FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_wapbl.c
1 /* $NetBSD: vfs_wapbl.c,v 1.3.8.5 2011/03/07 04:09:28 riz 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 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.3.8.5 2011/03/07 04:09:28 riz Exp $");
37
38 #include <sys/param.h>
39
40 #ifdef _KERNEL
41 #include <sys/param.h>
42 #include <sys/namei.h>
43 #include <sys/proc.h>
44 #include <sys/sysctl.h>
45 #include <sys/uio.h>
46 #include <sys/vnode.h>
47 #include <sys/file.h>
48 #include <sys/malloc.h>
49 #include <sys/resourcevar.h>
50 #include <sys/conf.h>
51 #include <sys/mount.h>
52 #include <sys/kernel.h>
53 #include <sys/kauth.h>
54 #include <sys/mutex.h>
55 #include <sys/atomic.h>
56 #include <sys/wapbl.h>
57
58 #if WAPBL_UVM_ALLOC
59 #include <uvm/uvm.h>
60 #endif
61
62 #include <miscfs/specfs/specdev.h>
63
64 MALLOC_JUSTDEFINE(M_WAPBL, "wapbl", "write-ahead physical block logging");
65 #define wapbl_malloc(s) malloc((s), M_WAPBL, M_WAITOK)
66 #define wapbl_free(a) free((a), M_WAPBL)
67 #define wapbl_calloc(n, s) malloc((n)*(s), M_WAPBL, M_WAITOK | M_ZERO)
68
69 static struct sysctllog *wapbl_sysctl;
70 static int wapbl_flush_disk_cache = 1;
71 static int wapbl_verbose_commit = 0;
72
73 #else /* !_KERNEL */
74 #include <assert.h>
75 #include <errno.h>
76 #include <stdio.h>
77 #include <stdbool.h>
78 #include <stdlib.h>
79 #include <string.h>
80
81 #include <sys/time.h>
82 #include <sys/wapbl.h>
83
84 #define KDASSERT(x) assert(x)
85 #define KASSERT(x) assert(x)
86 #define wapbl_malloc(s) malloc(s)
87 #define wapbl_free(a) free(a)
88 #define wapbl_calloc(n, s) calloc((n), (s))
89
90 #endif /* !_KERNEL */
91
92 /*
93 * INTERNAL DATA STRUCTURES
94 */
95
96 /*
97 * This structure holds per-mount log information.
98 *
99 * Legend: a = atomic access only
100 * r = read-only after init
101 * l = rwlock held
102 * m = mutex held
103 * lm = rwlock held writing or mutex held
104 * u = unlocked access ok
105 * b = bufcache_lock held
106 */
107 struct wapbl {
108 struct vnode *wl_logvp; /* r: log here */
109 struct vnode *wl_devvp; /* r: log on this device */
110 struct mount *wl_mount; /* r: mountpoint wl is associated with */
111 daddr_t wl_logpbn; /* r: Physical block number of start of log */
112 int wl_log_dev_bshift; /* r: logarithm of device block size of log
113 device */
114 int wl_fs_dev_bshift; /* r: logarithm of device block size of
115 filesystem device */
116
117 unsigned wl_lock_count; /* m: Count of transactions in progress */
118
119 size_t wl_circ_size; /* r: Number of bytes in buffer of log */
120 size_t wl_circ_off; /* r: Number of bytes reserved at start */
121
122 size_t wl_bufcount_max; /* r: Number of buffers reserved for log */
123 size_t wl_bufbytes_max; /* r: Number of buf bytes reserved for log */
124
125 off_t wl_head; /* l: Byte offset of log head */
126 off_t wl_tail; /* l: Byte offset of log tail */
127 /*
128 * head == tail == 0 means log is empty
129 * head == tail != 0 means log is full
130 * see assertions in wapbl_advance() for other boundary conditions.
131 * only truncate moves the tail, except when flush sets it to
132 * wl_header_size only flush moves the head, except when truncate
133 * sets it to 0.
134 */
135
136 struct wapbl_wc_header *wl_wc_header; /* l */
137 void *wl_wc_scratch; /* l: scratch space (XXX: por que?!?) */
138
139 kmutex_t wl_mtx; /* u: short-term lock */
140 krwlock_t wl_rwlock; /* u: File system transaction lock */
141
142 /*
143 * Must be held while accessing
144 * wl_count or wl_bufs or head or tail
145 */
146
147 /*
148 * Callback called from within the flush routine to flush any extra
149 * bits. Note that flush may be skipped without calling this if
150 * there are no outstanding buffers in the transaction.
151 */
152 wapbl_flush_fn_t wl_flush; /* r */
153 wapbl_flush_fn_t wl_flush_abort;/* r */
154
155 size_t wl_bufbytes; /* m: Byte count of pages in wl_bufs */
156 size_t wl_bufcount; /* m: Count of buffers in wl_bufs */
157 size_t wl_bcount; /* m: Total bcount of wl_bufs */
158
159 LIST_HEAD(, buf) wl_bufs; /* m: Buffers in current transaction */
160
161 kcondvar_t wl_reclaimable_cv; /* m (obviously) */
162 size_t wl_reclaimable_bytes; /* m: Amount of space available for
163 reclamation by truncate */
164 int wl_error_count; /* m: # of wl_entries with errors */
165 size_t wl_reserved_bytes; /* never truncate log smaller than this */
166
167 #ifdef WAPBL_DEBUG_BUFBYTES
168 size_t wl_unsynced_bufbytes; /* Byte count of unsynced buffers */
169 #endif
170
171 daddr_t *wl_deallocblks;/* lm: address of block */
172 int *wl_dealloclens; /* lm: size of block */
173 int wl_dealloccnt; /* lm: total count */
174 int wl_dealloclim; /* l: max count */
175
176 /* hashtable of inode numbers for allocated but unlinked inodes */
177 /* synch ??? */
178 LIST_HEAD(wapbl_ino_head, wapbl_ino) *wl_inohash;
179 u_long wl_inohashmask;
180 int wl_inohashcnt;
181
182 SIMPLEQ_HEAD(, wapbl_entry) wl_entries; /* On disk transaction
183 accounting */
184 };
185
186 #ifdef WAPBL_DEBUG_PRINT
187 int wapbl_debug_print = WAPBL_DEBUG_PRINT;
188 #endif
189
190 /****************************************************************/
191 #ifdef _KERNEL
192
193 #ifdef WAPBL_DEBUG
194 struct wapbl *wapbl_debug_wl;
195 #endif
196
197 static int wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail);
198 static int wapbl_write_blocks(struct wapbl *wl, off_t *offp);
199 static int wapbl_write_revocations(struct wapbl *wl, off_t *offp);
200 static int wapbl_write_inodes(struct wapbl *wl, off_t *offp);
201 #endif /* _KERNEL */
202
203 static int wapbl_replay_prescan(struct wapbl_replay *wr);
204 static int wapbl_replay_get_inodes(struct wapbl_replay *wr);
205
206 static __inline size_t wapbl_space_free(size_t avail, off_t head,
207 off_t tail);
208 static __inline size_t wapbl_space_used(size_t avail, off_t head,
209 off_t tail);
210
211 #ifdef _KERNEL
212
213 #define WAPBL_INODETRK_SIZE 83
214 static int wapbl_ino_pool_refcount;
215 static struct pool wapbl_ino_pool;
216 struct wapbl_ino {
217 LIST_ENTRY(wapbl_ino) wi_hash;
218 ino_t wi_ino;
219 mode_t wi_mode;
220 };
221
222 static void wapbl_inodetrk_init(struct wapbl *wl, u_int size);
223 static void wapbl_inodetrk_free(struct wapbl *wl);
224 static struct wapbl_ino *wapbl_inodetrk_get(struct wapbl *wl, ino_t ino);
225
226 static size_t wapbl_transaction_len(struct wapbl *wl);
227 static __inline size_t wapbl_transaction_inodes_len(struct wapbl *wl);
228
229 /*
230 * This is useful for debugging. If set, the log will
231 * only be truncated when necessary.
232 */
233 int wapbl_lazy_truncate = 0;
234
235 struct wapbl_ops wapbl_ops = {
236 .wo_wapbl_discard = wapbl_discard,
237 .wo_wapbl_replay_isopen = wapbl_replay_isopen1,
238 .wo_wapbl_replay_read = wapbl_replay_read,
239 .wo_wapbl_add_buf = wapbl_add_buf,
240 .wo_wapbl_remove_buf = wapbl_remove_buf,
241 .wo_wapbl_resize_buf = wapbl_resize_buf,
242 .wo_wapbl_begin = wapbl_begin,
243 .wo_wapbl_end = wapbl_end,
244 .wo_wapbl_junlock_assert= wapbl_junlock_assert,
245
246 /* XXX: the following is only used to say "this is a wapbl buf" */
247 .wo_wapbl_biodone = wapbl_biodone,
248 };
249
250 void
251 wapbl_init()
252 {
253 int rv;
254 const struct sysctlnode *rnode, *cnode;
255
256 malloc_type_attach(M_WAPBL);
257
258 wapbl_sysctl = NULL;
259
260 rv = sysctl_createv(&wapbl_sysctl, 0, NULL, &rnode,
261 CTLFLAG_PERMANENT,
262 CTLTYPE_NODE, "vfs", NULL,
263 NULL, 0, NULL, 0,
264 CTL_VFS, CTL_EOL);
265 if (rv)
266 return;
267
268 rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &rnode,
269 CTLFLAG_PERMANENT,
270 CTLTYPE_NODE, "wapbl",
271 SYSCTL_DESCR("WAPBL journaling options"),
272 NULL, 0, NULL, 0,
273 CTL_CREATE, CTL_EOL);
274 if (rv)
275 return;
276
277 rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &cnode,
278 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
279 CTLTYPE_INT, "flush_disk_cache",
280 SYSCTL_DESCR("flush disk cache"),
281 NULL, 0, &wapbl_flush_disk_cache, 0,
282 CTL_CREATE, CTL_EOL);
283 if (rv)
284 return;
285
286 rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &cnode,
287 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
288 CTLTYPE_INT, "verbose_commit",
289 SYSCTL_DESCR("show time and size of wapbl log commits"),
290 NULL, 0, &wapbl_verbose_commit, 0,
291 CTL_CREATE, CTL_EOL);
292 }
293
294 int
295 wapbl_start(struct wapbl ** wlp, struct mount *mp, struct vnode *vp,
296 daddr_t off, size_t count, size_t blksize, struct wapbl_replay *wr,
297 wapbl_flush_fn_t flushfn, wapbl_flush_fn_t flushabortfn)
298 {
299 struct wapbl *wl;
300 struct vnode *devvp;
301 daddr_t logpbn;
302 int error;
303 int log_dev_bshift = DEV_BSHIFT;
304 int fs_dev_bshift = DEV_BSHIFT;
305 int run;
306
307 WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_start: vp=%p off=%" PRId64
308 " count=%zu blksize=%zu\n", vp, off, count, blksize));
309
310 if (log_dev_bshift > fs_dev_bshift) {
311 WAPBL_PRINTF(WAPBL_PRINT_OPEN,
312 ("wapbl: log device's block size cannot be larger "
313 "than filesystem's\n"));
314 /*
315 * Not currently implemented, although it could be if
316 * needed someday.
317 */
318 return ENOSYS;
319 }
320
321 if (off < 0)
322 return EINVAL;
323
324 if (blksize < DEV_BSIZE)
325 return EINVAL;
326 if (blksize % DEV_BSIZE)
327 return EINVAL;
328
329 /* XXXTODO: verify that the full load is writable */
330
331 /*
332 * XXX check for minimum log size
333 * minimum is governed by minimum amount of space
334 * to complete a transaction. (probably truncate)
335 */
336 /* XXX for now pick something minimal */
337 if ((count * blksize) < MAXPHYS) {
338 return ENOSPC;
339 }
340
341 if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, &run)) != 0) {
342 return error;
343 }
344
345 wl = wapbl_calloc(1, sizeof(*wl));
346 rw_init(&wl->wl_rwlock);
347 mutex_init(&wl->wl_mtx, MUTEX_DEFAULT, IPL_NONE);
348 cv_init(&wl->wl_reclaimable_cv, "wapblrec");
349 LIST_INIT(&wl->wl_bufs);
350 SIMPLEQ_INIT(&wl->wl_entries);
351
352 wl->wl_logvp = vp;
353 wl->wl_devvp = devvp;
354 wl->wl_mount = mp;
355 wl->wl_logpbn = logpbn;
356 wl->wl_log_dev_bshift = log_dev_bshift;
357 wl->wl_fs_dev_bshift = fs_dev_bshift;
358
359 wl->wl_flush = flushfn;
360 wl->wl_flush_abort = flushabortfn;
361
362 /* Reserve two log device blocks for the commit headers */
363 wl->wl_circ_off = 2<<wl->wl_log_dev_bshift;
364 wl->wl_circ_size = ((count * blksize) - wl->wl_circ_off);
365 /* truncate the log usage to a multiple of log_dev_bshift */
366 wl->wl_circ_size >>= wl->wl_log_dev_bshift;
367 wl->wl_circ_size <<= wl->wl_log_dev_bshift;
368
369 /*
370 * wl_bufbytes_max limits the size of the in memory transaction space.
