1 /* $NetBSD: lfs_vfsops.c,v 1.345 2015/09/01 06:16:59 dholland Exp $ */
2
3 /*-
4 * Copyright (c) 1999, 2000, 2001, 2002, 2003, 2007, 2007
5 * The NetBSD Foundation, Inc.
6 * All rights reserved.
7 *
8 * This code is derived from software contributed to The NetBSD Foundation
9 * by Konrad E. Schroder <perseant@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 1.345 2015/09/01 06:16:59 dholland Exp $");
65
66 #if defined(_KERNEL_OPT)
67 #include "opt_lfs.h"
68 #include "opt_quota.h"
69 #endif
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/namei.h>
74 #include <sys/proc.h>
75 #include <sys/kernel.h>
76 #include <sys/vnode.h>
77 #include <sys/mount.h>
78 #include <sys/kthread.h>
79 #include <sys/buf.h>
80 #include <sys/device.h>
81 #include <sys/mbuf.h>
82 #include <sys/file.h>
83 #include <sys/disklabel.h>
84 #include <sys/ioctl.h>
85 #include <sys/errno.h>
86 #include <sys/malloc.h>
87 #include <sys/pool.h>
88 #include <sys/socket.h>
89 #include <sys/syslog.h>
90 #include <uvm/uvm_extern.h>
91 #include <sys/sysctl.h>
92 #include <sys/conf.h>
93 #include <sys/kauth.h>
94 #include <sys/module.h>
95 #include <sys/syscallvar.h>
96 #include <sys/syscall.h>
97 #include <sys/syscallargs.h>
98
99 #include <miscfs/specfs/specdev.h>
100
101 #include <ufs/lfs/ulfs_quotacommon.h>
102 #include <ufs/lfs/ulfs_inode.h>
103 #include <ufs/lfs/ulfsmount.h>
104 #include <ufs/lfs/ulfs_bswap.h>
105 #include <ufs/lfs/ulfs_extern.h>
106
107 #include <uvm/uvm.h>
108 #include <uvm/uvm_stat.h>
109 #include <uvm/uvm_pager.h>
110 #include <uvm/uvm_pdaemon.h>
111
112 #include <ufs/lfs/lfs.h>
113 #include <ufs/lfs/lfs_accessors.h>
114 #include <ufs/lfs/lfs_kernel.h>
115 #include <ufs/lfs/lfs_extern.h>
116
117 #include <miscfs/genfs/genfs.h>
118 #include <miscfs/genfs/genfs_node.h>
119
120 MODULE(MODULE_CLASS_VFS, lfs, NULL);
121
122 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
123 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *);
124
125 static struct sysctllog *lfs_sysctl_log;
126
127 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
128 extern const struct vnodeopv_desc lfs_specop_opv_desc;
129 extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
130
131 pid_t lfs_writer_daemon = 0;
132 lwpid_t lfs_writer_lid = 0;
133 int lfs_do_flush = 0;
134 #ifdef LFS_KERNEL_RFW
135 int lfs_do_rfw = 0;
136 #endif
137
138 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
139 &lfs_vnodeop_opv_desc,
140 &lfs_specop_opv_desc,
141 &lfs_fifoop_opv_desc,
142 NULL,
143 };
144
145 struct vfsops lfs_vfsops = {
146 .vfs_name = MOUNT_LFS,
147 .vfs_min_mount_data = sizeof (struct ulfs_args),
148 .vfs_mount = lfs_mount,
149 .vfs_start = ulfs_start,
150 .vfs_unmount = lfs_unmount,
151 .vfs_root = ulfs_root,
152 .vfs_quotactl = ulfs_quotactl,
153 .vfs_statvfs = lfs_statvfs,
154 .vfs_sync = lfs_sync,
155 .vfs_vget = lfs_vget,
156 .vfs_loadvnode = lfs_loadvnode,
157 .vfs_newvnode = lfs_newvnode,
158 .vfs_fhtovp = lfs_fhtovp,
159 .vfs_vptofh = lfs_vptofh,
160 .vfs_init = lfs_init,
161 .vfs_reinit = lfs_reinit,
162 .vfs_done = lfs_done,
163 .vfs_mountroot = lfs_mountroot,
164 .vfs_snapshot = (void *)eopnotsupp,
165 .vfs_extattrctl = lfs_extattrctl,
166 .vfs_suspendctl = (void *)eopnotsupp,
167 .vfs_renamelock_enter = genfs_renamelock_enter,
168 .vfs_renamelock_exit = genfs_renamelock_exit,
169 .vfs_fsync = (void *)eopnotsupp,
170 .vfs_opv_descs = lfs_vnodeopv_descs
171 };
172
173 const struct genfs_ops lfs_genfsops = {
174 .gop_size = lfs_gop_size,
175 .gop_alloc = ulfs_gop_alloc,
176 .gop_write = lfs_gop_write,
177 .gop_markupdate = ulfs_gop_markupdate,
178 };
179
180 struct shortlong {
181 const char *sname;
182 const char *lname;
183 };
184
185 static int
sysctl_lfs_dostats(SYSCTLFN_ARGS)186 sysctl_lfs_dostats(SYSCTLFN_ARGS)
187 {
188 extern struct lfs_stats lfs_stats;
189 extern int lfs_dostats;
190 int error;
191
192 error = sysctl_lookup(SYSCTLFN_CALL(rnode));
193 if (error || newp == NULL)
194 return (error);
195
196 if (lfs_dostats == 0)
197 memset(&lfs_stats, 0, sizeof(lfs_stats));
198
199 return (0);
200 }
201
202 static void
lfs_sysctl_setup(struct sysctllog ** clog)203 lfs_sysctl_setup(struct sysctllog **clog)
204 {
205 int i;
206 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead,
207 lfs_fs_pagetrip, lfs_ignore_lazy_sync;
208 #ifdef DEBUG
209 extern int lfs_debug_log_subsys[DLOG_MAX];
210 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */
211 { "rollforward", "Debug roll-forward code" },
212 { "alloc", "Debug inode allocation and free list" },
213 { "avail", "Debug space-available-now accounting" },
214 { "flush", "Debug flush triggers" },
215 { "lockedlist", "Debug locked list accounting" },
216 { "vnode_verbose", "Verbose per-vnode-written debugging" },
217 { "vnode", "Debug vnode use during segment write" },
218 { "segment", "Debug segment writing" },
219 { "seguse", "Debug segment used-bytes accounting" },
220 { "cleaner", "Debug cleaning routines" },
221 { "mount", "Debug mount/unmount routines" },
222 { "pagecache", "Debug UBC interactions" },
223 { "dirop", "Debug directory-operation accounting" },
224 { "malloc", "Debug private malloc accounting" },
225 };
226 #endif /* DEBUG */
227 struct shortlong stat_names[] = { /* Must match lfs.h! */
228 { "segsused", "Number of new segments allocated" },
229 { "psegwrites", "Number of partial-segment writes" },
230 { "psyncwrites", "Number of synchronous partial-segment"
231 " writes" },
232 { "pcleanwrites", "Number of partial-segment writes by the"
233 " cleaner" },
234 { "blocktot", "Number of blocks written" },
235 { "cleanblocks", "Number of blocks written by the cleaner" },
236 { "ncheckpoints", "Number of checkpoints made" },
237 { "nwrites", "Number of whole writes" },
238 { "nsync_writes", "Number of synchronous writes" },
239 { "wait_exceeded", "Number of times writer waited for"
240 " cleaner" },
241 { "write_exceeded", "Number of times writer invoked flush" },
242 { "flush_invoked", "Number of times flush was invoked" },
243 { "vflush_invoked", "Number of time vflush was called" },
244 { "clean_inlocked", "Number of vnodes skipped for being dead" },
245 { "clean_vnlocked", "Number of vnodes skipped for vget failure" },
246 { "segs_reclaimed", "Number of segments reclaimed" },
247 };
248
249 sysctl_createv(clog, 0, NULL, NULL,
250 CTLFLAG_PERMANENT,
251 CTLTYPE_NODE, "lfs",
252 SYSCTL_DESCR("Log-structured file system"),
253 NULL, 0, NULL, 0,
254 CTL_VFS, 5, CTL_EOL);
255 /*
256 * XXX the "5" above could be dynamic, thereby eliminating one
257 * more instance of the "number to vfs" mapping problem, but
258 * "5" is the order as taken from sys/mount.h
259 */
260
261 sysctl_createv(clog, 0, NULL, NULL,
262 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
263 CTLTYPE_INT, "flushindir", NULL,
264 NULL, 0, &lfs_writeindir, 0,
265 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
266 sysctl_createv(clog, 0, NULL, NULL,
267 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
268 CTLTYPE_INT, "clean_vnhead", NULL,
269 NULL, 0, &lfs_clean_vnhead, 0,
270 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
271 sysctl_createv(clog, 0, NULL, NULL,
272 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
273 CTLTYPE_INT, "dostats",
274 SYSCTL_DESCR("Maintain statistics on LFS operations"),
275 sysctl_lfs_dostats, 0, &lfs_dostats, 0,
276 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
277 sysctl_createv(clog, 0, NULL, NULL,
278 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
279 CTLTYPE_INT, "pagetrip",
280 SYSCTL_DESCR("How many dirty pages in fs triggers"
281 " a flush"),
282 NULL, 0, &lfs_fs_pagetrip, 0,
283 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL);
284 sysctl_createv(clog, 0, NULL, NULL,
285 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
286 CTLTYPE_INT, "ignore_lazy_sync",
287 SYSCTL_DESCR("Lazy Sync is ignored entirely"),
288 NULL, 0, &lfs_ignore_lazy_sync, 0,
289 CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL);
290 #ifdef LFS_KERNEL_RFW
291 sysctl_createv(clog, 0, NULL, NULL,
292 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
293 CTLTYPE_INT, "rfw",
294 SYSCTL_DESCR("Use in-kernel roll-forward on mount"),
295 NULL, 0, &lfs_do_rfw, 0,
296 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL);
297 #endif
298
299 sysctl_createv(clog, 0, NULL, NULL,
300 CTLFLAG_PERMANENT,
301 CTLTYPE_NODE, "stats",
302 SYSCTL_DESCR("Debugging options"),
303 NULL, 0, NULL, 0,
304 CTL_VFS, 5, LFS_STATS, CTL_EOL);
305 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) {
306 sysctl_createv(clog, 0, NULL, NULL,
307 CTLFLAG_PERMANENT|CTLFLAG_READONLY,
308 CTLTYPE_INT, stat_names[i].sname,
309 SYSCTL_DESCR(stat_names[i].lname),
310 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]),
311 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL);
312 }
313
314 #ifdef DEBUG
315 sysctl_createv(clog, 0, NULL, NULL,
316 CTLFLAG_PERMANENT,
317 CTLTYPE_NODE, "debug",
318 SYSCTL_DESCR("Debugging options"),
319 NULL, 0, NULL, 0,
320 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL);
321 for (i = 0; i < DLOG_MAX; i++) {
322 sysctl_createv(clog, 0, NULL, NULL,
323 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
324 CTLTYPE_INT, dlog_names[i].sname,
325 SYSCTL_DESCR(dlog_names[i].lname),
326 NULL, 0, &(lfs_debug_log_subsys[i]), 0,
327 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL);
328 }
329 #endif
330 }
331
332 /* old cleaner syscall interface. see VOP_FCNTL() */
333 static const struct syscall_package lfs_syscalls[] = {
334 { SYS_lfs_bmapv, 0, (sy_call_t *)sys_lfs_bmapv },
335 { SYS_lfs_markv, 0, (sy_call_t *)sys_lfs_markv },
336 { SYS___lfs_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
341 static int
lfs_modcmd(modcmd_t cmd,void * arg)342 lfs_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 */
376 struct pool lfs_inode_pool;
377 struct pool lfs_dinode_pool;
378 struct pool lfs_inoext_pool;
379 struct 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 */
386 static void
lfs_writerd(void * arg)387 lfs_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 */
537 void
lfs_init(void)538 lfs_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
564 void
lfs_reinit(void)565 lfs_reinit(void)
566 {
567 ulfs_reinit();
568 }
569
570 void
lfs_done(void)571 lfs_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 */
587 int
lfs_mountroot(void)588 lfs_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 */
626 int
lfs_mount(struct mount * mp,const char * path,void * data,size_t * data_len)627 lfs_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 ump = VFSTOULFS(mp);
677 if (devvp != ump->um_devvp) {
678 if (devvp->v_rdev != ump->um_devvp->v_rdev)
679 error = EINVAL;
680 else {
681 vrele(devvp);
682 devvp = ump->um_devvp;
683 vref(devvp);
684 }
685 }
686 }
687 } else {
688 if (!update) {
689 /* New mounts must have a filename for the device */
690 return (EINVAL);
691 } else {
692 /* Use the extant mount */
693 ump = VFSTOULFS(mp);
694 devvp = ump->um_devvp;
695 vref(devvp);
696 }
697 }
698
699
700 /*
701 * If mount by non-root, then verify that user has necessary
702 * permissions on the device.
