xref: /netbsd-src/sys/kern/vfs_wapbl.c (revision 413d532bcc3f62d122e56d92e13ac64825a40baf)
1 /*	$NetBSD: vfs_wapbl.c,v 1.59 2014/02/25 18:30:11 pooka Exp $	*/
2 
3 /*-
4  * Copyright (c) 2003, 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Wasabi Systems, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * This implements file system independent write ahead filesystem logging.
34  */
35 
36 #define WAPBL_INTERNAL
37 
38 #include <sys/cdefs.h>
39 __KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.59 2014/02/25 18:30:11 pooka Exp $");
40 
41 #include <sys/param.h>
42 #include <sys/bitops.h>
43 
44 #ifdef _KERNEL
45 #include <sys/param.h>
46 #include <sys/namei.h>
47 #include <sys/proc.h>
48 #include <sys/sysctl.h>
49 #include <sys/uio.h>
50 #include <sys/vnode.h>
51 #include <sys/file.h>
52 #include <sys/module.h>
53 #include <sys/resourcevar.h>
54 #include <sys/conf.h>
55 #include <sys/mount.h>
56 #include <sys/kernel.h>
57 #include <sys/kauth.h>
58 #include <sys/mutex.h>
59 #include <sys/atomic.h>
60 #include <sys/wapbl.h>
61 #include <sys/wapbl_replay.h>
62 
63 #include <miscfs/specfs/specdev.h>
64 
65 #define	wapbl_alloc(s) kmem_alloc((s), KM_SLEEP)
66 #define	wapbl_free(a, s) kmem_free((a), (s))
67 #define	wapbl_calloc(n, s) kmem_zalloc((n)*(s), KM_SLEEP)
68 
69 static struct sysctllog *wapbl_sysctl;
70 static int wapbl_flush_disk_cache = 1;
71 static int wapbl_verbose_commit = 0;
72 
73 static inline size_t wapbl_space_free(size_t, off_t, off_t);
74 
75 #else /* !_KERNEL */
76 #include <assert.h>
77 #include <errno.h>
78 #include <stdio.h>
79 #include <stdbool.h>
80 #include <stdlib.h>
81 #include <string.h>
82 
83 #include <sys/time.h>
84 #include <sys/wapbl.h>
85 #include <sys/wapbl_replay.h>
86 
87 #define	KDASSERT(x) assert(x)
88 #define	KASSERT(x) assert(x)
89 #define	wapbl_alloc(s) malloc(s)
90 #define	wapbl_free(a, s) free(a)
91 #define	wapbl_calloc(n, s) calloc((n), (s))
92 
93 #endif /* !_KERNEL */
94 
95 /*
96  * INTERNAL DATA STRUCTURES
97  */
98 
99 /*
100  * This structure holds per-mount log information.
101  *
102  * Legend:	a = atomic access only
103  *		r = read-only after init
104  *		l = rwlock held
105  *		m = mutex held
106  *		lm = rwlock held writing or mutex held
107  *		u = unlocked access ok
108  *		b = bufcache_lock held
109  */
110 struct wapbl {
111 	struct vnode *wl_logvp;	/* r:	log here */
112 	struct vnode *wl_devvp;	/* r:	log on this device */
113 	struct mount *wl_mount;	/* r:	mountpoint wl is associated with */
114 	daddr_t wl_logpbn;	/* r:	Physical block number of start of log */
115 	int wl_log_dev_bshift;	/* r:	logarithm of device block size of log
116 					device */
117 	int wl_fs_dev_bshift;	/* r:	logarithm of device block size of
118 					filesystem device */
119 
120 	unsigned wl_lock_count;	/* m:	Count of transactions in progress */
121 
122 	size_t wl_circ_size; 	/* r:	Number of bytes in buffer of log */
123 	size_t wl_circ_off;	/* r:	Number of bytes reserved at start */
124 
125 	size_t wl_bufcount_max;	/* r:	Number of buffers reserved for log */
126 	size_t wl_bufbytes_max;	/* r:	Number of buf bytes reserved for log */
127 
128 	off_t wl_head;		/* l:	Byte offset of log head */
129 	off_t wl_tail;		/* l:	Byte offset of log tail */
130 	/*
131 	 * head == tail == 0 means log is empty
132 	 * head == tail != 0 means log is full
133 	 * see assertions in wapbl_advance() for other boundary conditions.
134 	 * only truncate moves the tail, except when flush sets it to
135 	 * wl_header_size only flush moves the head, except when truncate
136 	 * sets it to 0.
137 	 */
138 
139 	struct wapbl_wc_header *wl_wc_header;	/* l	*/
140 	void *wl_wc_scratch;	/* l:	scratch space (XXX: por que?!?) */
141 
142 	kmutex_t wl_mtx;	/* u:	short-term lock */
143 	krwlock_t wl_rwlock;	/* u:	File system transaction lock */
144 
145 	/*
146 	 * Must be held while accessing
147 	 * wl_count or wl_bufs or head or tail
148 	 */
149 
150 	/*
151 	 * Callback called from within the flush routine to flush any extra
152 	 * bits.  Note that flush may be skipped without calling this if
153 	 * there are no outstanding buffers in the transaction.
154 	 */
155 #if _KERNEL
156 	wapbl_flush_fn_t wl_flush;	/* r	*/
157 	wapbl_flush_fn_t wl_flush_abort;/* r	*/
158 #endif
159 
160 	size_t wl_bufbytes;	/* m:	Byte count of pages in wl_bufs */
161 	size_t wl_bufcount;	/* m:	Count of buffers in wl_bufs */
162 	size_t wl_bcount;	/* m:	Total bcount of wl_bufs */
163 
164 	LIST_HEAD(, buf) wl_bufs; /* m:	Buffers in current transaction */
165 
166 	kcondvar_t wl_reclaimable_cv;	/* m (obviously) */
167 	size_t wl_reclaimable_bytes; /* m:	Amount of space available for
168 						reclamation by truncate */
169 	int wl_error_count;	/* m:	# of wl_entries with errors */
170 	size_t wl_reserved_bytes; /* never truncate log smaller than this */
171 
172 #ifdef WAPBL_DEBUG_BUFBYTES
173 	size_t wl_unsynced_bufbytes; /* Byte count of unsynced buffers */
174 #endif
175 
176 	daddr_t *wl_deallocblks;/* lm:	address of block */
177 	int *wl_dealloclens;	/* lm:	size of block */
178 	int wl_dealloccnt;	/* lm:	total count */
179 	int wl_dealloclim;	/* l:	max count */
180 
181 	/* hashtable of inode numbers for allocated but unlinked inodes */
182 	/* synch ??? */
183 	LIST_HEAD(wapbl_ino_head, wapbl_ino) *wl_inohash;
184 	u_long wl_inohashmask;
185 	int wl_inohashcnt;
186 
187 	SIMPLEQ_HEAD(, wapbl_entry) wl_entries; /* On disk transaction
188 						   accounting */
189 
190 	u_char *wl_buffer;	/* l:   buffer for wapbl_buffered_write() */
191 	daddr_t wl_buffer_dblk;	/* l:   buffer disk block address */
192 	size_t wl_buffer_used;	/* l:   buffer current use */
193 };
194 
195 #ifdef WAPBL_DEBUG_PRINT
196 int wapbl_debug_print = WAPBL_DEBUG_PRINT;
197 #endif
198 
199 /****************************************************************/
200 #ifdef _KERNEL
201 
202 #ifdef WAPBL_DEBUG
203 struct wapbl *wapbl_debug_wl;
204 #endif
205 
206 static int wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail);
207 static int wapbl_write_blocks(struct wapbl *wl, off_t *offp);
208 static int wapbl_write_revocations(struct wapbl *wl, off_t *offp);
209 static int wapbl_write_inodes(struct wapbl *wl, off_t *offp);
210 #endif /* _KERNEL */
211 
212 static int wapbl_replay_process(struct wapbl_replay *wr, off_t, off_t);
213 
214 static inline size_t wapbl_space_used(size_t avail, off_t head,
215 	off_t tail);
216 
217 #ifdef _KERNEL
218 
219 static struct pool wapbl_entry_pool;
220 
221 #define	WAPBL_INODETRK_SIZE 83
222 static int wapbl_ino_pool_refcount;
223 static struct pool wapbl_ino_pool;
224 struct wapbl_ino {
225 	LIST_ENTRY(wapbl_ino) wi_hash;
226 	ino_t wi_ino;
227 	mode_t wi_mode;
228 };
229 
230 static void wapbl_inodetrk_init(struct wapbl *wl, u_int size);
231 static void wapbl_inodetrk_free(struct wapbl *wl);
232 static struct wapbl_ino *wapbl_inodetrk_get(struct wapbl *wl, ino_t ino);
233 
234 static size_t wapbl_transaction_len(struct wapbl *wl);
235 static inline size_t wapbl_transaction_inodes_len(struct wapbl *wl);
236 
237 #if 0
238 int wapbl_replay_verify(struct wapbl_replay *, struct vnode *);
239 #endif
240 
241 static int wapbl_replay_isopen1(struct wapbl_replay *);
242 
243 /*
244  * This is useful for debugging.  If set, the log will
245  * only be truncated when necessary.
246  */
247 int wapbl_lazy_truncate = 0;
248 
249 struct wapbl_ops wapbl_ops = {
250 	.wo_wapbl_discard	= wapbl_discard,
251 	.wo_wapbl_replay_isopen	= wapbl_replay_isopen1,
252 	.wo_wapbl_replay_can_read = wapbl_replay_can_read,
253 	.wo_wapbl_replay_read	= wapbl_replay_read,
254 	.wo_wapbl_add_buf	= wapbl_add_buf,
255 	.wo_wapbl_remove_buf	= wapbl_remove_buf,
256 	.wo_wapbl_resize_buf	= wapbl_resize_buf,
257 	.wo_wapbl_begin		= wapbl_begin,
258 	.wo_wapbl_end		= wapbl_end,
259 	.wo_wapbl_junlock_assert= wapbl_junlock_assert,
260 
261 	/* XXX: the following is only used to say "this is a wapbl buf" */
262 	.wo_wapbl_biodone	= wapbl_biodone,
263 };
264 
265 static int
266 wapbl_sysctl_init(void)
267 {
268 	int rv;
269 	const struct sysctlnode *rnode, *cnode;
270 
271 	wapbl_sysctl = NULL;
272 
273 	rv = sysctl_createv(&wapbl_sysctl, 0, NULL, &rnode,
274 		       CTLFLAG_PERMANENT,
275 		       CTLTYPE_NODE, "wapbl",
276 		       SYSCTL_DESCR("WAPBL journaling options"),
277 		       NULL, 0, NULL, 0,
278 		       CTL_VFS, CTL_CREATE, CTL_EOL);
279 	if (rv)
280 		return rv;
281 
282 	rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &cnode,
283 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
284 		       CTLTYPE_INT, "flush_disk_cache",
285 		       SYSCTL_DESCR("flush disk cache"),
286 		       NULL, 0, &wapbl_flush_disk_cache, 0,
287 		       CTL_CREATE, CTL_EOL);
288 	if (rv)
289 		return rv;
290 
291 	rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &cnode,
292 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
293 		       CTLTYPE_INT, "verbose_commit",
294 		       SYSCTL_DESCR("show time and size of wapbl log commits"),
295 		       NULL, 0, &wapbl_verbose_commit, 0,
296 		       CTL_CREATE, CTL_EOL);
297 	return rv;
298 }
299 
300 static void
301 wapbl_init(void)
302 {
303 
304 	pool_init(&wapbl_entry_pool, sizeof(struct wapbl_entry), 0, 0, 0,
305 	    "wapblentrypl", &pool_allocator_kmem, IPL_VM);
306 
307 	wapbl_sysctl_init();
308 }
309 
310 #ifdef notyet
311 static int
312 wapbl_fini(bool interface)
313 {
314 
315 	if (aio_sysctl != NULL)
316 		 sysctl_teardown(&aio_sysctl);
317 
318 	pool_destroy(&wapbl_entry_pool);
319 
320 	return 0;
321 }
322 #endif
323 
324 static int
325 wapbl_start_flush_inodes(struct wapbl *wl, struct wapbl_replay *wr)
326 {
327 	int error, i;
328 
329 	WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
330 	    ("wapbl_start: reusing log with %d inodes\n", wr->wr_inodescnt));
331 
332 	/*
333 	 * Its only valid to reuse the replay log if its
334 	 * the same as the new log we just opened.
335 	 */
336 	KDASSERT(!wapbl_replay_isopen(wr));
337 	KASSERT(wl->wl_devvp->v_type == VBLK);
338 	KASSERT(wr->wr_devvp->v_type == VBLK);
339 	KASSERT(wl->wl_devvp->v_rdev == wr->wr_devvp->v_rdev);
340 	KASSERT(wl->wl_logpbn == wr->wr_logpbn);
341 	KASSERT(wl->wl_circ_size == wr->wr_circ_size);
342 	KASSERT(wl->wl_circ_off == wr->wr_circ_off);
343 	KASSERT(wl->wl_log_dev_bshift == wr->wr_log_dev_bshift);
344 	KASSERT(wl->wl_fs_dev_bshift == wr->wr_fs_dev_bshift);
345 
346 	wl->wl_wc_header->wc_generation = wr->wr_generation + 1;
347 
348 	for (i = 0; i < wr->wr_inodescnt; i++)
349 		wapbl_register_inode(wl, wr->wr_inodes[i].wr_inumber,
350 		    wr->wr_inodes[i].wr_imode);
351 
352 	/* Make sure new transaction won't overwrite old inodes list */
353 	KDASSERT(wapbl_transaction_len(wl) <=
354 	    wapbl_space_free(wl->wl_circ_size, wr->wr_inodeshead,
355 	    wr->wr_inodestail));
356 
357 	wl->wl_head = wl->wl_tail = wr->wr_inodeshead;
358 	wl->wl_reclaimable_bytes = wl->wl_reserved_bytes =
359 	    wapbl_transaction_len(wl);
360 
361 	error = wapbl_write_inodes(wl, &wl->wl_head);
362 	if (error)
363 		return error;
364 
365 	KASSERT(wl->wl_head != wl->wl_tail);
366 	KASSERT(wl->wl_head != 0);
367 
368 	return 0;
369 }
370 
371 int
372 wapbl_start(struct wapbl ** wlp, struct mount *mp, struct vnode *vp,
373 	daddr_t off, size_t count, size_t blksize, struct wapbl_replay *wr,
374 	wapbl_flush_fn_t flushfn, wapbl_flush_fn_t flushabortfn)
375 {
376 	struct wapbl *wl;
377 	struct vnode *devvp;
378 	daddr_t logpbn;
379 	int error;
380 	int log_dev_bshift = ilog2(blksize);
381 	int fs_dev_bshift = log_dev_bshift;
382 	int run;
383 
384 	WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_start: vp=%p off=%" PRId64
385 	    " count=%zu blksize=%zu\n", vp, off, count, blksize));
386 
387 	if (log_dev_bshift > fs_dev_bshift) {
388 		WAPBL_PRINTF(WAPBL_PRINT_OPEN,
389 			("wapbl: log device's block size cannot be larger "
390 			 "than filesystem's\n"));
391 		/*
392 		 * Not currently implemented, although it could be if
393 		 * needed someday.
