1 /* $NetBSD: lfs_vfsops.c,v 1.223 2006/10/12 01:32:51 christos Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 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. 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 1.223 2006/10/12 01:32:51 christos 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 boolean_t 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 lfs_vnodeopv_descs, 160 0, 161 { NULL, NULL }, 162 }; 163 VFS_ATTACH(lfs_vfsops); 164 165 const struct genfs_ops lfs_genfsops = { 166 .gop_size = lfs_gop_size, 167 .gop_alloc = ufs_gop_alloc, 168 .gop_write = lfs_gop_write, 169 .gop_markupdate = ufs_gop_markupdate, 170 }; 171 172 static const struct ufs_ops lfs_ufsops = { 173 .uo_itimes = NULL, 174 .uo_update = lfs_update, 175 .uo_truncate = lfs_truncate, 176 .uo_valloc = lfs_valloc, 177 .uo_vfree = lfs_vfree, 178 .uo_balloc = lfs_balloc, 179 }; 180 181 /* 182 * XXX Same structure as FFS inodes? Should we share a common pool? 183 */ 184 POOL_INIT(lfs_inode_pool, sizeof(struct inode), 0, 0, 0, "lfsinopl", 185 &pool_allocator_nointr); 186 POOL_INIT(lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, "lfsdinopl", 187 &pool_allocator_nointr); 188 POOL_INIT(lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, "lfsinoextpl", 189 &pool_allocator_nointr); 190 POOL_INIT(lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, "lfslbnpool", 191 &pool_allocator_nointr); 192 193 /* 194 * The writer daemon. UVM keeps track of how many dirty pages we are holding 195 * in lfs_subsys_pages; the daemon flushes the filesystem when this value 196 * crosses the (user-defined) threshhold LFS_MAX_PAGES. 197 */ 198 static void 199 lfs_writerd(void *arg __unused) 200 { 201 struct mount *mp, *nmp; 202 struct lfs *fs; 203 int loopcount; 204 205 lfs_writer_daemon = curproc->p_pid; 206 207 simple_lock(&lfs_subsys_lock); 208 for (;;) { 209 ltsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10, 210 &lfs_subsys_lock); 211 212 /* 213 * Look through the list of LFSs to see if any of them 214 * have requested pageouts. 215 */ 216 simple_lock(&mountlist_slock); 217 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 218 mp = nmp) { 219 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 220 nmp = CIRCLEQ_NEXT(mp, mnt_list); 221 continue; 222 } 223 if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS, 224 MFSNAMELEN) == 0) { 225 fs = VFSTOUFS(mp)->um_lfs; 226 simple_lock(&fs->lfs_interlock); 227 if (fs->lfs_pdflush) { 228 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n")); 229 fs->lfs_pdflush = 0; 230 lfs_flush_fs(fs, 0); 231 simple_unlock(&fs->lfs_interlock); 232 } else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) { 233 DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n")); 234 simple_unlock(&fs->lfs_interlock); 235 lfs_writer_enter(fs, "wrdirop"); 236 lfs_flush_pchain(fs); 237 lfs_writer_leave(fs); 238 } else 239 simple_unlock(&fs->lfs_interlock); 240 } 241 242 simple_lock(&mountlist_slock); 243 nmp = CIRCLEQ_NEXT(mp, mnt_list); 244 vfs_unbusy(mp); 245 } 246 simple_unlock(&mountlist_slock); 247 248 /* 249 * If global state wants a flush, flush everything. 250 */ 251 simple_lock(&lfs_subsys_lock); 252 loopcount = 0; 253 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS || 254 locked_queue_bytes > LFS_MAX_BYTES || 255 lfs_subsys_pages > LFS_MAX_PAGES) { 256 257 if (lfs_do_flush) { 258 DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n")); 259 } 260 if (locked_queue_count > LFS_MAX_BUFS) { 261 DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n", 262 locked_queue_count, LFS_MAX_BUFS)); 263 } 264 if (locked_queue_bytes > LFS_MAX_BYTES) { 265 DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n", 266 locked_queue_bytes, LFS_MAX_BYTES)); 267 } 268 if (lfs_subsys_pages > LFS_MAX_PAGES) { 269 DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n", 270 lfs_subsys_pages, LFS_MAX_PAGES)); 271 } 272 273 lfs_flush(NULL, SEGM_WRITERD, 0); 274 lfs_do_flush = 0; 275 } 276 } 277 /* NOTREACHED */ 278 } 279 280 /* 281 * Initialize the filesystem, most work done by ufs_init. 282 */ 283 void 284 lfs_init() 285 { 286 #ifdef _LKM 287 malloc_type_attach(M_SEGMENT); 288 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 289 "lfsinopl", &pool_allocator_nointr); 290 pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, 291 "lfsdinopl", &pool_allocator_nointr); 292 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 293 "lfsinoextpl", &pool_allocator_nointr); 294 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 295 "lfslbnpool", &pool_allocator_nointr); 296 #endif 297 ufs_init(); 298 299 #ifdef DEBUG 300 memset(lfs_log, 0, sizeof(lfs_log)); 301 #endif 302 simple_lock_init(&lfs_subsys_lock); 303 } 304 305 void 306 lfs_reinit() 307 { 308 ufs_reinit(); 309 } 310 311 void 312 lfs_done() 313 { 314 ufs_done(); 315 #ifdef _LKM 316 pool_destroy(&lfs_inode_pool); 317 pool_destroy(&lfs_dinode_pool); 318 pool_destroy(&lfs_inoext_pool); 319 pool_destroy(&lfs_lbnentry_pool); 320 malloc_type_detach(M_SEGMENT); 321 #endif 322 } 323 324 /* 325 * Called by main() when ufs is going to be mounted as root. 326 */ 327 int 328 lfs_mountroot() 329 { 330 extern struct vnode *rootvp; 331 struct mount *mp; 332 struct lwp *l = curlwp; /* XXX */ 333 int error; 334 335 if (device_class(root_device) != DV_DISK) 336 return (ENODEV); 337 338 if (rootdev == NODEV) 339 return (ENODEV); 340 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 341 vrele(rootvp); 342 return (error); 343 } 344 if ((error = lfs_mountfs(rootvp, mp, l))) { 345 mp->mnt_op->vfs_refcount--; 346 vfs_unbusy(mp); 347 free(mp, M_MOUNT); 348 return (error); 349 } 350 simple_lock(&mountlist_slock); 351 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 352 simple_unlock(&mountlist_slock); 353 (void)lfs_statvfs(mp, &mp->mnt_stat, l); 354 vfs_unbusy(mp); 355 setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp)); 356 return (0); 357 } 358 359 /* 360 * VFS Operations. 361 * 362 * mount system call 363 */ 364 int 365 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct lwp *l) 366 { 367 struct vnode *devvp; 368 struct ufs_args args; 369 struct ufsmount *ump = NULL; 370 struct lfs *fs = NULL; /* LFS */ 371 int error, update; 372 mode_t accessmode; 373 374 if (mp->mnt_flag & MNT_GETARGS) { 375 ump = VFSTOUFS(mp); 376 if (ump == NULL) 377 return EIO; 378 args.fspec = NULL; 379 return copyout(&args, data, sizeof(args)); 380 } 381 error = copyin(data, &args, sizeof (struct ufs_args)); 382 if (error) 383 return (error); 384 385 update = mp->mnt_flag & MNT_UPDATE; 386 387 /* Check arguments */ 388 if (args.fspec != NULL) { 389 /* 390 * Look up the name and verify that it's sane. 391 */ 392 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, l); 393 if ((error = namei(ndp)) != 0) 394 return (error); 395 devvp = ndp->ni_vp; 396 397 if (!update) { 398 /* 399 * Be sure this is a valid block device 400 */ 401 if (devvp->v_type != VBLK) 402 error = ENOTBLK; 403 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 404 error = ENXIO; 405 } else { 406 /* 407 * Be sure we're still naming the same device 408 * used for our initial mount 409 */ 410 ump = VFSTOUFS(mp); 411 if (devvp != ump->um_devvp) 412 error = EINVAL; 413 } 414 } else { 415 if (!update) { 416 /* New mounts must have a filename for the device */ 417 return (EINVAL); 418 } else { 419 /* Use the extant mount */ 420 ump = VFSTOUFS(mp); 421 devvp = ump->um_devvp; 422 vref(devvp); 423 } 424 } 425 426 427 /* 428 * If mount by non-root, then verify that user has necessary 429 * permissions on the device. 