1 /* $NetBSD: lfs_vfsops.c,v 1.83 2002/11/24 08:37:43 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant@hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /*- 39 * Copyright (c) 1989, 1991, 1993, 1994 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. All advertising materials mentioning features or use of this software 51 * must display the following acknowledgement: 52 * This product includes software developed by the University of 53 * California, Berkeley and its contributors. 54 * 4. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * @(#)lfs_vfsops.c 8.20 (Berkeley) 6/10/95 71 */ 72 73 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.83 2002/11/24 08:37:43 yamt Exp $"); 75 76 #if defined(_KERNEL_OPT) 77 #include "opt_quota.h" 78 #endif 79 80 #include <sys/param.h> 81 #include <sys/systm.h> 82 #include <sys/namei.h> 83 #include <sys/proc.h> 84 #include <sys/kernel.h> 85 #include <sys/vnode.h> 86 #include <sys/mount.h> 87 #include <sys/buf.h> 88 #include <sys/device.h> 89 #include <sys/mbuf.h> 90 #include <sys/file.h> 91 #include <sys/disklabel.h> 92 #include <sys/ioctl.h> 93 #include <sys/errno.h> 94 #include <sys/malloc.h> 95 #include <sys/pool.h> 96 #include <sys/socket.h> 97 #include <uvm/uvm_extern.h> 98 #include <sys/sysctl.h> 99 #include <sys/conf.h> 100 101 #include <miscfs/specfs/specdev.h> 102 103 #include <ufs/ufs/quota.h> 104 #include <ufs/ufs/inode.h> 105 #include <ufs/ufs/ufsmount.h> 106 #include <ufs/ufs/ufs_extern.h> 107 108 #include <ufs/lfs/lfs.h> 109 #include <ufs/lfs/lfs_extern.h> 110 111 int lfs_mountfs(struct vnode *, struct mount *, struct proc *); 112 113 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc; 114 extern const struct vnodeopv_desc lfs_specop_opv_desc; 115 extern const struct vnodeopv_desc lfs_fifoop_opv_desc; 116 117 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = { 118 &lfs_vnodeop_opv_desc, 119 &lfs_specop_opv_desc, 120 &lfs_fifoop_opv_desc, 121 NULL, 122 }; 123 124 struct vfsops lfs_vfsops = { 125 MOUNT_LFS, 126 lfs_mount, 127 ufs_start, 128 lfs_unmount, 129 ufs_root, 130 ufs_quotactl, 131 lfs_statfs, 132 lfs_sync, 133 lfs_vget, 134 lfs_fhtovp, 135 lfs_vptofh, 136 lfs_init, 137 lfs_reinit, 138 lfs_done, 139 lfs_sysctl, 140 lfs_mountroot, 141 ufs_check_export, 142 lfs_vnodeopv_descs, 143 }; 144 145 struct genfs_ops lfs_genfsops = { 146 NULL, 147 NULL, 148 genfs_compat_gop_write, 149 }; 150 151 struct pool lfs_inode_pool; 152 153 extern int locked_queue_count; 154 extern long locked_queue_bytes; 155 156 /* 157 * Initialize the filesystem, most work done by ufs_init. 158 */ 159 void 160 lfs_init() 161 { 162 ufs_init(); 163 164 /* 165 * XXX Same structure as FFS inodes? Should we share a common pool? 166 */ 167 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 168 "lfsinopl", &pool_allocator_nointr); 169 #ifdef DEBUG 170 memset(lfs_log, 0, sizeof(lfs_log)); 171 #endif 172 } 173 174 void 175 lfs_reinit() 176 { 177 ufs_reinit(); 178 } 179 180 void 181 lfs_done() 182 { 183 ufs_done(); 184 pool_destroy(&lfs_inode_pool); 185 } 186 187 /* 188 * Called by main() when ufs is going to be mounted as root. 189 */ 190 int 191 lfs_mountroot() 192 { 193 extern struct vnode *rootvp; 194 struct mount *mp; 195 struct proc *p = curproc; /* XXX */ 196 int error; 197 198 if (root_device->dv_class != DV_DISK) 199 return (ENODEV); 200 201 if (rootdev == NODEV) 202 return (ENODEV); 203 /* 204 * Get vnodes for swapdev and rootdev. 205 */ 206 if ((error = bdevvp(rootdev, &rootvp))) { 207 printf("lfs_mountroot: can't setup bdevvp's"); 208 return (error); 209 } 210 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 211 vrele(rootvp); 212 return (error); 213 } 214 if ((error = lfs_mountfs(rootvp, mp, p))) { 215 mp->mnt_op->vfs_refcount--; 216 vfs_unbusy(mp); 217 free(mp, M_MOUNT); 218 vrele(rootvp); 219 return (error); 220 } 221 simple_lock(&mountlist_slock); 222 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 223 simple_unlock(&mountlist_slock); 224 (void)lfs_statfs(mp, &mp->mnt_stat, p); 225 vfs_unbusy(mp); 226 inittodr(VFSTOUFS(mp)->um_lfs->lfs_tstamp); 227 return (0); 228 } 229 230 /* 231 * VFS Operations. 232 * 233 * mount system call 234 */ 235 int 236 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct proc *p) 237 { 238 struct vnode *devvp; 239 struct ufs_args args; 240 struct ufsmount *ump = NULL; 241 struct lfs *fs = NULL; /* LFS */ 242 size_t size; 243 int error; 244 mode_t accessmode; 245 246 if (mp->mnt_flag & MNT_GETARGS) { 247 ump = VFSTOUFS(mp); 248 if (ump == NULL) 249 return EIO; 250 args.fspec = NULL; 251 vfs_showexport(mp, &args.export, &ump->um_export); 252 return copyout(&args, data, sizeof(args)); 253 } 254 error = copyin(data, (caddr_t)&args, sizeof (struct ufs_args)); 255 if (error) 256 return (error); 257 258 #if 0 259 /* Until LFS can do NFS right. XXX */ 260 if (args.export.ex_flags & MNT_EXPORTED) 261 return (EINVAL); 262 #endif 263 264 /* 265 * If updating, check whether changing from read-only to 266 * read/write; if there is no device name, that's all we do. 267 */ 268 if (mp->mnt_flag & MNT_UPDATE) { 269 ump = VFSTOUFS(mp); 270 fs = ump->um_lfs; 271 if (fs->lfs_ronly && (mp->mnt_flag & MNT_WANTRDWR)) { 272 /* 273 * If upgrade to read-write by non-root, then verify 274 * that user has necessary permissions on the device. 275 */ 276 if (p->p_ucred->cr_uid != 0) { 277 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 278 error = VOP_ACCESS(ump->um_devvp, VREAD|VWRITE, 279 p->p_ucred, p); 280 VOP_UNLOCK(ump->um_devvp, 0); 281 if (error) 282 return (error); 283 } 284 fs->lfs_ronly = 0; 285 } 286 if (args.fspec == 0) { 287 /* 288 * Process export requests. 