1 /* $NetBSD: lfs_vfsops.c,v 1.408 2026/10/30 01:03:04 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 1.408 2026/10/30 01:03:04 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/fstypes.h> 84 #include <sys/disklabel.h> 85 #include <sys/ioctl.h> 86 #include <sys/errno.h> 87 #include <sys/malloc.h> 88 #include <sys/pool.h> 89 #include <sys/socket.h> 90 #include <sys/syslog.h> 91 #include <sys/sysctl.h> 92 #include <sys/conf.h> 93 #include <sys/kauth.h> 94 #include <sys/module.h> 95 #include <sys/syscallvar.h> 96 #include <sys/syscall.h> 97 #include <sys/syscallargs.h> 98 99 #include <miscfs/specfs/specdev.h> 100 101 #include <ufs/lfs/ulfs_quotacommon.h> 102 #include <ufs/lfs/ulfs_inode.h> 103 #include <ufs/lfs/ulfsmount.h> 104 #include <ufs/lfs/ulfs_bswap.h> 105 #include <ufs/lfs/ulfs_extern.h> 106 107 #ifdef UVMHIST 108 #include <uvm/uvm.h> 109 #endif 110 #include <uvm/uvm_extern.h> 111 #include <uvm/uvm_object.h> 112 #include <uvm/uvm_page.h> 113 #include <uvm/uvm_stat.h> 114 115 #include <ufs/lfs/lfs.h> 116 #include <ufs/lfs/lfs_accessors.h> 117 #include <ufs/lfs/lfs_kernel.h> 118 #include <ufs/lfs/lfs_extern.h> 119 120 #include <miscfs/genfs/genfs.h> 121 #include <miscfs/genfs/genfs_node.h> 122 123 MODULE(MODULE_CLASS_VFS, lfs, NULL); 124 125 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int); 126 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *); 127 static int lfs_flushfiles(struct mount *, int); 128 static int compare_nodes_sd(void *, const void *, const void *); 129 static int compare_key_sd(void *, const void *, const void *); 130 static void lfs_reclaimd(void *); 131 132 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc; 133 extern const struct vnodeopv_desc lfs_specop_opv_desc; 134 extern const struct vnodeopv_desc lfs_fifoop_opv_desc; 135 136 extern int locked_queue_rcount; 137 extern long locked_queue_rbytes; 138 139 static const rb_tree_ops_t lfs_rbtree_ops = { 140 .rbto_compare_nodes = compare_nodes_sd, 141 .rbto_compare_key = compare_key_sd, 142 .rbto_node_offset = offsetof(struct segdelta, rb_entry), 143 .rbto_context = NULL 144 }; 145 146 struct lwp * lfs_writer_daemon = NULL; 147 kcondvar_t lfs_writerd_cv; 148 149 struct workqueue *lfs_cluster_wq = NULL; 150 struct workqueue *lfs_super_wq = NULL; 151 152 int lfs_do_flush = 0; 153 int lfs_do_rfw = 1; 154 155 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = { 156 &lfs_vnodeop_opv_desc, 157 &lfs_specop_opv_desc, 158 &lfs_fifoop_opv_desc, 159 NULL, 160 }; 161 162 struct vfsops lfs_vfsops = { 163 .vfs_name = MOUNT_LFS, 164 .vfs_min_mount_data = sizeof (struct ulfs_args), 165 .vfs_mount = lfs_mount, 166 .vfs_start = ulfs_start, 167 .vfs_unmount = lfs_unmount, 168 .vfs_root = ulfs_root, 169 .vfs_quotactl = ulfs_quotactl, 170 .vfs_statvfs = lfs_statvfs, 171 .vfs_sync = lfs_sync, 172 .vfs_vget = lfs_vget, 173 .vfs_loadvnode = lfs_loadvnode, 174 .vfs_newvnode = lfs_newvnode, 175 .vfs_fhtovp = lfs_fhtovp, 176 .vfs_vptofh = lfs_vptofh, 177 .vfs_init = lfs_init, 178 .vfs_reinit = lfs_reinit, 179 .vfs_done = lfs_done, 180 .vfs_mountroot = lfs_mountroot, 181 .vfs_snapshot = (void *)eopnotsupp, 182 .vfs_extattrctl = lfs_extattrctl, 183 .vfs_suspendctl = genfs_suspendctl, 184 .vfs_renamelock_enter = genfs_renamelock_enter, 185 .vfs_renamelock_exit = genfs_renamelock_exit, 186 .vfs_fsync = (void *)eopnotsupp, 187 .vfs_opv_descs = lfs_vnodeopv_descs 188 }; 189 190 const struct genfs_ops lfs_genfsops = { 191 .gop_size = lfs_gop_size, 192 .gop_alloc = ulfs_gop_alloc, 193 .gop_write = lfs_gop_write, 194 .gop_markupdate = ulfs_gop_markupdate, 195 .gop_putrange = genfs_gop_putrange, 196 }; 197 198 struct shortlong { 199 const char *sname; 200 const char *lname; 201 }; 202 203 static int 204 sysctl_lfs_dostats(SYSCTLFN_ARGS) 205 { 206 extern struct lfs_stats lfs_stats; 207 extern int lfs_dostats; 208 int error; 209 210 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 211 if (error || newp == NULL) 212 return (error); 213 214 if (lfs_dostats == 0) 215 memset(&lfs_stats, 0, sizeof(lfs_stats)); 216 217 return (0); 218 } 219 220 SYSCTL_SETUP(lfs_sysctl_setup, "lfs sysctl") 221 { 222 int i; 223 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 224 lfs_fs_pagetrip, lfs_ignore_lazy_sync; 225 #ifdef DEBUG 226 extern int lfs_debug_log_subsys[DLOG_MAX]; 227 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 228 { "rollforward", "Debug roll-forward code" }, 229 { "alloc", "Debug inode allocation and free list" }, 230 { "avail", "Debug space-available-now accounting" }, 231 { "flush", "Debug flush triggers" }, 232 { "lockedlist", "Debug locked list accounting" }, 233 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 234 { "vnode", "Debug vnode use during segment write" }, 235 { "segment", "Debug segment writing" }, 236 { "seguse", "Debug segment used-bytes accounting" }, 237 { "cleaner", "Debug cleaning routines" }, 238 { "mount", "Debug mount/unmount routines" }, 239 { "pagecache", "Debug UBC interactions" }, 240 { "dirop", "Debug directory-operation accounting" }, 241 { "malloc", "Debug private malloc accounting" }, 242 }; 243 #endif /* DEBUG */ 244 struct shortlong stat_names[] = { /* Must match lfs.h! */ 245 { "segsused", "Number of new segments allocated" }, 246 { "psegwrites", "Number of partial-segment writes" }, 247 { "psyncwrites", "Number of synchronous partial-segment" 248 " writes" }, 249 { "pcleanwrites", "Number of partial-segment writes by the" 250 " cleaner" }, 251 { "blocktot", "Number of blocks written" }, 252 { "cleanblocks", "Number of blocks written by the cleaner" }, 253 { "ncheckpoints", "Number of checkpoints made" }, 254 { "nwrites", "Number of whole writes" }, 255 { "nsync_writes", "Number of synchronous writes" }, 256 { "wait_exceeded", "Number of times writer waited for" 257 " cleaner" }, 258 { "write_exceeded", "Number of times writer invoked flush" }, 259 { "flush_invoked", "Number of times flush was invoked" }, 260 { "vflush_invoked", "Number of time vflush was called" }, 261 { "clean_inlocked", "Number of vnodes skipped for being dead" }, 262 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 263 { "segs_reclaimed", "Number of segments reclaimed" }, 264 }; 265 266 sysctl_createv(clog, 0, NULL, NULL, 267 CTLFLAG_PERMANENT, 268 CTLTYPE_NODE, "lfs", 269 SYSCTL_DESCR("Log-structured file system"), 270 NULL, 0, NULL, 0, 271 CTL_VFS, 5, CTL_EOL); 272 /* 273 * XXX the "5" above could be dynamic, thereby eliminating one 274 * more instance of the "number to vfs" mapping problem, but 275 * "5" is the order as taken from sys/mount.h 276 */ 277 278 sysctl_createv(clog, 0, NULL, NULL, 279 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 280 CTLTYPE_INT, "flushindir", NULL, 281 NULL, 0, &lfs_writeindir, 0, 282 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 283 sysctl_createv(clog, 0, NULL, NULL, 284 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 285 CTLTYPE_INT, "clean_vnhead", NULL, 286 NULL, 0, &lfs_clean_vnhead, 0, 287 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 288 sysctl_createv(clog, 0, NULL, NULL, 289 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 290 CTLTYPE_INT, "dostats", 291 SYSCTL_DESCR("Maintain statistics on LFS operations"), 292 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 293 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 294 sysctl_createv(clog, 0, NULL, NULL, 295 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 296 CTLTYPE_INT, "pagetrip", 297 SYSCTL_DESCR("How many dirty pages in fs triggers" 298 " a flush"), 299 NULL, 0, &lfs_fs_pagetrip, 0, 300 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 301 sysctl_createv(clog, 0, NULL, NULL, 302 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 303 CTLTYPE_INT, "ignore_lazy_sync", 304 SYSCTL_DESCR("Lazy Sync is ignored entirely"), 305 NULL, 0, &lfs_ignore_lazy_sync, 0, 306 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL); 307 sysctl_createv(clog, 0, NULL, NULL, 308 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 309 CTLTYPE_INT, "rfw", 310 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 311 NULL, 0, &lfs_do_rfw, 0, 312 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 313 sysctl_createv(clog, 0, NULL, NULL, 314 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 315 CTLTYPE_INT, "rfw_limit", 316 SYSCTL_DESCR("Maximum number of partial segments" 317 " to roll forward"), 318 NULL, 0, &lfs_rfw_max_psegs, 0, 319 CTL_VFS, 5, LFS_RFW_LIMIT, CTL_EOL); 320 321 sysctl_createv(clog, 0, NULL, NULL, 322 CTLFLAG_PERMANENT, 323 CTLTYPE_NODE, "stats", 324 SYSCTL_DESCR("Statistics"), 325 NULL, 0, NULL, 0, 326 CTL_VFS, 5, LFS_STATS, CTL_EOL); 327 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 328 sysctl_createv(clog, 0, NULL, NULL, 329 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 330 CTLTYPE_INT, stat_names[i].sname, 331 SYSCTL_DESCR(stat_names[i].lname), 332 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 333 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 334 } 335 336 #ifdef DEBUG 337 sysctl_createv(clog, 0, NULL, NULL, 338 CTLFLAG_PERMANENT, 339 CTLTYPE_NODE, "debug", 340 SYSCTL_DESCR("Debug options"), 341 NULL, 0, NULL, 0, 342 CTL_VFS, 5, LFS_DEBUG, CTL_EOL); 343 sysctl_createv(clog, 0, NULL, NULL, 344 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 345 CTLTYPE_INT, "freelist", 346 SYSCTL_DESCR("Track consistency of inode freelist"), 347 NULL, 0, &lfs_do_check_freelist, 0, 348 CTL_VFS, 5, LFS_DEBUG, LFS_DEBUG_FREELIST, CTL_EOL); 349 sysctl_createv(clog, 0, NULL, NULL, 350 CTLFLAG_PERMANENT, 351 CTLTYPE_NODE, "log", 352 SYSCTL_DESCR("Verbose logging"), 353 NULL, 0, NULL, 0, 354 CTL_VFS, 5, LFS_DEBUG, LFS_DEBUG_LOG, CTL_EOL); 355 for (i = 0; i < DLOG_MAX; i++) { 356 sysctl_createv(clog, 0, NULL, NULL, 357 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 358 CTLTYPE_INT, dlog_names[i].sname, 359 SYSCTL_DESCR(dlog_names[i].lname), 360 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 361 CTL_VFS, 5, LFS_DEBUG, LFS_DEBUG_LOG, i, CTL_EOL); 362 } 363 #endif 364 } 365 366 /* old cleaner syscall interface. see VOP_FCNTL() */ 367 static const struct syscall_package lfs_syscalls[] = { 368 { SYS_lfs_bmapv, 0, (sy_call_t *)sys_lfs_bmapv }, 369 { SYS_lfs_markv, 0, (sy_call_t *)sys_lfs_markv }, 370 { SYS___lfs_segwait50, 0, (sy_call_t *)sys___lfs_segwait50 }, 371 { SYS_lfs_segclean, 0, (sy_call_t *)sys_lfs_segclean }, 372 { 0, 0, NULL }, 373 }; 374 375 static int 376 lfs_modcmd(modcmd_t cmd, void *arg) 377 { 378 int error; 379 380 switch (cmd) { 381 case MODULE_CMD_INIT: 382 error = syscall_establish(NULL, lfs_syscalls); 383 if (error) 384 return error; 385 error = vfs_attach(&lfs_vfsops); 386 if (error != 0) { 387 syscall_disestablish(NULL, lfs_syscalls); 388 break; 389 } 390 cv_init(&lfs_allclean_wakeup, "segment"); 391 break; 392 case MODULE_CMD_FINI: 393 error = vfs_detach(&lfs_vfsops); 394 if (error != 0) 395 break; 396 syscall_disestablish(NULL, lfs_syscalls); 397 cv_destroy(&lfs_allclean_wakeup); 398 break; 399 default: 400 error = ENOTTY; 401 break; 402 } 403 404 return (error); 405 } 406 407 /* 408 * XXX Same structure as FFS inodes? Should we share a common pool? 409 */ 410 struct pool lfs_inode_pool; 411 struct pool lfs_dinode_pool; 412 struct pool lfs_inoext_pool; 413 struct pool lfs_lbnentry_pool; 414 415 /* 416 * The writer daemon. UVM keeps track of how many dirty pages we are holding 417 * in lfs_subsys_pages; the daemon flushes the filesystem when this value 418 * crosses the (user-defined) threshold LFS_MAX_PAGES. 