ffs_alloc.c revision 1.119
1 /* $NetBSD: ffs_alloc.c,v 1.119 2008/12/06 20:05:55 joerg Exp $ */ 2 3 /*- 4 * Copyright (c) 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Wasabi Systems, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (c) 2002 Networks Associates Technology, Inc. 34 * All rights reserved. 35 * 36 * This software was developed for the FreeBSD Project by Marshall 37 * Kirk McKusick and Network Associates Laboratories, the Security 38 * Research Division of Network Associates, Inc. under DARPA/SPAWAR 39 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS 40 * research program 41 * 42 * Copyright (c) 1982, 1986, 1989, 1993 43 * The Regents of the University of California. All rights reserved. 44 * 45 * Redistribution and use in source and binary forms, with or without 46 * modification, are permitted provided that the following conditions 47 * are met: 48 * 1. Redistributions of source code must retain the above copyright 49 * notice, this list of conditions and the following disclaimer. 50 * 2. Redistributions in binary form must reproduce the above copyright 51 * notice, this list of conditions and the following disclaimer in the 52 * documentation and/or other materials provided with the distribution. 53 * 3. Neither the name of the University nor the names of its contributors 54 * may be used to endorse or promote products derived from this software 55 * without specific prior written permission. 56 * 57 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 58 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 59 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 60 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 61 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 62 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 63 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 64 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 65 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 66 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 67 * SUCH DAMAGE. 68 * 69 * @(#)ffs_alloc.c 8.19 (Berkeley) 7/13/95 70 */ 71 72 #include <sys/cdefs.h> 73 __KERNEL_RCSID(0, "$NetBSD: ffs_alloc.c,v 1.119 2008/12/06 20:05:55 joerg Exp $"); 74 75 #if defined(_KERNEL_OPT) 76 #include "opt_ffs.h" 77 #include "opt_quota.h" 78 #endif 79 80 #include <sys/param.h> 81 #include <sys/systm.h> 82 #include <sys/buf.h> 83 #include <sys/fstrans.h> 84 #include <sys/kauth.h> 85 #include <sys/kernel.h> 86 #include <sys/mount.h> 87 #include <sys/proc.h> 88 #include <sys/syslog.h> 89 #include <sys/vnode.h> 90 #include <sys/wapbl.h> 91 92 #include <miscfs/specfs/specdev.h> 93 #include <ufs/ufs/quota.h> 94 #include <ufs/ufs/ufsmount.h> 95 #include <ufs/ufs/inode.h> 96 #include <ufs/ufs/ufs_extern.h> 97 #include <ufs/ufs/ufs_bswap.h> 98 #include <ufs/ufs/ufs_wapbl.h> 99 100 #include <ufs/ffs/fs.h> 101 #include <ufs/ffs/ffs_extern.h> 102 103 static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int, int); 104 static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t, int); 105 static ino_t ffs_dirpref(struct inode *); 106 static daddr_t ffs_fragextend(struct inode *, int, daddr_t, int, int); 107 static void ffs_fserr(struct fs *, u_int, const char *); 108 static daddr_t ffs_hashalloc(struct inode *, int, daddr_t, int, int, 109 daddr_t (*)(struct inode *, int, daddr_t, int, int)); 110 static daddr_t ffs_nodealloccg(struct inode *, int, daddr_t, int, int); 111 static int32_t ffs_mapsearch(struct fs *, struct cg *, 112 daddr_t, int); 113 static void ffs_blkfree_common(struct ufsmount *, struct fs *, dev_t, struct buf *, 114 daddr_t, long, bool); 115 static void ffs_freefile_common(struct ufsmount *, struct fs *, dev_t, struct buf *, ino_t, 116 int, bool); 117 118 /* if 1, changes in optimalization strategy are logged */ 119 int ffs_log_changeopt = 0; 120 121 /* in ffs_tables.c */ 122 extern const int inside[], around[]; 123 extern const u_char * const fragtbl[]; 124 125 /* Basic consistency check for block allocations */ 126 static int 127 ffs_check_bad_allocation(const char *func, struct fs *fs, daddr_t bno, 128 long size, dev_t dev, ino_t inum) 129 { 130 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || 131 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { 132 printf("dev = 0x%x, bno = %" PRId64 " bsize = %d, " 133 "size = %ld, fs = %s\n", 134 dev, bno, fs->fs_bsize, size, fs->fs_fsmnt); 135 panic("%s: bad size", func); 136 } 137 138 if (bno >= fs->fs_size) { 139 printf("bad block %" PRId64 ", ino %llu\n", bno, 140 (unsigned long long)inum); 141 ffs_fserr(fs, inum, "bad block"); 142 return EINVAL; 143 } 144 return 0; 145 } 146 147 /* 148 * Allocate a block in the file system. 149 * 150 * The size of the requested block is given, which must be some 151 * multiple of fs_fsize and <= fs_bsize. 152 * A preference may be optionally specified. If a preference is given 153 * the following hierarchy is used to allocate a block: 154 * 1) allocate the requested block. 155 * 2) allocate a rotationally optimal block in the same cylinder. 156 * 3) allocate a block in the same cylinder group. 157 * 4) quadradically rehash into other cylinder groups, until an 158 * available block is located. 159 * If no block preference is given the following hierarchy is used 160 * to allocate a block: 161 * 1) allocate a block in the cylinder group that contains the 162 * inode for the file. 163 * 2) quadradically rehash into other cylinder groups, until an 164 * available block is located. 165 * 166 * => called with um_lock held 167 * => releases um_lock before returning 168 */ 169 int 170 ffs_alloc(struct inode *ip, daddr_t lbn, daddr_t bpref, int size, int flags, 171 kauth_cred_t cred, daddr_t *bnp) 172 { 173 struct ufsmount *ump; 174 struct fs *fs; 175 daddr_t bno; 176 int cg; 177 #ifdef QUOTA 178 int error; 179 #endif 180 181 fs = ip->i_fs; 182 ump = ip->i_ump; 183 184 KASSERT(mutex_owned(&ump->um_lock)); 185 186 #ifdef UVM_PAGE_TRKOWN 187 if (ITOV(ip)->v_type == VREG && 188 lblktosize(fs, (voff_t)lbn) < round_page(ITOV(ip)->v_size)) { 189 struct vm_page *pg; 190 struct uvm_object *uobj = &ITOV(ip)->v_uobj; 191 voff_t off = trunc_page(lblktosize(fs, lbn)); 192 voff_t endoff = round_page(lblktosize(fs, lbn) + size); 193 194 mutex_enter(&uobj->vmobjlock); 195 while (off < endoff) { 196 pg = uvm_pagelookup(uobj, off); 197 KASSERT(pg != NULL); 198 KASSERT(pg->owner == curproc->p_pid); 199 off += PAGE_SIZE; 200 } 201 mutex_exit(&uobj->vmobjlock); 202 } 203 #endif 204 205 *bnp = 0; 206 #ifdef DIAGNOSTIC 207 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 208 printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n", 209 ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt); 210 panic("ffs_alloc: bad size"); 211 } 212 if (cred == NOCRED) 213 panic("ffs_alloc: missing credential"); 214 #endif /* DIAGNOSTIC */ 215 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) 216 goto nospace; 217 if (freespace(fs, fs->fs_minfree) <= 0 && 218 kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, NULL) != 0) 219 goto nospace; 220 #ifdef QUOTA 221 mutex_exit(&ump->um_lock); 222 if ((error = chkdq(ip, btodb(size), cred, 0)) != 0) 223 return (error); 224 mutex_enter(&ump->um_lock); 225 #endif 226 227 if (bpref >= fs->fs_size) 228 bpref = 0; 229 if (bpref == 0) 230 cg = ino_to_cg(fs, ip->i_number); 231 else 232 cg = dtog(fs, bpref); 233 bno = ffs_hashalloc(ip, cg, bpref, size, flags, ffs_alloccg); 234 if (bno > 0) { 235 DIP_ADD(ip, blocks, btodb(size)); 236 ip->i_flag |= IN_CHANGE | IN_UPDATE; 237 *bnp = bno; 238 return (0); 239 } 240 #ifdef QUOTA 241 /* 242 * Restore user's disk quota because allocation failed. 243 */ 244 (void) chkdq(ip, -btodb(size), cred, FORCE); 245 #endif 246 if (flags & B_CONTIG) { 247 /* 248 * XXX ump->um_lock handling is "suspect" at best. 249 * For the case where ffs_hashalloc() fails early 250 * in the B_CONTIG case we reach here with um_lock 251 * already unlocked, so we can't release it again 252 * like in the normal error path. See kern/39206. 