1 /* $NetBSD: lfs_vfsops.c,v 1.379 2020/08/04 03:00:47 riastradh 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@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 1.379 2020/08/04 03:00:47 riastradh 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/file.h> 82 #include <sys/disklabel.h> 83 #include <sys/ioctl.h> 84 #include <sys/errno.h> 85 #include <sys/malloc.h> 86 #include <sys/pool.h> 87 #include <sys/socket.h> 88 #include <sys/syslog.h> 89 #include <uvm/uvm_extern.h> 90 #include <sys/sysctl.h> 91 #include <sys/conf.h> 92 #include <sys/kauth.h> 93 #include <sys/module.h> 94 #include <sys/syscallvar.h> 95 #include <sys/syscall.h> 96 #include <sys/syscallargs.h> 97 98 #include <miscfs/specfs/specdev.h> 99 100 #include <ufs/lfs/ulfs_quotacommon.h> 101 #include <ufs/lfs/ulfs_inode.h> 102 #include <ufs/lfs/ulfsmount.h> 103 #include <ufs/lfs/ulfs_bswap.h> 104 #include <ufs/lfs/ulfs_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_accessors.h> 113 #include <ufs/lfs/lfs_kernel.h> 114 #include <ufs/lfs/lfs_extern.h> 115 116 #include <miscfs/genfs/genfs.h> 117 #include <miscfs/genfs/genfs_node.h> 118 119 MODULE(MODULE_CLASS_VFS, lfs, NULL); 120 121 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int); 122 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *); 123 static int lfs_flushfiles(struct mount *, int); 124 125 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc; 126 extern const struct vnodeopv_desc lfs_specop_opv_desc; 127 extern const struct vnodeopv_desc lfs_fifoop_opv_desc; 128 129 struct lwp * lfs_writer_daemon = NULL; 130 kcondvar_t lfs_writerd_cv; 131 132 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 .vfs_name = MOUNT_LFS, 146 .vfs_min_mount_data = sizeof (struct ulfs_args), 147 .vfs_mount = lfs_mount, 148 .vfs_start = ulfs_start, 149 .vfs_unmount = lfs_unmount, 150 .vfs_root = ulfs_root, 151 .vfs_quotactl = ulfs_quotactl, 152 .vfs_statvfs = lfs_statvfs, 153 .vfs_sync = lfs_sync, 154 .vfs_vget = lfs_vget, 155 .vfs_loadvnode = lfs_loadvnode, 156 .vfs_newvnode = lfs_newvnode, 157 .vfs_fhtovp = lfs_fhtovp, 158 .vfs_vptofh = lfs_vptofh, 159 .vfs_init = lfs_init, 160 .vfs_reinit = lfs_reinit, 161 .vfs_done = lfs_done, 162 .vfs_mountroot = lfs_mountroot, 163 .vfs_snapshot = (void *)eopnotsupp, 164 .vfs_extattrctl = lfs_extattrctl, 165 .vfs_suspendctl = genfs_suspendctl, 166 .vfs_renamelock_enter = genfs_renamelock_enter, 167 .vfs_renamelock_exit = genfs_renamelock_exit, 168 .vfs_fsync = (void *)eopnotsupp, 169 .vfs_opv_descs = lfs_vnodeopv_descs 170 }; 171 172 const struct genfs_ops lfs_genfsops = { 173 .gop_size = lfs_gop_size, 174 .gop_alloc = ulfs_gop_alloc, 175 .gop_write = lfs_gop_write, 176 .gop_markupdate = ulfs_gop_markupdate, 177 .gop_putrange = genfs_gop_putrange, 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 SYSCTL_SETUP(lfs_sysctl_setup, "lfs sysctl") 203 { 204 int i; 205 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 206 lfs_fs_pagetrip, lfs_ignore_lazy_sync; 207 #ifdef DEBUG 208 extern int lfs_debug_log_subsys[DLOG_MAX]; 209 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 210 { "rollforward", "Debug roll-forward code" }, 211 { "alloc", "Debug inode allocation and free list" }, 212 { "avail", "Debug space-available-now accounting" }, 213 { "flush", "Debug flush triggers" }, 214 { "lockedlist", "Debug locked list accounting" }, 215 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 216 { "vnode", "Debug vnode use during segment write" }, 217 { "segment", "Debug segment writing" }, 218 { "seguse", "Debug segment used-bytes accounting" }, 219 { "cleaner", "Debug cleaning routines" }, 220 { "mount", "Debug mount/unmount routines" }, 221 { "pagecache", "Debug UBC interactions" }, 222 { "dirop", "Debug directory-operation accounting" }, 223 { "malloc", "Debug private malloc accounting" }, 224 }; 225 #endif /* DEBUG */ 226 struct shortlong stat_names[] = { /* Must match lfs.h! */ 227 { "segsused", "Number of new segments allocated" }, 228 { "psegwrites", "Number of partial-segment writes" }, 229 { "psyncwrites", "Number of synchronous partial-segment" 230 " writes" }, 231 { "pcleanwrites", "Number of partial-segment writes by the" 232 " cleaner" }, 233 { "blocktot", "Number of blocks written" }, 234 { "cleanblocks", "Number of blocks written by the cleaner" }, 235 { "ncheckpoints", "Number of checkpoints made" }, 236 { "nwrites", "Number of whole writes" }, 237 { "nsync_writes", "Number of synchronous writes" }, 238 { "wait_exceeded", "Number of times writer waited for" 239 " cleaner" }, 240 { "write_exceeded", "Number of times writer invoked flush" }, 241 { "flush_invoked", "Number of times flush was invoked" }, 242 { "vflush_invoked", "Number of time vflush was called" }, 243 { "clean_inlocked", "Number of vnodes skipped for being dead" }, 244 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 245 { "segs_reclaimed", "Number of segments reclaimed" }, 246 }; 247 248 sysctl_createv(clog, 0, NULL, NULL, 249 CTLFLAG_PERMANENT, 250 CTLTYPE_NODE, "lfs", 251 SYSCTL_DESCR("Log-structured file system"), 252 NULL, 0, NULL, 0, 253 CTL_VFS, 5, CTL_EOL); 254 /* 255 * XXX the "5" above could be dynamic, thereby eliminating one 256 * more instance of the "number to vfs" mapping problem, but 257 * "5" is the order as taken from sys/mount.h 258 */ 259 260 sysctl_createv(clog, 0, NULL, NULL, 261 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 262 CTLTYPE_INT, "flushindir", NULL, 263 NULL, 0, &lfs_writeindir, 0, 264 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 265 sysctl_createv(clog, 0, NULL, NULL, 266 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 267 CTLTYPE_INT, "clean_vnhead", NULL, 268 NULL, 0, &lfs_clean_vnhead, 0, 269 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 270 sysctl_createv(clog, 0, NULL, NULL, 271 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 272 CTLTYPE_INT, "dostats", 273 SYSCTL_DESCR("Maintain statistics on LFS operations"), 274 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 275 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 276 sysctl_createv(clog, 0, NULL, NULL, 277 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 278 CTLTYPE_INT, "pagetrip", 279 SYSCTL_DESCR("How many dirty pages in fs triggers" 280 " a flush"), 281 NULL, 0, &lfs_fs_pagetrip, 0, 282 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 283 sysctl_createv(clog, 0, NULL, NULL, 284 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 285 CTLTYPE_INT, "ignore_lazy_sync", 286 SYSCTL_DESCR("Lazy Sync is ignored entirely"), 287 NULL, 0, &lfs_ignore_lazy_sync, 0, 288 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL); 289 #ifdef LFS_KERNEL_RFW 290 sysctl_createv(clog, 0, NULL, NULL, 291 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 292 CTLTYPE_INT, "rfw", 293 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 294 NULL, 0, &lfs_do_rfw, 0, 295 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 296 #endif 297 298 sysctl_createv(clog, 0, NULL, NULL, 299 CTLFLAG_PERMANENT, 300 CTLTYPE_NODE, "stats", 301 SYSCTL_DESCR("Debugging options"), 302 NULL, 0, NULL, 0, 303 CTL_VFS, 5, LFS_STATS, CTL_EOL); 304 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 305 sysctl_createv(clog, 0, NULL, NULL, 306 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 307 CTLTYPE_INT, stat_names[i].sname, 308 SYSCTL_DESCR(stat_names[i].lname), 309 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 310 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 311 } 312 313 #ifdef DEBUG 314 sysctl_createv(clog, 0, NULL, NULL, 315 CTLFLAG_PERMANENT, 316 CTLTYPE_NODE, "debug", 317 SYSCTL_DESCR("Debugging options"), 318 NULL, 0, NULL, 0, 319 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 320 for (i = 0; i < DLOG_MAX; i++) { 321 sysctl_createv(clog, 0, NULL, NULL, 322 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 323 CTLTYPE_INT, dlog_names[i].sname, 324 SYSCTL_DESCR(dlog_names[i].lname), 325 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 326 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 327 } 328 #endif 329 } 330 331 /* old cleaner syscall interface. see VOP_FCNTL() */ 332 static const struct syscall_package lfs_syscalls[] = { 333 { SYS_lfs_bmapv, 0, (sy_call_t *)sys_lfs_bmapv }, 334 { SYS_lfs_markv, 0, (sy_call_t *)sys_lfs_markv }, 335 { SYS___lfs_segwait50, 0, (sy_call_t *)sys___lfs_segwait50 }, 336 { SYS_lfs_segclean, 0, (sy_call_t *)sys_lfs_segclean }, 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 cv_init(&lfs_allclean_wakeup, "segment"); 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 cv_destroy(&lfs_allclean_wakeup); 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 mount_iterator_t *iter; 389 struct mount *mp; 390 struct lfs *fs; 391 struct vfsops *vfs = NULL; 392 int fsflags; 393 int lfsc; 394 int wrote_something = 0; 395 396 mutex_enter(&lfs_lock); 397 KASSERTMSG(lfs_writer_daemon == NULL, "more than one LFS writer daemon"); 398 lfs_writer_daemon = curlwp; 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 cv_timedwait(&lfs_writerd_cv, &lfs_lock, hz/10 + 1); 409 KASSERT(mutex_owned(&lfs_lock)); 410 wrote_something = 0; 411 412 /* 413 * If global state wants a flush, flush everything. 