ffs_alloc.c revision 1.130.8.1
1 /* $NetBSD: ffs_alloc.c,v 1.130.8.1 2012/11/20 03:02:53 tls Exp $ */ 2 3 /*- 4 * Copyright (c) 2008, 2009 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 1.130.8.1 2012/11/20 03:02:53 tls Exp $"); 74 75 #if defined(_KERNEL_OPT) 76 #include "opt_ffs.h" 77 #include "opt_quota.h" 78 #include "opt_uvm_page_trkown.h" 79 #endif 80 81 #include <sys/param.h> 82 #include <sys/systm.h> 83 #include <sys/buf.h> 84 #include <sys/cprng.h> 85 #include <sys/fstrans.h> 86 #include <sys/kauth.h> 87 #include <sys/kernel.h> 88 #include <sys/mount.h> 89 #include <sys/proc.h> 90 #include <sys/syslog.h> 91 #include <sys/vnode.h> 92 #include <sys/wapbl.h> 93 94 #include <miscfs/specfs/specdev.h> 95 #include <ufs/ufs/quota.h> 96 #include <ufs/ufs/ufsmount.h> 97 #include <ufs/ufs/inode.h> 98 #include <ufs/ufs/ufs_extern.h> 99 #include <ufs/ufs/ufs_bswap.h> 100 #include <ufs/ufs/ufs_wapbl.h> 101 102 #include <ufs/ffs/fs.h> 103 #include <ufs/ffs/ffs_extern.h> 104 105 #ifdef UVM_PAGE_TRKOWN 106 #include <uvm/uvm.h> 107 #endif 108 109 static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int, int); 110 static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t, int); 111 static ino_t ffs_dirpref(struct inode *); 112 static daddr_t ffs_fragextend(struct inode *, int, daddr_t, int, int); 113 static void ffs_fserr(struct fs *, u_int, const char *); 114 static daddr_t ffs_hashalloc(struct inode *, int, daddr_t, int, int, 115 daddr_t (*)(struct inode *, int, daddr_t, int, int)); 116 static daddr_t ffs_nodealloccg(struct inode *, int, daddr_t, int, int); 117 static int32_t ffs_mapsearch(struct fs *, struct cg *, 118 daddr_t, int); 119 static void ffs_blkfree_common(struct ufsmount *, struct fs *, dev_t, struct buf *, 120 daddr_t, long, bool); 121 static void ffs_freefile_common(struct ufsmount *, struct fs *, dev_t, struct buf *, ino_t, 122 int, bool); 123 124 /* if 1, changes in optimalization strategy are logged */ 125 int ffs_log_changeopt = 0; 126 127 /* in ffs_tables.c */ 128 extern const int inside[], around[]; 129 extern const u_char * const fragtbl[]; 130 131 /* Basic consistency check for block allocations */ 132 static int 133 ffs_check_bad_allocation(const char *func, struct fs *fs, daddr_t bno, 134 long size, dev_t dev, ino_t inum) 135 { 136 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || 137 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { 138 printf("dev = 0x%llx, bno = %" PRId64 " bsize = %d, " 139 "size = %ld, fs = %s\n", 140 (long long)dev, bno, fs->fs_bsize, size, fs->fs_fsmnt); 141 panic("%s: bad size", func); 142 } 143 144 if (bno >= fs->fs_size) { 145 printf("bad block %" PRId64 ", ino %llu\n", bno, 146 (unsigned long long)inum); 147 ffs_fserr(fs, inum, "bad block"); 148 return EINVAL; 149 } 150 return 0; 151 } 152 153 /* 154 * Allocate a block in the file system. 155 * 156 * The size of the requested block is given, which must be some 157 * multiple of fs_fsize and <= fs_bsize. 158 * A preference may be optionally specified. If a preference is given 159 * the following hierarchy is used to allocate a block: 160 * 1) allocate the requested block. 161 * 2) allocate a rotationally optimal block in the same cylinder. 162 * 3) allocate a block in the same cylinder group. 163 * 4) quadradically rehash into other cylinder groups, until an 164 * available block is located. 165 * If no block preference is given the following hierarchy is used 166 * to allocate a block: 167 * 1) allocate a block in the cylinder group that contains the 168 * inode for the file. 169 * 2) quadradically rehash into other cylinder groups, until an 170 * available block is located. 171 * 172 * => called with um_lock held 173 * => releases um_lock before returning 174 */ 175 int 176 ffs_alloc(struct inode *ip, daddr_t lbn, daddr_t bpref, int size, int flags, 177 kauth_cred_t cred, daddr_t *bnp) 178 { 179 struct ufsmount *ump; 180 struct fs *fs; 181 daddr_t bno; 182 int cg; 183 #if defined(QUOTA) || defined(QUOTA2) 184 int error; 185 #endif 186 187 fs = ip->i_fs; 188 ump = ip->i_ump; 189 190 KASSERT(mutex_owned(&ump->um_lock)); 191 192 #ifdef UVM_PAGE_TRKOWN 193 194 /* 195 * Sanity-check that allocations within the file size 196 * do not allow other threads to read the stale contents 197 * of newly allocated blocks. 198 * Usually pages will exist to cover the new allocation. 199 * There is an optimization in ffs_write() where we skip 200 * creating pages if several conditions are met: 201 * - the file must not be mapped (in any user address space). 202 * - the write must cover whole pages and whole blocks. 203 * If those conditions are not met then pages must exist and 204 * be locked by the current thread. 205 */ 206 207 if (ITOV(ip)->v_type == VREG && 208 lblktosize(fs, (voff_t)lbn) < round_page(ITOV(ip)->v_size)) { 209 struct vm_page *pg; 210 struct vnode *vp = ITOV(ip); 211 struct uvm_object *uobj = &vp->v_uobj; 212 voff_t off = trunc_page(lblktosize(fs, lbn)); 213 voff_t endoff = round_page(lblktosize(fs, lbn) + size); 214 215 mutex_enter(uobj->vmobjlock); 216 while (off < endoff) { 217 pg = uvm_pagelookup(uobj, off); 218 KASSERT((pg == NULL && (vp->v_vflag & VV_MAPPED) == 0 && 219 (size & PAGE_MASK) == 0 && 220 blkoff(fs, size) == 0) || 221 (pg != NULL && pg->owner == curproc->p_pid && 222 pg->lowner == curlwp->l_lid)); 223 off += PAGE_SIZE; 224 } 225 mutex_exit(uobj->vmobjlock); 226 } 227 #endif 228 229 *bnp = 0; 230 #ifdef DIAGNOSTIC 231 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { 232 printf("dev = 0x%llx, bsize = %d, size = %d, fs = %s\n", 233 (unsigned long long)ip->i_dev, fs->fs_bsize, size, 234 fs->fs_fsmnt); 235 panic("ffs_alloc: bad size"); 236 } 237 if (cred == NOCRED) 238 panic("ffs_alloc: missing credential"); 239 #endif /* DIAGNOSTIC */ 240 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) 241 goto nospace; 242 if (freespace(fs, fs->fs_minfree) <= 0 && 243 kauth_authorize_system(cred, KAUTH_SYSTEM_FS_RESERVEDSPACE, 0, NULL, 244 NULL, NULL) != 0) 245 goto nospace; 246 #if defined(QUOTA) || defined(QUOTA2) 247 mutex_exit(&ump->um_lock); 248 if ((error = chkdq(ip, btodb(size), cred, 0)) != 0) 249 return (error); 250 mutex_enter(&ump->um_lock); 251 #endif 252 253 if (bpref >= fs->fs_size) 254 bpref = 0; 255 if (bpref == 0) 256 cg = ino_to_cg(fs, ip->i_number); 257 else 258 cg = dtog(fs, bpref); 259 bno = ffs_hashalloc(ip, cg, bpref, size, flags, ffs_alloccg); 260 if (bno > 0) { 261 DIP_ADD(ip, blocks, btodb(size)); 262 ip->i_flag |= IN_CHANGE | IN_UPDATE; 263 *bnp = bno; 264 return (0); 265 } 266 #if defined(QUOTA) || defined(QUOTA2) 267 /* 268 * Restore user's disk quota because allocation failed. 269 */ 270 (void) chkdq(ip, -btodb(size), cred, FORCE); 271 #endif 272 if (flags & B_CONTIG) { 273 /* 274 * XXX ump->um_lock handling is "suspect" at best. 275 * For the case where ffs_hashalloc() fails early 276 * in the B_CONTIG case we reach here with um_lock 277 * already unlocked, so we can't release it again 278 * like in the normal error path. See kern/39206. 279 * 280 * 281 * Fail silently - it's up to our caller to report 282 * errors. 283 */ 284 return (ENOSPC); 285 } 286 nospace: 287 mutex_exit(&ump->um_lock); 288 ffs_fserr(fs, kauth_cred_geteuid(cred), "file system full"); 289 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 290 return (ENOSPC); 291 } 292 293 /* 294 * Reallocate a fragment to a bigger size 295 * 296 * The number and size of the old block is given, and a preference 297 * and new size is also specified. The allocator attempts to extend 298 * the original block. Failing that, the regular block allocator is 299 * invoked to get an appropriate block. 300 * 301 * => called with um_lock held 302 * => return with um_lock released 303 */ 304 int 305 ffs_realloccg(struct inode *ip, daddr_t lbprev, daddr_t bpref, int osize, 306 int nsize, kauth_cred_t cred, struct buf **bpp, daddr_t *blknop) 307 { 308 struct ufsmount *ump; 309 struct fs *fs; 310 struct buf *bp; 311 int cg, request, error; 312 daddr_t bprev, bno; 313 314 fs = ip->i_fs; 315 ump = ip->i_ump; 316 317 KASSERT(mutex_owned(&ump->um_lock)); 318 319 #ifdef UVM_PAGE_TRKOWN 320 321 /* 322 * Sanity-check that allocations within the file size 323 * do not allow other threads to read the stale contents 324 * of newly allocated blocks. 