253 * 254 * 255 * Fail silently - it's up to our caller to report 256 * errors. 257 */ 258 return (ENOSPC); 259 } 260 nospace: 261 mutex_exit(&ump->um_lock); 262 ffs_fserr(fs, kauth_cred_geteuid(cred), "file system full"); 263 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 264 return (ENOSPC); 265 } 266 267 /* 268 * Reallocate a fragment to a bigger size 269 * 270 * The number and size of the old block is given, and a preference 271 * and new size is also specified. The allocator attempts to extend 272 * the original block. Failing that, the regular block allocator is 273 * invoked to get an appropriate block. 274 * 275 * => called with um_lock held 276 * => return with um_lock released 277 */ 278 int 279 ffs_realloccg(struct inode *ip, daddr_t lbprev, daddr_t bpref, int osize, 280 int nsize, kauth_cred_t cred, struct buf **bpp, daddr_t *blknop) 281 { 282 struct ufsmount *ump; 283 struct fs *fs; 284 struct buf *bp; 285 int cg, request, error; 286 daddr_t bprev, bno; 287 288 fs = ip->i_fs; 289 ump = ip->i_ump; 290 291 KASSERT(mutex_owned(&ump->um_lock)); 292 293 #ifdef UVM_PAGE_TRKOWN 294 if (ITOV(ip)->v_type == VREG) { 295 struct vm_page *pg; 296 struct uvm_object *uobj = &ITOV(ip)->v_uobj; 297 voff_t off = trunc_page(lblktosize(fs, lbprev)); 298 voff_t endoff = round_page(lblktosize(fs, lbprev) + osize); 299 300 mutex_enter(&uobj->vmobjlock); 301 while (off < endoff) { 302 pg = uvm_pagelookup(uobj, off); 303 KASSERT(pg != NULL); 304 KASSERT(pg->owner == curproc->p_pid); 305 KASSERT((pg->flags & PG_CLEAN) == 0); 306 off += PAGE_SIZE; 307 } 308 mutex_exit(&uobj->vmobjlock); 309 } 310 #endif 311 312 #ifdef DIAGNOSTIC 313 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || 314 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { 315 printf( 316 "dev = 0x%x, bsize = %d, osize = %d, nsize = %d, fs = %s\n", 317 ip->i_dev, fs->fs_bsize, osize, nsize, fs->fs_fsmnt); 318 panic("ffs_realloccg: bad size"); 319 } 320 if (cred == NOCRED) 321 panic("ffs_realloccg: missing credential"); 322 #endif /* DIAGNOSTIC */ 323 if (freespace(fs, fs->fs_minfree) <= 0 && 324 kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, NULL) != 0) { 325 mutex_exit(&ump->um_lock); 326 goto nospace; 327 } 328 if (fs->fs_magic == FS_UFS2_MAGIC) 329 bprev = ufs_rw64(ip->i_ffs2_db[lbprev], UFS_FSNEEDSWAP(fs)); 330 else 331 bprev = ufs_rw32(ip->i_ffs1_db[lbprev], UFS_FSNEEDSWAP(fs)); 332 333 if (bprev == 0) { 334 printf("dev = 0x%x, bsize = %d, bprev = %" PRId64 ", fs = %s\n", 335 ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt); 336 panic("ffs_realloccg: bad bprev"); 337 } 338 mutex_exit(&ump->um_lock); 339 340 /* 341 * Allocate the extra space in the buffer. 342 */ 343 if (bpp != NULL && 344 (error = bread(ITOV(ip), lbprev, osize, NOCRED, 0, &bp)) != 0) { 345 brelse(bp, 0); 346 return (error); 347 } 348 #ifdef QUOTA 349 if ((error = chkdq(ip, btodb(nsize - osize), cred, 0)) != 0) { 350 if (bpp != NULL) { 351 brelse(bp, 0); 352 } 353 return (error); 354 } 355 #endif 356 /* 357 * Check for extension in the existing location. 358 */ 359 cg = dtog(fs, bprev); 360 mutex_enter(&ump->um_lock); 361 if ((bno = ffs_fragextend(ip, cg, bprev, osize, nsize)) != 0) { 362 DIP_ADD(ip, blocks, btodb(nsize - osize)); 363 ip->i_flag |= IN_CHANGE | IN_UPDATE; 364 365 if (bpp != NULL) { 366 if (bp->b_blkno != fsbtodb(fs, bno)) 367 panic("bad blockno"); 368 allocbuf(bp, nsize, 1); 369 memset((char *)bp->b_data + osize, 0, nsize - osize); 370 mutex_enter(bp->b_objlock); 371 KASSERT(!cv_has_waiters(&bp->b_done)); 372 bp->b_oflags |= BO_DONE; 373 mutex_exit(bp->b_objlock); 374 *bpp = bp; 375 } 376 if (blknop != NULL) { 377 *blknop = bno; 378 } 379 return (0); 380 } 381 /* 382 * Allocate a new disk location. 383 */ 384 if (bpref >= fs->fs_size) 385 bpref = 0; 386 switch ((int)fs->fs_optim) { 387 case FS_OPTSPACE: 388 /* 389 * Allocate an exact sized fragment. Although this makes 390 * best use of space, we will waste time relocating it if 391 * the file continues to grow. If the fragmentation is 392 * less than half of the minimum free reserve, we choose 393 * to begin optimizing for time. 394 */ 395 request = nsize; 396 if (fs->fs_minfree < 5 || 397 fs->fs_cstotal.cs_nffree > 398 fs->fs_dsize * fs->fs_minfree / (2 * 100)) 399 break; 400 401 if (ffs_log_changeopt) { 402 log(LOG_NOTICE, 403 "%s: optimization changed from SPACE to TIME\n", 404 fs->fs_fsmnt); 405 } 406 407 fs->fs_optim = FS_OPTTIME; 408 break; 409 case FS_OPTTIME: 410 /* 411 * At this point we have discovered a file that is trying to 412 * grow a small fragment to a larger fragment. To save time, 413 * we allocate a full sized block, then free the unused portion. 414 * If the file continues to grow, the `ffs_fragextend' call 415 * above will be able to grow it in place without further 416 * copying. If aberrant programs cause disk fragmentation to 417 * grow within 2% of the free reserve, we choose to begin 418 * optimizing for space. 419 */ 420 request = fs->fs_bsize; 421 if (fs->fs_cstotal.cs_nffree < 422 fs->fs_dsize * (fs->fs_minfree - 2) / 100) 423 break; 424 425 if (ffs_log_changeopt) { 426 log(LOG_NOTICE, 427 "%s: optimization changed from TIME to SPACE\n", 428 fs->fs_fsmnt); 429 } 430 431 fs->fs_optim = FS_OPTSPACE; 432 break; 433 default: 434 printf("dev = 0x%x, optim = %d, fs = %s\n", 435 ip->i_dev, fs->fs_optim, fs->fs_fsmnt); 436 panic("ffs_realloccg: bad optim"); 437 /* NOTREACHED */ 438 } 439 bno = ffs_hashalloc(ip, cg, bpref, request, 0, ffs_alloccg); 440 if (bno > 0) { 441 if (!DOINGSOFTDEP(ITOV(ip))) { 442 if ((ip->i_ump->um_mountp->mnt_wapbl) && 443 (ITOV(ip)->v_type != VREG)) { 444 UFS_WAPBL_REGISTER_DEALLOCATION( 445 ip->i_ump->um_mountp, fsbtodb(fs, bprev), 446 osize); 447 } else 448 ffs_blkfree(fs, ip->i_devvp, bprev, (long)osize, 449 ip->i_number); 450 } 451 if (nsize < request) { 452 if ((ip->i_ump->um_mountp->mnt_wapbl) && 453 (ITOV(ip)->v_type != VREG)) { 454 UFS_WAPBL_REGISTER_DEALLOCATION( 455 ip->i_ump->um_mountp, 456 fsbtodb(fs, (bno + numfrags(fs, nsize))), 457 request - nsize); 458 } else 459 ffs_blkfree(fs, ip->i_devvp, 460 bno + numfrags(fs, nsize), 461 (long)(request - nsize), ip->i_number); 462 } 463 DIP_ADD(ip, blocks, btodb(nsize - osize)); 464 ip->i_flag |= IN_CHANGE | IN_UPDATE; 465 if (bpp != NULL) { 466 bp->b_blkno = fsbtodb(fs, bno); 467 allocbuf(bp, nsize, 1); 468 memset((char *)bp->b_data + osize, 0, (u_int)nsize - osize); 469 mutex_enter(bp->b_objlock); 470 KASSERT(!cv_has_waiters(&bp->b_done)); 471 bp->b_oflags |= BO_DONE; 472 mutex_exit(bp->b_objlock); 473 *bpp = bp; 474 } 475 if (blknop != NULL) { 476 *blknop = bno; 477 } 478 return (0); 479 } 480 mutex_exit(&ump->um_lock); 481 482 #ifdef QUOTA 483 /* 484 * Restore user's disk quota because allocation failed. 485 */ 486 (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE); 487 #endif 488 if (bpp != NULL) { 489 brelse(bp, 0); 490 } 491 492 nospace: 493 /* 494 * no space available 495 */ 496 ffs_fserr(fs, kauth_cred_geteuid(cred), "file system full"); 497 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 498 return (ENOSPC); 499 } 500 501 /* 502 * Allocate an inode in the file system. 503 * 504 * If allocating a directory, use ffs_dirpref to select the inode. 505 * If allocating in a directory, the following hierarchy is followed: 506 * 1) allocate the preferred inode. 507 * 2) allocate an inode in the same cylinder group. 508 * 3) quadradically rehash into other cylinder groups, until an 509 * available inode is located. 510 * If no inode preference is given the following hierarchy is used 511 * to allocate an inode: 512 * 1) allocate an inode in cylinder group 0. 513 * 2) quadradically rehash into other cylinder groups, until an 514 * available inode is located. 