lfs_vfsops.c revision 1.386
1 /* $NetBSD: lfs_vfsops.c,v 1.386 2025/09/17 04:01:53 perseant 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 1.386 2025/09/17 04:01:53 perseant Exp $"); 65 66 #if defined(_KERNEL_OPT) 67 #include "opt_lfs.h" 68 #include "opt_quota.h" 69 #include "opt_uvmhist.h" 70 #endif 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/namei.h> 75 #include <sys/proc.h> 76 #include <sys/kernel.h> 77 #include <sys/vnode.h> 78 #include <sys/mount.h> 79 #include <sys/kthread.h> 80 #include <sys/buf.h> 81 #include <sys/device.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 <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 #ifdef UVMHIST 107 #include <uvm/uvm.h> 108 #endif 109 #include <uvm/uvm_extern.h> 110 #include <uvm/uvm_object.h> 111 #include <uvm/uvm_page.h> 112 #include <uvm/uvm_stat.h> 113 114 #include <ufs/lfs/lfs.h> 115 #include <ufs/lfs/lfs_accessors.h> 116 #include <ufs/lfs/lfs_kernel.h> 117 #include <ufs/lfs/lfs_extern.h> 118 119 #include <miscfs/genfs/genfs.h> 120 #include <miscfs/genfs/genfs_node.h> 121 122 MODULE(MODULE_CLASS_VFS, lfs, NULL); 123 124 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int); 125 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *); 126 static int lfs_flushfiles(struct mount *, int); 127 128 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc; 129 extern const struct vnodeopv_desc lfs_specop_opv_desc; 130 extern const struct vnodeopv_desc lfs_fifoop_opv_desc; 131 132 extern int locked_queue_rcount; 133 extern long locked_queue_rbytes; 134 135 struct lwp * lfs_writer_daemon = NULL; 136 kcondvar_t lfs_writerd_cv; 137 138 struct workqueue *lfs_cluster_wq = NULL; 139 struct workqueue *lfs_super_wq = NULL; 140 141 int lfs_do_flush = 0; 142 #ifdef LFS_KERNEL_RFW 143 int lfs_do_rfw = 0; 144 #endif 145 146 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = { 147 &lfs_vnodeop_opv_desc, 148 &lfs_specop_opv_desc, 149 &lfs_fifoop_opv_desc, 150 NULL, 151 }; 152 153 struct vfsops lfs_vfsops = { 154 .vfs_name = MOUNT_LFS, 155 .vfs_min_mount_data = sizeof (struct ulfs_args), 156 .vfs_mount = lfs_mount, 157 .vfs_start = ulfs_start, 158 .vfs_unmount = lfs_unmount, 159 .vfs_root = ulfs_root, 160 .vfs_quotactl = ulfs_quotactl, 161 .vfs_statvfs = lfs_statvfs, 162 .vfs_sync = lfs_sync, 163 .vfs_vget = lfs_vget, 164 .vfs_loadvnode = lfs_loadvnode, 165 .vfs_newvnode = lfs_newvnode, 166 .vfs_fhtovp = lfs_fhtovp, 167 .vfs_vptofh = lfs_vptofh, 168 .vfs_init = lfs_init, 169 .vfs_reinit = lfs_reinit, 170 .vfs_done = lfs_done, 171 .vfs_mountroot = lfs_mountroot, 172 .vfs_snapshot = (void *)eopnotsupp, 173 .vfs_extattrctl = lfs_extattrctl, 174 .vfs_suspendctl = genfs_suspendctl, 175 .vfs_renamelock_enter = genfs_renamelock_enter, 176 .vfs_renamelock_exit = genfs_renamelock_exit, 177 .vfs_fsync = (void *)eopnotsupp, 178 .vfs_opv_descs = lfs_vnodeopv_descs 179 }; 180 181 const struct genfs_ops lfs_genfsops = { 182 .gop_size = lfs_gop_size, 183 .gop_alloc = ulfs_gop_alloc, 184 .gop_write = lfs_gop_write, 185 .gop_markupdate = ulfs_gop_markupdate, 186 .gop_putrange = genfs_gop_putrange, 187 }; 188 189 struct shortlong { 190 const char *sname; 191 const char *lname; 192 }; 193 194 static int 195 sysctl_lfs_dostats(SYSCTLFN_ARGS) 196 { 197 extern struct lfs_stats lfs_stats; 198 extern int lfs_dostats; 199 int error; 200 201 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 202 if (error || newp == NULL) 203 return (error); 204 205 if (lfs_dostats == 0) 206 memset(&lfs_stats, 0, sizeof(lfs_stats)); 207 208 return (0); 209 } 210 211 SYSCTL_SETUP(lfs_sysctl_setup, "lfs sysctl") 212 { 213 int i; 214 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 215 lfs_fs_pagetrip, lfs_ignore_lazy_sync; 216 #ifdef DEBUG 217 extern int lfs_debug_log_subsys[DLOG_MAX]; 218 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 219 { "rollforward", "Debug roll-forward code" }, 220 { "alloc", "Debug inode allocation and free list" }, 221 { "avail", "Debug space-available-now accounting" }, 222 { "flush", "Debug flush triggers" }, 223 { "lockedlist", "Debug locked list accounting" }, 224 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 225 { "vnode", "Debug vnode use during segment write" }, 226 { "segment", "Debug segment writing" }, 227 { "seguse", "Debug segment used-bytes accounting" }, 228 { "cleaner", "Debug cleaning routines" }, 229 { "mount", "Debug mount/unmount routines" }, 230 { "pagecache", "Debug UBC interactions" }, 231 { "dirop", "Debug directory-operation accounting" }, 232 { "malloc", "Debug private malloc accounting" }, 233 }; 234 #endif /* DEBUG */ 235 struct shortlong stat_names[] = { /* Must match lfs.h! */ 236 { "segsused", "Number of new segments allocated" }, 237 { "psegwrites", "Number of partial-segment writes" }, 238 { "psyncwrites", "Number of synchronous partial-segment" 239 " writes" }, 240 { "pcleanwrites", "Number of partial-segment writes by the" 241 " cleaner" }, 242 { "blocktot", "Number of blocks written" }, 243 { "cleanblocks", "Number of blocks written by the cleaner" }, 244 { "ncheckpoints", "Number of checkpoints made" }, 245 { "nwrites", "Number of whole writes" }, 246 { "nsync_writes", "Number of synchronous writes" }, 247 { "wait_exceeded", "Number of times writer waited for" 248 " cleaner" }, 249 { "write_exceeded", "Number of times writer invoked flush" }, 250 { "flush_invoked", "Number of times flush was invoked" }, 251 { "vflush_invoked", "Number of time vflush was called" }, 252 { "clean_inlocked", "Number of vnodes skipped for being dead" }, 253 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 254 { "segs_reclaimed", "Number of segments reclaimed" }, 255 }; 256 257 sysctl_createv(clog, 0, NULL, NULL, 258 CTLFLAG_PERMANENT, 259 CTLTYPE_NODE, "lfs", 260 SYSCTL_DESCR("Log-structured file system"), 261 NULL, 0, NULL, 0, 262 CTL_VFS, 5, CTL_EOL); 263 /* 264 * XXX the "5" above could be dynamic, thereby eliminating one 265 * more instance of the "number to vfs" mapping problem, but 266 * "5" is the order as taken from sys/mount.h 267 */ 268 269 sysctl_createv(clog, 0, NULL, NULL, 270 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 271 CTLTYPE_INT, "flushindir", NULL, 272 NULL, 0, &lfs_writeindir, 0, 273 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 274 sysctl_createv(clog, 0, NULL, NULL, 275 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 276 CTLTYPE_INT, "clean_vnhead", NULL, 277 NULL, 0, &lfs_clean_vnhead, 0, 278 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 279 sysctl_createv(clog, 0, NULL, NULL, 280 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 281 CTLTYPE_INT, "dostats", 282 SYSCTL_DESCR("Maintain statistics on LFS operations"), 283 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 284 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 285 sysctl_createv(clog, 0, NULL, NULL, 286 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 287 CTLTYPE_INT, "pagetrip", 288 SYSCTL_DESCR("How many dirty pages in fs triggers" 289 " a flush"), 290 NULL, 0, &lfs_fs_pagetrip, 0, 291 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 292 sysctl_createv(clog, 0, NULL, NULL, 293 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 294 CTLTYPE_INT, "ignore_lazy_sync", 295 SYSCTL_DESCR("Lazy Sync is ignored entirely"), 296 NULL, 0, &lfs_ignore_lazy_sync, 0, 297 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL); 298 #ifdef LFS_KERNEL_RFW 299 sysctl_createv(clog, 0, NULL, NULL, 300 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 301 CTLTYPE_INT, "rfw", 302 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 303 NULL, 0, &lfs_do_rfw, 0, 304 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 305 #endif 306 307 sysctl_createv(clog, 0, NULL, NULL, 308 CTLFLAG_PERMANENT, 309 CTLTYPE_NODE, "stats", 310 SYSCTL_DESCR("Debugging options"), 311 NULL, 0, NULL, 0, 312 CTL_VFS, 5, LFS_STATS, CTL_EOL); 313 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 314 sysctl_createv(clog, 0, NULL, NULL, 315 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 316 CTLTYPE_INT, stat_names[i].sname, 317 SYSCTL_DESCR(stat_names[i].lname), 318 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 319 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 320 } 321 322 #ifdef DEBUG 323 sysctl_createv(clog, 0, NULL, NULL, 324 CTLFLAG_PERMANENT, 325 CTLTYPE_NODE, "debug", 326 SYSCTL_DESCR("Debugging options"), 327 NULL, 0, NULL, 0, 328 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 329 for (i = 0; i < DLOG_MAX; i++) { 330 sysctl_createv(clog, 0, NULL, NULL, 331 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 332 CTLTYPE_INT, dlog_names[i].sname, 333 SYSCTL_DESCR(dlog_names[i].lname), 334 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 335 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 336 } 337 #endif 338 } 339 340 /* old cleaner syscall interface. see VOP_FCNTL() */ 341 static const struct syscall_package lfs_syscalls[] = { 342 { SYS_lfs_bmapv, 0, (sy_call_t *)sys_lfs_bmapv }, 343 { SYS_lfs_markv, 0, (sy_call_t *)sys_lfs_markv }, 344 { SYS___lfs_segwait50, 0, (sy_call_t *)sys___lfs_segwait50 }, 345 { SYS_lfs_segclean, 0, (sy_call_t *)sys_lfs_segclean }, 346 { 0, 0, NULL }, 347 }; 348 349 static int 350 lfs_modcmd(modcmd_t cmd, void *arg) 351 { 352 int error; 353 354 switch (cmd) { 355 case MODULE_CMD_INIT: 356 error = syscall_establish(NULL, lfs_syscalls); 357 if (error) 358 return error; 359 error = vfs_attach(&lfs_vfsops); 360 if (error != 0) { 361 syscall_disestablish(NULL, lfs_syscalls); 362 break; 363 } 364 cv_init(&lfs_allclean_wakeup, "segment"); 365 break; 366 case MODULE_CMD_FINI: 367 error = vfs_detach(&lfs_vfsops); 368 if (error != 0) 369 break; 370 syscall_disestablish(NULL, lfs_syscalls); 371 cv_destroy(&lfs_allclean_wakeup); 372 break; 373 default: 374 error = ENOTTY; 375 break; 376 } 377 378 return (error); 379 } 380 381 /* 382 * XXX Same structure as FFS inodes? Should we share a common pool? 383 */ 384 struct pool lfs_inode_pool; 385 struct pool lfs_dinode_pool; 386 struct pool lfs_inoext_pool; 387 struct pool lfs_lbnentry_pool; 388 389 /* 390 * The writer daemon. UVM keeps track of how many dirty pages we are holding 391 * in lfs_subsys_pages; the daemon flushes the filesystem when this value 392 * crosses the (user-defined) threshold LFS_MAX_PAGES. 