325 * Unlike in ffs_alloc(), here pages must always exist 326 * for such allocations, because only the last block of a file 327 * can be a fragment and ffs_write() will reallocate the 328 * fragment to the new size using ufs_balloc_range(), 329 * which always creates pages to cover blocks it allocates. 330 */ 331 332 if (ITOV(ip)->v_type == VREG) { 333 struct vm_page *pg; 334 struct uvm_object *uobj = &ITOV(ip)->v_uobj; 335 voff_t off = trunc_page(lblktosize(fs, lbprev)); 336 voff_t endoff = round_page(lblktosize(fs, lbprev) + osize); 337 338 mutex_enter(uobj->vmobjlock); 339 while (off < endoff) { 340 pg = uvm_pagelookup(uobj, off); 341 KASSERT(pg->owner == curproc->p_pid && 342 pg->lowner == curlwp->l_lid); 343 off += PAGE_SIZE; 344 } 345 mutex_exit(uobj->vmobjlock); 346 } 347 #endif 348 349 #ifdef DIAGNOSTIC 350 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || 351 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { 352 printf( 353 "dev = 0x%llx, bsize = %d, osize = %d, nsize = %d, fs = %s\n", 354 (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize, 355 fs->fs_fsmnt); 356 panic("ffs_realloccg: bad size"); 357 } 358 if (cred == NOCRED) 359 panic("ffs_realloccg: missing credential"); 360 #endif /* DIAGNOSTIC */ 361 if (freespace(fs, fs->fs_minfree) <= 0 && 362 kauth_authorize_system(cred, KAUTH_SYSTEM_FS_RESERVEDSPACE, 0, NULL, 363 NULL, NULL) != 0) { 364 mutex_exit(&ump->um_lock); 365 goto nospace; 366 } 367 if (fs->fs_magic == FS_UFS2_MAGIC) 368 bprev = ufs_rw64(ip->i_ffs2_db[lbprev], UFS_FSNEEDSWAP(fs)); 369 else 370 bprev = ufs_rw32(ip->i_ffs1_db[lbprev], UFS_FSNEEDSWAP(fs)); 371 372 if (bprev == 0) { 373 printf("dev = 0x%llx, bsize = %d, bprev = %" PRId64 ", fs = %s\n", 374 (unsigned long long)ip->i_dev, fs->fs_bsize, bprev, 375 fs->fs_fsmnt); 376 panic("ffs_realloccg: bad bprev"); 377 } 378 mutex_exit(&ump->um_lock); 379 380 /* 381 * Allocate the extra space in the buffer. 382 */ 383 if (bpp != NULL && 384 (error = bread(ITOV(ip), lbprev, osize, NOCRED, 0, &bp)) != 0) { 385 brelse(bp, 0); 386 return (error); 387 } 388 #if defined(QUOTA) || defined(QUOTA2) 389 if ((error = chkdq(ip, btodb(nsize - osize), cred, 0)) != 0) { 390 if (bpp != NULL) { 391 brelse(bp, 0); 392 } 393 return (error); 394 } 395 #endif 396 /* 397 * Check for extension in the existing location. 398 */ 399 cg = dtog(fs, bprev); 400 mutex_enter(&ump->um_lock); 401 if ((bno = ffs_fragextend(ip, cg, bprev, osize, nsize)) != 0) { 402 DIP_ADD(ip, blocks, btodb(nsize - osize)); 403 ip->i_flag |= IN_CHANGE | IN_UPDATE; 404 405 if (bpp != NULL) { 406 if (bp->b_blkno != fsbtodb(fs, bno)) 407 panic("bad blockno"); 408 allocbuf(bp, nsize, 1); 409 memset((char *)bp->b_data + osize, 0, nsize - osize); 410 mutex_enter(bp->b_objlock); 411 KASSERT(!cv_has_waiters(&bp->b_done)); 412 bp->b_oflags |= BO_DONE; 413 mutex_exit(bp->b_objlock); 414 *bpp = bp; 415 } 416 if (blknop != NULL) { 417 *blknop = bno; 418 } 419 return (0); 420 } 421 /* 422 * Allocate a new disk location. 423 */ 424 if (bpref >= fs->fs_size) 425 bpref = 0; 426 switch ((int)fs->fs_optim) { 427 case FS_OPTSPACE: 428 /* 429 * Allocate an exact sized fragment. Although this makes 430 * best use of space, we will waste time relocating it if 431 * the file continues to grow. If the fragmentation is 432 * less than half of the minimum free reserve, we choose 433 * to begin optimizing for time. 434 */ 435 request = nsize; 436 if (fs->fs_minfree < 5 || 437 fs->fs_cstotal.cs_nffree > 438 fs->fs_dsize * fs->fs_minfree / (2 * 100)) 439 break; 440 441 if (ffs_log_changeopt) { 442 log(LOG_NOTICE, 443 "%s: optimization changed from SPACE to TIME\n", 444 fs->fs_fsmnt); 445 } 446 447 fs->fs_optim = FS_OPTTIME; 448 break; 449 case FS_OPTTIME: 450 /* 451 * At this point we have discovered a file that is trying to 452 * grow a small fragment to a larger fragment. To save time, 453 * we allocate a full sized block, then free the unused portion. 454 * If the file continues to grow, the `ffs_fragextend' call 455 * above will be able to grow it in place without further 456 * copying. If aberrant programs cause disk fragmentation to 457 * grow within 2% of the free reserve, we choose to begin 458 * optimizing for space. 459 */ 460 request = fs->fs_bsize; 461 if (fs->fs_cstotal.cs_nffree < 462 fs->fs_dsize * (fs->fs_minfree - 2) / 100) 463 break; 464 465 if (ffs_log_changeopt) { 466 log(LOG_NOTICE, 467 "%s: optimization changed from TIME to SPACE\n", 468 fs->fs_fsmnt); 469 } 470 471 fs->fs_optim = FS_OPTSPACE; 472 break; 473 default: 474 printf("dev = 0x%llx, optim = %d, fs = %s\n", 475 (unsigned long long)ip->i_dev, fs->fs_optim, fs->fs_fsmnt); 476 panic("ffs_realloccg: bad optim"); 477 /* NOTREACHED */ 478 } 479 bno = ffs_hashalloc(ip, cg, bpref, request, 0, ffs_alloccg); 480 if (bno > 0) { 481 if ((ip->i_ump->um_mountp->mnt_wapbl) && 482 (ITOV(ip)->v_type != VREG)) { 483 UFS_WAPBL_REGISTER_DEALLOCATION( 484 ip->i_ump->um_mountp, fsbtodb(fs, bprev), 485 osize); 486 } else { 487 ffs_blkfree(fs, ip->i_devvp, bprev, (long)osize, 488 ip->i_number); 489 } 490 if (nsize < request) { 491 if ((ip->i_ump->um_mountp->mnt_wapbl) && 492 (ITOV(ip)->v_type != VREG)) { 493 UFS_WAPBL_REGISTER_DEALLOCATION( 494 ip->i_ump->um_mountp, 495 fsbtodb(fs, (bno + numfrags(fs, nsize))), 496 request - nsize); 497 } else 498 ffs_blkfree(fs, ip->i_devvp, 499 bno + numfrags(fs, nsize), 500 (long)(request - nsize), ip->i_number); 501 } 502 DIP_ADD(ip, blocks, btodb(nsize - osize)); 503 ip->i_flag |= IN_CHANGE | IN_UPDATE; 504 if (bpp != NULL) { 505 bp->b_blkno = fsbtodb(fs, bno); 506 allocbuf(bp, nsize, 1); 507 memset((char *)bp->b_data + osize, 0, (u_int)nsize - osize); 508 mutex_enter(bp->b_objlock); 509 KASSERT(!cv_has_waiters(&bp->b_done)); 510 bp->b_oflags |= BO_DONE; 511 mutex_exit(bp->b_objlock); 512 *bpp = bp; 513 } 514 if (blknop != NULL) { 515 *blknop = bno; 516 } 517 return (0); 518 } 519 mutex_exit(&ump->um_lock); 520 521 #if defined(QUOTA) || defined(QUOTA2) 522 /* 523 * Restore user's disk quota because allocation failed. 524 */ 525 (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE); 526 #endif 527 if (bpp != NULL) { 528 brelse(bp, 0); 529 } 530 531 nospace: 532 /* 533 * no space available 534 */ 535 ffs_fserr(fs, kauth_cred_geteuid(cred), "file system full"); 536 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 537 return (ENOSPC); 538 } 539 540 /* 541 * Allocate an inode in the file system. 542 * 543 * If allocating a directory, use ffs_dirpref to select the inode. 544 * If allocating in a directory, the following hierarchy is followed: 545 * 1) allocate the preferred inode. 546 * 2) allocate an inode in the same cylinder group. 547 * 3) quadradically rehash into other cylinder groups, until an 548 * available inode is located. 549 * If no inode preference is given the following hierarchy is used 550 * to allocate an inode: 551 * 1) allocate an inode in cylinder group 0. 552 * 2) quadradically rehash into other cylinder groups, until an 553 * available inode is located. 554 * 555 * => um_lock not held upon entry or return 556 */ 557 int 558 ffs_valloc(struct vnode *pvp, int mode, kauth_cred_t cred, 559 struct vnode **vpp) 560 { 561 struct ufsmount *ump; 562 struct inode *pip; 563 struct fs *fs; 564 struct inode *ip; 565 struct timespec ts; 566 ino_t ino, ipref; 567 int cg, error; 568 569 UFS_WAPBL_JUNLOCK_ASSERT(pvp->v_mount); 570 571 *vpp = NULL; 572 pip = VTOI(pvp); 573 fs = pip->i_fs; 574 ump = pip->i_ump; 575 576 error = UFS_WAPBL_BEGIN(pvp->v_mount); 577 if (error) { 578 return error; 579 } 580 mutex_enter(&ump->um_lock); 581 if (fs->fs_cstotal.cs_nifree == 0) 582 goto noinodes; 583 584 if ((mode & IFMT) == IFDIR) 585 ipref = ffs_dirpref(pip); 586 else 587 ipref = pip->i_number; 588 if (ipref >= fs->fs_ncg * fs->fs_ipg) 589 ipref = 0; 590 cg = ino_to_cg(fs, ipref); 591 /* 592 * Track number of dirs created one after another 593 * in a same cg without intervening by files. 