1 /* lfs_inode.c 6.2 84/02/07 */ 2 3 #include "../h/param.h" 4 #include "../h/systm.h" 5 #include "../h/mount.h" 6 #include "../h/dir.h" 7 #include "../h/user.h" 8 #include "../h/inode.h" 9 #include "../h/fs.h" 10 #include "../h/conf.h" 11 #include "../h/buf.h" 12 #ifdef QUOTA 13 #include "../h/quota.h" 14 #endif 15 #include "../h/kernel.h" 16 17 #define INOHSZ 63 18 #if ((INOHSZ&(INOHSZ-1)) == 0) 19 #define INOHASH(dev,ino) (((dev)+(ino))&(INOHSZ-1)) 20 #else 21 #define INOHASH(dev,ino) (((unsigned)((dev)+(ino)))%INOHSZ) 22 #endif 23 24 union ihead { /* inode LRU cache, Chris Maltby */ 25 union ihead *ih_head[2]; 26 struct inode *ih_chain[2]; 27 } ihead[INOHSZ]; 28 29 struct inode *ifreeh, **ifreet; 30 31 /* 32 * Initialize hash links for inodes 33 * and build inode free list. 34 */ 35 ihinit() 36 { 37 register int i; 38 register struct inode *ip = inode; 39 register union ihead *ih = ihead; 40 41 for (i = INOHSZ; --i >= 0; ih++) { 42 ih->ih_head[0] = ih; 43 ih->ih_head[1] = ih; 44 } 45 ifreeh = ip; 46 ifreet = &ip->i_freef; 47 ip->i_freeb = &ifreeh; 48 ip->i_forw = ip; 49 ip->i_back = ip; 50 for (i = ninode; --i > 0; ) { 51 ++ip; 52 ip->i_forw = ip; 53 ip->i_back = ip; 54 *ifreet = ip; 55 ip->i_freeb = ifreet; 56 ifreet = &ip->i_freef; 57 } 58 ip->i_freef = NULL; 59 } 60 61 #ifdef notdef 62 /* 63 * Find an inode if it is incore. 64 * This is the equivalent, for inodes, 65 * of ``incore'' in bio.c or ``pfind'' in subr.c. 66 */ 67 struct inode * 68 ifind(dev, ino) 69 dev_t dev; 70 ino_t ino; 71 { 72 register struct inode *ip; 73 register union ihead *ih; 74 75 ih = &ihead[INOHASH(dev, ino)]; 76 for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw) 77 if (ino==ip->i_number && dev==ip->i_dev) 78 return (ip); 79 return ((struct inode *)0); 80 } 81 #endif notdef 82 83 /* 84 * Look up an inode by device,inumber. 85 * If it is in core (in the inode structure), 86 * honor the locking protocol. 87 * If it is not in core, read it in from the 88 * specified device. 89 * If the inode is mounted on, perform 90 * the indicated indirection. 91 * In all cases, a pointer to a locked 92 * inode structure is returned. 93 * 94 * panic: no imt -- if the mounted file 95 * system is not in the mount table. 96 * "cannot happen" 97 */ 98 struct inode * 99 iget(dev, fs, ino) 100 dev_t dev; 101 register struct fs *fs; 102 ino_t ino; 103 { 104 register struct inode *ip; 105 register union ihead *ih; 106 register struct mount *mp; 107 register struct buf *bp; 108 register struct dinode *dp; 109 register struct inode *iq; 110 111 loop: 112 if (getfs(dev) != fs) 113 panic("iget: bad fs"); 114 ih = &ihead[INOHASH(dev, ino)]; 115 for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw) 116 if (ino == ip->i_number && dev == ip->i_dev) { 117 if ((ip->i_flag&ILOCKED) != 0) { 118 ip->i_flag |= IWANT; 119 sleep((caddr_t)ip, PINOD); 120 goto loop; 121 } 122 if ((ip->i_flag&IMOUNT) != 0) { 123 for (mp = &mount[0]; mp < &mount[NMOUNT]; mp++) 124 if(mp->m_inodp == ip) { 125 dev = mp->m_dev; 126 fs = mp->m_bufp->b_un.b_fs; 127 ino = ROOTINO; 128 goto loop; 129 } 130 panic("no imt"); 131 } 132 if (ip->i_count == 0) { /* ino on free list */ 133 if (iq = ip->i_freef) 