1 /* $NetBSD: segwrite.c,v 1.10 2005/09/13 04:14:17 christos Exp $ */ 2 /*- 3 * Copyright (c) 2003 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Konrad E. Schroder <perseant@hhhh.org>. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 /* 38 * Copyright (c) 1991, 1993 39 * The Regents of the University of California. All rights reserved. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * @(#)lfs_segment.c 8.10 (Berkeley) 6/10/95 66 */ 67 68 /* 69 * Partial segment writer, taken from the kernel and adapted for userland. 70 */ 71 #include <sys/types.h> 72 #include <sys/param.h> 73 #include <sys/time.h> 74 #include <sys/buf.h> 75 #include <sys/mount.h> 76 77 #include <ufs/ufs/inode.h> 78 #include <ufs/ufs/ufsmount.h> 79 80 /* Override certain things to make <ufs/lfs/lfs.h> work */ 81 #define vnode uvnode 82 #define buf ubuf 83 #define panic call_panic 84 85 #include <ufs/lfs/lfs.h> 86 87 #include <assert.h> 88 #include <stdio.h> 89 #include <stdlib.h> 90 #include <string.h> 91 #include <err.h> 92 #include <errno.h> 93 94 #include "bufcache.h" 95 #include "vnode.h" 96 #include "lfs_user.h" 97 #include "segwrite.h" 98 99 /* Compatibility definitions */ 100 extern off_t locked_queue_bytes; 101 int locked_queue_count; 102 off_t written_bytes = 0; 103 off_t written_data = 0; 104 off_t written_indir = 0; 105 off_t written_dev = 0; 106 int written_inodes = 0; 107 108 /* Global variables */ 109 time_t write_time; 110 111 extern u_int32_t cksum(void *, size_t); 112 extern u_int32_t lfs_sb_cksum(struct dlfs *); 113 extern int preen; 114 115 /* 116 * Logical block number match routines used when traversing the dirty block 117 * chain. 118 */ 119 int 120 lfs_match_data(struct lfs * fs, struct ubuf * bp) 121 { 122 return (bp->b_lblkno >= 0); 123 } 124 125 int 126 lfs_match_indir(struct lfs * fs, struct ubuf * bp) 127 { 128 daddr_t lbn; 129 130 lbn = bp->b_lblkno; 131 return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0); 132 } 133 134 int 135 lfs_match_dindir(struct lfs * fs, struct ubuf * bp) 136 { 137 daddr_t lbn; 138 139 lbn = bp->b_lblkno; 140 return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1); 141 } 142 143 int 144 lfs_match_tindir(struct lfs * fs, struct ubuf * bp) 145 { 146 daddr_t lbn; 147 148 lbn = bp->b_lblkno; 149 return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2); 150 } 151 152 /* 153 * Do a checkpoint. 154 */ 155 int 156 lfs_segwrite(struct lfs * fs, int flags) 157 { 158 struct inode *ip; 159 struct segment *sp; 160 struct uvnode *vp; 161 int redo; 162 163 lfs_seglock(fs, flags | SEGM_CKP); 164 sp = fs->lfs_sp; 165 166 lfs_writevnodes(fs, sp, VN_REG); 167 lfs_writevnodes(fs, sp, VN_DIROP); 168 ((SEGSUM *) (sp->segsum))->ss_flags &= ~(SS_CONT); 169 170 do { 171 vp = fs->lfs_ivnode; 172 fs->lfs_flags &= ~LFS_IFDIRTY; 173 ip = VTOI(vp); 174 if (LIST_FIRST(&vp->v_dirtyblkhd) != NULL || fs->lfs_idaddr <= 0) 175 lfs_writefile(fs, sp, vp); 176 177 redo = lfs_writeinode(fs, sp, ip); 178 redo += lfs_writeseg(fs, sp); 179 redo += (fs->lfs_flags & LFS_IFDIRTY); 180 } while (redo); 181 182 lfs_segunlock(fs); 183 #if 0 184 printf("wrote %" PRId64 " bytes (%" PRId32 " fsb)\n", 185 written_bytes, (ufs_daddr_t)btofsb(fs, written_bytes)); 186 printf("wrote %" PRId64 " bytes data (%" PRId32 " fsb)\n", 187 written_data, (ufs_daddr_t)btofsb(fs, written_data)); 188 printf("wrote %" PRId64 " bytes indir (%" PRId32 " fsb)\n", 189 written_indir, (ufs_daddr_t)btofsb(fs, written_indir)); 190 printf("wrote %" PRId64 " bytes dev (%" PRId32 " fsb)\n", 191 written_dev, (ufs_daddr_t)btofsb(fs, written_dev)); 192 printf("wrote %d inodes (%" PRId32 " fsb)\n", 193 written_inodes, btofsb(fs, written_inodes * fs->lfs_ibsize)); 194 #endif 195 return 0; 196 } 197 198 /* 199 * Write the dirty blocks associated with a vnode. 