1/* $NetBSD: segwrite.c,v 1.19 2008/05/16 09:21:59 hannken 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 * 18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28 * POSSIBILITY OF SUCH DAMAGE. 29 */ 30/* 31 * Copyright (c) 1991, 1993 32 * The Regents of the University of California. All rights reserved. 33 * 34 * Redistribution and use in source and binary forms, with or without 35 * modification, are permitted provided that the following conditions 36 * are met: 37 * 1. Redistributions of source code must retain the above copyright 38 * notice, this list of conditions and the following disclaimer. 39 * 2. Redistributions in binary form must reproduce the above copyright 40 * notice, this list of conditions and the following disclaimer in the 41 * documentation and/or other materials provided with the distribution. 42 * 3. Neither the name of the University nor the names of its contributors 43 * may be used to endorse or promote products derived from this software 44 * without specific prior written permission. 45 * 46 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 47 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 48 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 49 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 50 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 51 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 52 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 53 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 54 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 55 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 56 * SUCH DAMAGE. 57 * 58 * @(#)lfs_segment.c 8.10 (Berkeley) 6/10/95 59 */ 60 61/* 62 * Partial segment writer, taken from the kernel and adapted for userland. 63 */ 64#include <sys/types.h> 65#include <sys/param.h> 66#include <sys/time.h> 67#include <sys/buf.h> 68#include <sys/mount.h> 69 70#include <ufs/ufs/inode.h> 71#include <ufs/ufs/ufsmount.h> 72 73/* Override certain things to make <ufs/lfs/lfs.h> work */ 74#define vnode uvnode 75#define buf ubuf 76#define panic call_panic 77 78#include <ufs/lfs/lfs.h> 79 80#include <assert.h> 81#include <stdio.h> 82#include <stdlib.h> 83#include <string.h> 84#include <err.h> 85#include <errno.h> 86#include <util.h> 87 88#include "bufcache.h" 89#include "vnode.h" 90#include "lfs_user.h" 91#include "segwrite.h" 92 93/* Compatibility definitions */ 94extern off_t locked_queue_bytes; 95int locked_queue_count; 96off_t written_bytes = 0; 97off_t written_data = 0; 98off_t written_indir = 0; 99off_t written_dev = 0; 100int written_inodes = 0; 101 102/* Global variables */ 103time_t write_time; 104 105extern u_int32_t cksum(void *, size_t); 106extern u_int32_t lfs_sb_cksum(struct dlfs *); 107extern int preen; 108 109/* 110 * Logical block number match routines used when traversing the dirty block 111 * chain. 112 */ 113int 114lfs_match_data(struct lfs * fs, struct ubuf * bp) 115{ 116 return (bp->b_lblkno >= 0); 117} 118 119int 120lfs_match_indir(struct lfs * fs, struct ubuf * bp) 121{ 122 daddr_t lbn; 123 124 lbn = bp->b_lblkno; 125 return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0); 126} 127 128int 129lfs_match_dindir(struct lfs * fs, struct ubuf * bp) 130{ 131 daddr_t lbn; 132 133 lbn = bp->b_lblkno; 134 return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1); 135} 136 137int 138lfs_match_tindir(struct lfs * fs, struct ubuf * bp) 139{ 140 daddr_t lbn; 141 142 lbn = bp->b_lblkno; 143 return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2); 144} 145 146/* 147 * Do a checkpoint. 