lfs_vfsops.c revision 1.220
1 /* $NetBSD: lfs_vfsops.c,v 1.220 2006/09/02 06:48:00 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 (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 1.220 2006/09/02 06:48:00 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) 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 if (locked_queue_count > LFS_MAX_BUFS) 260 DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n", 261 locked_queue_count, LFS_MAX_BUFS)); 262 if (locked_queue_bytes > LFS_MAX_BYTES) 263 DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n", 264 locked_queue_bytes, LFS_MAX_BYTES)); 265 if (lfs_subsys_pages > LFS_MAX_PAGES) 266 DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n", 267 lfs_subsys_pages, LFS_MAX_PAGES)); 268 269 lfs_flush(NULL, SEGM_WRITERD, 0); 270 lfs_do_flush = 0; 271 } 272 } 273 /* NOTREACHED */ 274 } 275 276 /* 277 * Initialize the filesystem, most work done by ufs_init. 278 */ 279 void 280 lfs_init() 281 { 282 #ifdef _LKM 283 malloc_type_attach(M_SEGMENT); 284 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 285 "lfsinopl", &pool_allocator_nointr); 286 pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, 287 "lfsdinopl", &pool_allocator_nointr); 288 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 289 "lfsinoextpl", &pool_allocator_nointr); 290 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 291 "lfslbnpool", &pool_allocator_nointr); 292 #endif 293 ufs_init(); 294 295 #ifdef DEBUG 296 memset(lfs_log, 0, sizeof(lfs_log)); 297 #endif 298 simple_lock_init(&lfs_subsys_lock); 299 } 300 301 void 302 lfs_reinit() 303 { 304 ufs_reinit(); 305 } 306 307 void 308 lfs_done() 309 { 310 ufs_done(); 311 #ifdef _LKM 312 pool_destroy(&lfs_inode_pool); 313 pool_destroy(&lfs_dinode_pool); 314 pool_destroy(&lfs_inoext_pool); 315 pool_destroy(&lfs_lbnentry_pool); 316 malloc_type_detach(M_SEGMENT); 317 #endif 318 } 319 320 /* 321 * Called by main() when ufs is going to be mounted as root. 322 */ 323 int 324 lfs_mountroot() 325 { 326 extern struct vnode *rootvp; 327 struct mount *mp; 328 struct lwp *l = curlwp; /* XXX */ 329 int error; 330 331 if (device_class(root_device) != DV_DISK) 332 return (ENODEV); 333 334 if (rootdev == NODEV) 335 return (ENODEV); 336 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 337 vrele(rootvp); 338 return (error); 339 } 340 if ((error = lfs_mountfs(rootvp, mp, l))) { 341 mp->mnt_op->vfs_refcount--; 342 vfs_unbusy(mp); 343 free(mp, M_MOUNT); 344 return (error); 345 } 346 simple_lock(&mountlist_slock); 347 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 348 simple_unlock(&mountlist_slock); 349 (void)lfs_statvfs(mp, &mp->mnt_stat, l); 350 vfs_unbusy(mp); 351 setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp)); 352 return (0); 353 } 354 355 /* 356 * VFS Operations. 357 * 358 * mount system call 359 */ 360 int 361 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct lwp *l) 362 { 363 struct vnode *devvp; 364 struct ufs_args args; 365 struct ufsmount *ump = NULL; 366 struct lfs *fs = NULL; /* LFS */ 367 int error, update; 368 mode_t accessmode; 369 370 if (mp->mnt_flag & MNT_GETARGS) { 371 ump = VFSTOUFS(mp); 372 if (ump == NULL) 373 return EIO; 374 args.fspec = NULL; 375 return copyout(&args, data, sizeof(args)); 376 } 377 error = copyin(data, &args, sizeof (struct ufs_args)); 378 if (error) 379 return (error); 380 381 update = mp->mnt_flag & MNT_UPDATE; 382 383 /* Check arguments */ 384 if (args.fspec != NULL) { 385 /* 386 * Look up the name and verify that it's sane. 387 */ 388 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, l); 389 if ((error = namei(ndp)) != 0) 390 return (error); 391 devvp = ndp->ni_vp; 392 393 if (!update) { 394 /* 395 * Be sure this is a valid block device 396 */ 397 if (devvp->v_type != VBLK) 398 error = ENOTBLK; 399 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 400 error = ENXIO; 401 } else { 402 /* 403 * Be sure we're still naming the same device 404 * used for our initial mount 405 */ 406 ump = VFSTOUFS(mp); 407 if (devvp != ump->um_devvp) 408 error = EINVAL; 409 } 410 } else { 411 if (!update) { 412 /* New mounts must have a filename for the device */ 413 return (EINVAL); 414 } else { 415 /* Use the extant mount */ 416 ump = VFSTOUFS(mp); 417 devvp = ump->um_devvp; 418 vref(devvp); 419 } 420 } 421 422 423 /* 424 * If mount by non-root, then verify that user has necessary 425 * permissions on the device. 426 */ 427 if (error == 0 && kauth_cred_geteuid(l->l_cred) != 0) { 428 accessmode = VREAD; 429 if (update ? 430 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 431 (mp->mnt_flag & MNT_RDONLY) == 0) 432 accessmode |= VWRITE; 433 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 434 error = VOP_ACCESS(devvp, accessmode, l->l_cred, l); 435 VOP_UNLOCK(devvp, 0); 436 } 437 438 if (error) { 439 vrele(devvp); 440 return (error); 441 } 442 443 if (!update) { 444 int flags; 445 446 /* 447 * Disallow multiple mounts of the same device. 448 * Disallow mounting of a device that is currently in use 449 * (except for root, which might share swap device for 450 * miniroot). 451 */ 452 error = vfs_mountedon(devvp); 453 if (error) 454 goto fail; 455 if (vcount(devvp) > 1 && devvp != rootvp) { 456 error = EBUSY; 457 goto fail; 458 } 459 if (mp->mnt_flag & MNT_RDONLY) 460 flags = FREAD; 461 else 462 flags = FREAD|FWRITE; 463 error = VOP_OPEN(devvp, flags, FSCRED, l); 464 if (error) 465 goto fail; 466 error = lfs_mountfs(devvp, mp, l); /* LFS */ 467 if (error) { 468 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 469 (void)VOP_CLOSE(devvp, flags, NOCRED, l); 470 VOP_UNLOCK(devvp, 0); 471 goto fail; 472 } 473 474 ump = VFSTOUFS(mp); 475 fs = ump->um_lfs; 476 } else { 477 /* 478 * Update the mount. 479 */ 480 481 /* 482 * The initial mount got a reference on this 483 * device, so drop the one obtained via 484 * namei(), above. 485 */ 486 vrele(devvp); 487 488 ump = VFSTOUFS(mp); 489 fs = ump->um_lfs; 490 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 491 /* 492 * Changing from read-only to read/write. 493 * Note in the superblocks that we're writing. 