Home | History | Annotate | Line # | Download | only in udf
udf_strat_sequential.c revision 1.18
      1  1.18     skrll /* $NetBSD: udf_strat_sequential.c,v 1.18 2022/08/27 05:31:59 skrll Exp $ */
      2   1.1   reinoud 
      3   1.1   reinoud /*
      4   1.1   reinoud  * Copyright (c) 2006, 2008 Reinoud Zandijk
      5   1.1   reinoud  * All rights reserved.
      6  1.18     skrll  *
      7   1.1   reinoud  * Redistribution and use in source and binary forms, with or without
      8   1.1   reinoud  * modification, are permitted provided that the following conditions
      9   1.1   reinoud  * are met:
     10   1.1   reinoud  * 1. Redistributions of source code must retain the above copyright
     11   1.1   reinoud  *    notice, this list of conditions and the following disclaimer.
     12   1.1   reinoud  * 2. Redistributions in binary form must reproduce the above copyright
     13   1.1   reinoud  *    notice, this list of conditions and the following disclaimer in the
     14   1.1   reinoud  *    documentation and/or other materials provided with the distribution.
     15  1.18     skrll  *
     16   1.1   reinoud  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     17   1.1   reinoud  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     18   1.1   reinoud  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     19   1.1   reinoud  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     20   1.1   reinoud  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     21   1.1   reinoud  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     22   1.1   reinoud  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     23   1.1   reinoud  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     24   1.1   reinoud  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     25   1.1   reinoud  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     26  1.18     skrll  *
     27   1.1   reinoud  */
     28   1.1   reinoud 
     29   1.1   reinoud #include <sys/cdefs.h>
     30   1.1   reinoud #ifndef lint
     31  1.18     skrll __KERNEL_RCSID(0, "$NetBSD: udf_strat_sequential.c,v 1.18 2022/08/27 05:31:59 skrll Exp $");
     32   1.1   reinoud #endif /* not lint */
     33   1.1   reinoud 
     34   1.1   reinoud 
     35   1.1   reinoud #if defined(_KERNEL_OPT)
     36   1.1   reinoud #include "opt_compat_netbsd.h"
     37   1.1   reinoud #endif
     38   1.1   reinoud 
     39   1.1   reinoud #include <sys/param.h>
     40   1.1   reinoud #include <sys/systm.h>
     41   1.1   reinoud #include <sys/sysctl.h>
     42   1.1   reinoud #include <sys/namei.h>
     43   1.1   reinoud #include <sys/proc.h>
     44   1.1   reinoud #include <sys/kernel.h>
     45   1.1   reinoud #include <sys/vnode.h>
     46   1.1   reinoud #include <miscfs/genfs/genfs_node.h>
     47   1.1   reinoud #include <sys/mount.h>
     48   1.1   reinoud #include <sys/buf.h>
     49   1.1   reinoud #include <sys/file.h>
     50   1.1   reinoud #include <sys/device.h>
     51   1.1   reinoud #include <sys/disklabel.h>
     52   1.1   reinoud #include <sys/ioctl.h>
     53   1.1   reinoud #include <sys/malloc.h>
     54   1.1   reinoud #include <sys/dirent.h>
     55   1.1   reinoud #include <sys/stat.h>
     56   1.1   reinoud #include <sys/conf.h>
     57   1.1   reinoud #include <sys/kauth.h>
     58   1.1   reinoud #include <sys/kthread.h>
     59   1.1   reinoud #include <dev/clock_subr.h>
     60   1.1   reinoud 
     61   1.1   reinoud #include <fs/udf/ecma167-udf.h>
     62   1.1   reinoud #include <fs/udf/udf_mount.h>
     63   1.1   reinoud 
     64   1.1   reinoud #include "udf.h"
     65   1.1   reinoud #include "udf_subr.h"
     66   1.1   reinoud #include "udf_bswap.h"
     67   1.1   reinoud 
     68   1.1   reinoud 
     69   1.1   reinoud #define VTOI(vnode) ((struct udf_node *) vnode->v_data)
     70   1.1   reinoud #define PRIV(ump) ((struct strat_private *) ump->strategy_private)
     71   1.1   reinoud 
     72   1.1   reinoud /* --------------------------------------------------------------------- */
     73   1.1   reinoud 
     74   1.1   reinoud /* BUFQ's */
     75   1.1   reinoud #define UDF_SHED_MAX 3
     76   1.1   reinoud 
     77   1.1   reinoud #define UDF_SHED_READING	0
     78   1.1   reinoud #define UDF_SHED_WRITING	1
     79   1.1   reinoud #define UDF_SHED_SEQWRITING	2
     80   1.1   reinoud 
     81   1.1   reinoud struct strat_private {
     82   1.1   reinoud 	struct pool		 desc_pool;	 	/* node descriptors */
     83   1.1   reinoud 
     84   1.1   reinoud 	lwp_t			*queue_lwp;
     85   1.1   reinoud 	kcondvar_t		 discstrat_cv;		/* to wait on       */
     86   1.1   reinoud 	kmutex_t		 discstrat_mutex;	/* disc strategy    */
     87   1.1   reinoud 
     88   1.1   reinoud 	int			 run_thread;		/* thread control */
     89  1.15   reinoud 	int			 sync_req;		/* thread control */
     90   1.1   reinoud 	int			 cur_queue;
     91   1.1   reinoud 
     92   1.1   reinoud 	struct disk_strategy	 old_strategy_setting;
     93   1.1   reinoud 	struct bufq_state	*queues[UDF_SHED_MAX];
     94   1.1   reinoud 	struct timespec		 last_queued[UDF_SHED_MAX];
     95   1.1   reinoud };
     96   1.1   reinoud 
     97   1.1   reinoud 
     98   1.1   reinoud /* --------------------------------------------------------------------- */