419 */ 420 static void 421 lfs_writerd(void *arg) 422 { 423 mount_iterator_t *iter; 424 struct mount *mp; 425 struct lfs *fs; 426 struct vfsops *vfs = NULL; 427 int fsflags; 428 int lfsc; 429 int wrote_something = 0; 430 431 mutex_enter(&lfs_lock); 432 KASSERTMSG(lfs_writer_daemon == NULL, "more than one LFS writer daemon"); 433 lfs_writer_daemon = curlwp; 434 mutex_exit(&lfs_lock); 435 436 /* Take an extra reference to the LFS vfsops. */ 437 vfs = vfs_getopsbyname(MOUNT_LFS); 438 439 mutex_enter(&lfs_lock); 440 for (;;) { 441 KASSERT(mutex_owned(&lfs_lock)); 442 if (wrote_something == 0) 443 cv_timedwait(&lfs_writerd_cv, &lfs_lock, 2 * hz); 444 KASSERT(mutex_owned(&lfs_lock)); 445 wrote_something = 0; 446 447 /* 448 * If global state wants a flush, flush everything. 449 */ 450 if (lfs_do_flush || locked_queue_count + locked_queue_rcount > LFS_MAX_BUFS || 451 locked_queue_bytes + locked_queue_rbytes > LFS_MAX_BYTES || 452 lfs_subsys_pages > LFS_MAX_PAGES) { 453 454 if (lfs_do_flush) { 455 DLOG((DLOG_FLUSH, "lfs_writerd: lfs_do_flush\n")); 456 } 457 if (locked_queue_count + locked_queue_rcount > LFS_MAX_BUFS) { 458 DLOG((DLOG_FLUSH, "lfs_writerd: lqc+lqrc = %d, max %d\n", 459 locked_queue_count, LFS_MAX_BUFS)); 460 } 461 if (locked_queue_bytes + locked_queue_rbytes > LFS_MAX_BYTES) { 462 DLOG((DLOG_FLUSH, "lfs_writerd: lqb + lqrb = %ld, max %ld\n", 463 locked_queue_bytes, LFS_MAX_BYTES)); 464 } 465 if (lfs_subsys_pages > LFS_MAX_PAGES) { 466 DLOG((DLOG_FLUSH, "lfs_writerd: lssp = %d, max %d\n", 467 lfs_subsys_pages, LFS_MAX_PAGES)); 468 } 469 470 lfs_flush(NULL, SEGM_WRITERD, 0); 471 lfs_do_flush = 0; 472 KASSERT(mutex_owned(&lfs_lock)); 473 continue; 474 } 475 KASSERT(mutex_owned(&lfs_lock)); 476 mutex_exit(&lfs_lock); 477 478 /* 479 * Look through the list of LFSs to see if any of them 480 * have requested pageouts. 481 */ 482 mountlist_iterator_init(&iter); 483 lfsc = 0; 484 while ((mp = mountlist_iterator_next(iter)) != NULL) { 485 KASSERT(!mutex_owned(&lfs_lock)); 486 if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS, 487 sizeof(mp->mnt_stat.f_fstypename)) == 0) { 488 ++lfsc; 489 fs = VFSTOULFS(mp)->um_lfs; 490 daddr_t ooffset = 0; 491 fsflags = SEGM_SINGLE; 492 493 mutex_enter(&lfs_lock); 494 ooffset = lfs_sb_getoffset(fs); 495 496 if (lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs) && fs->lfs_nowrap) { 497 /* Don't try to write if we're suspended */ 498 mutex_exit(&lfs_lock); 499 continue; 500 } 501 if (LFS_STARVED_FOR_SEGS(fs)) { 502 mutex_exit(&lfs_lock); 503 504 DLOG((DLOG_FLUSH, "lfs_writerd: need cleaning before writing possible\n")); 505 lfs_wakeup_cleaner(fs); 506 continue; 507 } 508 509 if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || 510 lfs_dirvcount > LFS_MAX_DIROP) && 511 fs->lfs_dirops == 0) { 512 fsflags &= ~SEGM_SINGLE; 513 fsflags |= SEGM_CKP; 514 DLOG((DLOG_FLUSH, "lfs_writerd: checkpoint\n")); 515 lfs_flush_fs(fs, fsflags); 516 } else if (fs->lfs_pdflush) { 517 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n")); 518 lfs_flush_fs(fs, fsflags); 519 } else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) { 520 DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n")); 521 mutex_exit(&lfs_lock); 522 lfs_writer_enter(fs, "wrdirop"); 523 lfs_flush_pchain(fs); 524 lfs_writer_leave(fs); 525 mutex_enter(&lfs_lock); 526 } 527 if (lfs_sb_getoffset(fs) != ooffset) 528 ++wrote_something; 529 mutex_exit(&lfs_lock); 530 } 531 KASSERT(!mutex_owned(&lfs_lock)); 532 } 533 if (lfsc == 0) { 534 mutex_enter(&lfs_lock); 535 lfs_writer_daemon = NULL; 536 mutex_exit(&lfs_lock); 537 mountlist_iterator_destroy(iter); 538 break; 539 } 540 mountlist_iterator_destroy(iter); 541 542 mutex_enter(&lfs_lock); 543 } 544 KASSERT(!mutex_owned(&lfs_lock)); 545 546 /* Give up our extra reference so the module can be unloaded. */ 547 mutex_enter(&vfs_list_lock); 548 if (vfs != NULL) 549 vfs->vfs_refcount--; 550 mutex_exit(&vfs_list_lock); 551 552 /* Done! */ 553 kthread_exit(0); 554 } 555 556 /* 557 * Initialize the filesystem, most work done by ulfs_init. 558 */ 559 void 560 lfs_init(void) 561 { 562 563 /* 564 * XXX: should we use separate pools for 32-bit and 64-bit 565 * dinodes? 566 */ 567 malloc_type_attach(M_SEGMENT); 568 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 569 "lfsinopl", &pool_allocator_nointr, IPL_NONE); 570 pool_init(&lfs_dinode_pool, sizeof(union lfs_dinode), 0, 0, 0, 571 "lfsdinopl", &pool_allocator_nointr, IPL_NONE); 572 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 573 "lfsinoextpl", &pool_allocator_nointr, IPL_NONE); 574 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 575 "lfslbnpool", &pool_allocator_nointr, IPL_NONE); 576 ulfs_init(); 577 578 #ifdef DEBUG 579 memset(lfs_log, 0, sizeof(lfs_log)); 580 #endif 581 mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE); 582 cv_init(&lfs_writerd_cv, "lfswrite"); 583 cv_init(&locked_queue_cv, "lfsbuf"); 584 cv_init(&lfs_writing_cv, "lfsflush"); 585 workqueue_create(&lfs_cluster_wq, "lfscwq", lfs_cluster_work, NULL, 586 PRI_BIO, IPL_BIO, WQ_MPSAFE); 587 workqueue_create(&lfs_super_wq, "lfsswq", lfs_super_work, NULL, 588 PRI_BIO, IPL_BIO, WQ_MPSAFE); 589 } 590 591 void 592 lfs_reinit(void) 593 { 594 ulfs_reinit(); 595 } 596 597 void 598 lfs_done(void) 599 { 600 ulfs_done(); 601 mutex_destroy(&lfs_lock); 602 cv_destroy(&lfs_writerd_cv); 603 cv_destroy(&locked_queue_cv); 604 cv_destroy(&lfs_writing_cv); 605 workqueue_destroy(lfs_cluster_wq); 606 workqueue_destroy(lfs_super_wq); 607 pool_destroy(&lfs_inode_pool); 608 pool_destroy(&lfs_dinode_pool); 609 pool_destroy(&lfs_inoext_pool); 610 pool_destroy(&lfs_lbnentry_pool); 611 malloc_type_detach(M_SEGMENT); 612 } 613 614 /* 615 * Called by main() when ulfs is going to be mounted as root. 616 */ 617 int 618 lfs_mountroot(void) 619 { 620 extern struct vnode *rootvp; 621 struct lfs *fs = NULL; /* LFS */ 622 struct mount *mp; 623 struct lwp *l = curlwp; 624 struct ulfsmount *ump; 625 int error; 626 627 if (device_class(root_device) != DV_DISK) 628 return (ENODEV); 629 630 if (rootdev == NODEV) 631 return (ENODEV); 632 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 633 vrele(rootvp); 634 return (error); 635 } 636 if ((error = lfs_mountfs(rootvp, mp, l))) { 637 vfs_unbusy(mp); 638 vfs_rele(mp); 639 return (error); 640 } 641 mountlist_append(mp); 642 ump = VFSTOULFS(mp); 643 fs = ump->um_lfs; 644 lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname); 645 (void)lfs_statvfs(mp, &mp->mnt_stat); 646 vfs_unbusy(mp); 647 setrootfstime((time_t)lfs_sb_gettstamp(VFSTOULFS(mp)->um_lfs)); 648 return (0); 649 } 650 651 /* 652 * VFS Operations. 653 * 654 * mount system call 655 */ 656 int 657 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 658 { 659 struct lwp *l = curlwp; 660 struct vnode *devvp; 661 struct ulfs_args *args = data; 662 struct ulfsmount *ump = NULL; 663 struct lfs *fs = NULL; /* LFS */ 664 int error = 0, update; 665 mode_t accessmode; 666 667 if (args == NULL) 668 return EINVAL; 669 if (*data_len < sizeof *args) 670 return EINVAL; 671 672 if (mp->mnt_flag & MNT_GETARGS) { 673 ump = VFSTOULFS(mp); 674 if (ump == NULL) 675 return EIO; 676 args->fspec = NULL; 677 *data_len = sizeof *args; 678 return 0; 679 } 680 681 update = mp->mnt_flag & MNT_UPDATE; 682 683 /* Check arguments */ 684 if (args->fspec != NULL) { 685 /* 686 * Look up the name and verify that it's sane. 687 */ 688 error = namei_simple_user(args->fspec, 689 NSM_FOLLOW_NOEMULROOT, &devvp); 690 if (error != 0) 691 return (error); 692 693 if (!update) { 694 /* 695 * Be sure this is a valid block device 696 */ 697 if (devvp->v_type != VBLK) 698 error = ENOTBLK; 699 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 700 error = ENXIO; 701 } else { 702 /* 703 * Be sure we're still naming the same device 704 * used for our initial mount 705 * 706 * XXX dholland 20151010: if namei gives us a 707 * different vnode for the same device, 708 * wouldn't it be better to use it going 709 * forward rather than ignore it in favor of 710 * the old one? 711 */ 712 ump = VFSTOULFS(mp); 713 fs = ump->um_lfs; 714 if (devvp != fs->lfs_devvp) { 715 if (devvp->v_rdev != fs->lfs_devvp->v_rdev) 716 error = EINVAL; 717 else { 718 vrele(devvp); 719 devvp = fs->lfs_devvp; 720 vref(devvp); 721 } 722 } 723 } 724 } else { 725 if (!update) { 726 /* New mounts must have a filename for the device */ 727 return (EINVAL); 728 } else { 729 /* Use the extant mount */ 730 ump = VFSTOULFS(mp); 731 fs = ump->um_lfs; 732 devvp = fs->lfs_devvp; 733 vref(devvp); 734 } 735 } 736 737 738 /* 739 * If mount by non-root, then verify that user has necessary 740 * permissions on the device. 741 */ 742 if (error == 0) { 743 accessmode = VREAD; 744 if (update ? 