494 */ 495 fs->lfs_ronly = 0; 496 if (fs->lfs_pflags & LFS_PF_CLEAN) { 497 fs->lfs_pflags &= ~LFS_PF_CLEAN; 498 lfs_writesuper(fs, fs->lfs_sboffs[0]); 499 lfs_writesuper(fs, fs->lfs_sboffs[1]); 500 } 501 } 502 if (args.fspec == NULL) 503 return EINVAL; 504 } 505 506 error = set_statvfs_info(path, UIO_USERSPACE, args.fspec, 507 UIO_USERSPACE, mp, l); 508 if (error == 0) 509 (void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname, 510 sizeof(fs->lfs_fsmnt)); 511 return error; 512 513 fail: 514 vrele(devvp); 515 return (error); 516 } 517 518 519 /* 520 * Common code for mount and mountroot 521 * LFS specific 522 */ 523 int 524 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 525 { 526 struct dlfs *tdfs, *dfs, *adfs; 527 struct lfs *fs; 528 struct ufsmount *ump; 529 struct vnode *vp; 530 struct buf *bp, *abp; 531 struct partinfo dpart; 532 dev_t dev; 533 int error, i, ronly, secsize, fsbsize; 534 kauth_cred_t cred; 535 CLEANERINFO *cip; 536 SEGUSE *sup; 537 daddr_t sb_addr; 538 539 cred = l ? l->l_cred : NOCRED; 540 541 /* 542 * Flush out any old buffers remaining from a previous use. 543 */ 544 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 545 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 546 VOP_UNLOCK(devvp, 0); 547 if (error) 548 return (error); 549 550 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 551 if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, l) != 0) 552 secsize = DEV_BSIZE; 553 else 554 secsize = dpart.disklab->d_secsize; 555 556 /* Don't free random space on error. */ 557 bp = NULL; 558 abp = NULL; 559 ump = NULL; 560 561 sb_addr = LFS_LABELPAD / secsize; 562 while (1) { 563 /* Read in the superblock. */ 564 error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp); 565 if (error) 566 goto out; 567 dfs = (struct dlfs *)bp->b_data; 568 569 /* Check the basics. */ 570 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE || 571 dfs->dlfs_version > LFS_VERSION || 572 dfs->dlfs_bsize < sizeof(struct dlfs)) { 573 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 574 error = EINVAL; /* XXX needs translation */ 575 goto out; 576 } 577 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) { 578 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 579 dfs->dlfs_inodefmt)); 580 error = EINVAL; 581 goto out; 582 } 583 584 if (dfs->dlfs_version == 1) 585 fsbsize = secsize; 586 else { 587 fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb + 588 dfs->dlfs_fsbtodb); 589 /* 590 * Could be, if the frag size is large enough, that we 591 * don't have the "real" primary superblock. If that's 592 * the case, get the real one, and try again. 593 */ 594 if (sb_addr != dfs->dlfs_sboffs[0] << 595 dfs->dlfs_fsbtodb) { 596 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 597 " 0x%llx is not right, trying 0x%llx\n", 598 (long long)sb_addr, 599 (long long)(dfs->dlfs_sboffs[0] << 600 dfs->dlfs_fsbtodb))); 601 sb_addr = dfs->dlfs_sboffs[0] << 602 dfs->dlfs_fsbtodb; 603 brelse(bp); 604 continue; 605 } 606 } 607 break; 608 } 609 610 /* 611 * Check the second superblock to see which is newer; then mount 612 * using the older of the two. This is necessary to ensure that 613 * the filesystem is valid if it was not unmounted cleanly. 614 */ 615 616 if (dfs->dlfs_sboffs[1] && 617 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 618 { 619 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize), 620 LFS_SBPAD, cred, &abp); 621 if (error) 622 goto out; 623 adfs = (struct dlfs *)abp->b_data; 624 625 if (dfs->dlfs_version == 1) { 626 /* 1s resolution comparison */ 627 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 628 tdfs = adfs; 629 else 630 tdfs = dfs; 631 } else { 632 /* monotonic infinite-resolution comparison */ 633 if (adfs->dlfs_serial < dfs->dlfs_serial) 634 tdfs = adfs; 635 else 636 tdfs = dfs; 637 } 638 639 /* Check the basics. */ 640 if (tdfs->dlfs_magic != LFS_MAGIC || 641 tdfs->dlfs_bsize > MAXBSIZE || 642 tdfs->dlfs_version > LFS_VERSION || 643 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 644 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 645 " sanity failed\n")); 646 error = EINVAL; /* XXX needs translation */ 647 goto out; 648 } 649 } else { 650 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock" 651 " daddr=0x%x\n", dfs->dlfs_sboffs[1])); 652 error = EINVAL; 653 goto out; 654 } 655 656 /* Allocate the mount structure, copy the superblock into it. */ 657 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO); 658 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 659 660 /* Compatibility */ 661 if (fs->lfs_version < 2) { 662 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 663 fs->lfs_ibsize = fs->lfs_bsize; 664 fs->lfs_start = fs->lfs_sboffs[0]; 665 fs->lfs_tstamp = fs->lfs_otstamp; 666 fs->lfs_fsbtodb = 0; 667 } 668 if (fs->lfs_resvseg == 0) 669 fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \ 670 MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1)); 671 672 /* 673 * If we aren't going to be able to write meaningfully to this 674 * filesystem, and were not mounted readonly, bomb out now. 675 */ 676 if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 677 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 678 " we need BUFPAGES >= %lld\n", 679 (long long)((bufmem_hiwater / bufmem_lowater) * 680 LFS_INVERSE_MAX_BYTES( 681 fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 682 free(fs, M_UFSMNT); 683 error = EFBIG; /* XXX needs translation */ 684 goto out; 685 } 686 687 /* Before rolling forward, lock so vget will sleep for other procs */ 688 if (l != NULL) { 689 fs->lfs_flags = LFS_NOTYET; 690 fs->lfs_rfpid = l->l_proc->p_pid; 691 } 692 693 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 694 ump->um_lfs = fs; 695 ump->um_ops = &lfs_ufsops; 696 ump->um_fstype = UFS1; 697 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 698 bp->b_flags |= B_INVAL; 699 abp->b_flags |= B_INVAL; 700 } 701 brelse(bp); 702 bp = NULL; 703 brelse(abp); 704 abp = NULL; 705 706 /* Set up the I/O information */ 707 fs->lfs_devbsize = secsize; 708 fs->lfs_iocount = 0; 709 fs->lfs_diropwait = 0; 710 fs->lfs_activesb = 0; 711 fs->lfs_uinodes = 0; 712 fs->lfs_ravail = 0; 713 fs->lfs_favail = 0; 714 fs->lfs_sbactive = 0; 715 716 /* Set up the ifile and lock aflags */ 717 fs->lfs_doifile = 0; 718 fs->lfs_writer = 0; 719 fs->lfs_dirops = 0; 720 fs->lfs_nadirop = 0; 721 fs->lfs_seglock = 0; 722 fs->lfs_pdflush = 0; 723 fs->lfs_sleepers = 0; 724 fs->lfs_pages = 0; 725 simple_lock_init(&fs->lfs_interlock); 726 lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0); 727 lockinit(&fs->lfs_iflock, PINOD, "lfs_iflock", 0, 0); 728 729 /* Set the file system readonly/modify bits. */ 730 fs->lfs_ronly = ronly; 731 if (ronly == 0) 732 fs->lfs_fmod = 1; 733 734 /* Initialize the mount structure. */ 735 dev = devvp->v_rdev; 736 mp->mnt_data = ump; 737 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 738 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 739 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 740 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 741 mp->mnt_stat.f_iosize = fs->lfs_bsize; 742 mp->mnt_flag |= MNT_LOCAL; 743 mp->mnt_fs_bshift = fs->lfs_bshift; 744 ump->um_flags = 0; 745 ump->um_mountp = mp; 746 ump->um_dev = dev; 747 ump->um_devvp = devvp; 748 ump->um_bptrtodb = fs->lfs_fsbtodb; 749 ump->um_seqinc = fragstofsb(fs, fs->lfs_frag); 750 ump->um_nindir = fs->lfs_nindir; 751 ump->um_lognindir = ffs(fs->lfs_nindir) - 1; 752 for (i = 0; i < MAXQUOTAS; i++) 753 ump->um_quotas[i] = NULLVP; 754 ump->um_maxsymlinklen = fs->lfs_maxsymlinklen; 755 ump->um_dirblksiz = DIRBLKSIZ; 756 ump->um_maxfilesize = fs->lfs_maxfilesize; 757 if (ump->um_maxsymlinklen > 0) 758 mp->mnt_iflag |= IMNT_DTYPE; 759 devvp->v_specmountpoint = mp; 760 761 /* Set up reserved memory for pageout */ 762 lfs_setup_resblks(fs); 763 /* Set up vdirop tailq */ 764 TAILQ_INIT(&fs->lfs_dchainhd); 765 /* and paging tailq */ 766 TAILQ_INIT(&fs->lfs_pchainhd); 767 /* and delayed segment accounting for truncation list */ 768 LIST_INIT(&fs->lfs_segdhd); 769 770 /* 771 * We use the ifile vnode for almost every operation. Instead of 772 * retrieving it from the hash table each time we retrieve it here, 773 * artificially increment the reference count and keep a pointer 774 * to it in the incore copy of the superblock. 