414 */ 415 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS || 416 locked_queue_bytes > LFS_MAX_BYTES || 417 lfs_subsys_pages > LFS_MAX_PAGES) { 418 419 if (lfs_do_flush) { 420 DLOG((DLOG_FLUSH, "lfs_writerd: lfs_do_flush\n")); 421 } 422 if (locked_queue_count > LFS_MAX_BUFS) { 423 DLOG((DLOG_FLUSH, "lfs_writerd: lqc = %d, max %d\n", 424 locked_queue_count, LFS_MAX_BUFS)); 425 } 426 if (locked_queue_bytes > LFS_MAX_BYTES) { 427 DLOG((DLOG_FLUSH, "lfs_writerd: lqb = %ld, max %ld\n", 428 locked_queue_bytes, LFS_MAX_BYTES)); 429 } 430 if (lfs_subsys_pages > LFS_MAX_PAGES) { 431 DLOG((DLOG_FLUSH, "lfs_writerd: lssp = %d, max %d\n", 432 lfs_subsys_pages, LFS_MAX_PAGES)); 433 } 434 435 lfs_flush(NULL, SEGM_WRITERD, 0); 436 lfs_do_flush = 0; 437 KASSERT(mutex_owned(&lfs_lock)); 438 continue; 439 } 440 KASSERT(mutex_owned(&lfs_lock)); 441 mutex_exit(&lfs_lock); 442 443 /* 444 * Look through the list of LFSs to see if any of them 445 * have requested pageouts. 446 */ 447 mountlist_iterator_init(&iter); 448 lfsc = 0; 449 while ((mp = mountlist_iterator_next(iter)) != NULL) { 450 KASSERT(!mutex_owned(&lfs_lock)); 451 if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS, 452 sizeof(mp->mnt_stat.f_fstypename)) == 0) { 453 ++lfsc; 454 fs = VFSTOULFS(mp)->um_lfs; 455 daddr_t ooffset = 0; 456 fsflags = SEGM_SINGLE; 457 458 mutex_enter(&lfs_lock); 459 ooffset = lfs_sb_getoffset(fs); 460 461 if (lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs) && fs->lfs_nowrap) { 462 /* Don't try to write if we're suspended */ 463 mutex_exit(&lfs_lock); 464 continue; 465 } 466 if (LFS_STARVED_FOR_SEGS(fs)) { 467 mutex_exit(&lfs_lock); 468 469 DLOG((DLOG_FLUSH, "lfs_writerd: need cleaning before writing possible\n")); 470 lfs_wakeup_cleaner(fs); 471 continue; 472 } 473 474 if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || 475 lfs_dirvcount > LFS_MAX_DIROP) && 476 fs->lfs_dirops == 0) { 477 fsflags &= ~SEGM_SINGLE; 478 fsflags |= SEGM_CKP; 479 DLOG((DLOG_FLUSH, "lfs_writerd: checkpoint\n")); 480 lfs_flush_fs(fs, fsflags); 481 } else if (fs->lfs_pdflush) { 482 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n")); 483 lfs_flush_fs(fs, fsflags); 484 } else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) { 485 DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n")); 486 mutex_exit(&lfs_lock); 487 lfs_writer_enter(fs, "wrdirop"); 488 lfs_flush_pchain(fs); 489 lfs_writer_leave(fs); 490 mutex_enter(&lfs_lock); 491 } 492 if (lfs_sb_getoffset(fs) != ooffset) 493 ++wrote_something; 494 mutex_exit(&lfs_lock); 495 } 496 KASSERT(!mutex_owned(&lfs_lock)); 497 } 498 if (lfsc == 0) { 499 mutex_enter(&lfs_lock); 500 lfs_writer_daemon = NULL; 501 mutex_exit(&lfs_lock); 502 mountlist_iterator_destroy(iter); 503 break; 504 } 505 mountlist_iterator_destroy(iter); 506 507 mutex_enter(&lfs_lock); 508 } 509 KASSERT(!mutex_owned(&lfs_lock)); 510 511 /* Give up our extra reference so the module can be unloaded. */ 512 mutex_enter(&vfs_list_lock); 513 if (vfs != NULL) 514 vfs->vfs_refcount--; 515 mutex_exit(&vfs_list_lock); 516 517 /* Done! */ 518 kthread_exit(0); 519 } 520 521 /* 522 * Initialize the filesystem, most work done by ulfs_init. 523 */ 524 void 525 lfs_init(void) 526 { 527 528 /* 529 * XXX: should we use separate pools for 32-bit and 64-bit 530 * dinodes? 531 */ 532 malloc_type_attach(M_SEGMENT); 533 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 534 "lfsinopl", &pool_allocator_nointr, IPL_NONE); 535 pool_init(&lfs_dinode_pool, sizeof(union lfs_dinode), 0, 0, 0, 536 "lfsdinopl", &pool_allocator_nointr, IPL_NONE); 537 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 538 "lfsinoextpl", &pool_allocator_nointr, IPL_NONE); 539 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 540 "lfslbnpool", &pool_allocator_nointr, IPL_NONE); 541 ulfs_init(); 542 543 #ifdef DEBUG 544 memset(lfs_log, 0, sizeof(lfs_log)); 545 #endif 546 mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE); 547 cv_init(&lfs_writerd_cv, "lfswrite"); 548 cv_init(&locked_queue_cv, "lfsbuf"); 549 cv_init(&lfs_writing_cv, "lfsflush"); 550 } 551 552 void 553 lfs_reinit(void) 554 { 555 ulfs_reinit(); 556 } 557 558 void 559 lfs_done(void) 560 { 561 ulfs_done(); 562 mutex_destroy(&lfs_lock); 563 cv_destroy(&lfs_writerd_cv); 564 cv_destroy(&locked_queue_cv); 565 cv_destroy(&lfs_writing_cv); 566 pool_destroy(&lfs_inode_pool); 567 pool_destroy(&lfs_dinode_pool); 568 pool_destroy(&lfs_inoext_pool); 569 pool_destroy(&lfs_lbnentry_pool); 570 malloc_type_detach(M_SEGMENT); 571 } 572 573 /* 574 * Called by main() when ulfs is going to be mounted as root. 575 */ 576 int 577 lfs_mountroot(void) 578 { 579 extern struct vnode *rootvp; 580 struct lfs *fs = NULL; /* LFS */ 581 struct mount *mp; 582 struct lwp *l = curlwp; 583 struct ulfsmount *ump; 584 int error; 585 586 if (device_class(root_device) != DV_DISK) 587 return (ENODEV); 588 589 if (rootdev == NODEV) 590 return (ENODEV); 591 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 592 vrele(rootvp); 593 return (error); 594 } 595 if ((error = lfs_mountfs(rootvp, mp, l))) { 596 vfs_unbusy(mp); 597 vfs_rele(mp); 598 return (error); 599 } 600 mountlist_append(mp); 601 ump = VFSTOULFS(mp); 602 fs = ump->um_lfs; 603 lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname); 604 (void)lfs_statvfs(mp, &mp->mnt_stat); 605 vfs_unbusy(mp); 606 setrootfstime((time_t)lfs_sb_gettstamp(VFSTOULFS(mp)->um_lfs)); 607 return (0); 608 } 609 610 /* 611 * VFS Operations. 612 * 613 * mount system call 614 */ 615 int 616 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 617 { 618 struct lwp *l = curlwp; 619 struct vnode *devvp; 620 struct ulfs_args *args = data; 621 struct ulfsmount *ump = NULL; 622 struct lfs *fs = NULL; /* LFS */ 623 int error = 0, update; 624 mode_t accessmode; 625 626 if (args == NULL) 627 return EINVAL; 628 if (*data_len < sizeof *args) 629 return EINVAL; 630 631 if (mp->mnt_flag & MNT_GETARGS) { 632 ump = VFSTOULFS(mp); 633 if (ump == NULL) 634 return EIO; 635 args->fspec = NULL; 636 *data_len = sizeof *args; 637 return 0; 638 } 639 640 update = mp->mnt_flag & MNT_UPDATE; 641 642 /* Check arguments */ 643 if (args->fspec != NULL) { 644 /* 645 * Look up the name and verify that it's sane. 646 */ 647 error = namei_simple_user(args->fspec, 648 NSM_FOLLOW_NOEMULROOT, &devvp); 649 if (error != 0) 650 return (error); 651 652 if (!update) { 653 /* 654 * Be sure this is a valid block device 655 */ 656 if (devvp->v_type != VBLK) 657 error = ENOTBLK; 658 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 659 error = ENXIO; 660 } else { 661 /* 662 * Be sure we're still naming the same device 663 * used for our initial mount 664 * 665 * XXX dholland 20151010: if namei gives us a 666 * different vnode for the same device, 667 * wouldn't it be better to use it going 668 * forward rather than ignore it in favor of 669 * the old one? 670 */ 671 ump = VFSTOULFS(mp); 672 fs = ump->um_lfs; 673 if (devvp != fs->lfs_devvp) { 674 if (devvp->v_rdev != fs->lfs_devvp->v_rdev) 675 error = EINVAL; 676 else { 677 vrele(devvp); 678 devvp = fs->lfs_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 fs = ump->um_lfs; 691 devvp = fs->lfs_devvp; 692 vref(devvp); 693 } 694 } 695 696 697 /* 698 * If mount by non-root, then verify that user has necessary 699 * permissions on the device. 700 */ 701 if (error == 0) { 702 accessmode = VREAD; 703 if (update ? 704 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 705 (mp->mnt_flag & MNT_RDONLY) == 0) 706 accessmode |= VWRITE; 707 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 708 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT, 709 KAUTH_REQ_SYSTEM_MOUNT_DEVICE, mp, devvp, 710 KAUTH_ARG(accessmode)); 711 VOP_UNLOCK(devvp); 712 } 713 714 if (error) { 715 vrele(devvp); 716 return (error); 717 } 718 719 if (!update) { 720 int flags; 721 722 if (mp->mnt_flag & MNT_RDONLY) 723 flags = FREAD; 724 else 725 flags = FREAD|FWRITE; 726 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 727 error = VOP_OPEN(devvp, flags, FSCRED); 728 VOP_UNLOCK(devvp); 729 if (error) 730 goto fail; 731 error = lfs_mountfs(devvp, mp, l); /* LFS */ 732 if (error) { 733 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 734 (void)VOP_CLOSE(devvp, flags, NOCRED); 735 VOP_UNLOCK(devvp); 736 goto fail; 737 } 738 739 ump = VFSTOULFS(mp); 740 fs = ump->um_lfs; 741 } else { 742 /* 743 * Update the mount. 744 */ 745 746 /* 747 * The initial mount got a reference on this 748 * device, so drop the one obtained via 749 * namei(), above. 750 */ 751 vrele(devvp); 752 753 ump = VFSTOULFS(mp); 754 fs = ump->um_lfs; 755 756 if (!fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDONLY)) { 757 /* 758 * Changing from read/write to read-only. 759 */ 760 int flags = WRITECLOSE; 761 if (mp->mnt_flag & MNT_FORCE) 762 flags |= FORCECLOSE; 763 error = lfs_flushfiles(mp, flags); 764 if (error) 765 return error; 766 fs->lfs_ronly = 1; 767 } else if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 768 /* 769 * Changing from read-only to read/write. 770 * Note in the superblocks that we're writing. 