xref: /netbsd-src/sys/kern/vfs_wapbl.c (revision 274254cdae52594c1aa480a736aef78313d15c9c)
1 /*	$NetBSD: vfs_wapbl.c,v 1.25 2009/04/05 11:48:02 lukem 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 1.25 2009/04/05 11:48:02 lukem Exp $");
40 
41 #include <sys/param.h>
42 
43 #ifdef _KERNEL
44 #include <sys/param.h>
45 #include <sys/namei.h>
46 #include <sys/proc.h>
47 #include <sys/uio.h>
48 #include <sys/vnode.h>
49 #include <sys/file.h>
50 #include <sys/malloc.h>
51 #include <sys/resourcevar.h>
52 #include <sys/conf.h>
53 #include <sys/mount.h>
54 #include <sys/kernel.h>
55 #include <sys/kauth.h>
56 #include <sys/mutex.h>
57 #include <sys/atomic.h>
58 #include <sys/wapbl.h>
59 #include <sys/wapbl_replay.h>
60 
61 #include <miscfs/specfs/specdev.h>
62 
63 #if 0 /* notyet */
64 #define	wapbl_malloc(s) kmem_alloc((s), KM_SLEEP)
65 #define	wapbl_free(a, s) kmem_free((a), (s))
66 #define	wapbl_calloc(n, s) kmem_zalloc((n)*(s), KM_SLEEP)
67 #else
68 MALLOC_JUSTDEFINE(M_WAPBL, "wapbl", "write-ahead physical block logging");
69 #define	wapbl_malloc(s) malloc((s), M_WAPBL, M_WAITOK)
70 #define	wapbl_free(a, s) free((a), M_WAPBL)
71 #define	wapbl_calloc(n, s) malloc((n)*(s), M_WAPBL, M_WAITOK | M_ZERO)
72 #endif
73 
74 #else /* !_KERNEL */
75 #include <assert.h>
76 #include <errno.h>
77 #include <stdio.h>
78 #include <stdbool.h>
79 #include <stdlib.h>
80 #include <string.h>
81 
82 #include <sys/time.h>
83 #include <sys/wapbl.h>
84 #include <sys/wapbl_replay.h>
85 
86 #define	KDASSERT(x) assert(x)
87 #define	KASSERT(x) assert(x)
88 #define	wapbl_malloc(s) malloc(s)
89 #define	wapbl_free(a, s) free(a)
90 #define	wapbl_calloc(n, s) calloc((n), (s))
91 
92 #endif /* !_KERNEL */
93 
94 /*
95  * INTERNAL DATA STRUCTURES
96  */
97 
98 /*
99  * This structure holds per-mount log information.
100  *
101  * Legend:	a = atomic access only
102  *		r = read-only after init
103  *		l = rwlock held
104  *		m = mutex held
105  *		u = unlocked access ok
106  *		b = bufcache_lock held
107  */
108 struct wapbl {
109 	struct vnode *wl_logvp;	/* r:	log here */
110 	struct vnode *wl_devvp;	/* r:	log on this device */
111 	struct mount *wl_mount;	/* r:	mountpoint wl is associated with */
112 	daddr_t wl_logpbn;	/* r:	Physical block number of start of log */
113 	int wl_log_dev_bshift;	/* r:	logarithm of device block size of log
114 					device */
115 	int wl_fs_dev_bshift;	/* r:	logarithm of device block size of
116 					filesystem device */
117 
118 	unsigned wl_lock_count;	/* m:	Count of transactions in progress */
119 
120 	size_t wl_circ_size; 	/* r:	Number of bytes in buffer of log */
121 	size_t wl_circ_off;	/* r:	Number of bytes reserved at start */
122 
123 	size_t wl_bufcount_max;	/* r:	Number of buffers reserved for log */
124 	size_t wl_bufbytes_max;	/* r:	Number of buf bytes reserved for log */
125 
126 	off_t wl_head;		/* l:	Byte offset of log head */
127 	off_t wl_tail;		/* l:	Byte offset of log tail */
128 	/*
129 	 * head == tail == 0 means log is empty
130 	 * head == tail != 0 means log is full
131 	 * see assertions in wapbl_advance() for other boundary conditions.
132 	 * only truncate moves the tail, except when flush sets it to
133 	 * wl_header_size only flush moves the head, except when truncate
134 	 * sets it to 0.
135 	 */
136 
137 	struct wapbl_wc_header *wl_wc_header;	/* l	*/
138 	void *wl_wc_scratch;	/* l:	scratch space (XXX: por que?!?) */
139 
140 	kmutex_t wl_mtx;	/* u:	short-term lock */
141 	krwlock_t wl_rwlock;	/* u:	File system transaction lock */
142 
143 	/*
144 	 * Must be held while accessing
145 	 * wl_count or wl_bufs or head or tail
146 	 */
147 
148 	/*
149 	 * Callback called from within the flush routine to flush any extra
150 	 * bits.  Note that flush may be skipped without calling this if
151 	 * there are no outstanding buffers in the transaction.
152 	 */
153 #if _KERNEL
154 	wapbl_flush_fn_t wl_flush;	/* r	*/
155 	wapbl_flush_fn_t wl_flush_abort;/* r	*/
156 #endif
157 
158 	size_t wl_bufbytes;	/* m:	Byte count of pages in wl_bufs */
159 	size_t wl_bufcount;	/* m:	Count of buffers in wl_bufs */
160 	size_t wl_bcount;	/* m:	Total bcount of wl_bufs */
161 
162 	LIST_HEAD(, buf) wl_bufs; /* m:	Buffers in current transaction */
163 
164 	kcondvar_t wl_reclaimable_cv;	/* m (obviously) */
165 	size_t wl_reclaimable_bytes; /* m:	Amount of space available for
166 						reclamation by truncate */
167 	int wl_error_count;	/* m:	# of wl_entries with errors */
168 	size_t wl_reserved_bytes; /* never truncate log smaller than this */
169 
170 #ifdef WAPBL_DEBUG_BUFBYTES
171 	size_t wl_unsynced_bufbytes; /* Byte count of unsynced buffers */
172 #endif
173 
174 	daddr_t *wl_deallocblks;/* l:	address of block */
175 	int *wl_dealloclens;	/* l:	size of block */
176 	int wl_dealloccnt;	/* l:	total count */
177 	int wl_dealloclim;	/* l:	max count */
178 
179 	/* hashtable of inode numbers for allocated but unlinked inodes */
180 	/* synch ??? */
181 	LIST_HEAD(wapbl_ino_head, wapbl_ino) *wl_inohash;
182 	u_long wl_inohashmask;
183 	int wl_inohashcnt;
184 
185 	SIMPLEQ_HEAD(, wapbl_entry) wl_entries; /* On disk transaction
186 						   accounting */
187 };
188 
189 #ifdef WAPBL_DEBUG_PRINT
190 int wapbl_debug_print = WAPBL_DEBUG_PRINT;
191 #endif
192 
193 /****************************************************************/
194 #ifdef _KERNEL
195 
196 #ifdef WAPBL_DEBUG
197 struct wapbl *wapbl_debug_wl;
198 #endif
199 
200 static int wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail);
201 static int wapbl_write_blocks(struct wapbl *wl, off_t *offp);
202 static int wapbl_write_revocations(struct wapbl *wl, off_t *offp);
203 static int wapbl_write_inodes(struct wapbl *wl, off_t *offp);
204 #endif /* _KERNEL */
205 
206 static int wapbl_replay_process(struct wapbl_replay *wr, off_t, off_t);
207 
208 static __inline size_t wapbl_space_free(size_t avail, off_t head,
209 	off_t tail);
210 static __inline size_t wapbl_space_used(size_t avail, off_t head,
211 	off_t tail);
212 
213 #ifdef _KERNEL
214 
215 #define	WAPBL_INODETRK_SIZE 83
216 static int wapbl_ino_pool_refcount;
217 static struct pool wapbl_ino_pool;
218 struct wapbl_ino {
219 	LIST_ENTRY(wapbl_ino) wi_hash;
220 	ino_t wi_ino;
221 	mode_t wi_mode;
222 };
223 
224 static void wapbl_inodetrk_init(struct wapbl *wl, u_int size);
225 static void wapbl_inodetrk_free(struct wapbl *wl);
226 static struct wapbl_ino *wapbl_inodetrk_get(struct wapbl *wl, ino_t ino);
227 
228 static size_t wapbl_transaction_len(struct wapbl *wl);
229 static __inline size_t wapbl_transaction_inodes_len(struct wapbl *wl);
230 
231 #if 0
232 int wapbl_replay_verify(struct wapbl_replay *, struct vnode *);
233 #endif
234 
235 static int wapbl_replay_isopen1(struct wapbl_replay *);
236 
237 /*
238  * This is useful for debugging.  If set, the log will
239  * only be truncated when necessary.
240  */
241 int wapbl_lazy_truncate = 0;
242 
243 struct wapbl_ops wapbl_ops = {
244 	.wo_wapbl_discard	= wapbl_discard,
245 	.wo_wapbl_replay_isopen	= wapbl_replay_isopen1,
246 	.wo_wapbl_replay_can_read = wapbl_replay_can_read,
247 	.wo_wapbl_replay_read	= wapbl_replay_read,
248 	.wo_wapbl_add_buf	= wapbl_add_buf,
249 	.wo_wapbl_remove_buf	= wapbl_remove_buf,
250 	.wo_wapbl_resize_buf	= wapbl_resize_buf,
251 	.wo_wapbl_begin		= wapbl_begin,
252 	.wo_wapbl_end		= wapbl_end,
253 	.wo_wapbl_junlock_assert= wapbl_junlock_assert,
254 
255 	/* XXX: the following is only used to say "this is a wapbl buf" */
256 	.wo_wapbl_biodone	= wapbl_biodone,
257 };
258 
259 void
260 wapbl_init(void)
261 {
262 
263 	malloc_type_attach(M_WAPBL);
264 }
265 
266 static int
267 wapbl_start_flush_inodes(struct wapbl *wl, struct wapbl_replay *wr)
268 {
269 	int error, i;
270 
271 	WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
272 	    ("wapbl_start: reusing log with %d inodes\n", wr->wr_inodescnt));
273 
274 	/*
275 	 * Its only valid to reuse the replay log if its
276 	 * the same as the new log we just opened.
277 	 */
278 	KDASSERT(!wapbl_replay_isopen(wr));
279 	KASSERT(wl->wl_devvp->v_rdev == wr->wr_devvp->v_rdev);
280 	KASSERT(wl->wl_logpbn == wr->wr_logpbn);
281 	KASSERT(wl->wl_circ_size == wr->wr_circ_size);
282 	KASSERT(wl->wl_circ_off == wr->wr_circ_off);
283 	KASSERT(wl->wl_log_dev_bshift == wr->wr_log_dev_bshift);
284 	KASSERT(wl->wl_fs_dev_bshift == wr->wr_fs_dev_bshift);
285 
286 	wl->wl_wc_header->wc_generation = wr->wr_generation + 1;
287 
288 	for (i = 0; i < wr->wr_inodescnt; i++)
289 		wapbl_register_inode(wl, wr->wr_inodes[i].wr_inumber,
290 		    wr->wr_inodes[i].wr_imode);
291 
292 	/* Make sure new transaction won't overwrite old inodes list */
293 	KDASSERT(wapbl_transaction_len(wl) <=
294 	    wapbl_space_free(wl->wl_circ_size, wr->wr_inodeshead,
295 	    wr->wr_inodestail));
296 
297 	wl->wl_head = wl->wl_tail = wr->wr_inodeshead;
298 	wl->wl_reclaimable_bytes = wl->wl_reserved_bytes =
299 	    wapbl_transaction_len(wl);
300 
301 	error = wapbl_write_inodes(wl, &wl->wl_head);
302 	if (error)
303 		return error;
304 
305 	KASSERT(wl->wl_head != wl->wl_tail);
306 	KASSERT(wl->wl_head != 0);
307 
308 	return 0;
309 }
310 
311 int
312 wapbl_start(struct wapbl ** wlp, struct mount *mp, struct vnode *vp,
313 	daddr_t off, size_t count, size_t blksize, struct wapbl_replay *wr,
314 	wapbl_flush_fn_t flushfn, wapbl_flush_fn_t flushabortfn)
315 {
316 	struct wapbl *wl;
317 	struct vnode *devvp;
318 	daddr_t logpbn;
319 	int error;
320 	int log_dev_bshift = DEV_BSHIFT;
321 	int fs_dev_bshift = DEV_BSHIFT;
322 	int run;
323 
324 	WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_start: vp=%p off=%" PRId64
325 	    " count=%zu blksize=%zu\n", vp, off, count, blksize));
326 
327 	if (log_dev_bshift > fs_dev_bshift) {
328 		WAPBL_PRINTF(WAPBL_PRINT_OPEN,
329 			("wapbl: log device's block size cannot be larger "
330 			 "than filesystem's\n"));
331 		/*
332 		 * Not currently implemented, although it could be if
333 		 * needed someday.
