lfs_vfsops.c revision 1.257
1 /* $NetBSD: lfs_vfsops.c,v 1.257 2008/04/29 18:18:09 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003, 2007, 2007 5 * The NetBSD Foundation, Inc. 6 * All rights reserved. 7 * 8 * This code is derived from software contributed to The NetBSD Foundation 9 * by Konrad E. Schroder <perseant (at) hhhh.org>. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 /*- 33 * Copyright (c) 1989, 1991, 1993, 1994 34 * The Regents of the University of California. All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. Neither the name of the University nor the names of its contributors 45 * may be used to endorse or promote products derived from this software 46 * without specific prior written permission. 47 * 48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 58 * SUCH DAMAGE. 59 * 60 * @(#)lfs_vfsops.c 8.20 (Berkeley) 6/10/95 61 */ 62 63 #include <sys/cdefs.h> 64 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.257 2008/04/29 18:18:09 ad Exp $"); 65 66 #if defined(_KERNEL_OPT) 67 #include "opt_lfs.h" 68 #include "opt_quota.h" 69 #endif 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/namei.h> 74 #include <sys/proc.h> 75 #include <sys/kernel.h> 76 #include <sys/vnode.h> 77 #include <sys/mount.h> 78 #include <sys/kthread.h> 79 #include <sys/buf.h> 80 #include <sys/device.h> 81 #include <sys/mbuf.h> 82 #include <sys/file.h> 83 #include <sys/disklabel.h> 84 #include <sys/ioctl.h> 85 #include <sys/errno.h> 86 #include <sys/malloc.h> 87 #include <sys/pool.h> 88 #include <sys/socket.h> 89 #include <sys/syslog.h> 90 #include <uvm/uvm_extern.h> 91 #include <sys/sysctl.h> 92 #include <sys/conf.h> 93 #include <sys/kauth.h> 94 95 #include <miscfs/specfs/specdev.h> 96 97 #include <ufs/ufs/quota.h> 98 #include <ufs/ufs/inode.h> 99 #include <ufs/ufs/ufsmount.h> 100 #include <ufs/ufs/ufs_extern.h> 101 102 #include <uvm/uvm.h> 103 #include <uvm/uvm_stat.h> 104 #include <uvm/uvm_pager.h> 105 #include <uvm/uvm_pdaemon.h> 106 107 #include <ufs/lfs/lfs.h> 108 #include <ufs/lfs/lfs_extern.h> 109 110 #include <miscfs/genfs/genfs.h> 111 #include <miscfs/genfs/genfs_node.h> 112 113 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int); 114 static bool lfs_issequential_hole(const struct ufsmount *, 115 daddr_t, daddr_t); 116 117 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *); 118 119 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc; 120 extern const struct vnodeopv_desc lfs_specop_opv_desc; 121 extern const struct vnodeopv_desc lfs_fifoop_opv_desc; 122 123 pid_t lfs_writer_daemon = 0; 124 int lfs_do_flush = 0; 125 #ifdef LFS_KERNEL_RFW 126 int lfs_do_rfw = 0; 127 #endif 128 129 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = { 130 &lfs_vnodeop_opv_desc, 131 &lfs_specop_opv_desc, 132 &lfs_fifoop_opv_desc, 133 NULL, 134 }; 135 136 struct vfsops lfs_vfsops = { 137 MOUNT_LFS, 138 sizeof (struct ufs_args), 139 lfs_mount, 140 ufs_start, 141 lfs_unmount, 142 ufs_root, 143 ufs_quotactl, 144 lfs_statvfs, 145 lfs_sync, 146 lfs_vget, 147 lfs_fhtovp, 148 lfs_vptofh, 149 lfs_init, 150 lfs_reinit, 151 lfs_done, 152 lfs_mountroot, 153 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 154 vfs_stdextattrctl, 155 (void *)eopnotsupp, /* vfs_suspendctl */ 156 genfs_renamelock_enter, 157 genfs_renamelock_exit, 158 (void *)eopnotsupp, 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 struct pool lfs_inode_pool; 185 struct pool lfs_dinode_pool; 186 struct pool lfs_inoext_pool; 187 struct pool lfs_lbnentry_pool; 188 189 /* 190 * The writer daemon. UVM keeps track of how many dirty pages we are holding 191 * in lfs_subsys_pages; the daemon flushes the filesystem when this value 192 * crosses the (user-defined) threshhold LFS_MAX_PAGES. 193 */ 194 static void 195 lfs_writerd(void *arg) 196 { 197 struct mount *mp, *nmp; 198 struct lfs *fs; 199 int fsflags; 200 int loopcount; 201 202 lfs_writer_daemon = curproc->p_pid; 203 204 mutex_enter(&lfs_lock); 205 for (;;) { 206 mtsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10, 207 &lfs_lock); 208 209 /* 210 * Look through the list of LFSs to see if any of them 211 * have requested pageouts. 212 */ 213 mutex_enter(&mountlist_lock); 214 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 215 mp = nmp) { 216 if (vfs_trybusy(mp, RW_WRITER, &mountlist_lock)) { 217 nmp = CIRCLEQ_NEXT(mp, mnt_list); 218 continue; 219 } 220 if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS, 221 sizeof(mp->mnt_stat.f_fstypename)) == 0) { 222 fs = VFSTOUFS(mp)->um_lfs; 223 mutex_enter(&lfs_lock); 224 fsflags = 0; 225 if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || 226 lfs_dirvcount > LFS_MAX_DIROP) && 227 fs->lfs_dirops == 0) 228 fsflags |= SEGM_CKP; 229 if (fs->lfs_pdflush) { 230 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n")); 231 fs->lfs_pdflush = 0; 232 lfs_flush_fs(fs, fsflags); 233 mutex_exit(&lfs_lock); 234 } else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) { 235 DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n")); 236 mutex_exit(&lfs_lock); 237 lfs_writer_enter(fs, "wrdirop"); 238 lfs_flush_pchain(fs); 239 lfs_writer_leave(fs); 240 } else 241 mutex_exit(&lfs_lock); 242 } 243 244 mutex_enter(&mountlist_lock); 245 nmp = CIRCLEQ_NEXT(mp, mnt_list); 246 vfs_unbusy(mp, false); 247 } 248 mutex_exit(&mountlist_lock); 249 250 /* 251 * If global state wants a flush, flush everything. 252 */ 253 mutex_enter(&lfs_lock); 254 loopcount = 0; 255 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS || 256 locked_queue_bytes > LFS_MAX_BYTES || 257 lfs_subsys_pages > LFS_MAX_PAGES) { 258 259 if (lfs_do_flush) { 260 DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n")); 261 } 262 if (locked_queue_count > LFS_MAX_BUFS) { 263 DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n", 264 locked_queue_count, LFS_MAX_BUFS)); 265 } 266 if (locked_queue_bytes > LFS_MAX_BYTES) { 267 DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n", 268 locked_queue_bytes, LFS_MAX_BYTES)); 269 } 270 if (lfs_subsys_pages > LFS_MAX_PAGES) { 271 DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n", 272 lfs_subsys_pages, LFS_MAX_PAGES)); 273 } 274 275 lfs_flush(NULL, SEGM_WRITERD, 0); 276 lfs_do_flush = 0; 277 } 278 } 279 /* NOTREACHED */ 280 } 281 282 /* 283 * Initialize the filesystem, most work done by ufs_init. 284 */ 285 void 286 lfs_init() 287 { 288 289 malloc_type_attach(M_SEGMENT); 290 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 291 "lfsinopl", &pool_allocator_nointr, IPL_NONE); 292 pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, 293 "lfsdinopl", &pool_allocator_nointr, IPL_NONE); 294 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 295 "lfsinoextpl", &pool_allocator_nointr, IPL_NONE); 296 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 297 "lfslbnpool", &pool_allocator_nointr, IPL_NONE); 298 ufs_init(); 299 300 #ifdef DEBUG 301 memset(lfs_log, 0, sizeof(lfs_log)); 302 #endif 303 mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE); 304 cv_init(&locked_queue_cv, "lfsbuf"); 305 cv_init(&lfs_writing_cv, "lfsflush"); 306 } 307 308 void 309 lfs_reinit() 310 { 311 ufs_reinit(); 312 } 313 314 void 315 lfs_done() 316 { 317 ufs_done(); 318 mutex_destroy(&lfs_lock); 319 cv_destroy(&locked_queue_cv); 320 cv_destroy(&lfs_writing_cv); 321 pool_destroy(&lfs_inode_pool); 322 pool_destroy(&lfs_dinode_pool); 323 pool_destroy(&lfs_inoext_pool); 324 pool_destroy(&lfs_lbnentry_pool); 325 malloc_type_detach(M_SEGMENT); 326 } 327 328 /* 329 * Called by main() when ufs is going to be mounted as root. 330 */ 331 int 332 lfs_mountroot() 333 { 334 extern struct vnode *rootvp; 335 struct mount *mp; 336 struct lwp *l = curlwp; 337 int error; 338 339 if (device_class(root_device) != DV_DISK) 340 return (ENODEV); 341 342 if (rootdev == NODEV) 343 return (ENODEV); 344 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 345 vrele(rootvp); 346 return (error); 347 } 348 if ((error = lfs_mountfs(rootvp, mp, l))) { 349 vfs_unbusy(mp, false); 350 vfs_destroy(mp); 351 return (error); 352 } 353 mutex_enter(&mountlist_lock); 354 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 355 mutex_exit(&mountlist_lock); 356 (void)lfs_statvfs(mp, &mp->mnt_stat); 357 vfs_unbusy(mp, false); 358 setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp)); 359 return (0); 360 } 361 362 /* 363 * VFS Operations. 364 * 365 * mount system call 366 */ 367 int 368 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 369 { 370 struct lwp *l = curlwp; 371 struct nameidata nd; 372 struct vnode *devvp; 373 struct ufs_args *args = data; 374 struct ufsmount *ump = NULL; 375 struct lfs *fs = NULL; /* LFS */ 376 int error = 0, update; 377 mode_t accessmode; 378 379 if (*data_len < sizeof *args) 380 return EINVAL; 381 382 if (mp->mnt_flag & MNT_GETARGS) { 383 ump = VFSTOUFS(mp); 384 if (ump == NULL) 385 return EIO; 386 args->fspec = NULL; 387 *data_len = sizeof *args; 388 return 0; 389 } 390 391 update = mp->mnt_flag & MNT_UPDATE; 392 393 /* Check arguments */ 394 if (args->fspec != NULL) { 395 /* 396 * Look up the name and verify that it's sane. 397 */ 398 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, args->fspec); 399 if ((error = namei(&nd)) != 0) 400 return (error); 401 devvp = nd.ni_vp; 402 403 if (!update) { 404 /* 405 * Be sure this is a valid block device 406 */ 407 if (devvp->v_type != VBLK) 408 error = ENOTBLK; 409 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 410 error = ENXIO; 411 } else { 412 /* 413 * Be sure we're still naming the same device 414 * used for our initial mount 415 */ 416 ump = VFSTOUFS(mp); 417 if (devvp != ump->um_devvp) 418 error = EINVAL; 419 } 420 } else { 421 if (!update) { 422 /* New mounts must have a filename for the device */ 423 return (EINVAL); 424 } else { 425 /* Use the extant mount */ 426 ump = VFSTOUFS(mp); 427 devvp = ump->um_devvp; 428 vref(devvp); 429 } 430 } 431 432 433 /* 434 * If mount by non-root, then verify that user has necessary 435 * permissions on the device. 436 */ 437 if (error == 0 && kauth_authorize_generic(l->l_cred, 438 KAUTH_GENERIC_ISSUSER, NULL) != 0) { 439 accessmode = VREAD; 440 if (update ? 441 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 442 (mp->mnt_flag & MNT_RDONLY) == 0) 443 accessmode |= VWRITE; 444 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 445 error = VOP_ACCESS(devvp, accessmode, l->l_cred); 446 VOP_UNLOCK(devvp, 0); 447 } 448 449 if (error) { 450 vrele(devvp); 451 return (error); 452 } 453 454 if (!update) { 455 int flags; 456 457 if (mp->mnt_flag & MNT_RDONLY) 458 flags = FREAD; 459 else 460 flags = FREAD|FWRITE; 461 error = VOP_OPEN(devvp, flags, FSCRED); 462 if (error) 463 goto fail; 464 error = lfs_mountfs(devvp, mp, l); /* LFS */ 465 if (error) { 466 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 467 (void)VOP_CLOSE(devvp, flags, NOCRED); 468 VOP_UNLOCK(devvp, 0); 469 goto fail; 470 } 471 472 ump = VFSTOUFS(mp); 473 fs = ump->um_lfs; 474 } else { 475 /* 476 * Update the mount. 477 */ 478 479 /* 480 * The initial mount got a reference on this 481 * device, so drop the one obtained via 482 * namei(), above. 483 */ 484 vrele(devvp); 485 486 ump = VFSTOUFS(mp); 487 fs = ump->um_lfs; 488 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 489 /* 490 * Changing from read-only to read/write. 491 * Note in the superblocks that we're writing. 492 */ 493 fs->lfs_ronly = 0; 494 if (fs->lfs_pflags & LFS_PF_CLEAN) { 495 fs->lfs_pflags &= ~LFS_PF_CLEAN; 496 lfs_writesuper(fs, fs->lfs_sboffs[0]); 497 lfs_writesuper(fs, fs->lfs_sboffs[1]); 498 } 499 } 500 if (args->fspec == NULL) 501 return EINVAL; 502 } 503 504 error = set_statvfs_info(path, UIO_USERSPACE, args->fspec, 505 UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l); 506 if (error == 0) 507 (void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname, 508 sizeof(fs->lfs_fsmnt)); 509 return error; 510 511 fail: 512 vrele(devvp); 513 return (error); 514 } 515 516 517 /* 518 * Common code for mount and mountroot 519 * LFS specific 520 */ 521 int 522 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 523 { 524 struct dlfs *tdfs, *dfs, *adfs; 525 struct lfs *fs; 526 struct ufsmount *ump; 527 struct vnode *vp; 528 struct buf *bp, *abp; 529 struct partinfo dpart; 530 dev_t dev; 531 int error, i, ronly, secsize, fsbsize; 532 kauth_cred_t cred; 533 CLEANERINFO *cip; 534 SEGUSE *sup; 535 daddr_t sb_addr; 536 537 cred = l ? l->l_cred : NOCRED; 538 539 /* 540 * Flush out any old buffers remaining from a previous use. 541 */ 542 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 543 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 544 VOP_UNLOCK(devvp, 0); 545 if (error) 546 return (error); 547 548 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 549 if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred) != 0) 550 secsize = DEV_BSIZE; 551 else 552 secsize = dpart.disklab->d_secsize; 553 554 /* Don't free random space on error. */ 555 bp = NULL; 556 abp = NULL; 557 ump = NULL; 558 559 sb_addr = LFS_LABELPAD / secsize; 560 while (1) { 561 /* Read in the superblock. */ 562 error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp); 563 if (error) 564 goto out; 565 dfs = (struct dlfs *)bp->b_data; 566 567 /* Check the basics. */ 568 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE || 569 dfs->dlfs_version > LFS_VERSION || 570 dfs->dlfs_bsize < sizeof(struct dlfs)) { 571 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 572 error = EINVAL; /* XXX needs translation */ 573 goto out; 574 } 575 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) { 576 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 577 dfs->dlfs_inodefmt)); 578 error = EINVAL; 579 goto out; 580 } 581 582 if (dfs->dlfs_version == 1) 583 fsbsize = secsize; 584 else { 585 fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb + 586 dfs->dlfs_fsbtodb); 587 /* 588 * Could be, if the frag size is large enough, that we 589 * don't have the "real" primary superblock. If that's 590 * the case, get the real one, and try again. 