430 */ 431 if (error == 0 && kauth_cred_geteuid(l->l_cred) != 0) { 432 accessmode = VREAD; 433 if (update ? 434 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 435 (mp->mnt_flag & MNT_RDONLY) == 0) 436 accessmode |= VWRITE; 437 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 438 error = VOP_ACCESS(devvp, accessmode, l->l_cred, l); 439 VOP_UNLOCK(devvp, 0); 440 } 441 442 if (error) { 443 vrele(devvp); 444 return (error); 445 } 446 447 if (!update) { 448 int flags; 449 450 /* 451 * Disallow multiple mounts of the same device. 452 * Disallow mounting of a device that is currently in use 453 * (except for root, which might share swap device for 454 * miniroot). 455 */ 456 error = vfs_mountedon(devvp); 457 if (error) 458 goto fail; 459 if (vcount(devvp) > 1 && devvp != rootvp) { 460 error = EBUSY; 461 goto fail; 462 } 463 if (mp->mnt_flag & MNT_RDONLY) 464 flags = FREAD; 465 else 466 flags = FREAD|FWRITE; 467 error = VOP_OPEN(devvp, flags, FSCRED, l); 468 if (error) 469 goto fail; 470 error = lfs_mountfs(devvp, mp, l); /* LFS */ 471 if (error) { 472 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 473 (void)VOP_CLOSE(devvp, flags, NOCRED, l); 474 VOP_UNLOCK(devvp, 0); 475 goto fail; 476 } 477 478 ump = VFSTOUFS(mp); 479 fs = ump->um_lfs; 480 } else { 481 /* 482 * Update the mount. 483 */ 484 485 /* 486 * The initial mount got a reference on this 487 * device, so drop the one obtained via 488 * namei(), above. 489 */ 490 vrele(devvp); 491 492 ump = VFSTOUFS(mp); 493 fs = ump->um_lfs; 494 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 495 /* 496 * Changing from read-only to read/write. 497 * Note in the superblocks that we're writing. 498 */ 499 fs->lfs_ronly = 0; 500 if (fs->lfs_pflags & LFS_PF_CLEAN) { 501 fs->lfs_pflags &= ~LFS_PF_CLEAN; 502 lfs_writesuper(fs, fs->lfs_sboffs[0]); 503 lfs_writesuper(fs, fs->lfs_sboffs[1]); 504 } 505 } 506 if (args.fspec == NULL) 507 return EINVAL; 508 } 509 510 error = set_statvfs_info(path, UIO_USERSPACE, args.fspec, 511 UIO_USERSPACE, mp, l); 512 if (error == 0) 513 (void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname, 514 sizeof(fs->lfs_fsmnt)); 515 return error; 516 517 fail: 518 vrele(devvp); 519 return (error); 520 } 521 522 523 /* 524 * Common code for mount and mountroot 525 * LFS specific 526 */ 527 int 528 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 529 { 530 struct dlfs *tdfs, *dfs, *adfs; 531 struct lfs *fs; 532 struct ufsmount *ump; 533 struct vnode *vp; 534 struct buf *bp, *abp; 535 struct partinfo dpart; 536 dev_t dev; 537 int error, i, ronly, secsize, fsbsize; 538 kauth_cred_t cred; 539 CLEANERINFO *cip; 540 SEGUSE *sup; 541 daddr_t sb_addr; 542 543 cred = l ? l->l_cred : NOCRED; 544 545 /* 546 * Flush out any old buffers remaining from a previous use. 547 */ 548 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 549 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 550 VOP_UNLOCK(devvp, 0); 551 if (error) 552 return (error); 553 554 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 555 if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, l) != 0) 556 secsize = DEV_BSIZE; 557 else 558 secsize = dpart.disklab->d_secsize; 559 560 /* Don't free random space on error. */ 561 bp = NULL; 562 abp = NULL; 563 ump = NULL; 564 565 sb_addr = LFS_LABELPAD / secsize; 566 while (1) { 567 /* Read in the superblock. */ 568 error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp); 569 if (error) 570 goto out; 571 dfs = (struct dlfs *)bp->b_data; 572 573 /* Check the basics. */ 574 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE || 575 dfs->dlfs_version > LFS_VERSION || 576 dfs->dlfs_bsize < sizeof(struct dlfs)) { 577 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 578 error = EINVAL; /* XXX needs translation */ 579 goto out; 580 } 581 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) { 582 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 583 dfs->dlfs_inodefmt)); 584 error = EINVAL; 585 goto out; 586 } 587 588 if (dfs->dlfs_version == 1) 589 fsbsize = secsize; 590 else { 591 fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb + 592 dfs->dlfs_fsbtodb); 593 /* 594 * Could be, if the frag size is large enough, that we 595 * don't have the "real" primary superblock. If that's 596 * the case, get the real one, and try again. 597 */ 598 if (sb_addr != dfs->dlfs_sboffs[0] << 599 dfs->dlfs_fsbtodb) { 600 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 601 " 0x%llx is not right, trying 0x%llx\n", 602 (long long)sb_addr, 603 (long long)(dfs->dlfs_sboffs[0] << 604 dfs->dlfs_fsbtodb))); 605 sb_addr = dfs->dlfs_sboffs[0] << 606 dfs->dlfs_fsbtodb; 607 brelse(bp); 608 continue; 609 } 610 } 611 break; 612 } 613 614 /* 615 * Check the second superblock to see which is newer; then mount 616 * using the older of the two. This is necessary to ensure that 617 * the filesystem is valid if it was not unmounted cleanly. 618 */ 619 620 if (dfs->dlfs_sboffs[1] && 621 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 622 { 623 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize), 624 LFS_SBPAD, cred, &abp); 625 if (error) 626 goto out; 627 adfs = (struct dlfs *)abp->b_data; 628 629 if (dfs->dlfs_version == 1) { 630 /* 1s resolution comparison */ 631 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 632 tdfs = adfs; 633 else 634 tdfs = dfs; 635 } else { 636 /* monotonic infinite-resolution comparison */ 637 if (adfs->dlfs_serial < dfs->dlfs_serial) 638 tdfs = adfs; 639 else 640 tdfs = dfs; 641 } 642 643 /* Check the basics. */ 644 if (tdfs->dlfs_magic != LFS_MAGIC || 645 tdfs->dlfs_bsize > MAXBSIZE || 646 tdfs->dlfs_version > LFS_VERSION || 647 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 648 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 649 " sanity failed\n")); 650 error = EINVAL; /* XXX needs translation */ 651 goto out; 652 } 653 } else { 654 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock" 655 " daddr=0x%x\n", dfs->dlfs_sboffs[1])); 656 error = EINVAL; 657 goto out; 658 } 659 660 /* Allocate the mount structure, copy the superblock into it. */ 661 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO); 662 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 663 664 /* Compatibility */ 665 if (fs->lfs_version < 2) { 666 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 667 fs->lfs_ibsize = fs->lfs_bsize; 668 fs->lfs_start = fs->lfs_sboffs[0]; 669 fs->lfs_tstamp = fs->lfs_otstamp; 670 fs->lfs_fsbtodb = 0; 671 } 672 if (fs->lfs_resvseg == 0) 673 fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \ 674 MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1)); 675 676 /* 677 * If we aren't going to be able to write meaningfully to this 678 * filesystem, and were not mounted readonly, bomb out now. 679 */ 680 if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 681 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 682 " we need BUFPAGES >= %lld\n", 683 (long long)((bufmem_hiwater / bufmem_lowater) * 684 LFS_INVERSE_MAX_BYTES( 