334 		 */
335 		return ENOSYS;
336 	}
337 
338 	if (off < 0)
339 		return EINVAL;
340 
341 	if (blksize < DEV_BSIZE)
342 		return EINVAL;
343 	if (blksize % DEV_BSIZE)
344 		return EINVAL;
345 
346 	/* XXXTODO: verify that the full load is writable */
347 
348 	/*
349 	 * XXX check for minimum log size
350 	 * minimum is governed by minimum amount of space
351 	 * to complete a transaction. (probably truncate)
352 	 */
353 	/* XXX for now pick something minimal */
354 	if ((count * blksize) < MAXPHYS) {
355 		return ENOSPC;
356 	}
357 
358 	if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, &run)) != 0) {
359 		return error;
360 	}
361 
362 	wl = wapbl_calloc(1, sizeof(*wl));
363 	rw_init(&wl->wl_rwlock);
364 	mutex_init(&wl->wl_mtx, MUTEX_DEFAULT, IPL_NONE);
365 	cv_init(&wl->wl_reclaimable_cv, "wapblrec");
366 	LIST_INIT(&wl->wl_bufs);
367 	SIMPLEQ_INIT(&wl->wl_entries);
368 
369 	wl->wl_logvp = vp;
370 	wl->wl_devvp = devvp;
371 	wl->wl_mount = mp;
372 	wl->wl_logpbn = logpbn;
373 	wl->wl_log_dev_bshift = log_dev_bshift;
374 	wl->wl_fs_dev_bshift = fs_dev_bshift;
375 
376 	wl->wl_flush = flushfn;
377 	wl->wl_flush_abort = flushabortfn;
378 
379 	/* Reserve two log device blocks for the commit headers */
380 	wl->wl_circ_off = 2<<wl->wl_log_dev_bshift;
381 	wl->wl_circ_size = ((count * blksize) - wl->wl_circ_off);
382 	/* truncate the log usage to a multiple of log_dev_bshift */
383 	wl->wl_circ_size >>= wl->wl_log_dev_bshift;
384 	wl->wl_circ_size <<= wl->wl_log_dev_bshift;
385 
386 	/*
387 	 * wl_bufbytes_max limits the size of the in memory transaction space.
388 	 * - Since buffers are allocated and accounted for in units of
389 	 *   PAGE_SIZE it is required to be a multiple of PAGE_SIZE
390 	 *   (i.e. 1<<PAGE_SHIFT)
391 	 * - Since the log device has to be written in units of
392 	 *   1<<wl_log_dev_bshift it is required to be a mulitple of
393 	 *   1<<wl_log_dev_bshift.
394 	 * - Since filesystem will provide data in units of 1<<wl_fs_dev_bshift,
395 	 *   it is convenient to be a multiple of 1<<wl_fs_dev_bshift.
396 	 * Therefore it must be multiple of the least common multiple of those
397 	 * three quantities.  Fortunately, all of those quantities are
398 	 * guaranteed to be a power of two, and the least common multiple of
399 	 * a set of numbers which are all powers of two is simply the maximum
400 	 * of those numbers.  Finally, the maximum logarithm of a power of two
401 	 * is the same as the log of the maximum power of two.  So we can do
402 	 * the following operations to size wl_bufbytes_max:
403 	 */
404 
405 	/* XXX fix actual number of pages reserved per filesystem. */
406 	wl->wl_bufbytes_max = MIN(wl->wl_circ_size, buf_memcalc() / 2);
407 
408 	/* Round wl_bufbytes_max to the largest power of two constraint */
409 	wl->wl_bufbytes_max >>= PAGE_SHIFT;
410 	wl->wl_bufbytes_max <<= PAGE_SHIFT;
411 	wl->wl_bufbytes_max >>= wl->wl_log_dev_bshift;
412 	wl->wl_bufbytes_max <<= wl->wl_log_dev_bshift;
413 	wl->wl_bufbytes_max >>= wl->wl_fs_dev_bshift;
414 	wl->wl_bufbytes_max <<= wl->wl_fs_dev_bshift;
415 
416 	/* XXX maybe use filesystem fragment size instead of 1024 */
417 	/* XXX fix actual number of buffers reserved per filesystem. */
418 	wl->wl_bufcount_max = (nbuf / 2) * 1024;
419 
420 	/* XXX tie this into resource estimation */
421 	wl->wl_dealloclim = 2 * btodb(wl->wl_bufbytes_max);
422 
423 	wl->wl_deallocblks = wapbl_malloc(sizeof(*wl->wl_deallocblks) *
424 	    wl->wl_dealloclim);
425 	wl->wl_dealloclens = wapbl_malloc(sizeof(*wl->wl_dealloclens) *
426 	    wl->wl_dealloclim);
427 
428 	wapbl_inodetrk_init(wl, WAPBL_INODETRK_SIZE);
429 
430 	/* Initialize the commit header */
431 	{
432 		struct wapbl_wc_header *wc;
433 		size_t len = 1 << wl->wl_log_dev_bshift;
434 		wc = wapbl_calloc(1, len);
435 		wc->wc_type = WAPBL_WC_HEADER;
436 		wc->wc_len = len;
437 		wc->wc_circ_off = wl->wl_circ_off;
438 		wc->wc_circ_size = wl->wl_circ_size;
439 		/* XXX wc->wc_fsid */
440 		wc->wc_log_dev_bshift = wl->wl_log_dev_bshift;
441 		wc->wc_fs_dev_bshift = wl->wl_fs_dev_bshift;
442 		wl->wl_wc_header = wc;
443 		wl->wl_wc_scratch = wapbl_malloc(len);
444 	}
445 
446 	/*
447 	 * if there was an existing set of unlinked but
448 	 * allocated inodes, preserve it in the new
449 	 * log.
450 	 */
451 	if (wr && wr->wr_inodescnt) {
452 		error = wapbl_start_flush_inodes(wl, wr);
453 		if (error)
454 			goto errout;
455 	}
456 
457 	error = wapbl_write_commit(wl, wl->wl_head, wl->wl_tail);
458 	if (error) {
459 		goto errout;
460 	}
461 
462 	*wlp = wl;
463 #if defined(WAPBL_DEBUG)
464 	wapbl_debug_wl = wl;
465 #endif
466 
467 	return 0;
468  errout:
469 	wapbl_discard(wl);
470 	wapbl_free(wl->wl_wc_scratch, wl->wl_wc_header->wc_len);
471 	wapbl_free(wl->wl_wc_header, wl->wl_wc_header->wc_len);
472 	wapbl_free(wl->wl_deallocblks,
473 	    sizeof(*wl->wl_deallocblks) * wl->wl_dealloclim);
474 	wapbl_free(wl->wl_dealloclens,
475 	    sizeof(*wl->wl_dealloclens) * wl->wl_dealloclim);
476 	wapbl_inodetrk_free(wl);
477 	wapbl_free(wl, sizeof(*wl));
478 
479 	return error;
480 }
481 
482 /*
483  * Like wapbl_flush, only discards the transaction
484  * completely
485  */
486 
487 void
488 wapbl_discard(struct wapbl *wl)
489 {
490 	struct wapbl_entry *we;
491 	struct buf *bp;
492 	int i;
493 
494 	/*
495 	 * XXX we may consider using upgrade here
496 	 * if we want to call flush from inside a transaction
497 	 */
498 	rw_enter(&wl->wl_rwlock, RW_WRITER);
499 	wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens,
500 	    wl->wl_dealloccnt);
501 
502 #ifdef WAPBL_DEBUG_PRINT
503 	{
504 		struct wapbl_entry *we;
505 		pid_t pid = -1;
506 		lwpid_t lid = -1;
507 		if (curproc)
508 			pid = curproc->p_pid;
509 		if (curlwp)
510 			lid = curlwp->l_lid;
511 #ifdef WAPBL_DEBUG_BUFBYTES
512 		WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
513 		    ("wapbl_discard: thread %d.%d discarding "
514 		    "transaction\n"
515 		    "\tbufcount=%zu bufbytes=%zu bcount=%zu "
516 		    "deallocs=%d inodes=%d\n"
517 		    "\terrcnt = %u, reclaimable=%zu reserved=%zu "
518 		    "unsynced=%zu\n",
519 		    pid, lid, wl->wl_bufcount, wl->wl_bufbytes,
520 		    wl->wl_bcount, wl->wl_dealloccnt,
521 		    wl->wl_inohashcnt, wl->wl_error_count,
522 		    wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
523 		    wl->wl_unsynced_bufbytes));
524 		SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
525 			WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
526 			    ("\tentry: bufcount = %zu, reclaimable = %zu, "
527 			     "error = %d, unsynced = %zu\n",
528 			     we->we_bufcount, we->we_reclaimable_bytes,
529 			     we->we_error, we->we_unsynced_bufbytes));
530 		}
531 #else /* !WAPBL_DEBUG_BUFBYTES */
532 		WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
533 		    ("wapbl_discard: thread %d.%d discarding transaction\n"
534 		    "\tbufcount=%zu bufbytes=%zu bcount=%zu "
535 		    "deallocs=%d inodes=%d\n"
536 		    "\terrcnt = %u, reclaimable=%zu reserved=%zu\n",
537 		    pid, lid, wl->wl_bufcount, wl->wl_bufbytes,
538 		    wl->wl_bcount, wl->wl_dealloccnt,
539 		    wl->wl_inohashcnt, wl->wl_error_count,
540 		    wl->wl_reclaimable_bytes, wl->wl_reserved_bytes));
541 		SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
542 			WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
543 			    ("\tentry: bufcount = %zu, reclaimable = %zu, "
544 			     "error = %d\n",
545 			     we->we_bufcount, we->we_reclaimable_bytes,
546 			     we->we_error));
547 		}
548 #endif /* !WAPBL_DEBUG_BUFBYTES */
549 	}
550 #endif /* WAPBL_DEBUG_PRINT */
551 
552 	for (i = 0; i <= wl->wl_inohashmask; i++) {
553 		struct wapbl_ino_head *wih;
554 		struct wapbl_ino *wi;
555 
556 		wih = &wl->wl_inohash[i];
557 		while ((wi = LIST_FIRST(wih)) != NULL) {
558 			LIST_REMOVE(wi, wi_hash);
559 			pool_put(&wapbl_ino_pool, wi);
560 			KASSERT(wl->wl_inohashcnt > 0);
561 			wl->wl_inohashcnt--;
562 		}
563 	}
564 
565 	/*
566 	 * clean buffer list
567 	 */
568 	mutex_enter(&bufcache_lock);
569 	mutex_enter(&wl->wl_mtx);
570 	while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) {
571 		if (bbusy(bp, 0, 0, &wl->wl_mtx) == 0) {
572 			/*
573 			 * The buffer will be unlocked and
574 			 * removed from the transaction in brelse
575 			 */
576 			mutex_exit(&wl->wl_mtx);
577 			brelsel(bp, 0);
578 			mutex_enter(&wl->wl_mtx);
579 		}
580 	}
581 	mutex_exit(&wl->wl_mtx);
582 	mutex_exit(&bufcache_lock);
583 
584 	/*
585 	 * Remove references to this wl from wl_entries, free any which
586 	 * no longer have buffers, others will be freed in wapbl_biodone
587 	 * when they no longer have any buffers.