371 * - Since buffers are allocated and accounted for in units of
372 * PAGE_SIZE it is required to be a multiple of PAGE_SIZE
373 * (i.e. 1<<PAGE_SHIFT)
374 * - Since the log device has to be written in units of
375 * 1<<wl_log_dev_bshift it is required to be a mulitple of
376 * 1<<wl_log_dev_bshift.
377 * - Since filesystem will provide data in units of 1<<wl_fs_dev_bshift,
378 * it is convenient to be a multiple of 1<<wl_fs_dev_bshift.
379 * Therefore it must be multiple of the least common multiple of those
380 * three quantities. Fortunately, all of those quantities are
381 * guaranteed to be a power of two, and the least common multiple of
382 * a set of numbers which are all powers of two is simply the maximum
383 * of those numbers. Finally, the maximum logarithm of a power of two
384 * is the same as the log of the maximum power of two. So we can do
385 * the following operations to size wl_bufbytes_max:
386 */
387
388 /* XXX fix actual number of pages reserved per filesystem. */
389 wl->wl_bufbytes_max = MIN(wl->wl_circ_size, buf_memcalc() / 2);
390
391 /* Round wl_bufbytes_max to the largest power of two constraint */
392 wl->wl_bufbytes_max >>= PAGE_SHIFT;
393 wl->wl_bufbytes_max <<= PAGE_SHIFT;
394 wl->wl_bufbytes_max >>= wl->wl_log_dev_bshift;
395 wl->wl_bufbytes_max <<= wl->wl_log_dev_bshift;
396 wl->wl_bufbytes_max >>= wl->wl_fs_dev_bshift;
397 wl->wl_bufbytes_max <<= wl->wl_fs_dev_bshift;
398
399 /* XXX maybe use filesystem fragment size instead of 1024 */
400 /* XXX fix actual number of buffers reserved per filesystem. */
401 wl->wl_bufcount_max = (nbuf / 2) * 1024;
402
403 /* XXX tie this into resource estimation */
404 wl->wl_dealloclim = 2 * btodb(wl->wl_bufbytes_max);
405
406 #if WAPBL_UVM_ALLOC
407 wl->wl_deallocblks = (void *) uvm_km_zalloc(kernel_map,
408 round_page(sizeof(*wl->wl_deallocblks) * wl->wl_dealloclim));
409 KASSERT(wl->wl_deallocblks != NULL);
410 wl->wl_dealloclens = (void *) uvm_km_zalloc(kernel_map,
411 round_page(sizeof(*wl->wl_dealloclens) * wl->wl_dealloclim));
412 KASSERT(wl->wl_dealloclens != NULL);
413 #else
414 wl->wl_deallocblks = wapbl_malloc(sizeof(*wl->wl_deallocblks) *
415 wl->wl_dealloclim);
416 wl->wl_dealloclens = wapbl_malloc(sizeof(*wl->wl_dealloclens) *
417 wl->wl_dealloclim);
418 #endif
419
420 wapbl_inodetrk_init(wl, WAPBL_INODETRK_SIZE);
421
422 /* Initialize the commit header */
423 {
424 struct wapbl_wc_header *wc;
425 size_t len = 1<<wl->wl_log_dev_bshift;
426 wc = wapbl_calloc(1, len);
427 wc->wc_type = WAPBL_WC_HEADER;
428 wc->wc_len = len;
429 wc->wc_circ_off = wl->wl_circ_off;
430 wc->wc_circ_size = wl->wl_circ_size;
431 /* XXX wc->wc_fsid */
432 wc->wc_log_dev_bshift = wl->wl_log_dev_bshift;
433 wc->wc_fs_dev_bshift = wl->wl_fs_dev_bshift;
434 wl->wl_wc_header = wc;
435 wl->wl_wc_scratch = wapbl_malloc(len);
436 }
437
438 /*
439 * if there was an existing set of unlinked but
440 * allocated inodes, preserve it in the new
441 * log.
442 */
443 if (wr && wr->wr_inodescnt) {
444 int i;
445
446 WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
447 ("wapbl_start: reusing log with %d inodes\n",
448 wr->wr_inodescnt));
449
450 /*
451 * Its only valid to reuse the replay log if its
452 * the same as the new log we just opened.
453 */
454 KDASSERT(!wapbl_replay_isopen(wr));
455 KASSERT(devvp->v_rdev == wr->wr_devvp->v_rdev);
456 KASSERT(logpbn == wr->wr_logpbn);
457 KASSERT(wl->wl_circ_size == wr->wr_wc_header.wc_circ_size);
458 KASSERT(wl->wl_circ_off == wr->wr_wc_header.wc_circ_off);
459 KASSERT(wl->wl_log_dev_bshift ==
460 wr->wr_wc_header.wc_log_dev_bshift);
461 KASSERT(wl->wl_fs_dev_bshift ==
462 wr->wr_wc_header.wc_fs_dev_bshift);
463
464 wl->wl_wc_header->wc_generation =
465 wr->wr_wc_header.wc_generation + 1;
466
467 for (i = 0; i < wr->wr_inodescnt; i++)
468 wapbl_register_inode(wl, wr->wr_inodes[i].wr_inumber,
469 wr->wr_inodes[i].wr_imode);
470
471 /* Make sure new transaction won't overwrite old inodes list */
472 KDASSERT(wapbl_transaction_len(wl) <=
473 wapbl_space_free(wl->wl_circ_size, wr->wr_inodeshead,
474 wr->wr_inodestail));
475
476 wl->wl_head = wl->wl_tail = wr->wr_inodeshead;
477 wl->wl_reclaimable_bytes = wl->wl_reserved_bytes =
478 wapbl_transaction_len(wl);
479
480 error = wapbl_write_inodes(wl, &wl->wl_head);
481 if (error)
482 goto errout;
483
484 KASSERT(wl->wl_head != wl->wl_tail);
485 KASSERT(wl->wl_head != 0);
486 }
487
488 error = wapbl_write_commit(wl, wl->wl_head, wl->wl_tail);
489 if (error) {
490 goto errout;
491 }
492
493 *wlp = wl;
494 #if defined(WAPBL_DEBUG)
495 wapbl_debug_wl = wl;
496 #endif
497
498 return 0;
499 errout:
500 wapbl_discard(wl);
501 wapbl_free(wl->wl_wc_scratch);
502 wapbl_free(wl->wl_wc_header);
503 #if WAPBL_UVM_ALLOC
504 uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_deallocblks,
505 round_page(sizeof(*wl->wl_deallocblks *
506 wl->wl_dealloclim)));
507 uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_dealloclens,
508 round_page(sizeof(*wl->wl_dealloclens *
509 wl->wl_dealloclim)));
510 #else
511 wapbl_free(wl->wl_deallocblks);
512 wapbl_free(wl->wl_dealloclens);
513 #endif
514 wapbl_inodetrk_free(wl);
515 wapbl_free(wl);
516
517 return error;
518 }
519
520 /*
521 * Like wapbl_flush, only discards the transaction
522 * completely
523 */
524
525 void
526 wapbl_discard(struct wapbl *wl)
527 {
528 struct wapbl_entry *we;
529 struct buf *bp;
530 int i;
531
532 /*
533 * XXX we may consider using upgrade here
534 * if we want to call flush from inside a transaction
535 */
536 rw_enter(&wl->wl_rwlock, RW_WRITER);
537 wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens,
538 wl->wl_dealloccnt);
539
540 #ifdef WAPBL_DEBUG_PRINT
541 {
542 struct wapbl_entry *we;
543 pid_t pid = -1;
544 lwpid_t lid = -1;
545 if (curproc)
546 pid = curproc->p_pid;
547 if (curlwp)
548 lid = curlwp->l_lid;
549 #ifdef WAPBL_DEBUG_BUFBYTES
550 WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
551 ("wapbl_discard: thread %d.%d discarding "
552 "transaction\n"
553 "\tbufcount=%zu bufbytes=%zu bcount=%zu "
554 "deallocs=%d inodes=%d\n"
555 "\terrcnt = %u, reclaimable=%zu reserved=%zu "
556 "unsynced=%zu\n",
557 pid, lid, wl->wl_bufcount, wl->wl_bufbytes,
558 wl->wl_bcount, wl->wl_dealloccnt,
559 wl->wl_inohashcnt, wl->wl_error_count,
560 wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
561 wl->wl_unsynced_bufbytes));
562 SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
563 WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
564 ("\tentry: bufcount = %zu, reclaimable = %zu, "
565 "error = %d, unsynced = %zu\n",
566 we->we_bufcount, we->we_reclaimable_bytes,
567 we->we_error, we->we_unsynced_bufbytes));
568 }
569 #else /* !WAPBL_DEBUG_BUFBYTES */
570 WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
571 ("wapbl_discard: thread %d.%d discarding transaction\n"
572 "\tbufcount=%zu bufbytes=%zu bcount=%zu "
573 "deallocs=%d inodes=%d\n"
574 "\terrcnt = %u, reclaimable=%zu reserved=%zu\n",
575 pid, lid, wl->wl_bufcount, wl->wl_bufbytes,
576 wl->wl_bcount, wl->wl_dealloccnt,
577 wl->wl_inohashcnt, wl->wl_error_count,
578 wl->wl_reclaimable_bytes, wl->wl_reserved_bytes));
579 SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
580 WAPBL_PRINTF(WAPBL_PRINT_DISCARD,
581 ("\tentry: bufcount = %zu, reclaimable = %zu, "
582 "error = %d\n",
583 we->we_bufcount, we->we_reclaimable_bytes,
584 we->we_error));
585 }
586 #endif /* !WAPBL_DEBUG_BUFBYTES */
587 }
588 #endif /* WAPBL_DEBUG_PRINT */
589
590 for (i = 0; i <= wl->wl_inohashmask; i++) {
591 struct wapbl_ino_head *wih;
592 struct wapbl_ino *wi;
593
594 wih = &wl->wl_inohash[i];
595 while ((wi = LIST_FIRST(wih)) != NULL) {
596 LIST_REMOVE(wi, wi_hash);
597 pool_put(&wapbl_ino_pool, wi);
598 KASSERT(wl->wl_inohashcnt > 0);
599 wl->wl_inohashcnt--;
600 }
601 }
602
603 /*
604 * clean buffer list
605 */
606 mutex_enter(&bufcache_lock);
607 mutex_enter(&wl->wl_mtx);
608 while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) {
609 if (bbusy(bp, 0, 0, &wl->wl_mtx) == 0) {
610 /*
611 * The buffer will be unlocked and
612 * removed from the transaction in brelse
613 */
614 mutex_exit(&wl->wl_mtx);
615 brelsel(bp, 0);
616 mutex_enter(&wl->wl_mtx);
617 }
618 }
619 mutex_exit(&wl->wl_mtx);
620 mutex_exit(&bufcache_lock);
621
622 /*
623 * Remove references to this wl from wl_entries, free any which
624 * no longer have buffers, others will be freed in wapbl_biodone
625 * when they no longer have any buffers.