     99   1.1   reinoud 
    100   1.1   reinoud static void
    101   1.1   reinoud udf_wr_nodedscr_callback(struct buf *buf)
    102   1.1   reinoud {
    103   1.1   reinoud 	struct udf_node *udf_node;
    104   1.1   reinoud 
    105   1.1   reinoud 	KASSERT(buf);
    106   1.1   reinoud 	KASSERT(buf->b_data);
    107   1.1   reinoud 
    108   1.1   reinoud 	/* called when write action is done */
    109   1.1   reinoud 	DPRINTF(WRITE, ("udf_wr_nodedscr_callback(): node written out\n"));
    110   1.1   reinoud 
    111   1.1   reinoud 	udf_node = VTOI(buf->b_vp);
    112   1.1   reinoud 	if (udf_node == NULL) {
    113   1.1   reinoud 		putiobuf(buf);
    114   1.1   reinoud 		printf("udf_wr_node_callback: NULL node?\n");
    115   1.1   reinoud 		return;
    116   1.1   reinoud 	}
    117   1.1   reinoud 
    118   1.1   reinoud 	/* XXX right flags to mark dirty again on error? */
    119   1.1   reinoud 	if (buf->b_error) {
    120   1.1   reinoud 		udf_node->i_flags |= IN_MODIFIED | IN_ACCESSED;
    121  1.17    andvar 		/* XXX TODO reschedule on error */
    122   1.1   reinoud 	}
    123   1.1   reinoud 
    124   1.2   reinoud 	/* decrement outstanding_nodedscr */
    125   1.2   reinoud 	KASSERT(udf_node->outstanding_nodedscr >= 1);
    126   1.2   reinoud 	udf_node->outstanding_nodedscr--;
    127   1.2   reinoud 	if (udf_node->outstanding_nodedscr == 0) {
    128   1.2   reinoud 		/* first unlock the node */
    129  1.10   reinoud 		UDF_UNLOCK_NODE(udf_node, 0);
    130   1.2   reinoud 		wakeup(&udf_node->outstanding_nodedscr);
    131   1.2   reinoud 	}
    132   1.1   reinoud 
    133   1.1   reinoud 	putiobuf(buf);
    134   1.1   reinoud }
    135   1.1   reinoud 
    136   1.1   reinoud /* --------------------------------------------------------------------- */
    137   1.1   reinoud 
    138   1.1   reinoud static int
    139   1.1   reinoud udf_create_logvol_dscr_seq(struct udf_strat_args *args)
    140   1.1   reinoud {
    141   1.1   reinoud 	union dscrptr   **dscrptr = &args->dscr;
    142   1.1   reinoud 	struct udf_mount *ump = args->ump;
    143   1.1   reinoud 	struct strat_private *priv = PRIV(ump);
    144   1.1   reinoud 	uint32_t lb_size;
    145   1.1   reinoud 
    146   1.1   reinoud 	lb_size = udf_rw32(ump->logical_vol->lb_size);
    147   1.1   reinoud 	*dscrptr = pool_get(&priv->desc_pool, PR_WAITOK);
    148   1.1   reinoud 	memset(*dscrptr, 0, lb_size);
    149   1.1   reinoud 
    150   1.1   reinoud 	return 0;
    151   1.1   reinoud }
    152   1.1   reinoud 
    153   1.1   reinoud 
    154   1.1   reinoud static void
    155   1.1   reinoud udf_free_logvol_dscr_seq(struct udf_strat_args *args)
    156   1.1   reinoud {
    157   1.1   reinoud 	union dscrptr    *dscr = args->dscr;
    158   1.1   reinoud 	struct udf_mount *ump  = args->ump;
    159   1.1   reinoud 	struct strat_private *priv = PRIV(ump);
    160   1.1   reinoud 
    161   1.1   reinoud 	pool_put(&priv->desc_pool, dscr);
    162   1.1   reinoud }
    163   1.1   reinoud 
    164   1.1   reinoud 
    165   1.1   reinoud static int
    166   1.1   reinoud udf_read_logvol_dscr_seq(struct udf_strat_args *args)
    167   1.1   reinoud {
    168   1.1   reinoud 	union dscrptr   **dscrptr = &args->dscr;
    169   1.1   reinoud 	union dscrptr    *tmpdscr;
    170   1.1   reinoud 	struct udf_mount *ump = args->ump;
    171   1.1   reinoud 	struct long_ad   *icb = args->icb;
    172   1.1   reinoud 	struct strat_private *priv = PRIV(ump);
    173   1.1   reinoud 	uint32_t lb_size;
    174   1.1   reinoud 	uint32_t sector, dummy;
    175   1.1   reinoud 	int error;
    176   1.1   reinoud 
    177   1.1   reinoud 	lb_size = udf_rw32(ump->logical_vol->lb_size);
    178   1.1   reinoud 
    179   1.1   reinoud 	error = udf_translate_vtop(ump, icb, &sector, &dummy);
    180   1.1   reinoud 	if (error)
    181   1.1   reinoud 		return error;
    182   1.1   reinoud 
    183   1.1   reinoud 	/* try to read in fe/efe */
    184   1.1   reinoud 	error = udf_read_phys_dscr(ump, sector, M_UDFTEMP, &tmpdscr);
    185   1.1   reinoud 	if (error)
    186   1.1   reinoud 		return error;
    187   1.1   reinoud 
    188   1.1   reinoud 	*dscrptr = pool_get(&priv->desc_pool, PR_WAITOK);
    189   1.1   reinoud 	memcpy(*dscrptr, tmpdscr, lb_size);
    190   1.1   reinoud 	free(tmpdscr, M_UDFTEMP);
    191   1.1   reinoud 
    192   1.1   reinoud 	return 0;
    193   1.1   reinoud }
    194   1.1   reinoud 
    195   1.1   reinoud 
    196   1.1   reinoud static int
    197   1.1   reinoud udf_write_logvol_dscr_seq(struct udf_strat_args *args)
    198   1.1   reinoud {
    199   1.1   reinoud 	union dscrptr    *dscr     = args->dscr;
    200   1.1   reinoud 	struct udf_mount *ump      = args->ump;
    201   1.1   reinoud 	struct udf_node  *udf_node = args->udf_node;
    202   1.1   reinoud 	struct long_ad   *icb      = args->icb;
    203   1.1   reinoud 	int               waitfor  = args->waitfor;
    204   1.1   reinoud 	uint32_t logsectornr, sectornr, dummy;
    205   1.1   reinoud 	int error, vpart;
    206   1.1   reinoud 
    207   1.1   reinoud 	/*
    208  1.18     skrll 	 * we have to decide if we write it out sequential or at its fixed
    209   1.1   reinoud 	 * position by examining the partition its (to be) written on.
