lfs_vfsops.c revision 1.237
1 /* $NetBSD: lfs_vfsops.c,v 1.237 2007/07/09 21:11:34 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003, 2007 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 (at) 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. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_vfsops.c 8.20 (Berkeley) 6/10/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.237 2007/07/09 21:11:34 ad Exp $"); 71 72 #if defined(_KERNEL_OPT) 73 #include "opt_quota.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/proc.h> 80 #include <sys/kernel.h> 81 #include <sys/vnode.h> 82 #include <sys/mount.h> 83 #include <sys/kthread.h> 84 #include <sys/buf.h> 85 #include <sys/device.h> 86 #include <sys/mbuf.h> 87 #include <sys/file.h> 88 #include <sys/disklabel.h> 89 #include <sys/ioctl.h> 90 #include <sys/errno.h> 91 #include <sys/malloc.h> 92 #include <sys/pool.h> 93 #include <sys/socket.h> 94 #include <sys/syslog.h> 95 #include <uvm/uvm_extern.h> 96 #include <sys/sysctl.h> 97 #include <sys/conf.h> 98 #include <sys/kauth.h> 99 100 #include <miscfs/specfs/specdev.h> 101 102 #include <ufs/ufs/quota.h> 103 #include <ufs/ufs/inode.h> 104 #include <ufs/ufs/ufsmount.h> 105 #include <ufs/ufs/ufs_extern.h> 106 107 #include <uvm/uvm.h> 108 #include <uvm/uvm_stat.h> 109 #include <uvm/uvm_pager.h> 110 #include <uvm/uvm_pdaemon.h> 111 112 #include <ufs/lfs/lfs.h> 113 #include <ufs/lfs/lfs_extern.h> 114 115 #include <miscfs/genfs/genfs.h> 116 #include <miscfs/genfs/genfs_node.h> 117 118 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int); 119 static bool lfs_issequential_hole(const struct ufsmount *, 120 daddr_t, daddr_t); 121 122 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *); 123 124 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc; 125 extern const struct vnodeopv_desc lfs_specop_opv_desc; 126 extern const struct vnodeopv_desc lfs_fifoop_opv_desc; 127 128 pid_t lfs_writer_daemon = 0; 129 int lfs_do_flush = 0; 130 #ifdef LFS_KERNEL_RFW 131 int lfs_do_rfw = 0; 132 #endif 133 134 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = { 135 &lfs_vnodeop_opv_desc, 136 &lfs_specop_opv_desc, 137 &lfs_fifoop_opv_desc, 138 NULL, 139 }; 140 141 struct vfsops lfs_vfsops = { 142 MOUNT_LFS, 143 lfs_mount, 144 ufs_start, 145 lfs_unmount, 146 ufs_root, 147 ufs_quotactl, 148 lfs_statvfs, 149 lfs_sync, 150 lfs_vget, 151 lfs_fhtovp, 152 lfs_vptofh, 153 lfs_init, 154 lfs_reinit, 155 lfs_done, 156 lfs_mountroot, 157 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 158 vfs_stdextattrctl, 159 vfs_stdsuspendctl, 160 lfs_vnodeopv_descs, 161 0, 162 { NULL, NULL }, 163 }; 164 VFS_ATTACH(lfs_vfsops); 165 166 const struct genfs_ops lfs_genfsops = { 167 .gop_size = lfs_gop_size, 168 .gop_alloc = ufs_gop_alloc, 169 .gop_write = lfs_gop_write, 170 .gop_markupdate = ufs_gop_markupdate, 171 }; 172 173 static const struct ufs_ops lfs_ufsops = { 174 .uo_itimes = NULL, 175 .uo_update = lfs_update, 176 .uo_truncate = lfs_truncate, 177 .uo_valloc = lfs_valloc, 178 .uo_vfree = lfs_vfree, 179 .uo_balloc = lfs_balloc, 180 }; 181 182 /* 183 * XXX Same structure as FFS inodes? Should we share a common pool? 184 */ 185 struct pool lfs_inode_pool; 186 struct pool lfs_dinode_pool; 187 struct pool lfs_inoext_pool; 188 struct pool lfs_lbnentry_pool; 189 190 /* 191 * The writer daemon. UVM keeps track of how many dirty pages we are holding 192 * in lfs_subsys_pages; the daemon flushes the filesystem when this value 193 * crosses the (user-defined) threshhold LFS_MAX_PAGES. 194 */ 195 static void 196 lfs_writerd(void *arg) 197 { 198 struct mount *mp, *nmp; 199 struct lfs *fs; 200 int fsflags; 201 int loopcount; 202 203 lfs_writer_daemon = curproc->p_pid; 204 205 simple_lock(&lfs_subsys_lock); 206 for (;;) { 207 ltsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10, 208 &lfs_subsys_lock); 209 210 /* 211 * Look through the list of LFSs to see if any of them 212 * have requested pageouts. 213 */ 214 simple_lock(&mountlist_slock); 215 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 216 mp = nmp) { 217 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 218 nmp = CIRCLEQ_NEXT(mp, mnt_list); 219 continue; 220 } 221 if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS, 222 MFSNAMELEN) == 0) { 223 fs = VFSTOUFS(mp)->um_lfs; 224 simple_lock(&fs->lfs_interlock); 225 fsflags = 0; 226 if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || 227 lfs_dirvcount > LFS_MAX_DIROP) && 228 fs->lfs_dirops == 0) 229 fsflags |= SEGM_CKP; 230 if (fs->lfs_pdflush) { 231 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n")); 232 fs->lfs_pdflush = 0; 233 lfs_flush_fs(fs, fsflags); 234 simple_unlock(&fs->lfs_interlock); 235 } else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) { 236 DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n")); 237 simple_unlock(&fs->lfs_interlock); 238 lfs_writer_enter(fs, "wrdirop"); 239 lfs_flush_pchain(fs); 240 lfs_writer_leave(fs); 241 } else 242 simple_unlock(&fs->lfs_interlock); 243 } 244 245 simple_lock(&mountlist_slock); 246 nmp = CIRCLEQ_NEXT(mp, mnt_list); 247 vfs_unbusy(mp); 248 } 249 simple_unlock(&mountlist_slock); 250 251 /* 252 * If global state wants a flush, flush everything. 253 */ 254 simple_lock(&lfs_subsys_lock); 255 loopcount = 0; 256 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS || 257 locked_queue_bytes > LFS_MAX_BYTES || 258 lfs_subsys_pages > LFS_MAX_PAGES) { 259 260 if (lfs_do_flush) { 261 DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n")); 262 } 263 if (locked_queue_count > LFS_MAX_BUFS) { 264 DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n", 265 locked_queue_count, LFS_MAX_BUFS)); 266 } 267 if (locked_queue_bytes > LFS_MAX_BYTES) { 268 DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n", 269 locked_queue_bytes, LFS_MAX_BYTES)); 270 } 271 if (lfs_subsys_pages > LFS_MAX_PAGES) { 272 DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n", 273 lfs_subsys_pages, LFS_MAX_PAGES)); 274 } 275 276 lfs_flush(NULL, SEGM_WRITERD, 0); 277 lfs_do_flush = 0; 278 } 279 } 280 /* NOTREACHED */ 281 } 282 283 /* 284 * Initialize the filesystem, most work done by ufs_init. 285 */ 286 void 287 lfs_init() 288 { 289 290 malloc_type_attach(M_SEGMENT); 291 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 292 "lfsinopl", &pool_allocator_nointr, IPL_NONE); 293 pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, 294 "lfsdinopl", &pool_allocator_nointr, IPL_NONE); 295 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 296 "lfsinoextpl", &pool_allocator_nointr, IPL_NONE); 297 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 298 "lfslbnpool", &pool_allocator_nointr, IPL_NONE); 299 ufs_init(); 300 301 #ifdef DEBUG 302 memset(lfs_log, 0, sizeof(lfs_log)); 303 #endif 304 simple_lock_init(&lfs_subsys_lock); 305 } 306 307 void 308 lfs_reinit() 309 { 310 ufs_reinit(); 311 } 312 313 void 314 lfs_done() 315 { 316 317 ufs_done(); 318 pool_destroy(&lfs_inode_pool); 319 pool_destroy(&lfs_dinode_pool); 320 pool_destroy(&lfs_inoext_pool); 321 pool_destroy(&lfs_lbnentry_pool); 322 malloc_type_detach(M_SEGMENT); 323 } 324 325 /* 326 * Called by main() when ufs is going to be mounted as root. 327 */ 328 int 329 lfs_mountroot() 330 { 331 extern struct vnode *rootvp; 332 struct mount *mp; 333 struct lwp *l = curlwp; /* XXX */ 334 int error; 335 336 if (device_class(root_device) != DV_DISK) 337 return (ENODEV); 338 339 if (rootdev == NODEV) 340 return (ENODEV); 341 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 342 vrele(rootvp); 343 return (error); 344 } 345 if ((error = lfs_mountfs(rootvp, mp, l))) { 346 mp->mnt_op->vfs_refcount--; 347 vfs_unbusy(mp); 348 free(mp, M_MOUNT); 349 return (error); 350 } 351 simple_lock(&mountlist_slock); 352 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 353 simple_unlock(&mountlist_slock); 354 (void)lfs_statvfs(mp, &mp->mnt_stat, l); 355 vfs_unbusy(mp); 356 setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp)); 357 return (0); 358 } 359 360 /* 361 * VFS Operations. 