lfs_vfsops.c revision 1.329
1 /* $NetBSD: lfs_vfsops.c,v 1.329 2015/07/28 05:09:35 dholland 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 1.329 2015/07/28 05:09:35 dholland 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/lfs/ulfs_quotacommon.h> 102 #include <ufs/lfs/ulfs_inode.h> 103 #include <ufs/lfs/ulfsmount.h> 104 #include <ufs/lfs/ulfs_bswap.h> 105 #include <ufs/lfs/ulfs_extern.h> 106 107 #include <uvm/uvm.h> 108 #include <uvm/uvm_stat.h> 109 #include <uvm/uvm_pager.h> 110 #include <uvm/uvm_pdaemon.h> 111 112 #include <ufs/lfs/lfs.h> 113 #include <ufs/lfs/lfs_accessors.h> 114 #include <ufs/lfs/lfs_kernel.h> 115 #include <ufs/lfs/lfs_extern.h> 116 117 #include <miscfs/genfs/genfs.h> 118 #include <miscfs/genfs/genfs_node.h> 119 120 MODULE(MODULE_CLASS_VFS, lfs, NULL); 121 122 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int); 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 lwpid_t lfs_writer_lid = 0; 133 int lfs_do_flush = 0; 134 #ifdef LFS_KERNEL_RFW 135 int lfs_do_rfw = 0; 136 #endif 137 138 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = { 139 &lfs_vnodeop_opv_desc, 140 &lfs_specop_opv_desc, 141 &lfs_fifoop_opv_desc, 142 NULL, 143 }; 144 145 struct vfsops lfs_vfsops = { 146 .vfs_name = MOUNT_LFS, 147 .vfs_min_mount_data = sizeof (struct ulfs_args), 148 .vfs_mount = lfs_mount, 149 .vfs_start = ulfs_start, 150 .vfs_unmount = lfs_unmount, 151 .vfs_root = ulfs_root, 152 .vfs_quotactl = ulfs_quotactl, 153 .vfs_statvfs = lfs_statvfs, 154 .vfs_sync = lfs_sync, 155 .vfs_vget = lfs_vget, 156 .vfs_loadvnode = lfs_loadvnode, 157 .vfs_newvnode = lfs_newvnode, 158 .vfs_fhtovp = lfs_fhtovp, 159 .vfs_vptofh = lfs_vptofh, 160 .vfs_init = lfs_init, 161 .vfs_reinit = lfs_reinit, 162 .vfs_done = lfs_done, 163 .vfs_mountroot = lfs_mountroot, 164 .vfs_snapshot = (void *)eopnotsupp, 165 .vfs_extattrctl = lfs_extattrctl, 166 .vfs_suspendctl = (void *)eopnotsupp, 167 .vfs_renamelock_enter = genfs_renamelock_enter, 168 .vfs_renamelock_exit = genfs_renamelock_exit, 169 .vfs_fsync = (void *)eopnotsupp, 170 .vfs_opv_descs = lfs_vnodeopv_descs 171 }; 172 173 const struct genfs_ops lfs_genfsops = { 174 .gop_size = lfs_gop_size, 175 .gop_alloc = ulfs_gop_alloc, 176 .gop_write = lfs_gop_write, 177 .gop_markupdate = ulfs_gop_markupdate, 178 }; 179 180 struct shortlong { 181 const char *sname; 182 const char *lname; 183 }; 184 185 static int 186 sysctl_lfs_dostats(SYSCTLFN_ARGS) 187 { 188 extern struct lfs_stats lfs_stats; 189 extern int lfs_dostats; 190 int error; 191 192 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 193 if (error || newp == NULL) 194 return (error); 195 196 if (lfs_dostats == 0) 197 memset(&lfs_stats, 0, sizeof(lfs_stats)); 198 199 return (0); 200 } 201 202 static void 203 lfs_sysctl_setup(struct sysctllog **clog) 204 { 205 int i; 206 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 207 lfs_fs_pagetrip, lfs_ignore_lazy_sync; 208 #ifdef DEBUG 209 extern int lfs_debug_log_subsys[DLOG_MAX]; 210 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 211 { "rollforward", "Debug roll-forward code" }, 212 { "alloc", "Debug inode allocation and free list" }, 213 { "avail", "Debug space-available-now accounting" }, 214 { "flush", "Debug flush triggers" }, 215 { "lockedlist", "Debug locked list accounting" }, 216 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 217 { "vnode", "Debug vnode use during segment write" }, 218 { "segment", "Debug segment writing" }, 219 { "seguse", "Debug segment used-bytes accounting" }, 220 { "cleaner", "Debug cleaning routines" }, 221 { "mount", "Debug mount/unmount routines" }, 222 { "pagecache", "Debug UBC interactions" }, 223 { "dirop", "Debug directory-operation accounting" }, 224 { "malloc", "Debug private malloc accounting" }, 225 }; 226 #endif /* DEBUG */ 227 struct shortlong stat_names[] = { /* Must match lfs.h! */ 228 { "segsused", "Number of new segments allocated" }, 229 { "psegwrites", "Number of partial-segment writes" }, 230 { "psyncwrites", "Number of synchronous partial-segment" 231 " writes" }, 232 { "pcleanwrites", "Number of partial-segment writes by the" 233 " cleaner" }, 234 { "blocktot", "Number of blocks written" }, 235 { "cleanblocks", "Number of blocks written by the cleaner" }, 236 { "ncheckpoints", "Number of checkpoints made" }, 237 { "nwrites", "Number of whole writes" }, 238 { "nsync_writes", "Number of synchronous writes" }, 239 { "wait_exceeded", "Number of times writer waited for" 240 " cleaner" }, 241 { "write_exceeded", "Number of times writer invoked flush" }, 242 { "flush_invoked", "Number of times flush was invoked" }, 243 { "vflush_invoked", "Number of time vflush was called" }, 244 { "clean_inlocked", "Number of vnodes skipped for being dead" }, 245 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 246 { "segs_reclaimed", "Number of segments reclaimed" }, 247 }; 248 249 sysctl_createv(clog, 0, NULL, NULL, 250 CTLFLAG_PERMANENT, 251 CTLTYPE_NODE, "lfs", 252 SYSCTL_DESCR("Log-structured file system"), 253 NULL, 0, NULL, 0, 254 CTL_VFS, 5, CTL_EOL); 255 /* 256 * XXX the "5" above could be dynamic, thereby eliminating one 257 * more instance of the "number to vfs" mapping problem, but 258 * "5" is the order as taken from sys/mount.h 259 */ 260 261 sysctl_createv(clog, 0, NULL, NULL, 262 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 263 CTLTYPE_INT, "flushindir", NULL, 264 NULL, 0, &lfs_writeindir, 0, 265 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 266 sysctl_createv(clog, 0, NULL, NULL, 267 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 268 CTLTYPE_INT, "clean_vnhead", NULL, 269 NULL, 0, &lfs_clean_vnhead, 0, 270 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 271 sysctl_createv(clog, 0, NULL, NULL, 272 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 273 CTLTYPE_INT, "dostats", 274 SYSCTL_DESCR("Maintain statistics on LFS operations"), 275 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 276 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 277 sysctl_createv(clog, 0, NULL, NULL, 278 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 279 CTLTYPE_INT, "pagetrip", 280 SYSCTL_DESCR("How many dirty pages in fs triggers" 281 " a flush"), 282 NULL, 0, &lfs_fs_pagetrip, 0, 283 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 284 sysctl_createv(clog, 0, NULL, NULL, 285 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 286 CTLTYPE_INT, "ignore_lazy_sync", 287 SYSCTL_DESCR("Lazy Sync is ignored entirely"), 288 NULL, 0, &lfs_ignore_lazy_sync, 0, 289 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL); 290 #ifdef LFS_KERNEL_RFW 291 sysctl_createv(clog, 0, NULL, NULL, 292 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 293 CTLTYPE_INT, "rfw", 294 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 295 NULL, 0, &lfs_do_rfw, 0, 296 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 297 #endif 298 299 sysctl_createv(clog, 0, NULL, NULL, 300 CTLFLAG_PERMANENT, 301 CTLTYPE_NODE, "stats", 302 SYSCTL_DESCR("Debugging options"), 303 NULL, 0, NULL, 0, 304 CTL_VFS, 5, LFS_STATS, CTL_EOL); 305 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 306 sysctl_createv(clog, 0, NULL, NULL, 307 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 308 CTLTYPE_INT, stat_names[i].sname, 309 SYSCTL_DESCR(stat_names[i].lname), 310 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 311 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 312 } 313 314 #ifdef DEBUG 315 sysctl_createv(clog, 0, NULL, NULL, 316 CTLFLAG_PERMANENT, 317 CTLTYPE_NODE, "debug", 318 SYSCTL_DESCR("Debugging options"), 319 NULL, 0, NULL, 0, 320 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 321 for (i = 0; i < DLOG_MAX; i++) { 322 sysctl_createv(clog, 0, NULL, NULL, 323 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 324 CTLTYPE_INT, dlog_names[i].sname, 325 SYSCTL_DESCR(dlog_names[i].lname), 326 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 327 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 328 } 329 #endif 330 } 331 332 /* old cleaner syscall interface. see VOP_FCNTL() */ 333 static const struct syscall_package lfs_syscalls[] = { 334 { SYS_lfs_bmapv, 0, (sy_call_t *)sys_lfs_bmapv }, 335 { SYS_lfs_markv, 0, (sy_call_t *)sys_lfs_markv }, 336 { SYS_lfs_segclean, 0, (sy_call_t *)sys___lfs_segwait50 }, 337 { 0, 0, NULL }, 338 }; 339 340 static int 341 lfs_modcmd(modcmd_t cmd, void *arg) 342 { 343 int error; 344 345 switch (cmd) { 346 case MODULE_CMD_INIT: 347 error = syscall_establish(NULL, lfs_syscalls); 348 if (error) 349 return error; 350 error = vfs_attach(&lfs_vfsops); 351 if (error != 0) { 352 syscall_disestablish(NULL, lfs_syscalls); 353 break; 354 } 355 lfs_sysctl_setup(&lfs_sysctl_log); 356 break; 357 case MODULE_CMD_FINI: 358 error = vfs_detach(&lfs_vfsops); 359 if (error != 0) 360 break; 361 syscall_disestablish(NULL, lfs_syscalls); 362 sysctl_teardown(&lfs_sysctl_log); 363 break; 364 default: 365 error = ENOTTY; 366 break; 367 } 368 369 return (error); 370 } 371 372 /* 373 * XXX Same structure as FFS inodes? Should we share a common pool? 374 */ 375 struct pool lfs_inode_pool; 376 struct pool lfs_dinode_pool; 377 struct pool lfs_inoext_pool; 378 struct pool lfs_lbnentry_pool; 379 380 /* 381 * The writer daemon. UVM keeps track of how many dirty pages we are holding 382 * in lfs_subsys_pages; the daemon flushes the filesystem when this value 383 * crosses the (user-defined) threshhold LFS_MAX_PAGES. 