394 		 */
395 		return ENOSYS;
396 	}
397 
398 	if (off < 0)
399 		return EINVAL;
400 
401 	if (blksize < DEV_BSIZE)
402 		return EINVAL;
403 	if (blksize % DEV_BSIZE)
404 		return EINVAL;
405 
406 	/* XXXTODO: verify that the full load is writable */
407 
408 	/*
409 	 * XXX check for minimum log size
410 	 * minimum is governed by minimum amount of space
411 	 * to complete a transaction. (probably truncate)
412 	 */
413 	/* XXX for now pick something minimal */
414 	if ((count * blksize) < MAXPHYS) {
415 		return ENOSPC;
416 	}
417 
418 	if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, &run)) != 0) {
419 		return error;
420 	}
421 
422 	wl = wapbl_calloc(1, sizeof(*wl));
423 	rw_init(&wl->wl_rwlock);
424 	mutex_init(&wl->wl_mtx, MUTEX_DEFAULT, IPL_NONE);
425 	cv_init(&wl->wl_reclaimable_cv, "wapblrec");
426 	LIST_INIT(&wl->wl_bufs);
427 	SIMPLEQ_INIT(&wl->wl_entries);
428 
429 	wl->wl_logvp = vp;
430 	wl->wl_devvp = devvp;
431 	wl->wl_mount = mp;
432 	wl->wl_logpbn = logpbn;
433 	wl->wl_log_dev_bshift = log_dev_bshift;
434 	wl->wl_fs_dev_bshift = fs_dev_bshift;
435 
436 	wl->wl_flush = flushfn;
437 	wl->wl_flush_abort = flushabortfn;
438 
439 	/* Reserve two log device blocks for the commit headers */
440 	wl->wl_circ_off = 2<<wl->wl_log_dev_bshift;
441 	wl->wl_circ_size = ((count * blksize) - wl->wl_circ_off);
442 	/* truncate the log usage to a multiple of log_dev_bshift */
443 	wl->wl_circ_size >>= wl->wl_log_dev_bshift;
444 	wl->wl_circ_size <<= wl->wl_log_dev_bshift;
445 
446 	/*
447 	 * wl_bufbytes_max limits the size of the in memory transaction space.
448 	 * - Since buffers are allocated and accounted for in units of
449 	 *   PAGE_SIZE it is required to be a multiple of PAGE_SIZE
450 	 *   (i.e. 1<<PAGE_SHIFT)
451 	 * - Since the log device has to be written in units of
452 	 *   1<<wl_log_dev_bshift it is required to be a mulitple of
453 	 *   1<<wl_log_dev_bshift.
454 	 * - Since filesystem will provide data in units of 1<<wl_fs_dev_bshift,
455 	 *   it is convenient to be a multiple of 1<<wl_fs_dev_bshift.
456 	 * Therefore it must be multiple of the least common multiple of those
457 	 * three quantities.  Fortunately, all of those quantities are
458 	 * guaranteed to be a power of two, and the least common multiple of
459 	 * a set of numbers which are all powers of two is simply the maximum
460 	 * of those numbers.  Finally, the maximum logarithm of a power of two
461 	 * is the same as the log of the maximum power of two.  So we can do
462 	 * the following operations to size wl_bufbytes_max:
463 	 */
464 
465 	/* XXX fix actual number of pages reserved per filesystem. */
466 	wl->wl_bufbytes_max = MIN(wl->wl_circ_size, buf_memcalc() / 2);
467 
468 	/* Round wl_bufbytes_max to the largest power of two constraint */
469 	wl->wl_bufbytes_max >>= PAGE_SHIFT;
470 	wl->wl_bufbytes_max <<= PAGE_SHIFT;
471 	wl->wl_bufbytes_max >>= wl->wl_log_dev_bshift;
472 	wl->wl_bufbytes_max <<= wl->wl_log_dev_bshift;
473 	wl->wl_bufbytes_max >>= wl->wl_fs_dev_bshift;
474 	wl->wl_bufbytes_max <<= wl->wl_fs_dev_bshift;
475 
476 	/* XXX maybe use filesystem fragment size instead of 1024 */
477 	/* XXX fix actual number of buffers reserved per filesystem. */
478 	wl->wl_bufcount_max = (nbuf / 2) * 1024;
479 
480 	/* XXX tie this into resource estimation */
481 	wl->wl_dealloclim = wl->wl_bufbytes_max / mp->mnt_stat.f_bsize / 2;
482 
483 	wl->wl_deallocblks = wapbl_alloc(sizeof(*wl->wl_deallocblks) *
484 	    wl->wl_dealloclim);
485 	wl->wl_dealloclens = wapbl_alloc(sizeof(*wl->wl_dealloclens) *
486 	    wl->wl_dealloclim);
487 
488 	wl->wl_buffer = wapbl_alloc(MAXPHYS);
489 	wl->wl_buffer_used = 0;
490 
491 	wapbl_inodetrk_init(wl, WAPBL_INODETRK_SIZE);
492 
493 	/* Initialize the commit header */
494 	{
495 		struct wapbl_wc_header *wc;
496 		size_t len = 1 << wl->wl_log_dev_bshift;
497 		wc = wapbl_calloc(1, len);
498 		wc->wc_type = WAPBL_WC_HEADER;
499 		wc->wc_len = len;
500 		wc->wc_circ_off = wl->wl_circ_off;
501 		wc->wc_circ_size = wl->wl_circ_size;
502 		/* XXX wc->wc_fsid */
503 		wc->wc_log_dev_bshift = wl->wl_log_dev_bshift;
504 		wc->wc_fs_dev_bshift = wl->wl_fs_dev_bshift;
505 		wl->wl_wc_header = wc;
506 		wl->wl_wc_scratch = wapbl_alloc(len);
507 	}
508 
509 	/*
510 	 * if there was an existing set of unlinked but
511 	 * allocated inodes, preserve it in the new
512 	 * log.
513 	 */
514 	if (wr && wr->wr_inodescnt) {
515 		error = wapbl_start_flush_inodes(wl, wr);
516 		if (error)
517 			goto errout;
518 	}
519 
520 	error = wapbl_write_commit(wl, wl->wl_head, wl->wl_tail);
521 	if (error) {
522 		goto errout;
523 	}
524 
525 	*wlp = wl;
526 #if defined(WAPBL_DEBUG)
527 	wapbl_debug_wl = wl;
528 #endif
529 
530 	return 0;
531  errout:
532 	wapbl_discard(wl);
533 	wapbl_free(wl->wl_wc_scratch, wl->wl_wc_header->wc_len);
534 	wapbl_free(wl->wl_wc_header, wl->wl_wc_header->wc_len);
535 	wapbl_free(wl->wl_deallocblks,
536 	    sizeof(*wl->wl_deallocblks) * wl->wl_dealloclim);
537 	wapbl_free(wl->wl_dealloclens,
538 	    sizeof(*wl->wl_dealloclens) * wl->wl_dealloclim);
539 	wapbl_free(wl->wl_buffer, MAXPHYS);
540 	wapbl_inodetrk_free(wl);
541 	wapbl_free(wl, sizeof(*wl));
542 
543 	return error;
544 }
545 
546 /*
547  * Like wapbl_flush, only discards the transaction
548  * completely
549  */
550 
551 void
552 wapbl_discard(struct wapbl *wl)
553 {
554 	struct wapbl_entry *we;
555 	struct buf *bp;
556 	int i;
557 
558 	/*
559 	 * XXX we may consider using upgrade here
560 	 * if we want to call flush from inside a transaction
561 	 */
562 	rw_enter(&wl->wl_rwlock, RW_WRITER);
563 	wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens,
564 	    wl->wl_dealloccnt);
565 
566 #ifdef WAPBL_DEBUG_PRINT
567 	{
568 		pid_t pid = -1;
569 		lwpid_t lid = -1;
570 		if (curproc)
571 			pid = curproc->p_pid;
572 		if (curlwp)
573 			lid = curlwp->l_lid;
574 #ifdef WAPBL_DEBUG_BUFBYTES
575 		WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
576 		    ("wapbl_discard: thread %d.%d discarding "
577 		    "transaction\n"
578 		    "\tbufcount=%zu bufbytes=%zu bcount=%zu "
579 		    "deallocs=%d inodes=%d\n"
580 		    "\terrcnt = %u, reclaimable=%zu reserved=%zu "
581 		    "unsynced=%zu\n",
582 		    pid, lid, wl->wl_bufcount, wl->wl_bufbytes,
583 		    wl->wl_bcount, wl->wl_dealloccnt,
584 		    wl->wl_inohashcnt, wl->wl_error_count,
585 		    wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
586 		    wl->wl_unsynced_bufbytes));
587 		SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
588 			WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
589 			    ("\tentry: bufcount = %zu, reclaimable = %zu, "
590 			     "error = %d, unsynced = %zu\n",
591 			     we->we_bufcount, we->we_reclaimable_bytes,
592 			     we->we_error, we->we_unsynced_bufbytes));
593 		}
594 #else /* !WAPBL_DEBUG_BUFBYTES */
595 		WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
596 		    ("wapbl_discard: thread %d.%d discarding transaction\n"
597 		    "\tbufcount=%zu bufbytes=%zu bcount=%zu "
598 		    "deallocs=%d inodes=%d\n"
599 		    "\terrcnt = %u, reclaimable=%zu reserved=%zu\n",
600 		    pid, lid, wl->wl_bufcount, wl->wl_bufbytes,
601 		    wl->wl_bcount, wl->wl_dealloccnt,
602 		    wl->wl_inohashcnt, wl->wl_error_count,
603 		    wl->wl_reclaimable_bytes, wl->wl_reserved_bytes));
604 		SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
605 			WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
606 			    ("\tentry: bufcount = %zu, reclaimable = %zu, "
607 			     "error = %d\n",
608 			     we->we_bufcount, we->we_reclaimable_bytes,
609 			     we->we_error));
610 		}
611 #endif /* !WAPBL_DEBUG_BUFBYTES */
612 	}
613 #endif /* WAPBL_DEBUG_PRINT */
614 
615 	for (i = 0; i <= wl->wl_inohashmask; i++) {
616 		struct wapbl_ino_head *wih;
617 		struct wapbl_ino *wi;
618 
619 		wih = &wl->wl_inohash[i];
620 		while ((wi = LIST_FIRST(wih)) != NULL) {
621 			LIST_REMOVE(wi, wi_hash);
622 			pool_put(&wapbl_ino_pool, wi);
623 			KASSERT(wl->wl_inohashcnt > 0);
624 			wl->wl_inohashcnt--;
625 		}
626 	}
627 
628 	/*
629 	 * clean buffer list
630 	 */
631 	mutex_enter(&bufcache_lock);
632 	mutex_enter(&wl->wl_mtx);
633 	while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) {
634 		if (bbusy(bp, 0, 0, &wl->wl_mtx) == 0) {
635 			/*
636 			 * The buffer will be unlocked and
637 			 * removed from the transaction in brelse
638 			 */
639 			mutex_exit(&wl->wl_mtx);
640 			brelsel(bp, 0);
641 			mutex_enter(&wl->wl_mtx);
642 		}
643 	}
644 	mutex_exit(&wl->wl_mtx);
645 	mutex_exit(&bufcache_lock);
646 
647 	/*
648 	 * Remove references to this wl from wl_entries, free any which
649 	 * no longer have buffers, others will be freed in wapbl_biodone
650 	 * when they no longer have any buffers.
651 	 */
652 	while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) != NULL) {
653 		SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries);
654 		/* XXX should we be accumulating wl_error_count
655 		 * and increasing reclaimable bytes ? */
656 		we->we_wapbl = NULL;
657 		if (we->we_bufcount == 0) {
658 #ifdef WAPBL_DEBUG_BUFBYTES
659 			KASSERT(we->we_unsynced_bufbytes == 0);
660 #endif
661 			pool_put(&wapbl_entry_pool, we);
662 		}
663 	}
664 
665 	/* Discard list of deallocs */
666 	wl->wl_dealloccnt = 0;
667 	/* XXX should we clear wl_reserved_bytes? */
668 
669 	KASSERT(wl->wl_bufbytes == 0);
670 	KASSERT(wl->wl_bcount == 0);
671 	KASSERT(wl->wl_bufcount == 0);
672 	KASSERT(LIST_EMPTY(&wl->wl_bufs));
673 	KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries));
674 	KASSERT(wl->wl_inohashcnt == 0);
675 
676 	rw_exit(&wl->wl_rwlock);
677 }
678 
679 int
680 wapbl_stop(struct wapbl *wl, int force)
681 {
682 	int error;
683 
684 	WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_stop called\n"));
685 	error = wapbl_flush(wl, 1);
686 	if (error) {
687 		if (force)
688 			wapbl_discard(wl);
689 		else
690 			return error;
691 	}
692 
693 	/* Unlinked inodes persist after a flush */
694 	if (wl->wl_inohashcnt) {
695 		if (force) {
696 			wapbl_discard(wl);
697 		} else {
698 			return EBUSY;
699 		}
700 	}
701 
702 	KASSERT(wl->wl_bufbytes == 0);
703 	KASSERT(wl->wl_bcount == 0);
704 	KASSERT(wl->wl_bufcount == 0);
705 	KASSERT(LIST_EMPTY(&wl->wl_bufs));
706 	KASSERT(wl->wl_dealloccnt == 0);
707 	KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries));
708 	KASSERT(wl->wl_inohashcnt == 0);
709 
710 	wapbl_free(wl->wl_wc_scratch, wl->wl_wc_header->wc_len);
711 	wapbl_free(wl->wl_wc_header, wl->wl_wc_header->wc_len);
712 	wapbl_free(wl->wl_deallocblks,
713 	    sizeof(*wl->wl_deallocblks) * wl->wl_dealloclim);
714 	wapbl_free(wl->wl_dealloclens,
715 	    sizeof(*wl->wl_dealloclens) * wl->wl_dealloclim);
716 	wapbl_free(wl->wl_buffer, MAXPHYS);
717 	wapbl_inodetrk_free(wl);
718 
719 	cv_destroy(&wl->wl_reclaimable_cv);
720 	mutex_destroy(&wl->wl_mtx);
721 	rw_destroy(&wl->wl_rwlock);
722 	wapbl_free(wl, sizeof(*wl));
723 
724 	return 0;
725 }
726 
727 static int
728 wapbl_doio(void *data, size_t len, struct vnode *devvp, daddr_t pbn, int flags)
729 {
730 	struct pstats *pstats = curlwp->l_proc->p_stats;
731 	struct buf *bp;
732 	int error;
733 
734 	KASSERT((flags & ~(B_WRITE | B_READ)) == 0);
735 	KASSERT(devvp->v_type == VBLK);
736 
737 	if ((flags & (B_WRITE | B_READ)) == B_WRITE) {
738 		mutex_enter(devvp->v_interlock);
739 		devvp->v_numoutput++;
740 		mutex_exit(devvp->v_interlock);
741 		pstats->p_ru.ru_oublock++;
742 	} else {
743 		pstats->p_ru.ru_inblock++;
744 	}
745 
746 	bp = getiobuf(devvp, true);
747 	bp->b_flags = flags;
748 	bp->b_cflags = BC_BUSY; /* silly & dubious */
749 	bp->b_dev = devvp->v_rdev;
750 	bp->b_data = data;
751 	bp->b_bufsize = bp->b_resid = bp->b_bcount = len;
752 	bp->b_blkno = pbn;
753 	BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
754 
755 	WAPBL_PRINTF(WAPBL_PRINT_IO,
756 	    ("wapbl_doio: %s %d bytes at block %"PRId64" on dev 0x%"PRIx64"\n",
757 	    BUF_ISWRITE(bp) ? "write" : "read", bp->b_bcount,
758 	    bp->b_blkno, bp->b_dev));
759 
760 	VOP_STRATEGY(devvp, bp);
761 
762 	error = biowait(bp);
763 	putiobuf(bp);
764 
765 	if (error) {
766 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
767 		    ("wapbl_doio: %s %zu bytes at block %" PRId64
768 		    " on dev 0x%"PRIx64" failed with error %d\n",
769 		    (((flags & (B_WRITE | B_READ)) == B_WRITE) ?
