xref: /netbsd-src/sys/fs/udf/udf_strat_sequential.c (revision fff57c5525bbe431aee7bdb3983954f0627a42cb)
1 /* $NetBSD: udf_strat_sequential.c,v 1.1 2008/05/14 16:49:48 reinoud 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_strat_sequential.c,v 1.1 2008/05/14 16:49:48 reinoud Exp $");
32 #endif /* not lint */
33 
34 
35 #if defined(_KERNEL_OPT)
36 #include "opt_quota.h"
37 #include "opt_compat_netbsd.h"
38 #endif
39 
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 #if defined(_KERNEL_OPT)
66 #include "opt_udf.h"
67 #endif
68 
69 #include "udf.h"
70 #include "udf_subr.h"
71 #include "udf_bswap.h"
72 
73 
74 #define VTOI(vnode) ((struct udf_node *) vnode->v_data)
75 #define PRIV(ump) ((struct strat_private *) ump->strategy_private)
76 
77 /* --------------------------------------------------------------------- */
78 
79 /* BUFQ's */
80 #define UDF_SHED_MAX 3
81 
82 #define UDF_SHED_READING	0
83 #define UDF_SHED_WRITING	1
84 #define UDF_SHED_SEQWRITING	2
85 
86 struct strat_private {
87 	struct pool		 desc_pool;	 	/* node descriptors */
88 
89 	lwp_t			*queue_lwp;
90 	kcondvar_t		 discstrat_cv;		/* to wait on       */
91 	kmutex_t		 discstrat_mutex;	/* disc strategy    */
92 
93 	int			 run_thread;		/* thread control */
94 	int			 cur_queue;
95 
96 	struct disk_strategy	 old_strategy_setting;
97 	struct bufq_state	*queues[UDF_SHED_MAX];
98 	struct timespec		 last_queued[UDF_SHED_MAX];
99 };
100 
101 
102 /* --------------------------------------------------------------------- */
103 
104 static void
105 udf_wr_nodedscr_callback(struct buf *buf)
106 {
107 	struct udf_node *udf_node;
108 
109 	KASSERT(buf);
110 	KASSERT(buf->b_data);
111 
112 	/* called when write action is done */
113 	DPRINTF(WRITE, ("udf_wr_nodedscr_callback(): node written out\n"));
114 
115 	udf_node = VTOI(buf->b_vp);
116 	if (udf_node == NULL) {
117 		putiobuf(buf);
118 		printf("udf_wr_node_callback: NULL node?\n");
119 		return;
120 	}
121 
122 	/* XXX noone is waiting on this outstanding_nodedscr */
123 	udf_node->outstanding_nodedscr--;
124 	if (udf_node->outstanding_nodedscr == 0)
125 		wakeup(&udf_node->outstanding_nodedscr);
126 
127 	/* XXX right flags to mark dirty again on error? */
128 	if (buf->b_error) {
129 		udf_node->i_flags |= IN_MODIFIED | IN_ACCESSED;
130 		/* XXX TODO reshedule on error */
131 	}
132 
133 	/* first unlock the node */
134 	KASSERT(udf_node->i_flags & IN_CALLBACK_ULK);
135 	UDF_UNLOCK_NODE(udf_node, IN_CALLBACK_ULK);
136 
137 	/* unreference the vnode so it can be recycled */
138 	holdrele(udf_node->vnode);
139 
140 	putiobuf(buf);
141 }
142 
143 /* --------------------------------------------------------------------- */
144 
145 static int
146 udf_create_logvol_dscr_seq(struct udf_strat_args *args)
147 {
148 	union dscrptr   **dscrptr = &args->dscr;
149 	struct udf_mount *ump = args->ump;
150 	struct strat_private *priv = PRIV(ump);
151 	uint32_t lb_size;
152 
153 	lb_size = udf_rw32(ump->logical_vol->lb_size);
154 	*dscrptr = pool_get(&priv->desc_pool, PR_WAITOK);
155 	memset(*dscrptr, 0, lb_size);
156 
157 	return 0;
158 }
159 
160 
161 static void
162 udf_free_logvol_dscr_seq(struct udf_strat_args *args)
163 {