685 fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 686 free(fs, M_UFSMNT); 687 error = EFBIG; /* XXX needs translation */ 688 goto out; 689 } 690 691 /* Before rolling forward, lock so vget will sleep for other procs */ 692 if (l != NULL) { 693 fs->lfs_flags = LFS_NOTYET; 694 fs->lfs_rfpid = l->l_proc->p_pid; 695 } 696 697 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 698 ump->um_lfs = fs; 699 ump->um_ops = &lfs_ufsops; 700 ump->um_fstype = UFS1; 701 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 702 bp->b_flags |= B_INVAL; 703 abp->b_flags |= B_INVAL; 704 } 705 brelse(bp); 706 bp = NULL; 707 brelse(abp); 708 abp = NULL; 709 710 /* Set up the I/O information */ 711 fs->lfs_devbsize = secsize; 712 fs->lfs_iocount = 0; 713 fs->lfs_diropwait = 0; 714 fs->lfs_activesb = 0; 715 fs->lfs_uinodes = 0; 716 fs->lfs_ravail = 0; 717 fs->lfs_favail = 0; 718 fs->lfs_sbactive = 0; 719 720 /* Set up the ifile and lock aflags */ 721 fs->lfs_doifile = 0; 722 fs->lfs_writer = 0; 723 fs->lfs_dirops = 0; 724 fs->lfs_nadirop = 0; 725 fs->lfs_seglock = 0; 726 fs->lfs_pdflush = 0; 727 fs->lfs_sleepers = 0; 728 fs->lfs_pages = 0; 729 simple_lock_init(&fs->lfs_interlock); 730 lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0); 731 lockinit(&fs->lfs_iflock, PINOD, "lfs_iflock", 0, 0); 732 lockinit(&fs->lfs_stoplock, PINOD, "lfs_stoplock", 0, 0); 733 734 /* Set the file system readonly/modify bits. */ 735 fs->lfs_ronly = ronly; 736 if (ronly == 0) 737 fs->lfs_fmod = 1; 738 739 /* Initialize the mount structure. */ 740 dev = devvp->v_rdev; 741 mp->mnt_data = ump; 742 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 743 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 744 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 745 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 746 mp->mnt_stat.f_iosize = fs->lfs_bsize; 747 mp->mnt_flag |= MNT_LOCAL; 748 mp->mnt_fs_bshift = fs->lfs_bshift; 749 ump->um_flags = 0; 750 ump->um_mountp = mp; 751 ump->um_dev = dev; 752 ump->um_devvp = devvp; 753 ump->um_bptrtodb = fs->lfs_fsbtodb; 754 ump->um_seqinc = fragstofsb(fs, fs->lfs_frag); 755 ump->um_nindir = fs->lfs_nindir; 756 ump->um_lognindir = ffs(fs->lfs_nindir) - 1; 757 for (i = 0; i < MAXQUOTAS; i++) 758 ump->um_quotas[i] = NULLVP; 759 ump->um_maxsymlinklen = fs->lfs_maxsymlinklen; 760 ump->um_dirblksiz = DIRBLKSIZ; 761 ump->um_maxfilesize = fs->lfs_maxfilesize; 762 if (ump->um_maxsymlinklen > 0) 763 mp->mnt_iflag |= IMNT_DTYPE; 764 devvp->v_specmountpoint = mp; 765 766 /* Set up reserved memory for pageout */ 767 lfs_setup_resblks(fs); 768 /* Set up vdirop tailq */ 769 TAILQ_INIT(&fs->lfs_dchainhd); 770 /* and paging tailq */ 771 TAILQ_INIT(&fs->lfs_pchainhd); 772 /* and delayed segment accounting for truncation list */ 773 LIST_INIT(&fs->lfs_segdhd); 774 775 /* 776 * We use the ifile vnode for almost every operation. Instead of 777 * retrieving it from the hash table each time we retrieve it here, 778 * artificially increment the reference count and keep a pointer 779 * to it in the incore copy of the superblock. 780 */ 781 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) { 782 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 783 goto out; 784 } 785 fs->lfs_ivnode = vp; 786 VREF(vp); 787 788 /* Set up inode bitmap and order free list */ 789 lfs_order_freelist(fs); 790 791 /* Set up segment usage flags for the autocleaner. */ 792 fs->lfs_nactive = 0; 793 fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *), 794 M_SEGMENT, M_WAITOK); 795 fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 796 M_SEGMENT, M_WAITOK); 797 fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 798 M_SEGMENT, M_WAITOK); 799 memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t)); 800 for (i = 0; i < fs->lfs_nseg; i++) { 801 int changed; 802 803 LFS_SEGENTRY(sup, fs, i, bp); 804 changed = 0; 805 if (!ronly) { 806 if (sup->su_nbytes == 0 && 807 !(sup->su_flags & SEGUSE_EMPTY)) { 808 sup->su_flags |= SEGUSE_EMPTY; 809 ++changed; 810 } else if (!(sup->su_nbytes == 0) && 811 (sup->su_flags & SEGUSE_EMPTY)) { 812 sup->su_flags &= ~SEGUSE_EMPTY; 813 ++changed; 814 } 815 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 816 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 817 ++changed; 818 } 819 } 820 fs->lfs_suflags[0][i] = sup->su_flags; 821 if (changed) 822 LFS_WRITESEGENTRY(sup, fs, i, bp); 823 else 824 brelse(bp); 825 } 826 827 #ifdef LFS_KERNEL_RFW 828 lfs_roll_forward(fs, mp, l); 829 #endif 830 831 /* If writing, sb is not clean; record in case of immediate crash */ 832 if (!fs->lfs_ronly) { 833 fs->lfs_pflags &= ~LFS_PF_CLEAN; 834 lfs_writesuper(fs, fs->lfs_sboffs[0]); 835 lfs_writesuper(fs, fs->lfs_sboffs[1]); 836 } 837 838 /* Allow vget now that roll-forward is complete */ 839 fs->lfs_flags &= ~(LFS_NOTYET); 840 wakeup(&fs->lfs_flags); 841 842 /* 843 * Initialize the ifile cleaner info with information from 844 * the superblock. 845 */ 846 LFS_CLEANERINFO(cip, fs, bp); 847 cip->clean = fs->lfs_nclean; 848 cip->dirty = fs->lfs_nseg - fs->lfs_nclean; 849 cip->avail = fs->lfs_avail; 850 cip->bfree = fs->lfs_bfree; 851 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 852 853 /* 854 * Mark the current segment as ACTIVE, since we're going to 855 * be writing to it. 856 */ 857 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); 858 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 859 fs->lfs_nactive++; 860 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); /* Ifile */ 861 862 /* Now that roll-forward is done, unlock the Ifile */ 863 vput(vp); 864 865 /* Start the pagedaemon-anticipating daemon */ 866 if (lfs_writer_daemon == 0 && 867 kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0) 868 panic("fork lfs_writer"); 869 870 return (0); 871 872 out: 873 if (bp) 874 brelse(bp); 875 if (abp) 876 brelse(abp); 877 if (ump) { 878 free(ump->um_lfs, M_UFSMNT); 879 free(ump, M_UFSMNT); 880 mp->mnt_data = NULL; 881 } 882 883 return (error); 884 } 885 886 /* 887 * unmount system call 888 */ 889 int 890 lfs_unmount(struct mount *mp, int mntflags, struct lwp *l) 891 { 892 struct ufsmount *ump; 893 struct lfs *fs; 894 int error, flags, ronly; 895 int s; 896 897 flags = 0; 898 if (mntflags & MNT_FORCE) 899 flags |= FORCECLOSE; 900 901 ump = VFSTOUFS(mp); 902 fs = ump->um_lfs; 903 904 /* Two checkpoints */ 905 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 906 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 907 908 /* wake up the cleaner so it can die */ 909 lfs_wakeup_cleaner(fs); 910 simple_lock(&fs->lfs_interlock); 911 while (fs->lfs_sleepers) 912 ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0, 913 &fs->lfs_interlock); 914 simple_unlock(&fs->lfs_interlock); 915 916 #ifdef QUOTA 917 if (mp->mnt_flag & MNT_QUOTA) { 918 int i; 919 error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags); 920 if (error) 921 return (error); 922 for (i = 0; i < MAXQUOTAS; i++) { 923 if (ump->um_quotas[i] == NULLVP) 924 continue; 925 quotaoff(l, mp, i); 926 } 927 /* 928 * Here we fall through to vflush again to ensure 929 * that we have gotten rid of all the system vnodes. 