lfs_balloc.c revision 1.51
1 /* $NetBSD: lfs_balloc.c,v 1.51 2005/03/02 21:16:09 perseant Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1989, 1991, 1993 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_balloc.c 8.4 (Berkeley) 5/8/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_balloc.c,v 1.51 2005/03/02 21:16:09 perseant Exp $"); 71 72 #if defined(_KERNEL_OPT) 73 #include "opt_quota.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/buf.h> 79 #include <sys/proc.h> 80 #include <sys/vnode.h> 81 #include <sys/mount.h> 82 #include <sys/resourcevar.h> 83 #include <sys/trace.h> 84 #include <sys/malloc.h> 85 86 #include <miscfs/specfs/specdev.h> 87 88 #include <ufs/ufs/quota.h> 89 #include <ufs/ufs/inode.h> 90 #include <ufs/ufs/ufsmount.h> 91 #include <ufs/ufs/ufs_extern.h> 92 93 #include <ufs/lfs/lfs.h> 94 #include <ufs/lfs/lfs_extern.h> 95 96 #include <uvm/uvm.h> 97 98 int lfs_fragextend(struct vnode *, int, int, daddr_t, struct buf **, struct ucred *); 99 100 u_int64_t locked_fakequeue_count; 101 102 /* 103 * Allocate a block, and to inode and filesystem block accounting for it 104 * and for any indirect blocks the may need to be created in order for 105 * this block to be created. 106 * 107 * Blocks which have never been accounted for (i.e., which "do not exist") 108 * have disk address 0, which is translated by ufs_bmap to the special value 109 * UNASSIGNED == -1, as in the historical UFS. 110 * 111 * Blocks which have been accounted for but which have not yet been written 112 * to disk are given the new special disk address UNWRITTEN == -2, so that 113 * they can be differentiated from completely new blocks. 114 */ 115 /* VOP_BWRITE NIADDR+2 times */ 116 int 117 lfs_balloc(void *v) 118 { 119 struct vop_balloc_args /* { 120 struct vnode *a_vp; 121 off_t a_startoffset; 122 int a_size; 123 struct ucred *a_cred; 124 int a_flags; 125 struct buf *a_bpp; 126 } */ *ap = v; 127 struct vnode *vp; 128 int offset; 129 u_long iosize; 130 daddr_t daddr, idaddr; 131 struct buf *ibp, *bp, **bpp; 132 struct inode *ip; 133 struct lfs *fs; 134 struct indir indirs[NIADDR+2], *idp; 135 daddr_t lbn, lastblock; 136 int bb, bcount; 137 int error, frags, i, nsize, osize, num; 138 139 vp = ap->a_vp; 140 ip = VTOI(vp); 141 fs = ip->i_lfs; 142 offset = blkoff(fs, ap->a_startoffset); 143 iosize = ap->a_size; 144 KASSERT(iosize <= fs->lfs_bsize); 145 lbn = lblkno(fs, ap->a_startoffset); 146 /* (void)lfs_check(vp, lbn, 0); */ 147 bpp = ap->a_bpp; 148 149 /* 150 * Three cases: it's a block beyond the end of file, it's a block in 151 * the file that may or may not have been assigned a disk address or 152 * we're writing an entire block. 153 * 154 * Note, if the daddr is UNWRITTEN, the block already exists in 155 * the cache (it was read or written earlier). If so, make sure 156 * we don't count it as a new block or zero out its contents. If 157 * it did not, make sure we allocate any necessary indirect 158 * blocks. 