289 */ 290 return (vfs_export(mp, &ump->um_export, &args.export)); 291 } 292 } 293 /* 294 * Not an update, or updating the name: look up the name 295 * and verify that it refers to a sensible block device. 296 */ 297 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p); 298 if ((error = namei(ndp)) != 0) 299 return (error); 300 devvp = ndp->ni_vp; 301 if (devvp->v_type != VBLK) { 302 vrele(devvp); 303 return (ENOTBLK); 304 } 305 if (bdevsw_lookup(devvp->v_rdev) == NULL) { 306 vrele(devvp); 307 return (ENXIO); 308 } 309 /* 310 * If mount by non-root, then verify that user has necessary 311 * permissions on the device. 312 */ 313 if (p->p_ucred->cr_uid != 0) { 314 accessmode = VREAD; 315 if ((mp->mnt_flag & MNT_RDONLY) == 0) 316 accessmode |= VWRITE; 317 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 318 error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p); 319 if (error) { 320 vput(devvp); 321 return (error); 322 } 323 VOP_UNLOCK(devvp, 0); 324 } 325 if ((mp->mnt_flag & MNT_UPDATE) == 0) 326 error = lfs_mountfs(devvp, mp, p); /* LFS */ 327 else { 328 if (devvp != ump->um_devvp) 329 error = EINVAL; /* needs translation */ 330 else 331 vrele(devvp); 332 } 333 if (error) { 334 vrele(devvp); 335 return (error); 336 } 337 ump = VFSTOUFS(mp); 338 fs = ump->um_lfs; /* LFS */ 339 (void)copyinstr(path, fs->lfs_fsmnt, sizeof(fs->lfs_fsmnt) - 1, &size); 340 bzero(fs->lfs_fsmnt + size, sizeof(fs->lfs_fsmnt) - size); 341 bcopy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname, MNAMELEN); 342 (void) copyinstr(args.fspec, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 343 &size); 344 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size); 345 return (0); 346 } 347 348 /* 349 * Roll-forward code. 350 */ 351 352 /* 353 * Load the appropriate indirect block, and change the appropriate pointer. 354 * Mark the block dirty. Do segment and avail accounting. 355 */ 356 static int 357 update_meta(struct lfs *fs, ino_t ino, int version, ufs_daddr_t lbn, 358 daddr_t ndaddr, size_t size, struct proc *p) 359 { 360 int error; 361 struct vnode *vp; 362 struct inode *ip; 363 daddr_t odaddr, ooff; 364 struct indir a[NIADDR], *ap; 365 struct buf *bp; 366 SEGUSE *sup; 367 int num; 368 369 if ((error = lfs_rf_valloc(fs, ino, version, p, &vp)) != 0) { 370 #ifdef DEBUG_LFS_RFW 371 printf("update_meta: ino %d: lfs_rf_valloc returned %d\n", ino, 372 error); 373 #endif 374 return error; 375 } 376 377 if ((error = VOP_BALLOC(vp, (lbn << fs->lfs_bshift), size, 378 NOCRED, 0, &bp)) != 0) { 379 vput(vp); 380 return (error); 381 } 382 /* No need to write, the block is already on disk */ 383 if (bp->b_flags & B_DELWRI) { 384 LFS_UNLOCK_BUF(bp); 385 fs->lfs_avail += btofsb(fs, bp->b_bcount); 386 } 387 bp->b_flags |= B_INVAL; 388 brelse(bp); 389 390 /* 391 * Extend the file, if it is not large enough already. 392 * XXX this is not exactly right, we don't know how much of the 393 * XXX last block is actually used. We hope that an inode will 394 * XXX appear later to give the correct size. 395 */ 396 ip = VTOI(vp); 397 if (ip->i_ffs_size <= (lbn << fs->lfs_bshift)) { 398 if (lbn < NDADDR) 399 ip->i_ffs_size = (lbn << fs->lfs_bshift) + 400 (size - fs->lfs_fsize) + 1; 401 else 402 ip->i_ffs_size = (lbn << fs->lfs_bshift) + 1; 403 } 404 405 error = ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL); 406 if (error) { 407 #ifdef DEBUG_LFS_RFW 408 printf("update_meta: ufs_bmaparray returned %d\n", error); 409 #endif 410 vput(vp); 411 return error; 412 } 413 switch (num) { 414 case 0: 415 ooff = ip->i_ffs_db[lbn]; 416 if (ooff == UNWRITTEN) 417 ip->i_ffs_blocks += btofsb(fs, size); 418 /* XXX what about fragment extension? */ 419 ip->i_ffs_db[lbn] = ndaddr; 420 break; 421 case 1: 422 ooff = ip->i_ffs_ib[a[0].in_off]; 423 if (ooff == UNWRITTEN) 424 ip->i_ffs_blocks += btofsb(fs, size); 425 ip->i_ffs_ib[a[0].in_off] = ndaddr; 426 break; 427 default: 428 ap = &a[num - 1]; 429 if (bread(vp, ap->in_lbn, fs->lfs_bsize, NOCRED, &bp)) 430 panic("update_meta: bread bno %d", ap->in_lbn); 431 432 ooff = ((ufs_daddr_t *)bp->b_data)[ap->in_off]; 433 if (ooff == UNWRITTEN) 434 ip->i_ffs_blocks += btofsb(fs, size); 435 ((ufs_daddr_t *)bp->b_data)[ap->in_off] = ndaddr; 436 (void) VOP_BWRITE(bp); 437 } 438 LFS_SET_UINO(ip, IN_CHANGE | IN_MODIFIED | IN_UPDATE); 439 440 /* Update segment usage information. */ 441 if (odaddr > 0) { 442 LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, odaddr)), bp); 443 #ifdef DIAGNOSTIC 444 if (sup->su_nbytes < size) { 445 panic("update_meta: negative bytes " 446 "(segment %d short by %ld)\n", 447 dtosn(fs, dbtofsb(fs, odaddr)), (long)size - sup->su_nbytes); 448 sup->su_nbytes = size; 449 } 450 #endif 451 sup->su_nbytes -= size; 452 LFS_BWRITE_LOG(bp); 453 } 454 LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 455 sup->su_nbytes += size; 456 LFS_BWRITE_LOG(bp); 457 458 /* Fix this so it can be released */ 459 /* ip->i_lfs_effnblks = ip->i_ffs_blocks; */ 460 461 #ifdef DEBUG_LFS_RFW 462 /* Now look again to make sure it worked */ 463 ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL ); 464 if (dbtofsb(fs, odaddr) != ndaddr) 465 printf("update_meta: failed setting ino %d lbn %d to %x\n", 466 ino, lbn, ndaddr); 467 #endif 468 vput(vp); 469 return 0; 470 } 471 472 static int 473 update_inoblk(struct lfs *fs, daddr_t offset, struct ucred *cred, 474 struct proc *p) 475 { 476 struct vnode *devvp, *vp; 477 struct inode *ip; 478 struct dinode *dip; 479 struct buf *dbp, *ibp; 480 int error; 481 daddr_t daddr; 482 IFILE *ifp; 483 SEGUSE *sup; 484 485 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 486 487 /* 488 * Get the inode, update times and perms. 