164 	union dscrptr    *dscr = args->dscr;
165 	struct udf_mount *ump  = args->ump;
166 	struct strat_private *priv = PRIV(ump);
167 
168 	pool_put(&priv->desc_pool, dscr);
169 }
170 
171 
172 static int
173 udf_read_logvol_dscr_seq(struct udf_strat_args *args)
174 {
175 	union dscrptr   **dscrptr = &args->dscr;
176 	union dscrptr    *tmpdscr;
177 	struct udf_mount *ump = args->ump;
178 	struct long_ad   *icb = args->icb;
179 	struct strat_private *priv = PRIV(ump);
180 	uint32_t lb_size;
181 	uint32_t sector, dummy;
182 	int error;
183 
184 	lb_size = udf_rw32(ump->logical_vol->lb_size);
185 
186 	error = udf_translate_vtop(ump, icb, &sector, &dummy);
187 	if (error)
188 		return error;
189 
190 	/* try to read in fe/efe */
191 	error = udf_read_phys_dscr(ump, sector, M_UDFTEMP, &tmpdscr);
192 	if (error)
193 		return error;
194 
195 	*dscrptr = pool_get(&priv->desc_pool, PR_WAITOK);
196 	memcpy(*dscrptr, tmpdscr, lb_size);
197 	free(tmpdscr, M_UDFTEMP);
198 
199 	return 0;
200 }
201 
202 
203 static int
204 udf_write_logvol_dscr_seq(struct udf_strat_args *args)
205 {
206 	union dscrptr    *dscr     = args->dscr;
207 	struct udf_mount *ump      = args->ump;
208 	struct udf_node  *udf_node = args->udf_node;
209 	struct long_ad   *icb      = args->icb;
210 	int               waitfor  = args->waitfor;
211 	uint32_t logsectornr, sectornr, dummy;
212 	int error, vpart;
213 
214 	/*
215 	 * we have to decide if we write it out sequential or at its fixed
216 	 * position by examining the partition its (to be) written on.
217 	 */
218 	vpart       = udf_rw16(udf_node->loc.loc.part_num);
219 	logsectornr = udf_rw32(icb->loc.lb_num);
220 	sectornr    = 0;
221 	if (ump->vtop_tp[vpart] != UDF_VTOP_TYPE_VIRT) {
222 		error = udf_translate_vtop(ump, icb, &sectornr, &dummy);
223 		if (error)
224 			return error;
225 	}
226 
227 	UDF_LOCK_NODE(udf_node, IN_CALLBACK_ULK);
228 
229 	if (waitfor) {
230 		DPRINTF(WRITE, ("udf_write_logvol_dscr: sync write\n"));
231 
232 		error = udf_write_phys_dscr_sync(ump, udf_node, UDF_C_NODE,
233 			dscr, sectornr, logsectornr);
234 		UDF_UNLOCK_NODE(udf_node, IN_CALLBACK_ULK);
235 	} else {
236 		DPRINTF(WRITE, ("udf_write_logvol_dscr: no wait, async write\n"));
237 
238 		/* add reference to the vnode to prevent recycling */
239 		vhold(udf_node->vnode);
240 
241 		udf_node->outstanding_nodedscr++;
242 
243 		error = udf_write_phys_dscr_async(ump, udf_node, UDF_C_NODE,
244 			dscr, sectornr, logsectornr, udf_wr_nodedscr_callback);
245 		/* will be UNLOCKED in call back */
246 	}
247 
248 	return error;
249 }
250 
251 /* --------------------------------------------------------------------- */
252 
253 /*
254  * Main file-system specific sheduler. Due to the nature of optical media
255  * sheduling can't be performed in the traditional way. Most OS
256  * implementations i've seen thus read or write a file atomically giving all
257  * kinds of side effects.
258  *
259  * This implementation uses a kernel thread to shedule the queued requests in
260  * such a way that is semi-optimal for optical media; this means aproximately
261  * (R*|(Wr*|Ws*))* since switching between reading and writing is expensive in
262  * time.