930 */ 931 } 932 #endif 933 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 934 return (error); 935 if ((error = VFS_SYNC(mp, 1, l->l_cred, l)) != 0) 936 return (error); 937 s = splbio(); 938 if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd)) 939 panic("lfs_unmount: still dirty blocks on ifile vnode"); 940 splx(s); 941 942 /* Explicitly write the superblock, to update serial and pflags */ 943 fs->lfs_pflags |= LFS_PF_CLEAN; 944 lfs_writesuper(fs, fs->lfs_sboffs[0]); 945 lfs_writesuper(fs, fs->lfs_sboffs[1]); 946 simple_lock(&fs->lfs_interlock); 947 while (fs->lfs_iocount) 948 ltsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 949 &fs->lfs_interlock); 950 simple_unlock(&fs->lfs_interlock); 951 952 /* Finish with the Ifile, now that we're done with it */ 953 vrele(fs->lfs_ivnode); 954 vgone(fs->lfs_ivnode); 955 956 ronly = !fs->lfs_ronly; 957 if (ump->um_devvp->v_type != VBAD) 958 ump->um_devvp->v_specmountpoint = NULL; 959 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 960 error = VOP_CLOSE(ump->um_devvp, 961 ronly ? FREAD : FREAD|FWRITE, NOCRED, l); 962 vput(ump->um_devvp); 963 964 /* Complain about page leakage */ 965 if (fs->lfs_pages > 0) 966 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 967 fs->lfs_pages, lfs_subsys_pages); 968 969 /* Free per-mount data structures */ 970 free(fs->lfs_ino_bitmap, M_SEGMENT); 971 free(fs->lfs_suflags[0], M_SEGMENT); 972 free(fs->lfs_suflags[1], M_SEGMENT); 973 free(fs->lfs_suflags, M_SEGMENT); 974 lfs_free_resblks(fs); 975 free(fs, M_UFSMNT); 976 free(ump, M_UFSMNT); 977 978 mp->mnt_data = NULL; 979 mp->mnt_flag &= ~MNT_LOCAL; 980 return (error); 981 } 982 983 /* 984 * Get file system statistics. 985 * 986 * NB: We don't lock to access the superblock here, because it's not 987 * really that important if we get it wrong. 988 */ 989 int 990 lfs_statvfs(struct mount *mp, struct statvfs *sbp, struct lwp *l __unused) 991 { 992 struct lfs *fs; 993 struct ufsmount *ump; 994 995 ump = VFSTOUFS(mp); 996 fs = ump->um_lfs; 997 if (fs->lfs_magic != LFS_MAGIC) 998 panic("lfs_statvfs: magic"); 999 1000 sbp->f_bsize = fs->lfs_bsize; 1001 sbp->f_frsize = fs->lfs_fsize; 1002 sbp->f_iosize = fs->lfs_bsize; 1003 sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks); 1004 1005 sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs)); 1006 KASSERT(sbp->f_bfree <= fs->lfs_dsize); 1007 #if 0 1008 if (sbp->f_bfree < 0) 1009 sbp->f_bfree = 0; 1010 #endif 1011 1012 sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs)); 1013 if (sbp->f_bfree > sbp->f_bresvd) 1014 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1015 else 1016 sbp->f_bavail = 0; 1017 1018 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs); 1019 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles; 1020 sbp->f_favail = sbp->f_ffree; 1021 sbp->f_fresvd = 0; 1022 copy_statvfs_info(sbp, mp); 1023 return (0); 1024 } 1025 1026 /* 1027 * Go through the disk queues to initiate sandbagged IO; 1028 * go through the inodes to write those that have been modified; 1029 * initiate the writing of the super block if it has been modified. 1030 * 1031 * Note: we are always called with the filesystem marked `MPBUSY'. 1032 */ 1033 int 1034 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred __unused, 1035 struct lwp *l __unused) 1036 { 1037 int error; 1038 struct lfs *fs; 1039 1040 fs = VFSTOUFS(mp)->um_lfs; 1041 if (fs->lfs_ronly) 1042 return 0; 1043 1044 /* Snapshots should not hose the syncer */ 1045 /* 1046 * XXX Sync can block here anyway, since we don't have a very 1047 * XXX good idea of how much data is pending. If it's more 1048 * XXX than a segment and lfs_nextseg is close to the end of 1049 * XXX the log, we'll likely block. 1050 */ 1051 simple_lock(&fs->lfs_interlock); 1052 if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) { 1053 simple_unlock(&fs->lfs_interlock); 1054 return 0; 1055 } 1056 simple_unlock(&fs->lfs_interlock); 1057 1058 lfs_writer_enter(fs, "lfs_dirops"); 1059 1060 /* All syncs must be checkpoints until roll-forward is implemented. */ 1061 DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset)); 1062 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1063 lfs_writer_leave(fs); 1064 #ifdef QUOTA 1065 qsync(mp); 1066 #endif 1067 return (error); 1068 } 1069 1070 extern struct lock ufs_hashlock; 1071 1072 /* 1073 * Look up an LFS dinode number to find its incore vnode. If not already 1074 * in core, read it in from the specified device. Return the inode locked. 1075 * Detection and handling of mount points must be done by the calling routine. 1076 */ 1077 int 1078 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1079 { 1080 struct lfs *fs; 1081 struct ufs1_dinode *dip; 1082 struct inode *ip; 1083 struct buf *bp; 1084 struct ifile *ifp; 1085 struct vnode *vp; 1086 struct ufsmount *ump; 1087 daddr_t daddr; 1088 dev_t dev; 1089 int error, retries; 1090 struct timespec ts; 1091 1092 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1093 1094 ump = VFSTOUFS(mp); 1095 dev = ump->um_dev; 1096 fs = ump->um_lfs; 1097 1098 /* 1099 * If the filesystem is not completely mounted yet, suspend 1100 * any access requests (wait for roll-forward to complete). 1101 */ 1102 simple_lock(&fs->lfs_interlock); 1103 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1104 ltsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1105 &fs->lfs_interlock); 1106 simple_unlock(&fs->lfs_interlock); 1107 1108 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) 1109 return (0); 1110 1111 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) { 1112 *vpp = NULL; 1113 return (error); 1114 } 1115 1116 do { 1117 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) { 1118 ungetnewvnode(vp); 1119 return (0); 1120 } 1121 } while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0)); 1122 1123 /* Translate the inode number to a disk address. */ 1124 if (ino == LFS_IFILE_INUM) 1125 daddr = fs->lfs_idaddr; 1126 else { 1127 /* XXX bounds-check this too */ 1128 LFS_IENTRY(ifp, fs, ino, bp); 1129 daddr = ifp->if_daddr; 1130 if (fs->lfs_version > 1) { 1131 ts.tv_sec = ifp->if_atime_sec; 1132 ts.tv_nsec = ifp->if_atime_nsec; 1133 } 1134 1135 brelse(bp); 1136 if (daddr == LFS_UNUSED_DADDR) { 1137 *vpp = NULLVP; 1138 ungetnewvnode(vp); 1139 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1140 return (ENOENT); 1141 } 1142 } 1143 1144 /* Allocate/init new vnode/inode. */ 1145 lfs_vcreate(mp, ino, vp); 1146 1147 /* 1148 * Put it onto its hash chain and lock it so that other requests for 1149 * this inode will block if they arrive while we are sleeping waiting 1150 * for old data structures to be purged or for the contents of the 1151 * disk portion of this inode to be read. 1152 */ 1153 ip = VTOI(vp); 1154 ufs_ihashins(ip); 1155 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1156 1157 /* 1158 * XXX 1159 * This may not need to be here, logically it should go down with 1160 * the i_devvp initialization. 1161 * Ask Kirk. 1162 */ 1163 ip->i_lfs = ump->um_lfs; 1164 1165 /* Read in the disk contents for the inode, copy into the inode. */ 1166 retries = 0; 1167 again: 1168 error = bread(ump->um_devvp, fsbtodb(fs, daddr), 1169 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize), 1170 NOCRED, &bp); 1171 if (error) { 1172 /* 1173 * The inode does not contain anything useful, so it would 1174 * be misleading to leave it on its hash chain. With mode 1175 * still zero, it will be unlinked and returned to the free 1176 * list by vput(). 