393 */ 394 static void 395 lfs_writerd(void *arg) 396 { 397 mount_iterator_t *iter; 398 struct mount *mp; 399 struct lfs *fs; 400 struct vfsops *vfs = NULL; 401 int fsflags; 402 int lfsc; 403 int wrote_something = 0; 404 405 mutex_enter(&lfs_lock); 406 KASSERTMSG(lfs_writer_daemon == NULL, "more than one LFS writer daemon"); 407 lfs_writer_daemon = curlwp; 408 mutex_exit(&lfs_lock); 409 410 /* Take an extra reference to the LFS vfsops. */ 411 vfs = vfs_getopsbyname(MOUNT_LFS); 412 413 mutex_enter(&lfs_lock); 414 for (;;) { 415 KASSERT(mutex_owned(&lfs_lock)); 416 if (wrote_something == 0) 417 cv_timedwait(&lfs_writerd_cv, &lfs_lock, hz/10 + 1); 418 KASSERT(mutex_owned(&lfs_lock)); 419 wrote_something = 0; 420 421 /* 422 * If global state wants a flush, flush everything. 423 */ 424 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS || 425 locked_queue_bytes > LFS_MAX_BYTES || 426 lfs_subsys_pages > LFS_MAX_PAGES) { 427 428 if (lfs_do_flush) { 429 DLOG((DLOG_FLUSH, "lfs_writerd: lfs_do_flush\n")); 430 } 431 if (locked_queue_count > LFS_MAX_BUFS) { 432 DLOG((DLOG_FLUSH, "lfs_writerd: lqc = %d, max %d\n", 433 locked_queue_count, LFS_MAX_BUFS)); 434 } 435 if (locked_queue_bytes > LFS_MAX_BYTES) { 436 DLOG((DLOG_FLUSH, "lfs_writerd: lqb = %ld, max %ld\n", 437 locked_queue_bytes, LFS_MAX_BYTES)); 438 } 439 if (lfs_subsys_pages > LFS_MAX_PAGES) { 440 DLOG((DLOG_FLUSH, "lfs_writerd: lssp = %d, max %d\n", 441 lfs_subsys_pages, LFS_MAX_PAGES)); 442 } 443 444 lfs_flush(NULL, SEGM_WRITERD, 0); 445 lfs_do_flush = 0; 446 KASSERT(mutex_owned(&lfs_lock)); 447 continue; 448 } 449 KASSERT(mutex_owned(&lfs_lock)); 450 mutex_exit(&lfs_lock); 451 452 /* 453 * Look through the list of LFSs to see if any of them 454 * have requested pageouts. 455 */ 456 mountlist_iterator_init(&iter); 457 lfsc = 0; 458 while ((mp = mountlist_iterator_next(iter)) != NULL) { 459 KASSERT(!mutex_owned(&lfs_lock)); 460 if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS, 461 sizeof(mp->mnt_stat.f_fstypename)) == 0) { 462 ++lfsc; 463 fs = VFSTOULFS(mp)->um_lfs; 464 daddr_t ooffset = 0; 465 fsflags = SEGM_SINGLE; 466 467 mutex_enter(&lfs_lock); 468 ooffset = lfs_sb_getoffset(fs); 469 470 if (lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs) && fs->lfs_nowrap) { 471 /* Don't try to write if we're suspended */ 472 mutex_exit(&lfs_lock); 473 continue; 474 } 475 if (LFS_STARVED_FOR_SEGS(fs)) { 476 mutex_exit(&lfs_lock); 477 478 DLOG((DLOG_FLUSH, "lfs_writerd: need cleaning before writing possible\n")); 479 lfs_wakeup_cleaner(fs); 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 } 507 if (lfsc == 0) { 508 mutex_enter(&lfs_lock); 509 lfs_writer_daemon = NULL; 510 mutex_exit(&lfs_lock); 511 mountlist_iterator_destroy(iter); 512 break; 513 } 514 mountlist_iterator_destroy(iter); 515 516 mutex_enter(&lfs_lock); 517 } 518 KASSERT(!mutex_owned(&lfs_lock)); 519 520 /* Give up our extra reference so the module can be unloaded. */ 521 mutex_enter(&vfs_list_lock); 522 if (vfs != NULL) 523 vfs->vfs_refcount--; 524 mutex_exit(&vfs_list_lock); 525 526 /* Done! */ 527 kthread_exit(0); 528 } 529 530 /* 531 * Initialize the filesystem, most work done by ulfs_init. 532 */ 533 void 534 lfs_init(void) 535 { 536 537 /* 538 * XXX: should we use separate pools for 32-bit and 64-bit 539 * dinodes? 540 */ 541 malloc_type_attach(M_SEGMENT); 542 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 543 "lfsinopl", &pool_allocator_nointr, IPL_NONE); 544 pool_init(&lfs_dinode_pool, sizeof(union lfs_dinode), 0, 0, 0, 545 "lfsdinopl", &pool_allocator_nointr, IPL_NONE); 546 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 547 "lfsinoextpl", &pool_allocator_nointr, IPL_NONE); 548 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 549 "lfslbnpool", &pool_allocator_nointr, IPL_NONE); 550 ulfs_init(); 551 552 #ifdef DEBUG 553 memset(lfs_log, 0, sizeof(lfs_log)); 554 #endif 555 mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE); 556 cv_init(&lfs_writerd_cv, "lfswrite"); 557 cv_init(&locked_queue_cv, "lfsbuf"); 558 cv_init(&lfs_writing_cv, "lfsflush"); 559 workqueue_create(&lfs_cluster_wq, "lfscwq", lfs_cluster_work, NULL, 560 PRI_BIO, IPL_BIO, WQ_MPSAFE); 561 workqueue_create(&lfs_super_wq, "lfsswq", lfs_super_work, NULL, 562 PRI_BIO, IPL_BIO, WQ_MPSAFE); 563 } 564 565 void 566 lfs_reinit(void) 567 { 568 ulfs_reinit(); 569 } 570 571 void 572 lfs_done(void) 573 { 574 ulfs_done(); 575 mutex_destroy(&lfs_lock); 576 cv_destroy(&lfs_writerd_cv); 577 cv_destroy(&locked_queue_cv); 578 cv_destroy(&lfs_writing_cv); 579 workqueue_destroy(lfs_cluster_wq); 580 workqueue_destroy(lfs_super_wq); 581 pool_destroy(&lfs_inode_pool); 582 pool_destroy(&lfs_dinode_pool); 583 pool_destroy(&lfs_inoext_pool); 584 pool_destroy(&lfs_lbnentry_pool); 585 malloc_type_detach(M_SEGMENT); 586 } 587 588 /* 589 * Called by main() when ulfs is going to be mounted as root. 590 */ 591 int 592 lfs_mountroot(void) 593 { 594 extern struct vnode *rootvp; 595 struct lfs *fs = NULL; /* LFS */ 596 struct mount *mp; 597 struct lwp *l = curlwp; 598 struct ulfsmount *ump; 599 int error; 600 601 if (device_class(root_device) != DV_DISK) 602 return (ENODEV); 603 604 if (rootdev == NODEV) 605 return (ENODEV); 606 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 607 vrele(rootvp); 608 return (error); 609 } 610 if ((error = lfs_mountfs(rootvp, mp, l))) { 611 vfs_unbusy(mp); 612 vfs_rele(mp); 613 return (error); 614 } 615 mountlist_append(mp); 616 ump = VFSTOULFS(mp); 617 fs = ump->um_lfs; 618 lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname); 619 (void)lfs_statvfs(mp, &mp->mnt_stat); 620 vfs_unbusy(mp); 621 setrootfstime((time_t)lfs_sb_gettstamp(VFSTOULFS(mp)->um_lfs)); 622 return (0); 623 } 624 625 /* 626 * VFS Operations. 627 * 628 * mount system call 629 */ 630 int 631 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 632 { 633 struct lwp *l = curlwp; 634 struct vnode *devvp; 635 struct ulfs_args *args = data; 636 struct ulfsmount *ump = NULL; 637 struct lfs *fs = NULL; /* LFS */ 638 int error = 0, update; 639 mode_t accessmode; 640 641 if (args == NULL) 642 return EINVAL; 643 if (*data_len < sizeof *args) 644 return EINVAL; 645 646 if (mp->mnt_flag & MNT_GETARGS) { 647 ump = VFSTOULFS(mp); 648 if (ump == NULL) 649 return EIO; 650 args->fspec = NULL; 651 *data_len = sizeof *args; 652 return 0; 653 } 654 655 update = mp->mnt_flag & MNT_UPDATE; 656 657 /* Check arguments */ 658 if (args->fspec != NULL) { 659 /* 660 * Look up the name and verify that it's sane. 661 */ 662 error = namei_simple_user(args->fspec, 663 NSM_FOLLOW_NOEMULROOT, &devvp); 664 if (error != 0) 665 return (error); 666 667 if (!update) { 668 /* 669 * Be sure this is a valid block device 670 */ 671 if (devvp->v_type != VBLK) 672 error = ENOTBLK; 673 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 674 error = ENXIO; 675 } else { 676 /* 677 * Be sure we're still naming the same device 678 * used for our initial mount 679 * 680 * XXX dholland 20151010: if namei gives us a 681 * different vnode for the same device, 682 * wouldn't it be better to use it going 683 * forward rather than ignore it in favor of 684 * the old one? 685 */ 686 ump = VFSTOULFS(mp); 687 fs = ump->um_lfs; 688 if (devvp != fs->lfs_devvp) { 689 if (devvp->v_rdev != fs->lfs_devvp->v_rdev) 690 error = EINVAL; 691 else { 692 vrele(devvp); 693 devvp = fs->lfs_devvp; 694 vref(devvp); 695 } 696 } 697 } 698 } else { 699 if (!update) { 700 /* New mounts must have a filename for the device */ 701 return (EINVAL); 702 } else { 703 /* Use the extant mount */ 704 ump = VFSTOULFS(mp); 705 fs = ump->um_lfs; 706 devvp = fs->lfs_devvp; 707 vref(devvp); 708 } 709 } 710 711 712 /* 713 * If mount by non-root, then verify that user has necessary 714 * permissions on the device. 715 */ 716 if (error == 0) { 717 accessmode = VREAD; 718 if (update ? 719 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 720 (mp->mnt_flag & MNT_RDONLY) == 0) 721 accessmode |= VWRITE; 722 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 723 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT, 724 KAUTH_REQ_SYSTEM_MOUNT_DEVICE, mp, devvp, 725 KAUTH_ARG(accessmode)); 726 VOP_UNLOCK(devvp); 727 } 728 729 if (error) { 730 vrele(devvp); 731 return (error); 732 } 733 734 if (!update) { 735 int flags; 736 737 if (mp->mnt_flag & MNT_RDONLY) 738 flags = FREAD; 739 else 740 flags = FREAD|FWRITE; 741 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 742 error = VOP_OPEN(devvp, flags, FSCRED); 743 VOP_UNLOCK(devvp); 744 if (error) 745 goto fail; 746 error = lfs_mountfs(devvp, mp, l); /* LFS */ 747 if (error) { 748 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 749 (void)VOP_CLOSE(devvp, flags, NOCRED); 750 VOP_UNLOCK(devvp); 751 goto fail; 752 } 753 754 ump = VFSTOULFS(mp); 755 fs = ump->um_lfs; 756 } else { 757 /* 758 * Update the mount. 