200 */ 201 void 202 lfs_writefile(struct lfs * fs, struct segment * sp, struct uvnode * vp) 203 { 204 struct ubuf *bp; 205 struct finfo *fip; 206 struct inode *ip; 207 IFILE *ifp; 208 209 ip = VTOI(vp); 210 211 if (sp->seg_bytes_left < fs->lfs_bsize || 212 sp->sum_bytes_left < sizeof(struct finfo)) 213 (void) lfs_writeseg(fs, sp); 214 215 sp->sum_bytes_left -= FINFOSIZE; 216 ++((SEGSUM *) (sp->segsum))->ss_nfinfo; 217 218 if (vp->v_flag & VDIROP) 219 ((SEGSUM *) (sp->segsum))->ss_flags |= (SS_DIROP | SS_CONT); 220 221 fip = sp->fip; 222 fip->fi_nblocks = 0; 223 fip->fi_ino = ip->i_number; 224 LFS_IENTRY(ifp, fs, fip->fi_ino, bp); 225 fip->fi_version = ifp->if_version; 226 brelse(bp); 227 228 lfs_gather(fs, sp, vp, lfs_match_data); 229 lfs_gather(fs, sp, vp, lfs_match_indir); 230 lfs_gather(fs, sp, vp, lfs_match_dindir); 231 lfs_gather(fs, sp, vp, lfs_match_tindir); 232 233 fip = sp->fip; 234 if (fip->fi_nblocks != 0) { 235 sp->fip = (FINFO *) ((caddr_t) fip + FINFOSIZE + 236 sizeof(ufs_daddr_t) * (fip->fi_nblocks)); 237 sp->start_lbp = &sp->fip->fi_blocks[0]; 238 } else { 239 sp->sum_bytes_left += FINFOSIZE; 240 --((SEGSUM *) (sp->segsum))->ss_nfinfo; 241 } 242 } 243 244 int 245 lfs_writeinode(struct lfs * fs, struct segment * sp, struct inode * ip) 246 { 247 struct ubuf *bp, *ibp; 248 struct ufs1_dinode *cdp; 249 IFILE *ifp; 250 SEGUSE *sup; 251 daddr_t daddr; 252 ino_t ino; 253 int error, i, ndx, fsb = 0; 254 int redo_ifile = 0; 255 struct timespec ts; 256 int gotblk = 0; 257 258 /* Allocate a new inode block if necessary. */ 259 if ((ip->i_number != LFS_IFILE_INUM || sp->idp == NULL) && 260 sp->ibp == NULL) { 261 /* Allocate a new segment if necessary. */ 262 if (sp->seg_bytes_left < fs->lfs_ibsize || 263 sp->sum_bytes_left < sizeof(ufs_daddr_t)) 264 (void) lfs_writeseg(fs, sp); 265 266 /* Get next inode block. */ 267 daddr = fs->lfs_offset; 268 fs->lfs_offset += btofsb(fs, fs->lfs_ibsize); 269 sp->ibp = *sp->cbpp++ = 270 getblk(fs->lfs_devvp, fsbtodb(fs, daddr), 271 fs->lfs_ibsize); 272 sp->ibp->b_flags |= B_GATHERED; 273 gotblk++; 274 275 /* Zero out inode numbers */ 276 for (i = 0; i < INOPB(fs); ++i) 277 ((struct ufs1_dinode *) sp->ibp->b_data)[i].di_inumber = 0; 278 279 ++sp->start_bpp; 280 fs->lfs_avail -= btofsb(fs, fs->lfs_ibsize); 281 /* Set remaining space counters. */ 282 sp->seg_bytes_left -= fs->lfs_ibsize; 283 sp->sum_bytes_left -= sizeof(ufs_daddr_t); 284 ndx = fs->lfs_sumsize / sizeof(ufs_daddr_t) - 285 sp->ninodes / INOPB(fs) - 1; 286 ((ufs_daddr_t *) (sp->segsum))[ndx] = daddr; 287 } 288 /* Update the inode times and copy the inode onto the inode page. */ 289 ts.tv_nsec = 0; 290 ts.tv_sec = write_time; 291 /* XXX kludge --- don't redirty the ifile just to put times on it */ 292 if (ip->i_number != LFS_IFILE_INUM) 293 LFS_ITIMES(ip, &ts, &ts, &ts); 294 295 /* 296 * If this is the Ifile, and we've already written the Ifile in this 297 * partial segment, just overwrite it (it's not on disk yet) and 298 * continue. 