Home | History | Annotate | Line # | Download | only in udf
udf_strat_sequential.c revision 1.6
      1 /* $NetBSD: udf_strat_sequential.c,v 1.6 2008/12/16 16:18:25 pooka 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 1.6 2008/12/16 16:18:25 pooka Exp $");
     32 #endif /* not lint */
     33 
     34 
     35 #if defined(_KERNEL_OPT)
     36 #include "opt_compat_netbsd.h"
     37 #endif
     38 
     39 #include <sys/param.h>
     40 #include <sys/systm.h>
     41 #include <sys/sysctl.h>
     42 #include <sys/namei.h>
     43 #include <sys/proc.h>
     44 #include <sys/kernel.h>
     45 #include <sys/vnode.h>
     46 #include <miscfs/genfs/genfs_node.h>
     47 #include <sys/mount.h>
     48 #include <sys/buf.h>
     49 #include <sys/file.h>
     50 #include <sys/device.h>
     51 #include <sys/disklabel.h>
     52 #include <sys/ioctl.h>
     53 #include <sys/malloc.h>
     54 #include <sys/dirent.h>
     55 #include <sys/stat.h>
     56 #include <sys/conf.h>
     57 #include <sys/kauth.h>
     58 #include <sys/kthread.h>
     59 #include <dev/clock_subr.h>
     60 
     61 #include <fs/udf/ecma167-udf.h>
     62 #include <fs/udf/udf_mount.h>
     63 
     64 #include "udf.h"
     65 #include "udf_subr.h"
     66 #include "udf_bswap.h"
     67 
     68 
     69 #define VTOI(vnode) ((struct udf_node *) vnode->v_data)
     70 #define PRIV(ump) ((struct strat_private *) ump->strategy_private)
     71 
     72 /* --------------------------------------------------------------------- */
     73 
     74 /* BUFQ's */
     75 #define UDF_SHED_MAX 3
     76 
     77 #define UDF_SHED_READING	0
     78 #define UDF_SHED_WRITING	1
     79 #define UDF_SHED_SEQWRITING	2
     80 
     81 struct strat_private {
     82 	struct pool		 desc_pool;	 	/* node descriptors */
     83 
     84 	lwp_t			*queue_lwp;
     85 	kcondvar_t		 discstrat_cv;		/* to wait on       */
     86 	kmutex_t		 discstrat_mutex;	/* disc strategy    */
     87 
     88 	int			 run_thread;		/* thread control */
     89 	int			 cur_queue;
     90 
     91 	struct disk_strategy	 old_strategy_setting;
     92 	struct bufq_state	*queues[UDF_SHED_MAX];
     93 	struct timespec		 last_queued[UDF_SHED_MAX];
     94 };
     95 
     96 
     97 /* --------------------------------------------------------------------- */
     98 
     99 static void
    100 udf_wr_nodedscr_callback(struct buf *buf)
    101 {
    102 	struct udf_node *udf_node;
    103 
    104 	KASSERT(buf);
    105 	KASSERT(buf->b_data);
    106 
    107 	/* called when write action is done */
    108 	DPRINTF(WRITE, ("udf_wr_nodedscr_callback(): node written out\n"));
    109 
    110 	udf_node = VTOI(buf->b_vp);
    111 	if (udf_node == NULL) {
    112 		putiobuf(buf);
    113 		printf("udf_wr_node_callback: NULL node?\n");
    114 		return;
    115 	}
    116 
    117 	/* XXX right flags to mark dirty again on error? */
    118 	if (buf->b_error) {
    119 		udf_node->i_flags |= IN_MODIFIED | IN_ACCESSED;
    120 		/* XXX TODO reshedule on error */
    121 	}
    122 
    123 	/* decrement outstanding_nodedscr */
    124 	KASSERT(udf_node->outstanding_nodedscr >= 1);
    125 	udf_node->outstanding_nodedscr--;
    126 	if (udf_node->outstanding_nodedscr == 0) {
    127 		/* first unlock the node */
    128 		KASSERT(udf_node->i_flags & IN_CALLBACK_ULK);
    129 		UDF_UNLOCK_NODE(udf_node, IN_CALLBACK_ULK);
    130 
    131 		wakeup(&udf_node->outstanding_nodedscr);
    132 	}
    133 
    134 	/* unreference the vnode so it can be recycled */
    135 	holdrele(udf_node->vnode);
    136 
    137 	putiobuf(buf);
    138 }
    139 
    140 /* --------------------------------------------------------------------- */
    141 
    142 static int
    143 udf_create_logvol_dscr_seq(struct udf_strat_args *args)
    144 {
    145 	union dscrptr   **dscrptr = &args->dscr;
