| 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 | * @(#)ffs_inode.c 7.7 (Berkeley) %G% |
| 18 | */ |
| 19 | |
| 20 | #include "param.h" |
| 21 | #include "systm.h" |
| 22 | #include "mount.h" |
| 23 | #include "user.h" |
| 24 | #include "file.h" |
| 25 | #include "buf.h" |
| 26 | #include "cmap.h" |
| 27 | #include "vnode.h" |
| 28 | #include "../ufs/inode.h" |
| 29 | #include "../ufs/fs.h" |
| 30 | #include "../ufs/ufsmount.h" |
| 31 | #ifdef QUOTA |
| 32 | #include "../ufs/quota.h" |
| 33 | #endif |
| 34 | #include "kernel.h" |
| 35 | #include "malloc.h" |
| 36 | |
| 37 | #define INOHSZ 512 |
| 38 | #if ((INOHSZ&(INOHSZ-1)) == 0) |
| 39 | #define INOHASH(dev,ino) (((dev)+(ino))&(INOHSZ-1)) |
| 40 | #else |
| 41 | #define INOHASH(dev,ino) (((unsigned)((dev)+(ino)))%INOHSZ) |
| 42 | #endif |
| 43 | |
| 44 | #define INSFREE(ip) {\ |
| 45 | if (ifreeh) { \ |
| 46 | *ifreet = (ip); \ |
| 47 | (ip)->i_freeb = ifreet; \ |
| 48 | } else { \ |
| 49 | ifreeh = (ip); \ |
| 50 | (ip)->i_freeb = &ifreeh; \ |
| 51 | } \ |
| 52 | (ip)->i_freef = NULL; \ |
| 53 | ifreet = &(ip)->i_freef; \ |
| 54 | } |
| 55 | |
| 56 | union ihead { /* inode LRU cache, Chris Maltby */ |
| 57 | union ihead *ih_head[2]; |
| 58 | struct inode *ih_chain[2]; |
| 59 | } ihead[INOHSZ]; |
| 60 | |
| 61 | struct inode *ifreeh, **ifreet, *bdevlisth; |
| 62 | |
| 63 | /* |
| 64 | * Initialize hash links for inodes |
| 65 | * and build inode free list. |
| 66 | */ |
| 67 | ihinit() |
| 68 | { |
| 69 | register int i; |
| 70 | register struct inode *ip = inode; |
| 71 | register union ihead *ih = ihead; |
| 72 | |
| 73 | for (i = INOHSZ; --i >= 0; ih++) { |
| 74 | ih->ih_head[0] = ih; |
| 75 | ih->ih_head[1] = ih; |
| 76 | } |
| 77 | ifreeh = ip; |
| 78 | ifreet = &ip->i_freef; |
| 79 | ip->i_freeb = &ifreeh; |
| 80 | ip->i_forw = ip; |
| 81 | ip->i_back = ip; |
| 82 | ITOV(ip)->v_data = (qaddr_t)ip; |
| 83 | for (i = ninode; --i > 0; ) { |
| 84 | ++ip; |
| 85 | ip->i_forw = ip; |
| 86 | ip->i_back = ip; |
| 87 | ITOV(ip)->v_data = (qaddr_t)ip; |
| 88 | *ifreet = ip; |
| 89 | ip->i_freeb = ifreet; |
| 90 | ifreet = &ip->i_freef; |
| 91 | } |
| 92 | ip->i_freef = NULL; |
| 93 | } |
| 94 | |
| 95 | /* |
| 96 | * Look up an vnode/inode by device,inumber. |
| 97 | * If it is in core (in the inode structure), |
| 98 | * honor the locking protocol. |
| 99 | * If it is not in core, read it in from the |
| 100 | * specified device. |
| 101 | * Callers must check for mount points!! |
| 102 | * In all cases, a pointer to a locked |
| 103 | * inode structure is returned. |
| 104 | */ |
| 105 | iget(xp, ino, ipp) |
| 106 | struct inode *xp; |
| 107 | ino_t ino; |
| 108 | struct inode **ipp; |
| 109 | { |
| 110 | dev_t dev = xp->i_dev; |
| 111 | struct mount *mntp = ITOV(xp)->v_mount; |
| 112 | register struct fs *fs = VFSTOUFS(mntp)->um_fs; |
| 113 | register struct inode *ip, *iq; |
| 114 | register struct vnode *vp; |
| 115 | struct inode *nip; |
| 116 | struct buf *bp; |
| 117 | struct dinode tdip, *dp; |
| 118 | union ihead *ih; |
| 119 | int error; |
| 120 | |
| 121 | loop: |
| 122 | ih = &ihead[INOHASH(dev, ino)]; |
| 123 | for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw) |
| 124 | if (ino == ip->i_number && dev == ip->i_dev) { |
| 125 | /* |
| 126 | * Following is essentially an inline expanded |
| 127 | * copy of igrab(), expanded inline for speed, |
| 128 | * and so that the test for a mounted on inode |
| 129 | * can be deferred until after we are sure that |
| 130 | * the inode isn't busy. |
