1/* $NetBSD: lfs_vfsops.c,v 1.351 2016/07/07 06:55:44 msaitoh Exp $ */
2
3/*-
4 * Copyright (c) 1999, 2000, 2001, 2002, 2003, 2007, 2007
5 * The NetBSD Foundation, Inc.
6 * All rights reserved.
7 *
8 * This code is derived from software contributed to The NetBSD Foundation
9 * by Konrad E. Schroder <perseant@hhhh.org>.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32/*-
33 * Copyright (c) 1989, 1991, 1993, 1994
34 * The Regents of the University of California. All rights reserved.
35 *
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
38 * are met:
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
44 * 3. Neither the name of the University nor the names of its contributors
45 * may be used to endorse or promote products derived from this software
46 * without specific prior written permission.
47 *
48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 * SUCH DAMAGE.
59 *
60 * @(#)lfs_vfsops.c 8.20 (Berkeley) 6/10/95
61 */
62
63#include <sys/cdefs.h>
64__KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.351 2016/07/07 06:55:44 msaitoh Exp $");
65
66#if defined(_KERNEL_OPT)
67#include "opt_lfs.h"
68#include "opt_quota.h"
69#endif
70
71#include <sys/param.h>
72#include <sys/systm.h>
73#include <sys/namei.h>
74#include <sys/proc.h>
75#include <sys/kernel.h>
76#include <sys/vnode.h>
77#include <sys/mount.h>
78#include <sys/kthread.h>
79#include <sys/buf.h>
80#include <sys/device.h>
81#include <sys/mbuf.h>
82#include <sys/file.h>
83#include <sys/disklabel.h>
84#include <sys/ioctl.h>
85#include <sys/errno.h>
86#include <sys/malloc.h>
87#include <sys/pool.h>
88#include <sys/socket.h>
89#include <sys/syslog.h>
90#include <uvm/uvm_extern.h>
91#include <sys/sysctl.h>
92#include <sys/conf.h>
93#include <sys/kauth.h>
94#include <sys/module.h>
95#include <sys/syscallvar.h>
96#include <sys/syscall.h>
97#include <sys/syscallargs.h>
98
99#include <miscfs/specfs/specdev.h>
100
101#include <ufs/lfs/ulfs_quotacommon.h>
102#include <ufs/lfs/ulfs_inode.h>
103#include <ufs/lfs/ulfsmount.h>
104#include <ufs/lfs/ulfs_bswap.h>
105#include <ufs/lfs/ulfs_extern.h>
106
107#include <uvm/uvm.h>
108#include <uvm/uvm_stat.h>
109#include <uvm/uvm_pager.h>
110#include <uvm/uvm_pdaemon.h>
111
112#include <ufs/lfs/lfs.h>
113#include <ufs/lfs/lfs_accessors.h>
114#include <ufs/lfs/lfs_kernel.h>
115#include <ufs/lfs/lfs_extern.h>
116
117#include <miscfs/genfs/genfs.h>
118#include <miscfs/genfs/genfs_node.h>
119
120MODULE(MODULE_CLASS_VFS, lfs, NULL);
121
122static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
123static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *);
124
125static struct sysctllog *lfs_sysctl_log;
126
127extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
128extern const struct vnodeopv_desc lfs_specop_opv_desc;
129extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
130
131pid_t lfs_writer_daemon = 0;
132lwpid_t lfs_writer_lid = 0;
133int lfs_do_flush = 0;
134#ifdef LFS_KERNEL_RFW
135int lfs_do_rfw = 0;
136#endif
137
138const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
139 &lfs_vnodeop_opv_desc,
140 &lfs_specop_opv_desc,
141 &lfs_fifoop_opv_desc,
142 NULL,
143};
144
145struct vfsops lfs_vfsops = {
146 .vfs_name = MOUNT_LFS,
147 .vfs_min_mount_data = sizeof (struct ulfs_args),
148 .vfs_mount = lfs_mount,
149 .vfs_start = ulfs_start,
150 .vfs_unmount = lfs_unmount,
151 .vfs_root = ulfs_root,
152 .vfs_quotactl = ulfs_quotactl,
153 .vfs_statvfs = lfs_statvfs,
154 .vfs_sync = lfs_sync,
155 .vfs_vget = lfs_vget,
156 .vfs_loadvnode = lfs_loadvnode,
157 .vfs_newvnode = lfs_newvnode,
158 .vfs_fhtovp = lfs_fhtovp,
159 .vfs_vptofh = lfs_vptofh,
160 .vfs_init = lfs_init,
161 .vfs_reinit = lfs_reinit,
162 .vfs_done = lfs_done,
163 .vfs_mountroot = lfs_mountroot,
164 .vfs_snapshot = (void *)eopnotsupp,
165 .vfs_extattrctl = lfs_extattrctl,
166 .vfs_suspendctl = (void *)eopnotsupp,
167 .vfs_renamelock_enter = genfs_renamelock_enter,
168 .vfs_renamelock_exit = genfs_renamelock_exit,
169 .vfs_fsync = (void *)eopnotsupp,
170 .vfs_opv_descs = lfs_vnodeopv_descs
171};
172
173const struct genfs_ops lfs_genfsops = {
174 .gop_size = lfs_gop_size,
175 .gop_alloc = ulfs_gop_alloc,
176 .gop_write = lfs_gop_write,
177 .gop_markupdate = ulfs_gop_markupdate,
178};
179
180struct shortlong {
181 const char *sname;
182 const char *lname;
183};
184
185static int
186sysctl_lfs_dostats(SYSCTLFN_ARGS)
187{
188 extern struct lfs_stats lfs_stats;
189 extern int lfs_dostats;
190 int error;
191
192 error = sysctl_lookup(SYSCTLFN_CALL(rnode));
193 if (error || newp == NULL)
194 return (error);
195
196 if (lfs_dostats == 0)
197 memset(&lfs_stats, 0, sizeof(lfs_stats));
198
199 return (0);
200}
201
202static void
203lfs_sysctl_setup(struct sysctllog **clog)
204{
205 int i;
206 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead,
207 lfs_fs_pagetrip, lfs_ignore_lazy_sync;
208#ifdef DEBUG
209 extern int lfs_debug_log_subsys[DLOG_MAX];
210 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */
211 { "rollforward", "Debug roll-forward code" },
212 { "alloc", "Debug inode allocation and free list" },
213 { "avail", "Debug space-available-now accounting" },
214 { "flush", "Debug flush triggers" },
215 { "lockedlist", "Debug locked list accounting" },
216 { "vnode_verbose", "Verbose per-vnode-written debugging" },
217 { "vnode", "Debug vnode use during segment write" },
218 { "segment", "Debug segment writing" },
219 { "seguse", "Debug segment used-bytes accounting" },
220 { "cleaner", "Debug cleaning routines" },
221 { "mount", "Debug mount/unmount routines" },
222 { "pagecache", "Debug UBC interactions" },
223 { "dirop", "Debug directory-operation accounting" },
224 { "malloc", "Debug private malloc accounting" },
225 };
226#endif /* DEBUG */
227 struct shortlong stat_names[] = { /* Must match lfs.h! */
228 { "segsused", "Number of new segments allocated" },
229 { "psegwrites", "Number of partial-segment writes" },
230 { "psyncwrites", "Number of synchronous partial-segment"
231 " writes" },
232 { "pcleanwrites", "Number of partial-segment writes by the"
233 " cleaner" },
234 { "blocktot", "Number of blocks written" },
235 { "cleanblocks", "Number of blocks written by the cleaner" },
236 { "ncheckpoints", "Number of checkpoints made" },
237 { "nwrites", "Number of whole writes" },
238 { "nsync_writes", "Number of synchronous writes" },
239 { "wait_exceeded", "Number of times writer waited for"
240 " cleaner" },
241 { "write_exceeded", "Number of times writer invoked flush" },
242 { "flush_invoked", "Number of times flush was invoked" },
243 { "vflush_invoked", "Number of time vflush was called" },
244 { "clean_inlocked", "Number of vnodes skipped for being dead" },
245 { "clean_vnlocked", "Number of vnodes skipped for vget failure" },
246 { "segs_reclaimed", "Number of segments reclaimed" },
247 };
248
249 sysctl_createv(clog, 0, NULL, NULL,
250 CTLFLAG_PERMANENT,
251 CTLTYPE_NODE, "lfs",
252 SYSCTL_DESCR("Log-structured file system"),
253 NULL, 0, NULL, 0,
254 CTL_VFS, 5, CTL_EOL);
255 /*
256 * XXX the "5" above could be dynamic, thereby eliminating one
257 * more instance of the "number to vfs" mapping problem, but
258 * "5" is the order as taken from sys/mount.h
259 */
260
261 sysctl_createv(clog, 0, NULL, NULL,
262 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
263 CTLTYPE_INT, "flushindir", NULL,
264 NULL, 0, &lfs_writeindir, 0,
265 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
266 sysctl_createv(clog, 0, NULL, NULL,
267 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
268 CTLTYPE_INT, "clean_vnhead", NULL,
269 NULL, 0, &lfs_clean_vnhead, 0,
270 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
271 sysctl_createv(clog, 0, NULL, NULL,
272 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
273 CTLTYPE_INT, "dostats",
274 SYSCTL_DESCR("Maintain statistics on LFS operations"),
275 sysctl_lfs_dostats, 0, &lfs_dostats, 0,
276 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
277 sysctl_createv(clog, 0, NULL, NULL,
278 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
279 CTLTYPE_INT, "pagetrip",
280 SYSCTL_DESCR("How many dirty pages in fs triggers"
281 " a flush"),
282 NULL, 0, &lfs_fs_pagetrip, 0,
283 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL);
284 sysctl_createv(clog, 0, NULL, NULL,
285 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
286 CTLTYPE_INT, "ignore_lazy_sync",
287 SYSCTL_DESCR("Lazy Sync is ignored entirely"),
288 NULL, 0, &lfs_ignore_lazy_sync, 0,
289 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL);
290#ifdef LFS_KERNEL_RFW
291 sysctl_createv(clog, 0, NULL, NULL,
292 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
293 CTLTYPE_INT, "rfw",
294 SYSCTL_DESCR("Use in-kernel roll-forward on mount"),
295 NULL, 0, &lfs_do_rfw, 0,
296 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL);
297#endif
298
299 sysctl_createv(clog, 0, NULL, NULL,
300 CTLFLAG_PERMANENT,
301 CTLTYPE_NODE, "stats",
302 SYSCTL_DESCR("Debugging options"),
303 NULL, 0, NULL, 0,
304 CTL_VFS, 5, LFS_STATS, CTL_EOL);
305 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) {
306 sysctl_createv(clog, 0, NULL, NULL,
307 CTLFLAG_PERMANENT|CTLFLAG_READONLY,
308 CTLTYPE_INT, stat_names[i].sname,
309 SYSCTL_DESCR(stat_names[i].lname),
310 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]),
311 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL);
312 }
313
314#ifdef DEBUG
315 sysctl_createv(clog, 0, NULL, NULL,
316 CTLFLAG_PERMANENT,
317 CTLTYPE_NODE, "debug",
318 SYSCTL_DESCR("Debugging options"),
319 NULL, 0, NULL, 0,
320 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL);
321 for (i = 0; i < DLOG_MAX; i++) {
322 sysctl_createv(clog, 0, NULL, NULL,
323 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
324 CTLTYPE_INT, dlog_names[i].sname,
325 SYSCTL_DESCR(dlog_names[i].lname),
326 NULL, 0, &(lfs_debug_log_subsys[i]), 0,
327 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL);
328 }
329#endif
330}
331
332/* old cleaner syscall interface. see VOP_FCNTL() */
333static const struct syscall_package lfs_syscalls[] = {
334 { SYS_lfs_bmapv, 0, (sy_call_t *)sys_lfs_bmapv },
335 { SYS_lfs_markv, 0, (sy_call_t *)sys_lfs_markv },
336 { SYS___lfs_segwait50, 0, (sy_call_t *)sys___lfs_segwait50 },
337 { SYS_lfs_segclean, 0, (sy_call_t *)sys_lfs_segclean },
338 { 0, 0, NULL },
339};
340
341static int
342lfs_modcmd(modcmd_t cmd, void *arg)
343{
344 int error;
345
346 switch (cmd) {
347 case MODULE_CMD_INIT:
348 error = syscall_establish(NULL, lfs_syscalls);
349 if (error)
350 return error;
351 error = vfs_attach(&lfs_vfsops);
352 if (error != 0) {
353 syscall_disestablish(NULL, lfs_syscalls);
354 break;
355 }
356 lfs_sysctl_setup(&lfs_sysctl_log);
357 break;
358 case MODULE_CMD_FINI:
359 error = vfs_detach(&lfs_vfsops);
360 if (error != 0)
361 break;
362 syscall_disestablish(NULL, lfs_syscalls);
363 sysctl_teardown(&lfs_sysctl_log);
364 break;
365 default:
366 error = ENOTTY;
367 break;
368 }
369
370 return (error);
371}
372
373/*
374 * XXX Same structure as FFS inodes? Should we share a common pool?