134 iq->i_freeb = ip->i_freeb; 135 else 136 ifreet = ip->i_freeb; 137 *ip->i_freeb = iq; 138 ip->i_freef = NULL; 139 ip->i_freeb = NULL; 140 } 141 ip->i_count++; 142 ip->i_flag |= ILOCKED; 143 return(ip); 144 } 145 146 if ((ip = ifreeh) == NULL) { 147 tablefull("inode"); 148 u.u_error = ENFILE; 149 return(NULL); 150 } 151 if (iq = ip->i_freef) 152 iq->i_freeb = &ifreeh; 153 ifreeh = iq; 154 ip->i_freef = NULL; 155 ip->i_freeb = NULL; 156 /* 157 * Now to take inode off the hash chain it was on 158 * (initially, or after an iflush, it is on a "hash chain" 159 * consisting entirely of itself, and pointed to by no-one, 160 * but that doesn't matter), and put it on the chain for 161 * its new (ino, dev) pair 162 */ 163 remque(ip); 164 insque(ip, ih); 165 #ifdef QUOTA 166 dqrele(ip->i_dquot); 167 #endif 168 ip->i_dev = dev; 169 ip->i_fs = fs; 170 ip->i_number = ino; 171 ip->i_flag = ILOCKED; 172 ip->i_count++; 173 ip->i_lastr = 0; 174 bp = bread(dev, fsbtodb(fs, itod(fs, ino)), (int)fs->fs_bsize); 175 /* 176 * Check I/O errors 177 */ 178 if ((bp->b_flags&B_ERROR) != 0) { 179 brelse(bp); 180 /* 181 * the inode doesn't contain anything useful, so it would 182 * be misleading to leave it on its hash chain. 183 * 'iput' will take care of putting it back on the free list. 184 */ 185 remque(ip); 186 ip->i_forw = ip; 187 ip->i_back = ip; 188 /* 189 * we also loose its inumber, just in case (as iput 190 * doesn't do that any more) - but as it isn't on its 191 * hash chain, I doubt if this is really necessary .. kre 192 * (probably the two methods are interchangable) 193 */ 194 ip->i_number = 0; 195 #ifdef QUOTA 196 ip->i_dquot = NODQUOT; 197 #endif 198 iput(ip); 199 return(NULL); 200 } 201 dp = bp->b_un.b_dino; 202 dp += itoo(fs, ino); 203 ip->i_ic = dp->di_ic; 204 brelse(bp); 205 #ifdef QUOTA 206 if (ip->i_mode == 0) 207 ip->i_dquot = NODQUOT; 208 else 209 ip->i_dquot = inoquota(ip); 210 #endif 211 return (ip); 212 } 213 214 /* 215 * Decrement reference count of 216 * an inode structure. 217 * On the last reference, 218 * write the inode out and if necessary, 219 * truncate and deallocate the file. 220 */ 221 iput(ip) 222 register struct inode *ip; 223 { 224 225 if ((ip->i_flag & ILOCKED) == 0) 226 panic("iput"); 227 iunlock(ip); 228 irele(ip); 229 } 230 231 irele(ip) 232 register struct inode *ip; 233 { 234 int mode; 235 236 if (ip->i_count == 1) { 237 ip->i_flag |= ILOCKED; 238 if (ip->i_nlink <= 0) { 239 itrunc(ip, (u_long)0); 240 mode = ip->i_mode; 241 ip->i_mode = 0; 242 ip->i_rdev = 0; 243 ip->i_flag |= IUPD|ICHG; 244 ifree(ip, ip->i_number, mode); 245 #ifdef QUOTA 246 (void) chkiq(ip->i_dev, ip, ip->i_uid, 0); 247 dqrele(ip->i_dquot); 248 ip->i_dquot = NODQUOT; 249 #endif 250 } 251 IUPDAT(ip, &time, &time, 0); 252 iunlock(ip); 253 ip->i_flag = 0; 254 /* 255 * Put the inode on the end of the free list. 256 * Possibly in some cases it would be better to 257 * put the inode at the head of the free list, 258 * (eg: where i_mode == 0 || i_number == 0) 259 * but I will think about that later .. kre 260 * (i_number is rarely 0 - only after an i/o error in iget, 261 * where i_mode == 0, the inode will probably be wanted 262 * again