703 */
704 if (error == 0) {
705 accessmode = VREAD;
706 if (update ?
707 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
708 (mp->mnt_flag & MNT_RDONLY) == 0)
709 accessmode |= VWRITE;
710 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
711 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
712 KAUTH_REQ_SYSTEM_MOUNT_DEVICE, mp, devvp,
713 KAUTH_ARG(accessmode));
714 VOP_UNLOCK(devvp);
715 }
716
717 if (error) {
718 vrele(devvp);
719 return (error);
720 }
721
722 if (!update) {
723 int flags;
724
725 if (mp->mnt_flag & MNT_RDONLY)
726 flags = FREAD;
727 else
728 flags = FREAD|FWRITE;
729 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
730 error = VOP_OPEN(devvp, flags, FSCRED);
731 VOP_UNLOCK(devvp);
732 if (error)
733 goto fail;
734 error = lfs_mountfs(devvp, mp, l); /* LFS */
735 if (error) {
736 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
737 (void)VOP_CLOSE(devvp, flags, NOCRED);
738 VOP_UNLOCK(devvp);
739 goto fail;
740 }
741
742 ump = VFSTOULFS(mp);
743 fs = ump->um_lfs;
744 } else {
745 /*
746 * Update the mount.
747 */
748
749 /*
750 * The initial mount got a reference on this
751 * device, so drop the one obtained via
752 * namei(), above.
753 */
754 vrele(devvp);
755
756 ump = VFSTOULFS(mp);
757 fs = ump->um_lfs;
758
759 if (fs->lfs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
760 /*
761 * Changing from read/write to read-only.
762 * XXX: shouldn't we sync here? or does vfs do that?
763 */
764 #ifdef LFS_QUOTA2
765 /* XXX: quotas should remain on when readonly */
766 if (fs->lfs_use_quota2) {
767 error = lfsquota2_umount(mp, 0);
768 if (error) {
769 return error;
770 }
771 }
772 #endif
773 }
774
775 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
776 /*
777 * Changing from read-only to read/write.
778 * Note in the superblocks that we're writing.
779 */
780
781 /* XXX: quotas should have been on even if readonly */
782 if (fs->lfs_use_quota2) {
783 #ifdef LFS_QUOTA2
784 error = lfs_quota2_mount(mp);
785 #else
786 uprintf("%s: no kernel support for this "
787 "filesystem's quotas\n",
788 mp->mnt_stat.f_mntonname);
789 if (mp->mnt_flag & MNT_FORCE) {
790 uprintf("%s: mounting anyway; "
791 "fsck afterwards\n",
792 mp->mnt_stat.f_mntonname);
793 } else {
794 error = EINVAL;
795 }
796 #endif
797 if (error) {
798 return error;
799 }
800 }
801
802 fs->lfs_ronly = 0;
803 if (lfs_sb_getpflags(fs) & LFS_PF_CLEAN) {
804 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN);
805 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
806 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
807 }
808 }
809 if (args->fspec == NULL)
810 return EINVAL;
811 }
812
813 error = set_statvfs_info(path, UIO_USERSPACE, args->fspec,
814 UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
815 if (error == 0)
816 lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname);
817 return error;
818
819 fail:
820 vrele(devvp);
821 return (error);
822 }
823
824
825 /*
826 * Common code for mount and mountroot
827 * LFS specific
828 */
829 int
lfs_mountfs(struct vnode * devvp,struct mount * mp,struct lwp * l)830 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
831 {
832 struct dlfs *tdfs, *dfs, *adfs;
833 struct lfs *fs;
834 struct ulfsmount *ump;
835 struct vnode *vp;
836 struct buf *bp, *abp;
837 dev_t dev;
838 int error, i, ronly, fsbsize;
839 kauth_cred_t cred;
840 CLEANERINFO *cip;
841 SEGUSE *sup;
842 daddr_t sb_addr;
843
844 cred = l ? l->l_cred : NOCRED;
845
846 /* The superblock is supposed to be 512 bytes. */
847 __CTASSERT(sizeof(struct dlfs) == DEV_BSIZE);
848
849 /*
850 * Flush out any old buffers remaining from a previous use.
851 */
852 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
853 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
854 VOP_UNLOCK(devvp);
855 if (error)
856 return (error);
857
858 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
859
860 /* Don't free random space on error. */
861 bp = NULL;
862 abp = NULL;
863 ump = NULL;
864
865 sb_addr = LFS_LABELPAD / DEV_BSIZE;
866 while (1) {
867 /* Read in the superblock. */
868 error = bread(devvp, sb_addr, LFS_SBPAD, 0, &bp);
869 if (error)
870 goto out;
871 dfs = (struct dlfs *)bp->b_data;
872
873 /* Check the basics. */
874 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE ||
875 dfs->dlfs_version > LFS_VERSION ||
876 dfs->dlfs_bsize < sizeof(struct dlfs)) {
877 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
878 error = EINVAL; /* XXX needs translation */
879 goto out;
880 }
881 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) {
882 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
883 dfs->dlfs_inodefmt));
884 error = EINVAL;
885 goto out;
886 }
887
888 if (dfs->dlfs_version == 1)
889 fsbsize = DEV_BSIZE;
890 else {
891 fsbsize = 1 << dfs->dlfs_ffshift;
892 /*
893 * Could be, if the frag size is large enough, that we
894 * don't have the "real" primary superblock. If that's
895 * the case, get the real one, and try again.
896 */
897 if (sb_addr != (dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT))) {
898 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
899 " 0x%llx is not right, trying 0x%llx\n",
900 (long long)sb_addr,
901 (long long)(dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT))));
902 sb_addr = dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT);
903 brelse(bp, 0);
904 continue;
905 }
906 }
907 break;
908 }
909
910 /*
911 * Check the second superblock to see which is newer; then mount
912 * using the older of the two. This is necessary to ensure that
913 * the filesystem is valid if it was not unmounted cleanly.
914 */
915
916 if (dfs->dlfs_sboffs[1] &&
917 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
918 {
919 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / DEV_BSIZE),
920 LFS_SBPAD, 0, &abp);
921 if (error)
922 goto out;
923 adfs = (struct dlfs *)abp->b_data;
924
925 if (dfs->dlfs_version == 1) {
926 /* 1s resolution comparison */
927 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
928 tdfs = adfs;
929 else
930 tdfs = dfs;
931 } else {
932 /* monotonic infinite-resolution comparison */
933 if (adfs->dlfs_serial < dfs->dlfs_serial)
934 tdfs = adfs;
935 else
936 tdfs = dfs;
937 }
938
939 /* Check the basics. */
940 if (tdfs->dlfs_magic != LFS_MAGIC ||
941 tdfs->dlfs_bsize > MAXBSIZE ||
942 tdfs->dlfs_version > LFS_VERSION ||
943 tdfs->dlfs_bsize < sizeof(struct dlfs)) {
944 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
945 " sanity failed\n"));
946 error = EINVAL; /* XXX needs translation */
947 goto out;
948 }
949 } else {
950 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock"
951 " daddr=0x%x\n", dfs->dlfs_sboffs[1]));
952 error = EINVAL;
953 goto out;
954 }
955
956 /* Allocate the mount structure, copy the superblock into it. */
957 fs = kmem_zalloc(sizeof(struct lfs), KM_SLEEP);
958 memcpy(&fs->lfs_dlfs_u.u_32, tdfs, sizeof(struct dlfs));
959 fs->lfs_is64 = false; /* XXX notyet */
960 fs->lfs_dobyteswap = false; /* XXX notyet */
961 fs->lfs_hasolddirfmt = false; /* set for real below */
962
963 /* Compatibility */
964 if (lfs_sb_getversion(fs) < 2) {
965 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE);
966 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs));
967 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0));
968 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs));
969 lfs_sb_setfsbtodb(fs, 0);
970 }
971 if (lfs_sb_getresvseg(fs) == 0)
972 lfs_sb_setresvseg(fs, MIN(lfs_sb_getminfreeseg(fs) - 1, \
973 MAX(MIN_RESV_SEGS, lfs_sb_getminfreeseg(fs) / 2 + 1)));
974
975 /*
976 * If we aren't going to be able to write meaningfully to this
977 * filesystem, and were not mounted readonly, bomb out now.