594 */ 595 if ((mode & IFMT) == IFDIR) { 596 if (fs->fs_contigdirs[cg] < 255) 597 fs->fs_contigdirs[cg]++; 598 } else { 599 if (fs->fs_contigdirs[cg] > 0) 600 fs->fs_contigdirs[cg]--; 601 } 602 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, ffs_nodealloccg); 603 if (ino == 0) 604 goto noinodes; 605 UFS_WAPBL_END(pvp->v_mount); 606 error = VFS_VGET(pvp->v_mount, ino, vpp); 607 if (error) { 608 int err; 609 err = UFS_WAPBL_BEGIN(pvp->v_mount); 610 if (err == 0) 611 ffs_vfree(pvp, ino, mode); 612 if (err == 0) 613 UFS_WAPBL_END(pvp->v_mount); 614 return (error); 615 } 616 KASSERT((*vpp)->v_type == VNON); 617 ip = VTOI(*vpp); 618 if (ip->i_mode) { 619 #if 0 620 printf("mode = 0%o, inum = %d, fs = %s\n", 621 ip->i_mode, ip->i_number, fs->fs_fsmnt); 622 #else 623 printf("dmode %x mode %x dgen %x gen %x\n", 624 DIP(ip, mode), ip->i_mode, 625 DIP(ip, gen), ip->i_gen); 626 printf("size %llx blocks %llx\n", 627 (long long)DIP(ip, size), (long long)DIP(ip, blocks)); 628 printf("ino %llu ipref %llu\n", (unsigned long long)ino, 629 (unsigned long long)ipref); 630 #if 0 631 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 632 (int)fs->fs_bsize, NOCRED, 0, &bp); 633 #endif 634 635 #endif 636 panic("ffs_valloc: dup alloc"); 637 } 638 if (DIP(ip, blocks)) { /* XXX */ 639 printf("free inode %s/%llu had %" PRId64 " blocks\n", 640 fs->fs_fsmnt, (unsigned long long)ino, DIP(ip, blocks)); 641 DIP_ASSIGN(ip, blocks, 0); 642 } 643 ip->i_flag &= ~IN_SPACECOUNTED; 644 ip->i_flags = 0; 645 DIP_ASSIGN(ip, flags, 0); 646 /* 647 * Set up a new generation number for this inode. 648 */ 649 ip->i_gen++; 650 DIP_ASSIGN(ip, gen, ip->i_gen); 651 if (fs->fs_magic == FS_UFS2_MAGIC) { 652 vfs_timestamp(&ts); 653 ip->i_ffs2_birthtime = ts.tv_sec; 654 ip->i_ffs2_birthnsec = ts.tv_nsec; 655 } 656 return (0); 657 noinodes: 658 mutex_exit(&ump->um_lock); 659 UFS_WAPBL_END(pvp->v_mount); 660 ffs_fserr(fs, kauth_cred_geteuid(cred), "out of inodes"); 661 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); 662 return (ENOSPC); 663 } 664 665 /* 666 * Find a cylinder group in which to place a directory. 667 * 668 * The policy implemented by this algorithm is to allocate a 669 * directory inode in the same cylinder group as its parent 670 * directory, but also to reserve space for its files inodes 671 * and data. Restrict the number of directories which may be 672 * allocated one after another in the same cylinder group 673 * without intervening allocation of files. 674 * 675 * If we allocate a first level directory then force allocation 676 * in another cylinder group. 677 */ 678 static ino_t 679 ffs_dirpref(struct inode *pip) 680 { 681 register struct fs *fs; 682 int cg, prefcg; 683 int64_t dirsize, cgsize, curdsz; 684 int avgifree, avgbfree, avgndir; 685 int minifree, minbfree, maxndir; 686 int mincg, minndir; 687 int maxcontigdirs; 688 689 KASSERT(mutex_owned(&pip->i_ump->um_lock)); 690 691 fs = pip->i_fs; 692 693 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; 694 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 695 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; 696 697 /* 698 * Force allocation in another cg if creating a first level dir. 699 */ 700 if (ITOV(pip)->v_vflag & VV_ROOT) { 701 prefcg = random() % fs->fs_ncg; 702 mincg = prefcg; 703 minndir = fs->fs_ipg; 704 for (cg = prefcg; cg < fs->fs_ncg; cg++) 705 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 706 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 707 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 708 mincg = cg; 709 minndir = fs->fs_cs(fs, cg).cs_ndir; 710 } 711 for (cg = 0; cg < prefcg; cg++) 712 if (fs->fs_cs(fs, cg).cs_ndir < minndir && 713 fs->fs_cs(fs, cg).cs_nifree >= avgifree && 714 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 715 mincg = cg; 716 minndir = fs->fs_cs(fs, cg).cs_ndir; 717 } 718 return ((ino_t)(fs->fs_ipg * mincg)); 719 } 720 721 /* 722 * Count various limits which used for 723 * optimal allocation of a directory inode. 724 */ 725 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); 726 minifree = avgifree - fs->fs_ipg / 4; 727 if (minifree < 0) 728 minifree = 0; 729 minbfree = avgbfree - fragstoblks(fs, fs->fs_fpg) / 4; 730 if (minbfree < 0) 731 minbfree = 0; 732 cgsize = (int64_t)fs->fs_fsize * fs->fs_fpg; 733 dirsize = (int64_t)fs->fs_avgfilesize * fs->fs_avgfpdir; 734 if (avgndir != 0) { 735 curdsz = (cgsize - (int64_t)avgbfree * fs->fs_bsize) / avgndir; 736 if (dirsize < curdsz) 737 dirsize = curdsz; 738 } 739 if (cgsize < dirsize * 255) 740 maxcontigdirs = cgsize / dirsize; 741 else 742 maxcontigdirs = 255; 743 if (fs->fs_avgfpdir > 0) 744 maxcontigdirs = min(maxcontigdirs, 745 fs->fs_ipg / fs->fs_avgfpdir); 746 if (maxcontigdirs == 0) 747 maxcontigdirs = 1; 748 749 /* 750 * Limit number of dirs in one cg and reserve space for 751 * regular files, but only if we have no deficit in 752 * inodes or space. 753 */ 754 prefcg = ino_to_cg(fs, pip->i_number); 755 for (cg = prefcg; cg < fs->fs_ncg; cg++) 756 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 757 fs->fs_cs(fs, cg).cs_nifree >= minifree && 758 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 759 if (fs->fs_contigdirs[cg] < maxcontigdirs) 760 return ((ino_t)(fs->fs_ipg * cg)); 761 } 762 for (cg = 0; cg < prefcg; cg++) 763 if (fs->fs_cs(fs, cg).cs_ndir < maxndir && 764 fs->fs_cs(fs, cg).cs_nifree >= minifree && 765 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { 766 if (fs->fs_contigdirs[cg] < maxcontigdirs) 767 return ((ino_t)(fs->fs_ipg * cg)); 768 } 769 /* 770 * This is a backstop when we are deficient in space. 771 */ 772 for (cg = prefcg; cg < fs->fs_ncg; cg++) 773 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 774 return ((ino_t)(fs->fs_ipg * cg)); 775 for (cg = 0; cg < prefcg; cg++) 776 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) 777 break; 778 return ((ino_t)(fs->fs_ipg * cg)); 779 } 780 781 /* 782 * Select the desired position for the next block in a file. The file is 783 * logically divided into sections. The first section is composed of the 784 * direct blocks. Each additional section contains fs_maxbpg blocks. 785 * 786 * If no blocks have been allocated in the first section, the policy is to 787 * request a block in the same cylinder group as the inode that describes 788 * the file. If no blocks have been allocated in any other section, the 789 * policy is to place the section in a cylinder group with a greater than 790 * average number of free blocks. An appropriate cylinder group is found 791 * by using a rotor that sweeps the cylinder groups. When a new group of 792 * blocks is needed, the sweep begins in the cylinder group following the 793 * cylinder group from which the previous allocation was made. The sweep 794 * continues until a cylinder group with greater than the average number 795 * of free blocks is found. If the allocation is for the first block in an 796 * indirect block, the information on the previous allocation is unavailable; 797 * here a best guess is made based upon the logical block number being 798 * allocated. 799 * 800 * If a section is already partially allocated, the policy is to 801 * contiguously allocate fs_maxcontig blocks. The end of one of these 802 * contiguous blocks and the beginning of the next is laid out 803 * contigously if possible. 804 * 805 * => um_lock held on entry and exit 806 */ 807 daddr_t 808 ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int flags, 809 int32_t *bap /* XXX ondisk32 */) 810 { 811 struct fs *fs; 812 int cg; 813 int avgbfree, startcg; 814 815 KASSERT(mutex_owned(&ip->i_ump->um_lock)); 816 817 fs = ip->i_fs; 818 819 /* 820 * If allocating a contiguous file with B_CONTIG, use the hints 821 * in the inode extentions to return the desired block. 822 * 823 * For metadata (indirect blocks) return the address of where 824 * the first indirect block resides - we'll scan for the next 825 * available slot if we need to allocate more than one indirect 826 * block. For data, return the address of the actual block 827 * relative to the address of the first data block. 828 */ 829 if (flags & B_CONTIG) { 830 KASSERT(ip->i_ffs_first_data_blk != 0); 831 KASSERT(ip->i_ffs_first_indir_blk != 0); 832 if (flags & B_METAONLY) 833 return ip->i_ffs_first_indir_blk; 834 else 835 return ip->i_ffs_first_data_blk + blkstofrags(fs, lbn); 836 } 837 838 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 839 if (lbn < NDADDR + NINDIR(fs)) { 840 cg = ino_to_cg(fs, ip->i_number); 841 return (cgbase(fs, cg) + fs->fs_frag); 842 } 843 /* 844 * Find a cylinder with greater than average number of 845 * unused data blocks. 