ 1973 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { 1974 IAGFREE_LOCK(imap); 1975 iagp->iagfree = cpu_to_le32(imap->im_freeiag); 1976 imap->im_freeiag = iagno; 1977 IAGFREE_UNLOCK(imap); 1978 } 1979 write_metapage(mp); 1980 return (rc); 1981 } 1982 1983 /* set the results of the allocation and write the iag. 1984 */ 1985 diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp); 1986 1987 write_metapage(mp); 1988 1989 return (0); 1990 } 1991 1992 1993 /* 1994 * NAME: diAllocBit(imap,iagp,ino) 1995 * 1996 * FUNCTION: allocate a backed inode from an iag. 1997 * 1998 * this routine performs the mechanics of allocating a 1999 * specified inode from a backed extent. 2000 * 2001 * if the inode to be allocated represents the last free 2002 * inode within the iag, the iag will be removed from the 2003 * ag free inode list. 2004 * 2005 * a careful update approach is used to provide consistency 2006 * in the face of updates to multiple buffers. under this 2007 * approach, all required buffers are obtained before making 2008 * any updates and are held all are updates are complete. 2009 * 2010 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on 2011 * this AG. Must have read lock on imap inode. 2012 * 2013 * PARAMETERS: 2014 * imap - pointer to inode map control structure. 2015 * iagp - pointer to iag. 2016 * ino - inode number to be allocated within the iag. 2017 * 2018 * RETURN VALUES: 2019 * 0 - success. 2020 * -ENOSPC - insufficient disk resources. 2021 * -EIO - i/o error. 2022 */ 2023 static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino) 2024 { 2025 int extno, bitno, agno, sword, rc; 2026 struct metapage *amp = NULL, *bmp = NULL; 2027 struct iag *aiagp = NULL, *biagp = NULL; 2028 u32 mask; 2029 2030 /* check if this is the last free inode within the iag. 2031 * if so, it will have to be removed from the ag free 2032 * inode list, so get the iags preceding and following 2033 * it on the list. 2034 */ 2035 if (iagp->nfreeinos == cpu_to_le32(1)) { 2036 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) { 2037 if ((rc = 2038 diIAGRead(imap, le32_to_cpu(iagp->inofreefwd), 2039 &))) 2040 return (rc); 2041 aiagp = (struct iag *) amp->data; 2042 } 2043 2044 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) { 2045 if ((rc = 2046 diIAGRead(imap, 2047 le32_to_cpu(iagp->inofreeback), 2048 &bmp))) { 2049 if (amp) 2050 release_metapage(amp); 2051 return (rc); 2052 } 2053 biagp = (struct iag *) bmp->data; 2054 } 2055 } 2056 2057 /* get the ag number, extent number, inode number within 2058 * the extent. 2059 */ 2060 agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb)); 2061 extno = ino >> L2INOSPEREXT; 2062 bitno = ino & (INOSPEREXT - 1); 2063 2064 /* compute the mask for setting the map. 2065 */ 2066 mask = HIGHORDER >> bitno; 2067 2068 /* the inode should be free and backed. 2069 */ 2070 if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) || 2071 ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) || 2072 (addressPXD(&iagp->inoext[extno]) == 0)) { 2073 if (amp) 2074 release_metapage(amp); 2075 if (bmp) 2076 release_metapage(bmp); 2077 2078 jfs_error(imap->im_ipimap->i_sb, "iag inconsistent\n"); 2079 return -EIO; 2080 } 2081 2082 /* mark the inode as allocated in the working map. 2083 */ 2084 iagp->wmap[extno] |= cpu_to_le32(mask); 2085 2086 /* check if all inodes within the extent are now 2087 * allocated. if so, update the free inode summary 2088 * map to reflect this. 2089 */ 2090 if (iagp->wmap[extno] == cpu_to_le32(ONES)) { 2091 sword = extno >> L2EXTSPERSUM; 2092 bitno = extno & (EXTSPERSUM - 1); 2093 iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno); 2094 } 2095 2096 /* if this was the last free inode in the iag, remove the 2097 * iag from the ag free inode list. 2098 */ 2099 if (iagp->nfreeinos == cpu_to_le32(1)) { 2100 if (amp) { 2101 aiagp->inofreeback = iagp->inofreeback; 2102 write_metapage(amp); 2103 } 2104 2105 if (bmp) { 2106 biagp->inofreefwd = iagp->inofreefwd; 2107 write_metapage(bmp); 2108 } else { 2109 imap->im_agctl[agno].inofree = 2110 le32_to_cpu(iagp->inofreefwd); 2111 } 2112 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); 2113 } 2114 2115 /* update the free inode count at the iag, ag, inode 2116 * map levels. 2117 */ 2118 le32_add_cpu(&iagp->nfreeinos, -1); 2119 imap->im_agctl[agno].numfree -= 1; 2120 atomic_dec(&imap->im_numfree); 2121 2122 return (0); 2123 } 2124 2125 2126 /* 2127 * NAME: diNewExt(imap,iagp,extno) 2128 * 2129 * FUNCTION: initialize a new extent of inodes for an iag, allocating 2130 * the first inode of the extent for use for the current 2131 * allocation request. 