suj.c revision 209716
1207141Sjeff/*- 2207141Sjeff * Copyright 2009, 2010 Jeffrey W. Roberson <jeff@FreeBSD.org> 3207141Sjeff * All rights reserved. 4207141Sjeff * 5207141Sjeff * Redistribution and use in source and binary forms, with or without 6207141Sjeff * modification, are permitted provided that the following conditions 7207141Sjeff * are met: 8207141Sjeff * 1. Redistributions of source code must retain the above copyright 9207141Sjeff * notice, this list of conditions and the following disclaimer. 10207141Sjeff * 2. Redistributions in binary form must reproduce the above copyright 11207141Sjeff * notice, this list of conditions and the following disclaimer in the 12207141Sjeff * documentation and/or other materials provided with the distribution. 13207141Sjeff * 14207141Sjeff * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15207141Sjeff * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16207141Sjeff * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17207141Sjeff * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18207141Sjeff * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19207141Sjeff * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20207141Sjeff * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21207141Sjeff * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22207141Sjeff * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23207141Sjeff * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24207141Sjeff * SUCH DAMAGE. 25207141Sjeff */ 26207141Sjeff 27207141Sjeff#include <sys/cdefs.h> 28207141Sjeff__FBSDID("$FreeBSD: head/sbin/fsck_ffs/suj.c 209716 2010-07-06 07:07:29Z jeff $"); 29207141Sjeff 30207141Sjeff#include <sys/param.h> 31207141Sjeff#include <sys/disklabel.h> 32207141Sjeff#include <sys/mount.h> 33207141Sjeff#include <sys/stat.h> 34207141Sjeff 35207141Sjeff#include <ufs/ufs/ufsmount.h> 36207141Sjeff#include <ufs/ufs/dinode.h> 37207141Sjeff#include <ufs/ufs/dir.h> 38207141Sjeff#include <ufs/ffs/fs.h> 39207141Sjeff 40209408Sdelphij#include <setjmp.h> 41209408Sdelphij#include <stdarg.h> 42207141Sjeff#include <stdio.h> 43207141Sjeff#include <stdlib.h> 44207141Sjeff#include <stdint.h> 45207141Sjeff#include <libufs.h> 46207141Sjeff#include <string.h> 47207141Sjeff#include <strings.h> 48209408Sdelphij#include <sysexits.h> 49207141Sjeff#include <err.h> 50207141Sjeff#include <assert.h> 51207141Sjeff 52207141Sjeff#include "fsck.h" 53207141Sjeff 54207141Sjeff#define DOTDOT_OFFSET DIRECTSIZ(1) 55207141Sjeff#define SUJ_HASHSIZE 2048 56207141Sjeff#define SUJ_HASHMASK (SUJ_HASHSIZE - 1) 57207141Sjeff#define SUJ_HASH(x) ((x * 2654435761) & SUJ_HASHMASK) 58207141Sjeff 59207141Sjeffstruct suj_seg { 60207141Sjeff TAILQ_ENTRY(suj_seg) ss_next; 61207141Sjeff struct jsegrec ss_rec; 62207141Sjeff uint8_t *ss_blk; 63207141Sjeff}; 64207141Sjeff 65207141Sjeffstruct suj_rec { 66207141Sjeff TAILQ_ENTRY(suj_rec) sr_next; 67207141Sjeff union jrec *sr_rec; 68207141Sjeff}; 69207141SjeffTAILQ_HEAD(srechd, suj_rec); 70207141Sjeff 71207141Sjeffstruct suj_ino { 72207141Sjeff LIST_ENTRY(suj_ino) si_next; 73207141Sjeff struct srechd si_recs; 74207141Sjeff struct srechd si_newrecs; 75207141Sjeff struct srechd si_movs; 76207141Sjeff struct jtrncrec *si_trunc; 77207141Sjeff ino_t si_ino; 78207141Sjeff char si_skipparent; 79207141Sjeff char si_hasrecs; 80207141Sjeff char si_blkadj; 81207141Sjeff char si_linkadj; 82207141Sjeff int si_mode; 83207141Sjeff nlink_t si_nlinkadj; 84207141Sjeff nlink_t si_nlink; 85207141Sjeff nlink_t si_dotlinks; 86207141Sjeff}; 87207141SjeffLIST_HEAD(inohd, suj_ino); 88207141Sjeff 89207141Sjeffstruct suj_blk { 90207141Sjeff LIST_ENTRY(suj_blk) sb_next; 91207141Sjeff struct srechd sb_recs; 92207141Sjeff ufs2_daddr_t sb_blk; 93207141Sjeff}; 94207141SjeffLIST_HEAD(blkhd, suj_blk); 95207141Sjeff 96207141Sjeffstruct data_blk { 97207141Sjeff LIST_ENTRY(data_blk) db_next; 98207141Sjeff uint8_t *db_buf; 99207141Sjeff ufs2_daddr_t db_blk; 100207141Sjeff int db_size; 101207141Sjeff int db_dirty; 102207141Sjeff}; 103207141Sjeff 104207141Sjeffstruct ino_blk { 105207141Sjeff LIST_ENTRY(ino_blk) ib_next; 106207141Sjeff uint8_t *ib_buf; 107207141Sjeff int ib_dirty; 108207141Sjeff ufs2_daddr_t ib_blk; 109207141Sjeff}; 110207141SjeffLIST_HEAD(iblkhd, ino_blk); 111207141Sjeff 112207141Sjeffstruct suj_cg { 113207141Sjeff LIST_ENTRY(suj_cg) sc_next; 114207141Sjeff struct blkhd sc_blkhash[SUJ_HASHSIZE]; 115207141Sjeff struct inohd sc_inohash[SUJ_HASHSIZE]; 116207141Sjeff struct iblkhd sc_iblkhash[SUJ_HASHSIZE]; 117207141Sjeff struct ino_blk *sc_lastiblk; 118207141Sjeff struct suj_ino *sc_lastino; 119207141Sjeff struct suj_blk *sc_lastblk; 120207141Sjeff uint8_t *sc_cgbuf; 121207141Sjeff struct cg *sc_cgp; 122207141Sjeff int sc_dirty; 123207141Sjeff int sc_cgx; 124207141Sjeff}; 125207141Sjeff 126207141SjeffLIST_HEAD(cghd, suj_cg) cghash[SUJ_HASHSIZE]; 127207141SjeffLIST_HEAD(dblkhd, data_blk) dbhash[SUJ_HASHSIZE]; 128207141Sjeffstruct suj_cg *lastcg; 129207141Sjeffstruct data_blk *lastblk; 130207141Sjeff 131207141SjeffTAILQ_HEAD(seghd, suj_seg) allsegs; 132207141Sjeffuint64_t oldseq; 133207141Sjeffstatic struct uufsd *disk = NULL; 134207141Sjeffstatic struct fs *fs = NULL; 135207141Sjeffino_t sujino; 136207141Sjeff 137207141Sjeff/* 138207141Sjeff * Summary statistics. 139207141Sjeff */ 140207141Sjeffuint64_t freefrags; 141207141Sjeffuint64_t freeblocks; 142207141Sjeffuint64_t freeinos; 143207141Sjeffuint64_t freedir; 144207141Sjeffuint64_t jbytes; 145207141Sjeffuint64_t jrecs; 146207141Sjeff 147209408Sdelphijstatic jmp_buf jmpbuf; 148209408Sdelphij 149207141Sjefftypedef void (*ino_visitor)(ino_t, ufs_lbn_t, ufs2_daddr_t, int); 150209408Sdelphijstatic void err_suj(const char *, ...) __dead2; 151207141Sjeffstatic void ino_trunc(ino_t, off_t); 152207141Sjeffstatic void ino_decr(ino_t); 153207141Sjeffstatic void ino_adjust(struct suj_ino *); 154207141Sjeffstatic void ino_build(struct suj_ino *); 155207141Sjeffstatic int blk_isfree(ufs2_daddr_t); 156207141Sjeff 157207141Sjeffstatic void * 158207141Sjefferrmalloc(size_t n) 159207141Sjeff{ 160207141Sjeff void *a; 161207141Sjeff 162207141Sjeff a = malloc(n); 163207141Sjeff if (a == NULL) 164209408Sdelphij err(EX_OSERR, "malloc(%zu)", n); 165207141Sjeff return (a); 166207141Sjeff} 167207141Sjeff 168207141Sjeff/* 169209408Sdelphij * When hit a fatal error in journalling check, print out 170209408Sdelphij * the error and then offer to fallback to normal fsck. 171209408Sdelphij */ 172209408Sdelphijstatic void 173209408Sdelphijerr_suj(const char * restrict fmt, ...) 174209408Sdelphij{ 175209408Sdelphij va_list ap; 176209408Sdelphij 177209408Sdelphij if (preen) 178209408Sdelphij (void)fprintf(stdout, "%s: ", cdevname); 179209408Sdelphij 180209408Sdelphij va_start(ap, fmt); 181209408Sdelphij (void)vfprintf(stdout, fmt, ap); 182209408Sdelphij va_end(ap); 183209408Sdelphij 184209408Sdelphij longjmp(jmpbuf, -1); 185209408Sdelphij} 186209408Sdelphij 187209408Sdelphij/* 188207141Sjeff * Open the given provider, load superblock. 189207141Sjeff */ 190207141Sjeffstatic void 191207141Sjeffopendisk(const char *devnam) 192207141Sjeff{ 193207141Sjeff if (disk != NULL) 194207141Sjeff return; 195207141Sjeff disk = malloc(sizeof(*disk)); 196207141Sjeff if (disk == NULL) 197209408Sdelphij err(EX_OSERR, "malloc(%zu)", sizeof(*disk)); 198207141Sjeff if (ufs_disk_fillout(disk, devnam) == -1) { 199209408Sdelphij err(EX_OSERR, "ufs_disk_fillout(%s) failed: %s", devnam, 200207141Sjeff disk->d_error); 201207141Sjeff } 202207141Sjeff fs = &disk->d_fs; 203207141Sjeff} 204207141Sjeff 205207141Sjeff/* 206207141Sjeff * Mark file system as clean, write the super-block back, close the disk. 207207141Sjeff */ 208207141Sjeffstatic void 209207141Sjeffclosedisk(const char *devnam) 210207141Sjeff{ 211207141Sjeff struct csum *cgsum; 212207141Sjeff int i; 213207141Sjeff 214207141Sjeff /* 215207141Sjeff * Recompute the fs summary info from correct cs summaries. 216207141Sjeff */ 217207141Sjeff bzero(&fs->fs_cstotal, sizeof(struct csum_total)); 218207141Sjeff for (i = 0; i < fs->fs_ncg; i++) { 219207141Sjeff cgsum = &fs->fs_cs(fs, i); 220207141Sjeff fs->fs_cstotal.cs_nffree += cgsum->cs_nffree; 221207141Sjeff fs->fs_cstotal.cs_nbfree += cgsum->cs_nbfree; 222207141Sjeff fs->fs_cstotal.cs_nifree += cgsum->cs_nifree; 223207141Sjeff fs->fs_cstotal.cs_ndir += cgsum->cs_ndir; 224207141Sjeff } 225207141Sjeff fs->fs_pendinginodes = 0; 226207141Sjeff fs->fs_pendingblocks = 0; 227207141Sjeff fs->fs_clean = 1; 228207141Sjeff fs->fs_time = time(NULL); 229207141Sjeff fs->fs_mtime = time(NULL); 230207141Sjeff if (sbwrite(disk, 0) == -1) 231209408Sdelphij err(EX_OSERR, "sbwrite(%s)", devnam); 232207141Sjeff if (ufs_disk_close(disk) == -1) 233209408Sdelphij err(EX_OSERR, "ufs_disk_close(%s)", devnam); 234207141Sjeff free(disk); 235207141Sjeff disk = NULL; 236207141Sjeff fs = NULL; 237207141Sjeff} 238207141Sjeff 239207141Sjeff/* 240207141Sjeff * Lookup a cg by number in the hash so we can keep track of which cgs 241207141Sjeff * need stats rebuilt. 242207141Sjeff */ 243207141Sjeffstatic struct suj_cg * 244207141Sjeffcg_lookup(int cgx) 245207141Sjeff{ 246207141Sjeff struct cghd *hd; 247207141Sjeff struct suj_cg *sc; 248207141Sjeff 249209408Sdelphij if (cgx < 0 || cgx >= fs->fs_ncg) 250209408Sdelphij err_suj("Bad cg number %d\n", cgx); 251207141Sjeff if (lastcg && lastcg->sc_cgx == cgx) 252207141Sjeff return (lastcg); 253207141Sjeff hd = &cghash[SUJ_HASH(cgx)]; 254207141Sjeff LIST_FOREACH(sc, hd, sc_next) 255207141Sjeff if (sc->sc_cgx == cgx) { 256207141Sjeff lastcg = sc; 257207141Sjeff return (sc); 258207141Sjeff } 259207141Sjeff sc = errmalloc(sizeof(*sc)); 260207141Sjeff bzero(sc, sizeof(*sc)); 261207141Sjeff sc->sc_cgbuf = errmalloc(fs->fs_bsize); 262207141Sjeff sc->sc_cgp = (struct cg *)sc->sc_cgbuf; 263207141Sjeff sc->sc_cgx = cgx; 264207141Sjeff LIST_INSERT_HEAD(hd, sc, sc_next); 265207141Sjeff if (bread(disk, fsbtodb(fs, cgtod(fs, sc->sc_cgx)), sc->sc_cgbuf, 266207141Sjeff fs->fs_bsize) == -1) 267209408Sdelphij err_suj("Unable to read cylinder group %d\n", sc->sc_cgx); 268207141Sjeff 269207141Sjeff return (sc); 270207141Sjeff} 271207141Sjeff 272207141Sjeff/* 273207141Sjeff * Lookup an inode number in the hash and allocate a suj_ino if it does 274207141Sjeff * not exist. 275207141Sjeff */ 276207141Sjeffstatic struct suj_ino * 277207141Sjeffino_lookup(ino_t ino, int creat) 278207141Sjeff{ 279207141Sjeff struct suj_ino *sino; 280207141Sjeff struct inohd *hd; 281207141Sjeff struct suj_cg *sc; 282207141Sjeff 283207141Sjeff sc = cg_lookup(ino_to_cg(fs, ino)); 284207141Sjeff if (sc->sc_lastino && sc->sc_lastino->si_ino == ino) 285207141Sjeff return (sc->sc_lastino); 286207141Sjeff hd = &sc->sc_inohash[SUJ_HASH(ino)]; 287207141Sjeff LIST_FOREACH(sino, hd, si_next) 288207141Sjeff if (sino->si_ino == ino) 289207141Sjeff return (sino); 290207141Sjeff if (creat == 0) 291207141Sjeff return (NULL); 292207141Sjeff sino = errmalloc(sizeof(*sino)); 293207141Sjeff bzero(sino, sizeof(*sino)); 294207141Sjeff sino->si_ino = ino; 295207141Sjeff TAILQ_INIT(&sino->si_recs); 296207141Sjeff TAILQ_INIT(&sino->si_newrecs); 297207141Sjeff TAILQ_INIT(&sino->si_movs); 298207141Sjeff LIST_INSERT_HEAD(hd, sino, si_next); 299207141Sjeff 300207141Sjeff return (sino); 301207141Sjeff} 302207141Sjeff 303207141Sjeff/* 304207141Sjeff * Lookup a block number in the hash and allocate a suj_blk if it does 305207141Sjeff * not exist. 306207141Sjeff */ 307207141Sjeffstatic struct suj_blk * 308207141Sjeffblk_lookup(ufs2_daddr_t blk, int creat) 309207141Sjeff{ 310207141Sjeff struct suj_blk *sblk; 311207141Sjeff struct suj_cg *sc; 312207141Sjeff struct blkhd *hd; 313207141Sjeff 314207141Sjeff sc = cg_lookup(dtog(fs, blk)); 315207141Sjeff if (sc->sc_lastblk && sc->sc_lastblk->sb_blk == blk) 316207141Sjeff return (sc->sc_lastblk); 317207141Sjeff hd = &sc->sc_blkhash[SUJ_HASH(fragstoblks(fs, blk))]; 318207141Sjeff LIST_FOREACH(sblk, hd, sb_next) 319207141Sjeff if (sblk->sb_blk == blk) 320207141Sjeff return (sblk); 321207141Sjeff if (creat == 0) 322207141Sjeff return (NULL); 323207141Sjeff sblk = errmalloc(sizeof(*sblk)); 324207141Sjeff bzero(sblk, sizeof(*sblk)); 325207141Sjeff sblk->sb_blk = blk; 326207141Sjeff TAILQ_INIT(&sblk->sb_recs); 327207141Sjeff LIST_INSERT_HEAD(hd, sblk, sb_next); 328207141Sjeff 329207141Sjeff return (sblk); 330207141Sjeff} 331207141Sjeff 332207141Sjeffstatic struct data_blk * 333207141Sjeffdblk_lookup(ufs2_daddr_t blk) 334207141Sjeff{ 335207141Sjeff struct data_blk *dblk; 336207141Sjeff struct dblkhd *hd; 337207141Sjeff 338207141Sjeff hd = &dbhash[SUJ_HASH(fragstoblks(fs, blk))]; 339207141Sjeff if (lastblk && lastblk->db_blk == blk) 340207141Sjeff return (lastblk); 341207141Sjeff LIST_FOREACH(dblk, hd, db_next) 342207141Sjeff if (dblk->db_blk == blk) 343207141Sjeff return (dblk); 344207141Sjeff /* 345207141Sjeff * The inode block wasn't located, allocate a new one. 346207141Sjeff */ 347207141Sjeff dblk = errmalloc(sizeof(*dblk)); 348207141Sjeff bzero(dblk, sizeof(*dblk)); 349207141Sjeff LIST_INSERT_HEAD(hd, dblk, db_next); 350207141Sjeff dblk->db_blk = blk; 351207141Sjeff return (dblk); 352207141Sjeff} 353207141Sjeff 354207141Sjeffstatic uint8_t * 355207141Sjeffdblk_read(ufs2_daddr_t blk, int size) 356207141Sjeff{ 357207141Sjeff struct data_blk *dblk; 358207141Sjeff 359207141Sjeff dblk = dblk_lookup(blk); 360207141Sjeff /* 361207141Sjeff * I doubt size mismatches can happen in practice but it is trivial 362207141Sjeff * to handle. 