148 */ 149int 150lfs_segwrite(struct lfs * fs, int flags) 151{ 152 struct inode *ip; 153 struct segment *sp; 154 struct uvnode *vp; 155 int redo; 156 157 lfs_seglock(fs, flags | SEGM_CKP); 158 sp = fs->lfs_sp; 159 160 lfs_writevnodes(fs, sp, VN_REG); 161 lfs_writevnodes(fs, sp, VN_DIROP); 162 ((SEGSUM *) (sp->segsum))->ss_flags &= ~(SS_CONT); 163 164 do { 165 vp = fs->lfs_ivnode; 166 fs->lfs_flags &= ~LFS_IFDIRTY; 167 ip = VTOI(vp); 168 if (LIST_FIRST(&vp->v_dirtyblkhd) != NULL || fs->lfs_idaddr <= 0) 169 lfs_writefile(fs, sp, vp); 170 171 redo = lfs_writeinode(fs, sp, ip); 172 redo += lfs_writeseg(fs, sp); 173 redo += (fs->lfs_flags & LFS_IFDIRTY); 174 } while (redo); 175 176 lfs_segunlock(fs); 177#if 0 178 printf("wrote %" PRId64 " bytes (%" PRId32 " fsb)\n", 179 written_bytes, (ufs_daddr_t)btofsb(fs, written_bytes)); 180 printf("wrote %" PRId64 " bytes data (%" PRId32 " fsb)\n", 181 written_data, (ufs_daddr_t)btofsb(fs, written_data)); 182 printf("wrote %" PRId64 " bytes indir (%" PRId32 " fsb)\n", 183 written_indir, (ufs_daddr_t)btofsb(fs, written_indir)); 184 printf("wrote %" PRId64 " bytes dev (%" PRId32 " fsb)\n", 185 written_dev, (ufs_daddr_t)btofsb(fs, written_dev)); 186 printf("wrote %d inodes (%" PRId32 " fsb)\n", 187 written_inodes, btofsb(fs, written_inodes * fs->lfs_ibsize)); 188#endif 189 return 0; 190} 191 192/* 193 * Write the dirty blocks associated with a vnode. 194 */ 195void 196lfs_writefile(struct lfs * fs, struct segment * sp, struct uvnode * vp) 197{ 198 struct ubuf *bp; 199 struct finfo *fip; 200 struct inode *ip; 201 IFILE *ifp; 202 203 ip = VTOI(vp); 204 205 if (sp->seg_bytes_left < fs->lfs_bsize || 206 sp->sum_bytes_left < sizeof(struct finfo)) 207 (void) lfs_writeseg(fs, sp); 208 209 sp->sum_bytes_left -= FINFOSIZE; 210 ++((SEGSUM *) (sp->segsum))->ss_nfinfo; 211 212 if (vp->v_uflag & VU_DIROP) 213 ((SEGSUM *) (sp->segsum))->ss_flags |= (SS_DIROP | SS_CONT); 214 215 fip = sp->fip; 216 fip->fi_nblocks = 0; 217 fip->fi_ino = ip->i_number; 218 LFS_IENTRY(ifp, fs, fip->fi_ino, bp); 219 fip->fi_version = ifp->if_version; 220 brelse(bp, 0); 221 222 lfs_gather(fs, sp, vp, lfs_match_data); 223 lfs_gather(fs, sp, vp, lfs_match_indir); 224 lfs_gather(fs, sp, vp, lfs_match_dindir); 225 lfs_gather(fs, sp, vp, lfs_match_tindir); 226 227 fip = sp->fip; 228 if (fip->fi_nblocks != 0) { 229 sp->fip = (FINFO *) ((caddr_t) fip + FINFOSIZE + 230 sizeof(ufs_daddr_t) * (fip->fi_nblocks)); 231 sp->start_lbp = &sp->fip->fi_blocks[0]; 232 } else { 233 sp->sum_bytes_left += FINFOSIZE; 234 --((SEGSUM *) (sp->segsum))->ss_nfinfo; 235 } 236} 237 238int 239lfs_writeinode(struct lfs * fs, struct segment * sp, struct inode * ip) 240{ 241 struct ubuf *bp, *ibp; 242 struct ufs1_dinode *cdp; 243 IFILE *ifp; 244 SEGUSE *sup; 245 daddr_t daddr; 246 ino_t ino; 247 int error, i, ndx, fsb = 0; 248 int redo_ifile = 0; 249 struct timespec ts; 250 int gotblk = 0; 251 252 /* Allocate a new inode block if necessary. */ 253 if ((ip->i_number != LFS_IFILE_INUM || sp->idp == NULL) && 254 sp->ibp == NULL) { 255 /* Allocate a new segment if necessary. */ 256 if (sp->seg_bytes_left < fs->lfs_ibsize || 257 sp->sum_bytes_left < sizeof(ufs_daddr_t)) 258 (void) lfs_writeseg(fs, sp); 259 260 /* Get next inode block. */ 261 daddr = fs->lfs_offset; 262 fs->lfs_offset += btofsb(fs, fs->lfs_ibsize); 263 sp->ibp = *sp->cbpp++ = 264 getblk(fs->lfs_devvp, fsbtodb(fs, daddr), 