1177 */ 1178 vput(vp); 1179 brelse(bp); 1180 *vpp = NULL; 1181 return (error); 1182 } 1183 1184 dip = lfs_ifind(fs, ino, bp); 1185 if (dip == NULL) { 1186 /* Assume write has not completed yet; try again */ 1187 bp->b_flags |= B_INVAL; 1188 brelse(bp); 1189 ++retries; 1190 if (retries > LFS_IFIND_RETRIES) { 1191 #ifdef DEBUG 1192 /* If the seglock is held look at the bpp to see 1193 what is there anyway */ 1194 simple_lock(&fs->lfs_interlock); 1195 if (fs->lfs_seglock > 0) { 1196 struct buf **bpp; 1197 struct ufs1_dinode *dp; 1198 int i; 1199 1200 for (bpp = fs->lfs_sp->bpp; 1201 bpp != fs->lfs_sp->cbpp; ++bpp) { 1202 if ((*bpp)->b_vp == fs->lfs_ivnode && 1203 bpp != fs->lfs_sp->bpp) { 1204 /* Inode block */ 1205 printf("lfs_vget: block 0x%" PRIx64 ": ", 1206 (*bpp)->b_blkno); 1207 dp = (struct ufs1_dinode *)(*bpp)->b_data; 1208 for (i = 0; i < INOPB(fs); i++) 1209 if (dp[i].di_u.inumber) 1210 printf("%d ", dp[i].di_u.inumber); 1211 printf("\n"); 1212 } 1213 } 1214 } 1215 simple_unlock(&fs->lfs_interlock); 1216 #endif /* DEBUG */ 1217 panic("lfs_vget: dinode not found"); 1218 } 1219 simple_lock(&fs->lfs_interlock); 1220 if (fs->lfs_iocount) { 1221 DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino)); 1222 (void)ltsleep(&fs->lfs_iocount, PRIBIO + 1, 1223 "lfs ifind", 1, &fs->lfs_interlock); 1224 } else 1225 retries = LFS_IFIND_RETRIES; 1226 simple_unlock(&fs->lfs_interlock); 1227 goto again; 1228 } 1229 *ip->i_din.ffs1_din = *dip; 1230 brelse(bp); 1231 1232 if (fs->lfs_version > 1) { 1233 ip->i_ffs1_atime = ts.tv_sec; 1234 ip->i_ffs1_atimensec = ts.tv_nsec; 1235 } 1236 1237 lfs_vinit(mp, &vp); 1238 1239 *vpp = vp; 1240 1241 KASSERT(VOP_ISLOCKED(vp)); 1242 1243 return (0); 1244 } 1245 1246 /* 1247 * File handle to vnode 1248 */ 1249 int 1250 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1251 { 1252 struct lfid lfh; 1253 struct buf *bp; 1254 IFILE *ifp; 1255 int32_t daddr; 1256 struct lfs *fs; 1257 1258 if (fhp->fid_len != sizeof(struct lfid)) 1259 return EINVAL; 1260 1261 memcpy(&lfh, fhp, sizeof(lfh)); 1262 if (lfh.lfid_ino < LFS_IFILE_INUM) 1263 return ESTALE; 1264 1265 fs = VFSTOUFS(mp)->um_lfs; 1266 if (lfh.lfid_ident != fs->lfs_ident) 1267 return ESTALE; 1268 1269 if (lfh.lfid_ino > 1270 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) - 1271 fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb) 1272 return ESTALE; 1273 1274 if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino) == NULLVP) { 1275 LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp); 1276 daddr = ifp->if_daddr; 1277 brelse(bp); 1278 if (daddr == LFS_UNUSED_DADDR) 1279 return ESTALE; 1280 } 1281 1282 return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp)); 1283 } 1284 1285 /* 1286 * Vnode pointer to File handle 1287 */ 1288 /* ARGSUSED */ 1289 int 1290 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 1291 { 1292 struct inode *ip; 1293 struct lfid lfh; 1294 1295 if (*fh_size < sizeof(struct lfid)) { 1296 *fh_size = sizeof(struct lfid); 1297 return E2BIG; 1298 } 1299 *fh_size = sizeof(struct lfid); 1300 ip = VTOI(vp); 1301 memset(&lfh, 0, sizeof(lfh)); 1302 lfh.lfid_len = sizeof(struct lfid); 1303 lfh.lfid_ino = ip->i_number; 1304 lfh.lfid_gen = ip->i_gen; 1305 lfh.lfid_ident = ip->i_lfs->lfs_ident; 1306 memcpy(fhp, &lfh, sizeof(lfh)); 1307 return (0); 1308 } 1309 1310 static int 1311 sysctl_lfs_dostats(SYSCTLFN_ARGS) 1312 { 1313 extern struct lfs_stats lfs_stats; 1314 extern int lfs_dostats; 1315 int error; 1316 1317 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 1318 if (error || newp == NULL) 1319 return (error); 1320 1321 if (lfs_dostats == 0) 1322 memset(&lfs_stats, 0, sizeof(lfs_stats)); 1323 1324 return (0); 1325 } 1326 1327 struct shortlong { 1328 const char *sname; 1329 const char *lname; 1330 }; 1331 1332 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup") 1333 { 1334 int i; 1335 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 1336 lfs_fs_pagetrip; 1337 #ifdef DEBUG 1338 extern int lfs_debug_log_subsys[DLOG_MAX]; 1339 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 1340 { "rollforward", "Debug roll-forward code" }, 1341 { "alloc", "Debug inode allocation and free list" }, 1342 { "avail", "Debug space-available-now accounting" }, 1343 { "flush", "Debug flush triggers" }, 1344 { "lockedlist", "Debug locked list accounting" }, 1345 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 1346 { "vnode", "Debug vnode use during segment write" }, 1347 { "segment", "Debug segment writing" }, 1348 { "seguse", "Debug segment used-bytes accounting" }, 1349 { "cleaner", "Debug cleaning routines" }, 1350 { "mount", "Debug mount/unmount routines" }, 1351 { "pagecache", "Debug UBC interactions" }, 1352 { "dirop", "Debug directory-operation accounting" }, 1353 { "malloc", "Debug private malloc accounting" }, 1354 }; 1355 #endif /* DEBUG */ 1356 struct shortlong stat_names[] = { /* Must match lfs.h! */ 1357 { "segsused", "Number of new segments allocated" }, 1358 { "psegwrites", "Number of partial-segment writes" }, 1359 { "psyncwrites", "Number of synchronous partial-segment" 1360 " writes" }, 1361 { "pcleanwrites", "Number of partial-segment writes by the" 1362 " cleaner" }, 1363 { "blocktot", "Number of blocks written" }, 1364 { "cleanblocks", "Number of blocks written by the cleaner" }, 1365 { "ncheckpoints", "Number of checkpoints made" }, 1366 { "nwrites", "Number of whole writes" }, 1367 { "nsync_writes", "Number of synchronous writes" }, 1368 { "wait_exceeded", "Number of times writer waited for" 1369 " cleaner" }, 1370 { "write_exceeded", "Number of times writer invoked flush" }, 1371 { "flush_invoked", "Number of times flush was invoked" }, 1372 { "vflush_invoked", "Number of time vflush was called" }, 1373 { "clean_inlocked", "Number of vnodes skipped for VXLOCK" }, 1374 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 1375 { "segs_reclaimed", "Number of segments reclaimed" }, 1376 }; 1377 1378 sysctl_createv(clog, 0, NULL, NULL, 1379 CTLFLAG_PERMANENT, 1380 CTLTYPE_NODE, "vfs", NULL, 1381 NULL, 0, NULL, 0, 1382 CTL_VFS, CTL_EOL); 1383 sysctl_createv(clog, 0, NULL, NULL, 1384 CTLFLAG_PERMANENT, 1385 CTLTYPE_NODE, "lfs", 1386 SYSCTL_DESCR("Log-structured file system"), 1387 NULL, 0, NULL, 0, 1388 CTL_VFS, 5, CTL_EOL); 1389 /* 1390 * XXX the "5" above could be dynamic, thereby eliminating one 1391 * more instance of the "number to vfs" mapping problem, but 1392 * "5" is the order as taken from sys/mount.h 1393 */ 1394 1395 sysctl_createv(clog, 0, NULL, NULL, 1396 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1397 CTLTYPE_INT, "flushindir", NULL, 1398 NULL, 0, &lfs_writeindir, 0, 1399 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 1400 sysctl_createv(clog, 0, NULL, NULL, 1401 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1402 CTLTYPE_INT, "clean_vnhead", NULL, 1403 NULL, 0, &lfs_clean_vnhead, 0, 1404 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 1405 sysctl_createv(clog, 0, NULL, NULL, 1406 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1407 CTLTYPE_INT, "dostats", 1408 SYSCTL_DESCR("Maintain statistics on LFS operations"), 1409 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 1410 