    210   1.1   reinoud 	 */
    211   1.1   reinoud 	vpart       = udf_rw16(udf_node->loc.loc.part_num);
    212   1.1   reinoud 	logsectornr = udf_rw32(icb->loc.lb_num);
    213   1.1   reinoud 	sectornr    = 0;
    214   1.1   reinoud 	if (ump->vtop_tp[vpart] != UDF_VTOP_TYPE_VIRT) {
    215   1.1   reinoud 		error = udf_translate_vtop(ump, icb, &sectornr, &dummy);
    216   1.1   reinoud 		if (error)
    217   1.2   reinoud 			goto out;
    218   1.1   reinoud 	}
    219   1.1   reinoud 
    220   1.1   reinoud 	if (waitfor) {
    221   1.1   reinoud 		DPRINTF(WRITE, ("udf_write_logvol_dscr: sync write\n"));
    222   1.1   reinoud 
    223   1.1   reinoud 		error = udf_write_phys_dscr_sync(ump, udf_node, UDF_C_NODE,
    224   1.1   reinoud 			dscr, sectornr, logsectornr);
    225   1.1   reinoud 	} else {
    226   1.1   reinoud 		DPRINTF(WRITE, ("udf_write_logvol_dscr: no wait, async write\n"));
    227   1.1   reinoud 
    228   1.1   reinoud 		error = udf_write_phys_dscr_async(ump, udf_node, UDF_C_NODE,
    229   1.1   reinoud 			dscr, sectornr, logsectornr, udf_wr_nodedscr_callback);
    230   1.1   reinoud 		/* will be UNLOCKED in call back */
    231   1.2   reinoud 		return error;
    232   1.2   reinoud 	}
    233   1.2   reinoud out:
    234   1.2   reinoud 	udf_node->outstanding_nodedscr--;
    235   1.2   reinoud 	if (udf_node->outstanding_nodedscr == 0) {
    236   1.2   reinoud 		UDF_UNLOCK_NODE(udf_node, 0);
    237   1.2   reinoud 		wakeup(&udf_node->outstanding_nodedscr);
    238   1.1   reinoud 	}
    239   1.1   reinoud 
    240   1.1   reinoud 	return error;
    241   1.1   reinoud }
    242   1.1   reinoud 
    243   1.1   reinoud /* --------------------------------------------------------------------- */
    244   1.1   reinoud 
    245   1.1   reinoud /*
    246  1.17    andvar  * Main file-system specific scheduler. Due to the nature of optical media
    247  1.17    andvar  * scheduling can't be performed in the traditional way. Most OS
    248   1.1   reinoud  * implementations i've seen thus read or write a file atomically giving all
    249   1.1   reinoud  * kinds of side effects.
    250   1.1   reinoud  *
    251  1.17    andvar  * This implementation uses a kernel thread to schedule the queued requests in
    252   1.1   reinoud  * such a way that is semi-optimal for optical media; this means aproximately
    253   1.1   reinoud  * (R*|(Wr*|Ws*))* since switching between reading and writing is expensive in
    254   1.1   reinoud  * time.