626 */
627 while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) != NULL) {
628 SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries);
629 /* XXX should we be accumulating wl_error_count
630 * and increasing reclaimable bytes ? */
631 we->we_wapbl = NULL;
632 if (we->we_bufcount == 0) {
633 #ifdef WAPBL_DEBUG_BUFBYTES
634 KASSERT(we->we_unsynced_bufbytes == 0);
635 #endif
636 wapbl_free(we);
637 }
638 }
639
640 /* Discard list of deallocs */
641 wl->wl_dealloccnt = 0;
642 /* XXX should we clear wl_reserved_bytes? */
643
644 KASSERT(wl->wl_bufbytes == 0);
645 KASSERT(wl->wl_bcount == 0);
646 KASSERT(wl->wl_bufcount == 0);
647 KASSERT(LIST_EMPTY(&wl->wl_bufs));
648 KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries));
649 KASSERT(wl->wl_inohashcnt == 0);
650
651 rw_exit(&wl->wl_rwlock);
652 }
653
654 int
655 wapbl_stop(struct wapbl *wl, int force)
656 {
657 struct vnode *vp;
658 int error;
659
660 WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_stop called\n"));
661 error = wapbl_flush(wl, 1);
662 if (error) {
663 if (force)
664 wapbl_discard(wl);
665 else
666 return error;
667 }
668
669 /* Unlinked inodes persist after a flush */
670 if (wl->wl_inohashcnt) {
671 if (force) {
672 wapbl_discard(wl);
673 } else {
674 return EBUSY;
675 }
676 }
677
678 KASSERT(wl->wl_bufbytes == 0);
679 KASSERT(wl->wl_bcount == 0);
680 KASSERT(wl->wl_bufcount == 0);
681 KASSERT(LIST_EMPTY(&wl->wl_bufs));
682 KASSERT(wl->wl_dealloccnt == 0);
683 KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries));
684 KASSERT(wl->wl_inohashcnt == 0);
685
686 vp = wl->wl_logvp;
687
688 wapbl_free(wl->wl_wc_scratch);
689 wapbl_free(wl->wl_wc_header);
690 #if WAPBL_UVM_ALLOC
691 uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_deallocblks,
692 round_page(sizeof(*wl->wl_deallocblks *
693 wl->wl_dealloclim)));
694 uvm_km_free_wakeup(kernel_map, (vaddr_t) wl->wl_dealloclens,
695 round_page(sizeof(*wl->wl_dealloclens *
696 wl->wl_dealloclim)));
697 #else
698 wapbl_free(wl->wl_deallocblks);
699 wapbl_free(wl->wl_dealloclens);
700 #endif
701 wapbl_inodetrk_free(wl);
702
703 cv_destroy(&wl->wl_reclaimable_cv);
704 mutex_destroy(&wl->wl_mtx);
705 rw_destroy(&wl->wl_rwlock);
706 wapbl_free(wl);
707
708 return 0;
709 }
710
711 static int
712 wapbl_doio(void *data, size_t len, struct vnode *devvp, daddr_t pbn, int flags)
713 {
714 struct pstats *pstats = curlwp->l_proc->p_stats;
715 struct buf *bp;
716 int error;
717
718 KASSERT((flags & ~(B_WRITE | B_READ)) == 0);
719 KASSERT(devvp->v_type == VBLK);
720
721 if ((flags & (B_WRITE | B_READ)) == B_WRITE) {
722 mutex_enter(&devvp->v_interlock);
723 devvp->v_numoutput++;
724 mutex_exit(&devvp->v_interlock);
725 pstats->p_ru.ru_oublock++;
726 } else {
727 pstats->p_ru.ru_inblock++;
728 }
729
730 bp = getiobuf(devvp, true);
731 bp->b_flags = flags;
732 bp->b_cflags = BC_BUSY; /* silly & dubious */
733 bp->b_dev = devvp->v_rdev;
734 bp->b_data = data;
735 bp->b_bufsize = bp->b_resid = bp->b_bcount = len;
736 bp->b_blkno = pbn;
737
738 WAPBL_PRINTF(WAPBL_PRINT_IO,
739 ("wapbl_doio: %s %d bytes at block %"PRId64" on dev 0x%x\n",
740 BUF_ISWRITE(bp) ? "write" : "read", bp->b_bcount,
741 bp->b_blkno, bp->b_dev));
742
743 VOP_STRATEGY(devvp, bp);
744
745 error = biowait(bp);
746 putiobuf(bp);
747
748 if (error) {
749 WAPBL_PRINTF(WAPBL_PRINT_ERROR,
750 ("wapbl_doio: %s %zu bytes at block %" PRId64
751 " on dev 0x%x failed with error %d\n",
752 (((flags & (B_WRITE | B_READ)) == B_WRITE) ?
753 "write" : "read"),
754 len, pbn, devvp->v_rdev, error));
755 }
756
757 return error;
758 }
759
760 int
761 wapbl_write(void *data, size_t len, struct vnode *devvp, daddr_t pbn)
762 {
763
764 return wapbl_doio(data, len, devvp, pbn, B_WRITE);
765 }
766
767 int
768 wapbl_read(void *data, size_t len, struct vnode *devvp, daddr_t pbn)
769 {
770
771 return wapbl_doio(data, len, devvp, pbn, B_READ);
772 }
773
774 /*
775 * Off is byte offset returns new offset for next write
776 * handles log wraparound
777 */
778 static int
779 wapbl_circ_write(struct wapbl *wl, void *data, size_t len, off_t *offp)
780 {
781 size_t slen;
782 off_t off = *offp;
783 int error;
784
785 KDASSERT(((len >> wl->wl_log_dev_bshift) <<
786 wl->wl_log_dev_bshift) == len);
787
788 if (off < wl->wl_circ_off)
789 off = wl->wl_circ_off;
790 slen = wl->wl_circ_off + wl->wl_circ_size - off;
791 if (slen < len) {
792 error = wapbl_write(data, slen, wl->wl_devvp,
793 wl->wl_logpbn + (off >> wl->wl_log_dev_bshift));
794 if (error)
795 return error;
796 data = (uint8_t *)data + slen;
797 len -= slen;
798 off = wl->wl_circ_off;
799 }
800 error = wapbl_write(data, len, wl->wl_devvp,
801 wl->wl_logpbn + (off >> wl->wl_log_dev_bshift));
802 if (error)
803 return error;
804 off += len;
805 if (off >= wl->wl_circ_off + wl->wl_circ_size)
806 off = wl->wl_circ_off;
807 *offp = off;
808 return 0;
809 }
810
811 /****************************************************************/
812
813 int
814 wapbl_begin(struct wapbl *wl, const char *file, int line)
815 {
816 int doflush;
817 unsigned lockcount;
818
819 KDASSERT(wl);
820
821 /*
822 * XXX this needs to be made much more sophisticated.
823 * perhaps each wapbl_begin could reserve a specified
824 * number of buffers and bytes.
825 */
826 mutex_enter(&wl->wl_mtx);
827 lockcount = wl->wl_lock_count;
828 doflush = ((wl->wl_bufbytes + (lockcount * MAXPHYS)) >
829 wl->wl_bufbytes_max / 2) ||
830 ((wl->wl_bufcount + (lockcount * 10)) >
831 wl->wl_bufcount_max / 2) ||
832 (wapbl_transaction_len(wl) > wl->wl_circ_size / 2) ||
833 (wl->wl_dealloccnt >=
834 (wl->wl_dealloclim - (wl->wl_dealloclim >> 8)));
835 mutex_exit(&wl->wl_mtx);
836
837 if (doflush) {
838 WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
839 ("force flush lockcnt=%d bufbytes=%zu "
840 "(max=%zu) bufcount=%zu (max=%zu) "
841 "dealloccnt %d (lim=%d)\n",
842 lockcount, wl->wl_bufbytes,
843 wl->wl_bufbytes_max, wl->wl_bufcount,
844 wl->wl_bufcount_max,
845 wl->wl_dealloccnt, wl->wl_dealloclim));
846 }
847
848 if (doflush) {
849 int error = wapbl_flush(wl, 0);
850 if (error)
851 return error;
852 }
853
854 rw_enter(&wl->wl_rwlock, RW_READER);
855 mutex_enter(&wl->wl_mtx);
856 wl->wl_lock_count++;
857 mutex_exit(&wl->wl_mtx);
858
859 #if defined(WAPBL_DEBUG_PRINT)
860 WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION,
861 ("wapbl_begin thread %d.%d with bufcount=%zu "
862 "bufbytes=%zu bcount=%zu at %s:%d\n",
863 curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
864 wl->wl_bufbytes, wl->wl_bcount, file, line));
865 #endif
866
867 return 0;
868 }
869
870 void
871 wapbl_end(struct wapbl *wl)
872 {
873
874 #if defined(WAPBL_DEBUG_PRINT)
875 WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION,
876 ("wapbl_end thread %d.%d with bufcount=%zu "
877 "bufbytes=%zu bcount=%zu\n",
878 curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
879 wl->wl_bufbytes, wl->wl_bcount));
880 #endif
881
882 mutex_enter(&wl->wl_mtx);
883 KASSERT(wl->wl_lock_count > 0);
884 wl->wl_lock_count--;
885 mutex_exit(&wl->wl_mtx);
886
887 rw_exit(&wl->wl_rwlock);
888 }
889
890 void
891 wapbl_add_buf(struct wapbl *wl, struct buf * bp)
892 {
893
894 KASSERT(bp->b_cflags & BC_BUSY);
895 KASSERT(bp->b_vp);
896
897 wapbl_jlock_assert(wl);
898
899 #if 0
900 /*
901 * XXX this might be an issue for swapfiles.
902 * see uvm_swap.c:1702
903 *
904 * XXX2 why require it then? leap of semantics?
905 */
906 KASSERT((bp->b_cflags & BC_NOCACHE) == 0);
907 #endif
908
909 mutex_enter(&wl->wl_mtx);
910 if (bp->b_flags & B_LOCKED) {
911 LIST_REMOVE(bp, b_wapbllist);
912 WAPBL_PRINTF(WAPBL_PRINT_BUFFER2,
913 ("wapbl_add_buf thread %d.%d re-adding buf %p "
914 "with %d bytes %d bcount\n",
915 curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize,
916 bp->b_bcount));
917 } else {
918 /* unlocked by dirty buffers shouldn't exist */
919 KASSERT(!(bp->b_oflags & BO_DELWRI));
920 wl->wl_bufbytes += bp->b_bufsize;
921 wl->wl_bcount += bp->b_bcount;
922 wl->wl_bufcount++;
923 WAPBL_PRINTF(WAPBL_PRINT_BUFFER,
924 ("wapbl_add_buf thread %d.%d adding buf %p "
925 "with %d bytes %d bcount\n",
926 curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize,
927 bp->b_bcount));
928 }
929 LIST_INSERT_HEAD(&wl->wl_bufs, bp, b_wapbllist);
930 mutex_exit(&wl->wl_mtx);
931
932 bp->b_flags |= B_LOCKED;
933 }
934
935 static void
936 wapbl_remove_buf_locked(struct wapbl * wl, struct buf *bp)
937 {
938
939 KASSERT(mutex_owned(&wl->wl_mtx));
940 KASSERT(bp->b_cflags & BC_BUSY);
941 wapbl_jlock_assert(wl);
942
943 #if 0
944 /*
945 * XXX this might be an issue for swapfiles.
946 * see uvm_swap.c:1725
947 *
948 * XXXdeux: see above
949 */
950 KASSERT((bp->b_flags & BC_NOCACHE) == 0);
951 #endif
952 KASSERT(bp->b_flags & B_LOCKED);
953
954 WAPBL_PRINTF(WAPBL_PRINT_BUFFER,
955 ("wapbl_remove_buf thread %d.%d removing buf %p with "
956 "%d bytes %d bcount\n",
957 curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, bp->b_bcount));
958
959 KASSERT(wl->wl_bufbytes >= bp->b_bufsize);
960 wl->wl_bufbytes -= bp->b_bufsize;
961 KASSERT(wl->wl_bcount >= bp->b_bcount);
962 wl->wl_bcount -= bp->b_bcount;
963 KASSERT(wl->wl_bufcount > 0);
964 wl->wl_bufcount--;
965 KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0));
966 KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0));
967 LIST_REMOVE(bp, b_wapbllist);
968
969 bp->b_flags &= ~B_LOCKED;
970 }
971
972 /* called from brelsel() in vfs_bio among other places */
973 void
974 wapbl_remove_buf(struct wapbl * wl, struct buf *bp)
975 {
976
977 mutex_enter(&wl->wl_mtx);
978 wapbl_remove_buf_locked(wl, bp);
979 mutex_exit(&wl->wl_mtx);
980 }
981
982 void
983 wapbl_resize_buf(struct wapbl *wl, struct buf *bp, long oldsz, long oldcnt)
984 {
985
986 KASSERT(bp->b_cflags & BC_BUSY);
987
988 /*
989 * XXX: why does this depend on B_LOCKED? otherwise the buf
990 * is not for a transaction? if so, why is this called in the
991 * first place?
992 */
993 if (bp->b_flags & B_LOCKED) {
994 mutex_enter(&wl->wl_mtx);
995 wl->wl_bufbytes += bp->b_bufsize - oldsz;
996 wl->wl_bcount += bp->b_bcount - oldcnt;
997 mutex_exit(&wl->wl_mtx);
998 }
999 }
1000
1001 #endif /* _KERNEL */
1002
1003 /****************************************************************/
1004 /* Some utility inlines */
1005
1006 /* This is used to advance the pointer at old to new value at old+delta */
1007 static __inline off_t
1008 wapbl_advance(size_t size, size_t off, off_t old, size_t delta)
1009 {
1010 off_t new;
1011
1012 /* Define acceptable ranges for inputs. */
1013 KASSERT(delta <= size);
1014 KASSERT((old == 0) || (old >= off));
1015 KASSERT(old < (size + off));
1016
1017 if ((old == 0) && (delta != 0))
1018 new = off + delta;
1019 else if ((old + delta) < (size + off))
1020 new = old + delta;
1021 else
1022 new = (old + delta) - size;
1023
1024 /* Note some interesting axioms */
1025 KASSERT((delta != 0) || (new == old));
1026 KASSERT((delta == 0) || (new != 0));
1027 KASSERT((delta != (size)) || (new == old));
1028
1029 /* Define acceptable ranges for output. */
1030 KASSERT((new == 0) || (new >= off));
1031 KASSERT(new < (size + off));
1032 return new;
1033 }
1034
1035 static __inline size_t
1036 wapbl_space_used(size_t avail, off_t head, off_t tail)
1037 {
1038
1039 if (tail == 0) {
1040 KASSERT(head == 0);
1041 return 0;
1042 }
1043 return ((head + (avail - 1) - tail) % avail) + 1;
1044 }
1045
1046 static __inline size_t
1047 wapbl_space_free(size_t avail, off_t head, off_t tail)
1048 {
1049
1050 return avail - wapbl_space_used(avail, head, tail);
1051 }
1052
1053 static __inline void
1054 wapbl_advance_head(size_t size, size_t off, size_t delta, off_t *headp,
1055 off_t *tailp)
1056 {
1057 off_t head = *headp;
1058 off_t tail = *tailp;
1059
1060 KASSERT(delta <= wapbl_space_free(size, head, tail));
1061 head = wapbl_advance(size, off, head, delta);
1062 if ((tail == 0) && (head != 0))
1063 tail = off;
1064 *headp = head;
1065 *tailp = tail;
1066 }
1067
1068 static __inline void
1069 wapbl_advance_tail(size_t size, size_t off, size_t delta, off_t *headp,
1070 off_t *tailp)
1071 {
1072 off_t head = *headp;
1073 off_t tail = *tailp;
1074
1075 KASSERT(delta <= wapbl_space_used(size, head, tail));
1076 tail = wapbl_advance(size, off, tail, delta);
1077 if (head == tail) {
1078 head = tail = 0;
1079 }
1080 *headp = head;
1081 *tailp = tail;
1082 }
1083
1084 #ifdef _KERNEL
1085
1086 /****************************************************************/
1087
1088 /*
1089 * Remove transactions whose buffers are completely flushed to disk.