775 */ 776 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) { 777 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 778 goto out; 779 } 780 fs->lfs_ivnode = vp; 781 VREF(vp); 782 783 /* Set up inode bitmap and order free list */ 784 lfs_order_freelist(fs); 785 786 /* Set up segment usage flags for the autocleaner. */ 787 fs->lfs_nactive = 0; 788 fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *), 789 M_SEGMENT, M_WAITOK); 790 fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 791 M_SEGMENT, M_WAITOK); 792 fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 793 M_SEGMENT, M_WAITOK); 794 memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t)); 795 for (i = 0; i < fs->lfs_nseg; i++) { 796 int changed; 797 798 LFS_SEGENTRY(sup, fs, i, bp); 799 changed = 0; 800 if (!ronly) { 801 if (sup->su_nbytes == 0 && 802 !(sup->su_flags & SEGUSE_EMPTY)) { 803 sup->su_flags |= SEGUSE_EMPTY; 804 ++changed; 805 } else if (!(sup->su_nbytes == 0) && 806 (sup->su_flags & SEGUSE_EMPTY)) { 807 sup->su_flags &= ~SEGUSE_EMPTY; 808 ++changed; 809 } 810 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 811 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 812 ++changed; 813 } 814 } 815 fs->lfs_suflags[0][i] = sup->su_flags; 816 if (changed) 817 LFS_WRITESEGENTRY(sup, fs, i, bp); 818 else 819 brelse(bp); 820 } 821 822 #ifdef LFS_KERNEL_RFW 823 lfs_roll_forward(fs, mp, l); 824 #endif 825 826 /* If writing, sb is not clean; record in case of immediate crash */ 827 if (!fs->lfs_ronly) { 828 fs->lfs_pflags &= ~LFS_PF_CLEAN; 829 lfs_writesuper(fs, fs->lfs_sboffs[0]); 830 lfs_writesuper(fs, fs->lfs_sboffs[1]); 831 } 832 833 /* Allow vget now that roll-forward is complete */ 834 fs->lfs_flags &= ~(LFS_NOTYET); 835 wakeup(&fs->lfs_flags); 836 837 /* 838 * Initialize the ifile cleaner info with information from 839 * the superblock. 840 */ 841 LFS_CLEANERINFO(cip, fs, bp); 842 cip->clean = fs->lfs_nclean; 843 cip->dirty = fs->lfs_nseg - fs->lfs_nclean; 844 cip->avail = fs->lfs_avail; 845 cip->bfree = fs->lfs_bfree; 846 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 847 848 /* 849 * Mark the current segment as ACTIVE, since we're going to 850 * be writing to it. 851 */ 852 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); 853 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 854 fs->lfs_nactive++; 855 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); /* Ifile */ 856 857 /* Now that roll-forward is done, unlock the Ifile */ 858 vput(vp); 859 860 /* Start the pagedaemon-anticipating daemon */ 861 if (lfs_writer_daemon == 0 && 862 kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0) 863 panic("fork lfs_writer"); 864 865 return (0); 866 867 out: 868 if (bp) 869 brelse(bp); 870 if (abp) 871 brelse(abp); 872 if (ump) { 873 free(ump->um_lfs, M_UFSMNT); 874 free(ump, M_UFSMNT); 875 mp->mnt_data = NULL; 876 } 877 878 return (error); 879 } 880 881 /* 882 * unmount system call 883 */ 884 int 885 lfs_unmount(struct mount *mp, int mntflags, struct lwp *l) 886 { 887 struct ufsmount *ump; 888 struct lfs *fs; 889 int error, flags, ronly; 890 int s; 891 892 flags = 0; 893 if (mntflags & MNT_FORCE) 894 flags |= FORCECLOSE; 895 896 ump = VFSTOUFS(mp); 897 fs = ump->um_lfs; 898 899 /* Two checkpoints */ 900 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 901 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 902 903 /* wake up the cleaner so it can die */ 904 lfs_wakeup_cleaner(fs); 905 simple_lock(&fs->lfs_interlock); 906 while (fs->lfs_sleepers) 907 ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0, 908 &fs->lfs_interlock); 909 simple_unlock(&fs->lfs_interlock); 910 911 #ifdef QUOTA 912 if (mp->mnt_flag & MNT_QUOTA) { 913 int i; 914 error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags); 915 if (error) 916 return (error); 917 for (i = 0; i < MAXQUOTAS; i++) { 918 if (ump->um_quotas[i] == NULLVP) 919 continue; 920 quotaoff(l, mp, i); 921 } 922 /* 923 * Here we fall through to vflush again to ensure 924 * that we have gotten rid of all the system vnodes. 925 */ 926 } 927 #endif 928 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 929 return (error); 930 if ((error = VFS_SYNC(mp, 1, l->l_cred, l)) != 0) 931 return (error); 932 s = splbio(); 933 if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd)) 934 panic("lfs_unmount: still dirty blocks on ifile vnode"); 935 splx(s); 936 937 /* Explicitly write the superblock, to update serial and pflags */ 938 fs->lfs_pflags |= LFS_PF_CLEAN; 939 lfs_writesuper(fs, fs->lfs_sboffs[0]); 940 lfs_writesuper(fs, fs->lfs_sboffs[1]); 941 simple_lock(&fs->lfs_interlock); 942 while (fs->lfs_iocount) 943 ltsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 944 &fs->lfs_interlock); 945 simple_unlock(&fs->lfs_interlock); 946 947 /* Finish with the Ifile, now that we're done with it */ 948 vrele(fs->lfs_ivnode); 949 vgone(fs->lfs_ivnode); 950 951 ronly = !fs->lfs_ronly; 952 if (ump->um_devvp->v_type != VBAD) 953 ump->um_devvp->v_specmountpoint = NULL; 954 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 955 error = VOP_CLOSE(ump->um_devvp, 956 ronly ? FREAD : FREAD|FWRITE, NOCRED, l); 957 vput(ump->um_devvp); 958 959 /* Complain about page leakage */ 960 if (fs->lfs_pages > 0) 961 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 962 fs->lfs_pages, lfs_subsys_pages); 963 964 /* Free per-mount data structures */ 965 free(fs->lfs_ino_bitmap, M_SEGMENT); 966 free(fs->lfs_suflags[0], M_SEGMENT); 967 free(fs->lfs_suflags[1], M_SEGMENT); 968 free(fs->lfs_suflags, M_SEGMENT); 969 lfs_free_resblks(fs); 970 free(fs, M_UFSMNT); 971 free(ump, M_UFSMNT); 972 973 mp->mnt_data = NULL; 974 mp->mnt_flag &= ~MNT_LOCAL; 975 return (error); 976 } 977 978 /* 979 * Get file system statistics. 980 * 981 * NB: We don't lock to access the superblock here, because it's not 982 * really that important if we get it wrong. 