2132 * 2133 * disk resources are allocated for the new extent of inodes 2134 * and the inodes themselves are initialized to reflect their 2135 * existence within the extent (i.e. their inode numbers and 2136 * inode extent addresses are set) and their initial state 2137 * (mode and link count are set to zero). 2138 * 2139 * if the iag is new, it is not yet on an ag extent free list 2140 * but will now be placed on this list. 2141 * 2142 * if the allocation of the new extent causes the iag to 2143 * have no free extent, the iag will be removed from the 2144 * ag extent free list. 2145 * 2146 * if the iag has no free backed inodes, it will be placed 2147 * on the ag free inode list, since the addition of the new 2148 * extent will now cause it to have free inodes. 2149 * 2150 * a careful update approach is used to provide consistency 2151 * (i.e. list consistency) in the face of updates to multiple 2152 * buffers. under this approach, all required buffers are 2153 * obtained before making any updates and are held until all 2154 * updates are complete. 2155 * 2156 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on 2157 * this AG. Must have read lock on imap inode. 2158 * 2159 * PARAMETERS: 2160 * imap - pointer to inode map control structure. 2161 * iagp - pointer to iag. 2162 * extno - extent number. 2163 * 2164 * RETURN VALUES: 2165 * 0 - success. 2166 * -ENOSPC - insufficient disk resources. 2167 * -EIO - i/o error. 2168 */ 2169 static int diNewExt(struct inomap * imap, struct iag * iagp, int extno) 2170 { 2171 int agno, iagno, fwd, back, freei = 0, sword, rc; 2172 struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL; 2173 struct metapage *amp, *bmp, *cmp, *dmp; 2174 struct inode *ipimap; 2175 s64 blkno, hint; 2176 int i, j; 2177 u32 mask; 2178 ino_t ino; 2179 struct dinode *dp; 2180 struct jfs_sb_info *sbi; 2181 2182 /* better have free extents. 2183 */ 2184 if (!iagp->nfreeexts) { 2185 jfs_error(imap->im_ipimap->i_sb, "no free extents\n"); 2186 return -EIO; 2187 } 2188 2189 /* get the inode map inode. 2190 */ 2191 ipimap = imap->im_ipimap; 2192 sbi = JFS_SBI(ipimap->i_sb); 2193 2194 amp = bmp = cmp = NULL; 2195 2196 /* get the ag and iag numbers for this iag. 2197 */ 2198 agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi); 2199 iagno = le32_to_cpu(iagp->iagnum); 2200 2201 /* check if this is the last free extent within the 2202 * iag. if so, the iag must be removed from the ag 2203 * free extent list, so get the iags preceding and 2204 * following the iag on this list. 2205 */ 2206 if (iagp->nfreeexts == cpu_to_le32(1)) { 2207 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) { 2208 if ((rc = diIAGRead(imap, fwd, &))) 2209 return (rc); 2210 aiagp = (struct iag *) amp->data; 2211 } 2212 2213 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) { 2214 if ((rc = diIAGRead(imap, back, &bmp))) 2215 goto error_out; 2216 biagp = (struct iag *) bmp->data; 2217 } 2218 } else { 2219 /* the iag has free extents. if all extents are free 2220 * (as is the case for a newly allocated iag), the iag 2221 * must be added to the ag free extent list, so get 2222 * the iag at the head of the list in preparation for 2223 * adding this iag to this list. 2224 */ 2225 fwd = back = -1; 2226 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { 2227 if ((fwd = imap->im_agctl[agno].extfree) >= 0) { 2228 if ((rc = diIAGRead(imap, fwd, &))) 2229 goto error_out; 2230 aiagp = (struct iag *) amp->data; 2231 } 2232 } 2233 } 2234 2235 /* check if the iag has no free inodes. if so, the iag 2236 * will have to be added to the ag free inode list, so get 2237 * the iag at the head of the list in preparation for 2238 * adding this iag to this list. in doing this, we must 2239 * check if we already have the iag at the head of 2240 * the list in hand. 2241 */ 2242 if (iagp->nfreeinos == 0) { 2243 freei = imap->im_agctl[agno].inofree; 2244 2245 if (freei >= 0) { 2246 if (freei == fwd) { 2247 ciagp = aiagp; 2248 } else if (freei == back) { 2249 ciagp = biagp; 2250 } else { 2251 if ((rc = diIAGRead(imap, freei, &cmp))) 2252 goto error_out; 2253 ciagp = (struct iag *) cmp->data; 2254 } 2255 if (ciagp == NULL) { 2256 jfs_error(imap->im_ipimap->i_sb, 2257 "ciagp == NULL\n"); 2258 rc = -EIO; 2259 goto error_out; 2260 } 2261 } 2262 } 2263 2264 /* allocate disk space for the inode extent. 2265 */ 2266 if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0)) 2267 hint = ((s64) agno << sbi->bmap->db_agl2size) - 1; 2268 else 2269 hint = addressPXD(&iagp->inoext[extno - 1]) + 2270 lengthPXD(&iagp->inoext[extno - 1]) - 1; 2271 2272 if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno))) 2273 goto error_out; 2274 2275 /* compute the inode number of the first inode within the 2276 * extent. 