363207141Sjeff */ 364207141Sjeff if (size != dblk->db_size) { 365207141Sjeff if (dblk->db_buf) 366207141Sjeff free(dblk->db_buf); 367207141Sjeff dblk->db_buf = errmalloc(size); 368207141Sjeff dblk->db_size = size; 369207141Sjeff if (bread(disk, fsbtodb(fs, blk), dblk->db_buf, size) == -1) 370209408Sdelphij err_suj("Failed to read data block %jd\n", blk); 371207141Sjeff } 372207141Sjeff return (dblk->db_buf); 373207141Sjeff} 374207141Sjeff 375207141Sjeffstatic void 376207141Sjeffdblk_dirty(ufs2_daddr_t blk) 377207141Sjeff{ 378207141Sjeff struct data_blk *dblk; 379207141Sjeff 380207141Sjeff dblk = dblk_lookup(blk); 381207141Sjeff dblk->db_dirty = 1; 382207141Sjeff} 383207141Sjeff 384207141Sjeffstatic void 385207141Sjeffdblk_write(void) 386207141Sjeff{ 387207141Sjeff struct data_blk *dblk; 388207141Sjeff int i; 389207141Sjeff 390207141Sjeff for (i = 0; i < SUJ_HASHSIZE; i++) { 391207141Sjeff LIST_FOREACH(dblk, &dbhash[i], db_next) { 392207141Sjeff if (dblk->db_dirty == 0 || dblk->db_size == 0) 393207141Sjeff continue; 394207141Sjeff if (bwrite(disk, fsbtodb(fs, dblk->db_blk), 395207141Sjeff dblk->db_buf, dblk->db_size) == -1) 396209408Sdelphij err_suj("Unable to write block %jd\n", 397207141Sjeff dblk->db_blk); 398207141Sjeff } 399207141Sjeff } 400207141Sjeff} 401207141Sjeff 402207141Sjeffstatic union dinode * 403207141Sjeffino_read(ino_t ino) 404207141Sjeff{ 405207141Sjeff struct ino_blk *iblk; 406207141Sjeff struct iblkhd *hd; 407207141Sjeff struct suj_cg *sc; 408207141Sjeff ufs2_daddr_t blk; 409207141Sjeff int off; 410207141Sjeff 411207141Sjeff blk = ino_to_fsba(fs, ino); 412207141Sjeff sc = cg_lookup(ino_to_cg(fs, ino)); 413207141Sjeff iblk = sc->sc_lastiblk; 414207141Sjeff if (iblk && iblk->ib_blk == blk) 415207141Sjeff goto found; 416207141Sjeff hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))]; 417207141Sjeff LIST_FOREACH(iblk, hd, ib_next) 418207141Sjeff if (iblk->ib_blk == blk) 419207141Sjeff goto found; 420207141Sjeff /* 421207141Sjeff * The inode block wasn't located, allocate a new one. 422207141Sjeff */ 423207141Sjeff iblk = errmalloc(sizeof(*iblk)); 424207141Sjeff bzero(iblk, sizeof(*iblk)); 425207141Sjeff iblk->ib_buf = errmalloc(fs->fs_bsize); 426207141Sjeff iblk->ib_blk = blk; 427207141Sjeff LIST_INSERT_HEAD(hd, iblk, ib_next); 428207141Sjeff if (bread(disk, fsbtodb(fs, blk), iblk->ib_buf, fs->fs_bsize) == -1) 429209408Sdelphij err_suj("Failed to read inode block %jd\n", blk); 430207141Sjefffound: 431207141Sjeff sc->sc_lastiblk = iblk; 432207141Sjeff off = ino_to_fsbo(fs, ino); 433207141Sjeff if (fs->fs_magic == FS_UFS1_MAGIC) 434207141Sjeff return (union dinode *)&((struct ufs1_dinode *)iblk->ib_buf)[off]; 435207141Sjeff else 436207141Sjeff return (union dinode *)&((struct ufs2_dinode *)iblk->ib_buf)[off]; 437207141Sjeff} 438207141Sjeff 439207141Sjeffstatic void 440207141Sjeffino_dirty(ino_t ino) 441207141Sjeff{ 442207141Sjeff struct ino_blk *iblk; 443207141Sjeff struct iblkhd *hd; 444207141Sjeff struct suj_cg *sc; 445207141Sjeff ufs2_daddr_t blk; 446207141Sjeff 447207141Sjeff blk = ino_to_fsba(fs, ino); 448207141Sjeff sc = cg_lookup(ino_to_cg(fs, ino)); 449207141Sjeff iblk = sc->sc_lastiblk; 450207141Sjeff if (iblk && iblk->ib_blk == blk) { 451207141Sjeff iblk->ib_dirty = 1; 452207141Sjeff return; 453207141Sjeff } 454207141Sjeff hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))]; 455207141Sjeff LIST_FOREACH(iblk, hd, ib_next) { 456207141Sjeff if (iblk->ib_blk == blk) { 457207141Sjeff iblk->ib_dirty = 1; 458207141Sjeff return; 459207141Sjeff } 460207141Sjeff } 461207141Sjeff ino_read(ino); 462207141Sjeff ino_dirty(ino); 463207141Sjeff} 464207141Sjeff 465207141Sjeffstatic void 466207141Sjeffiblk_write(struct ino_blk *iblk) 467207141Sjeff{ 468207141Sjeff 469207141Sjeff if (iblk->ib_dirty == 0) 470207141Sjeff return; 471207141Sjeff if (bwrite(disk, fsbtodb(fs, iblk->ib_blk), iblk->ib_buf, 472207141Sjeff fs->fs_bsize) == -1) 473209408Sdelphij err_suj("Failed to write inode block %jd\n", iblk->ib_blk); 474207141Sjeff} 475207141Sjeff 476207141Sjeffstatic int 477207141Sjeffblk_overlaps(struct jblkrec *brec, ufs2_daddr_t start, int frags) 478207141Sjeff{ 479207141Sjeff ufs2_daddr_t bstart; 480207141Sjeff ufs2_daddr_t bend; 481207141Sjeff ufs2_daddr_t end; 482207141Sjeff 483207141Sjeff end = start + frags; 484207141Sjeff bstart = brec->jb_blkno + brec->jb_oldfrags; 485207141Sjeff bend = bstart + brec->jb_frags; 486207141Sjeff if (start < bend && end > bstart) 487207141Sjeff return (1); 488207141Sjeff return (0); 489207141Sjeff} 490207141Sjeff 491207141Sjeffstatic int 492207141Sjeffblk_equals(struct jblkrec *brec, ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t start, 493207141Sjeff int frags) 494207141Sjeff{ 495207141Sjeff 496207141Sjeff if (brec->jb_ino != ino || brec->jb_lbn != lbn) 497207141Sjeff return (0); 498207141Sjeff if (brec->jb_blkno + brec->jb_oldfrags != start) 499207141Sjeff return (0); 500207141Sjeff if (brec->jb_frags != frags) 501207141Sjeff return (0); 502207141Sjeff return (1); 503207141Sjeff} 504207141Sjeff 505207141Sjeffstatic void 506207141Sjeffblk_setmask(struct jblkrec *brec, int *mask) 507207141Sjeff{ 508207141Sjeff int i; 509207141Sjeff 510207141Sjeff for (i = brec->jb_oldfrags; i < brec->jb_oldfrags + brec->jb_frags; i++) 511207141Sjeff *mask |= 1 << i; 512207141Sjeff} 513207141Sjeff 514207141Sjeff/* 515207141Sjeff * Determine whether a given block has been reallocated to a new location. 516207141Sjeff * Returns a mask of overlapping bits if any frags have been reused or 517207141Sjeff * zero if the block has not been re-used and the contents can be trusted. 518207141Sjeff * 519207141Sjeff * This is used to ensure that an orphaned pointer due to truncate is safe 520207141Sjeff * to be freed. The mask value can be used to free partial blocks. 521207141Sjeff */ 522207141Sjeffstatic int 523207141Sjeffblk_freemask(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags) 524207141Sjeff{ 525207141Sjeff struct suj_blk *sblk; 526207141Sjeff struct suj_rec *srec; 527207141Sjeff struct jblkrec *brec; 528207141Sjeff int mask; 529207141Sjeff int off; 530207141Sjeff 531207141Sjeff /* 532207141Sjeff * To be certain we're not freeing a reallocated block we lookup 533207141Sjeff * this block in the blk hash and see if there is an allocation 534207141Sjeff * journal record that overlaps with any fragments in the block 535207141Sjeff * we're concerned with. If any fragments have ben reallocated 536207141Sjeff * the block has already been freed and re-used for another purpose. 537207141Sjeff */ 538207141Sjeff mask = 0; 539207141Sjeff sblk = blk_lookup(blknum(fs, blk), 0); 540207141Sjeff if (sblk == NULL) 541207141Sjeff return (0); 542207141Sjeff off = blk - sblk->sb_blk; 543207141Sjeff TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) { 544207141Sjeff brec = (struct jblkrec *)srec->sr_rec; 545207141Sjeff /* 546207141Sjeff * If the block overlaps but does not match 547207141Sjeff * exactly it's a new allocation. If it matches 548207141Sjeff * exactly this record refers to the current 549207141Sjeff * location. 550207141Sjeff */ 551207141Sjeff if (blk_overlaps(brec, blk, frags) == 0) 552207141Sjeff continue; 553207141Sjeff if (blk_equals(brec, ino, lbn, blk, frags) == 1) 554207141Sjeff mask = 0; 555207141Sjeff else 556207141Sjeff blk_setmask(brec, &mask); 557207141Sjeff } 558207141Sjeff if (debug) 559207141Sjeff printf("blk_freemask: blk %jd sblk %jd off %d mask 0x%X\n", 560207141Sjeff blk, sblk->sb_blk, off, mask); 561207141Sjeff return (mask >> off); 562207141Sjeff} 563207141Sjeff 564207141Sjeff/* 565207141Sjeff * Determine whether it is safe to follow an indirect. It is not safe 566207141Sjeff * if any part of the indirect has been reallocated or the last journal 567207141Sjeff * entry was an allocation. Just allocated indirects may not have valid 568207141Sjeff * pointers yet and all of their children will have their own records. 569207141Sjeff * It is also not safe to follow an indirect if the cg bitmap has been 570207141Sjeff * cleared as a new allocation may write to the block prior to the journal 571207141Sjeff * being written. 572207141Sjeff * 573207141Sjeff * Returns 1 if it's safe to follow the indirect and 0 otherwise. 574207141Sjeff */ 575207141Sjeffstatic int 576207141Sjeffblk_isindir(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn) 577207141Sjeff{ 578207141Sjeff struct suj_blk *sblk; 579207141Sjeff struct jblkrec *brec; 580207141Sjeff 581207141Sjeff sblk = blk_lookup(blk, 0); 582207141Sjeff if (sblk == NULL) 583207141Sjeff return (1); 584207141Sjeff if (TAILQ_EMPTY(&sblk->sb_recs)) 585207141Sjeff return (1); 586207141Sjeff brec = (struct jblkrec *)TAILQ_LAST(&sblk->sb_recs, srechd)->sr_rec; 587207141Sjeff if (blk_equals(brec, ino, lbn, blk, fs->fs_frag)) 588207141Sjeff if (brec->jb_op == JOP_FREEBLK) 589207141Sjeff return (!blk_isfree(blk)); 590207141Sjeff return (0); 591207141Sjeff} 592207141Sjeff 593207141Sjeff/* 594207141Sjeff * Clear an inode from the cg bitmap. If the inode was already clear return 595207141Sjeff * 0 so the caller knows it does not have to check the inode contents. 596207141Sjeff */ 597207141Sjeffstatic int 598207141Sjeffino_free(ino_t ino, int mode) 599207141Sjeff{ 600207141Sjeff struct suj_cg *sc; 601207141Sjeff uint8_t *inosused; 602207141Sjeff struct cg *cgp; 603207141Sjeff int cg; 604207141Sjeff 605207141Sjeff cg = ino_to_cg(fs, ino); 606207141Sjeff ino = ino % fs->fs_ipg; 607207141Sjeff sc = cg_lookup(cg); 608207141Sjeff cgp = sc->sc_cgp; 609207141Sjeff inosused = cg_inosused(cgp); 610207141Sjeff /* 611207141Sjeff * The bitmap may never have made it to the disk so we have to 612207141Sjeff * conditionally clear. We can avoid writing the cg in this case. 613207141Sjeff */ 614207141Sjeff if (isclr(inosused, ino)) 615207141Sjeff return (0); 616207141Sjeff freeinos++; 617207141Sjeff clrbit(inosused, ino); 618207141Sjeff if (ino < cgp->cg_irotor) 619207141Sjeff cgp->cg_irotor = ino; 620207141Sjeff cgp->cg_cs.cs_nifree++; 621207141Sjeff if ((mode & IFMT) == IFDIR) { 622207141Sjeff freedir++; 623207141Sjeff cgp->cg_cs.cs_ndir--; 624207141Sjeff } 625207141Sjeff sc->sc_dirty = 1; 626207141Sjeff 627207141Sjeff return (1); 628207141Sjeff} 629207141Sjeff 630207141Sjeff/* 631207141Sjeff * Free 'frags' frags starting at filesystem block 'bno' skipping any frags 632207141Sjeff * set in the mask. 633207141Sjeff */ 634207141Sjeffstatic void 635207141Sjeffblk_free(ufs2_daddr_t bno, int mask, int frags) 636207141Sjeff{ 637207141Sjeff ufs1_daddr_t fragno, cgbno; 638207141Sjeff struct suj_cg *sc; 639207141Sjeff struct cg *cgp; 640207141Sjeff int i, cg; 641207141Sjeff uint8_t *blksfree; 642207141Sjeff 643207141Sjeff if (debug) 644207141Sjeff printf("Freeing %d frags at blk %jd\n", frags, bno); 645207141Sjeff cg = dtog(fs, bno); 646207141Sjeff sc = cg_lookup(cg); 647207141Sjeff cgp = sc->sc_cgp; 648207141Sjeff cgbno = dtogd(fs, bno); 649207141Sjeff blksfree = cg_blksfree(cgp); 650207141Sjeff 651207141Sjeff /* 652207141Sjeff * If it's not allocated we only wrote the journal entry 653207141Sjeff * and never the bitmaps. Here we unconditionally clear and 654207141Sjeff * resolve the cg summary later. 655207141Sjeff */ 656207141Sjeff if (frags == fs->fs_frag && mask == 0) { 657207141Sjeff fragno = fragstoblks(fs, cgbno); 658207141Sjeff ffs_setblock(fs, blksfree, fragno); 659207141Sjeff freeblocks++; 660207141Sjeff } else { 661207141Sjeff /* 662207141Sjeff * deallocate the fragment 663207141Sjeff */ 664207141Sjeff for (i = 0; i < frags; i++) 665207141Sjeff if ((mask & (1 << i)) == 0 && isclr(blksfree, cgbno +i)) { 666207141Sjeff freefrags++; 667207141Sjeff setbit(blksfree, cgbno + i); 668207141Sjeff } 669207141Sjeff } 670207141Sjeff sc->sc_dirty = 1; 671207141Sjeff} 672207141Sjeff 673207141Sjeff/* 674207141Sjeff * Returns 1 if the whole block starting at 'bno' is marked free and 0 675207141Sjeff * otherwise. 676207141Sjeff */ 677207141Sjeffstatic int 678207141Sjeffblk_isfree(ufs2_daddr_t bno) 679207141Sjeff{ 680207141Sjeff struct suj_cg *sc; 681207141Sjeff 682207141Sjeff sc = cg_lookup(dtog(fs, bno)); 683207141Sjeff return ffs_isblock(fs, cg_blksfree(sc->sc_cgp), dtogd(fs, bno)); 684207141Sjeff} 685207141Sjeff 686207141Sjeff/* 687207141Sjeff * Fetch an indirect block to find the block at a given lbn. The lbn 688207141Sjeff * may be negative to fetch a specific indirect block pointer or positive 689207141Sjeff * to fetch a specific block. 690207141Sjeff */ 691207141Sjeffstatic ufs2_daddr_t 692207141Sjeffindir_blkatoff(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t cur, ufs_lbn_t lbn) 693207141Sjeff{ 694207141Sjeff ufs2_daddr_t *bap2; 695207141Sjeff ufs2_daddr_t *bap1; 696207141Sjeff ufs_lbn_t lbnadd; 697207141Sjeff ufs_lbn_t base; 698207141Sjeff int level; 699207141Sjeff int i; 700207141Sjeff 701207141Sjeff if (blk == 0) 702207141Sjeff return (0); 703207141Sjeff level = lbn_level(cur); 704207141Sjeff if (level == -1) 705209408Sdelphij err_suj("Invalid indir lbn %jd\n", lbn); 706207141Sjeff if (level == 0 && lbn < 0) 707209408Sdelphij err_suj("Invalid lbn %jd\n", lbn); 708207141Sjeff bap2 = (void *)dblk_read(blk, fs->fs_bsize); 709207141Sjeff bap1 = (void *)bap2; 710207141Sjeff lbnadd = 1; 711207141Sjeff base = -(cur + level); 712207141Sjeff for (i = level; i > 0; i--) 713207141Sjeff lbnadd *= NINDIR(fs); 714207141Sjeff if (lbn > 0) 715207141Sjeff i = (lbn - base) / lbnadd; 716207141Sjeff else 717207141Sjeff i = (-lbn - base) / lbnadd; 718207141Sjeff if (i < 0 || i >= NINDIR(fs)) 719209408Sdelphij err_suj("Invalid indirect index %d produced by lbn %jd\n", 720207141Sjeff i, lbn); 721207141Sjeff if (level == 0) 722207141Sjeff cur = base + (i * lbnadd); 723207141Sjeff else 724207141Sjeff cur = -(base + (i * lbnadd)) - (level - 1); 725207141Sjeff if (fs->fs_magic == FS_UFS1_MAGIC) 726207141Sjeff blk = bap1[i]; 727207141Sjeff else 728207141Sjeff blk = bap2[i]; 729207141Sjeff if (cur == lbn) 730207141Sjeff return (blk); 731209408Sdelphij if (level == 0) 732209408Sdelphij err_suj("Invalid lbn %jd at level 0\n", lbn); 733207141Sjeff return indir_blkatoff(blk, ino, cur, lbn); 734207141Sjeff} 735207141Sjeff 736207141Sjeff/* 737207141Sjeff * Finds the disk block address at the specified lbn within the inode 738207141Sjeff * specified by ip. This follows the whole tree and honors di_size and 739207141Sjeff * di_extsize so it is a true test of reachability. The lbn may be 740207141Sjeff * negative if an extattr or indirect block is requested. 