384 */ 385 static void 386 lfs_writerd(void *arg) 387 { 388 struct mount *mp, *nmp; 389 struct lfs *fs; 390 struct vfsops *vfs = NULL; 391 int fsflags; 392 int skipc; 393 int lfsc; 394 int wrote_something = 0; 395 396 mutex_enter(&lfs_lock); 397 lfs_writer_daemon = curproc->p_pid; 398 lfs_writer_lid = curlwp->l_lid; 399 mutex_exit(&lfs_lock); 400 401 /* Take an extra reference to the LFS vfsops. */ 402 vfs = vfs_getopsbyname(MOUNT_LFS); 403 404 mutex_enter(&lfs_lock); 405 for (;;) { 406 KASSERT(mutex_owned(&lfs_lock)); 407 if (wrote_something == 0) 408 mtsleep(&lfs_writer_daemon, PVM, "lfswriter", hz/10 + 1, 409 &lfs_lock); 410 411 KASSERT(mutex_owned(&lfs_lock)); 412 wrote_something = 0; 413 414 /* 415 * If global state wants a flush, flush everything. 416 */ 417 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS || 418 locked_queue_bytes > LFS_MAX_BYTES || 419 lfs_subsys_pages > LFS_MAX_PAGES) { 420 421 if (lfs_do_flush) { 422 DLOG((DLOG_FLUSH, "lfs_writerd: lfs_do_flush\n")); 423 } 424 if (locked_queue_count > LFS_MAX_BUFS) { 425 DLOG((DLOG_FLUSH, "lfs_writerd: lqc = %d, max %d\n", 426 locked_queue_count, LFS_MAX_BUFS)); 427 } 428 if (locked_queue_bytes > LFS_MAX_BYTES) { 429 DLOG((DLOG_FLUSH, "lfs_writerd: lqb = %ld, max %ld\n", 430 locked_queue_bytes, LFS_MAX_BYTES)); 431 } 432 if (lfs_subsys_pages > LFS_MAX_PAGES) { 433 DLOG((DLOG_FLUSH, "lfs_writerd: lssp = %d, max %d\n", 434 lfs_subsys_pages, LFS_MAX_PAGES)); 435 } 436 437 lfs_flush(NULL, SEGM_WRITERD, 0); 438 lfs_do_flush = 0; 439 KASSERT(mutex_owned(&lfs_lock)); 440 continue; 441 } 442 KASSERT(mutex_owned(&lfs_lock)); 443 mutex_exit(&lfs_lock); 444 445 /* 446 * Look through the list of LFSs to see if any of them 447 * have requested pageouts. 448 */ 449 mutex_enter(&mountlist_lock); 450 lfsc = 0; 451 skipc = 0; 452 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 453 if (vfs_busy(mp, &nmp)) { 454 ++skipc; 455 continue; 456 } 457 KASSERT(!mutex_owned(&lfs_lock)); 458 if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS, 459 sizeof(mp->mnt_stat.f_fstypename)) == 0) { 460 ++lfsc; 461 fs = VFSTOULFS(mp)->um_lfs; 462 int32_t ooffset = 0; 463 fsflags = SEGM_SINGLE; 464 465 mutex_enter(&lfs_lock); 466 ooffset = lfs_sb_getoffset(fs); 467 468 if (lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs) && fs->lfs_nowrap) { 469 /* Don't try to write if we're suspended */ 470 mutex_exit(&lfs_lock); 471 vfs_unbusy(mp, false, &nmp); 472 continue; 473 } 474 if (LFS_STARVED_FOR_SEGS(fs)) { 475 mutex_exit(&lfs_lock); 476 477 DLOG((DLOG_FLUSH, "lfs_writerd: need cleaning before writing possible\n")); 478 lfs_wakeup_cleaner(fs); 479 vfs_unbusy(mp, false, &nmp); 480 continue; 481 } 482 483 if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || 484 lfs_dirvcount > LFS_MAX_DIROP) && 485 fs->lfs_dirops == 0) { 486 fsflags &= ~SEGM_SINGLE; 487 fsflags |= SEGM_CKP; 488 DLOG((DLOG_FLUSH, "lfs_writerd: checkpoint\n")); 489 lfs_flush_fs(fs, fsflags); 490 } else if (fs->lfs_pdflush) { 491 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n")); 492 lfs_flush_fs(fs, fsflags); 493 } else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) { 494 DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n")); 495 mutex_exit(&lfs_lock); 496 lfs_writer_enter(fs, "wrdirop"); 497 lfs_flush_pchain(fs); 498 lfs_writer_leave(fs); 499 mutex_enter(&lfs_lock); 500 } 501 if (lfs_sb_getoffset(fs) != ooffset) 502 ++wrote_something; 503 mutex_exit(&lfs_lock); 504 } 505 KASSERT(!mutex_owned(&lfs_lock)); 506 vfs_unbusy(mp, false, &nmp); 507 } 508 if (lfsc + skipc == 0) { 509 mutex_enter(&lfs_lock); 510 lfs_writer_daemon = 0; 511 lfs_writer_lid = 0; 512 mutex_exit(&lfs_lock); 513 mutex_exit(&mountlist_lock); 514 break; 515 } 516 mutex_exit(&mountlist_lock); 517 518 mutex_enter(&lfs_lock); 519 } 520 KASSERT(!mutex_owned(&lfs_lock)); 521 KASSERT(!mutex_owned(&mountlist_lock)); 522 523 /* Give up our extra reference so the module can be unloaded. */ 524 mutex_enter(&vfs_list_lock); 525 if (vfs != NULL) 526 vfs->vfs_refcount--; 527 mutex_exit(&vfs_list_lock); 528 529 /* Done! */ 530 kthread_exit(0); 531 } 532 533 /* 534 * Initialize the filesystem, most work done by ulfs_init. 535 */ 536 void 537 lfs_init(void) 538 { 539 540 malloc_type_attach(M_SEGMENT); 541 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 542 "lfsinopl", &pool_allocator_nointr, IPL_NONE); 543 pool_init(&lfs_dinode_pool, sizeof(struct ulfs1_dinode), 0, 0, 0, 544 "lfsdinopl", &pool_allocator_nointr, IPL_NONE); 545 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 546 "lfsinoextpl", &pool_allocator_nointr, IPL_NONE); 547 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 548 "lfslbnpool", &pool_allocator_nointr, IPL_NONE); 549 ulfs_init(); 550 551 #ifdef DEBUG 552 memset(lfs_log, 0, sizeof(lfs_log)); 553 #endif 554 mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE); 555 cv_init(&locked_queue_cv, "lfsbuf"); 556 cv_init(&lfs_writing_cv, "lfsflush"); 557 } 558 559 void 560 lfs_reinit(void) 561 { 562 ulfs_reinit(); 563 } 564 565 void 566 lfs_done(void) 567 { 568 ulfs_done(); 569 mutex_destroy(&lfs_lock); 570 cv_destroy(&locked_queue_cv); 571 cv_destroy(&lfs_writing_cv); 572 pool_destroy(&lfs_inode_pool); 573 pool_destroy(&lfs_dinode_pool); 574 pool_destroy(&lfs_inoext_pool); 575 pool_destroy(&lfs_lbnentry_pool); 576 malloc_type_detach(M_SEGMENT); 577 } 578 579 /* 580 * Called by main() when ulfs is going to be mounted as root. 581 */ 582 int 583 lfs_mountroot(void) 584 { 585 extern struct vnode *rootvp; 586 struct lfs *fs = NULL; /* LFS */ 587 struct mount *mp; 588 struct lwp *l = curlwp; 589 struct ulfsmount *ump; 590 int error; 591 592 if (device_class(root_device) != DV_DISK) 593 return (ENODEV); 594 595 if (rootdev == NODEV) 596 return (ENODEV); 597 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 598 vrele(rootvp); 599 return (error); 600 } 601 if ((error = lfs_mountfs(rootvp, mp, l))) { 602 vfs_unbusy(mp, false, NULL); 603 vfs_destroy(mp); 604 return (error); 605 } 606 mountlist_append(mp); 607 ump = VFSTOULFS(mp); 608 fs = ump->um_lfs; 609 memset(fs->lfs_dlfs.dlfs_fsmnt, 0, sizeof(fs->lfs_dlfs.dlfs_fsmnt)); 610 (void)copystr(mp->mnt_stat.f_mntonname, fs->lfs_dlfs.dlfs_fsmnt, sizeof(fs->lfs_dlfs.dlfs_fsmnt), 0); 611 (void)lfs_statvfs(mp, &mp->mnt_stat); 612 vfs_unbusy(mp, false, NULL); 613 setrootfstime((time_t)lfs_sb_gettstamp(VFSTOULFS(mp)->um_lfs)); 614 return (0); 615 } 616 617 /* 618 * VFS Operations. 619 * 620 * mount system call 621 */ 622 int 623 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 624 { 625 struct lwp *l = curlwp; 626 struct vnode *devvp; 627 struct ulfs_args *args = data; 628 struct ulfsmount *ump = NULL; 629 struct lfs *fs = NULL; /* LFS */ 630 int error = 0, update; 631 mode_t accessmode; 632 633 if (args == NULL) 634 return EINVAL; 635 if (*data_len < sizeof *args) 636 return EINVAL; 637 638 if (mp->mnt_flag & MNT_GETARGS) { 639 ump = VFSTOULFS(mp); 640 if (ump == NULL) 641 return EIO; 642 args->fspec = NULL; 643 *data_len = sizeof *args; 644 return 0; 645 } 646 647 update = mp->mnt_flag & MNT_UPDATE; 648 649 /* Check arguments */ 650 if (args->fspec != NULL) { 651 /* 652 * Look up the name and verify that it's sane. 653 */ 654 error = namei_simple_user(args->fspec, 655 NSM_FOLLOW_NOEMULROOT, &devvp); 656 if (error != 0) 657 return (error); 658 659 if (!update) { 660 /* 661 * Be sure this is a valid block device 662 */ 663 if (devvp->v_type != VBLK) 664 error = ENOTBLK; 665 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 666 error = ENXIO; 667 } else { 668 /* 669 * Be sure we're still naming the same device 670 * used for our initial mount 671 */ 672 ump = VFSTOULFS(mp); 673 if (devvp != ump->um_devvp) { 674 if (devvp->v_rdev != ump->um_devvp->v_rdev) 675 error = EINVAL; 676 else { 677 vrele(devvp); 678 devvp = ump->um_devvp; 679 vref(devvp); 680 } 681 } 682 } 683 } else { 684 if (!update) { 685 /* New mounts must have a filename for the device */ 686 return (EINVAL); 687 } else { 688 /* Use the extant mount */ 689 ump = VFSTOULFS(mp); 690 devvp = ump->um_devvp; 691 vref(devvp); 692 } 693 } 694 695 696 /* 697 * If mount by non-root, then verify that user has necessary 698 * permissions on the device. 699 */ 700 if (error == 0) { 701 accessmode = VREAD; 702 if (update ? 