lfs_vfsops.c revision 1.288
1 /* $NetBSD: lfs_vfsops.c,v 1.288 2011/03/06 17:08:39 bouyer 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 1.288 2011/03/06 17:08:39 bouyer 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 #include <sys/module.h> 95 #include <sys/syscallvar.h> 96 #include <sys/syscall.h> 97 #include <sys/syscallargs.h> 98 99 #include <miscfs/specfs/specdev.h> 100 101 #include <ufs/ufs/quota.h> 102 #include <ufs/ufs/inode.h> 103 #include <ufs/ufs/ufsmount.h> 104 #include <ufs/ufs/ufs_extern.h> 105 106 #include <uvm/uvm.h> 107 #include <uvm/uvm_stat.h> 108 #include <uvm/uvm_pager.h> 109 #include <uvm/uvm_pdaemon.h> 110 111 #include <ufs/lfs/lfs.h> 112 #include <ufs/lfs/lfs_extern.h> 113 114 #include <miscfs/genfs/genfs.h> 115 #include <miscfs/genfs/genfs_node.h> 116 117 MODULE(MODULE_CLASS_VFS, lfs, "ffs"); 118 119 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int); 120 static bool lfs_issequential_hole(const struct ufsmount *, 121 daddr_t, daddr_t); 122 123 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *); 124 125 static struct sysctllog *lfs_sysctl_log; 126 127 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc; 128 extern const struct vnodeopv_desc lfs_specop_opv_desc; 129 extern const struct vnodeopv_desc lfs_fifoop_opv_desc; 130 131 pid_t lfs_writer_daemon = 0; 132 int lfs_do_flush = 0; 133 #ifdef LFS_KERNEL_RFW 134 int lfs_do_rfw = 0; 135 #endif 136 137 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = { 138 &lfs_vnodeop_opv_desc, 139 &lfs_specop_opv_desc, 140 &lfs_fifoop_opv_desc, 141 NULL, 142 }; 143 144 struct vfsops lfs_vfsops = { 145 MOUNT_LFS, 146 sizeof (struct ufs_args), 147 lfs_mount, 148 ufs_start, 149 lfs_unmount, 150 ufs_root, 151 ufs_quotactl, 152 lfs_statvfs, 153 lfs_sync, 154 lfs_vget, 155 lfs_fhtovp, 156 lfs_vptofh, 157 lfs_init, 158 lfs_reinit, 159 lfs_done, 160 lfs_mountroot, 161 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 162 vfs_stdextattrctl, 163 (void *)eopnotsupp, /* vfs_suspendctl */ 164 genfs_renamelock_enter, 165 genfs_renamelock_exit, 166 (void *)eopnotsupp, 167 lfs_vnodeopv_descs, 168 0, 169 { NULL, NULL }, 170 }; 171 172 const struct genfs_ops lfs_genfsops = { 173 .gop_size = lfs_gop_size, 174 .gop_alloc = ufs_gop_alloc, 175 .gop_write = lfs_gop_write, 176 .gop_markupdate = ufs_gop_markupdate, 177 }; 178 179 static const struct ufs_ops lfs_ufsops = { 180 .uo_itimes = NULL, 181 .uo_update = lfs_update, 182 .uo_truncate = lfs_truncate, 183 .uo_valloc = lfs_valloc, 184 .uo_vfree = lfs_vfree, 185 .uo_balloc = lfs_balloc, 186 .uo_unmark_vnode = lfs_unmark_vnode, 187 }; 188 189 struct shortlong { 190 const char *sname; 191 const char *lname; 192 }; 193 194 static int 195 sysctl_lfs_dostats(SYSCTLFN_ARGS) 196 { 197 extern struct lfs_stats lfs_stats; 198 extern int lfs_dostats; 199 int error; 200 201 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 202 if (error || newp == NULL) 203 return (error); 204 205 if (lfs_dostats == 0) 206 memset(&lfs_stats, 0, sizeof(lfs_stats)); 207 208 return (0); 209 } 210 211 static void 212 lfs_sysctl_setup(struct sysctllog **clog) 213 { 214 int i; 215 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 216 lfs_fs_pagetrip, lfs_ignore_lazy_sync; 217 #ifdef DEBUG 218 extern int lfs_debug_log_subsys[DLOG_MAX]; 219 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 220 { "rollforward", "Debug roll-forward code" }, 221 { "alloc", "Debug inode allocation and free list" }, 222 { "avail", "Debug space-available-now accounting" }, 223 { "flush", "Debug flush triggers" }, 224 { "lockedlist", "Debug locked list accounting" }, 225 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 226 { "vnode", "Debug vnode use during segment write" }, 227 { "segment", "Debug segment writing" }, 228 { "seguse", "Debug segment used-bytes accounting" }, 229 { "cleaner", "Debug cleaning routines" }, 230 { "mount", "Debug mount/unmount routines" }, 231 { "pagecache", "Debug UBC interactions" }, 232 { "dirop", "Debug directory-operation accounting" }, 233 { "malloc", "Debug private malloc accounting" }, 234 }; 235 #endif /* DEBUG */ 236 struct shortlong stat_names[] = { /* Must match lfs.h! */ 237 { "segsused", "Number of new segments allocated" }, 238 { "psegwrites", "Number of partial-segment writes" }, 239 { "psyncwrites", "Number of synchronous partial-segment" 240 " writes" }, 241 { "pcleanwrites", "Number of partial-segment writes by the" 242 " cleaner" }, 243 { "blocktot", "Number of blocks written" }, 244 { "cleanblocks", "Number of blocks written by the cleaner" }, 245 { "ncheckpoints", "Number of checkpoints made" }, 246 { "nwrites", "Number of whole writes" }, 247 { "nsync_writes", "Number of synchronous writes" }, 248 { "wait_exceeded", "Number of times writer waited for" 249 " cleaner" }, 250 { "write_exceeded", "Number of times writer invoked flush" }, 251 { "flush_invoked", "Number of times flush was invoked" }, 252 { "vflush_invoked", "Number of time vflush was called" }, 253 { "clean_inlocked", "Number of vnodes skipped for VI_XLOCK" }, 254 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 255 { "segs_reclaimed", "Number of segments reclaimed" }, 256 }; 257 258 sysctl_createv(clog, 0, NULL, NULL, 259 CTLFLAG_PERMANENT, 260 CTLTYPE_NODE, "vfs", NULL, 261 NULL, 0, NULL, 0, 262 CTL_VFS, CTL_EOL); 263 sysctl_createv(clog, 0, NULL, NULL, 264 CTLFLAG_PERMANENT, 265 CTLTYPE_NODE, "lfs", 266 SYSCTL_DESCR("Log-structured file system"), 267 NULL, 0, NULL, 0, 268 CTL_VFS, 5, CTL_EOL); 269 /* 270 * XXX the "5" above could be dynamic, thereby eliminating one 271 * more instance of the "number to vfs" mapping problem, but 272 * "5" is the order as taken from sys/mount.h 273 */ 274 275 sysctl_createv(clog, 0, NULL, NULL, 276 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 277 CTLTYPE_INT, "flushindir", NULL, 278 NULL, 0, &lfs_writeindir, 0, 279 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 280 sysctl_createv(clog, 0, NULL, NULL, 281 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 282 CTLTYPE_INT, "clean_vnhead", NULL, 283 NULL, 0, &lfs_clean_vnhead, 0, 284 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 285 sysctl_createv(clog, 0, NULL, NULL, 286 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 287 CTLTYPE_INT, "dostats", 288 SYSCTL_DESCR("Maintain statistics on LFS operations"), 289 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 290 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 291 sysctl_createv(clog, 0, NULL, NULL, 292 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 293 CTLTYPE_INT, "pagetrip", 294 SYSCTL_DESCR("How many dirty pages in fs triggers" 295 " a flush"), 296 NULL, 0, &lfs_fs_pagetrip, 0, 297 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 298 sysctl_createv(clog, 0, NULL, NULL, 299 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 300 CTLTYPE_INT, "ignore_lazy_sync", 301 SYSCTL_DESCR("Lazy Sync is ignored entirely"), 302 NULL, 0, &lfs_ignore_lazy_sync, 0, 303 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL); 304 #ifdef LFS_KERNEL_RFW 305 sysctl_createv(clog, 0, NULL, NULL, 306 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 307 CTLTYPE_INT, "rfw", 308 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 309 NULL, 0, &lfs_do_rfw, 0, 310 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 311 #endif 312 313 sysctl_createv(clog, 0, NULL, NULL, 314 CTLFLAG_PERMANENT, 315 CTLTYPE_NODE, "stats", 316 SYSCTL_DESCR("Debugging options"), 317 NULL, 0, NULL, 0, 318 CTL_VFS, 5, LFS_STATS, CTL_EOL); 319 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 320 sysctl_createv(clog, 