362 * 363 * mount system call 364 */ 365 int 366 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct lwp *l) 367 { 368 struct vnode *devvp; 369 struct ufs_args args; 370 struct ufsmount *ump = NULL; 371 struct lfs *fs = NULL; /* LFS */ 372 int error, update; 373 mode_t accessmode; 374 375 if (mp->mnt_flag & MNT_GETARGS) { 376 ump = VFSTOUFS(mp); 377 if (ump == NULL) 378 return EIO; 379 args.fspec = NULL; 380 return copyout(&args, data, sizeof(args)); 381 } 382 error = copyin(data, &args, sizeof (struct ufs_args)); 383 if (error) 384 return (error); 385 386 update = mp->mnt_flag & MNT_UPDATE; 387 388 /* Check arguments */ 389 if (args.fspec != NULL) { 390 /* 391 * Look up the name and verify that it's sane. 392 */ 393 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, l); 394 if ((error = namei(ndp)) != 0) 395 return (error); 396 devvp = ndp->ni_vp; 397 398 if (!update) { 399 /* 400 * Be sure this is a valid block device 401 */ 402 if (devvp->v_type != VBLK) 403 error = ENOTBLK; 404 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 405 error = ENXIO; 406 } else { 407 /* 408 * Be sure we're still naming the same device 409 * used for our initial mount 410 */ 411 ump = VFSTOUFS(mp); 412 if (devvp != ump->um_devvp) 413 error = EINVAL; 414 } 415 } else { 416 if (!update) { 417 /* New mounts must have a filename for the device */ 418 return (EINVAL); 419 } else { 420 /* Use the extant mount */ 421 ump = VFSTOUFS(mp); 422 devvp = ump->um_devvp; 423 vref(devvp); 424 } 425 } 426 427 428 /* 429 * If mount by non-root, then verify that user has necessary 430 * permissions on the device. 431 */ 432 if (error == 0 && kauth_authorize_generic(l->l_cred, 433 KAUTH_GENERIC_ISSUSER, NULL) != 0) { 434 accessmode = VREAD; 435 if (update ? 436 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 437 (mp->mnt_flag & MNT_RDONLY) == 0) 438 accessmode |= VWRITE; 439 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 440 error = VOP_ACCESS(devvp, accessmode, l->l_cred, l); 441 VOP_UNLOCK(devvp, 0); 442 } 443 444 if (error) { 445 vrele(devvp); 446 return (error); 447 } 448 449 if (!update) { 450 int flags; 451 452 /* 453 * Disallow multiple mounts of the same device. 454 * Disallow mounting of a device that is currently in use 455 * (except for root, which might share swap device for 456 * miniroot). 457 */ 458 error = vfs_mountedon(devvp); 459 if (error) 460 goto fail; 461 if (vcount(devvp) > 1 && devvp != rootvp) { 462 error = EBUSY; 463 goto fail; 464 } 465 if (mp->mnt_flag & MNT_RDONLY) 466 flags = FREAD; 467 else 468 flags = FREAD|FWRITE; 469 error = VOP_OPEN(devvp, flags, FSCRED, l); 470 if (error) 471 goto fail; 472 error = lfs_mountfs(devvp, mp, l); /* LFS */ 473 if (error) { 474 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 475 (void)VOP_CLOSE(devvp, flags, NOCRED, l); 476 VOP_UNLOCK(devvp, 0); 477 goto fail; 478 } 479 480 ump = VFSTOUFS(mp); 481 fs = ump->um_lfs; 482 } else { 483 /* 484 * Update the mount. 485 */ 486 487 /* 488 * The initial mount got a reference on this 489 * device, so drop the one obtained via 490 * namei(), above. 491 */ 492 vrele(devvp); 493 494 ump = VFSTOUFS(mp); 495 fs = ump->um_lfs; 496 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 497 /* 498 * Changing from read-only to read/write. 499 * Note in the superblocks that we're writing. 500 */ 501 fs->lfs_ronly = 0; 502 if (fs->lfs_pflags & LFS_PF_CLEAN) { 503 fs->lfs_pflags &= ~LFS_PF_CLEAN; 504 lfs_writesuper(fs, fs->lfs_sboffs[0]); 505 lfs_writesuper(fs, fs->lfs_sboffs[1]); 506 } 507 } 508 if (args.fspec == NULL) 509 return EINVAL; 510 } 511 512 error = set_statvfs_info(path, UIO_USERSPACE, args.fspec, 513 UIO_USERSPACE, mp, l); 514 if (error == 0) 515 (void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname, 516 sizeof(fs->lfs_fsmnt)); 517 return error; 518 519 fail: 520 vrele(devvp); 521 return (error); 522 } 523 524 525 /* 526 * Common code for mount and mountroot 527 * LFS specific 528 */ 529 int 530 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 531 { 532 struct dlfs *tdfs, *dfs, *adfs; 533 struct lfs *fs; 534 struct ufsmount *ump; 535 struct vnode *vp; 536 struct buf *bp, *abp; 537 struct partinfo dpart; 538 dev_t dev; 539 int error, i, ronly, secsize, fsbsize; 540 kauth_cred_t cred; 541 CLEANERINFO *cip; 542 SEGUSE *sup; 543 daddr_t sb_addr; 544 545 cred = l ? l->l_cred : NOCRED; 546 547 /* 548 * Flush out any old buffers remaining from a previous use. 549 */ 550 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 551 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 552 VOP_UNLOCK(devvp, 0); 553 if (error) 554 return (error); 555 556 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 557 if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, l) != 0) 558 secsize = DEV_BSIZE; 559 else 560 secsize = dpart.disklab->d_secsize; 561 562 /* Don't free random space on error. */ 563 bp = NULL; 564 abp = NULL; 565 ump = NULL; 566 567 sb_addr = LFS_LABELPAD / secsize; 568 while (1) { 569 /* Read in the superblock. */ 570 error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp); 571 if (error) 572 goto out; 573 dfs = (struct dlfs *)bp->b_data; 574 575 /* Check the basics. */ 576 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE || 577 dfs->dlfs_version > LFS_VERSION || 578 dfs->dlfs_bsize < sizeof(struct dlfs)) { 579 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 580 error = EINVAL; /* XXX needs translation */ 581 goto out; 582 } 583 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) { 584 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 585 dfs->dlfs_inodefmt)); 586 error = EINVAL; 587 goto out; 588 } 589 590 if (dfs->dlfs_version == 1) 591 fsbsize = secsize; 592 else { 593 fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb + 594 dfs->dlfs_fsbtodb); 595 /* 596 * Could be, if the frag size is large enough, that we 597 * don't have the "real" primary superblock. If that's 598 * the case, get the real one, and try again. 599 */ 600 if (sb_addr != dfs->dlfs_sboffs[0] << 601 dfs->dlfs_fsbtodb) { 602 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 603 " 0x%llx is not right, trying 0x%llx\n", 604 (long long)sb_addr, 605 (long long)(dfs->dlfs_sboffs[0] << 606 dfs->dlfs_fsbtodb))); 607 sb_addr = dfs->dlfs_sboffs[0] << 608 dfs->dlfs_fsbtodb; 609 brelse(bp); 610 continue; 611 } 612 } 613 break; 614 } 615 616 /* 617 * Check the second superblock to see which is newer; then mount 618 * using the older of the two. This is necessary to ensure that 619 * the filesystem is valid if it was not unmounted cleanly. 