703 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 704 (mp->mnt_flag & MNT_RDONLY) == 0) 705 accessmode |= VWRITE; 706 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 707 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT, 708 KAUTH_REQ_SYSTEM_MOUNT_DEVICE, mp, devvp, 709 KAUTH_ARG(accessmode)); 710 VOP_UNLOCK(devvp); 711 } 712 713 if (error) { 714 vrele(devvp); 715 return (error); 716 } 717 718 if (!update) { 719 int flags; 720 721 if (mp->mnt_flag & MNT_RDONLY) 722 flags = FREAD; 723 else 724 flags = FREAD|FWRITE; 725 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 726 error = VOP_OPEN(devvp, flags, FSCRED); 727 VOP_UNLOCK(devvp); 728 if (error) 729 goto fail; 730 error = lfs_mountfs(devvp, mp, l); /* LFS */ 731 if (error) { 732 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 733 (void)VOP_CLOSE(devvp, flags, NOCRED); 734 VOP_UNLOCK(devvp); 735 goto fail; 736 } 737 738 ump = VFSTOULFS(mp); 739 fs = ump->um_lfs; 740 } else { 741 /* 742 * Update the mount. 743 */ 744 745 /* 746 * The initial mount got a reference on this 747 * device, so drop the one obtained via 748 * namei(), above. 749 */ 750 vrele(devvp); 751 752 ump = VFSTOULFS(mp); 753 fs = ump->um_lfs; 754 755 if (fs->lfs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) { 756 /* 757 * Changing from read/write to read-only. 758 * XXX: shouldn't we sync here? or does vfs do that? 759 */ 760 #ifdef LFS_QUOTA2 761 /* XXX: quotas should remain on when readonly */ 762 if (fs->lfs_use_quota2) { 763 error = lfsquota2_umount(mp, 0); 764 if (error) { 765 return error; 766 } 767 } 768 #endif 769 } 770 771 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 772 /* 773 * Changing from read-only to read/write. 774 * Note in the superblocks that we're writing. 775 */ 776 777 /* XXX: quotas should have been on even if readonly */ 778 if (fs->lfs_use_quota2) { 779 #ifdef LFS_QUOTA2 780 error = lfs_quota2_mount(mp); 781 #else 782 uprintf("%s: no kernel support for this " 783 "filesystem's quotas\n", 784 mp->mnt_stat.f_mntonname); 785 if (mp->mnt_flag & MNT_FORCE) { 786 uprintf("%s: mounting anyway; " 787 "fsck afterwards\n", 788 mp->mnt_stat.f_mntonname); 789 } else { 790 error = EINVAL; 791 } 792 #endif 793 if (error) { 794 return error; 795 } 796 } 797 798 fs->lfs_ronly = 0; 799 if (lfs_sb_getpflags(fs) & LFS_PF_CLEAN) { 800 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN); 801 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 802 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 803 } 804 } 805 if (args->fspec == NULL) 806 return EINVAL; 807 } 808 809 error = set_statvfs_info(path, UIO_USERSPACE, args->fspec, 810 UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l); 811 if (error == 0) 812 (void)strncpy(fs->lfs_dlfs.dlfs_fsmnt, 813 mp->mnt_stat.f_mntonname, 814 sizeof(fs->lfs_dlfs.dlfs_fsmnt)); 815 return error; 816 817 fail: 818 vrele(devvp); 819 return (error); 820 } 821 822 823 /* 824 * Common code for mount and mountroot 825 * LFS specific 826 */ 827 int 828 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 829 { 830 struct dlfs *tdfs, *dfs, *adfs; 831 struct lfs *fs; 832 struct ulfsmount *ump; 833 struct vnode *vp; 834 struct buf *bp, *abp; 835 dev_t dev; 836 int error, i, ronly, fsbsize; 837 kauth_cred_t cred; 838 CLEANERINFO *cip; 839 SEGUSE *sup; 840 daddr_t sb_addr; 841 842 cred = l ? l->l_cred : NOCRED; 843 844 /* 845 * Flush out any old buffers remaining from a previous use. 846 */ 847 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 848 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 849 VOP_UNLOCK(devvp); 850 if (error) 851 return (error); 852 853 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 854 855 /* Don't free random space on error. */ 856 bp = NULL; 857 abp = NULL; 858 ump = NULL; 859 860 sb_addr = LFS_LABELPAD / DEV_BSIZE; 861 while (1) { 862 /* Read in the superblock. */ 863 error = bread(devvp, sb_addr, LFS_SBPAD, 0, &bp); 864 if (error) 865 goto out; 866 dfs = (struct dlfs *)bp->b_data; 867 868 /* Check the basics. */ 869 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE || 870 dfs->dlfs_version > LFS_VERSION || 871 dfs->dlfs_bsize < sizeof(struct dlfs)) { 872 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 873 error = EINVAL; /* XXX needs translation */ 874 goto out; 875 } 876 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) { 877 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 878 dfs->dlfs_inodefmt)); 879 error = EINVAL; 880 goto out; 881 } 882 883 if (dfs->dlfs_version == 1) 884 fsbsize = DEV_BSIZE; 885 else { 886 fsbsize = 1 << dfs->dlfs_ffshift; 887 /* 888 * Could be, if the frag size is large enough, that we 889 * don't have the "real" primary superblock. If that's 890 * the case, get the real one, and try again. 891 */ 892 if (sb_addr != (dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT))) { 893 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 894 " 0x%llx is not right, trying 0x%llx\n", 895 (long long)sb_addr, 896 (long long)(dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT)))); 897 sb_addr = dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT); 898 brelse(bp, 0); 899 continue; 900 } 901 } 902 break; 903 } 904 905 /* 906 * Check the second superblock to see which is newer; then mount 907 * using the older of the two. This is necessary to ensure that 908 * the filesystem is valid if it was not unmounted cleanly. 909 */ 910 911 if (dfs->dlfs_sboffs[1] && 912 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 913 { 914 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / DEV_BSIZE), 915 LFS_SBPAD, 0, &abp); 916 if (error) 917 goto out; 918 adfs = (struct dlfs *)abp->b_data; 919 920 if (dfs->dlfs_version == 1) { 921 /* 1s resolution comparison */ 922 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 923 tdfs = adfs; 924 else 925 tdfs = dfs; 926 } else { 927 /* monotonic infinite-resolution comparison */ 928 if (adfs->dlfs_serial < dfs->dlfs_serial) 929 tdfs = adfs; 930 else 931 tdfs = dfs; 932 } 933 934 /* Check the basics. */ 935 if (tdfs->dlfs_magic != LFS_MAGIC || 936 tdfs->dlfs_bsize > MAXBSIZE || 937 tdfs->dlfs_version > LFS_VERSION || 938 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 939 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 940 " sanity failed\n")); 941 error = EINVAL; /* XXX needs translation */ 942 goto out; 943 } 944 } else { 945 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock" 946 " daddr=0x%x\n", dfs->dlfs_sboffs[1])); 947 error = EINVAL; 948 goto out; 949 } 950 951 /* Allocate the mount structure, copy the superblock into it. */ 952 fs = kmem_zalloc(sizeof(struct lfs), KM_SLEEP); 953 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 954 955 /* Compatibility */ 956 if (fs->lfs_version < 2) { 957 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE); 958 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs)); 959 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0)); 960 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs)); 961 lfs_sb_setfsbtodb(fs, 0); 962 } 963 if (lfs_sb_getresvseg(fs) == 0) 964 lfs_sb_setresvseg(fs, MIN(lfs_sb_getminfreeseg(fs) - 1, \ 965 MAX(MIN_RESV_SEGS, lfs_sb_getminfreeseg(fs) / 2 + 1))); 966 967 /* 968 * If we aren't going to be able to write meaningfully to this 969 * filesystem, and were not mounted readonly, bomb out now. 970 */ 971 if (lfs_fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 972 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 973 " we need BUFPAGES >= %lld\n", 974 (long long)((bufmem_hiwater / bufmem_lowater) * 975 LFS_INVERSE_MAX_BYTES( 976 lfs_fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 977 kmem_free(fs, sizeof(struct lfs)); 978 error = EFBIG; /* XXX needs translation */ 979 goto out; 980 } 981 982 /* Before rolling forward, lock so vget will sleep for other procs */ 983 if (l != NULL) { 984 fs->lfs_flags = LFS_NOTYET; 985 fs->lfs_rfpid = l->l_proc->p_pid; 986 } 987 988 ump = kmem_zalloc(sizeof(*ump), KM_SLEEP); 989 ump->um_lfs = fs; 990 ump->um_fstype = ULFS1; 991 /* ump->um_cleaner_thread = NULL; */ 992 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 993 brelse(bp, BC_INVAL); 994 brelse(abp, BC_INVAL); 995 } else { 996 brelse(bp, 0); 997 brelse(abp, 0); 998 } 999 bp = NULL; 1000 abp = NULL; 1001 1002 1003 /* Set up the I/O information */ 1004 fs->lfs_devbsize = DEV_BSIZE; 1005 fs->lfs_iocount = 0; 1006 fs->lfs_diropwait = 0; 1007 fs->lfs_activesb = 0; 1008 lfs_sb_setuinodes(fs, 0); 1009 fs->lfs_ravail = 0; 1010 fs->lfs_favail = 0; 1011 fs->lfs_sbactive = 0; 1012 1013 /* Set up the ifile and lock aflags */ 1014 fs->lfs_doifile = 0; 1015 fs->lfs_writer = 0; 1016 fs->lfs_dirops = 0; 1017 fs->lfs_nadirop = 0; 1018 fs->lfs_seglock = 0; 1019 fs->lfs_pdflush = 0; 1020 fs->lfs_sleepers = 0; 1021 fs->lfs_pages = 0; 1022 rw_init(&fs->lfs_fraglock); 1023 rw_init(&fs->lfs_iflock); 1024 cv_init(&fs->lfs_stopcv, "lfsstop"); 1025 1026 /* Set the file system