771 */ 772 773 /* XXX: quotas should have been on even if readonly */ 774 if (fs->lfs_use_quota2) { 775 #ifdef LFS_QUOTA2 776 error = lfs_quota2_mount(mp); 777 #else 778 uprintf("%s: no kernel support for this " 779 "filesystem's quotas\n", 780 mp->mnt_stat.f_mntonname); 781 if (mp->mnt_flag & MNT_FORCE) { 782 uprintf("%s: mounting anyway; " 783 "fsck afterwards\n", 784 mp->mnt_stat.f_mntonname); 785 } else { 786 error = EINVAL; 787 } 788 #endif 789 if (error) { 790 return error; 791 } 792 } 793 794 fs->lfs_ronly = 0; 795 if (lfs_sb_getpflags(fs) & LFS_PF_CLEAN) { 796 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN); 797 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 798 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 799 } 800 } 801 802 if (args->fspec == NULL) 803 return 0; 804 } 805 806 error = set_statvfs_info(path, UIO_USERSPACE, args->fspec, 807 UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l); 808 if (error == 0) 809 lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname); 810 return error; 811 812 fail: 813 vrele(devvp); 814 return (error); 815 } 816 817 /* 818 * Helper for mountfs. Note that the fs pointer may be a dummy one 819 * pointing into a superblock buffer. (Which is gross; see below.) 820 */ 821 static int 822 lfs_checkmagic(struct lfs *fs) 823 { 824 switch (fs->lfs_dlfs_u.u_32.dlfs_magic) { 825 case LFS_MAGIC: 826 fs->lfs_is64 = false; 827 fs->lfs_dobyteswap = false; 828 break; 829 case LFS64_MAGIC: 830 fs->lfs_is64 = true; 831 fs->lfs_dobyteswap = false; 832 break; 833 #ifdef LFS_EI 834 case LFS_MAGIC_SWAPPED: 835 fs->lfs_is64 = false; 836 fs->lfs_dobyteswap = true; 837 break; 838 case LFS64_MAGIC_SWAPPED: 839 fs->lfs_is64 = true; 840 fs->lfs_dobyteswap = true; 841 break; 842 #endif 843 default: 844 /* XXX needs translation */ 845 return EINVAL; 846 } 847 return 0; 848 } 849 850 /* 851 * Common code for mount and mountroot 852 * LFS specific 853 */ 854 int 855 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 856 { 857 struct lfs *primarysb, *altsb, *thesb; 858 struct buf *primarybuf, *altbuf; 859 struct lfs *fs; 860 struct ulfsmount *ump; 861 struct vnode *vp; 862 dev_t dev; 863 int error, i, ronly, fsbsize; 864 kauth_cred_t cred; 865 CLEANERINFO *cip; 866 SEGUSE *sup; 867 daddr_t sb_addr; 868 ino_t *orphan; 869 size_t norphan; 870 871 cred = l ? l->l_cred : NOCRED; 872 873 /* The superblock is supposed to be 512 bytes. */ 874 __CTASSERT(sizeof(struct dlfs) == DEV_BSIZE); 875 876 /* 877 * Flush out any old buffers remaining from a previous use. 878 */ 879 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 880 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 881 VOP_UNLOCK(devvp); 882 if (error) 883 return (error); 884 885 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 886 887 /* Don't free random space on error. */ 888 primarybuf = NULL; 889 altbuf = NULL; 890 ump = NULL; 891 892 sb_addr = LFS_LABELPAD / DEV_BSIZE; 893 while (1) { 894 /* 895 * Read in the superblock. 896 * 897 * Note that because LFS_SBPAD is substantially larger 898 * (8K) than the actual on-disk superblock (512 bytes) 899 * the buffer contains enough space to be used as a 900 * whole struct lfs (in-memory superblock) - we do this 901 * only so we can set and use the is64 and dobyteswap 902 * members. XXX this is gross and the logic here should 903 * be reworked. 904 */ 905 error = bread(devvp, sb_addr, LFS_SBPAD, 0, &primarybuf); 906 if (error) 907 goto out; 908 primarysb = (struct lfs *)primarybuf->b_data; 909 910 /* Check the basics. */ 911 error = lfs_checkmagic(primarysb); 912 if (error) { 913 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock wrong magic\n")); 914 goto out; 915 } 916 if (lfs_sb_getbsize(primarysb) > MAXBSIZE || 917 lfs_sb_getversion(primarysb) > LFS_VERSION || 918 lfs_sb_getbsize(primarysb) < sizeof(struct dlfs)) { 919 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 920 /* XXX needs translation */ 921 error = EINVAL; 922 goto out; 923 } 924 if (lfs_sb_getinodefmt(primarysb) > LFS_MAXINODEFMT) { 925 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 926 lfs_sb_getinodefmt(primarysb))); 927 error = EINVAL; 928 goto out; 929 } 930 931 if (lfs_sb_getversion(primarysb) == 1) 932 fsbsize = DEV_BSIZE; 933 else { 934 fsbsize = 1 << lfs_sb_getffshift(primarysb); 935 /* 936 * Could be, if the frag size is large enough, that we 937 * don't have the "real" primary superblock. If that's 938 * the case, get the real one, and try again. 939 */ 940 if (sb_addr != (lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT))) { 941 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 942 " 0x%llx is not right, trying 0x%llx\n", 943 (long long)sb_addr, 944 (long long)(lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT)))); 945 sb_addr = lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT); 946 brelse(primarybuf, BC_INVAL); 947 continue; 948 } 949 } 950 break; 951 } 952 953 /* 954 * Check the second superblock to see which is newer; then mount 955 * using the older of the two. This is necessary to ensure that 956 * the filesystem is valid if it was not unmounted cleanly. 957 */ 958 959 if (lfs_sb_getsboff(primarysb, 1) && 960 lfs_sb_getsboff(primarysb, 1) - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 961 { 962 error = bread(devvp, lfs_sb_getsboff(primarysb, 1) * (fsbsize / DEV_BSIZE), 963 LFS_SBPAD, 0, &altbuf); 964 if (error) 965 goto out; 966 altsb = (struct lfs *)altbuf->b_data; 967 968 /* 969 * Note: this used to do the sanity check only if the 970 * timestamp/serial comparison required use of altsb; 971 * this way is less tolerant, but if altsb is corrupted 972 * enough that the magic number, version, and blocksize 973 * are bogus, why would the timestamp or serial fields 974 * mean anything either? If this kind of thing happens, 975 * you need to fsck anyway. 976 */ 977 978 error = lfs_checkmagic(altsb); 979 if (error) 980 goto out; 981 982 /* Check the basics. */ 983 if (lfs_sb_getbsize(altsb) > MAXBSIZE || 984 lfs_sb_getversion(altsb) > LFS_VERSION || 985 lfs_sb_getbsize(altsb) < sizeof(struct dlfs)) { 986 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 987 " sanity failed\n")); 988 error = EINVAL; /* XXX needs translation */ 989 goto out; 990 } 991 992 if (lfs_sb_getversion(primarysb) == 1) { 993 /* 1s resolution comparison */ 994 if (lfs_sb_gettstamp(altsb) < lfs_sb_gettstamp(primarysb)) 995 thesb = altsb; 996 else 997 thesb = primarysb; 998 } else { 999 /* monotonic infinite-resolution comparison */ 1000 if (lfs_sb_getserial(altsb) < lfs_sb_getserial(primarysb)) 1001 thesb = altsb; 1002 else 1003 thesb = primarysb; 1004 } 1005 } else { 1006 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock location" 1007 " daddr=0x%x\n", lfs_sb_getsboff(primarysb, 1))); 1008 error = EINVAL; 1009 goto out; 1010 } 1011 1012 /* 1013 * Allocate the mount structure, copy the superblock into it. 1014 * Note that the 32-bit and 64-bit superblocks are the same size. 1015 */ 1016 fs = kmem_zalloc(sizeof(struct lfs), KM_SLEEP); 1017 memcpy(&fs->lfs_dlfs_u.u_32, &thesb->lfs_dlfs_u.u_32, 1018 sizeof(struct dlfs)); 1019 fs->lfs_is64 = thesb->lfs_is64; 1020 fs->lfs_dobyteswap = thesb->lfs_dobyteswap; 1021 fs->lfs_hasolddirfmt = false; /* set for real below */ 1022 1023 /* Compatibility */ 1024 if (lfs_sb_getversion(fs) < 2) { 1025 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE); 1026 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs)); 1027 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0)); 1028 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs)); 1029 lfs_sb_setfsbtodb(fs, 0); 1030 } 1031 if (lfs_sb_getresvseg(fs) == 0) 1032 lfs_sb_setresvseg(fs, MIN(lfs_sb_getminfreeseg(fs) - 1, \ 1033 MAX(MIN_RESV_SEGS, lfs_sb_getminfreeseg(fs) / 2 + 1))); 1034 1035 /* 1036 * If we aren't going to be able to write meaningfully to this 1037 * filesystem, and were not mounted readonly, bomb out now. 1038 */ 1039 if (lfs_fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 1040 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 1041 " we need BUFPAGES >= %lld\n", 1042 (long long)((bufmem_hiwater / bufmem_lowater) * 1043 LFS_INVERSE_MAX_BYTES( 1044 lfs_fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 1045 kmem_free(fs, sizeof(struct lfs)); 1046 error = EFBIG; /* XXX needs translation */ 1047 goto out; 1048 } 1049 1050 /* Before rolling forward, lock so vget will sleep for other procs */ 1051 if (l != NULL) { 1052 fs->lfs_flags = LFS_NOTYET; 1053 fs->lfs_rfpid = l->l_proc->p_pid; 1054 } 1055 1056 ump = kmem_zalloc(sizeof(*ump), KM_SLEEP); 1057 ump->um_lfs = fs; 1058 ump->um_fstype = fs->lfs_is64 ? ULFS2 : ULFS1; 1059 /* ump->um_cleaner_thread = NULL; */ 1060 brelse(primarybuf, BC_INVAL); 1061 brelse(altbuf, BC_INVAL); 1062 primarybuf = NULL; 1063 altbuf = NULL; 1064 1065 1066 /* Set up the I/O information */ 1067 fs->lfs_devbsize = DEV_BSIZE; 1068 fs->lfs_iocount = 0; 1069 fs->lfs_diropwait = 0; 1070 fs->lfs_activesb = 0; 1071 lfs_sb_setuinodes(fs, 0); 1072 fs->lfs_ravail = 0; 1073 fs->lfs_favail = 0; 1074 fs->lfs_sbactive = 0; 1075 1076 /* Set up the ifile and lock aflags */ 1077 fs->lfs_doifile = 0; 1078 fs->lfs_writer = 0; 1079 fs->lfs_dirops = 0; 1080 fs->lfs_nadirop = 0; 1081 fs->lfs_seglock = 0; 1082 fs->lfs_pdflush = 0; 1083 fs->lfs_sleepers = 0; 1084 fs->lfs_pages = 0; 1085 rw_init(&fs->lfs_fraglock); 1086 rw_init(&fs->lfs_iflock); 