770 		     "write" : "read"),
771 		    len, pbn, devvp->v_rdev, error));
772 	}
773 
774 	return error;
775 }
776 
777 int
778 wapbl_write(void *data, size_t len, struct vnode *devvp, daddr_t pbn)
779 {
780 
781 	return wapbl_doio(data, len, devvp, pbn, B_WRITE);
782 }
783 
784 int
785 wapbl_read(void *data, size_t len, struct vnode *devvp, daddr_t pbn)
786 {
787 
788 	return wapbl_doio(data, len, devvp, pbn, B_READ);
789 }
790 
791 /*
792  * Flush buffered data if any.
793  */
794 static int
795 wapbl_buffered_flush(struct wapbl *wl)
796 {
797 	int error;
798 
799 	if (wl->wl_buffer_used == 0)
800 		return 0;
801 
802 	error = wapbl_doio(wl->wl_buffer, wl->wl_buffer_used,
803 	    wl->wl_devvp, wl->wl_buffer_dblk, B_WRITE);
804 	wl->wl_buffer_used = 0;
805 
806 	return error;
807 }
808 
809 /*
810  * Write data to the log.
811  * Try to coalesce writes and emit MAXPHYS aligned blocks.
812  */
813 static int
814 wapbl_buffered_write(void *data, size_t len, struct wapbl *wl, daddr_t pbn)
815 {
816 	int error;
817 	size_t resid;
818 
819 	/*
820 	 * If not adjacent to buffered data flush first.  Disk block
821 	 * address is always valid for non-empty buffer.
822 	 */
823 	if (wl->wl_buffer_used > 0 &&
824 	    pbn != wl->wl_buffer_dblk + btodb(wl->wl_buffer_used)) {
825 		error = wapbl_buffered_flush(wl);
826 		if (error)
827 			return error;
828 	}
829 	/*
830 	 * If this write goes to an empty buffer we have to
831 	 * save the disk block address first.
832 	 */
833 	if (wl->wl_buffer_used == 0)
834 		wl->wl_buffer_dblk = pbn;
835 	/*
836 	 * Remaining space so this buffer ends on a MAXPHYS boundary.
837 	 *
838 	 * Cannot become less or equal zero as the buffer would have been
839 	 * flushed on the last call then.
840 	 */
841 	resid = MAXPHYS - dbtob(wl->wl_buffer_dblk % btodb(MAXPHYS)) -
842 	    wl->wl_buffer_used;
843 	KASSERT(resid > 0);
844 	KASSERT(dbtob(btodb(resid)) == resid);
845 	if (len >= resid) {
846 		memcpy(wl->wl_buffer + wl->wl_buffer_used, data, resid);
847 		wl->wl_buffer_used += resid;
848 		error = wapbl_doio(wl->wl_buffer, wl->wl_buffer_used,
849 		    wl->wl_devvp, wl->wl_buffer_dblk, B_WRITE);
850 		data = (uint8_t *)data + resid;
851 		len -= resid;
852 		wl->wl_buffer_dblk = pbn + btodb(resid);
853 		wl->wl_buffer_used = 0;
854 		if (error)
855 			return error;
856 	}
857 	KASSERT(len < MAXPHYS);
858 	if (len > 0) {
859 		memcpy(wl->wl_buffer + wl->wl_buffer_used, data, len);
860 		wl->wl_buffer_used += len;
861 	}
862 
863 	return 0;
864 }
865 
866 /*
867  * Off is byte offset returns new offset for next write
868  * handles log wraparound
869  */
870 static int
871 wapbl_circ_write(struct wapbl *wl, void *data, size_t len, off_t *offp)
872 {
873 	size_t slen;
874 	off_t off = *offp;
875 	int error;
876 	daddr_t pbn;
877 
878 	KDASSERT(((len >> wl->wl_log_dev_bshift) <<
879 	    wl->wl_log_dev_bshift) == len);
880 
881 	if (off < wl->wl_circ_off)
882 		off = wl->wl_circ_off;
883 	slen = wl->wl_circ_off + wl->wl_circ_size - off;
884 	if (slen < len) {
885 		pbn = wl->wl_logpbn + (off >> wl->wl_log_dev_bshift);
886 #ifdef _KERNEL
887 		pbn = btodb(pbn << wl->wl_log_dev_bshift);
888 #endif
889 		error = wapbl_buffered_write(data, slen, wl, pbn);
890 		if (error)
891 			return error;
892 		data = (uint8_t *)data + slen;
893 		len -= slen;
894 		off = wl->wl_circ_off;
895 	}
896 	pbn = wl->wl_logpbn + (off >> wl->wl_log_dev_bshift);
897 #ifdef _KERNEL
898 	pbn = btodb(pbn << wl->wl_log_dev_bshift);
899 #endif
900 	error = wapbl_buffered_write(data, len, wl, pbn);
901 	if (error)
902 		return error;
903 	off += len;
904 	if (off >= wl->wl_circ_off + wl->wl_circ_size)
905 		off = wl->wl_circ_off;
906 	*offp = off;
907 	return 0;
908 }
909 
910 /****************************************************************/
911 
912 int
913 wapbl_begin(struct wapbl *wl, const char *file, int line)
914 {
915 	int doflush;
916 	unsigned lockcount;
917 
918 	KDASSERT(wl);
919 
920 	/*
921 	 * XXX this needs to be made much more sophisticated.
922 	 * perhaps each wapbl_begin could reserve a specified
923 	 * number of buffers and bytes.
924 	 */
925 	mutex_enter(&wl->wl_mtx);
926 	lockcount = wl->wl_lock_count;
927 	doflush = ((wl->wl_bufbytes + (lockcount * MAXPHYS)) >
928 		   wl->wl_bufbytes_max / 2) ||
929 		  ((wl->wl_bufcount + (lockcount * 10)) >
930 		   wl->wl_bufcount_max / 2) ||
931 		  (wapbl_transaction_len(wl) > wl->wl_circ_size / 2) ||
932 		  (wl->wl_dealloccnt >= (wl->wl_dealloclim / 2));
933 	mutex_exit(&wl->wl_mtx);
934 
935 	if (doflush) {
936 		WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
937 		    ("force flush lockcnt=%d bufbytes=%zu "
938 		    "(max=%zu) bufcount=%zu (max=%zu) "
939 		    "dealloccnt %d (lim=%d)\n",
940 		    lockcount, wl->wl_bufbytes,
941 		    wl->wl_bufbytes_max, wl->wl_bufcount,
942 		    wl->wl_bufcount_max,
943 		    wl->wl_dealloccnt, wl->wl_dealloclim));
944 	}
945 
946 	if (doflush) {
947 		int error = wapbl_flush(wl, 0);
948 		if (error)
949 			return error;
950 	}
951 
952 	rw_enter(&wl->wl_rwlock, RW_READER);
953 	mutex_enter(&wl->wl_mtx);
954 	wl->wl_lock_count++;
955 	mutex_exit(&wl->wl_mtx);
956 
957 #if defined(WAPBL_DEBUG_PRINT)
958 	WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION,
959 	    ("wapbl_begin thread %d.%d with bufcount=%zu "
960 	    "bufbytes=%zu bcount=%zu at %s:%d\n",
961 	    curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
962 	    wl->wl_bufbytes, wl->wl_bcount, file, line));
963 #endif
964 
965 	return 0;
966 }
967 
968 void
969 wapbl_end(struct wapbl *wl)
970 {
971 
972 #if defined(WAPBL_DEBUG_PRINT)
973 	WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION,
974 	     ("wapbl_end thread %d.%d with bufcount=%zu "
975 	      "bufbytes=%zu bcount=%zu\n",
976 	      curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
977 	      wl->wl_bufbytes, wl->wl_bcount));
978 #endif
979 
980 #ifdef DIAGNOSTIC
981 	size_t flushsize = wapbl_transaction_len(wl);
982 	if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) {
983 		/*
984 		 * XXX this could be handled more gracefully, perhaps place
985 		 * only a partial transaction in the log and allow the
986 		 * remaining to flush without the protection of the journal.
987 		 */
988 		panic("wapbl_end: current transaction too big to flush\n");
989 	}
990 #endif
991 
992 	mutex_enter(&wl->wl_mtx);
993 	KASSERT(wl->wl_lock_count > 0);
994 	wl->wl_lock_count--;
995 	mutex_exit(&wl->wl_mtx);
996 
997 	rw_exit(&wl->wl_rwlock);
998 }
999 
1000 void
1001 wapbl_add_buf(struct wapbl *wl, struct buf * bp)
1002 {
1003 
1004 	KASSERT(bp->b_cflags & BC_BUSY);
1005 	KASSERT(bp->b_vp);
1006 
1007 	wapbl_jlock_assert(wl);
1008 
1009 #if 0
1010 	/*
1011 	 * XXX this might be an issue for swapfiles.
1012 	 * see uvm_swap.c:1702
1013 	 *
1014 	 * XXX2 why require it then?  leap of semantics?
1015 	 */
1016 	KASSERT((bp->b_cflags & BC_NOCACHE) == 0);
1017 #endif
1018 
1019 	mutex_enter(&wl->wl_mtx);
1020 	if (bp->b_flags & B_LOCKED) {
1021 		LIST_REMOVE(bp, b_wapbllist);
1022 		WAPBL_PRINTF(WAPBL_PRINT_BUFFER2,
1023 		   ("wapbl_add_buf thread %d.%d re-adding buf %p "
1024 		    "with %d bytes %d bcount\n",
1025 		    curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize,
1026 		    bp->b_bcount));
1027 	} else {
1028 		/* unlocked by dirty buffers shouldn't exist */
1029 		KASSERT(!(bp->b_oflags & BO_DELWRI));
1030 		wl->wl_bufbytes += bp->b_bufsize;
1031 		wl->wl_bcount += bp->b_bcount;
1032 		wl->wl_bufcount++;
1033 		WAPBL_PRINTF(WAPBL_PRINT_BUFFER,
1034 		   ("wapbl_add_buf thread %d.%d adding buf %p "
1035 		    "with %d bytes %d bcount\n",
1036 		    curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize,
1037 		    bp->b_bcount));
1038 	}
1039 	LIST_INSERT_HEAD(&wl->wl_bufs, bp, b_wapbllist);
1040 	mutex_exit(&wl->wl_mtx);
1041 
1042 	bp->b_flags |= B_LOCKED;
1043 }
1044 
1045 static void
1046 wapbl_remove_buf_locked(struct wapbl * wl, struct buf *bp)
1047 {
1048 
1049 	KASSERT(mutex_owned(&wl->wl_mtx));
1050 	KASSERT(bp->b_cflags & BC_BUSY);
1051 	wapbl_jlock_assert(wl);
1052 
1053 #if 0
1054 	/*
1055 	 * XXX this might be an issue for swapfiles.
1056 	 * see uvm_swap.c:1725
1057 	 *
1058 	 * XXXdeux: see above
1059 	 */
1060 	KASSERT((bp->b_flags & BC_NOCACHE) == 0);
1061 #endif
1062 	KASSERT(bp->b_flags & B_LOCKED);
1063 
1064 	WAPBL_PRINTF(WAPBL_PRINT_BUFFER,
1065 	   ("wapbl_remove_buf thread %d.%d removing buf %p with "
1066 	    "%d bytes %d bcount\n",
1067 	    curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, bp->b_bcount));
1068 
1069 	KASSERT(wl->wl_bufbytes >= bp->b_bufsize);
1070 	wl->wl_bufbytes -= bp->b_bufsize;
1071 	KASSERT(wl->wl_bcount >= bp->b_bcount);
1072 	wl->wl_bcount -= bp->b_bcount;
1073 	KASSERT(wl->wl_bufcount > 0);
1074 	wl->wl_bufcount--;
1075 	KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0));
1076 	KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0));
1077 	LIST_REMOVE(bp, b_wapbllist);
1078 
1079 	bp->b_flags &= ~B_LOCKED;
1080 }
1081 
1082 /* called from brelsel() in vfs_bio among other places */
1083 void
1084 wapbl_remove_buf(struct wapbl * wl, struct buf *bp)
1085 {
1086 
1087 	mutex_enter(&wl->wl_mtx);
1088 	wapbl_remove_buf_locked(wl, bp);
1089 	mutex_exit(&wl->wl_mtx);
1090 }
1091 
1092 void
1093 wapbl_resize_buf(struct wapbl *wl, struct buf *bp, long oldsz, long oldcnt)
1094 {
1095 
1096 	KASSERT(bp->b_cflags & BC_BUSY);
1097 
1098 	/*
1099 	 * XXX: why does this depend on B_LOCKED?  otherwise the buf
1100 	 * is not for a transaction?  if so, why is this called in the
1101 	 * first place?