591 */ 592 if (sb_addr != dfs->dlfs_sboffs[0] << 593 dfs->dlfs_fsbtodb) { 594 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 595 " 0x%llx is not right, trying 0x%llx\n", 596 (long long)sb_addr, 597 (long long)(dfs->dlfs_sboffs[0] << 598 dfs->dlfs_fsbtodb))); 599 sb_addr = dfs->dlfs_sboffs[0] << 600 dfs->dlfs_fsbtodb; 601 brelse(bp, 0); 602 continue; 603 } 604 } 605 break; 606 } 607 608 /* 609 * Check the second superblock to see which is newer; then mount 610 * using the older of the two. This is necessary to ensure that 611 * the filesystem is valid if it was not unmounted cleanly. 612 */ 613 614 if (dfs->dlfs_sboffs[1] && 615 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 616 { 617 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize), 618 LFS_SBPAD, cred, &abp); 619 if (error) 620 goto out; 621 adfs = (struct dlfs *)abp->b_data; 622 623 if (dfs->dlfs_version == 1) { 624 /* 1s resolution comparison */ 625 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 626 tdfs = adfs; 627 else 628 tdfs = dfs; 629 } else { 630 /* monotonic infinite-resolution comparison */ 631 if (adfs->dlfs_serial < dfs->dlfs_serial) 632 tdfs = adfs; 633 else 634 tdfs = dfs; 635 } 636 637 /* Check the basics. */ 638 if (tdfs->dlfs_magic != LFS_MAGIC || 639 tdfs->dlfs_bsize > MAXBSIZE || 640 tdfs->dlfs_version > LFS_VERSION || 641 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 642 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 643 " sanity failed\n")); 644 error = EINVAL; /* XXX needs translation */ 645 goto out; 646 } 647 } else { 648 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock" 649 " daddr=0x%x\n", dfs->dlfs_sboffs[1])); 650 error = EINVAL; 651 goto out; 652 } 653 654 /* Allocate the mount structure, copy the superblock into it. */ 655 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO); 656 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 657 658 /* Compatibility */ 659 if (fs->lfs_version < 2) { 660 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 661 fs->lfs_ibsize = fs->lfs_bsize; 662 fs->lfs_start = fs->lfs_sboffs[0]; 663 fs->lfs_tstamp = fs->lfs_otstamp; 664 fs->lfs_fsbtodb = 0; 665 } 666 if (fs->lfs_resvseg == 0) 667 fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \ 668 MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1)); 669 670 /* 671 * If we aren't going to be able to write meaningfully to this 672 * filesystem, and were not mounted readonly, bomb out now. 673 */ 674 if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 675 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 676 " we need BUFPAGES >= %lld\n", 677 (long long)((bufmem_hiwater / bufmem_lowater) * 678 LFS_INVERSE_MAX_BYTES( 679 fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 680 free(fs, M_UFSMNT); 681 error = EFBIG; /* XXX needs translation */ 682 goto out; 683 } 684 685 /* Before rolling forward, lock so vget will sleep for other procs */ 686 if (l != NULL) { 687 fs->lfs_flags = LFS_NOTYET; 688 fs->lfs_rfpid = l->l_proc->p_pid; 689 } 690 691 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 692 ump->um_lfs = fs; 693 ump->um_ops = &lfs_ufsops; 694 ump->um_fstype = UFS1; 695 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 696 brelse(bp, BC_INVAL); 697 brelse(abp, BC_INVAL); 698 } else { 699 brelse(bp, 0); 700 brelse(abp, 0); 701 } 702 bp = NULL; 703 abp = NULL; 704 705 706 /* 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 rw_init(&fs->lfs_fraglock); 726 rw_init(&fs->lfs_iflock); 727 cv_init(&fs->lfs_stopcv, "lfsstop"); 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, 0); 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 && kthread_create(PRI_BIO, 0, NULL, 862 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, 0); 870 if (abp) 871 brelse(abp, 0); 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) 886 { 887 struct lwp *l = curlwp; 888 struct ufsmount *ump; 889 struct lfs *fs; 890 int error, flags, ronly; 891 vnode_t *vp; 892 893 flags = 0; 894 if (mntflags & MNT_FORCE) 895 flags |= FORCECLOSE; 896 897 ump = VFSTOUFS(mp); 898 fs = ump->um_lfs; 899 900 /* Two checkpoints */ 901 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 902 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 903 904 /* wake up the cleaner so it can die */ 905 lfs_wakeup_cleaner(fs); 906 mutex_enter(&lfs_lock); 907 while (fs->lfs_sleepers) 908 mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0, 909 &lfs_lock); 910 mutex_exit(&lfs_lock); 911 912 #ifdef QUOTA 913 if (mp->mnt_flag & MNT_QUOTA) { 914 int i; 915 error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags); 916 if (error) 917 return (error); 918 for (i = 0; i < MAXQUOTAS; i++) { 919 if (ump->um_quotas[i] == NULLVP) 920 continue; 921 quotaoff(l, mp, i); 922 } 923 /* 924 * Here we fall through to vflush again to ensure 925 * that we have gotten rid of all the system vnodes. 926 */ 927 } 928 #endif 929 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 930 return (error); 931 if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0) 932 return (error); 933 vp = fs->lfs_ivnode; 934 mutex_enter(&vp->v_interlock); 935 if (LIST_FIRST(&vp->v_dirtyblkhd)) 936 panic("lfs_unmount: still dirty blocks on ifile vnode"); 937 mutex_exit(&vp->v_interlock); 938 939 /* Explicitly write the superblock, to update serial and pflags */ 940 fs->lfs_pflags |= LFS_PF_CLEAN; 941 lfs_writesuper(fs, fs->lfs_sboffs[0]); 942 lfs_writesuper(fs, fs->lfs_sboffs[1]); 943 mutex_enter(&lfs_lock); 944 while (fs->lfs_iocount) 945 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 946 &lfs_lock); 947 mutex_exit(&lfs_lock); 948 949 /* Finish with the Ifile, now that we're done with it */ 950 vgone(fs->lfs_ivnode); 951 952 ronly = !fs->lfs_ronly; 953 if (ump->um_devvp->v_type != VBAD) 954 ump->um_devvp->v_specmountpoint = NULL; 955 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 956 error = VOP_CLOSE(ump->um_devvp, 957 ronly ? FREAD : FREAD|FWRITE, NOCRED); 958 vput(ump->um_devvp); 959 960 /* Complain about page leakage */ 961 if (fs->lfs_pages > 0) 962 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 963 fs->lfs_pages, lfs_subsys_pages); 964 965 /* Free per-mount data structures */ 966 free(fs->lfs_ino_bitmap, M_SEGMENT); 967 free(fs->lfs_suflags[0], M_SEGMENT); 968 free(fs->lfs_suflags[1], M_SEGMENT); 969 free(fs->lfs_suflags, M_SEGMENT); 970 lfs_free_resblks(fs); 971 cv_destroy(&fs->lfs_stopcv); 972 rw_destroy(&fs->lfs_fraglock); 973 rw_destroy(&fs->lfs_iflock); 974 free(fs, M_UFSMNT); 975 free(ump, M_UFSMNT); 976 977 mp->mnt_data = NULL; 978 mp->mnt_flag &= ~MNT_LOCAL; 979 return (error); 980 } 981 982 /* 983 * Get file system statistics. 984 * 985 * NB: We don't lock to access the superblock here, because it's not 986 * really that important if we get it wrong. 