620 */ 621 622 if (dfs->dlfs_sboffs[1] && 623 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 624 { 625 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize), 626 LFS_SBPAD, cred, &abp); 627 if (error) 628 goto out; 629 adfs = (struct dlfs *)abp->b_data; 630 631 if (dfs->dlfs_version == 1) { 632 /* 1s resolution comparison */ 633 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 634 tdfs = adfs; 635 else 636 tdfs = dfs; 637 } else { 638 /* monotonic infinite-resolution comparison */ 639 if (adfs->dlfs_serial < dfs->dlfs_serial) 640 tdfs = adfs; 641 else 642 tdfs = dfs; 643 } 644 645 /* Check the basics. */ 646 if (tdfs->dlfs_magic != LFS_MAGIC || 647 tdfs->dlfs_bsize > MAXBSIZE || 648 tdfs->dlfs_version > LFS_VERSION || 649 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 650 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 651 " sanity failed\n")); 652 error = EINVAL; /* XXX needs translation */ 653 goto out; 654 } 655 } else { 656 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock" 657 " daddr=0x%x\n", dfs->dlfs_sboffs[1])); 658 error = EINVAL; 659 goto out; 660 } 661 662 /* Allocate the mount structure, copy the superblock into it. */ 663 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO); 664 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 665 666 /* Compatibility */ 667 if (fs->lfs_version < 2) { 668 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 669 fs->lfs_ibsize = fs->lfs_bsize; 670 fs->lfs_start = fs->lfs_sboffs[0]; 671 fs->lfs_tstamp = fs->lfs_otstamp; 672 fs->lfs_fsbtodb = 0; 673 } 674 if (fs->lfs_resvseg == 0) 675 fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \ 676 MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1)); 677 678 /* 679 * If we aren't going to be able to write meaningfully to this 680 * filesystem, and were not mounted readonly, bomb out now. 681 */ 682 if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 683 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 684 " we need BUFPAGES >= %lld\n", 685 (long long)((bufmem_hiwater / bufmem_lowater) * 686 LFS_INVERSE_MAX_BYTES( 687 fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 688 free(fs, M_UFSMNT); 689 error = EFBIG; /* XXX needs translation */ 690 goto out; 691 } 692 693 /* Before rolling forward, lock so vget will sleep for other procs */ 694 if (l != NULL) { 695 fs->lfs_flags = LFS_NOTYET; 696 fs->lfs_rfpid = l->l_proc->p_pid; 697 } 698 699 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 700 ump->um_lfs = fs; 701 ump->um_ops = &lfs_ufsops; 702 ump->um_fstype = UFS1; 703 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 704 bp->b_flags |= B_INVAL; 705 abp->b_flags |= B_INVAL; 706 } 707 brelse(bp); 708 bp = NULL; 709 brelse(abp); 710 abp = NULL; 711 712 /* Set up the I/O information */ 713 fs->lfs_devbsize = secsize; 714 fs->lfs_iocount = 0; 715 fs->lfs_diropwait = 0; 716 fs->lfs_activesb = 0; 717 fs->lfs_uinodes = 0; 718 fs->lfs_ravail = 0; 719 fs->lfs_favail = 0; 720 fs->lfs_sbactive = 0; 721 722 /* Set up the ifile and lock aflags */ 723 fs->lfs_doifile = 0; 724 fs->lfs_writer = 0; 725 fs->lfs_dirops = 0; 726 fs->lfs_nadirop = 0; 727 fs->lfs_seglock = 0; 728 fs->lfs_pdflush = 0; 729 fs->lfs_sleepers = 0; 730 fs->lfs_pages = 0; 731 simple_lock_init(&fs->lfs_interlock); 732 rw_init(&fs->lfs_fraglock); 733 lockinit(&fs->lfs_iflock, PINOD, "lfs_iflock", 0, 0); 734 lockinit(&fs->lfs_stoplock, PINOD, "lfs_stoplock", 0, 0); 735 736 /* Set the file system readonly/modify bits. */ 737 fs->lfs_ronly = ronly; 738 if (ronly == 0) 739 fs->lfs_fmod = 1; 740 741 /* Initialize the mount structure. */ 742 dev = devvp->v_rdev; 743 mp->mnt_data = ump; 744 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 745 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 746 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 747 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 748 mp->mnt_stat.f_iosize = fs->lfs_bsize; 749 mp->mnt_flag |= MNT_LOCAL; 750 mp->mnt_fs_bshift = fs->lfs_bshift; 751 ump->um_flags = 0; 752 ump->um_mountp = mp; 753 ump->um_dev = dev; 754 ump->um_devvp = devvp; 755 ump->um_bptrtodb = fs->lfs_fsbtodb; 756 ump->um_seqinc = fragstofsb(fs, fs->lfs_frag); 757 ump->um_nindir = fs->lfs_nindir; 758 ump->um_lognindir = ffs(fs->lfs_nindir) - 1; 759 for (i = 0; i < MAXQUOTAS; i++) 760 ump->um_quotas[i] = NULLVP; 761 ump->um_maxsymlinklen = fs->lfs_maxsymlinklen; 762 ump->um_dirblksiz = DIRBLKSIZ; 763 ump->um_maxfilesize = fs->lfs_maxfilesize; 764 if (ump->um_maxsymlinklen > 0) 765 mp->mnt_iflag |= IMNT_DTYPE; 766 devvp->v_specmountpoint = mp; 767 768 /* Set up reserved memory for pageout */ 769 lfs_setup_resblks(fs); 770 /* Set up vdirop tailq */ 771 TAILQ_INIT(&fs->lfs_dchainhd); 772 /* and paging tailq */ 773 TAILQ_INIT(&fs->lfs_pchainhd); 774 /* and delayed segment accounting for truncation list */ 775 LIST_INIT(&fs->lfs_segdhd); 776 777 /* 778 * We use the ifile vnode for almost every operation. Instead of 779 * retrieving it from the hash table each time we retrieve it here, 780 * artificially increment the reference count and keep a pointer 781 * to it in the incore copy of the superblock. 782 */ 783 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) { 784 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 785 goto out; 786 } 787 fs->lfs_ivnode = vp; 788 VREF(vp); 789 790 /* Set up inode bitmap and order free list */ 791 lfs_order_freelist(fs); 792 793 /* Set up segment usage flags for the autocleaner. */ 794 fs->lfs_nactive = 0; 795 fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *), 796 M_SEGMENT, M_WAITOK); 797 fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 798 M_SEGMENT, M_WAITOK); 799 fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 800 M_SEGMENT, M_WAITOK); 801 memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t)); 802 for (i = 0; i < fs->lfs_nseg; i++) { 803 int changed; 804 805 LFS_SEGENTRY(sup, fs, i, bp); 806 changed = 0; 807 if (!ronly) { 808 if (sup->su_nbytes == 0 && 809 !(sup->su_flags & SEGUSE_EMPTY)) { 810 sup->su_flags |= SEGUSE_EMPTY; 811 ++changed; 812 } else if (!(sup->su_nbytes == 0) && 813 (sup->su_flags & SEGUSE_EMPTY)) { 814 sup->su_flags &= ~SEGUSE_EMPTY; 815 ++changed; 816 } 817 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 818 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 819 ++changed; 820 } 821 } 822 fs->lfs_suflags[0][i] = sup->su_flags; 823 if (changed) 824 LFS_WRITESEGENTRY(sup, fs, i, bp); 825 else 826 brelse(bp); 827 } 828 829 #ifdef LFS_KERNEL_RFW 830 lfs_roll_forward(fs, mp, l); 831 #endif 832 833 /* If writing, sb is not clean; record in case of immediate crash */ 834 if (!fs->lfs_ronly) { 835 fs->lfs_pflags &= ~LFS_PF_CLEAN; 836 lfs_writesuper(fs, fs->lfs_sboffs[0]); 837 lfs_writesuper(fs, fs->lfs_sboffs[1]); 838 } 839 840 /* Allow vget now that roll-forward is complete */ 841 fs->lfs_flags &= ~(LFS_NOTYET); 842 wakeup(&fs->lfs_flags); 843 844 /* 845 * Initialize the ifile cleaner info with information from 846 * the superblock. 847 */ 848 LFS_CLEANERINFO(cip, fs, bp); 849 cip->clean = fs->lfs_nclean; 850 cip->dirty = fs->lfs_nseg - fs->lfs_nclean; 851 cip->avail = fs->lfs_avail; 852 cip->bfree = fs->lfs_bfree; 853 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 854 855 /* 856 * Mark the current segment as ACTIVE, since we're going to 857 * be writing to it. 858 */ 859 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); 860 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 861 fs->lfs_nactive++; 862 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); /* Ifile */ 863 864 /* Now that roll-forward is done, unlock the Ifile */ 865 vput(vp); 866 867 /* Start the pagedaemon-anticipating daemon */ 868 if (lfs_writer_daemon == 0 && 869 kthread_create(PRI_NONE, 0, NULL, lfs_writerd, NULL, NULL, 870 "lfs_writer") != 0) 871 panic("fork lfs_writer"); 872 873 return (0); 874 875 out: 876 if (bp) 877 brelse(bp); 878 if (abp) 879 brelse(abp); 880 if (ump) { 881 free(ump->um_lfs, M_UFSMNT); 882 free(ump, M_UFSMNT); 883 mp->mnt_data = NULL; 884 } 885 886 return (error); 887 } 888 889 /* 890 * unmount system call 891 */ 892 int 893 lfs_unmount(struct mount *mp, int mntflags, struct lwp *l) 894 { 895 struct ufsmount *ump; 896 struct lfs *fs; 897 int error, flags, ronly; 898 int s; 899 900 flags = 0; 901 if (mntflags & MNT_FORCE) 902 flags |= FORCECLOSE; 903 904 ump = VFSTOUFS(mp); 905 fs = ump->um_lfs; 906 907 /* Two checkpoints */ 908 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 909 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 910 911 /* wake up the cleaner so it can die */ 912 lfs_wakeup_cleaner(fs); 913 simple_lock(&fs->lfs_interlock); 914 while (fs->lfs_sleepers) 915 ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0, 916 &fs->lfs_interlock); 917 simple_unlock(&fs->lfs_interlock); 918 919 #ifdef QUOTA 920 if (mp->mnt_flag & MNT_QUOTA) { 921 int i; 922 error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags); 923 if (error) 924 return (error); 925 for (i = 0; i < MAXQUOTAS; i++) { 926 if (ump->um_quotas[i] == NULLVP) 927 continue; 928 quotaoff(l, mp, i); 929 } 930 /* 931 * Here we fall through to vflush again to ensure 932 * that we have gotten rid of all the system vnodes. 