978 */
979 if (lfs_fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
980 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
981 " we need BUFPAGES >= %lld\n",
982 (long long)((bufmem_hiwater / bufmem_lowater) *
983 LFS_INVERSE_MAX_BYTES(
984 lfs_fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
985 kmem_free(fs, sizeof(struct lfs));
986 error = EFBIG; /* XXX needs translation */
987 goto out;
988 }
989
990 /* Before rolling forward, lock so vget will sleep for other procs */
991 if (l != NULL) {
992 fs->lfs_flags = LFS_NOTYET;
993 fs->lfs_rfpid = l->l_proc->p_pid;
994 }
995
996 ump = kmem_zalloc(sizeof(*ump), KM_SLEEP);
997 ump->um_lfs = fs;
998 ump->um_fstype = ULFS1;
999 /* ump->um_cleaner_thread = NULL; */
1000 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */
1001 brelse(bp, BC_INVAL);
1002 brelse(abp, BC_INVAL);
1003 } else {
1004 brelse(bp, 0);
1005 brelse(abp, 0);
1006 }
1007 bp = NULL;
1008 abp = NULL;
1009
1010
1011 /* Set up the I/O information */
1012 fs->lfs_devbsize = DEV_BSIZE;
1013 fs->lfs_iocount = 0;
1014 fs->lfs_diropwait = 0;
1015 fs->lfs_activesb = 0;
1016 lfs_sb_setuinodes(fs, 0);
1017 fs->lfs_ravail = 0;
1018 fs->lfs_favail = 0;
1019 fs->lfs_sbactive = 0;
1020
1021 /* Set up the ifile and lock aflags */
1022 fs->lfs_doifile = 0;
1023 fs->lfs_writer = 0;
1024 fs->lfs_dirops = 0;
1025 fs->lfs_nadirop = 0;
1026 fs->lfs_seglock = 0;
1027 fs->lfs_pdflush = 0;
1028 fs->lfs_sleepers = 0;
1029 fs->lfs_pages = 0;
1030 rw_init(&fs->lfs_fraglock);
1031 rw_init(&fs->lfs_iflock);
1032 cv_init(&fs->lfs_stopcv, "lfsstop");
1033
1034 /* Set the file system readonly/modify bits. */
1035 fs->lfs_ronly = ronly;
1036 if (ronly == 0)
1037 fs->lfs_fmod = 1;
1038
1039 /* ulfs-level information */
1040 fs->um_flags = 0;
1041 fs->um_bptrtodb = lfs_sb_getffshift(fs) - DEV_BSHIFT;
1042 fs->um_seqinc = lfs_sb_getfrag(fs);
1043 fs->um_nindir = lfs_sb_getnindir(fs);
1044 fs->um_lognindir = ffs(lfs_sb_getnindir(fs)) - 1;
1045 fs->um_maxsymlinklen = lfs_sb_getmaxsymlinklen(fs);
1046 fs->um_dirblksiz = LFS_DIRBLKSIZ;
1047 fs->um_maxfilesize = lfs_sb_getmaxfilesize(fs);
1048
1049 /* quota stuff */
1050 /* XXX: these need to come from the on-disk superblock to be used */
1051 fs->lfs_use_quota2 = 0;
1052 fs->lfs_quota_magic = 0;
1053 fs->lfs_quota_flags = 0;
1054 fs->lfs_quotaino[0] = 0;
1055 fs->lfs_quotaino[1] = 0;
1056
1057 /* Initialize the mount structure. */
1058 dev = devvp->v_rdev;
1059 mp->mnt_data = ump;
1060 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
1061 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
1062 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
1063 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
1064 mp->mnt_stat.f_iosize = lfs_sb_getbsize(fs);
1065 mp->mnt_flag |= MNT_LOCAL;
1066 mp->mnt_fs_bshift = lfs_sb_getbshift(fs);
1067 if (fs->um_maxsymlinklen > 0)
1068 mp->mnt_iflag |= IMNT_DTYPE;
1069 else
1070 fs->lfs_hasolddirfmt = true;
1071
1072 ump->um_mountp = mp;
1073 ump->um_dev = dev;
1074 ump->um_devvp = devvp;
1075 for (i = 0; i < ULFS_MAXQUOTAS; i++)
1076 ump->um_quotas[i] = NULLVP;
1077 spec_node_setmountedfs(devvp, mp);
1078
1079 /* Set up reserved memory for pageout */
1080 lfs_setup_resblks(fs);
1081 /* Set up vdirop tailq */
1082 TAILQ_INIT(&fs->lfs_dchainhd);
1083 /* and paging tailq */
1084 TAILQ_INIT(&fs->lfs_pchainhd);
1085 /* and delayed segment accounting for truncation list */
1086 LIST_INIT(&fs->lfs_segdhd);
1087
1088 /*
1089 * We use the ifile vnode for almost every operation. Instead of
1090 * retrieving it from the hash table each time we retrieve it here,
1091 * artificially increment the reference count and keep a pointer
1092 * to it in the incore copy of the superblock.
1093 */
1094 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1095 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
1096 goto out;
1097 }
1098 fs->lfs_ivnode = vp;
1099 vref(vp);
1100
1101 /* Set up inode bitmap and order free list */
1102 lfs_order_freelist(fs);
1103
1104 /* Set up segment usage flags for the autocleaner. */
1105 fs->lfs_nactive = 0;
1106 fs->lfs_suflags = malloc(2 * sizeof(u_int32_t *),
1107 M_SEGMENT, M_WAITOK);
1108 fs->lfs_suflags[0] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t),
1109 M_SEGMENT, M_WAITOK);
1110 fs->lfs_suflags[1] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t),
1111 M_SEGMENT, M_WAITOK);
1112 memset(fs->lfs_suflags[1], 0, lfs_sb_getnseg(fs) * sizeof(u_int32_t));
1113 for (i = 0; i < lfs_sb_getnseg(fs); i++) {
1114 int changed;
1115
1116 LFS_SEGENTRY(sup, fs, i, bp);
1117 changed = 0;
1118 if (!ronly) {
1119 if (sup->su_nbytes == 0 &&
1120 !(sup->su_flags & SEGUSE_EMPTY)) {
1121 sup->su_flags |= SEGUSE_EMPTY;
1122 ++changed;
1123 } else if (!(sup->su_nbytes == 0) &&
1124 (sup->su_flags & SEGUSE_EMPTY)) {
1125 sup->su_flags &= ~SEGUSE_EMPTY;
1126 ++changed;
1127 }
1128 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
1129 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
1130 ++changed;
1131 }
1132 }
1133 fs->lfs_suflags[0][i] = sup->su_flags;
1134 if (changed)
1135 LFS_WRITESEGENTRY(sup, fs, i, bp);
1136 else
1137 brelse(bp, 0);
1138 }
1139
1140 /*
1141 * XXX: if the fs has quotas, quotas should be on even if
1142 * readonly. Otherwise you can't query the quota info!
1143 * However, that's not how the quota2 code got written and I
1144 * don't know if it'll behave itself if enabled while
1145 * readonly, so for now use the same enable logic as ffs.
1146 *
1147 * XXX: also, if you use the -f behavior allowed here (and
1148 * equivalently above for remount) it will corrupt the fs. It
1149 * ought not to allow that. It should allow mounting readonly
1150 * if there are quotas and the kernel doesn't have the quota
1151 * code, but only readonly.
1152 *
1153 * XXX: and if you use the -f behavior allowed here it will
1154 * likely crash at unmount time (or remount time) because we
1155 * think quotas are active.
1156 *
1157 * Although none of this applies until there's a way to set
1158 * lfs_use_quota2 and have quotas in the fs at all.
1159 */
1160 if (!ronly && fs->lfs_use_quota2) {
1161 #ifdef LFS_QUOTA2
1162 error = lfs_quota2_mount(mp);
1163 #else
1164 uprintf("%s: no kernel support for this filesystem's quotas\n",
1165 mp->mnt_stat.f_mntonname);
1166 if (mp->mnt_flag & MNT_FORCE) {
1167 uprintf("%s: mounting anyway; fsck afterwards\n",
1168 mp->mnt_stat.f_mntonname);
1169 } else {
1170 error = EINVAL;
1171 }
1172 #endif
1173 if (error) {
1174 /* XXX XXX must clean up the stuff immediately above */
1175 printf("lfs_mountfs: sorry, leaking some memory\n");
1176 goto out;
1177 }
1178 }
1179
1180 #ifdef LFS_EXTATTR
1181 /*
1182 * Initialize file-backed extended attributes for ULFS1 file
1183 * systems.
1184 *
1185 * XXX: why is this limited to ULFS1?
1186 */
1187 if (ump->um_fstype == ULFS1) {
1188 ulfs_extattr_uepm_init(&ump->um_extattr);
1189 }
1190 #endif
1191
1192 #ifdef LFS_KERNEL_RFW
1193 lfs_roll_forward(fs, mp, l);
1194 #endif
1195
1196 /* If writing, sb is not clean; record in case of immediate crash */
1197 if (!fs->lfs_ronly) {
1198 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN);
1199 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
1200 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
1201 }
1202
1203 /* Allow vget now that roll-forward is complete */
1204 fs->lfs_flags &= ~(LFS_NOTYET);
1205 wakeup(&fs->lfs_flags);
1206
1207 /*
1208 * Initialize the ifile cleaner info with information from
1209 * the superblock.