1090 * Will block until at least minfree space is available.
1091 * only intended to be called from inside wapbl_flush and therefore
1092 * does not protect against commit races with itself or with flush.
1093 */
1094 static int
1095 wapbl_truncate(struct wapbl *wl, size_t minfree, int waitonly)
1096 {
1097 size_t delta;
1098 size_t avail;
1099 off_t head;
1100 off_t tail;
1101 int error = 0;
1102
1103 KASSERT(minfree <= (wl->wl_circ_size - wl->wl_reserved_bytes));
1104 KASSERT(rw_write_held(&wl->wl_rwlock));
1105
1106 mutex_enter(&wl->wl_mtx);
1107
1108 /*
1109 * First check to see if we have to do a commit
1110 * at all.
1111 */
1112 avail = wapbl_space_free(wl->wl_circ_size, wl->wl_head, wl->wl_tail);
1113 if (minfree < avail) {
1114 mutex_exit(&wl->wl_mtx);
1115 return 0;
1116 }
1117 minfree -= avail;
1118 while ((wl->wl_error_count == 0) &&
1119 (wl->wl_reclaimable_bytes < minfree)) {
1120 WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE,
1121 ("wapbl_truncate: sleeping on %p wl=%p bytes=%zd "
1122 "minfree=%zd\n",
1123 &wl->wl_reclaimable_bytes, wl, wl->wl_reclaimable_bytes,
1124 minfree));
1125
1126 cv_wait(&wl->wl_reclaimable_cv, &wl->wl_mtx);
1127 }
1128 if (wl->wl_reclaimable_bytes < minfree) {
1129 KASSERT(wl->wl_error_count);
1130 /* XXX maybe get actual error from buffer instead someday? */
1131 error = EIO;
1132 }
1133 head = wl->wl_head;
1134 tail = wl->wl_tail;
1135 delta = wl->wl_reclaimable_bytes;
1136
1137 /* If all of of the entries are flushed, then be sure to keep
1138 * the reserved bytes reserved. Watch out for discarded transactions,
1139 * which could leave more bytes reserved than are reclaimable.
1140 */
1141 if (SIMPLEQ_EMPTY(&wl->wl_entries) &&
1142 (delta >= wl->wl_reserved_bytes)) {
1143 delta -= wl->wl_reserved_bytes;
1144 }
1145 wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, &head,
1146 &tail);
1147 KDASSERT(wl->wl_reserved_bytes <=
1148 wapbl_space_used(wl->wl_circ_size, head, tail));
1149 mutex_exit(&wl->wl_mtx);
1150
1151 if (error)
1152 return error;
1153
1154 if (waitonly)
1155 return 0;
1156
1157 /*
1158 * This is where head, tail and delta are unprotected
1159 * from races against itself or flush. This is ok since
1160 * we only call this routine from inside flush itself.
1161 *
1162 * XXX: how can it race against itself when accessed only
1163 * from behind the write-locked rwlock?
1164 */
1165 error = wapbl_write_commit(wl, head, tail);
1166 if (error)
1167 return error;
1168
1169 wl->wl_head = head;
1170 wl->wl_tail = tail;
1171
1172 mutex_enter(&wl->wl_mtx);
1173 KASSERT(wl->wl_reclaimable_bytes >= delta);
1174 wl->wl_reclaimable_bytes -= delta;
1175 mutex_exit(&wl->wl_mtx);
1176 WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE,
1177 ("wapbl_truncate thread %d.%d truncating %zu bytes\n",
1178 curproc->p_pid, curlwp->l_lid, delta));
1179
1180 return 0;
1181 }
1182
1183 /****************************************************************/
1184
1185 void
1186 wapbl_biodone(struct buf *bp)
1187 {
1188 struct wapbl_entry *we = bp->b_private;
1189 struct wapbl *wl = we->we_wapbl;
1190
1191 /*
1192 * Handle possible flushing of buffers after log has been
1193 * decomissioned.
1194 */
1195 if (!wl) {
1196 KASSERT(we->we_bufcount > 0);
1197 we->we_bufcount--;
1198 #ifdef WAPBL_DEBUG_BUFBYTES
1199 KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize);
1200 we->we_unsynced_bufbytes -= bp->b_bufsize;
1201 #endif
1202
1203 if (we->we_bufcount == 0) {
1204 #ifdef WAPBL_DEBUG_BUFBYTES
1205 KASSERT(we->we_unsynced_bufbytes == 0);
1206 #endif
1207 wapbl_free(we);
1208 }
1209
1210 brelse(bp, 0);
1211 return;
1212 }
1213
1214 #ifdef ohbother
1215 KDASSERT(bp->b_flags & B_DONE);
1216 KDASSERT(!(bp->b_flags & B_DELWRI));
1217 KDASSERT(bp->b_flags & B_ASYNC);
1218 KDASSERT(bp->b_flags & B_BUSY);
1219 KDASSERT(!(bp->b_flags & B_LOCKED));
1220 KDASSERT(!(bp->b_flags & B_READ));
1221 KDASSERT(!(bp->b_flags & B_INVAL));
1222 KDASSERT(!(bp->b_flags & B_NOCACHE));
1223 #endif
1224
1225 if (bp->b_error) {
1226 #ifdef notyet /* Can't currently handle possible dirty buffer reuse */
1227 XXXpooka: interfaces not fully updated
1228 Note: this was not enabled in the original patch
1229 against netbsd4 either. I don't know if comment
1230 above is true or not.
1231
1232 /*
1233 * If an error occurs, report the error and leave the
1234 * buffer as a delayed write on the LRU queue.
1235 * restarting the write would likely result in
1236 * an error spinloop, so let it be done harmlessly
1237 * by the syncer.
1238 */
1239 bp->b_flags &= ~(B_DONE);
1240 simple_unlock(&bp->b_interlock);
1241
1242 if (we->we_error == 0) {
1243 mutex_enter(&wl->wl_mtx);
1244 wl->wl_error_count++;
1245 mutex_exit(&wl->wl_mtx);
1246 cv_broadcast(&wl->wl_reclaimable_cv);
1247 }
1248 we->we_error = bp->b_error;
1249 bp->b_error = 0;
1250 brelse(bp);
1251 return;
1252 #else
1253 /* For now, just mark the log permanently errored out */
1254
1255 mutex_enter(&wl->wl_mtx);
1256 if (wl->wl_error_count == 0) {
1257 wl->wl_error_count++;
1258 cv_broadcast(&wl->wl_reclaimable_cv);
1259 }
1260 mutex_exit(&wl->wl_mtx);
1261 #endif
1262 }
1263
1264 mutex_enter(&wl->wl_mtx);
1265
1266 KASSERT(we->we_bufcount > 0);
1267 we->we_bufcount--;
1268 #ifdef WAPBL_DEBUG_BUFBYTES
1269 KASSERT(we->we_unsynced_bufbytes >= bp->b_bufsize);
1270 we->we_unsynced_bufbytes -= bp->b_bufsize;
1271 KASSERT(wl->wl_unsynced_bufbytes >= bp->b_bufsize);
1272 wl->wl_unsynced_bufbytes -= bp->b_bufsize;
1273 #endif
1274
1275 /*
1276 * If the current transaction can be reclaimed, start
1277 * at the beginning and reclaim any consecutive reclaimable
1278 * transactions. If we successfully reclaim anything,
1279 * then wakeup anyone waiting for the reclaim.
1280 */
1281 if (we->we_bufcount == 0) {
1282 size_t delta = 0;
1283 int errcnt = 0;
1284 #ifdef WAPBL_DEBUG_BUFBYTES
1285 KDASSERT(we->we_unsynced_bufbytes == 0);
1286 #endif
1287 /*
1288 * clear any posted error, since the buffer it came from
1289 * has successfully flushed by now
1290 */
1291 while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) &&
1292 (we->we_bufcount == 0)) {
1293 delta += we->we_reclaimable_bytes;
1294 if (we->we_error)
1295 errcnt++;
1296 SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries);
1297 wapbl_free(we);
1298 }
1299
1300 if (delta) {
1301 wl->wl_reclaimable_bytes += delta;
1302 KASSERT(wl->wl_error_count >= errcnt);
1303 wl->wl_error_count -= errcnt;
1304 cv_broadcast(&wl->wl_reclaimable_cv);
1305 }
1306 }
1307
1308 mutex_exit(&wl->wl_mtx);
1309 brelse(bp, 0);
1310 }
1311
1312 /*
1313 * Write transactions to disk + start I/O for contents
1314 */
1315 int
1316 wapbl_flush(struct wapbl *wl, int waitfor)
1317 {
1318 struct buf *bp;
1319 struct wapbl_entry *we;
1320 off_t off;
1321 off_t head;
1322 off_t tail;
1323 size_t delta = 0;
1324 size_t flushsize;
1325 size_t reserved;
1326 int error = 0;
1327
1328 /*
1329 * Do a quick check to see if a full flush can be skipped
1330 * This assumes that the flush callback does not need to be called
1331 * unless there are other outstanding bufs.
1332 */
1333 if (!waitfor) {
1334 size_t nbufs;
1335 mutex_enter(&wl->wl_mtx); /* XXX need mutex here to
1336 protect the KASSERTS */
1337 nbufs = wl->wl_bufcount;
1338 KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0));
1339 KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0));
1340 mutex_exit(&wl->wl_mtx);
1341 if (nbufs == 0)
1342 return 0;
1343 }
1344
1345 /*
1346 * XXX we may consider using LK_UPGRADE here
1347 * if we want to call flush from inside a transaction
1348 */
1349 rw_enter(&wl->wl_rwlock, RW_WRITER);
1350 wl->wl_flush(wl->wl_mount, wl->wl_deallocblks, wl->wl_dealloclens,
1351 wl->wl_dealloccnt);
1352
1353 /*
1354 * Now that we are fully locked and flushed,
1355 * do another check for nothing to do.
1356 */
1357 if (wl->wl_bufcount == 0) {
1358 goto out;
1359 }
1360
1361 #if 0
1362 WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1363 ("wapbl_flush thread %d.%d flushing entries with "
1364 "bufcount=%zu bufbytes=%zu\n",
1365 curproc->p_pid, curlwp->l_lid, wl->wl_bufcount,
1366 wl->wl_bufbytes));
1367 #endif
1368
1369 /* Calculate amount of space needed to flush */
1370 flushsize = wapbl_transaction_len(wl);
1371 if (wapbl_verbose_commit) {
1372 struct timespec ts;
1373 getnanotime(&ts);
1374 printf("%s: %lld.%06ld this transaction = %zu bytes\n",
1375 __func__, (long long)ts.tv_sec,
1376 (long)ts.tv_nsec, flushsize);
1377 }
1378
1379 if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) {
1380 /*
1381 * XXX this could be handled more gracefully, perhaps place
1382 * only a partial transaction in the log and allow the
1383 * remaining to flush without the protection of the journal.
