1/* $NetBSD: lfs_balloc.c,v 1.69 2010/02/16 23:20:30 mlelstv 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@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 * 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 * Copyright (c) 1989, 1991, 1993 33 * The Regents of the University of California. All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. Neither the name of the University nor the names of its contributors 44 * may be used to endorse or promote products derived from this software 45 * without specific prior written permission. 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 57 * SUCH DAMAGE. 58 * 59 * @(#)lfs_balloc.c 8.4 (Berkeley) 5/8/95 60 */ 61 62#include <sys/cdefs.h> 63__KERNEL_RCSID(0, "$NetBSD: lfs_balloc.c,v 1.69 2010/02/16 23:20:30 mlelstv Exp $"); 64 65#if defined(_KERNEL_OPT) 66#include "opt_quota.h" 67#endif 68 69#include <sys/param.h> 70#include <sys/systm.h> 71#include <sys/buf.h> 72#include <sys/proc.h> 73#include <sys/vnode.h> 74#include <sys/mount.h> 75#include <sys/resourcevar.h> 76#include <sys/tree.h> 77#include <sys/trace.h> 78#include <sys/kauth.h> 79 80#include <miscfs/specfs/specdev.h> 81 82#include <ufs/ufs/quota.h> 83#include <ufs/ufs/inode.h> 84#include <ufs/ufs/ufsmount.h> 85#include <ufs/ufs/ufs_extern.h> 86 87#include <ufs/lfs/lfs.h> 88#include <ufs/lfs/lfs_extern.h> 89 90#include <uvm/uvm.h> 91 92int lfs_fragextend(struct vnode *, int, int, daddr_t, struct buf **, kauth_cred_t); 93 94u_int64_t locked_fakequeue_count; 95 96/* 97 * Allocate a block, and to inode and filesystem block accounting for it 98 * and for any indirect blocks the may need to be created in order for 99 * this block to be created. 100 * 101 * Blocks which have never been accounted for (i.e., which "do not exist") 102 * have disk address 0, which is translated by ufs_bmap to the special value 103 * UNASSIGNED == -1, as in the historical UFS. 104 * 105 * Blocks which have been accounted for but which have not yet been written 106 * to disk are given the new special disk address UNWRITTEN == -2, so that 107 * they can be differentiated from completely new blocks. 108 */ 109/* VOP_BWRITE NIADDR+2 times */ 110int 111lfs_balloc(struct vnode *vp, off_t startoffset, int iosize, kauth_cred_t cred, 112 int flags, struct buf **bpp) 113{ 114 int offset; 115 daddr_t daddr, idaddr; 116 struct buf *ibp, *bp; 117 struct inode *ip; 118 struct lfs *fs; 119 struct indir indirs[NIADDR+2], *idp; 120 daddr_t lbn, lastblock; 121 int bcount; 122 int error, frags, i, nsize, osize, num; 123 124 ip = VTOI(vp); 125 fs = ip->i_lfs; 126 offset = blkoff(fs, startoffset); 127 KASSERT(iosize <= fs->lfs_bsize); 128 lbn = lblkno(fs, startoffset); 129 /* (void)lfs_check(vp, lbn, 0); */ 130 131 ASSERT_MAYBE_SEGLOCK(fs); 132 133 /* 134 * Three cases: it's a block beyond the end of file, it's a block in 135 * the file that may or may not have been assigned a disk address or 136 * we're writing an entire block. 