2277 */ 2278 ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT); 2279 2280 /* initialize the inodes within the newly allocated extent a 2281 * page at a time. 2282 */ 2283 for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) { 2284 /* get a buffer for this page of disk inodes. 2285 */ 2286 dmp = get_metapage(ipimap, blkno + i, PSIZE, 1); 2287 if (dmp == NULL) { 2288 rc = -EIO; 2289 goto error_out; 2290 } 2291 dp = (struct dinode *) dmp->data; 2292 2293 /* initialize the inode number, mode, link count and 2294 * inode extent address. 2295 */ 2296 for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) { 2297 dp->di_inostamp = cpu_to_le32(sbi->inostamp); 2298 dp->di_number = cpu_to_le32(ino); 2299 dp->di_fileset = cpu_to_le32(FILESYSTEM_I); 2300 dp->di_mode = 0; 2301 dp->di_nlink = 0; 2302 PXDaddress(&(dp->di_ixpxd), blkno); 2303 PXDlength(&(dp->di_ixpxd), imap->im_nbperiext); 2304 } 2305 write_metapage(dmp); 2306 } 2307 2308 /* if this is the last free extent within the iag, remove the 2309 * iag from the ag free extent list. 2310 */ 2311 if (iagp->nfreeexts == cpu_to_le32(1)) { 2312 if (fwd >= 0) 2313 aiagp->extfreeback = iagp->extfreeback; 2314 2315 if (back >= 0) 2316 biagp->extfreefwd = iagp->extfreefwd; 2317 else 2318 imap->im_agctl[agno].extfree = 2319 le32_to_cpu(iagp->extfreefwd); 2320 2321 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); 2322 } else { 2323 /* if the iag has all free extents (newly allocated iag), 2324 * add the iag to the ag free extent list. 2325 */ 2326 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { 2327 if (fwd >= 0) 2328 aiagp->extfreeback = cpu_to_le32(iagno); 2329 2330 iagp->extfreefwd = cpu_to_le32(fwd); 2331 iagp->extfreeback = cpu_to_le32(-1); 2332 imap->im_agctl[agno].extfree = iagno; 2333 } 2334 } 2335 2336 /* if the iag has no free inodes, add the iag to the 2337 * ag free inode list. 2338 */ 2339 if (iagp->nfreeinos == 0) { 2340 if (freei >= 0) 2341 ciagp->inofreeback = cpu_to_le32(iagno); 2342 2343 iagp->inofreefwd = 2344 cpu_to_le32(imap->im_agctl[agno].inofree); 2345 iagp->inofreeback = cpu_to_le32(-1); 2346 imap->im_agctl[agno].inofree = iagno; 2347 } 2348 2349 /* initialize the extent descriptor of the extent. */ 2350 PXDlength(&iagp->inoext[extno], imap->im_nbperiext); 2351 PXDaddress(&iagp->inoext[extno], blkno); 2352 2353 /* initialize the working and persistent map of the extent. 2354 * the working map will be initialized such that 2355 * it indicates the first inode of the extent is allocated. 2356 */ 2357 iagp->wmap[extno] = cpu_to_le32(HIGHORDER); 2358 iagp->pmap[extno] = 0; 2359 2360 /* update the free inode and free extent summary maps 2361 * for the extent to indicate the extent has free inodes 2362 * and no longer represents a free extent. 2363 */ 2364 sword = extno >> L2EXTSPERSUM; 2365 mask = HIGHORDER >> (extno & (EXTSPERSUM - 1)); 2366 iagp->extsmap[sword] |= cpu_to_le32(mask); 2367 iagp->inosmap[sword] &= cpu_to_le32(~mask); 2368 2369 /* update the free inode and free extent counts for the 2370 * iag. 2371 */ 2372 le32_add_cpu(&iagp->nfreeinos, (INOSPEREXT - 1)); 2373 le32_add_cpu(&iagp->nfreeexts, -1); 2374 2375 /* update the free and backed inode counts for the ag. 2376 */ 2377 imap->im_agctl[agno].numfree += (INOSPEREXT - 1); 2378 imap->im_agctl[agno].numinos += INOSPEREXT; 2379 2380 /* update the free and backed inode counts for the inode map. 2381 */ 2382 atomic_add(INOSPEREXT - 1, &imap->im_numfree); 2383 atomic_add(INOSPEREXT, &imap->im_numinos); 2384 2385 /* write the iags. 2386 */ 2387 if (amp) 2388 write_metapage(amp); 2389 if (bmp) 2390 write_metapage(bmp); 2391 if (cmp) 2392 write_metapage(cmp); 2393 2394 return (0); 2395 2396 error_out: 2397 2398 /* release the iags. 2399 */ 2400 if (amp) 2401 release_metapage(amp); 2402 if (bmp) 2403 release_metapage(bmp); 2404 if (cmp) 2405 release_metapage(cmp); 2406 2407 return (rc); 2408 } 2409 2410 2411 /* 2412 * NAME: diNewIAG(imap,iagnop,agno) 2413 * 2414 * FUNCTION: allocate a new iag for an allocation group. 2415 * 2416 * first tries to allocate the iag from the inode map 2417 * iagfree list: 2418 * if the list has free iags, the head of the list is removed 2419 * and returned to satisfy the request. 2420 * if the inode map's iag free list is empty, the inode map 2421 * is extended to hold a new iag. this new iag is initialized 2422 * and returned to satisfy the request. 