1 // File system implementation. Five layers: 2 // + Blocks: allocator for raw disk blocks. 3 // + Log: crash recovery for multi-step updates. 4 // + Files: inode allocator, reading, writing, metadata. 5 // + Directories: inode with special contents (list of other inodes!) 6 // + Names: paths like /usr/rtm/xv6/fs.c for convenient naming. 7 // 8 // This file contains the low-level file system manipulation 9 // routines. The (higher-level) system call implementations 10 // are in sysfile.c. 11 12 #include "types.h" 13 #include "defs.h" 14 #include "param.h" 15 #include "stat.h" 16 #include "mmu.h" 17 #include "proc.h" 18 #include "spinlock.h" 19 #include "sleeplock.h" 20 #include "fs.h" 21 #include "buf.h" 22 #include "file.h" 23 24 #define min(a, b) ((a) < (b) ? (a) : (b)) 25 static void itrunc(struct inode*); 26 // there should be one superblock per disk device, but we run with 27 // only one device 28 struct superblock sb; 29 30 // Read the super block. 31 void 32 readsb(int dev, struct superblock *sb) 33 { 34 struct buf *bp; 35 36 bp = bread(dev, 1); 37 memmove(sb, bp->data, sizeof(*sb)); 38 brelse(bp); 39 } 40 41 // Zero a block. 42 static void 43 bzero(int dev, int bno) 44 { 45 struct buf *bp; 46 47 bp = bread(dev, bno); 48 memset(bp->data, 0, BSIZE); 49 log_write(bp); 50 brelse(bp); 51 } 52 53 // Blocks. 54 55 // Allocate a zeroed disk block. 56 static uint 57 balloc(uint dev) 58 { 59 int b, bi, m; 60 struct buf *bp; 61 62 bp = 0; 63 for(b = 0; b < sb.size; b += BPB){ 64 bp = bread(dev, BBLOCK(b, sb)); 65 for(bi = 0; bi < BPB && b + bi < sb.size; bi++){ 66 m = 1 << (bi % 8); 67 if((bp->data[bi/8] & m) == 0){ // Is block free? 68 bp->data[bi/8] |= m; // Mark block in use. 69 log_write(bp); 70 brelse(bp); 71 bzero(dev, b + bi); 72 return b + bi; 73 } 74 } 75 brelse(bp); 76 } 77 panic("balloc: out of blocks"); 78 } 79 80 // Free a disk block. 81 static void 82 bfree(int dev, uint b) 83 { 84 struct buf *bp; 85 int bi, m; 86 87 readsb(dev, &sb); 88 bp = bread(dev, BBLOCK(b, sb)); 89 bi = b % BPB; 90 m = 1 << (bi % 8); 91 if((bp->data[bi/8] & m) == 0) 92 panic("freeing free block"); 93 bp->data[bi/8] &= ~m; 94 log_write(bp); 95 brelse(bp); 96 } 97 98 // Inodes. 99 // 100 // An inode describes a single unnamed file. 101 // The inode disk structure holds metadata: the file's type, 102 // its size, the number of links referring to it, and the 103 // list of blocks holding the file's content. 104 // 105 // The inodes are laid out sequentially on disk at 106 // sb.startinode. Each inode has a number, indicating its 107 // position on the disk. 108 // 109 // The kernel keeps a cache of in-use inodes in memory 110 // to provide a place for synchronizing access 111 // to inodes used by multiple processes. The cached 112 // inodes include book-keeping information that is 113 // not stored on disk: ip->ref and ip->valid. 114 // 115 // An inode and its in-memory representation go through a 116 // sequence of states before they can be used by the 117 // rest of the file system code. 118 // 119 // * Allocation: an inode is allocated if its type (on disk) 120 // is non-zero. ialloc() allocates, and iput() frees if 121 // the reference and link counts have fallen to zero. 122 // 123 // * Referencing in cache: an entry in the inode cache 124 // is free if ip->ref is zero. Otherwise ip->ref tracks 125 // the number of in-memory pointers to the entry (open 126 // files and current directories). iget() finds or 127 // creates a cache entry and increments its ref; iput() 128 // decrements ref. 129 // 130 // * Valid: the information (type, size, &c) in an inode 131 // cache entry is only correct when ip->valid is 1. 