| 131 | */ |
| 132 | if ((ip->i_flag&ILOCKED) != 0) { |
| 133 | ip->i_flag |= IWANT; |
| 134 | sleep((caddr_t)ip, PINOD); |
| 135 | goto loop; |
| 136 | } |
| 137 | vp = ITOV(ip); |
| 138 | if (vp->v_count == 0) { /* ino on free list */ |
| 139 | if (iq = ip->i_freef) |
| 140 | iq->i_freeb = ip->i_freeb; |
| 141 | else |
| 142 | ifreet = ip->i_freeb; |
| 143 | *ip->i_freeb = iq; |
| 144 | ip->i_freef = NULL; |
| 145 | ip->i_freeb = NULL; |
| 146 | } |
| 147 | ILOCK(ip); |
| 148 | vp->v_count++; |
| 149 | *ipp = ip; |
| 150 | return(0); |
| 151 | } |
| 152 | if (error = getnewino(dev, ino, &nip)) { |
| 153 | *ipp = 0; |
| 154 | return (error); |
| 155 | } |
| 156 | ip = nip; |
| 157 | /* |
| 158 | * Read in the disk contents for the inode. |
| 159 | */ |
| 160 | if (error = bread(VFSTOUFS(mntp)->um_devvp, fsbtodb(fs, itod(fs, ino)), |
| 161 | (int)fs->fs_bsize, &bp)) { |
| 162 | /* |
| 163 | * The inode doesn't contain anything useful, so it would |
| 164 | * be misleading to leave it on its hash chain. Iput() will |
| 165 | * take care of putting it back on the free list. We also |
| 166 | * lose its inumber, just in case. |
| 167 | */ |
| 168 | remque(ip); |
| 169 | ip->i_forw = ip; |
| 170 | ip->i_back = ip; |
| 171 | ip->i_number = 0; |
| 172 | INSFREE(ip); |
| 173 | iunlock(ip); |
| 174 | ip->i_flag = 0; |
| 175 | brelse(bp); |
| 176 | *ipp = 0; |
| 177 | return(error); |
| 178 | } |
| 179 | /* |
| 180 | * Check to see if the new inode represents a block device |
| 181 | * for which we already have an inode (either because of |
| 182 | * bdevvp() or because of a different inode representing |
| 183 | * the same block device). If such an alias exists, put the |
| 184 | * just allocated inode back on the free list, and replace |
| 185 | * the contents of the existing inode with the contents of |
| 186 | * the new inode. |
| 187 | */ |
| 188 | dp = bp->b_un.b_dino; |
| 189 | dp += itoo(fs, ino); |
| 190 | if ((dp->di_mode & IFMT) != IFBLK) { |
| 191 | ip->i_ic = dp->di_ic; |
| 192 | brelse(bp); |
| 193 | } else { |
| 194 | again: |
| 195 | for (iq = bdevlisth; iq; iq = iq->i_devlst) { |
| 196 | if (dp->di_rdev != ITOV(iq)->v_rdev) |
| 197 | continue; |
| 198 | igrab(iq); |
| 199 | if (dp->di_rdev != ITOV(iq)->v_rdev) { |
| 200 | iput(iq); |
| 201 | goto again; |
| 202 | } |
| 203 | /* |
| 204 | * Discard unneeded inode. |
| 205 | */ |
| 206 | remque(ip); |
| 207 | ip->i_forw = ip; |
| 208 | ip->i_back = ip; |
| 209 | ip->i_number = 0; |
| 210 | INSFREE(ip); |
| 211 | iunlock(ip); |
| 212 | ip->i_flag = 0; |
| 213 | /* |
| 214 | * Reinitialize aliased inode. |
| 215 | * We must release the buffer that we just read |
| 216 | * before doing the iupdat() to avoid a possible |
| 217 | * deadlock with updating an inode in the same |
| 218 | * disk block. |
| 219 | */ |
| 220 | ip = iq; |
| 221 | vp = ITOV(iq); |
| 222 | tdip.di_ic = dp->di_ic; |
| 223 | brelse(bp); |
| 224 | error = iupdat(ip, &time, &time, 1); |
| 225 | ip->i_ic = tdip.di_ic; |
| 226 | remque(ip); |
| 227 | insque(ip, ih); |
| 228 | ip->i_dev = dev; |
| 229 | ip->i_number = ino; |
| 230 | if (ip->i_devvp) { |
| 231 | vrele(ip->i_devvp); |
| 232 | ip->i_devvp = 0; |
| 233 | } |
| 234 | cache_purge(vp); |
| 235 | break; |
| 236 | } |
| 237 | if (iq == 0) { |
| 238 | ip->i_ic = dp->di_ic; |
| 239 | brelse(bp); |
| 240 | ip->i_devlst = bdevlisth; |
| 241 | bdevlisth = ip; |
| 242 | } |
| 243 | } |
| 244 | /* |
| 245 | * Finish inode initialization. |
| 246 | */ |
| 247 | ip->i_fs = fs; |