933 */ 934 } 935 #endif 936 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 937 return (error); 938 if ((error = VFS_SYNC(mp, 1, l->l_cred, l)) != 0) 939 return (error); 940 s = splbio(); 941 if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd)) 942 panic("lfs_unmount: still dirty blocks on ifile vnode"); 943 splx(s); 944 945 /* Explicitly write the superblock, to update serial and pflags */ 946 fs->lfs_pflags |= LFS_PF_CLEAN; 947 lfs_writesuper(fs, fs->lfs_sboffs[0]); 948 lfs_writesuper(fs, fs->lfs_sboffs[1]); 949 simple_lock(&fs->lfs_interlock); 950 while (fs->lfs_iocount) 951 ltsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 952 &fs->lfs_interlock); 953 simple_unlock(&fs->lfs_interlock); 954 955 /* Finish with the Ifile, now that we're done with it */ 956 vrele(fs->lfs_ivnode); 957 vgone(fs->lfs_ivnode); 958 959 ronly = !fs->lfs_ronly; 960 if (ump->um_devvp->v_type != VBAD) 961 ump->um_devvp->v_specmountpoint = NULL; 962 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 963 error = VOP_CLOSE(ump->um_devvp, 964 ronly ? FREAD : FREAD|FWRITE, NOCRED, l); 965 vput(ump->um_devvp); 966 967 /* Complain about page leakage */ 968 if (fs->lfs_pages > 0) 969 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 970 fs->lfs_pages, lfs_subsys_pages); 971 972 /* Free per-mount data structures */ 973 free(fs->lfs_ino_bitmap, M_SEGMENT); 974 free(fs->lfs_suflags[0], M_SEGMENT); 975 free(fs->lfs_suflags[1], M_SEGMENT); 976 free(fs->lfs_suflags, M_SEGMENT); 977 lfs_free_resblks(fs); 978 rw_destroy(&fs->lfs_fraglock); 979 free(fs, M_UFSMNT); 980 free(ump, M_UFSMNT); 981 982 mp->mnt_data = NULL; 983 mp->mnt_flag &= ~MNT_LOCAL; 984 return (error); 985 } 986 987 /* 988 * Get file system statistics. 989 * 990 * NB: We don't lock to access the superblock here, because it's not 991 * really that important if we get it wrong. 992 */ 993 int 994 lfs_statvfs(struct mount *mp, struct statvfs *sbp, struct lwp *l) 995 { 996 struct lfs *fs; 997 struct ufsmount *ump; 998 999 ump = VFSTOUFS(mp); 1000 fs = ump->um_lfs; 1001 if (fs->lfs_magic != LFS_MAGIC) 1002 panic("lfs_statvfs: magic"); 1003 1004 sbp->f_bsize = fs->lfs_bsize; 1005 sbp->f_frsize = fs->lfs_fsize; 1006 sbp->f_iosize = fs->lfs_bsize; 1007 sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks); 1008 1009 sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs)); 1010 KASSERT(sbp->f_bfree <= fs->lfs_dsize); 1011 #if 0 1012 if (sbp->f_bfree < 0) 1013 sbp->f_bfree = 0; 1014 #endif 1015 1016 sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs)); 1017 if (sbp->f_bfree > sbp->f_bresvd) 1018 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1019 else 1020 sbp->f_bavail = 0; 1021 1022 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs); 1023 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles; 1024 sbp->f_favail = sbp->f_ffree; 1025 sbp->f_fresvd = 0; 1026 copy_statvfs_info(sbp, mp); 1027 return (0); 1028 } 1029 1030 /* 1031 * Go through the disk queues to initiate sandbagged IO; 1032 * go through the inodes to write those that have been modified; 1033 * initiate the writing of the super block if it has been modified. 1034 * 1035 * Note: we are always called with the filesystem marked `MPBUSY'. 1036 */ 1037 int 1038 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred, 1039 struct lwp *l) 1040 { 1041 int error; 1042 struct lfs *fs; 1043 1044 fs = VFSTOUFS(mp)->um_lfs; 1045 if (fs->lfs_ronly) 1046 return 0; 1047 1048 /* Snapshots should not hose the syncer */ 1049 /* 1050 * XXX Sync can block here anyway, since we don't have a very 1051 * XXX good idea of how much data is pending. If it's more 1052 * XXX than a segment and lfs_nextseg is close to the end of 1053 * XXX the log, we'll likely block. 1054 */ 1055 simple_lock(&fs->lfs_interlock); 1056 if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) { 1057 simple_unlock(&fs->lfs_interlock); 1058 return 0; 1059 } 1060 simple_unlock(&fs->lfs_interlock); 1061 1062 lfs_writer_enter(fs, "lfs_dirops"); 1063 1064 /* All syncs must be checkpoints until roll-forward is implemented. */ 1065 DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset)); 1066 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1067 lfs_writer_leave(fs); 1068 #ifdef QUOTA 1069 qsync(mp); 1070 #endif 1071 return (error); 1072 } 1073 1074 extern kmutex_t ufs_hashlock; 1075 1076 /* 1077 * Look up an LFS dinode number to find its incore vnode. If not already 1078 * in core, read it in from the specified device. Return the inode locked. 1079 * Detection and handling of mount points must be done by the calling routine. 1080 */ 1081 int 1082 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1083 { 1084 struct lfs *fs; 1085 struct ufs1_dinode *dip; 1086 struct inode *ip; 1087 struct buf *bp; 1088 struct ifile *ifp; 1089 struct vnode *vp; 1090 struct ufsmount *ump; 1091 daddr_t daddr; 1092 dev_t dev; 1093 int error, retries; 1094 struct timespec ts; 1095 1096 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1097 1098 ump = VFSTOUFS(mp); 1099 dev = ump->um_dev; 1100 fs = ump->um_lfs; 1101 1102 /* 1103 * If the filesystem is not completely mounted yet, suspend 1104 * any access requests (wait for roll-forward to complete). 1105 */ 1106 simple_lock(&fs->lfs_interlock); 1107 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1108 ltsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1109 &fs->lfs_interlock); 1110 simple_unlock(&fs->lfs_interlock); 1111 1112 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) 1113 return (0); 1114 1115 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) { 1116 *vpp = NULL; 1117 return (error); 1118 } 1119 1120 mutex_enter(&ufs_hashlock); 1121 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) { 1122 mutex_exit(&ufs_hashlock); 1123 ungetnewvnode(vp); 1124 return (0); 1125 } 1126 1127 /* Translate the inode number to a disk address. */ 1128 if (ino == LFS_IFILE_INUM) 1129 daddr = fs->lfs_idaddr; 1130 else { 1131 /* XXX bounds-check this too */ 1132 LFS_IENTRY(ifp, fs, ino, bp); 1133 daddr = ifp->if_daddr; 1134 if (fs->lfs_version > 1) { 1135 ts.tv_sec = ifp->if_atime_sec; 1136 ts.tv_nsec = ifp->if_atime_nsec; 1137 } 1138 1139 brelse(bp); 1140 if (daddr == LFS_UNUSED_DADDR) { 1141 *vpp = NULLVP; 1142 mutex_exit(&ufs_hashlock); 1143 ungetnewvnode(vp); 1144 return (ENOENT); 1145 } 1146 } 1147 1148 /* Allocate/init new vnode/inode. */ 1149 lfs_vcreate(mp, ino, vp); 1150 1151 /* 1152 * Put it onto its hash chain and lock it so that other requests for 1153 * this inode will block if they arrive while we are sleeping waiting 1154 * for old data structures to be purged or for the contents of the 1155 * disk portion of this inode to be read. 1156 */ 1157 ip = VTOI(vp); 1158 ufs_ihashins(ip); 1159 mutex_exit(&ufs_hashlock); 1160 1161 /* 1162 * XXX 1163 * This may not need to be here, logically it should go down with 1164 * the i_devvp initialization. 