0, NULL, NULL, 321 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 322 CTLTYPE_INT, stat_names[i].sname, 323 SYSCTL_DESCR(stat_names[i].lname), 324 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 325 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 326 } 327 328 #ifdef DEBUG 329 sysctl_createv(clog, 0, NULL, NULL, 330 CTLFLAG_PERMANENT, 331 CTLTYPE_NODE, "debug", 332 SYSCTL_DESCR("Debugging options"), 333 NULL, 0, NULL, 0, 334 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 335 for (i = 0; i < DLOG_MAX; i++) { 336 sysctl_createv(clog, 0, NULL, NULL, 337 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 338 CTLTYPE_INT, dlog_names[i].sname, 339 SYSCTL_DESCR(dlog_names[i].lname), 340 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 341 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 342 } 343 #endif 344 } 345 346 /* old cleaner syscall interface. see VOP_FCNTL() */ 347 static const struct syscall_package lfs_syscalls[] = { 348 { SYS_lfs_bmapv, 0, (sy_call_t *)sys_lfs_bmapv }, 349 { SYS_lfs_markv, 0, (sy_call_t *)sys_lfs_markv }, 350 { SYS_lfs_segclean, 0, (sy_call_t *)sys___lfs_segwait50 }, 351 { 0, 0, NULL }, 352 }; 353 354 static int 355 lfs_modcmd(modcmd_t cmd, void *arg) 356 { 357 int error; 358 359 switch (cmd) { 360 case MODULE_CMD_INIT: 361 error = syscall_establish(NULL, lfs_syscalls); 362 if (error) 363 return error; 364 error = vfs_attach(&lfs_vfsops); 365 if (error != 0) { 366 syscall_disestablish(NULL, lfs_syscalls); 367 break; 368 } 369 lfs_sysctl_setup(&lfs_sysctl_log); 370 break; 371 case MODULE_CMD_FINI: 372 error = vfs_detach(&lfs_vfsops); 373 if (error != 0) 374 break; 375 syscall_disestablish(NULL, lfs_syscalls); 376 sysctl_teardown(&lfs_sysctl_log); 377 break; 378 default: 379 error = ENOTTY; 380 break; 381 } 382 383 return (error); 384 } 385 386 /* 387 * XXX Same structure as FFS inodes? Should we share a common pool? 388 */ 389 struct pool lfs_inode_pool; 390 struct pool lfs_dinode_pool; 391 struct pool lfs_inoext_pool; 392 struct pool lfs_lbnentry_pool; 393 394 /* 395 * The writer daemon. UVM keeps track of how many dirty pages we are holding 396 * in lfs_subsys_pages; the daemon flushes the filesystem when this value 397 * crosses the (user-defined) threshhold LFS_MAX_PAGES. 398 */ 399 static void 400 lfs_writerd(void *arg) 401 { 402 struct mount *mp, *nmp; 403 struct lfs *fs; 404 int fsflags; 405 int loopcount; 406 407 lfs_writer_daemon = curproc->p_pid; 408 409 mutex_enter(&lfs_lock); 410 for (;;) { 411 mtsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10, 412 &lfs_lock); 413 414 /* 415 * Look through the list of LFSs to see if any of them 416 * have requested pageouts. 417 */ 418 mutex_enter(&mountlist_lock); 419 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 420 mp = nmp) { 421 if (vfs_busy(mp, &nmp)) { 422 continue; 423 } 424 if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS, 425 sizeof(mp->mnt_stat.f_fstypename)) == 0) { 426 fs = VFSTOUFS(mp)->um_lfs; 427 mutex_enter(&lfs_lock); 428 fsflags = 0; 429 if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || 430 lfs_dirvcount > LFS_MAX_DIROP) && 431 fs->lfs_dirops == 0) 432 fsflags |= SEGM_CKP; 433 if (fs->lfs_pdflush) { 434 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n")); 435 fs->lfs_pdflush = 0; 436 lfs_flush_fs(fs, fsflags); 437 mutex_exit(&lfs_lock); 438 } else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) { 439 DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n")); 440 mutex_exit(&lfs_lock); 441 lfs_writer_enter(fs, "wrdirop"); 442 lfs_flush_pchain(fs); 443 lfs_writer_leave(fs); 444 } else 445 mutex_exit(&lfs_lock); 446 } 447 vfs_unbusy(mp, false, &nmp); 448 } 449 mutex_exit(&mountlist_lock); 450 451 /* 452 * If global state wants a flush, flush everything. 453 */ 454 mutex_enter(&lfs_lock); 455 loopcount = 0; 456 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS || 457 locked_queue_bytes > LFS_MAX_BYTES || 458 lfs_subsys_pages > LFS_MAX_PAGES) { 459 460 if (lfs_do_flush) { 461 DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n")); 462 } 463 if (locked_queue_count > LFS_MAX_BUFS) { 464 DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n", 465 locked_queue_count, LFS_MAX_BUFS)); 466 } 467 if (locked_queue_bytes > LFS_MAX_BYTES) { 468 DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n", 469 locked_queue_bytes, LFS_MAX_BYTES)); 470 } 471 if (lfs_subsys_pages > LFS_MAX_PAGES) { 472 DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n", 473 lfs_subsys_pages, LFS_MAX_PAGES)); 474 } 475 476 lfs_flush(NULL, SEGM_WRITERD, 0); 477 lfs_do_flush = 0; 478 } 479 } 480 /* NOTREACHED */ 481 } 482 483 /* 484 * Initialize the filesystem, most work done by ufs_init. 485 */ 486 void 487 lfs_init(void) 488 { 489 490 malloc_type_attach(M_SEGMENT); 491 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 492 "lfsinopl", &pool_allocator_nointr, IPL_NONE); 493 pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, 494 "lfsdinopl", &pool_allocator_nointr, IPL_NONE); 495 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 496 "lfsinoextpl", &pool_allocator_nointr, IPL_NONE); 497 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 498 "lfslbnpool", &pool_allocator_nointr, IPL_NONE); 499 ufs_init(); 500 501 #ifdef DEBUG 502 memset(lfs_log, 0, sizeof(lfs_log)); 503 #endif 504 mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE); 505 cv_init(&locked_queue_cv, "lfsbuf"); 506 cv_init(&lfs_writing_cv, "lfsflush"); 507 } 508 509 void 510 lfs_reinit(void) 511 { 512 ufs_reinit(); 513 } 514 515 void 516 lfs_done(void) 517 { 518 ufs_done(); 519 mutex_destroy(&lfs_lock); 520 cv_destroy(&locked_queue_cv); 521 cv_destroy(&lfs_writing_cv); 522 pool_destroy(&lfs_inode_pool); 523 pool_destroy(&lfs_dinode_pool); 524 pool_destroy(&lfs_inoext_pool); 525 pool_destroy(&lfs_lbnentry_pool); 526 malloc_type_detach(M_SEGMENT); 527 } 528 529 /* 530 * Called by main() when ufs is going to be mounted as root. 531 */ 532 int 533 lfs_mountroot(void) 534 { 535 extern struct vnode *rootvp; 536 struct lfs *fs = NULL; /* LFS */ 537 struct mount *mp; 538 struct lwp *l = curlwp; 539 struct ufsmount *ump; 540 int error; 541 542 if (device_class(root_device) != DV_DISK) 543 return (ENODEV); 544 545 if (rootdev == NODEV) 546 return (ENODEV); 547 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 548 vrele(rootvp); 549 return (error); 550 } 551 if ((error = lfs_mountfs(rootvp, mp, l))) { 552 vfs_unbusy(mp, false, NULL); 553 vfs_destroy(mp); 554 return (error); 555 } 556 mutex_enter(&mountlist_lock); 557 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 558 mutex_exit(&mountlist_lock); 559 ump = VFSTOUFS(mp); 560 fs = ump->um_lfs; 561 memset(fs->lfs_fsmnt, 0, sizeof(fs->lfs_fsmnt)); 562 (void)copystr(mp->mnt_stat.f_mntonname, fs->lfs_fsmnt, MNAMELEN - 1, 0); 563 (void)lfs_statvfs(mp, &mp->mnt_stat); 564 vfs_unbusy(mp, false, NULL); 565 setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp)); 566 return (0); 567 } 568 569 /* 570 * VFS Operations. 