745 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 746 (mp->mnt_flag & MNT_RDONLY) == 0) 747 accessmode |= VWRITE; 748 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 749 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT, 750 KAUTH_REQ_SYSTEM_MOUNT_DEVICE, mp, devvp, 751 KAUTH_ARG(accessmode)); 752 VOP_UNLOCK(devvp); 753 } 754 755 if (error) { 756 vrele(devvp); 757 return (error); 758 } 759 760 if (!update) { 761 int flags; 762 763 if (mp->mnt_flag & MNT_RDONLY) 764 flags = FREAD; 765 else 766 flags = FREAD|FWRITE; 767 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 768 error = VOP_OPEN(devvp, flags, FSCRED); 769 VOP_UNLOCK(devvp); 770 if (error) 771 goto fail; 772 error = lfs_mountfs(devvp, mp, l); /* LFS */ 773 if (error) { 774 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 775 (void)VOP_CLOSE(devvp, flags, NOCRED); 776 VOP_UNLOCK(devvp); 777 goto fail; 778 } 779 780 ump = VFSTOULFS(mp); 781 fs = ump->um_lfs; 782 } else { 783 /* 784 * Update the mount. 785 */ 786 787 /* 788 * The initial mount got a reference on this 789 * device, so drop the one obtained via 790 * namei(), above. 791 */ 792 vrele(devvp); 793 794 ump = VFSTOULFS(mp); 795 fs = ump->um_lfs; 796 797 if (!fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDONLY)) { 798 /* 799 * Changing from read/write to read-only. 800 */ 801 int flags = WRITECLOSE; 802 if (mp->mnt_flag & MNT_FORCE) 803 flags |= FORCECLOSE; 804 error = lfs_flushfiles(mp, flags); 805 if (error) 806 return error; 807 fs->lfs_ronly = 1; 808 809 /* Wake up reclaimd so it can die */ 810 cv_broadcast(&fs->lfs_reclaim_cv); 811 kthread_join(fs->lfs_reclaimd_lwp); 812 813 /* 814 * Drain pending I/O now that we have declared 815 * read-only. We compare against 1 because 816 * seglock increments by one. 817 */ 818 lfs_prelock(fs, 0); 819 mutex_enter(&lfs_lock); 820 while (fs->lfs_iocount > 1) 821 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, 822 "lfs_roio", 0, &lfs_lock); 823 mutex_exit(&lfs_lock); 824 lfs_preunlock(fs); 825 } else if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 826 /* 827 * Changing from read-only to read/write. 828 * Note in the superblocks that we're writing. 829 */ 830 831 /* Start reclaimd */ 832 error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 833 lfs_reclaimd, mp, &fs->lfs_reclaimd_lwp, 834 "lfs_reclaim"); 835 if (error) 836 return error; 837 838 /* XXX: quotas should have been on even if readonly */ 839 if (fs->lfs_use_quota2) { 840 #ifdef LFS_QUOTA2 841 error = lfs_quota2_mount(mp); 842 #else 843 uprintf("%s: no kernel support for this " 844 "filesystem's quotas\n", 845 mp->mnt_stat.f_mntonname); 846 if (mp->mnt_flag & MNT_FORCE) { 847 uprintf("%s: mounting anyway; " 848 "fsck afterwards\n", 849 mp->mnt_stat.f_mntonname); 850 } else { 851 error = EINVAL; 852 } 853 #endif 854 if (error) { 855 return error; 856 } 857 } 858 859 fs->lfs_ronly = 0; 860 if (lfs_sb_getpflags(fs) & LFS_PF_CLEAN) { 861 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN); 862 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 863 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 864 } 865 } 866 867 if (args->fspec == NULL) 868 return 0; 869 } 870 871 error = set_statvfs_info(path, UIO_USERSPACE, args->fspec, 872 UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l); 873 if (error == 0) 874 lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname); 875 return error; 876 877 fail: 878 vrele(devvp); 879 return (error); 880 } 881 882 /* 883 * Helper for mountfs. Note that the fs pointer may be a dummy one 884 * pointing into a superblock buffer. (Which is gross; see below.) 885 */ 886 static int 887 lfs_checkmagic(struct lfs *fs) 888 { 889 switch (fs->lfs_dlfs_u.u_32.dlfs_magic) { 890 case LFS_MAGIC: 891 fs->lfs_is64 = false; 892 fs->lfs_dobyteswap = false; 893 break; 894 case LFS64_MAGIC: 895 fs->lfs_is64 = true; 896 fs->lfs_dobyteswap = false; 897 break; 898 #ifdef LFS_EI 899 case LFS_MAGIC_SWAPPED: 900 fs->lfs_is64 = false; 901 fs->lfs_dobyteswap = true; 902 break; 903 case LFS64_MAGIC_SWAPPED: 904 fs->lfs_is64 = true; 905 fs->lfs_dobyteswap = true; 906 break; 907 #endif 908 default: 909 /* XXX needs translation */ 910 return EINVAL; 911 } 912 return 0; 913 } 914 915 /* 916 * Common code for mount and mountroot 917 * LFS specific 918 */ 919 int 920 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 921 { 922 struct lfs *primarysb, *altsb, *thesb; 923 struct buf *primarybuf, *altbuf; 924 struct lfs *fs; 925 struct ulfsmount *ump; 926 struct vnode *vp; 927 dev_t dev; 928 int error, i, ronly, fsbsize, rfw; 929 kauth_cred_t cred; 930 CLEANERINFO *cip; 931 SEGUSE *sup; 932 daddr_t sb_addr; 933 934 cred = l ? l->l_cred : NOCRED; 935 936 /* The superblock is supposed to be 512 bytes. */ 937 __CTASSERT(sizeof(struct dlfs) == DEV_BSIZE); 938 939 /* 940 * Flush out any old buffers remaining from a previous use. 941 */ 942 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 943 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 944 VOP_UNLOCK(devvp); 945 if (error) 946 return (error); 947 948 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 949 rfw = (mp->mnt_flag & MNT_LOG) != 0; 950 951 /* Don't free random space on error. */ 952 primarybuf = NULL; 953 altbuf = NULL; 954 ump = NULL; 955 956 sb_addr = LFS_LABELPAD / DEV_BSIZE; 957 while (1) { 958 /* 959 * Read in the superblock. 960 * 961 * Note that because LFS_SBPAD is substantially larger 962 * (8K) than the actual on-disk superblock (512 bytes) 963 * the buffer contains enough space to be used as a 964 * whole struct lfs (in-memory superblock) - we do this 965 * only so we can set and use the is64 and dobyteswap 966 * members. XXX this is gross and the logic here should 967 * be reworked. 968 */ 969 error = bread(devvp, sb_addr, LFS_SBPAD, 0, &primarybuf); 970 if (error) 971 goto out; 972 primarysb = (struct lfs *)primarybuf->b_data; 973 974 /* Check the basics. */ 975 error = lfs_checkmagic(primarysb); 976 if (error) { 977 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock wrong magic\n")); 978 goto out; 979 } 980 if (lfs_sb_getbsize(primarysb) > MAXBSIZE || 981 lfs_sb_getversion(primarysb) > LFS_VERSION || 982 lfs_sb_getbsize(primarysb) < sizeof(struct dlfs)) { 983 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 984 /* XXX needs translation */ 985 error = EINVAL; 986 goto out; 987 } 988 if (lfs_sb_getinodefmt(primarysb) > LFS_MAXINODEFMT) { 989 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 990 lfs_sb_getinodefmt(primarysb))); 991 error = EINVAL; 992 goto out; 993 } 994 995 if (lfs_sb_getversion(primarysb) == 1) 996 fsbsize = DEV_BSIZE; 997 else { 998 fsbsize = 1 << lfs_sb_getffshift(primarysb); 999 /* 1000 * Could be, if the frag size is large enough, that we 1001 * don't have the "real" primary superblock. If that's 1002 * the case, get the real one, and try again. 1003 */ 1004 if (sb_addr != (lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT))) { 1005 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 1006 " 0x%llx is not right, trying 0x%llx\n", 1007 (long long)sb_addr, 1008 (long long)(lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT)))); 1009 sb_addr = lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT); 1010 brelse(primarybuf, BC_INVAL); 1011 continue; 1012 } 1013 } 1014 break; 1015 } 1016 1017 /* 1018 * Check the second superblock to see which is newer; then mount 1019 * using the older of the two. This is necessary to ensure that 1020 * the filesystem is valid if it was not unmounted cleanly. 1021 */ 1022 1023 if (lfs_sb_getsboff(primarysb, 1) && 1024 lfs_sb_getsboff(primarysb, 1) - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 1025 { 1026 error = bread(devvp, lfs_sb_getsboff(primarysb, 1) * (fsbsize / DEV_BSIZE), 1027 LFS_SBPAD, 0, &altbuf); 1028 if (error) 1029 goto out; 1030 altsb = (struct lfs *)altbuf->b_data; 1031 1032 /* 1033 * Note: this used to do the sanity check only if the 1034 * timestamp/serial comparison required use of altsb; 1035 * this way is less tolerant, but if altsb is corrupted 1036 * enough that the magic number, version, and blocksize 1037 * are bogus, why would the timestamp or serial fields 1038 * mean anything either? If this kind of thing happens, 1039 * you need to fsck anyway. 1040 */ 1041 1042 error = lfs_checkmagic(altsb); 1043 if (error) 1044 goto out; 1045 1046 /* Check the basics. */ 1047 if (lfs_sb_getbsize(altsb) > MAXBSIZE || 1048 lfs_sb_getversion(altsb) > LFS_VERSION || 1049 lfs_sb_getbsize(altsb) < sizeof(struct dlfs)) { 1050 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 1051 " sanity failed\n")); 1052 error = EINVAL; /* XXX needs translation */ 1053 goto out; 1054 } 1055 1056 if (lfs_sb_getversion(primarysb) == 1) { 1057 /* 1s resolution comparison */ 1058 if (lfs_sb_gettstamp(altsb) < lfs_sb_gettstamp(primarysb)) 1059 thesb = altsb; 1060 else 1061 thesb = primarysb; 1062 } else { 1063 /* monotonic infinite-resolution comparison */ 1064 if (lfs_sb_getserial(altsb) < lfs_sb_getserial(primarysb)) 1065 thesb = altsb; 1066 else 1067 thesb = primarysb; 1068 } 1069 } else { 1070 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock location" 1071 " daddr=0x%x\n", lfs_sb_getsboff(primarysb, 1))); 1072 error = EINVAL; 1073 goto out; 1074 } 1075 1076 /* 1077 * Allocate the mount structure, copy the superblock into it. 1078 * Note that the 32-bit and 64-bit superblocks are the same size. 1079 */ 1080 fs = kmem_zalloc(sizeof(struct lfs), KM_SLEEP); 1081 memcpy(&fs->lfs_dlfs_u.u_32, &thesb->lfs_dlfs_u.u_32, 1082 sizeof(struct dlfs)); 1083 fs->lfs_is64 = thesb->lfs_is64; 1084 fs->lfs_dobyteswap = thesb->lfs_dobyteswap; 1085 fs->lfs_hasolddirfmt = false; /* set for real below */ 1086 1087 /* Compatibility */ 1088 if (lfs_sb_getversion(fs) < 2) { 1089 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE); 1090 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs)); 1091 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0)); 1092 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs)); 1093 lfs_sb_setfsbtodb(fs, 0); 1094 } 1095 if (lfs_sb_getresvseg(fs) == 0) 1096 lfs_sb_setresvseg(fs, MIN(lfs_sb_getminfreeseg(fs) - 1, \ 1097 MAX(MIN_RESV_SEGS, lfs_sb_getminfreeseg(fs) / 2 + 1))); 1098 1099 /* 1100 * If we aren't going to be able to write meaningfully to this 1101 * filesystem, and were not mounted readonly, bomb out now. 1102 */ 1103 if (lfs_fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 1104 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 1105 " we need BUFPAGES >= %lld\n", 1106 (long long)((bufmem_hiwater / bufmem_lowater) * 