1102 	 */
1103 	if (bp->b_flags & B_LOCKED) {
1104 		mutex_enter(&wl->wl_mtx);
1105 		wl->wl_bufbytes += bp->b_bufsize - oldsz;
1106 		wl->wl_bcount += bp->b_bcount - oldcnt;
1107 		mutex_exit(&wl->wl_mtx);
1108 	}
1109 }
1110 
1111 #endif /* _KERNEL */
1112 
1113 /****************************************************************/
1114 /* Some utility inlines */
1115 
1116 static inline size_t
1117 wapbl_space_used(size_t avail, off_t head, off_t tail)
1118 {
1119 
1120 	if (tail == 0) {
1121 		KASSERT(head == 0);
1122 		return 0;
1123 	}
1124 	return ((head + (avail - 1) - tail) % avail) + 1;
1125 }
1126 
1127 #ifdef _KERNEL
1128 /* This is used to advance the pointer at old to new value at old+delta */
1129 static inline off_t
1130 wapbl_advance(size_t size, size_t off, off_t old, size_t delta)
1131 {
1132 	off_t new;
1133 
1134 	/* Define acceptable ranges for inputs. */
1135 	KASSERT(delta <= (size_t)size);
1136 	KASSERT((old == 0) || ((size_t)old >= off));
1137 	KASSERT(old < (off_t)(size + off));
1138 
1139 	if ((old == 0) && (delta != 0))
1140 		new = off + delta;
1141 	else if ((old + delta) < (size + off))
1142 		new = old + delta;
1143 	else
1144 		new = (old + delta) - size;
1145 
1146 	/* Note some interesting axioms */
1147 	KASSERT((delta != 0) || (new == old));
1148 	KASSERT((delta == 0) || (new != 0));
1149 	KASSERT((delta != (size)) || (new == old));
1150 
1151 	/* Define acceptable ranges for output. */
1152 	KASSERT((new == 0) || ((size_t)new >= off));
1153 	KASSERT((size_t)new < (size + off));
1154 	return new;
1155 }
1156 
1157 static inline size_t
1158 wapbl_space_free(size_t avail, off_t head, off_t tail)
1159 {
1160 
1161 	return avail - wapbl_space_used(avail, head, tail);
1162 }
1163 
1164 static inline void
1165 wapbl_advance_head(size_t size, size_t off, size_t delta, off_t *headp,
1166 		   off_t *tailp)
1167 {
1168 	off_t head = *headp;
1169 	off_t tail = *tailp;
1170 
1171 	KASSERT(delta <= wapbl_space_free(size, head, tail));
1172 	head = wapbl_advance(size, off, head, delta);
1173 	if ((tail == 0) && (head != 0))
1174 		tail = off;
1175 	*headp = head;
1176 	*tailp = tail;
1177 }
1178 
1179 static inline void
1180 wapbl_advance_tail(size_t size, size_t off, size_t delta, off_t *headp,
1181 		   off_t *tailp)
1182 {
1183 	off_t head = *headp;
1184 	off_t tail = *tailp;
1185 
1186 	KASSERT(delta <= wapbl_space_used(size, head, tail));
1187 	tail = wapbl_advance(size, off, tail, delta);
1188 	if (head == tail) {
1189 		head = tail = 0;
1190 	}
1191 	*headp = head;
1192 	*tailp = tail;
1193 }
1194 
1195 
1196 /****************************************************************/
1197 
1198 /*
1199  * Remove transactions whose buffers are completely flushed to disk.
1200  * Will block until at least minfree space is available.
1201  * only intended to be called from inside wapbl_flush and therefore
1202  * does not protect against commit races with itself or with flush.
1203  */
1204 static int
1205 wapbl_truncate(struct wapbl *wl, size_t minfree, int waitonly)
1206 {
1207 	size_t delta;
1208 	size_t avail;
1209 	off_t head;
1210 	off_t tail;
1211 	int error = 0;
1212 
1213 	KASSERT(minfree <= (wl->wl_circ_size - wl->wl_reserved_bytes));
1214 	KASSERT(rw_write_held(&wl->wl_rwlock));
1215 
1216 	mutex_enter(&wl->wl_mtx);
1217 
1218 	/*
1219 	 * First check to see if we have to do a commit
1220 	 * at all.
1221 	 */
1222 	avail = wapbl_space_free(wl->wl_circ_size, wl->wl_head, wl->wl_tail);
1223 	if (minfree < avail) {
1224 		mutex_exit(&wl->wl_mtx);
1225 		return 0;
1226 	}
1227 	minfree -= avail;
1228 	while ((wl->wl_error_count == 0) &&
1229 	    (wl->wl_reclaimable_bytes < minfree)) {
1230         	WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE,
1231                    ("wapbl_truncate: sleeping on %p wl=%p bytes=%zd "
1232 		    "minfree=%zd\n",
1233                     &wl->wl_reclaimable_bytes, wl, wl->wl_reclaimable_bytes,
1234 		    minfree));
1235 
1236 		cv_wait(&wl->wl_reclaimable_cv, &wl->wl_mtx);
1237 	}
1238 	if (wl->wl_reclaimable_bytes < minfree) {
1239 		KASSERT(wl->wl_error_count);
1240 		/* XXX maybe get actual error from buffer instead someday? */
1241 		error = EIO;
1242 	}
1243 	head = wl->wl_head;
1244 	tail = wl->wl_tail;
1245 	delta = wl->wl_reclaimable_bytes;
1246 
1247 	/* If all of of the entries are flushed, then be sure to keep
1248 	 * the reserved bytes reserved.  Watch out for discarded transactions,
1249 	 * which could leave more bytes reserved than are reclaimable.
1250 	 */
1251 	if (SIMPLEQ_EMPTY(&wl->wl_entries) &&
1252 	    (delta >= wl->wl_reserved_bytes)) {
1253 		delta -= wl->wl_reserved_bytes;
1254 	}
1255 	wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, &head,
1256 			   &tail);
1257 	KDASSERT(wl->wl_reserved_bytes <=
1258 		wapbl_space_used(wl->wl_circ_size, head, tail));
1259 	mutex_exit(&wl->wl_mtx);
1260 
1261 	if (error)
1262 		return error;
1263 
1264 	if (waitonly)
1265 		return 0;
1266 
1267 	/*
1268 	 * This is where head, tail and delta are unprotected
1269 	 * from races against itself or flush.  This is ok since
1270 	 * we only call this routine from inside flush itself.
1271 	 *
1272 	 * XXX: how can it race against itself when accessed only
1273 	 * from behind the write-locked rwlock?
1274 	 */
1275 	error = wapbl_write_commit(wl, head, tail);
1276 	if (error)
1277 		return error;
1278 
1279 	wl->wl_head = head;
1280 	wl->wl_tail = tail;
1281 
1282 	mutex_enter(&wl->wl_mtx);
1283 	KASSERT(wl->wl_reclaimable_bytes >= delta);
1284 	wl->wl_reclaimable_bytes -= delta;
1285 	mutex_exit(&wl->wl_mtx);
1286 	WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE,
1287 	    ("wapbl_truncate thread %d.%d truncating %zu bytes\n",
1288 	    curproc->p_pid, curlwp->l_lid, delta));
1289 
1290 	return 0;
1291 }
1292 
1293 /****************************************************************/
1294 
1295 void
1296 wapbl_biodone(struct buf *bp)
1297 {
1298 	struct wapbl_entry *we = bp->b_private;
1299 	struct wapbl *wl = we->we_wapbl;
1300 #ifdef WAPBL_DEBUG_BUFBYTES
1301 	const int bufsize = bp->b_bufsize;
1302 #endif
1303 
1304 	/*
1305 	 * Handle possible flushing of buffers after log has been
1306 	 * decomissioned.
1307 	 */
1308 	if (!wl) {
1309 		KASSERT(we->we_bufcount > 0);
1310 		we->we_bufcount--;
1311 #ifdef WAPBL_DEBUG_BUFBYTES
1312 		KASSERT(we->we_unsynced_bufbytes >= bufsize);
1313 		we->we_unsynced_bufbytes -= bufsize;
1314 #endif
1315 
1316 		if (we->we_bufcount == 0) {
1317 #ifdef WAPBL_DEBUG_BUFBYTES
1318 			KASSERT(we->we_unsynced_bufbytes == 0);
1319 #endif
1320 			pool_put(&wapbl_entry_pool, we);
1321 		}
1322 
1323 		brelse(bp, 0);
1324 		return;
1325 	}
1326 
1327 #ifdef ohbother
1328 	KDASSERT(bp->b_oflags & BO_DONE);
1329 	KDASSERT(!(bp->b_oflags & BO_DELWRI));
1330 	KDASSERT(bp->b_flags & B_ASYNC);
1331 	KDASSERT(bp->b_cflags & BC_BUSY);
1332 	KDASSERT(!(bp->b_flags & B_LOCKED));
1333 	KDASSERT(!(bp->b_flags & B_READ));
1334 	KDASSERT(!(bp->b_cflags & BC_INVAL));
1335 	KDASSERT(!(bp->b_cflags & BC_NOCACHE));
1336 #endif
1337 
1338 	if (bp->b_error) {
1339 #ifdef notyet /* Can't currently handle possible dirty buffer reuse */
1340 		/*
1341 		 * XXXpooka: interfaces not fully updated
1342 		 * Note: this was not enabled in the original patch
1343 		 * against netbsd4 either.  I don't know if comment
1344 		 * above is true or not.
1345 		 */
1346 
1347 		/*
1348 		 * If an error occurs, report the error and leave the
1349 		 * buffer as a delayed write on the LRU queue.
1350 		 * restarting the write would likely result in
1351 		 * an error spinloop, so let it be done harmlessly
1352 		 * by the syncer.
1353 		 */
1354 		bp->b_flags &= ~(B_DONE);
1355 		simple_unlock(&bp->b_interlock);
1356 
1357 		if (we->we_error == 0) {
1358 			mutex_enter(&wl->wl_mtx);
1359 			wl->wl_error_count++;
1360 			mutex_exit(&wl->wl_mtx);
1361 			cv_broadcast(&wl->wl_reclaimable_cv);
1362 		}
1363 		we->we_error = bp->b_error;
1364 		bp->b_error = 0;
1365 		brelse(bp);
1366 		return;
1367 #else
1368 		/* For now, just mark the log permanently errored out */
1369 
1370 		mutex_enter(&wl->wl_mtx);
1371 		if (wl->wl_error_count == 0) {
1372 			wl->wl_error_count++;
1373 			cv_broadcast(&wl->wl_reclaimable_cv);
1374 		}
1375 		mutex_exit(&wl->wl_mtx);
1376 #endif
1377 	}
1378 
1379 	/*
1380 	 * Release the buffer here. wapbl_flush() may wait for the
1381 	 * log to become empty and we better unbusy the buffer before
1382 	 * wapbl_flush() returns.
1383 	 */
1384 	brelse(bp, 0);
1385 
1386 	mutex_enter(&wl->wl_mtx);
1387 
1388 	KASSERT(we->we_bufcount > 0);
1389 	we->we_bufcount--;
1390 #ifdef WAPBL_DEBUG_BUFBYTES
1391 	KASSERT(we->we_unsynced_bufbytes >= bufsize);
1392 	we->we_unsynced_bufbytes -= bufsize;
1393 	KASSERT(wl->wl_unsynced_bufbytes >= bufsize);
1394 	wl->wl_unsynced_bufbytes -= bufsize;
1395 #endif
1396 
1397 	/*
1398 	 * If the current transaction can be reclaimed, start
1399 	 * at the beginning and reclaim any consecutive reclaimable
1400 	 * transactions.  If we successfully reclaim anything,
1401 	 * then wakeup anyone waiting for the reclaim.
1402 	 */
1403 	if (we->we_bufcount == 0) {
1404 		size_t delta = 0;
1405 		int errcnt = 0;
1406 #ifdef WAPBL_DEBUG_BUFBYTES
1407 		KDASSERT(we->we_unsynced_bufbytes == 0);
1408 #endif
1409 		/*
1410 		 * clear any posted error, since the buffer it came from
1411 		 * has successfully flushed by now
1412 		 */
1413 		while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) &&
1414 		       (we->we_bufcount == 0)) {
1415 			delta += we->we_reclaimable_bytes;
1416 			if (we->we_error)
1417 				errcnt++;
1418 			SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries);
1419 			pool_put(&wapbl_entry_pool, we);
1420 		}
1421 
1422 		if (delta) {
1423 			wl->wl_reclaimable_bytes += delta;
1424 			KASSERT(wl->wl_error_count >= errcnt);
1425 			wl->wl_error_count -= errcnt;
1426 			cv_broadcast(&wl->wl_reclaimable_cv);
1427 		}
1428 	}
1429 
1430 	mutex_exit(&wl->wl_mtx);
1431 }
1432 
1433 /*
1434  * Write transactions to disk + start I/O for contents
1435  */
1436 int
1437 wapbl_flush(struct wapbl *wl, int waitfor)
1438 {
1439 	struct buf *bp;
1440 	struct wapbl_entry *we;
1441 	off_t off;
1442 	off_t head;
1443 	off_t tail;
1444 	size_t delta = 0;
1445 	size_t flushsize;
1446 	size_t reserved;
1447 	int error = 0;
1448 
1449 	/*
1450 	 * Do a quick check to see if a full flush can be skipped
1451 	 * This assumes that the flush callback does not need to be called
1452 	 * unless there are other outstanding bufs.
1453 	 */
1454 	if (!waitfor) {
1455 		size_t nbufs;
1456 		mutex_enter(&wl->wl_mtx);	/* XXX need mutex here to
1457 						   protect the KASSERTS */
1458 		nbufs = wl->wl_bufcount;
1459 		KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0));
1460 		KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0));
1461 		mutex_exit(&wl->wl_mtx);
1462 		if (nbufs == 0)
1463 			return 0;
1464 	}
1465 
1466 	/*
1467 	 * XXX we may consider using LK_UPGRADE here
1468 	 * if we want to call flush from inside a transaction
1469 	 */
1470 	rw_enter(&wl->wl_rwlock, RW_WRITER);
1471 	wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens,
1472 	    wl->wl_dealloccnt);
1473 
1474 	/*
1475 	 * Now that we are fully locked and flushed,
1476 	 * do another check for nothing to do.