588 	 */
589 	while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) != NULL) {
590 		SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries);
591 		/* XXX should we be accumulating wl_error_count
592 		 * and increasing reclaimable bytes ? */
593 		we->we_wapbl = NULL;
594 		if (we->we_bufcount == 0) {
595 #ifdef WAPBL_DEBUG_BUFBYTES
596 			KASSERT(we->we_unsynced_bufbytes == 0);
597 #endif
598 			wapbl_free(we, sizeof(*we));
599 		}
600 	}
601 
602 	/* Discard list of deallocs */
603 	wl->wl_dealloccnt = 0;
604 	/* XXX should we clear wl_reserved_bytes? */
605 
606 	KASSERT(wl->wl_bufbytes == 0);
607 	KASSERT(wl->wl_bcount == 0);
608 	KASSERT(wl->wl_bufcount == 0);
609 	KASSERT(LIST_EMPTY(&wl->wl_bufs));
610 	KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries));
611 	KASSERT(wl->wl_inohashcnt == 0);
612 
613 	rw_exit(&wl->wl_rwlock);
614 }
615 
616 int
617 wapbl_stop(struct wapbl *wl, int force)
618 {
619 	struct vnode *vp;
620 	int error;
621 
622 	WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_stop called\n"));
623 	error = wapbl_flush(wl, 1);
624 	if (error) {
625 		if (force)
626 			wapbl_discard(wl);
627 		else
628 			return error;
629 	}
630 
631 	/* Unlinked inodes persist after a flush */
632 	if (wl->wl_inohashcnt) {
633 		if (force) {
634 			wapbl_discard(wl);
635 		} else {
636 			return EBUSY;
637 		}
638 	}
639 
640 	KASSERT(wl->wl_bufbytes == 0);
641 	KASSERT(wl->wl_bcount == 0);
642 	KASSERT(wl->wl_bufcount == 0);
643 	KASSERT(LIST_EMPTY(&wl->wl_bufs));
644 	KASSERT(wl->wl_dealloccnt == 0);
645 	KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries));
646 	KASSERT(wl->wl_inohashcnt == 0);
647 
648 	vp = wl->wl_logvp;
649 
650 	wapbl_free(wl->wl_wc_scratch, wl->wl_wc_header->wc_len);
651 	wapbl_free(wl->wl_wc_header, wl->wl_wc_header->wc_len);
652 	wapbl_free(wl->wl_deallocblks,
653 	    sizeof(*wl->wl_deallocblks) * wl->wl_dealloclim);
654 	wapbl_free(wl->wl_dealloclens,
655 	    sizeof(*wl->wl_dealloclens) * wl->wl_dealloclim);
656 	wapbl_inodetrk_free(wl);
657 
658 	cv_destroy(&wl->wl_reclaimable_cv);
659 	mutex_destroy(&wl->wl_mtx);
660 	rw_destroy(&wl->wl_rwlock);
661 	wapbl_free(wl, sizeof(*wl));
662 
663 	return 0;
664 }
665 
666 static int
667 wapbl_doio(void *data, size_t len, struct vnode *devvp, daddr_t pbn, int flags)
668 {
669 	struct pstats *pstats = curlwp->l_proc->p_stats;
670 	struct buf *bp;
671 	int error;
672 
673 	KASSERT((flags & ~(B_WRITE | B_READ)) == 0);
674 	KASSERT(devvp->v_type == VBLK);
675 
676 	if ((flags & (B_WRITE | B_READ)) == B_WRITE) {
677 		mutex_enter(&devvp->v_interlock);
678 		devvp->v_numoutput++;
679 		mutex_exit(&devvp->v_interlock);
680 		pstats->p_ru.ru_oublock++;
681 	} else {
682 		pstats->p_ru.ru_inblock++;
683 	}
684 
685 	bp = getiobuf(devvp, true);
686 	bp->b_flags = flags;
687 	bp->b_cflags = BC_BUSY; /* silly & dubious */
688 	bp->b_dev = devvp->v_rdev;
689 	bp->b_data = data;
690 	bp->b_bufsize = bp->b_resid = bp->b_bcount = len;
691 	bp->b_blkno = pbn;
692 
693 	WAPBL_PRINTF(WAPBL_PRINT_IO,
694 	    ("wapbl_doio: %s %d bytes at block %"PRId64" on dev 0x%x\n",
695 	    BUF_ISWRITE(bp) ? "write" : "read", bp->b_bcount,
696 	    bp->b_blkno, bp->b_dev));
697 
698 	VOP_STRATEGY(devvp, bp);
699 
700 	error = biowait(bp);
701 	putiobuf(bp);
702 
703 	if (error) {
704 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
705 		    ("wapbl_doio: %s %zu bytes at block %" PRId64
706 		    " on dev 0x%x failed with error %d\n",
707 		    (((flags & (B_WRITE | B_READ)) == B_WRITE) ?
708 		     "write" : "read"),
709 		    len, pbn, devvp->v_rdev, error));
710 	}
711 
712 	return error;
713 }
714 
715 int
716 wapbl_write(void *data, size_t len, struct vnode *devvp, daddr_t pbn)
717 {
718 
719 	return wapbl_doio(data, len, devvp, pbn, B_WRITE);
720 }
721 
722 int
723 wapbl_read(void *data, size_t len, struct vnode *devvp, daddr_t pbn)
724 {
725 
726 	return wapbl_doio(data, len, devvp, pbn, B_READ);
727 }
728 
729 /*
730  * Off is byte offset returns new offset for next write
731  * handles log wraparound
732  */
733 static int
734 wapbl_circ_write(struct wapbl *wl, void *data, size_t len, off_t *offp)
735 {
736 	size_t slen;
737 	off_t off = *offp;
738 	int error;
739 
740 	KDASSERT(((len >> wl->wl_log_dev_bshift) <<
741 	    wl->wl_log_dev_bshift) == len);
742 
743 	if (off < wl->wl_circ_off)
744 		off = wl->wl_circ_off;
745 	slen = wl->wl_circ_off + wl->wl_circ_size - off;
746 	if (slen < len) {
747 		error = wapbl_write(data, slen, wl->wl_devvp,
748 		    wl->wl_logpbn + (off >> wl->wl_log_dev_bshift));
749 		if (error)
750 			return error;
751 		data = (uint8_t *)data + slen;
752 		len -= slen;
753 		off = wl->wl_circ_off;
754 	}
755 	error = wapbl_write(data, len, wl->wl_devvp,
756 			    wl->wl_logpbn + (off >> wl->wl_log_dev_bshift));
757 	if (error)
758 		return error;
759 	off += len;
760 	if (off >= wl->wl_circ_off + wl->wl_circ_size)
761 		off = wl->wl_circ_off;
762 	*offp = off;
763 	return 0;
764 }
765 
766 /****************************************************************/
767 
768 int
769 wapbl_begin(struct wapbl *wl, const char *file, int line)
770 {
771 	int doflush;
772 	unsigned lockcount;
773 
774 	KDASSERT(wl);
775 
776 	/*
777 	 * XXX this needs to be made much more sophisticated.
778 	 * perhaps each wapbl_begin could reserve a specified
779 	 * number of buffers and bytes.
780 	 */
781 	mutex_enter(&wl->wl_mtx);
782 	lockcount = wl->wl_lock_count;
783 	doflush = ((wl->wl_bufbytes + (lockcount * MAXPHYS)) >
784 		   wl->wl_bufbytes_max / 2) ||
785 		  ((wl->wl_bufcount + (lockcount * 10)) >
786 		   wl->wl_bufcount_max / 2) ||
787 		  (wapbl_transaction_len(wl) > wl->wl_circ_size / 2);
788 	mutex_exit(&wl->wl_mtx);
789 
790 	if (doflush) {
791 		WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
792 		    ("force flush lockcnt=%d bufbytes=%zu "
793 		    "(max=%zu) bufcount=%zu (max=%zu)\n",
794 		    lockcount, wl->wl_bufbytes,
795 		    wl->wl_bufbytes_max, wl->wl_bufcount,
796 		    wl->wl_bufcount_max));
797 	}
798 
799 	if (doflush) {
800 		int error = wapbl_flush(wl, 0);
801 		if (error)
802 			return error;
803 	}
804 
805 	rw_enter(&wl->wl_rwlock, RW_READER);
806 	mutex_enter(&wl->wl_mtx);
807 	wl->wl_lock_count++;
808 	mutex_exit(&wl->wl_mtx);
809 
810 #if defined(WAPBL_DEBUG_PRINT)
811 	WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION,
812 	    ("wapbl_begin thread %d.%d with bufcount=%zu "
813 	    "bufbytes=%zu bcount=%zu at %s:%d\n",
814 	    curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
815 	    wl->wl_bufbytes, wl->wl_bcount, file, line));
816 #endif
817 
818 	return 0;
819 }
820 
821 void
822 wapbl_end(struct wapbl *wl)
823 {
824 
825 #if defined(WAPBL_DEBUG_PRINT)
826 	WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION,
827 	     ("wapbl_end thread %d.%d with bufcount=%zu "
828 	      "bufbytes=%zu bcount=%zu\n",
829 	      curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
830 	      wl->wl_bufbytes, wl->wl_bcount));
831 #endif
832 
833 	mutex_enter(&wl->wl_mtx);
834 	KASSERT(wl->wl_lock_count > 0);
835 	wl->wl_lock_count--;
836 	mutex_exit(&wl->wl_mtx);
837 
838 	rw_exit(&wl->wl_rwlock);
839 }
840 
841 void
842 wapbl_add_buf(struct wapbl *wl, struct buf * bp)
843 {
844 
845 	KASSERT(bp->b_cflags & BC_BUSY);
846 	KASSERT(bp->b_vp);
847 
848 	wapbl_jlock_assert(wl);
849 
850 #if 0
851 	/*
852 	 * XXX this might be an issue for swapfiles.
853 	 * see uvm_swap.c:1702
854 	 *
855 	 * XXX2 why require it then?  leap of semantics?
856 	 */
857 	KASSERT((bp->b_cflags & BC_NOCACHE) == 0);
858 #endif
859 
860 	mutex_enter(&wl->wl_mtx);
861 	if (bp->b_flags & B_LOCKED) {
862 		LIST_REMOVE(bp, b_wapbllist);
863 		WAPBL_PRINTF(WAPBL_PRINT_BUFFER2,
864 		   ("wapbl_add_buf thread %d.%d re-adding buf %p "
865 		    "with %d bytes %d bcount\n",
866 		    curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize,
867 		    bp->b_bcount));
868 	} else {
869 		/* unlocked by dirty buffers shouldn't exist */
870 		KASSERT(!(bp->b_oflags & BO_DELWRI));
871 		wl->wl_bufbytes += bp->b_bufsize;
872 		wl->wl_bcount += bp->b_bcount;
873 		wl->wl_bufcount++;
874 		WAPBL_PRINTF(WAPBL_PRINT_BUFFER,
875 		   ("wapbl_add_buf thread %d.%d adding buf %p "
876 		    "with %d bytes %d bcount\n",
877 		    curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize,
878 		    bp->b_bcount));
879 	}
880 	LIST_INSERT_HEAD(&wl->wl_bufs, bp, b_wapbllist);
881 	mutex_exit(&wl->wl_mtx);
882 
883 	bp->b_flags |= B_LOCKED;
884 }
885 
886 static void
887 wapbl_remove_buf_locked(struct wapbl * wl, struct buf *bp)
888 {
889 
890 	KASSERT(mutex_owned(&wl->wl_mtx));
891 	KASSERT(bp->b_cflags & BC_BUSY);
892 	wapbl_jlock_assert(wl);
893 
894 #if 0
895 	/*
896 	 * XXX this might be an issue for swapfiles.
897 	 * see uvm_swap.c:1725
898 	 *
899 	 * XXXdeux: see above
900 	 */
901 	KASSERT((bp->b_flags & BC_NOCACHE) == 0);
902 #endif
903 	KASSERT(bp->b_flags & B_LOCKED);
904 
905 	WAPBL_PRINTF(WAPBL_PRINT_BUFFER,
906 	   ("wapbl_remove_buf thread %d.%d removing buf %p with "
907 	    "%d bytes %d bcount\n",
908 	    curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, bp->b_bcount));
909 
910 	KASSERT(wl->wl_bufbytes >= bp->b_bufsize);
911 	wl->wl_bufbytes -= bp->b_bufsize;
912 	KASSERT(wl->wl_bcount >= bp->b_bcount);
913 	wl->wl_bcount -= bp->b_bcount;
914 	KASSERT(wl->wl_bufcount > 0);
915 	wl->wl_bufcount--;
916 	KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0));
917 	KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0));
918 	LIST_REMOVE(bp, b_wapbllist);
919 
920 	bp->b_flags &= ~B_LOCKED;
921 }
922 
923 /* called from brelsel() in vfs_bio among other places */
924 void
925 wapbl_remove_buf(struct wapbl * wl, struct buf *bp)
926 {
927 
928 	mutex_enter(&wl->wl_mtx);
929 	wapbl_remove_buf_locked(wl, bp);
930 	mutex_exit(&wl->wl_mtx);
931 }
932 
933 void
934 wapbl_resize_buf(struct wapbl *wl, struct buf *bp, long oldsz, long oldcnt)
935 {
936 
937 	KASSERT(bp->b_cflags & BC_BUSY);
938 
939 	/*
940 	 * XXX: why does this depend on B_LOCKED?  otherwise the buf
941 	 * is not for a transaction?  if so, why is this called in the
942 	 * first place?
943 	 */
944 	if (bp->b_flags & B_LOCKED) {
945 		mutex_enter(&wl->wl_mtx);
946 		wl->wl_bufbytes += bp->b_bufsize - oldsz;
947 		wl->wl_bcount += bp->b_bcount - oldcnt;
948 		mutex_exit(&wl->wl_mtx);
949 	}
950 }
951 
952 #endif /* _KERNEL */
953 
954 /****************************************************************/
955 /* Some utility inlines */
956 
957 /* This is used to advance the pointer at old to new value at old+delta */
958 static __inline off_t
959 wapbl_advance(size_t size, size_t off, off_t old, size_t delta)
960 {
961 	off_t new;
962 
963 	/* Define acceptable ranges for inputs. */
964 	KASSERT(delta <= size);
965 	KASSERT((old == 0) || (old >= off));
966 	KASSERT(old < (size + off));
967 
968 	if ((old == 0) && (delta != 0))
969 		new = off + delta;
970 	else if ((old + delta) < (size + off))
971 		new = old + delta;
972 	else
973 		new = (old + delta) - size;
974 
975 	/* Note some interesting axioms */
976 	KASSERT((delta != 0) || (new == old));
977 	KASSERT((delta == 0) || (new != 0));
978 	KASSERT((delta != (size)) || (new == old));
979 
980 	/* Define acceptable ranges for output. */
981 	KASSERT((new == 0) || (new >= off));
982 	KASSERT(new < (size + off));
983 	return new;
984 }
985 
986 static __inline size_t
987 wapbl_space_used(size_t avail, off_t head, off_t tail)
988 {
989 
990 	if (tail == 0) {
991 		KASSERT(head == 0);
992 		return 0;
993 	}
994 	return ((head + (avail - 1) - tail) % avail) + 1;
995 }
996 
997 static __inline size_t
998 wapbl_space_free(size_t avail, off_t head, off_t tail)
999 {
1000 
1001 	return avail - wapbl_space_used(avail, head, tail);
1002 }
1003 
1004 static __inline void
1005 wapbl_advance_head(size_t size, size_t off, size_t delta, off_t *headp,
1006 		   off_t *tailp)
1007 {
1008 	off_t head = *headp;
1009 	off_t tail = *tailp;
1010 
1011 	KASSERT(delta <= wapbl_space_free(size, head, tail));
1012 	head = wapbl_advance(size, off, head, delta);
1013 	if ((tail == 0) && (head != 0))
1014 		tail = off;
1015 	*headp = head;
1016 	*tailp = tail;
1017 }
1018 
1019 static __inline void
1020 wapbl_advance_tail(size_t size, size_t off, size_t delta, off_t *headp,
1021 		   off_t *tailp)
1022 {
1023 	off_t head = *headp;
1024 	off_t tail = *tailp;
1025 
1026 	KASSERT(delta <= wapbl_space_used(size, head, tail));
1027 	tail = wapbl_advance(size, off, tail, delta);
1028 	if (head == tail) {
1029 		head = tail = 0;
1030 	}
1031 	*headp = head;
1032 	*tailp = tail;
1033 }
1034 
1035 #ifdef _KERNEL
1036 
1037 /****************************************************************/
1038 
1039 /*
1040  * Remove transactions whose buffers are completely flushed to disk.