    146 	struct udf_mount *ump = args->ump;
    147 	struct strat_private *priv = PRIV(ump);
    148 	uint32_t lb_size;
    149 
    150 	lb_size = udf_rw32(ump->logical_vol->lb_size);
    151 	*dscrptr = pool_get(&priv->desc_pool, PR_WAITOK);
    152 	memset(*dscrptr, 0, lb_size);
    153 
    154 	return 0;
    155 }
    156 
    157 
    158 static void
    159 udf_free_logvol_dscr_seq(struct udf_strat_args *args)
    160 {
    161 	union dscrptr    *dscr = args->dscr;
    162 	struct udf_mount *ump  = args->ump;
    163 	struct strat_private *priv = PRIV(ump);
    164 
    165 	pool_put(&priv->desc_pool, dscr);
    166 }
    167 
    168 
    169 static int
    170 udf_read_logvol_dscr_seq(struct udf_strat_args *args)
    171 {
    172 	union dscrptr   **dscrptr = &args->dscr;
    173 	union dscrptr    *tmpdscr;
    174 	struct udf_mount *ump = args->ump;
    175 	struct long_ad   *icb = args->icb;
    176 	struct strat_private *priv = PRIV(ump);
    177 	uint32_t lb_size;
    178 	uint32_t sector, dummy;
    179 	int error;
    180 
    181 	lb_size = udf_rw32(ump->logical_vol->lb_size);
    182 
    183 	error = udf_translate_vtop(ump, icb, &sector, &dummy);
    184 	if (error)
    185 		return error;
    186 
    187 	/* try to read in fe/efe */
    188 	error = udf_read_phys_dscr(ump, sector, M_UDFTEMP, &tmpdscr);
    189 	if (error)
    190 		return error;
    191 
    192 	*dscrptr = pool_get(&priv->desc_pool, PR_WAITOK);
    193 	memcpy(*dscrptr, tmpdscr, lb_size);
    194 	free(tmpdscr, M_UDFTEMP);
    195 
    196 	return 0;
    197 }
    198 
    199 
    200 static int
    201 udf_write_logvol_dscr_seq(struct udf_strat_args *args)
    202 {
    203 	union dscrptr    *dscr     = args->dscr;
    204 	struct udf_mount *ump      = args->ump;
    205 	struct udf_node  *udf_node = args->udf_node;
    206 	struct long_ad   *icb      = args->icb;
    207 	int               waitfor  = args->waitfor;
    208 	uint32_t logsectornr, sectornr, dummy;
    209 	int error, vpart;
    210 
    211 	/*
    212 	 * we have to decide if we write it out sequential or at its fixed
    213 	 * position by examining the partition its (to be) written on.
    214 	 */
    215 	vpart       = udf_rw16(udf_node->loc.loc.part_num);
    216 	logsectornr = udf_rw32(icb->loc.lb_num);
    217 	sectornr    = 0;
    218 	if (ump->vtop_tp[vpart] != UDF_VTOP_TYPE_VIRT) {
    219 		error = udf_translate_vtop(ump, icb, &sectornr, &dummy);
    220 		if (error)
    221 			goto out;
    222 	}
    223 
    224 	/* add reference to the vnode to prevent recycling */
    225 	vhold(udf_node->vnode);
    226 
    227 	if (waitfor) {
    228 		DPRINTF(WRITE, ("udf_write_logvol_dscr: sync write\n"));
    229 
    230 		error = udf_write_phys_dscr_sync(ump, udf_node, UDF_C_NODE,
    231 			dscr, sectornr, logsectornr);
    232 	} else {
    233 		DPRINTF(WRITE, ("udf_write_logvol_dscr: no wait, async write\n"));
    234 
    235 		error = udf_write_phys_dscr_async(ump, udf_node, UDF_C_NODE,
    236 			dscr, sectornr, logsectornr, udf_wr_nodedscr_callback);
    237 		/* will be UNLOCKED in call back */
    238 		return error;
    239 	}
    240 
    241 	holdrele(udf_node->vnode);
    242 out:
    243 	udf_node->outstanding_nodedscr--;
    244 	if (udf_node->outstanding_nodedscr == 0) {
    245 		UDF_UNLOCK_NODE(udf_node, 0);
    246 		wakeup(&udf_node->outstanding_nodedscr);
    247 	}
    248 
    249 	return error;
    250 }
    251 
    252 /* --------------------------------------------------------------------- */
    253 
    254 /*
    255  * Main file-system specific sheduler. Due to the nature of optical media
    256  * sheduling can't be performed in the traditional way. Most OS
    257  * implementations i've seen thus read or write a file atomically giving all
    258  * kinds of side effects.