137 * 138 * Note, if the daddr is UNWRITTEN, the block already exists in 139 * the cache (it was read or written earlier). If so, make sure 140 * we don't count it as a new block or zero out its contents. If 141 * it did not, make sure we allocate any necessary indirect 142 * blocks. 143 * 144 * If we are writing a block beyond the end of the file, we need to 145 * check if the old last block was a fragment. If it was, we need 146 * to rewrite it. 147 */ 148 149 if (bpp) 150 *bpp = NULL; 151 152 /* Check for block beyond end of file and fragment extension needed. */ 153 lastblock = lblkno(fs, ip->i_size); 154 if (lastblock < NDADDR && lastblock < lbn) { 155 osize = blksize(fs, ip, lastblock); 156 if (osize < fs->lfs_bsize && osize > 0) { 157 if ((error = lfs_fragextend(vp, osize, fs->lfs_bsize, 158 lastblock, 159 (bpp ? &bp : NULL), cred))) 160 return (error); 161 ip->i_ffs1_size = ip->i_size = 162 (lastblock + 1) * fs->lfs_bsize; 163 uvm_vnp_setsize(vp, ip->i_size); 164 ip->i_flag |= IN_CHANGE | IN_UPDATE; 165 if (bpp) 166 (void) VOP_BWRITE(bp->b_vp, bp); 167 } 168 } 169 170 /* 171 * If the block we are writing is a direct block, it's the last 172 * block in the file, and offset + iosize is less than a full 173 * block, we can write one or more fragments. There are two cases: 174 * the block is brand new and we should allocate it the correct 175 * size or it already exists and contains some fragments and 176 * may need to extend it. 177 */ 178 if (lbn < NDADDR && lblkno(fs, ip->i_size) <= lbn) { 179 osize = blksize(fs, ip, lbn); 180 nsize = fragroundup(fs, offset + iosize); 181 if (lblktosize(fs, lbn) >= ip->i_size) { 182 /* Brand new block or fragment */ 183 frags = numfrags(fs, nsize); 184 if (!ISSPACE(fs, frags, cred)) 185 return ENOSPC; 186 if (bpp) { 187 *bpp = bp = getblk(vp, lbn, nsize, 0, 0); 188 bp->b_blkno = UNWRITTEN; 189 if (flags & B_CLRBUF) 190 clrbuf(bp); 191 } 192 ip->i_lfs_effnblks += frags; 193 mutex_enter(&lfs_lock); 194 fs->lfs_bfree -= frags; 195 mutex_exit(&lfs_lock); 196 ip->i_ffs1_db[lbn] = UNWRITTEN; 197 } else { 198 if (nsize <= osize) { 199 /* No need to extend */ 200 if (bpp && (error = bread(vp, lbn, osize, 201 NOCRED, 0, &bp))) 202 return error; 203 } else { 204 /* Extend existing block */ 205 if ((error = 206 lfs_fragextend(vp, osize, nsize, lbn, 207 (bpp ? &bp : NULL), cred))) 208 return error; 209 } 210 if (bpp) 211 *bpp = bp; 212 } 213 return 0; 214 } 215 216 error = ufs_bmaparray(vp, lbn, &daddr, &indirs[0], &num, NULL, NULL); 217 if (error) 218 return (error); 219 220 daddr = (daddr_t)((int32_t)daddr); /* XXX ondisk32 */ 221 KASSERT(daddr <= LFS_MAX_DADDR); 222 223 /* 224 * Do byte accounting all at once, so we can gracefully fail *before* 225 * we start assigning blocks. 