1210 */
1211 LFS_CLEANERINFO(cip, fs, bp);
1212 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs));
1213 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs));
1214 lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs));
1215 lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs));
1216 (void) LFS_BWRITE_LOG(bp); /* Ifile */
1217
1218 /*
1219 * Mark the current segment as ACTIVE, since we're going to
1220 * be writing to it.
1221 */
1222 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp);
1223 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1224 fs->lfs_nactive++;
1225 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp); /* Ifile */
1226
1227 /* Now that roll-forward is done, unlock the Ifile */
1228 vput(vp);
1229
1230 /* Start the pagedaemon-anticipating daemon */
1231 mutex_enter(&lfs_lock);
1232 if (lfs_writer_daemon == 0 && lfs_writer_lid == 0 &&
1233 kthread_create(PRI_BIO, 0, NULL,
1234 lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1235 panic("fork lfs_writer");
1236 mutex_exit(&lfs_lock);
1237
1238 printf("WARNING: the log-structured file system is experimental\n"
1239 "WARNING: it may cause system crashes and/or corrupt data\n");
1240
1241 return (0);
1242
1243 out:
1244 if (bp)
1245 brelse(bp, 0);
1246 if (abp)
1247 brelse(abp, 0);
1248 if (ump) {
1249 kmem_free(ump->um_lfs, sizeof(struct lfs));
1250 kmem_free(ump, sizeof(*ump));
1251 mp->mnt_data = NULL;
1252 }
1253
1254 return (error);
1255 }
1256
1257 /*
1258 * unmount system call
1259 */
1260 int
lfs_unmount(struct mount * mp,int mntflags)1261 lfs_unmount(struct mount *mp, int mntflags)
1262 {
1263 struct lwp *l = curlwp;
1264 struct ulfsmount *ump;
1265 struct lfs *fs;
1266 int error, flags, ronly;
1267 vnode_t *vp;
1268
1269 flags = 0;
1270 if (mntflags & MNT_FORCE)
1271 flags |= FORCECLOSE;
1272
1273 ump = VFSTOULFS(mp);
1274 fs = ump->um_lfs;
1275
1276 /* Two checkpoints */
1277 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1278 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1279
1280 /* wake up the cleaner so it can die */
1281 /* XXX: shouldn't this be *after* the error cases below? */
1282 lfs_wakeup_cleaner(fs);
1283 mutex_enter(&lfs_lock);
1284 while (fs->lfs_sleepers)
1285 mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1286 &lfs_lock);
1287 mutex_exit(&lfs_lock);
1288
1289 #ifdef LFS_EXTATTR
1290 if (ump->um_fstype == ULFS1) {
1291 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) {
1292 ulfs_extattr_stop(mp, curlwp);
1293 }
1294 if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED) {
1295 ulfs_extattr_uepm_destroy(&ump->um_extattr);
1296 }
1297 }
1298 #endif
1299 #ifdef LFS_QUOTA
1300 if ((error = lfsquota1_umount(mp, flags)) != 0)
1301 return (error);
1302 #endif
1303 #ifdef LFS_QUOTA2
1304 if ((error = lfsquota2_umount(mp, flags)) != 0)
1305 return (error);
1306 #endif
1307 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1308 return (error);
1309 if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0)
1310 return (error);
1311 vp = fs->lfs_ivnode;
1312 mutex_enter(vp->v_interlock);
1313 if (LIST_FIRST(&vp->v_dirtyblkhd))
1314 panic("lfs_unmount: still dirty blocks on ifile vnode");
1315 mutex_exit(vp->v_interlock);
1316
1317 /* Explicitly write the superblock, to update serial and pflags */
1318 lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) | LFS_PF_CLEAN);
1319 lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
1320 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
1321 mutex_enter(&lfs_lock);
1322 while (fs->lfs_iocount)
1323 mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
1324 &lfs_lock);
1325 mutex_exit(&lfs_lock);
1326
1327 /* Finish with the Ifile, now that we're done with it */
1328 vgone(fs->lfs_ivnode);
1329
1330 ronly = !fs->lfs_ronly;
1331 if (ump->um_devvp->v_type != VBAD)
1332 spec_node_setmountedfs(ump->um_devvp, NULL);
1333 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1334 error = VOP_CLOSE(ump->um_devvp,
1335 ronly ? FREAD : FREAD|FWRITE, NOCRED);
1336 vput(ump->um_devvp);
1337
1338 /* Complain about page leakage */
1339 if (fs->lfs_pages > 0)
1340 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
1341 fs->lfs_pages, lfs_subsys_pages);
1342
1343 /* Free per-mount data structures */
1344 free(fs->lfs_ino_bitmap, M_SEGMENT);
1345 free(fs->lfs_suflags[0], M_SEGMENT);
1346 free(fs->lfs_suflags[1], M_SEGMENT);
1347 free(fs->lfs_suflags, M_SEGMENT);
1348 lfs_free_resblks(fs);
1349 cv_destroy(&fs->lfs_stopcv);
1350 rw_destroy(&fs->lfs_fraglock);
1351 rw_destroy(&fs->lfs_iflock);
1352
1353 kmem_free(fs, sizeof(struct lfs));
1354 kmem_free(ump, sizeof(*ump));
1355
1356 mp->mnt_data = NULL;
1357 mp->mnt_flag &= ~MNT_LOCAL;
1358 return (error);
1359 }
1360
1361 /*
1362 * Get file system statistics.
1363 *
1364 * NB: We don't lock to access the superblock here, because it's not
1365 * really that important if we get it wrong.
1366 */
1367 int
lfs_statvfs(struct mount * mp,struct statvfs * sbp)1368 lfs_statvfs(struct mount *mp, struct statvfs *sbp)
1369 {
1370 struct lfs *fs;
1371 struct ulfsmount *ump;
1372
1373 ump = VFSTOULFS(mp);
1374 fs = ump->um_lfs;
1375
1376 sbp->f_bsize = lfs_sb_getbsize(fs);
1377 sbp->f_frsize = lfs_sb_getfsize(fs);
1378 sbp->f_iosize = lfs_sb_getbsize(fs);
1379 sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks;
1380
1381 sbp->f_bfree = LFS_EST_BFREE(fs);
1382 /*
1383 * XXX this should be lfs_sb_getsize (measured in frags)
1384 * rather than dsize (measured in diskblocks). However,
1385 * getsize needs a format version check (for version 1 it
1386 * needs to be blockstofrags'd) so for the moment I'm going to
1387 * leave this... it won't fire wrongly as frags are at least
1388 * as big as diskblocks.
1389 */
1390 KASSERT(sbp->f_bfree <= lfs_sb_getdsize(fs));
1391 #if 0
1392 if (sbp->f_bfree < 0)
1393 sbp->f_bfree = 0;
1394 #endif
1395
1396 sbp->f_bresvd = LFS_EST_RSVD(fs);
1397 if (sbp->f_bfree > sbp->f_bresvd)
1398 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1399 else
1400 sbp->f_bavail = 0;
1401
1402 /* XXX: huh? - dholland 20150728 */
1403 sbp->f_files = lfs_sb_getbfree(fs) / lfs_btofsb(fs, lfs_sb_getibsize(fs))
1404 * LFS_INOPB(fs);
1405 sbp->f_ffree = sbp->f_files - lfs_sb_getnfiles(fs);
1406 sbp->f_favail = sbp->f_ffree;
1407 sbp->f_fresvd = 0;
1408 copy_statvfs_info(sbp, mp);
1409 return (0);
1410 }
1411
1412 /*
1413 * Go through the disk queues to initiate sandbagged IO;
1414 * go through the inodes to write those that have been modified;
1415 * initiate the writing of the super block if it has been modified.
1416 *
1417 * Note: we are always called with the filesystem marked `MPBUSY'.
1418 */
1419 int
lfs_sync(struct mount * mp,int waitfor,kauth_cred_t cred)1420 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
1421 {
1422 int error;
1423 struct lfs *fs;
1424
1425 fs = VFSTOULFS(mp)->um_lfs;
1426 if (fs->lfs_ronly)
1427 return 0;
1428
1429 /* Snapshots should not hose the syncer */
1430 /*
1431 * XXX Sync can block here anyway, since we don't have a very
1432 * XXX good idea of how much data is pending. If it's more
1433 * XXX than a segment and lfs_nextseg is close to the end of
1434 * XXX the log, we'll likely block.
1435 */
1436 mutex_enter(&lfs_lock);
1437 if (fs->lfs_nowrap && lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs)) {
1438 mutex_exit(&lfs_lock);
1439 return 0;
1440 }
1441 mutex_exit(&lfs_lock);
1442
1443 lfs_writer_enter(fs, "lfs_dirops");
1444
1445 /* All syncs must be checkpoints until roll-forward is implemented. */
1446 DLOG((DLOG_FLUSH, "lfs_sync at 0x%jx\n",
1447 (uintmax_t)lfs_sb_getoffset(fs)));
1448 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1449 lfs_writer_leave(fs);
1450 #ifdef LFS_QUOTA
1451 lfs_qsync(mp);
1452 #endif
1453 return (error);
1454 }
1455
1456 /*
1457 * Look up an LFS dinode number to find its incore vnode. If not already
1458 * in core, read it in from the specified device. Return the inode locked.
1459 * Detection and handling of mount points must be done by the calling routine.
1460 */
1461 int
lfs_vget(struct mount * mp,ino_t ino,struct vnode ** vpp)1462 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1463 {
1464 int error;
1465
1466 error = vcache_get(mp, &ino, sizeof(ino), vpp);
1467 if (error)
1468 return error;
1469 error = vn_lock(*vpp, LK_EXCLUSIVE);
1470 if (error) {
1471 vrele(*vpp);
1472 *vpp = NULL;
1473 return error;
1474 }
1475
1476 return 0;
1477 }
1478
1479 /*
1480 * Create a new vnode/inode pair and initialize what fields we can.