soon for an ialloc, so possibly we should keep it) 263 */ 264 if (ifreeh) { 265 *ifreet = ip; 266 ip->i_freeb = ifreet; 267 } else { 268 ifreeh = ip; 269 ip->i_freeb = &ifreeh; 270 } 271 ip->i_freef = NULL; 272 ifreet = &ip->i_freef; 273 } else if ((ip->i_flag & (IUPD|IACC|ICHG)) && 274 ((ip->i_flag & ILOCKED) == 0)) { 275 ip->i_flag |= ILOCKED; 276 iupdat(ip, &time, &time, 0); 277 iunlock(ip); 278 } 279 ip->i_count--; 280 } 281 282 /* 283 * Check accessed and update flags on 284 * an inode structure. 285 * If any is on, update the inode 286 * with the current time. 287 * If waitfor is given, then must insure 288 * i/o order so wait for write to complete. 289 */ 290 iupdat(ip, ta, tm, waitfor) 291 register struct inode *ip; 292 struct timeval *ta, *tm; 293 int waitfor; 294 { 295 register struct buf *bp; 296 struct dinode *dp; 297 register struct fs *fp; 298 299 fp = ip->i_fs; 300 if ((ip->i_flag & (IUPD|IACC|ICHG)) != 0) { 301 if (fp->fs_ronly) 302 return; 303 bp = bread(ip->i_dev, fsbtodb(fp, itod(fp, ip->i_number)), 304 (int)fp->fs_bsize); 305 if (bp->b_flags & B_ERROR) { 306 brelse(bp); 307 return; 308 } 309 if (ip->i_flag&IACC) 310 ip->i_atime = ta->tv_sec; 311 if (ip->i_flag&IUPD) 312 ip->i_mtime = tm->tv_sec; 313 if (ip->i_flag&ICHG) 314 ip->i_ctime = time.tv_sec; 315 ip->i_flag &= ~(IUPD|IACC|ICHG); 316 dp = bp->b_un.b_dino + itoo(fp, ip->i_number); 317 dp->di_ic = ip->i_ic; 318 if (waitfor) 319 bwrite(bp); 320 else 321 bdwrite(bp); 322 } 323 } 324 325 #define SINGLE 0 /* index of single indirect block */ 326 #define DOUBLE 1 /* index of double indirect block */ 327 #define TRIPLE 2 /* index of triple indirect block */ 328 /* 329 * Truncate the inode ip to at most 330 * length size. Free affected disk 331 * blocks -- the blocks of the file 332 * are removed in reverse order. 333 * 334 * NB: triple indirect blocks are untested. 335 */ 336 itrunc(oip, length) 337 struct inode *oip; 338 u_long length; 339 { 340 register i; 341 register daddr_t lastblock; 342 daddr_t bn, lastiblock[NIADDR]; 343 register struct fs *fs; 344 register struct inode *ip; 345 struct inode tip; 346 long blocksreleased = 0, nblocks; 347 long indirtrunc(); 348 int level; 349 350 if (oip->i_size <= length) { 351 oip->i_flag |= ICHG|IUPD; 352 iupdat(oip, &time, &time, 1); 353 return; 354 } 355 /* 356 * Calculate index into inode's block list of 357 * last direct and indirect blocks (if any) 358 * which we want to keep. Lastblock is -1 when 359 * the file is truncated to 0. 360 */ 361 fs = oip->i_fs; 362 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 363 lastiblock[SINGLE] = lastblock - NDADDR; 364 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 365 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 366 nblocks = btodb(fs->fs_bsize); 367 /* 368 * Update size of file and block pointers 369 * on disk before we start freeing blocks. 370 * If we crash before free'ing blocks below, 371 * the blocks will be returned to the free list. 372 * lastiblock values are also normalized to -1 373 * for calls to indirtrunc below. 