xref: /dragonfly/sbin/hammer/ondisk.c (revision 235099c3)
1 /*
2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * $DragonFly: src/sbin/hammer/ondisk.c,v 1.25 2008/08/21 23:28:43 thomas Exp $
35  */
36 
37 #include <sys/types.h>
38 #include <assert.h>
39 #include <stdio.h>
40 #include <stdlib.h>
41 #include <stdarg.h>
42 #include <string.h>
43 #include <unistd.h>
44 #include <stddef.h>
45 #include <err.h>
46 #include <fcntl.h>
47 #include "hammer_util.h"
48 
49 static void *alloc_blockmap(int zone, int bytes, hammer_off_t *result_offp,
50 			struct buffer_info **bufferp);
51 static hammer_off_t alloc_bigblock(struct volume_info *volume, int zone);
52 static void get_buffer_readahead(struct buffer_info *base);
53 #if 0
54 static void init_fifo_head(hammer_fifo_head_t head, u_int16_t hdr_type);
55 static hammer_off_t hammer_alloc_fifo(int32_t base_bytes, int32_t ext_bytes,
56 			struct buffer_info **bufp, u_int16_t hdr_type);
57 static void readhammerbuf(struct volume_info *vol, void *data,
58 			int64_t offset);
59 #endif
60 static void writehammerbuf(struct volume_info *vol, const void *data,
61 			int64_t offset);
62 
63 int DebugOpt;
64 
65 uuid_t Hammer_FSType;
66 uuid_t Hammer_FSId;
67 int64_t BootAreaSize;
68 int64_t MemAreaSize;
69 int64_t UndoBufferSize;
70 int     UsingSuperClusters;
71 int     NumVolumes;
72 int	RootVolNo = -1;
73 int	UseReadBehind = -4;
74 int	UseReadAhead = 4;
75 struct volume_list VolList = TAILQ_HEAD_INITIALIZER(VolList);
76 
77 static __inline
78 int
79 buffer_hash(hammer_off_t buf_offset)
80 {
81 	int hi;
82 
83 	hi = (int)(buf_offset / HAMMER_BUFSIZE) & HAMMER_BUFLISTMASK;
84 	return(hi);
85 }
86 
87 /*
88  * Lookup the requested information structure and related on-disk buffer.
89  * Missing structures are created.
90  */
91 struct volume_info *
92 setup_volume(int32_t vol_no, const char *filename, int isnew, int oflags)
93 {
94 	struct volume_info *vol;
95 	struct volume_info *scan;
96 	struct hammer_volume_ondisk *ondisk;
97 	int i, n;
98 
99 	/*
100 	 * Allocate the volume structure
101 	 */
102 	vol = malloc(sizeof(*vol));
103 	bzero(vol, sizeof(*vol));
104 	for (i = 0; i < HAMMER_BUFLISTS; ++i)
105 		TAILQ_INIT(&vol->buffer_lists[i]);
106 	vol->name = strdup(filename);
107 	vol->fd = open(filename, oflags);
108 	if (vol->fd < 0) {
109 		free(vol->name);
110 		free(vol);
111 		err(1, "setup_volume: %s: Open failed", filename);
112 	}
113 
114 	/*
115 	 * Read or initialize the volume header
116 	 */
117 	vol->ondisk = ondisk = malloc(HAMMER_BUFSIZE);
118 	if (isnew > 0) {
119 		bzero(ondisk, HAMMER_BUFSIZE);
120 	} else {
121 		n = pread(vol->fd, ondisk, HAMMER_BUFSIZE, 0);
122 		if (n != HAMMER_BUFSIZE) {
123 			err(1, "setup_volume: %s: Read failed at offset 0",
124 			    filename);
125 		}
126 		vol_no = ondisk->vol_no;
127 		if (RootVolNo < 0) {
128 			RootVolNo = ondisk->vol_rootvol;
129 		} else if (RootVolNo != (int)ondisk->vol_rootvol) {
130 			errx(1, "setup_volume: %s: root volume disagreement: "
131 				"%d vs %d",
132 				vol->name, RootVolNo, ondisk->vol_rootvol);
133 		}
134 