1087 cv_init(&fs->lfs_sleeperscv, "lfs_slp"); 1088 cv_init(&fs->lfs_diropscv, "lfs_dirop"); 1089 cv_init(&fs->lfs_stopcv, "lfsstop"); 1090 cv_init(&fs->lfs_nextsegsleep, "segment"); 1091 1092 /* Set the file system readonly/modify bits. */ 1093 fs->lfs_ronly = ronly; 1094 if (ronly == 0) 1095 fs->lfs_fmod = 1; 1096 1097 /* Device we're using */ 1098 dev = devvp->v_rdev; 1099 fs->lfs_dev = dev; 1100 fs->lfs_devvp = devvp; 1101 1102 /* ulfs-level information */ 1103 fs->um_flags = 0; 1104 fs->um_bptrtodb = lfs_sb_getffshift(fs) - DEV_BSHIFT; 1105 fs->um_seqinc = lfs_sb_getfrag(fs); 1106 fs->um_nindir = lfs_sb_getnindir(fs); 1107 fs->um_lognindir = ffs(lfs_sb_getnindir(fs)) - 1; 1108 fs->um_maxsymlinklen = lfs_sb_getmaxsymlinklen(fs); 1109 fs->um_dirblksiz = LFS_DIRBLKSIZ; 1110 fs->um_maxfilesize = lfs_sb_getmaxfilesize(fs); 1111 1112 /* quota stuff */ 1113 /* XXX: these need to come from the on-disk superblock to be used */ 1114 fs->lfs_use_quota2 = 0; 1115 fs->lfs_quota_magic = 0; 1116 fs->lfs_quota_flags = 0; 1117 fs->lfs_quotaino[0] = 0; 1118 fs->lfs_quotaino[1] = 0; 1119 1120 /* Initialize the mount structure. */ 1121 mp->mnt_data = ump; 1122 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 1123 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 1124 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 1125 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 1126 mp->mnt_stat.f_iosize = lfs_sb_getbsize(fs); 1127 mp->mnt_flag |= MNT_LOCAL; 1128 mp->mnt_iflag |= IMNT_SHRLOOKUP; 1129 mp->mnt_fs_bshift = lfs_sb_getbshift(fs); 1130 mp->mnt_iflag |= IMNT_CAN_RWTORO; 1131 if (fs->um_maxsymlinklen > 0) 1132 mp->mnt_iflag |= IMNT_DTYPE; 1133 else 1134 fs->lfs_hasolddirfmt = true; 1135 1136 ump->um_mountp = mp; 1137 for (i = 0; i < ULFS_MAXQUOTAS; i++) 1138 ump->um_quotas[i] = NULLVP; 1139 spec_node_setmountedfs(devvp, mp); 1140 1141 /* Set up reserved memory for pageout */ 1142 lfs_setup_resblks(fs); 1143 /* Set up vdirop tailq */ 1144 TAILQ_INIT(&fs->lfs_dchainhd); 1145 /* and paging tailq */ 1146 TAILQ_INIT(&fs->lfs_pchainhd); 1147 /* and delayed segment accounting for truncation list */ 1148 LIST_INIT(&fs->lfs_segdhd); 1149 1150 /* 1151 * We use the ifile vnode for almost every operation. Instead of 1152 * retrieving it from the hash table each time we retrieve it here, 1153 * artificially increment the reference count and keep a pointer 1154 * to it in the incore copy of the superblock. 1155 */ 1156 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, LK_EXCLUSIVE, &vp)) != 0) { 1157 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 1158 goto out; 1159 } 1160 fs->lfs_ivnode = vp; 1161 vref(vp); 1162 1163 /* Set up inode bitmap, order free list, and gather orphans. */ 1164 lfs_order_freelist(fs, &orphan, &norphan); 1165 1166 /* Set up segment usage flags for the autocleaner. */ 1167 fs->lfs_nactive = 0; 1168 fs->lfs_suflags = malloc(2 * sizeof(u_int32_t *), 1169 M_SEGMENT, M_WAITOK); 1170 fs->lfs_suflags[0] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t), 1171 M_SEGMENT, M_WAITOK); 1172 fs->lfs_suflags[1] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t), 1173 M_SEGMENT, M_WAITOK); 1174 memset(fs->lfs_suflags[1], 0, lfs_sb_getnseg(fs) * sizeof(u_int32_t)); 1175 for (i = 0; i < lfs_sb_getnseg(fs); i++) { 1176 int changed; 1177 struct buf *bp; 1178 1179 LFS_SEGENTRY(sup, fs, i, bp); 1180 changed = 0; 1181 if (!ronly) { 1182 if (sup->su_nbytes == 0 && 1183 !(sup->su_flags & SEGUSE_EMPTY)) { 1184 sup->su_flags |= SEGUSE_EMPTY; 1185 ++changed; 1186 } else if (!(sup->su_nbytes == 0) && 1187 (sup->su_flags & SEGUSE_EMPTY)) { 1188 sup->su_flags &= ~SEGUSE_EMPTY; 1189 ++changed; 1190 } 1191 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 1192 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 1193 ++changed; 1194 } 1195 } 1196 fs->lfs_suflags[0][i] = sup->su_flags; 1197 if (changed) 1198 LFS_WRITESEGENTRY(sup, fs, i, bp); 1199 else 1200 brelse(bp, 0); 1201 } 1202 1203 /* Free the orphans we discovered while ordering the freelist. */ 1204 lfs_free_orphans(fs, orphan, norphan); 1205 1206 /* 1207 * XXX: if the fs has quotas, quotas should be on even if 1208 * readonly. Otherwise you can't query the quota info! 1209 * However, that's not how the quota2 code got written and I 1210 * don't know if it'll behave itself if enabled while 1211 * readonly, so for now use the same enable logic as ffs. 1212 * 1213 * XXX: also, if you use the -f behavior allowed here (and 1214 * equivalently above for remount) it will corrupt the fs. It 1215 * ought not to allow that. It should allow mounting readonly 1216 * if there are quotas and the kernel doesn't have the quota 1217 * code, but only readonly. 1218 * 1219 * XXX: and if you use the -f behavior allowed here it will 1220 * likely crash at unmount time (or remount time) because we 1221 * think quotas are active. 1222 * 1223 * Although none of this applies until there's a way to set 1224 * lfs_use_quota2 and have quotas in the fs at all. 1225 */ 1226 if (!ronly && fs->lfs_use_quota2) { 1227 #ifdef LFS_QUOTA2 1228 error = lfs_quota2_mount(mp); 1229 #else 1230 uprintf("%s: no kernel support for this filesystem's quotas\n", 1231 mp->mnt_stat.f_mntonname); 1232 if (mp->mnt_flag & MNT_FORCE) { 1233 uprintf("%s: mounting anyway; fsck afterwards\n", 1234 mp->mnt_stat.f_mntonname); 1235 } else { 1236 error = EINVAL; 1237 } 1238 #endif 1239 if (error) { 1240 /* XXX XXX must clean up the stuff immediately above */ 1241 printf("lfs_mountfs: sorry, leaking some memory\n"); 1242 goto out; 1243 } 1244 } 1245 1246 #ifdef LFS_KERNEL_RFW 1247 lfs_roll_forward(fs, mp, l); 1248 #endif 1249 1250 /* If writing, sb is not clean; record in case of immediate crash */ 1251 if (!fs->lfs_ronly) { 1252 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN); 1253 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 1254 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 1255 } 1256 1257 /* Allow vget now that roll-forward is complete */ 1258 fs->lfs_flags &= ~(LFS_NOTYET); 1259 wakeup(&fs->lfs_flags); 1260 1261 /* 1262 * Initialize the ifile cleaner info with information from 1263 * the superblock. 1264 */ 1265 { 1266 struct buf *bp; 1267 1268 LFS_CLEANERINFO(cip, fs, bp); 1269 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs)); 1270 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs)); 1271 lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs)); 1272 lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs)); 1273 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1274 } 1275 1276 /* 1277 * Mark the current segment as ACTIVE, since we're going to 1278 * be writing to it. 1279 */ 1280 { 1281 struct buf *bp; 1282 1283 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); 1284 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1285 fs->lfs_nactive++; 1286 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); /* Ifile */ 1287 } 1288 1289 /* Now that roll-forward is done, unlock the Ifile */ 1290 vput(vp); 1291 1292 /* Start the pagedaemon-anticipating daemon */ 1293 mutex_enter(&lfs_lock); 1294 if (lfs_writer_daemon == NULL && 1295 kthread_create(PRI_BIO, 0, NULL, 1296 lfs_writerd, NULL, NULL, "lfs_writer") != 0) 1297 panic("fork lfs_writer"); 1298 mutex_exit(&lfs_lock); 1299 1300 printf("WARNING: the log-structured file system is experimental\n" 1301 "WARNING: it may cause system crashes and/or corrupt data\n"); 1302 1303 return (0); 1304 1305 out: 1306 if (primarybuf) 1307 brelse(primarybuf, BC_INVAL); 1308 if (altbuf) 1309 brelse(altbuf, BC_INVAL); 1310 if (ump) { 1311 kmem_free(ump->um_lfs, sizeof(struct lfs)); 1312 kmem_free(ump, sizeof(*ump)); 1313 mp->mnt_data = NULL; 1314 } 1315 1316 return (error); 1317 } 1318 1319 /* 1320 * unmount system call 1321 */ 1322 int 1323 lfs_unmount(struct mount *mp, int mntflags) 1324 { 1325 struct ulfsmount *ump; 1326 struct lfs *fs; 1327 int error, ronly; 1328 1329 ump = VFSTOULFS(mp); 1330 fs = ump->um_lfs; 1331 1332 error = lfs_flushfiles(mp, mntflags & MNT_FORCE ? FORCECLOSE : 0); 1333 if (error) 1334 return error; 1335 1336 /* Finish with the Ifile, now that we're done with it */ 1337 vgone(fs->lfs_ivnode); 1338 1339 ronly = !fs->lfs_ronly; 1340 if (fs->lfs_devvp->v_type != VBAD) 1341 spec_node_setmountedfs(fs->lfs_devvp, NULL); 1342 vn_lock(fs->lfs_devvp, LK_EXCLUSIVE | LK_RETRY); 1343 error = VOP_CLOSE(fs->lfs_devvp, 1344 ronly ? FREAD : FREAD|FWRITE, NOCRED); 1345 vput(fs->lfs_devvp); 1346 1347 /* Complain about page leakage */ 1348 if (fs->lfs_pages > 0) 1349 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 1350 fs->lfs_pages, lfs_subsys_pages); 1351 1352 /* Free per-mount data structures */ 1353 free(fs->lfs_ino_bitmap, M_SEGMENT); 1354 free(fs->lfs_suflags[0], M_SEGMENT); 1355 free(fs->lfs_suflags[1], M_SEGMENT); 1356 free(fs->lfs_suflags, M_SEGMENT); 1357 lfs_free_resblks(fs); 1358 cv_destroy(&fs->lfs_sleeperscv); 1359 cv_destroy(&fs->lfs_diropscv); 1360 cv_destroy(&fs->lfs_stopcv); 1361 cv_destroy(&fs->lfs_nextsegsleep); 1362 1363 rw_destroy(&fs->lfs_fraglock); 1364 rw_destroy(&fs->lfs_iflock); 1365 1366 kmem_free(fs, sizeof(struct lfs)); 1367 kmem_free(ump, sizeof(*ump)); 1368 1369 mp->mnt_data = NULL; 1370 mp->mnt_flag &= ~MNT_LOCAL; 