1477 	 */
1478 	if (wl->wl_bufcount == 0) {
1479 		goto out;
1480 	}
1481 
1482 #if 0
1483 	WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1484 		     ("wapbl_flush thread %d.%d flushing entries with "
1485 		      "bufcount=%zu bufbytes=%zu\n",
1486 		      curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
1487 		      wl->wl_bufbytes));
1488 #endif
1489 
1490 	/* Calculate amount of space needed to flush */
1491 	flushsize = wapbl_transaction_len(wl);
1492 	if (wapbl_verbose_commit) {
1493 		struct timespec ts;
1494 		getnanotime(&ts);
1495 		printf("%s: %lld.%09ld this transaction = %zu bytes\n",
1496 		    __func__, (long long)ts.tv_sec,
1497 		    (long)ts.tv_nsec, flushsize);
1498 	}
1499 
1500 	if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) {
1501 		/*
1502 		 * XXX this could be handled more gracefully, perhaps place
1503 		 * only a partial transaction in the log and allow the
1504 		 * remaining to flush without the protection of the journal.
1505 		 */
1506 		panic("wapbl_flush: current transaction too big to flush\n");
1507 	}
1508 
1509 	error = wapbl_truncate(wl, flushsize, 0);
1510 	if (error)
1511 		goto out2;
1512 
1513 	off = wl->wl_head;
1514 	KASSERT((off == 0) || ((off >= wl->wl_circ_off) &&
1515 	                      (off < wl->wl_circ_off + wl->wl_circ_size)));
1516 	error = wapbl_write_blocks(wl, &off);
1517 	if (error)
1518 		goto out2;
1519 	error = wapbl_write_revocations(wl, &off);
1520 	if (error)
1521 		goto out2;
1522 	error = wapbl_write_inodes(wl, &off);
1523 	if (error)
1524 		goto out2;
1525 
1526 	reserved = 0;
1527 	if (wl->wl_inohashcnt)
1528 		reserved = wapbl_transaction_inodes_len(wl);
1529 
1530 	head = wl->wl_head;
1531 	tail = wl->wl_tail;
1532 
1533 	wapbl_advance_head(wl->wl_circ_size, wl->wl_circ_off, flushsize,
1534 	    &head, &tail);
1535 #ifdef WAPBL_DEBUG
1536 	if (head != off) {
1537 		panic("lost head! head=%"PRIdMAX" tail=%" PRIdMAX
1538 		      " off=%"PRIdMAX" flush=%zu\n",
1539 		      (intmax_t)head, (intmax_t)tail, (intmax_t)off,
1540 		      flushsize);
1541 	}
1542 #else
1543 	KASSERT(head == off);
1544 #endif
1545 
1546 	/* Opportunistically move the tail forward if we can */
1547 	if (!wapbl_lazy_truncate) {
1548 		mutex_enter(&wl->wl_mtx);
1549 		delta = wl->wl_reclaimable_bytes;
1550 		mutex_exit(&wl->wl_mtx);
1551 		wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta,
1552 		    &head, &tail);
1553 	}
1554 
1555 	error = wapbl_write_commit(wl, head, tail);
1556 	if (error)
1557 		goto out2;
1558 
1559 	we = pool_get(&wapbl_entry_pool, PR_WAITOK);
1560 
1561 #ifdef WAPBL_DEBUG_BUFBYTES
1562 	WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1563 		("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu"
1564 		 " unsynced=%zu"
1565 		 "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d "
1566 		 "inodes=%d\n",
1567 		 curproc->p_pid, curlwp->l_lid, flushsize, delta,
1568 		 wapbl_space_used(wl->wl_circ_size, head, tail),
1569 		 wl->wl_unsynced_bufbytes, wl->wl_bufcount,
1570 		 wl->wl_bufbytes, wl->wl_bcount, wl->wl_dealloccnt,
1571 		 wl->wl_inohashcnt));
1572 #else
1573 	WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1574 		("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu"
1575 		 "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d "
1576 		 "inodes=%d\n",
1577 		 curproc->p_pid, curlwp->l_lid, flushsize, delta,
1578 		 wapbl_space_used(wl->wl_circ_size, head, tail),
1579 		 wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount,
1580 		 wl->wl_dealloccnt, wl->wl_inohashcnt));
1581 #endif
1582 
1583 
1584 	mutex_enter(&bufcache_lock);
1585 	mutex_enter(&wl->wl_mtx);
1586 
1587 	wl->wl_reserved_bytes = reserved;
1588 	wl->wl_head = head;
1589 	wl->wl_tail = tail;
1590 	KASSERT(wl->wl_reclaimable_bytes >= delta);
1591 	wl->wl_reclaimable_bytes -= delta;
1592 	wl->wl_dealloccnt = 0;
1593 #ifdef WAPBL_DEBUG_BUFBYTES
1594 	wl->wl_unsynced_bufbytes += wl->wl_bufbytes;
1595 #endif
1596 
1597 	we->we_wapbl = wl;
1598 	we->we_bufcount = wl->wl_bufcount;
1599 #ifdef WAPBL_DEBUG_BUFBYTES
1600 	we->we_unsynced_bufbytes = wl->wl_bufbytes;
1601 #endif
1602 	we->we_reclaimable_bytes = flushsize;
1603 	we->we_error = 0;
1604 	SIMPLEQ_INSERT_TAIL(&wl->wl_entries, we, we_entries);
1605 
1606 	/*
1607 	 * this flushes bufs in reverse order than they were queued
1608 	 * it shouldn't matter, but if we care we could use TAILQ instead.
1609 	 * XXX Note they will get put on the lru queue when they flush
1610 	 * so we might actually want to change this to preserve order.
1611 	 */
1612 	while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) {
1613 		if (bbusy(bp, 0, 0, &wl->wl_mtx)) {
1614 			continue;
1615 		}
1616 		bp->b_iodone = wapbl_biodone;
1617 		bp->b_private = we;
1618 		bremfree(bp);
1619 		wapbl_remove_buf_locked(wl, bp);
1620 		mutex_exit(&wl->wl_mtx);
1621 		mutex_exit(&bufcache_lock);
1622 		bawrite(bp);
1623 		mutex_enter(&bufcache_lock);
1624 		mutex_enter(&wl->wl_mtx);
1625 	}
1626 	mutex_exit(&wl->wl_mtx);
1627 	mutex_exit(&bufcache_lock);
1628 
1629 #if 0
1630 	WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1631 		     ("wapbl_flush thread %d.%d done flushing entries...\n",
1632 		     curproc->p_pid, curlwp->l_lid));
1633 #endif
1634 
1635  out:
1636 
1637 	/*
1638 	 * If the waitfor flag is set, don't return until everything is
1639 	 * fully flushed and the on disk log is empty.
1640 	 */
1641 	if (waitfor) {
1642 		error = wapbl_truncate(wl, wl->wl_circ_size -
1643 			wl->wl_reserved_bytes, wapbl_lazy_truncate);
1644 	}
1645 
1646  out2:
1647 	if (error) {
1648 		wl->wl_flush_abort(wl->wl_mount, wl->wl_deallocblks,
1649 		    wl->wl_dealloclens, wl->wl_dealloccnt);
1650 	}
1651 
1652 #ifdef WAPBL_DEBUG_PRINT
1653 	if (error) {
1654 		pid_t pid = -1;
1655 		lwpid_t lid = -1;
1656 		if (curproc)
1657 			pid = curproc->p_pid;
1658 		if (curlwp)
1659 			lid = curlwp->l_lid;
1660 		mutex_enter(&wl->wl_mtx);
1661 #ifdef WAPBL_DEBUG_BUFBYTES
1662 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1663 		    ("wapbl_flush: thread %d.%d aborted flush: "
1664 		    "error = %d\n"
1665 		    "\tbufcount=%zu bufbytes=%zu bcount=%zu "
1666 		    "deallocs=%d inodes=%d\n"
1667 		    "\terrcnt = %d, reclaimable=%zu reserved=%zu "
1668 		    "unsynced=%zu\n",
1669 		    pid, lid, error, wl->wl_bufcount,
1670 		    wl->wl_bufbytes, wl->wl_bcount,
1671 		    wl->wl_dealloccnt, wl->wl_inohashcnt,
1672 		    wl->wl_error_count, wl->wl_reclaimable_bytes,
1673 		    wl->wl_reserved_bytes, wl->wl_unsynced_bufbytes));
1674 		SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1675 			WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1676 			    ("\tentry: bufcount = %zu, reclaimable = %zu, "
1677 			     "error = %d, unsynced = %zu\n",
1678 			     we->we_bufcount, we->we_reclaimable_bytes,
1679 			     we->we_error, we->we_unsynced_bufbytes));
1680 		}
1681 #else
1682 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1683 		    ("wapbl_flush: thread %d.%d aborted flush: "
1684 		     "error = %d\n"
1685 		     "\tbufcount=%zu bufbytes=%zu bcount=%zu "
1686 		     "deallocs=%d inodes=%d\n"
1687 		     "\terrcnt = %d, reclaimable=%zu reserved=%zu\n",
1688 		     pid, lid, error, wl->wl_bufcount,
1689 		     wl->wl_bufbytes, wl->wl_bcount,
1690 		     wl->wl_dealloccnt, wl->wl_inohashcnt,
1691 		     wl->wl_error_count, wl->wl_reclaimable_bytes,
1692 		     wl->wl_reserved_bytes));
1693 		SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1694 			WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1695 			    ("\tentry: bufcount = %zu, reclaimable = %zu, "
1696 			     "error = %d\n", we->we_bufcount,
1697 			     we->we_reclaimable_bytes, we->we_error));
1698 		}
1699 #endif
1700 		mutex_exit(&wl->wl_mtx);
1701 	}
1702 #endif
1703 
1704 	rw_exit(&wl->wl_rwlock);
1705 	return error;
1706 }
1707 
1708 /****************************************************************/
1709 
1710 void
1711 wapbl_jlock_assert(struct wapbl *wl)
1712 {
1713 
1714 	KASSERT(rw_lock_held(&wl->wl_rwlock));
1715 }
1716 
1717 void
1718 wapbl_junlock_assert(struct wapbl *wl)
1719 {
1720 
1721 	KASSERT(!rw_write_held(&wl->wl_rwlock));
1722 }
1723 
1724 /****************************************************************/
1725 
1726 /* locks missing */
1727 void
1728 wapbl_print(struct wapbl *wl,
1729 		int full,
1730 		void (*pr)(const char *, ...))
1731 {
1732 	struct buf *bp;
1733 	struct wapbl_entry *we;
1734 	(*pr)("wapbl %p", wl);
1735 	(*pr)("\nlogvp = %p, devvp = %p, logpbn = %"PRId64"\n",
1736 	      wl->wl_logvp, wl->wl_devvp, wl->wl_logpbn);
1737 	(*pr)("circ = %zu, header = %zu, head = %"PRIdMAX" tail = %"PRIdMAX"\n",
1738 	      wl->wl_circ_size, wl->wl_circ_off,
1739 	      (intmax_t)wl->wl_head, (intmax_t)wl->wl_tail);
1740 	(*pr)("fs_dev_bshift = %d, log_dev_bshift = %d\n",
1741 	      wl->wl_log_dev_bshift, wl->wl_fs_dev_bshift);
1742 #ifdef WAPBL_DEBUG_BUFBYTES
1743 	(*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu "
1744 	      "reserved = %zu errcnt = %d unsynced = %zu\n",
1745 	      wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount,
1746 	      wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
1747 				wl->wl_error_count, wl->wl_unsynced_bufbytes);
1748 #else
1749 	(*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu "
1750 	      "reserved = %zu errcnt = %d\n", wl->wl_bufcount, wl->wl_bufbytes,
1751 	      wl->wl_bcount, wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
1752 				wl->wl_error_count);
1753 #endif
1754 	(*pr)("\tdealloccnt = %d, dealloclim = %d\n",
1755 	      wl->wl_dealloccnt, wl->wl_dealloclim);
1756 	(*pr)("\tinohashcnt = %d, inohashmask = 0x%08x\n",
1757 	      wl->wl_inohashcnt, wl->wl_inohashmask);
1758 	(*pr)("entries:\n");
1759 	SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1760 #ifdef WAPBL_DEBUG_BUFBYTES
1761 		(*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d, "
1762 		      "unsynced = %zu\n",
1763 		      we->we_bufcount, we->we_reclaimable_bytes,
1764 		      we->we_error, we->we_unsynced_bufbytes);
1765 #else
1766 		(*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d\n",
1767 		      we->we_bufcount, we->we_reclaimable_bytes, we->we_error);
1768 #endif
1769 	}
1770 	if (full) {
1771 		int cnt = 0;
1772 		(*pr)("bufs =");
1773 		LIST_FOREACH(bp, &wl->wl_bufs, b_wapbllist) {
1774 			if (!LIST_NEXT(bp, b_wapbllist)) {
1775 				(*pr)(" %p", bp);
1776 			} else if ((++cnt % 6) == 0) {
1777 				(*pr)(" %p,\n\t", bp);
1778 			} else {
1779 				(*pr)(" %p,", bp);
1780 			}
1781 		}
1782 		(*pr)("\n");
1783 
1784 		(*pr)("dealloced blks = ");
1785 		{
1786 			int i;
1787 			cnt = 0;
1788 			for (i = 0; i < wl->wl_dealloccnt; i++) {
1789 				(*pr)(" %"PRId64":%d,",
1790 				      wl->wl_deallocblks[i],
1791 				      wl->wl_dealloclens[i]);
1792 				if ((++cnt % 4) == 0) {
1793 					(*pr)("\n\t");
1794 				}
1795 			}
1796 		}
1797 		(*pr)("\n");
1798 
1799 		(*pr)("registered inodes = ");
1800 		{
1801 			int i;
1802 			cnt = 0;
1803 			for (i = 0; i <= wl->wl_inohashmask; i++) {
1804 				struct wapbl_ino_head *wih;
1805 				struct wapbl_ino *wi;
1806 
1807 				wih = &wl->wl_inohash[i];
1808 				LIST_FOREACH(wi, wih, wi_hash) {
1809 					if (wi->wi_ino == 0)
1810 						continue;
1811 					(*pr)(" %"PRIu64"/0%06"PRIo32",",
1812 					    wi->wi_ino, wi->wi_mode);
1813 					if ((++cnt % 4) == 0) {
1814 						(*pr)("\n\t");
1815 					}
1816 				}
1817 			}
1818 			(*pr)("\n");
1819 		}
1820 	}
1821 }
1822 
1823 #if defined(WAPBL_DEBUG) || defined(DDB)
1824 void
1825 wapbl_dump(struct wapbl *wl)
1826 {
1827 #if defined(WAPBL_DEBUG)
1828 	if (!wl)
1829 		wl = wapbl_debug_wl;
1830 #endif
1831 	if (!wl)
1832 		return;
1833 	wapbl_print(wl, 1, printf);
1834 }
1835 #endif
1836 
1837 /****************************************************************/
1838 
1839 void
1840 wapbl_register_deallocation(struct wapbl *wl, daddr_t blk, int len)
1841 {
1842 
1843 	wapbl_jlock_assert(wl);
1844 
1845 	mutex_enter(&wl->wl_mtx);
1846 	/* XXX should eventually instead tie this into resource estimation */
1847 	/*
1848 	 * XXX this panic needs locking/mutex analysis and the
1849 	 * ability to cope with the failure.