515 * 516 * => um_lock not held upon entry or return 517 */ 518 int 519 ffs_valloc(struct vnode *pvp, int mode, kauth_cred_t cred, 520 struct vnode **vpp) 521 { 522 struct ufsmount *ump; 523 struct inode *pip; 524 struct fs *fs; 525 struct inode *ip; 526 struct timespec ts; 527 ino_t ino, ipref; 528 int cg, error; 529 530 UFS_WAPBL_JUNLOCK_ASSERT(pvp->v_mount); 531 532 *vpp = NULL; 533 pip = VTOI(pvp); 534 fs = pip->i_fs; 535 ump = pip->i_ump; 536 537 error = UFS_WAPBL_BEGIN(pvp->v_mount); 538 if (error) { 539 return error; 540 } 541 mutex_enter(&ump->um_lock); 542 if (fs->fs_cstotal.cs_nifree == 0) 543 goto noinodes; 544 545 if ((mode & IFMT) == IFDIR) 546 ipref = ffs_dirpref(pip); 547 else 548 ipref = pip->i_number; 549 if (ipref >= fs->fs_ncg * fs->fs_ipg) 550 ipref = 0; 551 cg = ino_to_cg(fs, ipref); 552 /* 553 * Track number of dirs created one after another 554 * in a same cg without intervening by files. 555 */ 556 if ((mode & IFMT) == IFDIR) { 557 if (fs->fs_contigdirs[cg] < 255) 558 fs->fs_contigdirs[cg]++; 559 } else { 560 if (fs->fs_contigdirs[cg] > 0) 561 fs->fs_contigdirs[cg]--; 562 } 563 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, ffs_nodealloccg); 564 if (ino == 0) 565 goto noinodes; 566 UFS_WAPBL_END(pvp->v_mount); 567 error = VFS_VGET(pvp->v_mount, ino, vpp); 568 if (error) { 569 int err; 570 err = UFS_WAPBL_BEGIN(pvp->v_mount); 571 if (err == 0) 572 ffs_vfree(pvp, ino, mode); 573 if (err == 0) 574 UFS_WAPBL_END(pvp->v_mount); 575 return (error); 576 } 577 KASSERT((*vpp)->v_type == VNON); 578 ip = VTOI(*vpp); 579 if (ip->i_mode) { 580 #if 0 581 printf("mode = 0%o, inum = %d, fs = %s\n", 582 ip->i_mode, ip->i_number, fs->fs_fsmnt); 583 #else 584 printf("dmode %x mode %x dgen %x gen %x\n", 585 DIP(ip, mode), ip->i_mode, 586 DIP(ip, gen), ip->i_gen); 587 printf("size %llx blocks %llx\n", 588 (long long)DIP(ip, size), (long long)DIP(ip, blocks)); 589 printf("ino %llu ipref %llu\n", (unsigned long long)ino, 590 (unsigned long long)ipref); 591 #if 0 592 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 593 (int)fs->fs_bsize, NOCRED, 0, &bp); 594 #endif 595 596 #endif 597 panic("ffs_valloc: dup alloc"); 598 } 599 if (DIP(ip, blocks)) { /* XXX */ 600 printf("free inode %s/%llu had %" PRId64 " blocks\n", 601 fs->fs_fsmnt, (unsigned long long)ino, DIP(ip, blocks)); 602 DIP_ASSIGN(ip, blocks, 0); 603 } 604 ip->i_flag &= ~IN_SPACECOUNTED; 605 ip->i_flags = 0; 606 DIP_ASSIGN(ip, flags, 0); 607 /* 608 * Set up a new generation number for this inode. 609 */ 610 ip->i_gen++; 611 DIP_ASSIGN(ip, gen, ip->i_gen); 612 if (fs->fs_magic == FS_UFS2_MAGIC) { 613 vfs_timestamp(&ts); 614 ip->i_ffs2_birthtime = ts.tv_sec; 615 ip->i_ffs2_birthnsec = ts.tv_nsec; 616 } 617 return (0); 618 noinodes: 619 mutex_exit(&ump->um_lock); 620 UFS_WAPBL_END(pvp->v_mount); 621 ffs_fserr(fs, kauth_cred_geteuid(cred), "out of inodes"); 622 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); 623 return (ENOSPC); 624 } 625 626 /* 627 * Find a cylinder group in which to place a directory. 628 * 629 * The policy implemented by this algorithm is to allocate a 630 * directory inode in the same cylinder group as its parent 631 * directory, but also to reserve space for its files inodes 632 * and data. Restrict the number of directories which may be 633 * allocated one after another in the same cylinder group 634 * without intervening allocation of files. 635 * 636 * If we allocate a first level directory then force allocation 637 * in another cylinder group. 638 */ 639 static ino_t 640 ffs_dirpref(struct inode *pip) 641 { 642 register struct fs *fs; 643 int cg, prefcg; 644 int64_t dirsize, cgsize, curdsz; 645 int avgifree, avgbfree, avgndir; 646 int minifree, minbfree, maxndir; 647 int mincg, minndir; 648 int maxcontigdirs; 649 650 KASSERT(mutex_owned(&pip->i_ump->um_lock)); 651 652 fs = pip->i_fs; 653 654 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; 655 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 656 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; 657 658 /* 659 * Force allocation in another cg if creating a first level dir. 660 */ 661 if (ITOV(pip)->v_vflag & VV_ROOT) { 662 prefcg = random() % fs->fs_ncg; 663 mincg = prefcg; 664 minndir = fs->fs_ipg; 665 for (cg = prefcg; cg < fs->fs_ncg; cg++) 666 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 667 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 668 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 669 mincg = cg; 670 minndir = fs->fs_cs(fs, cg).cs_ndir; 671 } 672 for (cg = 0; cg < prefcg; cg++) 673 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 674 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 675 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 676 mincg = cg; 677 minndir = fs->fs_cs(fs, cg).cs_ndir; 678 } 679 return ((ino_t)(fs->fs_ipg * mincg)); 680 } 681 682 /* 683 * Count various limits which used for 684 * optimal allocation of a directory inode. 685 */ 686 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); 687 minifree = avgifree - fs->fs_ipg / 4; 688 if (minifree < 0) 689 minifree = 0; 690 minbfree = avgbfree - fragstoblks(fs, fs->fs_fpg) / 4; 691 if (minbfree < 0) 692 minbfree = 0; 693 cgsize = (int64_t)fs->fs_fsize * fs->fs_fpg; 694 dirsize = (int64_t)fs->fs_avgfilesize * fs->fs_avgfpdir; 695 if (avgndir != 0) { 696 curdsz = (cgsize - (int64_t)avgbfree * fs->fs_bsize) / avgndir; 697 if (dirsize < curdsz) 698 dirsize = curdsz; 699 } 700 if (cgsize < dirsize * 255) 701 maxcontigdirs = cgsize / dirsize; 702 else 703 maxcontigdirs = 255; 704 if (fs->fs_avgfpdir > 0) 705 maxcontigdirs = min(maxcontigdirs, 706 fs->fs_ipg / fs->fs_avgfpdir); 707 if (maxcontigdirs == 0) 708 maxcontigdirs = 1; 709 710 /* 711 * Limit number of dirs in one cg and reserve space for 712 * regular files, but only if we have no deficit in 713 * inodes or space. 714 */ 715 prefcg = ino_to_cg(fs, pip->i_number); 716 for (cg = prefcg; cg < fs->fs_ncg; cg++) 717 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 718 fs->fs_cs(fs, cg).cs_nifree >= minifree && 719 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 720 if (fs->fs_contigdirs[cg] < maxcontigdirs) 721 return ((ino_t)(fs->fs_ipg * cg)); 722 } 723 for (cg = 0; cg < prefcg; cg++) 724 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 725 fs->fs_cs(fs, cg).cs_nifree >= minifree && 726 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 727 if (fs->fs_contigdirs[cg] < maxcontigdirs) 728 return ((ino_t)(fs->fs_ipg * cg)); 729 } 730 /* 731 * This is a backstop when we are deficient in space. 732 */ 733 for (cg = prefcg; cg < fs->fs_ncg; cg++) 734 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 735 return ((ino_t)(fs->fs_ipg * cg)); 736 for (cg = 0; cg < prefcg; cg++) 737 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 738 break; 739 return ((ino_t)(fs->fs_ipg * cg)); 740 } 741 742 /* 743 * Select the desired position for the next block in a file. The file is 744 * logically divided into sections. The first section is composed of the 745 * direct blocks. Each additional section contains fs_maxbpg blocks. 746 * 747 * If no blocks have been allocated in the first section, the policy is to 748 * request a block in the same cylinder group as the inode that describes 749 * the file. If no blocks have been allocated in any other section, the 750 * policy is to place the section in a cylinder group with a greater than 751 * average number of free blocks. An appropriate cylinder group is found 752 * by using a rotor that sweeps the cylinder groups. When a new group of 753 * blocks is needed, the sweep begins in the cylinder group following the 754 * cylinder group from which the previous allocation was made. The sweep 755 * continues until a cylinder group with greater than the average number 756 * of free blocks is found. If the allocation is for the first block in an 757 * indirect block, the information on the previous allocation is unavailable; 758 * here a best guess is made based upon the logical block number being 759 * allocated. 760 * 761 * If a section is already partially allocated, the policy is to 762 * contiguously allocate fs_maxcontig blocks. The end of one of these 763 * contiguous blocks and the beginning of the next is laid out 764 * contigously if possible. 