983 */ 984 int 985 lfs_statvfs(struct mount *mp, struct statvfs *sbp, struct lwp *l) 986 { 987 struct lfs *fs; 988 struct ufsmount *ump; 989 990 ump = VFSTOUFS(mp); 991 fs = ump->um_lfs; 992 if (fs->lfs_magic != LFS_MAGIC) 993 panic("lfs_statvfs: magic"); 994 995 sbp->f_bsize = fs->lfs_bsize; 996 sbp->f_frsize = fs->lfs_fsize; 997 sbp->f_iosize = fs->lfs_bsize; 998 sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks); 999 1000 sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs)); 1001 KASSERT(sbp->f_bfree <= fs->lfs_dsize); 1002 #if 0 1003 if (sbp->f_bfree < 0) 1004 sbp->f_bfree = 0; 1005 #endif 1006 1007 sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs)); 1008 if (sbp->f_bfree > sbp->f_bresvd) 1009 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1010 else 1011 sbp->f_bavail = 0; 1012 1013 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs); 1014 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles; 1015 sbp->f_favail = sbp->f_ffree; 1016 sbp->f_fresvd = 0; 1017 copy_statvfs_info(sbp, mp); 1018 return (0); 1019 } 1020 1021 /* 1022 * Go through the disk queues to initiate sandbagged IO; 1023 * go through the inodes to write those that have been modified; 1024 * initiate the writing of the super block if it has been modified. 1025 * 1026 * Note: we are always called with the filesystem marked `MPBUSY'. 1027 */ 1028 int 1029 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred, struct lwp *l) 1030 { 1031 int error; 1032 struct lfs *fs; 1033 1034 fs = VFSTOUFS(mp)->um_lfs; 1035 if (fs->lfs_ronly) 1036 return 0; 1037 1038 /* Snapshots should not hose the syncer */ 1039 /* 1040 * XXX Sync can block here anyway, since we don't have a very 1041 * XXX good idea of how much data is pending. If it's more 1042 * XXX than a segment and lfs_nextseg is close to the end of 1043 * XXX the log, we'll likely block. 1044 */ 1045 simple_lock(&fs->lfs_interlock); 1046 if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) { 1047 simple_unlock(&fs->lfs_interlock); 1048 return 0; 1049 } 1050 simple_unlock(&fs->lfs_interlock); 1051 1052 lfs_writer_enter(fs, "lfs_dirops"); 1053 1054 /* All syncs must be checkpoints until roll-forward is implemented. */ 1055 DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset)); 1056 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1057 lfs_writer_leave(fs); 1058 #ifdef QUOTA 1059 qsync(mp); 1060 #endif 1061 return (error); 1062 } 1063 1064 extern struct lock ufs_hashlock; 1065 1066 /* 1067 * Look up an LFS dinode number to find its incore vnode. If not already 1068 * in core, read it in from the specified device. Return the inode locked. 1069 * Detection and handling of mount points must be done by the calling routine. 1070 */ 1071 int 1072 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1073 { 1074 struct lfs *fs; 1075 struct ufs1_dinode *dip; 1076 struct inode *ip; 1077 struct buf *bp; 1078 struct ifile *ifp; 1079 struct vnode *vp; 1080 struct ufsmount *ump; 1081 daddr_t daddr; 1082 dev_t dev; 1083 int error, retries; 1084 struct timespec ts; 1085 1086 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1087 1088 ump = VFSTOUFS(mp); 1089 dev = ump->um_dev; 1090 fs = ump->um_lfs; 1091 1092 /* 1093 * If the filesystem is not completely mounted yet, suspend 1094 * any access requests (wait for roll-forward to complete). 1095 */ 1096 simple_lock(&fs->lfs_interlock); 1097 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1098 ltsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1099 &fs->lfs_interlock); 1100 simple_unlock(&fs->lfs_interlock); 1101 1102 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) 1103 return (0); 1104 1105 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) { 1106 *vpp = NULL; 1107 return (error); 1108 } 1109 1110 do { 1111 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) { 1112 ungetnewvnode(vp); 1113 return (0); 1114 } 1115 } while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0)); 1116 1117 /* Translate the inode number to a disk address. */ 1118 if (ino == LFS_IFILE_INUM) 1119 daddr = fs->lfs_idaddr; 1120 else { 1121 /* XXX bounds-check this too */ 1122 LFS_IENTRY(ifp, fs, ino, bp); 1123 daddr = ifp->if_daddr; 1124 if (fs->lfs_version > 1) { 1125 ts.tv_sec = ifp->if_atime_sec; 1126 ts.tv_nsec = ifp->if_atime_nsec; 1127 } 1128 1129 brelse(bp); 1130 if (daddr == LFS_UNUSED_DADDR) { 1131 *vpp = NULLVP; 1132 ungetnewvnode(vp); 1133 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1134 return (ENOENT); 1135 } 1136 } 1137 1138 /* Allocate/init new vnode/inode. */ 1139 lfs_vcreate(mp, ino, vp); 1140 1141 /* 1142 * Put it onto its hash chain and lock it so that other requests for 1143 * this inode will block if they arrive while we are sleeping waiting 1144 * for old data structures to be purged or for the contents of the 1145 * disk portion of this inode to be read. 1146 */ 1147 ip = VTOI(vp); 1148 ufs_ihashins(ip); 1149 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1150 1151 /* 1152 * XXX 1153 * This may not need to be here, logically it should go down with 1154 * the i_devvp initialization. 1155 * Ask Kirk. 1156 */ 1157 ip->i_lfs = ump->um_lfs; 1158 1159 /* Read in the disk contents for the inode, copy into the inode. */ 1160 retries = 0; 1161 again: 1162 error = bread(ump->um_devvp, fsbtodb(fs, daddr), 1163 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize), 1164 NOCRED, &bp); 1165 if (error) { 1166 /* 1167 * The inode does not contain anything useful, so it would 1168 * be misleading to leave it on its hash chain. With mode 1169 * still zero, it will be unlinked and returned to the free 1170 * list by vput(). 1171 */ 1172 vput(vp); 1173 brelse(bp); 1174 *vpp = NULL; 1175 return (error); 1176 } 1177 1178 dip = lfs_ifind(fs, ino, bp); 1179 if (dip == NULL) { 1180 /* Assume write has not completed yet; try again */ 1181 bp->b_flags |= B_INVAL; 1182 brelse(bp); 1183 ++retries; 1184 if (retries > LFS_IFIND_RETRIES) { 1185 #ifdef DEBUG 1186 /* If the seglock is held look at the bpp to see 1187 what is there anyway */ 1188 simple_lock(&fs->lfs_interlock); 1189 if (fs->lfs_seglock > 0) { 1190 struct buf **bpp; 1191 struct ufs1_dinode *dp; 1192 int i; 1193 1194 for (bpp = fs->lfs_sp->bpp; 1195 bpp != fs->lfs_sp->cbpp; ++bpp) { 1196 if ((*bpp)->b_vp == fs->lfs_ivnode && 1197 bpp != fs->lfs_sp->bpp) { 1198 /* Inode block */ 1199 printf("lfs_vget: block 0x%" PRIx64 ": ", 1200 (*bpp)->b_blkno); 1201 dp = (struct ufs1_dinode *)(*bpp)->b_data; 1202 for (i = 0; i < INOPB(fs); i++) 1203 if (dp[i].di_u.inumber) 1204 printf("%d ", dp[i].di_u.inumber); 1205 printf("\n"); 1206 } 1207 } 1208 } 1209 simple_unlock(&fs->lfs_interlock); 1210 #endif /* DEBUG */ 1211 panic("lfs_vget: dinode not found"); 1212 } 1213 simple_lock(&fs->lfs_interlock); 1214 if (fs->lfs_iocount) { 1215 DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino)); 1216 (void)ltsleep(&fs->lfs_iocount, PRIBIO + 1, 1217 "lfs ifind", 1, &fs->lfs_interlock); 1218 } else 1219 retries = LFS_IFIND_RETRIES; 1220 simple_unlock(&fs->lfs_interlock); 1221 goto again; 1222 } 1223 *ip->i_din.ffs1_din = *dip; 1224 brelse(bp); 1225 1226 if (fs->lfs_version > 1) { 1227 ip->i_ffs1_atime = ts.tv_sec; 