265 fs->lfs_ibsize); 266 sp->ibp->b_flags |= B_GATHERED; 267 gotblk++; 268 269 /* Zero out inode numbers */ 270 for (i = 0; i < INOPB(fs); ++i) 271 ((struct ufs1_dinode *) sp->ibp->b_data)[i].di_inumber = 0; 272 273 ++sp->start_bpp; 274 fs->lfs_avail -= btofsb(fs, fs->lfs_ibsize); 275 /* Set remaining space counters. */ 276 sp->seg_bytes_left -= fs->lfs_ibsize; 277 sp->sum_bytes_left -= sizeof(ufs_daddr_t); 278 ndx = fs->lfs_sumsize / sizeof(ufs_daddr_t) - 279 sp->ninodes / INOPB(fs) - 1; 280 ((ufs_daddr_t *) (sp->segsum))[ndx] = daddr; 281 } 282 /* Update the inode times and copy the inode onto the inode page. */ 283 ts.tv_nsec = 0; 284 ts.tv_sec = write_time; 285 /* XXX kludge --- don't redirty the ifile just to put times on it */ 286 if (ip->i_number != LFS_IFILE_INUM) 287 LFS_ITIMES(ip, &ts, &ts, &ts); 288 289 /* 290 * If this is the Ifile, and we've already written the Ifile in this 291 * partial segment, just overwrite it (it's not on disk yet) and 292 * continue. 293 * 294 * XXX we know that the bp that we get the second time around has 295 * already been gathered. 296 */ 297 if (ip->i_number == LFS_IFILE_INUM && sp->idp) { 298 *(sp->idp) = *ip->i_din.ffs1_din; 299 ip->i_lfs_osize = ip->i_ffs1_size; 300 return 0; 301 } 302 bp = sp->ibp; 303 cdp = ((struct ufs1_dinode *) bp->b_data) + (sp->ninodes % INOPB(fs)); 304 *cdp = *ip->i_din.ffs1_din; 305 306 /* If all blocks are goig to disk, update the "size on disk" */ 307 ip->i_lfs_osize = ip->i_ffs1_size; 308 309 if (ip->i_number == LFS_IFILE_INUM) /* We know sp->idp == NULL */ 310 sp->idp = ((struct ufs1_dinode *) bp->b_data) + 311 (sp->ninodes % INOPB(fs)); 312 if (gotblk) { 313 LFS_LOCK_BUF(bp); 314 assert(!(bp->b_flags & B_INVAL)); 315 brelse(bp, 0); 316 } 317 /* Increment inode count in segment summary block. */ 318 ++((SEGSUM *) (sp->segsum))->ss_ninos; 319 320 /* If this page is full, set flag to allocate a new page. */ 321 if (++sp->ninodes % INOPB(fs) == 0) 322 sp->ibp = NULL; 323 324 /* 325 * If updating the ifile, update the super-block. Update the disk 326 * address and access times for this inode in the ifile. 327 */ 328 ino = ip->i_number; 329 if (ino == LFS_IFILE_INUM) { 330 daddr = fs->lfs_idaddr; 331 fs->lfs_idaddr = dbtofsb(fs, bp->b_blkno); 332 sbdirty(); 333 } else { 334 LFS_IENTRY(ifp, fs, ino, ibp); 335 daddr = ifp->if_daddr; 336 ifp->if_daddr = dbtofsb(fs, bp->b_blkno) + fsb; 337 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 338 } 339 340 /* 341 * Account the inode: it no longer belongs to its former segment, 342 * though it will not belong to the new segment until that segment 343 * is actually written. 344 */ 345 if (daddr != LFS_UNUSED_DADDR) { 346 u_int32_t oldsn = dtosn(fs, daddr); 347 LFS_SEGENTRY(sup, fs, oldsn, bp); 348 sup->su_nbytes -= DINODE1_SIZE; 349 redo_ifile = 350 (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED)); 351 if (redo_ifile) 352 fs->lfs_flags |= LFS_IFDIRTY; 353 LFS_WRITESEGENTRY(sup, fs, oldsn, bp); /* Ifile */ 354 } 355 return redo_ifile; 356} 357 358int 359lfs_gatherblock(struct segment * sp, struct ubuf * bp) 360{ 361 struct lfs *fs; 362 int version; 363 int j, blksinblk; 364 365 /* 366 * If full, finish this segment. We may be doing I/O, so 367 * release and reacquire the splbio(). 