765 * 766 * => um_lock held on entry and exit 767 */ 768 daddr_t 769 ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int flags, 770 int32_t *bap /* XXX ondisk32 */) 771 { 772 struct fs *fs; 773 int cg; 774 int avgbfree, startcg; 775 776 KASSERT(mutex_owned(&ip->i_ump->um_lock)); 777 778 fs = ip->i_fs; 779 780 /* 781 * If allocating a contiguous file with B_CONTIG, use the hints 782 * in the inode extentions to return the desired block. 783 * 784 * For metadata (indirect blocks) return the address of where 785 * the first indirect block resides - we'll scan for the next 786 * available slot if we need to allocate more than one indirect 787 * block. For data, return the address of the actual block 788 * relative to the address of the first data block. 789 */ 790 if (flags & B_CONTIG) { 791 KASSERT(ip->i_ffs_first_data_blk != 0); 792 KASSERT(ip->i_ffs_first_indir_blk != 0); 793 if (flags & B_METAONLY) 794 return ip->i_ffs_first_indir_blk; 795 else 796 return ip->i_ffs_first_data_blk + blkstofrags(fs, lbn); 797 } 798 799 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 800 if (lbn < NDADDR + NINDIR(fs)) { 801 cg = ino_to_cg(fs, ip->i_number); 802 return (cgbase(fs, cg) + fs->fs_frag); 803 } 804 /* 805 * Find a cylinder with greater than average number of 806 * unused data blocks. 807 */ 808 if (indx == 0 || bap[indx - 1] == 0) 809 startcg = 810 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 811 else 812 startcg = dtog(fs, 813 ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1); 814 startcg %= fs->fs_ncg; 815 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 816 for (cg = startcg; cg < fs->fs_ncg; cg++) 817 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 818 return (cgbase(fs, cg) + fs->fs_frag); 819 } 820 for (cg = 0; cg < startcg; cg++) 821 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 822 return (cgbase(fs, cg) + fs->fs_frag); 823 } 824 return (0); 825 } 826 /* 827 * We just always try to lay things out contiguously. 828 */ 829 return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag; 830 } 831 832 daddr_t 833 ffs_blkpref_ufs2(struct inode *ip, daddr_t lbn, int indx, int flags, 834 int64_t *bap) 835 { 836 struct fs *fs; 837 int cg; 838 int avgbfree, startcg; 839 840 KASSERT(mutex_owned(&ip->i_ump->um_lock)); 841 842 fs = ip->i_fs; 843 844 /* 845 * If allocating a contiguous file with B_CONTIG, use the hints 846 * in the inode extentions to return the desired block. 847 * 848 * For metadata (indirect blocks) return the address of where 849 * the first indirect block resides - we'll scan for the next 850 * available slot if we need to allocate more than one indirect 851 * block. For data, return the address of the actual block 852 * relative to the address of the first data block. 853 */ 854 if (flags & B_CONTIG) { 855 KASSERT(ip->i_ffs_first_data_blk != 0); 856 KASSERT(ip->i_ffs_first_indir_blk != 0); 857 if (flags & B_METAONLY) 858 return ip->i_ffs_first_indir_blk; 859 else 860 return ip->i_ffs_first_data_blk + blkstofrags(fs, lbn); 861 } 862 863 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 864 if (lbn < NDADDR + NINDIR(fs)) { 865 cg = ino_to_cg(fs, ip->i_number); 866 return (cgbase(fs, cg) + fs->fs_frag); 867 } 868 /* 869 * Find a cylinder with greater than average number of 870 * unused data blocks. 871 */ 872 if (indx == 0 || bap[indx - 1] == 0) 873 startcg = 874 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 875 else 876 startcg = dtog(fs, 877 ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1); 878 startcg %= fs->fs_ncg; 879 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 880 for (cg = startcg; cg < fs->fs_ncg; cg++) 881 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 882 return (cgbase(fs, cg) + fs->fs_frag); 883 } 884 for (cg = 0; cg < startcg; cg++) 885 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 886 return (cgbase(fs, cg) + fs->fs_frag); 887 } 888 return (0); 889 } 890 /* 891 * We just always try to lay things out contiguously. 892 */ 893 return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag; 894 } 895 896 897 /* 898 * Implement the cylinder overflow algorithm. 899 * 900 * The policy implemented by this algorithm is: 901 * 1) allocate the block in its requested cylinder group. 902 * 2) quadradically rehash on the cylinder group number. 903 * 3) brute force search for a free block. 904 * 905 * => called with um_lock held 906 * => returns with um_lock released on success, held on failure 907 * (*allocator releases lock on success, retains lock on failure) 908 */ 909 /*VARARGS5*/ 910 static daddr_t 911 ffs_hashalloc(struct inode *ip, int cg, daddr_t pref, 912 int size /* size for data blocks, mode for inodes */, 913 int flags, daddr_t (*allocator)(struct inode *, int, daddr_t, int, int)) 914 { 915 struct fs *fs; 916 daddr_t result; 917 int i, icg = cg; 918 919 fs = ip->i_fs; 920 /* 921 * 1: preferred cylinder group 922 */ 923 result = (*allocator)(ip, cg, pref, size, flags); 924 if (result) 925 return (result); 926 927 if (flags & B_CONTIG) 928 return (result); 929 /* 930 * 2: quadratic rehash 931 */ 932 for (i = 1; i < fs->fs_ncg; i *= 2) { 933 cg += i; 934 if (cg >= fs->fs_ncg) 935 cg -= fs->fs_ncg; 936 result = (*allocator)(ip, cg, 0, size, flags); 937 if (result) 938 return (result); 939 } 940 /* 941 * 3: brute force search 942 * Note that we start at i == 2, since 0 was checked initially, 943 * and 1 is always checked in the quadratic rehash. 944 */ 945 cg = (icg + 2) % fs->fs_ncg; 946 for (i = 2; i < fs->fs_ncg; i++) { 947 result = (*allocator)(ip, cg, 0, size, flags); 948 if (result) 949 return (result); 950 cg++; 951 if (cg == fs->fs_ncg) 952 cg = 0; 953 } 954 return (0); 955 } 956 957 /* 958 * Determine whether a fragment can be extended. 959 * 960 * Check to see if the necessary fragments are available, and 961 * if they are, allocate them. 962 * 963 * => called with um_lock held 964 * => returns with um_lock released on success, held on failure 965 */ 966 static daddr_t 967 ffs_fragextend(struct inode *ip, int cg, daddr_t bprev, int osize, int nsize) 968 { 969 struct ufsmount *ump; 970 struct fs *fs; 971 struct cg *cgp; 972 struct buf *bp; 973 daddr_t bno; 974 int frags, bbase; 975 int i, error; 976 u_int8_t *blksfree; 977 978 fs = ip->i_fs; 979 ump = ip->i_ump; 980 981 KASSERT(mutex_owned(&ump->um_lock)); 982 983 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) 984 return (0); 985 frags = numfrags(fs, nsize); 986 bbase = fragnum(fs, bprev); 987 if (bbase > fragnum(fs, (bprev + frags - 1))) { 988 /* cannot extend across a block boundary */ 989 return (0); 990 } 991 mutex_exit(&ump->um_lock); 992 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 993 (int)fs->fs_cgsize, NOCRED, B_MODIFY, &bp); 994 if (error) 995 goto fail; 996 cgp = (struct cg *)bp->b_data; 997 if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs))) 998 goto fail; 999 cgp->cg_old_time = ufs_rw32(time_second, UFS_FSNEEDSWAP(fs)); 1000 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1001 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1002 cgp->cg_time = ufs_rw64(time_second, UFS_FSNEEDSWAP(fs)); 1003 bno = dtogd(fs, bprev); 1004 blksfree = cg_blksfree(cgp, UFS_FSNEEDSWAP(fs)); 1005 for (i = numfrags(fs, osize); i < frags; i++) 1006 if (isclr(blksfree, bno + i)) 1007 goto fail; 1008 /* 1009 * the current fragment can be extended 1010 * deduct the count on fragment being extended into 1011 * increase the count on the remaining fragment (if any) 1012 * allocate the extended piece 1013 */ 1014 for (i = frags; i < fs->fs_frag - bbase; i++) 1015 if (isclr(blksfree, bno + i)) 1016 break; 1017 ufs_add32(cgp->cg_frsum[i - numfrags(fs, osize)], -1, UFS_FSNEEDSWAP(fs)); 1018 if (i != frags) 1019 ufs_add32(cgp->cg_frsum[i - frags], 1, UFS_FSNEEDSWAP(fs)); 1020 mutex_enter(&ump->um_lock); 1021 for (i = numfrags(fs, osize); i < frags; i++) { 1022 clrbit(blksfree, bno + i); 1023 ufs_add32(cgp->cg_cs.cs_nffree, -1, UFS_FSNEEDSWAP(fs)); 1024 fs->fs_cstotal.cs_nffree--; 1025 fs->fs_cs(fs, cg).cs_nffree--; 1026 } 1027 fs->fs_fmod = 1; 1028 ACTIVECG_CLR(fs, cg); 1029 mutex_exit(&ump->um_lock); 1030 if (DOINGSOFTDEP(ITOV(ip))) 1031 softdep_setup_blkmapdep(bp, fs, bprev); 1032 bdwrite(bp); 1033 return (bprev); 1034 1035 fail: 1036 brelse(bp, 0); 1037 mutex_enter(&ump->um_lock); 1038 return (0); 1039 } 1040 1041 /* 1042 * Determine whether a block can be allocated. 