987 */ 988 int 989 lfs_statvfs(struct mount *mp, struct statvfs *sbp) 990 { 991 struct lfs *fs; 992 struct ufsmount *ump; 993 994 ump = VFSTOUFS(mp); 995 fs = ump->um_lfs; 996 if (fs->lfs_magic != LFS_MAGIC) 997 panic("lfs_statvfs: magic"); 998 999 sbp->f_bsize = fs->lfs_bsize; 1000 sbp->f_frsize = fs->lfs_fsize; 1001 sbp->f_iosize = fs->lfs_bsize; 1002 sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks); 1003 1004 sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs)); 1005 KASSERT(sbp->f_bfree <= fs->lfs_dsize); 1006 #if 0 1007 if (sbp->f_bfree < 0) 1008 sbp->f_bfree = 0; 1009 #endif 1010 1011 sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs)); 1012 if (sbp->f_bfree > sbp->f_bresvd) 1013 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1014 else 1015 sbp->f_bavail = 0; 1016 1017 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs); 1018 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles; 1019 sbp->f_favail = sbp->f_ffree; 1020 sbp->f_fresvd = 0; 1021 copy_statvfs_info(sbp, mp); 1022 return (0); 1023 } 1024 1025 /* 1026 * Go through the disk queues to initiate sandbagged IO; 1027 * go through the inodes to write those that have been modified; 1028 * initiate the writing of the super block if it has been modified. 1029 * 1030 * Note: we are always called with the filesystem marked `MPBUSY'. 1031 */ 1032 int 1033 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 1034 { 1035 int error; 1036 struct lfs *fs; 1037 1038 fs = VFSTOUFS(mp)->um_lfs; 1039 if (fs->lfs_ronly) 1040 return 0; 1041 1042 /* Snapshots should not hose the syncer */ 1043 /* 1044 * XXX Sync can block here anyway, since we don't have a very 1045 * XXX good idea of how much data is pending. If it's more 1046 * XXX than a segment and lfs_nextseg is close to the end of 1047 * XXX the log, we'll likely block. 1048 */ 1049 mutex_enter(&lfs_lock); 1050 if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) { 1051 mutex_exit(&lfs_lock); 1052 return 0; 1053 } 1054 mutex_exit(&lfs_lock); 1055 1056 lfs_writer_enter(fs, "lfs_dirops"); 1057 1058 /* All syncs must be checkpoints until roll-forward is implemented. */ 1059 DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset)); 1060 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1061 lfs_writer_leave(fs); 1062 #ifdef QUOTA 1063 qsync(mp); 1064 #endif 1065 return (error); 1066 } 1067 1068 extern kmutex_t ufs_hashlock; 1069 1070 /* 1071 * Look up an LFS dinode number to find its incore vnode. If not already 1072 * in core, read it in from the specified device. Return the inode locked. 1073 * Detection and handling of mount points must be done by the calling routine. 1074 */ 1075 int 1076 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1077 { 1078 struct lfs *fs; 1079 struct ufs1_dinode *dip; 1080 struct inode *ip; 1081 struct buf *bp; 1082 struct ifile *ifp; 1083 struct vnode *vp; 1084 struct ufsmount *ump; 1085 daddr_t daddr; 1086 dev_t dev; 1087 int error, retries; 1088 struct timespec ts; 1089 1090 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1091 1092 ump = VFSTOUFS(mp); 1093 dev = ump->um_dev; 1094 fs = ump->um_lfs; 1095 1096 /* 1097 * If the filesystem is not completely mounted yet, suspend 1098 * any access requests (wait for roll-forward to complete). 1099 */ 1100 mutex_enter(&lfs_lock); 1101 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1102 mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1103 &lfs_lock); 1104 mutex_exit(&lfs_lock); 1105 1106 retry: 1107 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) 1108 return (0); 1109 1110 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) { 1111 *vpp = NULL; 1112 return (error); 1113 } 1114 1115 mutex_enter(&ufs_hashlock); 1116 if (ufs_ihashget(dev, ino, 0) != NULL) { 1117 mutex_exit(&ufs_hashlock); 1118 ungetnewvnode(vp); 1119 goto retry; 1120 } 1121 1122 /* Translate the inode number to a disk address. */ 1123 if (ino == LFS_IFILE_INUM) 1124 daddr = fs->lfs_idaddr; 1125 else { 1126 /* XXX bounds-check this too */ 1127 LFS_IENTRY(ifp, fs, ino, bp); 1128 daddr = ifp->if_daddr; 1129 if (fs->lfs_version > 1) { 1130 ts.tv_sec = ifp->if_atime_sec; 1131 ts.tv_nsec = ifp->if_atime_nsec; 1132 } 1133 1134 brelse(bp, 0); 1135 if (daddr == LFS_UNUSED_DADDR) { 1136 *vpp = NULLVP; 1137 mutex_exit(&ufs_hashlock); 1138 ungetnewvnode(vp); 1139 return (ENOENT); 1140 } 1141 } 1142 1143 /* Allocate/init new vnode/inode. */ 1144 lfs_vcreate(mp, ino, vp); 1145 1146 /* 1147 * Put it onto its hash chain and lock it so that other requests for 1148 * this inode will block if they arrive while we are sleeping waiting 1149 * for old data structures to be purged or for the contents of the 1150 * disk portion of this inode to be read. 1151 */ 1152 ip = VTOI(vp); 1153 ufs_ihashins(ip); 1154 mutex_exit(&ufs_hashlock); 1155 1156 /* 1157 * XXX 1158 * This may not need to be here, logically it should go down with 1159 * the i_devvp initialization. 1160 * Ask Kirk. 1161 */ 1162 ip->i_lfs = ump->um_lfs; 1163 1164 /* Read in the disk contents for the inode, copy into the inode. */ 1165 retries = 0; 1166 again: 1167 error = bread(ump->um_devvp, fsbtodb(fs, daddr), 1168 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize), 1169 NOCRED, &bp); 1170 if (error) { 1171 /* 1172 * The inode does not contain anything useful, so it would 1173 * be misleading to leave it on its hash chain. With mode 1174 * still zero, it will be unlinked and returned to the free 1175 * list by vput(). 1176 */ 1177 vput(vp); 1178 brelse(bp, 0); 1179 *vpp = NULL; 1180 return (error); 1181 } 1182 1183 dip = lfs_ifind(fs, ino, bp); 1184 if (dip == NULL) { 1185 /* Assume write has not completed yet; try again */ 1186 brelse(bp, BC_INVAL); 1187 ++retries; 1188 if (retries > LFS_IFIND_RETRIES) { 1189 #ifdef DEBUG 1190 /* If the seglock is held look at the bpp to see 1191 what is there anyway */ 1192 mutex_enter(&lfs_lock); 1193 if (fs->lfs_seglock > 0) { 1194 struct buf **bpp; 1195 struct ufs1_dinode *dp; 1196 int i; 1197 1198 for (bpp = fs->lfs_sp->bpp; 1199 bpp != fs->lfs_sp->cbpp; ++bpp) { 1200 if ((*bpp)->b_vp == fs->lfs_ivnode && 1201 bpp != fs->lfs_sp->bpp) { 1202 /* Inode block */ 1203 printf("lfs_vget: block 0x%" PRIx64 ": ", 1204 (*bpp)->b_blkno); 1205 dp = (struct ufs1_dinode *)(*bpp)->b_data; 1206 for (i = 0; i < INOPB(fs); i++) 1207 if (dp[i].di_u.inumber) 1208 printf("%d ", dp[i].di_u.inumber); 1209 printf("\n"); 1210 } 1211 } 1212 } 1213 mutex_exit(&lfs_lock); 1214 #endif /* DEBUG */ 1215 panic("lfs_vget: dinode not found"); 1216 } 1217 mutex_enter(&lfs_lock); 1218 if (fs->lfs_iocount) { 1219 DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino)); 1220 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, 1221 "lfs ifind", 1, &lfs_lock); 1222 } else 1223 retries = LFS_IFIND_RETRIES; 1224 mutex_exit(&lfs_lock); 1225 goto again; 1226 } 1227 *ip->i_din.ffs1_din = *dip; 1228 brelse(bp, 0); 1229 1230 if (fs->lfs_version > 1) { 1231 ip->i_ffs1_atime = ts.tv_sec; 