1481 */
1482 static void
lfs_init_vnode(struct ulfsmount * ump,ino_t ino,struct vnode * vp)1483 lfs_init_vnode(struct ulfsmount *ump, ino_t ino, struct vnode *vp)
1484 {
1485 struct lfs *fs = ump->um_lfs;
1486 struct inode *ip;
1487 union lfs_dinode *dp;
1488
1489 ASSERT_NO_SEGLOCK(ump->um_lfs);
1490
1491 /* Initialize the inode. */
1492 ip = pool_get(&lfs_inode_pool, PR_WAITOK);
1493 memset(ip, 0, sizeof(*ip));
1494 dp = pool_get(&lfs_dinode_pool, PR_WAITOK);
1495 memset(dp, 0, sizeof(*dp));
1496 ip->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK);
1497 memset(ip->inode_ext.lfs, 0, sizeof(*ip->inode_ext.lfs));
1498 ip->i_din = dp;
1499 ip->i_ump = ump;
1500 ip->i_vnode = vp;
1501 ip->i_dev = ump->um_dev;
1502 lfs_dino_setinumber(fs, dp, ino);
1503 ip->i_number = ino;
1504 ip->i_lfs = ump->um_lfs;
1505 ip->i_lfs_effnblks = 0;
1506 SPLAY_INIT(&ip->i_lfs_lbtree);
1507 ip->i_lfs_nbtree = 0;
1508 LIST_INIT(&ip->i_lfs_segdhd);
1509
1510 vp->v_tag = VT_LFS;
1511 vp->v_op = lfs_vnodeop_p;
1512 vp->v_data = ip;
1513 }
1514
1515 /*
1516 * Undo lfs_init_vnode().
1517 */
1518 static void
lfs_deinit_vnode(struct ulfsmount * ump,struct vnode * vp)1519 lfs_deinit_vnode(struct ulfsmount *ump, struct vnode *vp)
1520 {
1521 struct inode *ip = VTOI(vp);
1522
1523 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs);
1524 pool_put(&lfs_dinode_pool, ip->i_din);
1525 pool_put(&lfs_inode_pool, ip);
1526 vp->v_data = NULL;
1527 }
1528
1529 /*
1530 * Read an inode from disk and initialize this vnode / inode pair.
1531 * Caller assures no other thread will try to load this inode.
1532 */
1533 int
lfs_loadvnode(struct mount * mp,struct vnode * vp,const void * key,size_t key_len,const void ** new_key)1534 lfs_loadvnode(struct mount *mp, struct vnode *vp,
1535 const void *key, size_t key_len, const void **new_key)
1536 {
1537 struct lfs *fs;
1538 union lfs_dinode *dip;
1539 struct inode *ip;
1540 struct buf *bp;
1541 IFILE *ifp;
1542 struct ulfsmount *ump;
1543 ino_t ino;
1544 daddr_t daddr;
1545 int error, retries;
1546 struct timespec ts;
1547
1548 KASSERT(key_len == sizeof(ino));
1549 memcpy(&ino, key, key_len);
1550
1551 memset(&ts, 0, sizeof ts); /* XXX gcc */
1552
1553 ump = VFSTOULFS(mp);
1554 fs = ump->um_lfs;
1555
1556 /*
1557 * If the filesystem is not completely mounted yet, suspend
1558 * any access requests (wait for roll-forward to complete).
1559 */
1560 mutex_enter(&lfs_lock);
1561 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1562 mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1563 &lfs_lock);
1564 mutex_exit(&lfs_lock);
1565
1566 /* Translate the inode number to a disk address. */
1567 if (ino == LFS_IFILE_INUM)
1568 daddr = lfs_sb_getidaddr(fs);
1569 else {
1570 /* XXX bounds-check this too */
1571 LFS_IENTRY(ifp, fs, ino, bp);
1572 daddr = lfs_if_getdaddr(fs, ifp);
1573 if (lfs_sb_getversion(fs) > 1) {
1574 ts.tv_sec = lfs_if_getatime_sec(fs, ifp);
1575 ts.tv_nsec = lfs_if_getatime_nsec(fs, ifp);
1576 }
1577
1578 brelse(bp, 0);
1579 if (daddr == LFS_UNUSED_DADDR)
1580 return (ENOENT);
1581 }
1582
1583 /* Allocate/init new vnode/inode. */
1584 lfs_init_vnode(ump, ino, vp);
1585 ip = VTOI(vp);
1586
1587 /* If the cleaner supplied the inode, use it. */
1588 if (curlwp == ump->um_cleaner_thread && ump->um_cleaner_hint != NULL &&
1589 ump->um_cleaner_hint->bi_lbn == LFS_UNUSED_LBN) {
1590 dip = ump->um_cleaner_hint->bi_bp;
1591 if (fs->lfs_is64) {
1592 error = copyin(dip, &ip->i_din->u_64,
1593 sizeof(struct lfs64_dinode));
1594 } else {
1595 error = copyin(dip, &ip->i_din->u_32,
1596 sizeof(struct lfs32_dinode));
1597 }
1598 if (error) {
1599 lfs_deinit_vnode(ump, vp);
1600 return error;
1601 }
1602 KASSERT(ip->i_number == ino);
1603 goto out;
1604 }
1605
1606 /* Read in the disk contents for the inode, copy into the inode. */
1607 retries = 0;
1608 again:
1609 error = bread(ump->um_devvp, LFS_FSBTODB(fs, daddr),
1610 (lfs_sb_getversion(fs) == 1 ? lfs_sb_getbsize(fs) : lfs_sb_getibsize(fs)),
1611 0, &bp);
1612 if (error) {
1613 lfs_deinit_vnode(ump, vp);
1614 return error;
1615 }
1616
1617 dip = lfs_ifind(fs, ino, bp);
1618 if (dip == NULL) {
1619 /* Assume write has not completed yet; try again */
1620 brelse(bp, BC_INVAL);
1621 ++retries;
1622 if (retries <= LFS_IFIND_RETRIES) {
1623 mutex_enter(&lfs_lock);
1624 if (fs->lfs_iocount) {
1625 DLOG((DLOG_VNODE,
1626 "%s: dinode %d not found, retrying...\n",
1627 __func__, ino));
1628 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
1629 "lfs ifind", 1, &lfs_lock);
1630 } else
1631 retries = LFS_IFIND_RETRIES;
1632 mutex_exit(&lfs_lock);
1633 goto again;
1634 }
1635 #ifdef DEBUG
1636 /* If the seglock is held look at the bpp to see
1637 what is there anyway */
1638 mutex_enter(&lfs_lock);
1639 if (fs->lfs_seglock > 0) {
1640 struct buf **bpp;
1641 union lfs_dinode *dp;
1642 int i;
1643
1644 for (bpp = fs->lfs_sp->bpp;
1645 bpp != fs->lfs_sp->cbpp; ++bpp) {
1646 if ((*bpp)->b_vp == fs->lfs_ivnode &&
1647 bpp != fs->lfs_sp->bpp) {
1648 /* Inode block */
1649 printf("%s: block 0x%" PRIx64 ": ",
1650 __func__, (*bpp)->b_blkno);
1651 for (i = 0; i < LFS_INOPB(fs); i++) {
1652 dp = DINO_IN_BLOCK(fs,
1653 (*bpp)->b_data, i);
1654 if (lfs_dino_getinumber(fs, dp))
1655 printf("%ju ",
1656 (uintmax_t)lfs_dino_getinumber(fs, dp));
1657 }
1658 printf("\n");
1659 }
1660 }
1661 }
1662 mutex_exit(&lfs_lock);
1663 #endif /* DEBUG */
1664 panic("lfs_loadvnode: dinode not found");
1665 }
1666 lfs_copy_dinode(fs, ip->i_din, dip);
1667 brelse(bp, 0);
1668
1669 out:
1670 if (lfs_sb_getversion(fs) > 1) {
1671 lfs_dino_setatime(fs, ip->i_din, ts.tv_sec);
1672 lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec);
1673 }
1674
1675 lfs_vinit(mp, &vp);
1676
1677 *new_key = &ip->i_number;
1678 return 0;
1679 }
1680
1681 /*
1682 * Create a new inode and initialize this vnode / inode pair.
1683 */
1684 int
lfs_newvnode(struct mount * mp,struct vnode * dvp,struct vnode * vp,struct vattr * vap,kauth_cred_t cred,size_t * key_len,const void ** new_key)1685 lfs_newvnode(struct mount *mp, struct vnode *dvp, struct vnode *vp,
1686 struct vattr *vap, kauth_cred_t cred,
1687 size_t *key_len, const void **new_key)
1688 {
1689 ino_t ino;
1690 struct inode *ip;
1691 struct ulfsmount *ump;
1692 struct lfs *fs;
1693 int error, mode, gen;
1694
1695 KASSERT(dvp != NULL || vap->va_fileid > 0);
1696 KASSERT(dvp != NULL && dvp->v_mount == mp);
1697 KASSERT(vap->va_type != VNON);
1698
1699 *key_len = sizeof(ino);
1700 ump = VFSTOULFS(mp);
1701 fs = ump->um_lfs;
1702 mode = MAKEIMODE(vap->va_type, vap->va_mode);
1703
1704 /*
1705 * Allocate fresh inode. With "dvp == NULL" take the inode number
1706 * and version from "vap".