846 */ 847 if (indx == 0 || bap[indx - 1] == 0) 848 startcg = 849 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 850 else 851 startcg = dtog(fs, 852 ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1); 853 startcg %= fs->fs_ncg; 854 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 855 for (cg = startcg; cg < fs->fs_ncg; cg++) 856 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 857 return (cgbase(fs, cg) + fs->fs_frag); 858 } 859 for (cg = 0; cg < startcg; cg++) 860 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 861 return (cgbase(fs, cg) + fs->fs_frag); 862 } 863 return (0); 864 } 865 /* 866 * We just always try to lay things out contiguously. 867 */ 868 return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag; 869 } 870 871 daddr_t 872 ffs_blkpref_ufs2(struct inode *ip, daddr_t lbn, int indx, int flags, 873 int64_t *bap) 874 { 875 struct fs *fs; 876 int cg; 877 int avgbfree, startcg; 878 879 KASSERT(mutex_owned(&ip->i_ump->um_lock)); 880 881 fs = ip->i_fs; 882 883 /* 884 * If allocating a contiguous file with B_CONTIG, use the hints 885 * in the inode extentions to return the desired block. 886 * 887 * For metadata (indirect blocks) return the address of where 888 * the first indirect block resides - we'll scan for the next 889 * available slot if we need to allocate more than one indirect 890 * block. For data, return the address of the actual block 891 * relative to the address of the first data block. 892 */ 893 if (flags & B_CONTIG) { 894 KASSERT(ip->i_ffs_first_data_blk != 0); 895 KASSERT(ip->i_ffs_first_indir_blk != 0); 896 if (flags & B_METAONLY) 897 return ip->i_ffs_first_indir_blk; 898 else 899 return ip->i_ffs_first_data_blk + blkstofrags(fs, lbn); 900 } 901 902 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { 903 if (lbn < NDADDR + NINDIR(fs)) { 904 cg = ino_to_cg(fs, ip->i_number); 905 return (cgbase(fs, cg) + fs->fs_frag); 906 } 907 /* 908 * Find a cylinder with greater than average number of 909 * unused data blocks. 910 */ 911 if (indx == 0 || bap[indx - 1] == 0) 912 startcg = 913 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg; 914 else 915 startcg = dtog(fs, 916 ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1); 917 startcg %= fs->fs_ncg; 918 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; 919 for (cg = startcg; cg < fs->fs_ncg; cg++) 920 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 921 return (cgbase(fs, cg) + fs->fs_frag); 922 } 923 for (cg = 0; cg < startcg; cg++) 924 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { 925 return (cgbase(fs, cg) + fs->fs_frag); 926 } 927 return (0); 928 } 929 /* 930 * We just always try to lay things out contiguously. 931 */ 932 return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag; 933 } 934 935 936 /* 937 * Implement the cylinder overflow algorithm. 938 * 939 * The policy implemented by this algorithm is: 940 * 1) allocate the block in its requested cylinder group. 941 * 2) quadradically rehash on the cylinder group number. 942 * 3) brute force search for a free block. 943 * 944 * => called with um_lock held 945 * => returns with um_lock released on success, held on failure 946 * (*allocator releases lock on success, retains lock on failure) 947 */ 948 /*VARARGS5*/ 949 static daddr_t 950 ffs_hashalloc(struct inode *ip, int cg, daddr_t pref, 951 int size /* size for data blocks, mode for inodes */, 952 int flags, daddr_t (*allocator)(struct inode *, int, daddr_t, int, int)) 953 { 954 struct fs *fs; 955 daddr_t result; 956 int i, icg = cg; 957 958 fs = ip->i_fs; 959 /* 960 * 1: preferred cylinder group 961 */ 962 result = (*allocator)(ip, cg, pref, size, flags); 963 if (result) 964 return (result); 965 966 if (flags & B_CONTIG) 967 return (result); 968 /* 969 * 2: quadratic rehash 970 */ 971 for (i = 1; i < fs->fs_ncg; i *= 2) { 972 cg += i; 973 if (cg >= fs->fs_ncg) 974 cg -= fs->fs_ncg; 975 result = (*allocator)(ip, cg, 0, size, flags); 976 if (result) 977 return (result); 978 } 979 /* 980 * 3: brute force search 981 * Note that we start at i == 2, since 0 was checked initially, 982 * and 1 is always checked in the quadratic rehash. 983 */ 984 cg = (icg + 2) % fs->fs_ncg; 985 for (i = 2; i < fs->fs_ncg; i++) { 986 result = (*allocator)(ip, cg, 0, size, flags); 987 if (result) 988 return (result); 989 cg++; 990 if (cg == fs->fs_ncg) 991 cg = 0; 992 } 993 return (0); 994 } 995 996 /* 997 * Determine whether a fragment can be extended. 998 * 999 * Check to see if the necessary fragments are available, and 1000 * if they are, allocate them. 1001 * 1002 * => called with um_lock held 1003 * => returns with um_lock released on success, held on failure 1004 */ 1005 static daddr_t 1006 ffs_fragextend(struct inode *ip, int cg, daddr_t bprev, int osize, int nsize) 1007 { 1008 struct ufsmount *ump; 1009 struct fs *fs; 1010 struct cg *cgp; 1011 struct buf *bp; 1012 daddr_t bno; 1013 int frags, bbase; 1014 int i, error; 1015 u_int8_t *blksfree; 1016 1017 fs = ip->i_fs; 1018 ump = ip->i_ump; 1019 1020 KASSERT(mutex_owned(&ump->um_lock)); 1021 1022 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) 1023 return (0); 1024 frags = numfrags(fs, nsize); 1025 bbase = fragnum(fs, bprev); 1026 if (bbase > fragnum(fs, (bprev + frags - 1))) { 1027 /* cannot extend across a block boundary */ 1028 return (0); 1029 } 1030 mutex_exit(&ump->um_lock); 1031 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1032 (int)fs->fs_cgsize, NOCRED, B_MODIFY, &bp); 1033 if (error) 1034 goto fail; 1035 cgp = (struct cg *)bp->b_data; 1036 if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs))) 1037 goto fail; 1038 cgp->cg_old_time = ufs_rw32(time_second, UFS_FSNEEDSWAP(fs)); 1039 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1040 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1041 cgp->cg_time = ufs_rw64(time_second, UFS_FSNEEDSWAP(fs)); 1042 bno = dtogd(fs, bprev); 1043 blksfree = cg_blksfree(cgp, UFS_FSNEEDSWAP(fs)); 1044 for (i = numfrags(fs, osize); i < frags; i++) 1045 if (isclr(blksfree, bno + i)) 1046 goto fail; 1047 /* 1048 * the current fragment can be extended 1049 * deduct the count on fragment being extended into 1050 * increase the count on the remaining fragment (if any) 1051 * allocate the extended piece 1052 */ 1053 for (i = frags; i < fs->fs_frag - bbase; i++) 1054 if (isclr(blksfree, bno + i)) 1055 break; 1056 ufs_add32(cgp->cg_frsum[i - numfrags(fs, osize)], -1, UFS_FSNEEDSWAP(fs)); 1057 if (i != frags) 1058 ufs_add32(cgp->cg_frsum[i - frags], 1, UFS_FSNEEDSWAP(fs)); 1059 mutex_enter(&ump->um_lock); 1060 for (i = numfrags(fs, osize); i < frags; i++) { 1061 clrbit(blksfree, bno + i); 1062 ufs_add32(cgp->cg_cs.cs_nffree, -1, UFS_FSNEEDSWAP(fs)); 1063 fs->fs_cstotal.cs_nffree--; 1064 fs->fs_cs(fs, cg).cs_nffree--; 1065 } 1066 fs->fs_fmod = 1; 1067 ACTIVECG_CLR(fs, cg); 1068 mutex_exit(&ump->um_lock); 1069 bdwrite(bp); 1070 return (bprev); 1071 1072 fail: 1073 brelse(bp, 0); 1074 mutex_enter(&ump->um_lock); 1075 return (0); 1076 } 1077 1078 /* 1079 * Determine whether a block can be allocated. 