759 */ 760 761 /* 762 * The initial mount got a reference on this 763 * device, so drop the one obtained via 764 * namei(), above. 765 */ 766 vrele(devvp); 767 768 ump = VFSTOULFS(mp); 769 fs = ump->um_lfs; 770 771 if (!fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDONLY)) { 772 /* 773 * Changing from read/write to read-only. 774 */ 775 int flags = WRITECLOSE; 776 if (mp->mnt_flag & MNT_FORCE) 777 flags |= FORCECLOSE; 778 error = lfs_flushfiles(mp, flags); 779 if (error) 780 return error; 781 fs->lfs_ronly = 1; 782 } else if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 783 /* 784 * Changing from read-only to read/write. 785 * Note in the superblocks that we're writing. 786 */ 787 788 /* XXX: quotas should have been on even if readonly */ 789 if (fs->lfs_use_quota2) { 790 #ifdef LFS_QUOTA2 791 error = lfs_quota2_mount(mp); 792 #else 793 uprintf("%s: no kernel support for this " 794 "filesystem's quotas\n", 795 mp->mnt_stat.f_mntonname); 796 if (mp->mnt_flag & MNT_FORCE) { 797 uprintf("%s: mounting anyway; " 798 "fsck afterwards\n", 799 mp->mnt_stat.f_mntonname); 800 } else { 801 error = EINVAL; 802 } 803 #endif 804 if (error) { 805 return error; 806 } 807 } 808 809 fs->lfs_ronly = 0; 810 if (lfs_sb_getpflags(fs) & LFS_PF_CLEAN) { 811 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN); 812 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 813 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 814 } 815 } 816 817 if (args->fspec == NULL) 818 return 0; 819 } 820 821 error = set_statvfs_info(path, UIO_USERSPACE, args->fspec, 822 UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l); 823 if (error == 0) 824 lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname); 825 return error; 826 827 fail: 828 vrele(devvp); 829 return (error); 830 } 831 832 /* 833 * Helper for mountfs. Note that the fs pointer may be a dummy one 834 * pointing into a superblock buffer. (Which is gross; see below.) 835 */ 836 static int 837 lfs_checkmagic(struct lfs *fs) 838 { 839 switch (fs->lfs_dlfs_u.u_32.dlfs_magic) { 840 case LFS_MAGIC: 841 fs->lfs_is64 = false; 842 fs->lfs_dobyteswap = false; 843 break; 844 case LFS64_MAGIC: 845 fs->lfs_is64 = true; 846 fs->lfs_dobyteswap = false; 847 break; 848 #ifdef LFS_EI 849 case LFS_MAGIC_SWAPPED: 850 fs->lfs_is64 = false; 851 fs->lfs_dobyteswap = true; 852 break; 853 case LFS64_MAGIC_SWAPPED: 854 fs->lfs_is64 = true; 855 fs->lfs_dobyteswap = true; 856 break; 857 #endif 858 default: 859 /* XXX needs translation */ 860 return EINVAL; 861 } 862 return 0; 863 } 864 865 /* 866 * Common code for mount and mountroot 867 * LFS specific 868 */ 869 int 870 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 871 { 872 struct lfs *primarysb, *altsb, *thesb; 873 struct buf *primarybuf, *altbuf; 874 struct lfs *fs; 875 struct ulfsmount *ump; 876 struct vnode *vp; 877 dev_t dev; 878 int error, i, ronly, fsbsize; 879 kauth_cred_t cred; 880 CLEANERINFO *cip; 881 SEGUSE *sup; 882 daddr_t sb_addr; 883 ino_t *orphan; 884 size_t norphan; 885 886 cred = l ? l->l_cred : NOCRED; 887 888 /* The superblock is supposed to be 512 bytes. */ 889 __CTASSERT(sizeof(struct dlfs) == DEV_BSIZE); 890 891 /* 892 * Flush out any old buffers remaining from a previous use. 893 */ 894 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 895 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 896 VOP_UNLOCK(devvp); 897 if (error) 898 return (error); 899 900 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 901 902 /* Don't free random space on error. */ 903 primarybuf = NULL; 904 altbuf = NULL; 905 ump = NULL; 906 907 sb_addr = LFS_LABELPAD / DEV_BSIZE; 908 while (1) { 909 /* 910 * Read in the superblock. 911 * 912 * Note that because LFS_SBPAD is substantially larger 913 * (8K) than the actual on-disk superblock (512 bytes) 914 * the buffer contains enough space to be used as a 915 * whole struct lfs (in-memory superblock) - we do this 916 * only so we can set and use the is64 and dobyteswap 917 * members. XXX this is gross and the logic here should 918 * be reworked. 919 */ 920 error = bread(devvp, sb_addr, LFS_SBPAD, 0, &primarybuf); 921 if (error) 922 goto out; 923 primarysb = (struct lfs *)primarybuf->b_data; 924 925 /* Check the basics. */ 926 error = lfs_checkmagic(primarysb); 927 if (error) { 928 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock wrong magic\n")); 929 goto out; 930 } 931 if (lfs_sb_getbsize(primarysb) > MAXBSIZE || 932 lfs_sb_getversion(primarysb) > LFS_VERSION || 933 lfs_sb_getbsize(primarysb) < sizeof(struct dlfs)) { 934 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 935 /* XXX needs translation */ 936 error = EINVAL; 937 goto out; 938 } 939 if (lfs_sb_getinodefmt(primarysb) > LFS_MAXINODEFMT) { 940 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 941 lfs_sb_getinodefmt(primarysb))); 942 error = EINVAL; 943 goto out; 944 } 945 946 if (lfs_sb_getversion(primarysb) == 1) 947 fsbsize = DEV_BSIZE; 948 else { 949 fsbsize = 1 << lfs_sb_getffshift(primarysb); 950 /* 951 * Could be, if the frag size is large enough, that we 952 * don't have the "real" primary superblock. If that's 953 * the case, get the real one, and try again. 954 */ 955 if (sb_addr != (lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT))) { 956 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 957 " 0x%llx is not right, trying 0x%llx\n", 958 (long long)sb_addr, 959 (long long)(lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT)))); 960 sb_addr = lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT); 961 brelse(primarybuf, BC_INVAL); 962 continue; 963 } 964 } 965 break; 966 } 967 968 /* 969 * Check the second superblock to see which is newer; then mount 970 * using the older of the two. This is necessary to ensure that 971 * the filesystem is valid if it was not unmounted cleanly. 972 */ 973 974 if (lfs_sb_getsboff(primarysb, 1) && 975 lfs_sb_getsboff(primarysb, 1) - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 976 { 977 error = bread(devvp, lfs_sb_getsboff(primarysb, 1) * (fsbsize / DEV_BSIZE), 978 LFS_SBPAD, 0, &altbuf); 979 if (error) 980 goto out; 981 altsb = (struct lfs *)altbuf->b_data; 982 983 /* 984 * Note: this used to do the sanity check only if the 985 * timestamp/serial comparison required use of altsb; 986 * this way is less tolerant, but if altsb is corrupted 987 * enough that the magic number, version, and blocksize 988 * are bogus, why would the timestamp or serial fields 989 * mean anything either? If this kind of thing happens, 990 * you need to fsck anyway. 991 */ 992 993 error = lfs_checkmagic(altsb); 994 if (error) 995 goto out; 996 997 /* Check the basics. */ 998 if (lfs_sb_getbsize(altsb) > MAXBSIZE || 999 lfs_sb_getversion(altsb) > LFS_VERSION || 1000 lfs_sb_getbsize(altsb) < sizeof(struct dlfs)) { 1001 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 1002 " sanity failed\n")); 1003 error = EINVAL; /* XXX needs translation */ 1004 goto out; 1005 } 1006 1007 if (lfs_sb_getversion(primarysb) == 1) { 1008 /* 1s resolution comparison */ 1009 if (lfs_sb_gettstamp(altsb) < lfs_sb_gettstamp(primarysb)) 1010 thesb = altsb; 1011 else 1012 thesb = primarysb; 1013 } else { 1014 /* monotonic infinite-resolution comparison */ 1015 if (lfs_sb_getserial(altsb) < lfs_sb_getserial(primarysb)) 1016 thesb = altsb; 1017 else 1018 thesb = primarysb; 1019 } 1020 } else { 1021 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock location" 1022 " daddr=0x%x\n", lfs_sb_getsboff(primarysb, 1))); 1023 error = EINVAL; 1024 goto out; 1025 } 1026 1027 /* 1028 * Allocate the mount structure, copy the superblock into it. 1029 * Note that the 32-bit and 64-bit superblocks are the same size. 1030 */ 1031 fs = kmem_zalloc(sizeof(struct lfs), KM_SLEEP); 1032 memcpy(&fs->lfs_dlfs_u.u_32, &thesb->lfs_dlfs_u.u_32, 1033 sizeof(struct dlfs)); 1034 fs->lfs_is64 = thesb->lfs_is64; 1035 fs->lfs_dobyteswap = thesb->lfs_dobyteswap; 1036 fs->lfs_hasolddirfmt = false; /* set for real below */ 1037 1038 /* Compatibility */ 1039 if (lfs_sb_getversion(fs) < 2) { 1040 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE); 1041 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs)); 1042 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0)); 1043 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs)); 1044 lfs_sb_setfsbtodb(fs, 0); 1045 } 1046 if (lfs_sb_getresvseg(fs) == 0) 1047 lfs_sb_setresvseg(fs, MIN(lfs_sb_getminfreeseg(fs) - 1, \ 1048 MAX(MIN_RESV_SEGS, lfs_sb_getminfreeseg(fs) / 2 + 1))); 1049 1050 /* 1051 * If we aren't going to be able to write meaningfully to this 1052 * filesystem, and were not mounted readonly, bomb out now. 1053 */ 1054 if (lfs_fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 1055 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 1056 " we need BUFPAGES >= %lld\n", 1057 (long long)((bufmem_hiwater / bufmem_lowater) * 1058 LFS_INVERSE_MAX_BYTES( 1059 lfs_fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 1060 kmem_free(fs, sizeof(struct lfs)); 1061 error = EFBIG; /* XXX needs translation */ 1062 goto out; 1063 } 1064 1065 /* Before rolling forward, lock so vget will