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 1411 sysctl_createv(clog, 0, NULL, NULL, 1412 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1413 CTLTYPE_INT, "pagetrip", 1414 SYSCTL_DESCR("How many dirty pages in fs triggers" 1415 " a flush"), 1416 NULL, 0, &lfs_fs_pagetrip, 0, 1417 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 1418 #ifdef LFS_KERNEL_RFW 1419 sysctl_createv(clog, 0, NULL, NULL, 1420 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1421 CTLTYPE_INT, "rfw", 1422 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 1423 NULL, 0, &lfs_do_rfw, 0, 1424 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 1425 #endif 1426 1427 sysctl_createv(clog, 0, NULL, NULL, 1428 CTLFLAG_PERMANENT, 1429 CTLTYPE_NODE, "stats", 1430 SYSCTL_DESCR("Debugging options"), 1431 NULL, 0, NULL, 0, 1432 CTL_VFS, 5, LFS_STATS, CTL_EOL); 1433 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 1434 sysctl_createv(clog, 0, NULL, NULL, 1435 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1436 CTLTYPE_INT, stat_names[i].sname, 1437 SYSCTL_DESCR(stat_names[i].lname), 1438 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 1439 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 1440 } 1441 1442 #ifdef DEBUG 1443 sysctl_createv(clog, 0, NULL, NULL, 1444 CTLFLAG_PERMANENT, 1445 CTLTYPE_NODE, "debug", 1446 SYSCTL_DESCR("Debugging options"), 1447 NULL, 0, NULL, 0, 1448 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 1449 for (i = 0; i < DLOG_MAX; i++) { 1450 sysctl_createv(clog, 0, NULL, NULL, 1451 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1452 CTLTYPE_INT, dlog_names[i].sname, 1453 SYSCTL_DESCR(dlog_names[i].lname), 1454 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 1455 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 1456 } 1457 #endif 1458 } 1459 1460 /* 1461 * ufs_bmaparray callback function for writing. 1462 * 1463 * Since blocks will be written to the new segment anyway, 1464 * we don't care about current daddr of them. 1465 */ 1466 static boolean_t 1467 lfs_issequential_hole(const struct ufsmount *ump __unused, 1468 daddr_t daddr0, daddr_t daddr1) 1469 { 1470 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */ 1471 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */ 1472 1473 KASSERT(daddr0 == UNWRITTEN || 1474 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR)); 1475 KASSERT(daddr1 == UNWRITTEN || 1476 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR)); 1477 1478 /* NOTE: all we want to know here is 'hole or not'. */ 1479 /* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */ 1480 1481 /* 1482 * treat UNWRITTENs and all resident blocks as 'contiguous' 1483 */ 1484 if (daddr0 != 0 && daddr1 != 0) 1485 return TRUE; 1486 1487 /* 1488 * both are in hole? 1489 */ 1490 if (daddr0 == 0 && daddr1 == 0) 1491 return TRUE; /* all holes are 'contiguous' for us. */ 1492 1493 return FALSE; 1494 } 1495 1496 /* 1497 * lfs_gop_write functions exactly like genfs_gop_write, except that 1498 * (1) it requires the seglock to be held by its caller, and sp->fip 1499 * to be properly initialized (it will return without re-initializing 1500 * sp->fip, and without calling lfs_writeseg). 1501 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1502 * to determine how large a block it can write at once (though it does 1503 * still use VOP_BMAP to find holes in the file); 1504 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1505 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1506 * now have clusters of clusters, ick.) 1507 */ 1508 static int 1509 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, 1510 int flags __unused) 1511 { 1512 int i, s, error, run, haveeof = 0; 1513 int fs_bshift; 1514 vaddr_t kva; 1515 off_t eof, offset, startoffset = 0; 1516 size_t bytes, iobytes, skipbytes; 1517 daddr_t lbn, blkno; 1518 struct vm_page *pg; 1519 struct buf *mbp, *bp; 1520 struct vnode *devvp = VTOI(vp)->i_devvp; 1521 struct inode *ip = VTOI(vp); 1522 struct lfs *fs = ip->i_lfs; 1523 struct segment *sp = fs->lfs_sp; 1524 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1525 1526 ASSERT_SEGLOCK(fs); 1527 1528 /* The Ifile lives in the buffer cache */ 1529 KASSERT(vp != fs->lfs_ivnode); 1530 1531 /* 1532 * We don't want to fill the disk before the cleaner has a chance 1533 * to make room for us. If we're in danger of doing that, fail 1534 * with EAGAIN. The caller will have to notice this, unlock 1535 * so the cleaner can run, relock and try again. 1536 * 1537 * We must write everything, however, if our vnode is being 1538 * reclaimed. 1539 */ 1540 if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp) 1541 goto tryagain; 1542 1543 /* 1544 * Sometimes things slip past the filters in lfs_putpages, 1545 * and the pagedaemon tries to write pages---problem is 1546 * that the pagedaemon never acquires the segment lock. 1547 * 1548 * Alternatively, pages that were clean when we called 1549 * genfs_putpages may have become dirty in the meantime. In this 1550 * case the segment header is not properly set up for blocks 1551 * to be added to it. 1552 * 1553 * Unbusy and unclean the pages, and put them on the ACTIVE 1554 * queue under the hypothesis that they couldn't have got here 1555 * unless they were modified *quite* recently. 1556 * 1557 * XXXUBC that last statement is an oversimplification of course. 1558 */ 1559 if (!LFS_SEGLOCK_HELD(fs) || 1560 (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) || 1561 (pgs[0]->offset & fs->lfs_bmask) != 0) { 1562 goto tryagain; 1563 } 1564 1565 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x", 1566 vp, pgs, npages, flags); 1567 1568 GOP_SIZE(vp, vp->v_size, &eof, 0); 1569 haveeof = 1; 1570 1571 if (vp->v_type == VREG) 1572 fs_bshift = vp->v_mount->mnt_fs_bshift; 1573 else 1574 fs_bshift = DEV_BSHIFT; 1575 error = 0; 1576 pg = pgs[0]; 1577 startoffset = pg->offset; 1578 KASSERT(eof >= 0); 1579 if (startoffset >= eof) { 1580 goto tryagain; 1581 } else 1582 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 1583 skipbytes = 0; 1584 1585 KASSERT(bytes != 0); 1586 1587 /* Swap PG_DELWRI for PG_PAGEOUT */ 1588 for (i = 0; i < npages; i++) 1589 if (pgs[i]->flags & PG_DELWRI) { 1590 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 1591 pgs[i]->flags &= ~PG_DELWRI; 1592 pgs[i]->flags |= PG_PAGEOUT; 1593 uvmexp.paging++; 1594 uvm_lock_pageq(); 1595 uvm_pageunwire(pgs[i]); 1596 uvm_unlock_pageq(); 1597 } 1598 1599 /* 1600 * Check to make sure we're starting on a block boundary. 1601 * We'll check later to make sure we always write entire 1602 * blocks (or fragments). 1603 */ 1604 if (startoffset & fs->lfs_bmask) 1605 printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n", 1606 startoffset, fs->lfs_bmask, 1607 startoffset & fs->lfs_bmask); 1608 KASSERT((startoffset & fs->lfs_bmask) == 0); 1609 if (bytes & fs->lfs_ffmask) { 1610 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 1611 panic("lfs_gop_write: non-integer blocks"); 1612 } 1613 1614 /* 1615 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 1616 * If we would, write what we have and try again. If we don't 1617 * have anything to write, we'll have to sleep. 1618 */ 1619 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1620 (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ? 