1107 LFS_INVERSE_MAX_BYTES( 1108 lfs_fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 1109 kmem_free(fs, sizeof(struct lfs)); 1110 error = EFBIG; /* XXX needs translation */ 1111 goto out; 1112 } 1113 1114 /* Before rolling forward, lock so vget will sleep for other procs */ 1115 if (l != NULL) { 1116 fs->lfs_flags = LFS_NOTYET; 1117 fs->lfs_rfpid = l->l_proc->p_pid; 1118 } 1119 1120 ump = kmem_zalloc(sizeof(*ump), KM_SLEEP); 1121 ump->um_lfs = fs; 1122 ump->um_fstype = fs->lfs_is64 ? ULFS2 : ULFS1; 1123 /* ump->um_cleaner_thread = NULL; */ 1124 brelse(primarybuf, BC_INVAL); 1125 brelse(altbuf, BC_INVAL); 1126 primarybuf = NULL; 1127 altbuf = NULL; 1128 1129 1130 /* Set up the I/O information */ 1131 fs->lfs_devbsize = DEV_BSIZE; 1132 fs->lfs_iocount = 0; 1133 fs->lfs_diropwait = 0; 1134 fs->lfs_activesb = 0; 1135 lfs_sb_setuinodes(fs, 0); 1136 fs->lfs_ravail = 0; 1137 fs->lfs_favail = 0; 1138 fs->lfs_sbactive = 0; 1139 1140 /* Set up the ifile and lock aflags */ 1141 fs->lfs_doifile = 0; 1142 fs->lfs_writer = 0; 1143 fs->lfs_dirops = 0; 1144 fs->lfs_nadirop = 0; 1145 fs->lfs_prelock = 0; 1146 fs->lfs_seglock = 0; 1147 fs->lfs_cleanlock = 0; 1148 fs->lfs_pdflush = 0; 1149 fs->lfs_sleepers = 0; 1150 fs->lfs_pages = 0; 1151 fs->lfs_prelocklwp = NULL; 1152 rw_init(&fs->lfs_fraglock); 1153 rw_init(&fs->lfs_iflock); 1154 cv_init(&fs->lfs_sleeperscv, "lfs_slp"); 1155 cv_init(&fs->lfs_diropscv, "lfs_dirop"); 1156 cv_init(&fs->lfs_stopcv, "lfsstop"); 1157 cv_init(&fs->lfs_nextsegsleep, "segment"); 1158 cv_init(&fs->lfs_cleanercv, "cleancv"); 1159 cv_init(&fs->lfs_prelockcv, "prelockcv"); 1160 cv_init(&fs->lfs_cleanquitcv, "cleanquit"); 1161 cv_init(&fs->lfs_reclaim_cv, "seg reclaim"); 1162 1163 /* Set the file system readonly/modify bits. */ 1164 fs->lfs_ronly = ronly; 1165 if (ronly == 0) 1166 fs->lfs_fmod = 1; 1167 1168 /* Device we're using */ 1169 dev = devvp->v_rdev; 1170 fs->lfs_dev = dev; 1171 fs->lfs_devvp = devvp; 1172 1173 /* ulfs-level information */ 1174 fs->um_flags = 0; 1175 fs->um_bptrtodb = lfs_sb_getffshift(fs) - DEV_BSHIFT; 1176 fs->um_seqinc = lfs_sb_getfrag(fs); 1177 fs->um_nindir = lfs_sb_getnindir(fs); 1178 fs->um_lognindir = ffs(lfs_sb_getnindir(fs)) - 1; 1179 fs->um_maxsymlinklen = lfs_sb_getmaxsymlinklen(fs); 1180 fs->um_dirblksiz = LFS_DIRBLKSIZ; 1181 fs->um_maxfilesize = lfs_sb_getmaxfilesize(fs); 1182 1183 /* quota stuff */ 1184 /* XXX: these need to come from the on-disk superblock to be used */ 1185 fs->lfs_use_quota2 = 0; 1186 fs->lfs_quota_magic = 0; 1187 fs->lfs_quota_flags = 0; 1188 fs->lfs_quotaino[0] = 0; 1189 fs->lfs_quotaino[1] = 0; 1190 1191 /* Initialize the mount structure. */ 1192 mp->mnt_data = ump; 1193 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 1194 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 1195 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 1196 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 1197 mp->mnt_stat.f_iosize = lfs_sb_getbsize(fs); 1198 mp->mnt_flag |= MNT_LOCAL; 1199 mp->mnt_iflag |= IMNT_SHRLOOKUP; 1200 mp->mnt_fs_bshift = lfs_sb_getbshift(fs); 1201 mp->mnt_iflag |= IMNT_CAN_RWTORO; 1202 if (fs->um_maxsymlinklen > 0) 1203 mp->mnt_iflag |= IMNT_DTYPE; 1204 else 1205 fs->lfs_hasolddirfmt = true; 1206 1207 ump->um_mountp = mp; 1208 for (i = 0; i < ULFS_MAXQUOTAS; i++) 1209 ump->um_quotas[i] = NULLVP; 1210 spec_node_setmountedfs(devvp, mp); 1211 1212 /* Set up reserved memory for pageout */ 1213 lfs_setup_resblks(fs); 1214 /* Set up vdirop tailq */ 1215 TAILQ_INIT(&fs->lfs_dchainhd); 1216 /* and cleaning vnode tailq */ 1217 TAILQ_INIT(&fs->lfs_cleanhd); 1218 /* and paging tailq */ 1219 TAILQ_INIT(&fs->lfs_pchainhd); 1220 /* and delayed segment accounting for truncation list */ 1221 rb_tree_init(&fs->lfs_segdhd, &lfs_rbtree_ops); 1222 1223 /* 1224 * We use the ifile vnode for almost every operation. Instead of 1225 * retrieving it from the hash table each time we retrieve it here, 1226 * artificially increment the reference count and keep a pointer 1227 * to it in the incore copy of the superblock. 1228 */ 1229 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, LK_EXCLUSIVE, &vp)) != 0) { 1230 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 1231 goto out; 1232 } 1233 fs->lfs_ivnode = vp; 1234 vref(vp); 1235 1236 /* Set up segment usage flags for the autocleaner. */ 1237 fs->lfs_nactive = 0; 1238 for (i = 0; i < lfs_sb_getnseg(fs); i++) { 1239 int changed; 1240 struct buf *bp; 1241 1242 LFS_SEGENTRY(sup, fs, i, bp); 1243 changed = 0; 1244 if (!ronly) { 1245 uint32_t nflags = sup->su_flags; 1246 nflags &= ~(SEGUSE_ACTIVE | SEGUSE_EMPTY 1247 | SEGUSE_READY | SEGUSE_INVAL 1248 | SEGUSE_ERROR); 1249 if (sup->su_nbytes == 0) { 1250 if (sup->su_flags & SEGUSE_DIRTY) 1251 nflags |= SEGUSE_EMPTY | SEGUSE_READY; 1252 } else { 1253 nflags &= ~(SEGUSE_EMPTY | SEGUSE_READY); 1254 } 1255 if (nflags != sup->su_flags) { 1256 sup->su_flags = nflags; 1257 ++changed; 1258 } 1259 } 1260 if (changed) 1261 LFS_WRITESEGENTRY(sup, fs, i, bp); 1262 else 1263 LFS_RELEASESEGENTRY(sup, fs, i, bp); 1264 } 1265 /* Set lastcleaned to an invalid value */ 1266 fs->lfs_lastcleaned = (uint32_t)-1; 1267 1268 if (rfw && !ronly) { 1269 /* Roll forward */ 1270 lfs_roll_forward(fs, mp, l); 1271 lfs_reset_avail(fs); 1272 } 1273 1274 /* Init reclaimd */ 1275 if (!ronly) { 1276 error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 1277 lfs_reclaimd, mp, &fs->lfs_reclaimd_lwp, "lfs_reclaim"); 1278 if (error != 0) 1279 goto out; 1280 } 1281 fs->lfs_rfpid = 0; 1282 1283 /* 1284 * XXX: if the fs has quotas, quotas should be on even if 1285 * readonly. Otherwise you can't query the quota info! 1286 * However, that's not how the quota2 code got written and I 1287 * don't know if it'll behave itself if enabled while 1288 * readonly, so for now use the same enable logic as ffs. 1289 * 1290 * XXX: also, if you use the -f behavior allowed here (and 1291 * equivalently above for remount) it will corrupt the fs. It 1292 * ought not to allow that. It should allow mounting readonly 1293 * if there are quotas and the kernel doesn't have the quota 1294 * code, but only readonly. 1295 * 1296 * XXX: and if you use the -f behavior allowed here it will 1297 * likely crash at unmount time (or remount time) because we 1298 * think quotas are active. 1299 * 1300 * Although none of this applies until there's a way to set 1301 * lfs_use_quota2 and have quotas in the fs at all. 1302 */ 1303 if (!ronly && fs->lfs_use_quota2) { 1304 #ifdef LFS_QUOTA2 1305 error = lfs_quota2_mount(mp); 1306 #else 1307 uprintf("%s: no kernel support for this filesystem's quotas\n", 1308 mp->mnt_stat.f_mntonname); 1309 if (mp->mnt_flag & MNT_FORCE) { 1310 uprintf("%s: mounting anyway; fsck afterwards\n", 1311 mp->mnt_stat.f_mntonname); 1312 } else { 1313 error = EINVAL; 1314 } 1315 #endif 1316 if (error) { 1317 /* XXX XXX must clean up the stuff immediately above */ 1318 printf("lfs_mountfs: sorry, leaking some memory\n"); 1319 goto out; 1320 } 1321 } 1322 1323 /* If writing, sb is not clean; record in case of immediate crash */ 1324 if (!fs->lfs_ronly) { 1325 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN); 1326 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 1327 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 1328 } 1329 1330 /* Allow vget now that roll-forward is complete */ 1331 fs->lfs_flags &= ~(LFS_NOTYET); 1332 wakeup(&fs->lfs_flags); 1333 1334 /* 1335 * Initialize the ifile cleaner info with information from 1336 * the superblock. 1337 */ 1338 { 1339 struct buf *bp; 1340 1341 LFS_CLEANERINFO(cip, fs, bp); 1342 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs)); 1343 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs)); 1344 lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs)); 1345 lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs)); 1346 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1347 } 1348 1349 /* 1350 * Mark the current segment as ACTIVE, since we're going to 1351 * be writing to it. 1352 */ 1353 { 1354 struct buf *bp; 1355 1356 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); 1357 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1358 fs->lfs_nactive++; 1359 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); /* Ifile */ 1360 } 1361 1362 /* Now that roll-forward is done, unlock the Ifile */ 1363 vput(vp); 1364 1365 /* Free any file orphaned during a crash */ 1366 lfs_free_orphans(fs); 1367 1368 /* Start the pagedaemon-anticipating daemon */ 1369 mutex_enter(&lfs_lock); 1370 if (lfs_writer_daemon == NULL && 1371 kthread_create(PRI_BIO, 0, NULL, 1372 lfs_writerd, NULL, NULL, "lfs_writer") != 0) 1373 panic("fork lfs_writer"); 1374 mutex_exit(&lfs_lock); 1375 1376 printf("WARNING: the log-structured file system is experimental\n" 1377 "WARNING: it may cause system crashes and/or corrupt data\n"); 1378 1379 return (0); 1380 1381 out: 1382 if (primarybuf) 1383 brelse(primarybuf, BC_INVAL); 1384 if (altbuf) 1385 brelse(altbuf, BC_INVAL); 1386 if (ump) { 1387 kmem_free(ump->um_lfs, sizeof(struct lfs)); 1388 kmem_free(ump, sizeof(*ump)); 1389 mp->mnt_data = NULL; 1390 } 1391 1392 return (error); 1393 } 1394 1395 static void 1396 lfs_reclaimd(void *v) 1397 { 1398 struct mount *mp = (struct mount *)v; 1399 struct lfs *fs = VFSTOULFS(mp)->um_lfs; 1400 1401 for (;;) { 1402 mutex_enter(&lfs_lock); 1403 cv_wait(&fs->lfs_reclaim_cv, &lfs_lock); 1404 if (mp->mnt_iflag & (IMNT_UNMOUNT|IMNT_WANTRDONLY)) 1405 goto out; 1406 mutex_exit(&lfs_lock); 1407 1408 EXCL_IFLOCK(fs); 1409 lfs_auto_segclean(fs); 1410 UNEXCL_IFLOCK(fs); 1411 } 1412 1413 out: 1414 mutex_exit(&lfs_lock); 1415 kthread_exit(0); 1416 } 1417 1418 void 1419 lfs_reset_avail(struct lfs *fs) 1420 { 1421 daddr_t avail; 1422 int sn, nclean, nsb, labelcorrect, curr, mfs; 1423 SEGUSE *sup; 1424 struct buf *bp; 1425 1426 KASSERT(!fs->lfs_ronly); 1427 1428 avail = nclean = nsb = labelcorrect = 0; 1429 for (sn = 0; sn < lfs_sb_getnseg(fs); ++sn) { 1430 LFS_SEGENTRY(sup, fs, sn, bp); 1431 1432 /* Count all clean segments and the remainder of this one */ 1433 if (!