368 */ 369 fs = sp->fs; 370 blksinblk = howmany(bp->b_bcount, fs->lfs_bsize); 371 if (sp->sum_bytes_left < sizeof(ufs_daddr_t) * blksinblk || 372 sp->seg_bytes_left < bp->b_bcount) { 373 lfs_updatemeta(sp); 374 375 version = sp->fip->fi_version; 376 (void) lfs_writeseg(fs, sp); 377 378 sp->fip->fi_version = version; 379 sp->fip->fi_ino = VTOI(sp->vp)->i_number; 380 /* Add the current file to the segment summary. */ 381 ++((SEGSUM *) (sp->segsum))->ss_nfinfo; 382 sp->sum_bytes_left -= FINFOSIZE; 383 384 return 1; 385 } 386 /* Insert into the buffer list, update the FINFO block. */ 387 bp->b_flags |= B_GATHERED; 388 /* bp->b_flags &= ~B_DONE; */ 389 390 *sp->cbpp++ = bp; 391 for (j = 0; j < blksinblk; j++) 392 sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno + j; 393 394 sp->sum_bytes_left -= sizeof(ufs_daddr_t) * blksinblk; 395 sp->seg_bytes_left -= bp->b_bcount; 396 return 0; 397} 398 399int 400lfs_gather(struct lfs * fs, struct segment * sp, struct uvnode * vp, int (*match) (struct lfs *, struct ubuf *)) 401{ 402 struct ubuf *bp, *nbp; 403 int count = 0; 404 405 sp->vp = vp; 406loop: 407 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 408 nbp = LIST_NEXT(bp, b_vnbufs); 409 410 assert(bp->b_flags & B_DELWRI); 411 if ((bp->b_flags & (B_BUSY | B_GATHERED)) || !match(fs, bp)) { 412 continue; 413 } 414 if (lfs_gatherblock(sp, bp)) { 415 goto loop; 416 } 417 count++; 418 } 419 420 lfs_updatemeta(sp); 421 sp->vp = NULL; 422 return count; 423} 424 425 426/* 427 * Change the given block's address to ndaddr, finding its previous 428 * location using ufs_bmaparray(). 429 * 430 * Account for this change in the segment table. 431 */ 432void 433lfs_update_single(struct lfs * fs, struct segment * sp, daddr_t lbn, 434 ufs_daddr_t ndaddr, int size) 435{ 436 SEGUSE *sup; 437 struct ubuf *bp; 438 struct indir a[NIADDR + 2], *ap; 439 struct inode *ip; 440 struct uvnode *vp; 441 daddr_t daddr, ooff; 442 int num, error; 443 int osize; 444 int frags, ofrags; 445 446 vp = sp->vp; 447 ip = VTOI(vp); 448 449 error = ufs_bmaparray(fs, vp, lbn, &daddr, a, &num); 450 if (error) 451 errx(1, "lfs_updatemeta: ufs_bmaparray returned %d looking up lbn %" PRId64 "\n", error, lbn); 452 if (daddr > 0) 453 daddr = dbtofsb(fs, daddr); 454 455 frags = numfrags(fs, size); 456 switch (num) { 457 case 0: 458 ooff = ip->i_ffs1_db[lbn]; 459 if (ooff == UNWRITTEN) 460 ip->i_ffs1_blocks += frags; 461 else { 462 /* possible fragment truncation or extension */ 463 ofrags = btofsb(fs, ip->i_lfs_fragsize[lbn]); 464 ip->i_ffs1_blocks += (frags - ofrags); 465 } 466 ip->i_ffs1_db[lbn] = ndaddr; 467 break; 468 case 1: 469 ooff = ip->i_ffs1_ib[a[0].in_off]; 470 if (ooff == UNWRITTEN) 471 ip->i_ffs1_blocks += frags; 472 ip->i_ffs1_ib[a[0].in_off] = ndaddr; 473 break; 474 default: 475 ap = &a[num - 1]; 476 if (bread(vp, ap->in_lbn, fs->lfs_bsize, NULL, 0, &bp)) 477 errx(1, "lfs_updatemeta: bread bno %" PRId64, 478 ap->in_lbn); 479 480 ooff = ((ufs_daddr_t *) bp->b_data)[ap->in_off]; 481 if (ooff == UNWRITTEN) 482 ip->i_ffs1_blocks += frags; 483 ((ufs_daddr_t *) bp->b_data)[ap->in_off] = ndaddr; 484 (void) VOP_BWRITE(bp); 485 } 486 487 /* 488 * Update segment usage information, based on old size 489 * and location. 490 */ 491 if (daddr > 0) { 492 u_int32_t oldsn = dtosn(fs, daddr); 493 if (lbn >= 0 && lbn < NDADDR) 494 osize = ip->i_lfs_fragsize[lbn]; 495 else 496 osize = fs->lfs_bsize; 497 LFS_SEGENTRY(sup, fs, oldsn, bp); 498 sup->su_nbytes -= osize; 499 if (!