1043 * 1044 * Check to see if a block of the appropriate size is available, 1045 * and if it is, allocate it. 1046 */ 1047 static daddr_t 1048 ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size, int flags) 1049 { 1050 struct ufsmount *ump; 1051 struct fs *fs = ip->i_fs; 1052 struct cg *cgp; 1053 struct buf *bp; 1054 int32_t bno; 1055 daddr_t blkno; 1056 int error, frags, allocsiz, i; 1057 u_int8_t *blksfree; 1058 #ifdef FFS_EI 1059 const int needswap = UFS_FSNEEDSWAP(fs); 1060 #endif 1061 1062 ump = ip->i_ump; 1063 1064 KASSERT(mutex_owned(&ump->um_lock)); 1065 1066 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) 1067 return (0); 1068 mutex_exit(&ump->um_lock); 1069 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1070 (int)fs->fs_cgsize, NOCRED, B_MODIFY, &bp); 1071 if (error) 1072 goto fail; 1073 cgp = (struct cg *)bp->b_data; 1074 if (!cg_chkmagic(cgp, needswap) || 1075 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) 1076 goto fail; 1077 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1078 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1079 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1080 cgp->cg_time = ufs_rw64(time_second, needswap); 1081 if (size == fs->fs_bsize) { 1082 mutex_enter(&ump->um_lock); 1083 blkno = ffs_alloccgblk(ip, bp, bpref, flags); 1084 ACTIVECG_CLR(fs, cg); 1085 mutex_exit(&ump->um_lock); 1086 bdwrite(bp); 1087 return (blkno); 1088 } 1089 /* 1090 * check to see if any fragments are already available 1091 * allocsiz is the size which will be allocated, hacking 1092 * it down to a smaller size if necessary 1093 */ 1094 blksfree = cg_blksfree(cgp, needswap); 1095 frags = numfrags(fs, size); 1096 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) 1097 if (cgp->cg_frsum[allocsiz] != 0) 1098 break; 1099 if (allocsiz == fs->fs_frag) { 1100 /* 1101 * no fragments were available, so a block will be 1102 * allocated, and hacked up 1103 */ 1104 if (cgp->cg_cs.cs_nbfree == 0) 1105 goto fail; 1106 mutex_enter(&ump->um_lock); 1107 blkno = ffs_alloccgblk(ip, bp, bpref, flags); 1108 bno = dtogd(fs, blkno); 1109 for (i = frags; i < fs->fs_frag; i++) 1110 setbit(blksfree, bno + i); 1111 i = fs->fs_frag - frags; 1112 ufs_add32(cgp->cg_cs.cs_nffree, i, needswap); 1113 fs->fs_cstotal.cs_nffree += i; 1114 fs->fs_cs(fs, cg).cs_nffree += i; 1115 fs->fs_fmod = 1; 1116 ufs_add32(cgp->cg_frsum[i], 1, needswap); 1117 ACTIVECG_CLR(fs, cg); 1118 mutex_exit(&ump->um_lock); 1119 bdwrite(bp); 1120 return (blkno); 1121 } 1122 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); 1123 #if 0 1124 /* 1125 * XXX fvdl mapsearch will panic, and never return -1 1126 * also: returning NULL as daddr_t ? 1127 */ 1128 if (bno < 0) 1129 goto fail; 1130 #endif 1131 for (i = 0; i < frags; i++) 1132 clrbit(blksfree, bno + i); 1133 mutex_enter(&ump->um_lock); 1134 ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap); 1135 fs->fs_cstotal.cs_nffree -= frags; 1136 fs->fs_cs(fs, cg).cs_nffree -= frags; 1137 fs->fs_fmod = 1; 1138 ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap); 1139 if (frags != allocsiz) 1140 ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap); 1141 blkno = cg * fs->fs_fpg + bno; 1142 ACTIVECG_CLR(fs, cg); 1143 mutex_exit(&ump->um_lock); 1144 if (DOINGSOFTDEP(ITOV(ip))) 1145 softdep_setup_blkmapdep(bp, fs, blkno); 1146 bdwrite(bp); 1147 return blkno; 1148 1149 fail: 1150 brelse(bp, 0); 1151 mutex_enter(&ump->um_lock); 1152 return (0); 1153 } 1154 1155 /* 1156 * Allocate a block in a cylinder group. 1157 * 1158 * This algorithm implements the following policy: 1159 * 1) allocate the requested block. 1160 * 2) allocate a rotationally optimal block in the same cylinder. 1161 * 3) allocate the next available block on the block rotor for the 1162 * specified cylinder group. 1163 * Note that this routine only allocates fs_bsize blocks; these 1164 * blocks may be fragmented by the routine that allocates them. 1165 */ 1166 static daddr_t 1167 ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref, int flags) 1168 { 1169 struct ufsmount *ump; 1170 struct fs *fs = ip->i_fs; 1171 struct cg *cgp; 1172 daddr_t blkno; 1173 int32_t bno; 1174 u_int8_t *blksfree; 1175 #ifdef FFS_EI 1176 const int needswap = UFS_FSNEEDSWAP(fs); 1177 #endif 1178 1179 ump = ip->i_ump; 1180 1181 KASSERT(mutex_owned(&ump->um_lock)); 1182 1183 cgp = (struct cg *)bp->b_data; 1184 blksfree = cg_blksfree(cgp, needswap); 1185 if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) { 1186 bpref = ufs_rw32(cgp->cg_rotor, needswap); 1187 } else { 1188 bpref = blknum(fs, bpref); 1189 bno = dtogd(fs, bpref); 1190 /* 1191 * if the requested block is available, use it 1192 */ 1193 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno))) 1194 goto gotit; 1195 /* 1196 * if the requested data block isn't available and we are 1197 * trying to allocate a contiguous file, return an error. 1198 */ 1199 if ((flags & (B_CONTIG | B_METAONLY)) == B_CONTIG) 1200 return (0); 1201 } 1202 1203 /* 1204 * Take the next available block in this cylinder group. 1205 */ 1206 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); 1207 if (bno < 0) 1208 return (0); 1209 cgp->cg_rotor = ufs_rw32(bno, needswap); 1210 gotit: 1211 blkno = fragstoblks(fs, bno); 1212 ffs_clrblock(fs, blksfree, blkno); 1213 ffs_clusteracct(fs, cgp, blkno, -1); 1214 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap); 1215 fs->fs_cstotal.cs_nbfree--; 1216 fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--; 1217 if ((fs->fs_magic == FS_UFS1_MAGIC) && 1218 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) { 1219 int cylno; 1220 cylno = old_cbtocylno(fs, bno); 1221 KASSERT(cylno >= 0); 1222 KASSERT(cylno < fs->fs_old_ncyl); 1223 KASSERT(old_cbtorpos(fs, bno) >= 0); 1224 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bno) < fs->fs_old_nrpos); 1225 ufs_add16(old_cg_blks(fs, cgp, cylno, needswap)[old_cbtorpos(fs, bno)], -1, 1226 needswap); 1227 ufs_add32(old_cg_blktot(cgp, needswap)[cylno], -1, needswap); 1228 } 1229 fs->fs_fmod = 1; 1230 blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno; 1231 if (DOINGSOFTDEP(ITOV(ip))) { 1232 mutex_exit(&ump->um_lock); 1233 softdep_setup_blkmapdep(bp, fs, blkno); 1234 mutex_enter(&ump->um_lock); 1235 } 1236 return (blkno); 1237 } 1238 1239 /* 1240 * Determine whether an inode can be allocated. 1241 * 1242 * Check to see if an inode is available, and if it is, 1243 * allocate it using the following policy: 1244 * 1) allocate the requested inode. 1245 * 2) allocate the next available inode after the requested 1246 * inode in the specified cylinder group. 1247 */ 1248 static daddr_t 1249 ffs_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode, int flags) 1250 { 1251 struct ufsmount *ump = ip->i_ump; 1252 struct fs *fs = ip->i_fs; 1253 struct cg *cgp; 1254 struct buf *bp, *ibp; 1255 u_int8_t *inosused; 1256 int error, start, len, loc, map, i; 1257 int32_t initediblk; 1258 daddr_t nalloc; 1259 struct ufs2_dinode *dp2; 1260 #ifdef FFS_EI 1261 const int needswap = UFS_FSNEEDSWAP(fs); 1262 #endif 1263 1264 KASSERT(mutex_owned(&ump->um_lock)); 1265 UFS_WAPBL_JLOCK_ASSERT(ip->i_ump->um_mountp); 1266 1267 if (fs->fs_cs(fs, cg).cs_nifree == 0) 1268 return (0); 1269 mutex_exit(&ump->um_lock); 1270 ibp = NULL; 1271 initediblk = -1; 1272 retry: 1273 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1274 (int)fs->fs_cgsize, NOCRED, B_MODIFY, &bp); 1275 if (error) 1276 goto fail; 1277 cgp = (struct cg *)bp->b_data; 1278 if (!cg_chkmagic(cgp, needswap) || cgp->cg_cs.cs_nifree == 0) 1279 goto fail; 1280 1281 if (ibp != NULL && 1282 initediblk != ufs_rw32(cgp->cg_initediblk, needswap)) { 1283 /* Another thread allocated more inodes so we retry the test. */ 1284 brelse(ibp, BC_INVAL); 1285 ibp = NULL; 1286 } 1287 /* 1288 * Check to see if we need to initialize more inodes. 