375 */
376struct pool lfs_inode_pool;
377struct pool lfs_dinode_pool;
378struct pool lfs_inoext_pool;
379struct pool lfs_lbnentry_pool;
380
381/*
382 * The writer daemon. UVM keeps track of how many dirty pages we are holding
383 * in lfs_subsys_pages; the daemon flushes the filesystem when this value
384 * crosses the (user-defined) threshhold LFS_MAX_PAGES.
385 */
386static void
387lfs_writerd(void *arg)
388{
389 struct mount *mp, *nmp;
390 struct lfs *fs;
391 struct vfsops *vfs = NULL;
392 int fsflags;
393 int skipc;
394 int lfsc;
395 int wrote_something = 0;
396
397 mutex_enter(&lfs_lock);
398 lfs_writer_daemon = curproc->p_pid;
399 lfs_writer_lid = curlwp->l_lid;
400 mutex_exit(&lfs_lock);
401
402 /* Take an extra reference to the LFS vfsops. */
403 vfs = vfs_getopsbyname(MOUNT_LFS);
404
405 mutex_enter(&lfs_lock);
406 for (;;) {
407 KASSERT(mutex_owned(&lfs_lock));
408 if (wrote_something == 0)
409 mtsleep(&lfs_writer_daemon, PVM, "lfswriter", hz/10 + 1,
410 &lfs_lock);
411
412 KASSERT(mutex_owned(&lfs_lock));
413 wrote_something = 0;
414
415 /*
416 * If global state wants a flush, flush everything.
417 */
418 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
419 locked_queue_bytes > LFS_MAX_BYTES ||
420 lfs_subsys_pages > LFS_MAX_PAGES) {
421
422 if (lfs_do_flush) {
423 DLOG((DLOG_FLUSH, "lfs_writerd: lfs_do_flush\n"));
424 }
425 if (locked_queue_count > LFS_MAX_BUFS) {
426 DLOG((DLOG_FLUSH, "lfs_writerd: lqc = %d, max %d\n",
427 locked_queue_count, LFS_MAX_BUFS));
428 }
429 if (locked_queue_bytes > LFS_MAX_BYTES) {
430 DLOG((DLOG_FLUSH, "lfs_writerd: lqb = %ld, max %ld\n",
431 locked_queue_bytes, LFS_MAX_BYTES));
432 }
433 if (lfs_subsys_pages > LFS_MAX_PAGES) {
434 DLOG((DLOG_FLUSH, "lfs_writerd: lssp = %d, max %d\n",
435 lfs_subsys_pages, LFS_MAX_PAGES));
436 }
437
438 lfs_flush(NULL, SEGM_WRITERD, 0);
439 lfs_do_flush = 0;
440 KASSERT(mutex_owned(&lfs_lock));
441 continue;
442 }
443 KASSERT(mutex_owned(&lfs_lock));
444 mutex_exit(&lfs_lock);
445
446 /*
447 * Look through the list of LFSs to see if any of them
448 * have requested pageouts.
449 */
450 mutex_enter(&mountlist_lock);
451 lfsc = 0;
452 skipc = 0;
453 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
454 if (vfs_busy(mp, &nmp)) {
455 ++skipc;
456 continue;
457 }
458 KASSERT(!mutex_owned(&lfs_lock));
459 if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS,
460 sizeof(mp->mnt_stat.f_fstypename)) == 0) {
461 ++lfsc;
462 fs = VFSTOULFS(mp)->um_lfs;
463 daddr_t ooffset = 0;
464 fsflags = SEGM_SINGLE;
465
466 mutex_enter(&lfs_lock);
467 ooffset = lfs_sb_getoffset(fs);
468
469 if (lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs) && fs->lfs_nowrap) {
470 /* Don't try to write if we're suspended */
471 mutex_exit(&lfs_lock);
472 vfs_unbusy(mp, false, &nmp);
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 vfs_unbusy(mp, false, &nmp);
481 continue;
482 }
483
484 if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) ||
485 lfs_dirvcount > LFS_MAX_DIROP) &&
486 fs->lfs_dirops == 0) {
487 fsflags &= ~SEGM_SINGLE;
488 fsflags |= SEGM_CKP;
489 DLOG((DLOG_FLUSH, "lfs_writerd: checkpoint\n"));
490 lfs_flush_fs(fs, fsflags);
491 } else if (fs->lfs_pdflush) {
492 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n"));
493 lfs_flush_fs(fs, fsflags);
494 } else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) {
495 DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n"));
496 mutex_exit(&lfs_lock);
497 lfs_writer_enter(fs, "wrdirop");
498 lfs_flush_pchain(fs);
499 lfs_writer_leave(fs);
500 mutex_enter(&lfs_lock);
501 }
502 if (lfs_sb_getoffset(fs) != ooffset)
503 ++wrote_something;
504 mutex_exit(&lfs_lock);
505 }
506 KASSERT(!mutex_owned(&lfs_lock));
507 vfs_unbusy(mp, false, &nmp);
508 }
509 if (lfsc + skipc == 0) {
510 mutex_enter(&lfs_lock);
511 lfs_writer_daemon = 0;
512 lfs_writer_lid = 0;
513 mutex_exit(&lfs_lock);
514 mutex_exit(&mountlist_lock);
515 break;
516 }
517 mutex_exit(&mountlist_lock);
518
519 mutex_enter(&lfs_lock);
520 }
521 KASSERT(!mutex_owned(&lfs_lock));
522 KASSERT(!mutex_owned(&mountlist_lock));
523
524 /* Give up our extra reference so the module can be unloaded. */
525 mutex_enter(&vfs_list_lock);
526 if (vfs != NULL)
527 vfs->vfs_refcount--;
528 mutex_exit(&vfs_list_lock);
529
530 /* Done! */
531 kthread_exit(0);
532}
533
534/*
535 * Initialize the filesystem, most work done by ulfs_init.
536 */
537void
538lfs_init(void)
539{
540
541 /*
542 * XXX: should we use separate pools for 32-bit and 64-bit
543 * dinodes?
544 */
545 malloc_type_attach(M_SEGMENT);
546 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
547 "lfsinopl", &pool_allocator_nointr, IPL_NONE);
548 pool_init(&lfs_dinode_pool, sizeof(union lfs_dinode), 0, 0, 0,
549 "lfsdinopl", &pool_allocator_nointr, IPL_NONE);
550 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
551 "lfsinoextpl", &pool_allocator_nointr, IPL_NONE);
552 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0,
553 "lfslbnpool", &pool_allocator_nointr, IPL_NONE);
554 ulfs_init();
555
556#ifdef DEBUG
557 memset(lfs_log, 0, sizeof(lfs_log));
558#endif
559 mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE);
560 cv_init(&locked_queue_cv, "lfsbuf");
561 cv_init(&lfs_writing_cv, "lfsflush");
562}
563
564void
565lfs_reinit(void)
566{
567 ulfs_reinit();
568}
569
570void
571lfs_done(void)
572{
573 ulfs_done();
574 mutex_destroy(&lfs_lock);
575 cv_destroy(&locked_queue_cv);
576 cv_destroy(&lfs_writing_cv);
577 pool_destroy(&lfs_inode_pool);
578 pool_destroy(&lfs_dinode_pool);
579 pool_destroy(&lfs_inoext_pool);
580 pool_destroy(&lfs_lbnentry_pool);
581 malloc_type_detach(M_SEGMENT);
582}
583
584/*
585 * Called by main() when ulfs is going to be mounted as root.
586 */
587int
588lfs_mountroot(void)
589{
590 extern struct vnode *rootvp;
591 struct lfs *fs = NULL; /* LFS */
592 struct mount *mp;
593 struct lwp *l = curlwp;
594 struct ulfsmount *ump;
595 int error;
596
597 if (device_class(root_device) != DV_DISK)
598 return (ENODEV);
599
600 if (rootdev == NODEV)
601 return (ENODEV);
602 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
603 vrele(rootvp);
604 return (error);
605 }
606 if ((error = lfs_mountfs(rootvp, mp, l))) {
607 vfs_unbusy(mp, false, NULL);
608 vfs_destroy(mp);
609 return (error);
610 }
611 mountlist_append(mp);
612 ump = VFSTOULFS(mp);
613 fs = ump->um_lfs;
614 lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname);
615 (void)lfs_statvfs(mp, &mp->mnt_stat);
616 vfs_unbusy(mp, false, NULL);
617 setrootfstime((time_t)lfs_sb_gettstamp(VFSTOULFS(mp)->um_lfs));
618 return (0);
619}
620
621/*
622 * VFS Operations.
623 *
624 * mount system call
625 */
626int
627lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
628{
629 struct lwp *l = curlwp;
630 struct vnode *devvp;
631 struct ulfs_args *args = data;
632 struct ulfsmount *ump = NULL;
633 struct lfs *fs = NULL; /* LFS */
634 int error = 0, update;
635 mode_t accessmode;
636
637 if (args == NULL)
638 return EINVAL;
639 if (*data_len < sizeof *args)
640 return EINVAL;
641
642 if (mp->mnt_flag & MNT_GETARGS) {
643 ump = VFSTOULFS(mp);
644 if (ump == NULL)
645 return EIO;
646 args->fspec = NULL;
647 *data_len = sizeof *args;
648 return 0;
649 }
650
651 update = mp->mnt_flag & MNT_UPDATE;
652
653 /* Check arguments */
654 if (args->fspec != NULL) {
655 /*
656 * Look up the name and verify that it's sane.