2423 * 2424 * PARAMETERS: 2425 * imap - pointer to inode map control structure. 2426 * iagnop - pointer to an iag number set with the number of the 2427 * newly allocated iag upon successful return. 2428 * agno - allocation group number. 2429 * bpp - Buffer pointer to be filled in with new IAG's buffer 2430 * 2431 * RETURN VALUES: 2432 * 0 - success. 2433 * -ENOSPC - insufficient disk resources. 2434 * -EIO - i/o error. 2435 * 2436 * serialization: 2437 * AG lock held on entry/exit; 2438 * write lock on the map is held inside; 2439 * read lock on the map is held on successful completion; 2440 * 2441 * note: new iag transaction: 2442 * . synchronously write iag; 2443 * . write log of xtree and inode of imap; 2444 * . commit; 2445 * . synchronous write of xtree (right to left, bottom to top); 2446 * . at start of logredo(): init in-memory imap with one additional iag page; 2447 * . at end of logredo(): re-read imap inode to determine 2448 * new imap size; 2449 */ 2450 static int 2451 diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp) 2452 { 2453 int rc; 2454 int iagno, i, xlen; 2455 struct inode *ipimap; 2456 struct super_block *sb; 2457 struct jfs_sb_info *sbi; 2458 struct metapage *mp; 2459 struct iag *iagp; 2460 s64 xaddr = 0; 2461 s64 blkno; 2462 tid_t tid; 2463 struct inode *iplist[1]; 2464 2465 /* pick up pointers to the inode map and mount inodes */ 2466 ipimap = imap->im_ipimap; 2467 sb = ipimap->i_sb; 2468 sbi = JFS_SBI(sb); 2469 2470 /* acquire the free iag lock */ 2471 IAGFREE_LOCK(imap); 2472 2473 /* if there are any iags on the inode map free iag list, 2474 * allocate the iag from the head of the list. 2475 */ 2476 if (imap->im_freeiag >= 0) { 2477 /* pick up the iag number at the head of the list */ 2478 iagno = imap->im_freeiag; 2479 2480 /* determine the logical block number of the iag */ 2481 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage); 2482 } else { 2483 /* no free iags. the inode map will have to be extented 2484 * to include a new iag. 2485 */ 2486 2487 /* acquire inode map lock */ 2488 IWRITE_LOCK(ipimap, RDWRLOCK_IMAP); 2489 2490 if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) { 2491 IWRITE_UNLOCK(ipimap); 2492 IAGFREE_UNLOCK(imap); 2493 jfs_error(imap->im_ipimap->i_sb, 2494 "ipimap->i_size is wrong\n"); 2495 return -EIO; 2496 } 2497 2498 2499 /* get the next available iag number */ 2500 iagno = imap->im_nextiag; 2501 2502 /* make sure that we have not exceeded the maximum inode 2503 * number limit. 2504 */ 2505 if (iagno > (MAXIAGS - 1)) { 2506 /* release the inode map lock */ 2507 IWRITE_UNLOCK(ipimap); 2508 2509 rc = -ENOSPC; 2510 goto out; 2511 } 2512 2513 /* 2514 * synchronously append new iag page. 2515 */ 2516 /* determine the logical address of iag page to append */ 2517 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage); 2518 2519 /* Allocate extent for new iag page */ 2520 xlen = sbi->nbperpage; 2521 if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) { 2522 /* release the inode map lock */ 2523 IWRITE_UNLOCK(ipimap); 2524 2525 goto out; 2526 } 2527 2528 /* 2529 * start transaction of update of the inode map 2530 * addressing structure pointing to the new iag page; 2531 */ 2532 tid = txBegin(sb, COMMIT_FORCE); 2533 mutex_lock(&JFS_IP(ipimap)->commit_mutex); 2534 2535 /* update the inode map addressing structure to point to it */ 2536 if ((rc = 2537 xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) { 2538 txEnd(tid); 2539 mutex_unlock(&JFS_IP(ipimap)->commit_mutex); 2540 /* Free the blocks allocated for the iag since it was 2541 * not successfully added to the inode map 2542 */ 2543 dbFree(ipimap, xaddr, (s64) xlen); 2544 2545 /* release the inode map lock */ 2546 IWRITE_UNLOCK(ipimap); 2547 2548 goto out; 2549 } 2550 2551 /* update the inode map's inode to reflect the extension */ 2552 ipimap->i_size += PSIZE; 2553 inode_add_bytes(ipimap, PSIZE); 2554 2555 /* assign a buffer for the page */ 2556 mp = get_metapage(ipimap, blkno, PSIZE, 0); 2557 if (!mp) { 2558 /* 2559 * This is very unlikely since we just created the 2560 * extent, but let's try to handle it correctly 2561 */ 2562 xtTruncate(tid, ipimap, ipimap->i_size - PSIZE, 2563 COMMIT_PWMAP); 2564 2565 txAbort(tid, 0); 2566 txEnd(tid); 2567 mutex_unlock(&JFS_IP(ipimap)->commit_mutex); 2568 2569 /* release the inode map lock */ 2570 IWRITE_UNLOCK(ipimap); 2571 2572 rc = -EIO; 2573 goto out; 2574 } 2575 iagp = (struct iag *) mp->data; 2576 2577 /* init the iag */ 2578 memset(iagp, 0, sizeof(struct iag)); 2579 iagp->iagnum = cpu_to_le32(iagno); 2580 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); 2581 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); 2582 iagp->iagfree = cpu_to_le32(-1); 2583 iagp->nfreeinos = 0; 2584 iagp->nfreeexts = cpu_to_le32(EXTSPERIAG); 2585 2586 /* initialize the free inode summary map (free extent 2587 * summary map initialization handled by bzero). 