1165 * Ask Kirk. 1166 */ 1167 ip->i_lfs = ump->um_lfs; 1168 1169 /* Read in the disk contents for the inode, copy into the inode. */ 1170 retries = 0; 1171 again: 1172 error = bread(ump->um_devvp, fsbtodb(fs, daddr), 1173 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize), 1174 NOCRED, &bp); 1175 if (error) { 1176 /* 1177 * The inode does not contain anything useful, so it would 1178 * be misleading to leave it on its hash chain. With mode 1179 * still zero, it will be unlinked and returned to the free 1180 * list by vput(). 1181 */ 1182 vput(vp); 1183 brelse(bp); 1184 *vpp = NULL; 1185 return (error); 1186 } 1187 1188 dip = lfs_ifind(fs, ino, bp); 1189 if (dip == NULL) { 1190 /* Assume write has not completed yet; try again */ 1191 bp->b_flags |= B_INVAL; 1192 brelse(bp); 1193 ++retries; 1194 if (retries > LFS_IFIND_RETRIES) { 1195 #ifdef DEBUG 1196 /* If the seglock is held look at the bpp to see 1197 what is there anyway */ 1198 simple_lock(&fs->lfs_interlock); 1199 if (fs->lfs_seglock > 0) { 1200 struct buf **bpp; 1201 struct ufs1_dinode *dp; 1202 int i; 1203 1204 for (bpp = fs->lfs_sp->bpp; 1205 bpp != fs->lfs_sp->cbpp; ++bpp) { 1206 if ((*bpp)->b_vp == fs->lfs_ivnode && 1207 bpp != fs->lfs_sp->bpp) { 1208 /* Inode block */ 1209 printf("lfs_vget: block 0x%" PRIx64 ": ", 1210 (*bpp)->b_blkno); 1211 dp = (struct ufs1_dinode *)(*bpp)->b_data; 1212 for (i = 0; i < INOPB(fs); i++) 1213 if (dp[i].di_u.inumber) 1214 printf("%d ", dp[i].di_u.inumber); 1215 printf("\n"); 1216 } 1217 } 1218 } 1219 simple_unlock(&fs->lfs_interlock); 1220 #endif /* DEBUG */ 1221 panic("lfs_vget: dinode not found"); 1222 } 1223 simple_lock(&fs->lfs_interlock); 1224 if (fs->lfs_iocount) { 1225 DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino)); 1226 (void)ltsleep(&fs->lfs_iocount, PRIBIO + 1, 1227 "lfs ifind", 1, &fs->lfs_interlock); 1228 } else 1229 retries = LFS_IFIND_RETRIES; 1230 simple_unlock(&fs->lfs_interlock); 1231 goto again; 1232 } 1233 *ip->i_din.ffs1_din = *dip; 1234 brelse(bp); 1235 1236 if (fs->lfs_version > 1) { 1237 ip->i_ffs1_atime = ts.tv_sec; 1238 ip->i_ffs1_atimensec = ts.tv_nsec; 1239 } 1240 1241 lfs_vinit(mp, &vp); 1242 1243 *vpp = vp; 1244 1245 KASSERT(VOP_ISLOCKED(vp)); 1246 1247 return (0); 1248 } 1249 1250 /* 1251 * File handle to vnode 1252 */ 1253 int 1254 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1255 { 1256 struct lfid lfh; 1257 struct buf *bp; 1258 IFILE *ifp; 1259 int32_t daddr; 1260 struct lfs *fs; 1261 1262 if (fhp->fid_len != sizeof(struct lfid)) 1263 return EINVAL; 1264 1265 memcpy(&lfh, fhp, sizeof(lfh)); 1266 if (lfh.lfid_ino < LFS_IFILE_INUM) 1267 return ESTALE; 1268 1269 fs = VFSTOUFS(mp)->um_lfs; 1270 if (lfh.lfid_ident != fs->lfs_ident) 1271 return ESTALE; 1272 1273 if (lfh.lfid_ino > 1274 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) - 1275 fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb) 1276 return ESTALE; 1277 1278 if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino) == NULLVP) { 1279 LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp); 1280 daddr = ifp->if_daddr; 1281 brelse(bp); 1282 if (daddr == LFS_UNUSED_DADDR) 1283 return ESTALE; 1284 } 1285 1286 return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp)); 1287 } 1288 1289 /* 1290 * Vnode pointer to File handle 1291 */ 1292 /* ARGSUSED */ 1293 int 1294 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 1295 { 1296 struct inode *ip; 1297 struct lfid lfh; 1298 1299 if (*fh_size < sizeof(struct lfid)) { 1300 *fh_size = sizeof(struct lfid); 1301 return E2BIG; 1302 } 1303 *fh_size = sizeof(struct lfid); 1304 ip = VTOI(vp); 1305 memset(&lfh, 0, sizeof(lfh)); 1306 lfh.lfid_len = sizeof(struct lfid); 1307 lfh.lfid_ino = ip->i_number; 1308 lfh.lfid_gen = ip->i_gen; 1309 lfh.lfid_ident = ip->i_lfs->lfs_ident; 1310 memcpy(fhp, &lfh, sizeof(lfh)); 1311 return (0); 1312 } 1313 1314 static int 1315 sysctl_lfs_dostats(SYSCTLFN_ARGS) 1316 { 1317 extern struct lfs_stats lfs_stats; 1318 extern int lfs_dostats; 1319 int error; 1320 1321 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 1322 if (error || newp == NULL) 1323 return (error); 1324 1325 if (lfs_dostats == 0) 1326 memset(&lfs_stats, 0, sizeof(lfs_stats)); 1327 1328 return (0); 1329 } 1330 1331 struct shortlong { 1332 const char *sname; 1333 const char *lname; 1334 }; 1335 1336 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup") 1337 { 1338 int i; 1339 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 1340 lfs_fs_pagetrip, lfs_ignore_lazy_sync; 1341 #ifdef DEBUG 1342 extern int lfs_debug_log_subsys[DLOG_MAX]; 1343 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 1344 { "rollforward", "Debug roll-forward code" }, 1345 { "alloc", "Debug inode allocation and free list" }, 1346 { "avail", "Debug space-available-now accounting" }, 1347 { "flush", "Debug flush triggers" }, 1348 { "lockedlist", "Debug locked list accounting" }, 1349 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 1350 { "vnode", "Debug vnode use during segment write" }, 1351 { "segment", "Debug segment writing" }, 1352 { "seguse", "Debug segment used-bytes accounting" }, 1353 { "cleaner", "Debug cleaning routines" }, 1354 { "mount", "Debug mount/unmount routines" }, 1355 { "pagecache", "Debug UBC interactions" }, 1356 { "dirop", "Debug directory-operation accounting" }, 1357 { "malloc", "Debug private malloc accounting" }, 1358 }; 1359 #endif /* DEBUG */ 1360 struct shortlong stat_names[] = { /* Must match lfs.h! */ 1361 { "segsused", "Number of new segments allocated" }, 1362 { "psegwrites", "Number of partial-segment writes" }, 1363 { "psyncwrites", "Number of synchronous partial-segment" 1364 " writes" }, 1365 { "pcleanwrites", "Number of partial-segment writes by the" 1366 " cleaner" }, 1367 { "blocktot", "Number of blocks written" }, 1368 { "cleanblocks", "Number of blocks written by the cleaner" }, 1369 { "ncheckpoints", "Number of checkpoints made" }, 1370 { "nwrites", "Number of whole writes" }, 1371 { "nsync_writes", "Number of synchronous writes" }, 1372 { "wait_exceeded", "Number of times writer waited for" 1373 " cleaner" }, 1374 { "write_exceeded", "Number of times writer invoked flush" }, 1375 { "flush_invoked", "Number of times flush was invoked" }, 1376 { "vflush_invoked", "Number of time vflush was called" }, 1377 { "clean_inlocked", "Number of vnodes skipped for VXLOCK" }, 1378 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 1379 { "segs_reclaimed", "Number of segments reclaimed" }, 1380 }; 1381 1382 sysctl_createv(clog, 0, NULL, NULL, 1383 CTLFLAG_PERMANENT, 1384 CTLTYPE_NODE, "vfs", NULL, 1385 NULL, 0, NULL, 0, 1386 CTL_VFS, CTL_EOL); 1387 sysctl_createv(clog, 0, NULL, NULL, 1388 CTLFLAG_PERMANENT, 1389 CTLTYPE_NODE, "lfs", 1390 SYSCTL_DESCR("Log-structured file system"), 1391 NULL, 0, NULL, 0, 1392 CTL_VFS, 5, CTL_EOL); 1393 /* 1394 * XXX the "5" above could be dynamic, thereby eliminating one 1395 * more instance of the "number to vfs" mapping problem, but 1396 * "5" is the order as taken from sys/mount.h 1397 */ 1398 1399 sysctl_createv(clog, 0, NULL, NULL, 1400 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1401 CTLTYPE_INT, "flushindir", NULL, 1402 NULL, 0, &lfs_writeindir, 