1 /* 2 * Copyright (c) 1982, 1986, 1989 Regents of the University of California. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms are permitted 6 * provided that the above copyright notice and this paragraph are 7 * duplicated in all such forms and that any documentation, 8 * advertising materials, and other materials related to such 9 * distribution and use acknowledge that the software was developed 10 * by the University of California, Berkeley. The name of the 11 * University may not be used to endorse or promote products derived 12 * from this software without specific prior written permission. 13 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR 14 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED 15 * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. 16 * 17 * @(#)ufs_inode.c 7.25 (Berkeley) 12/29/89 18 */ 19 20 #include "param.h" 21 #include "systm.h" 22 #include "mount.h" 23 #include "user.h" 24 #include "proc.h" 25 #include "file.h" 26 #include "buf.h" 27 #include "cmap.h" 28 #include "vnode.h" 29 #include "../ufs/inode.h" 30 #include "../ufs/fs.h" 31 #include "../ufs/ufsmount.h" 32 #ifdef QUOTA 33 #include "../ufs/quota.h" 34 #endif 35 #include "kernel.h" 36 #include "malloc.h" 37 38 #define INOHSZ 512 39 #if ((INOHSZ&(INOHSZ-1)) == 0) 40 #define INOHASH(dev,ino) (((dev)+(ino))&(INOHSZ-1)) 41 #else 42 #define INOHASH(dev,ino) (((unsigned)((dev)+(ino)))%INOHSZ) 43 #endif 44 45 union ihead { 46 union ihead *ih_head[2]; 47 struct inode *ih_chain[2]; 48 } ihead[INOHSZ]; 49 50 int prtactive; /* 1 => print out reclaim of active vnodes */ 51 52 /* 53 * Initialize hash links for inodes. 54 */ 55 ufs_init() 56 { 57 register int i; 58 register union ihead *ih = ihead; 59 60 #ifndef lint 61 if (VN_MAXPRIVATE < sizeof(struct inode)) 62 panic("ihinit: too small"); 63 #endif /* not lint */ 64 for (i = INOHSZ; --i >= 0; ih++) { 65 ih->ih_head[0] = ih; 66 ih->ih_head[1] = ih; 67 } 68 } 69 70 /* 71 * Look up an vnode/inode by device,inumber. 72 * If it is in core (in the inode structure), 73 * honor the locking protocol. 74 * If it is not in core, read it in from the 75 * specified device. 76 * Callers must check for mount points!! 77 * In all cases, a pointer to a locked 78 * inode structure is returned. 79 */ 80 iget(xp, ino, ipp) 81 struct inode *xp; 82 ino_t ino; 83 struct inode **ipp; 84 { 85 dev_t dev = xp->i_dev; 86 struct mount *mntp = ITOV(xp)->v_mount; 87 register struct fs *fs = VFSTOUFS(mntp)->um_fs; 88 extern struct vnodeops ufs_vnodeops, spec_inodeops; 89 register struct inode *ip, *iq; 90 register struct vnode *vp; 91 struct vnode *nvp; 92 struct buf *bp; 93 struct dinode *dp; 94 union ihead *ih; 95 int error; 96 97 ih = &ihead[INOHASH(dev, ino)]; 98 loop: 99 for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw) { 100 if (ino != ip->i_number || dev != ip->i_dev) 101 continue; 102 if ((ip->i_flag&ILOCKED) != 0) { 103 ip->i_flag |= IWANT; 104 sleep((caddr_t)ip, PINOD); 105 goto loop; 106 } 107 if (vget(ITOV(ip))) 108 goto loop; 109 *ipp = ip; 110 return(0); 111 } 112 /* 113 * Allocate a new inode. 114 */ 115 if (error = getnewvnode(VT_UFS, mntp, &ufs_vnodeops, &nvp)) { 116 *ipp = 0; 117 return (error); 118 } 119 ip = VTOI(nvp); 120 ip->i_vnode = nvp; 121 ip->i_flag = 0; 122 ip->i_devvp = 0; 123 ip->i_lastr = 0; 124 ip->i_mode = 0; 125 #ifdef QUOTA 126 ip->i_dquot = NODQUOT; 127 #endif 128 /* 129 * Put it onto its hash chain and lock it so that other requests for 130 * this inode will block if they arrive while we are sleeping waiting 131 * for old data structures to be purged or for the contents of the 132 * disk portion of this inode to be read. 133 */ 134 ip->i_dev = dev; 135 ip->i_number = ino; 136 insque(ip, ih); 137 ILOCK(ip); 138 /* 139 * Read in the disk contents for the inode. 