1707 */
1708 if (dvp == NULL) {
1709 ino = vap->va_fileid;
1710 gen = vap->va_gen;
1711 error = lfs_valloc_fixed(fs, ino, gen);
1712 } else {
1713 error = lfs_valloc(dvp, mode, cred, &ino, &gen);
1714 }
1715 if (error)
1716 return error;
1717
1718 /* Attach inode to vnode. */
1719 lfs_init_vnode(ump, ino, vp);
1720 ip = VTOI(vp);
1721
1722 mutex_enter(&lfs_lock);
1723 LFS_SET_UINO(ip, IN_CHANGE);
1724 mutex_exit(&lfs_lock);
1725
1726 /* Note no blocks yet */
1727 ip->i_lfs_hiblk = -1;
1728
1729 /* Set a new generation number for this inode. */
1730 ip->i_gen = gen;
1731 lfs_dino_setgen(fs, ip->i_din, gen);
1732
1733 memset(ip->i_lfs_fragsize, 0,
1734 ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
1735
1736 /* Set uid / gid. */
1737 if (cred == NOCRED || cred == FSCRED) {
1738 ip->i_gid = 0;
1739 ip->i_uid = 0;
1740 } else {
1741 ip->i_gid = VTOI(dvp)->i_gid;
1742 ip->i_uid = kauth_cred_geteuid(cred);
1743 }
1744 DIP_ASSIGN(ip, gid, ip->i_gid);
1745 DIP_ASSIGN(ip, uid, ip->i_uid);
1746
1747 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
1748 error = lfs_chkiq(ip, 1, cred, 0);
1749 if (error) {
1750 lfs_vfree(dvp, ino, mode);
1751 lfs_deinit_vnode(ump, vp);
1752
1753 return error;
1754 }
1755 #endif
1756
1757 /* Set type and finalize. */
1758 ip->i_flags = 0;
1759 DIP_ASSIGN(ip, flags, 0);
1760 ip->i_mode = mode;
1761 DIP_ASSIGN(ip, mode, mode);
1762 if (vap->va_rdev != VNOVAL) {
1763 /*
1764 * Want to be able to use this to make badblock
1765 * inodes, so don't truncate the dev number.
1766 */
1767 // XXX clean this up
1768 if (ump->um_fstype == ULFS1)
1769 ip->i_din->u_32.di_rdev = ulfs_rw32(vap->va_rdev,
1770 ULFS_MPNEEDSWAP(fs));
1771 else
1772 ip->i_din->u_64.di_rdev = ulfs_rw64(vap->va_rdev,
1773 ULFS_MPNEEDSWAP(fs));
1774 }
1775 lfs_vinit(mp, &vp);
1776
1777 *new_key = &ip->i_number;
1778 return 0;
1779 }
1780
1781 /*
1782 * File handle to vnode
1783 */
1784 int
lfs_fhtovp(struct mount * mp,struct fid * fhp,struct vnode ** vpp)1785 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1786 {
1787 struct lfid lfh;
1788 struct lfs *fs;
1789
1790 if (fhp->fid_len != sizeof(struct lfid))
1791 return EINVAL;
1792
1793 memcpy(&lfh, fhp, sizeof(lfh));
1794 if (lfh.lfid_ino < LFS_IFILE_INUM)
1795 return ESTALE;
1796
1797 fs = VFSTOULFS(mp)->um_lfs;
1798 if (lfh.lfid_ident != lfs_sb_getident(fs))
1799 return ESTALE;
1800
1801 if (lfh.lfid_ino >
1802 ((lfs_dino_getsize(fs, VTOI(fs->lfs_ivnode)->i_din) >> lfs_sb_getbshift(fs)) -
1803 lfs_sb_getcleansz(fs) - lfs_sb_getsegtabsz(fs)) * lfs_sb_getifpb(fs))
1804 return ESTALE;
1805
1806 return (ulfs_fhtovp(mp, &lfh.lfid_ufid, vpp));
1807 }
1808
1809 /*
1810 * Vnode pointer to File handle
1811 */
1812 /* ARGSUSED */
1813 int
lfs_vptofh(struct vnode * vp,struct fid * fhp,size_t * fh_size)1814 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
1815 {
1816 struct inode *ip;
1817 struct lfid lfh;
1818
1819 if (*fh_size < sizeof(struct lfid)) {
1820 *fh_size = sizeof(struct lfid);
1821 return E2BIG;
1822 }
1823 *fh_size = sizeof(struct lfid);
1824 ip = VTOI(vp);
1825 memset(&lfh, 0, sizeof(lfh));
1826 lfh.lfid_len = sizeof(struct lfid);
1827 lfh.lfid_ino = ip->i_number;
1828 lfh.lfid_gen = ip->i_gen;
1829 lfh.lfid_ident = lfs_sb_getident(ip->i_lfs);
1830 memcpy(fhp, &lfh, sizeof(lfh));
1831 return (0);
1832 }
1833
1834 /*
1835 * ulfs_bmaparray callback function for writing.
1836 *
1837 * Since blocks will be written to the new segment anyway,
1838 * we don't care about current daddr of them.
1839 */
1840 static bool
lfs_issequential_hole(const struct lfs * fs,daddr_t daddr0,daddr_t daddr1)1841 lfs_issequential_hole(const struct lfs *fs,
1842 daddr_t daddr0, daddr_t daddr1)
1843 {
1844 (void)fs; /* not used */
1845
1846 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */
1847 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */
1848
1849 KASSERT(daddr0 == UNWRITTEN ||
1850 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR(fs)));
1851 KASSERT(daddr1 == UNWRITTEN ||
1852 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR(fs)));
1853
1854 /* NOTE: all we want to know here is 'hole or not'. */
1855 /* NOTE: UNASSIGNED is converted to 0 by ulfs_bmaparray. */
1856
1857 /*
1858 * treat UNWRITTENs and all resident blocks as 'contiguous'
1859 */
1860 if (daddr0 != 0 && daddr1 != 0)
1861 return true;
1862
1863 /*
1864 * both are in hole?
1865 */
1866 if (daddr0 == 0 && daddr1 == 0)
1867 return true; /* all holes are 'contiguous' for us. */
1868
1869 return false;
1870 }
1871
1872 /*
1873 * lfs_gop_write functions exactly like genfs_gop_write, except that
1874 * (1) it requires the seglock to be held by its caller, and sp->fip
1875 * to be properly initialized (it will return without re-initializing
1876 * sp->fip, and without calling lfs_writeseg).
1877 * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1878 * to determine how large a block it can write at once (though it does
1879 * still use VOP_BMAP to find holes in the file);
1880 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1881 * (leaving lfs_writeseg to deal with the cluster blocks, so we might
1882 * now have clusters of clusters, ick.)
1883 */
1884 static int
lfs_gop_write(struct vnode * vp,struct vm_page ** pgs,int npages,int flags)1885 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1886 int flags)
1887 {
1888 int i, error, run, haveeof = 0;
1889 int fs_bshift;
1890 vaddr_t kva;
1891 off_t eof, offset, startoffset = 0;
1892 size_t bytes, iobytes, skipbytes;
1893 bool async = (flags & PGO_SYNCIO) == 0;
1894 daddr_t lbn, blkno;
1895 struct vm_page *pg;
1896 struct buf *mbp, *bp;
1897 struct vnode *devvp = VTOI(vp)->i_devvp;
1898 struct inode *ip = VTOI(vp);
1899 struct lfs *fs = ip->i_lfs;
1900 struct segment *sp = fs->lfs_sp;
1901 SEGSUM *ssp;
1902 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1903 const char * failreason = NULL;
1904
1905 ASSERT_SEGLOCK(fs);
1906
1907 /* The Ifile lives in the buffer cache */
1908 KASSERT(vp != fs->lfs_ivnode);
1909
1910 /*
1911 * We don't want to fill the disk before the cleaner has a chance
1912 * to make room for us. If we're in danger of doing that, fail
1913 * with EAGAIN. The caller will have to notice this, unlock
1914 * so the cleaner can run, relock and try again.
1915 *
1916 * We must write everything, however, if our vnode is being
1917 * reclaimed.
1918 */
1919 mutex_enter(vp->v_interlock);
1920 if (LFS_STARVED_FOR_SEGS(fs) && vdead_check(vp, VDEAD_NOWAIT) == 0) {
1921 mutex_exit(vp->v_interlock);
1922 failreason = "Starved for segs and not flushing vp";
1923 goto tryagain;
1924 }
1925 mutex_exit(vp->v_interlock);
1926
1927 /*
1928 * Sometimes things slip past the filters in lfs_putpages,
1929 * and the pagedaemon tries to write pages---problem is
1930 * that the pagedaemon never acquires the segment lock.
1931 *
1932 * Alternatively, pages that were clean when we called
1933 * genfs_putpages may have become dirty in the meantime. In this
1934 * case the segment header is not properly set up for blocks
1935 * to be added to it.
1936 *
1937 * Unbusy and unclean the pages, and put them on the ACTIVE
1938 * queue under the hypothesis that they couldn't have got here
1939 * unless they were modified *quite* recently.
1940 *
1941 * XXXUBC that last statement is an oversimplification of course.
1942 */
1943 if (!LFS_SEGLOCK_HELD(fs)) {
1944 failreason = "Seglock not held";
1945 goto tryagain;
1946 }
1947 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) {
1948 failreason = "Inode with no_gop_write";
1949 goto tryagain;
1950 }
1951 if ((pgs[0]->offset & lfs_sb_getbmask(fs)) != 0) {
1952 failreason = "Bad page offset";
1953 goto tryagain;
1954 }
1955
1956 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1957 vp, pgs, npages, flags);
1958
1959 GOP_SIZE(vp, vp->v_size, &eof, 0);
1960 haveeof = 1;
1961
1962 if (vp->v_type == VREG)
1963 fs_bshift = vp->v_mount->mnt_fs_bshift;
1964 else
1965 fs_bshift = DEV_BSHIFT;
1966 error = 0;
1967 pg = pgs[0];
1968 startoffset = pg->offset;
1969 KASSERT(eof >= 0);
1970
1971 if (startoffset >= eof) {
1972 failreason = "Offset beyond EOF";
1973 goto tryagain;
1974 } else
1975 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1976 skipbytes = 0;
1977
1978 KASSERT(bytes != 0);
1979
1980 /* Swap PG_DELWRI for PG_PAGEOUT */
1981 for (i = 0; i < npages; i++) {
1982 if (pgs[i]->flags & PG_DELWRI) {
1983 KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1984 pgs[i]->flags &= ~PG_DELWRI;
1985 pgs[i]->flags |= PG_PAGEOUT;
1986 uvm_pageout_start(1);
1987 mutex_enter(vp->v_interlock);
1988 mutex_enter(&uvm_pageqlock);
1989 uvm_pageunwire(pgs[i]);
1990 mutex_exit(&uvm_pageqlock);
1991 mutex_exit(vp->v_interlock);
1992 }
1993 }
1994
1995 /*
1996 * Check to make sure we're starting on a block boundary.
1997 * We'll check later to make sure we always write entire
1998 * blocks (or fragments).