    259  *
    260  * This implementation uses a kernel thread to shedule the queued requests in
    261  * such a way that is semi-optimal for optical media; this means aproximately
    262  * (R*|(Wr*|Ws*))* since switching between reading and writing is expensive in
    263  * time.
    264  */
    265 
    266 static void
    267 udf_queuebuf_seq(struct udf_strat_args *args)
    268 {
    269 	struct udf_mount *ump = args->ump;
    270 	struct buf *nestbuf = args->nestbuf;
    271 	struct strat_private *priv = PRIV(ump);
    272 	int queue;
    273 	int what;
    274 
    275 	KASSERT(ump);
    276 	KASSERT(nestbuf);
    277 	KASSERT(nestbuf->b_iodone == nestiobuf_iodone);
    278 
    279 	what = nestbuf->b_udf_c_type;
    280 	queue = UDF_SHED_READING;
    281 	if ((nestbuf->b_flags & B_READ) == 0) {
    282 		/* writing */
    283 		queue = UDF_SHED_SEQWRITING;
    284 		if (what == UDF_C_DSCR)
    285 			queue = UDF_SHED_WRITING;
    286 #if 0
    287 		if (queue == UDF_SHED_SEQWRITING) {
    288 			/* TODO do add sector to uncommitted space */
    289 		}
    290 #endif
    291 	}
    292 
    293 	/* use our own sheduler lists for more complex sheduling */
    294 	mutex_enter(&priv->discstrat_mutex);
    295 		BUFQ_PUT(priv->queues[queue], nestbuf);
    296 		vfs_timestamp(&priv->last_queued[queue]);
    297 	mutex_exit(&priv->discstrat_mutex);
    298 
    299 	/* signal our thread that there might be something to do */
    300 	cv_signal(&priv->discstrat_cv);
    301 }
    302 
    303 /* --------------------------------------------------------------------- */
    304 
    305 /* TODO convert to lb_size */
    306 static void
    307 udf_VAT_mapping_update(struct udf_mount *ump, struct buf *buf, uint32_t lb_map)
    308 {
    309 	union dscrptr    *fdscr = (union dscrptr *) buf->b_data;
    310 	struct vnode     *vp = buf->b_vp;
    311 	struct udf_node  *udf_node = VTOI(vp);
    312 	uint32_t lb_size, blks;
    313 	uint32_t lb_num;
    314 	uint32_t udf_rw32_lbmap;
    315 	int c_type = buf->b_udf_c_type;
    316 	int error;
    317 
    318 	/* only interested when we're using a VAT */
    319 	KASSERT(ump->vat_node);
    320 	KASSERT(ump->vtop_alloc[ump->node_part] == UDF_ALLOC_VAT);
    321 
    322 	/* only nodes are recorded in the VAT */
    323 	/* NOTE: and the fileset descriptor (FIXME ?) */
    324 	if (c_type != UDF_C_NODE)
    325 		return;
    326 
    327 	/* we now have an UDF FE/EFE node on media with VAT (or VAT itself) */
    328 	lb_size = udf_rw32(ump->logical_vol->lb_size);
    329 	blks = lb_size / DEV_BSIZE;
    330 
    331 	udf_rw32_lbmap = udf_rw32(lb_map);
    332 
    333 	/* if we're the VAT itself, only update our assigned sector number */
    334 	if (udf_node == ump->vat_node) {
    335 		fdscr->tag.tag_loc = udf_rw32_lbmap;
    336 		udf_validate_tag_sum(fdscr);
    337 		DPRINTF(TRANSLATE, ("VAT assigned to sector %u\n",
    338 			udf_rw32(udf_rw32_lbmap)));
    339 		/* no use mapping the VAT node in the VAT */
    340 		return;
    341 	}
    342 
    343 	/* record new position in VAT file */
    344 	lb_num = udf_rw32(fdscr->tag.tag_loc);
    345 