140 */ 141 if (error = bread(VFSTOUFS(mntp)->um_devvp, fsbtodb(fs, itod(fs, ino)), 142 (int)fs->fs_bsize, NOCRED, &bp)) { 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 == VCHR || vp->v_type == VBLK) { 161 vp->v_op = &spec_inodeops; 162 if (nvp = checkalias(vp, ip->i_rdev, mntp)) { 163 /* 164 * Reinitialize aliased inode. 165 */ 166 vp = nvp; 167 iq = VTOI(vp); 168 iq->i_vnode = vp; 169 iq->i_lastr = 0; 170 iq->i_flag = 0; 171 ILOCK(iq); 172 iq->i_din = ip->i_din; 173 iq->i_dev = dev; 174 iq->i_number = ino; 175 insque(iq, ih); 176 /* 177 * Discard unneeded vnode 178 */ 179 ip->i_mode = 0; 180 iput(ip); 181 ip = iq; 182 } 183 } 184 if (ino == ROOTINO) 185 vp->v_flag |= VROOT; 186 /* 187 * Finish inode initialization. 188 */ 189 ip->i_fs = fs; 190 ip->i_devvp = VFSTOUFS(mntp)->um_devvp; 191 VREF(ip->i_devvp); 192 #ifdef QUOTA 193 if (ip->i_mode != 0) 194 ip->i_dquot = inoquota(ip); 195 #endif 196 /* 197 * Set up a generation number for this inode if it does not 198 * already have one. This should only happen on old filesystems. 199 */ 200 if (ip->i_gen == 0) { 201 if (++nextgennumber < (u_long)time.tv_sec) 202 nextgennumber = time.tv_sec; 203 ip->i_gen = nextgennumber; 204 if ((vp->v_mount->m_flag & M_RDONLY) == 0) 205 ip->i_flag |= IMOD; 206 } 207 *ipp = ip; 208 return (0); 209 } 210 211 /* 212 * Unlock and decrement the reference count of an inode structure. 213 */ 214 iput(ip) 215 register struct inode *ip; 216 { 217 218 if ((ip->i_flag & ILOCKED) == 0) 219 panic("iput"); 220 IUNLOCK(ip); 221 vrele(ITOV(ip)); 222 } 223 224 /* 225 * Last reference to an inode, write the inode out and if necessary, 226 * truncate and deallocate the file. 227 */ 228 ufs_inactive(vp) 229 struct vnode *vp; 230 { 231 register struct inode *ip = VTOI(vp); 232 int mode, error = 0; 233 234 if (prtactive && vp->v_count != 0) 235 vprint("ufs_inactive: pushing active", vp); 236 /* 237 * Get rid of inodes related to stale file handles. 238 */ 239 if (ip->i_mode == 0) { 240 if ((vp->v_flag & VXLOCK) == 0) 241 vgone(vp); 242 return (0); 243 } 244 ILOCK(ip); 245 if (ip->i_nlink <= 0 && (vp->v_mount->m_flag & M_RDONLY) == 0) { 246 error = itrunc(ip, (u_long)0, 0); 247 mode = ip->i_mode; 248 ip->i_mode = 0; 249 ip->i_rdev = 0; 250 ip->i_flag |= IUPD|ICHG; 251 ifree(ip, ip->i_number, mode); 252 #ifdef QUOTA 253 (void) chkiq(ip->i_dev, ip, ip->i_uid, 0); 254 dqrele(ip->i_dquot); 255 ip->i_dquot = NODQUOT; 256 #endif 257 } 258 IUPDAT(ip, &time, &time, 0); 259 /* 260 * If we are done with the inode, reclaim it 261 * so that it can be reused immediately. 262 */ 263 if (vp->v_count == 0 && ip->i_mode == 0) { 264 vinvalbuf(vp, 0); 265 IUNLOCK(ip); 266 ip->i_flag = 0; 267 if ((vp->v_flag & VXLOCK) == 0) 268 vgone(vp); 269 return (error); 270 } 271 IUNLOCK(ip); 272 ip->i_flag = 0; 273 return (error); 274 } 275 276 /* 277 * Reclaim an inode so that it can be used for other purposes. 278 */ 279 ufs_reclaim(vp) 280 register struct vnode *vp; 281 { 282 register struct inode *ip = VTOI(vp); 283 284 if (prtactive && vp->v_count != 0) 285 vprint("ufs_reclaim: pushing active", vp); 286 /* 287 * Remove the inode from its hash chain. 288 */ 289 remque(ip); 290 ip->i_forw = ip; 291 ip->i_back = ip; 292 /* 293 * Purge old data structures associated with the inode. 294 */ 295 cache_purge(vp); 296 if (ip->i_devvp) { 297 vrele(ip->i_devvp); 298 ip->i_devvp = 0; 299 } 300 #ifdef QUOTA 301 dqrele(ip->i_dquot); 302 ip->i_dquot = NODQUOT; 303 #endif 304 ip->i_flag = 0; 305 return (0); 306 } 307 308 /* 309 * Check accessed and update flags on an inode structure. 