1041  * Will block until at least minfree space is available.
1042  * only intended to be called from inside wapbl_flush and therefore
1043  * does not protect against commit races with itself or with flush.
1044  */
1045 static int
1046 wapbl_truncate(struct wapbl *wl, size_t minfree, int waitonly)
1047 {
1048 	size_t delta;
1049 	size_t avail;
1050 	off_t head;
1051 	off_t tail;
1052 	int error = 0;
1053 
1054 	KASSERT(minfree <= (wl->wl_circ_size - wl->wl_reserved_bytes));
1055 	KASSERT(rw_write_held(&wl->wl_rwlock));
1056 
1057 	mutex_enter(&wl->wl_mtx);
1058 
1059 	/*
1060 	 * First check to see if we have to do a commit
1061 	 * at all.
1062 	 */
1063 	avail = wapbl_space_free(wl->wl_circ_size, wl->wl_head, wl->wl_tail);
1064 	if (minfree < avail) {
1065 		mutex_exit(&wl->wl_mtx);
1066 		return 0;
1067 	}
1068 	minfree -= avail;
1069 	while ((wl->wl_error_count == 0) &&
1070 	    (wl->wl_reclaimable_bytes < minfree)) {
1071         	WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE,
1072                    ("wapbl_truncate: sleeping on %p wl=%p bytes=%zd "
1073 		    "minfree=%zd\n",
1074                     &wl->wl_reclaimable_bytes, wl, wl->wl_reclaimable_bytes,
1075 		    minfree));
1076 
1077 		cv_wait(&wl->wl_reclaimable_cv, &wl->wl_mtx);
1078 	}
1079 	if (wl->wl_reclaimable_bytes < minfree) {
1080 		KASSERT(wl->wl_error_count);
1081 		/* XXX maybe get actual error from buffer instead someday? */
1082 		error = EIO;
1083 	}
1084 	head = wl->wl_head;
1085 	tail = wl->wl_tail;
1086 	delta = wl->wl_reclaimable_bytes;
1087 
1088 	/* If all of of the entries are flushed, then be sure to keep
1089 	 * the reserved bytes reserved.  Watch out for discarded transactions,
1090 	 * which could leave more bytes reserved than are reclaimable.
1091 	 */
1092 	if (SIMPLEQ_EMPTY(&wl->wl_entries) &&
1093 	    (delta >= wl->wl_reserved_bytes)) {
1094 		delta -= wl->wl_reserved_bytes;
1095 	}
1096 	wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, &head,
1097 			   &tail);
1098 	KDASSERT(wl->wl_reserved_bytes <=
1099 		wapbl_space_used(wl->wl_circ_size, head, tail));
1100 	mutex_exit(&wl->wl_mtx);
1101 
1102 	if (error)
1103 		return error;
1104 
1105 	if (waitonly)
1106 		return 0;
1107 
1108 	/*
1109 	 * This is where head, tail and delta are unprotected
1110 	 * from races against itself or flush.  This is ok since
1111 	 * we only call this routine from inside flush itself.
1112 	 *
1113 	 * XXX: how can it race against itself when accessed only
1114 	 * from behind the write-locked rwlock?
1115 	 */
1116 	error = wapbl_write_commit(wl, head, tail);
1117 	if (error)
1118 		return error;
1119 
1120 	wl->wl_head = head;
1121 	wl->wl_tail = tail;
1122 
1123 	mutex_enter(&wl->wl_mtx);
1124 	KASSERT(wl->wl_reclaimable_bytes >= delta);
1125 	wl->wl_reclaimable_bytes -= delta;
1126 	mutex_exit(&wl->wl_mtx);
1127 	WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE,
1128 	    ("wapbl_truncate thread %d.%d truncating %zu bytes\n",
1129 	    curproc->p_pid, curlwp->l_lid, delta));
1130 
1131 	return 0;
1132 }
1133 
1134 /****************************************************************/
1135 
1136 void
1137 wapbl_biodone(struct buf *bp)
1138 {
1139 	struct wapbl_entry *we = bp->b_private;
1140 	struct wapbl *wl = we->we_wapbl;
1141 
1142 	/*
1143 	 * Handle possible flushing of buffers after log has been
1144 	 * decomissioned.
1145 	 */
1146 	if (!wl) {
1147 		KASSERT(we->we_bufcount > 0);
1148 		we->we_bufcount--;
1149 #ifdef WAPBL_DEBUG_BUFBYTES
1150 		KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize);
1151 		we->we_unsynced_bufbytes -= bp->b_bufsize;
1152 #endif
1153 
1154 		if (we->we_bufcount == 0) {
1155 #ifdef WAPBL_DEBUG_BUFBYTES
1156 			KASSERT(we->we_unsynced_bufbytes == 0);
1157 #endif
1158 			wapbl_free(we, sizeof(*we));
1159 		}
1160 
1161 		brelse(bp, 0);
1162 		return;
1163 	}
1164 
1165 #ifdef ohbother
1166 	KDASSERT(bp->b_flags & B_DONE);
1167 	KDASSERT(!(bp->b_flags & B_DELWRI));
1168 	KDASSERT(bp->b_flags & B_ASYNC);
1169 	KDASSERT(bp->b_flags & B_BUSY);
1170 	KDASSERT(!(bp->b_flags & B_LOCKED));
1171 	KDASSERT(!(bp->b_flags & B_READ));
1172 	KDASSERT(!(bp->b_flags & B_INVAL));
1173 	KDASSERT(!(bp->b_flags & B_NOCACHE));
1174 #endif
1175 
1176 	if (bp->b_error) {
1177 #ifdef notyet /* Can't currently handle possible dirty buffer reuse */
1178 		XXXpooka: interfaces not fully updated
1179 		Note: this was not enabled in the original patch
1180 		against netbsd4 either.  I don't know if comment
1181 		above is true or not.
1182 
1183 		/*
1184 		 * If an error occurs, report the error and leave the
1185 		 * buffer as a delayed write on the LRU queue.
1186 		 * restarting the write would likely result in
1187 		 * an error spinloop, so let it be done harmlessly
1188 		 * by the syncer.
1189 		 */
1190 		bp->b_flags &= ~(B_DONE);
1191 		simple_unlock(&bp->b_interlock);
1192 
1193 		if (we->we_error == 0) {
1194 			mutex_enter(&wl->wl_mtx);
1195 			wl->wl_error_count++;
1196 			mutex_exit(&wl->wl_mtx);
1197 			cv_broadcast(&wl->wl_reclaimable_cv);
1198 		}
1199 		we->we_error = bp->b_error;
1200 		bp->b_error = 0;
1201 		brelse(bp);
1202 		return;
1203 #else
1204 		/* For now, just mark the log permanently errored out */
1205 
1206 		mutex_enter(&wl->wl_mtx);
1207 		if (wl->wl_error_count == 0) {
1208 			wl->wl_error_count++;
1209 			cv_broadcast(&wl->wl_reclaimable_cv);
1210 		}
1211 		mutex_exit(&wl->wl_mtx);
1212 #endif
1213 	}
1214 
1215 	mutex_enter(&wl->wl_mtx);
1216 
1217 	KASSERT(we->we_bufcount > 0);
1218 	we->we_bufcount--;
1219 #ifdef WAPBL_DEBUG_BUFBYTES
1220 	KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize);
1221 	we->we_unsynced_bufbytes -= bp->b_bufsize;
1222 	KASSERT(wl->wl_unsynced_bufbytes >= bp->b_bufsize);
1223 	wl->wl_unsynced_bufbytes -= bp->b_bufsize;
1224 #endif
1225 
1226 	/*
1227 	 * If the current transaction can be reclaimed, start
1228 	 * at the beginning and reclaim any consecutive reclaimable
1229 	 * transactions.  If we successfully reclaim anything,
1230 	 * then wakeup anyone waiting for the reclaim.
1231 	 */
1232 	if (we->we_bufcount == 0) {
1233 		size_t delta = 0;
1234 		int errcnt = 0;
1235 #ifdef WAPBL_DEBUG_BUFBYTES
1236 		KDASSERT(we->we_unsynced_bufbytes == 0);
1237 #endif
1238 		/*
1239 		 * clear any posted error, since the buffer it came from
1240 		 * has successfully flushed by now
1241 		 */
1242 		while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) &&
1243 		       (we->we_bufcount == 0)) {
1244 			delta += we->we_reclaimable_bytes;
1245 			if (we->we_error)
1246 				errcnt++;
1247 			SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries);
1248 			wapbl_free(we, sizeof(*we));
1249 		}
1250 
1251 		if (delta) {
1252 			wl->wl_reclaimable_bytes += delta;
1253 			KASSERT(wl->wl_error_count >= errcnt);
1254 			wl->wl_error_count -= errcnt;
1255 			cv_broadcast(&wl->wl_reclaimable_cv);
1256 		}
1257 	}
1258 
1259 	mutex_exit(&wl->wl_mtx);
1260 	brelse(bp, 0);
1261 }
1262 
1263 /*
1264  * Write transactions to disk + start I/O for contents
1265  */
1266 int
1267 wapbl_flush(struct wapbl *wl, int waitfor)
1268 {
1269 	struct buf *bp;
1270 	struct wapbl_entry *we;
1271 	off_t off;
1272 	off_t head;
1273 	off_t tail;
1274 	size_t delta = 0;
1275 	size_t flushsize;
1276 	size_t reserved;
1277 	int error = 0;
1278 
1279 	/*
1280 	 * Do a quick check to see if a full flush can be skipped
1281 	 * This assumes that the flush callback does not need to be called
1282 	 * unless there are other outstanding bufs.
1283 	 */
1284 	if (!waitfor) {
1285 		size_t nbufs;
1286 		mutex_enter(&wl->wl_mtx);	/* XXX need mutex here to
1287 						   protect the KASSERTS */
1288 		nbufs = wl->wl_bufcount;
1289 		KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0));
1290 		KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0));
1291 		mutex_exit(&wl->wl_mtx);
1292 		if (nbufs == 0)
1293 			return 0;
1294 	}
1295 
1296 	/*
1297 	 * XXX we may consider using LK_UPGRADE here
1298 	 * if we want to call flush from inside a transaction
1299 	 */
1300 	rw_enter(&wl->wl_rwlock, RW_WRITER);
1301 	wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens,
1302 	    wl->wl_dealloccnt);
1303 
1304 	/*
1305 	 * Now that we are fully locked and flushed,
1306 	 * do another check for nothing to do.
1307 	 */
1308 	if (wl->wl_bufcount == 0) {
1309 		goto out;
1310 	}
1311 
1312 #if 0
1313 	WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1314 		     ("wapbl_flush thread %d.%d flushing entries with "
1315 		      "bufcount=%zu bufbytes=%zu\n",
1316 		      curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
1317 		      wl->wl_bufbytes));
1318 #endif
1319 
1320 	/* Calculate amount of space needed to flush */
1321 	flushsize = wapbl_transaction_len(wl);
1322 
1323 	if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) {
1324 		/*
1325 		 * XXX this could be handled more gracefully, perhaps place
1326 		 * only a partial transaction in the log and allow the
1327 		 * remaining to flush without the protection of the journal.