| 248 | ip->i_devvp = VFSTOUFS(mntp)->um_devvp; |
| 249 | ip->i_devvp->v_count++; |
| 250 | /* |
| 251 | * Initialize the associated vnode |
| 252 | */ |
| 253 | vp = ITOV(ip); |
| 254 | vinit(vp, mntp, IFTOVT(ip->i_mode), &ufs_vnodeops); |
| 255 | if (vp->v_type == VCHR || vp->v_type == VBLK) { |
| 256 | vp->v_rdev = ip->i_rdev; |
| 257 | vp->v_op = &blk_vnodeops; |
| 258 | } |
| 259 | if (ino == ROOTINO) |
| 260 | vp->v_flag |= VROOT; |
| 261 | #ifdef QUOTA |
| 262 | if (ip->i_mode != 0) |
| 263 | ip->i_dquot = inoquota(ip); |
| 264 | #endif |
| 265 | *ipp = ip; |
| 266 | return (0); |
| 267 | } |
| 268 | |
| 269 | /* |
| 270 | * Allocate a new inode. |
| 271 | * |
| 272 | * Put it onto its hash chain and lock it so that other requests for |
| 273 | * this inode will block if they arrive while we are sleeping waiting |
| 274 | * for old data structures to be purged or for the contents of the disk |
| 275 | * portion of this inode to be read. |
| 276 | */ |
| 277 | getnewino(dev, ino, ipp) |
| 278 | dev_t dev; |
| 279 | ino_t ino; |
| 280 | struct inode **ipp; |
| 281 | { |
| 282 | union ihead *ih; |
| 283 | register struct inode *ip, *iq; |
| 284 | register struct vnode *vp; |
| 285 | |
| 286 | /* |
| 287 | * Remove the next inode from the free list. |
| 288 | */ |
| 289 | if ((ip = ifreeh) == NULL) { |
| 290 | tablefull("inode"); |
| 291 | *ipp = 0; |
| 292 | return(ENFILE); |
| 293 | } |
| 294 | vp = ITOV(ip); |
| 295 | if (vp->v_count) |
| 296 | panic("free inode isn't"); |
| 297 | if (iq = ip->i_freef) |
| 298 | iq->i_freeb = &ifreeh; |
| 299 | ifreeh = iq; |
| 300 | ip->i_freef = NULL; |
| 301 | ip->i_freeb = NULL; |
| 302 | /* |
| 303 | * Now to take inode off the hash chain it was on |
| 304 | * (initially, or after an iflush, it is on a "hash chain" |
| 305 | * consisting entirely of itself, and pointed to by no-one) |
| 306 | * and put it on the chain for its new (ino, dev) pair. |
| 307 | */ |
| 308 | remque(ip); |
| 309 | ip->i_dev = dev; |
| 310 | ip->i_number = ino; |
| 311 | if (dev != NODEV) { |
| 312 | ih = &ihead[INOHASH(dev, ino)]; |
| 313 | insque(ip, ih); |
| 314 | } |
| 315 | ip->i_flag = 0; |
| 316 | ILOCK(ip); |
| 317 | ip->i_lastr = 0; |
| 318 | #endif SECSIZE |
| 319 | /* |
| 320 | * Purge old data structures associated with the inode. |
| 321 | */ |
| 322 | cache_purge(vp); |
| 323 | if (ip->i_devvp) { |
| 324 | vrele(ip->i_devvp); |
| 325 | ip->i_devvp = 0; |
| 326 | } |
| 327 | #ifdef QUOTA |
| 328 | dqrele(ip->i_dquot); |
| 329 | ip->i_dquot = NODQUOT; |
| 330 | #endif |
| 331 | if (vp->v_type == VBLK) { |
| 332 | if (bdevlisth == ip) { |
| 333 | bdevlisth = ip->i_devlst; |
| 334 | } else { |
| 335 | for (iq = bdevlisth; iq; iq = iq->i_devlst) { |
| 336 | if (iq->i_devlst != ip) |
| 337 | continue; |
| 338 | iq->i_devlst = ip->i_devlst; |
| 339 | break; |
| 340 | } |
| 341 | if (iq == NULL) |
| 342 | panic("missing bdev"); |
| 343 | } |
| 344 | } |
| 345 | *ipp = ip; |
| 346 | return (0); |
| 347 | } |
| 348 | |
| 349 | /* |
| 350 | * Convert a pointer to an inode into a reference to an inode. |
| 351 | * |
| 352 | * This is basically the internal piece of iget (after the |
| 353 | * inode pointer is located) but without the test for mounted |
| 354 | * filesystems. It is caller's responsibility to check that |
| 355 | * the inode pointer is valid. |
| 356 | */ |
| 357 | igrab(ip) |
| 358 | register struct inode *ip; |
| 359 | { |
| 360 | register struct vnode *vp = ITOV(ip); |
| 361 | |
| 362 | while ((ip->i_flag&ILOCKED) != 0) { |
| 363 | ip->i_flag |= IWANT; |
