xref: /netbsd-src/sys/fs/udf/udf_allocation.c (revision a536ee5124e62c9a0051a252f7833dc8f50f44c9)
1 /* $NetBSD: udf_allocation.c,v 1.32 2011/06/16 09:21:02 hannken Exp $ */
2 
3 /*
4  * Copyright (c) 2006, 2008 Reinoud Zandijk
5  * All rights reserved.
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  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  *
27  */
28 
29 #include <sys/cdefs.h>
30 #ifndef lint
31 __KERNEL_RCSID(0, "$NetBSD: udf_allocation.c,v 1.32 2011/06/16 09:21:02 hannken Exp $");
32 #endif /* not lint */
33 
34 
35 #if defined(_KERNEL_OPT)
36 #include "opt_compat_netbsd.h"
37 #endif
38 
39 /* TODO strip */
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysctl.h>
43 #include <sys/namei.h>
44 #include <sys/proc.h>
45 #include <sys/kernel.h>
46 #include <sys/vnode.h>
47 #include <miscfs/genfs/genfs_node.h>
48 #include <sys/mount.h>
49 #include <sys/buf.h>
50 #include <sys/file.h>
51 #include <sys/device.h>
52 #include <sys/disklabel.h>
53 #include <sys/ioctl.h>
54 #include <sys/malloc.h>
55 #include <sys/dirent.h>
56 #include <sys/stat.h>
57 #include <sys/conf.h>
58 #include <sys/kauth.h>
59 #include <sys/kthread.h>
60 #include <dev/clock_subr.h>
61 
62 #include <fs/udf/ecma167-udf.h>
63 #include <fs/udf/udf_mount.h>
64 
65 #include "udf.h"
66 #include "udf_subr.h"
67 #include "udf_bswap.h"
68 
69 
70 #define VTOI(vnode) ((struct udf_node *) vnode->v_data)
71 
72 static void udf_record_allocation_in_node(struct udf_mount *ump,
73 	struct buf *buf, uint16_t vpart_num, uint64_t *mapping,
74 	struct long_ad *node_ad_cpy);
75 
76 static void udf_collect_free_space_for_vpart(struct udf_mount *ump,
77 	uint16_t vpart_num, uint32_t num_lb);
78 
79 static int udf_ads_merge(uint32_t max_len, uint32_t lb_size, struct long_ad *a1, struct long_ad *a2);
80 static void udf_wipe_adslots(struct udf_node *udf_node);
81 static void udf_count_alloc_exts(struct udf_node *udf_node);
82 
83 
84 /* --------------------------------------------------------------------- */
85 
86 #if 0
87 #if 1
88 static void
89 udf_node_dump(struct udf_node *udf_node) {
90 	struct file_entry    *fe;
91 	struct extfile_entry *efe;
92 	struct icb_tag *icbtag;
93 	struct long_ad s_ad;
94 	uint64_t inflen;
95 	uint32_t icbflags, addr_type;
96 	uint32_t len, lb_num;
97 	uint32_t flags;
98 	int part_num;
99 	int lb_size, eof, slot;
100 
101 	if ((udf_verbose & UDF_DEBUG_NODEDUMP) == 0)
102 		return;
103 
104 	lb_size = udf_rw32(udf_node->ump->logical_vol->lb_size);
105 
106 	fe  = udf_node->fe;
107 	efe = udf_node->efe;
108 	if (fe) {
109 		icbtag = &fe->icbtag;
110 		inflen = udf_rw64(fe->inf_len);
111 	} else {
112 		icbtag = &efe->icbtag;
113 		inflen = udf_rw64(efe->inf_len);
114 	}
115 
116 	icbflags   = udf_rw16(icbtag->flags);
117 	addr_type  = icbflags & UDF_ICB_TAG_FLAGS_ALLOC_MASK;
118 
119 	printf("udf_node_dump %p :\n", udf_node);
120 
121 	if (addr_type == UDF_ICB_INTERN_ALLOC) {
122 		printf("\tIntern alloc, len = %"PRIu64"\n", inflen);
123 		return;
124 	}
125 
126 	printf("\tInflen  = %"PRIu64"\n", inflen);
127 	printf("\t\t");
128 
129 	slot = 0;
130 	for (;;) {
131 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
132 		if (eof)
133 			break;
134 		part_num = udf_rw16(s_ad.loc.part_num);
135 		lb_num = udf_rw32(s_ad.loc.lb_num);
136 		len   = udf_rw32(s_ad.len);
137 		flags = UDF_EXT_FLAGS(len);
138 		len   = UDF_EXT_LEN(len);
139 
140 		printf("[");
141 		if (part_num >= 0)
142 			printf("part %d, ", part_num);
143 		printf("lb_num %d, len %d", lb_num, len);
144 		if (flags)
145 			printf(", flags %d", flags>>30);
146 		printf("] ");
147 
148 		if (flags == UDF_EXT_REDIRECT) {
149 			printf("\n\textent END\n\tallocation extent\n\t\t");
150 		}
151 
152 		slot++;
153 	}
154 	printf("\n\tl_ad END\n\n");
155 }
156 #else
157 #define udf_node_dump(a)
158 #endif
159 
160 
161 static void
162 udf_assert_allocated(struct udf_mount *ump, uint16_t vpart_num,
163 	uint32_t lb_num, uint32_t num_lb)
164 {
165 	struct udf_bitmap *bitmap;
166 	struct part_desc *pdesc;
167 	uint32_t ptov;
168 	uint32_t bitval;
169 	uint8_t *bpos;
170 	int bit;
171 	int phys_part;
172 	int ok;
173 
174 	DPRINTF(PARANOIA, ("udf_assert_allocated: check virt lbnum %d "
175 			  "part %d + %d sect\n", lb_num, vpart_num, num_lb));
176 
177 	/* get partition backing up this vpart_num */
178 	pdesc = ump->partitions[ump->vtop[vpart_num]];
179 
180 	switch (ump->vtop_tp[vpart_num]) {
181 	case UDF_VTOP_TYPE_PHYS :
182 	case UDF_VTOP_TYPE_SPARABLE :
183 		/* free space to freed or unallocated space bitmap */
184 		ptov      = udf_rw32(pdesc->start_loc);
185 		phys_part = ump->vtop[vpart_num];
186 
187 		/* use unallocated bitmap */
188 		bitmap = &ump->part_unalloc_bits[phys_part];
189 
190 		/* if no bitmaps are defined, bail out */
191 		if (bitmap->bits == NULL)
192 			break;
193 
194 		/* check bits */
195 		KASSERT(bitmap->bits);
196 		ok = 1;
197 		bpos = bitmap->bits + lb_num/8;
198 		bit  = lb_num % 8;
199 		while (num_lb > 0) {
200 			bitval = (1 << bit);
201 			DPRINTF(PARANOIA, ("XXX : check %d, %p, bit %d\n",
202 				lb_num, bpos, bit));
203 			KASSERT(bitmap->bits + lb_num/8 == bpos);
204 			if (*bpos & bitval) {
205 				printf("\tlb_num %d is NOT marked busy\n",
206 					lb_num);
207 				ok = 0;
208 			}
209 			lb_num++; num_lb--;
210 			bit = (bit + 1) % 8;
211 			if (bit == 0)
212 				bpos++;
213 		}
214 		if (!ok) {
215 			/* KASSERT(0); */
216 		}
217 
218 		break;
219 	case UDF_VTOP_TYPE_VIRT :
220 		/* TODO check space */
221 		KASSERT(num_lb == 1);
222 		break;
223 	case UDF_VTOP_TYPE_META :
224 		/* TODO check space in the metadata bitmap */
225 	default:
226 		/* not implemented */
227 		break;
228 	}
229 }
230 
231 
232 static void
233 udf_node_sanity_check(struct udf_node *udf_node,
234 		uint64_t *cnt_inflen, uint64_t *cnt_logblksrec)
235 {
236 	union dscrptr *dscr;
237 	struct file_entry    *fe;
238 	struct extfile_entry *efe;
239 	struct icb_tag *icbtag;
240 	struct long_ad  s_ad;
241 	uint64_t inflen, logblksrec;
242 	uint32_t icbflags, addr_type;
243 	uint32_t len, lb_num, l_ea, l_ad, max_l_ad;
244 	uint16_t part_num;
245 	uint8_t *data_pos;
246 	int dscr_size, lb_size, flags, whole_lb;
247 	int i, slot, eof;
248 
249 //	KASSERT(mutex_owned(&udf_node->ump->allocate_mutex));
250 
251 	if (1)
252 		udf_node_dump(udf_node);
253 
254 	lb_size = udf_rw32(udf_node->ump->logical_vol->lb_size);
255 
256 	fe  = udf_node->fe;
257 	efe = udf_node->efe;
258 	if (fe) {
259 		dscr       = (union dscrptr *) fe;
260 		icbtag     = &fe->icbtag;
261 		inflen     = udf_rw64(fe->inf_len);
262 		dscr_size  = sizeof(struct file_entry) -1;
263 		logblksrec = udf_rw64(fe->logblks_rec);
264 		l_ad       = udf_rw32(fe->l_ad);
265 		l_ea       = udf_rw32(fe->l_ea);
266 	} else {
267 		dscr       = (union dscrptr *) efe;
268 		icbtag     = &efe->icbtag;
269 		inflen     = udf_rw64(efe->inf_len);
270 		dscr_size  = sizeof(struct extfile_entry) -1;
271 		logblksrec = udf_rw64(efe->logblks_rec);
272 		l_ad       = udf_rw32(efe->l_ad);
273 		l_ea       = udf_rw32(efe->l_ea);
274 	}
275 	data_pos  = (uint8_t *) dscr + dscr_size + l_ea;
276 	max_l_ad   = lb_size - dscr_size - l_ea;
277 	icbflags   = udf_rw16(icbtag->flags);
278 	addr_type  = icbflags & UDF_ICB_TAG_FLAGS_ALLOC_MASK;
279 
280 	/* check if tail is zero */
281 	DPRINTF(PARANOIA, ("Sanity check blank tail\n"));
282 	for (i = l_ad; i < max_l_ad; i++) {
283 		if (data_pos[i] != 0)
284 			printf( "sanity_check: violation: node byte %d "
285 				"has value %d\n", i, data_pos[i]);
286 	}
287 
288 	/* reset counters */
289 	*cnt_inflen     = 0;
290 	*cnt_logblksrec = 0;
291 
292 	if (addr_type == UDF_ICB_INTERN_ALLOC) {
293 		KASSERT(l_ad <= max_l_ad);
294 		KASSERT(l_ad == inflen);
295 		*cnt_inflen = inflen;
296 		return;
297 	}
298 
299 	/* start counting */
300 	whole_lb = 1;
301 	slot = 0;
302 	for (;;) {
303 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
304 		if (eof)
305 			break;
306 		KASSERT(whole_lb == 1);
307 
308 		part_num = udf_rw16(s_ad.loc.part_num);
309 		lb_num = udf_rw32(s_ad.loc.lb_num);
310 		len   = udf_rw32(s_ad.len);
311 		flags = UDF_EXT_FLAGS(len);
312 		len   = UDF_EXT_LEN(len);
313 
314 		if (flags != UDF_EXT_REDIRECT) {
315 			*cnt_inflen += len;
316 			if (flags == UDF_EXT_ALLOCATED) {
317 				*cnt_logblksrec += (len + lb_size -1) / lb_size;
318 			}
319 		} else {
320 			KASSERT(len == lb_size);
321 		}
322 		/* check allocation */
323 		if (flags == UDF_EXT_ALLOCATED)
324 			udf_assert_allocated(udf_node->ump, part_num, lb_num,
325 				(len + lb_size - 1) / lb_size);
326 
327 		/* check whole lb */
328 		whole_lb = ((len % lb_size) == 0);
329 
330 		slot++;
331 	}
332 	/* rest should be zero (ad_off > l_ad < max_l_ad - adlen) */
333 
334 	KASSERT(*cnt_inflen == inflen);
335 	KASSERT(*cnt_logblksrec == logblksrec);
336 
337 //	KASSERT(mutex_owned(&udf_node->ump->allocate_mutex));
338 }
339 #else
340 static void
341 udf_node_sanity_check(struct udf_node *udf_node,
342 		uint64_t *cnt_inflen, uint64_t *cnt_logblksrec) {
343 	struct file_entry    *fe;
344 	struct extfile_entry *efe;
345 	struct icb_tag *icbtag;
346 	uint64_t inflen, logblksrec;
347 	int dscr_size, lb_size;
348 
349 	lb_size = udf_rw32(udf_node->ump->logical_vol->lb_size);
350 
351 	fe  = udf_node->fe;
352 	efe = udf_node->efe;
353 	if (fe) {
354 		icbtag = &fe->icbtag;
355 		inflen = udf_rw64(fe->inf_len);
356 		dscr_size  = sizeof(struct file_entry) -1;
357 		logblksrec = udf_rw64(fe->logblks_rec);
358 	} else {
359 		icbtag = &efe->icbtag;
360 		inflen = udf_rw64(efe->inf_len);
361 		dscr_size  = sizeof(struct extfile_entry) -1;
362 		logblksrec = udf_rw64(efe->logblks_rec);
363 	}
364 	*cnt_logblksrec = logblksrec;
365 	*cnt_inflen     = inflen;
366 }
367 #endif
368 
369 /* --------------------------------------------------------------------- */
370 
371 void
372 udf_calc_freespace(struct udf_mount *ump, uint64_t *sizeblks, uint64_t *freeblks)
373 {
374 	struct logvol_int_desc *lvid;
375 	uint32_t *pos1, *pos2;
376 	int vpart, num_vpart;
377 
378 	lvid = ump->logvol_integrity;
379 	*freeblks = *sizeblks = 0;
380 
381 	/*
382 	 * Sequentials media report free space directly (CD/DVD/BD-R), for the
383 	 * other media we need the logical volume integrity.
384 	 *
385 	 * We sum all free space up here regardless of type.
386 	 */
387 
388 	KASSERT(lvid);
389 	num_vpart = udf_rw32(lvid->num_part);
390 
391 	if (ump->discinfo.mmc_cur & MMC_CAP_SEQUENTIAL) {
392 		/* use track info directly summing if there are 2 open */
393 		/* XXX assumption at most two tracks open */
394 		*freeblks = ump->data_track.free_blocks;
395 		if (ump->data_track.tracknr != ump->metadata_track.tracknr)
396 			*freeblks += ump->metadata_track.free_blocks;
397 		*sizeblks = ump->discinfo.last_possible_lba;
398 	} else {
399 		/* free and used space for mountpoint based on logvol integrity */
400 		for (vpart = 0; vpart < num_vpart; vpart++) {
401 			pos1 = &lvid->tables[0] + vpart;
402 			pos2 = &lvid->tables[0] + num_vpart + vpart;
403 			if (udf_rw32(*pos1) != (uint32_t) -1) {
404 				*freeblks += udf_rw32(*pos1);
405 				*sizeblks += udf_rw32(*pos2);
406 			}
407 		}
408 	}
409 	/* adjust for accounted uncommitted blocks */
410 	for (vpart = 0; vpart < num_vpart; vpart++)
411 		*freeblks -= ump->uncommitted_lbs[vpart];
412 
413 	if (*freeblks > UDF_DISC_SLACK) {
414 		*freeblks -= UDF_DISC_SLACK;
415 	} else {
416 		*freeblks = 0;
417 	}
418 }
419 
420 
421 static void
422 udf_calc_vpart_freespace(struct udf_mount *ump, uint16_t vpart_num, uint64_t *freeblks)
423 {
424 	struct logvol_int_desc *lvid;
425 	uint32_t *pos1;
426 
427 	lvid = ump->logvol_integrity;
428 	*freeblks = 0;
429 
430 	/*
431 	 * Sequentials media report free space directly (CD/DVD/BD-R), for the
432 	 * other media we need the logical volume integrity.
433 	 *
434 	 * We sum all free space up here regardless of type.