(bp->b_flags & B_GATHERED)) 500 fs->lfs_flags |= LFS_IFDIRTY; 501 LFS_WRITESEGENTRY(sup, fs, oldsn, bp); 502 } 503 /* 504 * Now that this block has a new address, and its old 505 * segment no longer owns it, we can forget about its 506 * old size. 507 */ 508 if (lbn >= 0 && lbn < NDADDR) 509 ip->i_lfs_fragsize[lbn] = size; 510} 511 512/* 513 * Update the metadata that points to the blocks listed in the FINFO 514 * array. 515 */ 516void 517lfs_updatemeta(struct segment * sp) 518{ 519 struct ubuf *sbp; 520 struct lfs *fs; 521 struct uvnode *vp; 522 daddr_t lbn; 523 int i, nblocks, num; 524 int frags; 525 int bytesleft, size; 526 527 vp = sp->vp; 528 nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp; 529 530 if (vp == NULL || nblocks == 0) 531 return; 532 533 /* 534 * This count may be high due to oversize blocks from lfs_gop_write. 535 * Correct for this. (XXX we should be able to keep track of these.) 536 */ 537 fs = sp->fs; 538 for (i = 0; i < nblocks; i++) { 539 if (sp->start_bpp[i] == NULL) { 540 printf("nblocks = %d, not %d\n", i, nblocks); 541 nblocks = i; 542 break; 543 } 544 num = howmany(sp->start_bpp[i]->b_bcount, fs->lfs_bsize); 545 nblocks -= num - 1; 546 } 547 548 /* 549 * Sort the blocks. 550 */ 551 lfs_shellsort(sp->start_bpp, sp->start_lbp, nblocks, fs->lfs_bsize); 552 553 /* 554 * Record the length of the last block in case it's a fragment. 555 * If there are indirect blocks present, they sort last. An 556 * indirect block will be lfs_bsize and its presence indicates 557 * that you cannot have fragments. 558 */ 559 sp->fip->fi_lastlength = ((sp->start_bpp[nblocks - 1]->b_bcount - 1) & 560 fs->lfs_bmask) + 1; 561 562 /* 563 * Assign disk addresses, and update references to the logical 564 * block and the segment usage information. 565 */ 566 for (i = nblocks; i--; ++sp->start_bpp) { 567 sbp = *sp->start_bpp; 568 lbn = *sp->start_lbp; 569 570 sbp->b_blkno = fsbtodb(fs, fs->lfs_offset); 571 572 /* 573 * If we write a frag in the wrong place, the cleaner won't 574 * be able to correctly identify its size later, and the 575 * segment will be uncleanable. (Even worse, it will assume 576 * that the indirect block that actually ends the list 577 * is of a smaller size!) 578 */ 579 if ((sbp->b_bcount & fs->lfs_bmask) && i != 0) 580 errx(1, "lfs_updatemeta: fragment is not last block"); 581 582 /* 583 * For each subblock in this possibly oversized block, 584 * update its address on disk. 585 */ 586 for (bytesleft = sbp->b_bcount; bytesleft > 0; 587 bytesleft -= fs->lfs_bsize) { 588 size = MIN(bytesleft, fs->lfs_bsize); 589 frags = numfrags(fs, size); 590 lbn = *sp->start_lbp++; 591 lfs_update_single(fs, sp, lbn, fs->lfs_offset, size); 592 fs->lfs_offset += frags; 593 } 594 595 } 596} 597 598/* 599 * Start a new segment. 600 */ 601int 602lfs_initseg(struct lfs * fs) 603{ 604 struct segment *sp; 605 SEGUSE *sup; 606 SEGSUM *ssp; 607 struct ubuf *bp, *sbp; 608 int repeat; 609 610 sp = fs->lfs_sp; 611 612 repeat = 0; 613 614 /* Advance to the next segment. */ 615 if (!LFS_PARTIAL_FITS(fs)) { 616 /* lfs_avail eats the remaining space */ 617 fs->lfs_avail -= fs->lfs_fsbpseg - (fs->lfs_offset - 618 fs->lfs_curseg); 619 lfs_newseg(fs); 620 repeat = 1; 621 fs->lfs_offset = fs->lfs_curseg; 622 623 sp->seg_number = dtosn(fs, fs->lfs_curseg); 624 sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg); 625 626 /* 627 * If the segment contains a superblock, update the offset 628 * and summary address to skip over it. 629 */ 630 LFS_SEGENTRY(sup, fs, sp->seg_number, bp); 631 if (sup->su_flags & SEGUSE_SUPERBLOCK) { 632 fs->lfs_offset += btofsb(fs, LFS_SBPAD); 633 sp->seg_bytes_left -= LFS_SBPAD; 634 } 635 brelse(bp, 0); 636 /* Segment zero could also contain the labelpad */ 637 if (fs->lfs_version > 1 && sp->seg_number == 0 && 638 fs->lfs_start < btofsb(fs, LFS_LABELPAD)) { 639 fs->lfs_offset += btofsb(fs, LFS_LABELPAD) - fs->lfs_start; 640 sp->seg_bytes_left -= LFS_LABELPAD - fsbtob(fs, fs->lfs_start); 641 } 642 } else { 643 sp->seg_number = dtosn(fs, fs->lfs_curseg); 644 sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg - 645 (fs->lfs_offset - fs->lfs_curseg)); 646 } 647 fs->lfs_lastpseg = fs->lfs_offset; 648 649 sp->fs = fs; 650 sp->ibp = NULL; 651 sp->idp = NULL; 652 sp->ninodes = 0; 653 sp->ndupino = 0; 654 655 /* Get a new buffer for SEGSUM and enter it into the buffer list. */ 656 sp->cbpp = sp->bpp; 657 sbp = *sp->cbpp = getblk(fs->lfs_devvp, 658 fsbtodb(fs, fs->lfs_offset), fs->lfs_sumsize); 659 sp->segsum = sbp->b_data; 660 memset(sp->segsum, 0, fs->lfs_sumsize); 661 sp->start_bpp = ++sp->cbpp; 662 fs->lfs_offset += btofsb(fs, fs->lfs_sumsize); 663 664 /* Set point to SEGSUM, initialize it. */ 665 ssp = sp->segsum; 666 ssp->ss_next = fs->lfs_nextseg; 667 ssp->ss_nfinfo = ssp->ss_ninos = 0; 668 ssp->ss_magic = SS_MAGIC; 669 670 /* Set pointer to first FINFO, initialize it. */ 671 sp->fip = (struct finfo *) ((caddr_t) sp->segsum + SEGSUM_SIZE(fs)); 672 sp->fip->fi_nblocks = 0; 673 sp->start_lbp = &sp->fip->fi_blocks[0]; 674 sp->fip->fi_lastlength = 0; 675 676 sp->seg_bytes_left -= fs->lfs_sumsize; 677 sp->sum_bytes_left = fs->lfs_sumsize - SEGSUM_SIZE(fs); 678 679 LFS_LOCK_BUF(sbp); 680 brelse(sbp, 0); 681 return repeat; 682} 683 684/* 685 * Return the next segment to write. 686 */ 687void 688lfs_newseg(struct lfs * fs) 689{ 690 CLEANERINFO *cip; 691 SEGUSE *sup; 692 struct ubuf *bp; 693 int curseg, isdirty, sn; 694 695 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp); 696 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 697 sup->su_nbytes = 0; 698 sup->su_nsums = 0; 699 sup->su_ninos = 0; 700 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp); 701 702 LFS_CLEANERINFO(cip, fs, bp); 703 --cip->clean; 704 ++cip->dirty; 705 fs->lfs_nclean = cip->clean; 706 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 707 708 fs->lfs_lastseg = fs->lfs_curseg; 709 fs->lfs_curseg = fs->lfs_nextseg; 710 for (sn = curseg = dtosn(fs, fs->lfs_curseg) + fs->lfs_interleave;;) { 711 sn = (sn + 1) % fs->lfs_nseg; 712 if (sn == curseg) 713 errx(1, "lfs_nextseg: no clean segments"); 714 LFS_SEGENTRY(sup, fs, sn, bp); 715 isdirty = sup->su_flags & SEGUSE_DIRTY; 716 brelse(bp, 0); 717 718 if (!isdirty) 719 break; 720 } 721 722 ++fs->lfs_nactive; 723 fs->lfs_nextseg = sntod(fs, sn); 724} 725 726 727int 728lfs_writeseg(struct lfs * fs, struct segment * sp) 729{ 730 struct ubuf **bpp, *bp; 731 SEGUSE *sup; 732 SEGSUM *ssp; 733 char *datap, *dp; 734 int i; 735 int do_again, nblocks, byteoffset; 736 size_t el_size; 737 u_short ninos; 738 struct uvnode *devvp; 739 740 /* 741 * If there are no buffers other than the segment summary to write 742 * and it is not a checkpoint, don't do anything. On a checkpoint, 743 * even if there aren't any buffers, you need to write the superblock. 