readonly/modify bits. */ 1027 fs->lfs_ronly = ronly; 1028 if (ronly == 0) 1029 fs->lfs_fmod = 1; 1030 1031 /* ulfs-level information */ 1032 fs->um_flags = 0; 1033 fs->um_bptrtodb = lfs_sb_getffshift(fs) - DEV_BSHIFT; 1034 fs->um_seqinc = lfs_sb_getfrag(fs); 1035 fs->um_nindir = lfs_sb_getnindir(fs); 1036 fs->um_lognindir = ffs(lfs_sb_getnindir(fs)) - 1; 1037 fs->um_maxsymlinklen = lfs_sb_getmaxsymlinklen(fs); 1038 fs->um_dirblksiz = LFS_DIRBLKSIZ; 1039 fs->um_maxfilesize = lfs_sb_getmaxfilesize(fs); 1040 1041 /* quota stuff */ 1042 /* XXX: these need to come from the on-disk superblock to be used */ 1043 fs->lfs_use_quota2 = 0; 1044 fs->lfs_quota_magic = 0; 1045 fs->lfs_quota_flags = 0; 1046 fs->lfs_quotaino[0] = 0; 1047 fs->lfs_quotaino[1] = 0; 1048 1049 /* Initialize the mount structure. */ 1050 dev = devvp->v_rdev; 1051 mp->mnt_data = ump; 1052 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 1053 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 1054 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 1055 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 1056 mp->mnt_stat.f_iosize = lfs_sb_getbsize(fs); 1057 mp->mnt_flag |= MNT_LOCAL; 1058 mp->mnt_fs_bshift = lfs_sb_getbshift(fs); 1059 if (fs->um_maxsymlinklen > 0) 1060 mp->mnt_iflag |= IMNT_DTYPE; 1061 1062 ump->um_mountp = mp; 1063 ump->um_dev = dev; 1064 ump->um_devvp = devvp; 1065 for (i = 0; i < ULFS_MAXQUOTAS; i++) 1066 ump->um_quotas[i] = NULLVP; 1067 spec_node_setmountedfs(devvp, mp); 1068 1069 /* Set up reserved memory for pageout */ 1070 lfs_setup_resblks(fs); 1071 /* Set up vdirop tailq */ 1072 TAILQ_INIT(&fs->lfs_dchainhd); 1073 /* and paging tailq */ 1074 TAILQ_INIT(&fs->lfs_pchainhd); 1075 /* and delayed segment accounting for truncation list */ 1076 LIST_INIT(&fs->lfs_segdhd); 1077 1078 /* 1079 * We use the ifile vnode for almost every operation. Instead of 1080 * retrieving it from the hash table each time we retrieve it here, 1081 * artificially increment the reference count and keep a pointer 1082 * to it in the incore copy of the superblock. 1083 */ 1084 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) { 1085 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 1086 goto out; 1087 } 1088 fs->lfs_ivnode = vp; 1089 vref(vp); 1090 1091 /* Set up inode bitmap and order free list */ 1092 lfs_order_freelist(fs); 1093 1094 /* Set up segment usage flags for the autocleaner. */ 1095 fs->lfs_nactive = 0; 1096 fs->lfs_suflags = malloc(2 * sizeof(u_int32_t *), 1097 M_SEGMENT, M_WAITOK); 1098 fs->lfs_suflags[0] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t), 1099 M_SEGMENT, M_WAITOK); 1100 fs->lfs_suflags[1] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t), 1101 M_SEGMENT, M_WAITOK); 1102 memset(fs->lfs_suflags[1], 0, lfs_sb_getnseg(fs) * sizeof(u_int32_t)); 1103 for (i = 0; i < lfs_sb_getnseg(fs); i++) { 1104 int changed; 1105 1106 LFS_SEGENTRY(sup, fs, i, bp); 1107 changed = 0; 1108 if (!ronly) { 1109 if (sup->su_nbytes == 0 && 1110 !(sup->su_flags & SEGUSE_EMPTY)) { 1111 sup->su_flags |= SEGUSE_EMPTY; 1112 ++changed; 1113 } else if (!(sup->su_nbytes == 0) && 1114 (sup->su_flags & SEGUSE_EMPTY)) { 1115 sup->su_flags &= ~SEGUSE_EMPTY; 1116 ++changed; 1117 } 1118 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 1119 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 1120 ++changed; 1121 } 1122 } 1123 fs->lfs_suflags[0][i] = sup->su_flags; 1124 if (changed) 1125 LFS_WRITESEGENTRY(sup, fs, i, bp); 1126 else 1127 brelse(bp, 0); 1128 } 1129 1130 /* 1131 * XXX: if the fs has quotas, quotas should be on even if 1132 * readonly. Otherwise you can't query the quota info! 1133 * However, that's not how the quota2 code got written and I 1134 * don't know if it'll behave itself if enabled while 1135 * readonly, so for now use the same enable logic as ffs. 1136 * 1137 * XXX: also, if you use the -f behavior allowed here (and 1138 * equivalently above for remount) it will corrupt the fs. It 1139 * ought not to allow that. It should allow mounting readonly 1140 * if there are quotas and the kernel doesn't have the quota 1141 * code, but only readonly. 1142 * 1143 * XXX: and if you use the -f behavior allowed here it will 1144 * likely crash at unmount time (or remount time) because we 1145 * think quotas are active. 1146 * 1147 * Although none of this applies until there's a way to set 1148 * lfs_use_quota2 and have quotas in the fs at all. 1149 */ 1150 if (!ronly && fs->lfs_use_quota2) { 1151 #ifdef LFS_QUOTA2 1152 error = lfs_quota2_mount(mp); 1153 #else 1154 uprintf("%s: no kernel support for this filesystem's quotas\n", 1155 mp->mnt_stat.f_mntonname); 1156 if (mp->mnt_flag & MNT_FORCE) { 1157 uprintf("%s: mounting anyway; fsck afterwards\n", 1158 mp->mnt_stat.f_mntonname); 1159 } else { 1160 error = EINVAL; 1161 } 1162 #endif 1163 if (error) { 1164 /* XXX XXX must clean up the stuff immediately above */ 1165 printf("lfs_mountfs: sorry, leaking some memory\n"); 1166 goto out; 1167 } 1168 } 1169 1170 #ifdef LFS_EXTATTR 1171 /* 1172 * Initialize file-backed extended attributes for ULFS1 file 1173 * systems. 1174 * 1175 * XXX: why is this limited to ULFS1? 1176 */ 1177 if (ump->um_fstype == ULFS1) { 1178 ulfs_extattr_uepm_init(&ump->um_extattr); 1179 } 1180 #endif 1181 1182 #ifdef LFS_KERNEL_RFW 1183 lfs_roll_forward(fs, mp, l); 1184 #endif 1185 1186 /* If writing, sb is not clean; record in case of immediate crash */ 1187 if (!fs->lfs_ronly) { 1188 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN); 1189 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 1190 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 1191 } 1192 1193 /* Allow vget now that roll-forward is complete */ 1194 fs->lfs_flags &= ~(LFS_NOTYET); 1195 wakeup(&fs->lfs_flags); 1196 1197 /* 1198 * Initialize the ifile cleaner info with information from 1199 * the superblock. 1200 */ 1201 LFS_CLEANERINFO(cip, fs, bp); 1202 cip->clean = lfs_sb_getnclean(fs); 1203 cip->dirty = lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs); 1204 cip->avail = lfs_sb_getavail(fs); 1205 cip->bfree = lfs_sb_getbfree(fs); 1206 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1207 1208 /* 1209 * Mark the current segment as ACTIVE, since we're going to 1210 * be writing to it. 1211 */ 1212 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); 1213 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1214 fs->lfs_nactive++; 1215 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); /* Ifile */ 1216 1217 /* Now that roll-forward is done, unlock the Ifile */ 1218 vput(vp); 1219 1220 /* Start the pagedaemon-anticipating daemon */ 1221 mutex_enter(&lfs_lock); 1222 if (lfs_writer_daemon == 0 && lfs_writer_lid == 0 && 1223 kthread_create(PRI_BIO, 0, NULL, 1224 lfs_writerd, NULL, NULL, "lfs_writer") != 0) 1225 panic("fork lfs_writer"); 1226 mutex_exit(&lfs_lock); 1227 1228 printf("WARNING: the log-structured file system is experimental\n" 1229 "WARNING: it may cause system crashes and/or corrupt data\n"); 1230 1231 return (0); 1232 1233 out: 1234 if (bp) 1235 brelse(bp, 0); 1236 if (abp) 1237 brelse(abp, 0); 1238 if (ump) { 1239 kmem_free(ump->um_lfs, sizeof(struct lfs)); 1240 kmem_free(ump, sizeof(*ump)); 1241 mp->mnt_data = NULL; 1242 } 1243 1244 return (error); 1245 } 1246 1247 /* 1248 * unmount system call 1249 */ 1250 int 1251 lfs_unmount(struct mount *mp, int mntflags) 1252 { 1253 struct lwp *l = curlwp; 1254 struct ulfsmount *ump; 1255 struct lfs *fs; 1256 int error, flags, ronly; 1257 vnode_t *vp; 1258 1259 flags = 0; 1260 if (mntflags & MNT_FORCE) 1261 flags |= FORCECLOSE; 1262 1263 ump = VFSTOULFS(mp); 1264 fs = ump->um_lfs; 1265 1266 /* Two checkpoints */ 1267 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1268 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1269 1270 /* wake up the cleaner so it can die */ 1271 /* XXX: shouldn't this be *after* the error cases below? */ 1272 lfs_wakeup_cleaner(fs); 1273 mutex_enter(&lfs_lock); 1274 while (fs->lfs_sleepers) 1275 mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0, 1276 &lfs_lock); 1277 mutex_exit(&lfs_lock); 1278 1279 #ifdef LFS_EXTATTR 1280 if (ump->um_fstype == ULFS1) { 1281 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) { 1282 ulfs_extattr_stop(mp, curlwp); 1283 } 1284 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED) { 1285 ulfs_extattr_uepm_destroy(&ump->um_extattr); 1286 } 1287 } 1288 #endif 1289 #ifdef LFS_QUOTA 1290 if ((error = lfsquota1_umount(mp, flags)) != 0) 1291 return (error); 1292 #endif 1293 #ifdef LFS_QUOTA2 1294 if ((error = lfsquota2_umount(mp, flags)) != 0) 1295 return (error); 1296 #endif 1297 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 1298 return (error); 1299 if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0) 1300 return (error); 1301 vp = fs->lfs_ivnode; 1302 mutex_enter(vp->v_interlock); 1303 if (LIST_FIRST(&vp->v_dirtyblkhd)) 1304 