1 /* 2 * Copyright (c) 1982, 1986, 1989 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)ufs_inode.c 7.33 (Berkeley) 06/28/90 8 */ 9 10 #include "param.h" 11 #include "systm.h" 12 #include "mount.h" 13 #include "user.h" 14 #include "proc.h" 15 #include "file.h" 16 #include "buf.h" 17 #include "cmap.h" 18 #include "vnode.h" 19 #include "../ufs/quota.h" 20 #include "../ufs/inode.h" 21 #include "../ufs/fs.h" 22 #include "../ufs/ufsmount.h" 23 #include "kernel.h" 24 #include "malloc.h" 25 26 #define INOHSZ 512 27 #if ((INOHSZ&(INOHSZ-1)) == 0) 28 #define INOHASH(dev,ino) (((dev)+(ino))&(INOHSZ-1)) 29 #else 30 #define INOHASH(dev,ino) (((unsigned)((dev)+(ino)))%INOHSZ) 31 #endif 32 33 union ihead { 34 union ihead *ih_head[2]; 35 struct inode *ih_chain[2]; 36 } ihead[INOHSZ]; 37 38 int prtactive; /* 1 => print out reclaim of active vnodes */ 39 40 /* 41 * Initialize hash links for inodes. 42 */ 43 ufs_init() 44 { 45 register int i; 46 register union ihead *ih = ihead; 47 48 #ifndef lint 49 if (VN_MAXPRIVATE < sizeof(struct inode)) 50 panic("ihinit: too small"); 51 #endif /* not lint */ 52 for (i = INOHSZ; --i >= 0; ih++) { 53 ih->ih_head[0] = ih; 54 ih->ih_head[1] = ih; 55 } 56 #ifdef QUOTA 57 dqinit(); 58 #endif /* QUOTA */ 59 } 60 61 /* 62 * Look up an vnode/inode by device,inumber. 63 * If it is in core (in the inode structure), 64 * honor the locking protocol. 65 * If it is not in core, read it in from the 66 * specified device. 67 * Callers must check for mount points!! 68 * In all cases, a pointer to a locked 69 * inode structure is returned. 70 */ 71 iget(xp, ino, ipp) 72 struct inode *xp; 73 ino_t ino; 74 struct inode **ipp; 75 { 76 dev_t dev = xp->i_dev; 77 struct mount *mntp = ITOV(xp)->v_mount; 78 register struct fs *fs = VFSTOUFS(mntp)->um_fs; 79 extern struct vnodeops ufs_vnodeops, spec_inodeops; 80 register struct inode *ip, *iq; 81 register struct vnode *vp; 82 struct vnode *nvp; 83 struct buf *bp; 84 struct dinode *dp; 85 union ihead *ih; 86 int i, error; 87 88 ih = &ihead[INOHASH(dev, ino)]; 89 loop: 90 for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw) { 91 if (ino != ip->i_number || dev != ip->i_dev) 92 continue; 93 if ((ip->i_flag&ILOCKED) != 0) { 94 ip->i_flag |= IWANT; 95 sleep((caddr_t)ip, PINOD); 96 goto loop; 97 } 98 if (vget(ITOV(ip))) 99 goto loop; 100 *ipp = ip; 101 return(0); 102 } 103 /* 104 * Allocate a new inode. 105 */ 106 if (error = getnewvnode(VT_UFS, mntp, &ufs_vnodeops, &nvp)) { 107 *ipp = 0; 108 return (error); 109 } 110 ip = VTOI(nvp); 111 ip->i_vnode = nvp; 112 ip->i_flag = 0; 113 ip->i_devvp = 0; 114 ip->i_mode = 0; 115 ip->i_diroff = 0; 116 #ifdef QUOTA 117 for (i = 0; i < MAXQUOTAS; i++) 118 ip->i_dquot[i] = NODQUOT; 119 #endif 120 /* 121 * Put it onto its hash chain and lock it so that other requests for 122 * this inode will block if they arrive while we are sleeping waiting 123 * for old data structures to be purged or for the contents of the 124 * disk portion of this inode to be read. 125 */ 126 ip->i_dev = dev; 127 ip->i_number = ino; 128 insque(ip, ih); 129 ILOCK(ip); 130 /* 131 * Read in the disk contents for the inode. 132 */ 133 if (error = bread(VFSTOUFS(mntp)->um_devvp, fsbtodb(fs, itod(fs, ino)), 134 (int)fs->fs_bsize, NOCRED, &bp)) { 135 /* 136 * The inode does not contain anything useful, so it would 137 * be misleading to leave it on its hash chain. 