1384 */
1385 panic("wapbl_flush: current transaction too big to flush\n");
1386 }
1387
1388 error = wapbl_truncate(wl, flushsize, 0);
1389 if (error)
1390 goto out2;
1391
1392 off = wl->wl_head;
1393 KASSERT((off == 0) || ((off >= wl->wl_circ_off) &&
1394 (off < wl->wl_circ_off + wl->wl_circ_size)));
1395 error = wapbl_write_blocks(wl, &off);
1396 if (error)
1397 goto out2;
1398 error = wapbl_write_revocations(wl, &off);
1399 if (error)
1400 goto out2;
1401 error = wapbl_write_inodes(wl, &off);
1402 if (error)
1403 goto out2;
1404
1405 reserved = 0;
1406 if (wl->wl_inohashcnt)
1407 reserved = wapbl_transaction_inodes_len(wl);
1408
1409 head = wl->wl_head;
1410 tail = wl->wl_tail;
1411
1412 wapbl_advance_head(wl->wl_circ_size, wl->wl_circ_off, flushsize,
1413 &head, &tail);
1414 #ifdef WAPBL_DEBUG
1415 if (head != off) {
1416 panic("lost head! head=%"PRIdMAX" tail=%" PRIdMAX
1417 " off=%"PRIdMAX" flush=%zu\n",
1418 (intmax_t)head, (intmax_t)tail, (intmax_t)off,
1419 flushsize);
1420 }
1421 #else
1422 KASSERT(head == off);
1423 #endif
1424
1425 /* Opportunistically move the tail forward if we can */
1426 if (!wapbl_lazy_truncate) {
1427 mutex_enter(&wl->wl_mtx);
1428 delta = wl->wl_reclaimable_bytes;
1429 mutex_exit(&wl->wl_mtx);
1430 wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta,
1431 &head, &tail);
1432 }
1433
1434 error = wapbl_write_commit(wl, head, tail);
1435 if (error)
1436 goto out2;
1437
1438 /* poolme? or kmemme? */
1439 we = wapbl_calloc(1, sizeof(*we));
1440
1441 #ifdef WAPBL_DEBUG_BUFBYTES
1442 WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1443 ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu"
1444 " unsynced=%zu"
1445 "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d "
1446 "inodes=%d\n",
1447 curproc->p_pid, curlwp->l_lid, flushsize, delta,
1448 wapbl_space_used(wl->wl_circ_size, head, tail),
1449 wl->wl_unsynced_bufbytes, wl->wl_bufcount,
1450 wl->wl_bufbytes, wl->wl_bcount, wl->wl_dealloccnt,
1451 wl->wl_inohashcnt));
1452 #else
1453 WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1454 ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu"
1455 "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d "
1456 "inodes=%d\n",
1457 curproc->p_pid, curlwp->l_lid, flushsize, delta,
1458 wapbl_space_used(wl->wl_circ_size, head, tail),
1459 wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount,
1460 wl->wl_dealloccnt, wl->wl_inohashcnt));
1461 #endif
1462
1463
1464 mutex_enter(&bufcache_lock);
1465 mutex_enter(&wl->wl_mtx);
1466
1467 wl->wl_reserved_bytes = reserved;
1468 wl->wl_head = head;
1469 wl->wl_tail = tail;
1470 KASSERT(wl->wl_reclaimable_bytes >= delta);
1471 wl->wl_reclaimable_bytes -= delta;
1472 wl->wl_dealloccnt = 0;
1473 #ifdef WAPBL_DEBUG_BUFBYTES
1474 wl->wl_unsynced_bufbytes += wl->wl_bufbytes;
1475 #endif
1476
1477 we->we_wapbl = wl;
1478 we->we_bufcount = wl->wl_bufcount;
1479 #ifdef WAPBL_DEBUG_BUFBYTES
1480 we->we_unsynced_bufbytes = wl->wl_bufbytes;
1481 #endif
1482 we->we_reclaimable_bytes = flushsize;
1483 we->we_error = 0;
1484 SIMPLEQ_INSERT_TAIL(&wl->wl_entries, we, we_entries);
1485
1486 /*
1487 * this flushes bufs in reverse order than they were queued
1488 * it shouldn't matter, but if we care we could use TAILQ instead.
1489 * XXX Note they will get put on the lru queue when they flush
1490 * so we might actually want to change this to preserve order.
1491 */
1492 while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) {
1493 if (bbusy(bp, 0, 0, &wl->wl_mtx)) {
1494 continue;
1495 }
1496 bp->b_iodone = wapbl_biodone;
1497 bp->b_private = we;
1498 bremfree(bp);
1499 wapbl_remove_buf_locked(wl, bp);
1500 mutex_exit(&wl->wl_mtx);
1501 mutex_exit(&bufcache_lock);
1502 bawrite(bp);
1503 mutex_enter(&bufcache_lock);
1504 mutex_enter(&wl->wl_mtx);
1505 }
1506 mutex_exit(&wl->wl_mtx);
1507 mutex_exit(&bufcache_lock);
1508
1509 #if 0
1510 WAPBL_PRINTF(WAPBL_PRINT_FLUSH,
1511 ("wapbl_flush thread %d.%d done flushing entries...\n",
1512 curproc->p_pid, curlwp->l_lid));
1513 #endif
1514
1515 out:
1516
1517 /*
1518 * If the waitfor flag is set, don't return until everything is
1519 * fully flushed and the on disk log is empty.
1520 */
1521 if (waitfor) {
1522 error = wapbl_truncate(wl, wl->wl_circ_size -
1523 wl->wl_reserved_bytes, wapbl_lazy_truncate);
1524 }
1525
1526 out2:
1527 if (error) {
1528 wl->wl_flush_abort(wl->wl_mount, wl->wl_deallocblks,
1529 wl->wl_dealloclens, wl->wl_dealloccnt);
1530 }
1531
1532 #ifdef WAPBL_DEBUG_PRINT
1533 if (error) {
1534 pid_t pid = -1;
1535 lwpid_t lid = -1;
1536 if (curproc)
1537 pid = curproc->p_pid;
1538 if (curlwp)
1539 lid = curlwp->l_lid;
1540 mutex_enter(&wl->wl_mtx);
1541 #ifdef WAPBL_DEBUG_BUFBYTES
1542 WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1543 ("wapbl_flush: thread %d.%d aborted flush: "
1544 "error = %d\n"
1545 "\tbufcount=%zu bufbytes=%zu bcount=%zu "
1546 "deallocs=%d inodes=%d\n"
1547 "\terrcnt = %d, reclaimable=%zu reserved=%zu "
1548 "unsynced=%zu\n",
1549 pid, lid, error, wl->wl_bufcount,
1550 wl->wl_bufbytes, wl->wl_bcount,
1551 wl->wl_dealloccnt, wl->wl_inohashcnt,
1552 wl->wl_error_count, wl->wl_reclaimable_bytes,
1553 wl->wl_reserved_bytes, wl->wl_unsynced_bufbytes));
1554 SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1555 WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1556 ("\tentry: bufcount = %zu, reclaimable = %zu, "
1557 "error = %d, unsynced = %zu\n",
1558 we->we_bufcount, we->we_reclaimable_bytes,
1559 we->we_error, we->we_unsynced_bufbytes));
1560 }
1561 #else
1562 WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1563 ("wapbl_flush: thread %d.%d aborted flush: "
1564 "error = %d\n"
1565 "\tbufcount=%zu bufbytes=%zu bcount=%zu "
1566 "deallocs=%d inodes=%d\n"
1567 "\terrcnt = %d, reclaimable=%zu reserved=%zu\n",
1568 pid, lid, error, wl->wl_bufcount,
1569 wl->wl_bufbytes, wl->wl_bcount,
1570 wl->wl_dealloccnt, wl->wl_inohashcnt,
1571 wl->wl_error_count, wl->wl_reclaimable_bytes,
1572 wl->wl_reserved_bytes));
1573 SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1574 WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1575 ("\tentry: bufcount = %zu, reclaimable = %zu, "
1576 "error = %d\n", we->we_bufcount,
1577 we->we_reclaimable_bytes, we->we_error));
1578 }
1579 #endif
1580 mutex_exit(&wl->wl_mtx);
1581 }
1582 #endif
1583
1584 rw_exit(&wl->wl_rwlock);
1585 return error;
1586 }
1587
1588 /****************************************************************/
1589
1590 void
1591 wapbl_jlock_assert(struct wapbl *wl)
1592 {
1593
1594 KASSERT(rw_lock_held(&wl->wl_rwlock));
1595 }
1596
1597 void
1598 wapbl_junlock_assert(struct wapbl *wl)
1599 {
1600
1601 KASSERT(!rw_write_held(&wl->wl_rwlock));
1602 }
1603
1604 /****************************************************************/
1605
1606 /* locks missing */
1607 void
1608 wapbl_print(struct wapbl *wl,
1609 int full,
1610 void (*pr)(const char *, ...))
1611 {
1612 struct buf *bp;
1613 struct wapbl_entry *we;
1614 (*pr)("wapbl %p", wl);
1615 (*pr)("\nlogvp = %p, devvp = %p, logpbn = %"PRId64"\n",
1616 wl->wl_logvp, wl->wl_devvp, wl->wl_logpbn);
1617 (*pr)("circ = %zu, header = %zu, head = %"PRIdMAX" tail = %"PRIdMAX"\n",
1618 wl->wl_circ_size, wl->wl_circ_off,
1619 (intmax_t)wl->wl_head, (intmax_t)wl->wl_tail);
1620 (*pr)("fs_dev_bshift = %d, log_dev_bshift = %d\n",
1621 wl->wl_log_dev_bshift, wl->wl_fs_dev_bshift);
1622 #ifdef WAPBL_DEBUG_BUFBYTES
1623 (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu "
1624 "reserved = %zu errcnt = %d unsynced = %zu\n",
1625 wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount,
1626 wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
1627 wl->wl_error_count, wl->wl_unsynced_bufbytes);
1628 #else
1629 (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu "
1630 "reserved = %zu errcnt = %d\n", wl->wl_bufcount, wl->wl_bufbytes,
1631 wl->wl_bcount, wl->wl_reclaimable_bytes, wl->wl_reserved_bytes,
1632 wl->wl_error_count);
1633 #endif
1634 (*pr)("\tdealloccnt = %d, dealloclim = %d\n",
1635 wl->wl_dealloccnt, wl->wl_dealloclim);
1636 (*pr)("\tinohashcnt = %d, inohashmask = 0x%08x\n",
1637 wl->wl_inohashcnt, wl->wl_inohashmask);
1638 (*pr)("entries:\n");
1639 SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) {
1640 #ifdef WAPBL_DEBUG_BUFBYTES
1641 (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d, "
1642 "unsynced = %zu\n",
1643 we->we_bufcount, we->we_reclaimable_bytes,
1644 we->we_error, we->we_unsynced_bufbytes);
1645 #else
1646 (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d\n",
1647 we->we_bufcount, we->we_reclaimable_bytes, we->we_error);
1648 #endif
1649 }
1650 if (full) {
1651 int cnt = 0;
1652 (*pr)("bufs =");
1653 LIST_FOREACH(bp, &wl->wl_bufs, b_wapbllist) {
1654 if (!LIST_NEXT(bp, b_wapbllist)) {
1655 (*pr)(" %p", bp);
1656 } else if ((++cnt % 6) == 0) {
1657 (*pr)(" %p,\n\t", bp);
1658 } else {
1659 (*pr)(" %p,", bp);
1660 }
1661 }
1662 (*pr)("\n");
1663
1664 (*pr)("dealloced blks = ");
1665 {
1666 int i;
1667 cnt = 0;
1668 for (i = 0; i < wl->wl_dealloccnt; i++) {
1669 (*pr)(" %"PRId64":%d,",
1670 wl->wl_deallocblks[i],
1671 wl->wl_dealloclens[i]);
1672 if ((++cnt % 4) == 0) {
1673 (*pr)("\n\t");
1674 }
1675 }
1676 }
1677 (*pr)("\n");
1678
1679 (*pr)("registered inodes = ");
1680 {
1681 int i;
1682 cnt = 0;
1683 for (i = 0; i <= wl->wl_inohashmask; i++) {
1684 struct wapbl_ino_head *wih;
1685 struct wapbl_ino *wi;
1686
1687 wih = &wl->wl_inohash[i];
1688 LIST_FOREACH(wi, wih, wi_hash) {
1689 if (wi->wi_ino == 0)
1690 continue;
1691 (*pr)(" %"PRId32"/0%06"PRIo32",",
1692 wi->wi_ino, wi->wi_mode);
1693 if ((++cnt % 4) == 0) {
1694 (*pr)("\n\t");
1695 }
1696 }
1697 }
1698 (*pr)("\n");
1699 }
1700 }
1701 }
1702
1703 #if defined(WAPBL_DEBUG) || defined(DDB)
1704 void
1705 wapbl_dump(struct wapbl *wl)
1706 {
1707 #if defined(WAPBL_DEBUG)
1708 if (!wl)
1709 wl = wapbl_debug_wl;
1710 #endif
1711 if (!wl)
1712 return;
1713 wapbl_print(wl, 1, printf);
1714 }
1715 #endif
1716
1717 /****************************************************************/
1718
1719 void
1720 wapbl_register_deallocation(struct wapbl *wl, daddr_t blk, int len)
1721 {
1722
1723 wapbl_jlock_assert(wl);
1724
1725 mutex_enter(&wl->wl_mtx);
1726 /* XXX should eventually instead tie this into resource estimation */
1727 /*
1728 * XXX this panic needs locking/mutex analysis and the
1729 * ability to cope with the failure.