657 */
658 error = namei_simple_user(args->fspec,
659 NSM_FOLLOW_NOEMULROOT, &devvp);
660 if (error != 0)
661 return (error);
662
663 if (!update) {
664 /*
665 * Be sure this is a valid block device
666 */
667 if (devvp->v_type != VBLK)
668 error = ENOTBLK;
669 else if (bdevsw_lookup(devvp->v_rdev) == NULL)
670 error = ENXIO;
671 } else {
672 /*
673 * Be sure we're still naming the same device
674 * used for our initial mount
675 *
676 * XXX dholland 20151010: if namei gives us a
677 * different vnode for the same device,
678 * wouldn't it be better to use it going
679 * forward rather than ignore it in favor of
680 * the old one?
681 */
682 ump = VFSTOULFS(mp);
683 fs = ump->um_lfs;
684 if (devvp != fs->lfs_devvp) {
685 if (devvp->v_rdev != fs->lfs_devvp->v_rdev)
686 error = EINVAL;
687 else {
688 vrele(devvp);
689 devvp = fs->lfs_devvp;
690 vref(devvp);
691 }
692 }
693 }
694 } else {
695 if (!update) {
696 /* New mounts must have a filename for the device */
697 return (EINVAL);
698 } else {
699 /* Use the extant mount */
700 ump = VFSTOULFS(mp);
701 fs = ump->um_lfs;
702 devvp = fs->lfs_devvp;
703 vref(devvp);
704 }
705 }
706
707
708 /*
709 * If mount by non-root, then verify that user has necessary
710 * permissions on the device.
711 */
712 if (error == 0) {
713 accessmode = VREAD;
714 if (update ?
715 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
716 (mp->mnt_flag & MNT_RDONLY) == 0)
717 accessmode |= VWRITE;
718 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
719 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
720 KAUTH_REQ_SYSTEM_MOUNT_DEVICE, mp, devvp,
721 KAUTH_ARG(accessmode));
722 VOP_UNLOCK(devvp);
723 }
724
725 if (error) {
726 vrele(devvp);
727 return (error);
728 }
729
730 if (!update) {
731 int flags;
732
733 if (mp->mnt_flag & MNT_RDONLY)
734 flags = FREAD;
735 else
736 flags = FREAD|FWRITE;
737 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
738 error = VOP_OPEN(devvp, flags, FSCRED);
739 VOP_UNLOCK(devvp);
740 if (error)
741 goto fail;
742 error = lfs_mountfs(devvp, mp, l); /* LFS */
743 if (error) {
744 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
745 (void)VOP_CLOSE(devvp, flags, NOCRED);
746 VOP_UNLOCK(devvp);
747 goto fail;
748 }
749
750 ump = VFSTOULFS(mp);
751 fs = ump->um_lfs;
752 } else {
753 /*
754 * Update the mount.
755 */
756
757 /*
758 * The initial mount got a reference on this
759 * device, so drop the one obtained via
760 * namei(), above.
761 */
762 vrele(devvp);
763
764 ump = VFSTOULFS(mp);
765 fs = ump->um_lfs;
766
767 if (fs->lfs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
768 /*
769 * Changing from read/write to read-only.
770 * XXX: shouldn't we sync here? or does vfs do that?
771 */
772#ifdef LFS_QUOTA2
773 /* XXX: quotas should remain on when readonly */
774 if (fs->lfs_use_quota2) {
775 error = lfsquota2_umount(mp, 0);
776 if (error) {
777 return error;
778 }
779 }
780#endif
781 }
782
783 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
784 /*
785 * Changing from read-only to read/write.
786 * Note in the superblocks that we're writing.
787 */
788
789 /* XXX: quotas should have been on even if readonly */
790 if (fs->lfs_use_quota2) {
791#ifdef LFS_QUOTA2
792 error = lfs_quota2_mount(mp);
793#else
794 uprintf("%s: no kernel support for this "
795 "filesystem's quotas\n",
796 mp->mnt_stat.f_mntonname);
797 if (mp->mnt_flag & MNT_FORCE) {
798 uprintf("%s: mounting anyway; "
799 "fsck afterwards\n",
800 mp->mnt_stat.f_mntonname);
801 } else {
802 error = EINVAL;
803 }
804#endif
805 if (error) {
806 return error;
807 }
808 }
809
810 fs->lfs_ronly = 0;
811 if (lfs_sb_getpflags(fs) & LFS_PF_CLEAN) {
812 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN);
813 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
814 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
815 }
816 }
817 if (args->fspec == NULL)
818 return EINVAL;
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
827fail:
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 */
836static int
837lfs_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 */
869int
870lfs_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
884 cred = l ? l->l_cred : NOCRED;
885
886 /* The superblock is supposed to be 512 bytes. */
887 __CTASSERT(sizeof(struct dlfs) == DEV_BSIZE);
888
889 /*
890 * Flush out any old buffers remaining from a previous use.
891 */
892 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
893 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
894 VOP_UNLOCK(devvp);
895 if (error)
896 return (error);
897
898 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
899
900 /* Don't free random space on error. */
901 primarybuf = NULL;
902 altbuf = NULL;
903 ump = NULL;
904
905 sb_addr = LFS_LABELPAD / DEV_BSIZE;
906 while (1) {
907 /*
908 * Read in the superblock.
909 *
910 * Note that because LFS_SBPAD is substantially larger
911 * (8K) than the actual on-disk superblock (512 bytes)
912 * the buffer contains enough space to be used as a
913 * whole struct lfs (in-memory superblock) - we do this
914 * only so we can set and use the is64 and dobyteswap
915 * members. XXX this is gross and the logic here should
916 * be reworked.
917 */
918 error = bread(devvp, sb_addr, LFS_SBPAD, 0, &primarybuf);
919 if (error)
920 goto out;
921 primarysb = (struct lfs *)primarybuf->b_data;
922
923 /* Check the basics. */
924 error = lfs_checkmagic(primarysb);
925 if (error) {
926 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock wrong magic\n"));
927 goto out;
928 }
929 if (lfs_sb_getbsize(primarysb) > MAXBSIZE ||
930 lfs_sb_getversion(primarysb) > LFS_VERSION ||
931 lfs_sb_getbsize(primarysb) < sizeof(struct dlfs)) {
932 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
933 /* XXX needs translation */
934 error = EINVAL;
935 goto out;
936 }
937 if (lfs_sb_getinodefmt(primarysb) > LFS_MAXINODEFMT) {
938 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
939 lfs_sb_getinodefmt(primarysb)));
940 error = EINVAL;
941 goto out;
942 }
943
944 if (lfs_sb_getversion(primarysb) == 1)
945 fsbsize = DEV_BSIZE;
946 else {
947 fsbsize = 1 << lfs_sb_getffshift(primarysb);
948 /*
949 * Could be, if the frag size is large enough, that we
950 * don't have the "real" primary superblock. If that's
951 * the case, get the real one, and try again.
952 */
953 if (sb_addr != (lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT))) {
954 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
955 " 0x%llx is not right, trying 0x%llx\n",
956 (long long)sb_addr,
957 (long long)(lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT))));
958 sb_addr = lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT);
959 brelse(primarybuf, BC_INVAL);
960 continue;
961 }
962 }
963 break;
964 }
965
966 /*
967 * Check the second superblock to see which is newer; then mount
968 * using the older of the two. This is necessary to ensure that
969 * the filesystem is valid if it was not unmounted cleanly.
970 */
971
972 if (lfs_sb_getsboff(primarysb, 1) &&
973 lfs_sb_getsboff(primarysb, 1) - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
974 {
975 error = bread(devvp, lfs_sb_getsboff(primarysb, 1) * (fsbsize / DEV_BSIZE),
976 LFS_SBPAD, 0, &altbuf);
977 if (error)
978 goto out;
979 altsb = (struct lfs *)altbuf->b_data;
980
981 /*
982 * Note: this used to do the sanity check only if the
983 * timestamp/serial comparison required use of altsb;
984 * this way is less tolerant, but if altsb is corrupted
985 * enough that the magic number, version, and blocksize
986 * are bogus, why would the timestamp or serial fields
987 * mean anything either? If this kind of thing happens,
988 * you need to fsck anyway.
989 */
990
991 error = lfs_checkmagic(altsb);
992 if (error)
993 goto out;
994
995 /* Check the basics. */
996 if (lfs_sb_getbsize(altsb) > MAXBSIZE ||
997 lfs_sb_getversion(altsb) > LFS_VERSION ||
998 lfs_sb_getbsize(altsb) < sizeof(struct dlfs)) {
999 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
1000 " sanity failed\n"));
1001 error = EINVAL; /* XXX needs translation */
1002 goto out;
1003 }
1004
1005 if (lfs_sb_getversion(primarysb) == 1) {
1006 /* 1s resolution comparison */
1007 if (lfs_sb_gettstamp(altsb) < lfs_sb_gettstamp(primarysb))
1008 thesb = altsb;
1009 else
1010 thesb = primarysb;
1011 } else {
1012 /* monotonic infinite-resolution comparison */
1013 if (lfs_sb_getserial(altsb) < lfs_sb_getserial(primarysb))
1014 thesb = altsb;
1015 else
1016 thesb = primarysb;
1017 }
1018 } else {
1019 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock location"
1020 " daddr=0x%x\n", lfs_sb_getsboff(primarysb, 1)));
1021 error = EINVAL;
1022 goto out;
1023 }
1024
1025 /*
1026 * Allocate the mount structure, copy the superblock into it.
1027 * Note that the 32-bit and 64-bit superblocks are the same size.
1028 */
1029 fs = kmem_zalloc(sizeof(struct lfs), KM_SLEEP);
1030 memcpy(&fs->lfs_dlfs_u.u_32, &thesb->lfs_dlfs_u.u_32,
1031 sizeof(struct dlfs));
1032 fs->lfs_is64 = thesb->lfs_is64;
1033 fs->lfs_dobyteswap = thesb->lfs_dobyteswap;
1034 fs->lfs_hasolddirfmt = false; /* set for real below */
1035
1036 /* Compatibility */
1037 if (lfs_sb_getversion(fs) < 2) {
1038 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE);
1039 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs));
1040 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0));
1041 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs));
1042 lfs_sb_setfsbtodb(fs, 0);
1043 }
1044 if (lfs_sb_getresvseg(fs) == 0)
1045 lfs_sb_setresvseg(fs, MIN(lfs_sb_getminfreeseg(fs) - 1, \
1046 MAX(MIN_RESV_SEGS, lfs_sb_getminfreeseg(fs) / 2 + 1)));
1047
1048 /*
1049 * If we aren't going to be able to write meaningfully to this
1050 * filesystem, and were not mounted readonly, bomb out now.