571 * 572 * mount system call 573 */ 574 int 575 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 576 { 577 struct lwp *l = curlwp; 578 struct vnode *devvp; 579 struct ufs_args *args = data; 580 struct ufsmount *ump = NULL; 581 struct lfs *fs = NULL; /* LFS */ 582 int error = 0, update; 583 mode_t accessmode; 584 585 if (*data_len < sizeof *args) 586 return EINVAL; 587 588 if (mp->mnt_flag & MNT_GETARGS) { 589 ump = VFSTOUFS(mp); 590 if (ump == NULL) 591 return EIO; 592 args->fspec = NULL; 593 *data_len = sizeof *args; 594 return 0; 595 } 596 597 update = mp->mnt_flag & MNT_UPDATE; 598 599 /* Check arguments */ 600 if (args->fspec != NULL) { 601 /* 602 * Look up the name and verify that it's sane. 603 */ 604 error = namei_simple_user(args->fspec, 605 NSM_FOLLOW_NOEMULROOT, &devvp); 606 if (error != 0) 607 return (error); 608 609 if (!update) { 610 /* 611 * Be sure this is a valid block device 612 */ 613 if (devvp->v_type != VBLK) 614 error = ENOTBLK; 615 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 616 error = ENXIO; 617 } else { 618 /* 619 * Be sure we're still naming the same device 620 * used for our initial mount 621 */ 622 ump = VFSTOUFS(mp); 623 if (devvp != ump->um_devvp) { 624 if (devvp->v_rdev != ump->um_devvp->v_rdev) 625 error = EINVAL; 626 else { 627 vrele(devvp); 628 devvp = ump->um_devvp; 629 vref(devvp); 630 } 631 } 632 } 633 } else { 634 if (!update) { 635 /* New mounts must have a filename for the device */ 636 return (EINVAL); 637 } else { 638 /* Use the extant mount */ 639 ump = VFSTOUFS(mp); 640 devvp = ump->um_devvp; 641 vref(devvp); 642 } 643 } 644 645 646 /* 647 * If mount by non-root, then verify that user has necessary 648 * permissions on the device. 649 */ 650 if (error == 0) { 651 accessmode = VREAD; 652 if (update ? 653 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 654 (mp->mnt_flag & MNT_RDONLY) == 0) 655 accessmode |= VWRITE; 656 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 657 error = genfs_can_mount(devvp, accessmode, l->l_cred); 658 VOP_UNLOCK(devvp); 659 } 660 661 if (error) { 662 vrele(devvp); 663 return (error); 664 } 665 666 if (!update) { 667 int flags; 668 669 if (mp->mnt_flag & MNT_RDONLY) 670 flags = FREAD; 671 else 672 flags = FREAD|FWRITE; 673 error = VOP_OPEN(devvp, flags, FSCRED); 674 if (error) 675 goto fail; 676 error = lfs_mountfs(devvp, mp, l); /* LFS */ 677 if (error) { 678 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 679 (void)VOP_CLOSE(devvp, flags, NOCRED); 680 VOP_UNLOCK(devvp); 681 goto fail; 682 } 683 684 ump = VFSTOUFS(mp); 685 fs = ump->um_lfs; 686 } else { 687 /* 688 * Update the mount. 689 */ 690 691 /* 692 * The initial mount got a reference on this 693 * device, so drop the one obtained via 694 * namei(), above. 695 */ 696 vrele(devvp); 697 698 ump = VFSTOUFS(mp); 699 fs = ump->um_lfs; 700 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 701 /* 702 * Changing from read-only to read/write. 703 * Note in the superblocks that we're writing. 704 */ 705 fs->lfs_ronly = 0; 706 if (fs->lfs_pflags & LFS_PF_CLEAN) { 707 fs->lfs_pflags &= ~LFS_PF_CLEAN; 708 lfs_writesuper(fs, fs->lfs_sboffs[0]); 709 lfs_writesuper(fs, fs->lfs_sboffs[1]); 710 } 711 } 712 if (args->fspec == NULL) 713 return EINVAL; 714 } 715 716 error = set_statvfs_info(path, UIO_USERSPACE, args->fspec, 717 UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l); 718 if (error == 0) 719 (void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname, 720 sizeof(fs->lfs_fsmnt)); 721 return error; 722 723 fail: 724 vrele(devvp); 725 return (error); 726 } 727 728 729 /* 730 * Common code for mount and mountroot 731 * LFS specific 732 */ 733 int 734 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 735 { 736 struct dlfs *tdfs, *dfs, *adfs; 737 struct lfs *fs; 738 struct ufsmount *ump; 739 struct vnode *vp; 740 struct buf *bp, *abp; 741 dev_t dev; 742 int error, i, ronly, fsbsize; 743 kauth_cred_t cred; 744 CLEANERINFO *cip; 745 SEGUSE *sup; 746 daddr_t sb_addr; 747 748 cred = l ? l->l_cred : NOCRED; 749 750 /* 751 * Flush out any old buffers remaining from a previous use. 752 */ 753 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 754 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 755 VOP_UNLOCK(devvp); 756 if (error) 757 return (error); 758 759 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 760 761 /* Don't free random space on error. */ 762 bp = NULL; 763 abp = NULL; 764 ump = NULL; 765 766 sb_addr = LFS_LABELPAD / DEV_BSIZE; 767 while (1) { 768 /* Read in the superblock. */ 769 error = bread(devvp, sb_addr, LFS_SBPAD, cred, 0, &bp); 770 if (error) 771 goto out; 772 dfs = (struct dlfs *)bp->b_data; 773 774 /* Check the basics. */ 775 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE || 776 dfs->dlfs_version > LFS_VERSION || 777 dfs->dlfs_bsize < sizeof(struct dlfs)) { 778 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 779 error = EINVAL; /* XXX needs translation */ 780 goto out; 781 } 782 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) { 783 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 784 dfs->dlfs_inodefmt)); 785 error = EINVAL; 786 goto out; 787 } 788 789 if (dfs->dlfs_version == 1) 790 fsbsize = DEV_BSIZE; 791 else { 792 fsbsize = 1 << dfs->dlfs_ffshift; 793 /* 794 * Could be, if the frag size is large enough, that we 795 * don't have the "real" primary superblock. If that's 796 * the case, get the real one, and try again. 797 */ 798 if (sb_addr != (dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT))) { 799 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 800 " 0x%llx is not right, trying 0x%llx\n", 801 (long long)sb_addr, 802 (long long)(dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT)))); 803 sb_addr = dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT); 804 brelse(bp, 0); 805 continue; 806 } 807 } 808 break; 809 } 810 811 /* 812 * Check the second superblock to see which is newer; then mount 813 * using the older of the two. This is necessary to ensure that 814 * the filesystem is valid if it was not unmounted cleanly. 815 */ 816 817 if (dfs->dlfs_sboffs[1] && 818 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 819 { 820 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / DEV_BSIZE), 821 LFS_SBPAD, cred, 0, &abp); 822 if (error) 823 goto out; 824 adfs = (struct dlfs *)abp->b_data; 825 826 if (dfs->dlfs_version == 1) { 827 /* 1s resolution comparison */ 828 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 829 tdfs = adfs; 830 else 831 tdfs = dfs; 832 } else { 833 /* monotonic infinite-resolution comparison */ 834 if (adfs->dlfs_serial < dfs->dlfs_serial) 835 tdfs = adfs; 836 else 837 tdfs = dfs; 838 } 839 840 /* Check the basics. */ 841 if (tdfs->dlfs_magic != LFS_MAGIC || 842 tdfs->dlfs_bsize > MAXBSIZE || 843 tdfs->dlfs_version > LFS_VERSION || 844 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 845 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 846 " sanity failed\n")); 847 error = EINVAL; /* XXX needs translation */ 848 goto out; 849 } 850 } else { 851 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock" 852 " daddr=0x%x\n", dfs->dlfs_sboffs[1])); 853 error = EINVAL; 854 goto out; 855 } 856 857 /* Allocate the mount structure, copy the superblock into it. */ 858 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO); 859 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 860 861 /* Compatibility */ 862 if (fs->lfs_version < 2) { 863 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 864 fs->lfs_ibsize = fs->lfs_bsize; 865 fs->lfs_start = fs->lfs_sboffs[0]; 866 fs->lfs_tstamp = fs->lfs_otstamp; 867 fs->lfs_fsbtodb = 0; 868 } 869 if (fs->lfs_resvseg == 0) 870 fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \ 871 MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1)); 872 873 /* 874 * If we aren't going to be able to write meaningfully to this 875 * filesystem, and were not mounted readonly, bomb out now. 876 */ 877 if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 878 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 879 " we need BUFPAGES >= %lld\n", 880 (long long)((bufmem_hiwater / bufmem_lowater) * 881 LFS_INVERSE_MAX_BYTES( 882 fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 883 free(fs, M_UFSMNT); 884 error = EFBIG; /* XXX needs translation */ 885 goto out; 886 } 887 888 /* Before rolling forward, lock so vget will sleep for other procs */ 889 if (l != NULL) { 890 fs->lfs_flags = LFS_NOTYET; 891 fs->lfs_rfpid = l->l_proc->p_pid; 892 } 893 894 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 895 ump->um_lfs = fs; 896 ump->um_ops = &lfs_ufsops; 897 ump->um_fstype = UFS1; 898 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 899 brelse(bp, BC_INVAL); 900 brelse(abp, BC_INVAL); 901 } else { 902 brelse(bp, 0); 903 brelse(abp, 0); 904 } 905 bp = NULL; 906 abp = NULL; 907 908 909 /* Set up the I/O information */ 910 fs->lfs_devbsize = DEV_BSIZE; 911 fs->lfs_iocount = 0; 912 fs->lfs_diropwait = 0; 913 fs->lfs_activesb = 0; 914 fs->lfs_uinodes = 0; 915 fs->lfs_ravail = 0; 916 fs->lfs_favail = 0; 917 fs->lfs_sbactive = 0; 918 919 /* Set up the ifile and lock aflags */ 920 fs->lfs_doifile = 0; 921 fs->lfs_writer = 0; 922 fs->lfs_dirops = 0; 923 fs->lfs_nadirop = 0; 924 fs->lfs_seglock = 0; 925 fs->lfs_pdflush = 0; 926 fs->lfs_sleepers = 0; 927 fs->lfs_pages = 0; 928 rw_init(&fs->lfs_fraglock); 929 rw_init(&fs->lfs_iflock); 930 cv_init(&fs->lfs_stopcv, "lfsstop"); 931 932 /* Set the file system readonly/modify bits. */ 933 fs->lfs_ronly = ronly; 934 if (ronly == 0) 935 fs->lfs_fmod = 1; 936 937 /* Initialize the mount structure. */ 938 dev = devvp->v_rdev; 939 mp->mnt_data = ump; 940 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 941 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 942 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 943 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 944 mp->mnt_stat.f_iosize = fs->lfs_bsize; 945 mp->mnt_flag |= MNT_LOCAL; 946 mp->mnt_fs_bshift = fs->lfs_bshift; 947 ump->um_flags = 0; 948 ump->um_mountp = mp; 949 ump->um_dev = dev; 950 ump->um_devvp = devvp; 951 ump->um_bptrtodb = fs->lfs_ffshift - DEV_BSHIFT; 952 ump->um_seqinc = fs->lfs_frag; 953 ump->um_nindir = fs->lfs_nindir; 954 ump->um_lognindir = ffs(fs->lfs_nindir) - 1; 955 for (i = 0; i < MAXQUOTAS; i++) 956 ump->um_quotas[i] = NULLVP; 957 ump->um_maxsymlinklen = fs->lfs_maxsymlinklen; 958 ump->um_dirblksiz = DIRBLKSIZ; 959 ump->um_maxfilesize = fs->lfs_maxfilesize; 960 if (ump->um_maxsymlinklen > 0) 961 mp->mnt_iflag |= IMNT_DTYPE; 962 devvp->v_specmountpoint = mp; 963 964 /* Set up reserved memory for pageout */ 965 lfs_setup_resblks(fs); 966 /* Set up vdirop tailq */ 967 TAILQ_INIT(&fs->lfs_dchainhd); 968 /* and paging tailq */ 969 TAILQ_INIT(&fs->lfs_pchainhd); 970 /* and delayed segment accounting for truncation list */ 971 LIST_INIT(&fs->lfs_segdhd); 972 973 /* 974 * We use the ifile vnode for almost every operation. Instead of 975 * retrieving it from the hash table each time we retrieve it here, 976 * artificially increment the reference count and keep a pointer 977 * to it in the incore copy of the superblock. 978 */ 979 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) { 980 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 981 goto out; 982 } 983 fs->lfs_ivnode = vp; 984 vref(vp); 985 986 /* Set up inode bitmap and order free list */ 987 lfs_order_freelist(fs); 988 989 /* Set up segment usage flags for the autocleaner. */ 990 fs->lfs_nactive = 0; 991 fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *), 992 M_SEGMENT, M_WAITOK); 993 fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 994 M_SEGMENT, M_WAITOK); 995 fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 996 M_SEGMENT, M_WAITOK); 997 memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t)); 998 for (i = 0; i < fs->lfs_nseg; i++) { 999 int changed; 1000 1001 LFS_SEGENTRY(sup, fs, i, bp); 1002 changed = 0; 1003 if (!ronly) { 1004 if (sup->su_nbytes == 0 && 1005 !(sup->su_flags & SEGUSE_EMPTY)) { 1006 sup->su_flags |= SEGUSE_EMPTY; 1007 ++changed; 1008 } else if (!(sup->su_nbytes == 0) && 1009 (sup->su_flags & SEGUSE_EMPTY)) { 1010 sup->su_flags &= ~SEGUSE_EMPTY; 1011 ++changed; 1012 } 1013 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 1014 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 1015 ++changed; 1016 } 1017 } 1018 fs->lfs_suflags[0][i] = sup->su_flags; 1019 if (changed) 1020 LFS_WRITESEGENTRY(sup, fs, i, bp); 1021 else 1022 brelse(bp, 0); 1023 } 1024 1025 #ifdef LFS_KERNEL_RFW 1026 lfs_roll_forward(fs, mp, l); 1027 #endif 1028 1029 /* If writing, sb is not clean; record in case of immediate crash */ 1030 if (!fs->lfs_ronly) { 1031 fs->lfs_pflags &= ~LFS_PF_CLEAN; 1032 lfs_writesuper(fs, fs->lfs_sboffs[0]); 1033 lfs_writesuper(fs, fs->lfs_sboffs[1]); 1034 } 1035 1036 /* Allow vget now that roll-forward is complete */ 1037 fs->lfs_flags &= ~(LFS_NOTYET); 1038 wakeup(&fs->lfs_flags); 1039 1040 /* 1041 * Initialize the ifile cleaner info with information from 1042 * the superblock. 1043 */ 1044 LFS_CLEANERINFO(cip, fs, bp); 1045 cip->clean = fs->lfs_nclean; 1046 cip->dirty = fs->lfs_nseg - fs->lfs_nclean; 1047 cip->avail = fs->lfs_avail; 1048 cip->bfree = fs->lfs_bfree; 1049 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1050 1051 /* 1052 * Mark the current segment as ACTIVE, since we're going to 1053 * be writing to it. 1054 */ 1055 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); 1056 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1057 fs->lfs_nactive++; 1058 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); /* Ifile */ 1059 1060 /* Now that roll-forward is done, unlock the Ifile */ 1061 vput(vp); 1062 1063 /* Start the pagedaemon-anticipating daemon */ 1064 if (lfs_writer_daemon == 0 && kthread_create(PRI_BIO, 0, NULL, 1065 lfs_writerd, NULL, NULL, "lfs_writer") != 0) 1066 panic("fork lfs_writer"); 1067 /* 1068 * XXX: Get extra reference to LFS vfsops. This prevents unload, 1069 * but also prevents kernel panic due to text being unloaded 1070 * from below lfs_writerd. When lfs_writerd can exit, remove 1071 * this!!! 1072 */ 1073 vfs_getopsbyname(MOUNT_LFS); 1074 1075 printf("WARNING: the log-structured file system is experimental\n" 1076 "WARNING: it may cause system crashes and/or corrupt data\n"); 1077 1078 return (0); 1079 1080 out: 1081 if (bp) 1082 brelse(bp, 0); 1083 if (abp) 1084 brelse(abp, 0); 1085 if (ump) { 1086 free(ump->um_lfs, M_UFSMNT); 1087 free(ump, M_UFSMNT); 1088 mp->mnt_data = NULL; 1089 } 1090 1091 return (error); 1092 } 1093 1094 /* 1095 * unmount system call 1096 */ 1097 int 1098 lfs_unmount(struct mount *mp, int mntflags) 1099 { 1100 struct lwp *l = curlwp; 1101 struct ufsmount *ump; 1102 struct lfs *fs; 1103 int error, flags, ronly; 1104 vnode_t *vp; 1105 1106 flags = 0; 1107 if (mntflags & MNT_FORCE) 1108 flags |= FORCECLOSE; 1109 1110 ump = VFSTOUFS(mp); 1111 fs = ump->um_lfs; 1112 1113 /* Two checkpoints */ 1114 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1115 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1116 1117 /* wake up the cleaner so it can die */ 1118 lfs_wakeup_cleaner(fs); 1119 mutex_enter(&lfs_lock); 1120 while (fs->lfs_sleepers) 1121 mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0, 1122 &lfs_lock); 1123 mutex_exit(&lfs_lock); 1124 1125 #ifdef QUOTA 1126 if ((error = quota1_umount(mp, flags)) != 0) 1127 return (error); 1128 #endif 1129 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 1130 return (error); 1131 if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0) 1132 return (error); 1133 vp = fs->lfs_ivnode; 1134 mutex_enter(&vp->v_interlock); 1135 if (LIST_FIRST(&vp->v_dirtyblkhd)) 1136 panic("lfs_unmount: still dirty blocks on ifile vnode"); 1137 mutex_exit(&vp->v_interlock); 1138 1139 /* Explicitly write the superblock, to update serial and pflags */ 1140 fs->lfs_pflags |= LFS_PF_CLEAN; 1141 lfs_writesuper(fs, fs->lfs_sboffs[0]); 1142 lfs_writesuper(fs, fs->lfs_sboffs[1]); 1143 mutex_enter(&lfs_lock); 1144 while (fs->lfs_iocount) 1145 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 1146 &lfs_lock); 1147 mutex_exit(&lfs_lock); 1148 1149 /* Finish with the Ifile, now that we're done with it */ 1150 vgone(fs->lfs_ivnode); 1151 1152 ronly = !fs->lfs_ronly; 1153 if (ump->um_devvp->v_type != VBAD) 1154 ump->um_devvp->v_specmountpoint = NULL; 1155 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1156 error = VOP_CLOSE(ump->um_devvp, 1157 ronly ? FREAD : FREAD|FWRITE, NOCRED); 1158 vput(ump->um_devvp); 1159 1160 /* Complain about page leakage */ 1161 if (fs->lfs_pages > 0) 1162 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 1163 fs->lfs_pages, lfs_subsys_pages); 1164 1165 /* Free per-mount data structures */ 1166 free(fs->lfs_ino_bitmap, M_SEGMENT); 1167 free(fs->lfs_suflags[0], M_SEGMENT); 1168 free(fs->lfs_suflags[1], M_SEGMENT); 1169 free(fs->lfs_suflags, M_SEGMENT); 1170 lfs_free_resblks(fs); 1171 cv_destroy(&fs->lfs_stopcv); 1172 rw_destroy(&fs->lfs_fraglock); 1173 rw_destroy(&fs->lfs_iflock); 1174 free(fs, M_UFSMNT); 1175 free(ump, M_UFSMNT); 1176 1177 mp->mnt_data = NULL; 1178 mp->mnt_flag &= ~MNT_LOCAL; 1179 return (error); 1180 } 1181 1182 /* 1183 * Get file system statistics. 1184 * 1185 * NB: We don't lock to access the superblock here, because it's not 1186 * really that important if we get it wrong. 1187 */ 1188 int 1189 lfs_statvfs(struct mount *mp, struct statvfs *sbp) 1190 { 1191 struct lfs *fs; 1192 struct ufsmount *ump; 1193 1194 ump = VFSTOUFS(mp); 1195 fs = ump->um_lfs; 1196 if (fs->lfs_magic != LFS_MAGIC) 1197 panic("lfs_statvfs: magic"); 1198 1199 sbp->f_bsize = fs->lfs_bsize; 1200 sbp->f_frsize = fs->lfs_fsize; 1201 sbp->f_iosize = fs->lfs_bsize; 1202 sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks; 1203 1204 sbp->f_bfree = LFS_EST_BFREE(fs); 1205 KASSERT(sbp->f_bfree <= fs->lfs_dsize); 1206 #if 0 1207 if (sbp->f_bfree < 0) 1208 sbp->f_bfree = 0; 1209 #endif 1210 1211 sbp->f_bresvd = LFS_EST_RSVD(fs); 1212 if (sbp->f_bfree > sbp->f_bresvd) 1213 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1214 else 1215 sbp->f_bavail = 0; 1216 1217 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs); 1218 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles; 1219 sbp->f_favail = sbp->f_ffree; 1220 sbp->f_fresvd = 0; 1221 copy_statvfs_info(sbp, mp); 1222 return (0); 1223 } 1224 1225 /* 1226 * Go through the disk queues to initiate sandbagged IO; 1227 * go through the inodes to write those that have been modified; 1228 * initiate the writing of the super block if it has been modified. 1229 * 1230 * Note: we are always called with the filesystem marked `MPBUSY'. 1231 */ 1232 int 1233 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 1234 { 1235 int error; 1236 struct lfs *fs; 1237 1238 fs = VFSTOUFS(mp)->um_lfs; 1239 if (fs->lfs_ronly) 1240 return 0; 1241 1242 /* Snapshots should not hose the syncer */ 1243 /* 1244 * XXX Sync can block here anyway, since we don't have a very 1245 * XXX good idea of how much data is pending. If it's more 1246 * XXX than a segment and lfs_nextseg is close to the end of 1247 * XXX the log, we'll likely block. 1248 */ 1249 mutex_enter(&lfs_lock); 1250 if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) { 1251 mutex_exit(&lfs_lock); 1252 return 0; 1253 } 1254 mutex_exit(&lfs_lock); 1255 1256 lfs_writer_enter(fs, "lfs_dirops"); 1257 1258 /* All syncs must be checkpoints until roll-forward is implemented. */ 1259 DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset)); 1260 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1261 lfs_writer_leave(fs); 1262 #ifdef QUOTA 1263 qsync(mp); 1264 #endif 1265 return (error); 1266 } 1267 1268 /* 1269 * Look up an LFS dinode number to find its incore vnode. If not already 1270 * in core, read it in from the specified device. Return the inode locked. 1271 * Detection and handling of mount points must be done by the calling routine. 1272 */ 1273 int 1274 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1275 { 1276 struct lfs *fs; 1277 struct ufs1_dinode *dip; 1278 struct inode *ip; 1279 struct buf *bp; 1280 struct ifile *ifp; 1281 struct vnode *vp; 1282 struct ufsmount *ump; 1283 daddr_t daddr; 1284 dev_t dev; 1285 int error, retries; 1286 struct timespec ts; 1287 1288 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1289 1290 ump = VFSTOUFS(mp); 1291 dev = ump->um_dev; 1292 fs = ump->um_lfs; 1293 1294 /* 1295 * If the filesystem is not completely mounted yet, suspend 1296 * any access requests (wait for roll-forward to complete). 1297 */ 1298 mutex_enter(&lfs_lock); 1299 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1300 mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1301 &lfs_lock); 1302 mutex_exit(&lfs_lock); 1303 1304 retry: 1305 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) 1306 return (0); 1307 1308 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) { 1309 *vpp = NULL; 1310 return (error); 1311 } 1312 1313 mutex_enter(&ufs_hashlock); 1314 if (ufs_ihashget(dev, ino, 0) != NULL) { 1315 mutex_exit(&ufs_hashlock); 1316 ungetnewvnode(vp); 1317 goto retry; 1318 } 1319 1320 /* Translate the inode number to a disk address. */ 1321 if (ino == LFS_IFILE_INUM) 1322 daddr = fs->lfs_idaddr; 1323 else { 1324 /* XXX bounds-check this too */ 1325 LFS_IENTRY(ifp, fs, ino, bp); 1326 daddr = ifp->if_daddr; 1327 if (fs->lfs_version > 1) { 1328 ts.tv_sec = ifp->if_atime_sec; 1329 ts.tv_nsec = ifp->if_atime_nsec; 1330 } 1331 1332 brelse(bp, 0); 1333 if (daddr == LFS_UNUSED_DADDR) { 1334 *vpp = NULLVP; 1335 mutex_exit(&ufs_hashlock); 1336 ungetnewvnode(vp); 1337 return (ENOENT); 1338 } 1339 } 1340 1341 /* Allocate/init new vnode/inode. */ 1342 lfs_vcreate(mp, ino, vp); 1343 1344 /* 1345 * Put it onto its hash chain and lock it so that other requests for 1346 * this inode will block if they arrive while we are sleeping waiting 1347 * for old data structures to be purged or for the contents of the 1348 * disk portion of this inode to be read. 1349 */ 1350 ip = VTOI(vp); 1351 ufs_ihashins(ip); 1352 mutex_exit(&ufs_hashlock); 1353 1354 /* 1355 * XXX 1356 * This may not need to be here, logically it should go down with 1357 * the i_devvp initialization. 1358 * Ask Kirk. 1359 */ 1360 ip->i_lfs = ump->um_lfs; 1361 1362 /* Read in the disk contents for the inode, copy into the inode. */ 1363 retries = 0; 1364 again: 1365 error = bread(ump->um_devvp, fsbtodb(fs, daddr), 1366 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize), 1367 NOCRED, 0, &bp); 1368 if (error) { 1369 /* 1370 * The inode does not contain anything useful, so it would 1371 * be misleading to leave it on its hash chain. With mode 1372 * still zero, it will be unlinked and returned to the free 1373 * list by vput(). 