    255   1.1   reinoud  */
    256   1.1   reinoud 
    257   1.1   reinoud static void
    258   1.1   reinoud udf_queuebuf_seq(struct udf_strat_args *args)
    259   1.1   reinoud {
    260   1.1   reinoud 	struct udf_mount *ump = args->ump;
    261   1.1   reinoud 	struct buf *nestbuf = args->nestbuf;
    262   1.1   reinoud 	struct strat_private *priv = PRIV(ump);
    263   1.1   reinoud 	int queue;
    264   1.1   reinoud 	int what;
    265   1.1   reinoud 
    266   1.1   reinoud 	KASSERT(ump);
    267   1.1   reinoud 	KASSERT(nestbuf);
    268   1.1   reinoud 	KASSERT(nestbuf->b_iodone == nestiobuf_iodone);
    269   1.1   reinoud 
    270   1.1   reinoud 	what = nestbuf->b_udf_c_type;
    271   1.1   reinoud 	queue = UDF_SHED_READING;
    272   1.1   reinoud 	if ((nestbuf->b_flags & B_READ) == 0) {
    273   1.1   reinoud 		/* writing */
    274   1.1   reinoud 		queue = UDF_SHED_SEQWRITING;
    275   1.8   reinoud 		if (what == UDF_C_ABSOLUTE)
    276   1.1   reinoud 			queue = UDF_SHED_WRITING;
    277   1.1   reinoud 	}
    278   1.1   reinoud 
    279  1.17    andvar 	/* use our own scheduler lists for more complex scheduling */
    280   1.1   reinoud 	mutex_enter(&priv->discstrat_mutex);
    281   1.7      yamt 		bufq_put(priv->queues[queue], nestbuf);
    282   1.1   reinoud 		vfs_timestamp(&priv->last_queued[queue]);
    283   1.1   reinoud 	mutex_exit(&priv->discstrat_mutex);
    284   1.1   reinoud 
    285   1.1   reinoud 	/* signal our thread that there might be something to do */
    286   1.1   reinoud 	cv_signal(&priv->discstrat_cv);
    287   1.1   reinoud }
    288   1.1   reinoud 
    289   1.1   reinoud /* --------------------------------------------------------------------- */
    290   1.1   reinoud 
    291  1.15   reinoud static void
    292  1.15   reinoud udf_sync_caches_seq(struct udf_strat_args *args)
    293  1.15   reinoud {
    294  1.15   reinoud 	struct udf_mount *ump = args->ump;
    295  1.15   reinoud 	struct strat_private *priv = PRIV(ump);
    296  1.15   reinoud 
    297  1.15   reinoud 	/* we might be called during unmount inadvertedly, be on safe side */
    298  1.15   reinoud 	if (!priv)
    299  1.15   reinoud 		return;
    300  1.15   reinoud 
    301  1.15   reinoud 	/* signal our thread that there might be something to do */
    302  1.15   reinoud 	priv->sync_req = 1;
    303  1.15   reinoud 	cv_signal(&priv->discstrat_cv);
    304  1.15   reinoud 
    305  1.15   reinoud 	mutex_enter(&priv->discstrat_mutex);
    306  1.15   reinoud 		while (priv->sync_req) {
    307  1.15   reinoud 			cv_timedwait(&priv->discstrat_cv,
    308  1.15   reinoud 				&priv->discstrat_mutex, hz/8);
    309  1.15   reinoud 		}
    310  1.15   reinoud 	mutex_exit(&priv->discstrat_mutex);
    311  1.15   reinoud }
    312  1.15   reinoud 
    313  1.15   reinoud /* --------------------------------------------------------------------- */
    314  1.15   reinoud 
    315   1.1   reinoud /* TODO convert to lb_size */
    316   1.1   reinoud static void
    317   1.3   reinoud udf_VAT_mapping_update(struct udf_mount *ump, struct buf *buf, uint32_t lb_map)
    318   1.1   reinoud {
    319   1.1   reinoud 	union dscrptr    *fdscr = (union dscrptr *) buf->b_data;
    320   1.1   reinoud 	struct vnode     *vp = buf->b_vp;
    321   1.1   reinoud 	struct udf_node  *udf_node = VTOI(vp);
    322   1.3   reinoud 	uint32_t lb_num;
    323   1.1   reinoud 	uint32_t udf_rw32_lbmap;
    324   1.1   reinoud 	int c_type = buf->b_udf_c_type;
    325   1.1   reinoud 	int error;
    326   1.1   reinoud 
    327   1.1   reinoud 	/* only interested when we're using a VAT */
    328   1.1   reinoud 	KASSERT(ump->vat_node);
    329   1.4   reinoud 	KASSERT(ump->vtop_alloc[ump->node_part] == UDF_ALLOC_VAT);
    330   1.1   reinoud 
    331   1.1   reinoud 	/* only nodes are recorded in the VAT */
    332   1.1   reinoud 	/* NOTE: and the fileset descriptor (FIXME ?) */
    333   1.1   reinoud 	if (c_type != UDF_C_NODE)
    334   1.1   reinoud 		return;
    335   1.1   reinoud 
    336   1.1   reinoud 	udf_rw32_lbmap = udf_rw32(lb_map);
    337   1.1   reinoud 
    338   1.1   reinoud 	/* if we're the VAT itself, only update our assigned sector number */
    339   1.1   reinoud 	if (udf_node == ump->vat_node) {
    340   1.1   reinoud 		fdscr->tag.tag_loc = udf_rw32_lbmap;
    341   1.1   reinoud 		udf_validate_tag_sum(fdscr);
    342   1.1   reinoud 		DPRINTF(TRANSLATE, ("VAT assigned to sector %u\n",
    343   1.1   reinoud 			udf_rw32(udf_rw32_lbmap)));
    344   1.1   reinoud 		/* no use mapping the VAT node in the VAT */