1228 ip->i_ffs1_atimensec = ts.tv_nsec; 1229 } 1230 1231 lfs_vinit(mp, &vp); 1232 1233 *vpp = vp; 1234 1235 KASSERT(VOP_ISLOCKED(vp)); 1236 1237 return (0); 1238 } 1239 1240 /* 1241 * File handle to vnode 1242 */ 1243 int 1244 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1245 { 1246 struct lfid lfh; 1247 struct buf *bp; 1248 IFILE *ifp; 1249 int32_t daddr; 1250 struct lfs *fs; 1251 1252 if (fhp->fid_len != sizeof(struct lfid)) 1253 return EINVAL; 1254 1255 memcpy(&lfh, fhp, sizeof(lfh)); 1256 if (lfh.lfid_ino < LFS_IFILE_INUM) 1257 return ESTALE; 1258 1259 fs = VFSTOUFS(mp)->um_lfs; 1260 if (lfh.lfid_ident != fs->lfs_ident) 1261 return ESTALE; 1262 1263 if (lfh.lfid_ino > 1264 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) - 1265 fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb) 1266 return ESTALE; 1267 1268 if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino) == NULLVP) { 1269 LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp); 1270 daddr = ifp->if_daddr; 1271 brelse(bp); 1272 if (daddr == LFS_UNUSED_DADDR) 1273 return ESTALE; 1274 } 1275 1276 return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp)); 1277 } 1278 1279 /* 1280 * Vnode pointer to File handle 1281 */ 1282 /* ARGSUSED */ 1283 int 1284 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 1285 { 1286 struct inode *ip; 1287 struct lfid lfh; 1288 1289 if (*fh_size < sizeof(struct lfid)) { 1290 *fh_size = sizeof(struct lfid); 1291 return E2BIG; 1292 } 1293 *fh_size = sizeof(struct lfid); 1294 ip = VTOI(vp); 1295 memset(&lfh, 0, sizeof(lfh)); 1296 lfh.lfid_len = sizeof(struct lfid); 1297 lfh.lfid_ino = ip->i_number; 1298 lfh.lfid_gen = ip->i_gen; 1299 lfh.lfid_ident = ip->i_lfs->lfs_ident; 1300 memcpy(fhp, &lfh, sizeof(lfh)); 1301 return (0); 1302 } 1303 1304 static int 1305 sysctl_lfs_dostats(SYSCTLFN_ARGS) 1306 { 1307 extern struct lfs_stats lfs_stats; 1308 extern int lfs_dostats; 1309 int error; 1310 1311 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 1312 if (error || newp == NULL) 1313 return (error); 1314 1315 if (lfs_dostats == 0) 1316 memset(&lfs_stats, 0, sizeof(lfs_stats)); 1317 1318 return (0); 1319 } 1320 1321 struct shortlong { 1322 const char *sname; 1323 const char *lname; 1324 }; 1325 1326 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup") 1327 { 1328 int i; 1329 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 1330 lfs_fs_pagetrip; 1331 #ifdef DEBUG 1332 extern int lfs_debug_log_subsys[DLOG_MAX]; 1333 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 1334 { "rollforward", "Debug roll-forward code" }, 1335 { "alloc", "Debug inode allocation and free list" }, 1336 { "avail", "Debug space-available-now accounting" }, 1337 { "flush", "Debug flush triggers" }, 1338 { "lockedlist", "Debug locked list accounting" }, 1339 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 1340 { "vnode", "Debug vnode use during segment write" }, 1341 { "segment", "Debug segment writing" }, 1342 { "seguse", "Debug segment used-bytes accounting" }, 1343 { "cleaner", "Debug cleaning routines" }, 1344 { "mount", "Debug mount/unmount routines" }, 1345 { "pagecache", "Debug UBC interactions" }, 1346 { "dirop", "Debug directory-operation accounting" }, 1347 { "malloc", "Debug private malloc accounting" }, 1348 }; 1349 #endif /* DEBUG */ 1350 struct shortlong stat_names[] = { /* Must match lfs.h! */ 1351 { "segsused", "Number of new segments allocated" }, 1352 { "psegwrites", "Number of partial-segment writes" }, 1353 { "psyncwrites", "Number of synchronous partial-segment" 1354 " writes" }, 1355 { "pcleanwrites", "Number of partial-segment writes by the" 1356 " cleaner" }, 1357 { "blocktot", "Number of blocks written" }, 1358 { "cleanblocks", "Number of blocks written by the cleaner" }, 1359 { "ncheckpoints", "Number of checkpoints made" }, 1360 { "nwrites", "Number of whole writes" }, 1361 { "nsync_writes", "Number of synchronous writes" }, 1362 { "wait_exceeded", "Number of times writer waited for" 1363 " cleaner" }, 1364 { "write_exceeded", "Number of times writer invoked flush" }, 1365 { "flush_invoked", "Number of times flush was invoked" }, 1366 { "vflush_invoked", "Number of time vflush was called" }, 1367 { "clean_inlocked", "Number of vnodes skipped for VXLOCK" }, 1368 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 1369 { "segs_reclaimed", "Number of segments reclaimed" }, 1370 }; 1371 1372 sysctl_createv(clog, 0, NULL, NULL, 1373 CTLFLAG_PERMANENT, 1374 CTLTYPE_NODE, "vfs", NULL, 1375 NULL, 0, NULL, 0, 1376 CTL_VFS, CTL_EOL); 1377 sysctl_createv(clog, 0, NULL, NULL, 1378 CTLFLAG_PERMANENT, 1379 CTLTYPE_NODE, "lfs", 1380 SYSCTL_DESCR("Log-structured file system"), 1381 NULL, 0, NULL, 0, 1382 CTL_VFS, 5, CTL_EOL); 1383 /* 1384 * XXX the "5" above could be dynamic, thereby eliminating one 1385 * more instance of the "number to vfs" mapping problem, but 1386 * "5" is the order as taken from sys/mount.h 1387 */ 1388 1389 sysctl_createv(clog, 0, NULL, NULL, 1390 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1391 CTLTYPE_INT, "flushindir", NULL, 1392 NULL, 0, &lfs_writeindir, 0, 1393 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 1394 sysctl_createv(clog, 0, NULL, NULL, 1395 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1396 CTLTYPE_INT, "clean_vnhead", NULL, 1397 NULL, 0, &lfs_clean_vnhead, 0, 1398 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 1399 sysctl_createv(clog, 0, NULL, NULL, 1400 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1401 CTLTYPE_INT, "dostats", 1402 SYSCTL_DESCR("Maintain statistics on LFS operations"), 1403 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 1404 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 1405 sysctl_createv(clog, 0, NULL, NULL, 1406 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1407 CTLTYPE_INT, "pagetrip", 1408 SYSCTL_DESCR("How many dirty pages in fs triggers" 1409 " a flush"), 1410 NULL, 0, &lfs_fs_pagetrip, 0, 1411 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 1412 #ifdef LFS_KERNEL_RFW 1413 sysctl_createv(clog, 0, NULL, NULL, 1414 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1415 CTLTYPE_INT, "rfw", 1416 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 1417 NULL, 0, &lfs_do_rfw, 0, 1418 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 1419 #endif 1420 1421 sysctl_createv(clog, 0, NULL, NULL, 1422 CTLFLAG_PERMANENT, 1423 CTLTYPE_NODE, "stats", 1424 SYSCTL_DESCR("Debugging options"), 1425 NULL, 0, NULL, 0, 1426 CTL_VFS, 5, LFS_STATS, CTL_EOL); 1427 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 1428 sysctl_createv(clog, 0, NULL, NULL, 1429 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1430 CTLTYPE_INT, stat_names[i].sname, 1431 SYSCTL_DESCR(stat_names[i].lname), 1432 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 1433 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 1434 } 1435 1436 #ifdef DEBUG 1437 sysctl_createv(clog, 0, NULL, NULL, 1438 CTLFLAG_PERMANENT, 1439 CTLTYPE_NODE, "debug", 1440 SYSCTL_DESCR("Debugging options"), 1441 NULL, 0, NULL, 0, 1442 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 1443 for (i = 0; i < DLOG_MAX; i++) { 1444 sysctl_createv(clog, 0, NULL, NULL, 1445 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1446 CTLTYPE_INT, dlog_names[i].sname, 1447 SYSCTL_DESCR(dlog_names[i].lname), 1448 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 1449 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 1450 } 1451 #endif 1452 } 1453 1454 /* 1455 * ufs_bmaparray callback function for writing. 