1730 */
1731 /* XXX this XXX doesn't have enough XXX */
1732 if (__predict_false(wl->wl_dealloccnt >= wl->wl_dealloclim))
1733 panic("wapbl_register_deallocation: out of resources");
1734
1735 wl->wl_deallocblks[wl->wl_dealloccnt] = blk;
1736 wl->wl_dealloclens[wl->wl_dealloccnt] = len;
1737 wl->wl_dealloccnt++;
1738 WAPBL_PRINTF(WAPBL_PRINT_ALLOC,
1739 ("wapbl_register_deallocation: blk=%"PRId64" len=%d\n", blk, len));
1740 mutex_exit(&wl->wl_mtx);
1741 }
1742
1743 /****************************************************************/
1744
1745 static void
1746 wapbl_inodetrk_init(struct wapbl *wl, u_int size)
1747 {
1748
1749 wl->wl_inohash = hashinit(size, HASH_LIST, true, &wl->wl_inohashmask);
1750 if (atomic_inc_uint_nv(&wapbl_ino_pool_refcount) == 1) {
1751 pool_init(&wapbl_ino_pool, sizeof(struct wapbl_ino), 0, 0, 0,
1752 "wapblinopl", &pool_allocator_nointr, IPL_NONE);
1753 }
1754 }
1755
1756 static void
1757 wapbl_inodetrk_free(struct wapbl *wl)
1758 {
1759
1760 /* XXX this KASSERT needs locking/mutex analysis */
1761 KASSERT(wl->wl_inohashcnt == 0);
1762 hashdone(wl->wl_inohash, HASH_LIST, wl->wl_inohashmask);
1763 if (atomic_dec_uint_nv(&wapbl_ino_pool_refcount) == 0) {
1764 pool_destroy(&wapbl_ino_pool);
1765 }
1766 }
1767
1768 static struct wapbl_ino *
1769 wapbl_inodetrk_get(struct wapbl *wl, ino_t ino)
1770 {
1771 struct wapbl_ino_head *wih;
1772 struct wapbl_ino *wi;
1773
1774 KASSERT(mutex_owned(&wl->wl_mtx));
1775
1776 wih = &wl->wl_inohash[ino & wl->wl_inohashmask];
1777 LIST_FOREACH(wi, wih, wi_hash) {
1778 if (ino == wi->wi_ino)
1779 return wi;
1780 }
1781 return 0;
1782 }
1783
1784 void
1785 wapbl_register_inode(struct wapbl *wl, ino_t ino, mode_t mode)
1786 {
1787 struct wapbl_ino_head *wih;
1788 struct wapbl_ino *wi;
1789
1790 wi = pool_get(&wapbl_ino_pool, PR_WAITOK);
1791
1792 mutex_enter(&wl->wl_mtx);
1793 if (wapbl_inodetrk_get(wl, ino) == NULL) {
1794 wi->wi_ino = ino;
1795 wi->wi_mode = mode;
1796 wih = &wl->wl_inohash[ino & wl->wl_inohashmask];
1797 LIST_INSERT_HEAD(wih, wi, wi_hash);
1798 wl->wl_inohashcnt++;
1799 WAPBL_PRINTF(WAPBL_PRINT_INODE,
1800 ("wapbl_register_inode: ino=%"PRId64"\n", ino));
1801 mutex_exit(&wl->wl_mtx);
1802 } else {
1803 mutex_exit(&wl->wl_mtx);
1804 pool_put(&wapbl_ino_pool, wi);
1805 }
1806 }
1807
1808 void
1809 wapbl_unregister_inode(struct wapbl *wl, ino_t ino, mode_t mode)
1810 {
1811 struct wapbl_ino *wi;
1812
1813 mutex_enter(&wl->wl_mtx);
1814 wi = wapbl_inodetrk_get(wl, ino);
1815 if (wi) {
1816 WAPBL_PRINTF(WAPBL_PRINT_INODE,
1817 ("wapbl_unregister_inode: ino=%"PRId64"\n", ino));
1818 KASSERT(wl->wl_inohashcnt > 0);
1819 wl->wl_inohashcnt--;
1820 LIST_REMOVE(wi, wi_hash);
1821 mutex_exit(&wl->wl_mtx);
1822
1823 pool_put(&wapbl_ino_pool, wi);
1824 } else {
1825 mutex_exit(&wl->wl_mtx);
1826 }
1827 }
1828
1829 /****************************************************************/
1830
1831 static __inline size_t
1832 wapbl_transaction_inodes_len(struct wapbl *wl)
1833 {
1834 int blocklen = 1<<wl->wl_log_dev_bshift;
1835 int iph;
1836
1837 /* Calculate number of inodes described in a inodelist header */
1838 iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) /
1839 sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]);
1840
1841 KASSERT(iph > 0);
1842
1843 return MAX(1, howmany(wl->wl_inohashcnt, iph))*blocklen;
1844 }
1845
1846
1847 /* Calculate amount of space a transaction will take on disk */
1848 static size_t
1849 wapbl_transaction_len(struct wapbl *wl)
1850 {
1851 int blocklen = 1<<wl->wl_log_dev_bshift;
1852 size_t len;
1853 int bph;
1854
1855 /* Calculate number of blocks described in a blocklist header */
1856 bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
1857 sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
1858
1859 KASSERT(bph > 0);
1860
1861 len = wl->wl_bcount;
1862 len += howmany(wl->wl_bufcount, bph)*blocklen;
1863 len += howmany(wl->wl_dealloccnt, bph)*blocklen;
1864 len += wapbl_transaction_inodes_len(wl);
1865
1866 return len;
1867 }
1868
1869 /*
1870 * Perform commit operation
1871 *
1872 * Note that generation number incrementation needs to
1873 * be protected against racing with other invocations
1874 * of wapbl_commit. This is ok since this routine
1875 * is only invoked from wapbl_flush
1876 */
1877 static int
1878 wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail)
1879 {
1880 struct wapbl_wc_header *wc = wl->wl_wc_header;
1881 struct timespec ts;
1882 int error;
1883 int force = 1;
1884
1885 if (wapbl_flush_disk_cache) {
1886 /* XXX Calc checksum here, instead we do this for now */
1887 error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force,
1888 FWRITE, FSCRED);
1889 if (error) {
1890 WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1891 ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x "
1892 "returned %d\n", wl->wl_devvp->v_rdev, error));
1893 }
1894 }
1895
1896 wc->wc_head = head;
1897 wc->wc_tail = tail;
1898 wc->wc_checksum = 0;
1899 wc->wc_version = 1;
1900 getnanotime(&ts);
1901 wc->wc_time = ts.tv_sec;;
1902 wc->wc_timensec = ts.tv_nsec;
1903
1904 WAPBL_PRINTF(WAPBL_PRINT_WRITE,
1905 ("wapbl_write_commit: head = %"PRIdMAX "tail = %"PRIdMAX"\n",
1906 (intmax_t)head, (intmax_t)tail));
1907
1908 /*
1909 * XXX if generation will rollover, then first zero
1910 * over second commit header before trying to write both headers.
1911 */
1912
1913 error = wapbl_write(wc, wc->wc_len, wl->wl_devvp,
1914 wl->wl_logpbn + wc->wc_generation % 2);
1915 if (error)
1916 return error;
1917
1918 if (wapbl_flush_disk_cache) {
1919 error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force,
1920 FWRITE, FSCRED);
1921 if (error) {
1922 WAPBL_PRINTF(WAPBL_PRINT_ERROR,
1923 ("wapbl_write_commit: DIOCCACHESYNC on dev 0x%x "
1924 "returned %d\n", wl->wl_devvp->v_rdev, error));
1925 }
1926 }
1927
1928 /*
1929 * If the generation number was zero, write it out a second time.
1930 * This handles initialization and generation number rollover
1931 */
1932 if (wc->wc_generation++ == 0) {
1933 error = wapbl_write_commit(wl, head, tail);
1934 /*
1935 * This panic should be able to be removed if we do the
1936 * zero'ing mentioned above, and we are certain to roll
1937 * back generation number on failure.
1938 */
1939 if (error)
1940 panic("wapbl_write_commit: error writing duplicate "
1941 "log header: %d\n", error);
1942 }
1943 return 0;
1944 }
1945
1946 /* Returns new offset value */
1947 static int
1948 wapbl_write_blocks(struct wapbl *wl, off_t *offp)
1949 {
1950 struct wapbl_wc_blocklist *wc =
1951 (struct wapbl_wc_blocklist *)wl->wl_wc_scratch;
1952 int blocklen = 1<<wl->wl_log_dev_bshift;
1953 int bph;
1954 struct buf *bp;
1955 off_t off = *offp;
1956 int error;
1957
1958 KASSERT(rw_write_held(&wl->wl_rwlock));
1959
1960 bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
1961 sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
1962
1963 bp = LIST_FIRST(&wl->wl_bufs);
1964
1965 while (bp) {
1966 int cnt;
1967 struct buf *obp = bp;
1968
1969 KASSERT(bp->b_flags & B_LOCKED);
1970
1971 wc->wc_type = WAPBL_WC_BLOCKS;
1972 wc->wc_len = blocklen;
1973 wc->wc_blkcount = 0;
1974 while (bp && (wc->wc_blkcount < bph)) {
1975 /*
1976 * Make sure all the physical block numbers are up to
1977 * date. If this is not always true on a given
1978 * filesystem, then VOP_BMAP must be called. We
1979 * could call VOP_BMAP here, or else in the filesystem
1980 * specific flush callback, although neither of those
1981 * solutions allow us to take the vnode lock. If a
1982 * filesystem requires that we must take the vnode lock
1983 * to call VOP_BMAP, then we can probably do it in
1984 * bwrite when the vnode lock should already be held
1985 * by the invoking code.
1986 */
1987 KASSERT((bp->b_vp->v_type == VBLK) ||
1988 (bp->b_blkno != bp->b_lblkno));
1989 KASSERT(bp->b_blkno > 0);
1990
1991 wc->wc_blocks[wc->wc_blkcount].wc_daddr = bp->b_blkno;
1992 wc->wc_blocks[wc->wc_blkcount].wc_dlen = bp->b_bcount;
1993 wc->wc_len += bp->b_bcount;
1994 wc->wc_blkcount++;
1995 bp = LIST_NEXT(bp, b_wapbllist);
1996 }
1997 WAPBL_PRINTF(WAPBL_PRINT_WRITE,
1998 ("wapbl_write_blocks: len = %u off = %"PRIdMAX"\n",
1999 wc->wc_len, (intmax_t)off));
2000
2001 error = wapbl_circ_write(wl, wc, blocklen, &off);
2002 if (error)
2003 return error;
2004 bp = obp;
2005 cnt = 0;
2006 while (bp && (cnt++ < bph)) {
2007 error = wapbl_circ_write(wl, bp->b_data,
2008 bp->b_bcount, &off);
2009 if (error)
2010 return error;
2011 bp = LIST_NEXT(bp, b_wapbllist);
2012 }
2013 }
2014 *offp = off;
2015 return 0;
2016 }
2017
2018 static int
2019 wapbl_write_revocations(struct wapbl *wl, off_t *offp)
2020 {
2021 struct wapbl_wc_blocklist *wc =
2022 (struct wapbl_wc_blocklist *)wl->wl_wc_scratch;
2023 int i;
2024 int blocklen = 1<<wl->wl_log_dev_bshift;
2025 int bph;
2026 off_t off = *offp;
2027 int error;
2028
2029 if (wl->wl_dealloccnt == 0)
2030 return 0;
2031
2032 bph = (blocklen - offsetof(struct wapbl_wc_blocklist, wc_blocks)) /
2033 sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]);
2034
2035 i = 0;
2036 while (i < wl->wl_dealloccnt) {
2037 wc->wc_type = WAPBL_WC_REVOCATIONS;
2038 wc->wc_len = blocklen;
2039 wc->wc_blkcount = 0;
2040 while ((i < wl->wl_dealloccnt) && (wc->wc_blkcount < bph)) {
2041 wc->wc_blocks[wc->wc_blkcount].wc_daddr =
2042 wl->wl_deallocblks[i];
2043 wc->wc_blocks[wc->wc_blkcount].wc_dlen =
2044 wl->wl_dealloclens[i];
2045 wc->wc_blkcount++;
2046 i++;
2047 }
2048 WAPBL_PRINTF(WAPBL_PRINT_WRITE,
2049 ("wapbl_write_revocations: len = %u off = %"PRIdMAX"\n",
2050 wc->wc_len, (intmax_t)off));
2051 error = wapbl_circ_write(wl, wc, blocklen, &off);
2052 if (error)
2053 return error;
2054 }
2055 *offp = off;
2056 return 0;
2057 }
2058
2059 static int
2060 wapbl_write_inodes(struct wapbl *wl, off_t *offp)