sleep for other procs */ 1066 if (l != NULL) { 1067 fs->lfs_flags = LFS_NOTYET; 1068 fs->lfs_rfpid = l->l_proc->p_pid; 1069 } 1070 1071 ump = kmem_zalloc(sizeof(*ump), KM_SLEEP); 1072 ump->um_lfs = fs; 1073 ump->um_fstype = fs->lfs_is64 ? ULFS2 : ULFS1; 1074 /* ump->um_cleaner_thread = NULL; */ 1075 brelse(primarybuf, BC_INVAL); 1076 brelse(altbuf, BC_INVAL); 1077 primarybuf = NULL; 1078 altbuf = NULL; 1079 1080 1081 /* Set up the I/O information */ 1082 fs->lfs_devbsize = DEV_BSIZE; 1083 fs->lfs_iocount = 0; 1084 fs->lfs_diropwait = 0; 1085 fs->lfs_activesb = 0; 1086 lfs_sb_setuinodes(fs, 0); 1087 fs->lfs_ravail = 0; 1088 fs->lfs_favail = 0; 1089 fs->lfs_sbactive = 0; 1090 1091 /* Set up the ifile and lock aflags */ 1092 fs->lfs_doifile = 0; 1093 fs->lfs_writer = 0; 1094 fs->lfs_dirops = 0; 1095 fs->lfs_nadirop = 0; 1096 fs->lfs_seglock = 0; 1097 fs->lfs_pdflush = 0; 1098 fs->lfs_sleepers = 0; 1099 fs->lfs_pages = 0; 1100 rw_init(&fs->lfs_fraglock); 1101 rw_init(&fs->lfs_iflock); 1102 cv_init(&fs->lfs_sleeperscv, "lfs_slp"); 1103 cv_init(&fs->lfs_diropscv, "lfs_dirop"); 1104 cv_init(&fs->lfs_stopcv, "lfsstop"); 1105 cv_init(&fs->lfs_nextsegsleep, "segment"); 1106 1107 /* Set the file system readonly/modify bits. */ 1108 fs->lfs_ronly = ronly; 1109 if (ronly == 0) 1110 fs->lfs_fmod = 1; 1111 1112 /* Device we're using */ 1113 dev = devvp->v_rdev; 1114 fs->lfs_dev = dev; 1115 fs->lfs_devvp = devvp; 1116 1117 /* ulfs-level information */ 1118 fs->um_flags = 0; 1119 fs->um_bptrtodb = lfs_sb_getffshift(fs) - DEV_BSHIFT; 1120 fs->um_seqinc = lfs_sb_getfrag(fs); 1121 fs->um_nindir = lfs_sb_getnindir(fs); 1122 fs->um_lognindir = ffs(lfs_sb_getnindir(fs)) - 1; 1123 fs->um_maxsymlinklen = lfs_sb_getmaxsymlinklen(fs); 1124 fs->um_dirblksiz = LFS_DIRBLKSIZ; 1125 fs->um_maxfilesize = lfs_sb_getmaxfilesize(fs); 1126 1127 /* quota stuff */ 1128 /* XXX: these need to come from the on-disk superblock to be used */ 1129 fs->lfs_use_quota2 = 0; 1130 fs->lfs_quota_magic = 0; 1131 fs->lfs_quota_flags = 0; 1132 fs->lfs_quotaino[0] = 0; 1133 fs->lfs_quotaino[1] = 0; 1134 1135 /* Initialize the mount structure. */ 1136 mp->mnt_data = ump; 1137 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 1138 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 1139 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 1140 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 1141 mp->mnt_stat.f_iosize = lfs_sb_getbsize(fs); 1142 mp->mnt_flag |= MNT_LOCAL; 1143 mp->mnt_iflag |= IMNT_SHRLOOKUP; 1144 mp->mnt_fs_bshift = lfs_sb_getbshift(fs); 1145 mp->mnt_iflag |= IMNT_CAN_RWTORO; 1146 if (fs->um_maxsymlinklen > 0) 1147 mp->mnt_iflag |= IMNT_DTYPE; 1148 else 1149 fs->lfs_hasolddirfmt = true; 1150 1151 ump->um_mountp = mp; 1152 for (i = 0; i < ULFS_MAXQUOTAS; i++) 1153 ump->um_quotas[i] = NULLVP; 1154 spec_node_setmountedfs(devvp, mp); 1155 1156 /* Set up reserved memory for pageout */ 1157 lfs_setup_resblks(fs); 1158 /* Set up vdirop tailq */ 1159 TAILQ_INIT(&fs->lfs_dchainhd); 1160 /* and paging tailq */ 1161 TAILQ_INIT(&fs->lfs_pchainhd); 1162 /* and delayed segment accounting for truncation list */ 1163 LIST_INIT(&fs->lfs_segdhd); 1164 1165 /* 1166 * We use the ifile vnode for almost every operation. Instead of 1167 * retrieving it from the hash table each time we retrieve it here, 1168 * artificially increment the reference count and keep a pointer 1169 * to it in the incore copy of the superblock. 1170 */ 1171 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, LK_EXCLUSIVE, &vp)) != 0) { 1172 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 1173 goto out; 1174 } 1175 fs->lfs_ivnode = vp; 1176 vref(vp); 1177 1178 /* Set up inode bitmap, order free list, and gather orphans. */ 1179 lfs_order_freelist(fs, &orphan, &norphan); 1180 1181 /* Set up segment usage flags for the autocleaner. */ 1182 fs->lfs_nactive = 0; 1183 fs->lfs_suflags = malloc(2 * sizeof(u_int32_t *), 1184 M_SEGMENT, M_WAITOK); 1185 fs->lfs_suflags[0] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t), 1186 M_SEGMENT, M_WAITOK); 1187 fs->lfs_suflags[1] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t), 1188 M_SEGMENT, M_WAITOK); 1189 memset(fs->lfs_suflags[1], 0, lfs_sb_getnseg(fs) * sizeof(u_int32_t)); 1190 for (i = 0; i < lfs_sb_getnseg(fs); i++) { 1191 int changed; 1192 struct buf *bp; 1193 1194 LFS_SEGENTRY(sup, fs, i, bp); 1195 changed = 0; 1196 if (!ronly) { 1197 if (sup->su_nbytes == 0 && 1198 !(sup->su_flags & SEGUSE_EMPTY)) { 1199 sup->su_flags |= SEGUSE_EMPTY; 1200 ++changed; 1201 } else if (!(sup->su_nbytes == 0) && 1202 (sup->su_flags & SEGUSE_EMPTY)) { 1203 sup->su_flags &= ~SEGUSE_EMPTY; 1204 ++changed; 1205 } 1206 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 1207 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 1208 ++changed; 1209 } 1210 } 1211 fs->lfs_suflags[0][i] = sup->su_flags; 1212 if (changed) 1213 LFS_WRITESEGENTRY(sup, fs, i, bp); 1214 else 1215 brelse(bp, 0); 1216 } 1217 1218 /* Free the orphans we discovered while ordering the freelist. */ 1219 lfs_free_orphans(fs, orphan, norphan); 1220 1221 /* 1222 * XXX: if the fs has quotas, quotas should be on even if 1223 * readonly. Otherwise you can't query the quota info! 1224 * However, that's not how the quota2 code got written and I 1225 * don't know if it'll behave itself if enabled while 1226 * readonly, so for now use the same enable logic as ffs. 1227 * 1228 * XXX: also, if you use the -f behavior allowed here (and 1229 * equivalently above for remount) it will corrupt the fs. It 1230 * ought not to allow that. It should allow mounting readonly 1231 * if there are quotas and the kernel doesn't have the quota 1232 * code, but only readonly. 1233 * 1234 * XXX: and if you use the -f behavior allowed here it will 1235 * likely crash at unmount time (or remount time) because we 1236 * think quotas are active. 1237 * 1238 * Although none of this applies until there's a way to set 1239 * lfs_use_quota2 and have quotas in the fs at all. 1240 */ 1241 if (!ronly && fs->lfs_use_quota2) { 1242 #ifdef LFS_QUOTA2 1243 error = lfs_quota2_mount(mp); 1244 #else 1245 uprintf("%s: no kernel support for this filesystem's quotas\n", 1246 mp->mnt_stat.f_mntonname); 1247 if (mp->mnt_flag & MNT_FORCE) { 1248 uprintf("%s: mounting anyway; fsck afterwards\n", 1249 mp->mnt_stat.f_mntonname); 1250 } else { 1251 error = EINVAL; 1252 } 1253 #endif 1254 if (error) { 1255 /* XXX XXX must clean up the stuff immediately above */ 1256 printf("lfs_mountfs: sorry, leaking some memory\n"); 1257 goto out; 1258 } 1259 } 1260 1261 #ifdef LFS_KERNEL_RFW 1262 lfs_roll_forward(fs, mp, l); 1263 #endif 1264 1265 /* If writing, sb is not clean; record in case of immediate crash */ 1266 if (!fs->lfs_ronly) { 1267 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN); 1268 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 1269 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 1270 } 1271 1272 /* Allow vget now that roll-forward is complete */ 1273 fs->lfs_flags &= ~(LFS_NOTYET); 1274 wakeup(&fs->lfs_flags); 1275 1276 /* 1277 * Initialize the ifile cleaner info with information from 1278 * the superblock. 1279 */ 1280 { 1281 struct buf *bp; 1282 1283 LFS_CLEANERINFO(cip, fs, bp); 1284 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs)); 1285 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs)); 1286 lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs)); 1287 lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs)); 1288 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1289 } 1290 1291 /* 1292 * Mark the current segment as ACTIVE, since we're going to 1293 * be writing to it. 1294 */ 1295 { 1296 struct buf *bp; 1297 1298 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); 1299 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1300 fs->lfs_nactive++; 1301 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); /* Ifile */ 1302 } 1303 1304 /* Now that roll-forward is done, unlock the Ifile */ 1305 vput(vp); 1306 1307 /* Start the pagedaemon-anticipating daemon */ 1308 mutex_enter(&lfs_lock); 1309 if (lfs_writer_daemon == NULL && 1310 kthread_create(PRI_BIO, 0, NULL, 1311 lfs_writerd, NULL, NULL, "lfs_writer") != 0) 1312 panic("fork lfs_writer"); 1313 mutex_exit(&lfs_lock); 1314 1315 printf("WARNING: the log-structured file system is experimental\n" 1316 "WARNING: it may cause system crashes and/or corrupt data\n"); 1317 1318 return (0); 1319 1320 out: 1321 if (primarybuf) 1322 brelse(primarybuf, BC_INVAL); 1323 if (altbuf) 1324 brelse(altbuf, BC_INVAL); 1325 if (ump) { 1326 kmem_free(ump->um_lfs, sizeof(struct lfs)); 1327 kmem_free(ump, sizeof(*ump)); 1328 mp->mnt_data = NULL; 1329 } 1330 1331 return (error); 1332 } 1333 1334 /* 1335 * unmount system call 1336 */ 1337 int 1338 lfs_unmount(struct mount *mp, int mntflags) 1339 { 1340 struct ulfsmount *ump; 1341 struct lfs *fs; 1342 int error, ronly; 1343 1344 ump = VFSTOULFS(mp); 1345 fs = ump->um_lfs; 1346 1347 error = lfs_flushfiles(mp, mntflags & MNT_FORCE ? FORCECLOSE : 0); 1348 if (error) 1349 return error; 1350 1351 DEBUG_CHECK_FREELIST(fs); 1352 1353 /* Finish with the Ifile, now that we're done with it */ 1354 vgone(fs->lfs_ivnode); 1355 1356 ronly = !fs->lfs_ronly; 1357 if (fs->lfs_devvp->v_type != VBAD) 1358 spec_node_setmountedfs(fs->lfs_devvp, NULL); 1359 vn_lock(fs->lfs_devvp, LK_EXCLUSIVE | LK_RETRY); 1360 error = VOP_CLOSE(fs->lfs_devvp, 1361 ronly ? FREAD : FREAD|FWRITE, NOCRED); 1362 vput(fs->lfs_devvp); 1363 1364 /* Complain about page leakage */ 1365 if (fs->lfs_pages > 0) 1366 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 1367 fs->lfs_pages, lfs_subsys_pages); 1368 1369 /* Free per-mount data structures */ 1370 free(fs->lfs_ino_bitmap, M_SEGMENT); 1371 free(fs->lfs_suflags[0], M_SEGMENT); 1372 free(fs->lfs_suflags[1], M_SEGMENT); 1373 free(fs->lfs_suflags, M_SEGMENT); 1374 lfs_free_resblks(fs); 1375 cv_destroy(&fs->lfs_sleeperscv); 1376 cv_destroy(&fs->lfs_diropscv); 1377 cv_destroy(&fs->lfs_stopcv); 1378 cv_destroy(&fs->lfs_nextsegsleep); 1379 1380 rw_destroy(&fs->lfs_fraglock); 1381 rw_destroy(&fs->lfs_iflock); 1382 1383 kmem_free(fs, sizeof(struct lfs)); 1384 kmem_free(ump, sizeof(*ump)); 1385 1386 mp->mnt_data = NULL; 1387 mp->mnt_flag &= ~MNT_LOCAL; 1388 return (error); 1389 } 1390 1391 static int 1392 lfs_flushfiles(struct mount *mp, int flags) 1393 { 1394 struct lwp *l = curlwp; 1395 struct ulfsmount *ump; 1396 struct lfs *fs; 1397 struct vnode *vp; 1398 int error; 1399 1400 ump = VFSTOULFS(mp); 1401 fs = ump->um_lfs; 1402 1403 /* Two checkpoints */ 1404 if (!fs->lfs_ronly) { 1405 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1406 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1407 } 1408 1409 /* wake up the cleaner so it can die */ 1410 /* XXX: shouldn't this be *after* the error cases below? */ 1411 lfs_wakeup_cleaner(fs); 1412 mutex_enter(&lfs_lock); 1413 while (fs->lfs_sleepers) 1414 cv_wait(&fs->lfs_sleeperscv, &lfs_lock); 1415 mutex_exit(&lfs_lock); 1416 1417 #ifdef LFS_EXTATTR 1418 if (ump->um_fstype == ULFS1) { 1419 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) { 1420 ulfs_extattr_stop(mp, curlwp); 1421 } 1422 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED) { 1423 ulfs_extattr_uepm_destroy(&ump->um_extattr); 1424 mp->mnt_flag &= ~MNT_EXTATTR; 1425 } 1426 } 1427 #endif 1428 #ifdef LFS_QUOTA 1429 if ((error = lfsquota1_umount(mp, flags)) != 0) 1430 return (error); 1431 #endif 1432 #ifdef LFS_QUOTA2 1433 if ((error = lfsquota2_umount(mp, flags)) != 0) 1434 return (error); 1435 #endif 1436 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 1437 return (error); 1438 if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0) 1439 return (error); 1440 vp = fs->lfs_ivnode; 1441 mutex_enter(vp->v_interlock); 1442 if (LIST_FIRST(&vp->v_dirtyblkhd)) 1443 panic("lfs_unmount: still dirty blocks on ifile vnode"); 1444 mutex_exit(vp->v_interlock); 1445 1446 /* Explicitly write the superblock, to update serial and pflags */ 1447 if (!fs->lfs_ronly) { 1448 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) | LFS_PF_CLEAN); 1449 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 1450 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 1451 } 1452 mutex_enter(&lfs_lock); 1453 while (fs->lfs_iocount) 1454 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 1455 &lfs_lock); 1456 mutex_exit(&lfs_lock); 1457 1458 return 0; 1459 } 1460 1461 /* 1462 * Get file system statistics. 1463 * 1464 * NB: We don't lock to access the superblock here, because it's not 1465 * really that important if we get it wrong. 1466 */ 1467 int 1468 lfs_statvfs(struct mount *mp, struct statvfs *sbp) 1469 { 1470 struct lfs *fs; 1471 struct ulfsmount *ump; 1472 1473 ump = VFSTOULFS(mp); 1474 fs = ump->um_lfs; 1475 1476 sbp->f_bsize = lfs_sb_getbsize(fs); 1477 sbp->f_frsize = lfs_sb_getfsize(fs); 1478 sbp->f_iosize = lfs_sb_getbsize(fs); 1479 sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks; 1480 1481 sbp->f_bfree = LFS_EST_BFREE(fs); 1482 /* 1483 * XXX this should be lfs_sb_getsize (measured in frags) 1484 * rather than dsize (measured in diskblocks). However, 1485 * getsize needs a format version check (for version 1 it 1486 * needs to be blockstofrags'd) so for the moment I'm going to 1487 * leave this... it won't fire wrongly as frags are at least 1488 * as big as diskblocks. 1489 */ 1490 KASSERT(sbp->f_bfree <= lfs_sb_getdsize(fs)); 1491 #if 0 1492 if (sbp->f_bfree < 0) 1493 sbp->f_bfree = 0; 1494 #endif 1495 1496 sbp->f_bresvd = LFS_EST_RSVD(fs); 1497 if (sbp->f_bfree > sbp->f_bresvd) 1498 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1499 else 1500 sbp->f_bavail = 0; 1501 1502 /* XXX: huh? - dholland 20150728 */ 1503 sbp->f_files = lfs_sb_getbfree(fs) / lfs_btofsb(fs, lfs_sb_getibsize(fs)) 1504 * LFS_INOPB(fs); 1505 sbp->f_ffree = sbp->f_files - lfs_sb_getnfiles(fs); 1506 sbp->f_favail = sbp->f_ffree; 1507 sbp->f_fresvd = 0; 1508 copy_statvfs_info(sbp, mp); 1509 return (0); 1510 } 1511 1512 /* 1513 * Go through the disk queues to initiate sandbagged IO; 1514 * go through the inodes to write those that have been modified; 1515 * initiate the writing of the super block if it has been modified. 1516 */ 1517 int 1518 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 1519 { 1520 int error; 1521 struct lfs *fs; 1522 1523 fs = VFSTOULFS(mp)->um_lfs; 1524 if (fs->lfs_ronly) 1525 return 0; 1526 1527 /* Snapshots should not hose the syncer */ 1528 /* 1529 * XXX Sync can block here anyway, since we don't have a very 1530 * XXX good idea of how much data is pending. If it's more 1531 * XXX than a segment and lfs_nextseg is close to the end of 1532 * XXX the log, we'll likely block. 1533 */ 1534 mutex_enter(&lfs_lock); 1535 if (fs->lfs_nowrap && lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs)) { 1536 mutex_exit(&lfs_lock); 1537 return 0; 1538 } 1539 mutex_exit(&lfs_lock); 1540 1541 lfs_writer_enter(fs, "lfs_dirops"); 1542 1543 /* All syncs must be checkpoints until roll-forward is implemented. */ 1544 DLOG((DLOG_FLUSH, "lfs_sync at 0x%jx\n", 1545 (uintmax_t)lfs_sb_getoffset(fs))); 1546 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1547 lfs_writer_leave(fs); 1548 #ifdef LFS_QUOTA 1549 lfs_qsync(mp); 1550 #endif 1551 return (error); 1552 } 1553 1554 /* 1555 * Look up an LFS dinode number to find its incore vnode. If not already 1556 * in core, read it in from the specified device. Return the inode locked. 1557 * Detection and handling of mount points must be done by the calling routine. 1558 */ 1559 int 1560 lfs_vget(struct mount *mp, ino_t ino, int lktype, struct vnode **vpp) 1561 { 1562 int error; 1563 1564 error = vcache_get(mp, &ino, sizeof(ino), vpp); 1565 if (error) 1566 return error; 1567 error = vn_lock(*vpp, lktype); 1568 if (error) { 1569 vrele(*vpp); 1570 *vpp = NULL; 1571 return error; 1572 } 1573 1574 return 0; 1575 } 1576 1577 /* 1578 * Create a new vnode/inode pair and initialize what fields we can. 1579 */ 1580 static void 1581 lfs_init_vnode(struct ulfsmount *ump, ino_t ino, struct vnode *vp) 1582 { 1583 struct lfs *fs = ump->um_lfs; 1584 struct inode *ip; 1585 union lfs_dinode *dp; 1586 1587 ASSERT_NO_SEGLOCK(fs); 1588 1589 /* Initialize the inode. */ 1590 ip = pool_get(&lfs_inode_pool, PR_WAITOK); 1591 memset(ip, 0, sizeof(*ip)); 1592 dp = pool_get(&lfs_dinode_pool, PR_WAITOK); 1593 memset(dp, 0, sizeof(*dp)); 1594 ip->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK); 1595 memset(ip->inode_ext.lfs, 0, sizeof(*ip->inode_ext.lfs)); 1596 ip->i_din = dp; 1597 ip->i_ump = ump; 1598 ip->i_vnode = vp; 1599 ip->i_dev = fs->lfs_dev; 1600 lfs_dino_setinumber(fs, dp, ino); 1601 ip->i_number = ino; 1602 ip->i_lfs = fs; 1603 ip->i_lfs_effnblks = 0; 1604 SPLAY_INIT(&ip->i_lfs_lbtree); 1605 ip->i_lfs_nbtree = 0; 1606 LIST_INIT(&ip->i_lfs_segdhd); 1607 1608 vp->v_tag = VT_LFS; 1609 vp->v_op = lfs_vnodeop_p; 1610 vp->v_data = ip; 1611 } 1612 1613 /* 1614 * Undo lfs_init_vnode(). 1615 */ 1616 static void 1617 lfs_deinit_vnode(struct ulfsmount *ump, struct vnode *vp) 1618 { 1619 struct inode *ip = VTOI(vp); 1620 1621 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1622 pool_put(&lfs_dinode_pool, ip->i_din); 1623 pool_put(&lfs_inode_pool, ip); 1624 vp->v_data = NULL; 1625 } 1626 1627 /* 1628 * Read an inode from disk and initialize this vnode / inode pair. 1629 * Caller assures no other thread will try to load this inode. 1630 */ 1631 int 1632 lfs_loadvnode(struct mount *mp, struct vnode *vp, 1633 const void *key, size_t key_len, const void **new_key) 1634 { 1635 struct lfs *fs; 1636 union lfs_dinode *dip; 1637 struct inode *ip; 1638 struct buf *bp; 1639 IFILE *ifp; 1640 struct ulfsmount *ump; 1641 ino_t ino; 1642 daddr_t daddr; 1643 int error, retries; 1644 struct timespec ts; 1645 1646 KASSERT(key_len == sizeof(ino)); 1647 memcpy(&ino, key, key_len); 1648 1649 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1650 1651 ump = VFSTOULFS(mp); 1652 fs = ump->um_lfs; 1653 1654 /* 1655 * If the filesystem is not completely mounted yet, suspend 1656 * any access requests (wait for roll-forward to complete). 