panic("lfs_unmount: still dirty blocks on ifile vnode"); 1305 mutex_exit(vp->v_interlock); 1306 1307 /* Explicitly write the superblock, to update serial and pflags */ 1308 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) | LFS_PF_CLEAN); 1309 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 1310 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 1311 mutex_enter(&lfs_lock); 1312 while (fs->lfs_iocount) 1313 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 1314 &lfs_lock); 1315 mutex_exit(&lfs_lock); 1316 1317 /* Finish with the Ifile, now that we're done with it */ 1318 vgone(fs->lfs_ivnode); 1319 1320 ronly = !fs->lfs_ronly; 1321 if (ump->um_devvp->v_type != VBAD) 1322 spec_node_setmountedfs(ump->um_devvp, NULL); 1323 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1324 error = VOP_CLOSE(ump->um_devvp, 1325 ronly ? FREAD : FREAD|FWRITE, NOCRED); 1326 vput(ump->um_devvp); 1327 1328 /* Complain about page leakage */ 1329 if (fs->lfs_pages > 0) 1330 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 1331 fs->lfs_pages, lfs_subsys_pages); 1332 1333 /* Free per-mount data structures */ 1334 free(fs->lfs_ino_bitmap, M_SEGMENT); 1335 free(fs->lfs_suflags[0], M_SEGMENT); 1336 free(fs->lfs_suflags[1], M_SEGMENT); 1337 free(fs->lfs_suflags, M_SEGMENT); 1338 lfs_free_resblks(fs); 1339 cv_destroy(&fs->lfs_stopcv); 1340 rw_destroy(&fs->lfs_fraglock); 1341 rw_destroy(&fs->lfs_iflock); 1342 1343 kmem_free(fs, sizeof(struct lfs)); 1344 kmem_free(ump, sizeof(*ump)); 1345 1346 mp->mnt_data = NULL; 1347 mp->mnt_flag &= ~MNT_LOCAL; 1348 return (error); 1349 } 1350 1351 /* 1352 * Get file system statistics. 1353 * 1354 * NB: We don't lock to access the superblock here, because it's not 1355 * really that important if we get it wrong. 1356 */ 1357 int 1358 lfs_statvfs(struct mount *mp, struct statvfs *sbp) 1359 { 1360 struct lfs *fs; 1361 struct ulfsmount *ump; 1362 1363 ump = VFSTOULFS(mp); 1364 fs = ump->um_lfs; 1365 if (fs->lfs_magic != LFS_MAGIC) 1366 panic("lfs_statvfs: magic"); 1367 1368 sbp->f_bsize = lfs_sb_getbsize(fs); 1369 sbp->f_frsize = lfs_sb_getfsize(fs); 1370 sbp->f_iosize = lfs_sb_getbsize(fs); 1371 sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks; 1372 1373 sbp->f_bfree = LFS_EST_BFREE(fs); 1374 KASSERT(sbp->f_bfree <= lfs_sb_getdsize(fs)); 1375 #if 0 1376 if (sbp->f_bfree < 0) 1377 sbp->f_bfree = 0; 1378 #endif 1379 1380 sbp->f_bresvd = LFS_EST_RSVD(fs); 1381 if (sbp->f_bfree > sbp->f_bresvd) 1382 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1383 else 1384 sbp->f_bavail = 0; 1385 1386 sbp->f_files = lfs_sb_getbfree(fs) / lfs_btofsb(fs, lfs_sb_getibsize(fs)) 1387 * LFS_INOPB(fs); 1388 sbp->f_ffree = sbp->f_files - lfs_sb_getnfiles(fs); 1389 sbp->f_favail = sbp->f_ffree; 1390 sbp->f_fresvd = 0; 1391 copy_statvfs_info(sbp, mp); 1392 return (0); 1393 } 1394 1395 /* 1396 * Go through the disk queues to initiate sandbagged IO; 1397 * go through the inodes to write those that have been modified; 1398 * initiate the writing of the super block if it has been modified. 1399 * 1400 * Note: we are always called with the filesystem marked `MPBUSY'. 1401 */ 1402 int 1403 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 1404 { 1405 int error; 1406 struct lfs *fs; 1407 1408 fs = VFSTOULFS(mp)->um_lfs; 1409 if (fs->lfs_ronly) 1410 return 0; 1411 1412 /* Snapshots should not hose the syncer */ 1413 /* 1414 * XXX Sync can block here anyway, since we don't have a very 1415 * XXX good idea of how much data is pending. If it's more 1416 * XXX than a segment and lfs_nextseg is close to the end of 1417 * XXX the log, we'll likely block. 1418 */ 1419 mutex_enter(&lfs_lock); 1420 if (fs->lfs_nowrap && lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs)) { 1421 mutex_exit(&lfs_lock); 1422 return 0; 1423 } 1424 mutex_exit(&lfs_lock); 1425 1426 lfs_writer_enter(fs, "lfs_dirops"); 1427 1428 /* All syncs must be checkpoints until roll-forward is implemented. */ 1429 DLOG((DLOG_FLUSH, "lfs_sync at 0x%jx\n", 1430 (uintmax_t)lfs_sb_getoffset(fs))); 1431 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1432 lfs_writer_leave(fs); 1433 #ifdef LFS_QUOTA 1434 lfs_qsync(mp); 1435 #endif 1436 return (error); 1437 } 1438 1439 /* 1440 * Look up an LFS dinode number to find its incore vnode. If not already 1441 * in core, read it in from the specified device. Return the inode locked. 1442 * Detection and handling of mount points must be done by the calling routine. 1443 */ 1444 int 1445 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1446 { 1447 int error; 1448 1449 error = vcache_get(mp, &ino, sizeof(ino), vpp); 1450 if (error) 1451 return error; 1452 error = vn_lock(*vpp, LK_EXCLUSIVE); 1453 if (error) { 1454 vrele(*vpp); 1455 *vpp = NULL; 1456 return error; 1457 } 1458 1459 return 0; 1460 } 1461 1462 /* 1463 * Create a new vnode/inode pair and initialize what fields we can. 1464 */ 1465 static void 1466 lfs_init_vnode(struct ulfsmount *ump, ino_t ino, struct vnode *vp) 1467 { 1468 struct inode *ip; 1469 struct ulfs1_dinode *dp; 1470 1471 ASSERT_NO_SEGLOCK(ump->um_lfs); 1472 1473 /* Initialize the inode. */ 1474 ip = pool_get(&lfs_inode_pool, PR_WAITOK); 1475 memset(ip, 0, sizeof(*ip)); 1476 dp = pool_get(&lfs_dinode_pool, PR_WAITOK); 1477 memset(dp, 0, sizeof(*dp)); 1478 ip->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK); 1479 memset(ip->inode_ext.lfs, 0, sizeof(*ip->inode_ext.lfs)); 1480 ip->i_din.ffs1_din = dp; 1481 ip->i_ump = ump; 1482 ip->i_vnode = vp; 1483 ip->i_dev = ump->um_dev; 1484 ip->i_number = dp->di_inumber = ino; 1485 ip->i_lfs = ump->um_lfs; 1486 ip->i_lfs_effnblks = 0; 1487 SPLAY_INIT(&ip->i_lfs_lbtree); 1488 ip->i_lfs_nbtree = 0; 1489 LIST_INIT(&ip->i_lfs_segdhd); 1490 1491 vp->v_tag = VT_LFS; 1492 vp->v_op = lfs_vnodeop_p; 1493 vp->v_data = ip; 1494 } 1495 1496 /* 1497 * Undo lfs_init_vnode(). 1498 */ 1499 static void 1500 lfs_deinit_vnode(struct ulfsmount *ump, struct vnode *vp) 1501 { 1502 struct inode *ip = VTOI(vp); 1503 1504 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1505 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1506 pool_put(&lfs_inode_pool, ip); 1507 vp->v_data = NULL; 1508 } 1509 1510 /* 1511 * Read an inode from disk and initialize this vnode / inode pair. 1512 * Caller assures no other thread will try to load this inode. 1513 */ 1514 int 1515 lfs_loadvnode(struct mount *mp, struct vnode *vp, 1516 const void *key, size_t key_len, const void **new_key) 1517 { 1518 struct lfs *fs; 1519 struct ulfs1_dinode *dip; 1520 struct inode *ip; 1521 struct buf *bp; 1522 struct ifile *ifp; 1523 struct ulfsmount *ump; 1524 ino_t ino; 1525 daddr_t daddr; 1526 int error, retries; 1527 struct timespec ts; 1528 1529 KASSERT(key_len == sizeof(ino)); 1530 memcpy(&ino, key, key_len); 1531 1532 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1533 1534 ump = VFSTOULFS(mp); 1535 fs = ump->um_lfs; 1536 1537 /* 1538 * If the filesystem is not completely mounted yet, suspend 1539 * any access requests (wait for roll-forward to complete). 1540 */ 1541 mutex_enter(&lfs_lock); 1542 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1543 mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1544 &lfs_lock); 1545 mutex_exit(&lfs_lock); 1546 1547 /* Translate the inode number to a disk address. */ 1548 if (ino == LFS_IFILE_INUM) 1549 daddr = lfs_sb_getidaddr(fs); 1550 else { 1551 /* XXX bounds-check this too */ 1552 LFS_IENTRY(ifp, fs, ino, bp); 1553 daddr = ifp->if_daddr; 1554 if (fs->lfs_version > 1) { 1555 ts.tv_sec = ifp->if_atime_sec; 1556 ts.tv_nsec = ifp->if_atime_nsec; 1557 } 1558 1559 brelse(bp, 0); 1560 if (daddr == LFS_UNUSED_DADDR) 1561 return (ENOENT); 1562 } 1563 1564 /* Allocate/init new vnode/inode. */ 1565 lfs_init_vnode(ump, ino, vp); 1566 ip = VTOI(vp); 1567 1568 /* If the cleaner supplied the inode, use it. */ 1569 if (curlwp == ump->um_cleaner_thread && ump->um_cleaner_hint != NULL && 1570 ump->um_cleaner_hint->bi_lbn == LFS_UNUSED_LBN) { 1571 dip = ump->um_cleaner_hint->bi_bp; 1572 error = copyin(dip, ip->i_din.ffs1_din, 1573 sizeof(struct ulfs1_dinode)); 1574 if (error) { 1575 lfs_deinit_vnode(ump, vp); 1576 return error; 1577 } 1578 KASSERT(ip->i_number == ino); 1579 goto out; 1580 } 1581 1582 /* Read in the disk contents for the inode, copy into the inode. */ 1583 retries = 0; 1584 again: 1585 error = bread(ump->um_devvp, LFS_FSBTODB(fs, daddr), 1586 (fs->lfs_version == 1 ? lfs_sb_getbsize(fs) : lfs_sb_getibsize(fs)), 1587 0, &bp); 1588 if (error) { 1589 lfs_deinit_vnode(ump, vp); 1590 return error; 1591 } 1592 1593 dip = lfs_ifind(fs, ino, bp); 1594 if (dip == NULL) { 1595 /* Assume write has not completed yet; try again */ 1596 brelse(bp, BC_INVAL); 1597 ++retries; 1598 if (retries <= LFS_IFIND_RETRIES) { 1599 mutex_enter(&lfs_lock); 1600 if (fs->lfs_iocount) { 1601 DLOG((DLOG_VNODE, 1602 "%s: dinode %d not found, retrying...