2588 */ 2589 for (i = 0; i < SMAPSZ; i++) 2590 iagp->inosmap[i] = cpu_to_le32(ONES); 2591 2592 /* 2593 * Write and sync the metapage 2594 */ 2595 flush_metapage(mp); 2596 2597 /* 2598 * txCommit(COMMIT_FORCE) will synchronously write address 2599 * index pages and inode after commit in careful update order 2600 * of address index pages (right to left, bottom up); 2601 */ 2602 iplist[0] = ipimap; 2603 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE); 2604 2605 txEnd(tid); 2606 mutex_unlock(&JFS_IP(ipimap)->commit_mutex); 2607 2608 duplicateIXtree(sb, blkno, xlen, &xaddr); 2609 2610 /* update the next available iag number */ 2611 imap->im_nextiag += 1; 2612 2613 /* Add the iag to the iag free list so we don't lose the iag 2614 * if a failure happens now. 2615 */ 2616 imap->im_freeiag = iagno; 2617 2618 /* Until we have logredo working, we want the imap inode & 2619 * control page to be up to date. 2620 */ 2621 diSync(ipimap); 2622 2623 /* release the inode map lock */ 2624 IWRITE_UNLOCK(ipimap); 2625 } 2626 2627 /* obtain read lock on map */ 2628 IREAD_LOCK(ipimap, RDWRLOCK_IMAP); 2629 2630 /* read the iag */ 2631 if ((rc = diIAGRead(imap, iagno, &mp))) { 2632 IREAD_UNLOCK(ipimap); 2633 rc = -EIO; 2634 goto out; 2635 } 2636 iagp = (struct iag *) mp->data; 2637 2638 /* remove the iag from the iag free list */ 2639 imap->im_freeiag = le32_to_cpu(iagp->iagfree); 2640 iagp->iagfree = cpu_to_le32(-1); 2641 2642 /* set the return iag number and buffer pointer */ 2643 *iagnop = iagno; 2644 *mpp = mp; 2645 2646 out: 2647 /* release the iag free lock */ 2648 IAGFREE_UNLOCK(imap); 2649 2650 return (rc); 2651 } 2652 2653 /* 2654 * NAME: diIAGRead() 2655 * 2656 * FUNCTION: get the buffer for the specified iag within a fileset 2657 * or aggregate inode map. 2658 * 2659 * PARAMETERS: 2660 * imap - pointer to inode map control structure. 2661 * iagno - iag number. 2662 * bpp - point to buffer pointer to be filled in on successful 2663 * exit. 2664 * 2665 * SERIALIZATION: 2666 * must have read lock on imap inode 2667 * (When called by diExtendFS, the filesystem is quiesced, therefore 2668 * the read lock is unnecessary.) 2669 * 2670 * RETURN VALUES: 2671 * 0 - success. 2672 * -EIO - i/o error. 2673 */ 2674 static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp) 2675 { 2676 struct inode *ipimap = imap->im_ipimap; 2677 s64 blkno; 2678 2679 /* compute the logical block number of the iag. */ 2680 blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage); 2681 2682 /* read the iag. */ 2683 *mpp = read_metapage(ipimap, blkno, PSIZE, 0); 2684 if (*mpp == NULL) { 2685 return -EIO; 2686 } 2687 2688 return (0); 2689 } 2690 2691 /* 2692 * NAME: diFindFree() 2693 * 2694 * FUNCTION: find the first free bit in a word starting at 2695 * the specified bit position. 2696 * 2697 * PARAMETERS: 2698 * word - word to be examined. 2699 * start - starting bit position. 2700 * 2701 * RETURN VALUES: 2702 * bit position of first free bit in the word or 32 if 2703 * no free bits were found. 2704 */ 2705 static int diFindFree(u32 word, int start) 2706 { 2707 int bitno; 2708 assert(start < 32); 2709 /* scan the word for the first free bit. */ 2710 for (word <<= start, bitno = start; bitno < 32; 2711 bitno++, word <<= 1) { 2712 if ((word & HIGHORDER) == 0) 2713 break; 2714 } 2715 return (bitno); 2716 } 2717 2718 /* 2719 * NAME: diUpdatePMap() 2720 * 2721 * FUNCTION: Update the persistent map in an IAG for the allocation or 2722 * freeing of the specified inode. 2723 * 2724 * PRE CONDITIONS: Working map has already been updated for allocate. 2725 * 2726 * PARAMETERS: 2727 * ipimap - Incore inode map inode 2728 * inum - Number of inode to mark in permanent map 2729 * is_free - If 'true' indicates inode should be marked freed, otherwise 2730 * indicates inode should be marked allocated. 