1051 */
1052 if (lfs_fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
1053 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
1054 " we need BUFPAGES >= %lld\n",
1055 (long long)((bufmem_hiwater / bufmem_lowater) *
1056 LFS_INVERSE_MAX_BYTES(
1057 lfs_fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
1058 kmem_free(fs, sizeof(struct lfs));
1059 error = EFBIG; /* XXX needs translation */
1060 goto out;
1061 }
1062
1063 /* Before rolling forward, lock so vget will sleep for other procs */
1064 if (l != NULL) {
1065 fs->lfs_flags = LFS_NOTYET;
1066 fs->lfs_rfpid = l->l_proc->p_pid;
1067 }
1068
1069 ump = kmem_zalloc(sizeof(*ump), KM_SLEEP);
1070 ump->um_lfs = fs;
1071 ump->um_fstype = fs->lfs_is64 ? ULFS2 : ULFS1;
1072 /* ump->um_cleaner_thread = NULL; */
1073 brelse(primarybuf, BC_INVAL);
1074 brelse(altbuf, BC_INVAL);
1075 primarybuf = NULL;
1076 altbuf = NULL;
1077
1078
1079 /* Set up the I/O information */
1080 fs->lfs_devbsize = DEV_BSIZE;
1081 fs->lfs_iocount = 0;
1082 fs->lfs_diropwait = 0;
1083 fs->lfs_activesb = 0;
1084 lfs_sb_setuinodes(fs, 0);
1085 fs->lfs_ravail = 0;
1086 fs->lfs_favail = 0;
1087 fs->lfs_sbactive = 0;
1088
1089 /* Set up the ifile and lock aflags */
1090 fs->lfs_doifile = 0;
1091 fs->lfs_writer = 0;
1092 fs->lfs_dirops = 0;
1093 fs->lfs_nadirop = 0;
1094 fs->lfs_seglock = 0;
1095 fs->lfs_pdflush = 0;
1096 fs->lfs_sleepers = 0;
1097 fs->lfs_pages = 0;
1098 rw_init(&fs->lfs_fraglock);
1099 rw_init(&fs->lfs_iflock);
1100 cv_init(&fs->lfs_stopcv, "lfsstop");
1101
1102 /* Set the file system readonly/modify bits. */
1103 fs->lfs_ronly = ronly;
1104 if (ronly == 0)
1105 fs->lfs_fmod = 1;
1106
1107 /* Device we're using */
1108 dev = devvp->v_rdev;
1109 fs->lfs_dev = dev;
1110 fs->lfs_devvp = devvp;
1111
1112 /* ulfs-level information */
1113 fs->um_flags = 0;
1114 fs->um_bptrtodb = lfs_sb_getffshift(fs) - DEV_BSHIFT;
1115 fs->um_seqinc = lfs_sb_getfrag(fs);
1116 fs->um_nindir = lfs_sb_getnindir(fs);
1117 fs->um_lognindir = ffs(lfs_sb_getnindir(fs)) - 1;
1118 fs->um_maxsymlinklen = lfs_sb_getmaxsymlinklen(fs);
1119 fs->um_dirblksiz = LFS_DIRBLKSIZ;
1120 fs->um_maxfilesize = lfs_sb_getmaxfilesize(fs);
1121
1122 /* quota stuff */
1123 /* XXX: these need to come from the on-disk superblock to be used */
1124 fs->lfs_use_quota2 = 0;
1125 fs->lfs_quota_magic = 0;
1126 fs->lfs_quota_flags = 0;
1127 fs->lfs_quotaino[0] = 0;
1128 fs->lfs_quotaino[1] = 0;
1129
1130 /* Initialize the mount structure. */
1131 mp->mnt_data = ump;
1132 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
1133 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
1134 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
1135 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
1136 mp->mnt_stat.f_iosize = lfs_sb_getbsize(fs);
1137 mp->mnt_flag |= MNT_LOCAL;
1138 mp->mnt_fs_bshift = lfs_sb_getbshift(fs);
1139 if (fs->um_maxsymlinklen > 0)
1140 mp->mnt_iflag |= IMNT_DTYPE;
1141 else
1142 fs->lfs_hasolddirfmt = true;
1143
1144 ump->um_mountp = mp;
1145 for (i = 0; i < ULFS_MAXQUOTAS; i++)
1146 ump->um_quotas[i] = NULLVP;
1147 spec_node_setmountedfs(devvp, mp);
1148
1149 /* Set up reserved memory for pageout */
1150 lfs_setup_resblks(fs);
1151 /* Set up vdirop tailq */
1152 TAILQ_INIT(&fs->lfs_dchainhd);
1153 /* and paging tailq */
1154 TAILQ_INIT(&fs->lfs_pchainhd);
1155 /* and delayed segment accounting for truncation list */
1156 LIST_INIT(&fs->lfs_segdhd);
1157
1158 /*
1159 * We use the ifile vnode for almost every operation. Instead of
1160 * retrieving it from the hash table each time we retrieve it here,
1161 * artificially increment the reference count and keep a pointer
1162 * to it in the incore copy of the superblock.
1163 */
1164 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1165 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
1166 goto out;
1167 }
1168 fs->lfs_ivnode = vp;
1169 vref(vp);
1170
1171 /* Set up inode bitmap and order free list */
1172 lfs_order_freelist(fs);
1173
1174 /* Set up segment usage flags for the autocleaner. */
1175 fs->lfs_nactive = 0;
1176 fs->lfs_suflags = malloc(2 * sizeof(u_int32_t *),
1177 M_SEGMENT, M_WAITOK);
1178 fs->lfs_suflags[0] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t),
1179 M_SEGMENT, M_WAITOK);
1180 fs->lfs_suflags[1] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t),
1181 M_SEGMENT, M_WAITOK);
1182 memset(fs->lfs_suflags[1], 0, lfs_sb_getnseg(fs) * sizeof(u_int32_t));
1183 for (i = 0; i < lfs_sb_getnseg(fs); i++) {
1184 int changed;
1185 struct buf *bp;
1186
1187 LFS_SEGENTRY(sup, fs, i, bp);
1188 changed = 0;
1189 if (!ronly) {
1190 if (sup->su_nbytes == 0 &&
1191 !(sup->su_flags & SEGUSE_EMPTY)) {
1192 sup->su_flags |= SEGUSE_EMPTY;
1193 ++changed;
1194 } else if (!(sup->su_nbytes == 0) &&
1195 (sup->su_flags & SEGUSE_EMPTY)) {
1196 sup->su_flags &= ~SEGUSE_EMPTY;
1197 ++changed;
1198 }
1199 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
1200 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
1201 ++changed;
1202 }
1203 }
1204 fs->lfs_suflags[0][i] = sup->su_flags;
1205 if (changed)
1206 LFS_WRITESEGENTRY(sup, fs, i, bp);
1207 else
1208 brelse(bp, 0);
1209 }
1210
1211 /*
1212 * XXX: if the fs has quotas, quotas should be on even if
1213 * readonly. Otherwise you can't query the quota info!
1214 * However, that's not how the quota2 code got written and I
1215 * don't know if it'll behave itself if enabled while
1216 * readonly, so for now use the same enable logic as ffs.
1217 *
1218 * XXX: also, if you use the -f behavior allowed here (and
1219 * equivalently above for remount) it will corrupt the fs. It
1220 * ought not to allow that. It should allow mounting readonly
1221 * if there are quotas and the kernel doesn't have the quota
1222 * code, but only readonly.
1223 *
1224 * XXX: and if you use the -f behavior allowed here it will
1225 * likely crash at unmount time (or remount time) because we
1226 * think quotas are active.
1227 *
1228 * Although none of this applies until there's a way to set
1229 * lfs_use_quota2 and have quotas in the fs at all.
1230 */
1231 if (!ronly && fs->lfs_use_quota2) {
1232#ifdef LFS_QUOTA2
1233 error = lfs_quota2_mount(mp);
1234#else
1235 uprintf("%s: no kernel support for this filesystem's quotas\n",
1236 mp->mnt_stat.f_mntonname);
1237 if (mp->mnt_flag & MNT_FORCE) {
1238 uprintf("%s: mounting anyway; fsck afterwards\n",
1239 mp->mnt_stat.f_mntonname);
1240 } else {
1241 error = EINVAL;
1242 }
1243#endif
1244 if (error) {
1245 /* XXX XXX must clean up the stuff immediately above */
1246 printf("lfs_mountfs: sorry, leaking some memory\n");
1247 goto out;
1248 }
1249 }
1250
1251#ifdef LFS_KERNEL_RFW
1252 lfs_roll_forward(fs, mp, l);
1253#endif
1254
1255 /* If writing, sb is not clean; record in case of immediate crash */
1256 if (!fs->lfs_ronly) {
1257 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN);
1258 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
1259 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
1260 }
1261
1262 /* Allow vget now that roll-forward is complete */
1263 fs->lfs_flags &= ~(LFS_NOTYET);
1264 wakeup(&fs->lfs_flags);
1265
1266 /*
1267 * Initialize the ifile cleaner info with information from
1268 * the superblock.
1269 */
1270 {
1271 struct buf *bp;
1272
1273 LFS_CLEANERINFO(cip, fs, bp);
1274 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs));
1275 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs));
1276 lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs));
1277 lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs));
1278 (void) LFS_BWRITE_LOG(bp); /* Ifile */
1279 }
1280
1281 /*
1282 * Mark the current segment as ACTIVE, since we're going to
1283 * be writing to it.
1284 */
1285 {
1286 struct buf *bp;
1287
1288 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp);
1289 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1290 fs->lfs_nactive++;
1291 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); /* Ifile */
1292 }
1293
1294 /* Now that roll-forward is done, unlock the Ifile */
1295 vput(vp);
1296
1297 /* Start the pagedaemon-anticipating daemon */
1298 mutex_enter(&lfs_lock);
1299 if (lfs_writer_daemon == 0 && lfs_writer_lid == 0 &&
1300 kthread_create(PRI_BIO, 0, NULL,
1301 lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1302 panic("fork lfs_writer");
1303 mutex_exit(&lfs_lock);
1304
1305 printf("WARNING: the log-structured file system is experimental\n"
1306 "WARNING: it may cause system crashes and/or corrupt data\n");
1307
1308 return (0);
1309
1310out:
1311 if (primarybuf)
1312 brelse(primarybuf, BC_INVAL);
1313 if (altbuf)
1314 brelse(altbuf, BC_INVAL);
1315 if (ump) {
1316 kmem_free(ump->um_lfs, sizeof(struct lfs));
1317 kmem_free(ump, sizeof(*ump));
1318 mp->mnt_data = NULL;
1319 }
1320
1321 return (error);
1322}
1323
1324/*
1325 * unmount system call
1326 */
1327int
1328lfs_unmount(struct mount *mp, int mntflags)
1329{
1330 struct lwp *l = curlwp;
1331 struct ulfsmount *ump;
1332 struct lfs *fs;
1333 int error, flags, ronly;
1334 vnode_t *vp;
1335
1336 flags = 0;
1337 if (mntflags & MNT_FORCE)
1338 flags |= FORCECLOSE;
1339
1340 ump = VFSTOULFS(mp);
1341 fs = ump->um_lfs;
1342
1343 /* Two checkpoints */
1344 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1345 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1346
1347 /* wake up the cleaner so it can die */
1348 /* XXX: shouldn't this be *after* the error cases below? */
1349 lfs_wakeup_cleaner(fs);
1350 mutex_enter(&lfs_lock);
1351 while (fs->lfs_sleepers)
1352 mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1353 &lfs_lock);
1354 mutex_exit(&lfs_lock);
1355
1356#ifdef LFS_EXTATTR
1357 if (ump->um_fstype == ULFS1) {
1358 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) {
1359 ulfs_extattr_stop(mp, curlwp);
1360 }
1361 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED) {
1362 ulfs_extattr_uepm_destroy(&ump->um_extattr);
1363 mp->mnt_flag &= ~MNT_EXTATTR;
1364 }
1365 }
1366#endif
1367#ifdef LFS_QUOTA
1368 if ((error = lfsquota1_umount(mp, flags)) != 0)
1369 return (error);
1370#endif
1371#ifdef LFS_QUOTA2
1372 if ((error = lfsquota2_umount(mp, flags)) != 0)
1373 return (error);
1374#endif
1375 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1376 return (error);
1377 if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0)
1378 return (error);
1379 vp = fs->lfs_ivnode;
1380 mutex_enter(vp->v_interlock);
1381 if (LIST_FIRST(&vp->v_dirtyblkhd))
1382 panic("lfs_unmount: still dirty blocks on ifile vnode");
1383 mutex_exit(vp->v_interlock);
1384
1385 /* Explicitly write the superblock, to update serial and pflags */
1386 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) | LFS_PF_CLEAN);
1387 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
1388 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
1389 mutex_enter(&lfs_lock);
1390 while (fs->lfs_iocount)
1391 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
1392 &lfs_lock);
1393 mutex_exit(&lfs_lock);
1394
1395 /* Finish with the Ifile, now that we're done with it */
1396 vgone(fs->lfs_ivnode);
1397
1398 ronly = !fs->lfs_ronly;
1399 if (fs->lfs_devvp->v_type != VBAD)
1400 spec_node_setmountedfs(fs->lfs_devvp, NULL);
1401 vn_lock(fs->lfs_devvp, LK_EXCLUSIVE | LK_RETRY);
1402 error = VOP_CLOSE(fs->lfs_devvp,
1403 ronly ? FREAD : FREAD|FWRITE, NOCRED);
1404 vput(fs->lfs_devvp);
1405
1406 /* Complain about page leakage */
1407 if (fs->lfs_pages > 0)
1408 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
1409 fs->lfs_pages, lfs_subsys_pages);
1410
1411 /* Free per-mount data structures */
1412 free(fs->lfs_ino_bitmap, M_SEGMENT);
1413 free(fs->lfs_suflags[0], M_SEGMENT);
1414 free(fs->lfs_suflags[1], M_SEGMENT);
1415 free(fs->lfs_suflags, M_SEGMENT);
1416 lfs_free_resblks(fs);
1417 cv_destroy(&fs->lfs_stopcv);
1418 rw_destroy(&fs->lfs_fraglock);
1419 rw_destroy(&fs->lfs_iflock);
1420
1421 kmem_free(fs, sizeof(struct lfs));
1422 kmem_free(ump, sizeof(*ump));
1423
1424 mp->mnt_data = NULL;
1425 mp->mnt_flag &= ~MNT_LOCAL;
1426 return (error);
1427}
1428
1429/*
1430 * Get file system statistics.