135 		if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType)) != 0) {
136 			errx(1, "setup_volume: %s: Header does not indicate "
137 				"that this is a hammer volume", vol->name);
138 		}
139 		if (TAILQ_EMPTY(&VolList)) {
140 			Hammer_FSId = vol->ondisk->vol_fsid;
141 		} else if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId)) != 0) {
142 			errx(1, "setup_volume: %s: FSId does match other "
143 				"volumes!", vol->name);
144 		}
145 	}
146 	vol->vol_no = vol_no;
147 
148 	if (isnew > 0) {
149 		/*init_fifo_head(&ondisk->head, HAMMER_HEAD_TYPE_VOL);*/
150 		vol->cache.modified = 1;
151         }
152 
153 	/*
154 	 * Link the volume structure in
155 	 */
156 	TAILQ_FOREACH(scan, &VolList, entry) {
157 		if (scan->vol_no == vol_no) {
158 			errx(1, "setup_volume %s: Duplicate volume number %d "
159 				"against %s", filename, vol_no, scan->name);
160 		}
161 	}
162 	TAILQ_INSERT_TAIL(&VolList, vol, entry);
163 	return(vol);
164 }
165 
166 struct volume_info *
167 get_volume(int32_t vol_no)
168 {
169 	struct volume_info *vol;
170 
171 	TAILQ_FOREACH(vol, &VolList, entry) {
172 		if (vol->vol_no == vol_no)
173 			break;
174 	}
175 	if (vol == NULL)
176 		errx(1, "get_volume: Volume %d does not exist!", vol_no);
177 	++vol->cache.refs;
178 	/* not added to or removed from hammer cache */
179 	return(vol);
180 }
181 
182 void
183 rel_volume(struct volume_info *volume)
184 {
185 	/* not added to or removed from hammer cache */
186 	--volume->cache.refs;
187 }
188 
189 /*
190  * Acquire the specified buffer.
191  */
192 struct buffer_info *
193 get_buffer(hammer_off_t buf_offset, int isnew)
194 {
195 	void *ondisk;
196 	struct buffer_info *buf;
197 	struct volume_info *volume;
198 	hammer_off_t orig_offset = buf_offset;
199 	int vol_no;
200 	int zone;
201 	int hi, n;
202 	int dora = 0;
203 
204 	zone = HAMMER_ZONE_DECODE(buf_offset);
205 	if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX) {
206 		buf_offset = blockmap_lookup(buf_offset, NULL, NULL);
207 	}
208 	assert((buf_offset & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_RAW_BUFFER);
209 	vol_no = HAMMER_VOL_DECODE(buf_offset);
210 	volume = get_volume(vol_no);
211 	buf_offset &= ~HAMMER_BUFMASK64;
212 
213 	hi = buffer_hash(buf_offset);
214 
215 	TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry) {
216 		if (buf->buf_offset == buf_offset)
217 			break;
218 	}
219 	if (buf == NULL) {
220 		buf = malloc(sizeof(*buf));
221 		bzero(buf, sizeof(*buf));
222 		if (DebugOpt) {
223 			fprintf(stderr, "get_buffer %016llx %016llx\n",
224 				(long long)orig_offset, (long long)buf_offset);
225 		}
226 		buf->buf_offset = buf_offset;
227 		buf->raw_offset = volume->ondisk->vol_buf_beg +
228 				  (buf_offset & HAMMER_OFF_SHORT_MASK);
229 		buf->volume = volume;
230 		TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry);
231 		++volume->cache.refs;
232 		buf->cache.u.buffer = buf;
233 		hammer_cache_add(&buf->cache, ISBUFFER);
234 		dora = (isnew == 0);
235 		if (isnew < 0)
236 			buf->flags |= HAMMER_BUFINFO_READAHEAD;
237 	} else {
238 		if (isnew >= 0) {