1328 		 */
1329 		panic("wapbl_flush: current transaction too big to flush\n");
1330 	}
1331 
1332 	error = wapbl_truncate(wl, flushsize, 0);
1333 	if (error)
1334 		goto out2;
1335 
1336 	off = wl->wl_head;
1337 	KASSERT((off == 0) || ((off >= wl->wl_circ_off) &&
1338 	                      (off < wl->wl_circ_off + wl->wl_circ_size)));
1339 	error = wapbl_write_blocks(wl, &off);
1340 	if (error)
1341 		goto out2;
1342 	error = wapbl_write_revocations(wl, &off);
1343 	if (error)
1344 		goto out2;
1345 	error = wapbl_write_inodes(wl, &off);
1346 	if (error)
1347 		goto out2;
1348 
1349 	reserved = 0;
1350 	if (wl->wl_inohashcnt)
1351 		reserved = wapbl_transaction_inodes_len(wl);
1352 
1353 	head = wl->wl_head;
1354 	tail = wl->wl_tail;
1355 
1356 	wapbl_advance_head(wl->wl_circ_size, wl->wl_circ_off, flushsize,
1357 	    &head, &tail);
1358 #ifdef WAPBL_DEBUG
1359 	if (head != off) {
1360 		panic("lost head! head=%"PRIdMAX" tail=%" PRIdMAX
1361 		      " off=%"PRIdMAX" flush=%zu\n",
1362 		      (intmax_t)head, (intmax_t)tail, (intmax_t)off,
1363 		      flushsize);
1364 	}
1365 #else
1366 	KASSERT(head == off);
1367 #endif
1368 
1369 	/* Opportunistically move the tail forward if we can */
1370 	if (!wapbl_lazy_truncate) {
1371 		mutex_enter(&wl->wl_mtx);
1372 		delta = wl->wl_reclaimable_bytes;
1373 		mutex_exit(&wl->wl_mtx);
1374 		wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta,
1375 		    &head, &tail);
1376 	}
1377 
1378 	error = wapbl_write_commit(wl, head, tail);
1379 	if (error)
1380 		goto out2;
1381 
1382 	we = wapbl_calloc(1, sizeof(*we));
1383 
1384 #ifdef WAPBL_DEBUG_BUFBYTES
1385 	WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1386 		("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu"
1387 		 " unsynced=%zu"
1388 		 "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d "
1389 		 "inodes=%d\n",
1390 		 curproc->p_pid, curlwp->l_lid, flushsize, delta,
1391 		 wapbl_space_used(wl->wl_circ_size, head, tail),
1392 		 wl->wl_unsynced_bufbytes, wl->wl_bufcount,
1393 		 wl->wl_bufbytes, wl->wl_bcount, wl->wl_dealloccnt,
1394 		 wl->wl_inohashcnt));
1395 #else
1396 	WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1397 		("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu"
1398 		 "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d "
1399 		 "inodes=%d\n",
1400 		 curproc->p_pid, curlwp->l_lid, flushsize, delta,
1401 		 wapbl_space_used(wl->wl_circ_size, head, tail),
1402 		 wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount,
1403 		 wl->wl_dealloccnt, wl->wl_inohashcnt));
1404 #endif
1405 
1406 
1407 	mutex_enter(&bufcache_lock);
1408 	mutex_enter(&wl->wl_mtx);
1409 
1410 	wl->wl_reserved_bytes = reserved;
1411 	wl->wl_head = head;
1412 	wl->wl_tail = tail;
1413 	KASSERT(wl->wl_reclaimable_bytes >= delta);
1414 	wl->wl_reclaimable_bytes -= delta;
1415 	wl->wl_dealloccnt = 0;
1416 #ifdef WAPBL_DEBUG_BUFBYTES
1417 	wl->wl_unsynced_bufbytes += wl->wl_bufbytes;
1418 #endif
1419 
1420 	we->we_wapbl = wl;
1421 	we->we_bufcount = wl->wl_bufcount;
1422 #ifdef WAPBL_DEBUG_BUFBYTES
1423 	we->we_unsynced_bufbytes = wl->wl_bufbytes;
1424 #endif
1425 	we->we_reclaimable_bytes = flushsize;
1426 	we->we_error = 0;
1427 	SIMPLEQ_INSERT_TAIL(&wl->wl_entries, we, we_entries);
1428 
1429 	/*
1430 	 * this flushes bufs in reverse order than they were queued
1431 	 * it shouldn't matter, but if we care we could use TAILQ instead.
1432 	 * XXX Note they will get put on the lru queue when they flush
1433 	 * so we might actually want to change this to preserve order.
1434 	 */
1435 	while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) {
1436 		if (bbusy(bp, 0, 0, &wl->wl_mtx)) {
1437 			continue;
1438 		}
1439 		bp->b_iodone = wapbl_biodone;
1440 		bp->b_private = we;
1441 		bremfree(bp);
1442 		wapbl_remove_buf_locked(wl, bp);
1443 		mutex_exit(&wl->wl_mtx);
1444 		mutex_exit(&bufcache_lock);
1445 		bawrite(bp);
1446 		mutex_enter(&bufcache_lock);
1447 		mutex_enter(&wl->wl_mtx);
1448 	}
1449 	mutex_exit(&wl->wl_mtx);
1450 	mutex_exit(&bufcache_lock);
1451 
1452 #if 0
1453 	WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1454 		     ("wapbl_flush thread %d.%d done flushing entries...\n",
1455 		     curproc->p_pid, curlwp->l_lid));
1456 #endif
1457 
1458  out:
1459 
1460 	/*
1461 	 * If the waitfor flag is set, don't return until everything is
1462 	 * fully flushed and the on disk log is empty.
1463 	 */
1464 	if (waitfor) {
1465 		error = wapbl_truncate(wl, wl->wl_circ_size -
1466 			wl->wl_reserved_bytes, wapbl_lazy_truncate);
1467 	}
1468 
1469  out2:
1470 	if (error) {
1471 		wl->wl_flush_abort(wl->wl_mount, wl->wl_deallocblks,
1472 		    wl->wl_dealloclens, wl->wl_dealloccnt);
1473 	}
1474 
1475 #ifdef WAPBL_DEBUG_PRINT
1476 	if (error) {
1477 		pid_t pid = -1;
1478 		lwpid_t lid = -1;
1479 		if (curproc)
1480 			pid = curproc->p_pid;
1481 		if (curlwp)
1482 			lid = curlwp->l_lid;
1483 		mutex_enter(&wl->wl_mtx);
1484 #ifdef WAPBL_DEBUG_BUFBYTES
1485 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1486 		    ("wapbl_flush: thread %d.%d aborted flush: "
1487 		    "error = %d\n"
1488 		    "\tbufcount=%zu bufbytes=%zu bcount=%zu "
1489 		    "deallocs=%d inodes=%d\n"
1490 		    "\terrcnt = %d, reclaimable=%zu reserved=%zu "
1491 		    "unsynced=%zu\n",
1492 		    pid, lid, error, wl->wl_bufcount,
1493 		    wl->wl_bufbytes, wl->wl_bcount,
1494 		    wl->wl_dealloccnt, wl->wl_inohashcnt,
1495 		    wl->wl_error_count, wl->wl_reclaimable_bytes,
1496 		    wl->wl_reserved_bytes, wl->wl_unsynced_bufbytes));
1497 		SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1498 			WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1499 			    ("\tentry: bufcount = %zu, reclaimable = %zu, "
1500 			     "error = %d, unsynced = %zu\n",
1501 			     we->we_bufcount, we->we_reclaimable_bytes,
1502 			     we->we_error, we->we_unsynced_bufbytes));
1503 		}
1504 #else
1505 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1506 		    ("wapbl_flush: thread %d.%d aborted flush: "
1507 		     "error = %d\n"
1508 		     "\tbufcount=%zu bufbytes=%zu bcount=%zu "
1509 		     "deallocs=%d inodes=%d\n"
1510 		     "\terrcnt = %d, reclaimable=%zu reserved=%zu\n",
1511 		     pid, lid, error, wl->wl_bufcount,
1512 		     wl->wl_bufbytes, wl->wl_bcount,
1513 		     wl->wl_dealloccnt, wl->wl_inohashcnt,
1514 		     wl->wl_error_count, wl->wl_reclaimable_bytes,
1515 		     wl->wl_reserved_bytes));
1516 		SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1517 			WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1518 			    ("\tentry: bufcount = %zu, reclaimable = %zu, "
1519 			     "error = %d\n", we->we_bufcount,
1520 			     we->we_reclaimable_bytes, we->we_error));
1521 		}
1522 #endif
1523 		mutex_exit(&wl->wl_mtx);
1524 	}
1525 #endif
1526 
1527 	rw_exit(&wl->wl_rwlock);
1528 	return error;
1529 }
1530 
1531 /****************************************************************/
1532 
1533 void
1534 wapbl_jlock_assert(struct wapbl *wl)
1535 {
1536 
1537 	KASSERT(rw_lock_held(&wl->wl_rwlock));
1538 }
1539 
1540 void
1541 wapbl_junlock_assert(struct wapbl *wl)
1542 {
1543 
1544 	KASSERT(!rw_write_held(&wl->wl_rwlock));
1545 }
1546 
1547 /****************************************************************/
1548 
1549 /* locks missing */
1550 void
1551 wapbl_print(struct wapbl *wl,
1552 		int full,
1553 		void (*pr)(const char *, ...))
1554 {
1555 	struct buf *bp;
1556 	struct wapbl_entry *we;
1557 	(*pr)("wapbl %p", wl);
1558 	(*pr)("\nlogvp = %p, devvp = %p, logpbn = %"PRId64"\n",
1559 	      wl->wl_logvp, wl->wl_devvp, wl->wl_logpbn);
1560 	(*pr)("circ = %zu, header = %zu, head = %"PRIdMAX" tail = %"PRIdMAX"\n",
1561 	      wl->wl_circ_size, wl->wl_circ_off,
1562 	      (intmax_t)wl->wl_head, (intmax_t)wl->wl_tail);
1563 	(*pr)("fs_dev_bshift = %d, log_dev_bshift = %d\n",
1564 	      wl->wl_log_dev_bshift, wl->wl_fs_dev_bshift);
1565 #ifdef WAPBL_DEBUG_BUFBYTES
1566 	(*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu "
1567 	      "reserved = %zu errcnt = %d unsynced = %zu\n",
1568 	      wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount,
1569 	      wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
1570 				wl->wl_error_count, wl->wl_unsynced_bufbytes);
1571 #else
1572 	(*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu "
1573 	      "reserved = %zu errcnt = %d\n", wl->wl_bufcount, wl->wl_bufbytes,
1574 	      wl->wl_bcount, wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
1575 				wl->wl_error_count);
1576 #endif
1577 	(*pr)("\tdealloccnt = %d, dealloclim = %d\n",
1578 	      wl->wl_dealloccnt, wl->wl_dealloclim);
1579 	(*pr)("\tinohashcnt = %d, inohashmask = 0x%08x\n",
1580 	      wl->wl_inohashcnt, wl->wl_inohashmask);
1581 	(*pr)("entries:\n");
1582 	SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1583 #ifdef WAPBL_DEBUG_BUFBYTES
1584 		(*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d, "
1585 		      "unsynced = %zu\n",
1586 		      we->we_bufcount, we->we_reclaimable_bytes,
1587 		      we->we_error, we->we_unsynced_bufbytes);
1588 #else
1589 		(*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d\n",
1590 		      we->we_bufcount, we->we_reclaimable_bytes, we->we_error);
1591 #endif
1592 	}
1593 	if (full) {
1594 		int cnt = 0;
1595 		(*pr)("bufs =");
1596 		LIST_FOREACH(bp, &wl->wl_bufs, b_wapbllist) {
1597 			if (!LIST_NEXT(bp, b_wapbllist)) {
1598 				(*pr)(" %p", bp);
1599 			} else if ((++cnt % 6) == 0) {