1850 	 */
1851 	/* XXX this XXX doesn't have enough XXX */
1852 	if (__predict_false(wl->wl_dealloccnt >= wl->wl_dealloclim))
1853 		panic("wapbl_register_deallocation: out of resources");
1854 
1855 	wl->wl_deallocblks[wl->wl_dealloccnt] = blk;
1856 	wl->wl_dealloclens[wl->wl_dealloccnt] = len;
1857 	wl->wl_dealloccnt++;
1858 	WAPBL_PRINTF(WAPBL_PRINT_ALLOC,
1859 	    ("wapbl_register_deallocation: blk=%"PRId64" len=%d\n", blk, len));
1860 	mutex_exit(&wl->wl_mtx);
1861 }
1862 
1863 /****************************************************************/
1864 
1865 static void
1866 wapbl_inodetrk_init(struct wapbl *wl, u_int size)
1867 {
1868 
1869 	wl->wl_inohash = hashinit(size, HASH_LIST, true, &wl->wl_inohashmask);
1870 	if (atomic_inc_uint_nv(&wapbl_ino_pool_refcount) == 1) {
1871 		pool_init(&wapbl_ino_pool, sizeof(struct wapbl_ino), 0, 0, 0,
1872 		    "wapblinopl", &pool_allocator_nointr, IPL_NONE);
1873 	}
1874 }
1875 
1876 static void
1877 wapbl_inodetrk_free(struct wapbl *wl)
1878 {
1879 
1880 	/* XXX this KASSERT needs locking/mutex analysis */
1881 	KASSERT(wl->wl_inohashcnt == 0);
1882 	hashdone(wl->wl_inohash, HASH_LIST, wl->wl_inohashmask);
1883 	if (atomic_dec_uint_nv(&wapbl_ino_pool_refcount) == 0) {
1884 		pool_destroy(&wapbl_ino_pool);
1885 	}
1886 }
1887 
1888 static struct wapbl_ino *
1889 wapbl_inodetrk_get(struct wapbl *wl, ino_t ino)
1890 {
1891 	struct wapbl_ino_head *wih;
1892 	struct wapbl_ino *wi;
1893 
1894 	KASSERT(mutex_owned(&wl->wl_mtx));
1895 
1896 	wih = &wl->wl_inohash[ino & wl->wl_inohashmask];
1897 	LIST_FOREACH(wi, wih, wi_hash) {
1898 		if (ino == wi->wi_ino)
1899 			return wi;
1900 	}
1901 	return 0;
1902 }
1903 
1904 void
1905 wapbl_register_inode(struct wapbl *wl, ino_t ino, mode_t mode)
1906 {
1907 	struct wapbl_ino_head *wih;
1908 	struct wapbl_ino *wi;
1909 
1910 	wi = pool_get(&wapbl_ino_pool, PR_WAITOK);
1911 
1912 	mutex_enter(&wl->wl_mtx);
1913 	if (wapbl_inodetrk_get(wl, ino) == NULL) {
1914 		wi->wi_ino = ino;
1915 		wi->wi_mode = mode;
1916 		wih = &wl->wl_inohash[ino & wl->wl_inohashmask];
1917 		LIST_INSERT_HEAD(wih, wi, wi_hash);
1918 		wl->wl_inohashcnt++;
1919 		WAPBL_PRINTF(WAPBL_PRINT_INODE,
1920 		    ("wapbl_register_inode: ino=%"PRId64"\n", ino));
1921 		mutex_exit(&wl->wl_mtx);
1922 	} else {
1923 		mutex_exit(&wl->wl_mtx);
1924 		pool_put(&wapbl_ino_pool, wi);
1925 	}
1926 }
1927 
1928 void
1929 wapbl_unregister_inode(struct wapbl *wl, ino_t ino, mode_t mode)
1930 {
1931 	struct wapbl_ino *wi;
1932 
1933 	mutex_enter(&wl->wl_mtx);
1934 	wi = wapbl_inodetrk_get(wl, ino);
1935 	if (wi) {
1936 		WAPBL_PRINTF(WAPBL_PRINT_INODE,
1937 		    ("wapbl_unregister_inode: ino=%"PRId64"\n", ino));
1938 		KASSERT(wl->wl_inohashcnt > 0);
1939 		wl->wl_inohashcnt--;
1940 		LIST_REMOVE(wi, wi_hash);
1941 		mutex_exit(&wl->wl_mtx);
1942 
1943 		pool_put(&wapbl_ino_pool, wi);
1944 	} else {
1945 		mutex_exit(&wl->wl_mtx);
1946 	}
1947 }
1948 
1949 /****************************************************************/
1950 
1951 static inline size_t
1952 wapbl_transaction_inodes_len(struct wapbl *wl)
1953 {
1954 	int blocklen = 1<<wl->wl_log_dev_bshift;
1955 	int iph;
1956 
1957 	/* Calculate number of inodes described in a inodelist header */
1958 	iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) /
1959 	    sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]);
1960 
1961 	KASSERT(iph > 0);
1962 
1963 	return MAX(1, howmany(wl->wl_inohashcnt, iph)) * blocklen;
1964 }
1965 
1966 
1967 /* Calculate amount of space a transaction will take on disk */
1968 static size_t
1969 wapbl_transaction_len(struct wapbl *wl)
1970 {
1971 	int blocklen = 1<<wl->wl_log_dev_bshift;
1972 	size_t len;
1973 	int bph;
1974 
1975 	/* Calculate number of blocks described in a blocklist header */
1976 	bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
1977 	    sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
1978 
1979 	KASSERT(bph > 0);
1980 
1981 	len = wl->wl_bcount;
1982 	len += howmany(wl->wl_bufcount, bph) * blocklen;
1983 	len += howmany(wl->wl_dealloccnt, bph) * blocklen;
1984 	len += wapbl_transaction_inodes_len(wl);
1985 
1986 	return len;
1987 }
1988 
1989 /*
1990  * wapbl_cache_sync: issue DIOCCACHESYNC
1991  */
1992 static int
1993 wapbl_cache_sync(struct wapbl *wl, const char *msg)
1994 {
1995 	const bool verbose = wapbl_verbose_commit >= 2;
1996 	struct bintime start_time;
1997 	int force = 1;
1998 	int error;
1999 
2000 	if (!wapbl_flush_disk_cache) {
2001 		return 0;
2002 	}
2003 	if (verbose) {
2004 		bintime(&start_time);
2005 	}
2006 	error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force,
2007 	    FWRITE, FSCRED);
2008 	if (error) {
2009 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
2010 		    ("wapbl_cache_sync: DIOCCACHESYNC on dev 0x%x "
2011 		    "returned %d\n", wl->wl_devvp->v_rdev, error));
2012 	}
2013 	if (verbose) {
2014 		struct bintime d;
2015 		struct timespec ts;
2016 
2017 		bintime(&d);
2018 		bintime_sub(&d, &start_time);
2019 		bintime2timespec(&d, &ts);
2020 		printf("wapbl_cache_sync: %s: dev 0x%jx %ju.%09lu\n",
2021 		    msg, (uintmax_t)wl->wl_devvp->v_rdev,
2022 		    (uintmax_t)ts.tv_sec, ts.tv_nsec);
2023 	}
2024 	return error;
2025 }
2026 
2027 /*
2028  * Perform commit operation
2029  *
2030  * Note that generation number incrementation needs to
2031  * be protected against racing with other invocations
2032  * of wapbl_write_commit.  This is ok since this routine
2033  * is only invoked from wapbl_flush
2034  */
2035 static int
2036 wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail)
2037 {
2038 	struct wapbl_wc_header *wc = wl->wl_wc_header;
2039 	struct timespec ts;
2040 	int error;
2041 	daddr_t pbn;
2042 
2043 	error = wapbl_buffered_flush(wl);
2044 	if (error)
2045 		return error;
2046 	/*
2047 	 * flush disk cache to ensure that blocks we've written are actually
2048 	 * written to the stable storage before the commit header.
2049 	 *
2050 	 * XXX Calc checksum here, instead we do this for now
2051 	 */
2052 	wapbl_cache_sync(wl, "1");
2053 
2054 	wc->wc_head = head;
2055 	wc->wc_tail = tail;
2056 	wc->wc_checksum = 0;
2057 	wc->wc_version = 1;
2058 	getnanotime(&ts);
2059 	wc->wc_time = ts.tv_sec;
2060 	wc->wc_timensec = ts.tv_nsec;
2061 
2062 	WAPBL_PRINTF(WAPBL_PRINT_WRITE,
2063 	    ("wapbl_write_commit: head = %"PRIdMAX "tail = %"PRIdMAX"\n",
2064 	    (intmax_t)head, (intmax_t)tail));
2065 
2066 	/*
2067 	 * write the commit header.
2068 	 *
2069 	 * XXX if generation will rollover, then first zero
2070 	 * over second commit header before trying to write both headers.
2071 	 */
2072 
2073 	pbn = wl->wl_logpbn + (wc->wc_generation % 2);
2074 #ifdef _KERNEL
2075 	pbn = btodb(pbn << wc->wc_log_dev_bshift);
2076 #endif
2077 	error = wapbl_buffered_write(wc, wc->wc_len, wl, pbn);
2078 	if (error)
2079 		return error;
2080 	error = wapbl_buffered_flush(wl);
2081 	if (error)
2082 		return error;
2083 
2084 	/*
2085 	 * flush disk cache to ensure that the commit header is actually
2086 	 * written before meta data blocks.
2087 	 */
2088 	wapbl_cache_sync(wl, "2");
2089 
2090 	/*
2091 	 * If the generation number was zero, write it out a second time.
2092 	 * This handles initialization and generation number rollover
2093 	 */
2094 	if (wc->wc_generation++ == 0) {
2095 		error = wapbl_write_commit(wl, head, tail);
2096 		/*
2097 		 * This panic should be able to be removed if we do the
2098 		 * zero'ing mentioned above, and we are certain to roll
2099 		 * back generation number on failure.
2100 		 */
2101 		if (error)
2102 			panic("wapbl_write_commit: error writing duplicate "
2103 			      "log header: %d\n", error);
2104 	}
2105 	return 0;
2106 }
2107 
2108 /* Returns new offset value */
2109 static int
2110 wapbl_write_blocks(struct wapbl *wl, off_t *offp)
2111 {
2112 	struct wapbl_wc_blocklist *wc =
2113 	    (struct wapbl_wc_blocklist *)wl->wl_wc_scratch;
2114 	int blocklen = 1<<wl->wl_log_dev_bshift;
2115 	int bph;
2116 	struct buf *bp;
2117 	off_t off = *offp;
2118 	int error;
2119 	size_t padding;
2120 
2121 	KASSERT(rw_write_held(&wl->wl_rwlock));
2122 
2123 	bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
2124 	    sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
2125 
2126 	bp = LIST_FIRST(&wl->wl_bufs);
2127 
2128 	while (bp) {
2129 		int cnt;
2130 		struct buf *obp = bp;
2131 
2132 		KASSERT(bp->b_flags & B_LOCKED);
2133 
2134 		wc->wc_type = WAPBL_WC_BLOCKS;
2135 		wc->wc_len = blocklen;
2136 		wc->wc_blkcount = 0;
2137 		while (bp && (wc->wc_blkcount < bph)) {
2138 			/*
2139 			 * Make sure all the physical block numbers are up to
2140 			 * date.  If this is not always true on a given
2141 			 * filesystem, then VOP_BMAP must be called.  We
2142 			 * could call VOP_BMAP here, or else in the filesystem
2143 			 * specific flush callback, although neither of those
2144 			 * solutions allow us to take the vnode lock.  If a
2145 			 * filesystem requires that we must take the vnode lock
2146 			 * to call VOP_BMAP, then we can probably do it in
2147 			 * bwrite when the vnode lock should already be held
2148 			 * by the invoking code.