239 			buf->flags &= ~HAMMER_BUFINFO_READAHEAD;
240 			hammer_cache_used(&buf->cache);
241 		}
242 		++buf->use_count;
243 	}
244 	++buf->cache.refs;
245 	hammer_cache_flush();
246 	if ((ondisk = buf->ondisk) == NULL) {
247 		buf->ondisk = ondisk = malloc(HAMMER_BUFSIZE);
248 		if (isnew <= 0) {
249 			n = pread(volume->fd, ondisk, HAMMER_BUFSIZE,
250 				  buf->raw_offset);
251 			if (n != HAMMER_BUFSIZE) {
252 				err(1, "get_buffer: %s:%016llx Read failed at "
253 				       "offset %016llx",
254 				    volume->name,
255 				    (long long)buf->buf_offset,
256 				    (long long)buf->raw_offset);
257 			}
258 		}
259 	}
260 	if (isnew > 0) {
261 		bzero(ondisk, HAMMER_BUFSIZE);
262 		buf->cache.modified = 1;
263 	}
264 	if (dora)
265 		get_buffer_readahead(buf);
266 	return(buf);
267 }
268 
269 static void
270 get_buffer_readahead(struct buffer_info *base)
271 {
272 	struct buffer_info *buf;
273 	struct volume_info *vol;
274 	hammer_off_t buf_offset;
275 	int64_t raw_offset;
276 	int ri = UseReadBehind;
277 	int re = UseReadAhead;
278 	int hi;
279 
280 	raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
281 	vol = base->volume;
282 
283 	while (ri < re) {
284 		if (raw_offset >= vol->ondisk->vol_buf_end)
285 			break;
286 		if (raw_offset < vol->ondisk->vol_buf_beg) {
287 			++ri;
288 			raw_offset += HAMMER_BUFSIZE;
289 			continue;
290 		}
291 		buf_offset = HAMMER_VOL_ENCODE(vol->vol_no) |
292 			     HAMMER_ZONE_RAW_BUFFER |
293 			     (raw_offset - vol->ondisk->vol_buf_beg);
294 		hi = buffer_hash(raw_offset);
295 		TAILQ_FOREACH(buf, &vol->buffer_lists[hi], entry) {
296 			if (buf->raw_offset == raw_offset)
297 				break;
298 		}
299 		if (buf == NULL) {
300 			buf = get_buffer(buf_offset, -1);
301 			rel_buffer(buf);
302 		}
303 		++ri;
304 		raw_offset += HAMMER_BUFSIZE;
305 	}
306 }
307 
308 void
309 rel_buffer(struct buffer_info *buffer)
310 {
311 	struct volume_info *volume;
312 	int hi;
313 
314 	assert(buffer->cache.refs > 0);
315 	if (--buffer->cache.refs == 0) {
316 		if (buffer->cache.delete) {
317 			hi = buffer_hash(buffer->buf_offset);
318 			volume = buffer->volume;
319 			if (buffer->cache.modified)
320 				flush_buffer(buffer);
321 			TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
322 			hammer_cache_del(&buffer->cache);
323 			free(buffer->ondisk);
324 			free(buffer);
325 			rel_volume(volume);
326 		}
327 	}
328 }
329 
330 void *
331 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
332 		int isnew)
333 {
334 	struct buffer_info *buffer;
335 
336 	if ((buffer = *bufferp) != NULL) {
337 		if (isnew > 0 ||
338 		    ((buffer->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
339 			rel_buffer(buffer);
340 			buffer = *bufferp = NULL;
341 		}
342 	}
343 	if (buffer == NULL)
344 		buffer = *bufferp = get_buffer(buf_offset, isnew);
345 	return((char *)buffer->ondisk + ((int32_t)buf_offset & HAMMER_BUFMASK));
346 }
347 
348 /*
349  * Retrieve a pointer to a B-Tree node given a cluster offset.  The underlying
350  * bufp is freed if non-NULL and a referenced buffer is loaded into it.