1232 ip->i_ffs1_atimensec = ts.tv_nsec; 1233 } 1234 1235 lfs_vinit(mp, &vp); 1236 1237 *vpp = vp; 1238 1239 KASSERT(VOP_ISLOCKED(vp)); 1240 1241 return (0); 1242 } 1243 1244 /* 1245 * File handle to vnode 1246 */ 1247 int 1248 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1249 { 1250 struct lfid lfh; 1251 struct buf *bp; 1252 IFILE *ifp; 1253 int32_t daddr; 1254 struct lfs *fs; 1255 vnode_t *vp; 1256 1257 if (fhp->fid_len != sizeof(struct lfid)) 1258 return EINVAL; 1259 1260 memcpy(&lfh, fhp, sizeof(lfh)); 1261 if (lfh.lfid_ino < LFS_IFILE_INUM) 1262 return ESTALE; 1263 1264 fs = VFSTOUFS(mp)->um_lfs; 1265 if (lfh.lfid_ident != fs->lfs_ident) 1266 return ESTALE; 1267 1268 if (lfh.lfid_ino > 1269 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) - 1270 fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb) 1271 return ESTALE; 1272 1273 mutex_enter(&ufs_ihash_lock); 1274 vp = ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino); 1275 mutex_exit(&ufs_ihash_lock); 1276 if (vp == NULL) { 1277 LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp); 1278 daddr = ifp->if_daddr; 1279 brelse(bp, 0); 1280 if (daddr == LFS_UNUSED_DADDR) 1281 return ESTALE; 1282 } 1283 1284 return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp)); 1285 } 1286 1287 /* 1288 * Vnode pointer to File handle 1289 */ 1290 /* ARGSUSED */ 1291 int 1292 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 1293 { 1294 struct inode *ip; 1295 struct lfid lfh; 1296 1297 if (*fh_size < sizeof(struct lfid)) { 1298 *fh_size = sizeof(struct lfid); 1299 return E2BIG; 1300 } 1301 *fh_size = sizeof(struct lfid); 1302 ip = VTOI(vp); 1303 memset(&lfh, 0, sizeof(lfh)); 1304 lfh.lfid_len = sizeof(struct lfid); 1305 lfh.lfid_ino = ip->i_number; 1306 lfh.lfid_gen = ip->i_gen; 1307 lfh.lfid_ident = ip->i_lfs->lfs_ident; 1308 memcpy(fhp, &lfh, sizeof(lfh)); 1309 return (0); 1310 } 1311 1312 static int 1313 sysctl_lfs_dostats(SYSCTLFN_ARGS) 1314 { 1315 extern struct lfs_stats lfs_stats; 1316 extern int lfs_dostats; 1317 int error; 1318 1319 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 1320 if (error || newp == NULL) 1321 return (error); 1322 1323 if (lfs_dostats == 0) 1324 memset(&lfs_stats, 0, sizeof(lfs_stats)); 1325 1326 return (0); 1327 } 1328 1329 struct shortlong { 1330 const char *sname; 1331 const char *lname; 1332 }; 1333 1334 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup") 1335 { 1336 int i; 1337 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 1338 lfs_fs_pagetrip, lfs_ignore_lazy_sync; 1339 #ifdef DEBUG 1340 extern int lfs_debug_log_subsys[DLOG_MAX]; 1341 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 1342 { "rollforward", "Debug roll-forward code" }, 1343 { "alloc", "Debug inode allocation and free list" }, 1344 { "avail", "Debug space-available-now accounting" }, 1345 { "flush", "Debug flush triggers" }, 1346 { "lockedlist", "Debug locked list accounting" }, 1347 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 1348 { "vnode", "Debug vnode use during segment write" }, 1349 { "segment", "Debug segment writing" }, 1350 { "seguse", "Debug segment used-bytes accounting" }, 1351 { "cleaner", "Debug cleaning routines" }, 1352 { "mount", "Debug mount/unmount routines" }, 1353 { "pagecache", "Debug UBC interactions" }, 1354 { "dirop", "Debug directory-operation accounting" }, 1355 { "malloc", "Debug private malloc accounting" }, 1356 }; 1357 #endif /* DEBUG */ 1358 struct shortlong stat_names[] = { /* Must match lfs.h! */ 1359 { "segsused", "Number of new segments allocated" }, 1360 { "psegwrites", "Number of partial-segment writes" }, 1361 { "psyncwrites", "Number of synchronous partial-segment" 1362 " writes" }, 1363 { "pcleanwrites", "Number of partial-segment writes by the" 1364 " cleaner" }, 1365 { "blocktot", "Number of blocks written" }, 1366 { "cleanblocks", "Number of blocks written by the cleaner" }, 1367 { "ncheckpoints", "Number of checkpoints made" }, 1368 { "nwrites", "Number of whole writes" }, 1369 { "nsync_writes", "Number of synchronous writes" }, 1370 { "wait_exceeded", "Number of times writer waited for" 1371 " cleaner" }, 1372 { "write_exceeded", "Number of times writer invoked flush" }, 1373 { "flush_invoked", "Number of times flush was invoked" }, 1374 { "vflush_invoked", "Number of time vflush was called" }, 1375 { "clean_inlocked", "Number of vnodes skipped for VI_XLOCK" }, 1376 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 1377 { "segs_reclaimed", "Number of segments reclaimed" }, 1378 }; 1379 1380 sysctl_createv(clog, 0, NULL, NULL, 1381 CTLFLAG_PERMANENT, 1382 CTLTYPE_NODE, "vfs", NULL, 1383 NULL, 0, NULL, 0, 1384 CTL_VFS, CTL_EOL); 1385 sysctl_createv(clog, 0, NULL, NULL, 1386 CTLFLAG_PERMANENT, 1387 CTLTYPE_NODE, "lfs", 1388 SYSCTL_DESCR("Log-structured file system"), 1389 NULL, 0, NULL, 0, 1390 CTL_VFS, 5, CTL_EOL); 1391 /* 1392 * XXX the "5" above could be dynamic, thereby eliminating one 1393 * more instance of the "number to vfs" mapping problem, but 1394 * "5" is the order as taken from sys/mount.h 1395 */ 1396 1397 sysctl_createv(clog, 0, NULL, NULL, 1398 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1399 CTLTYPE_INT, "flushindir", NULL, 1400 NULL, 0, &lfs_writeindir, 0, 1401 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 1402 sysctl_createv(clog, 0, NULL, NULL, 1403 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1404 CTLTYPE_INT, "clean_vnhead", NULL, 1405 NULL, 0, &lfs_clean_vnhead, 0, 1406 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 1407 sysctl_createv(clog, 0, NULL, NULL, 1408 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1409 CTLTYPE_INT, "dostats", 1410 SYSCTL_DESCR("Maintain statistics on LFS operations"), 1411 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 1412 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 1413 sysctl_createv(clog, 0, NULL, NULL, 1414 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1415 CTLTYPE_INT, "pagetrip", 1416 SYSCTL_DESCR("How many dirty pages in fs triggers" 1417 " a flush"), 1418 NULL, 0, &lfs_fs_pagetrip, 0, 1419 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 1420 sysctl_createv(clog, 0, NULL, NULL, 1421 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1422 CTLTYPE_INT, "ignore_lazy_sync", 1423 SYSCTL_DESCR("Lazy Sync is ignored entirely"), 1424 NULL, 0, &lfs_ignore_lazy_sync, 0, 1425 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL); 1426 #ifdef LFS_KERNEL_RFW 1427 sysctl_createv(clog, 0, NULL, NULL, 1428 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1429 CTLTYPE_INT, "rfw", 1430 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 1431 NULL, 0, &lfs_do_rfw, 0, 1432 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 1433 #endif 1434 1435 sysctl_createv(clog, 0, NULL, NULL, 1436 CTLFLAG_PERMANENT, 1437 CTLTYPE_NODE, "stats", 1438 SYSCTL_DESCR("Debugging options"), 1439 NULL, 0, NULL, 0, 1440 CTL_VFS, 5, LFS_STATS, CTL_EOL); 1441 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 1442 sysctl_createv(clog, 0, NULL, NULL, 1443 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1444 CTLTYPE_INT, stat_names[i].sname, 1445 SYSCTL_DESCR(stat_names[i].lname), 1446 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 1447 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 1448 } 1449 1450 #ifdef DEBUG 1451 sysctl_createv(clog, 0, NULL, NULL, 1452 CTLFLAG_PERMANENT, 1453 CTLTYPE_NODE, "debug", 1454 SYSCTL_DESCR("Debugging options"), 1455 NULL, 0, NULL, 0, 1456 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 1457 for (i = 0; i < DLOG_MAX; i++) { 1458 sysctl_createv(clog, 0, NULL, NULL, 1459 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1460 CTLTYPE_INT, dlog_names[i].sname, 1461 SYSCTL_DESCR(dlog_names[i].lname), 1462 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 1463 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 1464 } 1465 #endif 1466 } 1467 1468 /* 1469 * ufs_bmaparray callback function for writing. 