1621 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 1622 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 1623 #if 0 1624 " with nfinfo=%d at offset 0x%x\n", 1625 (int)((SEGSUM *)(sp->segsum))->ss_nfinfo, 1626 (unsigned)fs->lfs_offset)); 1627 #endif 1628 lfs_updatemeta(sp); 1629 lfs_release_finfo(fs); 1630 (void) lfs_writeseg(fs, sp); 1631 1632 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 1633 1634 /* 1635 * Having given up all of the pager_map we were holding, 1636 * we can now wait for aiodoned to reclaim it for us 1637 * without fear of deadlock. 1638 */ 1639 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1640 UVMPAGER_MAPIN_WAITOK); 1641 } 1642 1643 s = splbio(); 1644 simple_lock(&global_v_numoutput_slock); 1645 vp->v_numoutput += 2; /* one for biodone, one for aiodone */ 1646 simple_unlock(&global_v_numoutput_slock); 1647 splx(s); 1648 1649 mbp = getiobuf(); 1650 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 1651 vp, mbp, vp->v_numoutput, bytes); 1652 mbp->b_bufsize = npages << PAGE_SHIFT; 1653 mbp->b_data = (void *)kva; 1654 mbp->b_resid = mbp->b_bcount = bytes; 1655 mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL; 1656 mbp->b_iodone = uvm_aio_biodone; 1657 mbp->b_vp = vp; 1658 1659 bp = NULL; 1660 for (offset = startoffset; 1661 bytes > 0; 1662 offset += iobytes, bytes -= iobytes) { 1663 lbn = offset >> fs_bshift; 1664 error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 1665 lfs_issequential_hole); 1666 if (error) { 1667 UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d", 1668 error,0,0,0); 1669 skipbytes += bytes; 1670 bytes = 0; 1671 break; 1672 } 1673 1674 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 1675 bytes); 1676 if (blkno == (daddr_t)-1) { 1677 skipbytes += iobytes; 1678 continue; 1679 } 1680 1681 /* 1682 * Discover how much we can really pack into this buffer. 1683 */ 1684 /* If no room in the current segment, finish it up */ 1685 if (sp->sum_bytes_left < sizeof(int32_t) || 1686 sp->seg_bytes_left < (1 << fs->lfs_bshift)) { 1687 int vers; 1688 1689 lfs_updatemeta(sp); 1690 vers = sp->fip->fi_version; 1691 lfs_release_finfo(fs); 1692 (void) lfs_writeseg(fs, sp); 1693 1694 lfs_acquire_finfo(fs, ip->i_number, vers); 1695 } 1696 /* Check both for space in segment and space in segsum */ 1697 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 1698 << fs_bshift); 1699 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 1700 << fs_bshift); 1701 KASSERT(iobytes > 0); 1702 1703 /* if it's really one i/o, don't make a second buf */ 1704 if (offset == startoffset && iobytes == bytes) { 1705 bp = mbp; 1706 /* correct overcount if there is no second buffer */ 1707 s = splbio(); 1708 simple_lock(&global_v_numoutput_slock); 1709 --vp->v_numoutput; 1710 simple_unlock(&global_v_numoutput_slock); 1711 splx(s); 1712 } else { 1713 bp = getiobuf(); 1714 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 1715 vp, bp, vp->v_numoutput, 0); 1716 bp->b_data = (char *)kva + 1717 (vaddr_t)(offset - pg->offset); 1718 bp->b_resid = bp->b_bcount = iobytes; 1719 bp->b_flags = B_BUSY|B_WRITE|B_CALL; 1720 bp->b_iodone = uvm_aio_biodone1; 1721 } 1722 1723 /* XXX This is silly ... is this necessary? */ 1724 bp->b_vp = NULL; 1725 s = splbio(); 1726 bgetvp(vp, bp); 1727 splx(s); 1728 1729 bp->b_lblkno = lblkno(fs, offset); 1730 bp->b_private = mbp; 1731 if (devvp->v_type == VBLK) { 1732 bp->b_dev = devvp->v_rdev; 1733 } 1734 VOP_BWRITE(bp); 1735 while (lfs_gatherblock(sp, bp, NULL)) 1736 continue; 1737 } 1738 1739 if (skipbytes) { 1740 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0); 1741 s = splbio(); 1742 if (error) { 1743 mbp->b_flags |= B_ERROR; 1744 mbp->b_error = error; 1745 } 1746 mbp->b_resid -= skipbytes; 1747 if (mbp->b_resid == 0) { 1748 biodone(mbp); 1749 } 1750 splx(s); 1751 } 1752 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 1753 return (0); 1754 1755 tryagain: 1756 /* 1757 * We can't write the pages, for whatever reason. 1758 * Clean up after ourselves, and make the caller try again. 1759 */ 1760 simple_lock(&vp->v_interlock); 1761 1762 /* Tell why we're here, if we know */ 1763 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) { 1764 DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n")); 1765 } else if ((pgs[0]->offset & fs->lfs_bmask) != 0) { 1766 DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n")); 1767 } else if (haveeof && startoffset >= eof) { 1768 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 1769 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 1770 pgs[0]->offset, eof, npages)); 1771 } else if (LFS_STARVED_FOR_SEGS(fs)) { 1772 DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n")); 1773 } else { 1774 DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n")); 1775 } 1776 1777 uvm_lock_pageq(); 1778 for (i = 0; i < npages; i++) { 1779 pg = pgs[i]; 1780 1781 if (pg->flags & PG_PAGEOUT) 1782 uvmexp.paging--; 1783 if (pg->flags & PG_DELWRI) { 1784 uvm_pageunwire(pg); 1785 } 1786 uvm_pageactivate(pg); 1787 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 1788 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 1789 vp, pg->offset)); 1790 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 1791 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 1792 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 1793 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 1794 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 1795 pg->wire_count)); 1796 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 1797 pg->loan_count)); 1798 } 1799 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */ 1800 uvm_page_unbusy(pgs, npages); 1801 uvm_unlock_pageq(); 1802 simple_unlock(&vp->v_interlock); 1803 return EAGAIN; 1804 } 1805 1806 /* 1807 * finish vnode/inode initialization. 1808 * used by lfs_vget and lfs_fastvget. 1809 */ 1810 void 1811 lfs_vinit(struct mount *mp, struct vnode **vpp) 1812 { 1813 struct vnode *vp = *vpp; 1814 struct inode *ip = VTOI(vp); 1815 struct ufsmount *ump = VFSTOUFS(mp); 1816 int i; 1817 1818 ip->i_mode = ip->i_ffs1_mode; 1819 ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink; 1820 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size; 1821 ip->i_flags = ip->i_ffs1_flags; 1822 ip->i_gen = ip->i_ffs1_gen; 1823 ip->i_uid = ip->i_ffs1_uid; 1824 ip->i_gid = ip->i_ffs1_gid; 1825 1826 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 1827 ip->i_lfs_odnlink = ip->i_ffs1_nlink; 1828 1829 /* 1830 * Initialize the vnode from the inode, check for aliases. In all 1831 * cases re-init ip, the underlying vnode/inode may have changed. 1832 */ 1833 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 1834 ip = VTOI(vp); 1835 1836 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 1837 if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) { 1838 struct lfs *fs = ump->um_lfs; 1839 #ifdef DEBUG 1840 for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift; 1841 i < NDADDR; i++) { 1842 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 1843 i == 0) 1844 continue; 1845 if (ip->i_ffs1_db[i] != 0) { 1846 inconsistent: 1847 lfs_dump_dinode(ip->i_din.ffs1_din); 1848 panic("inconsistent inode"); 1849 } 1850 } 1851 for ( ; i < NDADDR + NIADDR; i++) { 1852 if (ip->i_ffs1_ib[i - NDADDR] != 0) { 1853 goto inconsistent; 1854 } 1855 } 1856 #endif /* DEBUG */ 1857 for (i = 0; i < NDADDR; i++) 1858 if (ip->i_ffs1_db[i] != 0) 1859 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 1860 } 1861 1862 #ifdef DIAGNOSTIC 1863 if (vp->v_type == VNON) { 1864 # ifdef DEBUG 1865 lfs_dump_dinode(ip->i_din.ffs1_din); 1866 # endif 1867 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 1868 (unsigned long long)ip->i_number, 1869 (ip->i_mode & IFMT) >> 12); 1870 } 1871 #endif /* DIAGNOSTIC */ 1872 1873 /* 1874 * Finish inode initialization now that aliasing has been resolved. 