132 // ilock() reads the inode from 133 // the disk and sets ip->valid, while iput() clears 134 // ip->valid if ip->ref has fallen to zero. 135 // 136 // * Locked: file system code may only examine and modify 137 // the information in an inode and its content if it 138 // has first locked the inode. 139 // 140 // Thus a typical sequence is: 141 // ip = iget(dev, inum) 142 // ilock(ip) 143 // ... examine and modify ip->xxx ... 144 // iunlock(ip) 145 // iput(ip) 146 // 147 // ilock() is separate from iget() so that system calls can 148 // get a long-term reference to an inode (as for an open file) 149 // and only lock it for short periods (e.g., in read()). 150 // The separation also helps avoid deadlock and races during 151 // pathname lookup. iget() increments ip->ref so that the inode 152 // stays cached and pointers to it remain valid. 153 // 154 // Many internal file system functions expect the caller to 155 // have locked the inodes involved; this lets callers create 156 // multi-step atomic operations. 157 // 158 // The icache.lock spin-lock defends ip->ref, ip->dev, and ip->inum. 159 // Since ip->ref indicates whether an icache entry is free, the 160 // icache.lock defends icache allocation. icache.lock also defends 161 // all fields of an unallocated icache entry, during allocation. 162 // 163 // An ip->lock sleep-lock defends all ip-> fields other than ref, 164 // dev, and inum. One must hold ip->lock in order to 165 // read or write that inode's ip->valid, ip->size, ip->type, &c. 166 167 struct { 168 struct spinlock lock; 169 struct inode inode[NINODE]; 170 } icache; 171 172 void 173 iinit(int dev) 174 { 175 int i = 0; 176 177 initlock(&icache.lock, "icache"); 178 for(i = 0; i < NINODE; i++) { 179 initsleeplock(&icache.inode[i].lock, "inode"); 180 } 181 182 readsb(dev, &sb); 183 cprintf("sb: size %d nblocks %d ninodes %d nlog %d logstart %d\ 184 inodestart %d bmap start %d\n", sb.size, sb.nblocks, 185 sb.ninodes, sb.nlog, sb.logstart, sb.inodestart, 186 sb.bmapstart); 187 } 188 189 static struct inode* iget(uint dev, uint inum); 190 191 //PAGEBREAK! 192 // Allocate a new inode with the given type on device dev. 193 // A free inode has a type of zero. 194 struct inode* 195 ialloc(uint dev, short type) 196 { 197 int inum; 198 struct buf *bp; 199 struct dinode *dip; 200 201 for(inum = 1; inum < sb.ninodes; inum++){ 202 bp = bread(dev, IBLOCK(inum, sb)); 203 dip = (struct dinode*)bp->data + inum%IPB; 204 if(dip->type == 0){ // a free inode 205 memset(dip, 0, sizeof(*dip)); 206 dip->type = type; 207 log_write(bp); // mark it allocated on the disk 208 brelse(bp); 209 return iget(dev, inum); 210 } 211 brelse(bp); 212 } 213 panic("ialloc: no inodes"); 214 } 215 216 // Copy a modified in-memory inode to disk. 217 // Caller must hold ip->lock. 218 void 219 iupdate(struct inode *ip) 220 { 221 struct buf *bp; 222 struct dinode *dip; 223 224 bp = bread(ip->dev, IBLOCK(ip->inum, sb)); 225 dip = (struct dinode*)bp->data + ip->inum%IPB; 226 dip->type = ip->type; 227 dip->major = ip->major; 228 dip->minor = ip->minor; 229 dip->nlink = ip->nlink; 230 dip->size = ip->size; 231 memmove(dip->addrs, ip->addrs, sizeof(ip->addrs)); 232 log_write(bp); 233 brelse(bp); 234 } 235 236 // Find the inode with number inum on device dev 237 // and return the in-memory copy. Does not lock 238 // the inode and does not read it from disk. 239 static struct inode* 240 iget(uint dev, uint inum) 241 { 242 struct inode *ip, *empty; 243 244 acquire(&icache.lock); 245 246 // Is the inode already cached? 