1470 * 1471 * Since blocks will be written to the new segment anyway, 1472 * we don't care about current daddr of them. 1473 */ 1474 static bool 1475 lfs_issequential_hole(const struct ufsmount *ump, 1476 daddr_t daddr0, daddr_t daddr1) 1477 { 1478 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */ 1479 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */ 1480 1481 KASSERT(daddr0 == UNWRITTEN || 1482 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR)); 1483 KASSERT(daddr1 == UNWRITTEN || 1484 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR)); 1485 1486 /* NOTE: all we want to know here is 'hole or not'. */ 1487 /* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */ 1488 1489 /* 1490 * treat UNWRITTENs and all resident blocks as 'contiguous' 1491 */ 1492 if (daddr0 != 0 && daddr1 != 0) 1493 return true; 1494 1495 /* 1496 * both are in hole? 1497 */ 1498 if (daddr0 == 0 && daddr1 == 0) 1499 return true; /* all holes are 'contiguous' for us. */ 1500 1501 return false; 1502 } 1503 1504 /* 1505 * lfs_gop_write functions exactly like genfs_gop_write, except that 1506 * (1) it requires the seglock to be held by its caller, and sp->fip 1507 * to be properly initialized (it will return without re-initializing 1508 * sp->fip, and without calling lfs_writeseg). 1509 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1510 * to determine how large a block it can write at once (though it does 1511 * still use VOP_BMAP to find holes in the file); 1512 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1513 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1514 * now have clusters of clusters, ick.) 1515 */ 1516 static int 1517 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, 1518 int flags) 1519 { 1520 int i, error, run, haveeof = 0; 1521 int fs_bshift; 1522 vaddr_t kva; 1523 off_t eof, offset, startoffset = 0; 1524 size_t bytes, iobytes, skipbytes; 1525 daddr_t lbn, blkno; 1526 struct vm_page *pg; 1527 struct buf *mbp, *bp; 1528 struct vnode *devvp = VTOI(vp)->i_devvp; 1529 struct inode *ip = VTOI(vp); 1530 struct lfs *fs = ip->i_lfs; 1531 struct segment *sp = fs->lfs_sp; 1532 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1533 1534 ASSERT_SEGLOCK(fs); 1535 1536 /* The Ifile lives in the buffer cache */ 1537 KASSERT(vp != fs->lfs_ivnode); 1538 1539 /* 1540 * We don't want to fill the disk before the cleaner has a chance 1541 * to make room for us. If we're in danger of doing that, fail 1542 * with EAGAIN. The caller will have to notice this, unlock 1543 * so the cleaner can run, relock and try again. 1544 * 1545 * We must write everything, however, if our vnode is being 1546 * reclaimed. 1547 */ 1548 if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp) 1549 goto tryagain; 1550 1551 /* 1552 * Sometimes things slip past the filters in lfs_putpages, 1553 * and the pagedaemon tries to write pages---problem is 1554 * that the pagedaemon never acquires the segment lock. 1555 * 1556 * Alternatively, pages that were clean when we called 1557 * genfs_putpages may have become dirty in the meantime. In this 1558 * case the segment header is not properly set up for blocks 1559 * to be added to it. 1560 * 1561 * Unbusy and unclean the pages, and put them on the ACTIVE 1562 * queue under the hypothesis that they couldn't have got here 1563 * unless they were modified *quite* recently. 1564 * 1565 * XXXUBC that last statement is an oversimplification of course. 1566 */ 1567 if (!LFS_SEGLOCK_HELD(fs) || 1568 (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) || 1569 (pgs[0]->offset & fs->lfs_bmask) != 0) { 1570 goto tryagain; 1571 } 1572 1573 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x", 1574 vp, pgs, npages, flags); 1575 1576 GOP_SIZE(vp, vp->v_size, &eof, 0); 1577 haveeof = 1; 1578 1579 if (vp->v_type == VREG) 1580 fs_bshift = vp->v_mount->mnt_fs_bshift; 1581 else 1582 fs_bshift = DEV_BSHIFT; 1583 error = 0; 1584 pg = pgs[0]; 1585 startoffset = pg->offset; 1586 KASSERT(eof >= 0); 1587 1588 if (startoffset >= eof) { 1589 goto tryagain; 1590 } else 1591 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 1592 skipbytes = 0; 1593 1594 KASSERT(bytes != 0); 1595 1596 /* Swap PG_DELWRI for PG_PAGEOUT */ 1597 for (i = 0; i < npages; i++) { 1598 if (pgs[i]->flags & PG_DELWRI) { 1599 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 1600 pgs[i]->flags &= ~PG_DELWRI; 1601 pgs[i]->flags |= PG_PAGEOUT; 1602 uvm_pageout_start(1); 1603 mutex_enter(&uvm_pageqlock); 1604 uvm_pageunwire(pgs[i]); 1605 mutex_exit(&uvm_pageqlock); 1606 } 1607 } 1608 1609 /* 1610 * Check to make sure we're starting on a block boundary. 1611 * We'll check later to make sure we always write entire 1612 * blocks (or fragments). 1613 */ 1614 if (startoffset & fs->lfs_bmask) 1615 printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n", 1616 startoffset, fs->lfs_bmask, 1617 startoffset & fs->lfs_bmask); 1618 KASSERT((startoffset & fs->lfs_bmask) == 0); 1619 if (bytes & fs->lfs_ffmask) { 1620 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 1621 panic("lfs_gop_write: non-integer blocks"); 1622 } 1623 1624 /* 1625 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 1626 * If we would, write what we have and try again. If we don't 1627 * have anything to write, we'll have to sleep. 1628 */ 1629 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1630 (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ? 1631 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 1632 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 1633 #if 0 1634 " with nfinfo=%d at offset 0x%x\n", 1635 (int)((SEGSUM *)(sp->segsum))->ss_nfinfo, 1636 (unsigned)fs->lfs_offset)); 1637 #endif 1638 lfs_updatemeta(sp); 1639 lfs_release_finfo(fs); 1640 (void) lfs_writeseg(fs, sp); 1641 1642 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 1643 1644 /* 1645 * Having given up all of the pager_map we were holding, 1646 * we can now wait for aiodoned to reclaim it for us 1647 * without fear of deadlock. 