    345   1.1   reinoud 		return;
    346   1.1   reinoud 	}
    347   1.1   reinoud 
    348   1.2   reinoud 	/* record new position in VAT file */
    349   1.2   reinoud 	lb_num = udf_rw32(fdscr->tag.tag_loc);
    350   1.1   reinoud 
    351   1.2   reinoud 	/* lb_num = udf_rw32(udf_node->write_loc.loc.lb_num); */
    352   1.1   reinoud 
    353   1.1   reinoud 	DPRINTF(TRANSLATE, ("VAT entry change (log %u -> phys %u)\n",
    354   1.1   reinoud 			lb_num, lb_map));
    355   1.1   reinoud 
    356   1.1   reinoud 	/* VAT should be the longer than this write, can't go wrong */
    357   1.1   reinoud 	KASSERT(lb_num <= ump->vat_entries);
    358   1.1   reinoud 
    359   1.1   reinoud 	mutex_enter(&ump->allocate_mutex);
    360   1.1   reinoud 	error = udf_vat_write(ump->vat_node,
    361   1.1   reinoud 			(uint8_t *) &udf_rw32_lbmap, 4,
    362   1.1   reinoud 			ump->vat_offset + lb_num * 4);
    363   1.1   reinoud 	mutex_exit(&ump->allocate_mutex);
    364   1.1   reinoud 
    365   1.1   reinoud 	if (error)
    366   1.1   reinoud 		panic( "udf_VAT_mapping_update: HELP! i couldn't "
    367   1.1   reinoud 			"write in the VAT file ?\n");
    368   1.1   reinoud }
    369   1.1   reinoud 
    370   1.1   reinoud 
    371   1.1   reinoud static void
    372   1.1   reinoud udf_issue_buf(struct udf_mount *ump, int queue, struct buf *buf)
    373   1.1   reinoud {
    374   1.8   reinoud 	union dscrptr *dscr;
    375   1.1   reinoud 	struct long_ad *node_ad_cpy;
    376   1.3   reinoud 	struct part_desc *pdesc;
    377  1.12  christos 	uint64_t *lmapping, *lmappos;
    378  1.12  christos 	uint32_t sectornr, bpos;
    379   1.3   reinoud 	uint32_t ptov;
    380   1.3   reinoud 	uint16_t vpart_num;
    381   1.1   reinoud 	uint8_t *fidblk;
    382   1.1   reinoud 	int sector_size = ump->discinfo.sector_size;
    383   1.1   reinoud 	int blks = sector_size / DEV_BSIZE;
    384   1.1   reinoud 	int len, buf_len;
    385   1.1   reinoud 
    386   1.1   reinoud 	/* if reading, just pass to the device's STRATEGY */
    387   1.1   reinoud 	if (queue == UDF_SHED_READING) {
    388   1.1   reinoud 		DPRINTF(SHEDULE, ("\nudf_issue_buf READ %p : sector %d type %d,"
    389   1.1   reinoud 			"b_resid %d, b_bcount %d, b_bufsize %d\n",
    390   1.1   reinoud 			buf, (uint32_t) buf->b_blkno / blks, buf->b_udf_c_type,
    391   1.1   reinoud 			buf->b_resid, buf->b_bcount, buf->b_bufsize));
    392   1.1   reinoud 		VOP_STRATEGY(ump->devvp, buf);
    393   1.1   reinoud 		return;
    394   1.1   reinoud 	}
    395   1.1   reinoud 
    396   1.1   reinoud 	if (queue == UDF_SHED_WRITING) {
    397   1.1   reinoud 		DPRINTF(SHEDULE, ("\nudf_issue_buf WRITE %p : sector %d "
    398   1.1   reinoud 			"type %d, b_resid %d, b_bcount %d, b_bufsize %d\n",
    399   1.1   reinoud 			buf, (uint32_t) buf->b_blkno / blks, buf->b_udf_c_type,
    400   1.1   reinoud 			buf->b_resid, buf->b_bcount, buf->b_bufsize));
    401   1.8   reinoud 		KASSERT(buf->b_udf_c_type == UDF_C_ABSOLUTE);
    402   1.8   reinoud 
    403   1.8   reinoud 		// udf_fixup_node_internals(ump, buf->b_data, buf->b_udf_c_type);
    404   1.1   reinoud 		VOP_STRATEGY(ump->devvp, buf);
    405   1.1   reinoud 		return;
    406   1.1   reinoud 	}
    407   1.1   reinoud 
    408   1.1   reinoud 	KASSERT(queue == UDF_SHED_SEQWRITING);
    409   1.1   reinoud 	DPRINTF(SHEDULE, ("\nudf_issue_buf SEQWRITE %p : sector XXXX "
    410   1.1   reinoud 		"type %d, b_resid %d, b_bcount %d, b_bufsize %d\n",
    411   1.1   reinoud 		buf, buf->b_udf_c_type, buf->b_resid, buf->b_bcount,
    412   1.1   reinoud 		buf->b_bufsize));
    413   1.1   reinoud 
    414   1.1   reinoud 	/*
    415   1.1   reinoud 	 * Buffers should not have been allocated to disc addresses yet on
    416   1.1   reinoud 	 * this queue. Note that a buffer can get multiple extents allocated.
    417   1.1   reinoud 	 *
    418   1.1   reinoud 	 * lmapping contains lb_num relative to base partition.
    419   1.1   reinoud 	 */
    420   1.1   reinoud 	lmapping    = ump->la_lmapping;
    421   1.1   reinoud 	node_ad_cpy = ump->la_node_ad_cpy;
    422   1.1   reinoud 
    423   1.3   reinoud 	/* logically allocate buf and map it in the file */
    424   1.3   reinoud 	udf_late_allocate_buf(ump, buf, lmapping, node_ad_cpy, &vpart_num);
    425   1.3   reinoud 
    426   1.3   reinoud 	/*
    427   1.3   reinoud 	 * NOTE We are using the knowledge here that sequential media will
    428   1.3   reinoud 	 * always be mapped linearly. Thus no use to explicitly translate the
    429   1.3   reinoud 	 * lmapping list.