2731 * 2732 * RETURN VALUES: 2733 * 0 for success 2734 */ 2735 int 2736 diUpdatePMap(struct inode *ipimap, 2737 unsigned long inum, bool is_free, struct tblock * tblk) 2738 { 2739 int rc; 2740 struct iag *iagp; 2741 struct metapage *mp; 2742 int iagno, ino, extno, bitno; 2743 struct inomap *imap; 2744 u32 mask; 2745 struct jfs_log *log; 2746 int lsn, difft, diffp; 2747 unsigned long flags; 2748 2749 imap = JFS_IP(ipimap)->i_imap; 2750 /* get the iag number containing the inode */ 2751 iagno = INOTOIAG(inum); 2752 /* make sure that the iag is contained within the map */ 2753 if (iagno >= imap->im_nextiag) { 2754 jfs_error(ipimap->i_sb, "the iag is outside the map\n"); 2755 return -EIO; 2756 } 2757 /* read the iag */ 2758 IREAD_LOCK(ipimap, RDWRLOCK_IMAP); 2759 rc = diIAGRead(imap, iagno, &mp); 2760 IREAD_UNLOCK(ipimap); 2761 if (rc) 2762 return (rc); 2763 metapage_wait_for_io(mp); 2764 iagp = (struct iag *) mp->data; 2765 /* get the inode number and extent number of the inode within 2766 * the iag and the inode number within the extent. 2767 */ 2768 ino = inum & (INOSPERIAG - 1); 2769 extno = ino >> L2INOSPEREXT; 2770 bitno = ino & (INOSPEREXT - 1); 2771 mask = HIGHORDER >> bitno; 2772 /* 2773 * mark the inode free in persistent map: 2774 */ 2775 if (is_free) { 2776 /* The inode should have been allocated both in working 2777 * map and in persistent map; 2778 * the inode will be freed from working map at the release 2779 * of last reference release; 2780 */ 2781 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { 2782 jfs_error(ipimap->i_sb, 2783 "inode %ld not marked as allocated in wmap!\n", 2784 inum); 2785 } 2786 if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) { 2787 jfs_error(ipimap->i_sb, 2788 "inode %ld not marked as allocated in pmap!\n", 2789 inum); 2790 } 2791 /* update the bitmap for the extent of the freed inode */ 2792 iagp->pmap[extno] &= cpu_to_le32(~mask); 2793 } 2794 /* 2795 * mark the inode allocated in persistent map: 2796 */ 2797 else { 2798 /* The inode should be already allocated in the working map 2799 * and should be free in persistent map; 2800 */ 2801 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { 2802 release_metapage(mp); 2803 jfs_error(ipimap->i_sb, 2804 "the inode is not allocated in the working map\n"); 2805 return -EIO; 2806 } 2807 if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) { 2808 release_metapage(mp); 2809 jfs_error(ipimap->i_sb, 2810 "the inode is not free in the persistent map\n"); 2811 return -EIO; 2812 } 2813 /* update the bitmap for the extent of the allocated inode */ 2814 iagp->pmap[extno] |= cpu_to_le32(mask); 2815 } 2816 /* 2817 * update iag lsn 2818 */ 2819 lsn = tblk->lsn; 2820 log = JFS_SBI(tblk->sb)->log; 2821 LOGSYNC_LOCK(log, flags); 2822 if (mp->lsn != 0) { 2823 /* inherit older/smaller lsn */ 2824 logdiff(difft, lsn, log); 2825 logdiff(diffp, mp->lsn, log); 2826 if (difft < diffp) { 2827 mp->lsn = lsn; 2828 /* move mp after tblock in logsync list */ 2829 list_move(&mp->synclist, &tblk->synclist); 2830 } 2831 /* inherit younger/larger clsn */ 2832 assert(mp->clsn); 2833 logdiff(difft, tblk->clsn, log); 2834 logdiff(diffp, mp->clsn, log); 2835 if (difft > diffp) 2836 mp->clsn = tblk->clsn; 2837 } else { 2838 mp->log = log; 2839 mp->lsn = lsn; 2840 /* insert mp after tblock in logsync list */ 2841 log->count++; 2842 list_add(&mp->synclist, &tblk->synclist); 2843 mp->clsn = tblk->clsn; 2844 } 2845 LOGSYNC_UNLOCK(log, flags); 2846 write_metapage(mp); 2847 return (0); 2848 } 2849 2850 /* 2851 * diExtendFS() 2852 * 2853 * function: update imap for extendfs(); 2854 * 2855 * note: AG size has been increased s.t. each k old contiguous AGs are 2856 * coalesced into a new AG; 2857 */ 2858 int diExtendFS(struct inode *ipimap, struct inode *ipbmap) 2859 { 2860 int rc, rcx = 0; 2861 struct inomap *imap = JFS_IP(ipimap)->i_imap; 2862 struct iag *iagp = NULL, *hiagp = NULL; 2863 struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap; 2864 struct metapage *bp, *hbp; 2865 int i, n, head; 2866 int numinos, xnuminos = 0, xnumfree = 0; 2867 s64 agstart; 2868 2869 jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d", 2870 imap->im_nextiag, atomic_read(&imap->im_numinos), 2871 atomic_read(&imap->im_numfree)); 2872 2873 /* 2874 * reconstruct imap 2875 * 2876 * coalesce contiguous k (newAGSize/oldAGSize) AGs; 2877 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn; 2878 * note: new AG size = old AG size * (2**x). 