435 	 */
436 
437 	KASSERT(lvid);
438 	if (ump->discinfo.mmc_cur & MMC_CAP_SEQUENTIAL) {
439 		/* XXX assumption at most two tracks open */
440 		if (vpart_num == ump->data_part) {
441 			*freeblks = ump->data_track.free_blocks;
442 		} else {
443 			*freeblks = ump->metadata_track.free_blocks;
444 		}
445 	} else {
446 		/* free and used space for mountpoint based on logvol integrity */
447 		pos1 = &lvid->tables[0] + vpart_num;
448 		if (udf_rw32(*pos1) != (uint32_t) -1)
449 			*freeblks += udf_rw32(*pos1);
450 	}
451 
452 	/* adjust for accounted uncommitted blocks */
453 	if (*freeblks > ump->uncommitted_lbs[vpart_num]) {
454 		*freeblks -= ump->uncommitted_lbs[vpart_num];
455 	} else {
456 		*freeblks = 0;
457 	}
458 }
459 
460 /* --------------------------------------------------------------------- */
461 
462 int
463 udf_translate_vtop(struct udf_mount *ump, struct long_ad *icb_loc,
464 		   uint32_t *lb_numres, uint32_t *extres)
465 {
466 	struct part_desc       *pdesc;
467 	struct spare_map_entry *sme;
468 	struct long_ad s_icb_loc;
469 	uint64_t foffset, end_foffset;
470 	uint32_t lb_size, len;
471 	uint32_t lb_num, lb_rel, lb_packet;
472 	uint32_t udf_rw32_lbmap, ext_offset;
473 	uint16_t vpart;
474 	int rel, part, error, eof, slot, flags;
475 
476 	assert(ump && icb_loc && lb_numres);
477 
478 	vpart  = udf_rw16(icb_loc->loc.part_num);
479 	lb_num = udf_rw32(icb_loc->loc.lb_num);
480 	if (vpart > UDF_VTOP_RAWPART)
481 		return EINVAL;
482 
483 translate_again:
484 	part = ump->vtop[vpart];
485 	pdesc = ump->partitions[part];
486 
487 	switch (ump->vtop_tp[vpart]) {
488 	case UDF_VTOP_TYPE_RAW :
489 		/* 1:1 to the end of the device */
490 		*lb_numres = lb_num;
491 		*extres = INT_MAX;
492 		return 0;
493 	case UDF_VTOP_TYPE_PHYS :
494 		/* transform into its disc logical block */
495 		if (lb_num > udf_rw32(pdesc->part_len))
496 			return EINVAL;
497 		*lb_numres = lb_num + udf_rw32(pdesc->start_loc);
498 
499 		/* extent from here to the end of the partition */
500 		*extres = udf_rw32(pdesc->part_len) - lb_num;
501 		return 0;
502 	case UDF_VTOP_TYPE_VIRT :
503 		/* only maps one logical block, lookup in VAT */
504 		if (lb_num >= ump->vat_entries)		/* XXX > or >= ? */
505 			return EINVAL;
506 
507 		/* lookup in virtual allocation table file */
508 		mutex_enter(&ump->allocate_mutex);
509 		error = udf_vat_read(ump->vat_node,
510 				(uint8_t *) &udf_rw32_lbmap, 4,
511 				ump->vat_offset + lb_num * 4);
512 		mutex_exit(&ump->allocate_mutex);
513 
514 		if (error)
515 			return error;
516 
517 		lb_num = udf_rw32(udf_rw32_lbmap);
518 
519 		/* transform into its disc logical block */
520 		if (lb_num > udf_rw32(pdesc->part_len))
521 			return EINVAL;
522 		*lb_numres = lb_num + udf_rw32(pdesc->start_loc);
523 
524 		/* just one logical block */
525 		*extres = 1;
526 		return 0;
527 	case UDF_VTOP_TYPE_SPARABLE :
528 		/* check if the packet containing the lb_num is remapped */
529 		lb_packet = lb_num / ump->sparable_packet_size;
530 		lb_rel    = lb_num % ump->sparable_packet_size;
531 
532 		for (rel = 0; rel < udf_rw16(ump->sparing_table->rt_l); rel++) {
533 			sme = &ump->sparing_table->entries[rel];
534 			if (lb_packet == udf_rw32(sme->org)) {
535 				/* NOTE maps to absolute disc logical block! */
536 				*lb_numres = udf_rw32(sme->map) + lb_rel;
537 				*extres    = ump->sparable_packet_size - lb_rel;
538 				return 0;
539 			}
540 		}
541 
542 		/* transform into its disc logical block */
543 		if (lb_num > udf_rw32(pdesc->part_len))
544 			return EINVAL;
545 		*lb_numres = lb_num + udf_rw32(pdesc->start_loc);
546 
547 		/* rest of block */
548 		*extres = ump->sparable_packet_size - lb_rel;
549 		return 0;
550 	case UDF_VTOP_TYPE_META :
551 		/* we have to look into the file's allocation descriptors */
552 
553 		/* use metadatafile allocation mutex */
554 		lb_size = udf_rw32(ump->logical_vol->lb_size);
555 
556 		UDF_LOCK_NODE(ump->metadata_node, 0);
557 
558 		/* get first overlapping extent */
559 		foffset = 0;
560 		slot    = 0;
561 		for (;;) {
562 			udf_get_adslot(ump->metadata_node,
563 				slot, &s_icb_loc, &eof);
564 			DPRINTF(ADWLK, ("slot %d, eof = %d, flags = %d, "
565 				"len = %d, lb_num = %d, part = %d\n",
566 				slot, eof,
567 				UDF_EXT_FLAGS(udf_rw32(s_icb_loc.len)),
568 				UDF_EXT_LEN(udf_rw32(s_icb_loc.len)),
569 				udf_rw32(s_icb_loc.loc.lb_num),
570 				udf_rw16(s_icb_loc.loc.part_num)));
571 			if (eof) {
572 				DPRINTF(TRANSLATE,
573 					("Meta partition translation "
574 					 "failed: can't seek location\n"));
575 				UDF_UNLOCK_NODE(ump->metadata_node, 0);
576 				return EINVAL;
577 			}
578 			len   = udf_rw32(s_icb_loc.len);
579 			flags = UDF_EXT_FLAGS(len);
580 			len   = UDF_EXT_LEN(len);
581 
582 			if (flags == UDF_EXT_REDIRECT) {
583 				slot++;
584 				continue;
585 			}
586 
587 			end_foffset = foffset + len;
588 
589 			if (end_foffset > lb_num * lb_size)
590 				break;	/* found */
591 			foffset = end_foffset;
592 			slot++;
593 		}
594 		/* found overlapping slot */
595 		ext_offset = lb_num * lb_size - foffset;
596 
597 		/* process extent offset */
598 		lb_num   = udf_rw32(s_icb_loc.loc.lb_num);
599 		vpart    = udf_rw16(s_icb_loc.loc.part_num);
600 		lb_num  += (ext_offset + lb_size -1) / lb_size;
601 		ext_offset = 0;
602 
603 		UDF_UNLOCK_NODE(ump->metadata_node, 0);
604 		if (flags != UDF_EXT_ALLOCATED) {
605 			DPRINTF(TRANSLATE, ("Metadata partition translation "
606 					    "failed: not allocated\n"));
607 			return EINVAL;
608 		}
609 
610 		/*
611 		 * vpart and lb_num are updated, translate again since we
612 		 * might be mapped on sparable media
613 		 */
614 		goto translate_again;
615 	default:
616 		printf("UDF vtop translation scheme %d unimplemented yet\n",
617 			ump->vtop_tp[vpart]);
618 	}
619 
620 	return EINVAL;
621 }
622 
623 
624 /* XXX  provisional primitive braindead version */
625 /* TODO use ext_res */
626 void
627 udf_translate_vtop_list(struct udf_mount *ump, uint32_t sectors,
628 	uint16_t vpart_num, uint64_t *lmapping, uint64_t *pmapping)
629 {
630 	struct long_ad loc;
631 	uint32_t lb_numres, ext_res;
632 	int sector;
633 
634 	for (sector = 0; sector < sectors; sector++) {
635 		memset(&loc, 0, sizeof(struct long_ad));
636 		loc.loc.part_num = udf_rw16(vpart_num);
637 		loc.loc.lb_num   = udf_rw32(*lmapping);
638 		udf_translate_vtop(ump, &loc, &lb_numres, &ext_res);
639 		*pmapping = lb_numres;
640 		lmapping++; pmapping++;
641 	}
642 }
643 
644 
645 /* --------------------------------------------------------------------- */
646 
647 /*
648  * Translate an extent (in logical_blocks) into logical block numbers; used
649  * for read and write operations. DOESNT't check extents.
650  */
651 
652 int
653 udf_translate_file_extent(struct udf_node *udf_node,
654 		          uint32_t from, uint32_t num_lb,
655 			  uint64_t *map)
656 {
657 	struct udf_mount *ump;
658 	struct icb_tag *icbtag;
659 	struct long_ad t_ad, s_ad;
660 	uint64_t transsec;
661 	uint64_t foffset, end_foffset;
662 	uint32_t transsec32;
663 	uint32_t lb_size;
664 	uint32_t ext_offset;
665 	uint32_t lb_num, len;
666 	uint32_t overlap, translen;
667 	uint16_t vpart_num;
668 	int eof, error, flags;
669 	int slot, addr_type, icbflags;
670 
671 	if (!udf_node)
672 		return ENOENT;
673 
674 	KASSERT(num_lb > 0);
675 
676 	UDF_LOCK_NODE(udf_node, 0);
677 
678 	/* initialise derivative vars */
679 	ump = udf_node->ump;
680 	lb_size = udf_rw32(ump->logical_vol->lb_size);
681 
682 	if (udf_node->fe) {
683 		icbtag = &udf_node->fe->icbtag;
684 	} else {
685 		icbtag = &udf_node->efe->icbtag;
686 	}
687 	icbflags  = udf_rw16(icbtag->flags);
688 	addr_type = icbflags & UDF_ICB_TAG_FLAGS_ALLOC_MASK;
689 
690 	/* do the work */
691 	if (addr_type == UDF_ICB_INTERN_ALLOC) {
692 		*map = UDF_TRANS_INTERN;
693 		UDF_UNLOCK_NODE(udf_node, 0);
694 		return 0;
695 	}
696 
697 	/* find first overlapping extent */
698 	foffset = 0;
699 	slot    = 0;
700 	for (;;) {
701 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
702 		DPRINTF(ADWLK, ("slot %d, eof = %d, flags = %d, len = %d, "
703 			"lb_num = %d, part = %d\n", slot, eof,
704 			UDF_EXT_FLAGS(udf_rw32(s_ad.len)),
705 			UDF_EXT_LEN(udf_rw32(s_ad.len)),
706 			udf_rw32(s_ad.loc.lb_num),
707 			udf_rw16(s_ad.loc.part_num)));
708 		if (eof) {
709 			DPRINTF(TRANSLATE,
710 				("Translate file extent "
711 				 "failed: can't seek location\n"));
712 			UDF_UNLOCK_NODE(udf_node, 0);
713 			return EINVAL;
714 		}
715 		len    = udf_rw32(s_ad.len);
716 		flags  = UDF_EXT_FLAGS(len);
717 		len    = UDF_EXT_LEN(len);
718 		lb_num = udf_rw32(s_ad.loc.lb_num);
719 
720 		if (flags == UDF_EXT_REDIRECT) {
721 			slot++;
722 			continue;
723 		}
724 
725 		end_foffset = foffset + len;
726 
727 		if (end_foffset > from * lb_size)
728 			break;	/* found */
729 		foffset = end_foffset;
730 		slot++;
731 	}
732 	/* found overlapping slot */
733 	ext_offset = from * lb_size - foffset;
734 
735 	for (;;) {
736 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
737 		DPRINTF(ADWLK, ("slot %d, eof = %d, flags = %d, len = %d, "
738 			"lb_num = %d, part = %d\n", slot, eof,
739 			UDF_EXT_FLAGS(udf_rw32(s_ad.len)),
740 			UDF_EXT_LEN(udf_rw32(s_ad.len)),
741 			udf_rw32(s_ad.loc.lb_num),
742 			udf_rw16(s_ad.loc.part_num)));
743 		if (eof) {
744 			DPRINTF(TRANSLATE,
745 				("Translate file extent "
746 				 "failed: past eof\n"));
747 			UDF_UNLOCK_NODE(udf_node, 0);
748 			return EINVAL;
749 		}
750 
751 		len    = udf_rw32(s_ad.len);
752 		flags  = UDF_EXT_FLAGS(len);
753 		len    = UDF_EXT_LEN(len);
754 
755 		lb_num    = udf_rw32(s_ad.loc.lb_num);
756 		vpart_num = udf_rw16(s_ad.loc.part_num);
757 
758 		end_foffset = foffset + len;
759 
760 		/* process extent, don't forget to advance on ext_offset! */
761 		lb_num  += (ext_offset + lb_size -1) / lb_size;
762 		overlap  = (len - ext_offset + lb_size -1) / lb_size;
763 		ext_offset = 0;
764 
765 		/*
766 		 * note that the while(){} is nessisary for the extent that
767 		 * the udf_translate_vtop() returns doens't have to span the
768 		 * whole extent.
769 		 */
770 
771 		overlap = MIN(overlap, num_lb);
772 		while (overlap && (flags != UDF_EXT_REDIRECT)) {
773 			switch (flags) {
774 			case UDF_EXT_FREE :
775 			case UDF_EXT_ALLOCATED_BUT_NOT_USED :
776 				transsec = UDF_TRANS_ZERO;
777 				translen = overlap;
778 				while (overlap && num_lb && translen) {
779 					*map++ = transsec;
780 					lb_num++;
781 					overlap--; num_lb--; translen--;
782 				}
783 				break;
784 			case UDF_EXT_ALLOCATED :
785 				t_ad.loc.lb_num   = udf_rw32(lb_num);
786 				t_ad.loc.part_num = udf_rw16(vpart_num);
787 				error = udf_translate_vtop(ump,
788 						&t_ad, &transsec32, &translen);
789 				transsec = transsec32;
790 				if (error) {
791 					UDF_UNLOCK_NODE(udf_node, 0);
792 					return error;
793 				}
794 				while (overlap && num_lb && translen) {
795 					*map++ = transsec;
796 					lb_num++; transsec++;
797 					overlap--; num_lb--; translen--;
798 				}
799 				break;
800 			default:
801 				DPRINTF(TRANSLATE,
802 					("Translate file extent "
803 					 "failed: bad flags %x\n", flags));
804 				UDF_UNLOCK_NODE(udf_node, 0);
805 				return EINVAL;
806 			}
807 		}
808 		if (num_lb == 0)
809 			break;
810 
811 		if (flags != UDF_EXT_REDIRECT)
812 			foffset = end_foffset;
813 		slot++;
814 	}
815 	UDF_UNLOCK_NODE(udf_node, 0);
816 
817 	return 0;
818 }
819 
820 /* --------------------------------------------------------------------- */
821 
822 static int
823 udf_search_free_vatloc(struct udf_mount *ump, uint32_t *lbnumres)
824 {
825 	uint32_t lb_size, lb_num, lb_map, udf_rw32_lbmap;
826 	uint8_t *blob;
827 	int entry, chunk, found, error;
828 
829 	KASSERT(ump);
830 	KASSERT(ump->logical_vol);
831 
832 	lb_size = udf_rw32(ump->logical_vol->lb_size);
833 	blob = malloc(lb_size, M_UDFTEMP, M_WAITOK);
834 
835 	/* TODO static allocation of search chunk */
836 
837 	lb_num = MIN(ump->vat_entries, ump->vat_last_free_lb);
838 	found  = 0;
839 	error  = 0;
840 	entry  = 0;
841 	do {
842 		chunk = MIN(lb_size, (ump->vat_entries - lb_num) * 4);
843 		if (chunk <= 0)
844 			break;
845 		/* load in chunk */
846 		error = udf_vat_read(ump->vat_node, blob, chunk,
847 				ump->vat_offset + lb_num * 4);
848 
849 		if (error)
850 			break;
851 
852 		/* search this chunk */
853 		for (entry=0; entry < chunk /4; entry++, lb_num++) {
854 			udf_rw32_lbmap = *((uint32_t *) (blob + entry * 4));
855 			lb_map = udf_rw32(udf_rw32_lbmap);
856 			if (lb_map == 0xffffffff) {
857 				found = 1;
858 				break;
859 			}
860 		}
861 	} while (!found);
862 	if (error) {
863 		printf("udf_search_free_vatloc: error reading in vat chunk "
864 			"(lb %d, size %d)\n", lb_num, chunk);
865 	}
866 
867 	if (!found) {
868 		/* extend VAT */
869 		DPRINTF(WRITE, ("udf_search_free_vatloc: extending\n"));
870 		lb_num = ump->vat_entries;
871 		ump->vat_entries++;
872 	}
873 
874 	/* mark entry with initialiser just in case */
875 	lb_map = udf_rw32(0xfffffffe);
876 	udf_vat_write(ump->vat_node, (uint8_t *) &lb_map, 4,
877 		ump->vat_offset + lb_num *4);
878 	ump->vat_last_free_lb = lb_num;
879 
880 	free(blob, M_UDFTEMP);
881 	*lbnumres = lb_num;
882 	return 0;
883 }
884 
885 
886 static void
887 udf_bitmap_allocate(struct udf_bitmap *bitmap, int ismetadata,
888 	uint32_t *num_lb, uint64_t *lmappos)
889 {
890 	uint32_t offset, lb_num, bit;
891 	int32_t  diff;
892 	uint8_t *bpos;
893 	int pass;
894 
895 	if (!ismetadata) {
896 		/* heuristic to keep the two pointers not too close */
897 		diff = bitmap->data_pos - bitmap->metadata_pos;
898 		if ((diff >= 0) && (diff < 1024))
899 			bitmap->data_pos = bitmap->metadata_pos + 1024;
900 	}
901 	offset = ismetadata ? bitmap->metadata_pos : bitmap->data_pos;
902 	offset &= ~7;
903 	for (pass = 0; pass < 2; pass++) {
904 		if (offset >= bitmap->max_offset)
905 			offset = 0;
906 
907 		while (offset < bitmap->max_offset) {
908 			if (*num_lb == 0)
909 				break;
910 
911 			/* use first bit not set */
912 			bpos  = bitmap->bits + offset/8;
913 			bit = ffs(*bpos);	/* returns 0 or 1..8 */
914 			if (bit == 0) {
915 				offset += 8;
916 				continue;
917 			}
918 
919 			/* check for ffs overshoot */
920 			if (offset + bit-1 >= bitmap->max_offset) {
921 				offset = bitmap->max_offset;
922 				break;
923 			}
924 
925 			DPRINTF(PARANOIA, ("XXX : allocate %d, %p, bit %d\n",
926 				offset + bit -1, bpos, bit-1));
927 			*bpos &= ~(1 << (bit-1));
928 			lb_num = offset + bit-1;
929 			*lmappos++ = lb_num;
930 			*num_lb = *num_lb - 1;
931 			// offset = (offset & ~7);
932 		}
933 	}
934 
935 	if (ismetadata) {
936 		bitmap->metadata_pos = offset;
937 	} else {
938 		bitmap->data_pos = offset;
939 	}
940 }
941 
942 
943 static void
944 udf_bitmap_free(struct udf_bitmap *bitmap, uint32_t lb_num, uint32_t num_lb)
945 {
946 	uint32_t offset;
947 	uint32_t bit, bitval;
948 	uint8_t *bpos;
949 
950 	offset = lb_num;
951 
952 	/* starter bits */
953 	bpos = bitmap->bits + offset/8;
954 	bit = offset % 8;
955 	while ((bit != 0) && (num_lb > 0)) {
956 		bitval = (1 << bit);
957 		KASSERT((*bpos & bitval) == 0);
958 		DPRINTF(PARANOIA, ("XXX : free %d, %p, %d\n",
959 			offset, bpos, bit));
960 		*bpos |= bitval;
961 		offset++; num_lb--;
962 		bit = (bit + 1) % 8;
963 	}
964 	if (num_lb == 0)
965 		return;
966 
967 	/* whole bytes */
968 	KASSERT(bit == 0);
969 	bpos = bitmap->bits + offset / 8;
970 	while (num_lb >= 8) {
971 		KASSERT((*bpos == 0));
972 		DPRINTF(PARANOIA, ("XXX : free %d + 8, %p\n", offset, bpos));
973 		*bpos = 255;
974 		offset += 8; num_lb -= 8;
975 		bpos++;
976 	}
977 
978 	/* stop bits */
979 	KASSERT(num_lb < 8);
980 	bit = 0;
981 	while (num_lb > 0) {
982 		bitval = (1 << bit);
983 		KASSERT((*bpos & bitval) == 0);
984 		DPRINTF(PARANOIA, ("XXX : free %d, %p, %d\n",
985 			offset, bpos, bit));
986 		*bpos |= bitval;
987 		offset++; num_lb--;
988 		bit = (bit + 1) % 8;
989 	}
990 }
991 
992 
993 static uint32_t
994 udf_bitmap_check_trunc_free(struct udf_bitmap *bitmap, uint32_t to_trunc)
995 {
996 	uint32_t seq_free, offset;
997 	uint8_t *bpos;
998 	uint8_t  bit, bitval;
999 
1000 	DPRINTF(RESERVE, ("\ttrying to trunc %d bits from bitmap\n", to_trunc));
1001 	offset = bitmap->max_offset - to_trunc;
1002 
1003 	/* starter bits (if any) */
1004 	bpos = bitmap->bits + offset/8;
1005 	bit = offset % 8;
1006 	seq_free = 0;
1007 	while (to_trunc > 0) {
1008 		seq_free++;
1009 		bitval = (1 << bit);
1010 		if (!(*bpos & bitval))
1011 			seq_free = 0;
1012 		offset++; to_trunc--;
1013 		bit++;
1014 		if (bit == 8) {
1015 			bpos++;
1016 			bit = 0;
1017 		}
1018 	}
1019 
1020 	DPRINTF(RESERVE, ("\tfound %d sequential free bits in bitmap\n", seq_free));
1021 	return seq_free;
1022 }
1023 
1024 /* --------------------------------------------------------------------- */
1025 
1026 /*
1027  * We check for overall disc space with a margin to prevent critical
1028  * conditions.  If disc space is low we try to force a sync() to improve our
1029  * estimates.  When confronted with meta-data partition size shortage we know
1030  * we have to check if it can be extended and we need to extend it when
1031  * needed.