744 */ 745 nblocks = sp->cbpp - sp->bpp; 746#if 0 747 printf("write %d blocks at 0x%x\n", 748 nblocks, (int)dbtofsb(fs, (*sp->bpp)->b_blkno)); 749#endif 750 if (nblocks == 1) 751 return 0; 752 753 devvp = fs->lfs_devvp; 754 755 /* Update the segment usage information. */ 756 LFS_SEGENTRY(sup, fs, sp->seg_number, bp); 757 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 758 759 /* Loop through all blocks, except the segment summary. */ 760 for (bpp = sp->bpp; ++bpp < sp->cbpp;) { 761 if ((*bpp)->b_vp != devvp) { 762 sup->su_nbytes += (*bpp)->b_bcount; 763 } 764 assert(dtosn(fs, dbtofsb(fs, (*bpp)->b_blkno)) == sp->seg_number); 765 } 766 767 ssp = (SEGSUM *) sp->segsum; 768 ssp->ss_flags |= SS_RFW; 769 770 ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs); 771 sup->su_nbytes += ssp->ss_ninos * DINODE1_SIZE; 772 773 if (fs->lfs_version == 1) 774 sup->su_olastmod = write_time; 775 else 776 sup->su_lastmod = write_time; 777 sup->su_ninos += ninos; 778 ++sup->su_nsums; 779 fs->lfs_dmeta += (btofsb(fs, fs->lfs_sumsize) + btofsb(fs, ninos * 780 fs->lfs_ibsize)); 781 fs->lfs_avail -= btofsb(fs, fs->lfs_sumsize); 782 783 do_again = !(bp->b_flags & B_GATHERED); 784 LFS_WRITESEGENTRY(sup, fs, sp->seg_number, bp); /* Ifile */ 785 786 /* 787 * Compute checksum across data and then across summary; the first 788 * block (the summary block) is skipped. Set the create time here 789 * so that it's guaranteed to be later than the inode mod times. 790 */ 791 if (fs->lfs_version == 1) 792 el_size = sizeof(u_long); 793 else 794 el_size = sizeof(u_int32_t); 795 datap = dp = emalloc(nblocks * el_size); 796 for (bpp = sp->bpp, i = nblocks - 1; i--;) { 797 ++bpp; 798 /* Loop through gop_write cluster blocks */ 799 for (byteoffset = 0; byteoffset < (*bpp)->b_bcount; 800 byteoffset += fs->lfs_bsize) { 801 memcpy(dp, (*bpp)->b_data + byteoffset, el_size); 802 dp += el_size; 803 } 804 bremfree(*bpp); 805 (*bpp)->b_flags |= B_BUSY; 806 } 807 if (fs->lfs_version == 1) 808 ssp->ss_ocreate = write_time; 809 else { 810 ssp->ss_create = write_time; 811 ssp->ss_serial = ++fs->lfs_serial; 812 ssp->ss_ident = fs->lfs_ident; 813 } 814 /* Set the summary block busy too */ 815 bremfree(*(sp->bpp)); 816 (*(sp->bpp))->b_flags |= B_BUSY; 817 818 ssp->ss_datasum = cksum(datap, (nblocks - 1) * el_size); 819 ssp->ss_sumsum = 820 cksum(&ssp->ss_datasum, fs->lfs_sumsize - sizeof(ssp->ss_sumsum)); 821 free(datap); 822 datap = dp = NULL; 823 fs->lfs_bfree -= (btofsb(fs, ninos * fs->lfs_ibsize) + 824 btofsb(fs, fs->lfs_sumsize)); 825 826 if (devvp == NULL) 827 errx(1, "devvp is NULL"); 828 for (bpp = sp->bpp, i = nblocks; i; bpp++, i--) { 829 bp = *bpp; 830#if 0 831 printf("i = %d, bp = %p, flags %lx, bn = %" PRIx64 "\n", 832 nblocks - i, bp, bp->b_flags, bp->b_blkno); 833 printf(" vp = %p\n", bp->b_vp); 834 if (bp->b_vp != fs->lfs_devvp) 835 printf(" ino = %d lbn = %" PRId64 "\n", 836 VTOI(bp->b_vp)->i_number, bp->b_lblkno); 837#endif 838 if (bp->b_vp == fs->lfs_devvp) 839 written_dev += bp->b_bcount; 840 else { 841 if (bp->b_lblkno >= 0) 842 written_data += bp->b_bcount; 843 else 844 written_indir += bp->b_bcount; 845 } 846 bp->b_flags &= ~(B_DELWRI | B_READ | B_GATHERED | B_ERROR | 847 B_LOCKED); 848 bwrite(bp); 849 written_bytes += bp->b_bcount; 850 } 851 written_inodes += ninos; 852 853 return (lfs_initseg(fs) || do_again); 854} 855 856/* 857 * Our own copy of shellsort. XXX use qsort or heapsort. 