2061 {
2062 struct wapbl_wc_inodelist *wc =
2063 (struct wapbl_wc_inodelist *)wl->wl_wc_scratch;
2064 int i;
2065 int blocklen = 1<<wl->wl_log_dev_bshift;
2066 off_t off = *offp;
2067 int error;
2068
2069 struct wapbl_ino_head *wih;
2070 struct wapbl_ino *wi;
2071 int iph;
2072
2073 iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) /
2074 sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]);
2075
2076 i = 0;
2077 wih = &wl->wl_inohash[0];
2078 wi = 0;
2079 do {
2080 wc->wc_type = WAPBL_WC_INODES;
2081 wc->wc_len = blocklen;
2082 wc->wc_inocnt = 0;
2083 wc->wc_clear = (i == 0);
2084 while ((i < wl->wl_inohashcnt) && (wc->wc_inocnt < iph)) {
2085 while (!wi) {
2086 KASSERT((wih - &wl->wl_inohash[0])
2087 <= wl->wl_inohashmask);
2088 wi = LIST_FIRST(wih++);
2089 }
2090 wc->wc_inodes[wc->wc_inocnt].wc_inumber = wi->wi_ino;
2091 wc->wc_inodes[wc->wc_inocnt].wc_imode = wi->wi_mode;
2092 wc->wc_inocnt++;
2093 i++;
2094 wi = LIST_NEXT(wi, wi_hash);
2095 }
2096 WAPBL_PRINTF(WAPBL_PRINT_WRITE,
2097 ("wapbl_write_inodes: len = %u off = %"PRIdMAX"\n",
2098 wc->wc_len, (intmax_t)off));
2099 error = wapbl_circ_write(wl, wc, blocklen, &off);
2100 if (error)
2101 return error;
2102 } while (i < wl->wl_inohashcnt);
2103
2104 *offp = off;
2105 return 0;
2106 }
2107
2108 #endif /* _KERNEL */
2109
2110 /****************************************************************/
2111
2112 #ifdef _KERNEL
2113 static struct pool wapbl_blk_pool;
2114 static int wapbl_blk_pool_refcount;
2115 #endif
2116 struct wapbl_blk {
2117 LIST_ENTRY(wapbl_blk) wb_hash;
2118 daddr_t wb_blk;
2119 off_t wb_off; /* Offset of this block in the log */
2120 };
2121 #define WAPBL_BLKPOOL_MIN 83
2122
2123 static void
2124 wapbl_blkhash_init(struct wapbl_replay *wr, u_int size)
2125 {
2126 if (size < WAPBL_BLKPOOL_MIN)
2127 size = WAPBL_BLKPOOL_MIN;
2128 KASSERT(wr->wr_blkhash == 0);
2129 #ifdef _KERNEL
2130 wr->wr_blkhash = hashinit(size, HASH_LIST, true, &wr->wr_blkhashmask);
2131 if (atomic_inc_uint_nv(&wapbl_blk_pool_refcount) == 1) {
2132 pool_init(&wapbl_blk_pool, sizeof(struct wapbl_blk), 0, 0, 0,
2133 "wapblblkpl", &pool_allocator_nointr, IPL_NONE);
2134 }
2135 #else /* ! _KERNEL */
2136 /* Manually implement hashinit */
2137 {
2138 int i;
2139 unsigned long hashsize;
2140 for (hashsize = 1; hashsize < size; hashsize <<= 1)
2141 continue;
2142 wr->wr_blkhash = wapbl_malloc(hashsize * sizeof(*wr->wr_blkhash));
2143 for (i = 0; i < hashsize; i++)
2144 LIST_INIT(&wr->wr_blkhash[i]);
2145 wr->wr_blkhashmask = hashsize - 1;
2146 }
2147 #endif /* ! _KERNEL */
2148 }
2149
2150 static void
2151 wapbl_blkhash_free(struct wapbl_replay *wr)
2152 {
2153 KASSERT(wr->wr_blkhashcnt == 0);
2154 #ifdef _KERNEL
2155 hashdone(wr->wr_blkhash, HASH_LIST, wr->wr_blkhashmask);
2156 if (atomic_dec_uint_nv(&wapbl_blk_pool_refcount) == 0) {
2157 pool_destroy(&wapbl_blk_pool);
2158 }
2159 #else /* ! _KERNEL */
2160 wapbl_free(wr->wr_blkhash);
2161 #endif /* ! _KERNEL */
2162 }
2163
2164 static struct wapbl_blk *
2165 wapbl_blkhash_get(struct wapbl_replay *wr, daddr_t blk)
2166 {
2167 struct wapbl_blk_head *wbh;
2168 struct wapbl_blk *wb;
2169 wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask];
2170 LIST_FOREACH(wb, wbh, wb_hash) {
2171 if (blk == wb->wb_blk)
2172 return wb;
2173 }
2174 return 0;
2175 }
2176
2177 static void
2178 wapbl_blkhash_ins(struct wapbl_replay *wr, daddr_t blk, off_t off)
2179 {
2180 struct wapbl_blk_head *wbh;
2181 struct wapbl_blk *wb;
2182 wb = wapbl_blkhash_get(wr, blk);
2183 if (wb) {
2184 KASSERT(wb->wb_blk == blk);
2185 wb->wb_off = off;
2186 } else {
2187 #ifdef _KERNEL
2188 wb = pool_get(&wapbl_blk_pool, PR_WAITOK);
2189 #else /* ! _KERNEL */
2190 wb = wapbl_malloc(sizeof(*wb));
2191 #endif /* ! _KERNEL */
2192 wb->wb_blk = blk;
2193 wb->wb_off = off;
2194 wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask];
2195 LIST_INSERT_HEAD(wbh, wb, wb_hash);
2196 wr->wr_blkhashcnt++;
2197 }
2198 }
2199
2200 static void
2201 wapbl_blkhash_rem(struct wapbl_replay *wr, daddr_t blk)
2202 {
2203 struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk);
2204 if (wb) {
2205 KASSERT(wr->wr_blkhashcnt > 0);
2206 wr->wr_blkhashcnt--;
2207 LIST_REMOVE(wb, wb_hash);
2208 #ifdef _KERNEL
2209 pool_put(&wapbl_blk_pool, wb);
2210 #else /* ! _KERNEL */
2211 wapbl_free(wb);
2212 #endif /* ! _KERNEL */
2213 }
2214 }
2215
2216 static void
2217 wapbl_blkhash_clear(struct wapbl_replay *wr)
2218 {
2219 int i;
2220 for (i = 0; i <= wr->wr_blkhashmask; i++) {
2221 struct wapbl_blk *wb;
2222
2223 while ((wb = LIST_FIRST(&wr->wr_blkhash[i]))) {
2224 KASSERT(wr->wr_blkhashcnt > 0);
2225 wr->wr_blkhashcnt--;
2226 LIST_REMOVE(wb, wb_hash);
2227 #ifdef _KERNEL
2228 pool_put(&wapbl_blk_pool, wb);
2229 #else /* ! _KERNEL */
2230 wapbl_free(wb);
2231 #endif /* ! _KERNEL */
2232 }
2233 }
2234 KASSERT(wr->wr_blkhashcnt == 0);
2235 }
2236
2237 /****************************************************************/
2238
2239 static int
2240 wapbl_circ_read(struct wapbl_replay *wr, void *data, size_t len, off_t *offp)
2241 {
2242 size_t slen;
2243 struct wapbl_wc_header *wc = &wr->wr_wc_header;
2244 off_t off = *offp;
2245 int error;
2246
2247 KASSERT(((len >> wc->wc_log_dev_bshift) <<
2248 wc->wc_log_dev_bshift) == len);
2249 if (off < wc->wc_circ_off)
2250 off = wc->wc_circ_off;
2251 slen = wc->wc_circ_off + wc->wc_circ_size - off;
2252 if (slen < len) {
2253 error = wapbl_read(data, slen, wr->wr_devvp,
2254 wr->wr_logpbn + (off >> wc->wc_log_dev_bshift));
2255 if (error)
2256 return error;
2257 data = (uint8_t *)data + slen;
2258 len -= slen;
2259 off = wc->wc_circ_off;
2260 }
2261 error = wapbl_read(data, len, wr->wr_devvp,
2262 wr->wr_logpbn + (off >> wc->wc_log_dev_bshift));
2263 if (error)
2264 return error;
2265 off += len;
2266 if (off >= wc->wc_circ_off + wc->wc_circ_size)
2267 off = wc->wc_circ_off;
2268 *offp = off;
2269 return 0;
2270 }
2271
2272 static void
2273 wapbl_circ_advance(struct wapbl_replay *wr, size_t len, off_t *offp)
2274 {
2275 size_t slen;
2276 struct wapbl_wc_header *wc = &wr->wr_wc_header;
2277 off_t off = *offp;
2278
2279 KASSERT(((len >> wc->wc_log_dev_bshift) <<
2280 wc->wc_log_dev_bshift) == len);
2281
2282 if (off < wc->wc_circ_off)
2283 off = wc->wc_circ_off;
2284 slen = wc->wc_circ_off + wc->wc_circ_size - off;
2285 if (slen < len) {
2286 len -= slen;
2287 off = wc->wc_circ_off;
2288 }
2289 off += len;
2290 if (off >= wc->wc_circ_off + wc->wc_circ_size)
2291 off = wc->wc_circ_off;
2292 *offp = off;
2293 }
2294
2295 /****************************************************************/
2296
2297 int
2298 wapbl_replay_start(struct wapbl_replay **wrp, struct vnode *vp,
2299 daddr_t off, size_t count, size_t blksize)
2300 {
2301 struct wapbl_replay *wr;
2302 int error;
2303 struct vnode *devvp;
2304 daddr_t logpbn;
2305 uint8_t *scratch;
2306 struct wapbl_wc_header *wch;
2307 struct wapbl_wc_header *wch2;
2308 /* Use this until we read the actual log header */
2309 int log_dev_bshift = DEV_BSHIFT;
2310 size_t used;
2311
2312 WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
2313 ("wapbl_replay_start: vp=%p off=%"PRId64 " count=%zu blksize=%zu\n",
2314 vp, off, count, blksize));
2315
2316 if (off < 0)
2317 return EINVAL;
2318
2319 if (blksize < DEV_BSIZE)
2320 return EINVAL;
2321 if (blksize % DEV_BSIZE)
2322 return EINVAL;
2323
2324 #ifdef _KERNEL
2325 #if 0
2326 /* XXX vp->v_size isn't reliably set for VBLK devices,
2327 * especially root. However, we might still want to verify
2328 * that the full load is readable */
2329 if ((off + count) * blksize > vp->v_size)
2330 return EINVAL;
2331 #endif
2332
2333 if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, 0)) != 0) {
2334 return error;
2335 }
2336 #else /* ! _KERNEL */
2337 devvp = vp;
2338 logpbn = off;
2339 #endif /* ! _KERNEL */
2340
2341 scratch = wapbl_malloc(MAXBSIZE);
2342
2343 error = wapbl_read(scratch, 2<<log_dev_bshift, devvp, logpbn);
2344 if (error)
2345 goto errout;
2346
2347 wch = (struct wapbl_wc_header *)scratch;
2348 wch2 =
2349 (struct wapbl_wc_header *)(scratch + (1<<log_dev_bshift));
2350 /* XXX verify checksums and magic numbers */
2351 if (wch->wc_type != WAPBL_WC_HEADER) {
2352 printf("Unrecognized wapbl magic: 0x%08x\n", wch->wc_type);
2353 error = EFTYPE;
2354 goto errout;
2355 }
2356
2357 if (wch2->wc_generation > wch->wc_generation)
2358 wch = wch2;
2359
2360 wr = wapbl_calloc(1, sizeof(*wr));
2361
2362 wr->wr_logvp = vp;
2363 wr->wr_devvp = devvp;
2364 wr->wr_logpbn = logpbn;
2365
2366 wr->wr_scratch = scratch;
2367
2368 memcpy(&wr->wr_wc_header, wch, sizeof(wr->wr_wc_header));
2369
2370 used = wapbl_space_used(wch->wc_circ_size, wch->wc_head, wch->wc_tail);
2371
2372 WAPBL_PRINTF(WAPBL_PRINT_REPLAY,
2373 ("wapbl_replay: head=%"PRId64" tail=%"PRId64" off=%"PRId64
2374 " len=%"PRId64" used=%zu\n",
2375 wch->wc_head, wch->wc_tail, wch->wc_circ_off,
2376 wch->wc_circ_size, used));
2377
2378 wapbl_blkhash_init(wr, (used >> wch->wc_fs_dev_bshift));
2379 error = wapbl_replay_prescan(wr);
2380 if (error) {
2381 wapbl_replay_stop(wr);
2382 wapbl_replay_free(wr);
2383 return error;
2384 }
2385
2386 error = wapbl_replay_get_inodes(wr);
2387 if (error) {
2388 wapbl_replay_stop(wr);
2389 wapbl_replay_free(wr);
2390 return error;
2391 }
2392
2393 *wrp = wr;
2394 return 0;
2395
2396 errout:
2397 wapbl_free(scratch);
2398 return error;
2399 }
2400
2401 void
2402 wapbl_replay_stop(struct wapbl_replay *wr)
2403 {
2404
2405 WAPBL_PRINTF(WAPBL_PRINT_REPLAY, ("wapbl_replay_stop called\n"));
2406
2407 KDASSERT(wapbl_replay_isopen(wr));
2408
2409 wapbl_free(wr->wr_scratch);
2410 wr->wr_scratch = 0;
2411
2412 wr->wr_logvp = 0;
2413
2414 wapbl_blkhash_clear(wr);
2415 wapbl_blkhash_free(wr);
2416 }
2417
2418 void
2419 wapbl_replay_free(struct wapbl_replay *wr)
2420 {