1289 */ 1290 if (fs->fs_magic == FS_UFS2_MAGIC && ibp == NULL) { 1291 initediblk = ufs_rw32(cgp->cg_initediblk, needswap); 1292 nalloc = fs->fs_ipg - ufs_rw32(cgp->cg_cs.cs_nifree, needswap); 1293 if (nalloc + INOPB(fs) > initediblk && 1294 initediblk < ufs_rw32(cgp->cg_niblk, needswap)) { 1295 /* 1296 * We have to release the cg buffer here to prevent 1297 * a deadlock when reading the inode block will 1298 * run a copy-on-write that might use this cg. 1299 */ 1300 brelse(bp, 0); 1301 bp = NULL; 1302 error = ffs_getblk(ip->i_devvp, fsbtodb(fs, 1303 ino_to_fsba(fs, cg * fs->fs_ipg + initediblk)), 1304 FFS_NOBLK, fs->fs_bsize, false, &ibp); 1305 if (error) 1306 goto fail; 1307 goto retry; 1308 } 1309 } 1310 1311 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1312 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1313 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1314 cgp->cg_time = ufs_rw64(time_second, needswap); 1315 inosused = cg_inosused(cgp, needswap); 1316 if (ipref) { 1317 ipref %= fs->fs_ipg; 1318 if (isclr(inosused, ipref)) 1319 goto gotit; 1320 } 1321 start = ufs_rw32(cgp->cg_irotor, needswap) / NBBY; 1322 len = howmany(fs->fs_ipg - ufs_rw32(cgp->cg_irotor, needswap), 1323 NBBY); 1324 loc = skpc(0xff, len, &inosused[start]); 1325 if (loc == 0) { 1326 len = start + 1; 1327 start = 0; 1328 loc = skpc(0xff, len, &inosused[0]); 1329 if (loc == 0) { 1330 printf("cg = %d, irotor = %d, fs = %s\n", 1331 cg, ufs_rw32(cgp->cg_irotor, needswap), 1332 fs->fs_fsmnt); 1333 panic("ffs_nodealloccg: map corrupted"); 1334 /* NOTREACHED */ 1335 } 1336 } 1337 i = start + len - loc; 1338 map = inosused[i]; 1339 ipref = i * NBBY; 1340 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) { 1341 if ((map & i) == 0) { 1342 cgp->cg_irotor = ufs_rw32(ipref, needswap); 1343 goto gotit; 1344 } 1345 } 1346 printf("fs = %s\n", fs->fs_fsmnt); 1347 panic("ffs_nodealloccg: block not in map"); 1348 /* NOTREACHED */ 1349 gotit: 1350 UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp, cg * fs->fs_ipg + ipref, 1351 mode); 1352 /* 1353 * Check to see if we need to initialize more inodes. 1354 */ 1355 if (ibp != NULL) { 1356 KASSERT(initediblk == ufs_rw32(cgp->cg_initediblk, needswap)); 1357 memset(ibp->b_data, 0, fs->fs_bsize); 1358 dp2 = (struct ufs2_dinode *)(ibp->b_data); 1359 for (i = 0; i < INOPB(fs); i++) { 1360 /* 1361 * Don't bother to swap, it's supposed to be 1362 * random, after all. 1363 */ 1364 dp2->di_gen = (arc4random() & INT32_MAX) / 2 + 1; 1365 dp2++; 1366 } 1367 initediblk += INOPB(fs); 1368 cgp->cg_initediblk = ufs_rw32(initediblk, needswap); 1369 } 1370 1371 mutex_enter(&ump->um_lock); 1372 ACTIVECG_CLR(fs, cg); 1373 setbit(inosused, ipref); 1374 ufs_add32(cgp->cg_cs.cs_nifree, -1, needswap); 1375 fs->fs_cstotal.cs_nifree--; 1376 fs->fs_cs(fs, cg).cs_nifree--; 1377 fs->fs_fmod = 1; 1378 if ((mode & IFMT) == IFDIR) { 1379 ufs_add32(cgp->cg_cs.cs_ndir, 1, needswap); 1380 fs->fs_cstotal.cs_ndir++; 1381 fs->fs_cs(fs, cg).cs_ndir++; 1382 } 1383 mutex_exit(&ump->um_lock); 1384 if (DOINGSOFTDEP(ITOV(ip))) 1385 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref); 1386 if (ibp != NULL) { 1387 bwrite(bp); 1388 bawrite(ibp); 1389 } else 1390 bdwrite(bp); 1391 return (cg * fs->fs_ipg + ipref); 1392 fail: 1393 if (bp != NULL) 1394 brelse(bp, 0); 1395 if (ibp != NULL) 1396 brelse(ibp, BC_INVAL); 1397 mutex_enter(&ump->um_lock); 1398 return (0); 1399 } 1400 1401 /* 1402 * Allocate a block or fragment. 1403 * 1404 * The specified block or fragment is removed from the 1405 * free map, possibly fragmenting a block in the process. 1406 * 1407 * This implementation should mirror fs_blkfree 1408 * 1409 * => um_lock not held on entry or exit 1410 */ 1411 int 1412 ffs_blkalloc(struct inode *ip, daddr_t bno, long size) 1413 { 1414 int error; 1415 1416 error = ffs_check_bad_allocation(__func__, ip->i_fs, bno, size, 1417 ip->i_dev, ip->i_uid); 1418 if (error) 1419 return error; 1420 1421 return ffs_blkalloc_ump(ip->i_ump, bno, size); 1422 } 1423 1424 int 1425 ffs_blkalloc_ump(struct ufsmount *ump, daddr_t bno, long size) 1426 { 1427 struct fs *fs = ump->um_fs; 1428 struct cg *cgp; 1429 struct buf *bp; 1430 int32_t fragno, cgbno; 1431 int i, error, cg, blk, frags, bbase; 1432 u_int8_t *blksfree; 1433 const int needswap = UFS_FSNEEDSWAP(fs); 1434 1435 KASSERT((u_int)size <= fs->fs_bsize && fragoff(fs, size) == 0 && 1436 fragnum(fs, bno) + numfrags(fs, size) <= fs->fs_frag); 1437 KASSERT(bno < fs->fs_size); 1438 1439 cg = dtog(fs, bno); 1440 error = bread(ump->um_devvp, fsbtodb(fs, cgtod(fs, cg)), 1441 (int)fs->fs_cgsize, NOCRED, B_MODIFY, &bp); 1442 if (error) { 1443 brelse(bp, 0); 1444 return error; 1445 } 1446 cgp = (struct cg *)bp->b_data; 1447 if (!cg_chkmagic(cgp, needswap)) { 1448 brelse(bp, 0); 1449 return EIO; 1450 } 1451 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1452 cgp->cg_time = ufs_rw64(time_second, needswap); 1453 cgbno = dtogd(fs, bno); 1454 blksfree = cg_blksfree(cgp, needswap); 1455 1456 mutex_enter(&ump->um_lock); 1457 if (size == fs->fs_bsize) { 1458 fragno = fragstoblks(fs, cgbno); 1459 if (!ffs_isblock(fs, blksfree, fragno)) { 1460 mutex_exit(&ump->um_lock); 1461 brelse(bp, 0); 1462 return EBUSY; 1463 } 1464 ffs_clrblock(fs, blksfree, fragno); 1465 ffs_clusteracct(fs, cgp, fragno, -1); 1466 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap); 1467 fs->fs_cstotal.cs_nbfree--; 1468 fs->fs_cs(fs, cg).cs_nbfree--; 1469 } else { 1470 bbase = cgbno - fragnum(fs, cgbno); 1471 1472 frags = numfrags(fs, size); 1473 for (i = 0; i < frags; i++) { 1474 if (isclr(blksfree, cgbno + i)) { 1475 mutex_exit(&ump->um_lock); 1476 brelse(bp, 0); 1477 return EBUSY; 1478 } 1479 } 1480 /* 1481 * if a complete block is being split, account for it 1482 */ 1483 fragno = fragstoblks(fs, bbase); 1484 if (ffs_isblock(fs, blksfree, fragno)) { 1485 ufs_add32(cgp->cg_cs.cs_nffree, fs->fs_frag, needswap); 1486 fs->fs_cstotal.cs_nffree += fs->fs_frag; 1487 fs->fs_cs(fs, cg).cs_nffree += fs->fs_frag; 1488 ffs_clusteracct(fs, cgp, fragno, -1); 1489 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap); 1490 fs->fs_cstotal.cs_nbfree--; 1491 fs->fs_cs(fs, cg).cs_nbfree--; 1492 } 1493 /* 1494 * decrement the counts associated with the old frags 1495 */ 1496 blk = blkmap(fs, blksfree, bbase); 1497 ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap); 1498 /* 1499 * allocate the fragment 1500 */ 1501 for (i = 0; i < frags; i++) { 1502 clrbit(blksfree, cgbno + i); 1503 } 1504 ufs_add32(cgp->cg_cs.cs_nffree, -i, needswap); 1505 fs->fs_cstotal.cs_nffree -= i; 1506 fs->fs_cs(fs, cg).cs_nffree -= i; 1507 /* 1508 * add back in counts associated with the new frags 1509 */ 1510 blk = blkmap(fs, blksfree, bbase); 1511 ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap); 1512 } 1513 fs->fs_fmod = 1; 1514 ACTIVECG_CLR(fs, cg); 1515 mutex_exit(&ump->um_lock); 1516 bdwrite(bp); 1517 return 0; 1518 } 1519 1520 /* 1521 * Free a block or fragment. 1522 * 1523 * The specified block or fragment is placed back in the 1524 * free map. If a fragment is deallocated, a possible 1525 * block reassembly is checked. 