| 364 | sleep((caddr_t)ip, PINOD); |
| 365 | } |
| 366 | if (vp->v_count == 0) { /* ino on free list */ |
| 367 | register struct inode *iq; |
| 368 | |
| 369 | if (iq = ip->i_freef) |
| 370 | iq->i_freeb = ip->i_freeb; |
| 371 | else |
| 372 | ifreet = ip->i_freeb; |
| 373 | *ip->i_freeb = iq; |
| 374 | ip->i_freef = NULL; |
| 375 | ip->i_freeb = NULL; |
| 376 | } |
| 377 | vp->v_count++; |
| 378 | ILOCK(ip); |
| 379 | } |
| 380 | |
| 381 | /* |
| 382 | * Create a vnode for a block device. |
| 383 | * Used for root filesystem, argdev, and swap areas. |
| 384 | */ |
| 385 | bdevvp(dev, vpp) |
| 386 | dev_t dev; |
| 387 | struct vnode **vpp; |
| 388 | { |
| 389 | register struct inode *ip; |
| 390 | register struct vnode *vp; |
| 391 | struct inode *nip; |
| 392 | int error; |
| 393 | |
| 394 | /* |
| 395 | * Check for the existence of an existing vnode. |
| 396 | */ |
| 397 | again: |
| 398 | for (ip = bdevlisth; ip; ip = ip->i_devlst) { |
| 399 | vp = ITOV(ip); |
| 400 | if (dev != vp->v_rdev) |
| 401 | continue; |
| 402 | igrab(ip); |
| 403 | if (dev != vp->v_rdev) { |
| 404 | iput(ip); |
| 405 | goto again; |
| 406 | } |
| 407 | IUNLOCK(ip); |
| 408 | *vpp = vp; |
| 409 | return (0); |
| 410 | } |
| 411 | if (error = getnewino(NODEV, (ino_t)0, &nip)) { |
| 412 | *vpp = 0; |
| 413 | return (error); |
| 414 | } |
| 415 | ip = nip; |
| 416 | ip->i_fs = 0; |
| 417 | ip->i_devlst = bdevlisth; |
| 418 | bdevlisth = ip; |
| 419 | vp = ITOV(ip); |
| 420 | vinit(vp, 0, VBLK, &blk_vnodeops); |
| 421 | vp->v_rdev = dev; |
| 422 | IUNLOCK(ip); |
| 423 | *vpp = vp; |
| 424 | return (0); |
| 425 | } |
| 426 | |
| 427 | /* |
| 428 | * Decrement reference count of |
| 429 | * an inode structure. |
| 430 | * On the last reference, |
| 431 | * write the inode out and if necessary, |
| 432 | * truncate and deallocate the file. |
| 433 | */ |
| 434 | iput(ip) |
| 435 | register struct inode *ip; |
| 436 | { |
| 437 | |
| 438 | if ((ip->i_flag & ILOCKED) == 0) |
| 439 | panic("iput"); |
| 440 | IUNLOCK(ip); |
| 441 | vrele(ITOV(ip)); |
| 442 | } |
| 443 | |
| 444 | |
| 445 | ufs_inactive(vp) |
| 446 | struct vnode *vp; |
| 447 | { |
| 448 | register struct inode *ip = VTOI(vp); |
| 449 | int mode, error; |
| 450 | |
| 451 | if (ITOV(ip)->v_count != 0) |
| 452 | panic("ufs_inactive: not inactive"); |
| 453 | ILOCK(ip); |
| 454 | if (ip->i_nlink <= 0 && (ITOV(ip)->v_mount->m_flag&M_RDONLY) == 0) { |
| 455 | error = itrunc(ip, (u_long)0); |
| 456 | mode = ip->i_mode; |
| 457 | ip->i_mode = 0; |
| 458 | ip->i_rdev = 0; |
| 459 | ip->i_flag |= IUPD|ICHG; |
| 460 | ifree(ip, ip->i_number, mode); |
| 461 | #ifdef QUOTA |
| 462 | (void) chkiq(ip->i_dev, ip, ip->i_uid, 0); |
| 463 | dqrele(ip->i_dquot); |
| 464 | ip->i_dquot = NODQUOT; |
| 465 | #endif |
| 466 | } |
| 467 | IUPDAT(ip, &time, &time, 0); |
| 468 | IUNLOCK(ip); |
| 469 | ip->i_flag = 0; |
| 470 | /* |
| 471 | * Put the inode on the end of the free list. |
| 472 | * Possibly in some cases it would be better to |
| 473 | * put the inode at the head of the free list, |
| 474 | * (eg: where i_mode == 0 || i_number == 0). |
| 475 | */ |
| 476 | INSFREE(ip); |
| 477 | return (error); |
| 478 | } |
| 479 | |
| 480 | /* |
| 481 | * Check accessed and update flags on |
| 482 | * an inode structure. |
| 483 | * If any is on, update the inode |
| 484 | * with the current time. |
| 485 | * If waitfor is given, then must insure |
| 486 | * i/o order so wait for write to complete. |
| 487 | */ |
| 488 | iupdat(ip, ta, tm, waitfor) |