2879 */ 2880 2881 /* init per AG control information im_agctl[] */ 2882 for (i = 0; i < MAXAG; i++) { 2883 imap->im_agctl[i].inofree = -1; 2884 imap->im_agctl[i].extfree = -1; 2885 imap->im_agctl[i].numinos = 0; /* number of backed inodes */ 2886 imap->im_agctl[i].numfree = 0; /* number of free backed inodes */ 2887 } 2888 2889 /* 2890 * process each iag page of the map. 2891 * 2892 * rebuild AG Free Inode List, AG Free Inode Extent List; 2893 */ 2894 for (i = 0; i < imap->im_nextiag; i++) { 2895 if ((rc = diIAGRead(imap, i, &bp))) { 2896 rcx = rc; 2897 continue; 2898 } 2899 iagp = (struct iag *) bp->data; 2900 if (le32_to_cpu(iagp->iagnum) != i) { 2901 release_metapage(bp); 2902 jfs_error(ipimap->i_sb, "unexpected value of iagnum\n"); 2903 return -EIO; 2904 } 2905 2906 /* leave free iag in the free iag list */ 2907 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { 2908 release_metapage(bp); 2909 continue; 2910 } 2911 2912 agstart = le64_to_cpu(iagp->agstart); 2913 n = agstart >> mp->db_agl2size; 2914 iagp->agstart = cpu_to_le64((s64)n << mp->db_agl2size); 2915 2916 /* compute backed inodes */ 2917 numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts)) 2918 << L2INOSPEREXT; 2919 if (numinos > 0) { 2920 /* merge AG backed inodes */ 2921 imap->im_agctl[n].numinos += numinos; 2922 xnuminos += numinos; 2923 } 2924 2925 /* if any backed free inodes, insert at AG free inode list */ 2926 if ((int) le32_to_cpu(iagp->nfreeinos) > 0) { 2927 if ((head = imap->im_agctl[n].inofree) == -1) { 2928 iagp->inofreefwd = cpu_to_le32(-1); 2929 iagp->inofreeback = cpu_to_le32(-1); 2930 } else { 2931 if ((rc = diIAGRead(imap, head, &hbp))) { 2932 rcx = rc; 2933 goto nextiag; 2934 } 2935 hiagp = (struct iag *) hbp->data; 2936 hiagp->inofreeback = iagp->iagnum; 2937 iagp->inofreefwd = cpu_to_le32(head); 2938 iagp->inofreeback = cpu_to_le32(-1); 2939 write_metapage(hbp); 2940 } 2941 2942 imap->im_agctl[n].inofree = 2943 le32_to_cpu(iagp->iagnum); 2944 2945 /* merge AG backed free inodes */ 2946 imap->im_agctl[n].numfree += 2947 le32_to_cpu(iagp->nfreeinos); 2948 xnumfree += le32_to_cpu(iagp->nfreeinos); 2949 } 2950 2951 /* if any free extents, insert at AG free extent list */ 2952 if (le32_to_cpu(iagp->nfreeexts) > 0) { 2953 if ((head = imap->im_agctl[n].extfree) == -1) { 2954 iagp->extfreefwd = cpu_to_le32(-1); 2955 iagp->extfreeback = cpu_to_le32(-1); 2956 } else { 2957 if ((rc = diIAGRead(imap, head, &hbp))) { 2958 rcx = rc; 2959 goto nextiag; 2960 } 2961 hiagp = (struct iag *) hbp->data; 2962 hiagp->extfreeback = iagp->iagnum; 2963 iagp->extfreefwd = cpu_to_le32(head); 2964 iagp->extfreeback = cpu_to_le32(-1); 2965 write_metapage(hbp); 2966 } 2967 2968 imap->im_agctl[n].extfree = 2969 le32_to_cpu(iagp->iagnum); 2970 } 2971 2972 nextiag: 2973 write_metapage(bp); 2974 } 2975 2976 if (xnuminos != atomic_read(&imap->im_numinos) || 2977 xnumfree != atomic_read(&imap->im_numfree)) { 2978 jfs_error(ipimap->i_sb, "numinos or numfree incorrect\n"); 2979 return -EIO; 2980 } 2981 2982 return rcx; 2983 } 2984 2985 2986 /* 2987 * duplicateIXtree() 2988 * 2989 * serialization: IWRITE_LOCK held on entry/exit 2990 * 2991 * note: shadow page with regular inode (rel.2); 2992 */ 2993 static void duplicateIXtree(struct super_block *sb, s64 blkno, 2994 int xlen, s64 *xaddr) 2995 { 2996 struct jfs_superblock *j_sb; 2997 struct buffer_head *bh; 2998 struct inode *ip; 2999 tid_t tid; 3000 3001 /* if AIT2 ipmap2 is bad, do not try to update it */ 3002 if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT) /* s_flag */ 3003 return; 3004 ip = diReadSpecial(sb, FILESYSTEM_I, 1); 3005 if (ip == NULL) { 3006 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT; 3007 if (readSuper(sb, &bh)) 3008 return; 3009 j_sb = (struct jfs_superblock *)bh->b_data; 3010 j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT); 3011 3012 mark_buffer_dirty(bh); 3013 sync_dirty_buffer(bh); 3014 brelse(bh); 3015 return; 3016 } 3017 3018 /* start transaction */ 3019 tid = txBegin(sb, COMMIT_FORCE); 3020 /* update the inode map addressing structure to point to it */ 3021 if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) { 3022 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT; 3023 txAbort(tid, 1); 3024 goto cleanup; 3025 3026 } 3027 /* update the inode map's inode to reflect the extension */ 3028 ip->i_size += PSIZE; 3029 inode_add_bytes(ip, PSIZE); 3030 txCommit(tid, 1, &ip, COMMIT_FORCE); 3031 cleanup: 3032 txEnd(tid); 3033 diFreeSpecial(ip); 3034 } 3035 3036 /* 3037 * NAME: copy_from_dinode() 3038 * 3039 * FUNCTION: Copies inode info from disk inode to in-memory inode 3040 * 3041 * RETURN VALUES: 3042 * 0 - success 3043 * -ENOMEM - insufficient memory 3044 */ 3045 static int copy_from_dinode(struct dinode * dip, struct inode *ip) 3046 { 3047 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 3048 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 3049 3050 jfs_ip->fileset = le32_to_cpu(dip->di_fileset); 3051 jfs_ip->mode2 = le32_to_cpu(dip->di_mode); 3052 jfs_set_inode_flags(ip); 3053 3054 ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff; 3055 if (sbi->umask != -1) { 3056 ip->i_mode = (ip->i_mode & ~0777) | (0777 & ~sbi->umask); 3057 /* For