1374 */ 1375 vput(vp); 1376 brelse(bp, 0); 1377 *vpp = NULL; 1378 return (error); 1379 } 1380 1381 dip = lfs_ifind(fs, ino, bp); 1382 if (dip == NULL) { 1383 /* Assume write has not completed yet; try again */ 1384 brelse(bp, BC_INVAL); 1385 ++retries; 1386 if (retries > LFS_IFIND_RETRIES) { 1387 #ifdef DEBUG 1388 /* If the seglock is held look at the bpp to see 1389 what is there anyway */ 1390 mutex_enter(&lfs_lock); 1391 if (fs->lfs_seglock > 0) { 1392 struct buf **bpp; 1393 struct ufs1_dinode *dp; 1394 int i; 1395 1396 for (bpp = fs->lfs_sp->bpp; 1397 bpp != fs->lfs_sp->cbpp; ++bpp) { 1398 if ((*bpp)->b_vp == fs->lfs_ivnode && 1399 bpp != fs->lfs_sp->bpp) { 1400 /* Inode block */ 1401 printf("lfs_vget: block 0x%" PRIx64 ": ", 1402 (*bpp)->b_blkno); 1403 dp = (struct ufs1_dinode *)(*bpp)->b_data; 1404 for (i = 0; i < INOPB(fs); i++) 1405 if (dp[i].di_u.inumber) 1406 printf("%d ", dp[i].di_u.inumber); 1407 printf("\n"); 1408 } 1409 } 1410 } 1411 mutex_exit(&lfs_lock); 1412 #endif /* DEBUG */ 1413 panic("lfs_vget: dinode not found"); 1414 } 1415 mutex_enter(&lfs_lock); 1416 if (fs->lfs_iocount) { 1417 DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino)); 1418 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, 1419 "lfs ifind", 1, &lfs_lock); 1420 } else 1421 retries = LFS_IFIND_RETRIES; 1422 mutex_exit(&lfs_lock); 1423 goto again; 1424 } 1425 *ip->i_din.ffs1_din = *dip; 1426 brelse(bp, 0); 1427 1428 if (fs->lfs_version > 1) { 1429 ip->i_ffs1_atime = ts.tv_sec; 1430 ip->i_ffs1_atimensec = ts.tv_nsec; 1431 } 1432 1433 lfs_vinit(mp, &vp); 1434 1435 *vpp = vp; 1436 1437 KASSERT(VOP_ISLOCKED(vp)); 1438 1439 return (0); 1440 } 1441 1442 /* 1443 * File handle to vnode 1444 */ 1445 int 1446 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1447 { 1448 struct lfid lfh; 1449 struct buf *bp; 1450 IFILE *ifp; 1451 int32_t daddr; 1452 struct lfs *fs; 1453 vnode_t *vp; 1454 1455 if (fhp->fid_len != sizeof(struct lfid)) 1456 return EINVAL; 1457 1458 memcpy(&lfh, fhp, sizeof(lfh)); 1459 if (lfh.lfid_ino < LFS_IFILE_INUM) 1460 return ESTALE; 1461 1462 fs = VFSTOUFS(mp)->um_lfs; 1463 if (lfh.lfid_ident != fs->lfs_ident) 1464 return ESTALE; 1465 1466 if (lfh.lfid_ino > 1467 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) - 1468 fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb) 1469 return ESTALE; 1470 1471 mutex_enter(&ufs_ihash_lock); 1472 vp = ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino); 1473 mutex_exit(&ufs_ihash_lock); 1474 if (vp == NULL) { 1475 LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp); 1476 daddr = ifp->if_daddr; 1477 brelse(bp, 0); 1478 if (daddr == LFS_UNUSED_DADDR) 1479 return ESTALE; 1480 } 1481 1482 return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp)); 1483 } 1484 1485 /* 1486 * Vnode pointer to File handle 1487 */ 1488 /* ARGSUSED */ 1489 int 1490 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 1491 { 1492 struct inode *ip; 1493 struct lfid lfh; 1494 1495 if (*fh_size < sizeof(struct lfid)) { 1496 *fh_size = sizeof(struct lfid); 1497 return E2BIG; 1498 } 1499 *fh_size = sizeof(struct lfid); 1500 ip = VTOI(vp); 1501 memset(&lfh, 0, sizeof(lfh)); 1502 lfh.lfid_len = sizeof(struct lfid); 1503 lfh.lfid_ino = ip->i_number; 1504 lfh.lfid_gen = ip->i_gen; 1505 lfh.lfid_ident = ip->i_lfs->lfs_ident; 1506 memcpy(fhp, &lfh, sizeof(lfh)); 1507 return (0); 1508 } 1509 1510 /* 1511 * ufs_bmaparray callback function for writing. 1512 * 1513 * Since blocks will be written to the new segment anyway, 1514 * we don't care about current daddr of them. 1515 */ 1516 static bool 1517 lfs_issequential_hole(const struct ufsmount *ump, 1518 daddr_t daddr0, daddr_t daddr1) 1519 { 1520 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */ 1521 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */ 1522 1523 KASSERT(daddr0 == UNWRITTEN || 1524 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR)); 1525 KASSERT(daddr1 == UNWRITTEN || 1526 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR)); 1527 1528 /* NOTE: all we want to know here is 'hole or not'. */ 1529 /* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */ 1530 1531 /* 1532 * treat UNWRITTENs and all resident blocks as 'contiguous' 1533 */ 1534 if (daddr0 != 0 && daddr1 != 0) 1535 return true; 1536 1537 /* 1538 * both are in hole? 1539 */ 1540 if (daddr0 == 0 && daddr1 == 0) 1541 return true; /* all holes are 'contiguous' for us. */ 1542 1543 return false; 1544 } 1545 1546 /* 1547 * lfs_gop_write functions exactly like genfs_gop_write, except that 1548 * (1) it requires the seglock to be held by its caller, and sp->fip 1549 * to be properly initialized (it will return without re-initializing 1550 * sp->fip, and without calling lfs_writeseg). 1551 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1552 * to determine how large a block it can write at once (though it does 1553 * still use VOP_BMAP to find holes in the file); 1554 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1555 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1556 * now have clusters of clusters, ick.) 1557 */ 1558 static int 1559 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, 1560 int flags) 1561 { 1562 int i, error, run, haveeof = 0; 1563 int fs_bshift; 1564 vaddr_t kva; 1565 off_t eof, offset, startoffset = 0; 1566 size_t bytes, iobytes, skipbytes; 1567 bool async = (flags & PGO_SYNCIO) == 0; 1568 daddr_t lbn, blkno; 1569 struct vm_page *pg; 1570 struct buf *mbp, *bp; 1571 struct vnode *devvp = VTOI(vp)->i_devvp; 1572 struct inode *ip = VTOI(vp); 1573 struct lfs *fs = ip->i_lfs; 1574 struct segment *sp = fs->lfs_sp; 1575 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1576 1577 ASSERT_SEGLOCK(fs); 1578 1579 /* The Ifile lives in the buffer cache */ 1580 KASSERT(vp != fs->lfs_ivnode); 1581 1582 /* 1583 * We don't want to fill the disk before the cleaner has a chance 1584 * to make room for us. If we're in danger of doing that, fail 1585 * with EAGAIN. The caller will have to notice this, unlock 1586 * so the cleaner can run, relock and try again. 1587 * 1588 * We must write everything, however, if our vnode is being 1589 * reclaimed. 1590 */ 1591 if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp) 1592 goto tryagain; 1593 1594 /* 1595 * Sometimes things slip past the filters in lfs_putpages, 1596 * and the pagedaemon tries to write pages---problem is 1597 * that the pagedaemon never acquires the segment lock. 1598 * 1599 * Alternatively, pages that were clean when we called 1600 * genfs_putpages may have become dirty in the meantime. In this 1601 * case the segment header is not properly set up for blocks 1602 * to be added to it. 1603 * 1604 * Unbusy and unclean the pages, and put them on the ACTIVE 1605 * queue under the hypothesis that they couldn't have got here 1606 * unless they were modified *quite* recently. 1607 * 1608 * XXXUBC that last statement is an oversimplification of course. 1609 */ 1610 if (!LFS_SEGLOCK_HELD(fs) || 1611 (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) || 1612 (pgs[0]->offset & fs->lfs_bmask) != 0) { 1613 goto tryagain; 1614 } 1615 1616 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x", 1617 vp, pgs, npages, flags); 1618 1619 GOP_SIZE(vp, vp->v_size, &eof, 0); 1620 haveeof = 1; 1621 1622 if (vp->v_type == VREG) 1623 fs_bshift = vp->v_mount->mnt_fs_bshift; 1624 else 1625 fs_bshift = DEV_BSHIFT; 1626 error = 0; 1627 pg = pgs[0]; 1628 startoffset = pg->offset; 1629 KASSERT(eof >= 0); 1630 1631 if (startoffset >= eof) { 1632 goto tryagain; 1633 } else 1634 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 1635 skipbytes = 0; 1636 1637 KASSERT(bytes != 0); 1638 1639 /* Swap PG_DELWRI for PG_PAGEOUT */ 1640 for (i = 0; i < npages; i++) { 1641 if (pgs[i]->flags & PG_DELWRI) { 1642 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 1643 pgs[i]->flags &= ~PG_DELWRI; 1644 pgs[i]->flags |= PG_PAGEOUT; 1645 uvm_pageout_start(1); 1646 mutex_enter(&uvm_pageqlock); 1647 uvm_pageunwire(pgs[i]); 1648 mutex_exit(&uvm_pageqlock); 1649 } 1650 } 1651 1652 /* 1653 * Check to make sure we're starting on a block boundary. 1654 * We'll check later to make sure we always write entire 1655 * blocks (or fragments). 