159 * 160 * If we are writing a block beyond the end of the file, we need to 161 * check if the old last block was a fragment. If it was, we need 162 * to rewrite it. 163 */ 164 165 if (bpp) 166 *bpp = NULL; 167 168 /* Check for block beyond end of file and fragment extension needed. */ 169 lastblock = lblkno(fs, ip->i_size); 170 if (lastblock < NDADDR && lastblock < lbn) { 171 osize = blksize(fs, ip, lastblock); 172 if (osize < fs->lfs_bsize && osize > 0) { 173 if ((error = lfs_fragextend(vp, osize, fs->lfs_bsize, 174 lastblock, 175 (bpp ? &bp : NULL), 176 ap->a_cred))) 177 return (error); 178 ip->i_ffs1_size = ip->i_size = 179 (lastblock + 1) * fs->lfs_bsize; 180 uvm_vnp_setsize(vp, ip->i_size); 181 ip->i_flag |= IN_CHANGE | IN_UPDATE; 182 if (bpp) 183 (void) VOP_BWRITE(bp); 184 } 185 } 186 187 /* 188 * If the block we are writing is a direct block, it's the last 189 * block in the file, and offset + iosize is less than a full 190 * block, we can write one or more fragments. There are two cases: 191 * the block is brand new and we should allocate it the correct 192 * size or it already exists and contains some fragments and 193 * may need to extend it. 194 */ 195 if (lbn < NDADDR && lblkno(fs, ip->i_size) <= lbn) { 196 osize = blksize(fs, ip, lbn); 197 nsize = fragroundup(fs, offset + iosize); 198 if (lblktosize(fs, lbn) >= ip->i_size) { 199 /* Brand new block or fragment */ 200 frags = numfrags(fs, nsize); 201 bb = fragstofsb(fs, frags); 202 if (!ISSPACE(fs, bb, ap->a_cred)) 203 return ENOSPC; 204 if (bpp) { 205 *ap->a_bpp = bp = getblk(vp, lbn, nsize, 0, 0); 206 bp->b_blkno = UNWRITTEN; 207 if (ap->a_flags & B_CLRBUF) 208 clrbuf(bp); 209 } 210 ip->i_lfs_effnblks += bb; 211 ip->i_lfs->lfs_bfree -= bb; 212 ip->i_ffs1_db[lbn] = UNWRITTEN; 213 } else { 214 if (nsize <= osize) { 215 /* No need to extend */ 216 if (bpp && (error = bread(vp, lbn, osize, NOCRED, &bp))) 217 return error; 218 } else { 219 /* Extend existing block */ 220 if ((error = 221 lfs_fragextend(vp, osize, nsize, lbn, 222 (bpp ? &bp : NULL), 223 ap->a_cred))) 224 return error; 225 } 226 if (bpp) 227 *bpp = bp; 228 } 229 return 0; 230 } 231 232 error = ufs_bmaparray(vp, lbn, &daddr, &indirs[0], &num, NULL, NULL); 233 if (error) 234 return (error); 235 236 daddr = (daddr_t)((int32_t)daddr); /* XXX ondisk32 */ 237 KASSERT(daddr <= LFS_MAX_DADDR); 238 239 /* 240 * Do byte accounting all at once, so we can gracefully fail *before* 241 * we start assigning blocks. 242 */ 243 bb = VFSTOUFS(vp->v_mount)->um_seqinc; 244 bcount = 0; 245 if (daddr == UNASSIGNED) { 246 bcount = bb; 247 } 248 for (i = 1; i < num; ++i) { 249 if (!indirs[i].in_exists) { 250 bcount += bb; 251 } 252 } 253 if (ISSPACE(fs, bcount, ap->a_cred)) { 254 ip->i_lfs->lfs_bfree -= bcount; 255 ip->i_lfs_effnblks += bcount; 256 } else { 257 return ENOSPC; 258 } 259 260 if (daddr == UNASSIGNED) { 261 if (num > 0 && ip->i_ffs1_ib[indirs[0].in_off] == 0) { 262 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 263 } 264 265 /* 266 * Create new indirect blocks if necessary 267 */ 268 if (num > 1) { 269 idaddr = ip->i_ffs1_ib[indirs[0].in_off]; 270 for (i = 1; i < num; ++i) { 271 ibp = getblk(vp, indirs[i].in_lbn, 272 fs->lfs_bsize, 0,0); 273 if (!indirs[i].in_exists) { 274 clrbuf(ibp); 275 ibp->b_blkno = UNWRITTEN; 276 } else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) { 277 ibp->b_blkno = fsbtodb(fs, idaddr); 278 ibp->b_flags |= B_READ; 279 VOP_STRATEGY(vp, ibp); 280 biowait(ibp); 281 } 282 /* 283 * This block exists, but the next one may not. 284 * If that is the case mark it UNWRITTEN to keep 285 * the accounting straight. 