directories, add x permission if r is allowed by umask */ 3058 if (S_ISDIR(ip->i_mode)) { 3059 if (ip->i_mode & 0400) 3060 ip->i_mode |= 0100; 3061 if (ip->i_mode & 0040) 3062 ip->i_mode |= 0010; 3063 if (ip->i_mode & 0004) 3064 ip->i_mode |= 0001; 3065 } 3066 } 3067 set_nlink(ip, le32_to_cpu(dip->di_nlink)); 3068 3069 jfs_ip->saved_uid = make_kuid(&init_user_ns, le32_to_cpu(dip->di_uid)); 3070 if (!uid_valid(sbi->uid)) 3071 ip->i_uid = jfs_ip->saved_uid; 3072 else { 3073 ip->i_uid = sbi->uid; 3074 } 3075 3076 jfs_ip->saved_gid = make_kgid(&init_user_ns, le32_to_cpu(dip->di_gid)); 3077 if (!gid_valid(sbi->gid)) 3078 ip->i_gid = jfs_ip->saved_gid; 3079 else { 3080 ip->i_gid = sbi->gid; 3081 } 3082 3083 ip->i_size = le64_to_cpu(dip->di_size); 3084 ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec); 3085 ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec); 3086 ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec); 3087 ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec); 3088 ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec); 3089 ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec); 3090 ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks)); 3091 ip->i_generation = le32_to_cpu(dip->di_gen); 3092 3093 jfs_ip->ixpxd = dip->di_ixpxd; /* in-memory pxd's are little-endian */ 3094 jfs_ip->acl = dip->di_acl; /* as are dxd's */ 3095 jfs_ip->ea = dip->di_ea; 3096 jfs_ip->next_index = le32_to_cpu(dip->di_next_index); 3097 jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec); 3098 jfs_ip->acltype = le32_to_cpu(dip->di_acltype); 3099 3100 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) { 3101 jfs_ip->dev = le32_to_cpu(dip->di_rdev); 3102 ip->i_rdev = new_decode_dev(jfs_ip->dev); 3103 } 3104 3105 if (S_ISDIR(ip->i_mode)) { 3106 memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384); 3107 } else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) { 3108 memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288); 3109 } else 3110 memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128); 3111 3112 /* Zero the in-memory-only stuff */ 3113 jfs_ip->cflag = 0; 3114 jfs_ip->btindex = 0; 3115 jfs_ip->btorder = 0; 3116 jfs_ip->bxflag = 0; 3117 jfs_ip->blid = 0; 3118 jfs_ip->atlhead = 0; 3119 jfs_ip->atltail = 0; 3120 jfs_ip->xtlid = 0; 3121 return (0); 3122 } 3123 3124 /* 3125 * NAME: copy_to_dinode() 3126 * 3127 * FUNCTION: Copies inode info from in-memory inode to disk inode 3128 */ 3129 static void copy_to_dinode(struct dinode * dip, struct inode *ip) 3130 { 3131 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 3132 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); 3133 3134 dip->di_fileset = cpu_to_le32(jfs_ip->fileset); 3135 dip->di_inostamp = cpu_to_le32(sbi->inostamp); 3136 dip->di_number = cpu_to_le32(ip->i_ino); 3137 dip->di_gen = cpu_to_le32(ip->i_generation); 3138 dip->di_size = cpu_to_le64(ip->i_size); 3139 dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks)); 3140 dip->di_nlink = cpu_to_le32(ip->i_nlink); 3141 if (!uid_valid(sbi->uid)) 3142 dip->di_uid = cpu_to_le32(i_uid_read(ip)); 3143 else 3144 dip->di_uid =cpu_to_le32(from_kuid(&init_user_ns, 3145 jfs_ip->saved_uid)); 3146 if (!gid_valid(sbi->gid)) 3147 dip->di_gid = cpu_to_le32(i_gid_read(ip)); 3148 else 3149 dip->di_gid = cpu_to_le32(from_kgid(&init_user_ns, 3150 jfs_ip->saved_gid)); 3151 jfs_get_inode_flags(jfs_ip); 3152 /* 3153 * mode2 is only needed for storing the higher order bits. 3154 * Trust i_mode for the lower order ones 3155 */ 3156 if (sbi->umask == -1) 3157 dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) | 3158 ip->i_mode); 3159 else /* Leave the original permissions alone */ 3160 dip->di_mode = cpu_to_le32(jfs_ip->mode2); 3161 3162 dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec); 3163 dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec); 3164 dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec); 3165 dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec); 3166 dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec); 3167 dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec); 3168 dip->di_ixpxd = jfs_ip->ixpxd; /* in-memory pxd's are little-endian */ 3169 dip->di_acl = jfs_ip->acl; /* as are dxd's */ 3170 dip->di_ea = jfs_ip->ea; 3171 dip->di_next_index = cpu_to_le32(jfs_ip->next_index); 3172 dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime); 3173 dip->di_otime.tv_nsec = 0; 3174 dip->di_acltype = cpu_to_le32(jfs_ip->acltype); 3175 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) 3176 dip->di_rdev = cpu_to_le32(jfs_ip->dev); 3177 } 3178