247 empty = 0; 248 for(ip = &icache.inode[0]; ip < &icache.inode[NINODE]; ip++){ 249 if(ip->ref > 0 && ip->dev == dev && ip->inum == inum){ 250 ip->ref++; 251 release(&icache.lock); 252 return ip; 253 } 254 if(empty == 0 && ip->ref == 0) // Remember empty slot. 255 empty = ip; 256 } 257 258 // Recycle an inode cache entry. 259 if(empty == 0) 260 panic("iget: no inodes"); 261 262 ip = empty; 263 ip->dev = dev; 264 ip->inum = inum; 265 ip->ref = 1; 266 ip->valid = 0; 267 release(&icache.lock); 268 269 return ip; 270 } 271 272 // Increment reference count for ip. 273 // Returns ip to enable ip = idup(ip1) idiom. 274 struct inode* 275 idup(struct inode *ip) 276 { 277 acquire(&icache.lock); 278 ip->ref++; 279 release(&icache.lock); 280 return ip; 281 } 282 283 // Lock the given inode. 284 // Reads the inode from disk if necessary. 285 void 286 ilock(struct inode *ip) 287 { 288 struct buf *bp; 289 struct dinode *dip; 290 291 if(ip == 0 || ip->ref < 1) 292 panic("ilock"); 293 294 acquiresleep(&ip->lock); 295 296 if(ip->valid == 0){ 297 bp = bread(ip->dev, IBLOCK(ip->inum, sb)); 298 dip = (struct dinode*)bp->data + ip->inum%IPB; 299 ip->type = dip->type; 300 ip->major = dip->major; 301 ip->minor = dip->minor; 302 ip->nlink = dip->nlink; 303 ip->size = dip->size; 304 memmove(ip->addrs, dip->addrs, sizeof(ip->addrs)); 305 brelse(bp); 306 ip->valid = 1; 307 if(ip->type == 0) 308 panic("ilock: no type"); 309 } 310 } 311 312 // Unlock the given inode. 313 void 314 iunlock(struct inode *ip) 315 { 316 if(ip == 0 || !holdingsleep(&ip->lock) || ip->ref < 1) 317 panic("iunlock"); 318 319 releasesleep(&ip->lock); 320 } 321 322 // Drop a reference to an in-memory inode. 323 // If that was the last reference, the inode cache entry can 324 // be recycled. 325 // If that was the last reference and the inode has no links 326 // to it, free the inode (and its content) on disk. 327 // All calls to iput() must be inside a transaction in 328 // case it has to free the inode. 329 void 330 iput(struct inode *ip) 331 { 332 acquire(&icache.lock); 333 if(ip->ref == 1){ 334 acquiresleep(&ip->lock); 335 if(ip->valid && ip->nlink == 0){ 336 // inode has no links and no other references: truncate and free. 337 release(&icache.lock); 338 itrunc(ip); 339 ip->type = 0; 340 iupdate(ip); 341 ip->valid = 0; 342 acquire(&icache.lock); 343 } 344 releasesleep(&ip->lock); 345 } 346 ip->ref--; 347 release(&icache.lock); 348 } 349 350 // Common idiom: unlock, then put. 351 void 352 iunlockput(struct inode *ip) 353 { 354 iunlock(ip); 355 iput(ip); 356 } 357 358 //PAGEBREAK! 359 // Inode content 360 // 361 // The content (data) associated with each inode is stored 362 // in blocks on the disk. The first NDIRECT block numbers 363 // are listed in ip->addrs[]. The next NINDIRECT blocks are 364 // listed in block ip->addrs[NDIRECT]. 