\n", 1603 __func__, ino)); 1604 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, 1605 "lfs ifind", 1, &lfs_lock); 1606 } else 1607 retries = LFS_IFIND_RETRIES; 1608 mutex_exit(&lfs_lock); 1609 goto again; 1610 } 1611 #ifdef DEBUG 1612 /* If the seglock is held look at the bpp to see 1613 what is there anyway */ 1614 mutex_enter(&lfs_lock); 1615 if (fs->lfs_seglock > 0) { 1616 struct buf **bpp; 1617 struct ulfs1_dinode *dp; 1618 int i; 1619 1620 for (bpp = fs->lfs_sp->bpp; 1621 bpp != fs->lfs_sp->cbpp; ++bpp) { 1622 if ((*bpp)->b_vp == fs->lfs_ivnode && 1623 bpp != fs->lfs_sp->bpp) { 1624 /* Inode block */ 1625 printf("%s: block 0x%" PRIx64 ": ", 1626 __func__, (*bpp)->b_blkno); 1627 dp = (struct ulfs1_dinode *) 1628 (*bpp)->b_data; 1629 for (i = 0; i < LFS_INOPB(fs); i++) 1630 if (dp[i].di_inumber) 1631 printf("%d ", 1632 dp[i].di_inumber); 1633 printf("\n"); 1634 } 1635 } 1636 } 1637 mutex_exit(&lfs_lock); 1638 #endif /* DEBUG */ 1639 panic("lfs_loadvnode: dinode not found"); 1640 } 1641 *ip->i_din.ffs1_din = *dip; 1642 brelse(bp, 0); 1643 1644 out: 1645 if (fs->lfs_version > 1) { 1646 ip->i_ffs1_atime = ts.tv_sec; 1647 ip->i_ffs1_atimensec = ts.tv_nsec; 1648 } 1649 1650 lfs_vinit(mp, &vp); 1651 1652 *new_key = &ip->i_number; 1653 return 0; 1654 } 1655 1656 /* 1657 * Create a new inode and initialize this vnode / inode pair. 1658 */ 1659 int 1660 lfs_newvnode(struct mount *mp, struct vnode *dvp, struct vnode *vp, 1661 struct vattr *vap, kauth_cred_t cred, 1662 size_t *key_len, const void **new_key) 1663 { 1664 ino_t ino; 1665 struct inode *ip; 1666 struct ulfsmount *ump; 1667 struct lfs *fs; 1668 int error, mode, gen; 1669 1670 KASSERT(dvp != NULL || vap->va_fileid > 0); 1671 KASSERT(dvp != NULL && dvp->v_mount == mp); 1672 KASSERT(vap->va_type != VNON); 1673 1674 *key_len = sizeof(ino); 1675 ump = VFSTOULFS(mp); 1676 fs = ump->um_lfs; 1677 mode = MAKEIMODE(vap->va_type, vap->va_mode); 1678 1679 /* 1680 * Allocate fresh inode. With "dvp == NULL" take the inode number 1681 * and version from "vap". 1682 */ 1683 if (dvp == NULL) { 1684 ino = vap->va_fileid; 1685 gen = vap->va_gen; 1686 error = lfs_valloc_fixed(fs, ino, gen); 1687 } else { 1688 error = lfs_valloc(dvp, mode, cred, &ino, &gen); 1689 } 1690 if (error) 1691 return error; 1692 1693 /* Attach inode to vnode. */ 1694 lfs_init_vnode(ump, ino, vp); 1695 ip = VTOI(vp); 1696 1697 mutex_enter(&lfs_lock); 1698 LFS_SET_UINO(ip, IN_CHANGE); 1699 mutex_exit(&lfs_lock); 1700 1701 /* Note no blocks yet */ 1702 ip->i_lfs_hiblk = -1; 1703 1704 /* Set a new generation number for this inode. */ 1705 ip->i_gen = gen; 1706 ip->i_ffs1_gen = gen; 1707 1708 memset(ip->i_lfs_fragsize, 0, 1709 ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 1710 1711 /* Set uid / gid. */ 1712 if (cred == NOCRED || cred == FSCRED) { 1713 ip->i_gid = 0; 1714 ip->i_uid = 0; 1715 } else { 1716 ip->i_gid = VTOI(dvp)->i_gid; 1717 ip->i_uid = kauth_cred_geteuid(cred); 1718 } 1719 DIP_ASSIGN(ip, gid, ip->i_gid); 1720 DIP_ASSIGN(ip, uid, ip->i_uid); 1721 1722 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 1723 error = lfs_chkiq(ip, 1, cred, 0); 1724 if (error) { 1725 lfs_vfree(dvp, ino, mode); 1726 lfs_deinit_vnode(ump, vp); 1727 1728 return error; 1729 } 1730 #endif 1731 1732 /* Set type and finalize. */ 1733 ip->i_flags = 0; 1734 DIP_ASSIGN(ip, flags, 0); 1735 ip->i_mode = mode; 1736 DIP_ASSIGN(ip, mode, mode); 1737 if (vap->va_rdev != VNOVAL) { 1738 /* 1739 * Want to be able to use this to make badblock 1740 * inodes, so don't truncate the dev number. 1741 */ 1742 if (ump->um_fstype == ULFS1) 1743 ip->i_ffs1_rdev = ulfs_rw32(vap->va_rdev, 1744 ULFS_MPNEEDSWAP(fs)); 1745 else 1746 ip->i_ffs2_rdev = ulfs_rw64(vap->va_rdev, 1747 ULFS_MPNEEDSWAP(fs)); 1748 } 1749 lfs_vinit(mp, &vp); 1750 1751 *new_key = &ip->i_number; 1752 return 0; 1753 } 1754 1755 /* 1756 * File handle to vnode 1757 */ 1758 int 1759 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1760 { 1761 struct lfid lfh; 1762 struct lfs *fs; 1763 1764 if (fhp->fid_len != sizeof(struct lfid)) 1765 return EINVAL; 1766 1767 memcpy(&lfh, fhp, sizeof(lfh)); 1768 if (lfh.lfid_ino < LFS_IFILE_INUM) 1769 return ESTALE; 1770 1771 fs = VFSTOULFS(mp)->um_lfs; 1772 if (lfh.lfid_ident != lfs_sb_getident(fs)) 1773 return ESTALE; 1774 1775 if (lfh.lfid_ino > 1776 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> lfs_sb_getbshift(fs)) - 1777 lfs_sb_getcleansz(fs) - lfs_sb_getsegtabsz(fs)) * lfs_sb_getifpb(fs)) 1778 return ESTALE; 1779 1780 return (ulfs_fhtovp(mp, &lfh.lfid_ufid, vpp)); 1781 } 1782 1783 /* 1784 * Vnode pointer to File handle 1785 */ 1786 /* ARGSUSED */ 1787 int 1788 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 1789 { 1790 struct inode *ip; 1791 struct lfid lfh; 1792 1793 if (*fh_size < sizeof(struct lfid)) { 1794 *fh_size = sizeof(struct lfid); 1795 return E2BIG; 1796 } 1797 *fh_size = sizeof(struct lfid); 1798 ip = VTOI(vp); 1799 memset(&lfh, 0, sizeof(lfh)); 1800 lfh.lfid_len = sizeof(struct lfid); 1801 lfh.lfid_ino = ip->i_number; 1802 lfh.lfid_gen = ip->i_gen; 1803 lfh.lfid_ident = lfs_sb_getident(ip->i_lfs); 1804 memcpy(fhp, &lfh, sizeof(lfh)); 1805 return (0); 1806 } 1807 1808 /* 1809 * ulfs_bmaparray callback function for writing. 1810 * 1811 * Since blocks will be written to the new segment anyway, 1812 * we don't care about current daddr of them. 1813 */ 1814 static bool 1815 lfs_issequential_hole(const struct lfs *fs, 1816 daddr_t daddr0, daddr_t daddr1) 1817 { 1818 (void)fs; /* not used */ 1819 1820 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */ 1821 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */ 1822 1823 KASSERT(daddr0 == UNWRITTEN || 1824 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR)); 1825 KASSERT(daddr1 == UNWRITTEN || 1826 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR)); 1827 1828 /* NOTE: all we want to know here is 'hole or not'. */ 1829 /* NOTE: UNASSIGNED is converted to 0 by ulfs_bmaparray. */ 1830 1831 /* 1832 * treat UNWRITTENs and all resident blocks as 'contiguous' 1833 */ 1834 if (daddr0 != 0 && daddr1 != 0) 1835 return true; 1836 1837 /* 1838 * both are in hole? 1839 */ 1840 if (daddr0 == 0 && daddr1 == 0) 1841 return true; /* all holes are 'contiguous' for us. */ 1842 1843 return false; 1844 } 1845 1846 /* 1847 * lfs_gop_write functions exactly like genfs_gop_write, except that 1848 * (1) it requires the seglock to be held by its caller, and sp->fip 1849 * to be properly initialized (it will return without re-initializing 1850 * sp->fip, and without calling lfs_writeseg). 1851 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1852 * to determine how large a block it can write at once (though it does 1853 * still use VOP_BMAP to find holes in the file); 1854 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1855 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1856 * now have clusters of clusters, ick.) 1857 */ 1858 static int 1859 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, 1860 int flags) 1861 { 1862 int i, error, run, haveeof = 0; 1863 int fs_bshift; 1864 vaddr_t kva; 1865 off_t eof, offset, startoffset = 0; 1866 size_t bytes, iobytes, skipbytes; 1867 bool async = (flags & PGO_SYNCIO) == 0; 1868 daddr_t lbn, blkno; 1869 struct vm_page *pg; 1870 struct buf *mbp, *bp; 1871 struct vnode *devvp = VTOI(vp)->i_devvp; 1872 struct inode *ip = VTOI(vp); 1873 struct lfs *fs = ip->i_lfs; 1874 struct segment *sp = fs->lfs_sp; 1875 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1876 const char * failreason = NULL; 1877 1878 ASSERT_SEGLOCK(fs); 1879 1880 /* The Ifile lives in the buffer cache */ 1881 KASSERT(vp != fs->lfs_ivnode); 1882 1883 /* 1884 * We don't want to fill the disk before the cleaner has a chance 1885 * to make room for us. If we're in danger of doing that, fail 1886 * with EAGAIN. The caller will have to notice this, unlock 1887 * so the cleaner can run, relock and try again. 1888 * 1889 * We must write everything, however, if our vnode is being 1890 * reclaimed. 1891 */ 1892 mutex_enter(vp->v_interlock); 1893 if (LFS_STARVED_FOR_SEGS(fs) && vdead_check(vp, VDEAD_NOWAIT) == 0) { 1894 mutex_exit(vp->v_interlock); 1895 failreason = "Starved for segs and not flushing vp"; 1896 goto tryagain; 1897 } 1898 mutex_exit(vp->v_interlock); 1899 1900 /* 1901 * Sometimes things slip past the filters in lfs_putpages, 1902 * and the pagedaemon tries to write pages---problem is 1903 * that the pagedaemon never acquires the segment lock. 1904 * 1905 * Alternatively, pages that were clean when we called 1906 * genfs_putpages may have become dirty in the meantime. In this 1907 * case the segment header is not properly set up for blocks 1908 * to be added to it. 1909 * 1910 * Unbusy and unclean the pages, and put them on the ACTIVE 1911 * queue under the hypothesis that they couldn't have got here 1912 * unless they were modified *quite* recently. 