299 * 300 * XXX we know that the bp that we get the second time around has 301 * already been gathered. 302 */ 303 if (ip->i_number == LFS_IFILE_INUM && sp->idp) { 304 *(sp->idp) = *ip->i_din.ffs1_din; 305 ip->i_lfs_osize = ip->i_ffs1_size; 306 return 0; 307 } 308 bp = sp->ibp; 309 cdp = ((struct ufs1_dinode *) bp->b_data) + (sp->ninodes % INOPB(fs)); 310 *cdp = *ip->i_din.ffs1_din; 311 312 /* If all blocks are goig to disk, update the "size on disk" */ 313 ip->i_lfs_osize = ip->i_ffs1_size; 314 315 if (ip->i_number == LFS_IFILE_INUM) /* We know sp->idp == NULL */ 316 sp->idp = ((struct ufs1_dinode *) bp->b_data) + 317 (sp->ninodes % INOPB(fs)); 318 if (gotblk) { 319 LFS_LOCK_BUF(bp); 320 brelse(bp); 321 } 322 /* Increment inode count in segment summary block. */ 323 ++((SEGSUM *) (sp->segsum))->ss_ninos; 324 325 /* If this page is full, set flag to allocate a new page. */ 326 if (++sp->ninodes % INOPB(fs) == 0) 327 sp->ibp = NULL; 328 329 /* 330 * If updating the ifile, update the super-block. Update the disk 331 * address and access times for this inode in the ifile. 332 */ 333 ino = ip->i_number; 334 if (ino == LFS_IFILE_INUM) { 335 daddr = fs->lfs_idaddr; 336 fs->lfs_idaddr = dbtofsb(fs, bp->b_blkno); 337 } else { 338 LFS_IENTRY(ifp, fs, ino, ibp); 339 daddr = ifp->if_daddr; 340 ifp->if_daddr = dbtofsb(fs, bp->b_blkno) + fsb; 341 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 342 } 343 344 /* 345 * Account the inode: it no longer belongs to its former segment, 346 * though it will not belong to the new segment until that segment 347 * is actually written. 348 */ 349 if (daddr != LFS_UNUSED_DADDR) { 350 u_int32_t oldsn = dtosn(fs, daddr); 351 LFS_SEGENTRY(sup, fs, oldsn, bp); 352 sup->su_nbytes -= DINODE1_SIZE; 353 redo_ifile = 354 (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED)); 355 if (redo_ifile) 356 fs->lfs_flags |= LFS_IFDIRTY; 357 LFS_WRITESEGENTRY(sup, fs, oldsn, bp); /* Ifile */ 358 } 359 return redo_ifile; 360 } 361 362 int 363 lfs_gatherblock(struct segment * sp, struct ubuf * bp) 364 { 365 struct lfs *fs; 366 int version; 367 int j, blksinblk; 368 369 /* 370 * If full, finish this segment. We may be doing I/O, so 371 * release and reacquire the splbio(). 372 */ 373 fs = sp->fs; 374 blksinblk = howmany(bp->b_bcount, fs->lfs_bsize); 375 if (sp->sum_bytes_left < sizeof(ufs_daddr_t) * blksinblk || 376 sp->seg_bytes_left < bp->b_bcount) { 377 lfs_updatemeta(sp); 378 379 version = sp->fip->fi_version; 380 (void) lfs_writeseg(fs, sp); 381 382 sp->fip->fi_version = version; 383 sp->fip->fi_ino = VTOI(sp->vp)->i_number; 384 /* Add the current file to the segment summary. */ 385 ++((SEGSUM *) (sp->segsum))->ss_nfinfo; 386 sp->sum_bytes_left -= FINFOSIZE; 387 388 return 1; 389 } 390 /* Insert into the buffer list, update the FINFO block. */ 391 bp->b_flags |= B_GATHERED; 392 /* bp->b_flags &= ~B_DONE; */ 393 394 *sp->cbpp++ = bp; 395 for (j = 0; j < blksinblk; j++) 396 sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno + j; 397 398 sp->sum_bytes_left -= sizeof(ufs_daddr_t) * blksinblk; 399 sp->seg_bytes_left -= bp->b_bcount; 400 return 