    430   1.3   reinoud 	 */
    431   1.3   reinoud 
    432   1.3   reinoud 	/* calculate offset from physical base partition */
    433   1.3   reinoud 	pdesc = ump->partitions[ump->vtop[vpart_num]];
    434   1.3   reinoud 	ptov  = udf_rw32(pdesc->start_loc);
    435   1.3   reinoud 
    436   1.3   reinoud 	/* set buffers blkno to the physical block number */
    437   1.3   reinoud 	buf->b_blkno = (*lmapping + ptov) * blks;
    438   1.1   reinoud 
    439   1.8   reinoud 	/* fixate floating descriptors */
    440   1.8   reinoud 	if (buf->b_udf_c_type == UDF_C_FLOAT_DSCR) {
    441   1.8   reinoud 		/* set our tag location to the absolute position */
    442   1.8   reinoud 		dscr = (union dscrptr *) buf->b_data;
    443   1.8   reinoud 		dscr->tag.tag_loc = udf_rw32(*lmapping + ptov);
    444   1.8   reinoud 		udf_validate_tag_and_crc_sums(dscr);
    445   1.8   reinoud 	}
    446   1.8   reinoud 
    447   1.8   reinoud 	/* update mapping in the VAT */
    448   1.8   reinoud 	if (buf->b_udf_c_type == UDF_C_NODE) {
    449   1.8   reinoud 		udf_VAT_mapping_update(ump, buf, *lmapping);
    450   1.8   reinoud 		udf_fixup_node_internals(ump, buf->b_data, buf->b_udf_c_type);
    451   1.8   reinoud 	}
    452   1.8   reinoud 
    453   1.1   reinoud 	/* if we have FIDs, fixup using the new allocation table */
    454   1.1   reinoud 	if (buf->b_udf_c_type == UDF_C_FIDS) {
    455   1.1   reinoud 		buf_len = buf->b_bcount;
    456   1.1   reinoud 		bpos = 0;
    457   1.1   reinoud 		lmappos = lmapping;
    458   1.1   reinoud 		while (buf_len) {
    459   1.1   reinoud 			sectornr = *lmappos++;
    460   1.1   reinoud 			len = MIN(buf_len, sector_size);
    461   1.1   reinoud 			fidblk = (uint8_t *) buf->b_data + bpos;
    462   1.1   reinoud 			udf_fixup_fid_block(fidblk, sector_size,
    463   1.1   reinoud 				0, len, sectornr);
    464   1.1   reinoud 			bpos += len;
    465   1.1   reinoud 			buf_len -= len;
    466   1.1   reinoud 		}
    467   1.1   reinoud 	}
    468   1.4   reinoud 
    469   1.1   reinoud 	VOP_STRATEGY(ump->devvp, buf);
    470   1.1   reinoud }
    471   1.1   reinoud 
    472   1.1   reinoud 
    473   1.1   reinoud static void
    474   1.1   reinoud udf_doshedule(struct udf_mount *ump)
    475   1.1   reinoud {
    476   1.1   reinoud 	struct buf *buf;
    477   1.1   reinoud 	struct timespec now, *last;
    478   1.1   reinoud 	struct strat_private *priv = PRIV(ump);
    479   1.1   reinoud 	void (*b_callback)(struct buf *);
    480   1.1   reinoud 	int new_queue;
    481   1.1   reinoud 	int error;
    482   1.1   reinoud 
    483   1.7      yamt 	buf = bufq_get(priv->queues[priv->cur_queue]);
    484   1.1   reinoud 	if (buf) {
    485   1.1   reinoud 		/* transfer from the current queue to the device queue */
    486   1.1   reinoud 		mutex_exit(&priv->discstrat_mutex);
    487   1.1   reinoud 
    488   1.1   reinoud 		/* transform buffer to synchronous; XXX needed? */
    489   1.1   reinoud 		b_callback = buf->b_iodone;
    490   1.1   reinoud 		buf->b_iodone = NULL;
    491   1.1   reinoud 		CLR(buf->b_flags, B_ASYNC);
    492   1.1   reinoud 
    493   1.1   reinoud 		/* issue and wait on completion */
    494   1.1   reinoud 		udf_issue_buf(ump, priv->cur_queue, buf);
    495   1.1   reinoud 		biowait(buf);
    496   1.1   reinoud 
    497   1.1   reinoud 		mutex_enter(&priv->discstrat_mutex);
    498   1.1   reinoud 
    499   1.1   reinoud 		/* if there is an error, repair this error, otherwise propagate */
    500   1.1   reinoud 		if (buf->b_error && ((buf->b_flags & B_READ) == 0)) {
    501   1.1   reinoud 			/* check what we need to do */
    502   1.1   reinoud 			panic("UDF write error, can't handle yet!\n");
    503   1.1   reinoud 		}
    504   1.1   reinoud 
    505   1.1   reinoud 		/* propagate result to higher layers */
    506   1.1   reinoud 		if (b_callback) {
    507   1.1   reinoud 			buf->b_iodone = b_callback;
    508   1.1   reinoud 			(*buf->b_iodone)(buf);
    509   1.1   reinoud 		}
    510   1.1   reinoud 
    511   1.1   reinoud 		return;
    512   1.1   reinoud 	}
    513   1.1   reinoud 
    514   1.1   reinoud 	/* Check if we're idling in this state */
    515   1.1   reinoud 	vfs_timestamp(&now);
    516   1.1   reinoud 	last = &priv->last_queued[priv->cur_queue];
    517   1.1   reinoud 	if (ump->discinfo.mmc_class == MMC_CLASS_CD) {
    518   1.1   reinoud 		/* dont switch too fast for CD media; its expensive in time */
    519   1.1   reinoud 		if (now.tv_sec - last->tv_sec < 3)
    520   1.1   reinoud 			return;
    521   1.1   reinoud 	}
    522   1.1   reinoud 
    523   1.1   reinoud 	/* check if we can/should switch */
    524   1.1   reinoud 	new_queue = priv->cur_queue;
    525   1.1   reinoud 
    526   1.7      yamt 	if (bufq_peek(priv->queues[UDF_SHED_READING]))
    527   1.1   reinoud 		new_queue = UDF_SHED_READING;
    528   1.9   reinoud 	if (bufq_peek(priv->queues[UDF_SHED_WRITING]))		/* only for unmount */