365 366 // Return the disk block address of the nth block in inode ip. 367 // If there is no such block, bmap allocates one. 368 static uint 369 bmap(struct inode *ip, uint bn) 370 { 371 uint addr, *a; 372 struct buf *bp; 373 374 if(bn < NDIRECT){ 375 if((addr = ip->addrs[bn]) == 0) 376 ip->addrs[bn] = addr = balloc(ip->dev); 377 return addr; 378 } 379 bn -= NDIRECT; 380 381 if(bn < NINDIRECT){ 382 // Load indirect block, allocating if necessary. 383 if((addr = ip->addrs[NDIRECT]) == 0) 384 ip->addrs[NDIRECT] = addr = balloc(ip->dev); 385 bp = bread(ip->dev, addr); 386 a = (uint*)bp->data; 387 if((addr = a[bn]) == 0){ 388 a[bn] = addr = balloc(ip->dev); 389 log_write(bp); 390 } 391 brelse(bp); 392 return addr; 393 } 394 395 panic("bmap: out of range"); 396 } 397 398 // Truncate inode (discard contents). 399 // Only called when the inode has no links 400 // to it (no directory entries referring to it) 401 // and has no in-memory reference to it (is 402 // not an open file or current directory). 403 static void 404 itrunc(struct inode *ip) 405 { 406 int i, j; 407 struct buf *bp; 408 uint *a; 409 410 for(i = 0; i < NDIRECT; i++){ 411 if(ip->addrs[i]){ 412 bfree(ip->dev, ip->addrs[i]); 413 ip->addrs[i] = 0; 414 } 415 } 416 417 if(ip->addrs[NDIRECT]){ 418 bp = bread(ip->dev, ip->addrs[NDIRECT]); 419 a = (uint*)bp->data; 420 for(j = 0; j < NINDIRECT; j++){ 421 if(a[j]) 422 bfree(ip->dev, a[j]); 423 } 424 brelse(bp); 425 bfree(ip->dev, ip->addrs[NDIRECT]); 426 ip->addrs[NDIRECT] = 0; 427 } 428 429 ip->size = 0; 430 iupdate(ip); 431 } 432 433 // Copy stat information from inode. 434 // Caller must hold ip->lock. 435 void 436 stati(struct inode *ip, struct stat *st) 437 { 438 st->dev = ip->dev; 439 st->ino = ip->inum; 440 st->type = ip->type; 441 st->nlink = ip->nlink; 442 st->size = ip->size; 443 } 444 445 //PAGEBREAK! 446 // Read data from inode. 447 // Caller must hold ip->lock. 448 int 449 readi(struct inode *ip, char *dst, uint off, uint n) 450 { 451 uint tot, m; 452 struct buf *bp; 453 454 if(ip->type == T_DEV){ 455 if(ip->major < 0 || ip->major >= NDEV || !devsw[ip->major].read) 456 return -1; 457 return devsw[ip->major].read(ip, dst, n); 458 } 459 460 if(off > ip->size || off + n < off) 461 return -1; 462 if(off + n > ip->size) 463 n = ip->size - off; 464 465 for(tot=0; tot<n; tot+=m, off+=m, dst+=m){ 466 bp = bread(ip->dev, bmap(ip, off/BSIZE)); 467 m = min(n - tot, BSIZE - off%BSIZE); 468 memmove(dst, bp->data + off%BSIZE, m); 469 brelse(bp); 470 } 471 return n; 472 } 473 474 // PAGEBREAK! 475 // Write data to inode. 476 // Caller must hold ip->lock. 477 int 478 writei(struct inode *ip, char *src, uint off, uint n) 479 { 480 uint tot, m; 481 struct buf *bp; 482 483 if(ip->type == T_DEV){ 484 if(ip->major < 0 || ip->major >= NDEV || !devsw[ip->major].write) 485 return -1; 486 return devsw[ip->major].write(ip, src, n); 487 } 488 489 if(off > ip->size || off + n < off) 490 return -1; 491 if(off + n > MAXFILE*BSIZE) 492 return -1; 493 494 for(tot=0; tot<n; tot+=m, off+=m, src+=m){ 495 bp = bread(ip->dev, bmap(ip, off/BSIZE)); 496 m = min(n - tot, BSIZE - off%BSIZE); 497 memmove(bp->data + off%BSIZE, src, m); 498 log_write(bp); 499 brelse(bp); 500 } 501 502 if(n > 0 && off > ip->size){ 503 ip->size = off; 504 iupdate(ip); 505 } 506 return n; 507 } 508 509 //PAGEBREAK! 510 // Directories 511 512 int 513 namecmp(const char *s, const char *t) 514 { 515 return strncmp(s, t, DIRSIZ); 516 } 517 518 // Look for a directory entry in a directory. 519 // If found, set *poff to byte offset of entry. 