351  */
352 hammer_node_ondisk_t
353 get_node(hammer_off_t node_offset, struct buffer_info **bufp)
354 {
355 	struct buffer_info *buf;
356 
357 	if (*bufp)
358 		rel_buffer(*bufp);
359 	*bufp = buf = get_buffer(node_offset, 0);
360 	return((void *)((char *)buf->ondisk +
361 			(int32_t)(node_offset & HAMMER_BUFMASK)));
362 }
363 
364 /*
365  * Allocate HAMMER elements - btree nodes, data storage, and record elements
366  *
367  * NOTE: hammer_alloc_fifo() initializes the fifo header for the returned
368  * item and zero's out the remainder, so don't bzero() it.
369  */
370 void *
371 alloc_btree_element(hammer_off_t *offp)
372 {
373 	struct buffer_info *buffer = NULL;
374 	hammer_node_ondisk_t node;
375 
376 	node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
377 			      offp, &buffer);
378 	bzero(node, sizeof(*node));
379 	/* XXX buffer not released, pointer remains valid */
380 	return(node);
381 }
382 
383 void *
384 alloc_data_element(hammer_off_t *offp, int32_t data_len,
385 		   struct buffer_info **data_bufferp)
386 {
387 	void *data;
388 
389 	if (data_len >= HAMMER_BUFSIZE) {
390 		assert(data_len <= HAMMER_BUFSIZE); /* just one buffer */
391 		data = alloc_blockmap(HAMMER_ZONE_LARGE_DATA_INDEX, data_len,
392 				      offp, data_bufferp);
393 		bzero(data, data_len);
394 	} else if (data_len) {
395 		data = alloc_blockmap(HAMMER_ZONE_SMALL_DATA_INDEX, data_len,
396 				      offp, data_bufferp);
397 		bzero(data, data_len);
398 	} else {
399 		data = NULL;
400 	}
401 	return (data);
402 }
403 
404 /*
405  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
406  * code will load each volume's freemap.
407  */
408 void
409 format_freemap(struct volume_info *root_vol, hammer_blockmap_t blockmap)
410 {
411 	struct buffer_info *buffer = NULL;
412 	hammer_off_t layer1_offset;
413 	struct hammer_blockmap_layer1 *layer1;
414 	int i, isnew;
415 
416 	layer1_offset = alloc_bigblock(root_vol, HAMMER_ZONE_FREEMAP_INDEX);
417 	for (i = 0; i < (int)HAMMER_BLOCKMAP_RADIX1; ++i) {
418 		isnew = ((i % HAMMER_BLOCKMAP_RADIX1_PERBUFFER) == 0);
419 		layer1 = get_buffer_data(layer1_offset + i * sizeof(*layer1),
420 					 &buffer, isnew);
421 		bzero(layer1, sizeof(*layer1));
422 		layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
423 		layer1->blocks_free = 0;
424 		layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
425 	}
426 	rel_buffer(buffer);
427 
428 	blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
429 	blockmap->phys_offset = layer1_offset;
430 	blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
431 	blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
432 	blockmap->reserved01 = 0;
433 	blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
434 	root_vol->cache.modified = 1;
435 }
436 
437 /*
438  * Load the volume's remaining free space into the freemap.
439  *
440  * Returns the number of bigblocks available.
441  */
442 int64_t
443 initialize_freemap(struct volume_info *vol)
444 {
445 	struct volume_info *root_vol;
446 	struct buffer_info *buffer1 = NULL;
447 	struct buffer_info *buffer2 = NULL;
448 	struct hammer_blockmap_layer1 *layer1;
449 	struct hammer_blockmap_layer2 *layer2;
450 	hammer_off_t layer1_base;
451 	hammer_off_t layer1_offset;
452 	hammer_off_t layer2_offset;
453 	hammer_off_t phys_offset;
454 	hammer_off_t aligned_vol_free_end;
455 	int64_t count = 0;
456 	int modified1 = 0;
457 
458 	root_vol = get_volume(RootVolNo);
459 	aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
460 				& ~HAMMER_BLOCKMAP_LAYER2_MASK;
461 
462 	printf("initialize freemap volume %d\n", vol->vol_no);
463 
464 	/*
465 	 * Initialize the freemap.  First preallocate the bigblocks required
466 	 * to implement layer2.   This preallocation is a bootstrap allocation
467 	 * using blocks from the target volume.