1371 return (error); 1372 } 1373 1374 static int 1375 lfs_flushfiles(struct mount *mp, int flags) 1376 { 1377 struct lwp *l = curlwp; 1378 struct ulfsmount *ump; 1379 struct lfs *fs; 1380 struct vnode *vp; 1381 int error; 1382 1383 ump = VFSTOULFS(mp); 1384 fs = ump->um_lfs; 1385 1386 /* Two checkpoints */ 1387 if (!fs->lfs_ronly) { 1388 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1389 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1390 } 1391 1392 /* wake up the cleaner so it can die */ 1393 /* XXX: shouldn't this be *after* the error cases below? */ 1394 lfs_wakeup_cleaner(fs); 1395 mutex_enter(&lfs_lock); 1396 while (fs->lfs_sleepers) 1397 cv_wait(&fs->lfs_sleeperscv, &lfs_lock); 1398 mutex_exit(&lfs_lock); 1399 1400 #ifdef LFS_EXTATTR 1401 if (ump->um_fstype == ULFS1) { 1402 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) { 1403 ulfs_extattr_stop(mp, curlwp); 1404 } 1405 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED) { 1406 ulfs_extattr_uepm_destroy(&ump->um_extattr); 1407 mp->mnt_flag &= ~MNT_EXTATTR; 1408 } 1409 } 1410 #endif 1411 #ifdef LFS_QUOTA 1412 if ((error = lfsquota1_umount(mp, flags)) != 0) 1413 return (error); 1414 #endif 1415 #ifdef LFS_QUOTA2 1416 if ((error = lfsquota2_umount(mp, flags)) != 0) 1417 return (error); 1418 #endif 1419 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 1420 return (error); 1421 if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0) 1422 return (error); 1423 vp = fs->lfs_ivnode; 1424 mutex_enter(vp->v_interlock); 1425 if (LIST_FIRST(&vp->v_dirtyblkhd)) 1426 panic("lfs_unmount: still dirty blocks on ifile vnode"); 1427 mutex_exit(vp->v_interlock); 1428 1429 /* Explicitly write the superblock, to update serial and pflags */ 1430 if (!fs->lfs_ronly) { 1431 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) | LFS_PF_CLEAN); 1432 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 1433 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 1434 } 1435 mutex_enter(&lfs_lock); 1436 while (fs->lfs_iocount) 1437 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 1438 &lfs_lock); 1439 mutex_exit(&lfs_lock); 1440 1441 return 0; 1442 } 1443 1444 /* 1445 * Get file system statistics. 1446 * 1447 * NB: We don't lock to access the superblock here, because it's not 1448 * really that important if we get it wrong. 1449 */ 1450 int 1451 lfs_statvfs(struct mount *mp, struct statvfs *sbp) 1452 { 1453 struct lfs *fs; 1454 struct ulfsmount *ump; 1455 1456 ump = VFSTOULFS(mp); 1457 fs = ump->um_lfs; 1458 1459 sbp->f_bsize = lfs_sb_getbsize(fs); 1460 sbp->f_frsize = lfs_sb_getfsize(fs); 1461 sbp->f_iosize = lfs_sb_getbsize(fs); 1462 sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks; 1463 1464 sbp->f_bfree = LFS_EST_BFREE(fs); 1465 /* 1466 * XXX this should be lfs_sb_getsize (measured in frags) 1467 * rather than dsize (measured in diskblocks). However, 1468 * getsize needs a format version check (for version 1 it 1469 * needs to be blockstofrags'd) so for the moment I'm going to 1470 * leave this... it won't fire wrongly as frags are at least 1471 * as big as diskblocks. 1472 */ 1473 KASSERT(sbp->f_bfree <= lfs_sb_getdsize(fs)); 1474 #if 0 1475 if (sbp->f_bfree < 0) 1476 sbp->f_bfree = 0; 1477 #endif 1478 1479 sbp->f_bresvd = LFS_EST_RSVD(fs); 1480 if (sbp->f_bfree > sbp->f_bresvd) 1481 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1482 else 1483 sbp->f_bavail = 0; 1484 1485 /* XXX: huh? - dholland 20150728 */ 1486 sbp->f_files = lfs_sb_getbfree(fs) / lfs_btofsb(fs, lfs_sb_getibsize(fs)) 1487 * LFS_INOPB(fs); 1488 sbp->f_ffree = sbp->f_files - lfs_sb_getnfiles(fs); 1489 sbp->f_favail = sbp->f_ffree; 1490 sbp->f_fresvd = 0; 1491 copy_statvfs_info(sbp, mp); 1492 return (0); 1493 } 1494 1495 /* 1496 * Go through the disk queues to initiate sandbagged IO; 1497 * go through the inodes to write those that have been modified; 1498 * initiate the writing of the super block if it has been modified. 1499 * 1500 * Note: we are always called with the filesystem marked `MPBUSY'. 1501 */ 1502 int 1503 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 1504 { 1505 int error; 1506 struct lfs *fs; 1507 1508 fs = VFSTOULFS(mp)->um_lfs; 1509 if (fs->lfs_ronly) 1510 return 0; 1511 1512 /* Snapshots should not hose the syncer */ 1513 /* 1514 * XXX Sync can block here anyway, since we don't have a very 1515 * XXX good idea of how much data is pending. If it's more 1516 * XXX than a segment and lfs_nextseg is close to the end of 1517 * XXX the log, we'll likely block. 1518 */ 1519 mutex_enter(&lfs_lock); 1520 if (fs->lfs_nowrap && lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs)) { 1521 mutex_exit(&lfs_lock); 1522 return 0; 1523 } 1524 mutex_exit(&lfs_lock); 1525 1526 lfs_writer_enter(fs, "lfs_dirops"); 1527 1528 /* All syncs must be checkpoints until roll-forward is implemented. */ 1529 DLOG((DLOG_FLUSH, "lfs_sync at 0x%jx\n", 1530 (uintmax_t)lfs_sb_getoffset(fs))); 1531 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1532 lfs_writer_leave(fs); 1533 #ifdef LFS_QUOTA 1534 lfs_qsync(mp); 1535 #endif 1536 return (error); 1537 } 1538 1539 /* 1540 * Look up an LFS dinode number to find its incore vnode. If not already 1541 * in core, read it in from the specified device. Return the inode locked. 1542 * Detection and handling of mount points must be done by the calling routine. 1543 */ 1544 int 1545 lfs_vget(struct mount *mp, ino_t ino, int lktype, struct vnode **vpp) 1546 { 1547 int error; 1548 1549 error = vcache_get(mp, &ino, sizeof(ino), vpp); 1550 if (error) 1551 return error; 1552 error = vn_lock(*vpp, lktype); 1553 if (error) { 1554 vrele(*vpp); 1555 *vpp = NULL; 1556 return error; 1557 } 1558 1559 return 0; 1560 } 1561 1562 /* 1563 * Create a new vnode/inode pair and initialize what fields we can. 1564 */ 1565 static void 1566 lfs_init_vnode(struct ulfsmount *ump, ino_t ino, struct vnode *vp) 1567 { 1568 struct lfs *fs = ump->um_lfs; 1569 struct inode *ip; 1570 union lfs_dinode *dp; 1571 1572 ASSERT_NO_SEGLOCK(fs); 1573 1574 /* Initialize the inode. */ 1575 ip = pool_get(&lfs_inode_pool, PR_WAITOK); 1576 memset(ip, 0, sizeof(*ip)); 1577 dp = pool_get(&lfs_dinode_pool, PR_WAITOK); 1578 memset(dp, 0, sizeof(*dp)); 1579 ip->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK); 1580 memset(ip->inode_ext.lfs, 0, sizeof(*ip->inode_ext.lfs)); 1581 ip->i_din = dp; 1582 ip->i_ump = ump; 1583 ip->i_vnode = vp; 1584 ip->i_dev = fs->lfs_dev; 1585 lfs_dino_setinumber(fs, dp, ino); 1586 ip->i_number = ino; 1587 ip->i_lfs = fs; 1588 ip->i_lfs_effnblks = 0; 1589 SPLAY_INIT(&ip->i_lfs_lbtree); 1590 ip->i_lfs_nbtree = 0; 1591 LIST_INIT(&ip->i_lfs_segdhd); 1592 1593 vp->v_tag = VT_LFS; 1594 vp->v_op = lfs_vnodeop_p; 1595 vp->v_vflag |= VV_LOCKSWORK; 1596 vp->v_data = ip; 1597 } 1598 1599 /* 1600 * Undo lfs_init_vnode(). 1601 */ 1602 static void 1603 lfs_deinit_vnode(struct ulfsmount *ump, struct vnode *vp) 1604 { 1605 struct inode *ip = VTOI(vp); 1606 1607 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1608 pool_put(&lfs_dinode_pool, ip->i_din); 1609 pool_put(&lfs_inode_pool, ip); 1610 vp->v_data = NULL; 1611 } 1612 1613 /* 1614 * Read an inode from disk and initialize this vnode / inode pair. 1615 * Caller assures no other thread will try to load this inode. 1616 */ 1617 int 1618 lfs_loadvnode(struct mount *mp, struct vnode *vp, 1619 const void *key, size_t key_len, const void **new_key) 1620 { 1621 struct lfs *fs; 1622 union lfs_dinode *dip; 1623 struct inode *ip; 1624 struct buf *bp; 1625 IFILE *ifp; 1626 struct ulfsmount *ump; 1627 ino_t ino; 1628 daddr_t daddr; 1629 int error, retries; 1630 struct timespec ts; 1631 1632 KASSERT(key_len == sizeof(ino)); 1633 memcpy(&ino, key, key_len); 1634 1635 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1636 1637 ump = VFSTOULFS(mp); 1638 fs = ump->um_lfs; 1639 1640 /* 1641 * If the filesystem is not completely mounted yet, suspend 1642 * any access requests (wait for roll-forward to complete). 1643 */ 1644 mutex_enter(&lfs_lock); 1645 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1646 mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1647 &lfs_lock); 1648 mutex_exit(&lfs_lock); 1649 1650 /* Translate the inode number to a disk address. */ 1651 if (ino == LFS_IFILE_INUM) 1652 daddr = lfs_sb_getidaddr(fs); 1653 else { 1654 /* XXX bounds-check this too */ 1655 LFS_IENTRY(ifp, fs, ino, bp); 1656 daddr = lfs_if_getdaddr(fs, ifp); 1657 if (lfs_sb_getversion(fs) > 1) { 1658 ts.tv_sec = lfs_if_getatime_sec(fs, ifp); 1659 ts.tv_nsec = lfs_if_getatime_nsec(fs, ifp); 1660 } 1661 1662 brelse(bp, 0); 1663 if (daddr == LFS_UNUSED_DADDR) 1664 return (ENOENT); 1665 } 1666 1667 /* Allocate/init new vnode/inode. */ 1668 lfs_init_vnode(ump, ino, vp); 1669 ip = VTOI(vp); 1670 1671 /* If the cleaner supplied the inode, use it. */ 1672 if (curlwp == fs->lfs_cleaner_thread && fs->lfs_cleaner_hint != NULL && 1673 fs->lfs_cleaner_hint->bi_lbn == LFS_UNUSED_LBN) { 1674 dip = fs->lfs_cleaner_hint->bi_bp; 1675 if (fs->lfs_is64) { 1676 error = copyin(dip, &ip->i_din->u_64, 1677 sizeof(struct lfs64_dinode)); 1678 } else { 1679 error = copyin(dip, &ip->i_din->u_32, 1680 sizeof(struct lfs32_dinode)); 1681 } 1682 if (error) { 1683 lfs_deinit_vnode(ump, vp); 1684 return error; 1685 } 1686 KASSERT(ip->i_number == ino); 1687 goto out; 1688 } 1689 1690 /* Read in the disk contents for the inode, copy into the inode. */ 1691 retries = 0; 1692 again: 1693 error = bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), 1694 (lfs_sb_getversion(fs) == 1 ? lfs_sb_getbsize(fs) : lfs_sb_getibsize(fs)), 1695 0, &bp); 1696 if (error) { 1697 lfs_deinit_vnode(ump, vp); 1698 return error; 1699 } 1700 1701 dip = lfs_ifind(fs, ino, bp); 1702 if (dip == NULL) { 1703 /* Assume write has not completed yet; try again */ 1704 brelse(bp, BC_INVAL); 1705 ++retries; 1706 if (retries <= LFS_IFIND_RETRIES) { 1707 mutex_enter(&lfs_lock); 1708 if (fs->lfs_iocount) { 1709 DLOG((DLOG_VNODE, 1710 "%s: dinode %d not found, retrying...