1032  *
1033  * A 2nd strategy we could use when disc space is getting low on a disc
1034  * formatted with a meta-data partition is to see if there are sparse areas in
1035  * the meta-data partition and free blocks there for extra data.
1036  */
1037 
1038 void
1039 udf_do_reserve_space(struct udf_mount *ump, struct udf_node *udf_node,
1040 	uint16_t vpart_num, uint32_t num_lb)
1041 {
1042 	ump->uncommitted_lbs[vpart_num] += num_lb;
1043 	if (udf_node)
1044 		udf_node->uncommitted_lbs += num_lb;
1045 }
1046 
1047 
1048 void
1049 udf_do_unreserve_space(struct udf_mount *ump, struct udf_node *udf_node,
1050 	uint16_t vpart_num, uint32_t num_lb)
1051 {
1052 	ump->uncommitted_lbs[vpart_num] -= num_lb;
1053 	if (ump->uncommitted_lbs[vpart_num] < 0) {
1054 		DPRINTF(RESERVE, ("UDF: underflow on partition reservation, "
1055 			"part %d: %d\n", vpart_num,
1056 			ump->uncommitted_lbs[vpart_num]));
1057 		ump->uncommitted_lbs[vpart_num] = 0;
1058 	}
1059 	if (udf_node) {
1060 		udf_node->uncommitted_lbs -= num_lb;
1061 		if (udf_node->uncommitted_lbs < 0) {
1062 			DPRINTF(RESERVE, ("UDF: underflow of node "
1063 				"reservation : %d\n",
1064 				udf_node->uncommitted_lbs));
1065 			udf_node->uncommitted_lbs = 0;
1066 		}
1067 	}
1068 }
1069 
1070 
1071 int
1072 udf_reserve_space(struct udf_mount *ump, struct udf_node *udf_node,
1073 	int udf_c_type, uint16_t vpart_num, uint32_t num_lb, int can_fail)
1074 {
1075 	uint64_t freeblks;
1076 	uint64_t slack;
1077 	int i, error;
1078 
1079 	slack = 0;
1080 	if (can_fail)
1081 		slack = UDF_DISC_SLACK;
1082 
1083 	error = 0;
1084 	mutex_enter(&ump->allocate_mutex);
1085 
1086 	/* check if there is enough space available */
1087 	for (i = 0; i < 3; i++) {	/* XXX arbitrary number */
1088 		udf_calc_vpart_freespace(ump, vpart_num, &freeblks);
1089 		if (num_lb + slack < freeblks)
1090 			break;
1091 		/* issue SYNC */
1092 		DPRINTF(RESERVE, ("udf_reserve_space: issuing sync\n"));
1093 		mutex_exit(&ump->allocate_mutex);
1094 		udf_do_sync(ump, FSCRED, 0);
1095 		mutex_enter(&mntvnode_lock);
1096 		/* 1/8 second wait */
1097 		cv_timedwait(&ump->dirtynodes_cv, &mntvnode_lock,
1098 			hz/8);
1099 		mutex_exit(&mntvnode_lock);
1100 		mutex_enter(&ump->allocate_mutex);
1101 	}
1102 
1103 	/* check if there is enough space available now */
1104 	udf_calc_vpart_freespace(ump, vpart_num, &freeblks);
1105 	if (num_lb + slack >= freeblks) {
1106 		DPRINTF(RESERVE, ("udf_reserve_space: try to redistribute "
1107 				  "partition space\n"));
1108 		DPRINTF(RESERVE, ("\tvpart %d, type %d is full\n",
1109 				vpart_num, ump->vtop_alloc[vpart_num]));
1110 		/* Try to redistribute space if possible */
1111 		udf_collect_free_space_for_vpart(ump, vpart_num, num_lb + slack);
1112 	}
1113 
1114 	/* check if there is enough space available now */
1115 	udf_calc_vpart_freespace(ump, vpart_num, &freeblks);
1116 	if (num_lb + slack <= freeblks) {
1117 		udf_do_reserve_space(ump, udf_node, vpart_num, num_lb);
1118 	} else {
1119 		DPRINTF(RESERVE, ("udf_reserve_space: out of disc space\n"));
1120 		error = ENOSPC;
1121 	}
1122 
1123 	mutex_exit(&ump->allocate_mutex);
1124 	return error;
1125 }
1126 
1127 
1128 void
1129 udf_cleanup_reservation(struct udf_node *udf_node)
1130 {
1131 	struct udf_mount *ump = udf_node->ump;
1132 	int vpart_num;
1133 
1134 	mutex_enter(&ump->allocate_mutex);
1135 
1136 	/* compensate for overlapping blocks */
1137 	DPRINTF(RESERVE, ("UDF: overlapped %d blocks in count\n", udf_node->uncommitted_lbs));
1138 
1139 	vpart_num = udf_get_record_vpart(ump, udf_get_c_type(udf_node));
1140 	udf_do_unreserve_space(ump, udf_node, vpart_num, udf_node->uncommitted_lbs);
1141 
1142 	DPRINTF(RESERVE, ("\ttotal now %d\n", ump->uncommitted_lbs[vpart_num]));
1143 
1144 	/* sanity */
1145 	if (ump->uncommitted_lbs[vpart_num] < 0)
1146 		ump->uncommitted_lbs[vpart_num] = 0;
1147 
1148 	mutex_exit(&ump->allocate_mutex);
1149 }
1150 
1151 /* --------------------------------------------------------------------- */
1152 
1153 /*
1154  * Allocate an extent of given length on given virt. partition. It doesn't
1155  * have to be one stretch.
1156  */
1157 
1158 int
1159 udf_allocate_space(struct udf_mount *ump, struct udf_node *udf_node,
1160 	int udf_c_type, uint16_t vpart_num, uint32_t num_lb, uint64_t *lmapping)
1161 {
1162 	struct mmc_trackinfo *alloc_track, *other_track;
1163 	struct udf_bitmap *bitmap;
1164 	struct part_desc *pdesc;
1165 	struct logvol_int_desc *lvid;
1166 	uint64_t *lmappos;
1167 	uint32_t ptov, lb_num, *freepos, free_lbs;
1168 	int lb_size, alloc_num_lb;
1169 	int alloc_type, error;
1170 	int is_node;
1171 
1172 	DPRINTF(CALL, ("udf_allocate_space(ctype %d, vpart %d, num_lb %d\n",
1173 		udf_c_type, vpart_num, num_lb));
1174 	mutex_enter(&ump->allocate_mutex);
1175 
1176 	lb_size = udf_rw32(ump->logical_vol->lb_size);
1177 	KASSERT(lb_size == ump->discinfo.sector_size);
1178 
1179 	alloc_type =  ump->vtop_alloc[vpart_num];
1180 	is_node    = (udf_c_type == UDF_C_NODE);
1181 
1182 	lmappos = lmapping;
1183 	error = 0;
1184 	switch (alloc_type) {
1185 	case UDF_ALLOC_VAT :
1186 		/* search empty slot in VAT file */
1187 		KASSERT(num_lb == 1);
1188 		error = udf_search_free_vatloc(ump, &lb_num);
1189 		if (!error) {
1190 			*lmappos = lb_num;
1191 
1192 			/* reserve on the backing sequential partition since
1193 			 * that partition is credited back later */
1194 			udf_do_reserve_space(ump, udf_node,
1195 				ump->vtop[vpart_num], num_lb);
1196 		}
1197 		break;
1198 	case UDF_ALLOC_SEQUENTIAL :
1199 		/* sequential allocation on recordable media */
1200 		/* get partition backing up this vpart_num_num */
1201 		pdesc = ump->partitions[ump->vtop[vpart_num]];
1202 
1203 		/* calculate offset from physical base partition */
1204 		ptov  = udf_rw32(pdesc->start_loc);
1205 
1206 		/* get our track descriptors */
1207 		if (vpart_num == ump->node_part) {
1208 			alloc_track = &ump->metadata_track;
1209 			other_track = &ump->data_track;
1210 		} else {
1211 			alloc_track = &ump->data_track;
1212 			other_track = &ump->metadata_track;
1213 		}
1214 
1215 		/* allocate */
1216 		for (lb_num = 0; lb_num < num_lb; lb_num++) {
1217 			*lmappos++ = alloc_track->next_writable - ptov;
1218 			alloc_track->next_writable++;
1219 			alloc_track->free_blocks--;
1220 		}
1221 
1222 		/* keep other track up-to-date */
1223 		if (alloc_track->tracknr == other_track->tracknr)
1224 			memcpy(other_track, alloc_track,
1225 				sizeof(struct mmc_trackinfo));
1226 		break;
1227 	case UDF_ALLOC_SPACEMAP :
1228 		/* try to allocate on unallocated bits */
1229 		alloc_num_lb = num_lb;
1230 		bitmap = &ump->part_unalloc_bits[vpart_num];
1231 		udf_bitmap_allocate(bitmap, is_node, &alloc_num_lb, lmappos);
1232 		ump->lvclose |= UDF_WRITE_PART_BITMAPS;
1233 
1234 		/* have we allocated all? */
1235 		if (alloc_num_lb) {
1236 			/* TODO convert freed to unalloc and try again */
1237 			/* free allocated piece for now */
1238 			lmappos = lmapping;
1239 			for (lb_num=0; lb_num < num_lb-alloc_num_lb; lb_num++) {
1240 				udf_bitmap_free(bitmap, *lmappos++, 1);
1241 			}
1242 			error = ENOSPC;
1243 		}
1244 		if (!error) {
1245 			/* adjust freecount */
1246 			lvid = ump->logvol_integrity;
1247 			freepos = &lvid->tables[0] + vpart_num;
1248 			free_lbs = udf_rw32(*freepos);
1249 			*freepos = udf_rw32(free_lbs - num_lb);
1250 		}
1251 		break;
1252 	case UDF_ALLOC_METABITMAP :		/* UDF 2.50, 2.60 BluRay-RE */
1253 		/* allocate on metadata unallocated bits */
1254 		alloc_num_lb = num_lb;
1255 		bitmap = &ump->metadata_unalloc_bits;
1256 		udf_bitmap_allocate(bitmap, is_node, &alloc_num_lb, lmappos);
1257 		ump->lvclose |= UDF_WRITE_PART_BITMAPS;
1258 
1259 		/* have we allocated all? */
1260 		if (alloc_num_lb) {
1261 			/* YIKES! TODO we need to extend the metadata partition */
1262 			/* free allocated piece for now */
1263 			lmappos = lmapping;
1264 			for (lb_num=0; lb_num < num_lb-alloc_num_lb; lb_num++) {
1265 				udf_bitmap_free(bitmap, *lmappos++, 1);
1266 			}
1267 			error = ENOSPC;
1268 		}
1269 		if (!error) {
1270 			/* adjust freecount */
1271 			lvid = ump->logvol_integrity;
1272 			freepos = &lvid->tables[0] + vpart_num;
1273 			free_lbs = udf_rw32(*freepos);
1274 			*freepos = udf_rw32(free_lbs - num_lb);
1275 		}
1276 		break;
1277 	case UDF_ALLOC_METASEQUENTIAL :		/* UDF 2.60       BluRay-R  */
1278 	case UDF_ALLOC_RELAXEDSEQUENTIAL :	/* UDF 2.50/~meta BluRay-R  */
1279 		printf("ALERT: udf_allocate_space : allocation %d "
1280 				"not implemented yet!\n", alloc_type);
1281 		/* TODO implement, doesn't have to be contiguous */
1282 		error = ENOSPC;
1283 		break;
1284 	}
1285 
1286 	if (!error) {
1287 		/* credit our partition since we have committed the space */
1288 		udf_do_unreserve_space(ump, udf_node, vpart_num, num_lb);
1289 	}
1290 
1291 #ifdef DEBUG
1292 	if (udf_verbose & UDF_DEBUG_ALLOC) {
1293 		lmappos = lmapping;
1294 		printf("udf_allocate_space, allocated logical lba :\n");
1295 		for (lb_num = 0; lb_num < num_lb; lb_num++) {
1296 			printf("%s %"PRIu64, (lb_num > 0)?",":"",
1297 				*lmappos++);
1298 		}
1299 		printf("\n");
1300 	}
1301 #endif
1302 	mutex_exit(&ump->allocate_mutex);
1303 
1304 	return error;
1305 }
1306 
1307 /* --------------------------------------------------------------------- */
1308 
1309 void
1310 udf_free_allocated_space(struct udf_mount *ump, uint32_t lb_num,
1311 	uint16_t vpart_num, uint32_t num_lb)
1312 {
1313 	struct udf_bitmap *bitmap;
1314 	struct part_desc *pdesc;
1315 	struct logvol_int_desc *lvid;
1316 	uint32_t ptov, lb_map, udf_rw32_lbmap;
1317 	uint32_t *freepos, free_lbs;
1318 	int phys_part;
1319 	int error;
1320 
1321 	DPRINTF(ALLOC, ("udf_free_allocated_space: freeing virt lbnum %d "
1322 			  "part %d + %d sect\n", lb_num, vpart_num, num_lb));
1323 
1324 	/* no use freeing zero length */
1325 	if (num_lb == 0)
1326 		return;
1327 
1328 	mutex_enter(&ump->allocate_mutex);
1329 
1330 	/* get partition backing up this vpart_num */
1331 	pdesc = ump->partitions[ump->vtop[vpart_num]];
1332 
1333 	switch (ump->vtop_tp[vpart_num]) {
1334 	case UDF_VTOP_TYPE_PHYS :
1335 	case UDF_VTOP_TYPE_SPARABLE :
1336 		/* free space to freed or unallocated space bitmap */
1337 		ptov      = udf_rw32(pdesc->start_loc);
1338 		phys_part = ump->vtop[vpart_num];
1339 
1340 		/* first try freed space bitmap */
1341 		bitmap    = &ump->part_freed_bits[phys_part];
1342 
1343 		/* if not defined, use unallocated bitmap */
1344 		if (bitmap->bits == NULL)
1345 			bitmap = &ump->part_unalloc_bits[phys_part];
1346 
1347 		/* if no bitmaps are defined, bail out; XXX OK? */
1348 		if (bitmap->bits == NULL)
1349 			break;
1350 
1351 		/* free bits if its defined */
1352 		KASSERT(bitmap->bits);
1353 		ump->lvclose |= UDF_WRITE_PART_BITMAPS;
1354 		udf_bitmap_free(bitmap, lb_num, num_lb);
1355 
1356 		/* adjust freecount */
1357 		lvid = ump->logvol_integrity;
1358 		freepos = &lvid->tables[0] + vpart_num;
1359 		free_lbs = udf_rw32(*freepos);
1360 		*freepos = udf_rw32(free_lbs + num_lb);
1361 		break;
1362 	case UDF_VTOP_TYPE_VIRT :
1363 		/* free this VAT entry */
1364 		KASSERT(num_lb == 1);
1365 
1366 		lb_map = 0xffffffff;
1367 		udf_rw32_lbmap = udf_rw32(lb_map);
1368 		error = udf_vat_write(ump->vat_node,
1369 			(uint8_t *) &udf_rw32_lbmap, 4,
1370 			ump->vat_offset + lb_num * 4);
1371 		KASSERT(error == 0);
1372 		ump->vat_last_free_lb = MIN(ump->vat_last_free_lb, lb_num);
1373 		break;
1374 	case UDF_VTOP_TYPE_META :
1375 		/* free space in the metadata bitmap */
1376 		bitmap = &ump->metadata_unalloc_bits;
1377 		KASSERT(bitmap->bits);
1378 
1379 		ump->lvclose |= UDF_WRITE_PART_BITMAPS;
1380 		udf_bitmap_free(bitmap, lb_num, num_lb);
1381 
1382 		/* adjust freecount */
1383 		lvid = ump->logvol_integrity;
1384 		freepos = &lvid->tables[0] + vpart_num;
1385 		free_lbs = udf_rw32(*freepos);
1386 		*freepos = udf_rw32(free_lbs + num_lb);
1387 		break;
1388 	default:
1389 		printf("ALERT: udf_free_allocated_space : allocation %d "
1390 			"not implemented yet!\n", ump->vtop_tp[vpart_num]);
1391 		break;
1392 	}
1393 
1394 	mutex_exit(&ump->allocate_mutex);
1395 }
1396 
1397 /* --------------------------------------------------------------------- */
1398 
1399 /*
1400  * Special function to synchronise the metadatamirror file when they change on
1401  * resizing. When the metadatafile is actually duplicated, this action is a
1402  * no-op since they describe different extents on the disc.