1456 * 1457 * Since blocks will be written to the new segment anyway, 1458 * we don't care about current daddr of them. 1459 */ 1460 static boolean_t 1461 lfs_issequential_hole(const struct ufsmount *ump, 1462 daddr_t daddr0, daddr_t daddr1) 1463 { 1464 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */ 1465 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */ 1466 1467 KASSERT(daddr0 == UNWRITTEN || 1468 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR)); 1469 KASSERT(daddr1 == UNWRITTEN || 1470 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR)); 1471 1472 /* NOTE: all we want to know here is 'hole or not'. */ 1473 /* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */ 1474 1475 /* 1476 * treat UNWRITTENs and all resident blocks as 'contiguous' 1477 */ 1478 if (daddr0 != 0 && daddr1 != 0) 1479 return TRUE; 1480 1481 /* 1482 * both are in hole? 1483 */ 1484 if (daddr0 == 0 && daddr1 == 0) 1485 return TRUE; /* all holes are 'contiguous' for us. */ 1486 1487 return FALSE; 1488 } 1489 1490 /* 1491 * lfs_gop_write functions exactly like genfs_gop_write, except that 1492 * (1) it requires the seglock to be held by its caller, and sp->fip 1493 * to be properly initialized (it will return without re-initializing 1494 * sp->fip, and without calling lfs_writeseg). 1495 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1496 * to determine how large a block it can write at once (though it does 1497 * still use VOP_BMAP to find holes in the file); 1498 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1499 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1500 * now have clusters of clusters, ick.) 1501 */ 1502 static int 1503 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags) 1504 { 1505 int i, s, error, run, haveeof = 0; 1506 int fs_bshift; 1507 vaddr_t kva; 1508 off_t eof, offset, startoffset = 0; 1509 size_t bytes, iobytes, skipbytes; 1510 daddr_t lbn, blkno; 1511 struct vm_page *pg; 1512 struct buf *mbp, *bp; 1513 struct vnode *devvp = VTOI(vp)->i_devvp; 1514 struct inode *ip = VTOI(vp); 1515 struct lfs *fs = ip->i_lfs; 1516 struct segment *sp = fs->lfs_sp; 1517 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1518 1519 ASSERT_SEGLOCK(fs); 1520 1521 /* The Ifile lives in the buffer cache */ 1522 KASSERT(vp != fs->lfs_ivnode); 1523 1524 /* 1525 * We don't want to fill the disk before the cleaner has a chance 1526 * to make room for us. If we're in danger of doing that, fail 1527 * with EAGAIN. The caller will have to notice this, unlock 1528 * so the cleaner can run, relock and try again. 1529 * 1530 * We must write everything, however, if our vnode is being 1531 * reclaimed. 1532 */ 1533 if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp) 1534 goto tryagain; 1535 1536 /* 1537 * Sometimes things slip past the filters in lfs_putpages, 1538 * and the pagedaemon tries to write pages---problem is 1539 * that the pagedaemon never acquires the segment lock. 1540 * 1541 * Alternatively, pages that were clean when we called 1542 * genfs_putpages may have become dirty in the meantime. In this 1543 * case the segment header is not properly set up for blocks 1544 * to be added to it. 1545 * 1546 * Unbusy and unclean the pages, and put them on the ACTIVE 1547 * queue under the hypothesis that they couldn't have got here 1548 * unless they were modified *quite* recently. 1549 * 1550 * XXXUBC that last statement is an oversimplification of course. 1551 */ 1552 if (!LFS_SEGLOCK_HELD(fs) || 1553 (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) || 1554 (pgs[0]->offset & fs->lfs_bmask) != 0) { 1555 goto tryagain; 1556 } 1557 1558 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x", 1559 vp, pgs, npages, flags); 1560 1561 GOP_SIZE(vp, vp->v_size, &eof, 0); 1562 haveeof = 1; 1563 1564 if (vp->v_type == VREG) 1565 fs_bshift = vp->v_mount->mnt_fs_bshift; 1566 else 1567 fs_bshift = DEV_BSHIFT; 1568 error = 0; 1569 pg = pgs[0]; 1570 startoffset = pg->offset; 1571 KASSERT(eof >= 0); 1572 if (startoffset >= eof) { 1573 goto tryagain; 1574 } else 1575 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 1576 skipbytes = 0; 1577 1578 KASSERT(bytes != 0); 1579 1580 /* Swap PG_DELWRI for PG_PAGEOUT */ 1581 for (i = 0; i < npages; i++) 1582 if (pgs[i]->flags & PG_DELWRI) { 1583 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 1584 pgs[i]->flags &= ~PG_DELWRI; 1585 pgs[i]->flags |= PG_PAGEOUT; 1586 uvmexp.paging++; 1587 uvm_lock_pageq(); 1588 uvm_pageunwire(pgs[i]); 1589 uvm_unlock_pageq(); 1590 } 1591 1592 /* 1593 * Check to make sure we're starting on a block boundary. 1594 * We'll check later to make sure we always write entire 1595 * blocks (or fragments). 1596 */ 1597 if (startoffset & fs->lfs_bmask) 1598 printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n", 1599 startoffset, fs->lfs_bmask, 1600 startoffset & fs->lfs_bmask); 1601 KASSERT((startoffset & fs->lfs_bmask) == 0); 1602 if (bytes & fs->lfs_ffmask) { 1603 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 1604 panic("lfs_gop_write: non-integer blocks"); 1605 } 1606 1607 /* 1608 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 1609 * If we would, write what we have and try again. If we don't 1610 * have anything to write, we'll have to sleep. 1611 */ 1612 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1613 (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ? 1614 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 1615 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 1616 #if 0 1617 " with nfinfo=%d at offset 0x%x\n", 1618 (int)((SEGSUM *)(sp->segsum))->ss_nfinfo, 1619 (unsigned)fs->lfs_offset)); 1620 #endif 1621 lfs_updatemeta(sp); 1622 lfs_release_finfo(fs); 1623 (void) lfs_writeseg(fs, sp); 1624 1625 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 1626 1627 /* 1628 * Having given up all of the pager_map we were holding, 1629 * we can now wait for aiodoned to reclaim it for us 1630 * without fear of deadlock. 