226 */ 227 frags = VFSTOUFS(vp->v_mount)->um_seqinc; 228 bcount = 0; 229 if (daddr == UNASSIGNED) { 230 bcount = frags; 231 } 232 for (i = 1; i < num; ++i) { 233 if (!indirs[i].in_exists) { 234 bcount += frags; 235 } 236 } 237 if (ISSPACE(fs, bcount, cred)) { 238 mutex_enter(&lfs_lock); 239 fs->lfs_bfree -= bcount; 240 mutex_exit(&lfs_lock); 241 ip->i_lfs_effnblks += bcount; 242 } else { 243 return ENOSPC; 244 } 245 246 if (daddr == UNASSIGNED) { 247 if (num > 0 && ip->i_ffs1_ib[indirs[0].in_off] == 0) { 248 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 249 } 250 251 /* 252 * Create new indirect blocks if necessary 253 */ 254 if (num > 1) { 255 idaddr = ip->i_ffs1_ib[indirs[0].in_off]; 256 for (i = 1; i < num; ++i) { 257 ibp = getblk(vp, indirs[i].in_lbn, 258 fs->lfs_bsize, 0,0); 259 if (!indirs[i].in_exists) { 260 clrbuf(ibp); 261 ibp->b_blkno = UNWRITTEN; 262 } else if (!(ibp->b_oflags & (BO_DELWRI | BO_DONE))) { 263 ibp->b_blkno = fsbtodb(fs, idaddr); 264 ibp->b_flags |= B_READ; 265 VOP_STRATEGY(vp, ibp); 266 biowait(ibp); 267 } 268 /* 269 * This block exists, but the next one may not. 270 * If that is the case mark it UNWRITTEN to keep 271 * the accounting straight. 272 */ 273 /* XXX ondisk32 */ 274 if (((int32_t *)ibp->b_data)[indirs[i].in_off] == 0) 275 ((int32_t *)ibp->b_data)[indirs[i].in_off] = 276 UNWRITTEN; 277 /* XXX ondisk32 */ 278 idaddr = ((int32_t *)ibp->b_data)[indirs[i].in_off]; 279#ifdef DEBUG 280 if (vp == fs->lfs_ivnode) { 281 LFS_ENTER_LOG("balloc", __FILE__, 282 __LINE__, indirs[i].in_lbn, 283 ibp->b_flags, curproc->p_pid); 284 } 285#endif 286 if ((error = VOP_BWRITE(ibp->b_vp, ibp))) 287 return error; 288 } 289 } 290 } 291 292 293 /* 294 * Get the existing block from the cache, if requested. 295 */ 296 if (bpp) 297 *bpp = bp = getblk(vp, lbn, blksize(fs, ip, lbn), 0, 0); 298 299 /* 300 * Do accounting on blocks that represent pages. 301 */ 302 if (!bpp) 303 lfs_register_block(vp, lbn); 304 305 /* 306 * The block we are writing may be a brand new block 307 * in which case we need to do accounting. 308 * 309 * We can tell a truly new block because ufs_bmaparray will say 310 * it is UNASSIGNED. Once we allocate it we will assign it the 311 * disk address UNWRITTEN. 312 */ 313 if (daddr == UNASSIGNED) { 314 if (bpp) { 315 if (flags & B_CLRBUF) 316 clrbuf(bp); 317 318 /* Note the new address */ 319 bp->b_blkno = UNWRITTEN; 320 } 321 322 switch (num) { 323 case 0: 324 ip->i_ffs1_db[lbn] = UNWRITTEN; 325 break; 326 case 1: 327 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 328 break; 329 default: 330 idp = &indirs[num - 1]; 331 if (bread(vp, idp->in_lbn, fs->lfs_bsize, NOCRED, 332 B_MODIFY, &ibp)) 333 panic("lfs_balloc: bread bno %lld", 334 (long long)idp->in_lbn); 335 /* XXX ondisk32 */ 336 ((int32_t *)ibp->b_data)[idp->in_off] = UNWRITTEN; 337#ifdef DEBUG 338 if (vp == fs->lfs_ivnode) { 339 LFS_ENTER_LOG("balloc", __FILE__, 340 __LINE__, idp->in_lbn, 341 ibp->b_flags, curproc->p_pid); 342 } 343#endif 344 VOP_BWRITE(ibp->b_vp, ibp); 345 } 346 } else if (bpp && !