1657 */ 1658 mutex_enter(&lfs_lock); 1659 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1660 mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1661 &lfs_lock); 1662 mutex_exit(&lfs_lock); 1663 1664 /* Translate the inode number to a disk address. */ 1665 if (ino == LFS_IFILE_INUM) 1666 daddr = lfs_sb_getidaddr(fs); 1667 else { 1668 /* XXX bounds-check this too */ 1669 LFS_IENTRY(ifp, fs, ino, bp); 1670 daddr = lfs_if_getdaddr(fs, ifp); 1671 if (lfs_sb_getversion(fs) > 1) { 1672 ts.tv_sec = lfs_if_getatime_sec(fs, ifp); 1673 ts.tv_nsec = lfs_if_getatime_nsec(fs, ifp); 1674 } 1675 1676 brelse(bp, 0); 1677 if (DADDR_IS_BAD(daddr)) 1678 return (ENOENT); 1679 } 1680 1681 /* Allocate/init new vnode/inode. */ 1682 lfs_init_vnode(ump, ino, vp); 1683 ip = VTOI(vp); 1684 1685 /* If the cleaner supplied the inode, use it. */ 1686 if (curlwp == fs->lfs_cleaner_thread && fs->lfs_cleaner_hint != NULL && 1687 fs->lfs_cleaner_hint->bi_lbn == LFS_UNUSED_LBN) { 1688 dip = fs->lfs_cleaner_hint->bi_bp; 1689 if (fs->lfs_is64) { 1690 error = copyin(dip, &ip->i_din->u_64, 1691 sizeof(struct lfs64_dinode)); 1692 } else { 1693 error = copyin(dip, &ip->i_din->u_32, 1694 sizeof(struct lfs32_dinode)); 1695 } 1696 if (error) { 1697 lfs_deinit_vnode(ump, vp); 1698 return error; 1699 } 1700 KASSERT(ip->i_number == ino); 1701 goto out; 1702 } 1703 1704 /* Read in the disk contents for the inode, copy into the inode. */ 1705 retries = 0; 1706 again: 1707 error = bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), 1708 (lfs_sb_getversion(fs) == 1 ? lfs_sb_getbsize(fs) : lfs_sb_getibsize(fs)), 1709 0, &bp); 1710 if (error) { 1711 lfs_deinit_vnode(ump, vp); 1712 return error; 1713 } 1714 1715 dip = lfs_ifind(fs, ino, bp); 1716 if (dip == NULL) { 1717 /* Assume write has not completed yet; try again */ 1718 brelse(bp, BC_INVAL); 1719 ++retries; 1720 if (retries <= LFS_IFIND_RETRIES) { 1721 mutex_enter(&lfs_lock); 1722 if (fs->lfs_iocount) { 1723 DLOG((DLOG_VNODE, 1724 "%s: dinode %d not found, retrying...\n", 1725 __func__, ino)); 1726 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, 1727 "lfs ifind", 1, &lfs_lock); 1728 } else 1729 retries = LFS_IFIND_RETRIES; 1730 mutex_exit(&lfs_lock); 1731 goto again; 1732 } 1733 #ifdef DEBUG 1734 /* If the seglock is held look at the bpp to see 1735 what is there anyway */ 1736 mutex_enter(&lfs_lock); 1737 if (fs->lfs_seglock > 0) { 1738 struct buf **bpp; 1739 union lfs_dinode *dp; 1740 int i; 1741 1742 for (bpp = fs->lfs_sp->bpp; 1743 bpp != fs->lfs_sp->cbpp; ++bpp) { 1744 if ((*bpp)->b_vp == fs->lfs_ivnode && 1745 bpp != fs->lfs_sp->bpp) { 1746 /* Inode block */ 1747 printf("%s: block 0x%" PRIx64 ": ", 1748 __func__, (*bpp)->b_blkno); 1749 for (i = 0; i < LFS_INOPB(fs); i++) { 1750 dp = DINO_IN_BLOCK(fs, 1751 (*bpp)->b_data, i); 1752 if (lfs_dino_getinumber(fs, dp)) 1753 printf("%ju ", 1754 (uintmax_t)lfs_dino_getinumber(fs, dp)); 1755 } 1756 printf("\n"); 1757 } 1758 } 1759 } 1760 mutex_exit(&lfs_lock); 1761 #endif /* DEBUG */ 1762 panic("lfs_loadvnode: dinode not found"); 1763 } 1764 lfs_copy_dinode(fs, ip->i_din, dip); 1765 brelse(bp, 0); 1766 1767 out: 1768 if (lfs_sb_getversion(fs) > 1) { 1769 lfs_dino_setatime(fs, ip->i_din, ts.tv_sec); 1770 lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec); 1771 } 1772 1773 lfs_vinit(mp, &vp); 1774 1775 *new_key = &ip->i_number; 1776 return 0; 1777 } 1778 1779 /* 1780 * Create a new inode and initialize this vnode / inode pair. 1781 */ 1782 int 1783 lfs_newvnode(struct mount *mp, struct vnode *dvp, struct vnode *vp, 1784 struct vattr *vap, kauth_cred_t cred, void *extra, 1785 size_t *key_len, const void **new_key) 1786 { 1787 ino_t ino; 1788 struct inode *ip; 1789 struct ulfsmount *ump; 1790 struct lfs *fs; 1791 int error, mode, gen; 1792 1793 KASSERT(dvp != NULL || vap->va_fileid > 0); 1794 KASSERT(dvp != NULL && dvp->v_mount == mp); 1795 KASSERT(vap->va_type != VNON); 1796 1797 *key_len = sizeof(ino); 1798 ump = VFSTOULFS(mp); 1799 fs = ump->um_lfs; 1800 mode = MAKEIMODE(vap->va_type, vap->va_mode); 1801 1802 /* 1803 * Allocate fresh inode. With "dvp == NULL" take the inode number 1804 * and version from "vap". 1805 */ 1806 if (dvp == NULL) { 1807 ino = vap->va_fileid; 1808 gen = vap->va_gen; 1809 error = lfs_valloc_fixed(fs, ino, gen); 1810 } else { 1811 error = lfs_valloc(dvp, mode, cred, &ino, &gen); 1812 } 1813 if (error) 1814 return error; 1815 1816 /* Attach inode to vnode. */ 1817 lfs_init_vnode(ump, ino, vp); 1818 ip = VTOI(vp); 1819 1820 mutex_enter(&lfs_lock); 1821 LFS_SET_UINO(ip, IN_CHANGE); 1822 mutex_exit(&lfs_lock); 1823 1824 /* Note no blocks yet */ 1825 ip->i_lfs_hiblk = -1; 1826 1827 /* Set a new generation number for this inode. */ 1828 ip->i_gen = gen; 1829 lfs_dino_setgen(fs, ip->i_din, gen); 1830 1831 memset(ip->i_lfs_fragsize, 0, 1832 ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 1833 1834 /* Set uid / gid. */ 1835 if (cred == NOCRED || cred == FSCRED) { 1836 ip->i_gid = 0; 1837 ip->i_uid = 0; 1838 } else { 1839 ip->i_gid = VTOI(dvp)->i_gid; 1840 ip->i_uid = kauth_cred_geteuid(cred); 1841 } 1842 DIP_ASSIGN(ip, gid, ip->i_gid); 1843 DIP_ASSIGN(ip, uid, ip->i_uid); 1844 1845 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 1846 error = lfs_chkiq(ip, 1, cred, 0); 1847 if (error) { 1848 lfs_vfree(dvp, ino, mode); 1849 lfs_deinit_vnode(ump, vp); 1850 1851 return error; 1852 } 1853 #endif 1854 1855 /* Set type and finalize. */ 1856 ip->i_flags = 0; 1857 DIP_ASSIGN(ip, flags, 0); 1858 ip->i_mode = mode; 1859 DIP_ASSIGN(ip, mode, mode); 1860 if (vap->va_rdev != VNOVAL) { 1861 /* 1862 * Want to be able to use this to make badblock 1863 * inodes, so don't truncate the dev number. 1864 */ 1865 // XXX clean this up 1866 if (ump->um_fstype == ULFS1) 1867 ip->i_din->u_32.di_rdev = ulfs_rw32(vap->va_rdev, 1868 ULFS_MPNEEDSWAP(fs)); 1869 else 1870 ip->i_din->u_64.di_rdev = ulfs_rw64(vap->va_rdev, 1871 ULFS_MPNEEDSWAP(fs)); 1872 } 1873 lfs_vinit(mp, &vp); 1874 1875 *new_key = &ip->i_number; 1876 return 0; 1877 } 1878 1879 /* 1880 * File handle to vnode 1881 */ 1882 int 1883 lfs_fhtovp(struct mount *mp, struct fid *fhp, int lktype, struct vnode **vpp) 1884 { 1885 struct lfid lfh; 1886 struct lfs *fs; 1887 1888 if (fhp->fid_len != sizeof(struct lfid)) 1889 return EINVAL; 1890 1891 memcpy(&lfh, fhp, sizeof(lfh)); 1892 if (lfh.lfid_ino < LFS_IFILE_INUM) 1893 return ESTALE; 1894 1895 fs = VFSTOULFS(mp)->um_lfs; 1896 if (lfh.lfid_ident != lfs_sb_getident(fs)) 1897 return ESTALE; 1898 1899 if (lfh.lfid_ino > 1900 ((lfs_dino_getsize(fs, VTOI(fs->lfs_ivnode)->i_din) >> lfs_sb_getbshift(fs)) - 1901 lfs_sb_getcleansz(fs) - lfs_sb_getsegtabsz(fs)) * lfs_sb_getifpb(fs)) 1902 return ESTALE; 1903 1904 return (ulfs_fhtovp(mp, &lfh.lfid_ufid, lktype, vpp)); 1905 } 1906 1907 /* 1908 * Vnode pointer to File handle 1909 */ 1910 /* ARGSUSED */ 1911 int 1912 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 1913 { 1914 struct inode *ip; 1915 struct lfid lfh; 1916 1917 if (*fh_size < sizeof(struct lfid)) { 1918 *fh_size = sizeof(struct lfid); 1919 return E2BIG; 1920 } 1921 *fh_size = sizeof(struct lfid); 1922 ip = VTOI(vp); 1923 memset(&lfh, 0, sizeof(lfh)); 1924 lfh.lfid_len = sizeof(struct lfid); 1925 lfh.lfid_ino = ip->i_number; 1926 lfh.lfid_gen = ip->i_gen; 1927 lfh.lfid_ident = lfs_sb_getident(ip->i_lfs); 1928 memcpy(fhp, &lfh, sizeof(lfh)); 1929 return (0); 1930 } 1931 1932 /* 1933 * ulfs_bmaparray callback function for writing. 1934 * 1935 * Since blocks will be written to the new segment anyway, 1936 * we don't care about current daddr of them. 1937 */ 1938 static bool 1939 lfs_issequential_hole(const struct lfs *fs, 1940 daddr_t daddr0, daddr_t daddr1) 1941 { 1942 (void)fs; /* not used */ 1943 1944 KASSERT(daddr0 == UNWRITTEN || 1945 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR(fs))); 1946 KASSERT(daddr1 == UNWRITTEN || 1947 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR(fs))); 1948 1949 /* NOTE: all we want to know here is 'hole or not'. */ 1950 /* NOTE: UNASSIGNED is converted to 0 by ulfs_bmaparray. */ 1951 1952 /* 1953 * treat UNWRITTENs and all resident blocks as 'contiguous' 1954 */ 1955 if (daddr0 != 0 && daddr1 != 0) 1956 return true; 1957 1958 /* 1959 * both are in hole? 1960 */ 1961 if (daddr0 == 0 && daddr1 == 0) 1962 return true; /* all holes are 'contiguous' for us. */ 1963 1964 return false; 1965 } 1966 1967 /* 1968 * lfs_gop_write functions exactly like genfs_gop_write, except that 1969 * (1) it requires the seglock to be held by its caller, and sp->fip 1970 * to be properly initialized (it will return without re-initializing 1971 * sp->fip, and without calling lfs_writeseg). 1972 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1973 * to determine how large a block it can write at once (though it does 1974 * still use VOP_BMAP to find holes in the file); 1975 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1976 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1977 * now have clusters of clusters, ick.) 1978 */ 1979 static int 1980 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, 1981 int flags) 1982 { 1983 int i, error, run, haveeof = 0; 1984 int fs_bshift; 1985 vaddr_t kva; 1986 off_t eof, offset, startoffset = 0; 1987 size_t bytes, iobytes, skipbytes; 1988 bool async = (flags & PGO_SYNCIO) == 0; 1989 daddr_t lbn, blkno; 1990 struct vm_page *pg; 1991 struct buf *mbp, *bp; 1992 struct vnode *devvp = VTOI(vp)->i_devvp; 1993 struct inode *ip = VTOI(vp); 1994 struct lfs *fs = ip->i_lfs; 1995 struct segment *sp = fs->lfs_sp; 1996 SEGSUM *ssp; 1997 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1998 const char * failreason = NULL; 1999 2000 ASSERT_SEGLOCK(fs); 2001 2002 /* The Ifile lives in the buffer cache */ 2003 KASSERT(vp != fs->lfs_ivnode); 2004 2005 /* 2006 * We don't want to fill the disk before the cleaner has a chance 2007 * to make room for us. If we're in danger of doing that, fail 2008 * with EAGAIN. The caller will have to notice this, unlock 2009 * so the cleaner can run, relock and try again. 2010 * 2011 * We must write everything, however, if our vnode is being 2012 * reclaimed. 