0; 401 } 402 403 int 404 lfs_gather(struct lfs * fs, struct segment * sp, struct uvnode * vp, int (*match) (struct lfs *, struct ubuf *)) 405 { 406 struct ubuf *bp, *nbp; 407 int count = 0; 408 409 sp->vp = vp; 410 loop: 411 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 412 nbp = LIST_NEXT(bp, b_vnbufs); 413 414 assert(bp->b_flags & B_DELWRI); 415 if ((bp->b_flags & (B_BUSY | B_GATHERED)) || !match(fs, bp)) { 416 continue; 417 } 418 if (lfs_gatherblock(sp, bp)) { 419 goto loop; 420 } 421 count++; 422 } 423 424 lfs_updatemeta(sp); 425 sp->vp = NULL; 426 return count; 427 } 428 429 430 /* 431 * Change the given block's address to ndaddr, finding its previous 432 * location using ufs_bmaparray(). 433 * 434 * Account for this change in the segment table. 435 */ 436 void 437 lfs_update_single(struct lfs * fs, struct segment * sp, daddr_t lbn, 438 ufs_daddr_t ndaddr, int size) 439 { 440 SEGUSE *sup; 441 struct ubuf *bp; 442 struct indir a[NIADDR + 2], *ap; 443 struct inode *ip; 444 struct uvnode *vp; 445 daddr_t daddr, ooff; 446 int num, error; 447 int bb, osize, obb; 448 449 vp = sp->vp; 450 ip = VTOI(vp); 451 452 error = ufs_bmaparray(fs, vp, lbn, &daddr, a, &num); 453 if (error) 454 errx(1, "lfs_updatemeta: ufs_bmaparray returned %d looking up lbn %" PRId64 "\n", error, lbn); 455 if (daddr > 0) 456 daddr = dbtofsb(fs, daddr); 457 458 bb = fragstofsb(fs, numfrags(fs, size)); 459 switch (num) { 460 case 0: 461 ooff = ip->i_ffs1_db[lbn]; 462 if (ooff == UNWRITTEN) 463 ip->i_ffs1_blocks += bb; 464 else { 465 /* possible fragment truncation or extension */ 466 obb = btofsb(fs, ip->i_lfs_fragsize[lbn]); 467 ip->i_ffs1_blocks += (bb - obb); 468 } 469 ip->i_ffs1_db[lbn] = ndaddr; 470 break; 471 case 1: 472 ooff = ip->i_ffs1_ib[a[0].in_off]; 473 if (ooff == UNWRITTEN) 474 ip->i_ffs1_blocks += bb; 475 ip->i_ffs1_ib[a[0].in_off] = ndaddr; 476 break; 477 default: 478 ap = &a[num - 1]; 479 if (bread(vp, ap->in_lbn, fs->lfs_bsize, NULL, &bp)) 480 errx(1, "lfs_updatemeta: bread bno %" PRId64, 481 ap->in_lbn); 482 483 ooff = ((ufs_daddr_t *) bp->b_data)[ap->in_off]; 484 if (ooff == UNWRITTEN) 485 ip->i_ffs1_blocks += bb; 486 ((ufs_daddr_t *) bp->b_data)[ap->in_off] = ndaddr; 487 (void) VOP_BWRITE(bp); 488 } 489 490 /* 491 * Update segment usage information, based on old size 492 * and location. 493 */ 494 if (daddr > 0) { 495 u_int32_t oldsn = dtosn(fs, daddr); 496 if (lbn >= 0 && lbn < NDADDR) 497 osize = ip->i_lfs_fragsize[lbn]; 498 else 499 osize = fs->lfs_bsize; 500 LFS_SEGENTRY(sup, fs, oldsn, bp); 501 sup->su_nbytes -= osize; 502 if (!(bp->b_flags & B_GATHERED)) 503 fs->lfs_flags |= LFS_IFDIRTY; 504 LFS_WRITESEGENTRY(sup, fs, oldsn, bp); 505 } 506 /* 507 * Now that this block has a new address, and its old 508 * segment no longer owns it, we can forget about its 509 * old size. 510 */ 511 if (lbn >= 0 && lbn < NDADDR) 512 ip->i_lfs_fragsize[lbn] = size; 513 } 514 515 /* 516 * Update the metadata that points to the blocks listed in the FINFO 517 * array. 518 */ 519 void 520 lfs_updatemeta(struct segment * sp) 521 { 522 struct ubuf *sbp; 523 struct lfs *fs; 524 struct uvnode *vp; 525 daddr_t lbn; 526 int i, nblocks, num; 527 int bb; 528 int bytesleft, size; 529 530 vp = sp->vp; 531 nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp; 532 533 if (vp == NULL || nblocks == 0) 534 return; 535 536 /* 537 * This count may be high due to oversize blocks from lfs_gop_write. 