468 	 */
469 	layer1_base = root_vol->ondisk->vol0_blockmap[
470 					HAMMER_ZONE_FREEMAP_INDEX].phys_offset;
471 	for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
472 	     phys_offset < aligned_vol_free_end;
473 	     phys_offset += HAMMER_BLOCKMAP_LAYER2) {
474 		layer1_offset = layer1_base +
475 				HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
476 		layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
477 		if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
478 			layer1->phys_offset = alloc_bigblock(vol,
479 						HAMMER_ZONE_FREEMAP_INDEX);
480 			layer1->blocks_free = 0;
481 			buffer1->cache.modified = 1;
482 			layer1->layer1_crc = crc32(layer1,
483 						   HAMMER_LAYER1_CRCSIZE);
484 		}
485 	}
486 
487 	/*
488 	 * Now fill everything in.
489 	 */
490 	for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
491 	     phys_offset < aligned_vol_free_end;
492 	     phys_offset += HAMMER_LARGEBLOCK_SIZE) {
493 		modified1 = 0;
494 		layer1_offset = layer1_base +
495 				HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
496 		layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
497 
498 		assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
499 		layer2_offset = layer1->phys_offset +
500 				HAMMER_BLOCKMAP_LAYER2_OFFSET(phys_offset);
501 
502 		layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
503 		bzero(layer2, sizeof(*layer2));
504 		if (phys_offset < vol->vol_free_off) {
505 			/*
506 			 * Fixups XXX - bigblocks already allocated as part
507 			 * of the freemap bootstrap.
508 			 */
509 			if (layer2->zone == 0) {
510 				layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
511 				layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
512 				layer2->bytes_free = 0;
513 			}
514 		} else if (phys_offset < vol->vol_free_end) {
515 			++layer1->blocks_free;
516 			buffer1->cache.modified = 1;
517 			layer2->zone = 0;
518 			layer2->append_off = 0;
519 			layer2->bytes_free = HAMMER_LARGEBLOCK_SIZE;
520 			++count;
521 			modified1 = 1;
522 		} else {
523 			layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
524 			layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
525 			layer2->bytes_free = 0;
526 		}
527 		layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
528 		buffer2->cache.modified = 1;
529 
530 		/*
531 		 * Finish-up layer 1
532 		 */
533 		if (modified1) {
534 			layer1->layer1_crc = crc32(layer1,
535 						   HAMMER_LAYER1_CRCSIZE);
536 			buffer1->cache.modified = 1;
537 		}
538 	}
539 	rel_buffer(buffer1);
540 	rel_buffer(buffer2);
541 	rel_volume(root_vol);
542 	return(count);
543 }
544 
545 /*
546  * Allocate big-blocks using our poor-man's volume->vol_free_off.
547  *
548  * If the zone is HAMMER_ZONE_FREEMAP_INDEX we are bootstrapping the freemap
549  * itself and cannot update it yet.
550  */
551 hammer_off_t
552 alloc_bigblock(struct volume_info *volume, int zone)
553 {
554 	struct buffer_info *buffer = NULL;
555 	struct volume_info *root_vol;
556 	hammer_off_t result_offset;
557 	hammer_off_t layer_offset;
558 	struct hammer_blockmap_layer1 *layer1;
559 	struct hammer_blockmap_layer2 *layer2;
560 	int didget;
561 
562 	if (volume == NULL) {
563 		volume = get_volume(RootVolNo);
564 		didget = 1;
565 	} else {
566 		didget = 0;
567 	}
568 	result_offset = volume->vol_free_off;
569 	if (result_offset >= volume->vol_free_end)
570 		panic("alloc_bigblock: Ran out of room, filesystem too small");
571 	volume->vol_free_off += HAMMER_LARGEBLOCK_SIZE;
572 
573 	/*
574 	 * Update the freemap.