1999 */
2000 if (startoffset & lfs_sb_getbmask(fs))
2001 printf("%" PRId64 " & %" PRIu64 " = %" PRId64 "\n",
2002 startoffset, lfs_sb_getbmask(fs),
2003 startoffset & lfs_sb_getbmask(fs));
2004 KASSERT((startoffset & lfs_sb_getbmask(fs)) == 0);
2005 if (bytes & lfs_sb_getffmask(fs)) {
2006 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
2007 panic("lfs_gop_write: non-integer blocks");
2008 }
2009
2010 /*
2011 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
2012 * If we would, write what we have and try again. If we don't
2013 * have anything to write, we'll have to sleep.
2014 */
2015 ssp = (SEGSUM *)sp->segsum;
2016 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
2017 (lfs_ss_getnfinfo(fs, ssp) < 1 ?
2018 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
2019 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
2020 #if 0
2021 " with nfinfo=%d at offset 0x%jx\n",
2022 (int)lfs_ss_getnfinfo(fs, ssp),
2023 (uintmax_t)lfs_sb_getoffset(fs)));
2024 #endif
2025 lfs_updatemeta(sp);
2026 lfs_release_finfo(fs);
2027 (void) lfs_writeseg(fs, sp);
2028
2029 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
2030
2031 /*
2032 * Having given up all of the pager_map we were holding,
2033 * we can now wait for aiodoned to reclaim it for us
2034 * without fear of deadlock.
2035 */
2036 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
2037 UVMPAGER_MAPIN_WAITOK);
2038 }
2039
2040 mbp = getiobuf(NULL, true);
2041 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
2042 vp, mbp, vp->v_numoutput, bytes);
2043 mbp->b_bufsize = npages << PAGE_SHIFT;
2044 mbp->b_data = (void *)kva;
2045 mbp->b_resid = mbp->b_bcount = bytes;
2046 mbp->b_cflags = BC_BUSY|BC_AGE;
2047 mbp->b_iodone = uvm_aio_biodone;
2048
2049 bp = NULL;
2050 for (offset = startoffset;
2051 bytes > 0;
2052 offset += iobytes, bytes -= iobytes) {
2053 lbn = offset >> fs_bshift;
2054 error = ulfs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
2055 lfs_issequential_hole);
2056 if (error) {
2057 UVMHIST_LOG(ubchist, "ulfs_bmaparray() -> %d",
2058 error,0,0,0);
2059 skipbytes += bytes;
2060 bytes = 0;
2061 break;
2062 }
2063
2064 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
2065 bytes);
2066 if (blkno == (daddr_t)-1) {
2067 skipbytes += iobytes;
2068 continue;
2069 }
2070
2071 /*
2072 * Discover how much we can really pack into this buffer.
2073 */
2074 /* If no room in the current segment, finish it up */
2075 if (sp->sum_bytes_left < sizeof(int32_t) ||
2076 sp->seg_bytes_left < (1 << lfs_sb_getbshift(fs))) {
2077 int vers;
2078
2079 lfs_updatemeta(sp);
2080 vers = lfs_fi_getversion(fs, sp->fip);
2081 lfs_release_finfo(fs);
2082 (void) lfs_writeseg(fs, sp);
2083
2084 lfs_acquire_finfo(fs, ip->i_number, vers);
2085 }
2086 /* Check both for space in segment and space in segsum */
2087 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
2088 << fs_bshift);
2089 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
2090 << fs_bshift);
2091 KASSERT(iobytes > 0);
2092
2093 /* if it's really one i/o, don't make a second buf */
2094 if (offset == startoffset && iobytes == bytes) {
2095 bp = mbp;
2096 /*
2097 * All the LFS output is done by the segwriter. It
2098 * will increment numoutput by one for all the bufs it
2099 * recieves. However this buffer needs one extra to
2100 * account for aiodone.
2101 */
2102 mutex_enter(vp->v_interlock);
2103 vp->v_numoutput++;
2104 mutex_exit(vp->v_interlock);
2105 } else {
2106 bp = getiobuf(NULL, true);
2107 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
2108 vp, bp, vp->v_numoutput, 0);
2109 nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes);
2110 /*
2111 * LFS doesn't like async I/O here, dies with
2112 * an assert in lfs_bwrite(). Is that assert
2113 * valid? I retained non-async behaviour when
2114 * converted this to use nestiobuf --pooka
2115 */
2116 bp->b_flags &= ~B_ASYNC;
2117 }
2118
2119 /* XXX This is silly ... is this necessary? */
2120 mutex_enter(&bufcache_lock);
2121 mutex_enter(vp->v_interlock);
2122 bgetvp(vp, bp);
2123 mutex_exit(vp->v_interlock);
2124 mutex_exit(&bufcache_lock);
2125
2126 bp->b_lblkno = lfs_lblkno(fs, offset);
2127 bp->b_private = mbp;
2128 if (devvp->v_type == VBLK) {
2129 bp->b_dev = devvp->v_rdev;
2130 }
2131 VOP_BWRITE(bp->b_vp, bp);
2132 while (lfs_gatherblock(sp, bp, NULL))
2133 continue;
2134 }
2135
2136 nestiobuf_done(mbp, skipbytes, error);
2137 if (skipbytes) {
2138 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
2139 }
2140 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
2141
2142 if (!async) {
2143 /* Start a segment write. */
2144 UVMHIST_LOG(ubchist, "flushing", 0,0,0,0);
2145 mutex_enter(&lfs_lock);
2146 lfs_flush(fs, 0, 1);
2147 mutex_exit(&lfs_lock);
2148 }
2149
2150 if ((sp->seg_flags & SEGM_SINGLE) && lfs_sb_getcurseg(fs) != fs->lfs_startseg)
2151 return EAGAIN;
2152
2153 return (0);
2154
2155 tryagain:
2156 /*
2157 * We can't write the pages, for whatever reason.
2158 * Clean up after ourselves, and make the caller try again.
2159 */
2160 mutex_enter(vp->v_interlock);
2161
2162 /* Tell why we're here, if we know */
2163 if (failreason != NULL) {
2164 DLOG((DLOG_PAGE, "lfs_gop_write: %s\n", failreason));
2165 }
2166 if (haveeof && startoffset >= eof) {
2167 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64
2168 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number,
2169 pgs[0]->offset, eof, npages));
2170 }
2171
2172 mutex_enter(&uvm_pageqlock);
2173 for (i = 0; i < npages; i++) {
2174 pg = pgs[i];
2175
2176 if (pg->flags & PG_PAGEOUT)
2177 uvm_pageout_done(1);
2178 if (pg->flags & PG_DELWRI) {
2179 uvm_pageunwire(pg);
2180 }
2181 uvm_pageactivate(pg);
2182 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
2183 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg,
2184 vp, pg->offset));
2185 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags));
2186 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags));
2187 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon));
2188 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject));
2189 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i,
2190 pg->wire_count));
2191 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i,
2192 pg->loan_count));
2193 }
2194 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
2195 uvm_page_unbusy(pgs, npages);
2196 mutex_exit(&uvm_pageqlock);
2197 mutex_exit(vp->v_interlock);
2198 return EAGAIN;
2199 }
2200
2201 /*
2202 * finish vnode/inode initialization.
2203 * used by lfs_vget.
2204 */
2205 void
lfs_vinit(struct mount * mp,struct vnode ** vpp)2206 lfs_vinit(struct mount *mp, struct vnode **vpp)
2207 {
2208 struct vnode *vp = *vpp;
2209 struct inode *ip = VTOI(vp);
2210 struct ulfsmount *ump = VFSTOULFS(mp);
2211 struct lfs *fs = ump->um_lfs;
2212 int i;
2213
2214 ip->i_mode = lfs_dino_getmode(fs, ip->i_din);
2215 ip->i_nlink = lfs_dino_getnlink(fs, ip->i_din);
2216 ip->i_lfs_osize = ip->i_size = lfs_dino_getsize(fs, ip->i_din);
2217 ip->i_flags = lfs_dino_getflags(fs, ip->i_din);
2218 ip->i_gen = lfs_dino_getgen(fs, ip->i_din);
2219 ip->i_uid = lfs_dino_getuid(fs, ip->i_din);
2220 ip->i_gid = lfs_dino_getgid(fs, ip->i_din);
2221
2222 ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din);
2223 ip->i_lfs_odnlink = lfs_dino_getnlink(fs, ip->i_din);
2224
2225 /*
2226 * Initialize the vnode from the inode, check for aliases. In all
2227 * cases re-init ip, the underlying vnode/inode may have changed.
2228 */
2229 ulfs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2230 ip = VTOI(vp);
2231
2232 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
2233 if (vp->v_type != VLNK || ip->i_size >= ip->i_lfs->um_maxsymlinklen) {
2234 #ifdef DEBUG
2235 for (i = (ip->i_size + lfs_sb_getbsize(fs) - 1) >> lfs_sb_getbshift(fs);
2236 i < ULFS_NDADDR; i++) {
2237 if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
2238 i == 0)
2239 continue;
2240 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) {
2241 lfs_dump_dinode(fs, ip->i_din);
2242 panic("inconsistent inode (direct)");
2243 }
2244 }
2245 for ( ; i < ULFS_NDADDR + ULFS_NIADDR; i++) {
2246 if (lfs_dino_getib(fs, ip->i_din, i - ULFS_NDADDR) != 0) {
2247 lfs_dump_dinode(fs, ip->i_din);
2248 panic("inconsistent inode (indirect)");
2249 }
2250 }
2251 #endif /* DEBUG */
2252 for (i = 0; i < ULFS_NDADDR; i++)
2253 if (lfs_dino_getdb(fs, ip->i_din, i) != 0)
2254 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i);
2255 }
2256
2257 #ifdef DIAGNOSTIC
2258 if (vp->v_type == VNON) {
2259 # ifdef DEBUG
2260 lfs_dump_dinode(fs, ip->i_din);
2261 # endif
2262 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n",
2263 (unsigned long long)ip->i_number,
2264 (ip->i_mode & LFS_IFMT) >> 12);
2265 }
2266 #endif /* DIAGNOSTIC */
2267
2268 /*
2269 * Finish inode initialization now that aliasing has been resolved.