    529   1.9   reinoud 		new_queue = UDF_SHED_WRITING;
    530   1.7      yamt 	if (bufq_peek(priv->queues[UDF_SHED_SEQWRITING]))
    531   1.1   reinoud 		new_queue = UDF_SHED_SEQWRITING;
    532   1.1   reinoud 	if (priv->cur_queue == UDF_SHED_READING) {
    533   1.1   reinoud 		if (new_queue == UDF_SHED_SEQWRITING) {
    534   1.1   reinoud 			/* TODO use flag to signal if this is needed */
    535   1.1   reinoud 			mutex_exit(&priv->discstrat_mutex);
    536   1.1   reinoud 
    537   1.1   reinoud 			/* update trackinfo for data and metadata */
    538   1.1   reinoud 			error = udf_update_trackinfo(ump,
    539   1.1   reinoud 					&ump->data_track);
    540   1.1   reinoud 			assert(error == 0);
    541   1.1   reinoud 			error = udf_update_trackinfo(ump,
    542   1.1   reinoud 					&ump->metadata_track);
    543   1.1   reinoud 			assert(error == 0);
    544   1.1   reinoud 			mutex_enter(&priv->discstrat_mutex);
    545  1.13  christos 			__USE(error);
    546   1.1   reinoud 		}
    547   1.1   reinoud 	}
    548   1.1   reinoud 
    549   1.1   reinoud 	if (new_queue != priv->cur_queue) {
    550   1.1   reinoud 		DPRINTF(SHEDULE, ("switching from %d to %d\n",
    551   1.1   reinoud 			priv->cur_queue, new_queue));
    552  1.16   reinoud 		if (new_queue == UDF_SHED_READING)
    553  1.16   reinoud 			udf_mmc_synchronise_caches(ump);
    554   1.1   reinoud 	}
    555   1.1   reinoud 
    556   1.1   reinoud 	priv->cur_queue = new_queue;
    557   1.1   reinoud }
    558   1.1   reinoud 
    559   1.1   reinoud 
    560   1.1   reinoud static void
    561   1.1   reinoud udf_discstrat_thread(void *arg)
    562   1.1   reinoud {
    563   1.1   reinoud 	struct udf_mount *ump = (struct udf_mount *) arg;
    564   1.1   reinoud 	struct strat_private *priv = PRIV(ump);
    565   1.1   reinoud 	int empty;
    566   1.1   reinoud 
    567   1.1   reinoud 	empty = 1;
    568   1.1   reinoud 	mutex_enter(&priv->discstrat_mutex);
    569  1.15   reinoud 	while (priv->run_thread || !empty || priv->sync_req) {
    570   1.1   reinoud 		/* process the current selected queue */
    571   1.1   reinoud 		udf_doshedule(ump);
    572   1.7      yamt 		empty  = (bufq_peek(priv->queues[UDF_SHED_READING]) == NULL);
    573   1.7      yamt 		empty &= (bufq_peek(priv->queues[UDF_SHED_WRITING]) == NULL);
    574   1.7      yamt 		empty &= (bufq_peek(priv->queues[UDF_SHED_SEQWRITING]) == NULL);
    575   1.1   reinoud 
    576   1.1   reinoud 		/* wait for more if needed */
    577  1.15   reinoud 		if (empty) {
    578  1.15   reinoud 			if (priv->sync_req) {
    579  1.15   reinoud 				/* on sync, we need to simulate a read->write transition */
    580  1.15   reinoud 				udf_mmc_synchronise_caches(ump);
    581  1.15   reinoud 				priv->cur_queue = UDF_SHED_READING;
    582  1.15   reinoud 				priv->sync_req = 0;
    583  1.15   reinoud 			}
    584   1.1   reinoud 			cv_timedwait(&priv->discstrat_cv,
    585   1.1   reinoud 				&priv->discstrat_mutex, hz/8);
    586  1.15   reinoud 		}
    587   1.1   reinoud 	}
    588   1.1   reinoud 	mutex_exit(&priv->discstrat_mutex);
    589   1.1   reinoud 
    590   1.1   reinoud 	wakeup(&priv->run_thread);
    591   1.1   reinoud 	kthread_exit(0);
    592   1.1   reinoud 	/* not reached */
    593   1.1   reinoud }
    594   1.1   reinoud 
    595   1.1   reinoud /* --------------------------------------------------------------------- */
    596   1.1   reinoud 
    597   1.1   reinoud static void
    598   1.1   reinoud udf_discstrat_init_seq(struct udf_strat_args *args)
    599   1.1   reinoud {
    600   1.1   reinoud 	struct udf_mount *ump = args->ump;
    601   1.1   reinoud 	struct strat_private *priv = PRIV(ump);
    602   1.1   reinoud 	struct disk_strategy dkstrat;
    603   1.1   reinoud 	uint32_t lb_size;
    604   1.1   reinoud 
    605   1.1   reinoud 	KASSERT(ump);
    606   1.1   reinoud 	KASSERT(ump->logical_vol);
    607   1.1   reinoud 	KASSERT(priv == NULL);
    608   1.1   reinoud 
    609   1.1   reinoud 	lb_size = udf_rw32(ump->logical_vol->lb_size);
    610   1.1   reinoud 	KASSERT(lb_size > 0);
    611   1.1   reinoud 
    612   1.1   reinoud 	/* initialise our memory space */
    613   1.1   reinoud 	ump->strategy_private = malloc(sizeof(struct strat_private),
    614   1.1   reinoud 		M_UDFTEMP, M_WAITOK);
    615   1.1   reinoud 	priv = ump->strategy_private;
    616   1.1   reinoud 	memset(priv, 0 , sizeof(struct strat_private));
    617   1.1   reinoud 
    618   1.1   reinoud 	/* initialise locks */
    619   1.1   reinoud 	cv_init(&priv->discstrat_cv, "udfstrat");
    620   1.1   reinoud 	mutex_init(&priv->discstrat_mutex, MUTEX_DEFAULT, IPL_NONE);
    621   1.1   reinoud 
    622   1.1   reinoud 	/*
    623   1.1   reinoud 	 * Initialise pool for descriptors associated with nodes. This is done
    624   1.1   reinoud 	 * in lb_size units though currently lb_size is dictated to be
    625   1.1   reinoud 	 * sector_size.