1631 */ 1632 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1633 UVMPAGER_MAPIN_WAITOK); 1634 } 1635 1636 s = splbio(); 1637 simple_lock(&global_v_numoutput_slock); 1638 vp->v_numoutput += 2; /* one for biodone, one for aiodone */ 1639 simple_unlock(&global_v_numoutput_slock); 1640 splx(s); 1641 1642 mbp = getiobuf(); 1643 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 1644 vp, mbp, vp->v_numoutput, bytes); 1645 mbp->b_bufsize = npages << PAGE_SHIFT; 1646 mbp->b_data = (void *)kva; 1647 mbp->b_resid = mbp->b_bcount = bytes; 1648 mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL; 1649 mbp->b_iodone = uvm_aio_biodone; 1650 mbp->b_vp = vp; 1651 1652 bp = NULL; 1653 for (offset = startoffset; 1654 bytes > 0; 1655 offset += iobytes, bytes -= iobytes) { 1656 lbn = offset >> fs_bshift; 1657 error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 1658 lfs_issequential_hole); 1659 if (error) { 1660 UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d", 1661 error,0,0,0); 1662 skipbytes += bytes; 1663 bytes = 0; 1664 break; 1665 } 1666 1667 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 1668 bytes); 1669 if (blkno == (daddr_t)-1) { 1670 skipbytes += iobytes; 1671 continue; 1672 } 1673 1674 /* 1675 * Discover how much we can really pack into this buffer. 1676 */ 1677 /* If no room in the current segment, finish it up */ 1678 if (sp->sum_bytes_left < sizeof(int32_t) || 1679 sp->seg_bytes_left < (1 << fs->lfs_bshift)) { 1680 int vers; 1681 1682 lfs_updatemeta(sp); 1683 vers = sp->fip->fi_version; 1684 lfs_release_finfo(fs); 1685 (void) lfs_writeseg(fs, sp); 1686 1687 lfs_acquire_finfo(fs, ip->i_number, vers); 1688 } 1689 /* Check both for space in segment and space in segsum */ 1690 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 1691 << fs_bshift); 1692 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 1693 << fs_bshift); 1694 KASSERT(iobytes > 0); 1695 1696 /* if it's really one i/o, don't make a second buf */ 1697 if (offset == startoffset && iobytes == bytes) { 1698 bp = mbp; 1699 /* correct overcount if there is no second buffer */ 1700 s = splbio(); 1701 simple_lock(&global_v_numoutput_slock); 1702 --vp->v_numoutput; 1703 simple_unlock(&global_v_numoutput_slock); 1704 splx(s); 1705 } else { 1706 bp = getiobuf(); 1707 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 1708 vp, bp, vp->v_numoutput, 0); 1709 bp->b_data = (char *)kva + 1710 (vaddr_t)(offset - pg->offset); 1711 bp->b_resid = bp->b_bcount = iobytes; 1712 bp->b_flags = B_BUSY|B_WRITE|B_CALL; 1713 bp->b_iodone = uvm_aio_biodone1; 1714 } 1715 1716 /* XXX This is silly ... is this necessary? */ 1717 bp->b_vp = NULL; 1718 s = splbio(); 1719 bgetvp(vp, bp); 1720 splx(s); 1721 1722 bp->b_lblkno = lblkno(fs, offset); 1723 bp->b_private = mbp; 1724 if (devvp->v_type == VBLK) { 1725 bp->b_dev = devvp->v_rdev; 1726 } 1727 VOP_BWRITE(bp); 1728 while (lfs_gatherblock(sp, bp, NULL)) 1729 continue; 1730 } 1731 1732 if (skipbytes) { 1733 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0); 1734 s = splbio(); 1735 if (error) { 1736 mbp->b_flags |= B_ERROR; 1737 mbp->b_error = error; 1738 } 1739 mbp->b_resid -= skipbytes; 1740 if (mbp->b_resid == 0) { 1741 biodone(mbp); 1742 } 1743 splx(s); 1744 } 1745 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 1746 return (0); 1747 1748 tryagain: 1749 /* 1750 * We can't write the pages, for whatever reason. 1751 * Clean up after ourselves, and make the caller try again. 1752 */ 1753 simple_lock(&vp->v_interlock); 1754 1755 /* Tell why we're here, if we know */ 1756 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) 1757 DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n")); 1758 else if ((pgs[0]->offset & fs->lfs_bmask) != 0) 1759 DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n")); 1760 else if (haveeof && startoffset >= eof) 1761 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 1762 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 1763 pgs[0]->offset, eof, npages)); 1764 else if (LFS_STARVED_FOR_SEGS(fs)) 1765 DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n")); 1766 else 1767 DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n")); 1768 1769 uvm_lock_pageq(); 1770 for (i = 0; i < npages; i++) { 1771 pg = pgs[i]; 1772 1773 if (pg->flags & PG_PAGEOUT) 1774 uvmexp.paging--; 1775 if (pg->flags & PG_DELWRI) { 1776 uvm_pageunwire(pg); 1777 } 1778 uvm_pageactivate(pg); 1779 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 1780 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 1781 vp, pg->offset)); 1782 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 1783 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 1784 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 1785 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 1786 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 1787 pg->wire_count)); 1788 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 1789 pg->loan_count)); 1790 } 1791 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */ 1792 uvm_page_unbusy(pgs, npages); 1793 uvm_unlock_pageq(); 1794 simple_unlock(&vp->v_interlock); 1795 return EAGAIN; 1796 } 1797 1798 /* 1799 * finish vnode/inode initialization. 1800 * used by lfs_vget and lfs_fastvget. 1801 */ 1802 void 1803 lfs_vinit(struct mount *mp, struct vnode **vpp) 1804 { 1805 struct vnode *vp = *vpp; 1806 struct inode *ip = VTOI(vp); 1807 struct ufsmount *ump = VFSTOUFS(mp); 1808 int i; 1809 1810 ip->i_mode = ip->i_ffs1_mode; 1811 ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink; 1812 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size; 1813 ip->i_flags = ip->i_ffs1_flags; 1814 ip->i_gen = ip->i_ffs1_gen; 1815 ip->i_uid = ip->i_ffs1_uid; 1816 ip->i_gid = ip->i_ffs1_gid; 1817 1818 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 1819 ip->i_lfs_odnlink = ip->i_ffs1_nlink; 1820 1821 /* 1822 * Initialize the vnode from the inode, check for aliases. In all 1823 * cases re-init ip, the underlying vnode/inode may have changed. 1824 */ 1825 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 1826 ip = VTOI(vp); 1827 1828 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 1829 if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) { 1830 struct lfs *fs = ump->um_lfs; 1831 #ifdef DEBUG 1832 for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift; 1833 i < NDADDR; i++) { 1834 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 1835 i == 0) 1836 continue; 1837 if (ip->i_ffs1_db[i] != 0) { 1838 inconsistent: 1839 lfs_dump_dinode(ip->i_din.ffs1_din); 1840 panic("inconsistent inode"); 1841 } 1842 } 1843 for ( ; i < NDADDR + NIADDR; i++) { 1844 if (ip->i_ffs1_ib[i - NDADDR] != 0) { 1845 goto inconsistent; 1846 } 1847 } 1848 #endif /* DEBUG */ 1849 for (i = 0; i < NDADDR; i++) 1850 if (ip->i_ffs1_db[i] != 0) 1851 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 1852 } 1853 1854 #ifdef DIAGNOSTIC 1855 if (vp->v_type == VNON) { 1856 # ifdef DEBUG 1857 lfs_dump_dinode(ip->i_din.ffs1_din); 1858 # endif 1859 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 1860 (unsigned long long)ip->i_number, 1861 (ip->i_mode & IFMT) >> 12); 1862 } 1863 #endif /* DIAGNOSTIC */ 1864 1865 /* 1866 * Finish inode initialization now that aliasing has been resolved. 