263  */
264 
265 static void
266 udf_queuebuf_seq(struct udf_strat_args *args)
267 {
268 	struct udf_mount *ump = args->ump;
269 	struct buf *nestbuf = args->nestbuf;
270 	struct strat_private *priv = PRIV(ump);
271 	int queue;
272 	int what;
273 
274 	KASSERT(ump);
275 	KASSERT(nestbuf);
276 	KASSERT(nestbuf->b_iodone == nestiobuf_iodone);
277 
278 	what = nestbuf->b_udf_c_type;
279 	queue = UDF_SHED_READING;
280 	if ((nestbuf->b_flags & B_READ) == 0) {
281 		/* writing */
282 		queue = UDF_SHED_SEQWRITING;
283 		if (what == UDF_C_DSCR)
284 			queue = UDF_SHED_WRITING;
285 		if (what == UDF_C_NODE) {
286 			if (ump->meta_alloc != UDF_ALLOC_VAT)
287 				queue = UDF_SHED_WRITING;
288 		}
289 #if 0
290 		if (queue == UDF_SHED_SEQWRITING) {
291 			/* TODO do add sector to uncommitted space */
292 		}
293 #endif
294 	}
295 
296 	/* use our own sheduler lists for more complex sheduling */
297 	mutex_enter(&priv->discstrat_mutex);
298 		BUFQ_PUT(priv->queues[queue], nestbuf);
299 		vfs_timestamp(&priv->last_queued[queue]);
300 	mutex_exit(&priv->discstrat_mutex);
301 
302 	/* signal our thread that there might be something to do */
303 	cv_signal(&priv->discstrat_cv);
304 }
305 
306 /* --------------------------------------------------------------------- */
307 
308 /* TODO convert to lb_size */
309 static void
310 udf_VAT_mapping_update(struct udf_mount *ump, struct buf *buf)
311 {
312 	union dscrptr    *fdscr = (union dscrptr *) buf->b_data;
313 	struct vnode     *vp = buf->b_vp;
314 	struct udf_node  *udf_node = VTOI(vp);
315 	struct part_desc *pdesc;
316 	uint32_t lb_size, blks;
317 	uint32_t lb_num, lb_map;
318 	uint32_t udf_rw32_lbmap;
319 	int c_type = buf->b_udf_c_type;
320 	int error;
321 
322 	/* only interested when we're using a VAT */
323 	if (ump->meta_alloc != UDF_ALLOC_VAT)
324 		return;
325 	KASSERT(ump->vat_node);
326 
327 	/* only nodes are recorded in the VAT */
328 	/* NOTE: and the fileset descriptor (FIXME ?) */
329 	if (c_type != UDF_C_NODE)
330 		return;
331 
332 	/* we now have an UDF FE/EFE node on media with VAT (or VAT itself) */
333 	lb_size = udf_rw32(ump->logical_vol->lb_size);
334 	blks = lb_size / DEV_BSIZE;
335 
336 	/* calculate offset from base partition */
337 	pdesc = ump->partitions[ump->vtop[ump->metadata_part]];
338 	lb_map  = buf->b_blkno / blks;
339 	lb_map -= udf_rw32(pdesc->start_loc);
340 
341 	udf_rw32_lbmap = udf_rw32(lb_map);
342 
343 	/* if we're the VAT itself, only update our assigned sector number */
344 	if (udf_node == ump->vat_node) {
345 		fdscr->tag.tag_loc = udf_rw32_lbmap;
346 		udf_validate_tag_sum(fdscr);
347 		DPRINTF(TRANSLATE, ("VAT assigned to sector %u\n",
348 			udf_rw32(udf_rw32_lbmap)));
349 		/* no use mapping the VAT node in the VAT */
350 		return;
351 	}
352 
353 	/* check for tag location is false for allocation extents */
354 	KASSERT(fdscr->tag.tag_loc == udf_node->write_loc.loc.lb_num);
355 
356 	/* record new position in VAT file */
357 	lb_num = udf_rw32(udf_node->write_loc.loc.lb_num);
358 
359 	DPRINTF(TRANSLATE, ("VAT entry change (log %u -> phys %u)\n",
360 			lb_num, lb_map));
361 
362 	/* VAT should be the longer than this write, can't go wrong */
363 	KASSERT(lb_num <= ump->vat_entries);
364 
365 	mutex_enter(&ump->allocate_mutex);
366 	error = udf_vat_write(ump->vat_node,
367 			(uint8_t *) &udf_rw32_lbmap, 4,
368 			ump->vat_offset + lb_num * 4);
369 	mutex_exit(&ump->allocate_mutex);
370 
371 	if (error)
372 		panic( "udf_VAT_mapping_update: HELP! i couldn't "
373 			"write in the VAT file ?\n");
374 }
375 
376 
377 static void
378 udf_issue_buf(struct udf_mount *ump, int queue, struct buf *buf)
379 {
380 	struct long_ad *node_ad_cpy;
381 	uint64_t *lmapping, *pmapping, *lmappos, blknr;