575 	 */
576 	if (zone != HAMMER_ZONE_FREEMAP_INDEX) {
577 		root_vol = get_volume(RootVolNo);
578 		layer_offset = root_vol->ondisk->vol0_blockmap[
579 					HAMMER_ZONE_FREEMAP_INDEX].phys_offset;
580 		layer_offset += HAMMER_BLOCKMAP_LAYER1_OFFSET(result_offset);
581 		layer1 = get_buffer_data(layer_offset, &buffer, 0);
582 		assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
583 		--layer1->blocks_free;
584 		layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
585 		buffer->cache.modified = 1;
586 		layer_offset = layer1->phys_offset +
587 			       HAMMER_BLOCKMAP_LAYER2_OFFSET(result_offset);
588 		layer2 = get_buffer_data(layer_offset, &buffer, 0);
589 		assert(layer2->zone == 0);
590 		layer2->zone = zone;
591 		layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
592 		layer2->bytes_free = 0;
593 		layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
594 		buffer->cache.modified = 1;
595 
596 		--root_vol->ondisk->vol0_stat_freebigblocks;
597 		root_vol->cache.modified = 1;
598 
599 		rel_buffer(buffer);
600 		rel_volume(root_vol);
601 	}
602 
603 	if (didget)
604 		rel_volume(volume);
605 	return(result_offset);
606 }
607 
608 /*
609  * Format the undo-map for the root volume.
610  */
611 void
612 format_undomap(hammer_volume_ondisk_t ondisk)
613 {
614 	const int undo_zone = HAMMER_ZONE_UNDO_INDEX;
615 	hammer_off_t undo_limit;
616 	hammer_blockmap_t blockmap;
617 	hammer_off_t scan;
618 	int n;
619 	int limit_index;
620 
621 	/*
622 	 * Size the undo buffer in multiples of HAMMER_LARGEBLOCK_SIZE,
623 	 * up to HAMMER_UNDO_LAYER2 large blocks.  Size to approximately
624 	 * 0.1% of the disk.
625 	 *
626 	 * The minimum UNDO fifo size is 100MB.
627 	 */
628 	undo_limit = UndoBufferSize;
629 	if (undo_limit == 0) {
630 		undo_limit = (ondisk->vol_buf_end - ondisk->vol_buf_beg) / 1000;
631 		if (undo_limit < 100*1024*1024)
632 			undo_limit = 100*1024*1024;
633 	}
634 	undo_limit = (undo_limit + HAMMER_LARGEBLOCK_MASK64) &
635 		     ~HAMMER_LARGEBLOCK_MASK64;
636 	if (undo_limit < HAMMER_LARGEBLOCK_SIZE)
637 		undo_limit = HAMMER_LARGEBLOCK_SIZE;
638 	if (undo_limit > HAMMER_LARGEBLOCK_SIZE * HAMMER_UNDO_LAYER2)
639 		undo_limit = HAMMER_LARGEBLOCK_SIZE * HAMMER_UNDO_LAYER2;
640 	UndoBufferSize = undo_limit;
641 
642 	blockmap = &ondisk->vol0_blockmap[undo_zone];
643 	bzero(blockmap, sizeof(*blockmap));
644 	blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
645 	blockmap->first_offset = HAMMER_ZONE_ENCODE(undo_zone, 0);
646 	blockmap->next_offset = blockmap->first_offset;
647 	blockmap->alloc_offset = HAMMER_ZONE_ENCODE(undo_zone, undo_limit);
648 	blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
649 
650 	n = 0;
651 	scan = blockmap->next_offset;
652 	limit_index = undo_limit / HAMMER_LARGEBLOCK_SIZE;
653 
654 	assert(limit_index <= HAMMER_UNDO_LAYER2);
655 
656 	for (n = 0; n < limit_index; ++n) {
657 		ondisk->vol0_undo_array[n] = alloc_bigblock(NULL,
658 							HAMMER_ZONE_UNDO_INDEX);
659 		scan += HAMMER_LARGEBLOCK_SIZE;
660 	}
661 	while (n < HAMMER_UNDO_LAYER2) {
662 		ondisk->vol0_undo_array[n] = HAMMER_BLOCKMAP_UNAVAIL;
663 		++n;
664 	}
665 }
666 
667 /*
668  * Format a new blockmap.  This is mostly a degenerate case because
669  * all allocations are now actually done from the freemap.