(sup->su_flags & SEGUSE_DIRTY)) { 1434 ++nclean; 1435 avail += lfs_segtod(fs, 1); 1436 1437 /* Correct for label and superblock, if present */ 1438 if (sup->su_flags & SEGUSE_SUPERBLOCK) 1439 ++nsb; 1440 if (sn == 0 && lfs_sb_getversion(fs) > 1 1441 && lfs_sb_gets0addr(fs) < lfs_btofsb(fs, LFS_LABELPAD)) 1442 labelcorrect = lfs_btofsb(fs, LFS_LABELPAD) 1443 - lfs_sb_gets0addr(fs); 1444 } 1445 1446 LFS_RELEASESEGENTRY(sup, fs, sn, bp); 1447 } 1448 1449 /* Also may be available bytes in current seg */ 1450 sn = lfs_dtosn(fs, lfs_sb_getoffset(fs)); 1451 curr = lfs_sntod(fs, sn + 1) - lfs_sb_getoffset(fs); 1452 /* But do not count minfreesegs */ 1453 mfs = lfs_segtod(fs, (lfs_sb_getminfreeseg(fs) - 1454 (lfs_sb_getminfreeseg(fs) / 2))); 1455 1456 avail = nclean * lfs_segtod(fs, 1); 1457 avail -= nsb * lfs_btofsb(fs, LFS_SBPAD); 1458 avail -= labelcorrect; 1459 avail += curr; 1460 avail -= mfs; 1461 1462 DLOG((DLOG_AVAIL, "avail := %jd*%jd-%jd*%jd-%jd+%jd-%jd=%jd\n", 1463 (intmax_t)nclean, 1464 (intmax_t)lfs_segtod(fs, 1), 1465 (intmax_t)nsb, 1466 (intmax_t)lfs_btofsb(fs, LFS_SBPAD), 1467 (intmax_t)labelcorrect, 1468 (intmax_t)curr, 1469 (intmax_t)mfs, 1470 (intmax_t)avail)); 1471 1472 lfs_sb_setavail(fs, avail); 1473 lfs_sb_setnclean(fs, nclean); 1474 } 1475 1476 /* 1477 * unmount system call 1478 */ 1479 int 1480 lfs_unmount(struct mount *mp, int mntflags) 1481 { 1482 struct ulfsmount *ump; 1483 struct lfs *fs; 1484 int error, ronly; 1485 1486 ump = VFSTOULFS(mp); 1487 fs = ump->um_lfs; 1488 1489 if (!fs->lfs_ronly) { 1490 /* Wake up reclaimd so it can die */ 1491 cv_broadcast(&fs->lfs_reclaim_cv); 1492 kthread_join(fs->lfs_reclaimd_lwp); 1493 } 1494 1495 error = lfs_flushfiles(mp, mntflags & MNT_FORCE ? FORCECLOSE : 0); 1496 if (error) 1497 return error; 1498 1499 DEBUG_CHECK_FREELIST(fs); 1500 1501 /* Check for dirty blocks on ifile */ 1502 KASSERT(LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd) == NULL); 1503 1504 /* Finish with the Ifile, now that we're done with it */ 1505 vgone(fs->lfs_ivnode); 1506 1507 ronly = !fs->lfs_ronly; 1508 if (fs->lfs_devvp->v_type != VBAD) 1509 spec_node_setmountedfs(fs->lfs_devvp, NULL); 1510 vn_lock(fs->lfs_devvp, LK_EXCLUSIVE | LK_RETRY); 1511 error = VOP_CLOSE(fs->lfs_devvp, 1512 ronly ? FREAD : FREAD|FWRITE, NOCRED); 1513 vput(fs->lfs_devvp); 1514 1515 /* Complain about page leakage */ 1516 if (fs->lfs_pages > 0) 1517 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 1518 fs->lfs_pages, lfs_subsys_pages); 1519 1520 /* 1521 * Make doubly sure we have no outstanding I/O before we 1522 * call lfs_free_resblks(), to avoid a busy pool panic. 1523 */ 1524 mutex_enter(&lfs_lock); 1525 while (fs->lfs_iocount) 1526 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount2", 0, 1527 &lfs_lock); 1528 mutex_exit(&lfs_lock); 1529 1530 /* Free per-mount data structures */ 1531 lfs_free_resblks(fs); 1532 cv_destroy(&fs->lfs_sleeperscv); 1533 cv_destroy(&fs->lfs_diropscv); 1534 cv_destroy(&fs->lfs_stopcv); 1535 cv_destroy(&fs->lfs_nextsegsleep); 1536 cv_destroy(&fs->lfs_reclaim_cv); 1537 1538 rw_destroy(&fs->lfs_fraglock); 1539 rw_destroy(&fs->lfs_iflock); 1540 1541 kmem_free(fs, sizeof(struct lfs)); 1542 kmem_free(ump, sizeof(*ump)); 1543 1544 mp->mnt_data = NULL; 1545 mp->mnt_flag &= ~MNT_LOCAL; 1546 return (error); 1547 } 1548 1549 static int 1550 lfs_flushfiles(struct mount *mp, int flags) 1551 { 1552 struct lwp *l = curlwp; 1553 struct ulfsmount *ump; 1554 struct lfs *fs; 1555 struct vnode *vp; 1556 int error; 1557 1558 ump = VFSTOULFS(mp); 1559 fs = ump->um_lfs; 1560 1561 /* Two checkpoints */ 1562 if (!fs->lfs_ronly) { 1563 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1564 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1565 } 1566 1567 /* wake up the cleaner so it can die */ 1568 /* XXX: shouldn't this be *after* the error cases below? */ 1569 lfs_wakeup_cleaner(fs); 1570 mutex_enter(&lfs_lock); 1571 while (fs->lfs_sleepers) 1572 cv_wait(&fs->lfs_sleeperscv, &lfs_lock); 1573 mutex_exit(&lfs_lock); 1574 1575 #ifdef LFS_EXTATTR 1576 if (ump->um_fstype == ULFS1) { 1577 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) { 1578 ulfs_extattr_stop(mp, curlwp); 1579 } 1580 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED) { 1581 ulfs_extattr_uepm_destroy(&ump->um_extattr); 1582 mp->mnt_flag &= ~MNT_EXTATTR; 1583 } 1584 } 1585 #endif 1586 #ifdef LFS_QUOTA 1587 if ((error = lfsquota1_umount(mp, flags)) != 0) 1588 return (error); 1589 #endif 1590 #ifdef LFS_QUOTA2 1591 if ((error = lfsquota2_umount(mp, flags)) != 0) 1592 return (error); 1593 #endif 1594 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 1595 return (error); 1596 if ((error = VFS_SYNC(mp, MNT_WAIT, l->l_cred)) != 0) 1597 return (error); 1598 vp = fs->lfs_ivnode; 1599 mutex_enter(vp->v_interlock); 1600 if (LIST_FIRST(&vp->v_dirtyblkhd)) 1601 panic("lfs_flushfiles: still dirty blocks on ifile vnode"); 1602 mutex_exit(vp->v_interlock); 1603 1604 /* Explicitly write the superblock, to update serial and pflags */ 1605 if (!fs->lfs_ronly) { 1606 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) | LFS_PF_CLEAN); 1607 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0)); 1608 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1)); 1609 } 1610 mutex_enter(&lfs_lock); 1611 while (fs->lfs_iocount) 1612 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 1613 &lfs_lock); 1614 mutex_exit(&lfs_lock); 1615 1616 return 0; 1617 } 1618 1619 /* 1620 * Get file system statistics. 1621 * 1622 * NB: We don't lock to access the superblock here, because it's not 1623 * really that important if we get it wrong. 1624 */ 1625 int 1626 lfs_statvfs(struct mount *mp, struct statvfs *sbp) 1627 { 1628 struct lfs *fs; 1629 struct ulfsmount *ump; 1630 1631 ump = VFSTOULFS(mp); 1632 fs = ump->um_lfs; 1633 1634 sbp->f_bsize = lfs_sb_getbsize(fs); 1635 sbp->f_frsize = lfs_sb_getfsize(fs); 1636 sbp->f_iosize = lfs_sb_getbsize(fs); 1637 sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks; 1638 1639 sbp->f_bfree = LFS_EST_BFREE(fs); 1640 /* 1641 * XXX this should be lfs_sb_getsize (measured in frags) 1642 * rather than dsize (measured in diskblocks). However, 1643 * getsize needs a format version check (for version 1 it 1644 * needs to be blockstofrags'd) so for the moment I'm going to 1645 * leave this... it won't fire wrongly as frags are at least 1646 * as big as diskblocks. 1647 */ 1648 KASSERT(sbp->f_bfree <= lfs_sb_getdsize(fs)); 1649 #if 0 1650 if (sbp->f_bfree < 0) 1651 sbp->f_bfree = 0; 1652 #endif 1653 1654 sbp->f_bresvd = LFS_EST_RSVD(fs); 1655 if (sbp->f_bfree > sbp->f_bresvd) 1656 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1657 else 1658 sbp->f_bavail = 0; 1659 1660 /* XXX: huh? - dholland 20150728 */ 1661 sbp->f_files = lfs_sb_getbfree(fs) / lfs_btofsb(fs, lfs_sb_getibsize(fs)) 1662 * LFS_INOPB(fs); 1663 sbp->f_ffree = sbp->f_files - lfs_sb_getnfiles(fs); 1664 sbp->f_favail = sbp->f_ffree; 1665 sbp->f_fresvd = 0; 1666 copy_statvfs_info(sbp, mp); 1667 return (0); 1668 } 1669 1670 /* 1671 * Go through the disk queues to initiate sandbagged IO; 1672 * go through the inodes to write those that have been modified; 1673 * initiate the writing of the super block if it has been modified. 1674 */ 1675 int 1676 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 1677 { 1678 int error, segflags; 1679 struct lfs *fs; 1680 1681 fs = VFSTOULFS(mp)->um_lfs; 1682 if (fs->lfs_ronly) 1683 return 0; 1684 1685 /* Snapshots should not hose the syncer */ 1686 /* 1687 * XXX Sync can block here anyway, since we don't have a very 1688 * XXX good idea of how much data is pending. If it's more 1689 * XXX than a segment and lfs_nextseg is close to the end of 1690 * XXX the log, we'll likely block. 1691 */ 1692 mutex_enter(&lfs_lock); 1693 if (fs->lfs_nowrap && lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs)) { 1694 mutex_exit(&lfs_lock); 1695 return 0; 1696 } 1697 mutex_exit(&lfs_lock); 1698 1699 lfs_writer_enter(fs, "lfs_wsync"); 1700 1701 DLOG((DLOG_FLUSH, "lfs_sync waitfor=%x at 0x%jx\n", waitfor, 1702 (uintmax_t)lfs_sb_getoffset(fs))); 1703 1704 segflags = 0; 1705 if (waitfor == MNT_LAZY) 1706 segflags = 0; 1707 else if (waitfor == MNT_NOWAIT) 1708 segflags = SEGM_CKP; 1709 else /* MNT_WAIT, or unknown value */ 1710 segflags = SEGM_SYNC | SEGM_CKP; 1711 1712 error = lfs_segwrite(mp, segflags); 1713 lfs_writer_leave(fs); 1714 #ifdef LFS_QUOTA 1715 lfs_qsync(mp); 1716 #endif 1717 return (error); 1718 } 1719 1720 /* 1721 * Look up an LFS dinode number to find its incore vnode. If not already 1722 * in core, read it in from the specified device. Return the inode locked. 1723 * Detection and handling of mount points must be done by the calling routine. 1724 */ 1725 int 1726 lfs_vget(struct mount *mp, ino_t ino, int lktype, struct vnode **vpp) 1727 { 1728 int error; 1729 1730 error = vcache_get(mp, &ino, sizeof(ino), vpp); 1731 if (error) 1732 return error; 1733 error = vn_lock(*vpp, lktype); 1734 if (error) { 1735 vrele(*vpp); 1736 *vpp = NULL; 1737 return error; 1738 } 1739 1740 return 0; 1741 } 1742 1743 static int 1744 compare_nodes_sd(void *context, const void *v1, const void *v2) 1745 { 1746 const struct segdelta *sd1, *sd2; 1747 1748 sd1 = (const struct segdelta *)v1; 1749 sd2 = (const struct segdelta *)v2; 1750 return sd1->segnum - sd2->segnum; 1751 } 1752 1753 static int 1754 compare_key_sd(void *context, const void *nv, const void *kv) 1755 { 1756 const struct segdelta *sd; 1757 long key; 1758 1759 sd = (const struct segdelta *)nv; 1760 key = *(const long *)kv; 1761 return sd->segnum - key; 1762 } 1763 1764 /* 1765 * Create a new vnode/inode pair and initialize what fields we can. 1766 */ 1767 static void 1768 lfs_init_vnode(struct ulfsmount *ump, ino_t ino, struct vnode *vp) 1769 { 1770 struct lfs *fs = ump->um_lfs; 1771 struct inode *ip; 1772 union lfs_dinode *dp; 1773 1774 ASSERT_NO_SEGLOCK(fs); 1775 1776 /* Initialize the inode. */ 1777 ip = pool_get(&lfs_inode_pool, PR_WAITOK); 1778 memset(ip, 0, sizeof(*ip)); 1779 dp = pool_get(&lfs_dinode_pool, PR_WAITOK); 1780 memset(dp, 0, sizeof(*dp)); 1781 ip->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK); 1782 memset(ip->inode_ext.lfs, 0, sizeof(*ip->inode_ext.lfs)); 1783 ip->i_din = dp; 1784 ip->i_ump = ump; 1785 ip->i_vnode = vp; 1786 ip->i_dev = fs->lfs_dev; 1787 lfs_dino_setinumber(fs, dp, ino); 1788 ip->i_number = ino; 1789 ip->i_lfs = fs; 1790 ip->i_lfs_effnblks = 0; 1791 SPLAY_INIT(&ip->i_lfs_lbtree); 1792 ip->i_lfs_nbtree = 0; 1793 rb_tree_init(&ip->i_lfs_segdhd, &lfs_rbtree_ops); 1794 1795 vp->v_tag = VT_LFS; 1796 vp->v_op = lfs_vnodeop_p; 1797 vp->v_data = ip; 1798 } 1799 1800 /* 1801 * Undo lfs_init_vnode(). 