489 * DO NOT update disk blocks, we do that separately. 490 */ 491 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp); 492 if (error) { 493 #ifdef DEBUG_LFS_RFW 494 printf("update_inoblk: bread returned %d\n", error); 495 #endif 496 return error; 497 } 498 dip = ((struct dinode *)(dbp->b_data)) + INOPB(fs); 499 while (--dip >= (struct dinode *)dbp->b_data) { 500 if (dip->di_inumber > LFS_IFILE_INUM) { 501 /* printf("ino %d version %d\n", dip->di_inumber, 502 dip->di_gen); */ 503 error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen, 504 p, &vp); 505 if (error) { 506 #ifdef DEBUG_LFS_RFW 507 printf("update_inoblk: lfs_rf_valloc returned %d\n", error); 508 #endif 509 continue; 510 } 511 ip = VTOI(vp); 512 if (dip->di_size != ip->i_ffs_size) 513 VOP_TRUNCATE(vp, dip->di_size, 0, NOCRED, p); 514 /* Get mode, link count, size, and times */ 515 memcpy(&ip->i_din.ffs_din, dip, 516 offsetof(struct dinode, di_db[0])); 517 518 /* Then the rest, except di_blocks */ 519 ip->i_ffs_flags = dip->di_flags; 520 ip->i_ffs_gen = dip->di_gen; 521 ip->i_ffs_uid = dip->di_uid; 522 ip->i_ffs_gid = dip->di_gid; 523 524 ip->i_ffs_effnlink = dip->di_nlink; 525 526 LFS_SET_UINO(ip, IN_CHANGE | IN_MODIFIED | IN_UPDATE); 527 528 /* Re-initialize to get type right */ 529 ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, 530 &vp); 531 vput(vp); 532 533 /* Record change in location */ 534 LFS_IENTRY(ifp, fs, dip->di_inumber, ibp); 535 daddr = ifp->if_daddr; 536 ifp->if_daddr = dbtofsb(fs, dbp->b_blkno); 537 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 538 /* And do segment accounting */ 539 if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) { 540 if (daddr > 0) { 541 LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), 542 ibp); 543 sup->su_nbytes -= DINODE_SIZE; 544 LFS_BWRITE_LOG(ibp); 545 } 546 LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 547 ibp); 548 sup->su_nbytes += DINODE_SIZE; 549 LFS_BWRITE_LOG(ibp); 550 } 551 } 552 } 553 dbp->b_flags |= B_AGE; 554 brelse(dbp); 555 556 return 0; 557 } 558 559 #define CHECK_CKSUM 0x0001 /* Check the checksum to make sure it's valid */ 560 #define CHECK_UPDATE 0x0002 /* Update Ifile for new data blocks / inodes */ 561 562 static daddr_t 563 check_segsum(struct lfs *fs, daddr_t offset, 564 struct ucred *cred, int flags, int *pseg_flags, struct proc *p) 565 { 566 struct vnode *devvp; 567 struct buf *bp, *dbp; 568 int error, nblocks, ninos, i, j; 569 SEGSUM *ssp; 570 u_long *dp, *datap; /* XXX u_int32_t */ 571 daddr_t *iaddr, oldoffset; 572 FINFO *fip; 573 SEGUSE *sup; 574 size_t size; 575 u_int64_t serial; 576 577 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 578 /* 579 * If the segment has a superblock and we're at the top 580 * of the segment, skip the superblock. 581 */ 582 if (sntod(fs, dtosn(fs, offset)) == offset) { 583 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 584 if (sup->su_flags & SEGUSE_SUPERBLOCK) 585 offset += btofsb(fs, LFS_SBPAD); 586 brelse(bp); 587 } 588 589 /* Read in the segment summary */ 590 error = bread(devvp, offset, fs->lfs_sumsize, cred, &bp); 591 if (error) 592 return -1; 593 594 /* Check summary checksum */ 595 ssp = (SEGSUM *)bp->b_data; 596 if (flags & CHECK_CKSUM) { 597 if (ssp->ss_sumsum != cksum(&ssp->ss_datasum, 598 fs->lfs_sumsize - 599 sizeof(ssp->ss_sumsum))) { 600 #ifdef DEBUG_LFS_RFW 601 printf("Sumsum error at 0x%x\n", offset); 602 #endif 603 offset = -1; 604 goto err1; 605 } 606 if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) { 607 #ifdef DEBUG_LFS_RFW 608 printf("Empty pseg at 0x%x\n", offset); 609 #endif 610 offset = -1; 611 goto err1; 612 } 613 if (ssp->ss_create < fs->lfs_tstamp) { 614 #ifdef DEBUG_LFS_RFW 615 printf("Old data at 0x%x\n", offset); 616 #endif 617 offset = -1; 618 goto err1; 619 } 620 } 621 if (fs->lfs_version > 1) { 622 serial = ssp->ss_serial; 623 if (serial != fs->lfs_serial + 1) { 624 #ifdef DEBUG_LFS_RFW 625 printf("Unexpected serial number at 0x%x\n", offset); 626 #endif 627 offset = -1; 628 goto err1; 629 } 630 if (ssp->ss_ident != fs->lfs_ident) { 631 #ifdef DEBUG_LFS_RFW 632 printf("Incorrect fsid (0x%x vs 0x%x) at 0x%x\n", 633 ssp->ss_ident, fs->lfs_ident, offset); 634 #endif 635 offset = -1; 636 goto err1; 637 } 638 } 639 if (pseg_flags) 640 *pseg_flags = ssp->ss_flags; 641 oldoffset = offset; 642 offset += btofsb(fs, fs->lfs_sumsize); 643 644 ninos = howmany(ssp->ss_ninos, INOPB(fs)); 645 iaddr = (daddr_t *)(bp->b_data + fs->lfs_sumsize - sizeof(daddr_t)); 646 if (flags & CHECK_CKSUM) { 647 /* Count blocks */ 648 nblocks = 0; 649 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs)); 650 for (i = 0; i < ssp->ss_nfinfo; ++i) { 651 nblocks += fip->fi_nblocks; 652 if (fip->fi_nblocks <= 0) 653 break; 654 fip = (FINFO *)(((char *)fip) + sizeof(FINFO) + 655 (fip->fi_nblocks - 1) * 656 sizeof(ufs_daddr_t)); 657 } 658 nblocks += ninos; 659 /* Create the sum array */ 660 datap = dp = (u_long *)malloc(nblocks * sizeof(u_long), 661 M_SEGMENT, M_WAITOK); 662 } 663 664 /* Handle individual blocks */ 665 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs)); 666 for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) { 