374 * (? fsck doesn't check validity of pointers in indirect blocks) 375 */ 376 tip = *oip; 377 for (level = TRIPLE; level >= SINGLE; level--) 378 if (lastiblock[level] < 0) { 379 oip->i_ib[level] = 0; 380 lastiblock[level] = -1; 381 } 382 for (i = NDADDR - 1; i > lastblock; i--) 383 oip->i_db[i] = 0; 384 oip->i_size = length; 385 oip->i_flag |= ICHG|IUPD; 386 iupdat(oip, &time, &time, 1); 387 ip = &tip; 388 389 /* 390 * Indirect blocks first. 391 */ 392 for (level = TRIPLE; level >= SINGLE; level--) { 393 bn = ip->i_ib[level]; 394 if (bn != 0) { 395 blocksreleased += 396 indirtrunc(ip, bn, lastiblock[level], level); 397 if (lastiblock[level] < 0) { 398 ip->i_ib[level] = 0; 399 free(ip, bn, (off_t)fs->fs_bsize); 400 blocksreleased += nblocks; 401 } 402 } 403 if (lastiblock[level] >= 0) 404 goto done; 405 } 406 407 /* 408 * All whole direct blocks or frags. 409 */ 410 for (i = NDADDR - 1; i > lastblock; i--) { 411 register int size; 412 413 bn = ip->i_db[i]; 414 if (bn == 0) 415 continue; 416 ip->i_db[i] = 0; 417 size = (off_t)blksize(fs, ip, i); 418 free(ip, bn, size); 419 blocksreleased += btodb(size); 420 } 421 if (lastblock < 0) 422 goto done; 423 424 /* 425 * Finally, look for a change in size of the 426 * last direct block; release any frags. 427 */ 428 bn = ip->i_db[lastblock]; 429 if (bn != 0) { 430 int oldspace, newspace; 431 432 /* 433 * Calculate amount of space we're giving 434 * back as old block size minus new block size. 435 */ 436 oldspace = blksize(fs, ip, lastblock); 437 ip->i_size = length; 438 newspace = blksize(fs, ip, lastblock); 439 if (newspace == 0) 440 panic("itrunc: newspace"); 441 if (oldspace - newspace > 0) { 442 /* 443 * Block number of space to be free'd is 444 * the old block # plus the number of frags 445 * required for the storage we're keeping. 446 */ 447 bn += numfrags(fs, newspace); 448 free(ip, bn, oldspace - newspace); 449 blocksreleased += btodb(oldspace - newspace); 450 } 451 } 452 done: 453 /* BEGIN PARANOIA */ 454 for (level = SINGLE; level <= TRIPLE; level++) 455 if (ip->i_ib[level] != oip->i_ib[level]) 456 panic("itrunc1"); 457 for (i = 0; i < NDADDR; i++) 458 if (ip->i_db[i] != oip->i_db[i]) 459 panic("itrunc2"); 460 /* END PARANOIA */ 461 oip->i_blocks -= blocksreleased; 462 if (oip->i_blocks < 0) /* sanity */ 463 oip->i_blocks = 0; 464 oip->i_flag |= ICHG; 465 #ifdef QUOTA 466 (void) chkdq(oip, -blocksreleased, 0); 467 #endif 468 } 469 470 /* 471 * Release blocks associated with the inode ip and 472 * stored in the indirect block bn. Blocks are free'd 473 * in LIFO order up to (but not including) lastbn. If 474 * level is greater than SINGLE, the block is an indirect 475 * block and recursive calls to indirtrunc must be used to 476 * cleanse other indirect blocks. 477 * 478 * NB: triple indirect blocks are untested. 479 */ 480 long 481 indirtrunc(ip, bn, lastbn, level) 482 register struct inode *ip; 483 daddr_t bn, lastbn; 484 int level; 485 { 486 register int i; 487 struct buf *bp, *copy; 488 register daddr_t *bap; 489 register struct fs *fs = ip->i_fs; 490 daddr_t nb, last; 491 long factor; 492 int blocksreleased = 0, nblocks; 493 494 /* 495 * Calculate index in current block of last 496 * block to be kept. -1 indicates the entire 497 * block so we need not calculate the index. 