    346 	/* lb_num = udf_rw32(udf_node->write_loc.loc.lb_num); */
    347 
    348 	DPRINTF(TRANSLATE, ("VAT entry change (log %u -> phys %u)\n",
    349 			lb_num, lb_map));
    350 
    351 	/* VAT should be the longer than this write, can't go wrong */
    352 	KASSERT(lb_num <= ump->vat_entries);
    353 
    354 	mutex_enter(&ump->allocate_mutex);
    355 	error = udf_vat_write(ump->vat_node,
    356 			(uint8_t *) &udf_rw32_lbmap, 4,
    357 			ump->vat_offset + lb_num * 4);
    358 	mutex_exit(&ump->allocate_mutex);
    359 
    360 	if (error)
    361 		panic( "udf_VAT_mapping_update: HELP! i couldn't "
    362 			"write in the VAT file ?\n");
    363 }
    364 
    365 
    366 static void
    367 udf_issue_buf(struct udf_mount *ump, int queue, struct buf *buf)
    368 {
    369 	struct long_ad *node_ad_cpy;
    370 	struct part_desc *pdesc;
    371 	uint64_t *lmapping, *lmappos, blknr;
    372 	uint32_t our_sectornr, sectornr, bpos;
    373 	uint32_t ptov;
    374 	uint16_t vpart_num;
    375 	uint8_t *fidblk;
    376 	int sector_size = ump->discinfo.sector_size;
    377 	int blks = sector_size / DEV_BSIZE;
    378 	int len, buf_len;
    379 
    380 	/* if reading, just pass to the device's STRATEGY */
    381 	if (queue == UDF_SHED_READING) {
    382 		DPRINTF(SHEDULE, ("\nudf_issue_buf READ %p : sector %d type %d,"
    383 			"b_resid %d, b_bcount %d, b_bufsize %d\n",
    384 			buf, (uint32_t) buf->b_blkno / blks, buf->b_udf_c_type,
    385 			buf->b_resid, buf->b_bcount, buf->b_bufsize));
    386 		VOP_STRATEGY(ump->devvp, buf);
    387 		return;
    388 	}
    389 
    390 	blknr        = buf->b_blkno;
    391 	our_sectornr = blknr / blks;
    392 
    393 	if (queue == UDF_SHED_WRITING) {
    394 		DPRINTF(SHEDULE, ("\nudf_issue_buf WRITE %p : sector %d "
    395 			"type %d, b_resid %d, b_bcount %d, b_bufsize %d\n",
    396 			buf, (uint32_t) buf->b_blkno / blks, buf->b_udf_c_type,
    397 			buf->b_resid, buf->b_bcount, buf->b_bufsize));
    398 		/* if we have FIDs fixup using buffer's sector number(s) */
    399 		if (buf->b_udf_c_type == UDF_C_FIDS) {
    400 			panic("UDF_C_FIDS in SHED_WRITING!\n");
    401 			buf_len = buf->b_bcount;
    402 			sectornr = our_sectornr;
    403 			bpos = 0;
    404 			while (buf_len) {
    405 				len = MIN(buf_len, sector_size);
    406 				fidblk = (uint8_t *) buf->b_data + bpos;
    407 				udf_fixup_fid_block(fidblk, sector_size,
    408 					0, len, sectornr);
    409 				sectornr++;
    410 				bpos += len;
    411 				buf_len -= len;
    412 			}
    413 		}
    414 		udf_fixup_node_internals(ump, buf->b_data, buf->b_udf_c_type);
    415 		VOP_STRATEGY(ump->devvp, buf);
    416 		return;
    417 	}
    418 
    419 	KASSERT(queue == UDF_SHED_SEQWRITING);
    420 	DPRINTF(SHEDULE, ("\nudf_issue_buf SEQWRITE %p : sector XXXX "
    421 		"type %d, b_resid %d, b_bcount %d, b_bufsize %d\n",
    422 		buf, buf->b_udf_c_type, buf->b_resid, buf->b_bcount,
    423 		buf->b_bufsize));
    424 
    425 	/*
    426 	 * Buffers should not have been allocated to disc addresses yet on
    427 	 * this queue. Note that a buffer can get multiple extents allocated.