1431 *
1432 * NB: We don't lock to access the superblock here, because it's not
1433 * really that important if we get it wrong.
1434 */
1435int
1436lfs_statvfs(struct mount *mp, struct statvfs *sbp)
1437{
1438 struct lfs *fs;
1439 struct ulfsmount *ump;
1440
1441 ump = VFSTOULFS(mp);
1442 fs = ump->um_lfs;
1443
1444 sbp->f_bsize = lfs_sb_getbsize(fs);
1445 sbp->f_frsize = lfs_sb_getfsize(fs);
1446 sbp->f_iosize = lfs_sb_getbsize(fs);
1447 sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks;
1448
1449 sbp->f_bfree = LFS_EST_BFREE(fs);
1450 /*
1451 * XXX this should be lfs_sb_getsize (measured in frags)
1452 * rather than dsize (measured in diskblocks). However,
1453 * getsize needs a format version check (for version 1 it
1454 * needs to be blockstofrags'd) so for the moment I'm going to
1455 * leave this... it won't fire wrongly as frags are at least
1456 * as big as diskblocks.
1457 */
1458 KASSERT(sbp->f_bfree <= lfs_sb_getdsize(fs));
1459#if 0
1460 if (sbp->f_bfree < 0)
1461 sbp->f_bfree = 0;
1462#endif
1463
1464 sbp->f_bresvd = LFS_EST_RSVD(fs);
1465 if (sbp->f_bfree > sbp->f_bresvd)
1466 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1467 else
1468 sbp->f_bavail = 0;
1469
1470 /* XXX: huh? - dholland 20150728 */
1471 sbp->f_files = lfs_sb_getbfree(fs) / lfs_btofsb(fs, lfs_sb_getibsize(fs))
1472 * LFS_INOPB(fs);
1473 sbp->f_ffree = sbp->f_files - lfs_sb_getnfiles(fs);
1474 sbp->f_favail = sbp->f_ffree;
1475 sbp->f_fresvd = 0;
1476 copy_statvfs_info(sbp, mp);
1477 return (0);
1478}
1479
1480/*
1481 * Go through the disk queues to initiate sandbagged IO;
1482 * go through the inodes to write those that have been modified;
1483 * initiate the writing of the super block if it has been modified.
1484 *
1485 * Note: we are always called with the filesystem marked `MPBUSY'.
1486 */
1487int
1488lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
1489{
1490 int error;
1491 struct lfs *fs;
1492
1493 fs = VFSTOULFS(mp)->um_lfs;
1494 if (fs->lfs_ronly)
1495 return 0;
1496
1497 /* Snapshots should not hose the syncer */
1498 /*
1499 * XXX Sync can block here anyway, since we don't have a very
1500 * XXX good idea of how much data is pending. If it's more
1501 * XXX than a segment and lfs_nextseg is close to the end of
1502 * XXX the log, we'll likely block.
1503 */
1504 mutex_enter(&lfs_lock);
1505 if (fs->lfs_nowrap && lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs)) {
1506 mutex_exit(&lfs_lock);
1507 return 0;
1508 }
1509 mutex_exit(&lfs_lock);
1510
1511 lfs_writer_enter(fs, "lfs_dirops");
1512
1513 /* All syncs must be checkpoints until roll-forward is implemented. */
1514 DLOG((DLOG_FLUSH, "lfs_sync at 0x%jx\n",
1515 (uintmax_t)lfs_sb_getoffset(fs)));
1516 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1517 lfs_writer_leave(fs);
1518#ifdef LFS_QUOTA
1519 lfs_qsync(mp);
1520#endif
1521 return (error);
1522}
1523
1524/*
1525 * Look up an LFS dinode number to find its incore vnode. If not already
1526 * in core, read it in from the specified device. Return the inode locked.
1527 * Detection and handling of mount points must be done by the calling routine.
1528 */
1529int
1530lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1531{
1532 int error;
1533
1534 error = vcache_get(mp, &ino, sizeof(ino), vpp);
1535 if (error)
1536 return error;
1537 error = vn_lock(*vpp, LK_EXCLUSIVE);
1538 if (error) {
1539 vrele(*vpp);
1540 *vpp = NULL;
1541 return error;
1542 }
1543
1544 return 0;
1545}
1546
1547/*
1548 * Create a new vnode/inode pair and initialize what fields we can.
1549 */
1550static void
1551lfs_init_vnode(struct ulfsmount *ump, ino_t ino, struct vnode *vp)
1552{
1553 struct lfs *fs = ump->um_lfs;
1554 struct inode *ip;
1555 union lfs_dinode *dp;
1556
1557 ASSERT_NO_SEGLOCK(fs);
1558
1559 /* Initialize the inode. */
1560 ip = pool_get(&lfs_inode_pool, PR_WAITOK);
1561 memset(ip, 0, sizeof(*ip));
1562 dp = pool_get(&lfs_dinode_pool, PR_WAITOK);
1563 memset(dp, 0, sizeof(*dp));
1564 ip->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK);
1565 memset(ip->inode_ext.lfs, 0, sizeof(*ip->inode_ext.lfs));
1566 ip->i_din = dp;
1567 ip->i_ump = ump;
1568 ip->i_vnode = vp;
1569 ip->i_dev = fs->lfs_dev;
1570 lfs_dino_setinumber(fs, dp, ino);
1571 ip->i_number = ino;
1572 ip->i_lfs = fs;
1573 ip->i_lfs_effnblks = 0;
1574 SPLAY_INIT(&ip->i_lfs_lbtree);
1575 ip->i_lfs_nbtree = 0;
1576 LIST_INIT(&ip->i_lfs_segdhd);
1577
1578 vp->v_tag = VT_LFS;
1579 vp->v_op = lfs_vnodeop_p;
1580 vp->v_data = ip;
1581}
1582
1583/*
1584 * Undo lfs_init_vnode().
1585 */
1586static void
1587lfs_deinit_vnode(struct ulfsmount *ump, struct vnode *vp)
1588{
1589 struct inode *ip = VTOI(vp);
1590
1591 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs);
1592 pool_put(&lfs_dinode_pool, ip->i_din);
1593 pool_put(&lfs_inode_pool, ip);
1594 vp->v_data = NULL;
1595}
1596
1597/*
1598 * Read an inode from disk and initialize this vnode / inode pair.
1599 * Caller assures no other thread will try to load this inode.
1600 */
1601int
1602lfs_loadvnode(struct mount *mp, struct vnode *vp,
1603 const void *key, size_t key_len, const void **new_key)
1604{
1605 struct lfs *fs;
1606 union lfs_dinode *dip;
1607 struct inode *ip;
1608 struct buf *bp;
1609 IFILE *ifp;
1610 struct ulfsmount *ump;
1611 ino_t ino;
1612 daddr_t daddr;
1613 int error, retries;
1614 struct timespec ts;
1615
1616 KASSERT(key_len == sizeof(ino));
1617 memcpy(&ino, key, key_len);
1618
1619 memset(&ts, 0, sizeof ts); /* XXX gcc */
1620
1621 ump = VFSTOULFS(mp);
1622 fs = ump->um_lfs;
1623
1624 /*
1625 * If the filesystem is not completely mounted yet, suspend
1626 * any access requests (wait for roll-forward to complete).