138 * Iput() will take care of putting it back on the free list. 139 */ 140 remque(ip); 141 ip->i_forw = ip; 142 ip->i_back = ip; 143 /* 144 * Unlock and discard unneeded inode. 145 */ 146 iput(ip); 147 brelse(bp); 148 *ipp = 0; 149 return (error); 150 } 151 dp = bp->b_un.b_dino; 152 dp += itoo(fs, ino); 153 ip->i_din = *dp; 154 brelse(bp); 155 /* 156 * Initialize the associated vnode 157 */ 158 vp = ITOV(ip); 159 vp->v_type = IFTOVT(ip->i_mode); 160 if (vp->v_type == VFIFO) { 161 #ifdef FIFO 162 extern struct vnodeops fifo_inodeops; 163 vp->v_op = &fifo_inodeops; 164 #else 165 iput(ip); 166 *ipp = 0; 167 return (EOPNOTSUPP); 168 #endif /* FIFO */ 169 } 170 if (vp->v_type == VCHR || vp->v_type == VBLK) { 171 vp->v_op = &spec_inodeops; 172 if (nvp = checkalias(vp, ip->i_rdev, mntp)) { 173 /* 174 * Reinitialize aliased inode. 175 */ 176 vp = nvp; 177 iq = VTOI(vp); 178 iq->i_vnode = vp; 179 iq->i_flag = 0; 180 ILOCK(iq); 181 iq->i_din = ip->i_din; 182 iq->i_dev = dev; 183 iq->i_number = ino; 184 insque(iq, ih); 185 /* 186 * Discard unneeded vnode 187 */ 188 ip->i_mode = 0; 189 iput(ip); 190 ip = iq; 191 } 192 } 193 if (ino == ROOTINO) 194 vp->v_flag |= VROOT; 195 /* 196 * Finish inode initialization. 197 */ 198 ip->i_fs = fs; 199 ip->i_devvp = VFSTOUFS(mntp)->um_devvp; 200 VREF(ip->i_devvp); 201 /* 202 * Set up a generation number for this inode if it does not 203 * already have one. This should only happen on old filesystems. 204 */ 205 if (ip->i_gen == 0) { 206 if (++nextgennumber < (u_long)time.tv_sec) 207 nextgennumber = time.tv_sec; 208 ip->i_gen = nextgennumber; 209 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) 210 ip->i_flag |= IMOD; 211 } 212 *ipp = ip; 213 return (0); 214 } 215 216 /* 217 * Unlock and decrement the reference count of an inode structure. 218 */ 219 iput(ip) 220 register struct inode *ip; 221 { 222 223 if ((ip->i_flag & ILOCKED) == 0) 224 panic("iput"); 225 IUNLOCK(ip); 226 vrele(ITOV(ip)); 227 } 228 229 /* 230 * Last reference to an inode, write the inode out and if necessary, 231 * truncate and deallocate the file. 232 */ 233 ufs_inactive(vp) 234 struct vnode *vp; 235 { 236 register struct inode *ip = VTOI(vp); 237 int mode, error = 0; 238 239 if (prtactive && vp->v_usecount != 0) 240 vprint("ufs_inactive: pushing active", vp); 241 /* 242 * Get rid of inodes related to stale file handles. 243 */ 244 if (ip->i_mode == 0) { 245 if ((vp->v_flag & VXLOCK) == 0) 246 vgone(vp); 247 return (0); 248 } 249 ILOCK(ip); 250 if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 251 #ifdef QUOTA 252 if (!getinoquota(ip)) 253 (void) chkiq(ip, -1, NOCRED, 0); 254 #endif 255 error = itrunc(ip, (u_long)0, 0); 256 mode = ip->i_mode; 257 ip->i_mode = 0; 258 ip->i_flag |= IUPD|ICHG; 259 ifree(ip, ip->i_number, mode); 260 } 261 IUPDAT(ip, &time, &time, 0); 262 IUNLOCK(ip); 263 ip->i_flag = 0; 264 /* 265 * If we are done with the inode, reclaim it 266 * so that it can be reused immediately. 267 */ 268 if (vp->v_usecount == 0 && ip->i_mode == 0) 269 vgone(vp); 270 return (error); 271 } 272 273 /* 274 * Reclaim an inode so that it can be used for other purposes. 275 */ 276 ufs_reclaim(vp) 277 register struct vnode *vp; 278 { 279 register struct inode *ip = VTOI(vp); 280 int i; 281 282 if (prtactive && vp->v_usecount != 0) 283 vprint("ufs_reclaim: pushing active", vp); 284 /* 285 * Remove the inode from its hash chain. 286 */ 287 remque(ip); 288 ip->i_forw = ip; 289 ip->i_back = ip; 290 /* 291 * Purge old data structures associated with the inode. 