\n", 1711 __func__, ino)); 1712 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, 1713 "lfs ifind", 1, &lfs_lock); 1714 } else 1715 retries = LFS_IFIND_RETRIES; 1716 mutex_exit(&lfs_lock); 1717 goto again; 1718 } 1719 #ifdef DEBUG 1720 /* If the seglock is held look at the bpp to see 1721 what is there anyway */ 1722 mutex_enter(&lfs_lock); 1723 if (fs->lfs_seglock > 0) { 1724 struct buf **bpp; 1725 union lfs_dinode *dp; 1726 int i; 1727 1728 for (bpp = fs->lfs_sp->bpp; 1729 bpp != fs->lfs_sp->cbpp; ++bpp) { 1730 if ((*bpp)->b_vp == fs->lfs_ivnode && 1731 bpp != fs->lfs_sp->bpp) { 1732 /* Inode block */ 1733 printf("%s: block 0x%" PRIx64 ": ", 1734 __func__, (*bpp)->b_blkno); 1735 for (i = 0; i < LFS_INOPB(fs); i++) { 1736 dp = DINO_IN_BLOCK(fs, 1737 (*bpp)->b_data, i); 1738 if (lfs_dino_getinumber(fs, dp)) 1739 printf("%ju ", 1740 (uintmax_t)lfs_dino_getinumber(fs, dp)); 1741 } 1742 printf("\n"); 1743 } 1744 } 1745 } 1746 mutex_exit(&lfs_lock); 1747 #endif /* DEBUG */ 1748 panic("lfs_loadvnode: dinode not found"); 1749 } 1750 lfs_copy_dinode(fs, ip->i_din, dip); 1751 brelse(bp, 0); 1752 1753 out: 1754 if (lfs_sb_getversion(fs) > 1) { 1755 lfs_dino_setatime(fs, ip->i_din, ts.tv_sec); 1756 lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec); 1757 } 1758 1759 lfs_vinit(mp, &vp); 1760 1761 *new_key = &ip->i_number; 1762 return 0; 1763 } 1764 1765 /* 1766 * Create a new inode and initialize this vnode / inode pair. 1767 */ 1768 int 1769 lfs_newvnode(struct mount *mp, struct vnode *dvp, struct vnode *vp, 1770 struct vattr *vap, kauth_cred_t cred, void *extra, 1771 size_t *key_len, const void **new_key) 1772 { 1773 ino_t ino; 1774 struct inode *ip; 1775 struct ulfsmount *ump; 1776 struct lfs *fs; 1777 int error, mode, gen; 1778 1779 KASSERT(dvp != NULL || vap->va_fileid > 0); 1780 KASSERT(dvp != NULL && dvp->v_mount == mp); 1781 KASSERT(vap->va_type != VNON); 1782 1783 *key_len = sizeof(ino); 1784 ump = VFSTOULFS(mp); 1785 fs = ump->um_lfs; 1786 mode = MAKEIMODE(vap->va_type, vap->va_mode); 1787 1788 /* 1789 * Allocate fresh inode. With "dvp == NULL" take the inode number 1790 * and version from "vap". 1791 */ 1792 if (dvp == NULL) { 1793 ino = vap->va_fileid; 1794 gen = vap->va_gen; 1795 error = lfs_valloc_fixed(fs, ino, gen); 1796 } else { 1797 error = lfs_valloc(dvp, mode, cred, &ino, &gen); 1798 } 1799 if (error) 1800 return error; 1801 1802 /* Attach inode to vnode. */ 1803 lfs_init_vnode(ump, ino, vp); 1804 ip = VTOI(vp); 1805 1806 mutex_enter(&lfs_lock); 1807 LFS_SET_UINO(ip, IN_CHANGE); 1808 mutex_exit(&lfs_lock); 1809 1810 /* Note no blocks yet */ 1811 ip->i_lfs_hiblk = -1; 1812 1813 /* Set a new generation number for this inode. */ 1814 ip->i_gen = gen; 1815 lfs_dino_setgen(fs, ip->i_din, gen); 1816 1817 memset(ip->i_lfs_fragsize, 0, 1818 ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 1819 1820 /* Set uid / gid. */ 1821 if (cred == NOCRED || cred == FSCRED) { 1822 ip->i_gid = 0; 1823 ip->i_uid = 0; 1824 } else { 1825 ip->i_gid = VTOI(dvp)->i_gid; 1826 ip->i_uid = kauth_cred_geteuid(cred); 1827 } 1828 DIP_ASSIGN(ip, gid, ip->i_gid); 1829 DIP_ASSIGN(ip, uid, ip->i_uid); 1830 1831 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 1832 error = lfs_chkiq(ip, 1, cred, 0); 1833 if (error) { 1834 lfs_vfree(dvp, ino, mode); 1835 lfs_deinit_vnode(ump, vp); 1836 1837 return error; 1838 } 1839 #endif 1840 1841 /* Set type and finalize. */ 1842 ip->i_flags = 0; 1843 DIP_ASSIGN(ip, flags, 0); 1844 ip->i_mode = mode; 1845 DIP_ASSIGN(ip, mode, mode); 1846 if (vap->va_rdev != VNOVAL) { 1847 /* 1848 * Want to be able to use this to make badblock 1849 * inodes, so don't truncate the dev number. 1850 */ 1851 // XXX clean this up 1852 if (ump->um_fstype == ULFS1) 1853 ip->i_din->u_32.di_rdev = ulfs_rw32(vap->va_rdev, 1854 ULFS_MPNEEDSWAP(fs)); 1855 else 1856 ip->i_din->u_64.di_rdev = ulfs_rw64(vap->va_rdev, 1857 ULFS_MPNEEDSWAP(fs)); 1858 } 1859 lfs_vinit(mp, &vp); 1860 1861 *new_key = &ip->i_number; 1862 return 0; 1863 } 1864 1865 /* 1866 * File handle to vnode 1867 */ 1868 int 1869 lfs_fhtovp(struct mount *mp, struct fid *fhp, int lktype, struct vnode **vpp) 1870 { 1871 struct lfid lfh; 1872 struct lfs *fs; 1873 1874 if (fhp->fid_len != sizeof(struct lfid)) 1875 return EINVAL; 1876 1877 memcpy(&lfh, fhp, sizeof(lfh)); 1878 if (lfh.lfid_ino < LFS_IFILE_INUM) 1879 return ESTALE; 1880 1881 fs = VFSTOULFS(mp)->um_lfs; 1882 if (lfh.lfid_ident != lfs_sb_getident(fs)) 1883 return ESTALE; 1884 1885 if (lfh.lfid_ino > 1886 ((lfs_dino_getsize(fs, VTOI(fs->lfs_ivnode)->i_din) >> lfs_sb_getbshift(fs)) - 1887 lfs_sb_getcleansz(fs) - lfs_sb_getsegtabsz(fs)) * lfs_sb_getifpb(fs)) 1888 return ESTALE; 1889 1890 return (ulfs_fhtovp(mp, &lfh.lfid_ufid, lktype, vpp)); 1891 } 1892 1893 /* 1894 * Vnode pointer to File handle 1895 */ 1896 /* ARGSUSED */ 1897 int 1898 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 1899 { 1900 struct inode *ip; 1901 struct lfid lfh; 1902 1903 if (*fh_size < sizeof(struct lfid)) { 1904 *fh_size = sizeof(struct lfid); 1905 return E2BIG; 1906 } 1907 *fh_size = sizeof(struct lfid); 1908 ip = VTOI(vp); 1909 memset(&lfh, 0, sizeof(lfh)); 1910 lfh.lfid_len = sizeof(struct lfid); 1911 lfh.lfid_ino = ip->i_number; 1912 lfh.lfid_gen = ip->i_gen; 1913 lfh.lfid_ident = lfs_sb_getident(ip->i_lfs); 1914 memcpy(fhp, &lfh, sizeof(lfh)); 1915 return (0); 1916 } 1917 1918 /* 1919 * ulfs_bmaparray callback function for writing. 1920 * 1921 * Since blocks will be written to the new segment anyway, 1922 * we don't care about current daddr of them. 1923 */ 1924 static bool 1925 lfs_issequential_hole(const struct lfs *fs, 1926 daddr_t daddr0, daddr_t daddr1) 1927 { 1928 (void)fs; /* not used */ 1929 1930 KASSERT(daddr0 == UNWRITTEN || 1931 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR(fs))); 1932 KASSERT(daddr1 == UNWRITTEN || 1933 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR(fs))); 1934 1935 /* NOTE: all we want to know here is 'hole or not'. */ 1936 /* NOTE: UNASSIGNED is converted to 0 by ulfs_bmaparray. */ 1937 1938 /* 1939 * treat UNWRITTENs and all resident blocks as 'contiguous' 1940 */ 1941 if (daddr0 != 0 && daddr1 != 0) 1942 return true; 1943 1944 /* 1945 * both are in hole? 1946 */ 1947 if (daddr0 == 0 && daddr1 == 0) 1948 return true; /* all holes are 'contiguous' for us. */ 1949 1950 return false; 1951 } 1952 1953 /* 1954 * lfs_gop_write functions exactly like genfs_gop_write, except that 1955 * (1) it requires the seglock to be held by its caller, and sp->fip 1956 * to be properly initialized (it will return without re-initializing 1957 * sp->fip, and without calling lfs_writeseg). 1958 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1959 * to determine how large a block it can write at once (though it does 1960 * still use VOP_BMAP to find holes in the file); 1961 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1962 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1963 * now have clusters of clusters, ick.) 1964 */ 1965 static int 1966 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, 1967 int flags) 1968 { 1969 int i, error, run, haveeof = 0; 1970 int fs_bshift; 1971 vaddr_t kva; 1972 off_t eof, offset, startoffset = 0; 1973 size_t bytes, iobytes, skipbytes; 1974 bool async = (flags & PGO_SYNCIO) == 0; 1975 daddr_t lbn, blkno; 1976 struct vm_page *pg; 1977 struct buf *mbp, *bp; 1978 struct vnode *devvp = VTOI(vp)->i_devvp; 1979 struct inode *ip = VTOI(vp); 1980 struct lfs *fs = ip->i_lfs; 1981 struct segment *sp = fs->lfs_sp; 1982 SEGSUM *ssp; 1983 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1984 const char * failreason = NULL; 1985 1986 ASSERT_SEGLOCK(fs); 1987 1988 /* The Ifile lives in the buffer cache */ 1989 KASSERT(vp != fs->lfs_ivnode); 1990 1991 /* 1992 * We don't want to fill the disk before the cleaner has a chance 1993 * to make room for us. If we're in danger of doing that, fail 1994 * with EAGAIN. The caller will have to notice this, unlock 1995 * so the cleaner can run, relock and try again. 1996 * 1997 * We must write everything, however, if our vnode is being 1998 * reclaimed. 