2270 */
2271
2272 ip->i_devvp = ump->um_devvp;
2273 vref(ip->i_devvp);
2274 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
2275 ulfsquota_init(ip);
2276 #endif
2277 genfs_node_init(vp, &lfs_genfsops);
2278 uvm_vnp_setsize(vp, ip->i_size);
2279
2280 /* Initialize hiblk from file size */
2281 ip->i_lfs_hiblk = lfs_lblkno(ip->i_lfs, ip->i_size + lfs_sb_getbsize(ip->i_lfs) - 1) - 1;
2282
2283 *vpp = vp;
2284 }
2285
2286 /*
2287 * Resize the filesystem to contain the specified number of segments.
2288 */
2289 int
lfs_resize_fs(struct lfs * fs,int newnsegs)2290 lfs_resize_fs(struct lfs *fs, int newnsegs)
2291 {
2292 SEGUSE *sup;
2293 CLEANERINFO *cip;
2294 struct buf *bp, *obp;
2295 daddr_t olast, nlast, ilast, noff, start, end;
2296 struct vnode *ivp;
2297 struct inode *ip;
2298 int error, badnews, inc, oldnsegs;
2299 int sbbytes, csbbytes, gain, cgain;
2300 int i;
2301
2302 /* Only support v2 and up */
2303 if (lfs_sb_getversion(fs) < 2)
2304 return EOPNOTSUPP;
2305
2306 /* If we're doing nothing, do it fast */
2307 oldnsegs = lfs_sb_getnseg(fs);
2308 if (newnsegs == oldnsegs)
2309 return 0;
2310
2311 /* We always have to have two superblocks */
2312 if (newnsegs <= lfs_dtosn(fs, lfs_sb_getsboff(fs, 1)))
2313 /* XXX this error code is rather nonsense */
2314 return EFBIG;
2315
2316 ivp = fs->lfs_ivnode;
2317 ip = VTOI(ivp);
2318 error = 0;
2319
2320 /* Take the segment lock so no one else calls lfs_newseg() */
2321 lfs_seglock(fs, SEGM_PROT);
2322
2323 /*
2324 * Make sure the segments we're going to be losing, if any,
2325 * are in fact empty. We hold the seglock, so their status
2326 * cannot change underneath us. Count the superblocks we lose,
2327 * while we're at it.
2328 */
2329 sbbytes = csbbytes = 0;
2330 cgain = 0;
2331 for (i = newnsegs; i < oldnsegs; i++) {
2332 LFS_SEGENTRY(sup, fs, i, bp);
2333 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL);
2334 if (sup->su_flags & SEGUSE_SUPERBLOCK)
2335 sbbytes += LFS_SBPAD;
2336 if (!(sup->su_flags & SEGUSE_DIRTY)) {
2337 ++cgain;
2338 if (sup->su_flags & SEGUSE_SUPERBLOCK)
2339 csbbytes += LFS_SBPAD;
2340 }
2341 brelse(bp, 0);
2342 if (badnews) {
2343 error = EBUSY;
2344 goto out;
2345 }
2346 }
2347
2348 /* Note old and new segment table endpoints, and old ifile size */
2349 olast = lfs_sb_getcleansz(fs) + lfs_sb_getsegtabsz(fs);
2350 nlast = howmany(newnsegs, lfs_sb_getsepb(fs)) + lfs_sb_getcleansz(fs);
2351 ilast = ivp->v_size >> lfs_sb_getbshift(fs);
2352 noff = nlast - olast;
2353
2354 /*
2355 * Make sure no one can use the Ifile while we change it around.
2356 * Even after taking the iflock we need to make sure no one still
2357 * is holding Ifile buffers, so we get each one, to drain them.
2358 * (XXX this could be done better.)
2359 */
2360 rw_enter(&fs->lfs_iflock, RW_WRITER);
2361 for (i = 0; i < ilast; i++) {
2362 /* XXX what to do if bread fails? */
2363 bread(ivp, i, lfs_sb_getbsize(fs), 0, &bp);
2364 brelse(bp, 0);
2365 }
2366
2367 /* Allocate new Ifile blocks */
2368 for (i = ilast; i < ilast + noff; i++) {
2369 if (lfs_balloc(ivp, i * lfs_sb_getbsize(fs), lfs_sb_getbsize(fs), NOCRED, 0,
2370 &bp) != 0)
2371 panic("balloc extending ifile");
2372 memset(bp->b_data, 0, lfs_sb_getbsize(fs));
2373 VOP_BWRITE(bp->b_vp, bp);
2374 }
2375
2376 /* Register new ifile size */
2377 ip->i_size += noff * lfs_sb_getbsize(fs);
2378 lfs_dino_setsize(fs, ip->i_din, ip->i_size);
2379 uvm_vnp_setsize(ivp, ip->i_size);
2380
2381 /* Copy the inode table to its new position */
2382 if (noff != 0) {
2383 if (noff < 0) {
2384 start = nlast;
2385 end = ilast + noff;
2386 inc = 1;
2387 } else {
2388 start = ilast + noff - 1;
2389 end = nlast - 1;
2390 inc = -1;
2391 }
2392 for (i = start; i != end; i += inc) {
2393 if (bread(ivp, i, lfs_sb_getbsize(fs),
2394 B_MODIFY, &bp) != 0)
2395 panic("resize: bread dst blk failed");
2396 if (bread(ivp, i - noff, lfs_sb_getbsize(fs),
2397 0, &obp))
2398 panic("resize: bread src blk failed");
2399 memcpy(bp->b_data, obp->b_data, lfs_sb_getbsize(fs));
2400 VOP_BWRITE(bp->b_vp, bp);
2401 brelse(obp, 0);
2402 }
2403 }
2404
2405 /* If we are expanding, write the new empty SEGUSE entries */
2406 if (newnsegs > oldnsegs) {
2407 for (i = oldnsegs; i < newnsegs; i++) {
2408 if ((error = bread(ivp, i / lfs_sb_getsepb(fs) +
2409 lfs_sb_getcleansz(fs), lfs_sb_getbsize(fs),
2410 B_MODIFY, &bp)) != 0)
2411 panic("lfs: ifile read: %d", error);
2412 while ((i + 1) % lfs_sb_getsepb(fs) && i < newnsegs) {
2413 sup = &((SEGUSE *)bp->b_data)[i % lfs_sb_getsepb(fs)];
2414 memset(sup, 0, sizeof(*sup));
2415 i++;
2416 }
2417 VOP_BWRITE(bp->b_vp, bp);
2418 }
2419 }
2420
2421 /* Zero out unused superblock offsets */
2422 for (i = 2; i < LFS_MAXNUMSB; i++)
2423 if (lfs_dtosn(fs, lfs_sb_getsboff(fs, i)) >= newnsegs)
2424 lfs_sb_setsboff(fs, i, 0x0);
2425
2426 /*
2427 * Correct superblock entries that depend on fs size.
2428 * The computations of these are as follows:
2429 *
2430 * size = lfs_segtod(fs, nseg)
2431 * dsize = lfs_segtod(fs, nseg - minfreeseg) - lfs_btofsb(#super * LFS_SBPAD)
2432 * bfree = dsize - lfs_btofsb(fs, bsize * nseg / 2) - blocks_actually_used
2433 * avail = lfs_segtod(fs, nclean) - lfs_btofsb(#clean_super * LFS_SBPAD)
2434 * + (lfs_segtod(fs, 1) - (offset - curseg))
2435 * - lfs_segtod(fs, minfreeseg - (minfreeseg / 2))
2436 *
2437 * XXX - we should probably adjust minfreeseg as well.
2438 */
2439 gain = (newnsegs - oldnsegs);
2440 lfs_sb_setnseg(fs, newnsegs);
2441 lfs_sb_setsegtabsz(fs, nlast - lfs_sb_getcleansz(fs));
2442 lfs_sb_addsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)));
2443 lfs_sb_adddsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes));
2444 lfs_sb_addbfree(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes)
2445 - gain * lfs_btofsb(fs, lfs_sb_getbsize(fs) / 2));
2446 if (gain > 0) {
2447 lfs_sb_addnclean(fs, gain);
2448 lfs_sb_addavail(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)));
2449 } else {
2450 lfs_sb_subnclean(fs, cgain);
2451 lfs_sb_subavail(fs, cgain * lfs_btofsb(fs, lfs_sb_getssize(fs)) -
2452 lfs_btofsb(fs, csbbytes));
2453 }
2454
2455 /* Resize segment flag cache */
2456 fs->lfs_suflags[0] = realloc(fs->lfs_suflags[0],
2457 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK);
2458 fs->lfs_suflags[1] = realloc(fs->lfs_suflags[1],
2459 lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK);
2460 for (i = oldnsegs; i < newnsegs; i++)
2461 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0;
2462
2463 /* Truncate Ifile if necessary */
2464 if (noff < 0)
2465 lfs_truncate(ivp, ivp->v_size + (noff << lfs_sb_getbshift(fs)), 0,
2466 NOCRED);
2467
2468 /* Update cleaner info so the cleaner can die */
2469 /* XXX what to do if bread fails? */
2470 bread(ivp, 0, lfs_sb_getbsize(fs), B_MODIFY, &bp);
2471 cip = bp->b_data;
2472 lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs));
2473 lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs));
2474 VOP_BWRITE(bp->b_vp, bp);
2475
2476 /* Let Ifile accesses proceed */
2477 rw_exit(&fs->lfs_iflock);
2478
2479 out:
2480 lfs_segunlock(fs);
2481 return error;
2482 }
2483
2484 /*
2485 * Extended attribute dispatch
2486 */
2487 int
lfs_extattrctl(struct mount * mp,int cmd,struct vnode * vp,int attrnamespace,const char * attrname)2488 lfs_extattrctl(struct mount *mp, int cmd, struct vnode *vp,
2489 int attrnamespace, const char *attrname)
2490 {
2491 #ifdef LFS_EXTATTR
2492 struct ulfsmount *ump;
2493
2494 ump = VFSTOULFS(mp);
2495 if (ump->um_fstype == ULFS1) {
2496 return ulfs_extattrctl(mp, cmd, vp, attrnamespace, attrname);
2497 }
2498 #endif
2499 return vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname);
2500 }
2501