1913 * 1914 * XXXUBC that last statement is an oversimplification of course. 1915 */ 1916 if (!LFS_SEGLOCK_HELD(fs)) { 1917 failreason = "Seglock not held"; 1918 goto tryagain; 1919 } 1920 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) { 1921 failreason = "Inode with no_gop_write"; 1922 goto tryagain; 1923 } 1924 if ((pgs[0]->offset & lfs_sb_getbmask(fs)) != 0) { 1925 failreason = "Bad page offset"; 1926 goto tryagain; 1927 } 1928 1929 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x", 1930 vp, pgs, npages, flags); 1931 1932 GOP_SIZE(vp, vp->v_size, &eof, 0); 1933 haveeof = 1; 1934 1935 if (vp->v_type == VREG) 1936 fs_bshift = vp->v_mount->mnt_fs_bshift; 1937 else 1938 fs_bshift = DEV_BSHIFT; 1939 error = 0; 1940 pg = pgs[0]; 1941 startoffset = pg->offset; 1942 KASSERT(eof >= 0); 1943 1944 if (startoffset >= eof) { 1945 failreason = "Offset beyond EOF"; 1946 goto tryagain; 1947 } else 1948 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 1949 skipbytes = 0; 1950 1951 KASSERT(bytes != 0); 1952 1953 /* Swap PG_DELWRI for PG_PAGEOUT */ 1954 for (i = 0; i < npages; i++) { 1955 if (pgs[i]->flags & PG_DELWRI) { 1956 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 1957 pgs[i]->flags &= ~PG_DELWRI; 1958 pgs[i]->flags |= PG_PAGEOUT; 1959 uvm_pageout_start(1); 1960 mutex_enter(vp->v_interlock); 1961 mutex_enter(&uvm_pageqlock); 1962 uvm_pageunwire(pgs[i]); 1963 mutex_exit(&uvm_pageqlock); 1964 mutex_exit(vp->v_interlock); 1965 } 1966 } 1967 1968 /* 1969 * Check to make sure we're starting on a block boundary. 1970 * We'll check later to make sure we always write entire 1971 * blocks (or fragments). 1972 */ 1973 if (startoffset & lfs_sb_getbmask(fs)) 1974 printf("%" PRId64 " & %" PRIu64 " = %" PRId64 "\n", 1975 startoffset, lfs_sb_getbmask(fs), 1976 startoffset & lfs_sb_getbmask(fs)); 1977 KASSERT((startoffset & lfs_sb_getbmask(fs)) == 0); 1978 if (bytes & lfs_sb_getffmask(fs)) { 1979 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 1980 panic("lfs_gop_write: non-integer blocks"); 1981 } 1982 1983 /* 1984 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 1985 * If we would, write what we have and try again. If we don't 1986 * have anything to write, we'll have to sleep. 1987 */ 1988 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 1989 (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ? 1990 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 1991 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 1992 #if 0 1993 " with nfinfo=%d at offset 0x%jx\n", 1994 (int)((SEGSUM *)(sp->segsum))->ss_nfinfo, 1995 (uintmax_t)lfs_sb_getoffset(fs))); 1996 #endif 1997 lfs_updatemeta(sp); 1998 lfs_release_finfo(fs); 1999 (void) lfs_writeseg(fs, sp); 2000 2001 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2002 2003 /* 2004 * Having given up all of the pager_map we were holding, 2005 * we can now wait for aiodoned to reclaim it for us 2006 * without fear of deadlock. 2007 */ 2008 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2009 UVMPAGER_MAPIN_WAITOK); 2010 } 2011 2012 mbp = getiobuf(NULL, true); 2013 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 2014 vp, mbp, vp->v_numoutput, bytes); 2015 mbp->b_bufsize = npages << PAGE_SHIFT; 2016 mbp->b_data = (void *)kva; 2017 mbp->b_resid = mbp->b_bcount = bytes; 2018 mbp->b_cflags = BC_BUSY|BC_AGE; 2019 mbp->b_iodone = uvm_aio_biodone; 2020 2021 bp = NULL; 2022 for (offset = startoffset; 2023 bytes > 0; 2024 offset += iobytes, bytes -= iobytes) { 2025 lbn = offset >> fs_bshift; 2026 error = ulfs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 2027 lfs_issequential_hole); 2028 if (error) { 2029 UVMHIST_LOG(ubchist, "ulfs_bmaparray() -> %d", 2030 error,0,0,0); 2031 skipbytes += bytes; 2032 bytes = 0; 2033 break; 2034 } 2035 2036 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 2037 bytes); 2038 if (blkno == (daddr_t)-1) { 2039 skipbytes += iobytes; 2040 continue; 2041 } 2042 2043 /* 2044 * Discover how much we can really pack into this buffer. 2045 */ 2046 /* If no room in the current segment, finish it up */ 2047 if (sp->sum_bytes_left < sizeof(int32_t) || 2048 sp->seg_bytes_left < (1 << lfs_sb_getbshift(fs))) { 2049 int vers; 2050 2051 lfs_updatemeta(sp); 2052 vers = sp->fip->fi_version; 2053 lfs_release_finfo(fs); 2054 (void) lfs_writeseg(fs, sp); 2055 2056 lfs_acquire_finfo(fs, ip->i_number, vers); 2057 } 2058 /* Check both for space in segment and space in segsum */ 2059 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 2060 << fs_bshift); 2061 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 2062 << fs_bshift); 2063 KASSERT(iobytes > 0); 2064 2065 /* if it's really one i/o, don't make a second buf */ 2066 if (offset == startoffset && iobytes == bytes) { 2067 bp = mbp; 2068 /* 2069 * All the LFS output is done by the segwriter. It 2070 * will increment numoutput by one for all the bufs it 2071 * recieves. However this buffer needs one extra to 2072 * account for aiodone. 2073 */ 2074 mutex_enter(vp->v_interlock); 2075 vp->v_numoutput++; 2076 mutex_exit(vp->v_interlock); 2077 } else { 2078 bp = getiobuf(NULL, true); 2079 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 2080 vp, bp, vp->v_numoutput, 0); 2081 nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes); 2082 /* 2083 * LFS doesn't like async I/O here, dies with 2084 * an assert in lfs_bwrite(). Is that assert 2085 * valid? I retained non-async behaviour when 2086 * converted this to use nestiobuf --pooka 2087 */ 2088 bp->b_flags &= ~B_ASYNC; 2089 } 2090 2091 /* XXX This is silly ... is this necessary? */ 2092 mutex_enter(&bufcache_lock); 2093 mutex_enter(vp->v_interlock); 2094 bgetvp(vp, bp); 2095 mutex_exit(vp->v_interlock); 2096 mutex_exit(&bufcache_lock); 2097 2098 bp->b_lblkno = lfs_lblkno(fs, offset); 2099 bp->b_private = mbp; 2100 if (devvp->v_type == VBLK) { 2101 bp->b_dev = devvp->v_rdev; 2102 } 2103 VOP_BWRITE(bp->b_vp, bp); 2104 while (lfs_gatherblock(sp, bp, NULL)) 2105 continue; 2106 } 2107 2108 nestiobuf_done(mbp, skipbytes, error); 2109 if (skipbytes) { 2110 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0); 2111 } 2112 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 2113 2114 if (!async) { 2115 /* Start a segment write. */ 2116 UVMHIST_LOG(ubchist, "flushing", 0,0,0,0); 2117 mutex_enter(&lfs_lock); 2118 lfs_flush(fs, 0, 1); 2119 mutex_exit(&lfs_lock); 2120 } 2121 2122 if ((sp->seg_flags & SEGM_SINGLE) && lfs_sb_getcurseg(fs) != fs->lfs_startseg) 2123 return EAGAIN; 2124 2125 return (0); 2126 2127 tryagain: 2128 /* 2129 * We can't write the pages, for whatever reason. 2130 * Clean up after ourselves, and make the caller try again. 2131 */ 2132 mutex_enter(vp->v_interlock); 2133 2134 /* Tell why we're here, if we know */ 2135 if (failreason != NULL) { 2136 DLOG((DLOG_PAGE, "lfs_gop_write: %s\n", failreason)); 2137 } 2138 if (haveeof && startoffset >= eof) { 2139 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 2140 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 2141 pgs[0]->offset, eof, npages)); 2142 } 2143 2144 mutex_enter(&uvm_pageqlock); 2145 for (i = 0; i < npages; i++) { 2146 pg = pgs[i]; 2147 2148 if (pg->flags & PG_PAGEOUT) 2149 uvm_pageout_done(1); 2150 if (pg->flags & PG_DELWRI) { 2151 uvm_pageunwire(pg); 2152 } 2153 uvm_pageactivate(pg); 2154 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 2155 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 2156 vp, pg->offset)); 2157 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 2158 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 2159 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 2160 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 2161 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 2162 pg->wire_count)); 2163 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 2164 pg->loan_count)); 2165 } 2166 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */ 2167 uvm_page_unbusy(pgs, npages); 2168 mutex_exit(&uvm_pageqlock); 2169 mutex_exit(vp->v_interlock); 2170 return EAGAIN; 2171 } 2172 2173 /* 2174 * finish vnode/inode initialization. 2175 * used by lfs_vget. 2176 */ 2177 void 2178 lfs_vinit(struct mount *mp, struct vnode **vpp) 2179 { 2180 struct vnode *vp = *vpp; 2181 struct inode *ip = VTOI(vp); 2182 struct ulfsmount *ump = VFSTOULFS(mp); 2183 struct lfs *fs = ump->um_lfs; 2184 int i; 2185 2186 ip->i_mode = ip->i_ffs1_mode; 2187 ip->i_nlink = ip->i_ffs1_nlink; 2188 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size; 2189 ip->i_flags = ip->i_ffs1_flags; 2190 ip->i_gen = ip->i_ffs1_gen; 2191 ip->i_uid = ip->i_ffs1_uid; 2192 ip->i_gid = ip->i_ffs1_gid; 2193 2194 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 2195 ip->i_lfs_odnlink = ip->i_ffs1_nlink; 2196 2197 /* 2198 * Initialize the vnode from the inode, check for aliases. In all 2199 * cases re-init ip, the underlying vnode/inode may have changed. 