1403  */
1404 
1405 void
1406 udf_synchronise_metadatamirror_node(struct udf_mount *ump)
1407 {
1408 	struct udf_node *meta_node, *metamirror_node;
1409 	struct long_ad s_ad;
1410 	uint32_t len, flags;
1411 	int slot, cpy_slot;
1412 	int error, eof;
1413 
1414 	if (ump->metadata_flags & METADATA_DUPLICATED)
1415 		return;
1416 
1417 	meta_node       = ump->metadata_node;
1418 	metamirror_node = ump->metadatamirror_node;
1419 
1420 	/* 1) wipe mirror node */
1421 	udf_wipe_adslots(metamirror_node);
1422 
1423 	/* 2) copy all node descriptors from the meta_node */
1424 	slot     = 0;
1425 	cpy_slot = 0;
1426 	for (;;) {
1427 		udf_get_adslot(meta_node, slot, &s_ad, &eof);
1428 		if (eof)
1429 			break;
1430 		len   = udf_rw32(s_ad.len);
1431 		flags = UDF_EXT_FLAGS(len);
1432 		len   = UDF_EXT_LEN(len);
1433 
1434 		if (flags == UDF_EXT_REDIRECT) {
1435 			slot++;
1436 			continue;
1437 		}
1438 
1439 		error = udf_append_adslot(metamirror_node, &cpy_slot, &s_ad);
1440 		if (error) {
1441 			/* WTF, this shouldn't happen, what to do now? */
1442 			panic("udf_synchronise_metadatamirror_node failed!");
1443 		}
1444 		cpy_slot++;
1445 		slot++;
1446 	}
1447 
1448 	/* 3) adjust metamirror_node size */
1449 	if (meta_node->fe) {
1450 		KASSERT(metamirror_node->fe);
1451 		metamirror_node->fe->inf_len = meta_node->fe->inf_len;
1452 	} else {
1453 		KASSERT(meta_node->efe);
1454 		KASSERT(metamirror_node->efe);
1455 		metamirror_node->efe->inf_len  = meta_node->efe->inf_len;
1456 		metamirror_node->efe->obj_size = meta_node->efe->obj_size;
1457 	}
1458 
1459 	/* for sanity */
1460 	udf_count_alloc_exts(metamirror_node);
1461 }
1462 
1463 /* --------------------------------------------------------------------- */
1464 
1465 /*
1466  * When faced with an out of space but there is still space available on other
1467  * partitions, try to redistribute the space. This is only defined for media
1468  * using Metadata partitions.
1469  *
1470  * There are two formats to deal with. Either its a `normal' metadata
1471  * partition and we can move blocks between a metadata bitmap and its
1472  * companion data spacemap OR its a UDF 2.60 formatted BluRay-R disc with POW
1473  * and a metadata partition.
1474  */
1475 
1476 /* implementation limit: ump->datapart is the companion partition */
1477 static uint32_t
1478 udf_trunc_metadatapart(struct udf_mount *ump, uint32_t num_lb)
1479 {
1480 	struct udf_node *bitmap_node;
1481 	struct udf_bitmap *bitmap;
1482 	struct space_bitmap_desc *sbd, *new_sbd;
1483 	struct logvol_int_desc *lvid;
1484 	uint64_t inf_len;
1485 	uint64_t meta_free_lbs, data_free_lbs, to_trunc;
1486 	uint32_t *freepos, *sizepos;
1487 	uint32_t unit, lb_size;
1488 	uint16_t meta_vpart_num, data_vpart_num, num_vpart;
1489 	int err;
1490 
1491 	unit = ump->metadata_alloc_unit_size;
1492 	lb_size = udf_rw32(ump->logical_vol->lb_size);
1493 	lvid = ump->logvol_integrity;
1494 
1495 	/* XXX
1496 	 *
1497 	 * the following checks will fail for BD-R UDF 2.60! but they are
1498 	 * read-only for now anyway! Its even doubtfull if it is to be allowed
1499 	 * for these discs.
1500 	 */
1501 
1502 	/* lookup vpart for metadata partition */
1503 	meta_vpart_num = ump->node_part;
1504 	KASSERT(ump->vtop_alloc[meta_vpart_num] == UDF_ALLOC_METABITMAP);
1505 
1506 	/* lookup vpart for data partition */
1507 	data_vpart_num = ump->data_part;
1508 	KASSERT(ump->vtop_alloc[data_vpart_num] == UDF_ALLOC_SPACEMAP);
1509 
1510 	udf_calc_vpart_freespace(ump, data_vpart_num, &data_free_lbs);
1511 	udf_calc_vpart_freespace(ump, meta_vpart_num, &meta_free_lbs);
1512 
1513 	DPRINTF(RESERVE, ("\tfree space on data partition     %"PRIu64" blks\n", data_free_lbs));
1514 	DPRINTF(RESERVE, ("\tfree space on metadata partition %"PRIu64" blks\n", meta_free_lbs));
1515 
1516 	/* give away some of the free meta space, in unit block sizes */
1517 	to_trunc = meta_free_lbs/4;			/* give out a quarter */
1518 	to_trunc = MAX(to_trunc, num_lb);
1519 	to_trunc = unit * ((to_trunc + unit-1) / unit);	/* round up */
1520 
1521 	/* scale down if needed and bail out when out of space */
1522 	if (to_trunc >= meta_free_lbs)
1523 		return num_lb;
1524 
1525 	/* check extent of bits marked free at the end of the map */
1526 	bitmap = &ump->metadata_unalloc_bits;
1527 	to_trunc = udf_bitmap_check_trunc_free(bitmap, to_trunc);
1528 	to_trunc = unit * (to_trunc / unit);		/* round down again */
1529 	if (to_trunc == 0)
1530 		return num_lb;
1531 
1532 	DPRINTF(RESERVE, ("\ttruncating %"PRIu64" lbs from the metadata bitmap\n",
1533 		to_trunc));
1534 
1535 	/* get length of the metadata bitmap node file */
1536 	bitmap_node = ump->metadatabitmap_node;
1537 	if (bitmap_node->fe) {
1538 		inf_len = udf_rw64(bitmap_node->fe->inf_len);
1539 	} else {
1540 		KASSERT(bitmap_node->efe);
1541 		inf_len = udf_rw64(bitmap_node->efe->inf_len);
1542 	}
1543 	inf_len -= to_trunc/8;
1544 
1545 	/* as per [UDF 2.60/2.2.13.6] : */
1546 	/* 1) update the SBD in the metadata bitmap file */
1547 	sbd = (struct space_bitmap_desc *) bitmap->blob;
1548 	sbd->num_bits  = udf_rw32(udf_rw32(sbd->num_bits)  - to_trunc);
1549 	sbd->num_bytes = udf_rw32(udf_rw32(sbd->num_bytes) - to_trunc/8);
1550 	bitmap->max_offset = udf_rw32(sbd->num_bits);
1551 
1552 	num_vpart = udf_rw32(lvid->num_part);
1553 	freepos = &lvid->tables[0] + meta_vpart_num;
1554 	sizepos = &lvid->tables[0] + num_vpart + meta_vpart_num;
1555 	*freepos = udf_rw32(*freepos) - to_trunc;
1556 	*sizepos = udf_rw32(*sizepos) - to_trunc;
1557 
1558 	/* realloc bitmap for better memory usage */
1559 	new_sbd = realloc(sbd, inf_len, M_UDFVOLD,
1560 		M_CANFAIL | M_WAITOK);
1561 	if (new_sbd) {
1562 		/* update pointers */
1563 		ump->metadata_unalloc_dscr = new_sbd;
1564 		bitmap->blob = (uint8_t *) new_sbd;
1565 	}
1566 	ump->lvclose |= UDF_WRITE_PART_BITMAPS;
1567 
1568 	/*
1569 	 * The truncated space is secured now and can't be allocated anymore.
1570 	 * Release the allocate mutex so we can shrink the nodes the normal
1571 	 * way.
1572 	 */
1573 	mutex_exit(&ump->allocate_mutex);
1574 
1575 	/* 2) trunc the metadata bitmap information file, freeing blocks */
1576 	err = udf_shrink_node(bitmap_node, inf_len);
1577 	KASSERT(err == 0);
1578 
1579 	/* 3) trunc the metadata file and mirror file, freeing blocks */
1580 	inf_len = (uint64_t) udf_rw32(sbd->num_bits) * lb_size;	/* [4/14.12.4] */
1581 	err = udf_shrink_node(ump->metadata_node, inf_len);
1582 	KASSERT(err == 0);
1583 	if (ump->metadatamirror_node) {
1584 		if (ump->metadata_flags & METADATA_DUPLICATED) {
1585 			err = udf_shrink_node(ump->metadatamirror_node, inf_len);
1586 		} else {
1587 			/* extents will be copied on writeout */
1588 		}
1589 		KASSERT(err == 0);
1590 	}
1591 	ump->lvclose |= UDF_WRITE_METAPART_NODES;
1592 
1593 	/* relock before exit */
1594 	mutex_enter(&ump->allocate_mutex);
1595 
1596 	if (to_trunc > num_lb)
1597 		return 0;
1598 	return num_lb - to_trunc;
1599 }
1600 
1601 
1602 static void
1603 udf_sparsify_metadatapart(struct udf_mount *ump, uint32_t num_lb)
1604 {
1605 	/* NOT IMPLEMENTED, fail */
1606 }
1607 
1608 
1609 static void
1610 udf_collect_free_space_for_vpart(struct udf_mount *ump,
1611 	uint16_t vpart_num, uint32_t num_lb)
1612 {
1613 	/* allocate mutex is helt */
1614 
1615 	/* only defined for metadata partitions */
1616 	if (ump->vtop_tp[ump->node_part] != UDF_VTOP_TYPE_META) {
1617 		DPRINTF(RESERVE, ("\tcan't grow/shrink; no metadata partitioning\n"));
1618 		return;
1619 	}
1620 
1621 	/* UDF 2.60 BD-R+POW? */
1622 	if (ump->vtop_alloc[ump->node_part] == UDF_ALLOC_METASEQUENTIAL) {
1623 		DPRINTF(RESERVE, ("\tUDF 2.60 BD-R+POW track grow not implemented yet\n"));
1624 		return;
1625 	}
1626 
1627 	if (ump->vtop_tp[vpart_num] == UDF_VTOP_TYPE_META) {
1628 		/* try to grow the meta partition */
1629 		DPRINTF(RESERVE, ("\ttrying to grow the meta partition\n"));
1630 		/* as per [UDF 2.60/2.2.13.5] : extend bitmap and metadata file(s) */
1631 		DPRINTF(NOTIMPL, ("\tgrowing meta partition not implemented yet\n"));
1632 	} else {
1633 		/* try to shrink the metadata partition */
1634 		DPRINTF(RESERVE, ("\ttrying to shrink the meta partition\n"));
1635 		/* as per [UDF 2.60/2.2.13.6] : either trunc or make sparse */
1636 		num_lb = udf_trunc_metadatapart(ump, num_lb);
1637 		if (num_lb)
1638 			udf_sparsify_metadatapart(ump, num_lb);
1639 	}
1640 
1641 	/* allocate mutex should still be helt */
1642 }
1643 
1644 /* --------------------------------------------------------------------- */
1645 
1646 /*
1647  * Allocate a buf on disc for direct write out. The space doesn't have to be
1648  * contiguous as the caller takes care of this.
1649  */
1650 
1651 void
1652 udf_late_allocate_buf(struct udf_mount *ump, struct buf *buf,
1653 	uint64_t *lmapping, struct long_ad *node_ad_cpy, uint16_t *vpart_nump)
1654 {
1655 	struct udf_node  *udf_node = VTOI(buf->b_vp);
1656 	int lb_size, blks, udf_c_type;
1657 	int vpart_num, num_lb;
1658 	int error, s;
1659 
1660 	/*
1661 	 * for each sector in the buf, allocate a sector on disc and record
1662 	 * its position in the provided mapping array.
1663 	 *
1664 	 * If its userdata or FIDs, record its location in its node.
1665 	 */
1666 
1667 	lb_size    = udf_rw32(ump->logical_vol->lb_size);
1668 	num_lb     = (buf->b_bcount + lb_size -1) / lb_size;
1669 	blks       = lb_size / DEV_BSIZE;
1670 	udf_c_type = buf->b_udf_c_type;
1671 
1672 	KASSERT(lb_size == ump->discinfo.sector_size);
1673 
1674 	/* select partition to record the buffer on */
1675 	vpart_num = *vpart_nump = udf_get_record_vpart(ump, udf_c_type);
1676 
1677 	if (udf_c_type == UDF_C_NODE) {
1678 		/* if not VAT, its allready allocated */
1679 		if (ump->vtop_alloc[ump->node_part] != UDF_ALLOC_VAT)
1680 			return;
1681 
1682 		/* allocate on its backing sequential partition */
1683 		vpart_num = ump->data_part;
1684 	}
1685 
1686 	/* XXX can this still happen? */
1687 	/* do allocation on the selected partition */
1688 	error = udf_allocate_space(ump, udf_node, udf_c_type,
1689 			vpart_num, num_lb, lmapping);
1690 	if (error) {
1691 		/*
1692 		 * ARGH! we haven't done our accounting right! it should
1693 		 * allways succeed.
1694 		 */
1695 		panic("UDF disc allocation accounting gone wrong");
1696 	}
1697 
1698 	/* If its userdata or FIDs, record its allocation in its node. */
1699 	if ((udf_c_type == UDF_C_USERDATA) ||
1700 	    (udf_c_type == UDF_C_FIDS) ||
1701 	    (udf_c_type == UDF_C_METADATA_SBM))
1702 	{
1703 		udf_record_allocation_in_node(ump, buf, vpart_num, lmapping,
1704 			node_ad_cpy);
1705 		/* decrement our outstanding bufs counter */
1706 		s = splbio();
1707 			udf_node->outstanding_bufs--;
1708 		splx(s);
1709 	}
1710 }
1711 
1712 /* --------------------------------------------------------------------- */
1713 
1714 /*
1715  * Try to merge a1 with the new piece a2. udf_ads_merge returns error when not
1716  * possible (anymore); a2 returns the rest piece.