1656 */ 1657 if (startoffset & fs->lfs_bmask) 1658 printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n", 1659 startoffset, fs->lfs_bmask, 1660 startoffset & fs->lfs_bmask); 1661 KASSERT((startoffset & fs->lfs_bmask) == 0); 1662 if (bytes & fs->lfs_ffmask) { 1663 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 1664 panic("lfs_gop_write: non-integer blocks"); 1665 } 1666 1667 /* 1668 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 1669 * If we would, write what we have and try again. If we don't 1670 * have anything to write, we'll have to sleep. 1671 */ 1672 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1673 (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ? 1674 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 1675 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 1676 #if 0 1677 " with nfinfo=%d at offset 0x%x\n", 1678 (int)((SEGSUM *)(sp->segsum))->ss_nfinfo, 1679 (unsigned)fs->lfs_offset)); 1680 #endif 1681 lfs_updatemeta(sp); 1682 lfs_release_finfo(fs); 1683 (void) lfs_writeseg(fs, sp); 1684 1685 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 1686 1687 /* 1688 * Having given up all of the pager_map we were holding, 1689 * we can now wait for aiodoned to reclaim it for us 1690 * without fear of deadlock. 1691 */ 1692 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1693 UVMPAGER_MAPIN_WAITOK); 1694 } 1695 1696 mbp = getiobuf(NULL, true); 1697 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 1698 vp, mbp, vp->v_numoutput, bytes); 1699 mbp->b_bufsize = npages << PAGE_SHIFT; 1700 mbp->b_data = (void *)kva; 1701 mbp->b_resid = mbp->b_bcount = bytes; 1702 mbp->b_cflags = BC_BUSY|BC_AGE; 1703 mbp->b_iodone = uvm_aio_biodone; 1704 1705 bp = NULL; 1706 for (offset = startoffset; 1707 bytes > 0; 1708 offset += iobytes, bytes -= iobytes) { 1709 lbn = offset >> fs_bshift; 1710 error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 1711 lfs_issequential_hole); 1712 if (error) { 1713 UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d", 1714 error,0,0,0); 1715 skipbytes += bytes; 1716 bytes = 0; 1717 break; 1718 } 1719 1720 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 1721 bytes); 1722 if (blkno == (daddr_t)-1) { 1723 skipbytes += iobytes; 1724 continue; 1725 } 1726 1727 /* 1728 * Discover how much we can really pack into this buffer. 1729 */ 1730 /* If no room in the current segment, finish it up */ 1731 if (sp->sum_bytes_left < sizeof(int32_t) || 1732 sp->seg_bytes_left < (1 << fs->lfs_bshift)) { 1733 int vers; 1734 1735 lfs_updatemeta(sp); 1736 vers = sp->fip->fi_version; 1737 lfs_release_finfo(fs); 1738 (void) lfs_writeseg(fs, sp); 1739 1740 lfs_acquire_finfo(fs, ip->i_number, vers); 1741 } 1742 /* Check both for space in segment and space in segsum */ 1743 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 1744 << fs_bshift); 1745 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 1746 << fs_bshift); 1747 KASSERT(iobytes > 0); 1748 1749 /* if it's really one i/o, don't make a second buf */ 1750 if (offset == startoffset && iobytes == bytes) { 1751 bp = mbp; 1752 /* 1753 * All the LFS output is done by the segwriter. It 1754 * will increment numoutput by one for all the bufs it 1755 * recieves. However this buffer needs one extra to 1756 * account for aiodone. 1757 */ 1758 mutex_enter(&vp->v_interlock); 1759 vp->v_numoutput++; 1760 mutex_exit(&vp->v_interlock); 1761 } else { 1762 bp = getiobuf(NULL, true); 1763 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 1764 vp, bp, vp->v_numoutput, 0); 1765 nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes); 1766 /* 1767 * LFS doesn't like async I/O here, dies with 1768 * and assert in lfs_bwrite(). Is that assert 1769 * valid? I retained non-async behaviour when 1770 * converted this to use nestiobuf --pooka 1771 */ 1772 bp->b_flags &= ~B_ASYNC; 1773 } 1774 1775 /* XXX This is silly ... is this necessary? */ 1776 mutex_enter(&bufcache_lock); 1777 mutex_enter(&vp->v_interlock); 1778 bgetvp(vp, bp); 1779 mutex_exit(&vp->v_interlock); 1780 mutex_exit(&bufcache_lock); 1781 1782 bp->b_lblkno = lblkno(fs, offset); 1783 bp->b_private = mbp; 1784 if (devvp->v_type == VBLK) { 1785 bp->b_dev = devvp->v_rdev; 1786 } 1787 VOP_BWRITE(bp); 1788 while (lfs_gatherblock(sp, bp, NULL)) 1789 continue; 1790 } 1791 1792 nestiobuf_done(mbp, skipbytes, error); 1793 if (skipbytes) { 1794 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0); 1795 } 1796 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 1797 1798 if (!async) { 1799 /* Start a segment write. */ 1800 UVMHIST_LOG(ubchist, "flushing", 0,0,0,0); 1801 mutex_enter(&lfs_lock); 1802 lfs_flush(fs, 0, 1); 1803 mutex_exit(&lfs_lock); 1804 } 1805 return (0); 1806 1807 tryagain: 1808 /* 1809 * We can't write the pages, for whatever reason. 1810 * Clean up after ourselves, and make the caller try again. 1811 */ 1812 mutex_enter(&vp->v_interlock); 1813 1814 /* Tell why we're here, if we know */ 1815 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) { 1816 DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n")); 1817 } else if ((pgs[0]->offset & fs->lfs_bmask) != 0) { 1818 DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n")); 1819 } else if (haveeof && startoffset >= eof) { 1820 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 1821 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 1822 pgs[0]->offset, eof, npages)); 1823 } else if (LFS_STARVED_FOR_SEGS(fs)) { 1824 DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n")); 1825 } else { 1826 DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n")); 1827 } 1828 1829 mutex_enter(&uvm_pageqlock); 1830 for (i = 0; i < npages; i++) { 1831 pg = pgs[i]; 1832 1833 if (pg->flags & PG_PAGEOUT) 1834 uvm_pageout_done(1); 1835 if (pg->flags & PG_DELWRI) { 1836 uvm_pageunwire(pg); 1837 } 1838 uvm_pageactivate(pg); 1839 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 1840 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 1841 vp, pg->offset)); 1842 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 1843 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 1844 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 1845 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 1846 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 1847 pg->wire_count)); 1848 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 1849 pg->loan_count)); 1850 } 1851 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */ 1852 uvm_page_unbusy(pgs, npages); 1853 mutex_exit(&uvm_pageqlock); 1854 mutex_exit(&vp->v_interlock); 1855 return EAGAIN; 1856 } 1857 1858 /* 1859 * finish vnode/inode initialization. 1860 * used by lfs_vget and lfs_fastvget. 1861 */ 1862 void 1863 lfs_vinit(struct mount *mp, struct vnode **vpp) 1864 { 1865 struct vnode *vp = *vpp; 1866 struct inode *ip = VTOI(vp); 1867 struct ufsmount *ump = VFSTOUFS(mp); 1868 struct lfs *fs = ump->um_lfs; 1869 int i; 1870 1871 ip->i_mode = ip->i_ffs1_mode; 1872 ip->i_nlink = ip->i_ffs1_nlink; 1873 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size; 1874 ip->i_flags = ip->i_ffs1_flags; 1875 ip->i_gen = ip->i_ffs1_gen; 1876 ip->i_uid = ip->i_ffs1_uid; 1877 ip->i_gid = ip->i_ffs1_gid; 1878 1879 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 1880 ip->i_lfs_odnlink = ip->i_ffs1_nlink; 1881 1882 /* 1883 * Initialize the vnode from the inode, check for aliases. In all 1884 * cases re-init ip, the underlying vnode/inode may have changed. 