741207141Sjeff */ 742207141Sjeffstatic ufs2_daddr_t 743207141Sjeffino_blkatoff(union dinode *ip, ino_t ino, ufs_lbn_t lbn, int *frags) 744207141Sjeff{ 745207141Sjeff ufs_lbn_t tmpval; 746207141Sjeff ufs_lbn_t cur; 747207141Sjeff ufs_lbn_t next; 748207141Sjeff int i; 749207141Sjeff 750207141Sjeff /* 751207141Sjeff * Handle extattr blocks first. 752207141Sjeff */ 753207141Sjeff if (lbn < 0 && lbn >= -NXADDR) { 754207141Sjeff lbn = -1 - lbn; 755207141Sjeff if (lbn > lblkno(fs, ip->dp2.di_extsize - 1)) 756207141Sjeff return (0); 757207141Sjeff *frags = numfrags(fs, sblksize(fs, ip->dp2.di_extsize, lbn)); 758207141Sjeff return (ip->dp2.di_extb[lbn]); 759207141Sjeff } 760207141Sjeff /* 761207141Sjeff * Now direct and indirect. 762207141Sjeff */ 763207141Sjeff if (DIP(ip, di_mode) == IFLNK && 764207141Sjeff DIP(ip, di_size) < fs->fs_maxsymlinklen) 765207141Sjeff return (0); 766207141Sjeff if (lbn >= 0 && lbn < NDADDR) { 767207141Sjeff *frags = numfrags(fs, sblksize(fs, DIP(ip, di_size), lbn)); 768207141Sjeff return (DIP(ip, di_db[lbn])); 769207141Sjeff } 770207141Sjeff *frags = fs->fs_frag; 771207141Sjeff 772207141Sjeff for (i = 0, tmpval = NINDIR(fs), cur = NDADDR; i < NIADDR; i++, 773207141Sjeff tmpval *= NINDIR(fs), cur = next) { 774207141Sjeff next = cur + tmpval; 775207141Sjeff if (lbn == -cur - i) 776207141Sjeff return (DIP(ip, di_ib[i])); 777207141Sjeff /* 778207141Sjeff * Determine whether the lbn in question is within this tree. 779207141Sjeff */ 780207141Sjeff if (lbn < 0 && -lbn >= next) 781207141Sjeff continue; 782207141Sjeff if (lbn > 0 && lbn >= next) 783207141Sjeff continue; 784207141Sjeff return indir_blkatoff(DIP(ip, di_ib[i]), ino, -cur - i, lbn); 785207141Sjeff } 786209408Sdelphij err_suj("lbn %jd not in ino\n", lbn); 787209408Sdelphij /* NOTREACHED */ 788207141Sjeff} 789207141Sjeff 790207141Sjeff/* 791207141Sjeff * Determine whether a block exists at a particular lbn in an inode. 792207141Sjeff * Returns 1 if found, 0 if not. lbn may be negative for indirects 793207141Sjeff * or ext blocks. 794207141Sjeff */ 795207141Sjeffstatic int 796207141Sjeffblk_isat(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int *frags) 797207141Sjeff{ 798207141Sjeff union dinode *ip; 799207141Sjeff ufs2_daddr_t nblk; 800207141Sjeff 801207141Sjeff ip = ino_read(ino); 802207141Sjeff 803207141Sjeff if (DIP(ip, di_nlink) == 0 || DIP(ip, di_mode) == 0) 804207141Sjeff return (0); 805207141Sjeff nblk = ino_blkatoff(ip, ino, lbn, frags); 806207141Sjeff 807207141Sjeff return (nblk == blk); 808207141Sjeff} 809207141Sjeff 810207141Sjeff/* 811209716Sjeff * Clear the directory entry at diroff that should point to child. Minimal 812209716Sjeff * checking is done and it is assumed that this path was verified with isat. 813209716Sjeff */ 814209716Sjeffstatic void 815209716Sjeffino_clrat(ino_t parent, off_t diroff, ino_t child) 816209716Sjeff{ 817209716Sjeff union dinode *dip; 818209716Sjeff struct direct *dp; 819209716Sjeff ufs2_daddr_t blk; 820209716Sjeff uint8_t *block; 821209716Sjeff ufs_lbn_t lbn; 822209716Sjeff int blksize; 823209716Sjeff int frags; 824209716Sjeff int doff; 825209716Sjeff 826209716Sjeff if (debug) 827209716Sjeff printf("Clearing inode %d from parent %d at offset %jd\n", 828209716Sjeff child, parent, diroff); 829209716Sjeff 830209716Sjeff lbn = lblkno(fs, diroff); 831209716Sjeff doff = blkoff(fs, diroff); 832209716Sjeff dip = ino_read(parent); 833209716Sjeff blk = ino_blkatoff(dip, parent, lbn, &frags); 834209716Sjeff blksize = sblksize(fs, DIP(dip, di_size), lbn); 835209716Sjeff block = dblk_read(blk, blksize); 836209716Sjeff dp = (struct direct *)&block[doff]; 837209716Sjeff if (dp->d_ino != child) 838209716Sjeff errx(1, "Inode %d does not exist in %d at %jd", 839209716Sjeff child, parent, diroff); 840209716Sjeff dp->d_ino = 0; 841209716Sjeff dblk_dirty(blk); 842209716Sjeff /* 843209716Sjeff * The actual .. reference count will already have been removed 844209716Sjeff * from the parent by the .. remref record. 845209716Sjeff */ 846209716Sjeff} 847209716Sjeff 848209716Sjeff/* 849207141Sjeff * Determines whether a pointer to an inode exists within a directory 850207141Sjeff * at a specified offset. Returns the mode of the found entry. 851207141Sjeff */ 852207141Sjeffstatic int 853207141Sjeffino_isat(ino_t parent, off_t diroff, ino_t child, int *mode, int *isdot) 854207141Sjeff{ 855207141Sjeff union dinode *dip; 856207141Sjeff struct direct *dp; 857207141Sjeff ufs2_daddr_t blk; 858207141Sjeff uint8_t *block; 859207141Sjeff ufs_lbn_t lbn; 860207141Sjeff int blksize; 861207141Sjeff int frags; 862207141Sjeff int dpoff; 863207141Sjeff int doff; 864207141Sjeff 865207141Sjeff *isdot = 0; 866207141Sjeff dip = ino_read(parent); 867207141Sjeff *mode = DIP(dip, di_mode); 868207141Sjeff if ((*mode & IFMT) != IFDIR) { 869207141Sjeff if (debug) { 870207141Sjeff /* 871207141Sjeff * This can happen if the parent inode 872207141Sjeff * was reallocated. 873207141Sjeff */ 874207141Sjeff if (*mode != 0) 875207141Sjeff printf("Directory %d has bad mode %o\n", 876207141Sjeff parent, *mode); 877207141Sjeff else 878207141Sjeff printf("Directory %d zero inode\n", parent); 879207141Sjeff } 880207141Sjeff return (0); 881207141Sjeff } 882207141Sjeff lbn = lblkno(fs, diroff); 883207141Sjeff doff = blkoff(fs, diroff); 884207141Sjeff blksize = sblksize(fs, DIP(dip, di_size), lbn); 885207141Sjeff if (diroff + DIRECTSIZ(1) > DIP(dip, di_size) || doff >= blksize) { 886207141Sjeff if (debug) 887207141Sjeff printf("ino %d absent from %d due to offset %jd" 888207141Sjeff " exceeding size %jd\n", 889207141Sjeff child, parent, diroff, DIP(dip, di_size)); 890207141Sjeff return (0); 891207141Sjeff } 892207141Sjeff blk = ino_blkatoff(dip, parent, lbn, &frags); 893207141Sjeff if (blk <= 0) { 894207141Sjeff if (debug) 895207141Sjeff printf("Sparse directory %d", parent); 896207141Sjeff return (0); 897207141Sjeff } 898207141Sjeff block = dblk_read(blk, blksize); 899207141Sjeff /* 900207141Sjeff * Walk through the records from the start of the block to be 901207141Sjeff * certain we hit a valid record and not some junk in the middle 902207141Sjeff * of a file name. Stop when we reach or pass the expected offset. 903207141Sjeff */ 904207141Sjeff dpoff = (doff / DIRBLKSIZ) * DIRBLKSIZ; 905207141Sjeff do { 906207141Sjeff dp = (struct direct *)&block[dpoff]; 907207141Sjeff if (dpoff == doff) 908207141Sjeff break; 909207141Sjeff if (dp->d_reclen == 0) 910207141Sjeff break; 911207141Sjeff dpoff += dp->d_reclen; 912207141Sjeff } while (dpoff <= doff); 913207141Sjeff if (dpoff > fs->fs_bsize) 914209408Sdelphij err_suj("Corrupt directory block in dir ino %d\n", parent); 915207141Sjeff /* Not found. */ 916207141Sjeff if (dpoff != doff) { 917207141Sjeff if (debug) 918207141Sjeff printf("ino %d not found in %d, lbn %jd, dpoff %d\n", 919207141Sjeff child, parent, lbn, dpoff); 920207141Sjeff return (0); 921207141Sjeff } 922207141Sjeff /* 923207141Sjeff * We found the item in question. Record the mode and whether it's 924207141Sjeff * a . or .. link for the caller. 925207141Sjeff */ 926207141Sjeff if (dp->d_ino == child) { 927207141Sjeff if (child == parent) 928207141Sjeff *isdot = 1; 929207141Sjeff else if (dp->d_namlen == 2 && 930207141Sjeff dp->d_name[0] == '.' && dp->d_name[1] == '.') 931207141Sjeff *isdot = 1; 932207141Sjeff *mode = DTTOIF(dp->d_type); 933207141Sjeff return (1); 934207141Sjeff } 935207141Sjeff if (debug) 936207141Sjeff printf("ino %d doesn't match dirent ino %d in parent %d\n", 937207141Sjeff child, dp->d_ino, parent); 938207141Sjeff return (0); 939207141Sjeff} 940207141Sjeff 941207141Sjeff#define VISIT_INDIR 0x0001 942207141Sjeff#define VISIT_EXT 0x0002 943207141Sjeff#define VISIT_ROOT 0x0004 /* Operation came via root & valid pointers. */ 944207141Sjeff 945207141Sjeff/* 946207141Sjeff * Read an indirect level which may or may not be linked into an inode. 947207141Sjeff */ 948207141Sjeffstatic void 949207141Sjeffindir_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, uint64_t *frags, 950207141Sjeff ino_visitor visitor, int flags) 951207141Sjeff{ 952207141Sjeff ufs2_daddr_t *bap2; 953207141Sjeff ufs1_daddr_t *bap1; 954207141Sjeff ufs_lbn_t lbnadd; 955207141Sjeff ufs2_daddr_t nblk; 956207141Sjeff ufs_lbn_t nlbn; 957207141Sjeff int level; 958207141Sjeff int i; 959207141Sjeff 960207141Sjeff /* 961207141Sjeff * Don't visit indirect blocks with contents we can't trust. This 962207141Sjeff * should only happen when indir_visit() is called to complete a 963207141Sjeff * truncate that never finished and not when a pointer is found via 964207141Sjeff * an inode. 965207141Sjeff */ 966207141Sjeff if (blk == 0) 967207141Sjeff return; 968207141Sjeff level = lbn_level(lbn); 969207141Sjeff if (level == -1) 970209408Sdelphij err_suj("Invalid level for lbn %jd\n", lbn); 971207141Sjeff if ((flags & VISIT_ROOT) == 0 && blk_isindir(blk, ino, lbn) == 0) { 972207141Sjeff if (debug) 973207141Sjeff printf("blk %jd ino %d lbn %jd(%d) is not indir.\n", 974207141Sjeff blk, ino, lbn, level); 975207141Sjeff goto out; 976207141Sjeff } 977207141Sjeff lbnadd = 1; 978207141Sjeff for (i = level; i > 0; i--) 979207141Sjeff lbnadd *= NINDIR(fs); 980207141Sjeff bap1 = (void *)dblk_read(blk, fs->fs_bsize); 981207141Sjeff bap2 = (void *)bap1; 982207141Sjeff for (i = 0; i < NINDIR(fs); i++) { 983207141Sjeff if (fs->fs_magic == FS_UFS1_MAGIC) 984207141Sjeff nblk = *bap1++; 985207141Sjeff else 986207141Sjeff nblk = *bap2++; 987207141Sjeff if (nblk == 0) 988207141Sjeff continue; 989207141Sjeff if (level == 0) { 990207141Sjeff nlbn = -lbn + i * lbnadd; 991207141Sjeff (*frags) += fs->fs_frag; 992207141Sjeff visitor(ino, nlbn, nblk, fs->fs_frag); 993207141Sjeff } else { 994207141Sjeff nlbn = (lbn + 1) - (i * lbnadd); 995207141Sjeff indir_visit(ino, nlbn, nblk, frags, visitor, flags); 996207141Sjeff } 997207141Sjeff } 998207141Sjeffout: 999207141Sjeff if (flags & VISIT_INDIR) { 1000207141Sjeff (*frags) += fs->fs_frag; 1001207141Sjeff visitor(ino, lbn, blk, fs->fs_frag); 1002207141Sjeff } 1003207141Sjeff} 1004207141Sjeff 1005207141Sjeff/* 1006207141Sjeff * Visit each block in an inode as specified by 'flags' and call a 1007207141Sjeff * callback function. The callback may inspect or free blocks. The 1008207141Sjeff * count of frags found according to the size in the file is returned. 1009207141Sjeff * This is not valid for sparse files but may be used to determine 1010207141Sjeff * the correct di_blocks for a file. 1011207141Sjeff */ 1012207141Sjeffstatic uint64_t 1013207141Sjeffino_visit(union dinode *ip, ino_t ino, ino_visitor visitor, int flags) 1014207141Sjeff{ 1015207141Sjeff ufs_lbn_t nextlbn; 1016207141Sjeff ufs_lbn_t tmpval; 1017207141Sjeff ufs_lbn_t lbn; 1018207141Sjeff uint64_t size; 1019207141Sjeff uint64_t fragcnt; 1020207141Sjeff int mode; 1021207141Sjeff int frags; 1022207141Sjeff int i; 1023207141Sjeff 1024207141Sjeff size = DIP(ip, di_size); 1025207141Sjeff mode = DIP(ip, di_mode) & IFMT; 1026207141Sjeff fragcnt = 0; 1027207141Sjeff if ((flags & VISIT_EXT) && 1028207141Sjeff fs->fs_magic == FS_UFS2_MAGIC && ip->dp2.di_extsize) { 1029207141Sjeff for (i = 0; i < NXADDR; i++) { 1030207141Sjeff if (ip->dp2.di_extb[i] == 0) 1031207141Sjeff continue; 1032207141Sjeff frags = sblksize(fs, ip->dp2.di_extsize, i); 1033207141Sjeff frags = numfrags(fs, frags); 1034207141Sjeff fragcnt += frags; 1035207141Sjeff visitor(ino, -1 - i, ip->dp2.di_extb[i], frags); 1036207141Sjeff } 1037207141Sjeff } 1038207141Sjeff /* Skip datablocks for short links and devices. */ 1039207141Sjeff if (mode == IFBLK || mode == IFCHR || 1040207141Sjeff (mode == IFLNK && size < fs->fs_maxsymlinklen)) 1041207141Sjeff return (fragcnt); 1042207141Sjeff for (i = 0; i < NDADDR; i++) { 1043207141Sjeff if (DIP(ip, di_db[i]) == 0) 1044207141Sjeff continue; 1045207141Sjeff frags = sblksize(fs, size, i); 1046207141Sjeff frags = numfrags(fs, frags); 1047207141Sjeff fragcnt += frags; 1048207141Sjeff visitor(ino, i, DIP(ip, di_db[i]), frags); 1049207141Sjeff } 1050207141Sjeff /* 1051207141Sjeff * We know the following indirects are real as we're following 1052207141Sjeff * real pointers to them. 1053207141Sjeff */ 1054207141Sjeff flags |= VISIT_ROOT; 1055207141Sjeff for (i = 0, tmpval = NINDIR(fs), lbn = NDADDR; i < NIADDR; i++, 1056207141Sjeff lbn = nextlbn) { 1057207141Sjeff nextlbn = lbn + tmpval; 1058207141Sjeff tmpval *= NINDIR(fs); 1059207141Sjeff if (DIP(ip, di_ib[i]) == 0) 1060207141Sjeff continue; 1061207141Sjeff indir_visit(ino, -lbn - i, DIP(ip, di_ib[i]), &fragcnt, visitor, 1062207141Sjeff flags); 1063207141Sjeff } 1064207141Sjeff return (fragcnt); 1065207141Sjeff} 1066207141Sjeff 1067207141Sjeff/* 1068207141Sjeff * Null visitor function used when we just want to count blocks and 1069207141Sjeff * record the lbn. 1070207141Sjeff */ 1071207141Sjeffufs_lbn_t visitlbn; 1072207141Sjeffstatic void 1073207141Sjeffnull_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 1074207141Sjeff{ 1075207141Sjeff if (lbn > 0) 1076207141Sjeff visitlbn = lbn; 1077207141Sjeff} 1078207141Sjeff 1079207141Sjeff/* 1080207141Sjeff * Recalculate di_blocks when we discover that a block allocation or 1081207141Sjeff * free was not successfully completed. The kernel does not roll this back 1082207141Sjeff * because it would be too expensive to compute which indirects were 1083207141Sjeff * reachable at the time the inode was written. 