1717  */
1718 
1719 static int
1720 udf_ads_merge(uint32_t max_len, uint32_t lb_size, struct long_ad *a1, struct long_ad *a2)
1721 {
1722 	uint32_t merge_len;
1723 	uint32_t a1_len, a2_len;
1724 	uint32_t a1_flags, a2_flags;
1725 	uint32_t a1_lbnum, a2_lbnum;
1726 	uint16_t a1_part, a2_part;
1727 
1728 	a1_flags = UDF_EXT_FLAGS(udf_rw32(a1->len));
1729 	a1_len   = UDF_EXT_LEN(udf_rw32(a1->len));
1730 	a1_lbnum = udf_rw32(a1->loc.lb_num);
1731 	a1_part  = udf_rw16(a1->loc.part_num);
1732 
1733 	a2_flags = UDF_EXT_FLAGS(udf_rw32(a2->len));
1734 	a2_len   = UDF_EXT_LEN(udf_rw32(a2->len));
1735 	a2_lbnum = udf_rw32(a2->loc.lb_num);
1736 	a2_part  = udf_rw16(a2->loc.part_num);
1737 
1738 	/* defines same space */
1739 	if (a1_flags != a2_flags)
1740 		return 1;
1741 
1742 	if (a1_flags != UDF_EXT_FREE) {
1743 		/* the same partition */
1744 		if (a1_part != a2_part)
1745 			return 1;
1746 
1747 		/* a2 is successor of a1 */
1748 		if (a1_lbnum * lb_size + a1_len != a2_lbnum * lb_size)
1749 			return 1;
1750 	}
1751 
1752 	/* merge as most from a2 if possible */
1753 	merge_len = MIN(a2_len, max_len - a1_len);
1754 	a1_len   += merge_len;
1755 	a2_len   -= merge_len;
1756 	a2_lbnum += merge_len/lb_size;
1757 
1758 	a1->len = udf_rw32(a1_len | a1_flags);
1759 	a2->len = udf_rw32(a2_len | a2_flags);
1760 	a2->loc.lb_num = udf_rw32(a2_lbnum);
1761 
1762 	if (a2_len > 0)
1763 		return 1;
1764 
1765 	/* there is space over to merge */
1766 	return 0;
1767 }
1768 
1769 /* --------------------------------------------------------------------- */
1770 
1771 static void
1772 udf_wipe_adslots(struct udf_node *udf_node)
1773 {
1774 	struct file_entry      *fe;
1775 	struct extfile_entry   *efe;
1776 	struct alloc_ext_entry *ext;
1777 	uint64_t inflen, objsize;
1778 	uint32_t lb_size, dscr_size, l_ea, l_ad, max_l_ad, crclen;
1779 	uint8_t *data_pos;
1780 	int extnr;
1781 
1782 	lb_size = udf_rw32(udf_node->ump->logical_vol->lb_size);
1783 
1784 	fe  = udf_node->fe;
1785 	efe = udf_node->efe;
1786 	if (fe) {
1787 		inflen  = udf_rw64(fe->inf_len);
1788 		objsize = inflen;
1789 		dscr_size  = sizeof(struct file_entry) -1;
1790 		l_ea       = udf_rw32(fe->l_ea);
1791 		l_ad       = udf_rw32(fe->l_ad);
1792 		data_pos = (uint8_t *) fe + dscr_size + l_ea;
1793 	} else {
1794 		inflen  = udf_rw64(efe->inf_len);
1795 		objsize = udf_rw64(efe->obj_size);
1796 		dscr_size  = sizeof(struct extfile_entry) -1;
1797 		l_ea       = udf_rw32(efe->l_ea);
1798 		l_ad       = udf_rw32(efe->l_ad);
1799 		data_pos = (uint8_t *) efe + dscr_size + l_ea;
1800 	}
1801 	max_l_ad = lb_size - dscr_size - l_ea;
1802 
1803 	/* wipe fe/efe */
1804 	memset(data_pos, 0, max_l_ad);
1805 	crclen = dscr_size - UDF_DESC_TAG_LENGTH + l_ea;
1806 	if (fe) {
1807 		fe->l_ad         = udf_rw32(0);
1808 		fe->logblks_rec  = udf_rw64(0);
1809 		fe->tag.desc_crc_len = udf_rw16(crclen);
1810 	} else {
1811 		efe->l_ad        = udf_rw32(0);
1812 		efe->logblks_rec = udf_rw64(0);
1813 		efe->tag.desc_crc_len = udf_rw16(crclen);
1814 	}
1815 
1816 	/* wipe all allocation extent entries */
1817 	for (extnr = 0; extnr < udf_node->num_extensions; extnr++) {
1818 		ext = udf_node->ext[extnr];
1819 		dscr_size  = sizeof(struct alloc_ext_entry) -1;
1820 		data_pos = (uint8_t *) ext->data;
1821 		max_l_ad = lb_size - dscr_size;
1822 		memset(data_pos, 0, max_l_ad);
1823 		ext->l_ad = udf_rw32(0);
1824 
1825 		crclen = dscr_size - UDF_DESC_TAG_LENGTH;
1826 		ext->tag.desc_crc_len = udf_rw16(crclen);
1827 	}
1828 	udf_node->i_flags |= IN_NODE_REBUILD;
1829 }
1830 
1831 /* --------------------------------------------------------------------- */
1832 
1833 void
1834 udf_get_adslot(struct udf_node *udf_node, int slot, struct long_ad *icb,
1835 	int *eof) {
1836 	struct file_entry      *fe;
1837 	struct extfile_entry   *efe;
1838 	struct alloc_ext_entry *ext;
1839 	struct icb_tag *icbtag;
1840 	struct short_ad *short_ad;
1841 	struct long_ad *long_ad, l_icb;
1842 	uint32_t offset;
1843 	uint32_t lb_size, dscr_size, l_ea, l_ad, flags;
1844 	uint8_t *data_pos;
1845 	int icbflags, addr_type, adlen, extnr;
1846 
1847 	/* determine what descriptor we are in */
1848 	lb_size = udf_rw32(udf_node->ump->logical_vol->lb_size);
1849 
1850 	fe  = udf_node->fe;
1851 	efe = udf_node->efe;
1852 	if (fe) {
1853 		icbtag  = &fe->icbtag;
1854 		dscr_size  = sizeof(struct file_entry) -1;
1855 		l_ea       = udf_rw32(fe->l_ea);
1856 		l_ad       = udf_rw32(fe->l_ad);
1857 		data_pos = (uint8_t *) fe + dscr_size + l_ea;
1858 	} else {
1859 		icbtag  = &efe->icbtag;
1860 		dscr_size  = sizeof(struct extfile_entry) -1;
1861 		l_ea       = udf_rw32(efe->l_ea);
1862 		l_ad       = udf_rw32(efe->l_ad);
1863 		data_pos = (uint8_t *) efe + dscr_size + l_ea;
1864 	}
1865 
1866 	icbflags  = udf_rw16(icbtag->flags);
1867 	addr_type = icbflags & UDF_ICB_TAG_FLAGS_ALLOC_MASK;
1868 
1869 	/* just in case we're called on an intern, its EOF */
1870 	if (addr_type == UDF_ICB_INTERN_ALLOC) {
1871 		memset(icb, 0, sizeof(struct long_ad));
1872 		*eof = 1;
1873 		return;
1874 	}
1875 
1876 	adlen = 0;
1877 	if (addr_type == UDF_ICB_SHORT_ALLOC) {
1878 		adlen = sizeof(struct short_ad);
1879 	} else if (addr_type == UDF_ICB_LONG_ALLOC) {
1880 		adlen = sizeof(struct long_ad);
1881 	}
1882 
1883 	/* if offset too big, we go to the allocation extensions */
1884 	offset = slot * adlen;
1885 	extnr  = -1;
1886 	while (offset >= l_ad) {
1887 		/* check if our last entry is a redirect */
1888 		if (addr_type == UDF_ICB_SHORT_ALLOC) {
1889 			short_ad = (struct short_ad *) (data_pos + l_ad-adlen);
1890 			l_icb.len          = short_ad->len;
1891 			l_icb.loc.part_num = udf_node->loc.loc.part_num;
1892 			l_icb.loc.lb_num   = short_ad->lb_num;
1893 		} else {
1894 			KASSERT(addr_type == UDF_ICB_LONG_ALLOC);
1895 			long_ad = (struct long_ad *) (data_pos + l_ad-adlen);
1896 			l_icb = *long_ad;
1897 		}
1898 		flags = UDF_EXT_FLAGS(udf_rw32(l_icb.len));
1899 		if (flags != UDF_EXT_REDIRECT) {
1900 			l_ad = 0;	/* force EOF */
1901 			break;
1902 		}
1903 
1904 		/* advance to next extent */
1905 		extnr++;
1906 		if (extnr >= udf_node->num_extensions) {
1907 			l_ad = 0;	/* force EOF */
1908 			break;
1909 		}
1910 		offset = offset - l_ad;
1911 		ext  = udf_node->ext[extnr];
1912 		dscr_size  = sizeof(struct alloc_ext_entry) -1;
1913 		l_ad = udf_rw32(ext->l_ad);
1914 		data_pos = (uint8_t *) ext + dscr_size;
1915 	}
1916 
1917 	/* XXX l_ad == 0 should be enough to check */
1918 	*eof = (offset >= l_ad) || (l_ad == 0);
1919 	if (*eof) {
1920 		DPRINTF(PARANOIDADWLK, ("returning EOF, extnr %d, offset %d, "
1921 			"l_ad %d\n", extnr, offset, l_ad));
1922 		memset(icb, 0, sizeof(struct long_ad));
1923 		return;
1924 	}
1925 
1926 	/* get the element */
1927 	if (addr_type == UDF_ICB_SHORT_ALLOC) {
1928 		short_ad = (struct short_ad *) (data_pos + offset);
1929 		icb->len          = short_ad->len;
1930 		icb->loc.part_num = udf_node->loc.loc.part_num;
1931 		icb->loc.lb_num   = short_ad->lb_num;
1932 	} else if (addr_type == UDF_ICB_LONG_ALLOC) {
1933 		long_ad = (struct long_ad *) (data_pos + offset);
1934 		*icb = *long_ad;
1935 	}
1936 	DPRINTF(PARANOIDADWLK, ("returning element : v %d, lb %d, len %d, "
1937 		"flags %d\n", icb->loc.part_num, icb->loc.lb_num,
1938 		UDF_EXT_LEN(icb->len), UDF_EXT_FLAGS(icb->len)));
1939 }
1940 
1941 /* --------------------------------------------------------------------- */
1942 
1943 int
1944 udf_append_adslot(struct udf_node *udf_node, int *slot, struct long_ad *icb) {
1945 	struct udf_mount *ump = udf_node->ump;
1946 	union dscrptr          *dscr, *extdscr;
1947 	struct file_entry      *fe;
1948 	struct extfile_entry   *efe;
1949 	struct alloc_ext_entry *ext;
1950 	struct icb_tag *icbtag;
1951 	struct short_ad *short_ad;
1952 	struct long_ad *long_ad, o_icb, l_icb;
1953 	uint64_t logblks_rec, *logblks_rec_p;
1954 	uint64_t lmapping;
1955 	uint32_t offset, rest, len, lb_num;
1956 	uint32_t lb_size, dscr_size, l_ea, l_ad, *l_ad_p, max_l_ad, crclen;
1957 	uint32_t flags;
1958 	uint16_t vpart_num;
1959 	uint8_t *data_pos;
1960 	int icbflags, addr_type, adlen, extnr;
1961 	int error;
1962 
1963 	lb_size = udf_rw32(ump->logical_vol->lb_size);
1964 	vpart_num = udf_rw16(udf_node->loc.loc.part_num);
1965 
1966 	/* determine what descriptor we are in */
1967 	fe  = udf_node->fe;
1968 	efe = udf_node->efe;
1969 	if (fe) {
1970 		icbtag  = &fe->icbtag;
1971 		dscr      = (union dscrptr *) fe;
1972 		dscr_size = sizeof(struct file_entry) -1;
1973 
1974 		l_ea      = udf_rw32(fe->l_ea);
1975 		l_ad_p    = &fe->l_ad;
1976 		logblks_rec_p = &fe->logblks_rec;
1977 	} else {
1978 		icbtag    = &efe->icbtag;
1979 		dscr      = (union dscrptr *) efe;
1980 		dscr_size = sizeof(struct extfile_entry) -1;
1981 
1982 		l_ea      = udf_rw32(efe->l_ea);
1983 		l_ad_p    = &efe->l_ad;
1984 		logblks_rec_p = &efe->logblks_rec;
1985 	}
1986 	data_pos  = (uint8_t *) dscr + dscr_size + l_ea;
1987 	max_l_ad = lb_size - dscr_size - l_ea;
1988 
1989 	icbflags  = udf_rw16(icbtag->flags);
1990 	addr_type = icbflags & UDF_ICB_TAG_FLAGS_ALLOC_MASK;
1991 
1992 	/* just in case we're called on an intern, its EOF */
1993 	if (addr_type == UDF_ICB_INTERN_ALLOC) {
1994 		panic("udf_append_adslot on UDF_ICB_INTERN_ALLOC\n");
1995 	}
1996 
1997 	adlen = 0;
1998 	if (addr_type == UDF_ICB_SHORT_ALLOC) {
1999 		adlen = sizeof(struct short_ad);
2000 	} else if (addr_type == UDF_ICB_LONG_ALLOC) {
2001 		adlen = sizeof(struct long_ad);
2002 	}
2003 
2004 	/* clean up given long_ad since it can be a synthesized one */
2005 	flags = UDF_EXT_FLAGS(udf_rw32(icb->len));
2006 	if (flags == UDF_EXT_FREE) {
2007 		icb->loc.part_num = udf_rw16(0);
2008 		icb->loc.lb_num   = udf_rw32(0);
2009 	}
2010 
2011 	/* if offset too big, we go to the allocation extensions */
2012 	l_ad   = udf_rw32(*l_ad_p);
2013 	offset = (*slot) * adlen;
2014 	extnr  = -1;
2015 	while (offset >= l_ad) {
2016 		/* check if our last entry is a redirect */
2017 		if (addr_type == UDF_ICB_SHORT_ALLOC) {
2018 			short_ad = (struct short_ad *) (data_pos + l_ad-adlen);
2019 			l_icb.len          = short_ad->len;
2020 			l_icb.loc.part_num = udf_node->loc.loc.part_num;
2021 			l_icb.loc.lb_num   = short_ad->lb_num;
2022 		} else {
2023 			KASSERT(addr_type == UDF_ICB_LONG_ALLOC);
2024 			long_ad = (struct long_ad *) (data_pos + l_ad-adlen);
2025 			l_icb = *long_ad;
2026 		}
2027 		flags = UDF_EXT_FLAGS(udf_rw32(l_icb.len));
2028 		if (flags != UDF_EXT_REDIRECT) {
2029 			/* only one past the last one is adressable */
2030 			break;
2031 		}
2032 
2033 		/* advance to next extent */
2034 		extnr++;
2035 		KASSERT(extnr < udf_node->num_extensions);
2036 		offset = offset - l_ad;
2037 
2038 		ext  = udf_node->ext[extnr];
2039 		dscr = (union dscrptr *) ext;
2040 		dscr_size  = sizeof(struct alloc_ext_entry) -1;
2041 		max_l_ad = lb_size - dscr_size;
2042 		l_ad_p = &ext->l_ad;
2043 		l_ad   = udf_rw32(*l_ad_p);
2044 		data_pos = (uint8_t *) ext + dscr_size;
2045 	}
2046 	DPRINTF(PARANOIDADWLK, ("append, ext %d, offset %d, l_ad %d\n",
2047 		extnr, offset, udf_rw32(*l_ad_p)));
2048 	KASSERT(l_ad == udf_rw32(*l_ad_p));
2049 
2050 	/* offset is offset within the current (E)FE/AED */
2051 	l_ad   = udf_rw32(*l_ad_p);
2052 	crclen = udf_rw16(dscr->tag.desc_crc_len);
2053 	logblks_rec = udf_rw64(*logblks_rec_p);
2054 
2055 	/* overwriting old piece? */
2056 	if (offset < l_ad) {
2057 		/* overwrite entry; compensate for the old element */
2058 		if (addr_type == UDF_ICB_SHORT_ALLOC) {
2059 			short_ad = (struct short_ad *) (data_pos + offset);
2060 			o_icb.len          = short_ad->len;
2061 			o_icb.loc.part_num = udf_rw16(0);	/* ignore */
2062 			o_icb.loc.lb_num   = short_ad->lb_num;
2063 		} else if (addr_type == UDF_ICB_LONG_ALLOC) {
2064 			long_ad = (struct long_ad *) (data_pos + offset);
2065 			o_icb = *long_ad;
2066 		} else {
2067 			panic("Invalid address type in udf_append_adslot\n");
2068 		}
2069 
2070 		len = udf_rw32(o_icb.len);
2071 		if (UDF_EXT_FLAGS(len) == UDF_EXT_ALLOCATED) {
2072 			/* adjust counts */
2073 			len = UDF_EXT_LEN(len);
2074 			logblks_rec -= (len + lb_size -1) / lb_size;
2075 		}
2076 	}
2077 
2078 	/* check if we're not appending a redirection */
2079 	flags = UDF_EXT_FLAGS(udf_rw32(icb->len));
2080 	KASSERT(flags != UDF_EXT_REDIRECT);
2081 
2082 	/* round down available space */
2083 	rest = adlen * ((max_l_ad - offset) / adlen);