| 489 | register struct inode *ip; |
| 490 | struct timeval *ta, *tm; |
| 491 | int waitfor; |
| 492 | { |
| 493 | struct buf *bp; |
| 494 | struct vnode *vp = ITOV(ip); |
| 495 | struct dinode *dp; |
| 496 | register struct fs *fs; |
| 497 | |
| 498 | fs = ip->i_fs; |
| 499 | if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0) |
| 500 | return (0); |
| 501 | if (vp->v_mount->m_flag & M_RDONLY) |
| 502 | return (0); |
| 503 | error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)), |
| 504 | (int)fs->fs_bsize, &bp); |
| 505 | if (error) { |
| 506 | brelse(bp); |
| 507 | return (error); |
| 508 | } |
| 509 | if (ip->i_flag&IACC) |
| 510 | ip->i_atime = ta->tv_sec; |
| 511 | if (ip->i_flag&IUPD) |
| 512 | ip->i_mtime = tm->tv_sec; |
| 513 | if (ip->i_flag&ICHG) |
| 514 | ip->i_ctime = time.tv_sec; |
| 515 | ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD); |
| 516 | dp = bp->b_un.b_dino + itoo(fs, ip->i_number); |
| 517 | dp->di_ic = ip->i_ic; |
| 518 | if (waitfor) { |
| 519 | return (bwrite(bp)); |
| 520 | } else { |
| 521 | bdwrite(bp); |
| 522 | return (0); |
| 523 | } |
| 524 | } |
| 525 | |
| 526 | #define SINGLE 0 /* index of single indirect block */ |
| 527 | #define DOUBLE 1 /* index of double indirect block */ |
| 528 | #define TRIPLE 2 /* index of triple indirect block */ |
| 529 | /* |
| 530 | * Truncate the inode ip to at most |
| 531 | * length size. Free affected disk |
| 532 | * blocks -- the blocks of the file |
| 533 | * are removed in reverse order. |
| 534 | * |
| 535 | * NB: triple indirect blocks are untested. |
| 536 | */ |
| 537 | itrunc(oip, length) |
| 538 | register struct inode *oip; |
| 539 | u_long length; |
| 540 | { |
| 541 | register daddr_t lastblock; |
| 542 | daddr_t bn, lbn, lastiblock[NIADDR]; |
| 543 | register struct fs *fs; |
| 544 | register struct inode *ip; |
| 545 | struct buf *bp; |
| 546 | int offset, osize, size, level; |
| 547 | long count, nblocks, blocksreleased = 0; |
| 548 | register int i; |
| 549 | int error, allerror = 0; |
| 550 | struct inode tip; |
| 551 | |
| 552 | if (oip->i_size <= length) { |
| 553 | oip->i_flag |= ICHG|IUPD; |
| 554 | error = iupdat(oip, &time, &time, 1); |
| 555 | return (error); |
| 556 | } |
| 557 | /* |
| 558 | * Calculate index into inode's block list of |
| 559 | * last direct and indirect blocks (if any) |
| 560 | * which we want to keep. Lastblock is -1 when |
| 561 | * the file is truncated to 0. |
| 562 | */ |
| 563 | fs = oip->i_fs; |
| 564 | lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; |
| 565 | lastiblock[SINGLE] = lastblock - NDADDR; |
| 566 | lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); |
| 567 | lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); |
| 568 | nblocks = btodb(fs->fs_bsize); |
| 569 | /* |
| 570 | * Update the size of the file. If the file is not being |
| 571 | * truncated to a block boundry, the contents of the |
| 572 | * partial block following the end of the file must be |
| 573 | * zero'ed in case it ever become accessable again because |
| 574 | * of subsequent file growth. |
| 575 | */ |
| 576 | osize = oip->i_size; |
| 577 | offset = blkoff(fs, length); |
| 578 | if (offset == 0) { |
| 579 | oip->i_size = length; |
| 580 | } else { |
| 581 | lbn = lblkno(fs, length); |
| 582 | error = balloc(oip, lbn, offset, &bn, B_CLRBUF); |
| 583 | if (error) |
| 584 | return (error); |
| 585 | if ((long)bn < 0) |
| 586 | panic("itrunc: hole"); |
| 587 | oip->i_size = length; |
| 588 | size = blksize(fs, oip, lbn); |