1084207141Sjeff */ 1085207141Sjeffstatic void 1086207141Sjeffino_adjblks(struct suj_ino *sino) 1087207141Sjeff{ 1088207141Sjeff union dinode *ip; 1089207141Sjeff uint64_t blocks; 1090207141Sjeff uint64_t frags; 1091207141Sjeff off_t isize; 1092207141Sjeff off_t size; 1093207141Sjeff ino_t ino; 1094207141Sjeff 1095207141Sjeff ino = sino->si_ino; 1096207141Sjeff ip = ino_read(ino); 1097207141Sjeff /* No need to adjust zero'd inodes. */ 1098207141Sjeff if (DIP(ip, di_mode) == 0) 1099207141Sjeff return; 1100207141Sjeff /* 1101207141Sjeff * Visit all blocks and count them as well as recording the last 1102207141Sjeff * valid lbn in the file. If the file size doesn't agree with the 1103207141Sjeff * last lbn we need to truncate to fix it. Otherwise just adjust 1104207141Sjeff * the blocks count. 1105207141Sjeff */ 1106207141Sjeff visitlbn = 0; 1107207141Sjeff frags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT); 1108207141Sjeff blocks = fsbtodb(fs, frags); 1109207141Sjeff /* 1110207141Sjeff * We assume the size and direct block list is kept coherent by 1111207141Sjeff * softdep. For files that have extended into indirects we truncate 1112207141Sjeff * to the size in the inode or the maximum size permitted by 1113207141Sjeff * populated indirects. 1114207141Sjeff */ 1115207141Sjeff if (visitlbn >= NDADDR) { 1116207141Sjeff isize = DIP(ip, di_size); 1117207141Sjeff size = lblktosize(fs, visitlbn + 1); 1118207141Sjeff if (isize > size) 1119207141Sjeff isize = size; 1120207141Sjeff /* Always truncate to free any unpopulated indirects. */ 1121207141Sjeff ino_trunc(sino->si_ino, isize); 1122207141Sjeff return; 1123207141Sjeff } 1124207141Sjeff if (blocks == DIP(ip, di_blocks)) 1125207141Sjeff return; 1126207141Sjeff if (debug) 1127207141Sjeff printf("ino %d adjusting block count from %jd to %jd\n", 1128207141Sjeff ino, DIP(ip, di_blocks), blocks); 1129207141Sjeff DIP_SET(ip, di_blocks, blocks); 1130207141Sjeff ino_dirty(ino); 1131207141Sjeff} 1132207141Sjeff 1133207141Sjeffstatic void 1134207141Sjeffblk_free_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 1135207141Sjeff{ 1136207141Sjeff int mask; 1137207141Sjeff 1138207141Sjeff mask = blk_freemask(blk, ino, lbn, frags); 1139207141Sjeff if (debug) 1140207141Sjeff printf("blk %jd freemask 0x%X\n", blk, mask); 1141207141Sjeff blk_free(blk, mask, frags); 1142207141Sjeff} 1143207141Sjeff 1144207141Sjeff/* 1145207141Sjeff * Free a block or tree of blocks that was previously rooted in ino at 1146207141Sjeff * the given lbn. If the lbn is an indirect all children are freed 1147207141Sjeff * recursively. 1148207141Sjeff */ 1149207141Sjeffstatic void 1150207141Sjeffblk_free_lbn(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags, int follow) 1151207141Sjeff{ 1152207141Sjeff uint64_t resid; 1153207141Sjeff int mask; 1154207141Sjeff 1155207141Sjeff mask = blk_freemask(blk, ino, lbn, frags); 1156207141Sjeff if (debug) 1157207141Sjeff printf("blk %jd freemask 0x%X\n", blk, mask); 1158207141Sjeff resid = 0; 1159207141Sjeff if (lbn <= -NDADDR && follow && mask == 0) 1160207141Sjeff indir_visit(ino, lbn, blk, &resid, blk_free_visit, VISIT_INDIR); 1161207141Sjeff else 1162207141Sjeff blk_free(blk, mask, frags); 1163207141Sjeff} 1164207141Sjeff 1165207141Sjeffstatic void 1166207141Sjeffino_setskip(struct suj_ino *sino, ino_t parent) 1167207141Sjeff{ 1168207141Sjeff int isdot; 1169207141Sjeff int mode; 1170207141Sjeff 1171207141Sjeff if (ino_isat(sino->si_ino, DOTDOT_OFFSET, parent, &mode, &isdot)) 1172207141Sjeff sino->si_skipparent = 1; 1173207141Sjeff} 1174207141Sjeff 1175209716Sjeffstatic void 1176209716Sjeffino_remref(ino_t parent, ino_t child, uint64_t diroff, int isdotdot) 1177209716Sjeff{ 1178209716Sjeff struct suj_ino *sino; 1179209716Sjeff struct suj_rec *srec; 1180209716Sjeff struct jrefrec *rrec; 1181209716Sjeff 1182209716Sjeff /* 1183209716Sjeff * Lookup this inode to see if we have a record for it. 1184209716Sjeff */ 1185209716Sjeff sino = ino_lookup(child, 0); 1186209716Sjeff /* 1187209716Sjeff * Tell any child directories we've already removed their 1188209716Sjeff * parent link cnt. Don't try to adjust our link down again. 1189209716Sjeff */ 1190209716Sjeff if (sino != NULL && isdotdot == 0) 1191209716Sjeff ino_setskip(sino, parent); 1192209716Sjeff /* 1193209716Sjeff * No valid record for this inode. Just drop the on-disk 1194209716Sjeff * link by one. 1195209716Sjeff */ 1196209716Sjeff if (sino == NULL || sino->si_hasrecs == 0) { 1197209716Sjeff ino_decr(child); 1198209716Sjeff return; 1199209716Sjeff } 1200209716Sjeff /* 1201209716Sjeff * Use ino_adjust() if ino_check() has already processed this 1202209716Sjeff * child. If we lose the last non-dot reference to a 1203209716Sjeff * directory it will be discarded. 1204209716Sjeff */ 1205209716Sjeff if (sino->si_linkadj) { 1206209716Sjeff sino->si_nlink--; 1207209716Sjeff if (isdotdot) 1208209716Sjeff sino->si_dotlinks--; 1209209716Sjeff ino_adjust(sino); 1210209716Sjeff return; 1211209716Sjeff } 1212209716Sjeff /* 1213209716Sjeff * If we haven't yet processed this inode we need to make 1214209716Sjeff * sure we will successfully discover the lost path. If not 1215209716Sjeff * use nlinkadj to remember. 1216209716Sjeff */ 1217209716Sjeff TAILQ_FOREACH(srec, &sino->si_recs, sr_next) { 1218209716Sjeff rrec = (struct jrefrec *)srec->sr_rec; 1219209716Sjeff if (rrec->jr_parent == parent && 1220209716Sjeff rrec->jr_diroff == diroff) 1221209716Sjeff return; 1222209716Sjeff } 1223209716Sjeff sino->si_nlinkadj++; 1224209716Sjeff} 1225209716Sjeff 1226207141Sjeff/* 1227207141Sjeff * Free the children of a directory when the directory is discarded. 1228207141Sjeff */ 1229207141Sjeffstatic void 1230207141Sjeffino_free_children(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 1231207141Sjeff{ 1232207141Sjeff struct suj_ino *sino; 1233207141Sjeff struct direct *dp; 1234207141Sjeff off_t diroff; 1235207141Sjeff uint8_t *block; 1236207141Sjeff int skipparent; 1237209716Sjeff int isdotdot; 1238207141Sjeff int dpoff; 1239207141Sjeff int size; 1240207141Sjeff 1241207141Sjeff sino = ino_lookup(ino, 0); 1242207141Sjeff if (sino) 1243207141Sjeff skipparent = sino->si_skipparent; 1244207141Sjeff else 1245207141Sjeff skipparent = 0; 1246207141Sjeff size = lfragtosize(fs, frags); 1247207141Sjeff block = dblk_read(blk, size); 1248207141Sjeff dp = (struct direct *)&block[0]; 1249207141Sjeff for (dpoff = 0; dpoff < size && dp->d_reclen; dpoff += dp->d_reclen) { 1250207141Sjeff dp = (struct direct *)&block[dpoff]; 1251207141Sjeff if (dp->d_ino == 0 || dp->d_ino == WINO) 1252207141Sjeff continue; 1253207141Sjeff if (dp->d_namlen == 1 && dp->d_name[0] == '.') 1254207141Sjeff continue; 1255209716Sjeff isdotdot = dp->d_namlen == 2 && dp->d_name[0] == '.' && 1256207141Sjeff dp->d_name[1] == '.'; 1257209716Sjeff if (isdotdot && skipparent == 1) 1258207141Sjeff continue; 1259207141Sjeff if (debug) 1260207141Sjeff printf("Directory %d removing ino %d name %s\n", 1261207141Sjeff ino, dp->d_ino, dp->d_name); 1262207141Sjeff diroff = lblktosize(fs, lbn) + dpoff; 1263209716Sjeff ino_remref(ino, dp->d_ino, diroff, isdotdot); 1264207141Sjeff } 1265207141Sjeff} 1266207141Sjeff 1267207141Sjeff/* 1268207141Sjeff * Reclaim an inode, freeing all blocks and decrementing all children's 1269207141Sjeff * link counts. Free the inode back to the cg. 1270207141Sjeff */ 1271207141Sjeffstatic void 1272207141Sjeffino_reclaim(union dinode *ip, ino_t ino, int mode) 1273207141Sjeff{ 1274207141Sjeff uint32_t gen; 1275207141Sjeff 1276207141Sjeff if (ino == ROOTINO) 1277209408Sdelphij err_suj("Attempting to free ROOTINO\n"); 1278207141Sjeff if (debug) 1279207141Sjeff printf("Truncating and freeing ino %d, nlink %d, mode %o\n", 1280207141Sjeff ino, DIP(ip, di_nlink), DIP(ip, di_mode)); 1281207141Sjeff 1282207141Sjeff /* We are freeing an inode or directory. */ 1283207141Sjeff if ((DIP(ip, di_mode) & IFMT) == IFDIR) 1284207141Sjeff ino_visit(ip, ino, ino_free_children, 0); 1285207141Sjeff DIP_SET(ip, di_nlink, 0); 1286207141Sjeff ino_visit(ip, ino, blk_free_visit, VISIT_EXT | VISIT_INDIR); 1287207141Sjeff /* Here we have to clear the inode and release any blocks it holds. */ 1288207141Sjeff gen = DIP(ip, di_gen); 1289207141Sjeff if (fs->fs_magic == FS_UFS1_MAGIC) 1290207141Sjeff bzero(ip, sizeof(struct ufs1_dinode)); 1291207141Sjeff else 1292207141Sjeff bzero(ip, sizeof(struct ufs2_dinode)); 1293207141Sjeff DIP_SET(ip, di_gen, gen); 1294207141Sjeff ino_dirty(ino); 1295207141Sjeff ino_free(ino, mode); 1296207141Sjeff return; 1297207141Sjeff} 1298207141Sjeff 1299207141Sjeff/* 1300207141Sjeff * Adjust an inode's link count down by one when a directory goes away. 1301207141Sjeff */ 1302207141Sjeffstatic void 1303207141Sjeffino_decr(ino_t ino) 1304207141Sjeff{ 1305207141Sjeff union dinode *ip; 1306207141Sjeff int reqlink; 1307207141Sjeff int nlink; 1308207141Sjeff int mode; 1309207141Sjeff 1310207141Sjeff ip = ino_read(ino); 1311207141Sjeff nlink = DIP(ip, di_nlink); 1312207141Sjeff mode = DIP(ip, di_mode); 1313207141Sjeff if (nlink < 1) 1314209408Sdelphij err_suj("Inode %d link count %d invalid\n", ino, nlink); 1315207141Sjeff if (mode == 0) 1316209408Sdelphij err_suj("Inode %d has a link of %d with 0 mode\n", ino, nlink); 1317207141Sjeff nlink--; 1318207141Sjeff if ((mode & IFMT) == IFDIR) 1319207141Sjeff reqlink = 2; 1320207141Sjeff else 1321207141Sjeff reqlink = 1; 1322207141Sjeff if (nlink < reqlink) { 1323207141Sjeff if (debug) 1324207141Sjeff printf("ino %d not enough links to live %d < %d\n", 1325207141Sjeff ino, nlink, reqlink); 1326207141Sjeff ino_reclaim(ip, ino, mode); 1327207141Sjeff return; 1328207141Sjeff } 1329207141Sjeff DIP_SET(ip, di_nlink, nlink); 1330207141Sjeff ino_dirty(ino); 1331207141Sjeff} 1332207141Sjeff 1333207141Sjeff/* 1334207141Sjeff * Adjust the inode link count to 'nlink'. If the count reaches zero 1335207141Sjeff * free it. 1336207141Sjeff */ 1337207141Sjeffstatic void 1338207141Sjeffino_adjust(struct suj_ino *sino) 1339207141Sjeff{ 1340207141Sjeff struct jrefrec *rrec; 1341207141Sjeff struct suj_rec *srec; 1342207141Sjeff struct suj_ino *stmp; 1343207141Sjeff union dinode *ip; 1344207141Sjeff nlink_t nlink; 1345209716Sjeff int recmode; 1346207141Sjeff int reqlink; 1347209716Sjeff int isdot; 1348207141Sjeff int mode; 1349207141Sjeff ino_t ino; 1350207141Sjeff 1351207141Sjeff nlink = sino->si_nlink; 1352207141Sjeff ino = sino->si_ino; 1353209716Sjeff mode = sino->si_mode & IFMT; 1354207141Sjeff /* 1355209716Sjeff * If it's a directory with no dot links, it was truncated before 1356209716Sjeff * the name was cleared. We need to clear the dirent that 1357209716Sjeff * points at it. 1358209716Sjeff */ 1359209716Sjeff if (mode == IFDIR && nlink == 1 && sino->si_dotlinks == 0) { 1360209716Sjeff sino->si_nlink = nlink = 0; 1361209716Sjeff TAILQ_FOREACH(srec, &sino->si_recs, sr_next) { 1362209716Sjeff rrec = (struct jrefrec *)srec->sr_rec; 1363209716Sjeff if (ino_isat(rrec->jr_parent, rrec->jr_diroff, ino, 1364209716Sjeff &recmode, &isdot) == 0) 1365209716Sjeff continue; 1366209716Sjeff ino_clrat(rrec->jr_parent, rrec->jr_diroff, ino); 1367209716Sjeff break; 1368209716Sjeff } 1369209716Sjeff if (srec == NULL) 1370209716Sjeff errx(1, "Directory %d name not found", ino); 1371209716Sjeff } 1372209716Sjeff /* 1373207141Sjeff * If it's a directory with no real names pointing to it go ahead 1374207141Sjeff * and truncate it. This will free any children. 1375207141Sjeff */ 1376209716Sjeff if (mode == IFDIR && nlink - sino->si_dotlinks == 0) { 1377207141Sjeff sino->si_nlink = nlink = 0; 1378207141Sjeff /* 1379207141Sjeff * Mark any .. links so they know not to free this inode 1380207141Sjeff * when they are removed. 1381207141Sjeff */ 1382207141Sjeff TAILQ_FOREACH(srec, &sino->si_recs, sr_next) { 1383207141Sjeff rrec = (struct jrefrec *)srec->sr_rec; 1384207141Sjeff if (rrec->jr_diroff == DOTDOT_OFFSET) { 1385207141Sjeff stmp = ino_lookup(rrec->jr_parent, 0); 1386207141Sjeff if (stmp) 1387207141Sjeff ino_setskip(stmp, ino); 1388207141Sjeff } 1389207141Sjeff } 1390207141Sjeff } 1391207141Sjeff ip = ino_read(ino); 1392207141Sjeff mode = DIP(ip, di_mode) & IFMT; 1393207141Sjeff if (nlink > LINK_MAX) 1394209408Sdelphij err_suj( 1395209408Sdelphij "ino %d nlink manipulation error, new link %d, old link %d\n", 1396207141Sjeff ino, nlink, DIP(ip, di_nlink)); 1397207141Sjeff if (debug) 1398207141Sjeff printf("Adjusting ino %d, nlink %d, old link %d lastmode %o\n", 1399207141Sjeff ino, nlink, DIP(ip, di_nlink), sino->si_mode); 1400207141Sjeff if (mode == 0) { 1401207141Sjeff if (debug) 1402207141Sjeff printf("ino %d, zero inode freeing bitmap\n", ino); 1403207141Sjeff ino_free(ino, sino->si_mode); 1404207141Sjeff return; 1405207141Sjeff } 1406207141Sjeff /* XXX Should be an assert? */ 1407207141Sjeff if (mode != sino->si_mode && debug) 1408207141Sjeff printf("ino %d, mode %o != %o\n", ino, mode, sino->si_mode); 1409207141Sjeff if ((mode & IFMT) == IFDIR) 1410207141Sjeff reqlink = 2; 1411207141Sjeff else 1412207141Sjeff reqlink = 1; 1413207141Sjeff /* If the inode doesn't have enough links to live, free it. */ 1414207141Sjeff if (nlink < reqlink) { 1415207141Sjeff if (debug) 1416207141Sjeff printf("ino %d not enough links to live %d < %d\n", 1417207141Sjeff ino, nlink, reqlink); 1418207141Sjeff ino_reclaim(ip, ino, mode); 1419207141Sjeff return; 1420207141Sjeff } 1421207141Sjeff /* If required write the updated link count. */ 1422207141Sjeff if (DIP(ip, di_nlink) == nlink) { 1423207141Sjeff if (debug) 1424207141Sjeff printf("ino %d, link matches, skipping.