286 */ 287 /* XXX ondisk32 */ 288 if (((int32_t *)ibp->b_data)[indirs[i].in_off] == 0) 289 ((int32_t *)ibp->b_data)[indirs[i].in_off] = 290 UNWRITTEN; 291 /* XXX ondisk32 */ 292 idaddr = ((int32_t *)ibp->b_data)[indirs[i].in_off]; 293 if ((error = VOP_BWRITE(ibp))) 294 return error; 295 } 296 } 297 } 298 299 300 /* 301 * Get the existing block from the cache, if requested. 302 */ 303 frags = fsbtofrags(fs, bb); 304 if (bpp) 305 *bpp = bp = getblk(vp, lbn, blksize(fs, ip, lbn), 0, 0); 306 307 /* 308 * Do accounting on blocks that represent pages. 309 */ 310 if (!bpp) 311 lfs_register_block(vp, lbn); 312 313 /* 314 * The block we are writing may be a brand new block 315 * in which case we need to do accounting. 316 * 317 * We can tell a truly new block because ufs_bmaparray will say 318 * it is UNASSIGNED. Once we allocate it we will assign it the 319 * disk address UNWRITTEN. 320 */ 321 if (daddr == UNASSIGNED) { 322 if (bpp) { 323 if (ap->a_flags & B_CLRBUF) 324 clrbuf(bp); 325 326 /* Note the new address */ 327 bp->b_blkno = UNWRITTEN; 328 } 329 330 switch (num) { 331 case 0: 332 ip->i_ffs1_db[lbn] = UNWRITTEN; 333 break; 334 case 1: 335 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 336 break; 337 default: 338 idp = &indirs[num - 1]; 339 if (bread(vp, idp->in_lbn, fs->lfs_bsize, NOCRED, 340 &ibp)) 341 panic("lfs_balloc: bread bno %lld", 342 (long long)idp->in_lbn); 343 /* XXX ondisk32 */ 344 ((int32_t *)ibp->b_data)[idp->in_off] = UNWRITTEN; 345 VOP_BWRITE(ibp); 346 } 347 } else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) { 348 /* 349 * Not a brand new block, also not in the cache; 350 * read it in from disk. 351 */ 352 if (iosize == fs->lfs_bsize) 353 /* Optimization: I/O is unnecessary. */ 354 bp->b_blkno = daddr; 355 else { 356 /* 357 * We need to read the block to preserve the 358 * existing bytes. 359 */ 360 bp->b_blkno = daddr; 361 bp->b_flags |= B_READ; 362 VOP_STRATEGY(vp, bp); 363 return (biowait(bp)); 364 } 365 } 366 367 return (0); 368 } 369 370 /* VOP_BWRITE 1 time */ 371 int 372 lfs_fragextend(struct vnode *vp, int osize, int nsize, daddr_t lbn, struct buf **bpp, struct ucred *cred) 373 { 374 struct inode *ip; 375 struct lfs *fs; 376 long bb; 377 int error; 378 extern long locked_queue_bytes; 379 size_t obufsize; 380 381 ip = VTOI(vp); 382 fs = ip->i_lfs; 383 bb = (long)fragstofsb(fs, numfrags(fs, nsize - osize)); 384 error = 0; 385 386 /* 387 * Get the seglock so we don't enlarge blocks while a segment 388 * is being written. If we're called with bpp==NULL, though, 389 * we are only pretending to change a buffer, so we don't have to 390 * lock. 391 */ 392 top: 393 if (bpp) { 394 lockmgr(&fs->lfs_fraglock, LK_SHARED, 0); 395 LFS_DEBUG_COUNTLOCKED("frag"); 396 } 397 398 if (!ISSPACE(fs, bb, cred)) { 399 error = ENOSPC; 400 goto out; 401 } 402 403 /* 404 * If we are not asked to actually return the block, all we need 405 * to do is allocate space for it. UBC will handle dirtying the 406 * appropriate things and making sure it all goes to disk. 407 * Don't bother to read in that case. 408 */ 409 if (bpp && (error = bread(vp, lbn, osize, NOCRED, bpp))) { 410 brelse(*bpp); 411 goto out; 412 } 413 #ifdef QUOTA 414 if ((error = chkdq(ip, bb, cred, 0))) { 415 if (bpp) 416 brelse(*bpp); 417 goto out; 418 } 419 #endif 420 /* 421 * Adjust accounting for lfs_avail. If there's not enough room, 422 * we will have to wait for the cleaner, which we can't do while 423 * holding a block busy or while holding the seglock. In that case, 424 * release both and start over after waiting. 