0, 1403 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 1404 sysctl_createv(clog, 0, NULL, NULL, 1405 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1406 CTLTYPE_INT, "clean_vnhead", NULL, 1407 NULL, 0, &lfs_clean_vnhead, 0, 1408 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 1409 sysctl_createv(clog, 0, NULL, NULL, 1410 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1411 CTLTYPE_INT, "dostats", 1412 SYSCTL_DESCR("Maintain statistics on LFS operations"), 1413 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 1414 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 1415 sysctl_createv(clog, 0, NULL, NULL, 1416 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1417 CTLTYPE_INT, "pagetrip", 1418 SYSCTL_DESCR("How many dirty pages in fs triggers" 1419 " a flush"), 1420 NULL, 0, &lfs_fs_pagetrip, 0, 1421 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 1422 sysctl_createv(clog, 0, NULL, NULL, 1423 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1424 CTLTYPE_INT, "ignore_lazy_sync", 1425 SYSCTL_DESCR("Lazy Sync is ignored entirely"), 1426 NULL, 0, &lfs_ignore_lazy_sync, 0, 1427 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL); 1428 #ifdef LFS_KERNEL_RFW 1429 sysctl_createv(clog, 0, NULL, NULL, 1430 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1431 CTLTYPE_INT, "rfw", 1432 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 1433 NULL, 0, &lfs_do_rfw, 0, 1434 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 1435 #endif 1436 1437 sysctl_createv(clog, 0, NULL, NULL, 1438 CTLFLAG_PERMANENT, 1439 CTLTYPE_NODE, "stats", 1440 SYSCTL_DESCR("Debugging options"), 1441 NULL, 0, NULL, 0, 1442 CTL_VFS, 5, LFS_STATS, CTL_EOL); 1443 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 1444 sysctl_createv(clog, 0, NULL, NULL, 1445 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1446 CTLTYPE_INT, stat_names[i].sname, 1447 SYSCTL_DESCR(stat_names[i].lname), 1448 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 1449 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 1450 } 1451 1452 #ifdef DEBUG 1453 sysctl_createv(clog, 0, NULL, NULL, 1454 CTLFLAG_PERMANENT, 1455 CTLTYPE_NODE, "debug", 1456 SYSCTL_DESCR("Debugging options"), 1457 NULL, 0, NULL, 0, 1458 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 1459 for (i = 0; i < DLOG_MAX; i++) { 1460 sysctl_createv(clog, 0, NULL, NULL, 1461 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1462 CTLTYPE_INT, dlog_names[i].sname, 1463 SYSCTL_DESCR(dlog_names[i].lname), 1464 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 1465 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 1466 } 1467 #endif 1468 } 1469 1470 /* 1471 * ufs_bmaparray callback function for writing. 1472 * 1473 * Since blocks will be written to the new segment anyway, 1474 * we don't care about current daddr of them. 1475 */ 1476 static bool 1477 lfs_issequential_hole(const struct ufsmount *ump, 1478 daddr_t daddr0, daddr_t daddr1) 1479 { 1480 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */ 1481 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */ 1482 1483 KASSERT(daddr0 == UNWRITTEN || 1484 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR)); 1485 KASSERT(daddr1 == UNWRITTEN || 1486 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR)); 1487 1488 /* NOTE: all we want to know here is 'hole or not'. */ 1489 /* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */ 1490 1491 /* 1492 * treat UNWRITTENs and all resident blocks as 'contiguous' 1493 */ 1494 if (daddr0 != 0 && daddr1 != 0) 1495 return true; 1496 1497 /* 1498 * both are in hole? 1499 */ 1500 if (daddr0 == 0 && daddr1 == 0) 1501 return true; /* all holes are 'contiguous' for us. */ 1502 1503 return false; 1504 } 1505 1506 /* 1507 * lfs_gop_write functions exactly like genfs_gop_write, except that 1508 * (1) it requires the seglock to be held by its caller, and sp->fip 1509 * to be properly initialized (it will return without re-initializing 1510 * sp->fip, and without calling lfs_writeseg). 1511 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1512 * to determine how large a block it can write at once (though it does 1513 * still use VOP_BMAP to find holes in the file); 1514 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1515 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1516 * now have clusters of clusters, ick.) 1517 */ 1518 static int 1519 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, 1520 int flags) 1521 { 1522 int i, s, error, run, haveeof = 0; 1523 int fs_bshift; 1524 vaddr_t kva; 1525 off_t eof, offset, startoffset = 0; 1526 size_t bytes, iobytes, skipbytes; 1527 daddr_t lbn, blkno; 1528 struct vm_page *pg; 1529 struct buf *mbp, *bp; 1530 struct vnode *devvp = VTOI(vp)->i_devvp; 1531 struct inode *ip = VTOI(vp); 1532 struct lfs *fs = ip->i_lfs; 1533 struct segment *sp = fs->lfs_sp; 1534 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1535 1536 ASSERT_SEGLOCK(fs); 1537 1538 /* The Ifile lives in the buffer cache */ 1539 KASSERT(vp != fs->lfs_ivnode); 1540 1541 /* 1542 * We don't want to fill the disk before the cleaner has a chance 1543 * to make room for us. If we're in danger of doing that, fail 1544 * with EAGAIN. The caller will have to notice this, unlock 1545 * so the cleaner can run, relock and try again. 1546 * 1547 * We must write everything, however, if our vnode is being 1548 * reclaimed. 1549 */ 1550 if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp) 1551 goto tryagain; 1552 1553 /* 1554 * Sometimes things slip past the filters in lfs_putpages, 1555 * and the pagedaemon tries to write pages---problem is 1556 * that the pagedaemon never acquires the segment lock. 1557 * 1558 * Alternatively, pages that were clean when we called 1559 * genfs_putpages may have become dirty in the meantime. In this 1560 * case the segment header is not properly set up for blocks 1561 * to be added to it. 1562 * 1563 * Unbusy and unclean the pages, and put them on the ACTIVE 1564 * queue under the hypothesis that they couldn't have got here 1565 * unless they were modified *quite* recently. 1566 * 1567 * XXXUBC that last statement is an oversimplification of course. 1568 */ 1569 if (!LFS_SEGLOCK_HELD(fs) || 1570 (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) || 1571 (pgs[0]->offset & fs->lfs_bmask) != 0) { 1572 goto tryagain; 1573 } 1574 1575 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x", 1576 vp, pgs, npages, flags); 1577 1578 GOP_SIZE(vp, vp->v_size, &eof, 0); 1579 haveeof = 1; 1580 1581 if (vp->v_type == VREG) 1582 fs_bshift = vp->v_mount->mnt_fs_bshift; 1583 else 1584 fs_bshift = DEV_BSHIFT; 1585 error = 0; 1586 pg = pgs[0]; 1587 startoffset = pg->offset; 1588 KASSERT(eof >= 0); 1589 1590 if (startoffset >= eof) { 1591 goto tryagain; 1592 } else 1593 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 1594 skipbytes = 0; 1595 1596 KASSERT(bytes != 0); 1597 1598 /* Swap PG_DELWRI for PG_PAGEOUT */ 1599 for (i = 0; i < npages; i++) { 1600 if (pgs[i]->flags & PG_DELWRI) { 1601 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 1602 pgs[i]->flags &= ~PG_DELWRI; 1603 pgs[i]->flags |= PG_PAGEOUT; 1604 uvmexp.paging++; 1605 uvm_lock_pageq(); 1606 uvm_pageunwire(pgs[i]); 1607 uvm_unlock_pageq(); 1608 } 1609 } 1610 1611 /* 1612 * Check to make sure we're starting on a block boundary. 1613 * We'll check later to make sure we always write entire 1614 * blocks (or fragments). 