382 	uint32_t our_sectornr, sectornr, bpos;
383 	uint8_t *fidblk;
384 	int sector_size = ump->discinfo.sector_size;
385 	int blks = sector_size / DEV_BSIZE;
386 	int len, buf_len;
387 
388 	/* if reading, just pass to the device's STRATEGY */
389 	if (queue == UDF_SHED_READING) {
390 		DPRINTF(SHEDULE, ("\nudf_issue_buf READ %p : sector %d type %d,"
391 			"b_resid %d, b_bcount %d, b_bufsize %d\n",
392 			buf, (uint32_t) buf->b_blkno / blks, buf->b_udf_c_type,
393 			buf->b_resid, buf->b_bcount, buf->b_bufsize));
394 		VOP_STRATEGY(ump->devvp, buf);
395 		return;
396 	}
397 
398 	blknr        = buf->b_blkno;
399 	our_sectornr = blknr / blks;
400 
401 	if (queue == UDF_SHED_WRITING) {
402 		DPRINTF(SHEDULE, ("\nudf_issue_buf WRITE %p : sector %d "
403 			"type %d, b_resid %d, b_bcount %d, b_bufsize %d\n",
404 			buf, (uint32_t) buf->b_blkno / blks, buf->b_udf_c_type,
405 			buf->b_resid, buf->b_bcount, buf->b_bufsize));
406 		/* if we have FIDs fixup using buffer's sector number(s) */
407 		if (buf->b_udf_c_type == UDF_C_FIDS) {
408 			panic("UDF_C_FIDS in SHED_WRITING!\n");
409 			buf_len = buf->b_bcount;
410 			sectornr = our_sectornr;
411 			bpos = 0;
412 			while (buf_len) {
413 				len = MIN(buf_len, sector_size);
414 				fidblk = (uint8_t *) buf->b_data + bpos;
415 				udf_fixup_fid_block(fidblk, sector_size,
416 					0, len, sectornr);
417 				sectornr++;
418 				bpos += len;
419 				buf_len -= len;
420 			}
421 		}
422 		udf_fixup_node_internals(ump, buf->b_data, buf->b_udf_c_type);
423 		VOP_STRATEGY(ump->devvp, buf);
424 		return;
425 	}
426 
427 	KASSERT(queue == UDF_SHED_SEQWRITING);
428 	DPRINTF(SHEDULE, ("\nudf_issue_buf SEQWRITE %p : sector XXXX "
429 		"type %d, b_resid %d, b_bcount %d, b_bufsize %d\n",
430 		buf, buf->b_udf_c_type, buf->b_resid, buf->b_bcount,
431 		buf->b_bufsize));
432 
433 	/*
434 	 * Buffers should not have been allocated to disc addresses yet on
435 	 * this queue. Note that a buffer can get multiple extents allocated.
436 	 *
437 	 * lmapping contains lb_num relative to base partition.
438 	 * pmapping contains lb_num as used for disc adressing.
439 	 */
440 	lmapping    = ump->la_lmapping;
441 	pmapping    = ump->la_pmapping;
442 	node_ad_cpy = ump->la_node_ad_cpy;
443 
444 	/* allocate buf and get its logical and physical mappings */
445 	udf_late_allocate_buf(ump, buf, lmapping, pmapping, node_ad_cpy);
446 	udf_VAT_mapping_update(ump, buf);	/* XXX could pass *lmapping */
447 
448 	/* if we have FIDs, fixup using the new allocation table */
449 	if (buf->b_udf_c_type == UDF_C_FIDS) {
450 		buf_len = buf->b_bcount;
451 		bpos = 0;
452 		lmappos = lmapping;
453 		while (buf_len) {
454 			sectornr = *lmappos++;
455 			len = MIN(buf_len, sector_size);
456 			fidblk = (uint8_t *) buf->b_data + bpos;
457 			udf_fixup_fid_block(fidblk, sector_size,
458 				0, len, sectornr);
459 			bpos += len;
460 			buf_len -= len;
461 		}
462 	}
463 	udf_fixup_node_internals(ump, buf->b_data, buf->b_udf_c_type);
464 	VOP_STRATEGY(ump->devvp, buf);
465 }
466 
467 
468 static void
469 udf_doshedule(struct udf_mount *ump)
470 {
471 	struct buf *buf;
472 	struct timespec now, *last;
473 	struct strat_private *priv = PRIV(ump);
474 	void (*b_callback)(struct buf *);
475 	int new_queue;
476 	int error;
477 
478 	buf = BUFQ_GET(priv->queues[priv->cur_queue]);
479 	if (buf) {
480 		/* transfer from the current queue to the device queue */
481 		mutex_exit(&priv->discstrat_mutex);
482 
483 		/* transform buffer to synchronous; XXX needed? */
484 		b_callback = buf->b_iodone;
485 		buf->b_iodone = NULL;
486 		CLR(buf->b_flags, B_ASYNC);
487 
488 		/* issue and wait on completion */
489 		udf_issue_buf(ump, priv->cur_queue, buf);