1999 */ 2000 mutex_enter(vp->v_interlock); 2001 if (LFS_STARVED_FOR_SEGS(fs) && vdead_check(vp, VDEAD_NOWAIT) == 0) { 2002 mutex_exit(vp->v_interlock); 2003 failreason = "Starved for segs and not flushing vp"; 2004 goto tryagain; 2005 } 2006 mutex_exit(vp->v_interlock); 2007 2008 /* 2009 * Sometimes things slip past the filters in lfs_putpages, 2010 * and the pagedaemon tries to write pages---problem is 2011 * that the pagedaemon never acquires the segment lock. 2012 * 2013 * Alternatively, pages that were clean when we called 2014 * genfs_putpages may have become dirty in the meantime. In this 2015 * case the segment header is not properly set up for blocks 2016 * to be added to it. 2017 * 2018 * Unbusy and unclean the pages, and put them on the ACTIVE 2019 * queue under the hypothesis that they couldn't have got here 2020 * unless they were modified *quite* recently. 2021 * 2022 * XXXUBC that last statement is an oversimplification of course. 2023 */ 2024 if (!LFS_SEGLOCK_HELD(fs)) { 2025 failreason = "Seglock not held"; 2026 goto tryagain; 2027 } 2028 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) { 2029 failreason = "Inode with no_gop_write"; 2030 goto tryagain; 2031 } 2032 if ((pgs[0]->offset & lfs_sb_getbmask(fs)) != 0) { 2033 failreason = "Bad page offset"; 2034 goto tryagain; 2035 } 2036 2037 UVMHIST_LOG(ubchist, "vp %#jx pgs %#jx npages %jd flags 0x%jx", 2038 (uintptr_t)vp, (uintptr_t)pgs, npages, flags); 2039 2040 GOP_SIZE(vp, vp->v_size, &eof, 0); 2041 haveeof = 1; 2042 2043 if (vp->v_type == VREG) 2044 fs_bshift = vp->v_mount->mnt_fs_bshift; 2045 else 2046 fs_bshift = DEV_BSHIFT; 2047 error = 0; 2048 pg = pgs[0]; 2049 startoffset = pg->offset; 2050 KASSERT(eof >= 0); 2051 2052 if (startoffset >= eof) { 2053 failreason = "Offset beyond EOF"; 2054 goto tryagain; 2055 } else 2056 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 2057 skipbytes = 0; 2058 2059 KASSERT(bytes != 0); 2060 2061 /* Swap PG_DELWRI for PG_PAGEOUT */ 2062 for (i = 0; i < npages; i++) { 2063 if (pgs[i]->flags & PG_DELWRI) { 2064 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 2065 pgs[i]->flags &= ~PG_DELWRI; 2066 pgs[i]->flags |= PG_PAGEOUT; 2067 uvm_pageout_start(1); 2068 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 2069 uvm_pagelock(pgs[i]); 2070 uvm_pageunwire(pgs[i]); 2071 uvm_pageunlock(pgs[i]); 2072 rw_exit(vp->v_uobj.vmobjlock); 2073 } 2074 } 2075 2076 /* 2077 * Check to make sure we're starting on a block boundary. 2078 * We'll check later to make sure we always write entire 2079 * blocks (or fragments). 2080 */ 2081 if (startoffset & lfs_sb_getbmask(fs)) 2082 printf("%" PRId64 " & %" PRIu64 " = %" PRId64 "\n", 2083 startoffset, lfs_sb_getbmask(fs), 2084 startoffset & lfs_sb_getbmask(fs)); 2085 KASSERT((startoffset & lfs_sb_getbmask(fs)) == 0); 2086 if (bytes & lfs_sb_getffmask(fs)) { 2087 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 2088 panic("lfs_gop_write: non-integer blocks"); 2089 } 2090 2091 /* 2092 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 2093 * If we would, write what we have and try again. If we don't 2094 * have anything to write, we'll have to sleep. 2095 */ 2096 ssp = (SEGSUM *)sp->segsum; 2097 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2098 (lfs_ss_getnfinfo(fs, ssp) < 1 ? 2099 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 2100 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 2101 #if 0 2102 " with nfinfo=%d at offset 0x%jx\n", 2103 (int)lfs_ss_getnfinfo(fs, ssp), 2104 (uintmax_t)lfs_sb_getoffset(fs))); 2105 #endif 2106 lfs_updatemeta(sp); 2107 lfs_release_finfo(fs); 2108 (void) lfs_writeseg(fs, sp); 2109 2110 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2111 2112 /* 2113 * Having given up all of the pager_map we were holding, 2114 * we can now wait for aiodoned to reclaim it for us 2115 * without fear of deadlock. 2116 */ 2117 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2118 UVMPAGER_MAPIN_WAITOK); 2119 } 2120 2121 mbp = getiobuf(NULL, true); 2122 UVMHIST_LOG(ubchist, "vp %#jx mbp %#jx num now %jd bytes 0x%jx", 2123 (uintptr_t)vp, (uintptr_t)mbp, vp->v_numoutput, bytes); 2124 mbp->b_bufsize = npages << PAGE_SHIFT; 2125 mbp->b_data = (void *)kva; 2126 mbp->b_resid = mbp->b_bcount = bytes; 2127 mbp->b_cflags |= BC_BUSY|BC_AGE; 2128 mbp->b_iodone = uvm_aio_aiodone; 2129 2130 bp = NULL; 2131 for (offset = startoffset; 2132 bytes > 0; 2133 offset += iobytes, bytes -= iobytes) { 2134 lbn = offset >> fs_bshift; 2135 error = ulfs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 2136 lfs_issequential_hole); 2137 if (error) { 2138 UVMHIST_LOG(ubchist, "ulfs_bmaparray() -> %jd", 2139 error,0,0,0); 2140 skipbytes += bytes; 2141 bytes = 0; 2142 break; 2143 } 2144 2145 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 2146 bytes); 2147 if (blkno == (daddr_t)-1) { 2148 skipbytes += iobytes; 2149 continue; 2150 } 2151 2152 /* 2153 * Discover how much we can really pack into this buffer. 2154 */ 2155 /* If no room in the current segment, finish it up */ 2156 if (sp->sum_bytes_left < sizeof(int32_t) || 2157 sp->seg_bytes_left < (1 << lfs_sb_getbshift(fs))) { 2158 int vers; 2159 2160 lfs_updatemeta(sp); 2161 vers = lfs_fi_getversion(fs, sp->fip); 2162 lfs_release_finfo(fs); 2163 (void) lfs_writeseg(fs, sp); 2164 2165 lfs_acquire_finfo(fs, ip->i_number, vers); 2166 } 2167 /* Check both for space in segment and space in segsum */ 2168 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 2169 << fs_bshift); 2170 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 2171 << fs_bshift); 2172 KASSERT(iobytes > 0); 2173 2174 /* if it's really one i/o, don't make a second buf */ 2175 if (offset == startoffset && iobytes == bytes) { 2176 bp = mbp; 2177 /* 2178 * All the LFS output is done by the segwriter. It 2179 * will increment numoutput by one for all the bufs it 2180 * receives. However this buffer needs one extra to 2181 * account for aiodone. 2182 */ 2183 mutex_enter(vp->v_interlock); 2184 vp->v_numoutput++; 2185 mutex_exit(vp->v_interlock); 2186 } else { 2187 bp = getiobuf(NULL, true); 2188 UVMHIST_LOG(ubchist, "vp %#jx bp %#jx num now %jd", 2189 (uintptr_t)vp, (uintptr_t)bp, vp->v_numoutput, 0); 2190 nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes); 2191 /* 2192 * LFS doesn't like async I/O here, dies with 2193 * an assert in lfs_bwrite(). Is that assert 2194 * valid? I retained non-async behaviour when 2195 * converted this to use nestiobuf --pooka 2196 */ 2197 bp->b_flags &= ~B_ASYNC; 2198 } 2199 2200 /* XXX This is silly ... is this necessary? */ 2201 mutex_enter(&bufcache_lock); 2202 mutex_enter(vp->v_interlock); 2203 bgetvp(vp, bp); 2204 mutex_exit(vp->v_interlock); 2205 mutex_exit(&bufcache_lock); 2206 2207 bp->b_lblkno = lfs_lblkno(fs, offset); 2208 bp->b_private = mbp; 2209 if (devvp->v_type == VBLK) { 2210 bp->b_dev = devvp->v_rdev; 2211 } 2212 VOP_BWRITE(bp->b_vp, bp); 2213 while (lfs_gatherblock(sp, bp, NULL)) 2214 continue; 2215 } 2216 2217 nestiobuf_done(mbp, skipbytes, error); 2218 if (skipbytes) { 2219 UVMHIST_LOG(ubchist, "skipbytes %jd", skipbytes, 0,0,0); 2220 } 2221 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 2222 2223 if (!async) { 2224 /* Start a segment write. */ 2225 UVMHIST_LOG(ubchist, "flushing", 0,0,0,0); 2226 mutex_enter(&lfs_lock); 2227 lfs_flush(fs, 0, 1); 2228 mutex_exit(&lfs_lock); 2229 } 2230 2231 if ((sp->seg_flags & SEGM_SINGLE) && lfs_sb_getcurseg(fs) != fs->lfs_startseg) 2232 return EAGAIN; 2233 2234 return (0); 2235 2236 tryagain: 2237 /* 2238 * We can't write the pages, for whatever reason. 2239 * Clean up after ourselves, and make the caller try again. 2240 */ 2241 mutex_enter(vp->v_interlock); 2242 2243 /* Tell why we're here, if we know */ 2244 if (failreason != NULL) { 2245 DLOG((DLOG_PAGE, "lfs_gop_write: %s\n", failreason)); 2246 } 2247 if (haveeof && startoffset >= eof) { 2248 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 2249 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 2250 pgs[0]->offset, eof, npages)); 2251 } 2252 2253 for (i = 0; i < npages; i++) { 2254 pg = pgs[i]; 2255 2256 if (pg->flags & PG_PAGEOUT) 2257 uvm_pageout_done(1); 2258 uvm_pagelock(pg); 2259 if (pg->flags & PG_DELWRI) { 2260 uvm_pageunwire(pg); 2261 } 2262 uvm_pageactivate(pg); 2263 uvm_pageunlock(pg); 2264 pg->flags &= ~(PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 2265 uvm_pagemarkdirty(pg, UVM_PAGE_STATUS_DIRTY); 2266 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 2267 vp, pg->offset)); 2268 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 2269 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 2270 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 2271 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 2272 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 2273 pg->wire_count)); 2274 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 2275 pg->loan_count)); 2276 } 2277 uvm_page_unbusy(pgs, npages); 2278 mutex_exit(vp->v_interlock); 2279 return EAGAIN; 2280 } 2281 2282 /* 2283 * finish vnode/inode initialization. 2284 * used by lfs_vget. 2285 */ 2286 void 2287 lfs_vinit(struct mount *mp, struct vnode **vpp) 2288 { 2289 struct vnode *vp = *vpp; 2290 struct inode *ip = VTOI(vp); 2291 struct ulfsmount *ump = VFSTOULFS(mp); 2292 struct lfs *fs = ump->um_lfs; 2293 int i; 2294 2295 ip->i_mode = lfs_dino_getmode(fs, ip->i_din); 2296 ip->i_nlink = lfs_dino_getnlink(fs, ip->i_din); 2297 ip->i_lfs_osize = ip->i_size = lfs_dino_getsize(fs, ip->i_din); 2298 ip->i_flags = lfs_dino_getflags(fs, ip->i_din); 2299 ip->i_gen = lfs_dino_getgen(fs, ip->i_din); 2300 ip->i_uid = lfs_dino_getuid(fs, ip->i_din); 2301 ip->i_gid = lfs_dino_getgid(fs, ip->i_din); 2302 2303 ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din); 2304 ip->i_lfs_odnlink = lfs_dino_getnlink(fs, ip->i_din); 2305 2306 /* 2307 * Initialize the vnode from the inode, check for aliases. In all 2308 * cases re-init ip, the underlying vnode/inode may have changed. 