1802 */ 1803 static void 1804 lfs_deinit_vnode(struct ulfsmount *ump, struct vnode *vp) 1805 { 1806 struct inode *ip = VTOI(vp); 1807 1808 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1809 pool_put(&lfs_dinode_pool, ip->i_din); 1810 pool_put(&lfs_inode_pool, ip); 1811 vp->v_data = NULL; 1812 } 1813 1814 /* 1815 * Read an inode from disk and initialize this vnode / inode pair. 1816 * Caller assures no other thread will try to load this inode. 1817 */ 1818 int 1819 lfs_loadvnode(struct mount *mp, struct vnode *vp, 1820 const void *key, size_t key_len, const void **new_key) 1821 { 1822 struct lfs *fs; 1823 union lfs_dinode *dip; 1824 struct inode *ip; 1825 struct buf *bp; 1826 IFILE *ifp; 1827 struct ulfsmount *ump; 1828 ino_t ino; 1829 daddr_t daddr; 1830 int error, retries; 1831 struct timespec ts; 1832 1833 KASSERT(key_len == sizeof(ino)); 1834 memcpy(&ino, key, key_len); 1835 1836 memset(&ts, 0, sizeof ts); /* XXX gcc */ 1837 1838 ump = VFSTOULFS(mp); 1839 fs = ump->um_lfs; 1840 1841 /* 1842 * If the filesystem is not completely mounted yet, suspend 1843 * any access requests (wait for roll-forward to complete). 1844 */ 1845 mutex_enter(&lfs_lock); 1846 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1847 mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1848 &lfs_lock); 1849 mutex_exit(&lfs_lock); 1850 1851 KASSERTMSG(ino >= LFS_IFILE_INUM, "ino=%lld", (long long)ino); 1852 /* LFS_ASSERT_MAXINO(fs, ino); */ 1853 1854 /* Translate the inode number to a disk address. */ 1855 if (ino == LFS_IFILE_INUM) 1856 daddr = lfs_sb_getidaddr(fs); 1857 else { 1858 LFS_IENTRY(ifp, fs, ino, bp); 1859 daddr = lfs_if_getdaddr(fs, ifp); 1860 if (lfs_sb_getversion(fs) > 1) { 1861 ts.tv_sec = lfs_if_getatime_sec(fs, ifp); 1862 ts.tv_nsec = lfs_if_getatime_nsec(fs, ifp); 1863 } 1864 LFS_RELEASEIENTRY(ifp, fs, ino, bp); 1865 1866 if (DADDR_IS_BAD(daddr)) 1867 return (ENOENT); 1868 } 1869 1870 /* Allocate/init new vnode/inode. */ 1871 lfs_init_vnode(ump, ino, vp); 1872 ip = VTOI(vp); 1873 1874 /* If the cleaner supplied the inode, use it. */ 1875 if (curlwp == fs->lfs_cleaner_thread && fs->lfs_cleaner_hint != NULL && 1876 fs->lfs_cleaner_hint->bi_lbn == LFS_UNUSED_LBN) { 1877 dip = fs->lfs_cleaner_hint->bi_bp; 1878 if (fs->lfs_is64) { 1879 error = copyin(dip, &ip->i_din->u_64, 1880 sizeof(struct lfs64_dinode)); 1881 } else { 1882 error = copyin(dip, &ip->i_din->u_32, 1883 sizeof(struct lfs32_dinode)); 1884 } 1885 if (error) { 1886 lfs_deinit_vnode(ump, vp); 1887 return error; 1888 } 1889 KASSERT(ip->i_number == ino); 1890 goto out; 1891 } 1892 1893 /* Read in the disk contents for the inode, copy into the inode. */ 1894 retries = 0; 1895 again: 1896 error = bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), 1897 (lfs_sb_getversion(fs) == 1 ? lfs_sb_getbsize(fs) : lfs_sb_getibsize(fs)), 1898 0, &bp); 1899 if (error) { 1900 lfs_deinit_vnode(ump, vp); 1901 return error; 1902 } 1903 1904 dip = lfs_ifind(fs, ino, bp); 1905 if (dip == NULL) { 1906 /* Assume write has not completed yet; try again */ 1907 brelse(bp, BC_INVAL); 1908 ++retries; 1909 if (retries <= LFS_IFIND_RETRIES) { 1910 mutex_enter(&lfs_lock); 1911 if (fs->lfs_iocount) { 1912 DLOG((DLOG_VNODE, 1913 "%s: dinode %d not found, retrying...\n", 1914 __func__, ino)); 1915 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, 1916 "lfs ifind", 1, &lfs_lock); 1917 } else 1918 retries = LFS_IFIND_RETRIES; 1919 mutex_exit(&lfs_lock); 1920 goto again; 1921 } 1922 #ifdef DEBUG 1923 /* If the seglock is held look at the bpp to see 1924 what is there anyway */ 1925 mutex_enter(&lfs_lock); 1926 if (fs->lfs_seglock > 0) { 1927 struct buf **bpp; 1928 union lfs_dinode *dp; 1929 int i; 1930 1931 for (bpp = fs->lfs_sp->bpp; 1932 bpp != fs->lfs_sp->cbpp; ++bpp) { 1933 if ((*bpp)->b_vp == fs->lfs_ivnode && 1934 bpp != fs->lfs_sp->bpp) { 1935 /* Inode block */ 1936 printf("%s: block 0x%" PRIx64 ": ", 1937 __func__, (*bpp)->b_blkno); 1938 for (i = 0; i < LFS_INOPB(fs); i++) { 1939 dp = DINO_IN_BLOCK(fs, 1940 (*bpp)->b_data, i); 1941 if (lfs_dino_getinumber(fs, dp)) 1942 printf("%ju ", 1943 (uintmax_t)lfs_dino_getinumber(fs, dp)); 1944 } 1945 printf("\n"); 1946 } 1947 } 1948 } 1949 mutex_exit(&lfs_lock); 1950 #endif /* DEBUG */ 1951 panic("lfs_loadvnode: dinode not found"); 1952 } 1953 lfs_copy_dinode(fs, ip->i_din, dip); 1954 brelse(bp, 0); 1955 1956 out: 1957 if (lfs_sb_getversion(fs) > 1) { 1958 lfs_dino_setatime(fs, ip->i_din, ts.tv_sec); 1959 lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec); 1960 } 1961 1962 lfs_vinit(mp, &vp); 1963 1964 *new_key = &ip->i_number; 1965 return 0; 1966 } 1967 1968 /* 1969 * Create a new inode and initialize this vnode / inode pair. 1970 */ 1971 int 1972 lfs_newvnode(struct mount *mp, struct vnode *dvp, struct vnode *vp, 1973 struct vattr *vap, kauth_cred_t cred, void *extra, 1974 size_t *key_len, const void **new_key) 1975 { 1976 ino_t ino; 1977 struct inode *ip; 1978 struct ulfsmount *ump; 1979 struct lfs *fs; 1980 int error, mode, gen; 1981 1982 ump = VFSTOULFS(mp); 1983 fs = ump->um_lfs; 1984 1985 KASSERT(dvp != NULL || vap->va_fileid > 0); 1986 KASSERT((fs->lfs_flags & LFS_NOTYET) || 1987 (dvp != NULL && dvp->v_mount == mp)); 1988 KASSERT(vap->va_type != VNON); 1989 1990 *key_len = sizeof(ino); 1991 mode = MAKEIMODE(vap->va_type, vap->va_mode); 1992 1993 /* 1994 * Allocate fresh inode. With "dvp == NULL" take the inode number 1995 * and version from "vap". 1996 */ 1997 if (dvp == NULL) { 1998 ino = vap->va_fileid; 1999 gen = vap->va_gen; 2000 error = lfs_valloc_fixed(fs, ino, gen); 2001 } else { 2002 error = lfs_valloc(dvp, mode, cred, &ino, &gen); 2003 } 2004 if (error) 2005 return error; 2006 2007 /* Attach inode to vnode. */ 2008 lfs_init_vnode(ump, ino, vp); 2009 ip = VTOI(vp); 2010 2011 mutex_enter(&lfs_lock); 2012 LFS_SET_UINO(ip, IN_CHANGE); 2013 mutex_exit(&lfs_lock); 2014 2015 /* Note no blocks yet */ 2016 ip->i_lfs_hiblk = -1; 2017 2018 /* Set a new generation number for this inode. */ 2019 ip->i_gen = gen; 2020 lfs_dino_setgen(fs, ip->i_din, gen); 2021 2022 memset(ip->i_lfs_fragsize, 0, 2023 ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 2024 2025 /* Set uid / gid. */ 2026 if (cred == NOCRED || cred == FSCRED) { 2027 ip->i_gid = 0; 2028 ip->i_uid = 0; 2029 } else { 2030 ip->i_gid = VTOI(dvp)->i_gid; 2031 ip->i_uid = kauth_cred_geteuid(cred); 2032 } 2033 DIP_ASSIGN(ip, gid, ip->i_gid); 2034 DIP_ASSIGN(ip, uid, ip->i_uid); 2035 2036 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 2037 error = lfs_chkiq(ip, 1, cred, 0); 2038 if (error) { 2039 lfs_vfree(dvp, ino, mode); 2040 lfs_deinit_vnode(ump, vp); 2041 2042 return error; 2043 } 2044 #endif 2045 2046 /* Set type and finalize. */ 2047 ip->i_flags = 0; 2048 DIP_ASSIGN(ip, flags, 0); 2049 ip->i_mode = mode; 2050 DIP_ASSIGN(ip, mode, mode); 2051 if (vap->va_rdev != VNOVAL) { 2052 /* 2053 * Want to be able to use this to make badblock 2054 * inodes, so don't truncate the dev number. 2055 */ 2056 // XXX clean this up 2057 if (ump->um_fstype == ULFS1) 2058 ip->i_din->u_32.di_rdev = ulfs_rw32(vap->va_rdev, 2059 ULFS_MPNEEDSWAP(fs)); 2060 else 2061 ip->i_din->u_64.di_rdev = ulfs_rw64(vap->va_rdev, 2062 ULFS_MPNEEDSWAP(fs)); 2063 } 2064 lfs_vinit(mp, &vp); 2065 2066 *new_key = &ip->i_number; 2067 return 0; 2068 } 2069 2070 /* 2071 * File handle to vnode 2072 */ 2073 int 2074 lfs_fhtovp(struct mount *mp, struct fid *fhp, int lktype, struct vnode **vpp) 2075 { 2076 struct lfid lfh; 2077 struct lfs *fs; 2078 2079 if (fhp->fid_len != sizeof(struct lfid)) 2080 return EINVAL; 2081 2082 memcpy(&lfh, fhp, sizeof(lfh)); 2083 if (lfh.lfid_ino < LFS_IFILE_INUM) 2084 return ESTALE; 2085 2086 fs = VFSTOULFS(mp)->um_lfs; 2087 if (lfh.lfid_ident != lfs_sb_getident(fs)) 2088 return ESTALE; 2089 2090 if (lfh.lfid_ino > 2091 ((lfs_dino_getsize(fs, VTOI(fs->lfs_ivnode)->i_din) >> lfs_sb_getbshift(fs)) - 2092 lfs_sb_getcleansz(fs) - lfs_sb_getsegtabsz(fs)) * lfs_sb_getifpb(fs)) 2093 return ESTALE; 2094 2095 return (ulfs_fhtovp(mp, &lfh.lfid_ufid, lktype, vpp)); 2096 } 2097 2098 /* 2099 * Vnode pointer to File handle 2100 */ 2101 /* ARGSUSED */ 2102 int 2103 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) 2104 { 2105 struct inode *ip; 2106 struct lfid lfh; 2107 2108 if (*fh_size < sizeof(struct lfid)) { 2109 *fh_size = sizeof(struct lfid); 2110 return E2BIG; 2111 } 2112 *fh_size = sizeof(struct lfid); 2113 ip = VTOI(vp); 2114 memset(&lfh, 0, sizeof(lfh)); 2115 lfh.lfid_len = sizeof(struct lfid); 2116 lfh.lfid_ino = ip->i_number; 2117 lfh.lfid_gen = ip->i_gen; 2118 lfh.lfid_ident = lfs_sb_getident(ip->i_lfs); 2119 memcpy(fhp, &lfh, sizeof(lfh)); 2120 return (0); 2121 } 2122 2123 /* 2124 * ulfs_bmaparray callback function for writing. 2125 * 2126 * Since blocks will be written to the new segment anyway, 2127 * we don't care about current daddr of them. 2128 */ 2129 static bool 2130 lfs_issequential_hole(const struct lfs *fs, 2131 daddr_t daddr0, daddr_t daddr1) 2132 { 2133 (void)fs; /* not used */ 2134 2135 KASSERT(daddr0 == UNWRITTEN || 2136 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR(fs))); 2137 KASSERT(daddr1 == UNWRITTEN || 2138 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR(fs))); 2139 2140 /* NOTE: all we want to know here is 'hole or not'. */ 2141 /* NOTE: UNASSIGNED is converted to 0 by ulfs_bmaparray. */ 2142 2143 /* 2144 * treat UNWRITTENs and all resident blocks as 'contiguous' 2145 */ 2146 if (daddr0 != 0 && daddr1 != 0) 2147 return true; 2148 2149 /* 2150 * both are in hole? 2151 */ 2152 if (daddr0 == 0 && daddr1 == 0) 2153 return true; /* all holes are 'contiguous' for us. */ 2154 2155 return false; 2156 } 2157 2158 /* 2159 * lfs_gop_write functions exactly like genfs_gop_write, except that 2160 * (1) it requires the seglock to be held by its caller, and sp->fip 2161 * to be properly initialized (it will return without re-initializing 2162 * sp->fip, and without calling lfs_writeseg). 