498 */ 499 factor = 1; 500 for (i = SINGLE; i < level; i++) 501 factor *= NINDIR(fs); 502 last = lastbn; 503 if (lastbn > 0) 504 last /= factor; 505 nblocks = btodb(fs->fs_bsize); 506 /* 507 * Get buffer of block pointers, zero those 508 * entries corresponding to blocks to be free'd, 509 * and update on disk copy first. 510 */ 511 copy = geteblk((int)fs->fs_bsize); 512 bp = bread(ip->i_dev, fsbtodb(fs, bn), (int)fs->fs_bsize); 513 if (bp->b_flags&B_ERROR) { 514 brelse(copy); 515 brelse(bp); 516 return (0); 517 } 518 bap = bp->b_un.b_daddr; 519 bcopy((caddr_t)bap, (caddr_t)copy->b_un.b_daddr, (u_int)fs->fs_bsize); 520 bzero((caddr_t)&bap[last + 1], 521 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t)); 522 bwrite(bp); 523 bp = copy, bap = bp->b_un.b_daddr; 524 525 /* 526 * Recursively free totally unused blocks. 527 */ 528 for (i = NINDIR(fs) - 1; i > last; i--) { 529 nb = bap[i]; 530 if (nb == 0) 531 continue; 532 if (level > SINGLE) 533 blocksreleased += 534 indirtrunc(ip, nb, (daddr_t)-1, level - 1); 535 free(ip, nb, (int)fs->fs_bsize); 536 blocksreleased += nblocks; 537 } 538 539 /* 540 * Recursively free last partial block. 541 */ 542 if (level > SINGLE && lastbn >= 0) { 543 last = lastbn % factor; 544 nb = bap[i]; 545 if (nb != 0) 546 blocksreleased += indirtrunc(ip, nb, last, level - 1); 547 } 548 brelse(bp); 549 return (blocksreleased); 550 } 551 552 /* 553 * remove any inodes in the inode cache belonging to dev 554 * 555 * There should not be any active ones, return error if any are found 556 * (nb: this is a user error, not a system err) 557 * 558 * Also, count the references to dev by block devices - this really 559 * has nothing to do with the object of the procedure, but as we have 560 * to scan the inode table here anyway, we might as well get the 561 * extra benefit. 562 * 563 * this is called from sumount()/sys3.c when dev is being unmounted 564 */ 565 #ifdef QUOTA 566 iflush(dev, iq) 567 dev_t dev; 568 struct inode *iq; 569 #else 570 iflush(dev) 571 dev_t dev; 572 #endif 573 { 574 register struct inode *ip; 575 register open = 0; 576 577 for (ip = inode; ip < inodeNINODE; ip++) { 578 #ifdef QUOTA 579 if (ip != iq && ip->i_dev == dev) 580 #else 581 if (ip->i_dev == dev) 582 #endif 583 if (ip->i_count) 584 return(-1); 585 else { 586 remque(ip); 587 ip->i_forw = ip; 588 ip->i_back = ip; 589 /* 590 * as i_count == 0, the inode was on the free 591 * list already, just leave it there, it will 592 * fall off the bottom eventually. We could 593 * perhaps move it to the head of the free 594 * list, but as umounts are done so 595 * infrequently, we would gain very little, 596 * while making the code bigger. 597 */ 598 #ifdef QUOTA 599 dqrele(ip->i_dquot); 600 ip->i_dquot = NODQUOT; 601 #endif 602 } 603 else if (ip->i_count && (ip->i_mode&IFMT)==IFBLK && 604 ip->i_rdev == dev) 605 open++; 606 } 607 return (open); 608 } 609 610 /* 611 * Lock an inode. If its already locked, set the WANT bit and sleep. 612 */ 613 ilock(ip) 614 register struct inode *ip; 615 { 616 617 ILOCK(ip); 618 } 619 620 /* 621 * Unlock an inode. If WANT bit is on, wakeup. 622 */ 623 iunlock(ip) 624 register struct inode *ip; 625 { 626 627 IUNLOCK(ip); 628 } 629