538 * Correct for this. (XXX we should be able to keep track of these.) 539 */ 540 fs = sp->fs; 541 for (i = 0; i < nblocks; i++) { 542 if (sp->start_bpp[i] == NULL) { 543 printf("nblocks = %d, not %d\n", i, nblocks); 544 nblocks = i; 545 break; 546 } 547 num = howmany(sp->start_bpp[i]->b_bcount, fs->lfs_bsize); 548 nblocks -= num - 1; 549 } 550 551 /* 552 * Sort the blocks. 553 */ 554 lfs_shellsort(sp->start_bpp, sp->start_lbp, nblocks, fs->lfs_bsize); 555 556 /* 557 * Record the length of the last block in case it's a fragment. 558 * If there are indirect blocks present, they sort last. An 559 * indirect block will be lfs_bsize and its presence indicates 560 * that you cannot have fragments. 561 */ 562 sp->fip->fi_lastlength = ((sp->start_bpp[nblocks - 1]->b_bcount - 1) & 563 fs->lfs_bmask) + 1; 564 565 /* 566 * Assign disk addresses, and update references to the logical 567 * block and the segment usage information. 568 */ 569 for (i = nblocks; i--; ++sp->start_bpp) { 570 sbp = *sp->start_bpp; 571 lbn = *sp->start_lbp; 572 573 sbp->b_blkno = fsbtodb(fs, fs->lfs_offset); 574 575 /* 576 * If we write a frag in the wrong place, the cleaner won't 577 * be able to correctly identify its size later, and the 578 * segment will be uncleanable. (Even worse, it will assume 579 * that the indirect block that actually ends the list 580 * is of a smaller size!) 581 */ 582 if ((sbp->b_bcount & fs->lfs_bmask) && i != 0) 583 errx(1, "lfs_updatemeta: fragment is not last block"); 584 585 /* 586 * For each subblock in this possibly oversized block, 587 * update its address on disk. 588 */ 589 for (bytesleft = sbp->b_bcount; bytesleft > 0; 590 bytesleft -= fs->lfs_bsize) { 591 size = MIN(bytesleft, fs->lfs_bsize); 592 bb = fragstofsb(fs, numfrags(fs, size)); 593 lbn = *sp->start_lbp++; 594 lfs_update_single(fs, sp, lbn, fs->lfs_offset, size); 595 fs->lfs_offset += bb; 596 } 597 598 } 599 } 600 601 /* 602 * Start a new segment. 603 */ 604 int 605 lfs_initseg(struct lfs * fs) 606 { 607 struct segment *sp; 608 SEGUSE *sup; 609 SEGSUM *ssp; 610 struct ubuf *bp, *sbp; 611 int repeat; 612 613 sp = fs->lfs_sp; 614 615 repeat = 0; 616 617 /* Advance to the next segment. */ 618 if (!LFS_PARTIAL_FITS(fs)) { 619 /* lfs_avail eats the remaining space */ 620 fs->lfs_avail -= fs->lfs_fsbpseg - (fs->lfs_offset - 621 fs->lfs_curseg); 622 lfs_newseg(fs); 623 repeat = 1; 624 fs->lfs_offset = fs->lfs_curseg; 625 626 sp->seg_number = dtosn(fs, fs->lfs_curseg); 627 sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg); 628 629 /* 630 * If the segment contains a superblock, update the offset 631 * and summary address to skip over it. 632 */ 633 LFS_SEGENTRY(sup, fs, sp->seg_number, bp); 634 if (sup->su_flags & SEGUSE_SUPERBLOCK) { 635 fs->lfs_offset += btofsb(fs, LFS_SBPAD); 636 sp->seg_bytes_left -= LFS_SBPAD; 637 } 638 brelse(bp); 639 /* Segment zero could also contain the labelpad */ 640 if (fs->lfs_version > 1 && sp->seg_number == 0 && 641 fs->lfs_start < btofsb(fs, LFS_LABELPAD)) { 642 fs->lfs_offset += btofsb(fs, LFS_LABELPAD) - fs->lfs_start; 643 sp->seg_bytes_left -= LFS_LABELPAD - fsbtob(fs, fs->lfs_start); 644 } 645 } else { 646 sp->seg_number = dtosn(fs, fs->lfs_curseg); 647 sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg - 648 (fs->lfs_offset - fs->lfs_curseg)); 649 } 650 fs->lfs_lastpseg = fs->lfs_offset; 651 652 sp->fs = fs; 653 sp->ibp = NULL; 654 sp->idp = NULL; 655 sp->ninodes = 0; 656 sp->ndupino = 0; 657 658 /* Get a new buffer for SEGSUM and enter it into the buffer list. */ 659 sp->cbpp = sp->bpp; 660 sbp = *sp->cbpp = getblk(fs->lfs_devvp, 661 fsbtodb(fs, fs->lfs_offset), fs->lfs_sumsize); 662 sp->segsum = sbp->b_data; 663 memset(sp->segsum, 0, fs->lfs_sumsize); 664 sp->start_bpp = ++sp->cbpp; 665 fs->lfs_offset += btofsb(fs, fs->lfs_sumsize); 666 667 /* Set point to SEGSUM, initialize it. */ 668 ssp = sp->segsum; 669 ssp->ss_next = fs->lfs_nextseg; 670 ssp->ss_nfinfo = ssp->ss_ninos = 0; 671 ssp->ss_magic = SS_MAGIC; 672 673 /* Set pointer to first FINFO, initialize it. */ 674 sp->fip = (struct finfo *) ((caddr_t) sp->segsum + SEGSUM_SIZE(fs)); 675 sp->fip->fi_nblocks = 0; 676 sp->start_lbp = &sp->fip->fi_blocks[0]; 677 sp->fip->fi_lastlength = 0; 678 679 sp->seg_bytes_left -= fs->lfs_sumsize; 680 sp->sum_bytes_left = fs->lfs_sumsize - SEGSUM_SIZE(fs); 681 682 LFS_LOCK_BUF(sbp); 683 brelse(sbp); 684 return repeat; 685 } 686 687 /* 688 * Return the next segment to write. 689 */ 690 void 691 lfs_newseg(struct lfs * fs) 692 { 693 CLEANERINFO *cip; 694 SEGUSE *sup; 695 struct ubuf *bp; 696 int curseg, isdirty, sn; 697 698 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp); 699 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 700 sup->su_nbytes = 0; 701 sup->su_nsums = 0; 702 sup->su_ninos = 0; 703 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp); 704 705 LFS_CLEANERINFO(cip, fs, bp); 706 --cip->clean; 707 ++cip->dirty; 708 fs->lfs_nclean = cip->clean; 709 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 710 711 fs->lfs_lastseg = fs->lfs_curseg; 712 fs->lfs_curseg = fs->lfs_nextseg; 713 for (sn = curseg = dtosn(fs, fs->lfs_curseg) + fs->lfs_interleave;;) { 714 sn = (sn + 1) % fs->lfs_nseg; 715 if (sn == curseg) 716 errx(1, "lfs_nextseg: no clean segments"); 717 LFS_SEGENTRY(sup, fs, sn, bp); 718 isdirty = sup->su_flags & SEGUSE_DIRTY; 719 brelse(bp); 720 721 if (!isdirty) 722 break; 723 } 724 725 ++fs->lfs_nactive; 726 fs->lfs_nextseg = sntod(fs, sn); 727 } 728 729 730 int 731 lfs_writeseg(struct lfs * fs, struct segment * sp) 732 { 733 struct ubuf **bpp, *bp; 734 SEGUSE *sup; 735 SEGSUM *ssp; 736 char *datap, *dp; 737 int i; 738 int do_again, nblocks, byteoffset; 739 size_t el_size; 740 u_short ninos; 741 struct uvnode *devvp; 742 743 /* 744 * If there are no buffers other than the segment summary to write 745 * and it is not a checkpoint, don't do anything. On a checkpoint, 746 * even if there aren't any buffers, you need to write the superblock. 747 */ 748 if ((nblocks = sp->cbpp - sp->bpp) == 1) 749 return 0; 750 751 devvp = fs->lfs_devvp; 752 753 /* Update the segment usage information. */ 754 LFS_SEGENTRY(sup, fs, sp->seg_number, bp); 755 756 /* Loop through all blocks, except the segment summary. */ 757 for (bpp = sp->bpp; ++bpp < sp->cbpp;) { 758 if ((*bpp)->b_vp != devvp) { 759 sup->su_nbytes += (*bpp)->b_bcount; 760 } 761 } 762 763 ssp = (SEGSUM *) sp->segsum; 764 765 ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs); 766 sup->su_nbytes += ssp->ss_ninos * DINODE1_SIZE; 767 768 if (fs->lfs_version == 1) 769 sup->su_olastmod = write_time; 770 else 771 sup->su_lastmod = write_time; 772 sup->su_ninos += ninos; 773 ++sup->su_nsums; 774 fs->lfs_dmeta += (btofsb(fs, fs->lfs_sumsize) + btofsb(fs, ninos * 775 fs->lfs_ibsize)); 776 fs->lfs_avail -= btofsb(fs, fs->lfs_sumsize); 777 778 do_again = !