292 */ 293 cache_purge(vp); 294 if (ip->i_devvp) { 295 vrele(ip->i_devvp); 296 ip->i_devvp = 0; 297 } 298 #ifdef QUOTA 299 for (i = 0; i < MAXQUOTAS; i++) { 300 if (ip->i_dquot[i] != NODQUOT) { 301 dqrele(vp, ip->i_dquot[i]); 302 ip->i_dquot[i] = NODQUOT; 303 } 304 } 305 #endif 306 ip->i_flag = 0; 307 return (0); 308 } 309 310 /* 311 * Check accessed and update flags on an inode structure. 312 * If any is on, update the inode with the current time. 313 * If waitfor is given, then must ensure I/O order, 314 * so wait for write to complete. 315 */ 316 iupdat(ip, ta, tm, waitfor) 317 register struct inode *ip; 318 struct timeval *ta, *tm; 319 int waitfor; 320 { 321 struct buf *bp; 322 struct vnode *vp = ITOV(ip); 323 struct dinode *dp; 324 register struct fs *fs; 325 int error; 326 327 fs = ip->i_fs; 328 if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0) 329 return (0); 330 if (vp->v_mount->mnt_flag & MNT_RDONLY) 331 return (0); 332 error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)), 333 (int)fs->fs_bsize, NOCRED, &bp); 334 if (error) { 335 brelse(bp); 336 return (error); 337 } 338 if (ip->i_flag&IACC) 339 ip->i_atime = ta->tv_sec; 340 if (ip->i_flag&IUPD) 341 ip->i_mtime = tm->tv_sec; 342 if (ip->i_flag&ICHG) 343 ip->i_ctime = time.tv_sec; 344 ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD); 345 dp = bp->b_un.b_dino + itoo(fs, ip->i_number); 346 *dp = ip->i_din; 347 if (waitfor) { 348 return (bwrite(bp)); 349 } else { 350 bdwrite(bp); 351 return (0); 352 } 353 } 354 355 #define SINGLE 0 /* index of single indirect block */ 356 #define DOUBLE 1 /* index of double indirect block */ 357 #define TRIPLE 2 /* index of triple indirect block */ 358 /* 359 * Truncate the inode ip to at most length size. Free affected disk 360 * blocks -- the blocks of the file are removed in reverse order. 361 * 362 * NB: triple indirect blocks are untested. 363 */ 364 itrunc(oip, length, flags) 365 register struct inode *oip; 366 u_long length; 367 int flags; 368 { 369 register daddr_t lastblock; 370 daddr_t bn, lbn, lastiblock[NIADDR]; 371 register struct fs *fs; 372 register struct inode *ip; 373 struct buf *bp; 374 int offset, osize, size, level; 375 long count, nblocks, blocksreleased = 0; 376 register int i; 377 int aflags, error, allerror; 378 struct inode tip; 379 380 if (oip->i_size <= length) { 381 oip->i_flag |= ICHG|IUPD; 382 error = iupdat(oip, &time, &time, 1); 383 return (error); 384 } 385 /* 386 * Calculate index into inode's block list of 387 * last direct and indirect blocks (if any) 388 * which we want to keep. Lastblock is -1 when 389 * the file is truncated to 0. 390 */ 391 fs = oip->i_fs; 392 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 393 lastiblock[SINGLE] = lastblock - NDADDR; 394 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 395 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 396 nblocks = btodb(fs->fs_bsize); 397 /* 398 * Update the size of the file. If the file is not being 399 * truncated to a block boundry, the contents of the 400 * partial block following the end of the file must be 401 * zero'ed in case it ever become accessable again because 402 * of subsequent file growth. 