2149 			 */
2150 			KASSERT((bp->b_vp->v_type == VBLK) ||
2151 				 (bp->b_blkno != bp->b_lblkno));
2152 			KASSERT(bp->b_blkno > 0);
2153 
2154 			wc->wc_blocks[wc->wc_blkcount].wc_daddr = bp->b_blkno;
2155 			wc->wc_blocks[wc->wc_blkcount].wc_dlen = bp->b_bcount;
2156 			wc->wc_len += bp->b_bcount;
2157 			wc->wc_blkcount++;
2158 			bp = LIST_NEXT(bp, b_wapbllist);
2159 		}
2160 		if (wc->wc_len % blocklen != 0) {
2161 			padding = blocklen - wc->wc_len % blocklen;
2162 			wc->wc_len += padding;
2163 		} else {
2164 			padding = 0;
2165 		}
2166 
2167 		WAPBL_PRINTF(WAPBL_PRINT_WRITE,
2168 		    ("wapbl_write_blocks: len = %u (padding %zu) off = %"PRIdMAX"\n",
2169 		    wc->wc_len, padding, (intmax_t)off));
2170 
2171 		error = wapbl_circ_write(wl, wc, blocklen, &off);
2172 		if (error)
2173 			return error;
2174 		bp = obp;
2175 		cnt = 0;
2176 		while (bp && (cnt++ < bph)) {
2177 			error = wapbl_circ_write(wl, bp->b_data,
2178 			    bp->b_bcount, &off);
2179 			if (error)
2180 				return error;
2181 			bp = LIST_NEXT(bp, b_wapbllist);
2182 		}
2183 		if (padding) {
2184 			void *zero;
2185 
2186 			zero = wapbl_alloc(padding);
2187 			memset(zero, 0, padding);
2188 			error = wapbl_circ_write(wl, zero, padding, &off);
2189 			wapbl_free(zero, padding);
2190 			if (error)
2191 				return error;
2192 		}
2193 	}
2194 	*offp = off;
2195 	return 0;
2196 }
2197 
2198 static int
2199 wapbl_write_revocations(struct wapbl *wl, off_t *offp)
2200 {
2201 	struct wapbl_wc_blocklist *wc =
2202 	    (struct wapbl_wc_blocklist *)wl->wl_wc_scratch;
2203 	int i;
2204 	int blocklen = 1<<wl->wl_log_dev_bshift;
2205 	int bph;
2206 	off_t off = *offp;
2207 	int error;
2208 
2209 	if (wl->wl_dealloccnt == 0)
2210 		return 0;
2211 
2212 	bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
2213 	    sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
2214 
2215 	i = 0;
2216 	while (i < wl->wl_dealloccnt) {
2217 		wc->wc_type = WAPBL_WC_REVOCATIONS;
2218 		wc->wc_len = blocklen;
2219 		wc->wc_blkcount = 0;
2220 		while ((i < wl->wl_dealloccnt) && (wc->wc_blkcount < bph)) {
2221 			wc->wc_blocks[wc->wc_blkcount].wc_daddr =
2222 			    wl->wl_deallocblks[i];
2223 			wc->wc_blocks[wc->wc_blkcount].wc_dlen =
2224 			    wl->wl_dealloclens[i];
2225 			wc->wc_blkcount++;
2226 			i++;
2227 		}
2228 		WAPBL_PRINTF(WAPBL_PRINT_WRITE,
2229 		    ("wapbl_write_revocations: len = %u off = %"PRIdMAX"\n",
2230 		    wc->wc_len, (intmax_t)off));
2231 		error = wapbl_circ_write(wl, wc, blocklen, &off);
2232 		if (error)
2233 			return error;
2234 	}
2235 	*offp = off;
2236 	return 0;
2237 }
2238 
2239 static int
2240 wapbl_write_inodes(struct wapbl *wl, off_t *offp)
2241 {
2242 	struct wapbl_wc_inodelist *wc =
2243 	    (struct wapbl_wc_inodelist *)wl->wl_wc_scratch;
2244 	int i;
2245 	int blocklen = 1 << wl->wl_log_dev_bshift;
2246 	off_t off = *offp;
2247 	int error;
2248 
2249 	struct wapbl_ino_head *wih;
2250 	struct wapbl_ino *wi;
2251 	int iph;
2252 
2253 	iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) /
2254 	    sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]);
2255 
2256 	i = 0;
2257 	wih = &wl->wl_inohash[0];
2258 	wi = 0;
2259 	do {
2260 		wc->wc_type = WAPBL_WC_INODES;
2261 		wc->wc_len = blocklen;
2262 		wc->wc_inocnt = 0;
2263 		wc->wc_clear = (i == 0);
2264 		while ((i < wl->wl_inohashcnt) && (wc->wc_inocnt < iph)) {
2265 			while (!wi) {
2266 				KASSERT((wih - &wl->wl_inohash[0])
2267 				    <= wl->wl_inohashmask);
2268 				wi = LIST_FIRST(wih++);
2269 			}
2270 			wc->wc_inodes[wc->wc_inocnt].wc_inumber = wi->wi_ino;
2271 			wc->wc_inodes[wc->wc_inocnt].wc_imode = wi->wi_mode;
2272 			wc->wc_inocnt++;
2273 			i++;
2274 			wi = LIST_NEXT(wi, wi_hash);
2275 		}
2276 		WAPBL_PRINTF(WAPBL_PRINT_WRITE,
2277 		    ("wapbl_write_inodes: len = %u off = %"PRIdMAX"\n",
2278 		    wc->wc_len, (intmax_t)off));
2279 		error = wapbl_circ_write(wl, wc, blocklen, &off);
2280 		if (error)
2281 			return error;
2282 	} while (i < wl->wl_inohashcnt);
2283 
2284 	*offp = off;
2285 	return 0;
2286 }
2287 
2288 #endif /* _KERNEL */
2289 
2290 /****************************************************************/
2291 
2292 struct wapbl_blk {
2293 	LIST_ENTRY(wapbl_blk) wb_hash;
2294 	daddr_t wb_blk;
2295 	off_t wb_off; /* Offset of this block in the log */
2296 };
2297 #define	WAPBL_BLKPOOL_MIN 83
2298 
2299 static void
2300 wapbl_blkhash_init(struct wapbl_replay *wr, u_int size)
2301 {
2302 	if (size < WAPBL_BLKPOOL_MIN)
2303 		size = WAPBL_BLKPOOL_MIN;
2304 	KASSERT(wr->wr_blkhash == 0);
2305 #ifdef _KERNEL
2306 	wr->wr_blkhash = hashinit(size, HASH_LIST, true, &wr->wr_blkhashmask);
2307 #else /* ! _KERNEL */
2308 	/* Manually implement hashinit */
2309 	{
2310 		unsigned long i, hashsize;
2311 		for (hashsize = 1; hashsize < size; hashsize <<= 1)
2312 			continue;
2313 		wr->wr_blkhash = wapbl_alloc(hashsize * sizeof(*wr->wr_blkhash));
2314 		for (i = 0; i < hashsize; i++)
2315 			LIST_INIT(&wr->wr_blkhash[i]);
2316 		wr->wr_blkhashmask = hashsize - 1;
2317 	}
2318 #endif /* ! _KERNEL */
2319 }
2320 
2321 static void
2322 wapbl_blkhash_free(struct wapbl_replay *wr)
2323 {
2324 	KASSERT(wr->wr_blkhashcnt == 0);
2325 #ifdef _KERNEL
2326 	hashdone(wr->wr_blkhash, HASH_LIST, wr->wr_blkhashmask);
2327 #else /* ! _KERNEL */
2328 	wapbl_free(wr->wr_blkhash,
2329 	    (wr->wr_blkhashmask + 1) * sizeof(*wr->wr_blkhash));
2330 #endif /* ! _KERNEL */
2331 }
2332 
2333 static struct wapbl_blk *
2334 wapbl_blkhash_get(struct wapbl_replay *wr, daddr_t blk)
2335 {
2336 	struct wapbl_blk_head *wbh;
2337 	struct wapbl_blk *wb;
2338 	wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask];
2339 	LIST_FOREACH(wb, wbh, wb_hash) {
2340 		if (blk == wb->wb_blk)
2341 			return wb;
2342 	}
2343 	return 0;
2344 }
2345 
2346 static void
2347 wapbl_blkhash_ins(struct wapbl_replay *wr, daddr_t blk, off_t off)
2348 {
2349 	struct wapbl_blk_head *wbh;
2350 	struct wapbl_blk *wb;
2351 	wb = wapbl_blkhash_get(wr, blk);
2352 	if (wb) {
2353 		KASSERT(wb->wb_blk == blk);
2354 		wb->wb_off = off;
2355 	} else {
2356 		wb = wapbl_alloc(sizeof(*wb));
2357 		wb->wb_blk = blk;
2358 		wb->wb_off = off;
2359 		wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask];
2360 		LIST_INSERT_HEAD(wbh, wb, wb_hash);
2361 		wr->wr_blkhashcnt++;
2362 	}
2363 }
2364 
2365 static void
2366 wapbl_blkhash_rem(struct wapbl_replay *wr, daddr_t blk)
2367 {
2368 	struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk);
2369 	if (wb) {
2370 		KASSERT(wr->wr_blkhashcnt > 0);
2371 		wr->wr_blkhashcnt--;
2372 		LIST_REMOVE(wb, wb_hash);
2373 		wapbl_free(wb, sizeof(*wb));
2374 	}
2375 }
2376 
2377 static void
2378 wapbl_blkhash_clear(struct wapbl_replay *wr)
2379 {
2380 	unsigned long i;
2381 	for (i = 0; i <= wr->wr_blkhashmask; i++) {
2382 		struct wapbl_blk *wb;
2383 
2384 		while ((wb = LIST_FIRST(&wr->wr_blkhash[i]))) {
2385 			KASSERT(wr->wr_blkhashcnt > 0);
2386 			wr->wr_blkhashcnt--;
2387 			LIST_REMOVE(wb, wb_hash);
2388 			wapbl_free(wb, sizeof(*wb));
2389 		}
2390 	}
2391 	KASSERT(wr->wr_blkhashcnt == 0);
2392 }
2393 
2394 /****************************************************************/
2395 
2396 static int
2397 wapbl_circ_read(struct wapbl_replay *wr, void *data, size_t len, off_t *offp)
2398 {
2399 	size_t slen;
2400 	off_t off = *offp;
2401 	int error;
2402 	daddr_t pbn;
2403 
2404 	KASSERT(((len >> wr->wr_log_dev_bshift) <<
2405 	    wr->wr_log_dev_bshift) == len);
2406 
2407 	if (off < wr->wr_circ_off)
2408 		off = wr->wr_circ_off;
2409 	slen = wr->wr_circ_off + wr->wr_circ_size - off;
2410 	if (slen < len) {
2411 		pbn = wr->wr_logpbn + (off >> wr->wr_log_dev_bshift);
2412 #ifdef _KERNEL
2413 		pbn = btodb(pbn << wr->wr_log_dev_bshift);
2414 #endif
2415 		error = wapbl_read(data, slen, wr->wr_devvp, pbn);
2416 		if (error)
2417 			return error;
2418 		data = (uint8_t *)data + slen;
2419 		len -= slen;
2420 		off = wr->wr_circ_off;
2421 	}
2422 	pbn = wr->wr_logpbn + (off >> wr->wr_log_dev_bshift);
2423 #ifdef _KERNEL
2424 	pbn = btodb(pbn << wr->wr_log_dev_bshift);
2425 #endif
2426 	error = wapbl_read(data, len, wr->wr_devvp, pbn);
2427 	if (error)
2428 		return error;
2429 	off += len;
2430 	if (off >= wr->wr_circ_off + wr->wr_circ_size)
2431 		off = wr->wr_circ_off;
2432 	*offp = off;
2433 	return 0;
2434 }
2435 
2436 static void
2437 wapbl_circ_advance(struct wapbl_replay *wr, size_t len, off_t *offp)
2438 {
2439 	size_t slen;
2440 	off_t off = *offp;
2441 
2442 	KASSERT(((len >> wr->wr_log_dev_bshift) <<
2443 	    wr->wr_log_dev_bshift) == len);
2444 
2445 	if (off < wr->wr_circ_off)
2446 		off = wr->wr_circ_off;
2447 	slen = wr->wr_circ_off + wr->wr_circ_size - off;
2448 	if (slen < len) {
2449 		len -= slen;
2450 		off = wr->wr_circ_off;
2451 	}
2452 	off += len;
2453 	if (off >= wr->wr_circ_off + wr->wr_circ_size)
2454 		off = wr->wr_circ_off;
2455 	*offp = off;
2456 }
2457 
2458 /****************************************************************/
2459 
2460 int
2461 wapbl_replay_start(struct wapbl_replay **wrp, struct vnode *vp,
2462 	daddr_t off, size_t count, size_t blksize)
2463 {
2464 	struct wapbl_replay *wr;
2465 	int error;
2466 	struct vnode *devvp;
2467 	daddr_t logpbn;
2468 	uint8_t *scratch;
2469 	struct wapbl_wc_header *wch;
2470 	struct wapbl_wc_header *wch2;
2471 	/* Use this until we read the actual log header */
2472 	int log_dev_bshift = ilog2(blksize);
2473 	size_t used;
2474 	daddr_t pbn;
2475 
2476 	WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
2477 	    ("wapbl_replay_start: vp=%p off=%"PRId64 " count=%zu blksize=%zu\n",
2478 	    vp, off, count, blksize));
2479 
2480 	if (off < 0)
2481 		return EINVAL;
2482 
2483 	if (blksize < DEV_BSIZE)
2484 		return EINVAL;
2485 	if (blksize % DEV_BSIZE)
2486 		return EINVAL;
2487 
2488 #ifdef _KERNEL
2489 #if 0
2490 	/* XXX vp->v_size isn't reliably set for VBLK devices,
2491 	 * especially root.  However, we might still want to verify
2492 	 * that the full load is readable */
2493 	if ((off + count) * blksize > vp->v_size)
2494 		return EINVAL;
2495 #endif
2496 	if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, 0)) != 0) {
2497 		return error;
2498 	}
2499 #else /* ! _KERNEL */
2500 	devvp = vp;
2501 	logpbn = off;
2502 #endif /* ! _KERNEL */
2503 
2504 	scratch = wapbl_alloc(MAXBSIZE);
2505 
2506 	pbn = logpbn;
2507 #ifdef _KERNEL
2508 	pbn = btodb(pbn << log_dev_bshift);
2509 #endif
2510 	error = wapbl_read(scratch, 2<<log_dev_bshift, devvp, pbn);
2511 	if (error)
2512 		goto errout;
2513 
2514 	wch = (struct wapbl_wc_header *)scratch;
2515 	wch2 =
2516 	    (struct wapbl_wc_header *)(scratch + (1<<log_dev_bshift));
2517 	/* XXX verify checksums and magic numbers */
2518 	if (wch->wc_type != WAPBL_WC_HEADER) {
2519 		printf("Unrecognized wapbl magic: 0x%08x\n", wch->wc_type);
2520 		error = EFTYPE;
2521 		goto errout;
2522 	}
2523 
2524 	if (wch2->wc_generation > wch->wc_generation)
2525 		wch = wch2;
2526 
2527 	wr = wapbl_calloc(1, sizeof(*wr));
2528 
2529 	wr->wr_logvp = vp;
2530 	wr->wr_devvp = devvp;
2531 	wr->wr_logpbn = logpbn;
2532 
2533 	wr->wr_scratch = scratch;
2534 
2535 	wr->wr_log_dev_bshift = wch->wc_log_dev_bshift;
2536 	wr->wr_fs_dev_bshift = wch->wc_fs_dev_bshift;
2537 	wr->wr_circ_off = wch->wc_circ_off;
2538 	wr->wr_circ_size = wch->wc_circ_size;
2539 	wr->wr_generation = wch->wc_generation;
2540 
2541 	used = wapbl_space_used(wch->wc_circ_size, wch->wc_head, wch->wc_tail);
2542 
2543 	WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
2544 	    ("wapbl_replay: head=%"PRId64" tail=%"PRId64" off=%"PRId64
2545 	    " len=%"PRId64" used=%zu\n",
2546 	    wch->wc_head, wch->wc_tail, wch->wc_circ_off,
2547 	    wch->wc_circ_size, used));
2548 
2549 	wapbl_blkhash_init(wr, (used >> wch->wc_fs_dev_bshift));
2550 
2551 	error = wapbl_replay_process(wr, wch->wc_head, wch->wc_tail);
2552 	if (error) {
2553 		wapbl_replay_stop(wr);
2554 		wapbl_replay_free(wr);
2555 		return error;
2556 	}
2557 
2558 	*wrp = wr;
2559 	return 0;
2560 
2561  errout:
2562 	wapbl_free(scratch, MAXBSIZE);
2563 	return error;
2564 }
2565 
2566 void
2567 wapbl_replay_stop(struct wapbl_replay *wr)
2568 {
2569 
2570 	if (!wapbl_replay_isopen(wr))
2571 		return;
2572 
2573 	WAPBL_PRINTF(WAPBL_PRINT_REPLAY, ("wapbl_replay_stop called\n"));
2574 
2575 	wapbl_free(wr->wr_scratch, MAXBSIZE);
2576 	wr->wr_scratch = NULL;
2577 
2578 	wr->wr_logvp = NULL;
2579 
2580 	wapbl_blkhash_clear(wr);
2581 	wapbl_blkhash_free(wr);
2582 }
2583 
2584 void
2585 wapbl_replay_free(struct wapbl_replay *wr)
2586 {
2587 
2588 	KDASSERT(!wapbl_replay_isopen(wr));
2589 
2590 	if (wr->wr_inodes)
2591 		wapbl_free(wr->wr_inodes,
2592 		    wr->wr_inodescnt * sizeof(wr->wr_inodes[0]));
2593 	wapbl_free(wr, sizeof(*wr));
2594 }
2595 
2596 #ifdef _KERNEL
2597 int
2598 wapbl_replay_isopen1(struct wapbl_replay *wr)
2599 {
2600 
2601 	return wapbl_replay_isopen(wr);
2602 }
2603 #endif
2604 
2605 static void
2606 wapbl_replay_process_blocks(struct wapbl_replay *wr, off_t *offp)
2607 {
2608 	struct wapbl_wc_blocklist *wc =
2609 	    (struct wapbl_wc_blocklist *)wr->wr_scratch;
2610 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2611 	int i, j, n;
2612 
2613 	for (i = 0; i < wc->wc_blkcount; i++) {
2614 		/*
2615 		 * Enter each physical block into the hashtable independently.