2084 	if (rest <= adlen) {
2085 		/* have to append aed, see if we already have a spare one */
2086 		extnr++;
2087 		ext = udf_node->ext[extnr];
2088 		l_icb = udf_node->ext_loc[extnr];
2089 		if (ext == NULL) {
2090 			DPRINTF(ALLOC,("adding allocation extent %d\n", extnr));
2091 
2092 			error = udf_reserve_space(ump, NULL, UDF_C_NODE,
2093 					vpart_num, 1, /* can fail */ false);
2094 			if (error) {
2095 				printf("UDF: couldn't reserve space for AED!\n");
2096 				return error;
2097 			}
2098 			error = udf_allocate_space(ump, NULL, UDF_C_NODE,
2099 					vpart_num, 1, &lmapping);
2100 			lb_num = lmapping;
2101 			if (error)
2102 				panic("UDF: couldn't allocate AED!\n");
2103 
2104 			/* initialise pointer to location */
2105 			memset(&l_icb, 0, sizeof(struct long_ad));
2106 			l_icb.len = udf_rw32(lb_size | UDF_EXT_REDIRECT);
2107 			l_icb.loc.lb_num   = udf_rw32(lb_num);
2108 			l_icb.loc.part_num = udf_rw16(vpart_num);
2109 
2110 			/* create new aed descriptor */
2111 			udf_create_logvol_dscr(ump, udf_node, &l_icb, &extdscr);
2112 			ext = &extdscr->aee;
2113 
2114 			udf_inittag(ump, &ext->tag, TAGID_ALLOCEXTENT, lb_num);
2115 			dscr_size  = sizeof(struct alloc_ext_entry) -1;
2116 			max_l_ad = lb_size - dscr_size;
2117 			memset(ext->data, 0, max_l_ad);
2118 			ext->l_ad = udf_rw32(0);
2119 			ext->tag.desc_crc_len =
2120 				udf_rw16(dscr_size - UDF_DESC_TAG_LENGTH);
2121 
2122 			/* declare aed */
2123 			udf_node->num_extensions++;
2124 			udf_node->ext_loc[extnr] = l_icb;
2125 			udf_node->ext[extnr] = ext;
2126 		}
2127 		/* add redirect and adjust l_ad and crclen for old descr */
2128 		if (addr_type == UDF_ICB_SHORT_ALLOC) {
2129 			short_ad = (struct short_ad *) (data_pos + offset);
2130 			short_ad->len    = l_icb.len;
2131 			short_ad->lb_num = l_icb.loc.lb_num;
2132 		} else if (addr_type == UDF_ICB_LONG_ALLOC) {
2133 			long_ad = (struct long_ad *) (data_pos + offset);
2134 			*long_ad = l_icb;
2135 		}
2136 		l_ad   += adlen;
2137 		crclen += adlen;
2138 		dscr->tag.desc_crc_len = udf_rw16(crclen);
2139 		*l_ad_p = udf_rw32(l_ad);
2140 
2141 		/* advance to the new extension */
2142 		KASSERT(ext != NULL);
2143 		dscr = (union dscrptr *) ext;
2144 		dscr_size  = sizeof(struct alloc_ext_entry) -1;
2145 		max_l_ad = lb_size - dscr_size;
2146 		data_pos = (uint8_t *) dscr + dscr_size;
2147 
2148 		l_ad_p = &ext->l_ad;
2149 		l_ad   = udf_rw32(*l_ad_p);
2150 		crclen = udf_rw16(dscr->tag.desc_crc_len);
2151 		offset = 0;
2152 
2153 		/* adjust callees slot count for link insert */
2154 		*slot += 1;
2155 	}
2156 
2157 	/* write out the element */
2158 	DPRINTF(PARANOIDADWLK, ("adding element : %p : v %d, lb %d, "
2159 			"len %d, flags %d\n", data_pos + offset,
2160 			icb->loc.part_num, icb->loc.lb_num,
2161 			UDF_EXT_LEN(icb->len), UDF_EXT_FLAGS(icb->len)));
2162 	if (addr_type == UDF_ICB_SHORT_ALLOC) {
2163 		short_ad = (struct short_ad *) (data_pos + offset);
2164 		short_ad->len    = icb->len;
2165 		short_ad->lb_num = icb->loc.lb_num;
2166 	} else if (addr_type == UDF_ICB_LONG_ALLOC) {
2167 		long_ad = (struct long_ad *) (data_pos + offset);
2168 		*long_ad = *icb;
2169 	}
2170 
2171 	/* adjust logblks recorded count */
2172 	len = udf_rw32(icb->len);
2173 	flags = UDF_EXT_FLAGS(len);
2174 	if (flags == UDF_EXT_ALLOCATED)
2175 		logblks_rec += (UDF_EXT_LEN(len) + lb_size -1) / lb_size;
2176 	*logblks_rec_p = udf_rw64(logblks_rec);
2177 
2178 	/* adjust l_ad and crclen when needed */
2179 	if (offset >= l_ad) {
2180 		l_ad   += adlen;
2181 		crclen += adlen;
2182 		dscr->tag.desc_crc_len = udf_rw16(crclen);
2183 		*l_ad_p = udf_rw32(l_ad);
2184 	}
2185 
2186 	return 0;
2187 }
2188 
2189 /* --------------------------------------------------------------------- */
2190 
2191 static void
2192 udf_count_alloc_exts(struct udf_node *udf_node)
2193 {
2194 	struct long_ad s_ad;
2195 	uint32_t lb_num, len, flags;
2196 	uint16_t vpart_num;
2197 	int slot, eof;
2198 	int num_extents, extnr;
2199 	int lb_size;
2200 
2201 	if (udf_node->num_extensions == 0)
2202 		return;
2203 
2204 	lb_size = udf_rw32(udf_node->ump->logical_vol->lb_size);
2205 	/* count number of allocation extents in use */
2206 	num_extents = 0;
2207 	slot = 0;
2208 	for (;;) {
2209 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
2210 		if (eof)
2211 			break;
2212 		len   = udf_rw32(s_ad.len);
2213 		flags = UDF_EXT_FLAGS(len);
2214 
2215 		if (flags == UDF_EXT_REDIRECT)
2216 			num_extents++;
2217 
2218 		slot++;
2219 	}
2220 
2221 	DPRINTF(ALLOC, ("udf_count_alloc_ext counted %d live extents\n",
2222 		num_extents));
2223 
2224 	/* XXX choice: we could delay freeing them on node writeout */
2225 	/* free excess entries */
2226 	extnr = num_extents;
2227 	for (;extnr < udf_node->num_extensions; extnr++) {
2228 		DPRINTF(ALLOC, ("freeing alloc ext %d\n", extnr));
2229 		/* free dscriptor */
2230 		s_ad = udf_node->ext_loc[extnr];
2231 		udf_free_logvol_dscr(udf_node->ump, &s_ad,
2232 			udf_node->ext[extnr]);
2233 		udf_node->ext[extnr] = NULL;
2234 
2235 		/* free disc space */
2236 		lb_num    = udf_rw32(s_ad.loc.lb_num);
2237 		vpart_num = udf_rw16(s_ad.loc.part_num);
2238 		udf_free_allocated_space(udf_node->ump, lb_num, vpart_num, 1);
2239 
2240 		memset(&udf_node->ext_loc[extnr], 0, sizeof(struct long_ad));
2241 	}
2242 
2243 	/* set our new number of allocation extents */
2244 	udf_node->num_extensions = num_extents;
2245 }
2246 
2247 
2248 /* --------------------------------------------------------------------- */
2249 
2250 /*
2251  * Adjust the node's allocation descriptors to reflect the new mapping; do
2252  * take note that we might glue to existing allocation descriptors.
2253  *
2254  * XXX Note there can only be one allocation being recorded/mount; maybe
2255  * explicit allocation in shedule thread?
2256  */
2257 
2258 static void
2259 udf_record_allocation_in_node(struct udf_mount *ump, struct buf *buf,
2260 	uint16_t vpart_num, uint64_t *mapping, struct long_ad *node_ad_cpy)
2261 {
2262 	struct vnode    *vp = buf->b_vp;
2263 	struct udf_node *udf_node = VTOI(vp);
2264 	struct file_entry      *fe;
2265 	struct extfile_entry   *efe;
2266 	struct icb_tag  *icbtag;
2267 	struct long_ad   s_ad, c_ad;
2268 	uint64_t inflen, from, till;
2269 	uint64_t foffset, end_foffset, restart_foffset;
2270 	uint64_t orig_inflen, orig_lbrec, new_inflen, new_lbrec;
2271 	uint32_t max_len;
2272 	uint32_t num_lb, len, flags, lb_num;
2273 	uint32_t run_start;
2274 	uint32_t slot_offset, replace_len, replace;
2275 	int addr_type, icbflags;
2276 //	int udf_c_type = buf->b_udf_c_type;
2277 	int lb_size, run_length, eof;
2278 	int slot, cpy_slot, cpy_slots, restart_slot;
2279 	int error;
2280 
2281 	DPRINTF(ALLOC, ("udf_record_allocation_in_node\n"));
2282 
2283 #if 0
2284 	/* XXX disable sanity check for now */
2285 	/* sanity check ... should be panic ? */
2286 	if ((udf_c_type != UDF_C_USERDATA) && (udf_c_type != UDF_C_FIDS))
2287 		return;
2288 #endif
2289 
2290 	lb_size = udf_rw32(udf_node->ump->logical_vol->lb_size);
2291 	max_len = ((UDF_EXT_MAXLEN / lb_size) * lb_size);
2292 
2293 	/* do the job */
2294 	UDF_LOCK_NODE(udf_node, 0);	/* XXX can deadlock ? */
2295 	udf_node_sanity_check(udf_node, &orig_inflen, &orig_lbrec);
2296 
2297 	fe  = udf_node->fe;
2298 	efe = udf_node->efe;
2299 	if (fe) {
2300 		icbtag = &fe->icbtag;
2301 		inflen = udf_rw64(fe->inf_len);
2302 	} else {
2303 		icbtag = &efe->icbtag;
2304 		inflen = udf_rw64(efe->inf_len);
2305 	}
2306 
2307 	/* do check if `till' is not past file information length */
2308 	from = buf->b_lblkno * lb_size;
2309 	till = MIN(inflen, from + buf->b_resid);
2310 
2311 	num_lb = (till - from + lb_size -1) / lb_size;
2312 
2313 	DPRINTF(ALLOC, ("record allocation from %"PRIu64" + %d\n", from, buf->b_bcount));
2314 
2315 	icbflags  = udf_rw16(icbtag->flags);
2316 	addr_type = icbflags & UDF_ICB_TAG_FLAGS_ALLOC_MASK;
2317 
2318 	if (addr_type == UDF_ICB_INTERN_ALLOC) {
2319 		/* nothing to do */
2320 		/* XXX clean up rest of node? just in case? */
2321 		UDF_UNLOCK_NODE(udf_node, 0);
2322 		return;
2323 	}
2324 
2325 	slot     = 0;
2326 	cpy_slot = 0;
2327 	foffset  = 0;
2328 
2329 	/* 1) copy till first overlap piece to the rewrite buffer */
2330 	for (;;) {
2331 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
2332 		if (eof) {
2333 			DPRINTF(WRITE,
2334 				("Record allocation in node "
2335 				 "failed: encountered EOF\n"));
2336 			UDF_UNLOCK_NODE(udf_node, 0);
2337 			buf->b_error = EINVAL;
2338 			return;
2339 		}
2340 		len   = udf_rw32(s_ad.len);
2341 		flags = UDF_EXT_FLAGS(len);
2342 		len   = UDF_EXT_LEN(len);
2343 
2344 		if (flags == UDF_EXT_REDIRECT) {
2345 			slot++;
2346 			continue;
2347 		}
2348 
2349 		end_foffset = foffset + len;
2350 		if (end_foffset > from)
2351 			break;	/* found */
2352 
2353 		node_ad_cpy[cpy_slot++] = s_ad;
2354 
2355 		DPRINTF(ALLOC, ("\t1: vp %d, lb %d, len %d, flags %d "
2356 			"-> stack\n",
2357 			udf_rw16(s_ad.loc.part_num),
2358 			udf_rw32(s_ad.loc.lb_num),
2359 			UDF_EXT_LEN(udf_rw32(s_ad.len)),
2360 			UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
2361 
2362 		foffset = end_foffset;
2363 		slot++;
2364 	}
2365 	restart_slot    = slot;
2366 	restart_foffset = foffset;
2367 
2368 	/* 2) trunc overlapping slot at overlap and copy it */
2369 	slot_offset = from - foffset;
2370 	if (slot_offset > 0) {
2371 		DPRINTF(ALLOC, ("\tslot_offset = %d, flags = %d (%d)\n",
2372 				slot_offset, flags >> 30, flags));
2373 
2374 		s_ad.len = udf_rw32(slot_offset | flags);
2375 		node_ad_cpy[cpy_slot++] = s_ad;
2376 
2377 		DPRINTF(ALLOC, ("\t2: vp %d, lb %d, len %d, flags %d "
2378 			"-> stack\n",
2379 			udf_rw16(s_ad.loc.part_num),
2380 			udf_rw32(s_ad.loc.lb_num),
2381 			UDF_EXT_LEN(udf_rw32(s_ad.len)),
2382 			UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
2383 	}
2384 	foffset += slot_offset;
2385 
2386 	/* 3) insert new mappings */
2387 	memset(&s_ad, 0, sizeof(struct long_ad));
2388 	lb_num = 0;
2389 	for (lb_num = 0; lb_num < num_lb; lb_num++) {
2390 		run_start  = mapping[lb_num];
2391 		run_length = 1;
2392 		while (lb_num < num_lb-1) {
2393 			if (mapping[lb_num+1] != mapping[lb_num]+1)
2394 				if (mapping[lb_num+1] != mapping[lb_num])
2395 					break;
2396 			run_length++;
2397 			lb_num++;
2398 		}
2399 		/* insert slot for this mapping */
2400 		len = run_length * lb_size;
2401 
2402 		/* bounds checking */
2403 		if (foffset + len > till)
2404 			len = till - foffset;
2405 		KASSERT(foffset + len <= inflen);
2406 
2407 		s_ad.len = udf_rw32(len | UDF_EXT_ALLOCATED);
2408 		s_ad.loc.part_num = udf_rw16(vpart_num);
2409 		s_ad.loc.lb_num   = udf_rw32(run_start);
2410 
2411 		foffset += len;
2412 
2413 		/* paranoia */
2414 		if (len == 0) {
2415 			DPRINTF(WRITE,
2416 				("Record allocation in node "
2417 				 "failed: insert failed\n"));
2418 			UDF_UNLOCK_NODE(udf_node, 0);
2419 			buf->b_error = EINVAL;
2420 			return;
2421 		}
2422 		node_ad_cpy[cpy_slot++] = s_ad;
2423 
2424 		DPRINTF(ALLOC, ("\t3: insert new mapping vp %d lb %d, len %d, "
2425 				"flags %d -> stack\n",
2426 			udf_rw16(s_ad.loc.part_num), udf_rw32(s_ad.loc.lb_num),
2427 			UDF_EXT_LEN(udf_rw32(s_ad.len)),
2428 			UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
2429 	}
2430 
2431 	/* 4) pop replaced length */
2432 	slot    = restart_slot;
2433 	foffset = restart_foffset;
2434 
2435 	replace_len = till - foffset;	/* total amount of bytes to pop */
2436 	slot_offset = from - foffset;	/* offset in first encounted slot */
2437 	KASSERT((slot_offset % lb_size) == 0);
2438 
2439 	for (;;) {
2440 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
2441 		if (eof)
2442 			break;
2443 
2444 		len    = udf_rw32(s_ad.len);
2445 		flags  = UDF_EXT_FLAGS(len);
2446 		len    = UDF_EXT_LEN(len);
2447 		lb_num = udf_rw32(s_ad.loc.lb_num);
2448 
2449 		if (flags == UDF_EXT_REDIRECT) {
2450 			slot++;
2451 			continue;
2452 		}
2453 
2454 		DPRINTF(ALLOC, ("\t4i: got slot %d, slot_offset %d, "
2455 				"replace_len %d, "
2456 				"vp %d, lb %d, len %d, flags %d\n",
2457 			slot, slot_offset, replace_len,
2458 			udf_rw16(s_ad.loc.part_num),
2459 			udf_rw32(s_ad.loc.lb_num),
2460 			UDF_EXT_LEN(udf_rw32(s_ad.len)),
2461 			UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
2462 
2463 		/* adjust for slot offset */
2464 		if (slot_offset) {