1867 */ 1868 1869 ip->i_devvp = ump->um_devvp; 1870 VREF(ip->i_devvp); 1871 genfs_node_init(vp, &lfs_genfsops); 1872 uvm_vnp_setsize(vp, ip->i_size); 1873 1874 /* Initialize hiblk from file size */ 1875 ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1; 1876 1877 *vpp = vp; 1878 } 1879 1880 /* 1881 * Resize the filesystem to contain the specified number of segments. 1882 */ 1883 int 1884 lfs_resize_fs(struct lfs *fs, int newnsegs) 1885 { 1886 SEGUSE *sup; 1887 struct buf *bp, *obp; 1888 daddr_t olast, nlast, ilast, noff, start, end; 1889 struct vnode *ivp; 1890 struct inode *ip; 1891 int error, badnews, inc, oldnsegs; 1892 int sbbytes, csbbytes, gain, cgain; 1893 int i; 1894 1895 /* Only support v2 and up */ 1896 if (fs->lfs_version < 2) 1897 return EOPNOTSUPP; 1898 1899 /* If we're doing nothing, do it fast */ 1900 oldnsegs = fs->lfs_nseg; 1901 if (newnsegs == oldnsegs) 1902 return 0; 1903 1904 /* We always have to have two superblocks */ 1905 if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1])) 1906 return EFBIG; 1907 1908 ivp = fs->lfs_ivnode; 1909 ip = VTOI(ivp); 1910 error = 0; 1911 1912 /* Take the segment lock so no one else calls lfs_newseg() */ 1913 lfs_seglock(fs, SEGM_PROT); 1914 1915 /* 1916 * Make sure the segments we're going to be losing, if any, 1917 * are in fact empty. We hold the seglock, so their status 1918 * cannot change underneath us. Count the superblocks we lose, 1919 * while we're at it. 1920 */ 1921 sbbytes = csbbytes = 0; 1922 cgain = 0; 1923 for (i = newnsegs; i < oldnsegs; i++) { 1924 LFS_SEGENTRY(sup, fs, i, bp); 1925 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 1926 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1927 sbbytes += LFS_SBPAD; 1928 if (!(sup->su_flags & SEGUSE_DIRTY)) { 1929 ++cgain; 1930 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1931 csbbytes += LFS_SBPAD; 1932 } 1933 brelse(bp); 1934 if (badnews) { 1935 error = EBUSY; 1936 goto out; 1937 } 1938 } 1939 1940 /* Note old and new segment table endpoints, and old ifile size */ 1941 olast = fs->lfs_cleansz + fs->lfs_segtabsz; 1942 nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz; 1943 ilast = ivp->v_size >> fs->lfs_bshift; 1944 noff = nlast - olast; 1945 1946 /* 1947 * Make sure no one can use the Ifile while we change it around. 1948 * Even after taking the iflock we need to make sure no one still 1949 * is holding Ifile buffers, so we get each one, to drain them. 1950 * (XXX this could be done better.) 1951 */ 1952 simple_lock(&fs->lfs_interlock); 1953 lockmgr(&fs->lfs_iflock, LK_EXCLUSIVE, &fs->lfs_interlock); 1954 simple_unlock(&fs->lfs_interlock); 1955 vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY); 1956 for (i = 0; i < ilast; i++) { 1957 bread(ivp, i, fs->lfs_bsize, NOCRED, &bp); 1958 brelse(bp); 1959 } 1960 1961 /* Allocate new Ifile blocks */ 1962 for (i = ilast; i < ilast + noff; i++) { 1963 if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0, 1964 &bp) != 0) 1965 panic("balloc extending ifile"); 1966 memset(bp->b_data, 0, fs->lfs_bsize); 1967 VOP_BWRITE(bp); 1968 } 1969 1970 /* Register new ifile size */ 1971 ip->i_size += noff * fs->lfs_bsize; 1972 ip->i_ffs1_size = ip->i_size; 1973 uvm_vnp_setsize(ivp, ip->i_size); 1974 1975 /* Copy the inode table to its new position */ 1976 if (noff != 0) { 1977 if (noff < 0) { 1978 start = nlast; 1979 end = ilast + noff; 1980 inc = 1; 1981 } else { 1982 start = ilast + noff - 1; 1983 end = nlast - 1; 1984 inc = -1; 1985 } 1986 for (i = start; i != end; i += inc) { 1987 if (bread(ivp, i, fs->lfs_bsize, NOCRED, &bp) != 0) 1988 panic("resize: bread dst blk failed"); 1989 if (bread(ivp, i - noff, fs->lfs_bsize, NOCRED, &obp)) 1990 panic("resize: bread src blk failed"); 1991 memcpy(bp->b_data, obp->b_data, fs->lfs_bsize); 1992 VOP_BWRITE(bp); 1993 brelse(obp); 1994 } 1995 } 1996 1997 /* If we are expanding, write the new empty SEGUSE entries */ 1998 if (newnsegs > oldnsegs) { 1999 for (i = oldnsegs; i < newnsegs; i++) { 2000 if ((error = bread(ivp, i / fs->lfs_sepb + 2001 fs->lfs_cleansz, 2002 fs->lfs_bsize, NOCRED, &bp)) != 0) 2003 panic("lfs: ifile read: %d", error); 2004 while ((i + 1) % fs->lfs_sepb && i < newnsegs) { 2005 sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb]; 2006 memset(sup, 0, sizeof(*sup)); 2007 i++; 2008 } 2009 VOP_BWRITE(bp); 2010 } 2011 } 2012 2013 /* Zero out unused superblock offsets */ 2014 for (i = 2; i < LFS_MAXNUMSB; i++) 2015 if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs) 2016 fs->lfs_sboffs[i] = 0x0; 2017 2018 /* 2019 * Correct superblock entries that depend on fs size. 2020 * The computations of these are as follows: 2021 * 2022 * size = segtod(fs, nseg) 2023 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD) 2024 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2025 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD) 2026 * + (segtod(fs, 1) - (offset - curseg)) 2027 * - segtod(fs, minfreeseg - (minfreeseg / 2)) 2028 * 2029 * XXX - we should probably adjust minfreeseg as well. 2030 */ 2031 gain = (newnsegs - oldnsegs); 2032 fs->lfs_nseg = newnsegs; 2033 fs->lfs_segtabsz = nlast - fs->lfs_cleansz; 2034 fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize); 2035 fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes); 2036 fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes) 2037 - gain * btofsb(fs, fs->lfs_bsize / 2); 2038 if (gain > 0) { 2039 fs->lfs_nclean += gain; 2040 fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize); 2041 } else { 2042 fs->lfs_nclean -= cgain; 2043 fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) - 2044 btofsb(fs, csbbytes); 2045 } 2046 2047 /* Resize segment flag cache */ 2048 fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2049 fs->lfs_nseg * sizeof(u_int32_t), 2050 M_SEGMENT, M_WAITOK); 2051 fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[1], 2052 fs->lfs_nseg * sizeof(u_int32_t), 2053 M_SEGMENT, M_WAITOK); 2054 for (i = oldnsegs; i < newnsegs; i++) 2055 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2056 2057 /* Truncate Ifile if necessary */ 2058 if (noff < 0) 2059 lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0, 2060 NOCRED, curlwp); 2061 2062 /* Update cleaner info so the cleaner can die */ 2063 bread(ivp, 0, fs->lfs_bsize, NOCRED, &bp); 2064 ((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean; 2065 ((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean; 2066 VOP_BWRITE(bp); 2067 2068 /* Let Ifile accesses proceed */ 2069 VOP_UNLOCK(ivp, 0); 2070 simple_lock(&fs->lfs_interlock); 2071 lockmgr(&fs->lfs_iflock, LK_RELEASE, &fs->lfs_interlock); 2072 simple_unlock(&fs->lfs_interlock); 2073 2074 out: 2075 lfs_segunlock(fs); 2076 return error; 2077 } 2078
Indexes created Mon Sep 29 21:09:56 GMT 2025