310 * If any is on, update the inode with the current time. 311 * If waitfor is given, then must ensure I/O order, 312 * so wait for write to complete. 313 */ 314 iupdat(ip, ta, tm, waitfor) 315 register struct inode *ip; 316 struct timeval *ta, *tm; 317 int waitfor; 318 { 319 struct buf *bp; 320 struct vnode *vp = ITOV(ip); 321 struct dinode *dp; 322 register struct fs *fs; 323 int error; 324 325 fs = ip->i_fs; 326 if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0) 327 return (0); 328 if (vp->v_mount->m_flag & M_RDONLY) 329 return (0); 330 error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)), 331 (int)fs->fs_bsize, NOCRED, &bp); 332 if (error) { 333 brelse(bp); 334 return (error); 335 } 336 if (ip->i_flag&IACC) 337 ip->i_atime = ta->tv_sec; 338 if (ip->i_flag&IUPD) 339 ip->i_mtime = tm->tv_sec; 340 if (ip->i_flag&ICHG) 341 ip->i_ctime = time.tv_sec; 342 ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD); 343 dp = bp->b_un.b_dino + itoo(fs, ip->i_number); 344 *dp = ip->i_din; 345 if (waitfor) { 346 return (bwrite(bp)); 347 } else { 348 bdwrite(bp); 349 return (0); 350 } 351 } 352 353 #define SINGLE 0 /* index of single indirect block */ 354 #define DOUBLE 1 /* index of double indirect block */ 355 #define TRIPLE 2 /* index of triple indirect block */ 356 /* 357 * Truncate the inode ip to at most length size. Free affected disk 358 * blocks -- the blocks of the file are removed in reverse order. 359 * 360 * NB: triple indirect blocks are untested. 361 */ 362 itrunc(oip, length, flags) 363 register struct inode *oip; 364 u_long length; 365 int flags; 366 { 367 register daddr_t lastblock; 368 daddr_t bn, lbn, lastiblock[NIADDR]; 369 register struct fs *fs; 370 register struct inode *ip; 371 struct buf *bp; 372 int offset, osize, size, level; 373 long count, nblocks, blocksreleased = 0; 374 register int i; 375 int aflags, error, allerror; 376 struct inode tip; 377 378 if (oip->i_size <= length) { 379 oip->i_flag |= ICHG|IUPD; 380 error = iupdat(oip, &time, &time, 1); 381 return (error); 382 } 383 /* 384 * Calculate index into inode's block list of 385 * last direct and indirect blocks (if any) 386 * which we want to keep. Lastblock is -1 when 387 * the file is truncated to 0. 388 */ 389 fs = oip->i_fs; 390 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 391 lastiblock[SINGLE] = lastblock - NDADDR; 392 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 393 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 394 nblocks = btodb(fs->fs_bsize); 395 /* 396 * Update the size of the file. If the file is not being 397 * truncated to a block boundry, the contents of the 398 * partial block following the end of the file must be 399 * zero'ed in case it ever become accessable again because 400 * of subsequent file growth. 401 */ 402 osize = oip->i_size; 403 offset = blkoff(fs, length); 404 if (offset == 0) { 405 oip->i_size = length; 406 } else { 407 lbn = lblkno(fs, length); 408 aflags = B_CLRBUF; 409 if (flags & IO_SYNC) 410 aflags |= B_SYNC; 411 if (error = balloc(oip, lbn, offset, &bp, aflags)) 412 return (error); 413 oip->i_size = length; 414 size = blksize(fs, oip, lbn); 415 bn = bp->b_blkno; 416 count = howmany(size, CLBYTES); 417 for (i = 0; i < count; i++) 418 munhash(oip->i_devvp, bn + i * CLBYTES / DEV_BSIZE); 419 bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset)); 420 brealloc(bp, size); 421 if (flags & IO_SYNC) 422 bwrite(bp); 423 else 424 bdwrite(bp); 425 } 426 /* 427 * Update file and block pointers 428 * on disk before we start freeing blocks. 