667 /* Inode block? */ 668 if (ninos && *iaddr == offset) { 669 if (flags & CHECK_CKSUM) { 670 /* Read in the head and add to the buffer */ 671 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize, 672 cred, &dbp); 673 if (error) { 674 offset = -1; 675 goto err2; 676 } 677 (*dp++) = ((u_long *)(dbp->b_data))[0]; 678 dbp->b_flags |= B_AGE; 679 brelse(dbp); 680 } 681 if (flags & CHECK_UPDATE) { 682 if ((error = update_inoblk(fs, offset, cred, p)) 683 != 0) { 684 offset = -1; 685 goto err2; 686 } 687 } 688 offset += btofsb(fs, fs->lfs_ibsize); 689 --iaddr; 690 --ninos; 691 --i; /* compensate */ 692 continue; 693 } 694 /* printf("check: blocks from ino %d version %d\n", 695 fip->fi_ino, fip->fi_version); */ 696 size = fs->lfs_bsize; 697 for (j = 0; j < fip->fi_nblocks; ++j) { 698 if (j == fip->fi_nblocks - 1) 699 size = fip->fi_lastlength; 700 if (flags & CHECK_CKSUM) { 701 error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp); 702 if (error) { 703 offset = -1; 704 goto err2; 705 } 706 (*dp++) = ((u_long *)(dbp->b_data))[0]; 707 dbp->b_flags |= B_AGE; 708 brelse(dbp); 709 } 710 /* Account for and update any direct blocks */ 711 if ((flags & CHECK_UPDATE) && 712 fip->fi_ino > LFS_IFILE_INUM && 713 fip->fi_blocks[j] >= 0) { 714 update_meta(fs, fip->fi_ino, fip->fi_version, 715 fip->fi_blocks[j], offset, size, p); 716 } 717 offset += btofsb(fs, size); 718 } 719 fip = (FINFO *)(((char *)fip) + sizeof(FINFO) 720 + (fip->fi_nblocks - 1) * sizeof(ufs_daddr_t)); 721 } 722 /* Checksum the array, compare */ 723 if ((flags & CHECK_CKSUM) && 724 ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long))) 725 { 726 #ifdef DEBUG_LFS_RFW 727 printf("Datasum error at 0x%x (wanted %x got %x)\n", offset, 728 ssp->ss_datasum, cksum(datap, nblocks * 729 sizeof(u_long))); 730 #endif 731 offset = -1; 732 goto err2; 733 } 734 735 /* If we're at the end of the segment, move to the next */ 736 if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) != 737 dtosn(fs, offset)) { 738 if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) { 739 offset = -1; 740 goto err2; 741 } 742 offset = ssp->ss_next; 743 #ifdef DEBUG_LFS_RFW 744 printf("LFS roll forward: moving on to offset 0x%x " 745 " -> segment %d\n", offset, dtosn(fs,offset)); 746 #endif 747 } 748 749 if (flags & CHECK_UPDATE) { 750 fs->lfs_avail -= (offset - oldoffset); 751 /* Don't clog the buffer queue */ 752 if (locked_queue_count > LFS_MAX_BUFS || 753 locked_queue_bytes > LFS_MAX_BYTES) { 754 ++fs->lfs_writer; 755 lfs_flush(fs, SEGM_CKP); 756 if (--fs->lfs_writer == 0) 757 wakeup(&fs->lfs_dirops); 758 } 759 } 760 761 err2: 762 if (flags & CHECK_CKSUM) 763 free(datap, M_SEGMENT); 764 err1: 765 bp->b_flags |= B_AGE; 766 brelse(bp); 767 768 /* XXX should we update the serial number even for bad psegs? */ 769 if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1) 770 fs->lfs_serial = serial; 771 return offset; 772 } 773 774 /* 775 * Common code for mount and mountroot 776 * LFS specific 777 */ 778 int 779 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct proc *p) 780 { 781 extern struct vnode *rootvp; 782 struct dlfs *tdfs, *dfs, *adfs; 783 struct lfs *fs; 784 struct ufsmount *ump; 785 struct vnode *vp; 786 struct buf *bp, *abp; 787 struct partinfo dpart; 788 dev_t dev; 789 int error, i, ronly, secsize, fsbsize; 790 struct ucred *cred; 791 CLEANERINFO *cip; 792 SEGUSE *sup; 793 int flags, dirty, do_rollforward; 794 daddr_t offset, oldoffset, lastgoodpseg, sb_addr; 795 int sn, curseg; 796 797 cred = p ? p->p_ucred : NOCRED; 798 /* 799 * Disallow multiple mounts of the same device. 800 * Disallow mounting of a device that is currently in use 801 * (except for root, which might share swap device for miniroot). 802 * Flush out any old buffers remaining from a previous use. 803 */ 804 if ((error = vfs_mountedon(devvp)) != 0) 805 return (error); 806 if (vcount(devvp) > 1 && devvp != rootvp) 807 return (EBUSY); 808 if ((error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0)) != 0) 809 return (error); 810 811 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 812 error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p); 813 if (error) 814 return (error); 815 if (VOP_IOCTL(devvp, DIOCGPART, (caddr_t)&dpart, FREAD, cred, p) != 0) 816 secsize = DEV_BSIZE; 817 else 818 secsize = dpart.disklab->d_secsize; 819 820 /* Don't free random space on error. */ 821 bp = NULL; 822 abp = NULL; 823 ump = NULL; 824 825 sb_addr = LFS_LABELPAD / secsize; 826 while (1) { 827 /* Read in the superblock. */ 828 error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp); 829 if (error) 830 goto out; 831 dfs = (struct dlfs *)bp->b_data; 832 833 /* Check the basics. */ 834 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize >= MAXBSIZE || 835 dfs->dlfs_version > LFS_VERSION || 836 dfs->dlfs_bsize < sizeof(struct dlfs)) { 837 #ifdef DEBUG_LFS 838 printf("lfs_mountfs: primary superblock sanity failed\n"); 839 #endif 840 error = EINVAL; /* XXX needs translation */ 841 goto out; 842 } 843 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) 844 printf("lfs_mountfs: warning: unknown inode format %d\n", 845 dfs->dlfs_inodefmt); 846 847 if (dfs->dlfs_version == 1) 848 fsbsize = secsize; 849 else { 850 fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb + 851 dfs->dlfs_fsbtodb); 852 /* 853 * Could be, if the frag size is large enough, that we 854 * don't have the "real" primary superblock. If that's 855 * the case, get the real one, and try again. 