    428 	 *
    429 	 * lmapping contains lb_num relative to base partition.
    430 	 */
    431 	lmapping    = ump->la_lmapping;
    432 	node_ad_cpy = ump->la_node_ad_cpy;
    433 
    434 	/* logically allocate buf and map it in the file */
    435 	udf_late_allocate_buf(ump, buf, lmapping, node_ad_cpy, &vpart_num);
    436 
    437 	/* update mapping in the VAT */
    438 	udf_VAT_mapping_update(ump, buf, *lmapping);
    439 
    440 	/*
    441 	 * NOTE We are using the knowledge here that sequential media will
    442 	 * always be mapped linearly. Thus no use to explicitly translate the
    443 	 * lmapping list.
    444 	 */
    445 
    446 	/* calculate offset from physical base partition */
    447 	pdesc = ump->partitions[ump->vtop[vpart_num]];
    448 	ptov  = udf_rw32(pdesc->start_loc);
    449 
    450 	/* set buffers blkno to the physical block number */
    451 	buf->b_blkno = (*lmapping + ptov) * blks;
    452 
    453 	/* if we have FIDs, fixup using the new allocation table */
    454 	if (buf->b_udf_c_type == UDF_C_FIDS) {
    455 		buf_len = buf->b_bcount;
    456 		bpos = 0;
    457 		lmappos = lmapping;
    458 		while (buf_len) {
    459 			sectornr = *lmappos++;
    460 			len = MIN(buf_len, sector_size);
    461 			fidblk = (uint8_t *) buf->b_data + bpos;
    462 			udf_fixup_fid_block(fidblk, sector_size,
    463 				0, len, sectornr);
    464 			bpos += len;
    465 			buf_len -= len;
    466 		}
    467 	}
    468 
    469 	/* NOTE we can't have metadata space bitmap descriptors here */
    470 
    471 	udf_fixup_node_internals(ump, buf->b_data, buf->b_udf_c_type);
    472 	VOP_STRATEGY(ump->devvp, buf);
    473 }
    474 
    475 
    476 static void
    477 udf_doshedule(struct udf_mount *ump)
    478 {
    479 	struct buf *buf;
    480 	struct timespec now, *last;
    481 	struct strat_private *priv = PRIV(ump);
    482 	void (*b_callback)(struct buf *);
    483 	int new_queue;
    484 	int error;
    485 
    486 	buf = BUFQ_GET(priv->queues[priv->cur_queue]);
    487 	if (buf) {
    488 		/* transfer from the current queue to the device queue */
    489 		mutex_exit(&priv->discstrat_mutex);
    490 
    491 		/* transform buffer to synchronous; XXX needed? */
    492 		b_callback = buf->b_iodone;
    493 		buf->b_iodone = NULL;
    494 		CLR(buf->b_flags, B_ASYNC);
    495 
    496 		/* issue and wait on completion */
    497 		udf_issue_buf(ump, priv->cur_queue, buf);
    498 		biowait(buf);
    499 
    500 		mutex_enter(&priv->discstrat_mutex);
    501 
    502 		/* if there is an error, repair this error, otherwise propagate */
    503 		if (buf->b_error && ((buf->b_flags & B_READ) == 0)) {
    504 			/* check what we need to do */
    505 			panic("UDF write error, can't handle yet!\n");
    506 		}
    507 
    508 		/* propagate result to higher layers */
    509 		if (b_callback) {
    510 			buf->b_iodone = b_callback;
    511 			(*buf->b_iodone)(buf);
    512 		}
    513 
    514 		return;
    515 	}
    516 
    517 	/* Check if we're idling in this state */
    518 	vfs_timestamp(&now);
    519 	last = &priv->last_queued[priv->cur_queue];
    520 	if (ump->discinfo.mmc_class == MMC_CLASS_CD) {
    521 		/* dont switch too fast for CD media; its expensive in time */
    522 		if (now.tv_sec - last->tv_sec < 3)
    523 			return;