2309 */ 2310 ulfs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 2311 ip = VTOI(vp); 2312 2313 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 2314 if (vp->v_type != VLNK || ip->i_size >= ip->i_lfs->um_maxsymlinklen) { 2315 #ifdef DEBUG 2316 for (i = (ip->i_size + lfs_sb_getbsize(fs) - 1) >> lfs_sb_getbshift(fs); 2317 i < ULFS_NDADDR; i++) { 2318 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 2319 i == 0) 2320 continue; 2321 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) { 2322 lfs_dump_dinode(fs, ip->i_din); 2323 panic("inconsistent inode (direct)"); 2324 } 2325 } 2326 for ( ; i < ULFS_NDADDR + ULFS_NIADDR; i++) { 2327 if (lfs_dino_getib(fs, ip->i_din, i - ULFS_NDADDR) != 0) { 2328 lfs_dump_dinode(fs, ip->i_din); 2329 panic("inconsistent inode (indirect)"); 2330 } 2331 } 2332 #endif /* DEBUG */ 2333 for (i = 0; i < ULFS_NDADDR; i++) 2334 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) 2335 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i); 2336 } 2337 2338 KASSERTMSG((vp->v_type != VNON), 2339 "lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 2340 (unsigned long long)ip->i_number, 2341 (ip->i_mode & LFS_IFMT) >> 12); 2342 2343 /* 2344 * Finish inode initialization now that aliasing has been resolved. 2345 */ 2346 2347 ip->i_devvp = fs->lfs_devvp; 2348 vref(ip->i_devvp); 2349 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 2350 ulfsquota_init(ip); 2351 #endif 2352 genfs_node_init(vp, &lfs_genfsops); 2353 uvm_vnp_setsize(vp, ip->i_size); 2354 2355 /* Initialize hiblk from file size */ 2356 ip->i_lfs_hiblk = lfs_lblkno(ip->i_lfs, ip->i_size + lfs_sb_getbsize(ip->i_lfs) - 1) - 1; 2357 2358 *vpp = vp; 2359 } 2360 2361 /* 2362 * Resize the filesystem to contain the specified number of segments. 2363 */ 2364 int 2365 lfs_resize_fs(struct lfs *fs, int newnsegs) 2366 { 2367 SEGUSE *sup; 2368 CLEANERINFO *cip; 2369 struct buf *bp, *obp; 2370 daddr_t olast, nlast, ilast, noff, start, end; 2371 struct vnode *ivp; 2372 struct inode *ip; 2373 int error, badnews, inc, oldnsegs; 2374 int sbbytes, csbbytes, gain, cgain; 2375 int i; 2376 2377 /* Only support v2 and up */ 2378 if (lfs_sb_getversion(fs) < 2) 2379 return EOPNOTSUPP; 2380 2381 /* If we're doing nothing, do it fast */ 2382 oldnsegs = lfs_sb_getnseg(fs); 2383 if (newnsegs == oldnsegs) 2384 return 0; 2385 2386 /* We always have to have two superblocks */ 2387 if (newnsegs <= lfs_dtosn(fs, lfs_sb_getsboff(fs, 1))) 2388 /* XXX this error code is rather nonsense */ 2389 return EFBIG; 2390 2391 ivp = fs->lfs_ivnode; 2392 ip = VTOI(ivp); 2393 error = 0; 2394 2395 /* Take the segment lock so no one else calls lfs_newseg() */ 2396 lfs_seglock(fs, SEGM_PROT); 2397 2398 /* 2399 * Make sure the segments we're going to be losing, if any, 2400 * are in fact empty. We hold the seglock, so their status 2401 * cannot change underneath us. Count the superblocks we lose, 2402 * while we're at it. 2403 */ 2404 sbbytes = csbbytes = 0; 2405 cgain = 0; 2406 for (i = newnsegs; i < oldnsegs; i++) { 2407 LFS_SEGENTRY(sup, fs, i, bp); 2408 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 2409 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2410 sbbytes += LFS_SBPAD; 2411 if (!(sup->su_flags & SEGUSE_DIRTY)) { 2412 ++cgain; 2413 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2414 csbbytes += LFS_SBPAD; 2415 } 2416 brelse(bp, 0); 2417 if (badnews) { 2418 error = EBUSY; 2419 goto out; 2420 } 2421 } 2422 2423 /* Note old and new segment table endpoints, and old ifile size */ 2424 olast = lfs_sb_getcleansz(fs) + lfs_sb_getsegtabsz(fs); 2425 nlast = howmany(newnsegs, lfs_sb_getsepb(fs)) + lfs_sb_getcleansz(fs); 2426 ilast = ivp->v_size >> lfs_sb_getbshift(fs); 2427 noff = nlast - olast; 2428 2429 /* 2430 * Make sure no one can use the Ifile while we change it around. 2431 * Even after taking the iflock we need to make sure no one still 2432 * is holding Ifile buffers, so we get each one, to drain them. 2433 * (XXX this could be done better.) 2434 */ 2435 rw_enter(&fs->lfs_iflock, RW_WRITER); 2436 for (i = 0; i < ilast; i++) { 2437 /* XXX what to do if bread fails? */ 2438 bread(ivp, i, lfs_sb_getbsize(fs), 0, &bp); 2439 brelse(bp, 0); 2440 } 2441 2442 /* Allocate new Ifile blocks */ 2443 for (i = ilast; i < ilast + noff; i++) { 2444 if (lfs_balloc(ivp, i * lfs_sb_getbsize(fs), lfs_sb_getbsize(fs), NOCRED, 0, 2445 &bp) != 0) 2446 panic("balloc extending ifile"); 2447 memset(bp->b_data, 0, lfs_sb_getbsize(fs)); 2448 VOP_BWRITE(bp->b_vp, bp); 2449 } 2450 2451 /* Register new ifile size */ 2452 ip->i_size += noff * lfs_sb_getbsize(fs); 2453 lfs_dino_setsize(fs, ip->i_din, ip->i_size); 2454 uvm_vnp_setsize(ivp, ip->i_size); 2455 2456 /* Copy the inode table to its new position */ 2457 if (noff != 0) { 2458 if (noff < 0) { 2459 start = nlast; 2460 end = ilast + noff; 2461 inc = 1; 2462 } else { 2463 start = ilast + noff - 1; 2464 end = nlast - 1; 2465 inc = -1; 2466 } 2467 for (i = start; i != end; i += inc) { 2468 if (bread(ivp, i, lfs_sb_getbsize(fs), 2469 B_MODIFY, &bp) != 0) 2470 panic("resize: bread dst blk failed"); 2471 if (bread(ivp, i - noff, lfs_sb_getbsize(fs), 2472 0, &obp)) 2473 panic("resize: bread src blk failed"); 2474 memcpy(bp->b_data, obp->b_data, lfs_sb_getbsize(fs)); 2475 VOP_BWRITE(bp->b_vp, bp); 2476 brelse(obp, 0); 2477 } 2478 } 2479 2480 /* If we are expanding, write the new empty SEGUSE entries */ 2481 if (newnsegs > oldnsegs) { 2482 for (i = oldnsegs; i < newnsegs; i++) { 2483 if ((error = bread(ivp, i / lfs_sb_getsepb(fs) + 2484 lfs_sb_getcleansz(fs), lfs_sb_getbsize(fs), 2485 B_MODIFY, &bp)) != 0) 2486 panic("lfs: ifile read: %d", error); 2487 while ((i + 1) % lfs_sb_getsepb(fs) && i < newnsegs) { 2488 sup = &((SEGUSE *)bp->b_data)[i % lfs_sb_getsepb(fs)]; 2489 memset(sup, 0, sizeof(*sup)); 2490 i++; 2491 } 2492 VOP_BWRITE(bp->b_vp, bp); 2493 } 2494 } 2495 2496 /* Zero out unused superblock offsets */ 2497 for (i = 2; i < LFS_MAXNUMSB; i++) 2498 if (lfs_dtosn(fs, lfs_sb_getsboff(fs, i)) >= newnsegs) 2499 lfs_sb_setsboff(fs, i, 0x0); 2500 2501 /* 2502 * Correct superblock entries that depend on fs size. 2503 * The computations of these are as follows: 2504 * 2505 * size = lfs_segtod(fs, nseg) 2506 * dsize = lfs_segtod(fs, nseg - minfreeseg) - lfs_btofsb(#super * LFS_SBPAD) 2507 * bfree = dsize - lfs_btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2508 * avail = lfs_segtod(fs, nclean) - lfs_btofsb(#clean_super * LFS_SBPAD) 2509 * + (lfs_segtod(fs, 1) - (offset - curseg)) 2510 * - lfs_segtod(fs, minfreeseg - (minfreeseg / 2)) 2511 * 2512 * XXX - we should probably adjust minfreeseg as well. 2513 */ 2514 gain = (newnsegs - oldnsegs); 2515 lfs_sb_setnseg(fs, newnsegs); 2516 lfs_sb_setsegtabsz(fs, nlast - lfs_sb_getcleansz(fs)); 2517 lfs_sb_addsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs))); 2518 lfs_sb_adddsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes)); 2519 lfs_sb_addbfree(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes) 2520 - gain * lfs_btofsb(fs, lfs_sb_getbsize(fs) / 2)); 2521 if (gain > 0) { 2522 lfs_sb_addnclean(fs, gain); 2523 lfs_sb_addavail(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs))); 2524 } else { 2525 lfs_sb_subnclean(fs, cgain); 2526 lfs_sb_subavail(fs, cgain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - 2527 lfs_btofsb(fs, csbbytes)); 2528 } 2529 2530 /* Resize segment flag cache */ 2531 fs->lfs_suflags[0] = realloc(fs->lfs_suflags[0], 2532 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK); 2533 fs->lfs_suflags[1] = realloc(fs->lfs_suflags[1], 2534 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK); 2535 for (i = oldnsegs; i < newnsegs; i++) 2536 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2537 2538 /* Truncate Ifile if necessary */ 2539 if (noff < 0) 2540 lfs_truncate(ivp, ivp->v_size + (noff << lfs_sb_getbshift(fs)), 0, 2541 NOCRED); 2542 2543 /* Update cleaner info so the cleaner can die */ 2544 /* XXX what to do if bread fails? */ 2545 bread(ivp, 0, lfs_sb_getbsize(fs), B_MODIFY, &bp); 2546 cip = bp->b_data; 2547 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs)); 2548 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs)); 2549 VOP_BWRITE(bp->b_vp, bp); 2550 2551 /* Let Ifile accesses proceed */ 2552 rw_exit(&fs->lfs_iflock); 2553 2554 out: 2555 lfs_segunlock(fs); 2556 return error; 2557 } 2558 2559 /* 2560 * Extended attribute dispatch 2561 */ 2562 int 2563 lfs_extattrctl(struct mount *mp, int cmd, struct vnode *vp, 2564 int attrnamespace, const char *attrname) 2565 { 2566 #ifdef LFS_EXTATTR 2567 struct ulfsmount *ump; 2568 2569 ump = VFSTOULFS(mp); 2570 if (ump->um_fstype == ULFS1) { 2571 return ulfs_extattrctl(mp, cmd, vp, attrnamespace, attrname); 2572 } 2573 #endif 2574 return vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname); 2575 } 2576