1600 				(*pr)(" %p,\n\t", bp);
1601 			} else {
1602 				(*pr)(" %p,", bp);
1603 			}
1604 		}
1605 		(*pr)("\n");
1606 
1607 		(*pr)("dealloced blks = ");
1608 		{
1609 			int i;
1610 			cnt = 0;
1611 			for (i = 0; i < wl->wl_dealloccnt; i++) {
1612 				(*pr)(" %"PRId64":%d,",
1613 				      wl->wl_deallocblks[i],
1614 				      wl->wl_dealloclens[i]);
1615 				if ((++cnt % 4) == 0) {
1616 					(*pr)("\n\t");
1617 				}
1618 			}
1619 		}
1620 		(*pr)("\n");
1621 
1622 		(*pr)("registered inodes = ");
1623 		{
1624 			int i;
1625 			cnt = 0;
1626 			for (i = 0; i <= wl->wl_inohashmask; i++) {
1627 				struct wapbl_ino_head *wih;
1628 				struct wapbl_ino *wi;
1629 
1630 				wih = &wl->wl_inohash[i];
1631 				LIST_FOREACH(wi, wih, wi_hash) {
1632 					if (wi->wi_ino == 0)
1633 						continue;
1634 					(*pr)(" %"PRId32"/0%06"PRIo32",",
1635 					    wi->wi_ino, wi->wi_mode);
1636 					if ((++cnt % 4) == 0) {
1637 						(*pr)("\n\t");
1638 					}
1639 				}
1640 			}
1641 			(*pr)("\n");
1642 		}
1643 	}
1644 }
1645 
1646 #if defined(WAPBL_DEBUG) || defined(DDB)
1647 void
1648 wapbl_dump(struct wapbl *wl)
1649 {
1650 #if defined(WAPBL_DEBUG)
1651 	if (!wl)
1652 		wl = wapbl_debug_wl;
1653 #endif
1654 	if (!wl)
1655 		return;
1656 	wapbl_print(wl, 1, printf);
1657 }
1658 #endif
1659 
1660 /****************************************************************/
1661 
1662 void
1663 wapbl_register_deallocation(struct wapbl *wl, daddr_t blk, int len)
1664 {
1665 
1666 	wapbl_jlock_assert(wl);
1667 
1668 	/* XXX should eventually instead tie this into resource estimation */
1669 	/* XXX this KASSERT needs locking/mutex analysis */
1670 	KASSERT(wl->wl_dealloccnt < wl->wl_dealloclim);
1671 	wl->wl_deallocblks[wl->wl_dealloccnt] = blk;
1672 	wl->wl_dealloclens[wl->wl_dealloccnt] = len;
1673 	wl->wl_dealloccnt++;
1674 	WAPBL_PRINTF(WAPBL_PRINT_ALLOC,
1675 	    ("wapbl_register_deallocation: blk=%"PRId64" len=%d\n", blk, len));
1676 }
1677 
1678 /****************************************************************/
1679 
1680 static void
1681 wapbl_inodetrk_init(struct wapbl *wl, u_int size)
1682 {
1683 
1684 	wl->wl_inohash = hashinit(size, HASH_LIST, true, &wl->wl_inohashmask);
1685 	if (atomic_inc_uint_nv(&wapbl_ino_pool_refcount) == 1) {
1686 		pool_init(&wapbl_ino_pool, sizeof(struct wapbl_ino), 0, 0, 0,
1687 		    "wapblinopl", &pool_allocator_nointr, IPL_NONE);
1688 	}
1689 }
1690 
1691 static void
1692 wapbl_inodetrk_free(struct wapbl *wl)
1693 {
1694 
1695 	/* XXX this KASSERT needs locking/mutex analysis */
1696 	KASSERT(wl->wl_inohashcnt == 0);
1697 	hashdone(wl->wl_inohash, HASH_LIST, wl->wl_inohashmask);
1698 	if (atomic_dec_uint_nv(&wapbl_ino_pool_refcount) == 0) {
1699 		pool_destroy(&wapbl_ino_pool);
1700 	}
1701 }
1702 
1703 static struct wapbl_ino *
1704 wapbl_inodetrk_get(struct wapbl *wl, ino_t ino)
1705 {
1706 	struct wapbl_ino_head *wih;
1707 	struct wapbl_ino *wi;
1708 
1709 	KASSERT(mutex_owned(&wl->wl_mtx));
1710 
1711 	wih = &wl->wl_inohash[ino & wl->wl_inohashmask];
1712 	LIST_FOREACH(wi, wih, wi_hash) {
1713 		if (ino == wi->wi_ino)
1714 			return wi;
1715 	}
1716 	return 0;
1717 }
1718 
1719 void
1720 wapbl_register_inode(struct wapbl *wl, ino_t ino, mode_t mode)
1721 {
1722 	struct wapbl_ino_head *wih;
1723 	struct wapbl_ino *wi;
1724 
1725 	wi = pool_get(&wapbl_ino_pool, PR_WAITOK);
1726 
1727 	mutex_enter(&wl->wl_mtx);
1728 	if (wapbl_inodetrk_get(wl, ino) == NULL) {
1729 		wi->wi_ino = ino;
1730 		wi->wi_mode = mode;
1731 		wih = &wl->wl_inohash[ino & wl->wl_inohashmask];
1732 		LIST_INSERT_HEAD(wih, wi, wi_hash);
1733 		wl->wl_inohashcnt++;
1734 		WAPBL_PRINTF(WAPBL_PRINT_INODE,
1735 		    ("wapbl_register_inode: ino=%"PRId64"\n", ino));
1736 		mutex_exit(&wl->wl_mtx);
1737 	} else {
1738 		mutex_exit(&wl->wl_mtx);
1739 		pool_put(&wapbl_ino_pool, wi);
1740 	}
1741 }
1742 
1743 void
1744 wapbl_unregister_inode(struct wapbl *wl, ino_t ino, mode_t mode)
1745 {
1746 	struct wapbl_ino *wi;
1747 
1748 	mutex_enter(&wl->wl_mtx);
1749 	wi = wapbl_inodetrk_get(wl, ino);
1750 	if (wi) {
1751 		WAPBL_PRINTF(WAPBL_PRINT_INODE,
1752 		    ("wapbl_unregister_inode: ino=%"PRId64"\n", ino));
1753 		KASSERT(wl->wl_inohashcnt > 0);
1754 		wl->wl_inohashcnt--;
1755 		LIST_REMOVE(wi, wi_hash);
1756 		mutex_exit(&wl->wl_mtx);
1757 
1758 		pool_put(&wapbl_ino_pool, wi);
1759 	} else {
1760 		mutex_exit(&wl->wl_mtx);
1761 	}
1762 }
1763 
1764 /****************************************************************/
1765 
1766 static __inline size_t
1767 wapbl_transaction_inodes_len(struct wapbl *wl)
1768 {
1769 	int blocklen = 1<<wl->wl_log_dev_bshift;
1770 	int iph;
1771 
1772 	/* Calculate number of inodes described in a inodelist header */
1773 	iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) /
1774 	    sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]);
1775 
1776 	KASSERT(iph > 0);
1777 
1778 	return MAX(1, howmany(wl->wl_inohashcnt, iph))*blocklen;
1779 }
1780 
1781 
1782 /* Calculate amount of space a transaction will take on disk */
1783 static size_t
1784 wapbl_transaction_len(struct wapbl *wl)
1785 {
1786 	int blocklen = 1<<wl->wl_log_dev_bshift;
1787 	size_t len;
1788 	int bph;
1789 
1790 	/* Calculate number of blocks described in a blocklist header */
1791 	bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
1792 	    sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
1793 
1794 	KASSERT(bph > 0);
1795 
1796 	len = wl->wl_bcount;
1797 	len += howmany(wl->wl_bufcount, bph)*blocklen;
1798 	len += howmany(wl->wl_dealloccnt, bph)*blocklen;
1799 	len += wapbl_transaction_inodes_len(wl);
1800 
1801 	return len;
1802 }
1803 
1804 /*
1805  * Perform commit operation
1806  *
1807  * Note that generation number incrementation needs to
1808  * be protected against racing with other invocations
1809  * of wapbl_commit.  This is ok since this routine
1810  * is only invoked from wapbl_flush
1811  */
1812 static int
1813 wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail)
1814 {
1815 	struct wapbl_wc_header *wc = wl->wl_wc_header;
1816 	struct timespec ts;
1817 	int error;
1818 	int force = 1;
1819 
1820 	/* XXX Calc checksum here, instead we do this for now */
1821 	error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, FWRITE, FSCRED);
1822 	if (error) {
1823 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1824 		    ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x "
1825 		    "returned %d\n", wl->wl_devvp->v_rdev, error));
1826 	}
1827 
1828 	wc->wc_head = head;
1829 	wc->wc_tail = tail;
1830 	wc->wc_checksum = 0;
1831 	wc->wc_version = 1;
1832 	getnanotime(&ts);
1833 	wc->wc_time = ts.tv_sec;
1834 	wc->wc_timensec = ts.tv_nsec;
1835 
1836 	WAPBL_PRINTF(WAPBL_PRINT_WRITE,
1837 	    ("wapbl_write_commit: head = %"PRIdMAX "tail = %"PRIdMAX"\n",
1838 	    (intmax_t)head, (intmax_t)tail));
1839 
1840 	/*
1841 	 * XXX if generation will rollover, then first zero
1842 	 * over second commit header before trying to write both headers.
1843 	 */
1844 
1845 	error = wapbl_write(wc, wc->wc_len, wl->wl_devvp,
1846 	    wl->wl_logpbn + wc->wc_generation % 2);
1847 	if (error)
1848 		return error;
1849 
1850 	error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, FWRITE, FSCRED);
1851 	if (error) {
1852 		WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1853 		    ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x "
1854 		    "returned %d\n", wl->wl_devvp->v_rdev, error));
1855 	}
1856 
1857 	/*
1858 	 * If the generation number was zero, write it out a second time.
1859 	 * This handles initialization and generation number rollover
1860 	 */
1861 	if (wc->wc_generation++ == 0) {
1862 		error = wapbl_write_commit(wl, head, tail);
1863 		/*
1864 		 * This panic should be able to be removed if we do the
1865 		 * zero'ing mentioned above, and we are certain to roll
1866 		 * back generation number on failure.
1867 		 */
1868 		if (error)
1869 			panic("wapbl_write_commit: error writing duplicate "
1870 			      "log header: %d\n", error);
1871 	}
1872 	return 0;
1873 }
1874 
1875 /* Returns new offset value */
1876 static int
1877 wapbl_write_blocks(struct wapbl *wl, off_t *offp)
1878 {
1879 	struct wapbl_wc_blocklist *wc =
1880 	    (struct wapbl_wc_blocklist *)wl->wl_wc_scratch;
1881 	int blocklen = 1<<wl->wl_log_dev_bshift;
1882 	int bph;
1883 	struct buf *bp;
1884 	off_t off = *offp;
1885 	int error;
1886 	size_t padding;
1887 
1888 	KASSERT(rw_write_held(&wl->wl_rwlock));
1889 
1890 	bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
1891 	    sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
1892 
1893 	bp = LIST_FIRST(&wl->wl_bufs);
1894 
1895 	while (bp) {
1896 		int cnt;
1897 		struct buf *obp = bp;
1898 
1899 		KASSERT(bp->b_flags & B_LOCKED);
1900 
1901 		wc->wc_type = WAPBL_WC_BLOCKS;
1902 		wc->wc_len = blocklen;
1903 		wc->wc_blkcount = 0;
1904 		while (bp && (wc->wc_blkcount < bph)) {
1905 			/*
1906 			 * Make sure all the physical block numbers are up to
1907 			 * date.  If this is not always true on a given
1908 			 * filesystem, then VOP_BMAP must be called.  We
1909 			 * could call VOP_BMAP here, or else in the filesystem
1910 			 * specific flush callback, although neither of those
1911 			 * solutions allow us to take the vnode lock.  If a
1912 			 * filesystem requires that we must take the vnode lock
1913 			 * to call VOP_BMAP, then we can probably do it in
1914 			 * bwrite when the vnode lock should already be held
1915 			 * by the invoking code.