1885 */ 1886 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 1887 ip = VTOI(vp); 1888 1889 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 1890 if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) { 1891 #ifdef DEBUG 1892 for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift; 1893 i < NDADDR; i++) { 1894 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 1895 i == 0) 1896 continue; 1897 if (ip->i_ffs1_db[i] != 0) { 1898 inconsistent: 1899 lfs_dump_dinode(ip->i_din.ffs1_din); 1900 panic("inconsistent inode"); 1901 } 1902 } 1903 for ( ; i < NDADDR + NIADDR; i++) { 1904 if (ip->i_ffs1_ib[i - NDADDR] != 0) { 1905 goto inconsistent; 1906 } 1907 } 1908 #endif /* DEBUG */ 1909 for (i = 0; i < NDADDR; i++) 1910 if (ip->i_ffs1_db[i] != 0) 1911 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 1912 } 1913 1914 #ifdef DIAGNOSTIC 1915 if (vp->v_type == VNON) { 1916 # ifdef DEBUG 1917 lfs_dump_dinode(ip->i_din.ffs1_din); 1918 # endif 1919 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 1920 (unsigned long long)ip->i_number, 1921 (ip->i_mode & IFMT) >> 12); 1922 } 1923 #endif /* DIAGNOSTIC */ 1924 1925 /* 1926 * Finish inode initialization now that aliasing has been resolved. 1927 */ 1928 1929 ip->i_devvp = ump->um_devvp; 1930 vref(ip->i_devvp); 1931 genfs_node_init(vp, &lfs_genfsops); 1932 uvm_vnp_setsize(vp, ip->i_size); 1933 1934 /* Initialize hiblk from file size */ 1935 ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1; 1936 1937 *vpp = vp; 1938 } 1939 1940 /* 1941 * Resize the filesystem to contain the specified number of segments. 1942 */ 1943 int 1944 lfs_resize_fs(struct lfs *fs, int newnsegs) 1945 { 1946 SEGUSE *sup; 1947 struct buf *bp, *obp; 1948 daddr_t olast, nlast, ilast, noff, start, end; 1949 struct vnode *ivp; 1950 struct inode *ip; 1951 int error, badnews, inc, oldnsegs; 1952 int sbbytes, csbbytes, gain, cgain; 1953 int i; 1954 1955 /* Only support v2 and up */ 1956 if (fs->lfs_version < 2) 1957 return EOPNOTSUPP; 1958 1959 /* If we're doing nothing, do it fast */ 1960 oldnsegs = fs->lfs_nseg; 1961 if (newnsegs == oldnsegs) 1962 return 0; 1963 1964 /* We always have to have two superblocks */ 1965 if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1])) 1966 return EFBIG; 1967 1968 ivp = fs->lfs_ivnode; 1969 ip = VTOI(ivp); 1970 error = 0; 1971 1972 /* Take the segment lock so no one else calls lfs_newseg() */ 1973 lfs_seglock(fs, SEGM_PROT); 1974 1975 /* 1976 * Make sure the segments we're going to be losing, if any, 1977 * are in fact empty. We hold the seglock, so their status 1978 * cannot change underneath us. Count the superblocks we lose, 1979 * while we're at it. 1980 */ 1981 sbbytes = csbbytes = 0; 1982 cgain = 0; 1983 for (i = newnsegs; i < oldnsegs; i++) { 1984 LFS_SEGENTRY(sup, fs, i, bp); 1985 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 1986 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1987 sbbytes += LFS_SBPAD; 1988 if (!(sup->su_flags & SEGUSE_DIRTY)) { 1989 ++cgain; 1990 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1991 csbbytes += LFS_SBPAD; 1992 } 1993 brelse(bp, 0); 1994 if (badnews) { 1995 error = EBUSY; 1996 goto out; 1997 } 1998 } 1999 2000 /* Note old and new segment table endpoints, and old ifile size */ 2001 olast = fs->lfs_cleansz + fs->lfs_segtabsz; 2002 nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz; 2003 ilast = ivp->v_size >> fs->lfs_bshift; 2004 noff = nlast - olast; 2005 2006 /* 2007 * Make sure no one can use the Ifile while we change it around. 2008 * Even after taking the iflock we need to make sure no one still 2009 * is holding Ifile buffers, so we get each one, to drain them. 2010 * (XXX this could be done better.) 2011 */ 2012 rw_enter(&fs->lfs_iflock, RW_WRITER); 2013 vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY); 2014 for (i = 0; i < ilast; i++) { 2015 bread(ivp, i, fs->lfs_bsize, NOCRED, 0, &bp); 2016 brelse(bp, 0); 2017 } 2018 2019 /* Allocate new Ifile blocks */ 2020 for (i = ilast; i < ilast + noff; i++) { 2021 if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0, 2022 &bp) != 0) 2023 panic("balloc extending ifile"); 2024 memset(bp->b_data, 0, fs->lfs_bsize); 2025 VOP_BWRITE(bp); 2026 } 2027 2028 /* Register new ifile size */ 2029 ip->i_size += noff * fs->lfs_bsize; 2030 ip->i_ffs1_size = ip->i_size; 2031 uvm_vnp_setsize(ivp, ip->i_size); 2032 2033 /* Copy the inode table to its new position */ 2034 if (noff != 0) { 2035 if (noff < 0) { 2036 start = nlast; 2037 end = ilast + noff; 2038 inc = 1; 2039 } else { 2040 start = ilast + noff - 1; 2041 end = nlast - 1; 2042 inc = -1; 2043 } 2044 for (i = start; i != end; i += inc) { 2045 if (bread(ivp, i, fs->lfs_bsize, NOCRED, 2046 B_MODIFY, &bp) != 0) 2047 panic("resize: bread dst blk failed"); 2048 if (bread(ivp, i - noff, fs->lfs_bsize, 2049 NOCRED, 0, &obp)) 2050 panic("resize: bread src blk failed"); 2051 memcpy(bp->b_data, obp->b_data, fs->lfs_bsize); 2052 VOP_BWRITE(bp); 2053 brelse(obp, 0); 2054 } 2055 } 2056 2057 /* If we are expanding, write the new empty SEGUSE entries */ 2058 if (newnsegs > oldnsegs) { 2059 for (i = oldnsegs; i < newnsegs; i++) { 2060 if ((error = bread(ivp, i / fs->lfs_sepb + 2061 fs->lfs_cleansz, fs->lfs_bsize, 2062 NOCRED, B_MODIFY, &bp)) != 0) 2063 panic("lfs: ifile read: %d", error); 2064 while ((i + 1) % fs->lfs_sepb && i < newnsegs) { 2065 sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb]; 2066 memset(sup, 0, sizeof(*sup)); 2067 i++; 2068 } 2069 VOP_BWRITE(bp); 2070 } 2071 } 2072 2073 /* Zero out unused superblock offsets */ 2074 for (i = 2; i < LFS_MAXNUMSB; i++) 2075 if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs) 2076 fs->lfs_sboffs[i] = 0x0; 2077 2078 /* 2079 * Correct superblock entries that depend on fs size. 2080 * The computations of these are as follows: 2081 * 2082 * size = segtod(fs, nseg) 2083 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD) 2084 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2085 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD) 2086 * + (segtod(fs, 1) - (offset - curseg)) 2087 * - segtod(fs, minfreeseg - (minfreeseg / 2)) 2088 * 2089 * XXX - we should probably adjust minfreeseg as well. 2090 */ 2091 gain = (newnsegs - oldnsegs); 2092 fs->lfs_nseg = newnsegs; 2093 fs->lfs_segtabsz = nlast - fs->lfs_cleansz; 2094 fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize); 2095 fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes); 2096 fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes) 2097 - gain * btofsb(fs, fs->lfs_bsize / 2); 2098 if (gain > 0) { 2099 fs->lfs_nclean += gain; 2100 fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize); 2101 } else { 2102 fs->lfs_nclean -= cgain; 2103 fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) - 2104 btofsb(fs, csbbytes); 2105 } 2106 2107 /* Resize segment flag cache */ 2108 fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2109 fs->lfs_nseg * sizeof(u_int32_t), 2110 M_SEGMENT, M_WAITOK); 2111 fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[1], 2112 fs->lfs_nseg * sizeof(u_int32_t), 2113 M_SEGMENT, M_WAITOK); 2114 for (i = oldnsegs; i < newnsegs; i++) 2115 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2116 2117 /* Truncate Ifile if necessary */ 2118 if (noff < 0) 2119 lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0, 2120 NOCRED); 2121 2122 /* Update cleaner info so the cleaner can die */ 2123 bread(ivp, 0, fs->lfs_bsize, NOCRED, B_MODIFY, &bp); 2124 ((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean; 2125 ((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean; 2126 VOP_BWRITE(bp); 2127 2128 /* Let Ifile accesses proceed */ 2129 VOP_UNLOCK(ivp); 2130 rw_exit(&fs->lfs_iflock); 2131 2132 out: 2133 lfs_segunlock(fs); 2134 return error; 2135 } 2136
Indexes created Tue Sep 30 07:10:03 GMT 2025