2200 */ 2201 ulfs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 2202 ip = VTOI(vp); 2203 2204 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 2205 if (vp->v_type != VLNK || ip->i_size >= ip->i_lfs->um_maxsymlinklen) { 2206 #ifdef DEBUG 2207 for (i = (ip->i_size + lfs_sb_getbsize(fs) - 1) >> lfs_sb_getbshift(fs); 2208 i < ULFS_NDADDR; i++) { 2209 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 2210 i == 0) 2211 continue; 2212 if (ip->i_ffs1_db[i] != 0) { 2213 lfs_dump_dinode(ip->i_din.ffs1_din); 2214 panic("inconsistent inode (direct)"); 2215 } 2216 } 2217 for ( ; i < ULFS_NDADDR + ULFS_NIADDR; i++) { 2218 if (ip->i_ffs1_ib[i - ULFS_NDADDR] != 0) { 2219 lfs_dump_dinode(ip->i_din.ffs1_din); 2220 panic("inconsistent inode (indirect)"); 2221 } 2222 } 2223 #endif /* DEBUG */ 2224 for (i = 0; i < ULFS_NDADDR; i++) 2225 if (ip->i_ffs1_db[i] != 0) 2226 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i); 2227 } 2228 2229 #ifdef DIAGNOSTIC 2230 if (vp->v_type == VNON) { 2231 # ifdef DEBUG 2232 lfs_dump_dinode(ip->i_din.ffs1_din); 2233 # endif 2234 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 2235 (unsigned long long)ip->i_number, 2236 (ip->i_mode & LFS_IFMT) >> 12); 2237 } 2238 #endif /* DIAGNOSTIC */ 2239 2240 /* 2241 * Finish inode initialization now that aliasing has been resolved. 2242 */ 2243 2244 ip->i_devvp = ump->um_devvp; 2245 vref(ip->i_devvp); 2246 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 2247 ulfsquota_init(ip); 2248 #endif 2249 genfs_node_init(vp, &lfs_genfsops); 2250 uvm_vnp_setsize(vp, ip->i_size); 2251 2252 /* Initialize hiblk from file size */ 2253 ip->i_lfs_hiblk = lfs_lblkno(ip->i_lfs, ip->i_size + lfs_sb_getbsize(ip->i_lfs) - 1) - 1; 2254 2255 *vpp = vp; 2256 } 2257 2258 /* 2259 * Resize the filesystem to contain the specified number of segments. 2260 */ 2261 int 2262 lfs_resize_fs(struct lfs *fs, int newnsegs) 2263 { 2264 SEGUSE *sup; 2265 struct buf *bp, *obp; 2266 daddr_t olast, nlast, ilast, noff, start, end; 2267 struct vnode *ivp; 2268 struct inode *ip; 2269 int error, badnews, inc, oldnsegs; 2270 int sbbytes, csbbytes, gain, cgain; 2271 int i; 2272 2273 /* Only support v2 and up */ 2274 if (fs->lfs_version < 2) 2275 return EOPNOTSUPP; 2276 2277 /* If we're doing nothing, do it fast */ 2278 oldnsegs = lfs_sb_getnseg(fs); 2279 if (newnsegs == oldnsegs) 2280 return 0; 2281 2282 /* We always have to have two superblocks */ 2283 if (newnsegs <= lfs_dtosn(fs, lfs_sb_getsboff(fs, 1))) 2284 /* XXX this error code is rather nonsense */ 2285 return EFBIG; 2286 2287 ivp = fs->lfs_ivnode; 2288 ip = VTOI(ivp); 2289 error = 0; 2290 2291 /* Take the segment lock so no one else calls lfs_newseg() */ 2292 lfs_seglock(fs, SEGM_PROT); 2293 2294 /* 2295 * Make sure the segments we're going to be losing, if any, 2296 * are in fact empty. We hold the seglock, so their status 2297 * cannot change underneath us. Count the superblocks we lose, 2298 * while we're at it. 2299 */ 2300 sbbytes = csbbytes = 0; 2301 cgain = 0; 2302 for (i = newnsegs; i < oldnsegs; i++) { 2303 LFS_SEGENTRY(sup, fs, i, bp); 2304 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 2305 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2306 sbbytes += LFS_SBPAD; 2307 if (!(sup->su_flags & SEGUSE_DIRTY)) { 2308 ++cgain; 2309 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2310 csbbytes += LFS_SBPAD; 2311 } 2312 brelse(bp, 0); 2313 if (badnews) { 2314 error = EBUSY; 2315 goto out; 2316 } 2317 } 2318 2319 /* Note old and new segment table endpoints, and old ifile size */ 2320 olast = lfs_sb_getcleansz(fs) + lfs_sb_getsegtabsz(fs); 2321 nlast = howmany(newnsegs, lfs_sb_getsepb(fs)) + lfs_sb_getcleansz(fs); 2322 ilast = ivp->v_size >> lfs_sb_getbshift(fs); 2323 noff = nlast - olast; 2324 2325 /* 2326 * Make sure no one can use the Ifile while we change it around. 2327 * Even after taking the iflock we need to make sure no one still 2328 * is holding Ifile buffers, so we get each one, to drain them. 2329 * (XXX this could be done better.) 2330 */ 2331 rw_enter(&fs->lfs_iflock, RW_WRITER); 2332 for (i = 0; i < ilast; i++) { 2333 /* XXX what to do if bread fails? */ 2334 bread(ivp, i, lfs_sb_getbsize(fs), 0, &bp); 2335 brelse(bp, 0); 2336 } 2337 2338 /* Allocate new Ifile blocks */ 2339 for (i = ilast; i < ilast + noff; i++) { 2340 if (lfs_balloc(ivp, i * lfs_sb_getbsize(fs), lfs_sb_getbsize(fs), NOCRED, 0, 2341 &bp) != 0) 2342 panic("balloc extending ifile"); 2343 memset(bp->b_data, 0, lfs_sb_getbsize(fs)); 2344 VOP_BWRITE(bp->b_vp, bp); 2345 } 2346 2347 /* Register new ifile size */ 2348 ip->i_size += noff * lfs_sb_getbsize(fs); 2349 ip->i_ffs1_size = ip->i_size; 2350 uvm_vnp_setsize(ivp, ip->i_size); 2351 2352 /* Copy the inode table to its new position */ 2353 if (noff != 0) { 2354 if (noff < 0) { 2355 start = nlast; 2356 end = ilast + noff; 2357 inc = 1; 2358 } else { 2359 start = ilast + noff - 1; 2360 end = nlast - 1; 2361 inc = -1; 2362 } 2363 for (i = start; i != end; i += inc) { 2364 if (bread(ivp, i, lfs_sb_getbsize(fs), 2365 B_MODIFY, &bp) != 0) 2366 panic("resize: bread dst blk failed"); 2367 if (bread(ivp, i - noff, lfs_sb_getbsize(fs), 2368 0, &obp)) 2369 panic("resize: bread src blk failed"); 2370 memcpy(bp->b_data, obp->b_data, lfs_sb_getbsize(fs)); 2371 VOP_BWRITE(bp->b_vp, bp); 2372 brelse(obp, 0); 2373 } 2374 } 2375 2376 /* If we are expanding, write the new empty SEGUSE entries */ 2377 if (newnsegs > oldnsegs) { 2378 for (i = oldnsegs; i < newnsegs; i++) { 2379 if ((error = bread(ivp, i / lfs_sb_getsepb(fs) + 2380 lfs_sb_getcleansz(fs), lfs_sb_getbsize(fs), 2381 B_MODIFY, &bp)) != 0) 2382 panic("lfs: ifile read: %d", error); 2383 while ((i + 1) % lfs_sb_getsepb(fs) && i < newnsegs) { 2384 sup = &((SEGUSE *)bp->b_data)[i % lfs_sb_getsepb(fs)]; 2385 memset(sup, 0, sizeof(*sup)); 2386 i++; 2387 } 2388 VOP_BWRITE(bp->b_vp, bp); 2389 } 2390 } 2391 2392 /* Zero out unused superblock offsets */ 2393 for (i = 2; i < LFS_MAXNUMSB; i++) 2394 if (lfs_dtosn(fs, lfs_sb_getsboff(fs, i)) >= newnsegs) 2395 lfs_sb_setsboff(fs, i, 0x0); 2396 2397 /* 2398 * Correct superblock entries that depend on fs size. 2399 * The computations of these are as follows: 2400 * 2401 * size = lfs_segtod(fs, nseg) 2402 * dsize = lfs_segtod(fs, nseg - minfreeseg) - lfs_btofsb(#super * LFS_SBPAD) 2403 * bfree = dsize - lfs_btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2404 * avail = lfs_segtod(fs, nclean) - lfs_btofsb(#clean_super * LFS_SBPAD) 2405 * + (lfs_segtod(fs, 1) - (offset - curseg)) 2406 * - lfs_segtod(fs, minfreeseg - (minfreeseg / 2)) 2407 * 2408 * XXX - we should probably adjust minfreeseg as well. 2409 */ 2410 gain = (newnsegs - oldnsegs); 2411 lfs_sb_setnseg(fs, newnsegs); 2412 lfs_sb_setsegtabsz(fs, nlast - lfs_sb_getcleansz(fs)); 2413 lfs_sb_addsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs))); 2414 lfs_sb_adddsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes)); 2415 lfs_sb_addbfree(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes) 2416 - gain * lfs_btofsb(fs, lfs_sb_getbsize(fs) / 2)); 2417 if (gain > 0) { 2418 lfs_sb_addnclean(fs, gain); 2419 lfs_sb_addavail(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs))); 2420 } else { 2421 lfs_sb_subnclean(fs, cgain); 2422 lfs_sb_subavail(fs, cgain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - 2423 lfs_btofsb(fs, csbbytes)); 2424 } 2425 2426 /* Resize segment flag cache */ 2427 fs->lfs_suflags[0] = realloc(fs->lfs_suflags[0], 2428 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK); 2429 fs->lfs_suflags[1] = realloc(fs->lfs_suflags[1], 2430 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK); 2431 for (i = oldnsegs; i < newnsegs; i++) 2432 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2433 2434 /* Truncate Ifile if necessary */ 2435 if (noff < 0) 2436 lfs_truncate(ivp, ivp->v_size + (noff << lfs_sb_getbshift(fs)), 0, 2437 NOCRED); 2438 2439 /* Update cleaner info so the cleaner can die */ 2440 /* XXX what to do if bread fails? */ 2441 bread(ivp, 0, lfs_sb_getbsize(fs), B_MODIFY, &bp); 2442 ((CLEANERINFO *)bp->b_data)->clean = lfs_sb_getnclean(fs); 2443 ((CLEANERINFO *)bp->b_data)->dirty = lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs); 2444 VOP_BWRITE(bp->b_vp, bp); 2445 2446 /* Let Ifile accesses proceed */ 2447 rw_exit(&fs->lfs_iflock); 2448 2449 out: 2450 lfs_segunlock(fs); 2451 return error; 2452 } 2453 2454 /* 2455 * Extended attribute dispatch 2456 */ 2457 int 2458 lfs_extattrctl(struct mount *mp, int cmd, struct vnode *vp, 2459 int attrnamespace, const char *attrname) 2460 { 2461 #ifdef LFS_EXTATTR 2462 struct ulfsmount *ump; 2463 2464 ump = VFSTOULFS(mp); 2465 if (ump->um_fstype == ULFS1) { 2466 return ulfs_extattrctl(mp, cmd, vp, attrnamespace, attrname); 2467 } 2468 #endif 2469 return vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname); 2470 } 2471
Indexes created Sun Sep 28 12:09:53 GMT 2025