670  */
671 void
672 format_blockmap(hammer_blockmap_t blockmap, hammer_off_t zone_base)
673 {
674 	blockmap->phys_offset = 0;
675 	blockmap->alloc_offset = zone_base | HAMMER_VOL_ENCODE(255) |
676 				 HAMMER_SHORT_OFF_ENCODE(-1);
677 	blockmap->first_offset = zone_base;
678 	blockmap->next_offset = zone_base;
679 	blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
680 }
681 
682 /*
683  * Allocate a chunk of data out of a blockmap.  This is a simplified
684  * version which uses next_offset as a simple allocation iterator.
685  */
686 static
687 void *
688 alloc_blockmap(int zone, int bytes, hammer_off_t *result_offp,
689 	       struct buffer_info **bufferp)
690 {
691 	struct buffer_info *buffer1 = NULL;
692 	struct buffer_info *buffer2 = NULL;
693 	struct volume_info *volume;
694 	hammer_blockmap_t blockmap;
695 	hammer_blockmap_t freemap;
696 	struct hammer_blockmap_layer1 *layer1;
697 	struct hammer_blockmap_layer2 *layer2;
698 	hammer_off_t layer1_offset;
699 	hammer_off_t layer2_offset;
700 	hammer_off_t zone2_offset;
701 	void *ptr;
702 
703 	volume = get_volume(RootVolNo);
704 
705 	blockmap = &volume->ondisk->vol0_blockmap[zone];
706 	freemap = &volume->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
707 
708 	/*
709 	 * Alignment and buffer-boundary issues.  If the allocation would
710 	 * cross a buffer boundary we have to skip to the next buffer.
711 	 */
712 	bytes = (bytes + 15) & ~15;
713 
714 again:
715 	if ((blockmap->next_offset ^ (blockmap->next_offset + bytes - 1)) &
716 	    ~HAMMER_BUFMASK64) {
717 		volume->cache.modified = 1;
718 		blockmap->next_offset = (blockmap->next_offset + bytes) &
719 				        ~HAMMER_BUFMASK64;
720 	}
721 
722 	/*
723 	 * Dive layer 1.  For now we can't allocate data outside of volume 0.
724 	 */
725 	layer1_offset = freemap->phys_offset +
726 			HAMMER_BLOCKMAP_LAYER1_OFFSET(blockmap->next_offset);
727 
728 	layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
729 
730 	if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
731 		fprintf(stderr, "alloc_blockmap: ran out of space!\n");
732 		exit(1);
733 	}
734 
735 	/*
736 	 * Dive layer 2
737 	 */
738 	layer2_offset = layer1->phys_offset +
739 			HAMMER_BLOCKMAP_LAYER2_OFFSET(blockmap->next_offset);
740 
741 	layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
742 
743 	if (layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX) {
744 		fprintf(stderr, "alloc_blockmap: ran out of space!\n");
745 		exit(1);
746 	}
747 
748 	/*
749 	 * If we are entering a new bigblock assign ownership to our
750 	 * zone.  If the bigblock is owned by another zone skip it.