858 */ 859void 860lfs_shellsort(struct ubuf ** bp_array, ufs_daddr_t * lb_array, int nmemb, int size) 861{ 862 static int __rsshell_increments[] = {4, 1, 0}; 863 int incr, *incrp, t1, t2; 864 struct ubuf *bp_temp; 865 866 for (incrp = __rsshell_increments; (incr = *incrp++) != 0;) 867 for (t1 = incr; t1 < nmemb; ++t1) 868 for (t2 = t1 - incr; t2 >= 0;) 869 if ((u_int32_t) bp_array[t2]->b_lblkno > 870 (u_int32_t) bp_array[t2 + incr]->b_lblkno) { 871 bp_temp = bp_array[t2]; 872 bp_array[t2] = bp_array[t2 + incr]; 873 bp_array[t2 + incr] = bp_temp; 874 t2 -= incr; 875 } else 876 break; 877 878 /* Reform the list of logical blocks */ 879 incr = 0; 880 for (t1 = 0; t1 < nmemb; t1++) { 881 for (t2 = 0; t2 * size < bp_array[t1]->b_bcount; t2++) { 882 lb_array[incr++] = bp_array[t1]->b_lblkno + t2; 883 } 884 } 885} 886 887 888/* 889 * lfs_seglock -- 890 * Single thread the segment writer. 891 */ 892int 893lfs_seglock(struct lfs * fs, unsigned long flags) 894{ 895 struct segment *sp; 896 897 if (fs->lfs_seglock) { 898 ++fs->lfs_seglock; 899 fs->lfs_sp->seg_flags |= flags; 900 return 0; 901 } 902 fs->lfs_seglock = 1; 903 904 sp = fs->lfs_sp = emalloc(sizeof(*sp)); 905 sp->bpp = emalloc(fs->lfs_ssize * sizeof(struct ubuf *)); 906 if (!sp->bpp) 907 errx(!preen, "Could not allocate %zu bytes: %s", 908 (size_t)(fs->lfs_ssize * sizeof(struct ubuf *)), 909 strerror(errno)); 910 sp->seg_flags = flags; 911 sp->vp = NULL; 912 sp->seg_iocount = 0; 913 (void) lfs_initseg(fs); 914 915 return 0; 916} 917 918/* 919 * lfs_segunlock -- 920 * Single thread the segment writer. 921 */ 922void 923lfs_segunlock(struct lfs * fs) 924{ 925 struct segment *sp; 926 struct ubuf *bp; 927 928 sp = fs->lfs_sp; 929 930 if (fs->lfs_seglock == 1) { 931 if (sp->bpp != sp->cbpp) { 932 /* Free allocated segment summary */ 933 fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize); 934 bp = *sp->bpp; 935 bremfree(bp); 936 bp->b_flags |= B_DONE | B_INVAL; 937 bp->b_flags &= ~B_DELWRI; 938 reassignbuf(bp, bp->b_vp); 939 bp->b_flags |= B_BUSY; /* XXX */ 940 brelse(bp, 0); 941 } else 942 printf("unlock to 0 with no summary"); 943 944 free(sp->bpp); 945 sp->bpp = NULL; 946 free(sp); 947 fs->lfs_sp = NULL; 948 949 fs->lfs_nactive = 0; 950 951 /* Since we *know* everything's on disk, write both sbs */ 952 lfs_writesuper(fs, fs->lfs_sboffs[0]); 953 lfs_writesuper(fs, fs->lfs_sboffs[1]); 954 955 --fs->lfs_seglock; 956 fs->lfs_lockpid = 0; 957 } else if (fs->lfs_seglock == 0) { 958 errx(1, "Seglock not held"); 959 } else { 960 --fs->lfs_seglock; 961 } 962} 963 964int 965lfs_writevnodes(struct lfs *fs, struct segment *sp, int op) 966{ 967 struct inode *ip; 968 struct uvnode *vp; 969 int inodes_written = 0; 970 971 LIST_FOREACH(vp, &vnodelist, v_mntvnodes) { 972 if (vp->v_bmap_op != lfs_vop_bmap) 973 continue; 974 975 ip = VTOI(vp); 976 977 if ((op == VN_DIROP && !(vp->v_uflag & VU_DIROP)) || 978 (op != VN_DIROP && (vp->v_uflag & VU_DIROP))) { 979 continue; 980 } 981 /* 982 * Write the inode/file if dirty and it's not the IFILE. 983 */ 984 if (ip->i_flag & IN_ALLMOD || !LIST_EMPTY(&vp->v_dirtyblkhd)) { 985 if (ip->i_number != LFS_IFILE_INUM) 986 lfs_writefile(fs, sp, vp); 987 (void) lfs_writeinode(fs, sp, ip); 988 inodes_written++; 989 } 990 } 991 return inodes_written; 992} 993 994void 995lfs_writesuper(struct lfs *fs, ufs_daddr_t daddr) 996{ 997 struct ubuf *bp; 998 999 /* Set timestamp of this version of the superblock */ 1000 if (fs->lfs_version == 1) 1001 fs->lfs_otstamp = write_time; 1002 fs->lfs_tstamp = write_time; 1003 1004 /* Checksum the superblock and copy it into a buffer. */ 1005 fs->lfs_cksum = lfs_sb_cksum(&(fs->lfs_dlfs)); 1006 assert(daddr > 0); 1007 bp = getblk(fs->lfs_devvp, fsbtodb(fs, daddr), LFS_SBPAD); 1008 memset(bp->b_data + sizeof(struct dlfs), 0, 1009 LFS_SBPAD - sizeof(struct dlfs)); 1010 *(struct dlfs *) bp->b_data = fs->lfs_dlfs; 1011 1012 bwrite(bp); 1013} 1014