856 */ 857 if (sb_addr != dfs->dlfs_sboffs[0] << 858 dfs->dlfs_fsbtodb) { 859 /* #ifdef DEBUG_LFS */ 860 printf("lfs_mountfs: sb daddr 0x%x is not right, trying 0x%x\n", 861 sb_addr, dfs->dlfs_sboffs[0] << 862 dfs->dlfs_fsbtodb); 863 /* #endif */ 864 sb_addr = dfs->dlfs_sboffs[0] << 865 dfs->dlfs_fsbtodb; 866 brelse(bp); 867 continue; 868 } 869 } 870 break; 871 } 872 873 /* 874 * Check the second superblock to see which is newer; then mount 875 * using the older of the two. This is necessary to ensure that 876 * the filesystem is valid if it was not unmounted cleanly. 877 */ 878 879 if (dfs->dlfs_sboffs[1] && 880 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 881 { 882 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize), 883 LFS_SBPAD, cred, &abp); 884 if (error) 885 goto out; 886 adfs = (struct dlfs *)abp->b_data; 887 888 if (dfs->dlfs_version == 1) { 889 /* 1s resolution comparison */ 890 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 891 tdfs = adfs; 892 else 893 tdfs = dfs; 894 } else { 895 /* monotonic infinite-resolution comparison */ 896 if (adfs->dlfs_serial < dfs->dlfs_serial) 897 tdfs = adfs; 898 else 899 tdfs = dfs; 900 } 901 902 /* Check the basics. */ 903 if (tdfs->dlfs_magic != LFS_MAGIC || 904 tdfs->dlfs_bsize > MAXBSIZE || 905 tdfs->dlfs_version > LFS_VERSION || 906 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 907 #ifdef DEBUG_LFS 908 printf("lfs_mountfs: alt superblock sanity failed\n"); 909 #endif 910 error = EINVAL; /* XXX needs translation */ 911 goto out; 912 } 913 } else { 914 #ifdef DEBUG_LFS 915 printf("lfs_mountfs: invalid alt superblock daddr=0x%x\n", 916 dfs->dlfs_sboffs[1]); 917 #endif 918 error = EINVAL; 919 goto out; 920 } 921 922 /* Allocate the mount structure, copy the superblock into it. */ 923 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK); 924 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 925 926 /* Compatibility */ 927 if (fs->lfs_version < 2) { 928 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 929 fs->lfs_ibsize = fs->lfs_bsize; 930 fs->lfs_start = fs->lfs_sboffs[0]; 931 fs->lfs_tstamp = fs->lfs_otstamp; 932 fs->lfs_fsbtodb = 0; 933 } 934 935 /* Before rolling forward, lock so vget will sleep for other procs */ 936 fs->lfs_flags = LFS_NOTYET; 937 fs->lfs_rfpid = p->p_pid; 938 939 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK); 940 memset((caddr_t)ump, 0, sizeof *ump); 941 ump->um_lfs = fs; 942 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 943 bp->b_flags |= B_INVAL; 944 abp->b_flags |= B_INVAL; 945 } 946 brelse(bp); 947 bp = NULL; 948 brelse(abp); 949 abp = NULL; 950 951 /* Set up the I/O information */ 952 fs->lfs_devbsize = secsize; 953 fs->lfs_iocount = 0; 954 fs->lfs_diropwait = 0; 955 fs->lfs_activesb = 0; 956 fs->lfs_uinodes = 0; 957 fs->lfs_ravail = 0; 958 fs->lfs_sbactive = 0; 959 960 /* Set up the ifile and lock aflags */ 961 fs->lfs_doifile = 0; 962 fs->lfs_writer = 0; 963 fs->lfs_dirops = 0; 964 fs->lfs_nadirop = 0; 965 fs->lfs_seglock = 0; 966 lockinit(&fs->lfs_freelock, PINOD, "lfs_freelock", 0, 0); 967 lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0); 968 969 /* Set the file system readonly/modify bits. */ 970 fs->lfs_ronly = ronly; 971 if (ronly == 0) 972 fs->lfs_fmod = 1; 973 974 /* Initialize the mount structure. */ 975 dev = devvp->v_rdev; 976 mp->mnt_data = ump; 977 mp->mnt_stat.f_fsid.val[0] = (long)dev; 978 mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_LFS); 979 mp->mnt_stat.f_iosize = fs->lfs_bsize; 980 mp->mnt_maxsymlinklen = fs->lfs_maxsymlinklen; 981 mp->mnt_flag |= MNT_LOCAL; 982 ump->um_flags = 0; 983 ump->um_mountp = mp; 984 ump->um_dev = dev; 985 ump->um_devvp = devvp; 986 ump->um_bptrtodb = fs->lfs_fsbtodb; 987 ump->um_seqinc = fragstofsb(fs, fs->lfs_frag); 988 ump->um_nindir = fs->lfs_nindir; 989 ump->um_lognindir = ffs(fs->lfs_nindir) - 1; 990 for (i = 0; i < MAXQUOTAS; i++) 991 ump->um_quotas[i] = NULLVP; 992 devvp->v_specmountpoint = mp; 993 994 /* 995 * We use the ifile vnode for almost every operation. Instead of 996 * retrieving it from the hash table each time we retrieve it here, 997 * artificially increment the reference count and keep a pointer 998 * to it in the incore copy of the superblock. 999 */ 1000 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) { 1001 #ifdef DEBUG 1002 printf("lfs_mountfs: ifile vget failed, error=%d\n", error); 1003 #endif 1004 goto out; 1005 } 1006 fs->lfs_ivnode = vp; 1007 VREF(vp); 1008 1009 /* 1010 * Roll forward. 1011 * 1012 * We don't automatically roll forward for v1 filesystems, because 1013 * of the danger that the clock was turned back between the last 1014 * checkpoint and crash. This would roll forward garbage. 1015 * 1016 * v2 filesystems don't have this problem because they use a 1017 * monotonically increasing serial number instead of a timestamp. 1018 */ 1019 #ifdef LFS_DO_ROLLFORWARD 1020 do_rollforward = !fs->lfs_ronly; 1021 #else 1022 do_rollforward = (fs->lfs_version > 1 && !fs->lfs_ronly && 1023 !(fs->lfs_pflags & LFS_PF_CLEAN)); 1024 #endif 1025 if (do_rollforward) { 1026 /* 1027 * Phase I: Find the address of the last good partial 1028 * segment that was written after the checkpoint. Mark 1029 * the segments in question dirty, so they won't be 1030 * reallocated. 1031 */ 1032 lastgoodpseg = oldoffset = offset = fs->lfs_offset; 1033 flags = 0x0; 1034 #ifdef DEBUG_LFS_RFW 1035 printf("LFS roll forward phase 1: starting at offset 0x%x\n", 1036 offset); 1037 #endif 1038 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 1039 if (!