(bp->b_flags & B_GATHERED); 779 LFS_WRITESEGENTRY(sup, fs, sp->seg_number, bp); /* Ifile */ 780 781 /* 782 * Compute checksum across data and then across summary; the first 783 * block (the summary block) is skipped. Set the create time here 784 * so that it's guaranteed to be later than the inode mod times. 785 */ 786 if (fs->lfs_version == 1) 787 el_size = sizeof(u_long); 788 else 789 el_size = sizeof(u_int32_t); 790 datap = dp = malloc(nblocks * el_size); 791 for (bpp = sp->bpp, i = nblocks - 1; i--;) { 792 ++bpp; 793 /* Loop through gop_write cluster blocks */ 794 for (byteoffset = 0; byteoffset < (*bpp)->b_bcount; 795 byteoffset += fs->lfs_bsize) { 796 memcpy(dp, (*bpp)->b_data + byteoffset, el_size); 797 dp += el_size; 798 } 799 bremfree(*bpp); 800 (*bpp)->b_flags |= B_BUSY; 801 } 802 if (fs->lfs_version == 1) 803 ssp->ss_ocreate = write_time; 804 else { 805 ssp->ss_create = write_time; 806 ssp->ss_serial = ++fs->lfs_serial; 807 ssp->ss_ident = fs->lfs_ident; 808 } 809 /* Set the summary block busy too */ 810 bremfree(*(sp->bpp)); 811 (*(sp->bpp))->b_flags |= B_BUSY; 812 813 ssp->ss_datasum = cksum(datap, (nblocks - 1) * el_size); 814 ssp->ss_sumsum = 815 cksum(&ssp->ss_datasum, fs->lfs_sumsize - sizeof(ssp->ss_sumsum)); 816 free(datap); 817 datap = dp = NULL; 818 fs->lfs_bfree -= (btofsb(fs, ninos * fs->lfs_ibsize) + 819 btofsb(fs, fs->lfs_sumsize)); 820 821 if (devvp == NULL) 822 errx(1, "devvp is NULL"); 823 for (bpp = sp->bpp, i = nblocks; i; bpp++, i--) { 824 bp = *bpp; 825 #if 0 826 printf("i = %d, bp = %p, flags %lx, bn = %" PRIx64 "\n", 827 nblocks - i, bp, bp->b_flags, bp->b_blkno); 828 printf(" vp = %p\n", bp->b_vp); 829 if (bp->b_vp != fs->lfs_devvp) 830 printf(" ino = %d lbn = %" PRId64 "\n", 831 VTOI(bp->b_vp)->i_number, bp->b_lblkno); 832 #endif 833 if (bp->b_vp == fs->lfs_devvp) 834 written_dev += bp->b_bcount; 835 else { 836 if (bp->b_lblkno >= 0) 837 written_data += bp->b_bcount; 838 else 839 written_indir += bp->b_bcount; 840 } 841 bp->b_flags &= ~(B_DELWRI | B_READ | B_GATHERED | B_ERROR | 842 B_LOCKED); 843 bwrite(bp); 844 written_bytes += bp->b_bcount; 845 } 846 written_inodes += ninos; 847 848 return (lfs_initseg(fs) || do_again); 849 } 850 851 /* 852 * Our own copy of shellsort. XXX use qsort or heapsort. 853 */ 854 void 855 lfs_shellsort(struct ubuf ** bp_array, ufs_daddr_t * lb_array, int nmemb, int size) 856 { 857 static int __rsshell_increments[] = {4, 1, 0}; 858 int incr, *incrp, t1, t2; 859 struct ubuf *bp_temp; 860 861 for (incrp = __rsshell_increments; (incr = *incrp++) != 0;) 862 for (t1 = incr; t1 < nmemb; ++t1) 863 for (t2 = t1 - incr; t2 >= 0;) 864 if ((u_int32_t) bp_array[t2]->b_lblkno > 865 (u_int32_t) bp_array[t2 + incr]->b_lblkno) { 866 bp_temp = bp_array[t2]; 867 bp_array[t2] = bp_array[t2 + incr]; 868 bp_array[t2 + incr] = bp_temp; 869 t2 -= incr; 870 } else 871 break; 872 873 /* Reform the list of logical blocks */ 874 incr = 0; 875 for (t1 = 0; t1 < nmemb; t1++) { 876 for (t2 = 0; t2 * size < bp_array[t1]->b_bcount; t2++) { 877 lb_array[incr++] = bp_array[t1]->b_lblkno + t2; 878 } 879 } 880 } 881 882 883 /* 884 * lfs_seglock -- 885 * Single thread the segment writer. 