490 		biowait(buf);
491 
492 		mutex_enter(&priv->discstrat_mutex);
493 
494 		/* if there is an error, repair this error, otherwise propagate */
495 		if (buf->b_error && ((buf->b_flags & B_READ) == 0)) {
496 			/* check what we need to do */
497 			panic("UDF write error, can't handle yet!\n");
498 		}
499 
500 		/* propagate result to higher layers */
501 		if (b_callback) {
502 			buf->b_iodone = b_callback;
503 			(*buf->b_iodone)(buf);
504 		}
505 
506 		return;
507 	}
508 
509 	/* Check if we're idling in this state */
510 	vfs_timestamp(&now);
511 	last = &priv->last_queued[priv->cur_queue];
512 	if (ump->discinfo.mmc_class == MMC_CLASS_CD) {
513 		/* dont switch too fast for CD media; its expensive in time */
514 		if (now.tv_sec - last->tv_sec < 3)
515 			return;
516 	}
517 
518 	/* check if we can/should switch */
519 	new_queue = priv->cur_queue;
520 
521 	if (BUFQ_PEEK(priv->queues[UDF_SHED_READING]))
522 		new_queue = UDF_SHED_READING;
523 	if (BUFQ_PEEK(priv->queues[UDF_SHED_SEQWRITING]))
524 		new_queue = UDF_SHED_SEQWRITING;
525 	if (BUFQ_PEEK(priv->queues[UDF_SHED_WRITING]))		/* only for unmount */
526 		new_queue = UDF_SHED_WRITING;
527 	if (priv->cur_queue == UDF_SHED_READING) {
528 		if (new_queue == UDF_SHED_SEQWRITING) {
529 			/* TODO use flag to signal if this is needed */
530 			mutex_exit(&priv->discstrat_mutex);
531 
532 			/* update trackinfo for data and metadata */
533 			error = udf_update_trackinfo(ump,
534 					&ump->data_track);
535 			assert(error == 0);
536 			error = udf_update_trackinfo(ump,
537 					&ump->metadata_track);
538 			assert(error == 0);
539 			mutex_enter(&priv->discstrat_mutex);
540 		}
541 	}
542 
543 	if (new_queue != priv->cur_queue) {
544 		DPRINTF(SHEDULE, ("switching from %d to %d\n",
545 			priv->cur_queue, new_queue));
546 	}
547 
548 	priv->cur_queue = new_queue;
549 }
550 
551 
552 static void
553 udf_discstrat_thread(void *arg)
554 {
555 	struct udf_mount *ump = (struct udf_mount *) arg;
556 	struct strat_private *priv = PRIV(ump);
557 	int empty;
558 
559 	empty = 1;
560 	mutex_enter(&priv->discstrat_mutex);
561 	while (priv->run_thread || !empty) {
562 		/* process the current selected queue */
563 		udf_doshedule(ump);
564 		empty  = (BUFQ_PEEK(priv->queues[UDF_SHED_READING]) == NULL);
565 		empty &= (BUFQ_PEEK(priv->queues[UDF_SHED_WRITING]) == NULL);
566 		empty &= (BUFQ_PEEK(priv->queues[UDF_SHED_SEQWRITING]) == NULL);
567 
568 		/* wait for more if needed */
569 		if (empty)
570 			cv_timedwait(&priv->discstrat_cv,
571 				&priv->discstrat_mutex, hz/8);
572 	}
573 	mutex_exit(&priv->discstrat_mutex);
574 
575 	wakeup(&priv->run_thread);
576 	kthread_exit(0);
577 	/* not reached */
578 }
579 
580 /* --------------------------------------------------------------------- */
581 
582 static void
583 udf_discstrat_init_seq(struct udf_strat_args *args)
584 {
585 	struct udf_mount *ump = args->ump;
586 	struct strat_private *priv = PRIV(ump);
587 	struct disk_strategy dkstrat;
588 	uint32_t lb_size;
589 
590 	KASSERT(ump);
591 	KASSERT(ump->logical_vol);
592 	KASSERT(priv == NULL);
593 
594 	lb_size = udf_rw32(ump->logical_vol->lb_size);
595 	KASSERT(lb_size > 0);
596 
597 	/* initialise our memory space */
598 	ump->strategy_private = malloc(sizeof(struct strat_private),
599 		M_UDFTEMP, M_WAITOK);
600 	priv = ump->strategy_private;
601 	memset(priv, 0 , sizeof(struct strat_private));
602 
603 	/* initialise locks */
604 	cv_init(&priv->discstrat_cv, "udfstrat");
605 	mutex_init(&priv->discstrat_mutex, MUTEX_DEFAULT, IPL_NONE);
606 
607 	/*
608 	 * Initialise pool for descriptors associated with nodes. This is done
609 	 * in lb_size units though currently lb_size is dictated to be
610 	 * sector_size.