1916 			 */
1917 			KASSERT((bp->b_vp->v_type == VBLK) ||
1918 				 (bp->b_blkno != bp->b_lblkno));
1919 			KASSERT(bp->b_blkno > 0);
1920 
1921 			wc->wc_blocks[wc->wc_blkcount].wc_daddr = bp->b_blkno;
1922 			wc->wc_blocks[wc->wc_blkcount].wc_dlen = bp->b_bcount;
1923 			wc->wc_len += bp->b_bcount;
1924 			wc->wc_blkcount++;
1925 			bp = LIST_NEXT(bp, b_wapbllist);
1926 		}
1927 		if (wc->wc_len % blocklen != 0) {
1928 			padding = blocklen - wc->wc_len % blocklen;
1929 			wc->wc_len += padding;
1930 		} else {
1931 			padding = 0;
1932 		}
1933 
1934 		WAPBL_PRINTF(WAPBL_PRINT_WRITE,
1935 		    ("wapbl_write_blocks: len = %u (padding %zu) off = %"PRIdMAX"\n",
1936 		    wc->wc_len, padding, (intmax_t)off));
1937 
1938 		error = wapbl_circ_write(wl, wc, blocklen, &off);
1939 		if (error)
1940 			return error;
1941 		bp = obp;
1942 		cnt = 0;
1943 		while (bp && (cnt++ < bph)) {
1944 			error = wapbl_circ_write(wl, bp->b_data,
1945 			    bp->b_bcount, &off);
1946 			if (error)
1947 				return error;
1948 			bp = LIST_NEXT(bp, b_wapbllist);
1949 		}
1950 		if (padding) {
1951 			void *zero;
1952 
1953 			zero = wapbl_malloc(padding);
1954 			memset(zero, 0, padding);
1955 			error = wapbl_circ_write(wl, zero, padding, &off);
1956 			wapbl_free(zero, padding);
1957 			if (error)
1958 				return error;
1959 		}
1960 	}
1961 	*offp = off;
1962 	return 0;
1963 }
1964 
1965 static int
1966 wapbl_write_revocations(struct wapbl *wl, off_t *offp)
1967 {
1968 	struct wapbl_wc_blocklist *wc =
1969 	    (struct wapbl_wc_blocklist *)wl->wl_wc_scratch;
1970 	int i;
1971 	int blocklen = 1<<wl->wl_log_dev_bshift;
1972 	int bph;
1973 	off_t off = *offp;
1974 	int error;
1975 
1976 	if (wl->wl_dealloccnt == 0)
1977 		return 0;
1978 
1979 	bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
1980 	    sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
1981 
1982 	i = 0;
1983 	while (i < wl->wl_dealloccnt) {
1984 		wc->wc_type = WAPBL_WC_REVOCATIONS;
1985 		wc->wc_len = blocklen;
1986 		wc->wc_blkcount = 0;
1987 		while ((i < wl->wl_dealloccnt) && (wc->wc_blkcount < bph)) {
1988 			wc->wc_blocks[wc->wc_blkcount].wc_daddr =
1989 			    wl->wl_deallocblks[i];
1990 			wc->wc_blocks[wc->wc_blkcount].wc_dlen =
1991 			    wl->wl_dealloclens[i];
1992 			wc->wc_blkcount++;
1993 			i++;
1994 		}
1995 		WAPBL_PRINTF(WAPBL_PRINT_WRITE,
1996 		    ("wapbl_write_revocations: len = %u off = %"PRIdMAX"\n",
1997 		    wc->wc_len, (intmax_t)off));
1998 		error = wapbl_circ_write(wl, wc, blocklen, &off);
1999 		if (error)
2000 			return error;
2001 	}
2002 	*offp = off;
2003 	return 0;
2004 }
2005 
2006 static int
2007 wapbl_write_inodes(struct wapbl *wl, off_t *offp)
2008 {
2009 	struct wapbl_wc_inodelist *wc =
2010 	    (struct wapbl_wc_inodelist *)wl->wl_wc_scratch;
2011 	int i;
2012 	int blocklen = 1 << wl->wl_log_dev_bshift;
2013 	off_t off = *offp;
2014 	int error;
2015 
2016 	struct wapbl_ino_head *wih;
2017 	struct wapbl_ino *wi;
2018 	int iph;
2019 
2020 	iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) /
2021 	    sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]);
2022 
2023 	i = 0;
2024 	wih = &wl->wl_inohash[0];
2025 	wi = 0;
2026 	do {
2027 		wc->wc_type = WAPBL_WC_INODES;
2028 		wc->wc_len = blocklen;
2029 		wc->wc_inocnt = 0;
2030 		wc->wc_clear = (i == 0);
2031 		while ((i < wl->wl_inohashcnt) && (wc->wc_inocnt < iph)) {
2032 			while (!wi) {
2033 				KASSERT((wih - &wl->wl_inohash[0])
2034 				    <= wl->wl_inohashmask);
2035 				wi = LIST_FIRST(wih++);
2036 			}
2037 			wc->wc_inodes[wc->wc_inocnt].wc_inumber = wi->wi_ino;
2038 			wc->wc_inodes[wc->wc_inocnt].wc_imode = wi->wi_mode;
2039 			wc->wc_inocnt++;
2040 			i++;
2041 			wi = LIST_NEXT(wi, wi_hash);
2042 		}
2043 		WAPBL_PRINTF(WAPBL_PRINT_WRITE,
2044 		    ("wapbl_write_inodes: len = %u off = %"PRIdMAX"\n",
2045 		    wc->wc_len, (intmax_t)off));
2046 		error = wapbl_circ_write(wl, wc, blocklen, &off);
2047 		if (error)
2048 			return error;
2049 	} while (i < wl->wl_inohashcnt);
2050 
2051 	*offp = off;
2052 	return 0;
2053 }
2054 
2055 #endif /* _KERNEL */
2056 
2057 /****************************************************************/
2058 
2059 struct wapbl_blk {
2060 	LIST_ENTRY(wapbl_blk) wb_hash;
2061 	daddr_t wb_blk;
2062 	off_t wb_off; /* Offset of this block in the log */
2063 };
2064 #define	WAPBL_BLKPOOL_MIN 83
2065 
2066 static void
2067 wapbl_blkhash_init(struct wapbl_replay *wr, u_int size)
2068 {
2069 	if (size < WAPBL_BLKPOOL_MIN)
2070 		size = WAPBL_BLKPOOL_MIN;
2071 	KASSERT(wr->wr_blkhash == 0);
2072 #ifdef _KERNEL
2073 	wr->wr_blkhash = hashinit(size, HASH_LIST, true, &wr->wr_blkhashmask);
2074 #else /* ! _KERNEL */
2075 	/* Manually implement hashinit */
2076 	{
2077 		unsigned long i, hashsize;
2078 		for (hashsize = 1; hashsize < size; hashsize <<= 1)
2079 			continue;
2080 		wr->wr_blkhash = wapbl_malloc(hashsize * sizeof(*wr->wr_blkhash));
2081 		for (i = 0; i < wr->wr_blkhashmask; i++)
2082 			LIST_INIT(&wr->wr_blkhash[i]);
2083 		wr->wr_blkhashmask = hashsize - 1;
2084 	}
2085 #endif /* ! _KERNEL */
2086 }
2087 
2088 static void
2089 wapbl_blkhash_free(struct wapbl_replay *wr)
2090 {
2091 	KASSERT(wr->wr_blkhashcnt == 0);
2092 #ifdef _KERNEL
2093 	hashdone(wr->wr_blkhash, HASH_LIST, wr->wr_blkhashmask);
2094 #else /* ! _KERNEL */
2095 	wapbl_free(wr->wr_blkhash,
2096 	    (wr->wr_blkhashmask + 1) * sizeof(*wr->wr_blkhash));
2097 #endif /* ! _KERNEL */
2098 }
2099 
2100 static struct wapbl_blk *
2101 wapbl_blkhash_get(struct wapbl_replay *wr, daddr_t blk)
2102 {
2103 	struct wapbl_blk_head *wbh;
2104 	struct wapbl_blk *wb;
2105 	wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask];
2106 	LIST_FOREACH(wb, wbh, wb_hash) {
2107 		if (blk == wb->wb_blk)
2108 			return wb;
2109 	}
2110 	return 0;
2111 }
2112 
2113 static void
2114 wapbl_blkhash_ins(struct wapbl_replay *wr, daddr_t blk, off_t off)
2115 {
2116 	struct wapbl_blk_head *wbh;
2117 	struct wapbl_blk *wb;
2118 	wb = wapbl_blkhash_get(wr, blk);
2119 	if (wb) {
2120 		KASSERT(wb->wb_blk == blk);
2121 		wb->wb_off = off;
2122 	} else {
2123 		wb = wapbl_malloc(sizeof(*wb));
2124 		wb->wb_blk = blk;
2125 		wb->wb_off = off;
2126 		wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask];
2127 		LIST_INSERT_HEAD(wbh, wb, wb_hash);
2128 		wr->wr_blkhashcnt++;
2129 	}
2130 }
2131 
2132 static void
2133 wapbl_blkhash_rem(struct wapbl_replay *wr, daddr_t blk)
2134 {
2135 	struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk);
2136 	if (wb) {
2137 		KASSERT(wr->wr_blkhashcnt > 0);
2138 		wr->wr_blkhashcnt--;
2139 		LIST_REMOVE(wb, wb_hash);
2140 		wapbl_free(wb, sizeof(*wb));
2141 	}
2142 }
2143 
2144 static void
2145 wapbl_blkhash_clear(struct wapbl_replay *wr)
2146 {
2147 	unsigned long i;
2148 	for (i = 0; i <= wr->wr_blkhashmask; i++) {
2149 		struct wapbl_blk *wb;
2150 
2151 		while ((wb = LIST_FIRST(&wr->wr_blkhash[i]))) {
2152 			KASSERT(wr->wr_blkhashcnt > 0);
2153 			wr->wr_blkhashcnt--;
2154 			LIST_REMOVE(wb, wb_hash);
2155 			wapbl_free(wb, sizeof(*wb));
2156 		}
2157 	}
2158 	KASSERT(wr->wr_blkhashcnt == 0);
2159 }
2160 
2161 /****************************************************************/
2162 
2163 static int
2164 wapbl_circ_read(struct wapbl_replay *wr, void *data, size_t len, off_t *offp)
2165 {
2166 	size_t slen;
2167 	off_t off = *offp;
2168 	int error;
2169 
2170 	KASSERT(((len >> wr->wr_log_dev_bshift) <<
2171 	    wr->wr_log_dev_bshift) == len);
2172 	if (off < wr->wr_circ_off)
2173 		off = wr->wr_circ_off;
2174 	slen = wr->wr_circ_off + wr->wr_circ_size - off;
2175 	if (slen < len) {
2176 		error = wapbl_read(data, slen, wr->wr_devvp,
2177 		    wr->wr_logpbn + (off >> wr->wr_log_dev_bshift));
2178 		if (error)
2179 			return error;
2180 		data = (uint8_t *)data + slen;
2181 		len -= slen;
2182 		off = wr->wr_circ_off;
2183 	}
2184 	error = wapbl_read(data, len, wr->wr_devvp,
2185 	    wr->wr_logpbn + (off >> wr->wr_log_dev_bshift));
2186 	if (error)
2187 		return error;
2188 	off += len;
2189 	if (off >= wr->wr_circ_off + wr->wr_circ_size)
2190 		off = wr->wr_circ_off;
2191 	*offp = off;
2192 	return 0;
2193 }
2194 
2195 static void
2196 wapbl_circ_advance(struct wapbl_replay *wr, size_t len, off_t *offp)
2197 {
2198 	size_t slen;
2199 	off_t off = *offp;
2200 
2201 	KASSERT(((len >> wr->wr_log_dev_bshift) <<
2202 	    wr->wr_log_dev_bshift) == len);
2203 
2204 	if (off < wr->wr_circ_off)
2205 		off = wr->wr_circ_off;
2206 	slen = wr->wr_circ_off + wr->wr_circ_size - off;
2207 	if (slen < len) {
2208 		len -= slen;
2209 		off = wr->wr_circ_off;
2210 	}
2211 	off += len;
2212 	if (off >= wr->wr_circ_off + wr->wr_circ_size)
2213 		off = wr->wr_circ_off;
2214 	*offp = off;
2215 }
2216 
2217 /****************************************************************/
2218 
2219 int
2220 wapbl_replay_start(struct wapbl_replay **wrp, struct vnode *vp,
2221 	daddr_t off, size_t count, size_t blksize)
2222 {
2223 	struct wapbl_replay *wr;
2224 	int error;
2225 	struct vnode *devvp;
2226 	daddr_t logpbn;
2227 	uint8_t *scratch;
2228 	struct wapbl_wc_header *wch;
2229 	struct wapbl_wc_header *wch2;
2230 	/* Use this until we read the actual log header */
2231 	int log_dev_bshift = DEV_BSHIFT;
2232 	size_t used;
2233 
2234 	WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
2235 	    ("wapbl_replay_start: vp=%p off=%"PRId64 " count=%zu blksize=%zu\n",
2236 	    vp, off, count, blksize));
2237 
2238 	if (off < 0)
2239 		return EINVAL;
2240 
2241 	if (blksize < DEV_BSIZE)
2242 		return EINVAL;
2243 	if (blksize % DEV_BSIZE)
2244 		return EINVAL;
2245 
2246 #ifdef _KERNEL
2247 #if 0
2248 	/* XXX vp->v_size isn't reliably set for VBLK devices,
2249 	 * especially root.  However, we might still want to verify
2250 	 * that the full load is readable */
2251 	if ((off + count) * blksize > vp->v_size)
2252 		return EINVAL;
2253 #endif
2254 
2255 	if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, 0)) != 0) {
2256 		return error;
2257 	}
2258 #else /* ! _KERNEL */
2259 	devvp = vp;
2260 	logpbn = off;
2261 #endif /* ! _KERNEL */
2262 
2263 	scratch = wapbl_malloc(MAXBSIZE);
2264 
2265 	error = wapbl_read(scratch, 2<<log_dev_bshift, devvp, logpbn);
2266 	if (error)
2267 		goto errout;
2268 
2269 	wch = (struct wapbl_wc_header *)scratch;
2270 	wch2 =
2271 	    (struct wapbl_wc_header *)(scratch + (1<<log_dev_bshift));
2272 	/* XXX verify checksums and magic numbers */
2273 	if (wch->wc_type != WAPBL_WC_HEADER) {
2274 		printf("Unrecognized wapbl magic: 0x%08x\n", wch->wc_type);
2275 		error = EFTYPE;
2276 		goto errout;
2277 	}
2278 
2279 	if (wch2->wc_generation > wch->wc_generation)
2280 		wch = wch2;
2281 
2282 	wr = wapbl_calloc(1, sizeof(*wr));
2283 
2284 	wr->wr_logvp = vp;
2285 	wr->wr_devvp = devvp;
2286 	wr->wr_logpbn = logpbn;
2287 
2288 	wr->wr_scratch = scratch;
2289 
2290 	wr->wr_log_dev_bshift = wch->wc_log_dev_bshift;
2291 	wr->wr_fs_dev_bshift = wch->wc_fs_dev_bshift;
2292 	wr->wr_circ_off = wch->wc_circ_off;
2293 	wr->wr_circ_size = wch->wc_circ_size;
2294 	wr->wr_generation = wch->wc_generation;
2295 
2296 	used = wapbl_space_used(wch->wc_circ_size, wch->wc_head, wch->wc_tail);
2297 
2298 	WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
2299 	    ("wapbl_replay: head=%"PRId64" tail=%"PRId64" off=%"PRId64
2300 	    " len=%"PRId64" used=%zu\n",
2301 	    wch->wc_head, wch->wc_tail, wch->wc_circ_off,
2302 	    wch->wc_circ_size, used));
2303 
2304 	wapbl_blkhash_init(wr, (used >> wch->wc_fs_dev_bshift));
2305 
2306 	error = wapbl_replay_process(wr, wch->wc_head, wch->wc_tail);
2307 	if (error) {
2308 		wapbl_replay_stop(wr);
2309 		wapbl_replay_free(wr);
2310 		return error;
2311 	}
2312 
2313 	*wrp = wr;
2314 	return 0;
2315 
2316  errout:
2317 	wapbl_free(scratch, MAXBSIZE);
2318 	return error;
2319 }
2320 
2321 void
2322 wapbl_replay_stop(struct wapbl_replay *wr)
2323 {
2324 
2325 	if (!wapbl_replay_isopen(wr))
2326 		return;
2327 
2328 	WAPBL_PRINTF(WAPBL_PRINT_REPLAY, ("wapbl_replay_stop called\n"));
2329 
2330 	wapbl_free(wr->wr_scratch, MAXBSIZE);
2331 	wr->wr_scratch = NULL;
2332 
2333 	wr->wr_logvp = NULL;
2334 
2335 	wapbl_blkhash_clear(wr);
2336 	wapbl_blkhash_free(wr);
2337 }
2338 
2339 void
2340 wapbl_replay_free(struct wapbl_replay *wr)
2341 {
2342 
2343 	KDASSERT(!wapbl_replay_isopen(wr));
2344 
2345 	if (wr->wr_inodes)
2346 		wapbl_free(wr->wr_inodes,
2347 		    wr->wr_inodescnt * sizeof(wr->wr_inodes[0]));
2348 	wapbl_free(wr, sizeof(*wr));
2349 }
2350 
2351 #ifdef _KERNEL
2352 int
2353 wapbl_replay_isopen1(struct wapbl_replay *wr)
2354 {
2355 
2356 	return wapbl_replay_isopen(wr);
2357 }
2358 #endif
2359 
2360 static void
2361 wapbl_replay_process_blocks(struct wapbl_replay *wr, off_t *offp)
2362 {
2363 	struct wapbl_wc_blocklist *wc =
2364 	    (struct wapbl_wc_blocklist *)wr->wr_scratch;
2365 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2366 	int i, j, n;
2367 
2368 	for (i = 0; i < wc->wc_blkcount; i++) {
2369 		/*
2370 		 * Enter each physical block into the hashtable independently.