\n", ino); 1425207141Sjeff return; 1426207141Sjeff } 1427207141Sjeff DIP_SET(ip, di_nlink, nlink); 1428207141Sjeff ino_dirty(ino); 1429207141Sjeff} 1430207141Sjeff 1431207141Sjeff/* 1432207141Sjeff * Truncate some or all blocks in an indirect, freeing any that are required 1433207141Sjeff * and zeroing the indirect. 1434207141Sjeff */ 1435207141Sjeffstatic void 1436207141Sjeffindir_trunc(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, ufs_lbn_t lastlbn) 1437207141Sjeff{ 1438207141Sjeff ufs2_daddr_t *bap2; 1439207141Sjeff ufs1_daddr_t *bap1; 1440207141Sjeff ufs_lbn_t lbnadd; 1441207141Sjeff ufs2_daddr_t nblk; 1442207141Sjeff ufs_lbn_t next; 1443207141Sjeff ufs_lbn_t nlbn; 1444207141Sjeff int dirty; 1445207141Sjeff int level; 1446207141Sjeff int i; 1447207141Sjeff 1448207141Sjeff if (blk == 0) 1449207141Sjeff return; 1450207141Sjeff dirty = 0; 1451207141Sjeff level = lbn_level(lbn); 1452207141Sjeff if (level == -1) 1453209408Sdelphij err_suj("Invalid level for lbn %jd\n", lbn); 1454207141Sjeff lbnadd = 1; 1455207141Sjeff for (i = level; i > 0; i--) 1456207141Sjeff lbnadd *= NINDIR(fs); 1457207141Sjeff bap1 = (void *)dblk_read(blk, fs->fs_bsize); 1458207141Sjeff bap2 = (void *)bap1; 1459207141Sjeff for (i = 0; i < NINDIR(fs); i++) { 1460207141Sjeff if (fs->fs_magic == FS_UFS1_MAGIC) 1461207141Sjeff nblk = *bap1++; 1462207141Sjeff else 1463207141Sjeff nblk = *bap2++; 1464207141Sjeff if (nblk == 0) 1465207141Sjeff continue; 1466207141Sjeff if (level != 0) { 1467207141Sjeff nlbn = (lbn + 1) - (i * lbnadd); 1468207141Sjeff /* 1469207141Sjeff * Calculate the lbn of the next indirect to 1470207141Sjeff * determine if any of this indirect must be 1471207141Sjeff * reclaimed. 1472207141Sjeff */ 1473207141Sjeff next = -(lbn + level) + ((i+1) * lbnadd); 1474207141Sjeff if (next <= lastlbn) 1475207141Sjeff continue; 1476207141Sjeff indir_trunc(ino, nlbn, nblk, lastlbn); 1477207141Sjeff /* If all of this indirect was reclaimed, free it. */ 1478207141Sjeff nlbn = next - lbnadd; 1479207141Sjeff if (nlbn < lastlbn) 1480207141Sjeff continue; 1481207141Sjeff } else { 1482207141Sjeff nlbn = -lbn + i * lbnadd; 1483207141Sjeff if (nlbn < lastlbn) 1484207141Sjeff continue; 1485207141Sjeff } 1486207141Sjeff dirty = 1; 1487207141Sjeff blk_free(nblk, 0, fs->fs_frag); 1488207141Sjeff if (fs->fs_magic == FS_UFS1_MAGIC) 1489207141Sjeff *(bap1 - 1) = 0; 1490207141Sjeff else 1491207141Sjeff *(bap2 - 1) = 0; 1492207141Sjeff } 1493207141Sjeff if (dirty) 1494207141Sjeff dblk_dirty(blk); 1495207141Sjeff} 1496207141Sjeff 1497207141Sjeff/* 1498207141Sjeff * Truncate an inode to the minimum of the given size or the last populated 1499207141Sjeff * block after any over size have been discarded. The kernel would allocate 1500207141Sjeff * the last block in the file but fsck does not and neither do we. This 1501207141Sjeff * code never extends files, only shrinks them. 1502207141Sjeff */ 1503207141Sjeffstatic void 1504207141Sjeffino_trunc(ino_t ino, off_t size) 1505207141Sjeff{ 1506207141Sjeff union dinode *ip; 1507207141Sjeff ufs2_daddr_t bn; 1508207141Sjeff uint64_t totalfrags; 1509207141Sjeff ufs_lbn_t nextlbn; 1510207141Sjeff ufs_lbn_t lastlbn; 1511207141Sjeff ufs_lbn_t tmpval; 1512207141Sjeff ufs_lbn_t lbn; 1513207141Sjeff ufs_lbn_t i; 1514207141Sjeff int frags; 1515207141Sjeff off_t cursize; 1516207141Sjeff off_t off; 1517207141Sjeff int mode; 1518207141Sjeff 1519207141Sjeff ip = ino_read(ino); 1520207141Sjeff mode = DIP(ip, di_mode) & IFMT; 1521207141Sjeff cursize = DIP(ip, di_size); 1522207141Sjeff if (debug) 1523207141Sjeff printf("Truncating ino %d, mode %o to size %jd from size %jd\n", 1524207141Sjeff ino, mode, size, cursize); 1525207141Sjeff 1526207141Sjeff /* Skip datablocks for short links and devices. */ 1527207141Sjeff if (mode == 0 || mode == IFBLK || mode == IFCHR || 1528207141Sjeff (mode == IFLNK && cursize < fs->fs_maxsymlinklen)) 1529207141Sjeff return; 1530207141Sjeff /* Don't extend. */ 1531207141Sjeff if (size > cursize) 1532207141Sjeff size = cursize; 1533207141Sjeff lastlbn = lblkno(fs, blkroundup(fs, size)); 1534207141Sjeff for (i = lastlbn; i < NDADDR; i++) { 1535207141Sjeff if (DIP(ip, di_db[i]) == 0) 1536207141Sjeff continue; 1537207141Sjeff frags = sblksize(fs, cursize, i); 1538207141Sjeff frags = numfrags(fs, frags); 1539207141Sjeff blk_free(DIP(ip, di_db[i]), 0, frags); 1540207141Sjeff DIP_SET(ip, di_db[i], 0); 1541207141Sjeff } 1542207141Sjeff /* 1543207141Sjeff * Follow indirect blocks, freeing anything required. 1544207141Sjeff */ 1545207141Sjeff for (i = 0, tmpval = NINDIR(fs), lbn = NDADDR; i < NIADDR; i++, 1546207141Sjeff lbn = nextlbn) { 1547207141Sjeff nextlbn = lbn + tmpval; 1548207141Sjeff tmpval *= NINDIR(fs); 1549207141Sjeff /* If we're not freeing any in this indirect range skip it. */ 1550207141Sjeff if (lastlbn >= nextlbn) 1551207141Sjeff continue; 1552207141Sjeff if (DIP(ip, di_ib[i]) == 0) 1553207141Sjeff continue; 1554207141Sjeff indir_trunc(ino, -lbn - i, DIP(ip, di_ib[i]), lastlbn); 1555207141Sjeff /* If we freed everything in this indirect free the indir. */ 1556207141Sjeff if (lastlbn > lbn) 1557207141Sjeff continue; 1558207141Sjeff blk_free(DIP(ip, di_ib[i]), 0, frags); 1559207141Sjeff DIP_SET(ip, di_ib[i], 0); 1560207141Sjeff } 1561207141Sjeff ino_dirty(ino); 1562207141Sjeff /* 1563207141Sjeff * Now that we've freed any whole blocks that exceed the desired 1564207141Sjeff * truncation size, figure out how many blocks remain and what the 1565207141Sjeff * last populated lbn is. We will set the size to this last lbn 1566207141Sjeff * rather than worrying about allocating the final lbn as the kernel 1567207141Sjeff * would've done. This is consistent with normal fsck behavior. 1568207141Sjeff */ 1569207141Sjeff visitlbn = 0; 1570207141Sjeff totalfrags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT); 1571207141Sjeff if (size > lblktosize(fs, visitlbn + 1)) 1572207141Sjeff size = lblktosize(fs, visitlbn + 1); 1573207141Sjeff /* 1574207141Sjeff * If we're truncating direct blocks we have to adjust frags 1575207141Sjeff * accordingly. 1576207141Sjeff */ 1577207141Sjeff if (visitlbn < NDADDR && totalfrags) { 1578207141Sjeff long oldspace, newspace; 1579207141Sjeff 1580207141Sjeff bn = DIP(ip, di_db[visitlbn]); 1581207141Sjeff if (bn == 0) 1582209408Sdelphij err_suj("Bad blk at ino %d lbn %jd\n", ino, visitlbn); 1583207141Sjeff oldspace = sblksize(fs, cursize, visitlbn); 1584207141Sjeff newspace = sblksize(fs, size, visitlbn); 1585207141Sjeff if (oldspace != newspace) { 1586207141Sjeff bn += numfrags(fs, newspace); 1587207141Sjeff frags = numfrags(fs, oldspace - newspace); 1588207141Sjeff blk_free(bn, 0, frags); 1589207141Sjeff totalfrags -= frags; 1590207141Sjeff } 1591207141Sjeff } 1592207141Sjeff DIP_SET(ip, di_blocks, fsbtodb(fs, totalfrags)); 1593207141Sjeff DIP_SET(ip, di_size, size); 1594207141Sjeff /* 1595207141Sjeff * If we've truncated into the middle of a block or frag we have 1596207141Sjeff * to zero it here. Otherwise the file could extend into 1597207141Sjeff * uninitialized space later. 1598207141Sjeff */ 1599207141Sjeff off = blkoff(fs, size); 1600207141Sjeff if (off) { 1601207141Sjeff uint8_t *buf; 1602207141Sjeff long clrsize; 1603207141Sjeff 1604207141Sjeff bn = ino_blkatoff(ip, ino, visitlbn, &frags); 1605207141Sjeff if (bn == 0) 1606209408Sdelphij err_suj("Block missing from ino %d at lbn %jd\n", 1607207141Sjeff ino, visitlbn); 1608207141Sjeff clrsize = frags * fs->fs_fsize; 1609207141Sjeff buf = dblk_read(bn, clrsize); 1610207141Sjeff clrsize -= off; 1611207141Sjeff buf += off; 1612207141Sjeff bzero(buf, clrsize); 1613207141Sjeff dblk_dirty(bn); 1614207141Sjeff } 1615207141Sjeff return; 1616207141Sjeff} 1617207141Sjeff 1618207141Sjeff/* 1619207141Sjeff * Process records available for one inode and determine whether the 1620207141Sjeff * link count is correct or needs adjusting. 1621207141Sjeff */ 1622207141Sjeffstatic void 1623207141Sjeffino_check(struct suj_ino *sino) 1624207141Sjeff{ 1625207141Sjeff struct suj_rec *srec; 1626207141Sjeff struct jrefrec *rrec; 1627207141Sjeff nlink_t dotlinks; 1628207141Sjeff int newlinks; 1629207141Sjeff int removes; 1630207141Sjeff int nlink; 1631207141Sjeff ino_t ino; 1632207141Sjeff int isdot; 1633207141Sjeff int isat; 1634207141Sjeff int mode; 1635207141Sjeff 1636207141Sjeff if (sino->si_hasrecs == 0) 1637207141Sjeff return; 1638207141Sjeff ino = sino->si_ino; 1639207141Sjeff rrec = (struct jrefrec *)TAILQ_FIRST(&sino->si_recs)->sr_rec; 1640207141Sjeff nlink = rrec->jr_nlink; 1641207141Sjeff newlinks = 0; 1642207141Sjeff dotlinks = 0; 1643207141Sjeff removes = sino->si_nlinkadj; 1644207141Sjeff TAILQ_FOREACH(srec, &sino->si_recs, sr_next) { 1645207141Sjeff rrec = (struct jrefrec *)srec->sr_rec; 1646207141Sjeff isat = ino_isat(rrec->jr_parent, rrec->jr_diroff, 1647207141Sjeff rrec->jr_ino, &mode, &isdot); 1648207141Sjeff if (isat && (mode & IFMT) != (rrec->jr_mode & IFMT)) 1649209408Sdelphij err_suj("Inode mode/directory type mismatch %o != %o\n", 1650207141Sjeff mode, rrec->jr_mode); 1651207141Sjeff if (debug) 1652207141Sjeff printf("jrefrec: op %d ino %d, nlink %d, parent %d, " 1653207141Sjeff "diroff %jd, mode %o, isat %d, isdot %d\n", 1654207141Sjeff rrec->jr_op, rrec->jr_ino, rrec->jr_nlink, 1655207141Sjeff rrec->jr_parent, rrec->jr_diroff, rrec->jr_mode, 1656207141Sjeff isat, isdot); 1657207141Sjeff mode = rrec->jr_mode & IFMT; 1658207141Sjeff if (rrec->jr_op == JOP_REMREF) 1659207141Sjeff removes++; 1660207141Sjeff newlinks += isat; 1661207141Sjeff if (isdot) 1662207141Sjeff dotlinks += isat; 1663207141Sjeff } 1664207141Sjeff /* 1665207141Sjeff * The number of links that remain are the starting link count 1666207141Sjeff * subtracted by the total number of removes with the total 1667207141Sjeff * links discovered back in. An incomplete remove thus 1668207141Sjeff * makes no change to the link count but an add increases 1669207141Sjeff * by one. 1670207141Sjeff */ 1671207141Sjeff if (debug) 1672207141Sjeff printf("ino %d nlink %d newlinks %d removes %d dotlinks %d\n", 1673207141Sjeff ino, nlink, newlinks, removes, dotlinks); 1674207141Sjeff nlink += newlinks; 1675207141Sjeff nlink -= removes; 1676207141Sjeff sino->si_linkadj = 1; 1677207141Sjeff sino->si_nlink = nlink; 1678207141Sjeff sino->si_dotlinks = dotlinks; 1679207141Sjeff sino->si_mode = mode; 1680207141Sjeff ino_adjust(sino); 1681207141Sjeff} 1682207141Sjeff 1683207141Sjeff/* 1684207141Sjeff * Process records available for one block and determine whether it is 1685207141Sjeff * still allocated and whether the owning inode needs to be updated or 1686207141Sjeff * a free completed. 1687207141Sjeff */ 1688207141Sjeffstatic void 1689207141Sjeffblk_check(struct suj_blk *sblk) 1690207141Sjeff{ 1691207141Sjeff struct suj_rec *srec; 1692207141Sjeff struct jblkrec *brec; 1693207141Sjeff struct suj_ino *sino; 1694207141Sjeff ufs2_daddr_t blk; 1695207141Sjeff int mask; 1696207141Sjeff int frags; 1697207141Sjeff int isat; 1698207141Sjeff 1699207141Sjeff /* 1700207141Sjeff * Each suj_blk actually contains records for any fragments in that 1701207141Sjeff * block. As a result we must evaluate each record individually. 1702207141Sjeff */ 1703207141Sjeff sino = NULL; 1704207141Sjeff TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) { 1705207141Sjeff brec = (struct jblkrec *)srec->sr_rec; 1706207141Sjeff frags = brec->jb_frags; 1707207141Sjeff blk = brec->jb_blkno + brec->jb_oldfrags; 1708207141Sjeff isat = blk_isat(brec->jb_ino, brec->jb_lbn, blk, &frags); 1709207141Sjeff if (sino == NULL || sino->si_ino != brec->jb_ino) { 1710207141Sjeff sino = ino_lookup(brec->jb_ino, 1); 1711207141Sjeff sino->si_blkadj = 1; 1712207141Sjeff } 1713207141Sjeff if (debug) 1714207141Sjeff printf("op %d blk %jd ino %d lbn %jd frags %d isat %d (%d)\n", 1715207141Sjeff brec->jb_op, blk, brec->jb_ino, brec->jb_lbn, 1716207141Sjeff brec->jb_frags, isat, frags); 1717207141Sjeff /* 1718207141Sjeff * If we found the block at this address we still have to 1719207141Sjeff * determine if we need to free the tail end that was 1720207141Sjeff * added by adding contiguous fragments from the same block. 1721207141Sjeff */ 1722207141Sjeff if (isat == 1) { 1723207141Sjeff if (frags == brec->jb_frags) 1724207141Sjeff continue; 1725207141Sjeff mask = blk_freemask(blk, brec->jb_ino, brec->jb_lbn, 1726207141Sjeff brec->jb_frags); 1727207141Sjeff mask >>= frags; 1728207141Sjeff blk += frags; 1729207141Sjeff frags = brec->jb_frags - frags; 1730207141Sjeff blk_free(blk, mask, frags); 1731207141Sjeff continue; 1732207141Sjeff } 1733207141Sjeff /* 1734207141Sjeff * The block wasn't found, attempt to free it. It won't be 1735207141Sjeff * freed if it was actually reallocated. If this was an 1736207141Sjeff * allocation we don't want to follow indirects as they 1737207141Sjeff * may not be written yet. Any children of the indirect will 1738207141Sjeff * have their own records. If it's a free we need to 1739207141Sjeff * recursively free children. 