1080 * 1081 * Check to see if a block of the appropriate size is available, 1082 * and if it is, allocate it. 1083 */ 1084 static daddr_t 1085 ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size, int flags) 1086 { 1087 struct ufsmount *ump; 1088 struct fs *fs = ip->i_fs; 1089 struct cg *cgp; 1090 struct buf *bp; 1091 int32_t bno; 1092 daddr_t blkno; 1093 int error, frags, allocsiz, i; 1094 u_int8_t *blksfree; 1095 #ifdef FFS_EI 1096 const int needswap = UFS_FSNEEDSWAP(fs); 1097 #endif 1098 1099 ump = ip->i_ump; 1100 1101 KASSERT(mutex_owned(&ump->um_lock)); 1102 1103 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) 1104 return (0); 1105 mutex_exit(&ump->um_lock); 1106 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1107 (int)fs->fs_cgsize, NOCRED, B_MODIFY, &bp); 1108 if (error) 1109 goto fail; 1110 cgp = (struct cg *)bp->b_data; 1111 if (!cg_chkmagic(cgp, needswap) || 1112 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) 1113 goto fail; 1114 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1115 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1116 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1117 cgp->cg_time = ufs_rw64(time_second, needswap); 1118 if (size == fs->fs_bsize) { 1119 mutex_enter(&ump->um_lock); 1120 blkno = ffs_alloccgblk(ip, bp, bpref, flags); 1121 ACTIVECG_CLR(fs, cg); 1122 mutex_exit(&ump->um_lock); 1123 bdwrite(bp); 1124 return (blkno); 1125 } 1126 /* 1127 * check to see if any fragments are already available 1128 * allocsiz is the size which will be allocated, hacking 1129 * it down to a smaller size if necessary 1130 */ 1131 blksfree = cg_blksfree(cgp, needswap); 1132 frags = numfrags(fs, size); 1133 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) 1134 if (cgp->cg_frsum[allocsiz] != 0) 1135 break; 1136 if (allocsiz == fs->fs_frag) { 1137 /* 1138 * no fragments were available, so a block will be 1139 * allocated, and hacked up 1140 */ 1141 if (cgp->cg_cs.cs_nbfree == 0) 1142 goto fail; 1143 mutex_enter(&ump->um_lock); 1144 blkno = ffs_alloccgblk(ip, bp, bpref, flags); 1145 bno = dtogd(fs, blkno); 1146 for (i = frags; i < fs->fs_frag; i++) 1147 setbit(blksfree, bno + i); 1148 i = fs->fs_frag - frags; 1149 ufs_add32(cgp->cg_cs.cs_nffree, i, needswap); 1150 fs->fs_cstotal.cs_nffree += i; 1151 fs->fs_cs(fs, cg).cs_nffree += i; 1152 fs->fs_fmod = 1; 1153 ufs_add32(cgp->cg_frsum[i], 1, needswap); 1154 ACTIVECG_CLR(fs, cg); 1155 mutex_exit(&ump->um_lock); 1156 bdwrite(bp); 1157 return (blkno); 1158 } 1159 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); 1160 #if 0 1161 /* 1162 * XXX fvdl mapsearch will panic, and never return -1 1163 * also: returning NULL as daddr_t ? 1164 */ 1165 if (bno < 0) 1166 goto fail; 1167 #endif 1168 for (i = 0; i < frags; i++) 1169 clrbit(blksfree, bno + i); 1170 mutex_enter(&ump->um_lock); 1171 ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap); 1172 fs->fs_cstotal.cs_nffree -= frags; 1173 fs->fs_cs(fs, cg).cs_nffree -= frags; 1174 fs->fs_fmod = 1; 1175 ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap); 1176 if (frags != allocsiz) 1177 ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap); 1178 blkno = cgbase(fs, cg) + bno; 1179 ACTIVECG_CLR(fs, cg); 1180 mutex_exit(&ump->um_lock); 1181 bdwrite(bp); 1182 return blkno; 1183 1184 fail: 1185 brelse(bp, 0); 1186 mutex_enter(&ump->um_lock); 1187 return (0); 1188 } 1189 1190 /* 1191 * Allocate a block in a cylinder group. 1192 * 1193 * This algorithm implements the following policy: 1194 * 1) allocate the requested block. 1195 * 2) allocate a rotationally optimal block in the same cylinder. 1196 * 3) allocate the next available block on the block rotor for the 1197 * specified cylinder group. 1198 * Note that this routine only allocates fs_bsize blocks; these 1199 * blocks may be fragmented by the routine that allocates them. 1200 */ 1201 static daddr_t 1202 ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref, int flags) 1203 { 1204 struct ufsmount *ump; 1205 struct fs *fs = ip->i_fs; 1206 struct cg *cgp; 1207 int cg; 1208 daddr_t blkno; 1209 int32_t bno; 1210 u_int8_t *blksfree; 1211 #ifdef FFS_EI 1212 const int needswap = UFS_FSNEEDSWAP(fs); 1213 #endif 1214 1215 ump = ip->i_ump; 1216 1217 KASSERT(mutex_owned(&ump->um_lock)); 1218 1219 cgp = (struct cg *)bp->b_data; 1220 blksfree = cg_blksfree(cgp, needswap); 1221 if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) { 1222 bpref = ufs_rw32(cgp->cg_rotor, needswap); 1223 } else { 1224 bpref = blknum(fs, bpref); 1225 bno = dtogd(fs, bpref); 1226 /* 1227 * if the requested block is available, use it 1228 */ 1229 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno))) 1230 goto gotit; 1231 /* 1232 * if the requested data block isn't available and we are 1233 * trying to allocate a contiguous file, return an error. 1234 */ 1235 if ((flags & (B_CONTIG | B_METAONLY)) == B_CONTIG) 1236 return (0); 1237 } 1238 1239 /* 1240 * Take the next available block in this cylinder group. 1241 */ 1242 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); 1243 if (bno < 0) 1244 return (0); 1245 cgp->cg_rotor = ufs_rw32(bno, needswap); 1246 gotit: 1247 blkno = fragstoblks(fs, bno); 1248 ffs_clrblock(fs, blksfree, blkno); 1249 ffs_clusteracct(fs, cgp, blkno, -1); 1250 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap); 1251 fs->fs_cstotal.cs_nbfree--; 1252 fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--; 1253 if ((fs->fs_magic == FS_UFS1_MAGIC) && 1254 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) { 1255 int cylno; 1256 cylno = old_cbtocylno(fs, bno); 1257 KASSERT(cylno >= 0); 1258 KASSERT(cylno < fs->fs_old_ncyl); 1259 KASSERT(old_cbtorpos(fs, bno) >= 0); 1260 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bno) < fs->fs_old_nrpos); 1261 ufs_add16(old_cg_blks(fs, cgp, cylno, needswap)[old_cbtorpos(fs, bno)], -1, 1262 needswap); 1263 ufs_add32(old_cg_blktot(cgp, needswap)[cylno], -1, needswap); 1264 } 1265 fs->fs_fmod = 1; 1266 cg = ufs_rw32(cgp->cg_cgx, needswap); 1267 blkno = cgbase(fs, cg) + bno; 1268 return (blkno); 1269 } 1270 1271 /* 1272 * Determine whether an inode can be allocated. 1273 * 1274 * Check to see if an inode is available, and if it is, 1275 * allocate it using the following policy: 1276 * 1) allocate the requested inode. 1277 * 2) allocate the next available inode after the requested 1278 * inode in the specified cylinder group. 1279 */ 1280 static daddr_t 1281 ffs_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode, int flags) 1282 { 1283 struct ufsmount *ump = ip->i_ump; 1284 struct fs *fs = ip->i_fs; 1285 struct cg *cgp; 1286 struct buf *bp, *ibp; 1287 u_int8_t *inosused; 1288 int error, start, len, loc, map, i; 1289 int32_t initediblk; 1290 daddr_t nalloc; 1291 struct ufs2_dinode *dp2; 1292 #ifdef FFS_EI 1293 const int needswap = UFS_FSNEEDSWAP(fs); 1294 #endif 1295 1296 KASSERT(mutex_owned(&ump->um_lock)); 1297 UFS_WAPBL_JLOCK_ASSERT(ip->i_ump->um_mountp); 1298 1299 if (fs->fs_cs(fs, cg).cs_nifree == 0) 1300 return (0); 1301 mutex_exit(&ump->um_lock); 1302 ibp = NULL; 1303 initediblk = -1; 1304 retry: 1305 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), 1306 (int)fs->fs_cgsize, NOCRED, B_MODIFY, &bp); 1307 if (error) 1308 goto fail; 1309 cgp = (struct cg *)bp->b_data; 1310 if (!cg_chkmagic(cgp, needswap) || cgp->cg_cs.cs_nifree == 0) 1311 goto fail; 1312 1313 if (ibp != NULL && 1314 initediblk != ufs_rw32(cgp->cg_initediblk, needswap)) { 1315 /* Another thread allocated more inodes so we retry the test. */ 1316 brelse(ibp, 0); 1317 ibp = NULL; 1318 } 1319 /* 1320 * Check to see if we need to initialize more inodes. 1321 */ 1322 if (fs->fs_magic == FS_UFS2_MAGIC && ibp == NULL) { 1323 initediblk = ufs_rw32(cgp->cg_initediblk, needswap); 1324 nalloc = fs->fs_ipg - ufs_rw32(cgp->cg_cs.cs_nifree, needswap); 1325 if (nalloc + INOPB(fs) > initediblk && 1326 initediblk < ufs_rw32(cgp->cg_niblk, needswap)) { 1327 /* 1328 * We have to release the cg buffer here to prevent 1329 * a deadlock when reading the inode block will 1330 * run a copy-on-write that might use this cg. 1331 */ 1332 brelse(bp, 0); 1333 bp = NULL; 1334 error = ffs_getblk(ip->i_devvp, fsbtodb(fs, 1335 ino_to_fsba(fs, cg * fs->fs_ipg + initediblk)), 1336 FFS_NOBLK, fs->fs_bsize, false, &ibp); 1337 if (error) 1338 goto fail; 1339 goto retry; 1340 } 1341 } 1342 1343 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1344 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1345 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1346 cgp->cg_time = ufs_rw64(time_second, needswap); 1347 inosused = cg_inosused(cgp, needswap); 1348 if (ipref) { 1349 ipref %= fs->fs_ipg; 1350 if (isclr(inosused, ipref)) 1351 goto gotit; 1352 } 1353 start = ufs_rw32(cgp->cg_irotor, needswap) / NBBY; 1354 len = howmany(fs->fs_ipg - ufs_rw32(cgp->cg_irotor, needswap), 1355 NBBY); 1356 loc = skpc(0xff, len, &inosused[start]); 1357 if (loc == 0) { 1358 len = start + 1; 1359 start = 0; 1360 loc = skpc(0xff, len, &inosused[0]); 1361 if (loc == 0) { 1362 printf("cg = %d, irotor = %d, fs = %s\n", 1363 cg, ufs_rw32(cgp->cg_irotor, needswap), 1364 fs->fs_fsmnt); 1365 panic("ffs_nodealloccg: map corrupted"); 1366 /* NOTREACHED */ 1367 } 1368 } 1369 i = start + len - loc; 1370 map = inosused[i] ^ 0xff; 1371 if (map == 0) { 1372 printf("fs = %s\n", fs->fs_fsmnt); 1373 panic("ffs_nodealloccg: block not in map"); 1374 } 1375 ipref = i * NBBY + ffs(map) - 1; 1376 cgp->cg_irotor = ufs_rw32(ipref, needswap); 1377 gotit: 1378 UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp, cg * fs->fs_ipg + ipref, 1379 mode); 1380 /* 1381 * Check to see if we need to initialize more inodes. 