403 */ 404 osize = oip->i_size; 405 offset = blkoff(fs, length); 406 if (offset == 0) { 407 oip->i_size = length; 408 } else { 409 lbn = lblkno(fs, length); 410 aflags = B_CLRBUF; 411 if (flags & IO_SYNC) 412 aflags |= B_SYNC; 413 #ifdef QUOTA 414 if (error = getinoquota(oip)) 415 return (error); 416 #endif 417 if (error = balloc(oip, lbn, offset, &bp, aflags)) 418 return (error); 419 oip->i_size = length; 420 size = blksize(fs, oip, lbn); 421 bn = bp->b_blkno; 422 count = howmany(size, CLBYTES); 423 for (i = 0; i < count; i++) 424 munhash(oip->i_devvp, bn + i * CLBYTES / DEV_BSIZE); 425 bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset)); 426 brealloc(bp, size); 427 if (flags & IO_SYNC) 428 bwrite(bp); 429 else 430 bdwrite(bp); 431 } 432 /* 433 * Update file and block pointers 434 * on disk before we start freeing blocks. 435 * If we crash before free'ing blocks below, 436 * the blocks will be returned to the free list. 437 * lastiblock values are also normalized to -1 438 * for calls to indirtrunc below. 439 */ 440 tip = *oip; 441 tip.i_size = osize; 442 for (level = TRIPLE; level >= SINGLE; level--) 443 if (lastiblock[level] < 0) { 444 oip->i_ib[level] = 0; 445 lastiblock[level] = -1; 446 } 447 for (i = NDADDR - 1; i > lastblock; i--) 448 oip->i_db[i] = 0; 449 oip->i_flag |= ICHG|IUPD; 450 vinvalbuf(ITOV(oip), (length > 0)); 451 allerror = iupdat(oip, &time, &time, MNT_WAIT); 452 453 /* 454 * Indirect blocks first. 455 */ 456 ip = &tip; 457 for (level = TRIPLE; level >= SINGLE; level--) { 458 bn = ip->i_ib[level]; 459 if (bn != 0) { 460 error = indirtrunc(ip, bn, lastiblock[level], level, 461 &count); 462 if (error) 463 allerror = error; 464 blocksreleased += count; 465 if (lastiblock[level] < 0) { 466 ip->i_ib[level] = 0; 467 blkfree(ip, bn, (off_t)fs->fs_bsize); 468 blocksreleased += nblocks; 469 } 470 } 471 if (lastiblock[level] >= 0) 472 goto done; 473 } 474 475 /* 476 * All whole direct blocks or frags. 477 */ 478 for (i = NDADDR - 1; i > lastblock; i--) { 479 register off_t bsize; 480 481 bn = ip->i_db[i]; 482 if (bn == 0) 483 continue; 484 ip->i_db[i] = 0; 485 bsize = (off_t)blksize(fs, ip, i); 486 blkfree(ip, bn, bsize); 487 blocksreleased += btodb(bsize); 488 } 489 if (lastblock < 0) 490 goto done; 491 492 /* 493 * Finally, look for a change in size of the 494 * last direct block; release any frags. 495 */ 496 bn = ip->i_db[lastblock]; 497 if (bn != 0) { 498 off_t oldspace, newspace; 499 500 /* 501 * Calculate amount of space we're giving 502 * back as old block size minus new block size. 503 */ 504 oldspace = blksize(fs, ip, lastblock); 505 ip->i_size = length; 506 newspace = blksize(fs, ip, lastblock); 507 if (newspace == 0) 508 panic("itrunc: newspace"); 509 if (oldspace - newspace > 0) { 510 /* 511 * Block number of space to be free'd is 512 * the old block # plus the number of frags 513 * required for the storage we're keeping. 514 */ 515 bn += numfrags(fs, newspace); 516 blkfree(ip, bn, oldspace - newspace); 517 blocksreleased += btodb(oldspace - newspace); 518 } 519 } 520 done: 521 /* BEGIN PARANOIA */ 522 for (level = SINGLE; level <= TRIPLE; level++) 523 if (ip->i_ib[level] != oip->i_ib[level]) 524 panic("itrunc1"); 525 for (i = 0; i < NDADDR; i++) 526 if (ip->i_db[i] != oip->i_db[i]) 527 panic("itrunc2"); 528 /* END PARANOIA */ 529 oip->i_blocks -= blocksreleased; 530 if (oip->i_blocks < 0) /* sanity */ 531 oip->i_blocks = 0; 532 oip->i_flag |= ICHG; 533 #ifdef QUOTA 534 if (!getinoquota(oip)) 535 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 536 #endif 537 return (allerror); 538 } 539 540 /* 541 * Release blocks associated with the inode ip and 542 * stored in the indirect block bn. Blocks are free'd 543 * in LIFO order up to (but not including) lastbn. If 544 * level is greater than SINGLE, the block is an indirect 545 * block and recursive calls to indirtrunc must be used to 546 * cleanse other indirect blocks. 