429 * If we crash before free'ing blocks below, 430 * the blocks will be returned to the free list. 431 * lastiblock values are also normalized to -1 432 * for calls to indirtrunc below. 433 */ 434 tip = *oip; 435 tip.i_size = osize; 436 for (level = TRIPLE; level >= SINGLE; level--) 437 if (lastiblock[level] < 0) { 438 oip->i_ib[level] = 0; 439 lastiblock[level] = -1; 440 } 441 for (i = NDADDR - 1; i > lastblock; i--) 442 oip->i_db[i] = 0; 443 oip->i_flag |= ICHG|IUPD; 444 vinvalbuf(ITOV(oip), (length > 0)); 445 allerror = iupdat(oip, &time, &time, MNT_WAIT); 446 447 /* 448 * Indirect blocks first. 449 */ 450 ip = &tip; 451 for (level = TRIPLE; level >= SINGLE; level--) { 452 bn = ip->i_ib[level]; 453 if (bn != 0) { 454 error = indirtrunc(ip, bn, lastiblock[level], level, 455 &count); 456 if (error) 457 allerror = error; 458 blocksreleased += count; 459 if (lastiblock[level] < 0) { 460 ip->i_ib[level] = 0; 461 blkfree(ip, bn, (off_t)fs->fs_bsize); 462 blocksreleased += nblocks; 463 } 464 } 465 if (lastiblock[level] >= 0) 466 goto done; 467 } 468 469 /* 470 * All whole direct blocks or frags. 471 */ 472 for (i = NDADDR - 1; i > lastblock; i--) { 473 register off_t bsize; 474 475 bn = ip->i_db[i]; 476 if (bn == 0) 477 continue; 478 ip->i_db[i] = 0; 479 bsize = (off_t)blksize(fs, ip, i); 480 blkfree(ip, bn, bsize); 481 blocksreleased += btodb(bsize); 482 } 483 if (lastblock < 0) 484 goto done; 485 486 /* 487 * Finally, look for a change in size of the 488 * last direct block; release any frags. 489 */ 490 bn = ip->i_db[lastblock]; 491 if (bn != 0) { 492 off_t oldspace, newspace; 493 494 /* 495 * Calculate amount of space we're giving 496 * back as old block size minus new block size. 497 */ 498 oldspace = blksize(fs, ip, lastblock); 499 ip->i_size = length; 500 newspace = blksize(fs, ip, lastblock); 501 if (newspace == 0) 502 panic("itrunc: newspace"); 503 if (oldspace - newspace > 0) { 504 /* 505 * Block number of space to be free'd is 506 * the old block # plus the number of frags 507 * required for the storage we're keeping. 508 */ 509 bn += numfrags(fs, newspace); 510 blkfree(ip, bn, oldspace - newspace); 511 blocksreleased += btodb(oldspace - newspace); 512 } 513 } 514 done: 515 /* BEGIN PARANOIA */ 516 for (level = SINGLE; level <= TRIPLE; level++) 517 if (ip->i_ib[level] != oip->i_ib[level]) 518 panic("itrunc1"); 519 for (i = 0; i < NDADDR; i++) 520 if (ip->i_db[i] != oip->i_db[i]) 521 panic("itrunc2"); 522 /* END PARANOIA */ 523 oip->i_blocks -= blocksreleased; 524 if (oip->i_blocks < 0) /* sanity */ 525 oip->i_blocks = 0; 526 oip->i_flag |= ICHG; 527 #ifdef QUOTA 528 (void) chkdq(oip, -blocksreleased, 0); 529 #endif 530 return (allerror); 531 } 532 533 /* 534 * Release blocks associated with the inode ip and 535 * stored in the indirect block bn. Blocks are free'd 536 * in LIFO order up to (but not including) lastbn. If 537 * level is greater than SINGLE, the block is an indirect 538 * block and recursive calls to indirtrunc must be used to 539 * cleanse other indirect blocks. 540 * 541 * NB: triple indirect blocks are untested. 542 */ 543 indirtrunc(ip, bn, lastbn, level, countp) 544 register struct inode *ip; 545 daddr_t bn, lastbn; 546 int level; 547 long *countp; 548 { 549 register int i; 550 struct buf *bp; 551 register struct fs *fs = ip->i_fs; 552 register daddr_t *bap; 553 daddr_t *copy, nb, last; 554 long blkcount, factor; 555 int nblocks, blocksreleased = 0; 556 int error, allerror = 0; 557 558 /* 559 * Calculate index in current block of last 560 * block to be kept. -1 indicates the entire 561 * block so we need not calculate the index. 