1382 */ 1383 if (ibp != NULL) { 1384 KASSERT(initediblk == ufs_rw32(cgp->cg_initediblk, needswap)); 1385 memset(ibp->b_data, 0, fs->fs_bsize); 1386 dp2 = (struct ufs2_dinode *)(ibp->b_data); 1387 for (i = 0; i < INOPB(fs); i++) { 1388 /* 1389 * Don't bother to swap, it's supposed to be 1390 * random, after all. 1391 */ 1392 dp2->di_gen = (cprng_fast32() & INT32_MAX) / 2 + 1; 1393 dp2++; 1394 } 1395 initediblk += INOPB(fs); 1396 cgp->cg_initediblk = ufs_rw32(initediblk, needswap); 1397 } 1398 1399 mutex_enter(&ump->um_lock); 1400 ACTIVECG_CLR(fs, cg); 1401 setbit(inosused, ipref); 1402 ufs_add32(cgp->cg_cs.cs_nifree, -1, needswap); 1403 fs->fs_cstotal.cs_nifree--; 1404 fs->fs_cs(fs, cg).cs_nifree--; 1405 fs->fs_fmod = 1; 1406 if ((mode & IFMT) == IFDIR) { 1407 ufs_add32(cgp->cg_cs.cs_ndir, 1, needswap); 1408 fs->fs_cstotal.cs_ndir++; 1409 fs->fs_cs(fs, cg).cs_ndir++; 1410 } 1411 mutex_exit(&ump->um_lock); 1412 if (ibp != NULL) { 1413 bwrite(bp); 1414 bawrite(ibp); 1415 } else 1416 bdwrite(bp); 1417 return (cg * fs->fs_ipg + ipref); 1418 fail: 1419 if (bp != NULL) 1420 brelse(bp, 0); 1421 if (ibp != NULL) 1422 brelse(ibp, 0); 1423 mutex_enter(&ump->um_lock); 1424 return (0); 1425 } 1426 1427 /* 1428 * Allocate a block or fragment. 1429 * 1430 * The specified block or fragment is removed from the 1431 * free map, possibly fragmenting a block in the process. 1432 * 1433 * This implementation should mirror fs_blkfree 1434 * 1435 * => um_lock not held on entry or exit 1436 */ 1437 int 1438 ffs_blkalloc(struct inode *ip, daddr_t bno, long size) 1439 { 1440 int error; 1441 1442 error = ffs_check_bad_allocation(__func__, ip->i_fs, bno, size, 1443 ip->i_dev, ip->i_uid); 1444 if (error) 1445 return error; 1446 1447 return ffs_blkalloc_ump(ip->i_ump, bno, size); 1448 } 1449 1450 int 1451 ffs_blkalloc_ump(struct ufsmount *ump, daddr_t bno, long size) 1452 { 1453 struct fs *fs = ump->um_fs; 1454 struct cg *cgp; 1455 struct buf *bp; 1456 int32_t fragno, cgbno; 1457 int i, error, cg, blk, frags, bbase; 1458 u_int8_t *blksfree; 1459 const int needswap = UFS_FSNEEDSWAP(fs); 1460 1461 KASSERT((u_int)size <= fs->fs_bsize && fragoff(fs, size) == 0 && 1462 fragnum(fs, bno) + numfrags(fs, size) <= fs->fs_frag); 1463 KASSERT(bno < fs->fs_size); 1464 1465 cg = dtog(fs, bno); 1466 error = bread(ump->um_devvp, fsbtodb(fs, cgtod(fs, cg)), 1467 (int)fs->fs_cgsize, NOCRED, B_MODIFY, &bp); 1468 if (error) { 1469 brelse(bp, 0); 1470 return error; 1471 } 1472 cgp = (struct cg *)bp->b_data; 1473 if (!cg_chkmagic(cgp, needswap)) { 1474 brelse(bp, 0); 1475 return EIO; 1476 } 1477 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1478 cgp->cg_time = ufs_rw64(time_second, needswap); 1479 cgbno = dtogd(fs, bno); 1480 blksfree = cg_blksfree(cgp, needswap); 1481 1482 mutex_enter(&ump->um_lock); 1483 if (size == fs->fs_bsize) { 1484 fragno = fragstoblks(fs, cgbno); 1485 if (!ffs_isblock(fs, blksfree, fragno)) { 1486 mutex_exit(&ump->um_lock); 1487 brelse(bp, 0); 1488 return EBUSY; 1489 } 1490 ffs_clrblock(fs, blksfree, fragno); 1491 ffs_clusteracct(fs, cgp, fragno, -1); 1492 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap); 1493 fs->fs_cstotal.cs_nbfree--; 1494 fs->fs_cs(fs, cg).cs_nbfree--; 1495 } else { 1496 bbase = cgbno - fragnum(fs, cgbno); 1497 1498 frags = numfrags(fs, size); 1499 for (i = 0; i < frags; i++) { 1500 if (isclr(blksfree, cgbno + i)) { 1501 mutex_exit(&ump->um_lock); 1502 brelse(bp, 0); 1503 return EBUSY; 1504 } 1505 } 1506 /* 1507 * if a complete block is being split, account for it 1508 */ 1509 fragno = fragstoblks(fs, bbase); 1510 if (ffs_isblock(fs, blksfree, fragno)) { 1511 ufs_add32(cgp->cg_cs.cs_nffree, fs->fs_frag, needswap); 1512 fs->fs_cstotal.cs_nffree += fs->fs_frag; 1513 fs->fs_cs(fs, cg).cs_nffree += fs->fs_frag; 1514 ffs_clusteracct(fs, cgp, fragno, -1); 1515 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap); 1516 fs->fs_cstotal.cs_nbfree--; 1517 fs->fs_cs(fs, cg).cs_nbfree--; 1518 } 1519 /* 1520 * decrement the counts associated with the old frags 1521 */ 1522 blk = blkmap(fs, blksfree, bbase); 1523 ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap); 1524 /* 1525 * allocate the fragment 1526 */ 1527 for (i = 0; i < frags; i++) { 1528 clrbit(blksfree, cgbno + i); 1529 } 1530 ufs_add32(cgp->cg_cs.cs_nffree, -i, needswap); 1531 fs->fs_cstotal.cs_nffree -= i; 1532 fs->fs_cs(fs, cg).cs_nffree -= i; 1533 /* 1534 * add back in counts associated with the new frags 1535 */ 1536 blk = blkmap(fs, blksfree, bbase); 1537 ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap); 1538 } 1539 fs->fs_fmod = 1; 1540 ACTIVECG_CLR(fs, cg); 1541 mutex_exit(&ump->um_lock); 1542 bdwrite(bp); 1543 return 0; 1544 } 1545 1546 /* 1547 * Free a block or fragment. 1548 * 1549 * The specified block or fragment is placed back in the 1550 * free map. If a fragment is deallocated, a possible 1551 * block reassembly is checked. 1552 * 1553 * => um_lock not held on entry or exit 1554 */ 1555 static void 1556 ffs_blkfree_cg(struct fs *fs, struct vnode *devvp, daddr_t bno, long size) 1557 { 1558 struct cg *cgp; 1559 struct buf *bp; 1560 struct ufsmount *ump; 1561 daddr_t cgblkno; 1562 int error, cg; 1563 dev_t dev; 1564 const bool devvp_is_snapshot = (devvp->v_type != VBLK); 1565 #ifdef FFS_EI 1566 const int needswap = UFS_FSNEEDSWAP(fs); 1567 #endif 1568 1569 KASSERT(!devvp_is_snapshot); 1570 1571 cg = dtog(fs, bno); 1572 dev = devvp->v_rdev; 1573 ump = VFSTOUFS(devvp->v_specmountpoint); 1574 KASSERT(fs == ump->um_fs); 1575 cgblkno = fsbtodb(fs, cgtod(fs, cg)); 1576 1577 error = bread(devvp, cgblkno, (int)fs->fs_cgsize, 1578 NOCRED, B_MODIFY, &bp); 1579 if (error) { 1580 brelse(bp, 0); 1581 return; 1582 } 1583 cgp = (struct cg *)bp->b_data; 1584 if (!cg_chkmagic(cgp, needswap)) { 1585 brelse(bp, 0); 1586 return; 1587 } 1588 1589 ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot); 1590 1591 bdwrite(bp); 1592 } 1593 1594 struct discardopdata { 1595 struct work wk; /* must be first */ 1596 struct vnode *devvp; 1597 daddr_t bno; 1598 long size; 1599 }; 1600 1601 struct discarddata { 1602 struct fs *fs; 1603 struct discardopdata *entry; 1604 long maxsize; 1605 kmutex_t entrylk; 1606 struct workqueue *wq; 1607 int wqcnt, wqdraining; 1608 kmutex_t wqlk; 1609 kcondvar_t wqcv; 1610 /* timer for flush? */ 1611 }; 1612 1613 static void 1614 ffs_blkfree_td(struct fs *fs, struct discardopdata *td) 1615 { 1616 long todo; 1617 1618 while (td->size) { 1619 todo = min(td->size, 1620 lfragtosize(fs, (fs->fs_frag - fragnum(fs, td->bno)))); 1621 ffs_blkfree_cg(fs, td->devvp, td->bno, todo); 1622 td->bno += numfrags(fs, todo); 1623 td->size -= todo; 1624 } 1625 } 1626 1627 static void 1628 ffs_discardcb(struct work *wk, void *arg) 1629 { 1630 struct discardopdata *td = (void *)wk; 1631 struct discarddata *ts = arg; 1632 struct fs *fs = ts->fs; 1633 struct disk_discard_range ta; 1634 int error; 1635 1636 ta.bno = fsbtodb(fs, td->bno); 1637 ta.size = td->size >> DEV_BSHIFT; 1638 error = VOP_IOCTL(td->devvp, DIOCDISCARD, &ta, FWRITE, FSCRED); 1639 #ifdef TRIMDEBUG 1640 printf("trim(%" PRId64 ",%ld):%d\n", td->bno, td->size, error); 1641 #endif 1642 1643 ffs_blkfree_td(fs, td); 1644 kmem_free(td, sizeof(*td)); 1645 mutex_enter(&ts->wqlk); 1646 ts->wqcnt--; 1647 if (ts->wqdraining && !ts->wqcnt) 1648 cv_signal(&ts->wqcv); 1649 mutex_exit(&ts->wqlk); 1650 } 1651 1652 void * 1653 ffs_discard_init(struct vnode *devvp, struct fs *fs) 1654 { 1655 struct disk_discard_params tp; 1656 struct discarddata *ts; 1657 int error; 1658 1659 error = VOP_IOCTL(devvp, DIOCGDISCARDPARAMS, &tp, FREAD, FSCRED); 1660 if (error) { 1661 printf("DIOCGDISCARDPARAMS: %d\n", error); 1662 return NULL; 1663 } 1664 if (tp.maxsize * DEV_BSIZE < fs->fs_bsize) { 1665 printf("tp.maxsize=%ld, fs_bsize=%d\n", tp.maxsize, fs->fs_bsize); 1666 return NULL; 1667 } 1668 1669 ts = kmem_zalloc(sizeof (*ts), KM_SLEEP); 1670 error = workqueue_create(&ts->wq, "trimwq", ffs_discardcb, ts, 1671 0, 0, 0); 1672 if (error) { 1673 kmem_free(ts, sizeof (*ts)); 1674 return NULL; 1675 } 1676 mutex_init(&ts->entrylk, MUTEX_DEFAULT, IPL_NONE); 1677 mutex_init(&ts->wqlk, MUTEX_DEFAULT, IPL_NONE); 1678 cv_init(&ts->wqcv, "trimwqcv"); 1679 ts->maxsize = max(tp.maxsize * DEV_BSIZE, 100*1024); /* XXX */ 1680 ts->fs = fs; 1681 return ts; 1682 } 1683 1684 void 1685 ffs_discard_finish(void *vts, int flags) 1686 { 1687 struct discarddata *ts = vts; 1688 struct discardopdata *td = NULL; 1689 int res = 0; 1690 1691 /* wait for workqueue to drain */ 1692 mutex_enter(&ts->wqlk); 1693 if (ts->wqcnt) { 1694 ts->wqdraining = 1; 1695 res = cv_timedwait(&ts->wqcv, &ts->wqlk, mstohz(5000)); 1696 } 1697 mutex_exit(&ts->wqlk); 1698 if (res) 1699 printf("ffs_discarddata drain timeout\n"); 1700 1701 mutex_enter(&ts->entrylk); 1702 if (ts->entry) { 1703 td = ts->entry; 1704 ts->entry = NULL; 1705 } 1706 mutex_exit(&ts->entrylk); 1707 if (td) { 1708 /* XXX don't tell disk, its optional */ 1709 ffs_blkfree_td(ts->fs, td); 1710 #ifdef TRIMDEBUG 1711 printf("finish(%" PRId64 ",%ld)\n", td->bno, td->size); 1712 #endif 1713 kmem_free(td, sizeof(*td)); 1714 } 1715 1716 cv_destroy(&ts->wqcv); 1717 mutex_destroy(&ts->entrylk); 1718 mutex_destroy(&ts->wqlk); 1719 workqueue_destroy(ts->wq); 1720 kmem_free(ts, sizeof(*ts)); 1721 } 1722 1723 void 1724 ffs_blkfree(struct fs *fs, struct vnode *devvp, daddr_t bno, long size, 1725 ino_t inum) 1726 { 1727 struct ufsmount *ump; 1728 int error; 1729 dev_t dev; 1730 struct discarddata *ts; 1731 struct discardopdata *td; 1732 1733 dev = devvp->v_rdev; 1734 ump = VFSTOUFS(devvp->v_specmountpoint); 1735 if (ffs_snapblkfree(fs, devvp, bno, size, inum)) 1736 return; 1737 1738 error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum); 1739 if (error) 1740 return; 1741 1742 if (!ump->um_discarddata) { 1743 ffs_blkfree_cg(fs, devvp, bno, size); 1744 return; 1745 } 1746 1747 #ifdef TRIMDEBUG 1748 printf("blkfree(%" PRId64 ",%ld)\n", bno, size); 1749 #endif 1750 ts = ump->um_discarddata; 1751 td = NULL; 1752 1753 mutex_enter(&ts->entrylk); 1754 if (ts->entry) { 1755 td = ts->entry; 1756 /* ffs deallocs backwards, check for prepend only */ 1757 if (td->bno == bno + numfrags(fs, size) 1758 && td->size + size <= ts->maxsize) { 1759 td->bno = bno; 1760 td->size += size; 1761 if (td->size < ts->maxsize) { 1762 #ifdef TRIMDEBUG 1763 printf("defer(%" PRId64 ",%ld)\n", td->bno, td->size); 1764 #endif 1765 mutex_exit(&ts->entrylk); 1766 return; 1767 } 1768 size = 0; /* mark done */ 1769 } 1770 ts->entry = NULL; 1771 } 1772 mutex_exit(&ts->entrylk); 1773 1774 if (td) { 1775 #ifdef TRIMDEBUG 1776 printf("enq old(%" PRId64 ",%ld)\n", td->bno, td->size); 1777 #endif 1778 mutex_enter(&ts->wqlk); 1779 ts->wqcnt++; 1780 mutex_exit(&ts->wqlk); 1781 workqueue_enqueue(ts->wq, &td->wk, NULL); 1782 } 1783 if (!size) 1784 return; 1785 1786 td = kmem_alloc(sizeof(*td), KM_SLEEP); 1787 td->devvp = devvp; 1788 td->bno = bno; 1789 td->size = size; 1790 1791 if (td->size < ts->maxsize) { /* XXX always the case */ 1792 mutex_enter(&ts->entrylk); 1793 if (!ts->entry) { /* possible race? */ 1794 #ifdef TRIMDEBUG 1795 printf("defer(%" PRId64 ",%ld)\n", td->bno, td->size); 1796 #endif 1797 ts->entry = td; 1798 td = NULL; 1799 } 1800 mutex_exit(&ts->entrylk); 1801 } 1802 if (td) { 1803 #ifdef TRIMDEBUG 1804 printf("enq new(%" PRId64 ",%ld)\n", td->bno, td->size); 1805 #endif 1806 mutex_enter(&ts->wqlk); 1807 ts->wqcnt++; 1808 mutex_exit(&ts->wqlk); 1809 workqueue_enqueue(ts->wq, &td->wk, NULL); 1810 } 1811 } 1812 1813 /* 1814 * Free a block or fragment from a snapshot cg copy. 1815 * 1816 * The specified block or fragment is placed back in the 1817 * free map. If a fragment is deallocated, a possible 1818 * block reassembly is checked. 1819 * 1820 * => um_lock not held on entry or exit 1821 */ 1822 void 1823 ffs_blkfree_snap(struct fs *fs, struct vnode *devvp, daddr_t bno, long size, 1824 ino_t inum) 1825 { 1826 struct cg *cgp; 1827 struct buf *bp; 1828 struct ufsmount *ump; 1829 daddr_t cgblkno; 1830 int error, cg; 1831 dev_t dev; 1832 const bool devvp_is_snapshot = (devvp->v_type != VBLK); 1833 #ifdef FFS_EI 1834 const int needswap = UFS_FSNEEDSWAP(fs); 1835 #endif 1836 1837 KASSERT(devvp_is_snapshot); 1838 1839 cg = dtog(fs, bno); 1840 dev = VTOI(devvp)->i_devvp->v_rdev; 1841 ump = VFSTOUFS(devvp->v_mount); 1842 cgblkno = fragstoblks(fs, cgtod(fs, cg)); 1843 1844 error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum); 1845 if (error) 1846 return; 1847 1848 error = bread(devvp, cgblkno, (int)fs->fs_cgsize, 1849 NOCRED, B_MODIFY, &bp); 1850 if (error) { 1851 brelse(bp, 0); 1852 return; 1853 } 1854 cgp = (struct cg *)bp->b_data; 1855 if (!cg_chkmagic(cgp, needswap)) { 1856 brelse(bp, 0); 1857 return; 1858 } 1859 1860 ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot); 1861 1862 bdwrite(bp); 1863 } 1864 1865 static void 1866 ffs_blkfree_common(struct ufsmount *ump, struct fs *fs, dev_t dev, 1867 struct buf *bp, daddr_t bno, long size, bool devvp_is_snapshot) 1868 { 1869 struct cg *cgp; 1870 int32_t fragno, cgbno; 1871 int i, cg, blk, frags, bbase; 1872 u_int8_t *blksfree; 1873 const int needswap = UFS_FSNEEDSWAP(fs); 1874 1875 cg = dtog(fs, bno); 1876 cgp = (struct cg *)bp->b_data; 1877 cgp->cg_old_time = ufs_rw32(time_second, needswap); 1878 if ((fs->fs_magic != FS_UFS1_MAGIC) || 1879 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 1880 cgp->cg_time = ufs_rw64(time_second, needswap); 1881 cgbno = dtogd(fs, bno); 1882 blksfree = cg_blksfree(cgp, needswap); 1883 mutex_enter(&ump->um_lock); 1884 if (size == fs->fs_bsize) { 1885 fragno = fragstoblks(fs, cgbno); 1886 if (!ffs_isfreeblock(fs, blksfree, fragno)) { 1887 if (devvp_is_snapshot) { 1888 mutex_exit(&ump->um_lock); 1889 return; 1890 } 1891 printf("dev = 0x%llx, block = %" PRId64 ", fs = %s\n", 1892 (unsigned long long)dev, bno, fs->fs_fsmnt); 1893 panic("blkfree: freeing free block"); 1894 } 1895 ffs_setblock(fs, blksfree, fragno); 1896 ffs_clusteracct(fs, cgp, fragno, 1); 1897 ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap); 1898 fs->fs_cstotal.cs_nbfree++; 1899 fs->fs_cs(fs, cg).cs_nbfree++; 1900 if ((fs->fs_magic == FS_UFS1_MAGIC) && 1901 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) { 1902 i = old_cbtocylno(fs, cgbno); 1903 KASSERT(i >= 0); 1904 KASSERT(i < fs->fs_old_ncyl); 1905 KASSERT(old_cbtorpos(fs, cgbno) >= 0); 1906 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, cgbno) < fs->fs_old_nrpos); 1907 ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs, cgbno)], 1, 1908 needswap); 1909 ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap); 1910 } 1911 } else { 1912 bbase = cgbno - fragnum(fs, cgbno); 1913 /* 1914 * decrement the counts associated with the old frags 1915 */ 1916 blk = blkmap(fs, blksfree, bbase); 1917 ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap); 1918 /* 1919 * deallocate the fragment 1920 */ 1921 frags = numfrags(fs, size); 1922 for (i = 0; i < frags; i++) { 1923 if (isset(blksfree, cgbno + i)) { 1924 printf("dev = 0x%llx, block = %" PRId64 1925 ", fs = %s\n", 1926 (unsigned long long)dev, bno + i, 1927 fs->fs_fsmnt); 1928 panic("blkfree: freeing free frag"); 1929 } 1930 setbit(blksfree, cgbno + i); 1931 } 1932 ufs_add32(cgp->cg_cs.cs_nffree, i, needswap); 1933 fs->fs_cstotal.cs_nffree += i; 1934 fs->fs_cs(fs, cg).cs_nffree += i; 1935 /* 1936 * add back in counts associated with the new frags 1937 */ 1938 blk = blkmap(fs, blksfree, bbase); 1939 ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap); 1940 /* 1941 * if a complete block has been reassembled, account for it 1942 */ 1943 fragno = fragstoblks(fs, bbase); 1944 if (ffs_isblock(fs, blksfree, fragno)) { 1945 ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap); 1946 fs->fs_cstotal.cs_nffree -= fs->fs_frag; 1947 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; 1948 ffs_clusteracct(fs, cgp, fragno, 1); 1949 ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap); 1950 fs->fs_cstotal.cs_nbfree++; 1951 fs->fs_cs(fs, cg).cs_nbfree++; 1952 if ((fs->fs_magic == FS_UFS1_MAGIC) && 1953 ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) { 1954 i = old_cbtocylno(fs, bbase); 1955 KASSERT(i >= 0); 1956 KASSERT(i < fs->fs_old_ncyl); 1957 KASSERT(old_cbtorpos(fs, bbase) >= 0); 1958 KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bbase) < fs->fs_old_nrpos); 1959 ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs, 1960 bbase)], 1, needswap); 1961 ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap); 1962 } 1963 } 1964 } 1965 fs->fs_fmod = 1; 1966 ACTIVECG_CLR(fs, cg); 1967 mutex_exit(&ump->um_lock); 1968 } 1969 1970 /* 1971 * Free an inode. 