2013 */ 2014 mutex_enter(vp->v_interlock); 2015 if (LFS_STARVED_FOR_SEGS(fs) && vdead_check(vp, VDEAD_NOWAIT) == 0) { 2016 mutex_exit(vp->v_interlock); 2017 failreason = "Starved for segs and not flushing vp"; 2018 goto tryagain; 2019 } 2020 mutex_exit(vp->v_interlock); 2021 2022 /* 2023 * Sometimes things slip past the filters in lfs_putpages, 2024 * and the pagedaemon tries to write pages---problem is 2025 * that the pagedaemon never acquires the segment lock. 2026 * 2027 * Alternatively, pages that were clean when we called 2028 * genfs_putpages may have become dirty in the meantime. In this 2029 * case the segment header is not properly set up for blocks 2030 * to be added to it. 2031 * 2032 * Unbusy and unclean the pages, and put them on the ACTIVE 2033 * queue under the hypothesis that they couldn't have got here 2034 * unless they were modified *quite* recently. 2035 * 2036 * XXXUBC that last statement is an oversimplification of course. 2037 */ 2038 if (!LFS_SEGLOCK_HELD(fs)) { 2039 failreason = "Seglock not held"; 2040 goto tryagain; 2041 } 2042 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) { 2043 failreason = "Inode with no_gop_write"; 2044 goto tryagain; 2045 } 2046 if ((pgs[0]->offset & lfs_sb_getbmask(fs)) != 0) { 2047 failreason = "Bad page offset"; 2048 goto tryagain; 2049 } 2050 2051 UVMHIST_LOG(ubchist, "vp %#jx pgs %#jx npages %jd flags 0x%jx", 2052 (uintptr_t)vp, (uintptr_t)pgs, npages, flags); 2053 2054 GOP_SIZE(vp, vp->v_size, &eof, 0); 2055 haveeof = 1; 2056 2057 if (vp->v_type == VREG) 2058 fs_bshift = vp->v_mount->mnt_fs_bshift; 2059 else 2060 fs_bshift = DEV_BSHIFT; 2061 error = 0; 2062 pg = pgs[0]; 2063 startoffset = pg->offset; 2064 KASSERT(eof >= 0); 2065 2066 if (startoffset >= eof) { 2067 failreason = "Offset beyond EOF"; 2068 goto tryagain; 2069 } else 2070 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 2071 skipbytes = 0; 2072 2073 KASSERT(bytes != 0); 2074 2075 /* Swap PG_DELWRI for PG_PAGEOUT */ 2076 for (i = 0; i < npages; i++) { 2077 if (pgs[i]->flags & PG_DELWRI) { 2078 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 2079 pgs[i]->flags &= ~PG_DELWRI; 2080 pgs[i]->flags |= PG_PAGEOUT; 2081 uvm_pageout_start(1); 2082 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 2083 uvm_pagelock(pgs[i]); 2084 uvm_pageunwire(pgs[i]); 2085 uvm_pageunlock(pgs[i]); 2086 rw_exit(vp->v_uobj.vmobjlock); 2087 } 2088 } 2089 2090 /* 2091 * Check to make sure we're starting on a block boundary. 2092 * We'll check later to make sure we always write entire 2093 * blocks (or fragments). 2094 */ 2095 if (startoffset & lfs_sb_getbmask(fs)) 2096 printf("%" PRId64 " & %" PRIu64 " = %" PRId64 "\n", 2097 startoffset, lfs_sb_getbmask(fs), 2098 startoffset & lfs_sb_getbmask(fs)); 2099 KASSERT((startoffset & lfs_sb_getbmask(fs)) == 0); 2100 if (bytes & lfs_sb_getffmask(fs)) { 2101 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 2102 panic("lfs_gop_write: non-integer blocks"); 2103 } 2104 2105 /* 2106 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 2107 * If we would, write what we have and try again. If we don't 2108 * have anything to write, we'll have to sleep. 2109 */ 2110 ssp = (SEGSUM *)sp->segsum; 2111 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2112 (lfs_ss_getnfinfo(fs, ssp) < 1 ? 2113 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 2114 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 2115 #if 0 2116 " with nfinfo=%d at offset 0x%jx\n", 2117 (int)lfs_ss_getnfinfo(fs, ssp), 2118 (uintmax_t)lfs_sb_getoffset(fs))); 2119 #endif 2120 lfs_updatemeta(sp); 2121 lfs_release_finfo(fs); 2122 (void) lfs_writeseg(fs, sp); 2123 2124 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2125 2126 /* 2127 * Having given up all of the pager_map we were holding, 2128 * we can now wait for aiodoned to reclaim it for us 2129 * without fear of deadlock. 2130 */ 2131 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2132 UVMPAGER_MAPIN_WAITOK); 2133 } 2134 2135 mbp = getiobuf(NULL, true); 2136 UVMHIST_LOG(ubchist, "vp %#jx mbp %#jx num now %jd bytes 0x%jx", 2137 (uintptr_t)vp, (uintptr_t)mbp, vp->v_numoutput, bytes); 2138 mbp->b_bufsize = npages << PAGE_SHIFT; 2139 mbp->b_data = (void *)kva; 2140 mbp->b_resid = mbp->b_bcount = bytes; 2141 mbp->b_cflags |= BC_BUSY|BC_AGE; 2142 mbp->b_iodone = uvm_aio_aiodone; 2143 2144 bp = NULL; 2145 for (offset = startoffset; 2146 bytes > 0; 2147 offset += iobytes, bytes -= iobytes) { 2148 lbn = offset >> fs_bshift; 2149 error = ulfs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 2150 lfs_issequential_hole); 2151 if (error) { 2152 UVMHIST_LOG(ubchist, "ulfs_bmaparray() -> %jd", 2153 error,0,0,0); 2154 skipbytes += bytes; 2155 bytes = 0; 2156 break; 2157 } 2158 2159 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 2160 bytes); 2161 if (blkno == (daddr_t)-1) { 2162 skipbytes += iobytes; 2163 continue; 2164 } 2165 2166 /* 2167 * Discover how much we can really pack into this buffer. 2168 */ 2169 /* If no room in the current segment, finish it up */ 2170 if (sp->sum_bytes_left < sizeof(int32_t) || 2171 sp->seg_bytes_left < (1 << lfs_sb_getbshift(fs))) { 2172 int vers; 2173 2174 lfs_updatemeta(sp); 2175 vers = lfs_fi_getversion(fs, sp->fip); 2176 lfs_release_finfo(fs); 2177 (void) lfs_writeseg(fs, sp); 2178 2179 lfs_acquire_finfo(fs, ip->i_number, vers); 2180 } 2181 /* Check both for space in segment and space in segsum */ 2182 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 2183 << fs_bshift); 2184 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 2185 << fs_bshift); 2186 KASSERT(iobytes > 0); 2187 2188 /* if it's really one i/o, don't make a second buf */ 2189 if (offset == startoffset && iobytes == bytes) { 2190 bp = mbp; 2191 /* 2192 * All the LFS output is done by the segwriter. It 2193 * will increment numoutput by one for all the bufs it 2194 * receives. However this buffer needs one extra to 2195 * account for aiodone. 2196 */ 2197 mutex_enter(vp->v_interlock); 2198 vp->v_numoutput++; 2199 mutex_exit(vp->v_interlock); 2200 } else { 2201 bp = getiobuf(NULL, true); 2202 UVMHIST_LOG(ubchist, "vp %#jx bp %#jx num now %jd", 2203 (uintptr_t)vp, (uintptr_t)bp, vp->v_numoutput, 0); 2204 nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes); 2205 /* 2206 * LFS doesn't like async I/O here, dies with 2207 * an assert in lfs_bwrite(). Is that assert 2208 * valid? I retained non-async behaviour when 2209 * converted this to use nestiobuf --pooka 2210 */ 2211 bp->b_flags &= ~B_ASYNC; 2212 } 2213 2214 /* XXX This is silly ... is this necessary? */ 2215 mutex_enter(&bufcache_lock); 2216 mutex_enter(vp->v_interlock); 2217 bgetvp(vp, bp); 2218 mutex_exit(vp->v_interlock); 2219 mutex_exit(&bufcache_lock); 2220 2221 bp->b_lblkno = lfs_lblkno(fs, offset); 2222 bp->b_private = mbp; 2223 if (devvp->v_type == VBLK) { 2224 bp->b_dev = devvp->v_rdev; 2225 } 2226 VOP_BWRITE(bp->b_vp, bp); 2227 while (lfs_gatherblock(sp, bp, NULL)) 2228 continue; 2229 } 2230 2231 nestiobuf_done(mbp, skipbytes, error); 2232 if (skipbytes) { 2233 UVMHIST_LOG(ubchist, "skipbytes %jd", skipbytes, 0,0,0); 2234 } 2235 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 2236 2237 if (!async) { 2238 /* Start a segment write. */ 2239 UVMHIST_LOG(ubchist, "flushing", 0,0,0,0); 2240 mutex_enter(&lfs_lock); 2241 lfs_flush(fs, 0, 1); 2242 mutex_exit(&lfs_lock); 2243 } 2244 2245 if ((sp->seg_flags & SEGM_SINGLE) && lfs_sb_getcurseg(fs) != fs->lfs_startseg) 2246 return EAGAIN; 2247 2248 return (0); 2249 2250 tryagain: 2251 /* 2252 * We can't write the pages, for whatever reason. 2253 * Clean up after ourselves, and make the caller try again. 2254 */ 2255 mutex_enter(vp->v_interlock); 2256 2257 /* Tell why we're here, if we know */ 2258 if (failreason != NULL) { 2259 DLOG((DLOG_PAGE, "lfs_gop_write: %s\n", failreason)); 2260 } 2261 if (haveeof && startoffset >= eof) { 2262 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 2263 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 2264 pgs[0]->offset, eof, npages)); 2265 } 2266 2267 for (i = 0; i < npages; i++) { 2268 pg = pgs[i]; 2269 2270 if (pg->flags & PG_PAGEOUT) 2271 uvm_pageout_done(1); 2272 uvm_pagelock(pg); 2273 if (pg->flags & PG_DELWRI) { 2274 uvm_pageunwire(pg); 2275 } 2276 uvm_pageactivate(pg); 2277 uvm_pageunlock(pg); 2278 pg->flags &= ~(PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 2279 uvm_pagemarkdirty(pg, UVM_PAGE_STATUS_DIRTY); 2280 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 2281 vp, pg->offset)); 2282 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 2283 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 2284 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 2285 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 2286 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 2287 pg->wire_count)); 2288 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 2289 pg->loan_count)); 2290 } 2291 uvm_page_unbusy(pgs, npages); 2292 mutex_exit(vp->v_interlock); 2293 return EAGAIN; 2294 } 2295 2296 /* 2297 * finish vnode/inode initialization. 