(bp->b_oflags & (BO_DONE|BO_DELWRI))) { 347 /* 348 * Not a brand new block, also not in the cache; 349 * read it in from disk. 350 */ 351 if (iosize == fs->lfs_bsize) 352 /* Optimization: I/O is unnecessary. */ 353 bp->b_blkno = daddr; 354 else { 355 /* 356 * We need to read the block to preserve the 357 * existing bytes. 358 */ 359 bp->b_blkno = daddr; 360 bp->b_flags |= B_READ; 361 VOP_STRATEGY(vp, bp); 362 return (biowait(bp)); 363 } 364 } 365 366 return (0); 367} 368 369/* VOP_BWRITE 1 time */ 370int 371lfs_fragextend(struct vnode *vp, int osize, int nsize, daddr_t lbn, struct buf **bpp, 372 kauth_cred_t cred) 373{ 374 struct inode *ip; 375 struct lfs *fs; 376 long frags; 377 int error; 378 extern long locked_queue_bytes; 379 size_t obufsize; 380 381 ip = VTOI(vp); 382 fs = ip->i_lfs; 383 frags = (long)numfrags(fs, nsize - osize); 384 error = 0; 385 386 ASSERT_NO_SEGLOCK(fs); 387 388 /* 389 * Get the seglock so we don't enlarge blocks while a segment 390 * is being written. If we're called with bpp==NULL, though, 391 * we are only pretending to change a buffer, so we don't have to 392 * lock. 393 */ 394 top: 395 if (bpp) { 396 rw_enter(&fs->lfs_fraglock, RW_READER); 397 LFS_DEBUG_COUNTLOCKED("frag"); 398 } 399 400 if (!ISSPACE(fs, frags, cred)) { 401 error = ENOSPC; 402 goto out; 403 } 404 405 /* 406 * If we are not asked to actually return the block, all we need 407 * to do is allocate space for it. UBC will handle dirtying the 408 * appropriate things and making sure it all goes to disk. 409 * Don't bother to read in that case. 410 */ 411 if (bpp && (error = bread(vp, lbn, osize, NOCRED, 0, bpp))) { 412 brelse(*bpp, 0); 413 goto out; 414 } 415#ifdef QUOTA 416 if ((error = chkdq(ip, frags, cred, 0))) { 417 if (bpp) 418 brelse(*bpp, 0); 419 goto out; 420 } 421#endif 422 /* 423 * Adjust accounting for lfs_avail. If there's not enough room, 424 * we will have to wait for the cleaner, which we can't do while 425 * holding a block busy or while holding the seglock. In that case, 426 * release both and start over after waiting. 427 */ 428 429 if (bpp && ((*bpp)->b_oflags & BO_DELWRI)) { 430 if (!lfs_fits(fs, frags)) { 431 if (bpp) 432 brelse(*bpp, 0); 433#ifdef QUOTA 434 chkdq(ip, -frags, cred, 0); 435#endif 436 rw_exit(&fs->lfs_fraglock); 437 lfs_availwait(fs, frags); 438 goto top; 439 } 440 fs->lfs_avail -= frags; 441 } 442 443 mutex_enter(&lfs_lock); 444 fs->lfs_bfree -= frags; 445 mutex_exit(&lfs_lock); 446 ip->i_lfs_effnblks += frags; 447 ip->i_flag |= IN_CHANGE | IN_UPDATE; 448 449 if (bpp) { 450 obufsize = (*bpp)->b_bufsize; 451 allocbuf(*bpp, nsize, 1); 452 453 /* Adjust locked-list accounting */ 454 if (((*bpp)->b_flags & B_LOCKED) != 0 && 455 (*bpp)->b_iodone == NULL) { 456 mutex_enter(&lfs_lock); 457 locked_queue_bytes += (*bpp)->b_bufsize - obufsize; 458 mutex_exit(&lfs_lock); 459 } 460 461 memset((char *)((*bpp)->b_data) + osize, 0, (u_int)(nsize - osize)); 462 } 463 464 out: 465 if (bpp) { 466 rw_exit(&fs->lfs_fraglock); 