425 */ 426 427 if (bpp && ((*bpp)->b_flags & B_DELWRI)) { 428 if (!lfs_fits(fs, bb)) { 429 if (bpp) 430 brelse(*bpp); 431 #ifdef QUOTA 432 chkdq(ip, -bb, cred, 0); 433 #endif 434 lockmgr(&fs->lfs_fraglock, LK_RELEASE, 0); 435 lfs_availwait(fs, bb); 436 goto top; 437 } 438 fs->lfs_avail -= bb; 439 } 440 441 fs->lfs_bfree -= bb; 442 ip->i_lfs_effnblks += bb; 443 ip->i_flag |= IN_CHANGE | IN_UPDATE; 444 445 if (bpp) { 446 obufsize = (*bpp)->b_bufsize; 447 allocbuf(*bpp, nsize, 1); 448 449 /* Adjust locked-list accounting */ 450 if (((*bpp)->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 451 locked_queue_bytes += (*bpp)->b_bufsize - obufsize; 452 453 bzero((char *)((*bpp)->b_data) + osize, (u_int)(nsize - osize)); 454 } 455 456 out: 457 if (bpp) { 458 lockmgr(&fs->lfs_fraglock, LK_RELEASE, 0); 459 } 460 return (error); 461 } 462 463 /* 464 * Record this lbn as being "write pending". We used to have this information 465 * on the buffer headers, but since pages don't have buffer headers we 466 * record it here instead. 467 */ 468 469 void 470 lfs_register_block(struct vnode *vp, daddr_t lbn) 471 { 472 struct lfs *fs; 473 struct inode *ip; 474 struct lbnentry *lbp; 475 int hash; 476 477 /* Don't count metadata */ 478 if (lbn < 0 || vp->v_type != VREG || VTOI(vp)->i_number == LFS_IFILE_INUM) 479 return; 480 481 ip = VTOI(vp); 482 fs = ip->i_lfs; 483 484 /* If no space, wait for the cleaner */ 485 lfs_availwait(fs, btofsb(fs, 1 << fs->lfs_bshift)); 486 487 hash = lbn % LFS_BLIST_HASH_WIDTH; 488 LIST_FOREACH(lbp, &(ip->i_lfs_blist[hash]), entry) { 489 if (lbp->lbn == lbn) 490 return; 491 } 492 493 lbp = (struct lbnentry *)pool_get(&lfs_lbnentry_pool, PR_WAITOK); 494 lbp->lbn = lbn; 495 LIST_INSERT_HEAD(&(ip->i_lfs_blist[hash]), lbp, entry); 496 fs->lfs_favail += btofsb(fs, (1 << fs->lfs_bshift)); 497 ++locked_fakequeue_count; 498 } 499 500 static void 501 lfs_do_deregister(struct lfs *fs, struct lbnentry *lbp) 502 { 503 LIST_REMOVE(lbp, entry); 504 pool_put(&lfs_lbnentry_pool, lbp); 505 if (fs->lfs_favail > btofsb(fs, (1 << fs->lfs_bshift))) 506 fs->lfs_favail -= btofsb(fs, (1 << fs->lfs_bshift)); 507 if (locked_fakequeue_count > 0) 508 --locked_fakequeue_count; 509 } 510 511 void 512 lfs_deregister_block(struct vnode *vp, daddr_t lbn) 513 { 514 struct lfs *fs; 515 struct inode *ip; 516 struct lbnentry *lbp; 517 int hash; 518 519 /* Don't count metadata */ 520 if (lbn < 0 || vp->v_type != VREG || VTOI(vp)->i_number == LFS_IFILE_INUM) 521 return; 522 523 ip = VTOI(vp); 524 fs = ip->i_lfs; 525 hash = lbn % LFS_BLIST_HASH_WIDTH; 526 LIST_FOREACH(lbp, &(ip->i_lfs_blist[hash]), entry) { 527 if (lbp->lbn == lbn) 528 break; 529 } 530 if (lbp == NULL) 531 return; 532 533 lfs_do_deregister(fs, lbp); 534 } 535 536 void 537 lfs_deregister_all(struct vnode *vp) 538 { 539 struct lbnentry *lbp; 540 struct lfs *fs; 541 struct inode *ip; 542 int i; 543 544 ip = VTOI(vp); 545 fs = ip->i_lfs; 546 for (i = 0; i < LFS_BLIST_HASH_WIDTH; i++) 547 while((lbp = LIST_FIRST(&(ip->i_lfs_blist[i]))) != NULL) 548 lfs_do_deregister(fs, lbp); 549 } 550
Indexes created Mon Sep 29 21:09:56 GMT 2025