2298 * used by lfs_vget. 2299 */ 2300 void 2301 lfs_vinit(struct mount *mp, struct vnode **vpp) 2302 { 2303 struct vnode *vp = *vpp; 2304 struct inode *ip = VTOI(vp); 2305 struct ulfsmount *ump = VFSTOULFS(mp); 2306 struct lfs *fs = ump->um_lfs; 2307 int i; 2308 2309 ip->i_mode = lfs_dino_getmode(fs, ip->i_din); 2310 ip->i_nlink = lfs_dino_getnlink(fs, ip->i_din); 2311 ip->i_lfs_osize = ip->i_size = lfs_dino_getsize(fs, ip->i_din); 2312 ip->i_flags = lfs_dino_getflags(fs, ip->i_din); 2313 ip->i_gen = lfs_dino_getgen(fs, ip->i_din); 2314 ip->i_uid = lfs_dino_getuid(fs, ip->i_din); 2315 ip->i_gid = lfs_dino_getgid(fs, ip->i_din); 2316 2317 ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din); 2318 ip->i_lfs_odnlink = lfs_dino_getnlink(fs, ip->i_din); 2319 2320 /* 2321 * Initialize the vnode from the inode, check for aliases. In all 2322 * cases re-init ip, the underlying vnode/inode may have changed. 2323 */ 2324 ulfs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 2325 ip = VTOI(vp); 2326 2327 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 2328 if (vp->v_type != VLNK || ip->i_size >= ip->i_lfs->um_maxsymlinklen) { 2329 #ifdef DEBUG 2330 for (i = (ip->i_size + lfs_sb_getbsize(fs) - 1) >> lfs_sb_getbshift(fs); 2331 i < ULFS_NDADDR; i++) { 2332 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 2333 i == 0) 2334 continue; 2335 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) { 2336 lfs_dump_dinode(fs, ip->i_din); 2337 panic("inconsistent inode (direct)"); 2338 } 2339 } 2340 for ( ; i < ULFS_NDADDR + ULFS_NIADDR; i++) { 2341 if (lfs_dino_getib(fs, ip->i_din, i - ULFS_NDADDR) != 0) { 2342 lfs_dump_dinode(fs, ip->i_din); 2343 panic("inconsistent inode (indirect)"); 2344 } 2345 } 2346 #endif /* DEBUG */ 2347 for (i = 0; i < ULFS_NDADDR; i++) 2348 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) 2349 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i); 2350 } 2351 2352 KASSERTMSG((vp->v_type != VNON), 2353 "lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 2354 (unsigned long long)ip->i_number, 2355 (ip->i_mode & LFS_IFMT) >> 12); 2356 2357 /* 2358 * Finish inode initialization now that aliasing has been resolved. 2359 */ 2360 2361 ip->i_devvp = fs->lfs_devvp; 2362 vref(ip->i_devvp); 2363 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 2364 ulfsquota_init(ip); 2365 #endif 2366 genfs_node_init(vp, &lfs_genfsops); 2367 uvm_vnp_setsize(vp, ip->i_size); 2368 2369 /* Initialize hiblk from file size */ 2370 ip->i_lfs_hiblk = lfs_lblkno(ip->i_lfs, ip->i_size + lfs_sb_getbsize(ip->i_lfs) - 1) - 1; 2371 2372 *vpp = vp; 2373 } 2374 2375 /* 2376 * Resize the filesystem to contain the specified number of segments. 2377 */ 2378 int 2379 lfs_resize_fs(struct lfs *fs, int newnsegs) 2380 { 2381 SEGUSE *sup; 2382 CLEANERINFO *cip; 2383 struct buf *bp, *obp; 2384 daddr_t olast, nlast, ilast, noff, start, end; 2385 struct vnode *ivp; 2386 struct inode *ip; 2387 int error, badnews, inc, oldnsegs; 2388 int sbbytes, csbbytes, gain, cgain; 2389 int i; 2390 2391 /* Only support v2 and up */ 2392 if (lfs_sb_getversion(fs) < 2) 2393 return EOPNOTSUPP; 2394 2395 /* If we're doing nothing, do it fast */ 2396 oldnsegs = lfs_sb_getnseg(fs); 2397 if (newnsegs == oldnsegs) 2398 return 0; 2399 2400 /* We always have to have two superblocks */ 2401 if (newnsegs <= lfs_dtosn(fs, lfs_sb_getsboff(fs, 1))) 2402 /* XXX this error code is rather nonsense */ 2403 return EFBIG; 2404 2405 ivp = fs->lfs_ivnode; 2406 ip = VTOI(ivp); 2407 error = 0; 2408 2409 /* Take the segment lock so no one else calls lfs_newseg() */ 2410 lfs_seglock(fs, SEGM_PROT); 2411 2412 /* 2413 * Make sure the segments we're going to be losing, if any, 2414 * are in fact empty. We hold the seglock, so their status 2415 * cannot change underneath us. Count the superblocks we lose, 2416 * while we're at it. 2417 */ 2418 sbbytes = csbbytes = 0; 2419 cgain = 0; 2420 for (i = newnsegs; i < oldnsegs; i++) { 2421 LFS_SEGENTRY(sup, fs, i, bp); 2422 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 2423 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2424 sbbytes += LFS_SBPAD; 2425 if (!(sup->su_flags & SEGUSE_DIRTY)) { 2426 ++cgain; 2427 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2428 csbbytes += LFS_SBPAD; 2429 } 2430 brelse(bp, 0); 2431 if (badnews) { 2432 error = EBUSY; 2433 goto out; 2434 } 2435 } 2436 2437 /* Note old and new segment table endpoints, and old ifile size */ 2438 olast = lfs_sb_getcleansz(fs) + lfs_sb_getsegtabsz(fs); 2439 nlast = howmany(newnsegs, lfs_sb_getsepb(fs)) + lfs_sb_getcleansz(fs); 2440 ilast = ivp->v_size >> lfs_sb_getbshift(fs); 2441 noff = nlast - olast; 2442 2443 /* 2444 * Make sure no one can use the Ifile while we change it around. 2445 * Even after taking the iflock we need to make sure no one still 2446 * is holding Ifile buffers, so we get each one, to drain them. 2447 * (XXX this could be done better.) 2448 */ 2449 rw_enter(&fs->lfs_iflock, RW_WRITER); 2450 for (i = 0; i < ilast; i++) { 2451 /* XXX what to do if bread fails? */ 2452 bread(ivp, i, lfs_sb_getbsize(fs), 0, &bp); 2453 brelse(bp, 0); 2454 } 2455 2456 /* Allocate new Ifile blocks */ 2457 for (i = ilast; i < ilast + noff; i++) { 2458 if (lfs_balloc(ivp, i * lfs_sb_getbsize(fs), lfs_sb_getbsize(fs), NOCRED, 0, 2459 &bp) != 0) 2460 panic("balloc extending ifile"); 2461 memset(bp->b_data, 0, lfs_sb_getbsize(fs)); 2462 VOP_BWRITE(bp->b_vp, bp); 2463 } 2464 2465 /* Register new ifile size */ 2466 ip->i_size += noff * lfs_sb_getbsize(fs); 2467 lfs_dino_setsize(fs, ip->i_din, ip->i_size); 2468 uvm_vnp_setsize(ivp, ip->i_size); 2469 2470 /* Copy the inode table to its new position */ 2471 if (noff != 0) { 2472 if (noff < 0) { 2473 start = nlast; 2474 end = ilast + noff; 2475 inc = 1; 2476 } else { 2477 start = ilast + noff - 1; 2478 end = nlast - 1; 2479 inc = -1; 2480 } 2481 for (i = start; i != end; i += inc) { 2482 if (bread(ivp, i, lfs_sb_getbsize(fs), 2483 B_MODIFY, &bp) != 0) 2484 panic("resize: bread dst blk failed"); 2485 if (bread(ivp, i - noff, lfs_sb_getbsize(fs), 2486 0, &obp)) 2487 panic("resize: bread src blk failed"); 2488 memcpy(bp->b_data, obp->b_data, lfs_sb_getbsize(fs)); 2489 VOP_BWRITE(bp->b_vp, bp); 2490 brelse(obp, 0); 2491 } 2492 } 2493 2494 /* If we are expanding, write the new empty SEGUSE entries */ 2495 if (newnsegs > oldnsegs) { 2496 for (i = oldnsegs; i < newnsegs; i++) { 2497 if ((error = bread(ivp, i / lfs_sb_getsepb(fs) + 2498 lfs_sb_getcleansz(fs), lfs_sb_getbsize(fs), 2499 B_MODIFY, &bp)) != 0) 2500 panic("lfs: ifile read: %d", error); 2501 while ((i + 1) % lfs_sb_getsepb(fs) && i < newnsegs) { 2502 sup = &((SEGUSE *)bp->b_data)[i % lfs_sb_getsepb(fs)]; 2503 memset(sup, 0, sizeof(*sup)); 2504 i++; 2505 } 2506 VOP_BWRITE(bp->b_vp, bp); 2507 } 2508 } 2509 2510 /* Zero out unused superblock offsets */ 2511 for (i = 2; i < LFS_MAXNUMSB; i++) 2512 if (lfs_dtosn(fs, lfs_sb_getsboff(fs, i)) >= newnsegs) 2513 lfs_sb_setsboff(fs, i, 0x0); 2514 2515 /* 2516 * Correct superblock entries that depend on fs size. 2517 * The computations of these are as follows: 2518 * 2519 * size = lfs_segtod(fs, nseg) 2520 * dsize = lfs_segtod(fs, nseg - minfreeseg) - lfs_btofsb(#super * LFS_SBPAD) 2521 * bfree = dsize - lfs_btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2522 * avail = lfs_segtod(fs, nclean) - lfs_btofsb(#clean_super * LFS_SBPAD) 2523 * + (lfs_segtod(fs, 1) - (offset - curseg)) 2524 * - lfs_segtod(fs, minfreeseg - (minfreeseg / 2)) 2525 * 2526 * XXX - we should probably adjust minfreeseg as well. 2527 */ 2528 gain = (newnsegs - oldnsegs); 2529 lfs_sb_setnseg(fs, newnsegs); 2530 lfs_sb_setsegtabsz(fs, nlast - lfs_sb_getcleansz(fs)); 2531 lfs_sb_addsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs))); 2532 lfs_sb_adddsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes)); 2533 lfs_sb_addbfree(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes) 2534 - gain * lfs_btofsb(fs, lfs_sb_getbsize(fs) / 2)); 2535 if (gain > 0) { 2536 lfs_sb_addnclean(fs, gain); 2537 lfs_sb_addavail(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs))); 2538 } else { 2539 lfs_sb_subnclean(fs, cgain); 2540 lfs_sb_subavail(fs, cgain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - 2541 lfs_btofsb(fs, csbbytes)); 2542 } 2543 2544 /* Resize segment flag cache */ 2545 fs->lfs_suflags[0] = realloc(fs->lfs_suflags[0], 2546 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK); 2547 fs->lfs_suflags[1] = realloc(fs->lfs_suflags[1], 2548 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK); 2549 for (i = oldnsegs; i < newnsegs; i++) 2550 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2551 2552 /* Truncate Ifile if necessary */ 2553 if (noff < 0) 2554 lfs_truncate(ivp, ivp->v_size + (noff << lfs_sb_getbshift(fs)), 0, 2555 NOCRED); 2556 2557 /* Update cleaner info so the cleaner can die */ 2558 /* XXX what to do if bread fails? */ 2559 bread(ivp, 0, lfs_sb_getbsize(fs), B_MODIFY, &bp); 2560 cip = bp->b_data; 2561 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs)); 2562 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs)); 2563 VOP_BWRITE(bp->b_vp, bp); 2564 2565 /* Let Ifile accesses proceed */ 2566 rw_exit(&fs->lfs_iflock); 2567 2568 out: 2569 lfs_segunlock(fs); 2570 return error; 2571 } 2572 2573 /* 2574 * Extended attribute dispatch 2575 */ 2576 int 2577 lfs_extattrctl(struct mount *mp, int cmd, struct vnode *vp, 2578 int attrnamespace, const char *attrname) 2579 { 2580 #ifdef LFS_EXTATTR 2581 struct ulfsmount *ump; 2582 2583 ump = VFSTOULFS(mp); 2584 if (ump->um_fstype == ULFS1) { 2585 return ulfs_extattrctl(mp, cmd, vp, attrnamespace, attrname); 2586 } 2587 #endif 2588 return vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname); 2589 } 2590
Indexes created Tue Sep 30 11:09:46 GMT 2025