1740207141Sjeff */ 1741207141Sjeff blk_free_lbn(blk, brec->jb_ino, brec->jb_lbn, brec->jb_frags, 1742207141Sjeff brec->jb_op == JOP_FREEBLK); 1743207141Sjeff } 1744207141Sjeff} 1745207141Sjeff 1746207141Sjeff/* 1747207141Sjeff * Walk the list of inode records for this cg and resolve moved and duplicate 1748207141Sjeff * inode references now that we have a complete picture. 1749207141Sjeff */ 1750207141Sjeffstatic void 1751207141Sjeffcg_build(struct suj_cg *sc) 1752207141Sjeff{ 1753207141Sjeff struct suj_ino *sino; 1754207141Sjeff int i; 1755207141Sjeff 1756207141Sjeff for (i = 0; i < SUJ_HASHSIZE; i++) 1757207141Sjeff LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) 1758207141Sjeff ino_build(sino); 1759207141Sjeff} 1760207141Sjeff 1761207141Sjeff/* 1762207141Sjeff * Handle inodes requiring truncation. This must be done prior to 1763207141Sjeff * looking up any inodes in directories. 1764207141Sjeff */ 1765207141Sjeffstatic void 1766207141Sjeffcg_trunc(struct suj_cg *sc) 1767207141Sjeff{ 1768207141Sjeff struct suj_ino *sino; 1769207141Sjeff int i; 1770207141Sjeff 1771207141Sjeff for (i = 0; i < SUJ_HASHSIZE; i++) 1772207141Sjeff LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) 1773207141Sjeff if (sino->si_trunc) { 1774207141Sjeff ino_trunc(sino->si_ino, 1775207141Sjeff sino->si_trunc->jt_size); 1776207141Sjeff sino->si_trunc = NULL; 1777207141Sjeff } 1778207141Sjeff} 1779207141Sjeff 1780207141Sjeff/* 1781207141Sjeff * Free any partially allocated blocks and then resolve inode block 1782207141Sjeff * counts. 1783207141Sjeff */ 1784207141Sjeffstatic void 1785207141Sjeffcg_check_blk(struct suj_cg *sc) 1786207141Sjeff{ 1787207141Sjeff struct suj_ino *sino; 1788207141Sjeff struct suj_blk *sblk; 1789207141Sjeff int i; 1790207141Sjeff 1791207141Sjeff 1792207141Sjeff for (i = 0; i < SUJ_HASHSIZE; i++) 1793207141Sjeff LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next) 1794207141Sjeff blk_check(sblk); 1795207141Sjeff /* 1796207141Sjeff * Now that we've freed blocks which are not referenced we 1797207141Sjeff * make a second pass over all inodes to adjust their block 1798207141Sjeff * counts. 1799207141Sjeff */ 1800207141Sjeff for (i = 0; i < SUJ_HASHSIZE; i++) 1801207141Sjeff LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) 1802207141Sjeff if (sino->si_blkadj) 1803207141Sjeff ino_adjblks(sino); 1804207141Sjeff} 1805207141Sjeff 1806207141Sjeff/* 1807207141Sjeff * Walk the list of inode records for this cg, recovering any 1808207141Sjeff * changes which were not complete at the time of crash. 1809207141Sjeff */ 1810207141Sjeffstatic void 1811207141Sjeffcg_check_ino(struct suj_cg *sc) 1812207141Sjeff{ 1813207141Sjeff struct suj_ino *sino; 1814207141Sjeff int i; 1815207141Sjeff 1816207141Sjeff for (i = 0; i < SUJ_HASHSIZE; i++) 1817207141Sjeff LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) 1818207141Sjeff ino_check(sino); 1819207141Sjeff} 1820207141Sjeff 1821207141Sjeff/* 1822207141Sjeff * Write a potentially dirty cg. Recalculate the summary information and 1823207141Sjeff * update the superblock summary. 1824207141Sjeff */ 1825207141Sjeffstatic void 1826207141Sjeffcg_write(struct suj_cg *sc) 1827207141Sjeff{ 1828207141Sjeff ufs1_daddr_t fragno, cgbno, maxbno; 1829207141Sjeff u_int8_t *blksfree; 1830207141Sjeff struct cg *cgp; 1831207141Sjeff int blk; 1832207141Sjeff int i; 1833207141Sjeff 1834207141Sjeff if (sc->sc_dirty == 0) 1835207141Sjeff return; 1836207141Sjeff /* 1837207141Sjeff * Fix the frag and cluster summary. 1838207141Sjeff */ 1839207141Sjeff cgp = sc->sc_cgp; 1840207141Sjeff cgp->cg_cs.cs_nbfree = 0; 1841207141Sjeff cgp->cg_cs.cs_nffree = 0; 1842207141Sjeff bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum)); 1843207141Sjeff maxbno = fragstoblks(fs, fs->fs_fpg); 1844207141Sjeff if (fs->fs_contigsumsize > 0) { 1845207141Sjeff for (i = 1; i <= fs->fs_contigsumsize; i++) 1846207141Sjeff cg_clustersum(cgp)[i] = 0; 1847207141Sjeff bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT)); 1848207141Sjeff } 1849207141Sjeff blksfree = cg_blksfree(cgp); 1850207141Sjeff for (cgbno = 0; cgbno < maxbno; cgbno++) { 1851207141Sjeff if (ffs_isfreeblock(fs, blksfree, cgbno)) 1852207141Sjeff continue; 1853207141Sjeff if (ffs_isblock(fs, blksfree, cgbno)) { 1854207141Sjeff ffs_clusteracct(fs, cgp, cgbno, 1); 1855207141Sjeff cgp->cg_cs.cs_nbfree++; 1856207141Sjeff continue; 1857207141Sjeff } 1858207141Sjeff fragno = blkstofrags(fs, cgbno); 1859207141Sjeff blk = blkmap(fs, blksfree, fragno); 1860207141Sjeff ffs_fragacct(fs, blk, cgp->cg_frsum, 1); 1861207141Sjeff for (i = 0; i < fs->fs_frag; i++) 1862207141Sjeff if (isset(blksfree, fragno + i)) 1863207141Sjeff cgp->cg_cs.cs_nffree++; 1864207141Sjeff } 1865207141Sjeff /* 1866207141Sjeff * Update the superblock cg summary from our now correct values 1867207141Sjeff * before writing the block. 1868207141Sjeff */ 1869207141Sjeff fs->fs_cs(fs, sc->sc_cgx) = cgp->cg_cs; 1870207141Sjeff if (bwrite(disk, fsbtodb(fs, cgtod(fs, sc->sc_cgx)), sc->sc_cgbuf, 1871207141Sjeff fs->fs_bsize) == -1) 1872209408Sdelphij err_suj("Unable to write cylinder group %d\n", sc->sc_cgx); 1873207141Sjeff} 1874207141Sjeff 1875207141Sjeff/* 1876207141Sjeff * Write out any modified inodes. 1877207141Sjeff */ 1878207141Sjeffstatic void 1879207141Sjeffcg_write_inos(struct suj_cg *sc) 1880207141Sjeff{ 1881207141Sjeff struct ino_blk *iblk; 1882207141Sjeff int i; 1883207141Sjeff 1884207141Sjeff for (i = 0; i < SUJ_HASHSIZE; i++) 1885207141Sjeff LIST_FOREACH(iblk, &sc->sc_iblkhash[i], ib_next) 1886207141Sjeff if (iblk->ib_dirty) 1887207141Sjeff iblk_write(iblk); 1888207141Sjeff} 1889207141Sjeff 1890207141Sjeffstatic void 1891207141Sjeffcg_apply(void (*apply)(struct suj_cg *)) 1892207141Sjeff{ 1893207141Sjeff struct suj_cg *scg; 1894207141Sjeff int i; 1895207141Sjeff 1896207141Sjeff for (i = 0; i < SUJ_HASHSIZE; i++) 1897207141Sjeff LIST_FOREACH(scg, &cghash[i], sc_next) 1898207141Sjeff apply(scg); 1899207141Sjeff} 1900207141Sjeff 1901207141Sjeff/* 1902207141Sjeff * Process the unlinked but referenced file list. Freeing all inodes. 1903207141Sjeff */ 1904207141Sjeffstatic void 1905207141Sjeffino_unlinked(void) 1906207141Sjeff{ 1907207141Sjeff union dinode *ip; 1908207141Sjeff uint16_t mode; 1909207141Sjeff ino_t inon; 1910207141Sjeff ino_t ino; 1911207141Sjeff 1912207141Sjeff ino = fs->fs_sujfree; 1913207141Sjeff fs->fs_sujfree = 0; 1914207141Sjeff while (ino != 0) { 1915207141Sjeff ip = ino_read(ino); 1916207141Sjeff mode = DIP(ip, di_mode) & IFMT; 1917207141Sjeff inon = DIP(ip, di_freelink); 1918207141Sjeff DIP_SET(ip, di_freelink, 0); 1919207141Sjeff /* 1920207141Sjeff * XXX Should this be an errx? 1921207141Sjeff */ 1922207141Sjeff if (DIP(ip, di_nlink) == 0) { 1923207141Sjeff if (debug) 1924207141Sjeff printf("Freeing unlinked ino %d mode %o\n", 1925207141Sjeff ino, mode); 1926207141Sjeff ino_reclaim(ip, ino, mode); 1927207141Sjeff } else if (debug) 1928207141Sjeff printf("Skipping ino %d mode %o with link %d\n", 1929207141Sjeff ino, mode, DIP(ip, di_nlink)); 1930207141Sjeff ino = inon; 1931207141Sjeff } 1932207141Sjeff} 1933207141Sjeff 1934207141Sjeff/* 1935207141Sjeff * Append a new record to the list of records requiring processing. 1936207141Sjeff */ 1937207141Sjeffstatic void 1938207141Sjeffino_append(union jrec *rec) 1939207141Sjeff{ 1940207141Sjeff struct jrefrec *refrec; 1941207141Sjeff struct jmvrec *mvrec; 1942207141Sjeff struct suj_ino *sino; 1943207141Sjeff struct suj_rec *srec; 1944207141Sjeff 1945207141Sjeff mvrec = &rec->rec_jmvrec; 1946207141Sjeff refrec = &rec->rec_jrefrec; 1947207141Sjeff if (debug && mvrec->jm_op == JOP_MVREF) 1948207141Sjeff printf("ino move: ino %d, parent %d, diroff %jd, oldoff %jd\n", 1949207141Sjeff mvrec->jm_ino, mvrec->jm_parent, mvrec->jm_newoff, 1950207141Sjeff mvrec->jm_oldoff); 1951207141Sjeff else if (debug && 1952207141Sjeff (refrec->jr_op == JOP_ADDREF || refrec->jr_op == JOP_REMREF)) 1953207141Sjeff printf("ino ref: op %d, ino %d, nlink %d, " 1954207141Sjeff "parent %d, diroff %jd\n", 1955207141Sjeff refrec->jr_op, refrec->jr_ino, refrec->jr_nlink, 1956207141Sjeff refrec->jr_parent, refrec->jr_diroff); 1957207141Sjeff /* 1958207141Sjeff * Lookup the ino and clear truncate if one is found. Partial 1959207141Sjeff * truncates are always done synchronously so if we discover 1960207141Sjeff * an operation that requires a lock the truncation has completed 1961207141Sjeff * and can be discarded. 1962207141Sjeff */ 1963207141Sjeff sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1); 1964207141Sjeff sino->si_trunc = NULL; 1965207141Sjeff sino->si_hasrecs = 1; 1966207141Sjeff srec = errmalloc(sizeof(*srec)); 1967207141Sjeff srec->sr_rec = rec; 1968207141Sjeff TAILQ_INSERT_TAIL(&sino->si_newrecs, srec, sr_next); 1969207141Sjeff} 1970207141Sjeff 1971207141Sjeff/* 1972207141Sjeff * Add a reference adjustment to the sino list and eliminate dups. The 1973207141Sjeff * primary loop in ino_build_ref() checks for dups but new ones may be 1974207141Sjeff * created as a result of offset adjustments. 1975207141Sjeff */ 1976207141Sjeffstatic void 1977207141Sjeffino_add_ref(struct suj_ino *sino, struct suj_rec *srec) 1978207141Sjeff{ 1979207141Sjeff struct jrefrec *refrec; 1980207141Sjeff struct suj_rec *srn; 1981207141Sjeff struct jrefrec *rrn; 1982207141Sjeff 1983207141Sjeff refrec = (struct jrefrec *)srec->sr_rec; 1984207141Sjeff /* 1985207141Sjeff * We walk backwards so that the oldest link count is preserved. If 1986207141Sjeff * an add record conflicts with a remove keep the remove. Redundant 1987207141Sjeff * removes are eliminated in ino_build_ref. Otherwise we keep the 1988207141Sjeff * oldest record at a given location. 1989207141Sjeff */ 1990207141Sjeff for (srn = TAILQ_LAST(&sino->si_recs, srechd); srn; 1991207141Sjeff srn = TAILQ_PREV(srn, srechd, sr_next)) { 1992207141Sjeff rrn = (struct jrefrec *)srn->sr_rec; 1993207141Sjeff if (rrn->jr_parent != refrec->jr_parent || 1994207141Sjeff rrn->jr_diroff != refrec->jr_diroff) 1995207141Sjeff continue; 1996207141Sjeff if (rrn->jr_op == JOP_REMREF || refrec->jr_op == JOP_ADDREF) { 1997207141Sjeff rrn->jr_mode = refrec->jr_mode; 1998207141Sjeff return; 1999207141Sjeff } 2000207141Sjeff /* 2001207141Sjeff * Adding a remove. 2002207141Sjeff * 2003207141Sjeff * Replace the record in place with the old nlink in case 2004207141Sjeff * we replace the head of the list. Abandon srec as a dup. 2005207141Sjeff */ 2006207141Sjeff refrec->jr_nlink = rrn->jr_nlink; 2007207141Sjeff srn->sr_rec = srec->sr_rec; 2008207141Sjeff return; 2009207141Sjeff } 2010207141Sjeff TAILQ_INSERT_TAIL(&sino->si_recs, srec, sr_next); 2011207141Sjeff} 2012207141Sjeff 2013207141Sjeff/* 2014207141Sjeff * Create a duplicate of a reference at a previous location. 2015207141Sjeff */ 2016207141Sjeffstatic void 2017207141Sjeffino_dup_ref(struct suj_ino *sino, struct jrefrec *refrec, off_t diroff) 2018207141Sjeff{ 2019207141Sjeff struct jrefrec *rrn; 2020207141Sjeff struct suj_rec *srn; 2021207141Sjeff 2022207141Sjeff rrn = errmalloc(sizeof(*refrec)); 2023207141Sjeff *rrn = *refrec; 2024207141Sjeff rrn->jr_op = JOP_ADDREF; 2025207141Sjeff rrn->jr_diroff = diroff; 2026207141Sjeff srn = errmalloc(sizeof(*srn)); 2027207141Sjeff srn->sr_rec = (union jrec *)rrn; 2028207141Sjeff ino_add_ref(sino, srn); 2029207141Sjeff} 2030207141Sjeff 2031207141Sjeff/* 2032207141Sjeff * Add a reference to the list at all known locations. We follow the offset 2033207141Sjeff * changes for a single instance and create duplicate add refs at each so 2034207141Sjeff * that we can tolerate any version of the directory block. Eliminate 2035207141Sjeff * removes which collide with adds that are seen in the journal. They should 2036207141Sjeff * not adjust the link count down. 2037207141Sjeff */ 2038207141Sjeffstatic void 2039207141Sjeffino_build_ref(struct suj_ino *sino, struct suj_rec *srec) 2040207141Sjeff{ 2041207141Sjeff struct jrefrec *refrec; 2042207141Sjeff struct jmvrec *mvrec; 2043207141Sjeff struct suj_rec *srp; 2044207141Sjeff struct suj_rec *srn; 2045207141Sjeff struct jrefrec *rrn; 2046207141Sjeff off_t diroff; 2047207141Sjeff 2048207141Sjeff refrec = (struct jrefrec *)srec->sr_rec; 2049207141Sjeff /* 2050207141Sjeff * Search for a mvrec that matches this offset. Whether it's an add 2051207141Sjeff * or a remove we can delete the mvref after creating a dup record in 2052207141Sjeff * the old location. 2053207141Sjeff */ 2054207141Sjeff if (!TAILQ_EMPTY(&sino->si_movs)) { 2055207141Sjeff diroff = refrec->jr_diroff; 2056207141Sjeff for (srn = TAILQ_LAST(&sino->si_movs, srechd); srn; srn = srp) { 2057207141Sjeff srp = TAILQ_PREV(srn, srechd, sr_next); 2058207141Sjeff mvrec = (struct jmvrec *)srn->sr_rec; 2059207141Sjeff if (mvrec->jm_parent != refrec->jr_parent || 2060207141Sjeff mvrec->jm_newoff != diroff) 2061207141Sjeff continue; 2062207141Sjeff diroff = mvrec->jm_oldoff; 2063207141Sjeff TAILQ_REMOVE(&sino->si_movs, srn, sr_next); 2064209408Sdelphij free(srn); 2065207141Sjeff ino_dup_ref(sino, refrec, diroff); 2066207141Sjeff } 2067207141Sjeff } 2068207141Sjeff /* 2069207141Sjeff * If a remove wasn't eliminated by an earlier add just append it to 2070207141Sjeff * the list. 2071207141Sjeff */ 2072207141Sjeff if (refrec->jr_op == JOP_REMREF) { 2073207141Sjeff ino_add_ref(sino, srec); 2074207141Sjeff return; 2075207141Sjeff } 2076207141Sjeff /* 2077207141Sjeff * Walk the list of records waiting to be added to the list. We 2078207141Sjeff * must check for moves that apply to our current offset and remove 2079207141Sjeff * them from the list. Remove any duplicates to eliminate removes 2080207141Sjeff * with corresponding adds. 2081207141Sjeff */ 2082207141Sjeff TAILQ_FOREACH_SAFE(srn, &sino->si_newrecs, sr_next, srp) { 2083207141Sjeff switch (srn->sr_rec->rec_jrefrec.jr_op) { 2084207141Sjeff case JOP_ADDREF: 2085207141Sjeff /* 2086207141Sjeff * This should actually be an error we should 2087207141Sjeff * have a remove for every add journaled. 