2163 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 2164 * to determine how large a block it can write at once (though it does 2165 * still use VOP_BMAP to find holes in the file); 2166 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 2167 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 2168 * now have clusters of clusters, ick.) 2169 */ 2170 static int 2171 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, 2172 int flags) 2173 { 2174 int i, error, run, haveeof = 0; 2175 int fs_bshift; 2176 vaddr_t kva; 2177 off_t eof, offset, startoffset = 0; 2178 size_t bytes, iobytes, skipbytes; 2179 bool async = (flags & PGO_SYNCIO) == 0; 2180 daddr_t lbn, blkno; 2181 struct vm_page *pg; 2182 struct buf *mbp, *bp; 2183 struct vnode *devvp = VTOI(vp)->i_devvp; 2184 struct inode *ip = VTOI(vp); 2185 struct lfs *fs = ip->i_lfs; 2186 struct segment *sp = fs->lfs_sp; 2187 SEGSUM *ssp; 2188 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 2189 const char * failreason = NULL; 2190 2191 ASSERT_SEGLOCK(fs); 2192 2193 /* The Ifile lives in the buffer cache */ 2194 KASSERT(vp != fs->lfs_ivnode); 2195 2196 /* 2197 * We don't want to fill the disk before the cleaner has a chance 2198 * to make room for us. If we're in danger of doing that, fail 2199 * with EAGAIN. The caller will have to notice this, unlock 2200 * so the cleaner can run, relock and try again. 2201 * 2202 * We must write everything, however, if our vnode is being 2203 * reclaimed. 2204 */ 2205 mutex_enter(vp->v_interlock); 2206 if (LFS_STARVED_FOR_SEGS(fs) && vdead_check(vp, VDEAD_NOWAIT) == 0) { 2207 mutex_exit(vp->v_interlock); 2208 failreason = "Starved for segs and not flushing vp"; 2209 goto tryagain; 2210 } 2211 mutex_exit(vp->v_interlock); 2212 2213 /* 2214 * Sometimes things slip past the filters in lfs_putpages, 2215 * and the pagedaemon tries to write pages---problem is 2216 * that the pagedaemon never acquires the segment lock. 2217 * 2218 * Alternatively, pages that were clean when we called 2219 * genfs_putpages may have become dirty in the meantime. In this 2220 * case the segment header is not properly set up for blocks 2221 * to be added to it. 2222 * 2223 * Unbusy and unclean the pages, and put them on the ACTIVE 2224 * queue under the hypothesis that they couldn't have got here 2225 * unless they were modified *quite* recently. 2226 * 2227 * XXXUBC that last statement is an oversimplification of course. 2228 */ 2229 if (!LFS_SEGLOCK_HELD(fs)) { 2230 failreason = "Seglock not held"; 2231 goto tryagain; 2232 } 2233 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) { 2234 failreason = "Inode with no_gop_write"; 2235 goto tryagain; 2236 } 2237 if ((pgs[0]->offset & lfs_sb_getbmask(fs)) != 0) { 2238 failreason = "Bad page offset"; 2239 goto tryagain; 2240 } 2241 2242 UVMHIST_LOG(ubchist, "vp %#jx pgs %#jx npages %jd flags 0x%jx", 2243 (uintptr_t)vp, (uintptr_t)pgs, npages, flags); 2244 2245 GOP_SIZE(vp, vp->v_size, &eof, 0); 2246 haveeof = 1; 2247 2248 if (vp->v_type == VREG) 2249 fs_bshift = vp->v_mount->mnt_fs_bshift; 2250 else 2251 fs_bshift = DEV_BSHIFT; 2252 error = 0; 2253 pg = pgs[0]; 2254 startoffset = pg->offset; 2255 KASSERT(eof >= 0); 2256 2257 if (startoffset >= eof) { 2258 failreason = "Offset beyond EOF"; 2259 goto tryagain; 2260 } else 2261 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 2262 skipbytes = 0; 2263 2264 KASSERT(bytes != 0); 2265 2266 /* Swap PG_DELWRI for PG_PAGEOUT */ 2267 for (i = 0; i < npages; i++) { 2268 if (pgs[i]->flags & PG_DELWRI) { 2269 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 2270 pgs[i]->flags &= ~PG_DELWRI; 2271 pgs[i]->flags |= PG_PAGEOUT; 2272 uvm_pageout_start(1); 2273 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 2274 uvm_pagelock(pgs[i]); 2275 uvm_pageunwire(pgs[i]); 2276 uvm_pageunlock(pgs[i]); 2277 rw_exit(vp->v_uobj.vmobjlock); 2278 } 2279 } 2280 2281 /* 2282 * Check to make sure we're starting on a block boundary. 2283 * We'll check later to make sure we always write entire 2284 * blocks (or fragments). 2285 */ 2286 if (startoffset & lfs_sb_getbmask(fs)) 2287 printf("%" PRId64 " & %" PRIu64 " = %" PRId64 "\n", 2288 startoffset, lfs_sb_getbmask(fs), 2289 startoffset & lfs_sb_getbmask(fs)); 2290 KASSERT((startoffset & lfs_sb_getbmask(fs)) == 0); 2291 if (bytes & lfs_sb_getffmask(fs)) { 2292 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 2293 panic("lfs_gop_write: non-integer blocks"); 2294 } 2295 2296 /* 2297 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 2298 * If we would, write what we have and try again. If we don't 2299 * have anything to write, we'll have to sleep. 2300 */ 2301 ssp = (SEGSUM *)sp->segsum; 2302 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2303 (lfs_ss_getnfinfo(fs, ssp) < 1 ? 2304 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 2305 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 2306 lfs_updatemeta(sp); 2307 lfs_release_finfo(fs); 2308 (void) lfs_writeseg(fs, sp); 2309 2310 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2311 2312 /* 2313 * Having given up all of the pager_map we were holding, 2314 * we can now wait for aiodoned to reclaim it for us 2315 * without fear of deadlock. 2316 */ 2317 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2318 UVMPAGER_MAPIN_WAITOK); 2319 } 2320 2321 mbp = getiobuf(NULL, true); 2322 UVMHIST_LOG(ubchist, "vp %#jx mbp %#jx num now %jd bytes 0x%jx", 2323 (uintptr_t)vp, (uintptr_t)mbp, vp->v_numoutput, bytes); 2324 mbp->b_bufsize = npages << PAGE_SHIFT; 2325 mbp->b_data = (void *)kva; 2326 mbp->b_resid = mbp->b_bcount = bytes; 2327 mbp->b_cflags |= BC_BUSY|BC_AGE; 2328 mbp->b_iodone = uvm_aio_aiodone; 2329 2330 bp = NULL; 2331 for (offset = startoffset; 2332 bytes > 0; 2333 offset += iobytes, bytes -= iobytes) { 2334 lbn = offset >> fs_bshift; 2335 error = ulfs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 2336 lfs_issequential_hole); 2337 if (error) { 2338 UVMHIST_LOG(ubchist, "ulfs_bmaparray() -> %jd", 2339 error,0,0,0); 2340 skipbytes += bytes; 2341 bytes = 0; 2342 break; 2343 } 2344 2345 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 2346 bytes); 2347 if (blkno == (daddr_t)-1) { 2348 skipbytes += iobytes; 2349 continue; 2350 } 2351 2352 /* 2353 * Discover how much we can really pack into this buffer. 2354 */ 2355 /* If no room in the current segment, finish it up */ 2356 if (sp->sum_bytes_left < sizeof(int32_t) || 2357 sp->seg_bytes_left < (1 << lfs_sb_getbshift(fs))) { 2358 int vers; 2359 2360 lfs_updatemeta(sp); 2361 vers = lfs_fi_getversion(fs, sp->fip); 2362 lfs_release_finfo(fs); 2363 (void) lfs_writeseg(fs, sp); 2364 2365 lfs_acquire_finfo(fs, ip->i_number, vers); 2366 } 2367 /* Check both for space in segment and space in segsum */ 2368 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 2369 << fs_bshift); 2370 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 2371 << fs_bshift); 2372 KASSERT(iobytes > 0); 2373 2374 /* if it's really one i/o, don't make a second buf */ 2375 if (offset == startoffset && iobytes == bytes) { 2376 bp = mbp; 2377 /* 2378 * All the LFS output is done by the segwriter. It 2379 * will increment numoutput by one for all the bufs it 2380 * receives. However this buffer needs one extra to 2381 * account for aiodone. 2382 */ 2383 mutex_enter(vp->v_interlock); 2384 vp->v_numoutput++; 2385 mutex_exit(vp->v_interlock); 2386 } else { 2387 bp = getiobuf(NULL, true); 2388 UVMHIST_LOG(ubchist, "vp %#jx bp %#jx num now %jd", 2389 (uintptr_t)vp, (uintptr_t)bp, vp->v_numoutput, 0); 2390 nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes); 2391 /* 2392 * LFS doesn't like async I/O here, dies with 2393 * an assert in lfs_bwrite(). Is that assert 2394 * valid? I retained non-async behaviour when 2395 * converted this to use nestiobuf --pooka 2396 */ 2397 bp->b_flags &= ~B_ASYNC; 2398 } 2399 2400 /* XXX This is silly ... is this necessary? */ 2401 mutex_enter(&bufcache_lock); 2402 mutex_enter(vp->v_interlock); 2403 bgetvp(vp, bp); 2404 mutex_exit(vp->v_interlock); 2405 mutex_exit(&bufcache_lock); 2406 2407 bp->b_lblkno = lfs_lblkno(fs, offset); 2408 bp->b_private = mbp; 2409 if (devvp->v_type == VBLK) { 2410 bp->b_dev = devvp->v_rdev; 2411 } 2412 VOP_BWRITE(bp->b_vp, bp); 2413 while (lfs_gatherblock(sp, bp, NULL)) 2414 continue; 2415 } 2416 2417 nestiobuf_done(mbp, skipbytes, error); 2418 if (skipbytes) { 2419 UVMHIST_LOG(ubchist, "skipbytes %jd", skipbytes, 0,0,0); 2420 } 2421 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 2422 2423 if (!async) { 2424 /* Start a segment write. */ 2425 UVMHIST_LOG(ubchist, "flushing", 0,0,0,0); 2426 mutex_enter(&lfs_lock); 2427 lfs_flush(fs, 0, 1); 2428 mutex_exit(&lfs_lock); 2429 } 2430 2431 if ((sp->seg_flags & SEGM_SINGLE) && lfs_sb_getcurseg(fs) != fs->lfs_startseg) 2432 return EAGAIN; 2433 2434 return (0); 2435 2436 tryagain: 2437 /* 2438 * We can't write the pages, for whatever reason. 2439 * Clean up after ourselves, and make the caller try again. 2440 */ 2441 2442 /* Tell why we're here, if we know */ 2443 if (failreason != NULL) { 2444 DLOG((DLOG_PAGE, "lfs_gop_write: %s\n", failreason)); 2445 } 2446 if (haveeof && startoffset >= eof) { 2447 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 2448 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 2449 pgs[0]->offset, eof, npages)); 2450 } 2451 2452 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 2453 for (i = 0; i < npages; i++) { 2454 pg = pgs[i]; 2455 2456 if (pg->flags & PG_PAGEOUT) 2457 uvm_pageout_done(1); 2458 uvm_pagelock(pg); 2459 if (pg->flags & PG_DELWRI) { 2460 uvm_pageunwire(pg); 2461 } 2462 uvm_pageactivate(pg); 2463 uvm_pageunlock(pg); 2464 pg->flags &= ~(PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 2465 uvm_pagemarkdirty(pg, UVM_PAGE_STATUS_DIRTY); 2466 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 2467 vp, pg->offset)); 2468 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 2469 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 2470 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 2471 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 2472 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 2473 pg->wire_count)); 2474 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 2475 pg->loan_count)); 2476 } 2477 uvm_page_unbusy(pgs, npages); 2478 rw_exit(vp->v_uobj.vmobjlock); 2479 return EAGAIN; 2480 } 2481 2482 /* 2483 * finish vnode/inode initialization. 