    524 	}
    525 
    526 	/* check if we can/should switch */
    527 	new_queue = priv->cur_queue;
    528 
    529 	if (BUFQ_PEEK(priv->queues[UDF_SHED_READING]))
    530 		new_queue = UDF_SHED_READING;
    531 	if (BUFQ_PEEK(priv->queues[UDF_SHED_SEQWRITING]))
    532 		new_queue = UDF_SHED_SEQWRITING;
    533 	if (BUFQ_PEEK(priv->queues[UDF_SHED_WRITING]))		/* only for unmount */
    534 		new_queue = UDF_SHED_WRITING;
    535 	if (priv->cur_queue == UDF_SHED_READING) {
    536 		if (new_queue == UDF_SHED_SEQWRITING) {
    537 			/* TODO use flag to signal if this is needed */
    538 			mutex_exit(&priv->discstrat_mutex);
    539 
    540 			/* update trackinfo for data and metadata */
    541 			error = udf_update_trackinfo(ump,
    542 					&ump->data_track);
    543 			assert(error == 0);
    544 			error = udf_update_trackinfo(ump,
    545 					&ump->metadata_track);
    546 			assert(error == 0);
    547 			mutex_enter(&priv->discstrat_mutex);
    548 		}
    549 	}
    550 
    551 	if (new_queue != priv->cur_queue) {
    552 		DPRINTF(SHEDULE, ("switching from %d to %d\n",
    553 			priv->cur_queue, new_queue));
    554 	}
    555 
    556 	priv->cur_queue = new_queue;
    557 }
    558 
    559 
    560 static void
    561 udf_discstrat_thread(void *arg)
    562 {
    563 	struct udf_mount *ump = (struct udf_mount *) arg;
    564 	struct strat_private *priv = PRIV(ump);
    565 	int empty;
    566 
    567 	empty = 1;
    568 	mutex_enter(&priv->discstrat_mutex);
    569 	while (priv->run_thread || !empty) {
    570 		/* process the current selected queue */
    571 		udf_doshedule(ump);
    572 		empty  = (BUFQ_PEEK(priv->queues[UDF_SHED_READING]) == NULL);
    573 		empty &= (BUFQ_PEEK(priv->queues[UDF_SHED_WRITING]) == NULL);
    574 		empty &= (BUFQ_PEEK(priv->queues[UDF_SHED_SEQWRITING]) == NULL);
    575 
    576 		/* wait for more if needed */
    577 		if (empty)
    578 			cv_timedwait(&priv->discstrat_cv,
    579 				&priv->discstrat_mutex, hz/8);
    580 	}
    581 	mutex_exit(&priv->discstrat_mutex);
    582 
    583 	wakeup(&priv->run_thread);
    584 	kthread_exit(0);
    585 	/* not reached */
    586 }
    587 
    588 /* --------------------------------------------------------------------- */
    589 
    590 static void
    591 udf_discstrat_init_seq(struct udf_strat_args *args)
    592 {
    593 	struct udf_mount *ump = args->ump;
    594 	struct strat_private *priv = PRIV(ump);
    595 	struct disk_strategy dkstrat;
    596 	uint32_t lb_size;
    597 
    598 	KASSERT(ump);
    599 	KASSERT(ump->logical_vol);
    600 	KASSERT(priv == NULL);
    601 
    602 	lb_size = udf_rw32(ump->logical_vol->lb_size);
    603 	KASSERT(lb_size > 0);
    604 
    605 	/* initialise our memory space */
    606 	ump->strategy_private = malloc(sizeof(struct strat_private),
    607 		M_UDFTEMP, M_WAITOK);
    608 	priv = ump->strategy_private;
    609 	memset(priv, 0 , sizeof(struct strat_private));
    610 
    611 	/* initialise locks */
    612 	cv_init(&priv->discstrat_cv, "udfstrat");
    613 	mutex_init(&priv->discstrat_mutex, MUTEX_DEFAULT, IPL_NONE);
    614 
    615 	/*
    616 	 * Initialise pool for descriptors associated with nodes. This is done
    617 	 * in lb_size units though currently lb_size is dictated to be
    618 	 * sector_size.