1875 */ 1876 1877 ip->i_devvp = ump->um_devvp; 1878 VREF(ip->i_devvp); 1879 genfs_node_init(vp, &lfs_genfsops); 1880 uvm_vnp_setsize(vp, ip->i_size); 1881 1882 /* Initialize hiblk from file size */ 1883 ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1; 1884 1885 *vpp = vp; 1886 } 1887 1888 /* 1889 * Resize the filesystem to contain the specified number of segments. 1890 */ 1891 int 1892 lfs_resize_fs(struct lfs *fs, int newnsegs) 1893 { 1894 SEGUSE *sup; 1895 struct buf *bp, *obp; 1896 daddr_t olast, nlast, ilast, noff, start, end; 1897 struct vnode *ivp; 1898 struct inode *ip; 1899 int error, badnews, inc, oldnsegs; 1900 int sbbytes, csbbytes, gain, cgain; 1901 int i; 1902 1903 /* Only support v2 and up */ 1904 if (fs->lfs_version < 2) 1905 return EOPNOTSUPP; 1906 1907 /* If we're doing nothing, do it fast */ 1908 oldnsegs = fs->lfs_nseg; 1909 if (newnsegs == oldnsegs) 1910 return 0; 1911 1912 /* We always have to have two superblocks */ 1913 if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1])) 1914 return EFBIG; 1915 1916 ivp = fs->lfs_ivnode; 1917 ip = VTOI(ivp); 1918 error = 0; 1919 1920 /* Take the segment lock so no one else calls lfs_newseg() */ 1921 lfs_seglock(fs, SEGM_PROT); 1922 1923 /* 1924 * Make sure the segments we're going to be losing, if any, 1925 * are in fact empty. We hold the seglock, so their status 1926 * cannot change underneath us. Count the superblocks we lose, 1927 * while we're at it. 1928 */ 1929 sbbytes = csbbytes = 0; 1930 cgain = 0; 1931 for (i = newnsegs; i < oldnsegs; i++) { 1932 LFS_SEGENTRY(sup, fs, i, bp); 1933 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 1934 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1935 sbbytes += LFS_SBPAD; 1936 if (!(sup->su_flags & SEGUSE_DIRTY)) { 1937 ++cgain; 1938 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1939 csbbytes += LFS_SBPAD; 1940 } 1941 brelse(bp); 1942 if (badnews) { 1943 error = EBUSY; 1944 goto out; 1945 } 1946 } 1947 1948 /* Note old and new segment table endpoints, and old ifile size */ 1949 olast = fs->lfs_cleansz + fs->lfs_segtabsz; 1950 nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz; 1951 ilast = ivp->v_size >> fs->lfs_bshift; 1952 noff = nlast - olast; 1953 1954 /* 1955 * Make sure no one can use the Ifile while we change it around. 1956 * Even after taking the iflock we need to make sure no one still 1957 * is holding Ifile buffers, so we get each one, to drain them. 1958 * (XXX this could be done better.) 1959 */ 1960 simple_lock(&fs->lfs_interlock); 1961 lockmgr(&fs->lfs_iflock, LK_EXCLUSIVE, &fs->lfs_interlock); 1962 simple_unlock(&fs->lfs_interlock); 1963 vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY); 1964 for (i = 0; i < ilast; i++) { 1965 bread(ivp, i, fs->lfs_bsize, NOCRED, &bp); 1966 brelse(bp); 1967 } 1968 1969 /* Allocate new Ifile blocks */ 1970 for (i = ilast; i < ilast + noff; i++) { 1971 if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0, 1972 &bp) != 0) 1973 panic("balloc extending ifile"); 1974 memset(bp->b_data, 0, fs->lfs_bsize); 1975 VOP_BWRITE(bp); 1976 } 1977 1978 /* Register new ifile size */ 1979 ip->i_size += noff * fs->lfs_bsize; 1980 ip->i_ffs1_size = ip->i_size; 1981 uvm_vnp_setsize(ivp, ip->i_size); 1982 1983 /* Copy the inode table to its new position */ 1984 if (noff != 0) { 1985 if (noff < 0) { 1986 start = nlast; 1987 end = ilast + noff; 1988 inc = 1; 1989 } else { 1990 start = ilast + noff - 1; 1991 end = nlast - 1; 1992 inc = -1; 1993 } 1994 for (i = start; i != end; i += inc) { 1995 if (bread(ivp, i, fs->lfs_bsize, NOCRED, &bp) != 0) 1996 panic("resize: bread dst blk failed"); 1997 if (bread(ivp, i - noff, fs->lfs_bsize, NOCRED, &obp)) 1998 panic("resize: bread src blk failed"); 1999 memcpy(bp->b_data, obp->b_data, fs->lfs_bsize); 2000 VOP_BWRITE(bp); 2001 brelse(obp); 2002 } 2003 } 2004 2005 /* If we are expanding, write the new empty SEGUSE entries */ 2006 if (newnsegs > oldnsegs) { 2007 for (i = oldnsegs; i < newnsegs; i++) { 2008 if ((error = bread(ivp, i / fs->lfs_sepb + 2009 fs->lfs_cleansz, 2010 fs->lfs_bsize, NOCRED, &bp)) != 0) 2011 panic("lfs: ifile read: %d", error); 2012 while ((i + 1) % fs->lfs_sepb && i < newnsegs) { 2013 sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb]; 2014 memset(sup, 0, sizeof(*sup)); 2015 i++; 2016 } 2017 VOP_BWRITE(bp); 2018 } 2019 } 2020 2021 /* Zero out unused superblock offsets */ 2022 for (i = 2; i < LFS_MAXNUMSB; i++) 2023 if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs) 2024 fs->lfs_sboffs[i] = 0x0; 2025 2026 /* 2027 * Correct superblock entries that depend on fs size. 2028 * The computations of these are as follows: 2029 * 2030 * size = segtod(fs, nseg) 2031 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD) 2032 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2033 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD) 2034 * + (segtod(fs, 1) - (offset - curseg)) 2035 * - segtod(fs, minfreeseg - (minfreeseg / 2)) 2036 * 2037 * XXX - we should probably adjust minfreeseg as well. 2038 */ 2039 gain = (newnsegs - oldnsegs); 2040 fs->lfs_nseg = newnsegs; 2041 fs->lfs_segtabsz = nlast - fs->lfs_cleansz; 2042 fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize); 2043 fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes); 2044 fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes) 2045 - gain * btofsb(fs, fs->lfs_bsize / 2); 2046 if (gain > 0) { 2047 fs->lfs_nclean += gain; 2048 fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize); 2049 } else { 2050 fs->lfs_nclean -= cgain; 2051 fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) - 2052 btofsb(fs, csbbytes); 2053 } 2054 2055 /* Resize segment flag cache */ 2056 fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2057 fs->lfs_nseg * sizeof(u_int32_t), 2058 M_SEGMENT, M_WAITOK); 2059 fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[1], 2060 fs->lfs_nseg * sizeof(u_int32_t), 2061 M_SEGMENT, M_WAITOK); 2062 for (i = oldnsegs; i < newnsegs; i++) 2063 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2064 2065 /* Truncate Ifile if necessary */ 2066 if (noff < 0) 2067 lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0, 2068 NOCRED, curlwp); 2069 2070 /* Update cleaner info so the cleaner can die */ 2071 bread(ivp, 0, fs->lfs_bsize, NOCRED, &bp); 2072 ((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean; 2073 ((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean; 2074 VOP_BWRITE(bp); 2075 2076 /* Let Ifile accesses proceed */ 2077 VOP_UNLOCK(ivp, 0); 2078 simple_lock(&fs->lfs_interlock); 2079 lockmgr(&fs->lfs_iflock, LK_RELEASE, &fs->lfs_interlock); 2080 simple_unlock(&fs->lfs_interlock); 2081 2082 out: 2083 lfs_segunlock(fs); 2084 return error; 2085 } 2086