2088207141Sjeff */ 2089207141Sjeff rrn = (struct jrefrec *)srn->sr_rec; 2090207141Sjeff if (rrn->jr_parent != refrec->jr_parent || 2091207141Sjeff rrn->jr_diroff != refrec->jr_diroff) 2092207141Sjeff break; 2093207141Sjeff TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next); 2094207141Sjeff break; 2095207141Sjeff case JOP_REMREF: 2096207141Sjeff /* 2097207141Sjeff * Once we remove the current iteration of the 2098207141Sjeff * record at this address we're done. 2099207141Sjeff */ 2100207141Sjeff rrn = (struct jrefrec *)srn->sr_rec; 2101207141Sjeff if (rrn->jr_parent != refrec->jr_parent || 2102207141Sjeff rrn->jr_diroff != refrec->jr_diroff) 2103207141Sjeff break; 2104207141Sjeff TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next); 2105207141Sjeff ino_add_ref(sino, srec); 2106207141Sjeff return; 2107207141Sjeff case JOP_MVREF: 2108207141Sjeff /* 2109207141Sjeff * Update our diroff based on any moves that match 2110207141Sjeff * and remove the move. 2111207141Sjeff */ 2112207141Sjeff mvrec = (struct jmvrec *)srn->sr_rec; 2113207141Sjeff if (mvrec->jm_parent != refrec->jr_parent || 2114207141Sjeff mvrec->jm_oldoff != refrec->jr_diroff) 2115207141Sjeff break; 2116207141Sjeff ino_dup_ref(sino, refrec, mvrec->jm_oldoff); 2117207141Sjeff refrec->jr_diroff = mvrec->jm_newoff; 2118207141Sjeff TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next); 2119207141Sjeff break; 2120207141Sjeff default: 2121209408Sdelphij err_suj("ino_build_ref: Unknown op %d\n", 2122207141Sjeff srn->sr_rec->rec_jrefrec.jr_op); 2123207141Sjeff } 2124207141Sjeff } 2125207141Sjeff ino_add_ref(sino, srec); 2126207141Sjeff} 2127207141Sjeff 2128207141Sjeff/* 2129207141Sjeff * Walk the list of new records and add them in-order resolving any 2130207141Sjeff * dups and adjusted offsets. 2131207141Sjeff */ 2132207141Sjeffstatic void 2133207141Sjeffino_build(struct suj_ino *sino) 2134207141Sjeff{ 2135207141Sjeff struct suj_rec *srec; 2136207141Sjeff 2137207141Sjeff while ((srec = TAILQ_FIRST(&sino->si_newrecs)) != NULL) { 2138207141Sjeff TAILQ_REMOVE(&sino->si_newrecs, srec, sr_next); 2139207141Sjeff switch (srec->sr_rec->rec_jrefrec.jr_op) { 2140207141Sjeff case JOP_ADDREF: 2141207141Sjeff case JOP_REMREF: 2142207141Sjeff ino_build_ref(sino, srec); 2143207141Sjeff break; 2144207141Sjeff case JOP_MVREF: 2145207141Sjeff /* 2146207141Sjeff * Add this mvrec to the queue of pending mvs. 2147207141Sjeff */ 2148207141Sjeff TAILQ_INSERT_TAIL(&sino->si_movs, srec, sr_next); 2149207141Sjeff break; 2150207141Sjeff default: 2151209408Sdelphij err_suj("ino_build: Unknown op %d\n", 2152207141Sjeff srec->sr_rec->rec_jrefrec.jr_op); 2153207141Sjeff } 2154207141Sjeff } 2155207141Sjeff if (TAILQ_EMPTY(&sino->si_recs)) 2156207141Sjeff sino->si_hasrecs = 0; 2157207141Sjeff} 2158207141Sjeff 2159207141Sjeff/* 2160207141Sjeff * Modify journal records so they refer to the base block number 2161207141Sjeff * and a start and end frag range. This is to facilitate the discovery 2162207141Sjeff * of overlapping fragment allocations. 2163207141Sjeff */ 2164207141Sjeffstatic void 2165207141Sjeffblk_build(struct jblkrec *blkrec) 2166207141Sjeff{ 2167207141Sjeff struct suj_rec *srec; 2168207141Sjeff struct suj_blk *sblk; 2169207141Sjeff struct jblkrec *blkrn; 2170207141Sjeff struct suj_ino *sino; 2171207141Sjeff ufs2_daddr_t blk; 2172207141Sjeff off_t foff; 2173207141Sjeff int frag; 2174207141Sjeff 2175207141Sjeff if (debug) 2176207141Sjeff printf("blk_build: op %d blkno %jd frags %d oldfrags %d " 2177207141Sjeff "ino %d lbn %jd\n", 2178207141Sjeff blkrec->jb_op, blkrec->jb_blkno, blkrec->jb_frags, 2179207141Sjeff blkrec->jb_oldfrags, blkrec->jb_ino, blkrec->jb_lbn); 2180207141Sjeff 2181207141Sjeff /* 2182207141Sjeff * Look up the inode and clear the truncate if any lbns after the 2183207141Sjeff * truncate lbn are freed or allocated. 2184207141Sjeff */ 2185207141Sjeff sino = ino_lookup(blkrec->jb_ino, 0); 2186207141Sjeff if (sino && sino->si_trunc) { 2187207141Sjeff foff = lblktosize(fs, blkrec->jb_lbn); 2188207141Sjeff foff += lfragtosize(fs, blkrec->jb_frags); 2189207141Sjeff if (foff > sino->si_trunc->jt_size) 2190207141Sjeff sino->si_trunc = NULL; 2191207141Sjeff } 2192207141Sjeff blk = blknum(fs, blkrec->jb_blkno); 2193207141Sjeff frag = fragnum(fs, blkrec->jb_blkno); 2194207141Sjeff sblk = blk_lookup(blk, 1); 2195207141Sjeff /* 2196207141Sjeff * Rewrite the record using oldfrags to indicate the offset into 2197207141Sjeff * the block. Leave jb_frags as the actual allocated count. 2198207141Sjeff */ 2199207141Sjeff blkrec->jb_blkno -= frag; 2200207141Sjeff blkrec->jb_oldfrags = frag; 2201207141Sjeff if (blkrec->jb_oldfrags + blkrec->jb_frags > fs->fs_frag) 2202209408Sdelphij err_suj("Invalid fragment count %d oldfrags %d\n", 2203207141Sjeff blkrec->jb_frags, frag); 2204207141Sjeff /* 2205207141Sjeff * Detect dups. If we detect a dup we always discard the oldest 2206207141Sjeff * record as it is superseded by the new record. This speeds up 2207207141Sjeff * later stages but also eliminates free records which are used 2208207141Sjeff * to indicate that the contents of indirects can be trusted. 2209207141Sjeff */ 2210207141Sjeff TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) { 2211207141Sjeff blkrn = (struct jblkrec *)srec->sr_rec; 2212207141Sjeff if (blkrn->jb_ino != blkrec->jb_ino || 2213207141Sjeff blkrn->jb_lbn != blkrec->jb_lbn || 2214207141Sjeff blkrn->jb_blkno != blkrec->jb_blkno || 2215207141Sjeff blkrn->jb_frags != blkrec->jb_frags || 2216207141Sjeff blkrn->jb_oldfrags != blkrec->jb_oldfrags) 2217207141Sjeff continue; 2218207141Sjeff if (debug) 2219207141Sjeff printf("Removed dup.\n"); 2220207141Sjeff /* Discard the free which is a dup with an alloc. */ 2221207141Sjeff if (blkrec->jb_op == JOP_FREEBLK) 2222207141Sjeff return; 2223207141Sjeff TAILQ_REMOVE(&sblk->sb_recs, srec, sr_next); 2224207141Sjeff free(srec); 2225207141Sjeff break; 2226207141Sjeff } 2227207141Sjeff srec = errmalloc(sizeof(*srec)); 2228207141Sjeff srec->sr_rec = (union jrec *)blkrec; 2229207141Sjeff TAILQ_INSERT_TAIL(&sblk->sb_recs, srec, sr_next); 2230207141Sjeff} 2231207141Sjeff 2232207141Sjeffstatic void 2233207141Sjeffino_build_trunc(struct jtrncrec *rec) 2234207141Sjeff{ 2235207141Sjeff struct suj_ino *sino; 2236207141Sjeff 2237207141Sjeff if (debug) 2238207141Sjeff printf("ino_build_trunc: ino %d, size %jd\n", 2239207141Sjeff rec->jt_ino, rec->jt_size); 2240207141Sjeff sino = ino_lookup(rec->jt_ino, 1); 2241207141Sjeff sino->si_trunc = rec; 2242207141Sjeff} 2243207141Sjeff 2244207141Sjeff/* 2245207141Sjeff * Build up tables of the operations we need to recover. 2246207141Sjeff */ 2247207141Sjeffstatic void 2248207141Sjeffsuj_build(void) 2249207141Sjeff{ 2250207141Sjeff struct suj_seg *seg; 2251207141Sjeff union jrec *rec; 2252207141Sjeff int off; 2253207141Sjeff int i; 2254207141Sjeff 2255207141Sjeff TAILQ_FOREACH(seg, &allsegs, ss_next) { 2256207141Sjeff if (debug) 2257207141Sjeff printf("seg %jd has %d records, oldseq %jd.\n", 2258207141Sjeff seg->ss_rec.jsr_seq, seg->ss_rec.jsr_cnt, 2259207141Sjeff seg->ss_rec.jsr_oldest); 2260207141Sjeff off = 0; 2261207141Sjeff rec = (union jrec *)seg->ss_blk; 2262207141Sjeff for (i = 0; i < seg->ss_rec.jsr_cnt; off += JREC_SIZE, rec++) { 2263207141Sjeff /* skip the segrec. */ 2264207141Sjeff if ((off % DEV_BSIZE) == 0) 2265207141Sjeff continue; 2266207141Sjeff switch (rec->rec_jrefrec.jr_op) { 2267207141Sjeff case JOP_ADDREF: 2268207141Sjeff case JOP_REMREF: 2269207141Sjeff case JOP_MVREF: 2270207141Sjeff ino_append(rec); 2271207141Sjeff break; 2272207141Sjeff case JOP_NEWBLK: 2273207141Sjeff case JOP_FREEBLK: 2274207141Sjeff blk_build((struct jblkrec *)rec); 2275207141Sjeff break; 2276207141Sjeff case JOP_TRUNC: 2277207141Sjeff ino_build_trunc((struct jtrncrec *)rec); 2278207141Sjeff break; 2279207141Sjeff default: 2280209408Sdelphij err_suj("Unknown journal operation %d (%d)\n", 2281207141Sjeff rec->rec_jrefrec.jr_op, off); 2282207141Sjeff } 2283207141Sjeff i++; 2284207141Sjeff } 2285207141Sjeff } 2286207141Sjeff} 2287207141Sjeff 2288207141Sjeff/* 2289207141Sjeff * Prune the journal segments to those we care about based on the 2290207141Sjeff * oldest sequence in the newest segment. Order the segment list 2291207141Sjeff * based on sequence number. 2292207141Sjeff */ 2293207141Sjeffstatic void 2294207141Sjeffsuj_prune(void) 2295207141Sjeff{ 2296207141Sjeff struct suj_seg *seg; 2297207141Sjeff struct suj_seg *segn; 2298207141Sjeff uint64_t newseq; 2299207141Sjeff int discard; 2300207141Sjeff 2301207141Sjeff if (debug) 2302207141Sjeff printf("Pruning up to %jd\n", oldseq); 2303207141Sjeff /* First free the expired segments. */ 2304207141Sjeff TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) { 2305207141Sjeff if (seg->ss_rec.jsr_seq >= oldseq) 2306207141Sjeff continue; 2307207141Sjeff TAILQ_REMOVE(&allsegs, seg, ss_next); 2308207141Sjeff free(seg->ss_blk); 2309207141Sjeff free(seg); 2310207141Sjeff } 2311207141Sjeff /* Next ensure that segments are ordered properly. */ 2312207141Sjeff seg = TAILQ_FIRST(&allsegs); 2313207141Sjeff if (seg == NULL) { 2314207141Sjeff if (debug) 2315207141Sjeff printf("Empty journal\n"); 2316207141Sjeff return; 2317207141Sjeff } 2318207141Sjeff newseq = seg->ss_rec.jsr_seq; 2319207141Sjeff for (;;) { 2320207141Sjeff seg = TAILQ_LAST(&allsegs, seghd); 2321207141Sjeff if (seg->ss_rec.jsr_seq >= newseq) 2322207141Sjeff break; 2323207141Sjeff TAILQ_REMOVE(&allsegs, seg, ss_next); 2324207141Sjeff TAILQ_INSERT_HEAD(&allsegs, seg, ss_next); 2325207141Sjeff newseq = seg->ss_rec.jsr_seq; 2326207141Sjeff 2327207141Sjeff } 2328209408Sdelphij if (newseq != oldseq) { 2329209408Sdelphij err_suj("Journal file sequence mismatch %jd != %jd\n", 2330207141Sjeff newseq, oldseq); 2331209408Sdelphij } 2332207141Sjeff /* 2333207141Sjeff * The kernel may asynchronously write segments which can create 2334207141Sjeff * gaps in the sequence space. Throw away any segments after the 2335207141Sjeff * gap as the kernel guarantees only those that are contiguously 2336207141Sjeff * reachable are marked as completed. 2337207141Sjeff */ 2338207141Sjeff discard = 0; 2339207141Sjeff TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) { 2340207141Sjeff if (!discard && newseq++ == seg->ss_rec.jsr_seq) { 2341207141Sjeff jrecs += seg->ss_rec.jsr_cnt; 2342207141Sjeff jbytes += seg->ss_rec.jsr_blocks * DEV_BSIZE; 2343207141Sjeff continue; 2344207141Sjeff } 2345207141Sjeff discard = 1; 2346207141Sjeff if (debug) 2347207141Sjeff printf("Journal order mismatch %jd != %jd pruning\n", 2348207141Sjeff newseq-1, seg->ss_rec.jsr_seq); 2349207141Sjeff TAILQ_REMOVE(&allsegs, seg, ss_next); 2350207141Sjeff free(seg->ss_blk); 2351207141Sjeff free(seg); 2352207141Sjeff } 2353207141Sjeff if (debug) 2354207141Sjeff printf("Processing journal segments from %jd to %jd\n", 2355207141Sjeff oldseq, newseq-1); 2356207141Sjeff} 2357207141Sjeff 2358207141Sjeff/* 2359207141Sjeff * Verify the journal inode before attempting to read records. 2360207141Sjeff */ 2361207141Sjeffstatic int 2362207141Sjeffsuj_verifyino(union dinode *ip) 2363207141Sjeff{ 2364207141Sjeff 2365207141Sjeff if (DIP(ip, di_nlink) != 1) { 2366207141Sjeff printf("Invalid link count %d for journal inode %d\n", 2367207141Sjeff DIP(ip, di_nlink), sujino); 2368207141Sjeff return (-1); 2369207141Sjeff } 2370207141Sjeff 2371207141Sjeff if ((DIP(ip, di_flags) & (SF_IMMUTABLE | SF_NOUNLINK)) != 2372207141Sjeff (SF_IMMUTABLE | SF_NOUNLINK)) { 2373207141Sjeff printf("Invalid flags 0x%X for journal inode %d\n", 2374207141Sjeff DIP(ip, di_flags), sujino); 2375207141Sjeff return (-1); 2376207141Sjeff } 2377207141Sjeff 2378207141Sjeff if (DIP(ip, di_mode) != (IFREG | IREAD)) { 2379207141Sjeff printf("Invalid mode %o for journal inode %d\n", 2380207141Sjeff DIP(ip, di_mode), sujino); 2381207141Sjeff return (-1); 2382207141Sjeff } 2383207141Sjeff 2384207141Sjeff if (DIP(ip, di_size) < SUJ_MIN || DIP(ip, di_size) > SUJ_MAX) { 2385207141Sjeff printf("Invalid size %jd for journal inode %d\n", 2386207141Sjeff DIP(ip, di_size), sujino); 2387207141Sjeff return (-1); 2388207141Sjeff } 2389207141Sjeff 2390207141Sjeff if (DIP(ip, di_modrev) != fs->fs_mtime) { 2391207141Sjeff printf("Journal timestamp does not match fs mount time\n"); 2392207141Sjeff return (-1); 2393207141Sjeff } 2394207141Sjeff 2395207141Sjeff return (0); 2396207141Sjeff} 2397207141Sjeff 2398207141Sjeffstruct jblocks { 2399207141Sjeff struct jextent *jb_extent; /* Extent array. */ 2400207141Sjeff int jb_avail; /* Available extents. */ 2401207141Sjeff int jb_used; /* Last used extent. */ 2402207141Sjeff int jb_head; /* Allocator head. */ 2403207141Sjeff int jb_off; /* Allocator extent offset. */ 2404207141Sjeff}; 2405207141Sjeffstruct jextent { 2406207141Sjeff ufs2_daddr_t je_daddr; /* Disk block address. */ 2407207141Sjeff int je_blocks; /* Disk block count. */ 2408207141Sjeff}; 2409207141Sjeff 2410207141Sjeffstruct jblocks *suj_jblocks; 2411207141Sjeff 2412207141Sjeffstatic struct jblocks * 2413207141Sjeffjblocks_create(void) 2414207141Sjeff{ 2415207141Sjeff struct jblocks *jblocks; 2416207141Sjeff int size; 2417207141Sjeff 2418207141Sjeff jblocks = errmalloc(sizeof(*jblocks)); 2419207141Sjeff jblocks->jb_avail = 10; 2420207141Sjeff jblocks->jb_used = 0; 2421207141Sjeff jblocks->jb_head = 0; 2422207141Sjeff jblocks->jb_off = 0; 2423207141Sjeff size = sizeof(struct jextent) * jblocks->jb_avail; 2424207141Sjeff jblocks->jb_extent = errmalloc(size); 2425207141Sjeff bzero(jblocks->jb_extent, size); 2426207141Sjeff 2427207141Sjeff return (jblocks); 2428207141Sjeff} 2429207141Sjeff 2430207141Sjeff/* 2431207141Sjeff * Return the next available disk block and the amount of contiguous 2432207141Sjeff * free space it contains. 