| 589 | count = howmany(size, CLBYTES); |
| 590 | munhash(oip->i_devvp, bn + i * CLBYTES / DEV_BSIZE); |
| 591 | error = bread(oip->i_devvp, bn, size, &bp); |
| 592 | if (error) { |
| 593 | oip->i_size = osize; |
| 594 | brelse(bp); |
| 595 | return (error); |
| 596 | } |
| 597 | bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset)); |
| 598 | bdwrite(bp); |
| 599 | } |
| 600 | /* |
| 601 | * Update file and block pointers |
| 602 | * on disk before we start freeing blocks. |
| 603 | * If we crash before free'ing blocks below, |
| 604 | * the blocks will be returned to the free list. |
| 605 | * lastiblock values are also normalized to -1 |
| 606 | * for calls to indirtrunc below. |
| 607 | */ |
| 608 | tip = *oip; |
| 609 | tip.i_size = osize; |
| 610 | for (level = TRIPLE; level >= SINGLE; level--) |
| 611 | if (lastiblock[level] < 0) { |
| 612 | oip->i_ib[level] = 0; |
| 613 | lastiblock[level] = -1; |
| 614 | } |
| 615 | for (i = NDADDR - 1; i > lastblock; i--) |
| 616 | oip->i_db[i] = 0; |
| 617 | oip->i_flag |= ICHG|IUPD; |
| 618 | allerror = syncip(oip); |
| 619 | |
| 620 | /* |
| 621 | * Indirect blocks first. |
| 622 | */ |
| 623 | ip = &tip; |
| 624 | for (level = TRIPLE; level >= SINGLE; level--) { |
| 625 | bn = ip->i_ib[level]; |
| 626 | if (bn != 0) { |
| 627 | error = indirtrunc(ip, bn, lastiblock[level], level, |
| 628 | &count); |
| 629 | if (error) |
| 630 | allerror = error; |
| 631 | blocksreleased += count; |
| 632 | if (lastiblock[level] < 0) { |
| 633 | ip->i_ib[level] = 0; |
| 634 | blkfree(ip, bn, (off_t)fs->fs_bsize); |
| 635 | blocksreleased += nblocks; |
| 636 | } |
| 637 | } |
| 638 | if (lastiblock[level] >= 0) |
| 639 | goto done; |
| 640 | } |
| 641 | |
| 642 | /* |
| 643 | * All whole direct blocks or frags. |
| 644 | */ |
| 645 | for (i = NDADDR - 1; i > lastblock; i--) { |
| 646 | register off_t bsize; |
| 647 | |
| 648 | bn = ip->i_db[i]; |
| 649 | if (bn == 0) |
| 650 | continue; |
| 651 | ip->i_db[i] = 0; |
| 652 | bsize = (off_t)blksize(fs, ip, i); |
| 653 | blkfree(ip, bn, bsize); |
| 654 | blocksreleased += btodb(bsize); |
| 655 | } |
| 656 | if (lastblock < 0) |
| 657 | goto done; |
| 658 | |
| 659 | /* |
| 660 | * Finally, look for a change in size of the |
| 661 | * last direct block; release any frags. |
| 662 | */ |
| 663 | bn = ip->i_db[lastblock]; |
| 664 | if (bn != 0) { |
| 665 | off_t oldspace, newspace; |
| 666 | |
| 667 | /* |
| 668 | * Calculate amount of space we're giving |
| 669 | * back as old block size minus new block size. |
| 670 | */ |
| 671 | oldspace = blksize(fs, ip, lastblock); |
| 672 | ip->i_size = length; |
| 673 | newspace = blksize(fs, ip, lastblock); |
| 674 | if (newspace == 0) |
| 675 | panic("itrunc: newspace"); |
| 676 | if (oldspace - newspace > 0) { |
| 677 | /* |
| 678 | * Block number of space to be free'd is |
| 679 | * the old block # plus the number of frags |
| 680 | * required for the storage we're keeping. |
| 681 | */ |
| 682 | bn += numfrags(fs, newspace); |
| 683 | blkfree(ip, bn, oldspace - newspace); |
| 684 | blocksreleased += btodb(oldspace - newspace); |
| 685 | } |
| 686 | } |
| 687 | done: |
| 688 | /* BEGIN PARANOIA */ |
| 689 | for (level = SINGLE; level <= TRIPLE; level++) |
| 690 | if (ip->i_ib[level] != oip->i_ib[level]) |
| 691 | panic("itrunc1"); |
| 692 | for (i = 0; i < NDADDR; i++) |
| 693 | if (ip->i_db[i] != oip->i_db[i]) |
| 694 | panic("itrunc2"); |
| 695 | /* END PARANOIA */ |
| 696 | oip->i_blocks -= blocksreleased; |
| 697 | if (oip->i_blocks < 0) /* sanity */ |