1615 */ 1616 if (startoffset & fs->lfs_bmask) 1617 printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n", 1618 startoffset, fs->lfs_bmask, 1619 startoffset & fs->lfs_bmask); 1620 KASSERT((startoffset & fs->lfs_bmask) == 0); 1621 if (bytes & fs->lfs_ffmask) { 1622 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 1623 panic("lfs_gop_write: non-integer blocks"); 1624 } 1625 1626 /* 1627 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 1628 * If we would, write what we have and try again. If we don't 1629 * have anything to write, we'll have to sleep. 1630 */ 1631 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1632 (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ? 1633 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 1634 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 1635 #if 0 1636 " with nfinfo=%d at offset 0x%x\n", 1637 (int)((SEGSUM *)(sp->segsum))->ss_nfinfo, 1638 (unsigned)fs->lfs_offset)); 1639 #endif 1640 lfs_updatemeta(sp); 1641 lfs_release_finfo(fs); 1642 (void) lfs_writeseg(fs, sp); 1643 1644 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 1645 1646 /* 1647 * Having given up all of the pager_map we were holding, 1648 * we can now wait for aiodoned to reclaim it for us 1649 * without fear of deadlock. 1650 */ 1651 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1652 UVMPAGER_MAPIN_WAITOK); 1653 } 1654 1655 s = splbio(); 1656 simple_lock(&global_v_numoutput_slock); 1657 vp->v_numoutput += 2; /* one for biodone, one for aiodone */ 1658 simple_unlock(&global_v_numoutput_slock); 1659 splx(s); 1660 1661 mbp = getiobuf(); 1662 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 1663 vp, mbp, vp->v_numoutput, bytes); 1664 mbp->b_bufsize = npages << PAGE_SHIFT; 1665 mbp->b_data = (void *)kva; 1666 mbp->b_resid = mbp->b_bcount = bytes; 1667 mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL; 1668 mbp->b_iodone = uvm_aio_biodone; 1669 mbp->b_vp = vp; 1670 1671 bp = NULL; 1672 for (offset = startoffset; 1673 bytes > 0; 1674 offset += iobytes, bytes -= iobytes) { 1675 lbn = offset >> fs_bshift; 1676 error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 1677 lfs_issequential_hole); 1678 if (error) { 1679 UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d", 1680 error,0,0,0); 1681 skipbytes += bytes; 1682 bytes = 0; 1683 break; 1684 } 1685 1686 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 1687 bytes); 1688 if (blkno == (daddr_t)-1) { 1689 skipbytes += iobytes; 1690 continue; 1691 } 1692 1693 /* 1694 * Discover how much we can really pack into this buffer. 1695 */ 1696 /* If no room in the current segment, finish it up */ 1697 if (sp->sum_bytes_left < sizeof(int32_t) || 1698 sp->seg_bytes_left < (1 << fs->lfs_bshift)) { 1699 int vers; 1700 1701 lfs_updatemeta(sp); 1702 vers = sp->fip->fi_version; 1703 lfs_release_finfo(fs); 1704 (void) lfs_writeseg(fs, sp); 1705 1706 lfs_acquire_finfo(fs, ip->i_number, vers); 1707 } 1708 /* Check both for space in segment and space in segsum */ 1709 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 1710 << fs_bshift); 1711 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 1712 << fs_bshift); 1713 KASSERT(iobytes > 0); 1714 1715 /* if it's really one i/o, don't make a second buf */ 1716 if (offset == startoffset && iobytes == bytes) { 1717 bp = mbp; 1718 /* correct overcount if there is no second buffer */ 1719 s = splbio(); 1720 simple_lock(&global_v_numoutput_slock); 1721 --vp->v_numoutput; 1722 simple_unlock(&global_v_numoutput_slock); 1723 splx(s); 1724 } else { 1725 bp = getiobuf(); 1726 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 1727 vp, bp, vp->v_numoutput, 0); 1728 bp->b_data = (char *)kva + 1729 (vaddr_t)(offset - pg->offset); 1730 bp->b_resid = bp->b_bcount = iobytes; 1731 bp->b_flags = B_BUSY|B_WRITE|B_CALL; 1732 bp->b_iodone = uvm_aio_biodone1; 1733 } 1734 1735 /* XXX This is silly ... is this necessary? */ 1736 bp->b_vp = NULL; 1737 s = splbio(); 1738 bgetvp(vp, bp); 1739 splx(s); 1740 1741 bp->b_lblkno = lblkno(fs, offset); 1742 bp->b_private = mbp; 1743 if (devvp->v_type == VBLK) { 1744 bp->b_dev = devvp->v_rdev; 1745 } 1746 VOP_BWRITE(bp); 1747 while (lfs_gatherblock(sp, bp, NULL)) 1748 continue; 1749 } 1750 1751 if (skipbytes) { 1752 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0); 1753 s = splbio(); 1754 if (error) { 1755 mbp->b_flags |= B_ERROR; 1756 mbp->b_error = error; 1757 } 1758 mbp->b_resid -= skipbytes; 1759 if (mbp->b_resid == 0) { 1760 biodone(mbp); 1761 } 1762 splx(s); 1763 } 1764 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 1765 return (0); 1766 1767 tryagain: 1768 /* 1769 * We can't write the pages, for whatever reason. 1770 * Clean up after ourselves, and make the caller try again. 1771 */ 1772 simple_lock(&vp->v_interlock); 1773 1774 /* Tell why we're here, if we know */ 1775 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) { 1776 DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n")); 1777 } else if ((pgs[0]->offset & fs->lfs_bmask) != 0) { 1778 DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n")); 1779 } else if (haveeof && startoffset >= eof) { 1780 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 1781 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 1782 pgs[0]->offset, eof, npages)); 1783 } else if (LFS_STARVED_FOR_SEGS(fs)) { 1784 DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n")); 1785 } else { 1786 DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n")); 1787 } 1788 1789 uvm_lock_pageq(); 1790 for (i = 0; i < npages; i++) { 1791 pg = pgs[i]; 1792 1793 if (pg->flags & PG_PAGEOUT) 1794 uvmexp.paging--; 1795 if (pg->flags & PG_DELWRI) { 1796 uvm_pageunwire(pg); 1797 } 1798 uvm_pageactivate(pg); 1799 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 1800 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 1801 vp, pg->offset)); 1802 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 1803 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 1804 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 1805 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 1806 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 1807 pg->wire_count)); 1808 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 1809 pg->loan_count)); 1810 } 1811 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */ 1812 uvm_page_unbusy(pgs, npages); 1813 uvm_unlock_pageq(); 1814 simple_unlock(&vp->v_interlock); 1815 return EAGAIN; 1816 } 1817 1818 /* 1819 * finish vnode/inode initialization. 1820 * used by lfs_vget and lfs_fastvget. 1821 */ 1822 void 1823 lfs_vinit(struct mount *mp, struct vnode **vpp) 1824 { 1825 struct vnode *vp = *vpp; 1826 struct inode *ip = VTOI(vp); 1827 struct ufsmount *ump = VFSTOUFS(mp); 1828 struct lfs *fs = ump->um_lfs; 1829 int i; 1830 1831 ip->i_mode = ip->i_ffs1_mode; 1832 ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink; 1833 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size; 1834 ip->i_flags = ip->i_ffs1_flags; 1835 ip->i_gen = ip->i_ffs1_gen; 1836 ip->i_uid = ip->i_ffs1_uid; 1837 ip->i_gid = ip->i_ffs1_gid; 1838 1839 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 1840 ip->i_lfs_odnlink = ip->i_ffs1_nlink; 1841 1842 /* 1843 * Initialize the vnode from the inode, check for aliases. In all 1844 * cases re-init ip, the underlying vnode/inode may have changed. 