2421
2422 KDASSERT(!wapbl_replay_isopen(wr));
2423
2424 if (wr->wr_inodes)
2425 wapbl_free(wr->wr_inodes);
2426 wapbl_free(wr);
2427 }
2428
2429 int
2430 wapbl_replay_isopen1(struct wapbl_replay *wr)
2431 {
2432
2433 return wapbl_replay_isopen(wr);
2434 }
2435
2436 static int
2437 wapbl_replay_prescan(struct wapbl_replay *wr)
2438 {
2439 off_t off;
2440 struct wapbl_wc_header *wch = &wr->wr_wc_header;
2441 int error;
2442
2443 int logblklen = 1<<wch->wc_log_dev_bshift;
2444 int fsblklen = 1<<wch->wc_fs_dev_bshift;
2445
2446 wapbl_blkhash_clear(wr);
2447
2448 off = wch->wc_tail;
2449 while (off != wch->wc_head) {
2450 struct wapbl_wc_null *wcn;
2451 off_t saveoff = off;
2452 error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off);
2453 if (error)
2454 goto errout;
2455 wcn = (struct wapbl_wc_null *)wr->wr_scratch;
2456 switch (wcn->wc_type) {
2457 case WAPBL_WC_BLOCKS:
2458 {
2459 struct wapbl_wc_blocklist *wc =
2460 (struct wapbl_wc_blocklist *)wr->wr_scratch;
2461 int i;
2462 for (i = 0; i < wc->wc_blkcount; i++) {
2463 int j, n;
2464 /*
2465 * Enter each physical block into the
2466 * hashtable independently
2467 */
2468 n = wc->wc_blocks[i].wc_dlen >>
2469 wch->wc_fs_dev_bshift;
2470 for (j = 0; j < n; j++) {
2471 wapbl_blkhash_ins(wr,
2472 wc->wc_blocks[i].wc_daddr + j,
2473 off);
2474 wapbl_circ_advance(wr,
2475 fsblklen, &off);
2476 }
2477 }
2478 }
2479 break;
2480
2481 case WAPBL_WC_REVOCATIONS:
2482 {
2483 struct wapbl_wc_blocklist *wc =
2484 (struct wapbl_wc_blocklist *)wr->wr_scratch;
2485 int i;
2486 for (i = 0; i < wc->wc_blkcount; i++) {
2487 int j, n;
2488 /*
2489 * Remove any blocks found from the
2490 * hashtable
2491 */
2492 n = wc->wc_blocks[i].wc_dlen >>
2493 wch->wc_fs_dev_bshift;
2494 for (j = 0; j < n; j++) {
2495 wapbl_blkhash_rem(wr,
2496 wc->wc_blocks[i].wc_daddr + j);
2497 }
2498 }
2499 }
2500 break;
2501
2502 case WAPBL_WC_INODES:
2503 {
2504 struct wapbl_wc_inodelist *wc =
2505 (struct wapbl_wc_inodelist *)wr->wr_scratch;
2506 /*
2507 * Keep track of where we found this so we
2508 * can use it later
2509 */
2510 if (wc->wc_clear) {
2511 wr->wr_inodestail = saveoff;
2512 wr->wr_inodescnt = 0;
2513 }
2514 if (wr->wr_inodestail)
2515 wr->wr_inodeshead = off;
2516 wr->wr_inodescnt += wc->wc_inocnt;
2517 }
2518 break;
2519 default:
2520 printf("Unrecognized wapbl type: 0x%08x\n",
2521 wcn->wc_type);
2522 error = EFTYPE;
2523 goto errout;
2524 }
2525 wapbl_circ_advance(wr, wcn->wc_len, &saveoff);
2526 if (off != saveoff) {
2527 printf("wapbl_replay: corrupted records\n");
2528 error = EFTYPE;
2529 goto errout;
2530 }
2531 }
2532 return 0;
2533
2534 errout:
2535 wapbl_blkhash_clear(wr);
2536 return error;
2537 }
2538
2539 static int
2540 wapbl_replay_get_inodes(struct wapbl_replay *wr)
2541 {
2542 off_t off;
2543 struct wapbl_wc_header *wch = &wr->wr_wc_header;
2544 int logblklen = 1<<wch->wc_log_dev_bshift;
2545 int cnt= 0;
2546
2547 KDASSERT(wapbl_replay_isopen(wr));
2548
2549 if (wr->wr_inodescnt == 0)
2550 return 0;
2551
2552 KASSERT(!wr->wr_inodes);
2553
2554 wr->wr_inodes = wapbl_malloc(wr->wr_inodescnt*sizeof(wr->wr_inodes[0]));
2555
2556 off = wr->wr_inodestail;
2557
2558 while (off != wr->wr_inodeshead) {
2559 struct wapbl_wc_null *wcn;
2560 int error;
2561 off_t saveoff = off;
2562 error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off);
2563 if (error) {
2564 wapbl_free(wr->wr_inodes);
2565 wr->wr_inodes = 0;
2566 return error;
2567 }
2568 wcn = (struct wapbl_wc_null *)wr->wr_scratch;
2569 switch (wcn->wc_type) {
2570 case WAPBL_WC_BLOCKS:
2571 case WAPBL_WC_REVOCATIONS:
2572 break;
2573 case WAPBL_WC_INODES:
2574 {
2575 struct wapbl_wc_inodelist *wc =
2576 (struct wapbl_wc_inodelist *)wr->wr_scratch;
2577 /*
2578 * Keep track of where we found this so we
2579 * can use it later
2580 */
2581 if (wc->wc_clear) {
2582 cnt = 0;
2583 }
2584 /* This memcpy assumes that wr_inodes is
2585 * laid out the same as wc_inodes. */
2586 memcpy(&wr->wr_inodes[cnt], wc->wc_inodes,
2587 wc->wc_inocnt*sizeof(wc->wc_inodes[0]));
2588 cnt += wc->wc_inocnt;
2589 }
2590 break;
2591 default:
2592 KASSERT(0);
2593 }
2594 off = saveoff;
2595 wapbl_circ_advance(wr, wcn->wc_len, &off);
2596 }
2597 KASSERT(cnt == wr->wr_inodescnt);
2598 return 0;
2599 }
2600
2601 #ifdef DEBUG
2602 int
2603 wapbl_replay_verify(struct wapbl_replay *wr, struct vnode *fsdevvp)
2604 {
2605 off_t off;
2606 struct wapbl_wc_header *wch = &wr->wr_wc_header;
2607 int mismatchcnt = 0;
2608 int logblklen = 1<<wch->wc_log_dev_bshift;
2609 int fsblklen = 1<<wch->wc_fs_dev_bshift;
2610 void *scratch1 = wapbl_malloc(MAXBSIZE);
2611 void *scratch2 = wapbl_malloc(MAXBSIZE);
2612 int error = 0;
2613
2614 KDASSERT(wapbl_replay_isopen(wr));
2615
2616 off = wch->wc_tail;
2617 while (off != wch->wc_head) {
2618 struct wapbl_wc_null *wcn;
2619 #ifdef DEBUG
2620 off_t saveoff = off;
2621 #endif
2622 error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off);
2623 if (error)
2624 goto out;
2625 wcn = (struct wapbl_wc_null *)wr->wr_scratch;
2626 switch (wcn->wc_type) {
2627 case WAPBL_WC_BLOCKS:
2628 {
2629 struct wapbl_wc_blocklist *wc =
2630 (struct wapbl_wc_blocklist *)wr->wr_scratch;
2631 int i;
2632 for (i = 0; i < wc->wc_blkcount; i++) {
2633 int foundcnt = 0;
2634 int dirtycnt = 0;
2635 int j, n;
2636 /*
2637 * Check each physical block into the
2638 * hashtable independently
2639 */
2640 n = wc->wc_blocks[i].wc_dlen >>
2641 wch->wc_fs_dev_bshift;
2642 for (j = 0; j < n; j++) {
2643 struct wapbl_blk *wb =
2644 wapbl_blkhash_get(wr,
2645 wc->wc_blocks[i].wc_daddr + j);
2646 if (wb && (wb->wb_off == off)) {
2647 foundcnt++;
2648 error =
2649 wapbl_circ_read(wr,
2650 scratch1, fsblklen,
2651 &off);
2652 if (error)
2653 goto out;
2654 error =
2655 wapbl_read(scratch2,
2656 fsblklen, fsdevvp,
2657 wb->wb_blk);
2658 if (error)
2659 goto out;
2660 if (memcmp(scratch1,
2661 scratch2,
2662 fsblklen)) {
2663 printf(
2664 "wapbl_verify: mismatch block %"PRId64" at off %"PRIdMAX"\n",
2665 wb->wb_blk, (intmax_t)off);
2666 dirtycnt++;
2667 mismatchcnt++;
2668 }
2669 } else {
2670 wapbl_circ_advance(wr,
2671 fsblklen, &off);
2672 }
2673 }
2674 #if 0
2675 /*
2676 * If all of the blocks in an entry
2677 * are clean, then remove all of its
2678 * blocks from the hashtable since they
2679 * never will need replay.
2680 */
2681 if ((foundcnt != 0) &&
2682 (dirtycnt == 0)) {
2683 off = saveoff;
2684 wapbl_circ_advance(wr,
2685 logblklen, &off);
2686 for (j = 0; j < n; j++) {
2687 struct wapbl_blk *wb =
2688 wapbl_blkhash_get(wr,
2689 wc->wc_blocks[i].wc_daddr + j);
2690 if (wb &&
2691 (wb->wb_off == off)) {
2692 wapbl_blkhash_rem(wr, wb->wb_blk);
2693 }
2694 wapbl_circ_advance(wr,
2695 fsblklen, &off);
2696 }
2697 }
2698 #endif
2699 }
2700 }
2701 break;
2702 case WAPBL_WC_REVOCATIONS:
2703 case WAPBL_WC_INODES:
2704 break;
2705 default:
2706 KASSERT(0);
2707 }
2708 #ifdef DEBUG
2709 wapbl_circ_advance(wr, wcn->wc_len, &saveoff);
2710 KASSERT(off == saveoff);
2711 #endif
2712 }
2713 out:
2714 wapbl_free(scratch1);
2715 wapbl_free(scratch2);
2716 if (!error && mismatchcnt)
2717 error = EFTYPE;
2718 return error;
2719 }
2720 #endif
2721
2722 int
2723 wapbl_replay_write(struct wapbl_replay *wr, struct vnode *fsdevvp)
2724 {
2725 off_t off;
2726 struct wapbl_wc_header *wch = &wr->wr_wc_header;
2727 int logblklen = 1<<wch->wc_log_dev_bshift;
2728 int fsblklen = 1<<wch->wc_fs_dev_bshift;
2729 void *scratch1 = wapbl_malloc(MAXBSIZE);
2730 int error = 0;
2731
2732 KDASSERT(wapbl_replay_isopen(wr));
2733
2734 /*
2735 * This parses the journal for replay, although it could
2736 * just as easily walk the hashtable instead.
2737 */
2738
2739 off = wch->wc_tail;
2740 while (off != wch->wc_head) {
2741 struct wapbl_wc_null *wcn;
2742 #ifdef DEBUG
2743 off_t saveoff = off;
2744 #endif
2745 error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off);
2746 if (error)
2747 goto out;
2748 wcn = (struct wapbl_wc_null *)wr->wr_scratch;
2749 switch (wcn->wc_type) {
2750 case WAPBL_WC_BLOCKS:
2751 {
2752 struct wapbl_wc_blocklist *wc =
2753 (struct wapbl_wc_blocklist *)wr->wr_scratch;
2754 int i;
2755 for (i = 0; i < wc->wc_blkcount; i++) {
2756 int j, n;
2757 /*
2758 * Check each physical block against
2759 * the hashtable independently
2760 */
2761 n = wc->wc_blocks[i].wc_dlen >>
2762 wch->wc_fs_dev_bshift;
2763 for (j = 0; j < n; j++) {
2764 struct wapbl_blk *wb =
2765 wapbl_blkhash_get(wr,
2766 wc->wc_blocks[i].wc_daddr + j);
2767 if (wb && (wb->wb_off == off)) {
2768 error = wapbl_circ_read(
2769 wr, scratch1,
2770 fsblklen, &off);
2771 if (error)
2772 goto out;
2773 error =
2774 wapbl_write(scratch1,
2775 fsblklen, fsdevvp,
2776 wb->wb_blk);
2777 if (error)
2778 goto out;
2779 } else {
2780 wapbl_circ_advance(wr,
2781 fsblklen, &off);
2782 }
2783 }
2784 }
2785 }
2786 break;
2787 case WAPBL_WC_REVOCATIONS:
2788 case WAPBL_WC_INODES:
2789 break;
2790 default:
2791 KASSERT(0);
2792 }
2793 #ifdef DEBUG
2794 wapbl_circ_advance(wr, wcn->wc_len, &saveoff);
2795 KASSERT(off == saveoff);
2796 #endif
2797 }
2798 out:
2799 wapbl_free(scratch1);
2800 return error;
2801 }
2802
2803 int
2804 wapbl_replay_read(struct wapbl_replay *wr, void *data, daddr_t blk, long len)
2805 {
2806 struct wapbl_wc_header *wch = &wr->wr_wc_header;
2807 int fsblklen = 1<<wch->wc_fs_dev_bshift;
2808
2809 KDASSERT(wapbl_replay_isopen(wr));
2810
2811 KASSERT((len % fsblklen) == 0);
2812
2813 while (len != 0) {
2814 struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk);
2815 if (wb) {
2816 off_t off = wb->wb_off;
2817 int error;
2818 error = wapbl_circ_read(wr, data, fsblklen, &off);
2819 if (error)
2820 return error;
2821 }
2822 data = (uint8_t *)data + fsblklen;
2823 len -= fsblklen;
2824 blk++;
2825 }
2826 return 0;
2827 }
Cache object: c4d4f9b48163750f99a1ecb936c8f07b
|