(sup->su_flags & SEGUSE_DIRTY)) 1040 --fs->lfs_nclean; 1041 sup->su_flags |= SEGUSE_DIRTY; 1042 (void) LFS_BWRITE_LOG(bp); 1043 while ((offset = check_segsum(fs, offset, cred, CHECK_CKSUM, 1044 &flags, p)) > 0) 1045 { 1046 if (sntod(fs, oldoffset) != sntod(fs, offset)) { 1047 LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset), 1048 bp); 1049 if (!(sup->su_flags & SEGUSE_DIRTY)) 1050 --fs->lfs_nclean; 1051 sup->su_flags |= SEGUSE_DIRTY; 1052 (void) LFS_BWRITE_LOG(bp); 1053 } 1054 1055 #ifdef DEBUG_LFS_RFW 1056 printf("LFS roll forward phase 1: offset=0x%x\n", 1057 offset); 1058 if (flags & SS_DIROP) { 1059 printf("lfs_mountfs: dirops at 0x%x\n", 1060 oldoffset); 1061 if (!(flags & SS_CONT)) 1062 printf("lfs_mountfs: dirops end " 1063 "at 0x%x\n", oldoffset); 1064 } 1065 #endif 1066 if (!(flags & SS_CONT)) 1067 lastgoodpseg = offset; 1068 oldoffset = offset; 1069 } 1070 #ifdef DEBUG_LFS_RFW 1071 if (flags & SS_CONT) { 1072 printf("LFS roll forward: warning: incomplete " 1073 "dirops discarded\n"); 1074 } 1075 printf("LFS roll forward phase 1: completed: " 1076 "lastgoodpseg=0x%x\n", lastgoodpseg); 1077 #endif 1078 oldoffset = fs->lfs_offset; 1079 if (fs->lfs_offset != lastgoodpseg) { 1080 /* Don't overwrite what we're trying to preserve */ 1081 offset = fs->lfs_offset; 1082 fs->lfs_offset = lastgoodpseg; 1083 fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset)); 1084 for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) { 1085 sn = (sn + 1) % fs->lfs_nseg; 1086 if (sn == curseg) 1087 panic("lfs_mountfs: no clean segments"); 1088 LFS_SEGENTRY(sup, fs, sn, bp); 1089 dirty = (sup->su_flags & SEGUSE_DIRTY); 1090 brelse(bp); 1091 if (!dirty) 1092 break; 1093 } 1094 fs->lfs_nextseg = sntod(fs, sn); 1095 1096 /* 1097 * Phase II: Roll forward from the first superblock. 1098 */ 1099 while (offset != lastgoodpseg) { 1100 #ifdef DEBUG_LFS_RFW 1101 printf("LFS roll forward phase 2: 0x%x\n", 1102 offset); 1103 #endif 1104 offset = check_segsum(fs, offset, cred, 1105 CHECK_UPDATE, NULL, p); 1106 } 1107 1108 /* 1109 * Finish: flush our changes to disk. 1110 */ 1111 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1112 printf("lfs_mountfs: roll forward recovered %d blocks\n", 1113 lastgoodpseg - oldoffset); 1114 } 1115 #ifdef DEBUG_LFS_RFW 1116 printf("LFS roll forward complete\n"); 1117 #endif 1118 } 1119 /* If writing, sb is not clean; record in case of immediate crash */ 1120 if (!fs->lfs_ronly) { 1121 fs->lfs_pflags &= ~LFS_PF_CLEAN; 1122 lfs_writesuper(fs, fs->lfs_sboffs[0]); 1123 } 1124 1125 /* Allow vget now that roll-forward is complete */ 1126 fs->lfs_flags &= ~(LFS_NOTYET); 1127 wakeup(&fs->lfs_flags); 1128 1129 /* 1130 * Initialize the ifile cleaner info with information from 1131 * the superblock. 1132 */ 1133 LFS_CLEANERINFO(cip, fs, bp); 1134 cip->clean = fs->lfs_nclean; 1135 cip->dirty = fs->lfs_nseg - fs->lfs_nclean; 1136 cip->avail = fs->lfs_avail; 1137 cip->bfree = fs->lfs_bfree; 1138 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1139 1140 /* 1141 * Mark the current segment as ACTIVE, since we're going to 1142 * be writing to it. 1143 */ 1144 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); 1145 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1146 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1147 1148 /* Now that roll-forward is done, unlock the Ifile */ 1149 vput(vp); 1150 1151 /* Comment on ifile size if it is too large */ 1152 if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS) { 1153 fs->lfs_flags |= LFS_WARNED; 1154 printf("lfs_mountfs: please consider increasing NBUF to at least %lld\n", 1155 (long long)(fs->lfs_ivnode->v_size / fs->lfs_bsize) * (nbuf / LFS_MAX_BUFS)); 1156 } 1157 if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES) { 1158 fs->lfs_flags |= LFS_WARNED; 1159 printf("lfs_mountfs: please consider increasing BUFPAGES to at least %lld\n", 1160 (long long)fs->lfs_ivnode->v_size * bufpages / LFS_MAX_BYTES); 1161 } 1162 1163 return (0); 1164 out: 1165 if (bp) 1166 brelse(bp); 1167 if (abp) 1168 brelse(abp); 1169 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1170 (void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p); 1171 VOP_UNLOCK(devvp, 0); 1172 if (ump) { 1173 free(ump->um_lfs, M_UFSMNT); 1174 free(ump, M_UFSMNT); 1175 mp->mnt_data = NULL; 1176 } 1177 return (error); 1178 } 1179 1180 /* 1181 * unmount system call 1182 */ 1183 int 1184 lfs_unmount(struct mount *mp, int mntflags, struct proc *p) 1185 { 1186 struct ufsmount *ump; 1187 struct lfs *fs; 1188 int error, flags, ronly; 1189 1190 flags = 0; 1191 if (mntflags & MNT_FORCE) 1192 flags |= FORCECLOSE; 1193 1194 ump = VFSTOUFS(mp); 1195 fs = ump->um_lfs; 1196 #ifdef QUOTA 1197 if (mp->mnt_flag & MNT_QUOTA) { 1198 int i; 1199 error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags); 1200 if (error) 1201 return (error); 1202 for (i = 0; i < MAXQUOTAS; i++) { 1203 if (ump->um_quotas[i] == NULLVP) 1204 continue; 1205 quotaoff(p, mp, i); 1206 } 1207 /* 1208 * Here we fall through to vflush again to ensure 1209 * that we have gotten rid of all the system vnodes. 1210 */ 1211 } 1212 #endif 1213 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 1214 return (error); 1215 if ((error = VFS_SYNC(mp, 1, p->p_ucred, p)) != 0) 1216 return (error); 1217 if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd)) 1218 panic("lfs_unmount: still dirty blocks on ifile vnode"); 1219 1220 /* Explicitly write the superblock, to update serial and pflags */ 1221 fs->lfs_pflags |= LFS_PF_CLEAN; 1222 lfs_writesuper(fs, fs->lfs_sboffs[0]); 1223 lfs_writesuper(fs, fs->lfs_sboffs[1]); 1224 1225 /* Comment on ifile size if it has become too large */ 1226 if (!