2465 			DPRINTF(ALLOC, ("\t4s: skipping %d\n", slot_offset));
2466 			lb_num += slot_offset / lb_size;
2467 			len    -= slot_offset;
2468 			foffset += slot_offset;
2469 			replace_len -= slot_offset;
2470 
2471 			/* mark adjusted */
2472 			slot_offset = 0;
2473 		}
2474 
2475 		/* advance for (the rest of) this slot */
2476 		replace = MIN(len, replace_len);
2477 		DPRINTF(ALLOC, ("\t4d: replacing %d\n", replace));
2478 
2479 		/* advance for this slot */
2480 		if (replace) {
2481 			/* note: dont round DOWN on num_lb since we then
2482 			 * forget the last partial one */
2483 			num_lb = (replace + lb_size - 1) / lb_size;
2484 			if (flags != UDF_EXT_FREE) {
2485 				udf_free_allocated_space(ump, lb_num,
2486 					udf_rw16(s_ad.loc.part_num), num_lb);
2487 			}
2488 			lb_num      += num_lb;
2489 			len         -= replace;
2490 			foffset     += replace;
2491 			replace_len -= replace;
2492 		}
2493 
2494 		/* do we have a slot tail ? */
2495 		if (len) {
2496 			KASSERT(foffset % lb_size == 0);
2497 
2498 			/* we arrived at our point, push remainder */
2499 			s_ad.len        = udf_rw32(len | flags);
2500 			s_ad.loc.lb_num = udf_rw32(lb_num);
2501 			if (flags == UDF_EXT_FREE)
2502 				s_ad.loc.lb_num = udf_rw32(0);
2503 			node_ad_cpy[cpy_slot++] = s_ad;
2504 			foffset += len;
2505 			slot++;
2506 
2507 			DPRINTF(ALLOC, ("\t4: vp %d, lb %d, len %d, flags %d "
2508 				"-> stack\n",
2509 				udf_rw16(s_ad.loc.part_num),
2510 				udf_rw32(s_ad.loc.lb_num),
2511 				UDF_EXT_LEN(udf_rw32(s_ad.len)),
2512 				UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
2513 			break;
2514 		}
2515 
2516 		slot++;
2517 	}
2518 
2519 	/* 5) copy remainder */
2520 	for (;;) {
2521 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
2522 		if (eof)
2523 			break;
2524 
2525 		len   = udf_rw32(s_ad.len);
2526 		flags = UDF_EXT_FLAGS(len);
2527 		len   = UDF_EXT_LEN(len);
2528 
2529 		if (flags == UDF_EXT_REDIRECT) {
2530 			slot++;
2531 			continue;
2532 		}
2533 
2534 		node_ad_cpy[cpy_slot++] = s_ad;
2535 
2536 		DPRINTF(ALLOC, ("\t5: insert new mapping "
2537 			"vp %d lb %d, len %d, flags %d "
2538 			"-> stack\n",
2539 		udf_rw16(s_ad.loc.part_num),
2540 		udf_rw32(s_ad.loc.lb_num),
2541 		UDF_EXT_LEN(udf_rw32(s_ad.len)),
2542 		UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
2543 
2544 		slot++;
2545 	}
2546 
2547 	/* 6) reset node descriptors */
2548 	udf_wipe_adslots(udf_node);
2549 
2550 	/* 7) copy back extents; merge when possible. Recounting on the fly */
2551 	cpy_slots = cpy_slot;
2552 
2553 	c_ad = node_ad_cpy[0];
2554 	slot = 0;
2555 	DPRINTF(ALLOC, ("\t7s: stack -> got mapping vp %d "
2556 		"lb %d, len %d, flags %d\n",
2557 	udf_rw16(c_ad.loc.part_num),
2558 	udf_rw32(c_ad.loc.lb_num),
2559 	UDF_EXT_LEN(udf_rw32(c_ad.len)),
2560 	UDF_EXT_FLAGS(udf_rw32(c_ad.len)) >> 30));
2561 
2562 	for (cpy_slot = 1; cpy_slot < cpy_slots; cpy_slot++) {
2563 		s_ad = node_ad_cpy[cpy_slot];
2564 
2565 		DPRINTF(ALLOC, ("\t7i: stack -> got mapping vp %d "
2566 			"lb %d, len %d, flags %d\n",
2567 		udf_rw16(s_ad.loc.part_num),
2568 		udf_rw32(s_ad.loc.lb_num),
2569 		UDF_EXT_LEN(udf_rw32(s_ad.len)),
2570 		UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
2571 
2572 		/* see if we can merge */
2573 		if (udf_ads_merge(max_len, lb_size, &c_ad, &s_ad)) {
2574 			/* not mergable (anymore) */
2575 			DPRINTF(ALLOC, ("\t7: appending vp %d lb %d, "
2576 				"len %d, flags %d\n",
2577 			udf_rw16(c_ad.loc.part_num),
2578 			udf_rw32(c_ad.loc.lb_num),
2579 			UDF_EXT_LEN(udf_rw32(c_ad.len)),
2580 			UDF_EXT_FLAGS(udf_rw32(c_ad.len)) >> 30));
2581 
2582 			error = udf_append_adslot(udf_node, &slot, &c_ad);
2583 			if (error) {
2584 				buf->b_error = error;
2585 				goto out;
2586 			}
2587 			c_ad = s_ad;
2588 			slot++;
2589 		}
2590 	}
2591 
2592 	/* 8) push rest slot (if any) */
2593 	if (UDF_EXT_LEN(c_ad.len) > 0) {
2594 		DPRINTF(ALLOC, ("\t8: last append vp %d lb %d, "
2595 				"len %d, flags %d\n",
2596 		udf_rw16(c_ad.loc.part_num),
2597 		udf_rw32(c_ad.loc.lb_num),
2598 		UDF_EXT_LEN(udf_rw32(c_ad.len)),
2599 		UDF_EXT_FLAGS(udf_rw32(c_ad.len)) >> 30));
2600 
2601 		error = udf_append_adslot(udf_node, &slot, &c_ad);
2602 		if (error) {
2603 			buf->b_error = error;
2604 			goto out;
2605 		}
2606 	}
2607 
2608 out:
2609 	udf_count_alloc_exts(udf_node);
2610 
2611 	/* the node's descriptors should now be sane */
2612 	udf_node_sanity_check(udf_node, &new_inflen, &new_lbrec);
2613 	UDF_UNLOCK_NODE(udf_node, 0);
2614 
2615 	KASSERT(orig_inflen == new_inflen);
2616 	KASSERT(new_lbrec >= orig_lbrec);
2617 
2618 	return;
2619 }
2620 
2621 /* --------------------------------------------------------------------- */
2622 
2623 int
2624 udf_grow_node(struct udf_node *udf_node, uint64_t new_size)
2625 {
2626 	union dscrptr *dscr;
2627 	struct vnode *vp = udf_node->vnode;
2628 	struct udf_mount *ump = udf_node->ump;
2629 	struct file_entry    *fe;
2630 	struct extfile_entry *efe;
2631 	struct icb_tag  *icbtag;
2632 	struct long_ad c_ad, s_ad;
2633 	uint64_t size_diff, old_size, inflen, objsize, chunk, append_len;
2634 	uint64_t foffset, end_foffset;
2635 	uint64_t orig_inflen, orig_lbrec, new_inflen, new_lbrec;
2636 	uint32_t lb_size, unit_size, dscr_size, crclen, lastblock_grow;
2637 	uint32_t icbflags, len, flags, max_len;
2638 	uint32_t max_l_ad, l_ad, l_ea;
2639 	uint16_t my_part, dst_part;
2640 	uint8_t *data_pos, *evacuated_data;
2641 	int addr_type;
2642 	int slot, cpy_slot;
2643 	int eof, error;
2644 
2645 	DPRINTF(ALLOC, ("udf_grow_node\n"));
2646 
2647 	UDF_LOCK_NODE(udf_node, 0);
2648 	udf_node_sanity_check(udf_node, &orig_inflen, &orig_lbrec);
2649 
2650 	lb_size = udf_rw32(ump->logical_vol->lb_size);
2651 
2652 	/* max_len in unit's IFF its a metadata node or metadata mirror node */
2653 	unit_size = lb_size;
2654 	if ((udf_node == ump->metadata_node) || (udf_node == ump->metadatamirror_node))
2655 		unit_size = ump->metadata_alloc_unit_size * lb_size;
2656 	max_len = ((UDF_EXT_MAXLEN / unit_size) * unit_size);
2657 
2658 	fe  = udf_node->fe;
2659 	efe = udf_node->efe;
2660 	if (fe) {
2661 		dscr       = (union dscrptr *) fe;
2662 		icbtag  = &fe->icbtag;
2663 		inflen  = udf_rw64(fe->inf_len);
2664 		objsize = inflen;
2665 		dscr_size  = sizeof(struct file_entry) -1;
2666 		l_ea       = udf_rw32(fe->l_ea);
2667 		l_ad       = udf_rw32(fe->l_ad);
2668 	} else {
2669 		dscr       = (union dscrptr *) efe;
2670 		icbtag  = &efe->icbtag;
2671 		inflen  = udf_rw64(efe->inf_len);
2672 		objsize = udf_rw64(efe->obj_size);
2673 		dscr_size  = sizeof(struct extfile_entry) -1;
2674 		l_ea       = udf_rw32(efe->l_ea);
2675 		l_ad       = udf_rw32(efe->l_ad);
2676 	}
2677 	data_pos  = (uint8_t *) dscr + dscr_size + l_ea;
2678 	max_l_ad = lb_size - dscr_size - l_ea;
2679 
2680 	icbflags   = udf_rw16(icbtag->flags);
2681 	addr_type  = icbflags & UDF_ICB_TAG_FLAGS_ALLOC_MASK;
2682 
2683 	old_size  = inflen;
2684 	size_diff = new_size - old_size;
2685 
2686 	DPRINTF(ALLOC, ("\tfrom %"PRIu64" to %"PRIu64"\n", old_size, new_size));
2687 
2688 	evacuated_data = NULL;
2689 	if (addr_type == UDF_ICB_INTERN_ALLOC) {
2690 		if (l_ad + size_diff <= max_l_ad) {
2691 			/* only reflect size change directly in the node */
2692 			inflen  += size_diff;
2693 			objsize += size_diff;
2694 			l_ad    += size_diff;
2695 			crclen = dscr_size - UDF_DESC_TAG_LENGTH + l_ea + l_ad;
2696 			if (fe) {
2697 				fe->inf_len   = udf_rw64(inflen);
2698 				fe->l_ad      = udf_rw32(l_ad);
2699 				fe->tag.desc_crc_len = udf_rw16(crclen);
2700 			} else {
2701 				efe->inf_len  = udf_rw64(inflen);
2702 				efe->obj_size = udf_rw64(objsize);
2703 				efe->l_ad     = udf_rw32(l_ad);
2704 				efe->tag.desc_crc_len = udf_rw16(crclen);
2705 			}
2706 			error = 0;
2707 
2708 			/* set new size for uvm */
2709 			uvm_vnp_setsize(vp, old_size);
2710 			uvm_vnp_setwritesize(vp, new_size);
2711 
2712 #if 0
2713 			/* zero append space in buffer */
2714 			ubc_zerorange(&vp->v_uobj, old_size,
2715 			    new_size - old_size, UBC_UNMAP_FLAG(vp));
2716 #endif
2717 
2718 			udf_node_sanity_check(udf_node, &new_inflen, &new_lbrec);
2719 
2720 			/* unlock */
2721 			UDF_UNLOCK_NODE(udf_node, 0);
2722 
2723 			KASSERT(new_inflen == orig_inflen + size_diff);
2724 			KASSERT(new_lbrec == orig_lbrec);
2725 			KASSERT(new_lbrec == 0);
2726 			return 0;
2727 		}
2728 
2729 		DPRINTF(ALLOC, ("\tCONVERT from internal\n"));
2730 
2731 		if (old_size > 0) {
2732 			/* allocate some space and copy in the stuff to keep */
2733 			evacuated_data = malloc(lb_size, M_UDFTEMP, M_WAITOK);
2734 			memset(evacuated_data, 0, lb_size);
2735 
2736 			/* node is locked, so safe to exit mutex */
2737 			UDF_UNLOCK_NODE(udf_node, 0);
2738 
2739 			/* read in using the `normal' vn_rdwr() */
2740 			error = vn_rdwr(UIO_READ, udf_node->vnode,
2741 					evacuated_data, old_size, 0,
2742 					UIO_SYSSPACE, IO_ALTSEMANTICS | IO_NODELOCKED,
2743 					FSCRED, NULL, NULL);
2744 
2745 			/* enter again */
2746 			UDF_LOCK_NODE(udf_node, 0);
2747 		}
2748 
2749 		/* convert to a normal alloc and select type */
2750 		my_part  = udf_rw16(udf_node->loc.loc.part_num);
2751 		dst_part = udf_get_record_vpart(ump, udf_get_c_type(udf_node));
2752 		addr_type = UDF_ICB_SHORT_ALLOC;
2753 		if (dst_part != my_part)
2754 			addr_type = UDF_ICB_LONG_ALLOC;
2755 
2756 		icbflags &= ~UDF_ICB_TAG_FLAGS_ALLOC_MASK;
2757 		icbflags |= addr_type;
2758 		icbtag->flags = udf_rw16(icbflags);
2759 
2760 		/* wipe old descriptor space */
2761 		udf_wipe_adslots(udf_node);
2762 
2763 		memset(&c_ad, 0, sizeof(struct long_ad));
2764 		c_ad.len          = udf_rw32(old_size | UDF_EXT_FREE);
2765 		c_ad.loc.part_num = udf_rw16(0); /* not relevant */
2766 		c_ad.loc.lb_num   = udf_rw32(0); /* not relevant */
2767 
2768 		slot = 0;
2769 	} else {
2770 		/* goto the last entry (if any) */
2771 		slot     = 0;
2772 		cpy_slot = 0;
2773 		foffset  = 0;
2774 		memset(&c_ad, 0, sizeof(struct long_ad));
2775 		for (;;) {
2776 			udf_get_adslot(udf_node, slot, &c_ad, &eof);
2777 			if (eof)
2778 				break;
2779 
2780 			len   = udf_rw32(c_ad.len);
2781 			flags = UDF_EXT_FLAGS(len);
2782 			len   = UDF_EXT_LEN(len);
2783 
2784 			end_foffset = foffset + len;
2785 			if (flags != UDF_EXT_REDIRECT)
2786 				foffset = end_foffset;
2787 
2788 			slot++;
2789 		}
2790 		/* at end of adslots */
2791 
2792 		/* special case if the old size was zero, then there is no last slot */
2793 		if (old_size == 0) {
2794 			c_ad.len          = udf_rw32(0 | UDF_EXT_FREE);
2795 			c_ad.loc.part_num = udf_rw16(0); /* not relevant */
2796 			c_ad.loc.lb_num   = udf_rw32(0); /* not relevant */
2797 		} else {
2798 			/* refetch last slot */
2799 			slot--;
2800 			udf_get_adslot(udf_node, slot, &c_ad, &eof);
2801 		}
2802 	}
2803 
2804 	/*
2805 	 * If the length of the last slot is not a multiple of lb_size, adjust
2806 	 * length so that it is; don't forget to adjust `append_len'! relevant for
2807 	 * extending existing files
2808 	 */
2809 	len   = udf_rw32(c_ad.len);
2810 	flags = UDF_EXT_FLAGS(len);
2811 	len   = UDF_EXT_LEN(len);
2812 
2813 	lastblock_grow = 0;
2814 	if (len % lb_size > 0) {
2815 		lastblock_grow = lb_size - (len % lb_size);
2816 		lastblock_grow = MIN(size_diff, lastblock_grow);
2817 		len += lastblock_grow;
2818 		c_ad.len = udf_rw32(len | flags);
2819 
2820 		/* TODO zero appened space in buffer! */
2821 		/* using ubc_zerorange(&vp->v_uobj, old_size, */
2822 		/*    new_size - old_size, UBC_UNMAP_FLAG(vp)); ? */
2823 	}
2824 	memset(&s_ad, 0, sizeof(struct long_ad));
2825 
2826 	/* size_diff can be bigger than allowed, so grow in chunks */
2827 	append_len = size_diff - lastblock_grow;
2828 	while (append_len > 0) {
2829 		chunk = MIN(append_len, max_len);
2830 		s_ad.len = udf_rw32(chunk | UDF_EXT_FREE);
2831 		s_ad.loc.part_num = udf_rw16(0);
2832 		s_ad.loc.lb_num   = udf_rw32(0);
2833 
2834 		if (udf_ads_merge(max_len, lb_size, &c_ad, &s_ad)) {
2835 			/* not mergable (anymore) */
2836 			error = udf_append_adslot(udf_node, &slot, &c_ad);
2837 			if (error)
2838 				goto errorout;
2839 			slot++;
2840 			c_ad = s_ad;
2841 			memset(&s_ad, 0, sizeof(struct long_ad));
2842 		}
2843 		append_len -= chunk;
2844 	}
2845 
2846 	/* if there is a rest piece in the accumulator, append it */
2847 	if (UDF_EXT_LEN(udf_rw32(c_ad.len)) > 0) {
2848 		error = udf_append_adslot(udf_node, &slot, &c_ad);