562 */ 563 factor = 1; 564 for (i = SINGLE; i < level; i++) 565 factor *= NINDIR(fs); 566 last = lastbn; 567 if (lastbn > 0) 568 last /= factor; 569 nblocks = btodb(fs->fs_bsize); 570 /* 571 * Get buffer of block pointers, zero those 572 * entries corresponding to blocks to be free'd, 573 * and update on disk copy first. 574 */ 575 error = bread(ip->i_devvp, fsbtodb(fs, bn), (int)fs->fs_bsize, 576 NOCRED, &bp); 577 if (error) { 578 brelse(bp); 579 *countp = 0; 580 return (error); 581 } 582 if ((bp->b_flags & B_CACHE) == 0) 583 reassignbuf(bp, ITOV(ip)); 584 bap = bp->b_un.b_daddr; 585 MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK); 586 bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize); 587 bzero((caddr_t)&bap[last + 1], 588 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t)); 589 if (last == -1) 590 bp->b_flags |= B_INVAL; 591 error = bwrite(bp); 592 if (error) 593 allerror = error; 594 bap = copy; 595 596 /* 597 * Recursively free totally unused blocks. 598 */ 599 for (i = NINDIR(fs) - 1; i > last; i--) { 600 nb = bap[i]; 601 if (nb == 0) 602 continue; 603 if (level > SINGLE) { 604 error = indirtrunc(ip, nb, (daddr_t)-1, level - 1, 605 &blkcount); 606 if (error) 607 allerror = error; 608 blocksreleased += blkcount; 609 } 610 blkfree(ip, nb, (off_t)fs->fs_bsize); 611 blocksreleased += nblocks; 612 } 613 614 /* 615 * Recursively free last partial block. 616 */ 617 if (level > SINGLE && lastbn >= 0) { 618 last = lastbn % factor; 619 nb = bap[i]; 620 if (nb != 0) { 621 error = indirtrunc(ip, nb, last, level - 1, &blkcount); 622 if (error) 623 allerror = error; 624 blocksreleased += blkcount; 625 } 626 } 627 FREE(copy, M_TEMP); 628 *countp = blocksreleased; 629 return (allerror); 630 } 631 632 /* 633 * Lock an inode. If its already locked, set the WANT bit and sleep. 634 */ 635 ilock(ip) 636 register struct inode *ip; 637 { 638 639 while (ip->i_flag & ILOCKED) { 640 ip->i_flag |= IWANT; 641 if (ip->i_spare0 == u.u_procp->p_pid) 642 panic("locking against myself"); 643 ip->i_spare1 = u.u_procp->p_pid; 644 (void) sleep((caddr_t)ip, PINOD); 645 } 646 ip->i_spare1 = 0; 647 ip->i_spare0 = u.u_procp->p_pid; 648 u.u_spare[0]++; 649 ip->i_flag |= ILOCKED; 650 } 651 652 /* 653 * Unlock an inode. If WANT bit is on, wakeup. 654 */ 655 iunlock(ip) 656 register struct inode *ip; 657 { 658 659 if ((ip->i_flag & ILOCKED) == 0) 660 vprint("iunlock: unlocked inode", ITOV(ip)); 661 ip->i_spare0 = 0; 662 u.u_spare[0]--; 663 ip->i_flag &= ~ILOCKED; 664 if (ip->i_flag&IWANT) { 665 ip->i_flag &= ~IWANT; 666 wakeup((caddr_t)ip); 667 } 668 } 669 670 /* 671 * Check mode permission on inode pointer. Mode is READ, WRITE or EXEC. 672 * The mode is shifted to select the owner/group/other fields. The 673 * super user is granted all permissions. 674 * 675 * NB: Called from vnode op table. It seems this could all be done 676 * using vattr's but... 677 */ 678 iaccess(ip, mode, cred) 679 register struct inode *ip; 680 register int mode; 681 struct ucred *cred; 682 { 683 register gid_t *gp; 684 int i; 685 686 /* 687 * If you're the super-user, you always get access. 688 */ 689 if (cred->cr_uid == 0) 690 return (0); 691 /* 692 * Access check is based on only one of owner, group, public. 693 * If not owner, then check group. If not a member of the 694 * group, then check public access. 695 */ 696 if (cred->cr_uid != ip->i_uid) { 697 mode >>= 3; 698 gp = cred->cr_groups; 699 for (i = 0; i < cred->cr_ngroups; i++, gp++) 700 if (ip->i_gid == *gp) 701 goto found; 702 mode >>= 3; 703 found: 704 ; 705 } 706 if ((ip->i_mode & mode) != 0) 707 return (0); 708 return (EACCES); 709 } 710