2484 * used by lfs_vget. 2485 */ 2486 void 2487 lfs_vinit(struct mount *mp, struct vnode **vpp) 2488 { 2489 struct vnode *vp = *vpp; 2490 struct inode *ip = VTOI(vp); 2491 struct ulfsmount *ump = VFSTOULFS(mp); 2492 struct lfs *fs = ump->um_lfs; 2493 int i; 2494 2495 ip->i_mode = lfs_dino_getmode(fs, ip->i_din); 2496 ip->i_nlink = lfs_dino_getnlink(fs, ip->i_din); 2497 ip->i_lfs_osize = ip->i_size = lfs_dino_getsize(fs, ip->i_din); 2498 ip->i_flags = lfs_dino_getflags(fs, ip->i_din); 2499 ip->i_gen = lfs_dino_getgen(fs, ip->i_din); 2500 ip->i_uid = lfs_dino_getuid(fs, ip->i_din); 2501 ip->i_gid = lfs_dino_getgid(fs, ip->i_din); 2502 2503 ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din); 2504 ip->i_lfs_odnlink = lfs_dino_getnlink(fs, ip->i_din); 2505 2506 /* 2507 * Initialize the vnode from the inode, check for aliases. In all 2508 * cases re-init ip, the underlying vnode/inode may have changed. 2509 */ 2510 ulfs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 2511 ip = VTOI(vp); 2512 2513 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 2514 if (vp->v_type != VLNK || ip->i_size >= ip->i_lfs->um_maxsymlinklen) { 2515 #ifdef DEBUG 2516 for (i = (ip->i_size + lfs_sb_getbsize(fs) - 1) >> lfs_sb_getbshift(fs); 2517 i < ULFS_NDADDR; i++) { 2518 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 2519 i == 0) 2520 continue; 2521 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) { 2522 lfs_dump_dinode(fs, ip->i_din); 2523 panic("inconsistent inode (direct)"); 2524 } 2525 } 2526 for ( ; i < ULFS_NDADDR + ULFS_NIADDR; i++) { 2527 if (lfs_dino_getib(fs, ip->i_din, i - ULFS_NDADDR) != 0) { 2528 lfs_dump_dinode(fs, ip->i_din); 2529 panic("inconsistent inode (indirect)"); 2530 } 2531 } 2532 #endif /* DEBUG */ 2533 for (i = 0; i < ULFS_NDADDR; i++) 2534 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) 2535 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i); 2536 } 2537 2538 KASSERTMSG((vp->v_type != VNON), 2539 "lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 2540 (unsigned long long)ip->i_number, 2541 (ip->i_mode & LFS_IFMT) >> 12); 2542 2543 /* 2544 * Finish inode initialization now that aliasing has been resolved. 2545 */ 2546 2547 ip->i_devvp = fs->lfs_devvp; 2548 vref(ip->i_devvp); 2549 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 2550 ulfsquota_init(ip); 2551 #endif 2552 genfs_node_init(vp, &lfs_genfsops); 2553 uvm_vnp_setsize(vp, ip->i_size); 2554 2555 /* Initialize hiblk from file size */ 2556 ip->i_lfs_hiblk = lfs_lblkno(ip->i_lfs, ip->i_size + lfs_sb_getbsize(ip->i_lfs) - 1) - 1; 2557 2558 *vpp = vp; 2559 } 2560 2561 /* 2562 * Resize the filesystem to contain the specified number of segments. 2563 */ 2564 int 2565 lfs_resize_fs(struct lfs *fs, int newnsegs) 2566 { 2567 SEGUSE *sup; 2568 CLEANERINFO *cip; 2569 struct buf *bp, *obp; 2570 daddr_t olast, nlast, ilast, noff, start, end, dirtysums; 2571 struct vnode *ivp; 2572 struct inode *ip; 2573 int error, badnews, inc, oldnsegs; 2574 int sbbytes, csbbytes, gain, cgain; 2575 int i, nexti; 2576 2577 /* Only support v2 and up */ 2578 if (lfs_sb_getversion(fs) < 2) 2579 return EOPNOTSUPP; 2580 2581 /* If we're doing nothing, do it fast */ 2582 oldnsegs = lfs_sb_getnseg(fs); 2583 if (newnsegs == oldnsegs) 2584 return 0; 2585 2586 /* We always have to have two superblocks */ 2587 if (newnsegs <= lfs_dtosn(fs, lfs_sb_getsboff(fs, 1))) 2588 return EINVAL; 2589 2590 ivp = fs->lfs_ivnode; 2591 ip = VTOI(ivp); 2592 error = 0; 2593 2594 /* Take the segment lock */ 2595 lfs_seglock(fs, SEGM_CKP); 2596 2597 /* If we're shrinking, clean out the upper segments */ 2598 if (newnsegs < oldnsegs) { 2599 lfs_rewind(fs, newnsegs); 2600 for (i = newnsegs; i < oldnsegs; i++) { 2601 uint32_t bytes; 2602 LFS_SEGENTRY(sup, fs, i, bp); 2603 bytes = sup->su_nbytes; 2604 LFS_RELEASESEGENTRY(sup, fs, i, bp); 2605 if (bytes > 0) { 2606 if ((error = lfs_rewrite_segment(fs, i, NULL, NOCRED, curlwp)) != 0) 2607 goto out; 2608 } 2609 2610 lfs_invalidate(fs, i); 2611 } 2612 } 2613 2614 /* 2615 * Make sure the segments we're going to be losing, if any, 2616 * are in fact empty. We hold the seglock, so their status 2617 * cannot change underneath us. Count the superblocks we lose, 2618 * while we're at it. 2619 */ 2620 sbbytes = csbbytes = cgain = dirtysums = 0; 2621 for (i = newnsegs; i < oldnsegs; i++) { 2622 LFS_SEGENTRY(sup, fs, i, bp); 2623 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 2624 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2625 sbbytes += LFS_SBPAD; 2626 if (!(sup->su_flags & SEGUSE_DIRTY)) { 2627 ++cgain; 2628 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2629 csbbytes += LFS_SBPAD; 2630 } else 2631 dirtysums += sup->su_nsums; 2632 LFS_RELEASESEGENTRY(sup, fs, i, bp); 2633 if (badnews) { 2634 error = EBUSY; 2635 goto out; 2636 } 2637 } 2638 2639 /* Note old and new segment table endpoints, and old ifile size */ 2640 olast = lfs_sb_getcleansz(fs) + lfs_sb_getsegtabsz(fs); 2641 nlast = howmany(newnsegs, lfs_sb_getsepb(fs)) + lfs_sb_getcleansz(fs); 2642 ilast = ivp->v_size >> lfs_sb_getbshift(fs); 2643 noff = nlast - olast; 2644 2645 /* 2646 * Make sure no one can use the Ifile while we change it around. 2647 * Even after taking the iflock we need to make sure no one still 2648 * is holding Ifile buffers, so we get each one, to drain them. 2649 * (XXX this could be done better.) 2650 */ 2651 for (i = 0; i < ilast; i++) { 2652 /* XXX what to do if bread fails? */ 2653 bread(ivp, i, lfs_sb_getbsize(fs), 0, &bp); 2654 brelse(bp, 0); 2655 } 2656 2657 /* Allocate new Ifile blocks */ 2658 for (i = ilast; i < ilast + noff; i++) { 2659 if (lfs_balloc(ivp, i * lfs_sb_getbsize(fs), 2660 lfs_sb_getbsize(fs), NOCRED, 0, 2661 &bp) != 0) 2662 panic("balloc extending ifile"); 2663 memset(bp->b_data, 0, lfs_sb_getbsize(fs)); 2664 VOP_BWRITE(bp->b_vp, bp); 2665 } 2666 2667 /* 2668 * If we are expanding, increase the Ifile size now 2669 * so we can allocate blocks. If we are shrinking 2670 * we need to wait until we have moved the data. 2671 */ 2672 if (noff > 0) { 2673 ip->i_size += noff * lfs_sb_getbsize(fs); 2674 lfs_dino_setsize(fs, ip->i_din, ip->i_size); 2675 uvm_vnp_setsize(ivp, ip->i_size); 2676 } 2677 2678 /* Copy the inode table to its new position */ 2679 if (noff != 0) { 2680 if (noff < 0) { 2681 start = nlast; 2682 end = ilast + noff; 2683 inc = 1; 2684 } else { 2685 start = ilast + noff - 1; 2686 end = nlast - 1; 2687 inc = -1; 2688 } 2689 for (i = start; i != end; i += inc) { 2690 if (bread(ivp, i, lfs_sb_getbsize(fs), 2691 B_MODIFY, &bp) != 0) 2692 panic("resize: bread dst blk failed"); 2693 if (bread(ivp, i - noff, lfs_sb_getbsize(fs), 2694 0, &obp)) 2695 panic("resize: bread src blk failed"); 2696 memcpy(bp->b_data, obp->b_data, lfs_sb_getbsize(fs)); 2697 VOP_BWRITE(bp->b_vp, bp); 2698 brelse(obp, 0); 2699 } 2700 } 2701 2702 /* If we are expanding, write the new empty SEGUSE entries */ 2703 for (i = oldnsegs, nexti = 0; i < newnsegs; i = nexti) { 2704 nexti = roundup(i + 1, lfs_sb_getsepb(fs)); 2705 if (nexti > newnsegs) 2706 nexti = newnsegs; 2707 2708 if ((error = bread(ivp, i / lfs_sb_getsepb(fs) + 2709 lfs_sb_getcleansz(fs), 2710 lfs_sb_getbsize(fs), 2711 (i > oldnsegs ? B_MODIFY : 0), &bp)) != 0) 2712 panic("lfs: ifile read: %d", error); 2713 sup = &((SEGUSE *)bp->b_data)[i % lfs_sb_getsepb(fs)]; 2714 memset(sup, 0, (nexti - i) * sizeof(SEGUSE)); 2715 VOP_BWRITE(bp->b_vp, bp); 2716 } 2717 2718 /* Zero out unused superblock offsets */ 2719 for (i = 2; i < LFS_MAXNUMSB; i++) 2720 if (lfs_dtosn(fs, lfs_sb_getsboff(fs, i)) >= newnsegs) 2721 lfs_sb_setsboff(fs, i, 0x0); 2722 2723 /* 2724 * Correct superblock entries that depend on fs size. 2725 * The computations of these are as follows: 2726 * 2727 * size = lfs_segtod(fs, nseg) 2728 * dsize = lfs_segtod(fs, nseg - minfreeseg) - lfs_btofsb(#super * LFS_SBPAD) 2729 * bfree = dsize - lfs_btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2730 * avail = lfs_segtod(fs, nclean) - lfs_btofsb(#clean_super * LFS_SBPAD) 2731 * + (lfs_segtod(fs, 1) - (offset - curseg)) 2732 * - lfs_segtod(fs, minfreeseg - (minfreeseg / 2)) 2733 * 2734 * XXX - we should probably adjust minfreeseg as well. 2735 */ 2736 gain = (newnsegs - oldnsegs); 2737 lfs_sb_setnseg(fs, newnsegs); 2738 lfs_sb_setsegtabsz(fs, nlast - lfs_sb_getcleansz(fs)); 2739 lfs_sb_addsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs))); 2740 lfs_sb_adddsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) 2741 + lfs_btofsb(fs, sbbytes)); 2742 lfs_sb_addbfree(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) 2743 + lfs_btofsb(fs, sbbytes) 2744 #if 0 /* This doesn't match fsck_lfs */ 2745 - gain * lfs_btofsb(fs, lfs_sb_getbsize(fs) / 2) 2746 + dirtysums * lfs_btofsb(fs, lfs_sb_getsumsize(fs)) 2747 #endif /* 0 */ 2748 ); 2749 if (gain > 0) { 2750 lfs_sb_addnclean(fs, gain); 2751 lfs_sb_addavail(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs))); 2752 } else { 2753 lfs_sb_subnclean(fs, cgain); 2754 lfs_sb_subavail(fs, cgain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - 2755 lfs_btofsb(fs, csbbytes)); 2756 } 2757 2758 /* Truncate Ifile if necessary */ 2759 if (noff < 0) { 2760 vn_lock(ivp, LK_EXCLUSIVE); 2761 lfs_truncate(ivp, ivp->v_size + (noff << lfs_sb_getbshift(fs)), 0, 2762 NOCRED); 2763 VOP_UNLOCK(ivp); 2764 } 2765 2766 /* Update cleaner info so the cleaner can die */ 2767 /* XXX what to do if bread fails? */ 2768 bread(ivp, 0, lfs_sb_getbsize(fs), B_MODIFY, &bp); 2769 cip = bp->b_data; 2770 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs)); 2771 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs)); 2772 VOP_BWRITE(bp->b_vp, bp); 2773 2774 out: 2775 lfs_segunlock(fs); 2776 return error; 2777 } 2778 2779 /* 2780 * Extended attribute dispatch 2781 */ 2782 int 2783 lfs_extattrctl(struct mount *mp, int cmd, struct vnode *vp, 2784 int attrnamespace, const char *attrname) 2785 { 2786 #ifdef LFS_EXTATTR 2787 struct ulfsmount *ump; 2788 2789 ump = VFSTOULFS(mp); 2790 if (ump->um_fstype == ULFS1) { 2791 return ulfs_extattrctl(mp, cmd, vp, attrnamespace, attrname); 2792 } 2793 #endif 2794 return vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname); 2795 } 2796
Indexes created Sun Nov 01 00:25:34 UTC 2026