547 * 548 * NB: triple indirect blocks are untested. 549 */ 550 indirtrunc(ip, bn, lastbn, level, countp) 551 register struct inode *ip; 552 daddr_t bn, lastbn; 553 int level; 554 long *countp; 555 { 556 register int i; 557 struct buf *bp; 558 register struct fs *fs = ip->i_fs; 559 register daddr_t *bap; 560 daddr_t *copy, nb, last; 561 long blkcount, factor; 562 int nblocks, blocksreleased = 0; 563 int error, allerror = 0; 564 565 /* 566 * Calculate index in current block of last 567 * block to be kept. -1 indicates the entire 568 * block so we need not calculate the index. 569 */ 570 factor = 1; 571 for (i = SINGLE; i < level; i++) 572 factor *= NINDIR(fs); 573 last = lastbn; 574 if (lastbn > 0) 575 last /= factor; 576 nblocks = btodb(fs->fs_bsize); 577 /* 578 * Get buffer of block pointers, zero those 579 * entries corresponding to blocks to be free'd, 580 * and update on disk copy first. 581 */ 582 error = bread(ip->i_devvp, fsbtodb(fs, bn), (int)fs->fs_bsize, 583 NOCRED, &bp); 584 if (error) { 585 brelse(bp); 586 *countp = 0; 587 return (error); 588 } 589 bap = bp->b_un.b_daddr; 590 MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK); 591 bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize); 592 bzero((caddr_t)&bap[last + 1], 593 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t)); 594 if (last == -1) 595 bp->b_flags |= B_INVAL; 596 error = bwrite(bp); 597 if (error) 598 allerror = error; 599 bap = copy; 600 601 /* 602 * Recursively free totally unused blocks. 603 */ 604 for (i = NINDIR(fs) - 1; i > last; i--) { 605 nb = bap[i]; 606 if (nb == 0) 607 continue; 608 if (level > SINGLE) { 609 error = indirtrunc(ip, nb, (daddr_t)-1, level - 1, 610 &blkcount); 611 if (error) 612 allerror = error; 613 blocksreleased += blkcount; 614 } 615 blkfree(ip, nb, (off_t)fs->fs_bsize); 616 blocksreleased += nblocks; 617 } 618 619 /* 620 * Recursively free last partial block. 621 */ 622 if (level > SINGLE && lastbn >= 0) { 623 last = lastbn % factor; 624 nb = bap[i]; 625 if (nb != 0) { 626 error = indirtrunc(ip, nb, last, level - 1, &blkcount); 627 if (error) 628 allerror = error; 629 blocksreleased += blkcount; 630 } 631 } 632 FREE(copy, M_TEMP); 633 *countp = blocksreleased; 634 return (allerror); 635 } 636 637 /* 638 * Lock an inode. If its already locked, set the WANT bit and sleep. 639 */ 640 ilock(ip) 641 register struct inode *ip; 642 { 643 644 while (ip->i_flag & ILOCKED) { 645 ip->i_flag |= IWANT; 646 if (ip->i_spare0 == u.u_procp->p_pid) 647 panic("locking against myself"); 648 ip->i_spare1 = u.u_procp->p_pid; 649 (void) sleep((caddr_t)ip, PINOD); 650 } 651 ip->i_spare1 = 0; 652 ip->i_spare0 = u.u_procp->p_pid; 653 u.u_spare[0]++; 654 ip->i_flag |= ILOCKED; 655 } 656 657 /* 658 * Unlock an inode. If WANT bit is on, wakeup. 659 */ 660 iunlock(ip) 661 register struct inode *ip; 662 { 663 664 if ((ip->i_flag & ILOCKED) == 0) 665 vprint("iunlock: unlocked inode", ITOV(ip)); 666 ip->i_spare0 = 0; 667 u.u_spare[0]--; 668 ip->i_flag &= ~ILOCKED; 669 if (ip->i_flag&IWANT) { 670 ip->i_flag &= ~IWANT; 671 wakeup((caddr_t)ip); 672 } 673 } 674