| 698 | oip->i_blocks = 0; |
| 699 | oip->i_flag |= ICHG; |
| 700 | #ifdef QUOTA |
| 701 | (void) chkdq(oip, -blocksreleased, 0); |
| 702 | #endif |
| 703 | return (allerror); |
| 704 | } |
| 705 | |
| 706 | /* |
| 707 | * Release blocks associated with the inode ip and |
| 708 | * stored in the indirect block bn. Blocks are free'd |
| 709 | * in LIFO order up to (but not including) lastbn. If |
| 710 | * level is greater than SINGLE, the block is an indirect |
| 711 | * block and recursive calls to indirtrunc must be used to |
| 712 | * cleanse other indirect blocks. |
| 713 | * |
| 714 | * NB: triple indirect blocks are untested. |
| 715 | */ |
| 716 | indirtrunc(ip, bn, lastbn, level, countp) |
| 717 | register struct inode *ip; |
| 718 | daddr_t bn, lastbn; |
| 719 | int level; |
| 720 | long *countp; |
| 721 | { |
| 722 | register int i; |
| 723 | struct buf *bp; |
| 724 | register struct fs *fs = ip->i_fs; |
| 725 | register daddr_t *bap; |
| 726 | daddr_t *copy, nb, last; |
| 727 | long blkcount, factor; |
| 728 | int nblocks, blocksreleased = 0; |
| 729 | int error, allerror = 0; |
| 730 | |
| 731 | /* |
| 732 | * Calculate index in current block of last |
| 733 | * block to be kept. -1 indicates the entire |
| 734 | * block so we need not calculate the index. |
| 735 | */ |
| 736 | factor = 1; |
| 737 | for (i = SINGLE; i < level; i++) |
| 738 | factor *= NINDIR(fs); |
| 739 | last = lastbn; |
| 740 | if (lastbn > 0) |
| 741 | last /= factor; |
| 742 | nblocks = btodb(fs->fs_bsize); |
| 743 | /* |
| 744 | * Get buffer of block pointers, zero those |
| 745 | * entries corresponding to blocks to be free'd, |
| 746 | * and update on disk copy first. |
| 747 | */ |
| 748 | #ifdef SECSIZE |
| 749 | bp = bread(ip->i_dev, fsbtodb(fs, bn), (int)fs->fs_bsize, |
| 750 | fs->fs_dbsize); |
| 751 | #else SECSIZE |
| 752 | error = bread(ip->i_devvp, fsbtodb(fs, bn), (int)fs->fs_bsize, &bp); |
| 753 | if (error) { |
| 754 | brelse(bp); |
| 755 | *countp = 0; |
| 756 | return (error); |
| 757 | } |
| 758 | bap = bp->b_un.b_daddr; |
| 759 | MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK); |
| 760 | bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize); |
| 761 | bzero((caddr_t)&bap[last + 1], |
| 762 | (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t)); |
| 763 | error = bwrite(bp); |
| 764 | if (error) |
| 765 | allerror = error; |
| 766 | bap = copy; |
| 767 | |
| 768 | /* |
| 769 | * Recursively free totally unused blocks. |
| 770 | */ |
| 771 | for (i = NINDIR(fs) - 1; i > last; i--) { |
| 772 | nb = bap[i]; |
| 773 | if (nb == 0) |
| 774 | continue; |
| 775 | if (level > SINGLE) { |
| 776 | error = indirtrunc(ip, nb, (daddr_t)-1, level - 1, |
| 777 | &blkcount); |
| 778 | if (error) |
| 779 | allerror = error; |
| 780 | blocksreleased += blkcount; |
| 781 | } |
| 782 | blkfree(ip, nb, (off_t)fs->fs_bsize); |
| 783 | blocksreleased += nblocks; |
| 784 | } |
| 785 | |
| 786 | /* |
| 787 | * Recursively free last partial block. |
| 788 | */ |
| 789 | if (level > SINGLE && lastbn >= 0) { |
| 790 | last = lastbn % factor; |
| 791 | nb = bap[i]; |
| 792 | if (nb != 0) { |
| 793 | error = indirtrunc(ip, nb, last, level - 1, &blkcount); |
| 794 | if (error) |
| 795 | allerror = error; |
| 796 | blocksreleased += blkcount; |
| 797 | } |
| 798 | } |
| 799 | FREE(copy, M_TEMP); |
| 800 | *countp = blocksreleased; |
| 801 | return (allerror); |
| 802 | } |
| 803 | |
| 804 | /* |
| 805 | * Remove any inodes in the inode cache belonging to dev. |
| 806 | * |