1627 */
1628 mutex_enter(&lfs_lock);
1629 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1630 mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1631 &lfs_lock);
1632 mutex_exit(&lfs_lock);
1633
1634 /* Translate the inode number to a disk address. */
1635 if (ino == LFS_IFILE_INUM)
1636 daddr = lfs_sb_getidaddr(fs);
1637 else {
1638 /* XXX bounds-check this too */
1639 LFS_IENTRY(ifp, fs, ino, bp);
1640 daddr = lfs_if_getdaddr(fs, ifp);
1641 if (lfs_sb_getversion(fs) > 1) {
1642 ts.tv_sec = lfs_if_getatime_sec(fs, ifp);
1643 ts.tv_nsec = lfs_if_getatime_nsec(fs, ifp);
1644 }
1645
1646 brelse(bp, 0);
1647 if (daddr == LFS_UNUSED_DADDR)
1648 return (ENOENT);
1649 }
1650
1651 /* Allocate/init new vnode/inode. */
1652 lfs_init_vnode(ump, ino, vp);
1653 ip = VTOI(vp);
1654
1655 /* If the cleaner supplied the inode, use it. */
1656 if (curlwp == fs->lfs_cleaner_thread && fs->lfs_cleaner_hint != NULL &&
1657 fs->lfs_cleaner_hint->bi_lbn == LFS_UNUSED_LBN) {
1658 dip = fs->lfs_cleaner_hint->bi_bp;
1659 if (fs->lfs_is64) {
1660 error = copyin(dip, &ip->i_din->u_64,
1661 sizeof(struct lfs64_dinode));
1662 } else {
1663 error = copyin(dip, &ip->i_din->u_32,
1664 sizeof(struct lfs32_dinode));
1665 }
1666 if (error) {
1667 lfs_deinit_vnode(ump, vp);
1668 return error;
1669 }
1670 KASSERT(ip->i_number == ino);
1671 goto out;
1672 }
1673
1674 /* Read in the disk contents for the inode, copy into the inode. */
1675 retries = 0;
1676again:
1677 error = bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr),
1678 (lfs_sb_getversion(fs) == 1 ? lfs_sb_getbsize(fs) : lfs_sb_getibsize(fs)),
1679 0, &bp);
1680 if (error) {
1681 lfs_deinit_vnode(ump, vp);
1682 return error;
1683 }
1684
1685 dip = lfs_ifind(fs, ino, bp);
1686 if (dip == NULL) {
1687 /* Assume write has not completed yet; try again */
1688 brelse(bp, BC_INVAL);
1689 ++retries;
1690 if (retries <= LFS_IFIND_RETRIES) {
1691 mutex_enter(&lfs_lock);
1692 if (fs->lfs_iocount) {
1693 DLOG((DLOG_VNODE,
1694 "%s: dinode %d not found, retrying...\n",
1695 __func__, ino));
1696 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
1697 "lfs ifind", 1, &lfs_lock);
1698 } else
1699 retries = LFS_IFIND_RETRIES;
1700 mutex_exit(&lfs_lock);
1701 goto again;
1702 }
1703#ifdef DEBUG
1704 /* If the seglock is held look at the bpp to see
1705 what is there anyway */
1706 mutex_enter(&lfs_lock);
1707 if (fs->lfs_seglock > 0) {
1708 struct buf **bpp;
1709 union lfs_dinode *dp;
1710 int i;
1711
1712 for (bpp = fs->lfs_sp->bpp;
1713 bpp != fs->lfs_sp->cbpp; ++bpp) {
1714 if ((*bpp)->b_vp == fs->lfs_ivnode &&
1715 bpp != fs->lfs_sp->bpp) {
1716 /* Inode block */
1717 printf("%s: block 0x%" PRIx64 ": ",
1718 __func__, (*bpp)->b_blkno);
1719 for (i = 0; i < LFS_INOPB(fs); i++) {
1720 dp = DINO_IN_BLOCK(fs,
1721 (*bpp)->b_data, i);
1722 if (lfs_dino_getinumber(fs, dp))
1723 printf("%ju ",
1724 (uintmax_t)lfs_dino_getinumber(fs, dp));
1725 }
1726 printf("\n");
1727 }
1728 }
1729 }
1730 mutex_exit(&lfs_lock);
1731#endif /* DEBUG */
1732 panic("lfs_loadvnode: dinode not found");
1733 }
1734 lfs_copy_dinode(fs, ip->i_din, dip);
1735 brelse(bp, 0);
1736
1737out:
1738 if (lfs_sb_getversion(fs) > 1) {
1739 lfs_dino_setatime(fs, ip->i_din, ts.tv_sec);
1740 lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec);
1741 }
1742
1743 lfs_vinit(mp, &vp);
1744
1745 *new_key = &ip->i_number;
1746 return 0;
1747}
1748
1749/*
1750 * Create a new inode and initialize this vnode / inode pair.
1751 */
1752int
1753lfs_newvnode(struct mount *mp, struct vnode *dvp, struct vnode *vp,
1754 struct vattr *vap, kauth_cred_t cred,
1755 size_t *key_len, const void **new_key)
1756{
1757 ino_t ino;
1758 struct inode *ip;
1759 struct ulfsmount *ump;
1760 struct lfs *fs;
1761 int error, mode, gen;
1762
1763 KASSERT(dvp != NULL || vap->va_fileid > 0);
1764 KASSERT(dvp != NULL && dvp->v_mount == mp);
1765 KASSERT(vap->va_type != VNON);
1766
1767 *key_len = sizeof(ino);
1768 ump = VFSTOULFS(mp);
1769 fs = ump->um_lfs;
1770 mode = MAKEIMODE(vap->va_type, vap->va_mode);
1771
1772 /*
1773 * Allocate fresh inode. With "dvp == NULL" take the inode number
1774 * and version from "vap".
1775 */
1776 if (dvp == NULL) {
1777 ino = vap->va_fileid;
1778 gen = vap->va_gen;
1779 error = lfs_valloc_fixed(fs, ino, gen);
1780 } else {
1781 error = lfs_valloc(dvp, mode, cred, &ino, &gen);
1782 }
1783 if (error)
1784 return error;
1785
1786 /* Attach inode to vnode. */
1787 lfs_init_vnode(ump, ino, vp);
1788 ip = VTOI(vp);
1789
1790 mutex_enter(&lfs_lock);
1791 LFS_SET_UINO(ip, IN_CHANGE);
1792 mutex_exit(&lfs_lock);
1793
1794 /* Note no blocks yet */
1795 ip->i_lfs_hiblk = -1;
1796
1797 /* Set a new generation number for this inode. */
1798 ip->i_gen = gen;
1799 lfs_dino_setgen(fs, ip->i_din, gen);
1800
1801 memset(ip->i_lfs_fragsize, 0,
1802 ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
1803
1804 /* Set uid / gid. */
1805 if (cred == NOCRED || cred == FSCRED) {
1806 ip->i_gid = 0;
1807 ip->i_uid = 0;
1808 } else {
1809 ip->i_gid = VTOI(dvp)->i_gid;
1810 ip->i_uid = kauth_cred_geteuid(cred);
1811 }
1812 DIP_ASSIGN(ip, gid, ip->i_gid);
1813 DIP_ASSIGN(ip, uid, ip->i_uid);
1814
1815#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
1816 error = lfs_chkiq(ip, 1, cred, 0);
1817 if (error) {
1818 lfs_vfree(dvp, ino, mode);
1819 lfs_deinit_vnode(ump, vp);
1820
1821 return error;
1822 }
1823#endif
1824
1825 /* Set type and finalize. */
1826 ip->i_flags = 0;
1827 DIP_ASSIGN(ip, flags, 0);
1828 ip->i_mode = mode;
1829 DIP_ASSIGN(ip, mode, mode);
1830 if (vap->va_rdev != VNOVAL) {
1831 /*
1832 * Want to be able to use this to make badblock
1833 * inodes, so don't truncate the dev number.
1834 */
1835 // XXX clean this up
1836 if (ump->um_fstype == ULFS1)
1837 ip->i_din->u_32.di_rdev = ulfs_rw32(vap->va_rdev,
1838 ULFS_MPNEEDSWAP(fs));
1839 else
1840 ip->i_din->u_64.di_rdev = ulfs_rw64(vap->va_rdev,
1841 ULFS_MPNEEDSWAP(fs));
1842 }
1843 lfs_vinit(mp, &vp);
1844
1845 *new_key = &ip->i_number;
1846 return 0;
1847}
1848
1849/*
1850 * File handle to vnode
1851 */
1852int
1853lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1854{
1855 struct lfid lfh;
1856 struct lfs *fs;
1857
1858 if (fhp->fid_len != sizeof(struct lfid))
1859 return EINVAL;
1860
1861 memcpy(&lfh, fhp, sizeof(lfh));
1862 if (lfh.lfid_ino < LFS_IFILE_INUM)
1863 return ESTALE;
1864
1865 fs = VFSTOULFS(mp)->um_lfs;
1866 if (lfh.lfid_ident != lfs_sb_getident(fs))
1867 return ESTALE;
1868
1869 if (lfh.lfid_ino >
1870 ((lfs_dino_getsize(fs, VTOI(fs->lfs_ivnode)->i_din) >> lfs_sb_getbshift(fs)) -
1871 lfs_sb_getcleansz(fs) - lfs_sb_getsegtabsz(fs)) * lfs_sb_getifpb(fs))
1872 return ESTALE;
1873
1874 return (ulfs_fhtovp(mp, &lfh.lfid_ufid, vpp));
1875}
1876
1877/*
1878 * Vnode pointer to File handle
1879 */
1880/* ARGSUSED */
1881int
1882lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
1883{
1884 struct inode *ip;
1885 struct lfid lfh;
1886
1887 if (*fh_size < sizeof(struct lfid)) {
1888 *fh_size = sizeof(struct lfid);
1889 return E2BIG;
1890 }
1891 *fh_size = sizeof(struct lfid);
1892 ip = VTOI(vp);
1893 memset(&lfh, 0, sizeof(lfh));
1894 lfh.lfid_len = sizeof(struct lfid);
1895 lfh.lfid_ino = ip->i_number;
1896 lfh.lfid_gen = ip->i_gen;
1897 lfh.lfid_ident = lfs_sb_getident(ip->i_lfs);
1898 memcpy(fhp, &lfh, sizeof(lfh));
1899 return (0);
1900}
1901
1902/*
1903 * ulfs_bmaparray callback function for writing.
1904 *
1905 * Since blocks will be written to the new segment anyway,
1906 * we don't care about current daddr of them.
1907 */
1908static bool
1909lfs_issequential_hole(const struct lfs *fs,
1910 daddr_t daddr0, daddr_t daddr1)
1911{
1912 (void)fs; /* not used */
1913
1914 KASSERT(daddr0 == UNWRITTEN ||
1915 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR(fs)));
1916 KASSERT(daddr1 == UNWRITTEN ||
1917 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR(fs)));
1918
1919 /* NOTE: all we want to know here is 'hole or not'. */
1920 /* NOTE: UNASSIGNED is converted to 0 by ulfs_bmaparray. */
1921
1922 /*
1923 * treat UNWRITTENs and all resident blocks as 'contiguous'
1924 */
1925 if (daddr0 != 0 && daddr1 != 0)
1926 return true;
1927
1928 /*
1929 * both are in hole?
1930 */
1931 if (daddr0 == 0 && daddr1 == 0)
1932 return true; /* all holes are 'contiguous' for us. */
1933
1934 return false;
1935}
1936
1937/*
1938 * lfs_gop_write functions exactly like genfs_gop_write, except that
1939 * (1) it requires the seglock to be held by its caller, and sp->fip
1940 * to be properly initialized (it will return without re-initializing
1941 * sp->fip, and without calling lfs_writeseg).
1942 * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1943 * to determine how large a block it can write at once (though it does
1944 * still use VOP_BMAP to find holes in the file);
1945 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1946 * (leaving lfs_writeseg to deal with the cluster blocks, so we might
1947 * now have clusters of clusters, ick.)
1948 */
1949static int
1950lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1951 int flags)
1952{
1953 int i, error, run, haveeof = 0;
1954 int fs_bshift;
1955 vaddr_t kva;
1956 off_t eof, offset, startoffset = 0;
1957 size_t bytes, iobytes, skipbytes;
1958 bool async = (flags & PGO_SYNCIO) == 0;
1959 daddr_t lbn, blkno;
1960 struct vm_page *pg;
1961 struct buf *mbp, *bp;
1962 struct vnode *devvp = VTOI(vp)->i_devvp;
1963 struct inode *ip = VTOI(vp);
1964 struct lfs *fs = ip->i_lfs;
1965 struct segment *sp = fs->lfs_sp;
1966 SEGSUM *ssp;
1967 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1968 const char * failreason = NULL;
1969
1970 ASSERT_SEGLOCK(fs);
1971
1972 /* The Ifile lives in the buffer cache */
1973 KASSERT(vp != fs->lfs_ivnode);
1974
1975 /*
1976 * We don't want to fill the disk before the cleaner has a chance
1977 * to make room for us. If we're in danger of doing that, fail
1978 * with EAGAIN. The caller will have to notice this, unlock
1979 * so the cleaner can run, relock and try again.
1980 *
1981 * We must write everything, however, if our vnode is being
1982 * reclaimed.