2371 		 */
2372 		n = wc->wc_blocks[i].wc_dlen >> wr->wr_fs_dev_bshift;
2373 		for (j = 0; j < n; j++) {
2374 			wapbl_blkhash_ins(wr, wc->wc_blocks[i].wc_daddr + j,
2375 			    *offp);
2376 			wapbl_circ_advance(wr, fsblklen, offp);
2377 		}
2378 	}
2379 }
2380 
2381 static void
2382 wapbl_replay_process_revocations(struct wapbl_replay *wr)
2383 {
2384 	struct wapbl_wc_blocklist *wc =
2385 	    (struct wapbl_wc_blocklist *)wr->wr_scratch;
2386 	int i, j, n;
2387 
2388 	for (i = 0; i < wc->wc_blkcount; i++) {
2389 		/*
2390 		 * Remove any blocks found from the hashtable.
2391 		 */
2392 		n = wc->wc_blocks[i].wc_dlen >> wr->wr_fs_dev_bshift;
2393 		for (j = 0; j < n; j++)
2394 			wapbl_blkhash_rem(wr, wc->wc_blocks[i].wc_daddr + j);
2395 	}
2396 }
2397 
2398 static void
2399 wapbl_replay_process_inodes(struct wapbl_replay *wr, off_t oldoff, off_t newoff)
2400 {
2401 	struct wapbl_wc_inodelist *wc =
2402 	    (struct wapbl_wc_inodelist *)wr->wr_scratch;
2403 	void *new_inodes;
2404 	const size_t oldsize = wr->wr_inodescnt * sizeof(wr->wr_inodes[0]);
2405 
2406 	KASSERT(sizeof(wr->wr_inodes[0]) == sizeof(wc->wc_inodes[0]));
2407 
2408 	/*
2409 	 * Keep track of where we found this so location won't be
2410 	 * overwritten.
2411 	 */
2412 	if (wc->wc_clear) {
2413 		wr->wr_inodestail = oldoff;
2414 		wr->wr_inodescnt = 0;
2415 		if (wr->wr_inodes != NULL) {
2416 			wapbl_free(wr->wr_inodes, oldsize);
2417 			wr->wr_inodes = NULL;
2418 		}
2419 	}
2420 	wr->wr_inodeshead = newoff;
2421 	if (wc->wc_inocnt == 0)
2422 		return;
2423 
2424 	new_inodes = wapbl_malloc((wr->wr_inodescnt + wc->wc_inocnt) *
2425 	    sizeof(wr->wr_inodes[0]));
2426 	if (wr->wr_inodes != NULL) {
2427 		memcpy(new_inodes, wr->wr_inodes, oldsize);
2428 		wapbl_free(wr->wr_inodes, oldsize);
2429 	}
2430 	wr->wr_inodes = new_inodes;
2431 	memcpy(&wr->wr_inodes[wr->wr_inodescnt], wc->wc_inodes,
2432 	    wc->wc_inocnt * sizeof(wr->wr_inodes[0]));
2433 	wr->wr_inodescnt += wc->wc_inocnt;
2434 }
2435 
2436 static int
2437 wapbl_replay_process(struct wapbl_replay *wr, off_t head, off_t tail)
2438 {
2439 	off_t off;
2440 	int error;
2441 
2442 	int logblklen = 1 << wr->wr_log_dev_bshift;
2443 
2444 	wapbl_blkhash_clear(wr);
2445 
2446 	off = tail;
2447 	while (off != head) {
2448 		struct wapbl_wc_null *wcn;
2449 		off_t saveoff = off;
2450 		error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off);
2451 		if (error)
2452 			goto errout;
2453 		wcn = (struct wapbl_wc_null *)wr->wr_scratch;
2454 		switch (wcn->wc_type) {
2455 		case WAPBL_WC_BLOCKS:
2456 			wapbl_replay_process_blocks(wr, &off);
2457 			break;
2458 
2459 		case WAPBL_WC_REVOCATIONS:
2460 			wapbl_replay_process_revocations(wr);
2461 			break;
2462 
2463 		case WAPBL_WC_INODES:
2464 			wapbl_replay_process_inodes(wr, saveoff, off);
2465 			break;
2466 
2467 		default:
2468 			printf("Unrecognized wapbl type: 0x%08x\n",
2469 			       wcn->wc_type);
2470  			error = EFTYPE;
2471 			goto errout;
2472 		}
2473 		wapbl_circ_advance(wr, wcn->wc_len, &saveoff);
2474 		if (off != saveoff) {
2475 			printf("wapbl_replay: corrupted records\n");
2476 			error = EFTYPE;
2477 			goto errout;
2478 		}
2479 	}
2480 	return 0;
2481 
2482  errout:
2483 	wapbl_blkhash_clear(wr);
2484 	return error;
2485 }
2486 
2487 #if 0
2488 int
2489 wapbl_replay_verify(struct wapbl_replay *wr, struct vnode *fsdevvp)
2490 {
2491 	off_t off;
2492 	int mismatchcnt = 0;
2493 	int logblklen = 1 << wr->wr_log_dev_bshift;
2494 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2495 	void *scratch1 = wapbl_malloc(MAXBSIZE);
2496 	void *scratch2 = wapbl_malloc(MAXBSIZE);
2497 	int error = 0;
2498 
2499 	KDASSERT(wapbl_replay_isopen(wr));
2500 
2501 	off = wch->wc_tail;
2502 	while (off != wch->wc_head) {
2503 		struct wapbl_wc_null *wcn;
2504 #ifdef DEBUG
2505 		off_t saveoff = off;
2506 #endif
2507 		error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off);
2508 		if (error)
2509 			goto out;
2510 		wcn = (struct wapbl_wc_null *)wr->wr_scratch;
2511 		switch (wcn->wc_type) {
2512 		case WAPBL_WC_BLOCKS:
2513 			{
2514 				struct wapbl_wc_blocklist *wc =
2515 				    (struct wapbl_wc_blocklist *)wr->wr_scratch;
2516 				int i;
2517 				for (i = 0; i < wc->wc_blkcount; i++) {
2518 					int foundcnt = 0;
2519 					int dirtycnt = 0;
2520 					int j, n;
2521 					/*
2522 					 * Check each physical block into the
2523 					 * hashtable independently
2524 					 */
2525 					n = wc->wc_blocks[i].wc_dlen >>
2526 					    wch->wc_fs_dev_bshift;
2527 					for (j = 0; j < n; j++) {
2528 						struct wapbl_blk *wb =
2529 						   wapbl_blkhash_get(wr,
2530 						   wc->wc_blocks[i].wc_daddr + j);
2531 						if (wb && (wb->wb_off == off)) {
2532 							foundcnt++;
2533 							error =
2534 							    wapbl_circ_read(wr,
2535 							    scratch1, fsblklen,
2536 							    &off);
2537 							if (error)
2538 								goto out;
2539 							error =
2540 							    wapbl_read(scratch2,
2541 							    fsblklen, fsdevvp,
2542 							    wb->wb_blk);
2543 							if (error)
2544 								goto out;
2545 							if (memcmp(scratch1,
2546 								   scratch2,
2547 								   fsblklen)) {
2548 								printf(
2549 		"wapbl_verify: mismatch block %"PRId64" at off %"PRIdMAX"\n",
2550 		wb->wb_blk, (intmax_t)off);
2551 								dirtycnt++;
2552 								mismatchcnt++;
2553 							}
2554 						} else {
2555 							wapbl_circ_advance(wr,
2556 							    fsblklen, &off);
2557 						}
2558 					}
2559 #if 0
2560 					/*
2561 					 * If all of the blocks in an entry
2562 					 * are clean, then remove all of its
2563 					 * blocks from the hashtable since they
2564 					 * never will need replay.
2565 					 */
2566 					if ((foundcnt != 0) &&
2567 					    (dirtycnt == 0)) {
2568 						off = saveoff;
2569 						wapbl_circ_advance(wr,
2570 						    logblklen, &off);
2571 						for (j = 0; j < n; j++) {
2572 							struct wapbl_blk *wb =
2573 							   wapbl_blkhash_get(wr,
2574 							   wc->wc_blocks[i].wc_daddr + j);
2575 							if (wb &&
2576 							  (wb->wb_off == off)) {
2577 								wapbl_blkhash_rem(wr, wb->wb_blk);
2578 							}
2579 							wapbl_circ_advance(wr,
2580 							    fsblklen, &off);
2581 						}
2582 					}
2583 #endif
2584 				}
2585 			}
2586 			break;
2587 		case WAPBL_WC_REVOCATIONS:
2588 		case WAPBL_WC_INODES:
2589 			break;
2590 		default:
2591 			KASSERT(0);
2592 		}
2593 #ifdef DEBUG
2594 		wapbl_circ_advance(wr, wcn->wc_len, &saveoff);
2595 		KASSERT(off == saveoff);
2596 #endif
2597 	}
2598  out:
2599 	wapbl_free(scratch1, MAXBSIZE);
2600 	wapbl_free(scratch2, MAXBSIZE);
2601 	if (!error && mismatchcnt)
2602 		error = EFTYPE;
2603 	return error;
2604 }
2605 #endif
2606 
2607 int
2608 wapbl_replay_write(struct wapbl_replay *wr, struct vnode *fsdevvp)
2609 {
2610 	struct wapbl_blk *wb;
2611 	size_t i;
2612 	off_t off;
2613 	void *scratch;
2614 	int error = 0;
2615 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2616 
2617 	KDASSERT(wapbl_replay_isopen(wr));
2618 
2619 	scratch = wapbl_malloc(MAXBSIZE);
2620 
2621 	for (i = 0; i < wr->wr_blkhashmask; ++i) {
2622 		LIST_FOREACH(wb, &wr->wr_blkhash[i], wb_hash) {
2623 			off = wb->wb_off;
2624 			error = wapbl_circ_read(wr, scratch, fsblklen, &off);
2625 			if (error)
2626 				break;
2627 			error = wapbl_write(scratch, fsblklen, fsdevvp,
2628 			    wb->wb_blk);
2629 			if (error)
2630 				break;
2631 		}
2632 	}
2633 
2634 	wapbl_free(scratch, MAXBSIZE);
2635 	return error;
2636 }
2637 
2638 int
2639 wapbl_replay_can_read(struct wapbl_replay *wr, daddr_t blk, long len)
2640 {
2641 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2642 
2643 	KDASSERT(wapbl_replay_isopen(wr));
2644 	KASSERT((len % fsblklen) == 0);
2645 
2646 	while (len != 0) {
2647 		struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk);
2648 		if (wb)
2649 			return 1;
2650 		len -= fsblklen;
2651 	}
2652 	return 0;
2653 }
2654 
2655 int
2656 wapbl_replay_read(struct wapbl_replay *wr, void *data, daddr_t blk, long len)
2657 {
2658 	int fsblklen = 1 << wr->wr_fs_dev_bshift;
2659 
2660 	KDASSERT(wapbl_replay_isopen(wr));
2661 
2662 	KASSERT((len % fsblklen) == 0);
2663 
2664 	while (len != 0) {
2665 		struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk);
2666 		if (wb) {
2667 			off_t off = wb->wb_off;
2668 			int error;
2669 			error = wapbl_circ_read(wr, data, fsblklen, &off);
2670 			if (error)
2671 				return error;
2672 		}
2673 		data = (uint8_t *)data + fsblklen;
2674 		len -= fsblklen;
2675 		blk++;
2676 	}
2677 	return 0;
2678 }
2679