520 struct inode* 521 dirlookup(struct inode *dp, char *name, uint *poff) 522 { 523 uint off, inum; 524 struct dirent de; 525 526 if(dp->type != T_DIR) 527 panic("dirlookup not DIR"); 528 529 for(off = 0; off < dp->size; off += sizeof(de)){ 530 if(readi(dp, (char*)&de, off, sizeof(de)) != sizeof(de)) 531 panic("dirlookup read"); 532 if(de.inum == 0) 533 continue; 534 if(namecmp(name, de.name) == 0){ 535 // entry matches path element 536 if(poff) 537 *poff = off; 538 inum = de.inum; 539 return iget(dp->dev, inum); 540 } 541 } 542 543 return 0; 544 } 545 546 // Write a new directory entry (name, inum) into the directory dp. 547 int 548 dirlink(struct inode *dp, char *name, uint inum) 549 { 550 int off; 551 struct dirent de; 552 struct inode *ip; 553 554 // Check that name is not present. 555 if((ip = dirlookup(dp, name, 0)) != 0){ 556 iput(ip); 557 return -1; 558 } 559 560 // Look for an empty dirent. 561 for(off = 0; off < dp->size; off += sizeof(de)){ 562 if(readi(dp, (char*)&de, off, sizeof(de)) != sizeof(de)) 563 panic("dirlink read"); 564 if(de.inum == 0) 565 break; 566 } 567 568 strncpy(de.name, name, DIRSIZ); 569 de.inum = inum; 570 if(writei(dp, (char*)&de, off, sizeof(de)) != sizeof(de)) 571 panic("dirlink"); 572 573 return 0; 574 } 575 576 //PAGEBREAK! 577 // Paths 578 579 // Copy the next path element from path into name. 580 // Return a pointer to the element following the copied one. 581 // The returned path has no leading slashes, 582 // so the caller can check *path=='\0' to see if the name is the last one. 583 // If no name to remove, return 0. 584 // 585 // Examples: 586 // skipelem("a/bb/c", name) = "bb/c", setting name = "a" 587 // skipelem("///a//bb", name) = "bb", setting name = "a" 588 // skipelem("a", name) = "", setting name = "a" 589 // skipelem("", name) = skipelem("////", name) = 0 590 // 591 static char* 592 skipelem(char *path, char *name) 593 { 594 char *s; 595 int len; 596 597 while(*path == '/') 598 path++; 599 if(*path == 0) 600 return 0; 601 s = path; 602 while(*path != '/' && *path != 0) 603 path++; 604 len = path - s; 605 if(len >= DIRSIZ) 606 memmove(name, s, DIRSIZ); 607 else { 608 memmove(name, s, len); 609 name[len] = 0; 610 } 611 while(*path == '/') 612 path++; 613 return path; 614 } 615 616 // Look up and return the inode for a path name. 617 // If parent != 0, return the inode for the parent and copy the final 618 // path element into name, which must have room for DIRSIZ bytes. 619 // Must be called inside a transaction since it calls iput(). 620 static struct inode* 621 namex(char *path, int nameiparent, char *name) 622 { 623 struct inode *ip, *next; 624 625 if(*path == '/') 626 ip = iget(ROOTDEV, ROOTINO); 627 else 628 ip = idup(myproc()->cwd); 629 630 while((path = skipelem(path, name)) != 0){ 631 ilock(ip); 632 if(ip->type != T_DIR){ 633 iunlockput(ip); 634 return 0; 635 } 636 if(nameiparent && *path == '\0'){ 637 // Stop one level early. 638 iunlock(ip); 639 return ip; 640 } 641 if((next = dirlookup(ip, name, 0)) == 0){ 642 iunlockput(ip); 643 return 0; 644 } 645 iunlockput(ip); 646 ip = next; 647 } 648 if(nameiparent){ 649 iput(ip); 650 return 0; 651 } 652 return ip; 653 } 654 655 struct inode* 656 namei(char *path) 657 { 658 char name[DIRSIZ]; 659 return namex(path, 0, name); 660 } 661 662 struct inode* 663 nameiparent(char *path, char *name) 664 { 665 return namex(path, 1, name); 666 } 667