1983 */
1984 mutex_enter(vp->v_interlock);
1985 if (LFS_STARVED_FOR_SEGS(fs) && vdead_check(vp, VDEAD_NOWAIT) == 0) {
1986 mutex_exit(vp->v_interlock);
1987 failreason = "Starved for segs and not flushing vp";
1988 goto tryagain;
1989 }
1990 mutex_exit(vp->v_interlock);
1991
1992 /*
1993 * Sometimes things slip past the filters in lfs_putpages,
1994 * and the pagedaemon tries to write pages---problem is
1995 * that the pagedaemon never acquires the segment lock.
1996 *
1997 * Alternatively, pages that were clean when we called
1998 * genfs_putpages may have become dirty in the meantime. In this
1999 * case the segment header is not properly set up for blocks
2000 * to be added to it.
2001 *
2002 * Unbusy and unclean the pages, and put them on the ACTIVE
2003 * queue under the hypothesis that they couldn't have got here
2004 * unless they were modified *quite* recently.
2005 *
2006 * XXXUBC that last statement is an oversimplification of course.
2007 */
2008 if (!LFS_SEGLOCK_HELD(fs)) {
2009 failreason = "Seglock not held";
2010 goto tryagain;
2011 }
2012 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) {
2013 failreason = "Inode with no_gop_write";
2014 goto tryagain;
2015 }
2016 if ((pgs[0]->offset & lfs_sb_getbmask(fs)) != 0) {
2017 failreason = "Bad page offset";
2018 goto tryagain;
2019 }
2020
2021 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
2022 vp, pgs, npages, flags);
2023
2024 GOP_SIZE(vp, vp->v_size, &eof, 0);
2025 haveeof = 1;
2026
2027 if (vp->v_type == VREG)
2028 fs_bshift = vp->v_mount->mnt_fs_bshift;
2029 else
2030 fs_bshift = DEV_BSHIFT;
2031 error = 0;
2032 pg = pgs[0];
2033 startoffset = pg->offset;
2034 KASSERT(eof >= 0);
2035
2036 if (startoffset >= eof) {
2037 failreason = "Offset beyond EOF";
2038 goto tryagain;
2039 } else
2040 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
2041 skipbytes = 0;
2042
2043 KASSERT(bytes != 0);
2044
2045 /* Swap PG_DELWRI for PG_PAGEOUT */
2046 for (i = 0; i < npages; i++) {
2047 if (pgs[i]->flags & PG_DELWRI) {
2048 KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
2049 pgs[i]->flags &= ~PG_DELWRI;
2050 pgs[i]->flags |= PG_PAGEOUT;
2051 uvm_pageout_start(1);
2052 mutex_enter(vp->v_interlock);
2053 mutex_enter(&uvm_pageqlock);
2054 uvm_pageunwire(pgs[i]);
2055 mutex_exit(&uvm_pageqlock);
2056 mutex_exit(vp->v_interlock);
2057 }
2058 }
2059
2060 /*
2061 * Check to make sure we're starting on a block boundary.
2062 * We'll check later to make sure we always write entire
2063 * blocks (or fragments).
2064 */
2065 if (startoffset & lfs_sb_getbmask(fs))
2066 printf("%" PRId64 " & %" PRIu64 " = %" PRId64 "\n",
2067 startoffset, lfs_sb_getbmask(fs),
2068 startoffset & lfs_sb_getbmask(fs));
2069 KASSERT((startoffset & lfs_sb_getbmask(fs)) == 0);
2070 if (bytes & lfs_sb_getffmask(fs)) {
2071 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
2072 panic("lfs_gop_write: non-integer blocks");
2073 }
2074
2075 /*
2076 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
2077 * If we would, write what we have and try again. If we don't
2078 * have anything to write, we'll have to sleep.
2079 */
2080 ssp = (SEGSUM *)sp->segsum;
2081 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
2082 (lfs_ss_getnfinfo(fs, ssp) < 1 ?
2083 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
2084 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
2085#if 0
2086 " with nfinfo=%d at offset 0x%jx\n",
2087 (int)lfs_ss_getnfinfo(fs, ssp),
2088 (uintmax_t)lfs_sb_getoffset(fs)));
2089#endif
2090 lfs_updatemeta(sp);
2091 lfs_release_finfo(fs);
2092 (void) lfs_writeseg(fs, sp);
2093
2094 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
2095
2096 /*
2097 * Having given up all of the pager_map we were holding,
2098 * we can now wait for aiodoned to reclaim it for us
2099 * without fear of deadlock.
2100 */
2101 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
2102 UVMPAGER_MAPIN_WAITOK);
2103 }
2104
2105 mbp = getiobuf(NULL, true);
2106 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
2107 vp, mbp, vp->v_numoutput, bytes);
2108 mbp->b_bufsize = npages << PAGE_SHIFT;
2109 mbp->b_data = (void *)kva;
2110 mbp->b_resid = mbp->b_bcount = bytes;
2111 mbp->b_cflags = BC_BUSY|BC_AGE;
2112 mbp->b_iodone = uvm_aio_biodone;
2113
2114 bp = NULL;
2115 for (offset = startoffset;
2116 bytes > 0;
2117 offset += iobytes, bytes -= iobytes) {
2118 lbn = offset >> fs_bshift;
2119 error = ulfs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
2120 lfs_issequential_hole);
2121 if (error) {
2122 UVMHIST_LOG(ubchist, "ulfs_bmaparray() -> %d",
2123 error,0,0,0);
2124 skipbytes += bytes;
2125 bytes = 0;
2126 break;
2127 }
2128
2129 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
2130 bytes);
2131 if (blkno == (daddr_t)-1) {
2132 skipbytes += iobytes;
2133 continue;
2134 }
2135
2136 /*
2137 * Discover how much we can really pack into this buffer.
2138 */
2139 /* If no room in the current segment, finish it up */
2140 if (sp->sum_bytes_left < sizeof(int32_t) ||
2141 sp->seg_bytes_left < (1 << lfs_sb_getbshift(fs))) {
2142 int vers;
2143
2144 lfs_updatemeta(sp);
2145 vers = lfs_fi_getversion(fs, sp->fip);
2146 lfs_release_finfo(fs);
2147 (void) lfs_writeseg(fs, sp);
2148
2149 lfs_acquire_finfo(fs, ip->i_number, vers);
2150 }
2151 /* Check both for space in segment and space in segsum */
2152 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
2153 << fs_bshift);
2154 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
2155 << fs_bshift);
2156 KASSERT(iobytes > 0);
2157
2158 /* if it's really one i/o, don't make a second buf */
2159 if (offset == startoffset && iobytes == bytes) {
2160 bp = mbp;
2161 /*
2162 * All the LFS output is done by the segwriter. It
2163 * will increment numoutput by one for all the bufs it
2164 * recieves. However this buffer needs one extra to
2165 * account for aiodone.
2166 */
2167 mutex_enter(vp->v_interlock);
2168 vp->v_numoutput++;
2169 mutex_exit(vp->v_interlock);
2170 } else {
2171 bp = getiobuf(NULL, true);
2172 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
2173 vp, bp, vp->v_numoutput, 0);
2174 nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes);
2175 /*
2176 * LFS doesn't like async I/O here, dies with
2177 * an assert in lfs_bwrite(). Is that assert
2178 * valid? I retained non-async behaviour when
2179 * converted this to use nestiobuf --pooka
2180 */
2181 bp->b_flags &= ~B_ASYNC;
2182 }
2183
2184 /* XXX This is silly ... is this necessary? */
2185 mutex_enter(&bufcache_lock);
2186 mutex_enter(vp->v_interlock);
2187 bgetvp(vp, bp);
2188 mutex_exit(vp->v_interlock);
2189 mutex_exit(&bufcache_lock);
2190
2191 bp->b_lblkno = lfs_lblkno(fs, offset);
2192 bp->b_private = mbp;
2193 if (devvp->v_type == VBLK) {
2194 bp->b_dev = devvp->v_rdev;
2195 }
2196 VOP_BWRITE(bp->b_vp, bp);
2197 while (lfs_gatherblock(sp, bp, NULL))
2198 continue;
2199 }
2200
2201 nestiobuf_done(mbp, skipbytes, error);
2202 if (skipbytes) {
2203 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
2204 }
2205 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
2206
2207 if (!async) {
2208 /* Start a segment write. */
2209 UVMHIST_LOG(ubchist, "flushing", 0,0,0,0);
2210 mutex_enter(&lfs_lock);
2211 lfs_flush(fs, 0, 1);
2212 mutex_exit(&lfs_lock);
2213 }
2214
2215 if ((sp->seg_flags & SEGM_SINGLE) && lfs_sb_getcurseg(fs) != fs->lfs_startseg)
2216 return EAGAIN;
2217
2218 return (0);
2219
2220 tryagain:
2221 /*
2222 * We can't write the pages, for whatever reason.
2223 * Clean up after ourselves, and make the caller try again.
2224 */
2225 mutex_enter(vp->v_interlock);
2226
2227 /* Tell why we're here, if we know */
2228 if (failreason != NULL) {
2229 DLOG((DLOG_PAGE, "lfs_gop_write: %s\n", failreason));
2230 }
2231 if (haveeof && startoffset >= eof) {
2232 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64
2233 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number,
2234 pgs[0]->offset, eof, npages));
2235 }
2236
2237 mutex_enter(&uvm_pageqlock);
2238 for (i = 0; i < npages; i++) {
2239 pg = pgs[i];
2240
2241 if (pg->flags & PG_PAGEOUT)
2242 uvm_pageout_done(1);
2243 if (pg->flags & PG_DELWRI) {
2244 uvm_pageunwire(pg);
2245 }
2246 uvm_pageactivate(pg);
2247 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
2248 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg,
2249 vp, pg->offset));
2250 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags));
2251 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags));
2252 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon));
2253 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject));
2254 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i,
2255 pg->wire_count));
2256 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i,
2257 pg->loan_count));
2258 }
2259 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
2260 uvm_page_unbusy(pgs, npages);
2261 mutex_exit(&uvm_pageqlock);
2262 mutex_exit(vp->v_interlock);
2263 return EAGAIN;
2264}
2265
2266/*
2267 * finish vnode/inode initialization.
2268 * used by lfs_vget.
2269 */
2270void
2271lfs_vinit(struct mount *mp, struct vnode **vpp)
2272{
2273 struct vnode *vp = *vpp;
2274 struct inode *ip = VTOI(vp);
2275 struct ulfsmount *ump = VFSTOULFS(mp);
2276 struct lfs *fs = ump->um_lfs;
2277 int i;
2278
2279 ip->i_mode = lfs_dino_getmode(fs, ip->i_din);
2280 ip->i_nlink = lfs_dino_getnlink(fs, ip->i_din);
2281 ip->i_lfs_osize = ip->i_size = lfs_dino_getsize(fs, ip->i_din);
2282 ip->i_flags = lfs_dino_getflags(fs, ip->i_din);
2283 ip->i_gen = lfs_dino_getgen(fs, ip->i_din);
2284 ip->i_uid = lfs_dino_getuid(fs, ip->i_din);
2285 ip->i_gid = lfs_dino_getgid(fs, ip->i_din);
2286
2287 ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din);
2288 ip->i_lfs_odnlink = lfs_dino_getnlink(fs, ip->i_din);
2289
2290 /*
2291 * Initialize the vnode from the inode, check for aliases. In all
2292 * cases re-init ip, the underlying vnode/inode may have changed.