2849 		if (error)
2850 			goto errorout;
2851 		slot++;
2852 	}
2853 
2854 	/* if there is a rest piece that didn't fit, append it */
2855 	if (UDF_EXT_LEN(udf_rw32(s_ad.len)) > 0) {
2856 		error = udf_append_adslot(udf_node, &slot, &s_ad);
2857 		if (error)
2858 			goto errorout;
2859 		slot++;
2860 	}
2861 
2862 	inflen  += size_diff;
2863 	objsize += size_diff;
2864 	if (fe) {
2865 		fe->inf_len   = udf_rw64(inflen);
2866 	} else {
2867 		efe->inf_len  = udf_rw64(inflen);
2868 		efe->obj_size = udf_rw64(objsize);
2869 	}
2870 	error = 0;
2871 
2872 	if (evacuated_data) {
2873 		/* set new write size for uvm */
2874 		uvm_vnp_setwritesize(vp, old_size);
2875 
2876 		/* write out evacuated data */
2877 		error = vn_rdwr(UIO_WRITE, udf_node->vnode,
2878 				evacuated_data, old_size, 0,
2879 				UIO_SYSSPACE, IO_ALTSEMANTICS | IO_NODELOCKED,
2880 				FSCRED, NULL, NULL);
2881 		uvm_vnp_setsize(vp, old_size);
2882 	}
2883 
2884 errorout:
2885 	if (evacuated_data)
2886 		free(evacuated_data, M_UDFTEMP);
2887 
2888 	udf_count_alloc_exts(udf_node);
2889 
2890 	udf_node_sanity_check(udf_node, &new_inflen, &new_lbrec);
2891 	UDF_UNLOCK_NODE(udf_node, 0);
2892 
2893 	KASSERT(new_inflen == orig_inflen + size_diff);
2894 	KASSERT(new_lbrec == orig_lbrec);
2895 
2896 	return error;
2897 }
2898 
2899 /* --------------------------------------------------------------------- */
2900 
2901 int
2902 udf_shrink_node(struct udf_node *udf_node, uint64_t new_size)
2903 {
2904 	struct vnode *vp = udf_node->vnode;
2905 	struct udf_mount *ump = udf_node->ump;
2906 	struct file_entry    *fe;
2907 	struct extfile_entry *efe;
2908 	struct icb_tag  *icbtag;
2909 	struct long_ad c_ad, s_ad, *node_ad_cpy;
2910 	uint64_t size_diff, old_size, inflen, objsize;
2911 	uint64_t foffset, end_foffset;
2912 	uint64_t orig_inflen, orig_lbrec, new_inflen, new_lbrec;
2913 	uint32_t lb_size, unit_size, dscr_size, crclen;
2914 	uint32_t slot_offset, slot_offset_lb;
2915 	uint32_t len, flags, max_len;
2916 	uint32_t num_lb, lb_num;
2917 	uint32_t max_l_ad, l_ad, l_ea;
2918 	uint16_t vpart_num;
2919 	uint8_t *data_pos;
2920 	int icbflags, addr_type;
2921 	int slot, cpy_slot, cpy_slots;
2922 	int eof, error;
2923 
2924 	DPRINTF(ALLOC, ("udf_shrink_node\n"));
2925 
2926 	UDF_LOCK_NODE(udf_node, 0);
2927 	udf_node_sanity_check(udf_node, &orig_inflen, &orig_lbrec);
2928 
2929 	lb_size = udf_rw32(ump->logical_vol->lb_size);
2930 
2931 	/* max_len in unit's IFF its a metadata node or metadata mirror node */
2932 	unit_size = lb_size;
2933 	if ((udf_node == ump->metadata_node) || (udf_node == ump->metadatamirror_node))
2934 		unit_size = ump->metadata_alloc_unit_size * lb_size;
2935 	max_len = ((UDF_EXT_MAXLEN / unit_size) * unit_size);
2936 
2937 	/* do the work */
2938 	fe  = udf_node->fe;
2939 	efe = udf_node->efe;
2940 	if (fe) {
2941 		icbtag  = &fe->icbtag;
2942 		inflen  = udf_rw64(fe->inf_len);
2943 		objsize = inflen;
2944 		dscr_size  = sizeof(struct file_entry) -1;
2945 		l_ea       = udf_rw32(fe->l_ea);
2946 		l_ad       = udf_rw32(fe->l_ad);
2947 		data_pos = (uint8_t *) fe + dscr_size + l_ea;
2948 	} else {
2949 		icbtag  = &efe->icbtag;
2950 		inflen  = udf_rw64(efe->inf_len);
2951 		objsize = udf_rw64(efe->obj_size);
2952 		dscr_size  = sizeof(struct extfile_entry) -1;
2953 		l_ea       = udf_rw32(efe->l_ea);
2954 		l_ad       = udf_rw32(efe->l_ad);
2955 		data_pos = (uint8_t *) efe + dscr_size + l_ea;
2956 	}
2957 	max_l_ad = lb_size - dscr_size - l_ea;
2958 
2959 	icbflags   = udf_rw16(icbtag->flags);
2960 	addr_type  = icbflags & UDF_ICB_TAG_FLAGS_ALLOC_MASK;
2961 
2962 	old_size  = inflen;
2963 	size_diff = old_size - new_size;
2964 
2965 	DPRINTF(ALLOC, ("\tfrom %"PRIu64" to %"PRIu64"\n", old_size, new_size));
2966 
2967 	/* shrink the node to its new size */
2968 	if (addr_type == UDF_ICB_INTERN_ALLOC) {
2969 		/* only reflect size change directly in the node */
2970 		KASSERT(new_size <= max_l_ad);
2971 		inflen  -= size_diff;
2972 		objsize -= size_diff;
2973 		l_ad    -= size_diff;
2974 		crclen = dscr_size - UDF_DESC_TAG_LENGTH + l_ea + l_ad;
2975 		if (fe) {
2976 			fe->inf_len   = udf_rw64(inflen);
2977 			fe->l_ad      = udf_rw32(l_ad);
2978 			fe->tag.desc_crc_len = udf_rw16(crclen);
2979 		} else {
2980 			efe->inf_len  = udf_rw64(inflen);
2981 			efe->obj_size = udf_rw64(objsize);
2982 			efe->l_ad     = udf_rw32(l_ad);
2983 			efe->tag.desc_crc_len = udf_rw16(crclen);
2984 		}
2985 		error = 0;
2986 
2987 		/* clear the space in the descriptor */
2988 		KASSERT(old_size > new_size);
2989 		memset(data_pos + new_size, 0, old_size - new_size);
2990 
2991 		/* TODO zero appened space in buffer! */
2992 		/* using ubc_zerorange(&vp->v_uobj, old_size, */
2993 		/*    old_size - new_size, UBC_UNMAP_FLAG(vp)); ? */
2994 
2995 		/* set new size for uvm */
2996 		uvm_vnp_setsize(vp, new_size);
2997 
2998 		udf_node_sanity_check(udf_node, &new_inflen, &new_lbrec);
2999 		UDF_UNLOCK_NODE(udf_node, 0);
3000 
3001 		KASSERT(new_inflen == orig_inflen - size_diff);
3002 		KASSERT(new_lbrec == orig_lbrec);
3003 		KASSERT(new_lbrec == 0);
3004 
3005 		return 0;
3006 	}
3007 
3008 	/* setup node cleanup extents copy space */
3009 	node_ad_cpy = malloc(lb_size * UDF_MAX_ALLOC_EXTENTS,
3010 		M_UDFMNT, M_WAITOK);
3011 	memset(node_ad_cpy, 0, lb_size * UDF_MAX_ALLOC_EXTENTS);
3012 
3013 	/*
3014 	 * Shrink the node by releasing the allocations and truncate the last
3015 	 * allocation to the new size. If the new size fits into the
3016 	 * allocation descriptor itself, transform it into an
3017 	 * UDF_ICB_INTERN_ALLOC.
3018 	 */
3019 	slot     = 0;
3020 	cpy_slot = 0;
3021 	foffset  = 0;
3022 
3023 	/* 1) copy till first overlap piece to the rewrite buffer */
3024 	for (;;) {
3025 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
3026 		if (eof) {
3027 			DPRINTF(WRITE,
3028 				("Shrink node failed: "
3029 				 "encountered EOF\n"));
3030 			error = EINVAL;
3031 			goto errorout; /* panic? */
3032 		}
3033 		len   = udf_rw32(s_ad.len);
3034 		flags = UDF_EXT_FLAGS(len);
3035 		len   = UDF_EXT_LEN(len);
3036 
3037 		if (flags == UDF_EXT_REDIRECT) {
3038 			slot++;
3039 			continue;
3040 		}
3041 
3042 		end_foffset = foffset + len;
3043 		if (end_foffset > new_size)
3044 			break;	/* found */
3045 
3046 		node_ad_cpy[cpy_slot++] = s_ad;
3047 
3048 		DPRINTF(ALLOC, ("\t1: vp %d, lb %d, len %d, flags %d "
3049 			"-> stack\n",
3050 			udf_rw16(s_ad.loc.part_num),
3051 			udf_rw32(s_ad.loc.lb_num),
3052 			UDF_EXT_LEN(udf_rw32(s_ad.len)),
3053 			UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
3054 
3055 		foffset = end_foffset;
3056 		slot++;
3057 	}
3058 	slot_offset = new_size - foffset;
3059 
3060 	/* 2) trunc overlapping slot at overlap and copy it */
3061 	if (slot_offset > 0) {
3062 		lb_num    = udf_rw32(s_ad.loc.lb_num);
3063 		vpart_num = udf_rw16(s_ad.loc.part_num);
3064 
3065 		if (flags == UDF_EXT_ALLOCATED) {
3066 			/* calculate extent in lb, and offset in lb */
3067 			num_lb = (len + lb_size -1) / lb_size;
3068 			slot_offset_lb = (slot_offset + lb_size -1) / lb_size;
3069 
3070 			/* adjust our slot */
3071 			lb_num += slot_offset_lb;
3072 			num_lb -= slot_offset_lb;
3073 
3074 			udf_free_allocated_space(ump, lb_num, vpart_num, num_lb);
3075 		}
3076 
3077 		s_ad.len = udf_rw32(slot_offset | flags);
3078 		node_ad_cpy[cpy_slot++] = s_ad;
3079 		slot++;
3080 
3081 		DPRINTF(ALLOC, ("\t2: vp %d, lb %d, len %d, flags %d "
3082 			"-> stack\n",
3083 			udf_rw16(s_ad.loc.part_num),
3084 			udf_rw32(s_ad.loc.lb_num),
3085 			UDF_EXT_LEN(udf_rw32(s_ad.len)),
3086 			UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
3087 	}
3088 
3089 	/* 3) delete remainder */
3090 	for (;;) {
3091 		udf_get_adslot(udf_node, slot, &s_ad, &eof);
3092 		if (eof)
3093 			break;
3094 
3095 		len       = udf_rw32(s_ad.len);
3096 		flags     = UDF_EXT_FLAGS(len);
3097 		len       = UDF_EXT_LEN(len);
3098 
3099 		if (flags == UDF_EXT_REDIRECT) {
3100 			slot++;
3101 			continue;
3102 		}
3103 
3104 		DPRINTF(ALLOC, ("\t3: delete remainder "
3105 			"vp %d lb %d, len %d, flags %d\n",
3106 		udf_rw16(s_ad.loc.part_num),
3107 		udf_rw32(s_ad.loc.lb_num),
3108 		UDF_EXT_LEN(udf_rw32(s_ad.len)),
3109 		UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
3110 
3111 		if (flags == UDF_EXT_ALLOCATED) {
3112 			lb_num    = udf_rw32(s_ad.loc.lb_num);
3113 			vpart_num = udf_rw16(s_ad.loc.part_num);
3114 			num_lb    = (len + lb_size - 1) / lb_size;
3115 
3116 			udf_free_allocated_space(ump, lb_num, vpart_num,
3117 				num_lb);
3118 		}
3119 
3120 		slot++;
3121 	}
3122 
3123 	/* 4) if it will fit into the descriptor then convert */
3124 	if (new_size < max_l_ad) {
3125 		/*
3126 		 * resque/evacuate old piece by reading it in, and convert it
3127 		 * to internal alloc.
3128 		 */
3129 		if (new_size == 0) {
3130 			/* XXX/TODO only for zero sizing now */
3131 			udf_wipe_adslots(udf_node);
3132 
3133 			icbflags &= ~UDF_ICB_TAG_FLAGS_ALLOC_MASK;
3134 			icbflags |=  UDF_ICB_INTERN_ALLOC;
3135 			icbtag->flags = udf_rw16(icbflags);
3136 
3137 			inflen  -= size_diff;	KASSERT(inflen == 0);
3138 			objsize -= size_diff;
3139 			l_ad     = new_size;
3140 			crclen = dscr_size - UDF_DESC_TAG_LENGTH + l_ea + l_ad;
3141 			if (fe) {
3142 				fe->inf_len   = udf_rw64(inflen);
3143 				fe->l_ad      = udf_rw32(l_ad);
3144 				fe->tag.desc_crc_len = udf_rw16(crclen);
3145 			} else {
3146 				efe->inf_len  = udf_rw64(inflen);
3147 				efe->obj_size = udf_rw64(objsize);
3148 				efe->l_ad     = udf_rw32(l_ad);
3149 				efe->tag.desc_crc_len = udf_rw16(crclen);
3150 			}
3151 			/* eventually copy in evacuated piece */
3152 			/* set new size for uvm */
3153 			uvm_vnp_setsize(vp, new_size);
3154 
3155 			free(node_ad_cpy, M_UDFMNT);
3156 			udf_node_sanity_check(udf_node, &new_inflen, &new_lbrec);
3157 
3158 			UDF_UNLOCK_NODE(udf_node, 0);
3159 
3160 			KASSERT(new_inflen == orig_inflen - size_diff);
3161 			KASSERT(new_inflen == 0);
3162 			KASSERT(new_lbrec == 0);
3163 
3164 			return 0;
3165 		}
3166 
3167 		printf("UDF_SHRINK_NODE: could convert to internal alloc!\n");
3168 	}
3169 
3170 	/* 5) reset node descriptors */
3171 	udf_wipe_adslots(udf_node);
3172 
3173 	/* 6) copy back extents; merge when possible. Recounting on the fly */
3174 	cpy_slots = cpy_slot;
3175 
3176 	c_ad = node_ad_cpy[0];
3177 	slot = 0;
3178 	for (cpy_slot = 1; cpy_slot < cpy_slots; cpy_slot++) {
3179 		s_ad = node_ad_cpy[cpy_slot];
3180 
3181 		DPRINTF(ALLOC, ("\t6: stack -> got mapping vp %d "
3182 			"lb %d, len %d, flags %d\n",
3183 		udf_rw16(s_ad.loc.part_num),
3184 		udf_rw32(s_ad.loc.lb_num),
3185 		UDF_EXT_LEN(udf_rw32(s_ad.len)),
3186 		UDF_EXT_FLAGS(udf_rw32(s_ad.len)) >> 30));
3187 
3188 		/* see if we can merge */
3189 		if (udf_ads_merge(max_len, lb_size, &c_ad, &s_ad)) {
3190 			/* not mergable (anymore) */
3191 			DPRINTF(ALLOC, ("\t6: appending vp %d lb %d, "
3192 				"len %d, flags %d\n",
3193 			udf_rw16(c_ad.loc.part_num),
3194 			udf_rw32(c_ad.loc.lb_num),
3195 			UDF_EXT_LEN(udf_rw32(c_ad.len)),
3196 			UDF_EXT_FLAGS(udf_rw32(c_ad.len)) >> 30));
3197 
3198 			error = udf_append_adslot(udf_node, &slot, &c_ad);
3199 			if (error)
3200 				goto errorout; /* panic? */
3201 			c_ad = s_ad;
3202 			slot++;
3203 		}
3204 	}
3205 
3206 	/* 7) push rest slot (if any) */
3207 	if (UDF_EXT_LEN(c_ad.len) > 0) {
3208 		DPRINTF(ALLOC, ("\t7: last append vp %d lb %d, "
3209 				"len %d, flags %d\n",
3210 		udf_rw16(c_ad.loc.part_num),
3211 		udf_rw32(c_ad.loc.lb_num),
3212 		UDF_EXT_LEN(udf_rw32(c_ad.len)),
3213 		UDF_EXT_FLAGS(udf_rw32(c_ad.len)) >> 30));
3214 
3215 		error = udf_append_adslot(udf_node, &slot, &c_ad);
3216 		if (error)
3217 			goto errorout; /* panic? */
3218 		;
3219 	}
3220 
3221 	inflen  -= size_diff;
3222 	objsize -= size_diff;
3223 	if (fe) {
3224 		fe->inf_len   = udf_rw64(inflen);
3225 	} else {
3226 		efe->inf_len  = udf_rw64(inflen);
3227 		efe->obj_size = udf_rw64(objsize);
3228 	}
3229 	error = 0;
3230 
3231 	/* set new size for uvm */
3232 	uvm_vnp_setsize(vp, new_size);
3233 
3234 errorout:
3235 	free(node_ad_cpy, M_UDFMNT);
3236 
3237 	udf_count_alloc_exts(udf_node);
3238 
3239 	udf_node_sanity_check(udf_node, &new_inflen, &new_lbrec);
3240 	UDF_UNLOCK_NODE(udf_node, 0);
3241 
3242 	KASSERT(new_inflen == orig_inflen - size_diff);
3243 
3244 	return error;
3245 }
3246 
3247