    619 	 */
    620 	pool_init(&priv->desc_pool, lb_size, 0, 0, 0, "udf_desc_pool", NULL,
    621 	    IPL_NONE);
    622 
    623 	/*
    624 	 * remember old device strategy method and explicit set method
    625 	 * `discsort' since we have our own more complex strategy that is not
    626 	 * implementable on the CD device and other strategies will get in the
    627 	 * way.
    628 	 */
    629 	memset(&priv->old_strategy_setting, 0,
    630 		sizeof(struct disk_strategy));
    631 	VOP_IOCTL(ump->devvp, DIOCGSTRATEGY, &priv->old_strategy_setting,
    632 		FREAD | FKIOCTL, NOCRED);
    633 	memset(&dkstrat, 0, sizeof(struct disk_strategy));
    634 	strcpy(dkstrat.dks_name, "discsort");
    635 	VOP_IOCTL(ump->devvp, DIOCSSTRATEGY, &dkstrat, FWRITE | FKIOCTL,
    636 		NOCRED);
    637 
    638 	/* initialise our internal sheduler */
    639 	priv->cur_queue = UDF_SHED_READING;
    640 	bufq_alloc(&priv->queues[UDF_SHED_READING], "disksort",
    641 		BUFQ_SORT_RAWBLOCK);
    642 	bufq_alloc(&priv->queues[UDF_SHED_WRITING], "disksort",
    643 		BUFQ_SORT_RAWBLOCK);
    644 	bufq_alloc(&priv->queues[UDF_SHED_SEQWRITING], "fcfs", 0);
    645 	vfs_timestamp(&priv->last_queued[UDF_SHED_READING]);
    646 	vfs_timestamp(&priv->last_queued[UDF_SHED_WRITING]);
    647 	vfs_timestamp(&priv->last_queued[UDF_SHED_SEQWRITING]);
    648 
    649 	/* create our disk strategy thread */
    650 	priv->run_thread = 1;
    651 	if (kthread_create(PRI_NONE, 0 /* KTHREAD_MPSAFE*/, NULL /* cpu_info*/,
    652 		udf_discstrat_thread, ump, &priv->queue_lwp,
    653 		"%s", "udf_rw")) {
    654 		panic("fork udf_rw");
    655 	}
    656 }
    657 
    658 
    659 static void
    660 udf_discstrat_finish_seq(struct udf_strat_args *args)
    661 {
    662 	struct udf_mount *ump = args->ump;
    663 	struct strat_private *priv = PRIV(ump);
    664 	int error;
    665 
    666 	if (ump == NULL)
    667 		return;
    668 
    669 	/* stop our sheduling thread */
    670 	KASSERT(priv->run_thread == 1);
    671 	priv->run_thread = 0;
    672 	wakeup(priv->queue_lwp);
    673 	do {
    674 		error = tsleep(&priv->run_thread, PRIBIO+1,
    675 			"udfshedfin", hz);
    676 	} while (error);
    677 	/* kthread should be finished now */
    678 
    679 	/* set back old device strategy method */
    680 	VOP_IOCTL(ump->devvp, DIOCSSTRATEGY, &priv->old_strategy_setting,
    681 			FWRITE, NOCRED);
    682 
    683 	/* destroy our pool */
    684 	pool_destroy(&priv->desc_pool);
    685 
    686 	/* free our private space */
    687 	free(ump->strategy_private, M_UDFTEMP);
    688 	ump->strategy_private = NULL;
    689 }
    690 
    691 /* --------------------------------------------------------------------- */
    692 
    693 struct udf_strategy udf_strat_sequential =
    694 {
    695 	udf_create_logvol_dscr_seq,
    696 	udf_free_logvol_dscr_seq,
    697 	udf_read_logvol_dscr_seq,
    698 	udf_write_logvol_dscr_seq,
    699 	udf_queuebuf_seq,
    700 	udf_discstrat_init_seq,
    701 	udf_discstrat_finish_seq
    702 };
    703 
    704 
    705