886 */ 887 int 888 lfs_seglock(struct lfs * fs, unsigned long flags) 889 { 890 struct segment *sp; 891 892 if (fs->lfs_seglock) { 893 ++fs->lfs_seglock; 894 fs->lfs_sp->seg_flags |= flags; 895 return 0; 896 } 897 fs->lfs_seglock = 1; 898 899 sp = fs->lfs_sp = (struct segment *) malloc(sizeof(*sp)); 900 sp->bpp = (struct ubuf **) malloc(fs->lfs_ssize * sizeof(struct ubuf *)); 901 if (!sp->bpp) 902 errx(!preen, "Could not allocate %zu bytes: %s", 903 (size_t)(fs->lfs_ssize * sizeof(struct ubuf *)), 904 strerror(errno)); 905 sp->seg_flags = flags; 906 sp->vp = NULL; 907 sp->seg_iocount = 0; 908 (void) lfs_initseg(fs); 909 910 return 0; 911 } 912 913 /* 914 * lfs_segunlock -- 915 * Single thread the segment writer. 916 */ 917 void 918 lfs_segunlock(struct lfs * fs) 919 { 920 struct segment *sp; 921 struct ubuf *bp; 922 923 sp = fs->lfs_sp; 924 925 if (fs->lfs_seglock == 1) { 926 if (sp->bpp != sp->cbpp) { 927 /* Free allocated segment summary */ 928 fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize); 929 bp = *sp->bpp; 930 bremfree(bp); 931 bp->b_flags |= B_DONE | B_INVAL; 932 bp->b_flags &= ~B_DELWRI; 933 reassignbuf(bp, bp->b_vp); 934 bp->b_flags |= B_BUSY; /* XXX */ 935 brelse(bp); 936 } else 937 printf("unlock to 0 with no summary"); 938 939 free(sp->bpp); 940 sp->bpp = NULL; 941 free(sp); 942 fs->lfs_sp = NULL; 943 944 fs->lfs_nactive = 0; 945 946 /* Since we *know* everything's on disk, write both sbs */ 947 lfs_writesuper(fs, fs->lfs_sboffs[0]); 948 lfs_writesuper(fs, fs->lfs_sboffs[1]); 949 950 --fs->lfs_seglock; 951 fs->lfs_lockpid = 0; 952 } else if (fs->lfs_seglock == 0) { 953 errx(1, "Seglock not held"); 954 } else { 955 --fs->lfs_seglock; 956 } 957 } 958 959 int 960 lfs_writevnodes(struct lfs *fs, struct segment *sp, int op) 961 { 962 struct inode *ip; 963 struct uvnode *vp; 964 int inodes_written = 0; 965 966 LIST_FOREACH(vp, &vnodelist, v_mntvnodes) { 967 if (vp->v_bmap_op != lfs_vop_bmap) 968 continue; 969 970 ip = VTOI(vp); 971 972 if ((op == VN_DIROP && !(vp->v_flag & VDIROP)) || 973 (op != VN_DIROP && (vp->v_flag & VDIROP))) { 974 continue; 975 } 976 /* 977 * Write the inode/file if dirty and it's not the IFILE. 978 */ 979 if (ip->i_flag & IN_ALLMOD || !LIST_EMPTY(&vp->v_dirtyblkhd)) { 980 if (ip->i_number != LFS_IFILE_INUM) 981 lfs_writefile(fs, sp, vp); 982 (void) lfs_writeinode(fs, sp, ip); 983 inodes_written++; 984 } 985 } 986 return inodes_written; 987 } 988 989 void 990 lfs_writesuper(struct lfs *fs, ufs_daddr_t daddr) 991 { 992 struct ubuf *bp; 993 994 /* Set timestamp of this version of the superblock */ 995 if (fs->lfs_version == 1) 996 fs->lfs_otstamp = write_time; 997 fs->lfs_tstamp = write_time; 998 999 /* Checksum the superblock and copy it into a buffer. */ 1000 fs->lfs_cksum = lfs_sb_cksum(&(fs->lfs_dlfs)); 1001 assert(daddr > 0); 1002 bp = getblk(fs->lfs_devvp, fsbtodb(fs, daddr), LFS_SBPAD); 1003 memset(bp->b_data + sizeof(struct dlfs), 0, 1004 LFS_SBPAD - sizeof(struct dlfs)); 1005 *(struct dlfs *) bp->b_data = fs->lfs_dlfs; 1006 1007 bwrite(bp); 1008 } 1009