1648 */ 1649 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1650 UVMPAGER_MAPIN_WAITOK); 1651 } 1652 1653 mutex_enter(&vp->v_interlock); 1654 vp->v_numoutput += 2; /* one for biodone, one for aiodone */ 1655 mutex_exit(&vp->v_interlock); 1656 1657 mbp = getiobuf(NULL, true); 1658 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 1659 vp, mbp, vp->v_numoutput, bytes); 1660 mbp->b_bufsize = npages << PAGE_SHIFT; 1661 mbp->b_data = (void *)kva; 1662 mbp->b_resid = mbp->b_bcount = bytes; 1663 mbp->b_cflags = BC_BUSY|BC_AGE; 1664 mbp->b_iodone = uvm_aio_biodone; 1665 1666 bp = NULL; 1667 for (offset = startoffset; 1668 bytes > 0; 1669 offset += iobytes, bytes -= iobytes) { 1670 lbn = offset >> fs_bshift; 1671 error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 1672 lfs_issequential_hole); 1673 if (error) { 1674 UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d", 1675 error,0,0,0); 1676 skipbytes += bytes; 1677 bytes = 0; 1678 break; 1679 } 1680 1681 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 1682 bytes); 1683 if (blkno == (daddr_t)-1) { 1684 skipbytes += iobytes; 1685 continue; 1686 } 1687 1688 /* 1689 * Discover how much we can really pack into this buffer. 1690 */ 1691 /* If no room in the current segment, finish it up */ 1692 if (sp->sum_bytes_left < sizeof(int32_t) || 1693 sp->seg_bytes_left < (1 << fs->lfs_bshift)) { 1694 int vers; 1695 1696 lfs_updatemeta(sp); 1697 vers = sp->fip->fi_version; 1698 lfs_release_finfo(fs); 1699 (void) lfs_writeseg(fs, sp); 1700 1701 lfs_acquire_finfo(fs, ip->i_number, vers); 1702 } 1703 /* Check both for space in segment and space in segsum */ 1704 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 1705 << fs_bshift); 1706 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 1707 << fs_bshift); 1708 KASSERT(iobytes > 0); 1709 1710 /* if it's really one i/o, don't make a second buf */ 1711 if (offset == startoffset && iobytes == bytes) { 1712 bp = mbp; 1713 /* correct overcount if there is no second buffer */ 1714 mutex_enter(&vp->v_interlock); 1715 --vp->v_numoutput; 1716 mutex_exit(&vp->v_interlock); 1717 } else { 1718 bp = getiobuf(NULL, true); 1719 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 1720 vp, bp, vp->v_numoutput, 0); 1721 bp->b_data = (char *)kva + 1722 (vaddr_t)(offset - pg->offset); 1723 bp->b_resid = bp->b_bcount = iobytes; 1724 bp->b_cflags = BC_BUSY; 1725 bp->b_iodone = uvm_aio_biodone1; 1726 } 1727 1728 /* XXX This is silly ... is this necessary? */ 1729 mutex_enter(&bufcache_lock); 1730 mutex_enter(&vp->v_interlock); 1731 bgetvp(vp, bp); 1732 mutex_exit(&vp->v_interlock); 1733 mutex_exit(&bufcache_lock); 1734 1735 bp->b_lblkno = lblkno(fs, offset); 1736 bp->b_private = mbp; 1737 if (devvp->v_type == VBLK) { 1738 bp->b_dev = devvp->v_rdev; 1739 } 1740 VOP_BWRITE(bp); 1741 while (lfs_gatherblock(sp, bp, NULL)) 1742 continue; 1743 } 1744 1745 if (skipbytes) { 1746 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0); 1747 mutex_enter(mbp->b_objlock); 1748 if (error) { 1749 mbp->b_error = error; 1750 } 1751 mbp->b_resid -= skipbytes; 1752 mutex_exit(mbp->b_objlock); 1753 if (mbp->b_resid == 0) { 1754 biodone(mbp); 1755 } 1756 } 1757 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 1758 return (0); 1759 1760 tryagain: 1761 /* 1762 * We can't write the pages, for whatever reason. 1763 * Clean up after ourselves, and make the caller try again. 1764 */ 1765 mutex_enter(&vp->v_interlock); 1766 1767 /* Tell why we're here, if we know */ 1768 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) { 1769 DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n")); 1770 } else if ((pgs[0]->offset & fs->lfs_bmask) != 0) { 1771 DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n")); 1772 } else if (haveeof && startoffset >= eof) { 1773 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 1774 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 1775 pgs[0]->offset, eof, npages)); 1776 } else if (LFS_STARVED_FOR_SEGS(fs)) { 1777 DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n")); 1778 } else { 1779 DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n")); 1780 } 1781 1782 mutex_enter(&uvm_pageqlock); 1783 for (i = 0; i < npages; i++) { 1784 pg = pgs[i]; 1785 1786 if (pg->flags & PG_PAGEOUT) 1787 uvm_pageout_done(1); 1788 if (pg->flags & PG_DELWRI) { 1789 uvm_pageunwire(pg); 1790 } 1791 uvm_pageactivate(pg); 1792 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 1793 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 1794 vp, pg->offset)); 1795 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 1796 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 1797 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 1798 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 1799 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 1800 pg->wire_count)); 1801 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 1802 pg->loan_count)); 1803 } 1804 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */ 1805 uvm_page_unbusy(pgs, npages); 1806 mutex_exit(&uvm_pageqlock); 1807 mutex_exit(&vp->v_interlock); 1808 return EAGAIN; 1809 } 1810 1811 /* 1812 * finish vnode/inode initialization. 1813 * used by lfs_vget and lfs_fastvget. 1814 */ 1815 void 1816 lfs_vinit(struct mount *mp, struct vnode **vpp) 1817 { 1818 struct vnode *vp = *vpp; 1819 struct inode *ip = VTOI(vp); 1820 struct ufsmount *ump = VFSTOUFS(mp); 1821 struct lfs *fs = ump->um_lfs; 1822 int i; 1823 1824 ip->i_mode = ip->i_ffs1_mode; 1825 ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink; 1826 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size; 1827 ip->i_flags = ip->i_ffs1_flags; 1828 ip->i_gen = ip->i_ffs1_gen; 1829 ip->i_uid = ip->i_ffs1_uid; 1830 ip->i_gid = ip->i_ffs1_gid; 1831 1832 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 1833 ip->i_lfs_odnlink = ip->i_ffs1_nlink; 1834 1835 /* 1836 * Initialize the vnode from the inode, check for aliases. In all 1837 * cases re-init ip, the underlying vnode/inode may have changed. 1838 */ 1839 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 1840 ip = VTOI(vp); 1841 1842 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 1843 if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) { 1844 #ifdef DEBUG 1845 for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift; 1846 i < NDADDR; i++) { 1847 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 1848 i == 0) 1849 continue; 1850 if (ip->i_ffs1_db[i] != 0) { 1851 inconsistent: 1852 lfs_dump_dinode(ip->i_din.ffs1_din); 1853 panic("inconsistent inode"); 1854 } 1855 } 1856 for ( ; i < NDADDR + NIADDR; i++) { 1857 if (ip->i_ffs1_ib[i - NDADDR] != 0) { 1858 goto inconsistent; 1859 } 1860 } 1861 #endif /* DEBUG */ 1862 for (i = 0; i < NDADDR; i++) 1863 if (ip->i_ffs1_db[i] != 0) 1864 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 1865 } 1866 1867 #ifdef DIAGNOSTIC 1868 if (vp->v_type == VNON) { 1869 # ifdef DEBUG 1870 lfs_dump_dinode(ip->i_din.ffs1_din); 1871 # endif 1872 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 1873 (unsigned long long)ip->i_number, 1874 (ip->i_mode & IFMT) >> 12); 1875 } 1876 #endif /* DIAGNOSTIC */ 1877 1878 /* 1879 * Finish inode initialization now that aliasing has been resolved. 