(fs->lfs_flags & LFS_WARNED)) { 1227 if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS) 1228 printf("lfs_unmount: please consider increasing" 1229 " NBUF to at least %lld\n", 1230 (long long)(fs->lfs_ivnode->v_size / 1231 fs->lfs_bsize) * 1232 (long long)(nbuf / LFS_MAX_BUFS)); 1233 if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES) 1234 printf("lfs_unmount: please consider increasing" 1235 " BUFPAGES to at least %lld\n", 1236 (long long)fs->lfs_ivnode->v_size * 1237 bufpages / LFS_MAX_BYTES); 1238 } 1239 1240 /* Finish with the Ifile, now that we're done with it */ 1241 vrele(fs->lfs_ivnode); 1242 vgone(fs->lfs_ivnode); 1243 1244 /* Wait for superblock writes to complete */ 1245 while (fs->lfs_iocount) 1246 tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0); 1247 1248 ronly = !fs->lfs_ronly; 1249 if (ump->um_devvp->v_type != VBAD) 1250 ump->um_devvp->v_specmountpoint = NULL; 1251 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1252 error = VOP_CLOSE(ump->um_devvp, 1253 ronly ? FREAD : FREAD|FWRITE, NOCRED, p); 1254 vput(ump->um_devvp); 1255 1256 /* XXX KS - wake up the cleaner so it can die */ 1257 wakeup(&fs->lfs_nextseg); 1258 wakeup(&lfs_allclean_wakeup); 1259 1260 free(fs, M_UFSMNT); 1261 free(ump, M_UFSMNT); 1262 mp->mnt_data = NULL; 1263 mp->mnt_flag &= ~MNT_LOCAL; 1264 return (error); 1265 } 1266 1267 /* 1268 * Get file system statistics. 1269 */ 1270 int 1271 lfs_statfs(struct mount *mp, struct statfs *sbp, struct proc *p) 1272 { 1273 struct lfs *fs; 1274 struct ufsmount *ump; 1275 1276 ump = VFSTOUFS(mp); 1277 fs = ump->um_lfs; 1278 if (fs->lfs_magic != LFS_MAGIC) 1279 panic("lfs_statfs: magic"); 1280 1281 sbp->f_type = 0; 1282 sbp->f_bsize = fs->lfs_fsize; 1283 sbp->f_iosize = fs->lfs_bsize; 1284 sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs)); 1285 sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs)); 1286 sbp->f_bavail = fsbtofrags(fs, (long)LFS_EST_BFREE(fs) - 1287 (long)LFS_EST_RSVD(fs)); 1288 1289 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs); 1290 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles; 1291 if (sbp != &mp->mnt_stat) { 1292 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN); 1293 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 1294 } 1295 strncpy(sbp->f_fstypename, mp->mnt_op->vfs_name, MFSNAMELEN); 1296 return (0); 1297 } 1298 1299 /* 1300 * Go through the disk queues to initiate sandbagged IO; 1301 * go through the inodes to write those that have been modified; 1302 * initiate the writing of the super block if it has been modified. 1303 * 1304 * Note: we are always called with the filesystem marked `MPBUSY'. 1305 */ 1306 int 1307 lfs_sync(struct mount *mp, int waitfor, struct ucred *cred, struct proc *p) 1308 { 1309 int error; 1310 struct lfs *fs; 1311 1312 fs = ((struct ufsmount *)mp->mnt_data)->ufsmount_u.lfs; 1313 if (fs->lfs_ronly) 1314 return 0; 1315 while (fs->lfs_dirops) 1316 error = tsleep(&fs->lfs_writer, PRIBIO + 1, "lfs_dirops", 0); 1317 fs->lfs_writer++; 1318 1319 /* All syncs must be checkpoints until roll-forward is implemented. */ 1320 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1321 if (--fs->lfs_writer == 0) 1322 wakeup(&fs->lfs_dirops); 1323 #ifdef QUOTA 1324 qsync(mp); 1325 #endif 1326 return (error); 1327 } 1328 1329 extern struct lock ufs_hashlock; 1330 1331 /* 1332 * Look up an LFS dinode number to find its incore vnode. If not already 1333 * in core, read it in from the specified device. Return the inode locked. 1334 * Detection and handling of mount points must be done by the calling routine. 1335 */ 1336 int 1337 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1338 { 1339 struct lfs *fs; 1340 struct dinode *dip; 1341 struct inode *ip; 1342 struct buf *bp; 1343 struct ifile *ifp; 1344 struct vnode *vp; 1345 struct ufsmount *ump; 1346 ufs_daddr_t daddr; 1347 dev_t dev; 1348 int i, error, retries; 1349 struct timespec ts; 1350 1351 ump = VFSTOUFS(mp); 1352 dev = ump->um_dev; 1353 fs = ump->um_lfs; 1354 1355 /* 1356 * If the filesystem is not completely mounted yet, suspend 1357 * any access requests (wait for roll-forward to complete). 1358 */ 1359 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1360 tsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0); 1361 1362 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) 1363 return (0); 1364 1365 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) { 1366 *vpp = NULL; 1367 return (error); 1368 } 1369 1370 do { 1371 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) { 1372 ungetnewvnode(vp); 1373 return (0); 1374 } 1375 } while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0)); 1376 1377 /* Translate the inode number to a disk address. */ 1378 if (ino == LFS_IFILE_INUM) 1379 daddr = fs->lfs_idaddr; 1380 else { 1381 /* XXX bounds-check this too */ 1382 LFS_IENTRY(ifp, fs, ino, bp); 1383 daddr = ifp->if_daddr; 1384 if (fs->lfs_version > 1) { 1385 ts.tv_sec = ifp->if_atime_sec; 1386 ts.tv_nsec = ifp->if_atime_nsec; 1387 } 1388 1389 brelse(bp); 1390 if (daddr == LFS_UNUSED_DADDR) { 1391 *vpp = NULLVP; 1392 ungetnewvnode(vp); 1393 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1394 return (ENOENT); 1395 } 1396 } 1397 1398 /* Allocate/init new vnode/inode. */ 1399 lfs_vcreate(mp, ino, vp); 1400 1401 /* 1402 * Put it onto its hash chain and lock it so that other requests for 1403 * this inode will block if they arrive while we are sleeping waiting 1404 * for old data structures to be purged or for the contents of the 1405 * disk portion of this inode to be read. 1406 */ 1407 ip = VTOI(vp); 1408 ufs_ihashins(ip); 1409 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1410 1411 /* 1412 * XXX 1413 * This may not need to be here, logically it should go down with 1414 * the i_devvp initialization. 