2433207141Sjeff */ 2434207141Sjeffstatic ufs2_daddr_t 2435207141Sjeffjblocks_next(struct jblocks *jblocks, int bytes, int *actual) 2436207141Sjeff{ 2437207141Sjeff struct jextent *jext; 2438207141Sjeff ufs2_daddr_t daddr; 2439207141Sjeff int freecnt; 2440207141Sjeff int blocks; 2441207141Sjeff 2442207141Sjeff blocks = bytes / DEV_BSIZE; 2443207141Sjeff jext = &jblocks->jb_extent[jblocks->jb_head]; 2444207141Sjeff freecnt = jext->je_blocks - jblocks->jb_off; 2445207141Sjeff if (freecnt == 0) { 2446207141Sjeff jblocks->jb_off = 0; 2447207141Sjeff if (++jblocks->jb_head > jblocks->jb_used) 2448207141Sjeff return (0); 2449207141Sjeff jext = &jblocks->jb_extent[jblocks->jb_head]; 2450207141Sjeff freecnt = jext->je_blocks; 2451207141Sjeff } 2452207141Sjeff if (freecnt > blocks) 2453207141Sjeff freecnt = blocks; 2454207141Sjeff *actual = freecnt * DEV_BSIZE; 2455207141Sjeff daddr = jext->je_daddr + jblocks->jb_off; 2456207141Sjeff 2457207141Sjeff return (daddr); 2458207141Sjeff} 2459207141Sjeff 2460207141Sjeff/* 2461207141Sjeff * Advance the allocation head by a specified number of bytes, consuming 2462207141Sjeff * one journal segment. 2463207141Sjeff */ 2464207141Sjeffstatic void 2465207141Sjeffjblocks_advance(struct jblocks *jblocks, int bytes) 2466207141Sjeff{ 2467207141Sjeff 2468207141Sjeff jblocks->jb_off += bytes / DEV_BSIZE; 2469207141Sjeff} 2470207141Sjeff 2471207141Sjeffstatic void 2472207141Sjeffjblocks_destroy(struct jblocks *jblocks) 2473207141Sjeff{ 2474207141Sjeff 2475207141Sjeff free(jblocks->jb_extent); 2476207141Sjeff free(jblocks); 2477207141Sjeff} 2478207141Sjeff 2479207141Sjeffstatic void 2480207141Sjeffjblocks_add(struct jblocks *jblocks, ufs2_daddr_t daddr, int blocks) 2481207141Sjeff{ 2482207141Sjeff struct jextent *jext; 2483207141Sjeff int size; 2484207141Sjeff 2485207141Sjeff jext = &jblocks->jb_extent[jblocks->jb_used]; 2486207141Sjeff /* Adding the first block. */ 2487207141Sjeff if (jext->je_daddr == 0) { 2488207141Sjeff jext->je_daddr = daddr; 2489207141Sjeff jext->je_blocks = blocks; 2490207141Sjeff return; 2491207141Sjeff } 2492207141Sjeff /* Extending the last extent. */ 2493207141Sjeff if (jext->je_daddr + jext->je_blocks == daddr) { 2494207141Sjeff jext->je_blocks += blocks; 2495207141Sjeff return; 2496207141Sjeff } 2497207141Sjeff /* Adding a new extent. */ 2498207141Sjeff if (++jblocks->jb_used == jblocks->jb_avail) { 2499207141Sjeff jblocks->jb_avail *= 2; 2500207141Sjeff size = sizeof(struct jextent) * jblocks->jb_avail; 2501207141Sjeff jext = errmalloc(size); 2502207141Sjeff bzero(jext, size); 2503207141Sjeff bcopy(jblocks->jb_extent, jext, 2504207141Sjeff sizeof(struct jextent) * jblocks->jb_used); 2505207141Sjeff free(jblocks->jb_extent); 2506207141Sjeff jblocks->jb_extent = jext; 2507207141Sjeff } 2508207141Sjeff jext = &jblocks->jb_extent[jblocks->jb_used]; 2509207141Sjeff jext->je_daddr = daddr; 2510207141Sjeff jext->je_blocks = blocks; 2511207141Sjeff 2512207141Sjeff return; 2513207141Sjeff} 2514207141Sjeff 2515207141Sjeff/* 2516207141Sjeff * Add a file block from the journal to the extent map. We can't read 2517207141Sjeff * each file block individually because the kernel treats it as a circular 2518207141Sjeff * buffer and segments may span mutliple contiguous blocks. 2519207141Sjeff */ 2520207141Sjeffstatic void 2521207141Sjeffsuj_add_block(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 2522207141Sjeff{ 2523207141Sjeff 2524207141Sjeff jblocks_add(suj_jblocks, fsbtodb(fs, blk), fsbtodb(fs, frags)); 2525207141Sjeff} 2526207141Sjeff 2527207141Sjeffstatic void 2528207141Sjeffsuj_read(void) 2529207141Sjeff{ 2530207141Sjeff uint8_t block[1 * 1024 * 1024]; 2531207141Sjeff struct suj_seg *seg; 2532207141Sjeff struct jsegrec *recn; 2533207141Sjeff struct jsegrec *rec; 2534207141Sjeff ufs2_daddr_t blk; 2535207141Sjeff int readsize; 2536207141Sjeff int blocks; 2537207141Sjeff int recsize; 2538207141Sjeff int size; 2539207141Sjeff int i; 2540207141Sjeff 2541207141Sjeff /* 2542207141Sjeff * Read records until we exhaust the journal space. If we find 2543207141Sjeff * an invalid record we start searching for a valid segment header 2544207141Sjeff * at the next block. This is because we don't have a head/tail 2545207141Sjeff * pointer and must recover the information indirectly. At the gap 2546207141Sjeff * between the head and tail we won't necessarily have a valid 2547207141Sjeff * segment. 2548207141Sjeff */ 2549207141Sjeffrestart: 2550207141Sjeff for (;;) { 2551207141Sjeff size = sizeof(block); 2552207141Sjeff blk = jblocks_next(suj_jblocks, size, &readsize); 2553207141Sjeff if (blk == 0) 2554207141Sjeff return; 2555207141Sjeff size = readsize; 2556207141Sjeff /* 2557207141Sjeff * Read 1MB at a time and scan for records within this block. 2558207141Sjeff */ 2559209408Sdelphij if (bread(disk, blk, &block, size) == -1) { 2560209408Sdelphij err_suj("Error reading journal block %jd\n", 2561207141Sjeff (intmax_t)blk); 2562209408Sdelphij } 2563207141Sjeff for (rec = (void *)block; size; size -= recsize, 2564207141Sjeff rec = (struct jsegrec *)((uintptr_t)rec + recsize)) { 2565207141Sjeff recsize = DEV_BSIZE; 2566207141Sjeff if (rec->jsr_time != fs->fs_mtime) { 2567207141Sjeff if (debug) 2568207141Sjeff printf("Rec time %jd != fs mtime %jd\n", 2569207141Sjeff rec->jsr_time, fs->fs_mtime); 2570207141Sjeff jblocks_advance(suj_jblocks, recsize); 2571207141Sjeff continue; 2572207141Sjeff } 2573207141Sjeff if (rec->jsr_cnt == 0) { 2574207141Sjeff if (debug) 2575207141Sjeff printf("Found illegal count %d\n", 2576207141Sjeff rec->jsr_cnt); 2577207141Sjeff jblocks_advance(suj_jblocks, recsize); 2578207141Sjeff continue; 2579207141Sjeff } 2580207141Sjeff blocks = rec->jsr_blocks; 2581207141Sjeff recsize = blocks * DEV_BSIZE; 2582207141Sjeff if (recsize > size) { 2583207141Sjeff /* 2584207141Sjeff * We may just have run out of buffer, restart 2585207141Sjeff * the loop to re-read from this spot. 2586207141Sjeff */ 2587207141Sjeff if (size < fs->fs_bsize && 2588207141Sjeff size != readsize && 2589207141Sjeff recsize <= fs->fs_bsize) 2590207141Sjeff goto restart; 2591207141Sjeff if (debug) 2592207141Sjeff printf("Found invalid segsize %d > %d\n", 2593207141Sjeff recsize, size); 2594207141Sjeff recsize = DEV_BSIZE; 2595207141Sjeff jblocks_advance(suj_jblocks, recsize); 2596207141Sjeff continue; 2597207141Sjeff } 2598207141Sjeff /* 2599207141Sjeff * Verify that all blocks in the segment are present. 2600207141Sjeff */ 2601207141Sjeff for (i = 1; i < blocks; i++) { 2602207141Sjeff recn = (void *) 2603207141Sjeff ((uintptr_t)rec) + i * DEV_BSIZE; 2604207141Sjeff if (recn->jsr_seq == rec->jsr_seq && 2605207141Sjeff recn->jsr_time == rec->jsr_time) 2606207141Sjeff continue; 2607207141Sjeff if (debug) 2608207141Sjeff printf("Incomplete record %jd (%d)\n", 2609207141Sjeff rec->jsr_seq, i); 2610207141Sjeff recsize = i * DEV_BSIZE; 2611207141Sjeff jblocks_advance(suj_jblocks, recsize); 2612207141Sjeff goto restart; 2613207141Sjeff } 2614207141Sjeff seg = errmalloc(sizeof(*seg)); 2615207141Sjeff seg->ss_blk = errmalloc(recsize); 2616207141Sjeff seg->ss_rec = *rec; 2617207141Sjeff bcopy((void *)rec, seg->ss_blk, recsize); 2618207141Sjeff if (rec->jsr_oldest > oldseq) 2619207141Sjeff oldseq = rec->jsr_oldest; 2620207141Sjeff TAILQ_INSERT_TAIL(&allsegs, seg, ss_next); 2621207141Sjeff jblocks_advance(suj_jblocks, recsize); 2622207141Sjeff } 2623207141Sjeff } 2624207141Sjeff} 2625207141Sjeff 2626207141Sjeff/* 2627207141Sjeff * Search a directory block for the SUJ_FILE. 2628207141Sjeff */ 2629207141Sjeffstatic void 2630207141Sjeffsuj_find(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 2631207141Sjeff{ 2632207141Sjeff char block[MAXBSIZE]; 2633207141Sjeff struct direct *dp; 2634207141Sjeff int bytes; 2635207141Sjeff int off; 2636207141Sjeff 2637207141Sjeff if (sujino) 2638207141Sjeff return; 2639207141Sjeff bytes = lfragtosize(fs, frags); 2640207141Sjeff if (bread(disk, fsbtodb(fs, blk), block, bytes) <= 0) 2641209408Sdelphij err_suj("Failed to read ROOTINO directory block %jd\n", blk); 2642207141Sjeff for (off = 0; off < bytes; off += dp->d_reclen) { 2643207141Sjeff dp = (struct direct *)&block[off]; 2644207141Sjeff if (dp->d_reclen == 0) 2645207141Sjeff break; 2646207141Sjeff if (dp->d_ino == 0) 2647207141Sjeff continue; 2648207141Sjeff if (dp->d_namlen != strlen(SUJ_FILE)) 2649207141Sjeff continue; 2650207141Sjeff if (bcmp(dp->d_name, SUJ_FILE, dp->d_namlen) != 0) 2651207141Sjeff continue; 2652207141Sjeff sujino = dp->d_ino; 2653207141Sjeff return; 2654207141Sjeff } 2655207141Sjeff} 2656207141Sjeff 2657207141Sjeff/* 2658207141Sjeff * Orchestrate the verification of a filesystem via the softupdates journal. 2659207141Sjeff */ 2660207141Sjeffint 2661207141Sjeffsuj_check(const char *filesys) 2662207141Sjeff{ 2663207141Sjeff union dinode *jip; 2664207141Sjeff union dinode *ip; 2665207141Sjeff uint64_t blocks; 2666209408Sdelphij int retval; 2667209408Sdelphij struct suj_seg *seg; 2668209408Sdelphij struct suj_seg *segn; 2669207141Sjeff 2670207141Sjeff opendisk(filesys); 2671207141Sjeff TAILQ_INIT(&allsegs); 2672209408Sdelphij 2673207141Sjeff /* 2674209408Sdelphij * Set an exit point when SUJ check failed 2675209408Sdelphij */ 2676209408Sdelphij retval = setjmp(jmpbuf); 2677209408Sdelphij if (retval != 0) { 2678209408Sdelphij pwarn("UNEXPECTED SU+J INCONSISTENCY\n"); 2679209408Sdelphij TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) { 2680209408Sdelphij TAILQ_REMOVE(&allsegs, seg, ss_next); 2681209408Sdelphij free(seg->ss_blk); 2682209408Sdelphij free(seg); 2683209408Sdelphij } 2684209408Sdelphij if (reply("FALLBACK TO FULL FSCK") == 0) { 2685209408Sdelphij ckfini(0); 2686209408Sdelphij exit(EEXIT); 2687209408Sdelphij } else 2688209408Sdelphij return (-1); 2689209408Sdelphij } 2690209408Sdelphij 2691209408Sdelphij /* 2692207141Sjeff * Find the journal inode. 2693207141Sjeff */ 2694207141Sjeff ip = ino_read(ROOTINO); 2695207141Sjeff sujino = 0; 2696207141Sjeff ino_visit(ip, ROOTINO, suj_find, 0); 2697209408Sdelphij if (sujino == 0) { 2698209408Sdelphij printf("Journal inode removed. Use tunefs to re-create.\n"); 2699209408Sdelphij sblock.fs_flags &= ~FS_SUJ; 2700209408Sdelphij sblock.fs_sujfree = 0; 2701209408Sdelphij return (-1); 2702209408Sdelphij } 2703207141Sjeff /* 2704207141Sjeff * Fetch the journal inode and verify it. 2705207141Sjeff */ 2706207141Sjeff jip = ino_read(sujino); 2707207141Sjeff printf("** SU+J Recovering %s\n", filesys); 2708207141Sjeff if (suj_verifyino(jip) != 0) 2709207141Sjeff return (-1); 2710207141Sjeff /* 2711207141Sjeff * Build a list of journal blocks in jblocks before parsing the 2712207141Sjeff * available journal blocks in with suj_read(). 2713207141Sjeff */ 2714207141Sjeff printf("** Reading %jd byte journal from inode %d.\n", 2715207141Sjeff DIP(jip, di_size), sujino); 2716207141Sjeff suj_jblocks = jblocks_create(); 2717207141Sjeff blocks = ino_visit(jip, sujino, suj_add_block, 0); 2718209408Sdelphij if (blocks != numfrags(fs, DIP(jip, di_size))) { 2719209408Sdelphij printf("Sparse journal inode %d.\n", sujino); 2720209408Sdelphij return (-1); 2721209408Sdelphij } 2722207141Sjeff suj_read(); 2723207141Sjeff jblocks_destroy(suj_jblocks); 2724207141Sjeff suj_jblocks = NULL; 2725207141Sjeff if (preen || reply("RECOVER")) { 2726207141Sjeff printf("** Building recovery table.\n"); 2727207141Sjeff suj_prune(); 2728207141Sjeff suj_build(); 2729207141Sjeff cg_apply(cg_build); 2730207141Sjeff printf("** Resolving unreferenced inode list.\n"); 2731207141Sjeff ino_unlinked(); 2732207141Sjeff printf("** Processing journal entries.\n"); 2733207141Sjeff cg_apply(cg_trunc); 2734207141Sjeff cg_apply(cg_check_blk); 2735207141Sjeff cg_apply(cg_check_ino); 2736207141Sjeff } 2737209408Sdelphij if (preen == 0 && (jrecs > 0 || jbytes > 0) && reply("WRITE CHANGES") == 0) 2738207141Sjeff return (0); 2739207141Sjeff /* 2740207141Sjeff * To remain idempotent with partial truncations the free bitmaps 2741207141Sjeff * must be written followed by indirect blocks and lastly inode 2742207141Sjeff * blocks. This preserves access to the modified pointers until 2743207141Sjeff * they are freed. 2744207141Sjeff */ 2745207141Sjeff cg_apply(cg_write); 2746207141Sjeff dblk_write(); 2747207141Sjeff cg_apply(cg_write_inos); 2748207141Sjeff /* Write back superblock. */ 2749207141Sjeff closedisk(filesys); 2750209408Sdelphij if (jrecs > 0 || jbytes > 0) { 2751209408Sdelphij printf("** %jd journal records in %jd bytes for %.2f%% utilization\n", 2752209408Sdelphij jrecs, jbytes, ((float)jrecs / (float)(jbytes / JREC_SIZE)) * 100); 2753209408Sdelphij printf("** Freed %jd inodes (%jd dirs) %jd blocks, and %jd frags.\n", 2754209408Sdelphij freeinos, freedir, freeblocks, freefrags); 2755209408Sdelphij } 2756207141Sjeff 2757207141Sjeff return (0); 2758207141Sjeff} 2759