1526 * 1527 * => um_lock not held on entry or exit 1528 */ 1529 void 1530 ffs_blkfree(struct fs *fs, struct vnode *devvp, daddr_t bno, long size, 1531 ino_t inum) 1532 { 1533 struct cg *cgp; 1534 struct buf *bp; 1535 struct ufsmount *ump; 1536 daddr_t cgblkno; 1537 int error, cg; 1538 dev_t dev; 1539 const bool devvp_is_snapshot = (devvp->v_type != VBLK); 1540 #ifdef FFS_EI 1541 const int needswap = UFS_FSNEEDSWAP(fs); 1542 #endif 1543 1544 KASSERT(!devvp_is_snapshot); 1545 1546 cg = dtog(fs, bno); 1547 dev = devvp->v_rdev; 1548 ump = VFSTOUFS(devvp->v_specmountpoint); 1549 KASSERT(fs == ump->um_fs); 1550 cgblkno = fsbtodb(fs, cgtod(fs, cg)); 1551 if (ffs_snapblkfree(fs, devvp, bno, size, inum)) 1552 return; 1553 1554 error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum); 1555 if (error) 1556 return; 1557 1558 error = bread(devvp, cgblkno, (int)fs->fs_cgsize, 1559 NOCRED, B_MODIFY, &bp); 1560 if (error) { 1561 brelse(bp, 0); 1562 return; 1563 } 1564 cgp = (struct cg *)bp->b_data; 1565 if (!cg_chkmagic(cgp, needswap)) { 1566 brelse(bp, 0); 1567 return; 1568 } 1569 1570 ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot); 1571 1572 bdwrite(bp); 1573 } 1574 1575 /* 1576 * Free a block or fragment from a snapshot cg copy. 1577 * 1578 * The specified block or fragment is placed back in the 1579 * free map. If a fragment is deallocated, a possible 1580 * block reassembly is checked. 1581 * 1582 * => um_lock not held on entry or exit 1583 */ 1584 void 1585 ffs_blkfree_snap(struct fs *fs, struct vnode *devvp, daddr_t bno, long size, 1586 ino_t inum) 1587 { 1588 struct cg *cgp; 1589 struct buf *bp; 1590 struct ufsmount *ump; 1591 daddr_t cgblkno; 1592 int error, cg; 1593 dev_t dev; 1594 const bool devvp_is_snapshot = (devvp->v_type != VBLK); 1595 #ifdef FFS_EI 1596 const int needswap = UFS_FSNEEDSWAP(fs); 1597 #endif 1598 1599 KASSERT(devvp_is_snapshot); 1600 1601 cg = dtog(fs, bno); 1602 dev = VTOI(devvp)->i_devvp->v_rdev; 1603 ump = VFSTOUFS(devvp->v_mount); 1604 cgblkno = fragstoblks(fs, cgtod(fs, cg)); 1605 1606 error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum); 1607 if (error) 1608 return; 1609 1610 error = bread(devvp, cgblkno, (int)fs->fs_cgsize, 1611 NOCRED, B_MODIFY, &bp); 1612 if (error) { 1613 brelse(bp, 0); 1614 return; 1615 } 1616 cgp = (struct cg *)bp->b_data; 1617 if (!cg_chkmagic(cgp, needswap)) { 1618 brelse(bp, 0); 1619 return; 1620 } 1621 1622 ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot); 1623 1624 bdwrite(bp); 1625 } 1626 1627 static void 1628 ffs_blkfree_common(struct ufsmount *ump, struct fs *fs, dev_t dev, 1629 struct buf *bp, daddr_t bno, long size, bool devvp_is_snapshot) 1630 { 1631 struct cg *cgp; 1632 int32_t fragno, cgbno; 1633 int i, cg, blk, frags, bbase; 1634 u_int8_t *blksfree; 1635 const int needswap = UFS_FSNEEDSWAP(fs); 1636 1637 cg = dtog(fs, bno); 1638 cgp = (struct cg *)bp->b_data; 1639 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1640 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1641 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1642 cgp->cg_time = ufs_rw64(time_second, needswap); 1643 cgbno = dtogd(fs, bno); 1644 blksfree = cg_blksfree(cgp, needswap); 1645 mutex_enter(&ump->um_lock); 1646 if (size == fs->fs_bsize) { 1647 fragno = fragstoblks(fs, cgbno); 1648 if (!ffs_isfreeblock(fs, blksfree, fragno)) { 1649 if (devvp_is_snapshot) { 1650 mutex_exit(&ump->um_lock); 1651 return; 1652 } 1653 printf("dev = 0x%x, block = %" PRId64 ", fs = %s\n", 1654 dev, bno, fs->fs_fsmnt); 1655 panic("blkfree: freeing free block"); 1656 } 1657 ffs_setblock(fs, blksfree, fragno); 1658 ffs_clusteracct(fs, cgp, fragno, 1); 1659 ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap); 1660 fs->fs_cstotal.cs_nbfree++; 1661 fs->fs_cs(fs, cg).cs_nbfree++; 1662 if ((fs->fs_magic == FS_UFS1_MAGIC) && 1663 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) { 1664 i = old_cbtocylno(fs, cgbno); 1665 KASSERT(i >= 0); 1666 KASSERT(i < fs->fs_old_ncyl); 1667 KASSERT(old_cbtorpos(fs, cgbno) >= 0); 1668 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, cgbno) < fs->fs_old_nrpos); 1669 ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs, cgbno)], 1, 1670 needswap); 1671 ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap); 1672 } 1673 } else { 1674 bbase = cgbno - fragnum(fs, cgbno); 1675 /* 1676 * decrement the counts associated with the old frags 1677 */ 1678 blk = blkmap(fs, blksfree, bbase); 1679 ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap); 1680 /* 1681 * deallocate the fragment 1682 */ 1683 frags = numfrags(fs, size); 1684 for (i = 0; i < frags; i++) { 1685 if (isset(blksfree, cgbno + i)) { 1686 printf("dev = 0x%x, block = %" PRId64 1687 ", fs = %s\n", 1688 dev, bno + i, fs->fs_fsmnt); 1689 panic("blkfree: freeing free frag"); 1690 } 1691 setbit(blksfree, cgbno + i); 1692 } 1693 ufs_add32(cgp->cg_cs.cs_nffree, i, needswap); 1694 fs->fs_cstotal.cs_nffree += i; 1695 fs->fs_cs(fs, cg).cs_nffree += i; 1696 /* 1697 * add back in counts associated with the new frags 1698 */ 1699 blk = blkmap(fs, blksfree, bbase); 1700 ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap); 1701 /* 1702 * if a complete block has been reassembled, account for it 1703 */ 1704 fragno = fragstoblks(fs, bbase); 1705 if (ffs_isblock(fs, blksfree, fragno)) { 1706 ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap); 1707 fs->fs_cstotal.cs_nffree -= fs->fs_frag; 1708 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; 1709 ffs_clusteracct(fs, cgp, fragno, 1); 1710 ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap); 1711 fs->fs_cstotal.cs_nbfree++; 1712 fs->fs_cs(fs, cg).cs_nbfree++; 1713 if ((fs->fs_magic == FS_UFS1_MAGIC) && 1714 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) { 1715 i = old_cbtocylno(fs, bbase); 1716 KASSERT(i >= 0); 1717 KASSERT(i < fs->fs_old_ncyl); 1718 KASSERT(old_cbtorpos(fs, bbase) >= 0); 1719 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bbase) < fs->fs_old_nrpos); 1720 ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs, 1721 bbase)], 1, needswap); 1722 ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap); 1723 } 1724 } 1725 } 1726 fs->fs_fmod = 1; 1727 ACTIVECG_CLR(fs, cg); 1728 mutex_exit(&ump->um_lock); 1729 } 1730 1731 /* 1732 * Free an inode. 1733 */ 1734 int 1735 ffs_vfree(struct vnode *vp, ino_t ino, int mode) 1736 { 1737 1738 if (DOINGSOFTDEP(vp)) { 1739 softdep_freefile(vp, ino, mode); 1740 return (0); 1741 } 1742 return ffs_freefile(vp->v_mount, ino, mode); 1743 } 1744 1745 /* 1746 * Do the actual free operation. 1747 * The specified inode is placed back in the free map. 1748 * 1749 * => um_lock not held on entry or exit 1750 */ 1751 int 1752 ffs_freefile(struct mount *mp, ino_t ino, int mode) 1753 { 1754 struct ufsmount *ump = VFSTOUFS(mp); 1755 struct fs *fs = ump->um_fs; 1756 struct vnode *devvp; 1757 struct cg *cgp; 1758 struct buf *bp; 1759 int error, cg; 1760 daddr_t cgbno; 1761 dev_t dev; 1762 #ifdef FFS_EI 1763 const int needswap = UFS_FSNEEDSWAP(fs); 1764 #endif 1765 1766 cg = ino_to_cg(fs, ino); 1767 devvp = ump->um_devvp; 1768 dev = devvp->v_rdev; 1769 cgbno = fsbtodb(fs, cgtod(fs, cg)); 1770 1771 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 1772 panic("ifree: range: dev = 0x%x, ino = %llu, fs = %s", 1773 dev, (unsigned long long)ino, fs->fs_fsmnt); 1774 error = bread(devvp, cgbno, (int)fs->fs_cgsize, 1775 NOCRED, B_MODIFY, &bp); 1776 if (error) { 1777 brelse(bp, 0); 1778 return (error); 1779 } 1780 cgp = (struct cg *)bp->b_data; 1781 if (!cg_chkmagic(cgp, needswap)) { 1782 brelse(bp, 0); 1783 return (0); 1784 } 1785 1786 ffs_freefile_common(ump, fs, dev, bp, ino, mode, false); 1787 1788 bdwrite(bp); 1789 1790 return 0; 1791 } 1792 1793 int 1794 ffs_freefile_snap(struct fs *fs, struct vnode *devvp, ino_t ino, int mode) 1795 { 1796 struct ufsmount *ump; 1797 struct cg *cgp; 1798 struct buf *bp; 1799 int error, cg; 1800 daddr_t cgbno; 1801 dev_t dev; 1802 #ifdef FFS_EI 1803 const int needswap = UFS_FSNEEDSWAP(fs); 1804 #endif 1805 1806 KASSERT(devvp->v_type != VBLK); 1807 1808 cg = ino_to_cg(fs, ino); 1809 dev = VTOI(devvp)->i_devvp->v_rdev; 1810 ump = VFSTOUFS(devvp->v_mount); 1811 cgbno = fragstoblks(fs, cgtod(fs, cg)); 1812 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 1813 panic("ifree: range: dev = 0x%x, ino = %llu, fs = %s", 1814 dev, (unsigned long long)ino, fs->fs_fsmnt); 1815 error = bread(devvp, cgbno, (int)fs->fs_cgsize, 1816 NOCRED, B_MODIFY, &bp); 1817 if (error) { 1818 brelse(bp, 0); 1819 return (error); 1820 } 1821 cgp = (struct cg *)bp->b_data; 1822 if (!cg_chkmagic(cgp, needswap)) { 1823 brelse(bp, 0); 1824 return (0); 1825 } 1826 ffs_freefile_common(ump, fs, dev, bp, ino, mode, true); 1827 1828 bdwrite(bp); 1829 1830 return 0; 1831 } 1832 1833 static void 1834 ffs_freefile_common(struct ufsmount *ump, struct fs *fs, dev_t dev, 1835 struct buf *bp, ino_t ino, int mode, bool devvp_is_snapshot) 1836 { 1837 int cg; 1838 struct cg *cgp; 1839 u_int8_t *inosused; 1840 #ifdef FFS_EI 1841 const int needswap = UFS_FSNEEDSWAP(fs); 1842 #endif 1843 1844 cg = ino_to_cg(fs, ino); 1845 cgp = (struct cg *)bp->b_data; 1846 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1847 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1848 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1849 cgp->cg_time = ufs_rw64(time_second, needswap); 1850 inosused = cg_inosused(cgp, needswap); 1851 ino %= fs->fs_ipg; 1852 if (isclr(inosused, ino)) { 1853 printf("ifree: dev = 0x%x, ino = %llu, fs = %s\n", 1854 dev, (unsigned long long)ino + cg * fs->fs_ipg, 1855 fs->fs_fsmnt); 1856 if (fs->fs_ronly == 0) 1857 panic("ifree: freeing free inode"); 1858 } 1859 clrbit(inosused, ino); 1860 if (!devvp_is_snapshot) 1861 UFS_WAPBL_UNREGISTER_INODE(ump->um_mountp, 1862 ino + cg * fs->fs_ipg, mode); 1863 if (ino < ufs_rw32(cgp->cg_irotor, needswap)) 1864 cgp->cg_irotor = ufs_rw32(ino, needswap); 1865 ufs_add32(cgp->cg_cs.cs_nifree, 1, needswap); 1866 mutex_enter(&ump->um_lock); 1867 fs->fs_cstotal.cs_nifree++; 1868 fs->fs_cs(fs, cg).cs_nifree++; 1869 if ((mode & IFMT) == IFDIR) { 1870 ufs_add32(cgp->cg_cs.cs_ndir, -1, needswap); 1871 fs->fs_cstotal.cs_ndir--; 1872 fs->fs_cs(fs, cg).cs_ndir--; 1873 } 1874 fs->fs_fmod = 1; 1875 ACTIVECG_CLR(fs, cg); 1876 mutex_exit(&ump->um_lock); 1877 } 1878 1879 /* 1880 * Check to see if a file is free. 1881 */ 1882 int 1883 ffs_checkfreefile(struct fs *fs, struct vnode *devvp, ino_t ino) 1884 { 1885 struct cg *cgp; 1886 struct buf *bp; 1887 daddr_t cgbno; 1888 int ret, cg; 1889 u_int8_t *inosused; 1890 const bool devvp_is_snapshot = (devvp->v_type != VBLK); 1891 1892 KASSERT(devvp_is_snapshot); 1893 1894 cg = ino_to_cg(fs, ino); 1895 if (devvp_is_snapshot) 1896 cgbno = fragstoblks(fs, cgtod(fs, cg)); 1897 else 1898 cgbno = fsbtodb(fs, cgtod(fs, cg)); 1899 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 1900 return 1; 1901 if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, 0, &bp)) { 1902 brelse(bp, 0); 1903 return 1; 1904 } 1905 cgp = (struct cg *)bp->b_data; 1906 if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs))) { 1907 brelse(bp, 0); 1908 return 1; 1909 } 1910 inosused = cg_inosused(cgp, UFS_FSNEEDSWAP(fs)); 1911 ino %= fs->fs_ipg; 1912 ret = isclr(inosused, ino); 1913 brelse(bp, 0); 1914 return ret; 1915 } 1916 1917 /* 1918 * Find a block of the specified size in the specified cylinder group. 1919 * 1920 * It is a panic if a request is made to find a block if none are 1921 * available. 1922 */ 1923 static int32_t 1924 ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz) 1925 { 1926 int32_t bno; 1927 int start, len, loc, i; 1928 int blk, field, subfield, pos; 1929 int ostart, olen; 1930 u_int8_t *blksfree; 1931 #ifdef FFS_EI 1932 const int needswap = UFS_FSNEEDSWAP(fs); 1933 #endif 1934 1935 /* KASSERT(mutex_owned(&ump->um_lock)); */ 1936 1937 /* 1938 * find the fragment by searching through the free block 1939 * map for an appropriate bit pattern 1940 */ 1941 if (bpref) 1942 start = dtogd(fs, bpref) / NBBY; 1943 else 1944 start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY; 1945 blksfree = cg_blksfree(cgp, needswap); 1946 len = howmany(fs->fs_fpg, NBBY) - start; 1947 ostart = start; 1948 olen = len; 1949 loc = scanc((u_int)len, 1950 (const u_char *)&blksfree[start], 1951 (const u_char *)fragtbl[fs->fs_frag], 1952 (1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1))))); 1953 if (loc == 0) { 1954 len = start + 1; 1955 start = 0; 1956 loc = scanc((u_int)len, 1957 (const u_char *)&blksfree[0], 1958 (const u_char *)fragtbl[fs->fs_frag], 1959 (1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1))))); 1960 if (loc == 0) { 1961 printf("start = %d, len = %d, fs = %s\n", 1962 ostart, olen, fs->fs_fsmnt); 1963 printf("offset=%d %ld\n", 1964 ufs_rw32(cgp->cg_freeoff, needswap), 1965 (long)blksfree - (long)cgp); 1966 printf("cg %d\n", cgp->cg_cgx); 1967 panic("ffs_alloccg: map corrupted"); 1968 /* NOTREACHED */ 1969 } 1970 } 1971 bno = (start + len - loc) * NBBY; 1972 cgp->cg_frotor = ufs_rw32(bno, needswap); 1973 /* 1974 * found the byte in the map 1975 * sift through the bits to find the selected frag 1976 */ 1977 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { 1978 blk = blkmap(fs, blksfree, bno); 1979 blk <<= 1; 1980 field = around[allocsiz]; 1981 subfield = inside[allocsiz]; 1982 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { 1983 if ((blk & field) == subfield) 1984 return (bno + pos); 1985 field <<= 1; 1986 subfield <<= 1; 1987 } 1988 } 1989 printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt); 1990 panic("ffs_alloccg: block not in map"); 1991 /* return (-1); */ 1992 } 1993 1994 /* 1995 * Update the cluster map because of an allocation or free. 1996 * 1997 * Cnt == 1 means free; cnt == -1 means allocating. 1998 */ 1999 void 2000 ffs_clusteracct(struct fs *fs, struct cg *cgp, int32_t blkno, int cnt) 2001 { 2002 int32_t *sump; 2003 int32_t *lp; 2004 u_char *freemapp, *mapp; 2005 int i, start, end, forw, back, map, bit; 2006 #ifdef FFS_EI 2007 const int needswap = UFS_FSNEEDSWAP(fs); 2008 #endif 2009 2010 /* KASSERT(mutex_owned(&ump->um_lock)); */ 2011 2012 if (fs->fs_contigsumsize <= 0) 2013 return; 2014 freemapp = cg_clustersfree(cgp, needswap); 2015 sump = cg_clustersum(cgp, needswap); 2016 /* 2017 * Allocate or clear the actual block. 2018 */ 2019 if (cnt > 0) 2020 setbit(freemapp, blkno); 2021 else 2022 clrbit(freemapp, blkno); 2023 /* 2024 * Find the size of the cluster going forward. 2025 */ 2026 start = blkno + 1; 2027 end = start + fs->fs_contigsumsize; 2028 if (end >= ufs_rw32(cgp->cg_nclusterblks, needswap)) 2029 end = ufs_rw32(cgp->cg_nclusterblks, needswap); 2030 mapp = &freemapp[start / NBBY]; 2031 map = *mapp++; 2032 bit = 1 << (start % NBBY); 2033 for (i = start; i < end; i++) { 2034 if ((map & bit) == 0) 2035 break; 2036 if ((i & (NBBY - 1)) != (NBBY - 1)) { 2037 bit <<= 1; 2038 } else { 2039 map = *mapp++; 2040 bit = 1; 2041 } 2042 } 2043 forw = i - start; 2044 /* 2045 * Find the size of the cluster going backward. 2046 */ 2047 start = blkno - 1; 2048 end = start - fs->fs_contigsumsize; 2049 if (end < 0) 2050 end = -1; 2051 mapp = &freemapp[start / NBBY]; 2052 map = *mapp--; 2053 bit = 1 << (start % NBBY); 2054 for (i = start; i > end; i--) { 2055 if ((map & bit) == 0) 2056 break; 2057 if ((i & (NBBY - 1)) != 0) { 2058 bit >>= 1; 2059 } else { 2060 map = *mapp--; 2061 bit = 1 << (NBBY - 1); 2062 } 2063 } 2064 back = start - i; 2065 /* 2066 * Account for old cluster and the possibly new forward and 2067 * back clusters. 2068 */ 2069 i = back + forw + 1; 2070 if (i > fs->fs_contigsumsize) 2071 i = fs->fs_contigsumsize; 2072 ufs_add32(sump[i], cnt, needswap); 2073 if (back > 0) 2074 ufs_add32(sump[back], -cnt, needswap); 2075 if (forw > 0) 2076 ufs_add32(sump[forw], -cnt, needswap); 2077 2078 /* 2079 * Update cluster summary information. 2080 */ 2081 lp = &sump[fs->fs_contigsumsize]; 2082 for (i = fs->fs_contigsumsize; i > 0; i--) 2083 if (ufs_rw32(*lp--, needswap) > 0) 2084 break; 2085 fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i; 2086 } 2087 2088 /* 2089 * Fserr prints the name of a file system with an error diagnostic. 2090 * 2091 * The form of the error message is: 2092 * fs: error message 2093 */ 2094 static void 2095 ffs_fserr(struct fs *fs, u_int uid, const char *cp) 2096 { 2097 2098 log(LOG_ERR, "uid %d, pid %d, command %s, on %s: %s\n", 2099 uid, curproc->p_pid, curproc->p_comm, fs->fs_fsmnt, cp); 2100 } 2101
Indexes created Mon Sep 29 21:09:56 GMT 2025