1415 * Ask Kirk. 1416 */ 1417 ip->i_lfs = ump->um_lfs; 1418 1419 /* Read in the disk contents for the inode, copy into the inode. */ 1420 retries = 0; 1421 again: 1422 error = bread(ump->um_devvp, fsbtodb(fs, daddr), 1423 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize), 1424 NOCRED, &bp); 1425 if (error) { 1426 /* 1427 * The inode does not contain anything useful, so it would 1428 * be misleading to leave it on its hash chain. With mode 1429 * still zero, it will be unlinked and returned to the free 1430 * list by vput(). 1431 */ 1432 vput(vp); 1433 brelse(bp); 1434 *vpp = NULL; 1435 return (error); 1436 } 1437 1438 dip = lfs_ifind(fs, ino, bp); 1439 if (dip == NULL) { 1440 /* Assume write has not completed yet; try again */ 1441 bp->b_flags |= B_INVAL; 1442 brelse(bp); 1443 ++retries; 1444 if (retries > LFS_IFIND_RETRIES) { 1445 #ifdef DEBUG 1446 /* If the seglock is held look at the bpp to see 1447 what is there anyway */ 1448 if (fs->lfs_seglock > 0) { 1449 struct buf **bpp; 1450 struct dinode *dp; 1451 int i; 1452 1453 for (bpp = fs->lfs_sp->bpp; 1454 bpp != fs->lfs_sp->cbpp; ++bpp) { 1455 if ((*bpp)->b_vp == fs->lfs_ivnode && 1456 bpp != fs->lfs_sp->bpp) { 1457 /* Inode block */ 1458 printf("block 0x%x: ", (*bpp)->b_blkno); 1459 dp = (struct dinode *)(*bpp)->b_data; 1460 for (i = 0; i < INOPB(fs); i++) 1461 if (dp[i].di_u.inumber) 1462 printf("%d ", dp[i].di_u.inumber); 1463 printf("\n"); 1464 } 1465 } 1466 } 1467 #endif 1468 panic("lfs_vget: dinode not found"); 1469 } 1470 printf("lfs_vget: dinode %d not found, retrying...\n", ino); 1471 (void)tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs ifind", 1); 1472 goto again; 1473 } 1474 ip->i_din.ffs_din = *dip; 1475 1476 ip->i_ffs_effnlink = ip->i_ffs_nlink; 1477 ip->i_lfs_effnblks = ip->i_ffs_blocks; 1478 ip->i_lfs_osize = ip->i_ffs_size; 1479 if (fs->lfs_version > 1) { 1480 ip->i_ffs_atime = ts.tv_sec; 1481 ip->i_ffs_atimensec = ts.tv_nsec; 1482 } 1483 brelse(bp); 1484 1485 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 1486 for (i = 0; i < NDADDR; i++) 1487 if (ip->i_ffs_db[i] != 0) 1488 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 1489 1490 /* 1491 * Initialize the vnode from the inode, check for aliases. In all 1492 * cases re-init ip, the underlying vnode/inode may have changed. 1493 */ 1494 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 1495 #ifdef DIAGNOSTIC 1496 if (vp->v_type == VNON) { 1497 panic("lfs_vget: ino %d is type VNON! (ifmt %o)", 1498 ip->i_number, (ip->i_ffs_mode & IFMT) >> 12); 1499 } 1500 #endif 1501 /* 1502 * Finish inode initialization now that aliasing has been resolved. 1503 */ 1504 1505 genfs_node_init(vp, &lfs_genfsops); 1506 ip->i_devvp = ump->um_devvp; 1507 VREF(ip->i_devvp); 1508 *vpp = vp; 1509 1510 uvm_vnp_setsize(vp, ip->i_ffs_size); 1511 1512 return (0); 1513 } 1514 1515 /* 1516 * File handle to vnode 1517 * 1518 * Have to be really careful about stale file handles: 1519 * - check that the inode number is valid 1520 * - call lfs_vget() to get the locked inode 1521 * - check for an unallocated inode (i_mode == 0) 1522 * 1523 * XXX 1524 * use ifile to see if inode is allocated instead of reading off disk 1525 * what is the relationship between my generational number and the NFS 1526 * generational number. 1527 */ 1528 int 1529 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1530 { 1531 struct ufid *ufhp; 1532 1533 ufhp = (struct ufid *)fhp; 1534 if (ufhp->ufid_ino < ROOTINO) 1535 return (ESTALE); 1536 return (ufs_fhtovp(mp, ufhp, vpp)); 1537 } 1538 1539 /* 1540 * Vnode pointer to File handle 1541 */ 1542 /* ARGSUSED */ 1543 int 1544 lfs_vptofh(struct vnode *vp, struct fid *fhp) 1545 { 1546 struct inode *ip; 1547 struct ufid *ufhp; 1548 1549 ip = VTOI(vp); 1550 ufhp = (struct ufid *)fhp; 1551 ufhp->ufid_len = sizeof(struct ufid); 1552 ufhp->ufid_ino = ip->i_number; 1553 ufhp->ufid_gen = ip->i_ffs_gen; 1554 return (0); 1555 } 1556 1557 int 1558 lfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp, size_t newlen, struct proc *p) 1559 { 1560 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead; 1561 extern struct lfs_stats lfs_stats; 1562 int error; 1563 1564 /* all sysctl names at this level are terminal */ 1565 if (namelen != 1) 1566 return (ENOTDIR); 1567 1568 switch (name[0]) { 1569 case LFS_WRITEINDIR: 1570 return (sysctl_int(oldp, oldlenp, newp, newlen, 1571 &lfs_writeindir)); 1572 case LFS_CLEAN_VNHEAD: 1573 return (sysctl_int(oldp, oldlenp, newp, newlen, 1574 &lfs_clean_vnhead)); 1575 case LFS_DOSTATS: 1576 if ((error = sysctl_int(oldp, oldlenp, newp, newlen, 1577 &lfs_dostats))) 1578 return error; 1579 if (lfs_dostats == 0) 1580 memset(&lfs_stats,0,sizeof(lfs_stats)); 1581 return 0; 1582 case LFS_STATS: 1583 return (sysctl_rdstruct(oldp, oldlenp, newp, 1584 &lfs_stats, sizeof(lfs_stats))); 1585 default: 1586 return (EOPNOTSUPP); 1587 } 1588 /* NOTREACHED */ 1589 } 1590