1845 */ 1846 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 1847 ip = VTOI(vp); 1848 1849 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 1850 if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) { 1851 #ifdef DEBUG 1852 for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift; 1853 i < NDADDR; i++) { 1854 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 1855 i == 0) 1856 continue; 1857 if (ip->i_ffs1_db[i] != 0) { 1858 inconsistent: 1859 lfs_dump_dinode(ip->i_din.ffs1_din); 1860 panic("inconsistent inode"); 1861 } 1862 } 1863 for ( ; i < NDADDR + NIADDR; i++) { 1864 if (ip->i_ffs1_ib[i - NDADDR] != 0) { 1865 goto inconsistent; 1866 } 1867 } 1868 #endif /* DEBUG */ 1869 for (i = 0; i < NDADDR; i++) 1870 if (ip->i_ffs1_db[i] != 0) 1871 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 1872 } 1873 1874 #ifdef DIAGNOSTIC 1875 if (vp->v_type == VNON) { 1876 # ifdef DEBUG 1877 lfs_dump_dinode(ip->i_din.ffs1_din); 1878 # endif 1879 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 1880 (unsigned long long)ip->i_number, 1881 (ip->i_mode & IFMT) >> 12); 1882 } 1883 #endif /* DIAGNOSTIC */ 1884 1885 /* 1886 * Finish inode initialization now that aliasing has been resolved. 1887 */ 1888 1889 ip->i_devvp = ump->um_devvp; 1890 VREF(ip->i_devvp); 1891 genfs_node_init(vp, &lfs_genfsops); 1892 uvm_vnp_setsize(vp, ip->i_size); 1893 1894 /* Initialize hiblk from file size */ 1895 ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1; 1896 1897 *vpp = vp; 1898 } 1899 1900 /* 1901 * Resize the filesystem to contain the specified number of segments. 1902 */ 1903 int 1904 lfs_resize_fs(struct lfs *fs, int newnsegs) 1905 { 1906 SEGUSE *sup; 1907 struct buf *bp, *obp; 1908 daddr_t olast, nlast, ilast, noff, start, end; 1909 struct vnode *ivp; 1910 struct inode *ip; 1911 int error, badnews, inc, oldnsegs; 1912 int sbbytes, csbbytes, gain, cgain; 1913 int i; 1914 1915 /* Only support v2 and up */ 1916 if (fs->lfs_version < 2) 1917 return EOPNOTSUPP; 1918 1919 /* If we're doing nothing, do it fast */ 1920 oldnsegs = fs->lfs_nseg; 1921 if (newnsegs == oldnsegs) 1922 return 0; 1923 1924 /* We always have to have two superblocks */ 1925 if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1])) 1926 return EFBIG; 1927 1928 ivp = fs->lfs_ivnode; 1929 ip = VTOI(ivp); 1930 error = 0; 1931 1932 /* Take the segment lock so no one else calls lfs_newseg() */ 1933 lfs_seglock(fs, SEGM_PROT); 1934 1935 /* 1936 * Make sure the segments we're going to be losing, if any, 1937 * are in fact empty. We hold the seglock, so their status 1938 * cannot change underneath us. Count the superblocks we lose, 1939 * while we're at it. 1940 */ 1941 sbbytes = csbbytes = 0; 1942 cgain = 0; 1943 for (i = newnsegs; i < oldnsegs; i++) { 1944 LFS_SEGENTRY(sup, fs, i, bp); 1945 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 1946 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1947 sbbytes += LFS_SBPAD; 1948 if (!(sup->su_flags & SEGUSE_DIRTY)) { 1949 ++cgain; 1950 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1951 csbbytes += LFS_SBPAD; 1952 } 1953 brelse(bp); 1954 if (badnews) { 1955 error = EBUSY; 1956 goto out; 1957 } 1958 } 1959 1960 /* Note old and new segment table endpoints, and old ifile size */ 1961 olast = fs->lfs_cleansz + fs->lfs_segtabsz; 1962 nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz; 1963 ilast = ivp->v_size >> fs->lfs_bshift; 1964 noff = nlast - olast; 1965 1966 /* 1967 * Make sure no one can use the Ifile while we change it around. 1968 * Even after taking the iflock we need to make sure no one still 1969 * is holding Ifile buffers, so we get each one, to drain them. 1970 * (XXX this could be done better.) 1971 */ 1972 simple_lock(&fs->lfs_interlock); 1973 lockmgr(&fs->lfs_iflock, LK_EXCLUSIVE, &fs->lfs_interlock); 1974 simple_unlock(&fs->lfs_interlock); 1975 vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY); 1976 for (i = 0; i < ilast; i++) { 1977 bread(ivp, i, fs->lfs_bsize, NOCRED, &bp); 1978 brelse(bp); 1979 } 1980 1981 /* Allocate new Ifile blocks */ 1982 for (i = ilast; i < ilast + noff; i++) { 1983 if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0, 1984 &bp) != 0) 1985 panic("balloc extending ifile"); 1986 memset(bp->b_data, 0, fs->lfs_bsize); 1987 VOP_BWRITE(bp); 1988 } 1989 1990 /* Register new ifile size */ 1991 ip->i_size += noff * fs->lfs_bsize; 1992 ip->i_ffs1_size = ip->i_size; 1993 uvm_vnp_setsize(ivp, ip->i_size); 1994 1995 /* Copy the inode table to its new position */ 1996 if (noff != 0) { 1997 if (noff < 0) { 1998 start = nlast; 1999 end = ilast + noff; 2000 inc = 1; 2001 } else { 2002 start = ilast + noff - 1; 2003 end = nlast - 1; 2004 inc = -1; 2005 } 2006 for (i = start; i != end; i += inc) { 2007 if (bread(ivp, i, fs->lfs_bsize, NOCRED, &bp) != 0) 2008 panic("resize: bread dst blk failed"); 2009 if (bread(ivp, i - noff, fs->lfs_bsize, NOCRED, &obp)) 2010 panic("resize: bread src blk failed"); 2011 memcpy(bp->b_data, obp->b_data, fs->lfs_bsize); 2012 VOP_BWRITE(bp); 2013 brelse(obp); 2014 } 2015 } 2016 2017 /* If we are expanding, write the new empty SEGUSE entries */ 2018 if (newnsegs > oldnsegs) { 2019 for (i = oldnsegs; i < newnsegs; i++) { 2020 if ((error = bread(ivp, i / fs->lfs_sepb + 2021 fs->lfs_cleansz, 2022 fs->lfs_bsize, NOCRED, &bp)) != 0) 2023 panic("lfs: ifile read: %d", error); 2024 while ((i + 1) % fs->lfs_sepb && i < newnsegs) { 2025 sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb]; 2026 memset(sup, 0, sizeof(*sup)); 2027 i++; 2028 } 2029 VOP_BWRITE(bp); 2030 } 2031 } 2032 2033 /* Zero out unused superblock offsets */ 2034 for (i = 2; i < LFS_MAXNUMSB; i++) 2035 if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs) 2036 fs->lfs_sboffs[i] = 0x0; 2037 2038 /* 2039 * Correct superblock entries that depend on fs size. 2040 * The computations of these are as follows: 2041 * 2042 * size = segtod(fs, nseg) 2043 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD) 2044 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2045 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD) 2046 * + (segtod(fs, 1) - (offset - curseg)) 2047 * - segtod(fs, minfreeseg - (minfreeseg / 2)) 2048 * 2049 * XXX - we should probably adjust minfreeseg as well. 2050 */ 2051 gain = (newnsegs - oldnsegs); 2052 fs->lfs_nseg = newnsegs; 2053 fs->lfs_segtabsz = nlast - fs->lfs_cleansz; 2054 fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize); 2055 fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes); 2056 fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes) 2057 - gain * btofsb(fs, fs->lfs_bsize / 2); 2058 if (gain > 0) { 2059 fs->lfs_nclean += gain; 2060 fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize); 2061 } else { 2062 fs->lfs_nclean -= cgain; 2063 fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) - 2064 btofsb(fs, csbbytes); 2065 } 2066 2067 /* Resize segment flag cache */ 2068 fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2069 fs->lfs_nseg * sizeof(u_int32_t), 2070 M_SEGMENT, M_WAITOK); 2071 fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[1], 2072 fs->lfs_nseg * sizeof(u_int32_t), 2073 M_SEGMENT, M_WAITOK); 2074 for (i = oldnsegs; i < newnsegs; i++) 2075 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2076 2077 /* Truncate Ifile if necessary */ 2078 if (noff < 0) 2079 lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0, 2080 NOCRED, curlwp); 2081 2082 /* Update cleaner info so the cleaner can die */ 2083 bread(ivp, 0, fs->lfs_bsize, NOCRED, &bp); 2084 ((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean; 2085 ((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean; 2086 VOP_BWRITE(bp); 2087 2088 /* Let Ifile accesses proceed */ 2089 VOP_UNLOCK(ivp, 0); 2090 simple_lock(&fs->lfs_interlock); 2091 lockmgr(&fs->lfs_iflock, LK_RELEASE, &fs->lfs_interlock); 2092 simple_unlock(&fs->lfs_interlock); 2093 2094 out: 2095 lfs_segunlock(fs); 2096 return error; 2097 } 2098
Indexes created Tue Sep 30 11:09:46 GMT 2025