2293 */
2294 ulfs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2295 ip = VTOI(vp);
2296
2297 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
2298 if (vp->v_type != VLNK || ip->i_size >= ip->i_lfs->um_maxsymlinklen) {
2299#ifdef DEBUG
2300 for (i = (ip->i_size + lfs_sb_getbsize(fs) - 1) >> lfs_sb_getbshift(fs);
2301 i < ULFS_NDADDR; i++) {
2302 if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
2303 i == 0)
2304 continue;
2305 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) {
2306 lfs_dump_dinode(fs, ip->i_din);
2307 panic("inconsistent inode (direct)");
2308 }
2309 }
2310 for ( ; i < ULFS_NDADDR + ULFS_NIADDR; i++) {
2311 if (lfs_dino_getib(fs, ip->i_din, i - ULFS_NDADDR) != 0) {
2312 lfs_dump_dinode(fs, ip->i_din);
2313 panic("inconsistent inode (indirect)");
2314 }
2315 }
2316#endif /* DEBUG */
2317 for (i = 0; i < ULFS_NDADDR; i++)
2318 if (lfs_dino_getdb(fs, ip->i_din, i) != 0)
2319 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i);
2320 }
2321
2322#ifdef DIAGNOSTIC
2323 if (vp->v_type == VNON) {
2324# ifdef DEBUG
2325 lfs_dump_dinode(fs, ip->i_din);
2326# endif
2327 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n",
2328 (unsigned long long)ip->i_number,
2329 (ip->i_mode & LFS_IFMT) >> 12);
2330 }
2331#endif /* DIAGNOSTIC */
2332
2333 /*
2334 * Finish inode initialization now that aliasing has been resolved.
2335 */
2336
2337 ip->i_devvp = fs->lfs_devvp;
2338 vref(ip->i_devvp);
2339#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
2340 ulfsquota_init(ip);
2341#endif
2342 genfs_node_init(vp, &lfs_genfsops);
2343 uvm_vnp_setsize(vp, ip->i_size);
2344
2345 /* Initialize hiblk from file size */
2346 ip->i_lfs_hiblk = lfs_lblkno(ip->i_lfs, ip->i_size + lfs_sb_getbsize(ip->i_lfs) - 1) - 1;
2347
2348 *vpp = vp;
2349}
2350
2351/*
2352 * Resize the filesystem to contain the specified number of segments.
2353 */
2354int
2355lfs_resize_fs(struct lfs *fs, int newnsegs)
2356{
2357 SEGUSE *sup;
2358 CLEANERINFO *cip;
2359 struct buf *bp, *obp;
2360 daddr_t olast, nlast, ilast, noff, start, end;
2361 struct vnode *ivp;
2362 struct inode *ip;
2363 int error, badnews, inc, oldnsegs;
2364 int sbbytes, csbbytes, gain, cgain;
2365 int i;
2366
2367 /* Only support v2 and up */
2368 if (lfs_sb_getversion(fs) < 2)
2369 return EOPNOTSUPP;
2370
2371 /* If we're doing nothing, do it fast */
2372 oldnsegs = lfs_sb_getnseg(fs);
2373 if (newnsegs == oldnsegs)
2374 return 0;
2375
2376 /* We always have to have two superblocks */
2377 if (newnsegs <= lfs_dtosn(fs, lfs_sb_getsboff(fs, 1)))
2378 /* XXX this error code is rather nonsense */
2379 return EFBIG;
2380
2381 ivp = fs->lfs_ivnode;
2382 ip = VTOI(ivp);
2383 error = 0;
2384
2385 /* Take the segment lock so no one else calls lfs_newseg() */
2386 lfs_seglock(fs, SEGM_PROT);
2387
2388 /*
2389 * Make sure the segments we're going to be losing, if any,
2390 * are in fact empty. We hold the seglock, so their status
2391 * cannot change underneath us. Count the superblocks we lose,
2392 * while we're at it.
2393 */
2394 sbbytes = csbbytes = 0;
2395 cgain = 0;
2396 for (i = newnsegs; i < oldnsegs; i++) {
2397 LFS_SEGENTRY(sup, fs, i, bp);
2398 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL);
2399 if (sup->su_flags & SEGUSE_SUPERBLOCK)
2400 sbbytes += LFS_SBPAD;
2401 if (!(sup->su_flags & SEGUSE_DIRTY)) {
2402 ++cgain;
2403 if (sup->su_flags & SEGUSE_SUPERBLOCK)
2404 csbbytes += LFS_SBPAD;
2405 }
2406 brelse(bp, 0);
2407 if (badnews) {
2408 error = EBUSY;
2409 goto out;
2410 }
2411 }
2412
2413 /* Note old and new segment table endpoints, and old ifile size */
2414 olast = lfs_sb_getcleansz(fs) + lfs_sb_getsegtabsz(fs);
2415 nlast = howmany(newnsegs, lfs_sb_getsepb(fs)) + lfs_sb_getcleansz(fs);
2416 ilast = ivp->v_size >> lfs_sb_getbshift(fs);
2417 noff = nlast - olast;
2418
2419 /*
2420 * Make sure no one can use the Ifile while we change it around.
2421 * Even after taking the iflock we need to make sure no one still
2422 * is holding Ifile buffers, so we get each one, to drain them.
2423 * (XXX this could be done better.)
2424 */
2425 rw_enter(&fs->lfs_iflock, RW_WRITER);
2426 for (i = 0; i < ilast; i++) {
2427 /* XXX what to do if bread fails? */
2428 bread(ivp, i, lfs_sb_getbsize(fs), 0, &bp);
2429 brelse(bp, 0);
2430 }
2431
2432 /* Allocate new Ifile blocks */
2433 for (i = ilast; i < ilast + noff; i++) {
2434 if (lfs_balloc(ivp, i * lfs_sb_getbsize(fs), lfs_sb_getbsize(fs), NOCRED, 0,
2435 &bp) != 0)
2436 panic("balloc extending ifile");
2437 memset(bp->b_data, 0, lfs_sb_getbsize(fs));
2438 VOP_BWRITE(bp->b_vp, bp);
2439 }
2440
2441 /* Register new ifile size */
2442 ip->i_size += noff * lfs_sb_getbsize(fs);
2443 lfs_dino_setsize(fs, ip->i_din, ip->i_size);
2444 uvm_vnp_setsize(ivp, ip->i_size);
2445
2446 /* Copy the inode table to its new position */
2447 if (noff != 0) {
2448 if (noff < 0) {
2449 start = nlast;
2450 end = ilast + noff;
2451 inc = 1;
2452 } else {
2453 start = ilast + noff - 1;
2454 end = nlast - 1;
2455 inc = -1;
2456 }
2457 for (i = start; i != end; i += inc) {
2458 if (bread(ivp, i, lfs_sb_getbsize(fs),
2459 B_MODIFY, &bp) != 0)
2460 panic("resize: bread dst blk failed");
2461 if (bread(ivp, i - noff, lfs_sb_getbsize(fs),
2462 0, &obp))
2463 panic("resize: bread src blk failed");
2464 memcpy(bp->b_data, obp->b_data, lfs_sb_getbsize(fs));
2465 VOP_BWRITE(bp->b_vp, bp);
2466 brelse(obp, 0);
2467 }
2468 }
2469
2470 /* If we are expanding, write the new empty SEGUSE entries */
2471 if (newnsegs > oldnsegs) {
2472 for (i = oldnsegs; i < newnsegs; i++) {
2473 if ((error = bread(ivp, i / lfs_sb_getsepb(fs) +
2474 lfs_sb_getcleansz(fs), lfs_sb_getbsize(fs),
2475 B_MODIFY, &bp)) != 0)
2476 panic("lfs: ifile read: %d", error);
2477 while ((i + 1) % lfs_sb_getsepb(fs) && i < newnsegs) {
2478 sup = &((SEGUSE *)bp->b_data)[i % lfs_sb_getsepb(fs)];
2479 memset(sup, 0, sizeof(*sup));
2480 i++;
2481 }
2482 VOP_BWRITE(bp->b_vp, bp);
2483 }
2484 }
2485
2486 /* Zero out unused superblock offsets */
2487 for (i = 2; i < LFS_MAXNUMSB; i++)
2488 if (lfs_dtosn(fs, lfs_sb_getsboff(fs, i)) >= newnsegs)
2489 lfs_sb_setsboff(fs, i, 0x0);
2490
2491 /*
2492 * Correct superblock entries that depend on fs size.
2493 * The computations of these are as follows:
2494 *
2495 * size = lfs_segtod(fs, nseg)
2496 * dsize = lfs_segtod(fs, nseg - minfreeseg) - lfs_btofsb(#super * LFS_SBPAD)
2497 * bfree = dsize - lfs_btofsb(fs, bsize * nseg / 2) - blocks_actually_used
2498 * avail = lfs_segtod(fs, nclean) - lfs_btofsb(#clean_super * LFS_SBPAD)
2499 * + (lfs_segtod(fs, 1) - (offset - curseg))
2500 * - lfs_segtod(fs, minfreeseg - (minfreeseg / 2))
2501 *
2502 * XXX - we should probably adjust minfreeseg as well.
2503 */
2504 gain = (newnsegs - oldnsegs);
2505 lfs_sb_setnseg(fs, newnsegs);
2506 lfs_sb_setsegtabsz(fs, nlast - lfs_sb_getcleansz(fs));
2507 lfs_sb_addsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)));
2508 lfs_sb_adddsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes));
2509 lfs_sb_addbfree(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes)
2510 - gain * lfs_btofsb(fs, lfs_sb_getbsize(fs) / 2));
2511 if (gain > 0) {
2512 lfs_sb_addnclean(fs, gain);
2513 lfs_sb_addavail(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)));
2514 } else {
2515 lfs_sb_subnclean(fs, cgain);
2516 lfs_sb_subavail(fs, cgain * lfs_btofsb(fs, lfs_sb_getssize(fs)) -
2517 lfs_btofsb(fs, csbbytes));
2518 }
2519
2520 /* Resize segment flag cache */
2521 fs->lfs_suflags[0] = realloc(fs->lfs_suflags[0],
2522 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK);
2523 fs->lfs_suflags[1] = realloc(fs->lfs_suflags[1],
2524 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK);
2525 for (i = oldnsegs; i < newnsegs; i++)
2526 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0;
2527
2528 /* Truncate Ifile if necessary */
2529 if (noff < 0)
2530 lfs_truncate(ivp, ivp->v_size + (noff << lfs_sb_getbshift(fs)), 0,
2531 NOCRED);
2532
2533 /* Update cleaner info so the cleaner can die */
2534 /* XXX what to do if bread fails? */
2535 bread(ivp, 0, lfs_sb_getbsize(fs), B_MODIFY, &bp);
2536 cip = bp->b_data;
2537 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs));
2538 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs));
2539 VOP_BWRITE(bp->b_vp, bp);
2540
2541 /* Let Ifile accesses proceed */
2542 rw_exit(&fs->lfs_iflock);
2543
2544 out:
2545 lfs_segunlock(fs);
2546 return error;
2547}
2548
2549/*
2550 * Extended attribute dispatch
2551 */
2552int
2553lfs_extattrctl(struct mount *mp, int cmd, struct vnode *vp,
2554 int attrnamespace, const char *attrname)
2555{
2556#ifdef LFS_EXTATTR
2557 struct ulfsmount *ump;
2558
2559 ump = VFSTOULFS(mp);
2560 if (ump->um_fstype == ULFS1) {
2561 return ulfs_extattrctl(mp, cmd, vp, attrnamespace, attrname);
2562 }
2563#endif
2564 return vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname);
2565}
2566