467 } 468 return (error); 469} 470 471static inline int 472lge(struct lbnentry *a, struct lbnentry *b) 473{ 474 return a->lbn - b->lbn; 475} 476 477SPLAY_PROTOTYPE(lfs_splay, lbnentry, entry, lge); 478 479SPLAY_GENERATE(lfs_splay, lbnentry, entry, lge); 480 481/* 482 * Record this lbn as being "write pending". We used to have this information 483 * on the buffer headers, but since pages don't have buffer headers we 484 * record it here instead. 485 */ 486void 487lfs_register_block(struct vnode *vp, daddr_t lbn) 488{ 489 struct lfs *fs; 490 struct inode *ip; 491 struct lbnentry *lbp; 492 493 ip = VTOI(vp); 494 495 /* Don't count metadata */ 496 if (lbn < 0 || vp->v_type != VREG || ip->i_number == LFS_IFILE_INUM) 497 return; 498 499 fs = ip->i_lfs; 500 501 ASSERT_NO_SEGLOCK(fs); 502 503 /* If no space, wait for the cleaner */ 504 lfs_availwait(fs, btofsb(fs, 1 << fs->lfs_bshift)); 505 506 lbp = (struct lbnentry *)pool_get(&lfs_lbnentry_pool, PR_WAITOK); 507 lbp->lbn = lbn; 508 mutex_enter(&lfs_lock); 509 if (SPLAY_INSERT(lfs_splay, &ip->i_lfs_lbtree, lbp) != NULL) { 510 mutex_exit(&lfs_lock); 511 /* Already there */ 512 pool_put(&lfs_lbnentry_pool, lbp); 513 return; 514 } 515 516 ++ip->i_lfs_nbtree; 517 fs->lfs_favail += btofsb(fs, (1 << fs->lfs_bshift)); 518 fs->lfs_pages += fs->lfs_bsize >> PAGE_SHIFT; 519 ++locked_fakequeue_count; 520 lfs_subsys_pages += fs->lfs_bsize >> PAGE_SHIFT; 521 mutex_exit(&lfs_lock); 522} 523 524static void 525lfs_do_deregister(struct lfs *fs, struct inode *ip, struct lbnentry *lbp) 526{ 527 ASSERT_MAYBE_SEGLOCK(fs); 528 529 mutex_enter(&lfs_lock); 530 --ip->i_lfs_nbtree; 531 SPLAY_REMOVE(lfs_splay, &ip->i_lfs_lbtree, lbp); 532 if (fs->lfs_favail > btofsb(fs, (1 << fs->lfs_bshift))) 533 fs->lfs_favail -= btofsb(fs, (1 << fs->lfs_bshift)); 534 fs->lfs_pages -= fs->lfs_bsize >> PAGE_SHIFT; 535 if (locked_fakequeue_count > 0) 536 --locked_fakequeue_count; 537 lfs_subsys_pages -= fs->lfs_bsize >> PAGE_SHIFT; 538 mutex_exit(&lfs_lock); 539 540 pool_put(&lfs_lbnentry_pool, lbp); 541} 542 543void 544lfs_deregister_block(struct vnode *vp, daddr_t lbn) 545{ 546 struct lfs *fs; 547 struct inode *ip; 548 struct lbnentry *lbp; 549 struct lbnentry tmp; 550 551 ip = VTOI(vp); 552 553 /* Don't count metadata */ 554 if (lbn < 0 || vp->v_type != VREG || ip->i_number == LFS_IFILE_INUM) 555 return; 556 557 fs = ip->i_lfs; 558 tmp.lbn = lbn; 559 lbp = SPLAY_FIND(lfs_splay, &ip->i_lfs_lbtree, &tmp); 560 if (lbp == NULL) 561 return; 562 563 lfs_do_deregister(fs, ip, lbp); 564} 565 566void 567lfs_deregister_all(struct vnode *vp) 568{ 569 struct lbnentry *lbp, *nlbp; 570 struct lfs_splay *hd; 571 struct lfs *fs; 572 struct inode *ip; 573 574 ip = VTOI(vp); 575 fs = ip->i_lfs; 576 hd = &ip->i_lfs_lbtree; 577 578 for (lbp = SPLAY_MIN(lfs_splay, hd); lbp != NULL; lbp = nlbp) { 579 nlbp = SPLAY_NEXT(lfs_splay, hd, lbp); 580 lfs_do_deregister(fs, ip, lbp); 581 } 582} 583