751 	 */
752 	if (layer2->zone == 0) {
753 		--layer1->blocks_free;
754 		layer2->zone = zone;
755 		assert(layer2->bytes_free == HAMMER_LARGEBLOCK_SIZE);
756 		assert(layer2->append_off == 0);
757 	}
758 	if (layer2->zone != zone) {
759 		blockmap->next_offset = (blockmap->next_offset + HAMMER_LARGEBLOCK_SIZE) &
760 					~HAMMER_LARGEBLOCK_MASK64;
761 		goto again;
762 	}
763 
764 	buffer1->cache.modified = 1;
765 	buffer2->cache.modified = 1;
766 	volume->cache.modified = 1;
767 	assert(layer2->append_off ==
768 	       (blockmap->next_offset & HAMMER_LARGEBLOCK_MASK));
769 	layer2->bytes_free -= bytes;
770 	*result_offp = blockmap->next_offset;
771 	blockmap->next_offset += bytes;
772 	layer2->append_off = (int)blockmap->next_offset &
773 			      HAMMER_LARGEBLOCK_MASK;
774 
775 	layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
776 	layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
777 
778 	zone2_offset = (*result_offp & ~HAMMER_OFF_ZONE_MASK) |
779 			HAMMER_ZONE_ENCODE(zone, 0);
780 
781 	ptr = get_buffer_data(zone2_offset, bufferp, 0);
782 	(*bufferp)->cache.modified = 1;
783 
784 	if (buffer1)
785 		rel_buffer(buffer1);
786 	if (buffer2)
787 		rel_buffer(buffer2);
788 
789 	rel_volume(volume);
790 	return(ptr);
791 }
792 
793 /*
794  * Flush various tracking structures to disk
795  */
796 
797 /*
798  * Flush various tracking structures to disk
799  */
800 void
801 flush_all_volumes(void)
802 {
803 	struct volume_info *vol;
804 
805 	TAILQ_FOREACH(vol, &VolList, entry)
806 		flush_volume(vol);
807 }
808 
809 void
810 flush_volume(struct volume_info *volume)
811 {
812 	struct buffer_info *buffer;
813 	int i;
814 
815 	for (i = 0; i < HAMMER_BUFLISTS; ++i) {
816 		TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
817 			flush_buffer(buffer);
818 	}
819 	writehammerbuf(volume, volume->ondisk, 0);
820 	volume->cache.modified = 0;
821 }
822 
823 void
824 flush_buffer(struct buffer_info *buffer)
825 {
826 	writehammerbuf(buffer->volume, buffer->ondisk, buffer->raw_offset);
827 	buffer->cache.modified = 0;
828 }
829 
830 #if 0
831 /*
832  * Generic buffer initialization
833  */
834 static void
835 init_fifo_head(hammer_fifo_head_t head, u_int16_t hdr_type)
836 {
837 	head->hdr_signature = HAMMER_HEAD_SIGNATURE;
838 	head->hdr_type = hdr_type;
839 	head->hdr_size = 0;
840 	head->hdr_crc = 0;
841 	head->hdr_seq = 0;
842 }
843 
844 #endif
845 
846 #if 0
847 /*
848  * Core I/O operations
849  */
850 static void
851 readhammerbuf(struct volume_info *vol, void *data, int64_t offset)
852 {
853 	ssize_t n;
854 
855 	n = pread(vol->fd, data, HAMMER_BUFSIZE, offset);
856 	if (n != HAMMER_BUFSIZE)
857 		err(1, "Read volume %d (%s)", vol->vol_no, vol->name);
858 }
859 
860 #endif
861 
862 static void
863 writehammerbuf(struct volume_info *vol, const void *data, int64_t offset)
864 {
865 	ssize_t n;
866 
867 	n = pwrite(vol->fd, data, HAMMER_BUFSIZE, offset);
868 	if (n != HAMMER_BUFSIZE)
869 		err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
870 }
871 
872 void
873 panic(const char *ctl, ...)
874 {
875 	va_list va;
876 
877 	va_start(va, ctl);
878 	vfprintf(stderr, ctl, va);
879 	va_end(va);
880 	fprintf(stderr, "\n");
881 	exit(1);
882 }
883 
884