1972 */ 1973 int 1974 ffs_vfree(struct vnode *vp, ino_t ino, int mode) 1975 { 1976 1977 return ffs_freefile(vp->v_mount, ino, mode); 1978 } 1979 1980 /* 1981 * Do the actual free operation. 1982 * The specified inode is placed back in the free map. 1983 * 1984 * => um_lock not held on entry or exit 1985 */ 1986 int 1987 ffs_freefile(struct mount *mp, ino_t ino, int mode) 1988 { 1989 struct ufsmount *ump = VFSTOUFS(mp); 1990 struct fs *fs = ump->um_fs; 1991 struct vnode *devvp; 1992 struct cg *cgp; 1993 struct buf *bp; 1994 int error, cg; 1995 daddr_t cgbno; 1996 dev_t dev; 1997 #ifdef FFS_EI 1998 const int needswap = UFS_FSNEEDSWAP(fs); 1999 #endif 2000 2001 cg = ino_to_cg(fs, ino); 2002 devvp = ump->um_devvp; 2003 dev = devvp->v_rdev; 2004 cgbno = fsbtodb(fs, cgtod(fs, cg)); 2005 2006 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 2007 panic("ifree: range: dev = 0x%llx, ino = %llu, fs = %s", 2008 (long long)dev, (unsigned long long)ino, fs->fs_fsmnt); 2009 error = bread(devvp, cgbno, (int)fs->fs_cgsize, 2010 NOCRED, B_MODIFY, &bp); 2011 if (error) { 2012 brelse(bp, 0); 2013 return (error); 2014 } 2015 cgp = (struct cg *)bp->b_data; 2016 if (!cg_chkmagic(cgp, needswap)) { 2017 brelse(bp, 0); 2018 return (0); 2019 } 2020 2021 ffs_freefile_common(ump, fs, dev, bp, ino, mode, false); 2022 2023 bdwrite(bp); 2024 2025 return 0; 2026 } 2027 2028 int 2029 ffs_freefile_snap(struct fs *fs, struct vnode *devvp, ino_t ino, int mode) 2030 { 2031 struct ufsmount *ump; 2032 struct cg *cgp; 2033 struct buf *bp; 2034 int error, cg; 2035 daddr_t cgbno; 2036 dev_t dev; 2037 #ifdef FFS_EI 2038 const int needswap = UFS_FSNEEDSWAP(fs); 2039 #endif 2040 2041 KASSERT(devvp->v_type != VBLK); 2042 2043 cg = ino_to_cg(fs, ino); 2044 dev = VTOI(devvp)->i_devvp->v_rdev; 2045 ump = VFSTOUFS(devvp->v_mount); 2046 cgbno = fragstoblks(fs, cgtod(fs, cg)); 2047 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 2048 panic("ifree: range: dev = 0x%llx, ino = %llu, fs = %s", 2049 (unsigned long long)dev, (unsigned long long)ino, 2050 fs->fs_fsmnt); 2051 error = bread(devvp, cgbno, (int)fs->fs_cgsize, 2052 NOCRED, B_MODIFY, &bp); 2053 if (error) { 2054 brelse(bp, 0); 2055 return (error); 2056 } 2057 cgp = (struct cg *)bp->b_data; 2058 if (!cg_chkmagic(cgp, needswap)) { 2059 brelse(bp, 0); 2060 return (0); 2061 } 2062 ffs_freefile_common(ump, fs, dev, bp, ino, mode, true); 2063 2064 bdwrite(bp); 2065 2066 return 0; 2067 } 2068 2069 static void 2070 ffs_freefile_common(struct ufsmount *ump, struct fs *fs, dev_t dev, 2071 struct buf *bp, ino_t ino, int mode, bool devvp_is_snapshot) 2072 { 2073 int cg; 2074 struct cg *cgp; 2075 u_int8_t *inosused; 2076 #ifdef FFS_EI 2077 const int needswap = UFS_FSNEEDSWAP(fs); 2078 #endif 2079 2080 cg = ino_to_cg(fs, ino); 2081 cgp = (struct cg *)bp->b_data; 2082 cgp->cg_old_time = ufs_rw32(time_second, needswap); 2083 if ((fs->fs_magic != FS_UFS1_MAGIC) || 2084 (fs->fs_old_flags & FS_FLAGS_UPDATED)) 2085 cgp->cg_time = ufs_rw64(time_second, needswap); 2086 inosused = cg_inosused(cgp, needswap); 2087 ino %= fs->fs_ipg; 2088 if (isclr(inosused, ino)) { 2089 printf("ifree: dev = 0x%llx, ino = %llu, fs = %s\n", 2090 (unsigned long long)dev, (unsigned long long)ino + 2091 cg * fs->fs_ipg, fs->fs_fsmnt); 2092 if (fs->fs_ronly == 0) 2093 panic("ifree: freeing free inode"); 2094 } 2095 clrbit(inosused, ino); 2096 if (!devvp_is_snapshot) 2097 UFS_WAPBL_UNREGISTER_INODE(ump->um_mountp, 2098 ino + cg * fs->fs_ipg, mode); 2099 if (ino < ufs_rw32(cgp->cg_irotor, needswap)) 2100 cgp->cg_irotor = ufs_rw32(ino, needswap); 2101 ufs_add32(cgp->cg_cs.cs_nifree, 1, needswap); 2102 mutex_enter(&ump->um_lock); 2103 fs->fs_cstotal.cs_nifree++; 2104 fs->fs_cs(fs, cg).cs_nifree++; 2105 if ((mode & IFMT) == IFDIR) { 2106 ufs_add32(cgp->cg_cs.cs_ndir, -1, needswap); 2107 fs->fs_cstotal.cs_ndir--; 2108 fs->fs_cs(fs, cg).cs_ndir--; 2109 } 2110 fs->fs_fmod = 1; 2111 ACTIVECG_CLR(fs, cg); 2112 mutex_exit(&ump->um_lock); 2113 } 2114 2115 /* 2116 * Check to see if a file is free. 2117 */ 2118 int 2119 ffs_checkfreefile(struct fs *fs, struct vnode *devvp, ino_t ino) 2120 { 2121 struct cg *cgp; 2122 struct buf *bp; 2123 daddr_t cgbno; 2124 int ret, cg; 2125 u_int8_t *inosused; 2126 const bool devvp_is_snapshot = (devvp->v_type != VBLK); 2127 2128 KASSERT(devvp_is_snapshot); 2129 2130 cg = ino_to_cg(fs, ino); 2131 if (devvp_is_snapshot) 2132 cgbno = fragstoblks(fs, cgtod(fs, cg)); 2133 else 2134 cgbno = fsbtodb(fs, cgtod(fs, cg)); 2135 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg) 2136 return 1; 2137 if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, 0, &bp)) { 2138 brelse(bp, 0); 2139 return 1; 2140 } 2141 cgp = (struct cg *)bp->b_data; 2142 if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs))) { 2143 brelse(bp, 0); 2144 return 1; 2145 } 2146 inosused = cg_inosused(cgp, UFS_FSNEEDSWAP(fs)); 2147 ino %= fs->fs_ipg; 2148 ret = isclr(inosused, ino); 2149 brelse(bp, 0); 2150 return ret; 2151 } 2152 2153 /* 2154 * Find a block of the specified size in the specified cylinder group. 2155 * 2156 * It is a panic if a request is made to find a block if none are 2157 * available. 2158 */ 2159 static int32_t 2160 ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz) 2161 { 2162 int32_t bno; 2163 int start, len, loc, i; 2164 int blk, field, subfield, pos; 2165 int ostart, olen; 2166 u_int8_t *blksfree; 2167 #ifdef FFS_EI 2168 const int needswap = UFS_FSNEEDSWAP(fs); 2169 #endif 2170 2171 /* KASSERT(mutex_owned(&ump->um_lock)); */ 2172 2173 /* 2174 * find the fragment by searching through the free block 2175 * map for an appropriate bit pattern 2176 */ 2177 if (bpref) 2178 start = dtogd(fs, bpref) / NBBY; 2179 else 2180 start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY; 2181 blksfree = cg_blksfree(cgp, needswap); 2182 len = howmany(fs->fs_fpg, NBBY) - start; 2183 ostart = start; 2184 olen = len; 2185 loc = scanc((u_int)len, 2186 (const u_char *)&blksfree[start], 2187 (const u_char *)fragtbl[fs->fs_frag], 2188 (1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1))))); 2189 if (loc == 0) { 2190 len = start + 1; 2191 start = 0; 2192 loc = scanc((u_int)len, 2193 (const u_char *)&blksfree[0], 2194 (const u_char *)fragtbl[fs->fs_frag], 2195 (1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1))))); 2196 if (loc == 0) { 2197 printf("start = %d, len = %d, fs = %s\n", 2198 ostart, olen, fs->fs_fsmnt); 2199 printf("offset=%d %ld\n", 2200 ufs_rw32(cgp->cg_freeoff, needswap), 2201 (long)blksfree - (long)cgp); 2202 printf("cg %d\n", cgp->cg_cgx); 2203 panic("ffs_alloccg: map corrupted"); 2204 /* NOTREACHED */ 2205 } 2206 } 2207 bno = (start + len - loc) * NBBY; 2208 cgp->cg_frotor = ufs_rw32(bno, needswap); 2209 /* 2210 * found the byte in the map 2211 * sift through the bits to find the selected frag 2212 */ 2213 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { 2214 blk = blkmap(fs, blksfree, bno); 2215 blk <<= 1; 2216 field = around[allocsiz]; 2217 subfield = inside[allocsiz]; 2218 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { 2219 if ((blk & field) == subfield) 2220 return (bno + pos); 2221 field <<= 1; 2222 subfield <<= 1; 2223 } 2224 } 2225 printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt); 2226 panic("ffs_alloccg: block not in map"); 2227 /* return (-1); */ 2228 } 2229 2230 /* 2231 * Fserr prints the name of a file system with an error diagnostic. 2232 * 2233 * The form of the error message is: 2234 * fs: error message 2235 */ 2236 static void 2237 ffs_fserr(struct fs *fs, u_int uid, const char *cp) 2238 { 2239 2240 log(LOG_ERR, "uid %d, pid %d, command %s, on %s: %s\n", 2241 uid, curproc->p_pid, curproc->p_comm, fs->fs_fsmnt, cp); 2242 } 2243
Indexes created Mon Oct 13 01:09:56 GMT 2025