1880 */ 1881 1882 ip->i_devvp = ump->um_devvp; 1883 VREF(ip->i_devvp); 1884 genfs_node_init(vp, &lfs_genfsops); 1885 uvm_vnp_setsize(vp, ip->i_size); 1886 1887 /* Initialize hiblk from file size */ 1888 ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1; 1889 1890 *vpp = vp; 1891 } 1892 1893 /* 1894 * Resize the filesystem to contain the specified number of segments. 1895 */ 1896 int 1897 lfs_resize_fs(struct lfs *fs, int newnsegs) 1898 { 1899 SEGUSE *sup; 1900 struct buf *bp, *obp; 1901 daddr_t olast, nlast, ilast, noff, start, end; 1902 struct vnode *ivp; 1903 struct inode *ip; 1904 int error, badnews, inc, oldnsegs; 1905 int sbbytes, csbbytes, gain, cgain; 1906 int i; 1907 1908 /* Only support v2 and up */ 1909 if (fs->lfs_version < 2) 1910 return EOPNOTSUPP; 1911 1912 /* If we're doing nothing, do it fast */ 1913 oldnsegs = fs->lfs_nseg; 1914 if (newnsegs == oldnsegs) 1915 return 0; 1916 1917 /* We always have to have two superblocks */ 1918 if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1])) 1919 return EFBIG; 1920 1921 ivp = fs->lfs_ivnode; 1922 ip = VTOI(ivp); 1923 error = 0; 1924 1925 /* Take the segment lock so no one else calls lfs_newseg() */ 1926 lfs_seglock(fs, SEGM_PROT); 1927 1928 /* 1929 * Make sure the segments we're going to be losing, if any, 1930 * are in fact empty. We hold the seglock, so their status 1931 * cannot change underneath us. Count the superblocks we lose, 1932 * while we're at it. 1933 */ 1934 sbbytes = csbbytes = 0; 1935 cgain = 0; 1936 for (i = newnsegs; i < oldnsegs; i++) { 1937 LFS_SEGENTRY(sup, fs, i, bp); 1938 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 1939 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1940 sbbytes += LFS_SBPAD; 1941 if (!(sup->su_flags & SEGUSE_DIRTY)) { 1942 ++cgain; 1943 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1944 csbbytes += LFS_SBPAD; 1945 } 1946 brelse(bp, 0); 1947 if (badnews) { 1948 error = EBUSY; 1949 goto out; 1950 } 1951 } 1952 1953 /* Note old and new segment table endpoints, and old ifile size */ 1954 olast = fs->lfs_cleansz + fs->lfs_segtabsz; 1955 nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz; 1956 ilast = ivp->v_size >> fs->lfs_bshift; 1957 noff = nlast - olast; 1958 1959 /* 1960 * Make sure no one can use the Ifile while we change it around. 1961 * Even after taking the iflock we need to make sure no one still 1962 * is holding Ifile buffers, so we get each one, to drain them. 1963 * (XXX this could be done better.) 1964 */ 1965 rw_enter(&fs->lfs_iflock, RW_WRITER); 1966 vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY); 1967 for (i = 0; i < ilast; i++) { 1968 bread(ivp, i, fs->lfs_bsize, NOCRED, &bp); 1969 brelse(bp, 0); 1970 } 1971 1972 /* Allocate new Ifile blocks */ 1973 for (i = ilast; i < ilast + noff; i++) { 1974 if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0, 1975 &bp) != 0) 1976 panic("balloc extending ifile"); 1977 memset(bp->b_data, 0, fs->lfs_bsize); 1978 VOP_BWRITE(bp); 1979 } 1980 1981 /* Register new ifile size */ 1982 ip->i_size += noff * fs->lfs_bsize; 1983 ip->i_ffs1_size = ip->i_size; 1984 uvm_vnp_setsize(ivp, ip->i_size); 1985 1986 /* Copy the inode table to its new position */ 1987 if (noff != 0) { 1988 if (noff < 0) { 1989 start = nlast; 1990 end = ilast + noff; 1991 inc = 1; 1992 } else { 1993 start = ilast + noff - 1; 1994 end = nlast - 1; 1995 inc = -1; 1996 } 1997 for (i = start; i != end; i += inc) { 1998 if (bread(ivp, i, fs->lfs_bsize, NOCRED, &bp) != 0) 1999 panic("resize: bread dst blk failed"); 2000 if (bread(ivp, i - noff, fs->lfs_bsize, NOCRED, &obp)) 2001 panic("resize: bread src blk failed"); 2002 memcpy(bp->b_data, obp->b_data, fs->lfs_bsize); 2003 VOP_BWRITE(bp); 2004 brelse(obp, 0); 2005 } 2006 } 2007 2008 /* If we are expanding, write the new empty SEGUSE entries */ 2009 if (newnsegs > oldnsegs) { 2010 for (i = oldnsegs; i < newnsegs; i++) { 2011 if ((error = bread(ivp, i / fs->lfs_sepb + 2012 fs->lfs_cleansz, 2013 fs->lfs_bsize, NOCRED, &bp)) != 0) 2014 panic("lfs: ifile read: %d", error); 2015 while ((i + 1) % fs->lfs_sepb && i < newnsegs) { 2016 sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb]; 2017 memset(sup, 0, sizeof(*sup)); 2018 i++; 2019 } 2020 VOP_BWRITE(bp); 2021 } 2022 } 2023 2024 /* Zero out unused superblock offsets */ 2025 for (i = 2; i < LFS_MAXNUMSB; i++) 2026 if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs) 2027 fs->lfs_sboffs[i] = 0x0; 2028 2029 /* 2030 * Correct superblock entries that depend on fs size. 2031 * The computations of these are as follows: 2032 * 2033 * size = segtod(fs, nseg) 2034 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD) 2035 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2036 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD) 2037 * + (segtod(fs, 1) - (offset - curseg)) 2038 * - segtod(fs, minfreeseg - (minfreeseg / 2)) 2039 * 2040 * XXX - we should probably adjust minfreeseg as well. 2041 */ 2042 gain = (newnsegs - oldnsegs); 2043 fs->lfs_nseg = newnsegs; 2044 fs->lfs_segtabsz = nlast - fs->lfs_cleansz; 2045 fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize); 2046 fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes); 2047 fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes) 2048 - gain * btofsb(fs, fs->lfs_bsize / 2); 2049 if (gain > 0) { 2050 fs->lfs_nclean += gain; 2051 fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize); 2052 } else { 2053 fs->lfs_nclean -= cgain; 2054 fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) - 2055 btofsb(fs, csbbytes); 2056 } 2057 2058 /* Resize segment flag cache */ 2059 fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2060 fs->lfs_nseg * sizeof(u_int32_t), 2061 M_SEGMENT, M_WAITOK); 2062 fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[1], 2063 fs->lfs_nseg * sizeof(u_int32_t), 2064 M_SEGMENT, M_WAITOK); 2065 for (i = oldnsegs; i < newnsegs; i++) 2066 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2067 2068 /* Truncate Ifile if necessary */ 2069 if (noff < 0) 2070 lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0, 2071 NOCRED); 2072 2073 /* Update cleaner info so the cleaner can die */ 2074 bread(ivp, 0, fs->lfs_bsize, NOCRED, &bp); 2075 ((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean; 2076 ((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean; 2077 VOP_BWRITE(bp); 2078 2079 /* Let Ifile accesses proceed */ 2080 VOP_UNLOCK(ivp, 0); 2081 rw_exit(&fs->lfs_iflock); 2082 2083 out: 2084 lfs_segunlock(fs); 2085 return error; 2086 } 2087
Indexes created Mon Sep 29 21:09:56 GMT 2025