2616 		 */
2617 		n = wc->wc_blocks[i].wc_dlen >> wr->wr_fs_dev_bshift;
2618 		for (j = 0; j < n; j++) {
2619 			wapbl_blkhash_ins(wr, wc->wc_blocks[i].wc_daddr + btodb(j * fsblklen),
2620 			    *offp);
2621 			wapbl_circ_advance(wr, fsblklen, offp);
2622 		}
2623 	}
2624 }
2625 
2626 static void
2627 wapbl_replay_process_revocations(struct wapbl_replay *wr)
2628 {
2629 	struct wapbl_wc_blocklist *wc =
2630 	    (struct wapbl_wc_blocklist *)wr->wr_scratch;
2631 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2632 	int i, j, n;
2633 
2634 	for (i = 0; i < wc->wc_blkcount; i++) {
2635 		/*
2636 		 * Remove any blocks found from the hashtable.
2637 		 */
2638 		n = wc->wc_blocks[i].wc_dlen >> wr->wr_fs_dev_bshift;
2639 		for (j = 0; j < n; j++)
2640 			wapbl_blkhash_rem(wr, wc->wc_blocks[i].wc_daddr + btodb(j * fsblklen));
2641 	}
2642 }
2643 
2644 static void
2645 wapbl_replay_process_inodes(struct wapbl_replay *wr, off_t oldoff, off_t newoff)
2646 {
2647 	struct wapbl_wc_inodelist *wc =
2648 	    (struct wapbl_wc_inodelist *)wr->wr_scratch;
2649 	void *new_inodes;
2650 	const size_t oldsize = wr->wr_inodescnt * sizeof(wr->wr_inodes[0]);
2651 
2652 	KASSERT(sizeof(wr->wr_inodes[0]) == sizeof(wc->wc_inodes[0]));
2653 
2654 	/*
2655 	 * Keep track of where we found this so location won't be
2656 	 * overwritten.
2657 	 */
2658 	if (wc->wc_clear) {
2659 		wr->wr_inodestail = oldoff;
2660 		wr->wr_inodescnt = 0;
2661 		if (wr->wr_inodes != NULL) {
2662 			wapbl_free(wr->wr_inodes, oldsize);
2663 			wr->wr_inodes = NULL;
2664 		}
2665 	}
2666 	wr->wr_inodeshead = newoff;
2667 	if (wc->wc_inocnt == 0)
2668 		return;
2669 
2670 	new_inodes = wapbl_alloc((wr->wr_inodescnt + wc->wc_inocnt) *
2671 	    sizeof(wr->wr_inodes[0]));
2672 	if (wr->wr_inodes != NULL) {
2673 		memcpy(new_inodes, wr->wr_inodes, oldsize);
2674 		wapbl_free(wr->wr_inodes, oldsize);
2675 	}
2676 	wr->wr_inodes = new_inodes;
2677 	memcpy(&wr->wr_inodes[wr->wr_inodescnt], wc->wc_inodes,
2678 	    wc->wc_inocnt * sizeof(wr->wr_inodes[0]));
2679 	wr->wr_inodescnt += wc->wc_inocnt;
2680 }
2681 
2682 static int
2683 wapbl_replay_process(struct wapbl_replay *wr, off_t head, off_t tail)
2684 {
2685 	off_t off;
2686 	int error;
2687 
2688 	int logblklen = 1 << wr->wr_log_dev_bshift;
2689 
2690 	wapbl_blkhash_clear(wr);
2691 
2692 	off = tail;
2693 	while (off != head) {
2694 		struct wapbl_wc_null *wcn;
2695 		off_t saveoff = off;
2696 		error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off);
2697 		if (error)
2698 			goto errout;
2699 		wcn = (struct wapbl_wc_null *)wr->wr_scratch;
2700 		switch (wcn->wc_type) {
2701 		case WAPBL_WC_BLOCKS:
2702 			wapbl_replay_process_blocks(wr, &off);
2703 			break;
2704 
2705 		case WAPBL_WC_REVOCATIONS:
2706 			wapbl_replay_process_revocations(wr);
2707 			break;
2708 
2709 		case WAPBL_WC_INODES:
2710 			wapbl_replay_process_inodes(wr, saveoff, off);
2711 			break;
2712 
2713 		default:
2714 			printf("Unrecognized wapbl type: 0x%08x\n",
2715 			       wcn->wc_type);
2716  			error = EFTYPE;
2717 			goto errout;
2718 		}
2719 		wapbl_circ_advance(wr, wcn->wc_len, &saveoff);
2720 		if (off != saveoff) {
2721 			printf("wapbl_replay: corrupted records\n");
2722 			error = EFTYPE;
2723 			goto errout;
2724 		}
2725 	}
2726 	return 0;
2727 
2728  errout:
2729 	wapbl_blkhash_clear(wr);
2730 	return error;
2731 }
2732 
2733 #if 0
2734 int
2735 wapbl_replay_verify(struct wapbl_replay *wr, struct vnode *fsdevvp)
2736 {
2737 	off_t off;
2738 	int mismatchcnt = 0;
2739 	int logblklen = 1 << wr->wr_log_dev_bshift;
2740 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2741 	void *scratch1 = wapbl_alloc(MAXBSIZE);
2742 	void *scratch2 = wapbl_alloc(MAXBSIZE);
2743 	int error = 0;
2744 
2745 	KDASSERT(wapbl_replay_isopen(wr));
2746 
2747 	off = wch->wc_tail;
2748 	while (off != wch->wc_head) {
2749 		struct wapbl_wc_null *wcn;
2750 #ifdef DEBUG
2751 		off_t saveoff = off;
2752 #endif
2753 		error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off);
2754 		if (error)
2755 			goto out;
2756 		wcn = (struct wapbl_wc_null *)wr->wr_scratch;
2757 		switch (wcn->wc_type) {
2758 		case WAPBL_WC_BLOCKS:
2759 			{
2760 				struct wapbl_wc_blocklist *wc =
2761 				    (struct wapbl_wc_blocklist *)wr->wr_scratch;
2762 				int i;
2763 				for (i = 0; i < wc->wc_blkcount; i++) {
2764 					int foundcnt = 0;
2765 					int dirtycnt = 0;
2766 					int j, n;
2767 					/*
2768 					 * Check each physical block into the
2769 					 * hashtable independently
2770 					 */
2771 					n = wc->wc_blocks[i].wc_dlen >>
2772 					    wch->wc_fs_dev_bshift;
2773 					for (j = 0; j < n; j++) {
2774 						struct wapbl_blk *wb =
2775 						   wapbl_blkhash_get(wr,
2776 						   wc->wc_blocks[i].wc_daddr + btodb(j * fsblklen));
2777 						if (wb && (wb->wb_off == off)) {
2778 							foundcnt++;
2779 							error =
2780 							    wapbl_circ_read(wr,
2781 							    scratch1, fsblklen,
2782 							    &off);
2783 							if (error)
2784 								goto out;
2785 							error =
2786 							    wapbl_read(scratch2,
2787 							    fsblklen, fsdevvp,
2788 							    wb->wb_blk);
2789 							if (error)
2790 								goto out;
2791 							if (memcmp(scratch1,
2792 								   scratch2,
2793 								   fsblklen)) {
2794 								printf(
2795 		"wapbl_verify: mismatch block %"PRId64" at off %"PRIdMAX"\n",
2796 		wb->wb_blk, (intmax_t)off);
2797 								dirtycnt++;
2798 								mismatchcnt++;
2799 							}
2800 						} else {
2801 							wapbl_circ_advance(wr,
2802 							    fsblklen, &off);
2803 						}
2804 					}
2805 #if 0
2806 					/*
2807 					 * If all of the blocks in an entry
2808 					 * are clean, then remove all of its
2809 					 * blocks from the hashtable since they
2810 					 * never will need replay.
2811 					 */
2812 					if ((foundcnt != 0) &&
2813 					    (dirtycnt == 0)) {
2814 						off = saveoff;
2815 						wapbl_circ_advance(wr,
2816 						    logblklen, &off);
2817 						for (j = 0; j < n; j++) {
2818 							struct wapbl_blk *wb =
2819 							   wapbl_blkhash_get(wr,
2820 							   wc->wc_blocks[i].wc_daddr + btodb(j * fsblklen));
2821 							if (wb &&
2822 							  (wb->wb_off == off)) {
2823 								wapbl_blkhash_rem(wr, wb->wb_blk);
2824 							}
2825 							wapbl_circ_advance(wr,
2826 							    fsblklen, &off);
2827 						}
2828 					}
2829 #endif
2830 				}
2831 			}
2832 			break;
2833 		case WAPBL_WC_REVOCATIONS:
2834 		case WAPBL_WC_INODES:
2835 			break;
2836 		default:
2837 			KASSERT(0);
2838 		}
2839 #ifdef DEBUG
2840 		wapbl_circ_advance(wr, wcn->wc_len, &saveoff);
2841 		KASSERT(off == saveoff);
2842 #endif
2843 	}
2844  out:
2845 	wapbl_free(scratch1, MAXBSIZE);
2846 	wapbl_free(scratch2, MAXBSIZE);
2847 	if (!error && mismatchcnt)
2848 		error = EFTYPE;
2849 	return error;
2850 }
2851 #endif
2852 
2853 int
2854 wapbl_replay_write(struct wapbl_replay *wr, struct vnode *fsdevvp)
2855 {
2856 	struct wapbl_blk *wb;
2857 	size_t i;
2858 	off_t off;
2859 	void *scratch;
2860 	int error = 0;
2861 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2862 
2863 	KDASSERT(wapbl_replay_isopen(wr));
2864 
2865 	scratch = wapbl_alloc(MAXBSIZE);
2866 
2867 	for (i = 0; i <= wr->wr_blkhashmask; ++i) {
2868 		LIST_FOREACH(wb, &wr->wr_blkhash[i], wb_hash) {
2869 			off = wb->wb_off;
2870 			error = wapbl_circ_read(wr, scratch, fsblklen, &off);
2871 			if (error)
2872 				break;
2873 			error = wapbl_write(scratch, fsblklen, fsdevvp,
2874 			    wb->wb_blk);
2875 			if (error)
2876 				break;
2877 		}
2878 	}
2879 
2880 	wapbl_free(scratch, MAXBSIZE);
2881 	return error;
2882 }
2883 
2884 int
2885 wapbl_replay_can_read(struct wapbl_replay *wr, daddr_t blk, long len)
2886 {
2887 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2888 
2889 	KDASSERT(wapbl_replay_isopen(wr));
2890 	KASSERT((len % fsblklen) == 0);
2891 
2892 	while (len != 0) {
2893 		struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk);
2894 		if (wb)
2895 			return 1;
2896 		len -= fsblklen;
2897 	}
2898 	return 0;
2899 }
2900 
2901 int
2902 wapbl_replay_read(struct wapbl_replay *wr, void *data, daddr_t blk, long len)
2903 {
2904 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2905 
2906 	KDASSERT(wapbl_replay_isopen(wr));
2907 
2908 	KASSERT((len % fsblklen) == 0);
2909 
2910 	while (len != 0) {
2911 		struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk);
2912 		if (wb) {
2913 			off_t off = wb->wb_off;
2914 			int error;
2915 			error = wapbl_circ_read(wr, data, fsblklen, &off);
2916 			if (error)
2917 				return error;
2918 		}
2919 		data = (uint8_t *)data + fsblklen;
2920 		len -= fsblklen;
2921 		blk++;
2922 	}
2923 	return 0;
2924 }
2925 
2926 #ifdef _KERNEL
2927 /*
2928  * This is not really a module now, but maybe on it's way to
2929  * being one some day.
2930  */
2931 MODULE(MODULE_CLASS_VFS, wapbl, NULL);
2932 
2933 static int
2934 wapbl_modcmd(modcmd_t cmd, void *arg)
2935 {
2936 
2937 	switch (cmd) {
2938 	case MODULE_CMD_INIT:
2939 		wapbl_init();
2940 		return 0;
2941 	case MODULE_CMD_FINI:
2942 #ifdef notyet
2943 		return wapbl_fini(true);
2944 #endif
2945 		return EOPNOTSUPP;
2946 	default:
2947 		return ENOTTY;
2948 	}
2949 }
2950 #endif /* _KERNEL */
2951