| 807 | * There should not be any active ones, return error if any are found |
| 808 | * (nb: this is a user error, not a system err). |
| 809 | */ |
| 810 | #ifdef QUOTA |
| 811 | iflush(dev, iq) |
| 812 | dev_t dev; |
| 813 | struct inode *iq; |
| 814 | #else |
| 815 | iflush(dev) |
| 816 | dev_t dev; |
| 817 | #endif |
| 818 | { |
| 819 | register struct inode *ip; |
| 820 | |
| 821 | for (ip = inode; ip < inodeNINODE; ip++) { |
| 822 | #ifdef QUOTA |
| 823 | if (ip != iq && ip->i_dev == dev) |
| 824 | #else |
| 825 | if (ip->i_dev == dev) |
| 826 | #endif |
| 827 | if (ITOV(ip)->v_count) |
| 828 | return (EBUSY); |
| 829 | else { |
| 830 | remque(ip); |
| 831 | ip->i_forw = ip; |
| 832 | ip->i_back = ip; |
| 833 | /* |
| 834 | * as v_count == 0, the inode was on the free |
| 835 | * list already, just leave it there, it will |
| 836 | * fall off the bottom eventually. We could |
| 837 | * perhaps move it to the head of the free |
| 838 | * list, but as umounts are done so |
| 839 | * infrequently, we would gain very little, |
| 840 | * while making the code bigger. |
| 841 | */ |
| 842 | #ifdef QUOTA |
| 843 | dqrele(ip->i_dquot); |
| 844 | ip->i_dquot = NODQUOT; |
| 845 | #endif |
| 846 | if (ip->i_devvp) { |
| 847 | vrele(ip->i_devvp); |
| 848 | ip->i_devvp = 0; |
| 849 | } |
| 850 | } |
| 851 | } |
| 852 | return (0); |
| 853 | } |
| 854 | |
| 855 | /* |
| 856 | * Lock an inode. If its already locked, set the WANT bit and sleep. |
| 857 | */ |
| 858 | ilock(ip) |
| 859 | register struct inode *ip; |
| 860 | { |
| 861 | |
| 862 | while (ip->i_flag & ILOCKED) { |
| 863 | ip->i_flag |= IWANT; |
| 864 | (void) sleep((caddr_t)ip, PINOD); |
| 865 | } |
| 866 | ip->i_flag |= ILOCKED; |
| 867 | } |
| 868 | |
| 869 | /* |
| 870 | * Unlock an inode. If WANT bit is on, wakeup. |
| 871 | */ |
| 872 | iunlock(ip) |
| 873 | register struct inode *ip; |
| 874 | { |
| 875 | |
| 876 | if ((ip->i_flag & ILOCKED) == 0) |
| 877 | printf("unlocking unlocked inode %d on dev 0x%x\n", |
| 878 | ip->i_number, ip->i_dev); |
| 879 | ip->i_flag &= ~ILOCKED; |
| 880 | if (ip->i_flag&IWANT) { |
| 881 | ip->i_flag &= ~IWANT; |
| 882 | wakeup((caddr_t)ip); |
| 883 | } |
| 884 | } |
| 885 | |
| 886 | /* |
| 887 | * Check mode permission on inode pointer. Mode is READ, WRITE or EXEC. |
| 888 | * The mode is shifted to select the owner/group/other fields. The |
| 889 | * super user is granted all permissions. |
| 890 | * |
| 891 | * NB: Called from vnode op table. It seems this could all be done |
| 892 | * using vattr's but... |
| 893 | */ |
| 894 | iaccess(ip, mode, cred) |
| 895 | register struct inode *ip; |
| 896 | register int mode; |
| 897 | struct ucred *cred; |
| 898 | { |
| 899 | register gid_t *gp; |
| 900 | register struct vnode *vp = ITOV(ip); |
| 901 | int i; |
| 902 | |
| 903 | /* |
| 904 | * If you're the super-user, |
| 905 | * you always get access. |
| 906 | */ |
| 907 | if (cred->cr_uid == 0) |
| 908 | return (0); |
| 909 | /* |
| 910 | * Access check is based on only one of owner, group, public. |
| 911 | * If not owner, then check group. If not a member of the |
| 912 | * group, then check public access. |
| 913 | */ |
| 914 | if (cred->cr_uid != ip->i_uid) { |
| 915 | mode >>= 3; |
| 916 | gp = cred->cr_groups; |
| 917 | for (i = 0; i < cred->cr_ngroups; i++, gp++) |
| 918 | if (ip->i_gid == *gp) |
| 919 | goto found; |
| 920 | mode >>= 3; |
| 921 | found: |
| 922 | ; |
| 923 | } |
| 924 | if ((ip->i_mode & mode) != 0) |
| 925 | return (0); |
| 926 | return (EACCES); |
| 927 | } |