611 	 */
612 	pool_init(&priv->desc_pool, lb_size, 0, 0, 0, "udf_desc_pool", NULL,
613 	    IPL_NONE);
614 
615 	/*
616 	 * remember old device strategy method and explicit set method
617 	 * `discsort' since we have our own more complex strategy that is not
618 	 * implementable on the CD device and other strategies will get in the
619 	 * way.
620 	 */
621 	memset(&priv->old_strategy_setting, 0,
622 		sizeof(struct disk_strategy));
623 	VOP_IOCTL(ump->devvp, DIOCGSTRATEGY, &priv->old_strategy_setting,
624 		FREAD | FKIOCTL, NOCRED);
625 	memset(&dkstrat, 0, sizeof(struct disk_strategy));
626 	strcpy(dkstrat.dks_name, "discsort");
627 	VOP_IOCTL(ump->devvp, DIOCSSTRATEGY, &dkstrat, FWRITE | FKIOCTL,
628 		NOCRED);
629 
630 	/* initialise our internal sheduler */
631 	priv->cur_queue = UDF_SHED_READING;
632 	bufq_alloc(&priv->queues[UDF_SHED_READING], "disksort",
633 		BUFQ_SORT_RAWBLOCK);
634 	bufq_alloc(&priv->queues[UDF_SHED_WRITING], "disksort",
635 		BUFQ_SORT_RAWBLOCK);
636 	bufq_alloc(&priv->queues[UDF_SHED_SEQWRITING], "fcfs", 0);
637 	vfs_timestamp(&priv->last_queued[UDF_SHED_READING]);
638 	vfs_timestamp(&priv->last_queued[UDF_SHED_WRITING]);
639 	vfs_timestamp(&priv->last_queued[UDF_SHED_SEQWRITING]);
640 
641 	/* create our disk strategy thread */
642 	priv->run_thread = 1;
643 	if (kthread_create(PRI_NONE, 0 /* KTHREAD_MPSAFE*/, NULL /* cpu_info*/,
644 		udf_discstrat_thread, ump, &priv->queue_lwp,
645 		"%s", "udf_rw")) {
646 		panic("fork udf_rw");
647 	}
648 }
649 
650 
651 static void
652 udf_discstrat_finish_seq(struct udf_strat_args *args)
653 {
654 	struct udf_mount *ump = args->ump;
655 	struct strat_private *priv = PRIV(ump);
656 	int error;
657 
658 	if (ump == NULL)
659 		return;
660 
661 	/* stop our sheduling thread */
662 	KASSERT(priv->run_thread == 1);
663 	priv->run_thread = 0;
664 	wakeup(priv->queue_lwp);
665 	do {
666 		error = tsleep(&priv->run_thread, PRIBIO+1,
667 			"udfshedfin", hz);
668 	} while (error);
669 	/* kthread should be finished now */
670 
671 	/* set back old device strategy method */
672 	VOP_IOCTL(ump->devvp, DIOCSSTRATEGY, &priv->old_strategy_setting,
673 			FWRITE, NOCRED);
674 
675 	/* destroy our pool */
676 	pool_destroy(&priv->desc_pool);
677 
678 	/* free our private space */
679 	free(ump->strategy_private, M_UDFTEMP);
680 	ump->strategy_private = NULL;
681 }
682 
683 /* --------------------------------------------------------------------- */
684 
685 struct udf_strategy udf_strat_sequential =
686 {
687 	udf_create_logvol_dscr_seq,
688 	udf_free_logvol_dscr_seq,
689 	udf_read_logvol_dscr_seq,
690 	udf_write_logvol_dscr_seq,
691 	udf_queuebuf_seq,
692 	udf_discstrat_init_seq,
693 	udf_discstrat_finish_seq
694 };
695 
696 
697