    626   1.1   reinoud 	 */
    627   1.1   reinoud 	pool_init(&priv->desc_pool, lb_size, 0, 0, 0, "udf_desc_pool", NULL,
    628   1.1   reinoud 	    IPL_NONE);
    629   1.1   reinoud 
    630   1.1   reinoud 	/*
    631   1.1   reinoud 	 * remember old device strategy method and explicit set method
    632   1.1   reinoud 	 * `discsort' since we have our own more complex strategy that is not
    633   1.1   reinoud 	 * implementable on the CD device and other strategies will get in the
    634   1.1   reinoud 	 * way.
    635   1.1   reinoud 	 */
    636   1.1   reinoud 	memset(&priv->old_strategy_setting, 0,
    637   1.1   reinoud 		sizeof(struct disk_strategy));
    638   1.1   reinoud 	VOP_IOCTL(ump->devvp, DIOCGSTRATEGY, &priv->old_strategy_setting,
    639   1.1   reinoud 		FREAD | FKIOCTL, NOCRED);
    640   1.1   reinoud 	memset(&dkstrat, 0, sizeof(struct disk_strategy));
    641   1.1   reinoud 	strcpy(dkstrat.dks_name, "discsort");
    642   1.1   reinoud 	VOP_IOCTL(ump->devvp, DIOCSSTRATEGY, &dkstrat, FWRITE | FKIOCTL,
    643   1.1   reinoud 		NOCRED);
    644   1.1   reinoud 
    645  1.17    andvar 	/* initialise our internal scheduler */
    646   1.1   reinoud 	priv->cur_queue = UDF_SHED_READING;
    647   1.1   reinoud 	bufq_alloc(&priv->queues[UDF_SHED_READING], "disksort",
    648   1.1   reinoud 		BUFQ_SORT_RAWBLOCK);
    649   1.1   reinoud 	bufq_alloc(&priv->queues[UDF_SHED_WRITING], "disksort",
    650   1.1   reinoud 		BUFQ_SORT_RAWBLOCK);
    651   1.1   reinoud 	bufq_alloc(&priv->queues[UDF_SHED_SEQWRITING], "fcfs", 0);
    652   1.1   reinoud 	vfs_timestamp(&priv->last_queued[UDF_SHED_READING]);
    653   1.1   reinoud 	vfs_timestamp(&priv->last_queued[UDF_SHED_WRITING]);
    654   1.1   reinoud 	vfs_timestamp(&priv->last_queued[UDF_SHED_SEQWRITING]);
    655   1.1   reinoud 
    656   1.1   reinoud 	/* create our disk strategy thread */
    657   1.1   reinoud 	priv->run_thread = 1;
    658  1.15   reinoud 	priv->sync_req   = 0;
    659   1.1   reinoud 	if (kthread_create(PRI_NONE, 0 /* KTHREAD_MPSAFE*/, NULL /* cpu_info*/,
    660   1.1   reinoud 		udf_discstrat_thread, ump, &priv->queue_lwp,
    661   1.1   reinoud 		"%s", "udf_rw")) {
    662   1.1   reinoud 		panic("fork udf_rw");
    663   1.1   reinoud 	}
    664   1.1   reinoud }
    665   1.1   reinoud 
    666   1.1   reinoud 
    667   1.1   reinoud static void
    668   1.1   reinoud udf_discstrat_finish_seq(struct udf_strat_args *args)
    669   1.1   reinoud {
    670   1.1   reinoud 	struct udf_mount *ump = args->ump;
    671   1.1   reinoud 	struct strat_private *priv = PRIV(ump);
    672   1.1   reinoud 	int error;
    673   1.1   reinoud 
    674   1.1   reinoud 	if (ump == NULL)
    675   1.1   reinoud 		return;
    676   1.1   reinoud 
    677  1.17    andvar 	/* stop our scheduling thread */
    678   1.1   reinoud 	KASSERT(priv->run_thread == 1);
    679   1.1   reinoud 	priv->run_thread = 0;
    680   1.1   reinoud 	wakeup(priv->queue_lwp);
    681   1.1   reinoud 	do {
    682   1.1   reinoud 		error = tsleep(&priv->run_thread, PRIBIO+1,
    683   1.1   reinoud 			"udfshedfin", hz);
    684   1.1   reinoud 	} while (error);
    685   1.1   reinoud 	/* kthread should be finished now */
    686   1.1   reinoud 
    687   1.1   reinoud 	/* set back old device strategy method */
    688   1.1   reinoud 	VOP_IOCTL(ump->devvp, DIOCSSTRATEGY, &priv->old_strategy_setting,
    689   1.1   reinoud 			FWRITE, NOCRED);
    690   1.1   reinoud 
    691   1.1   reinoud 	/* destroy our pool */
    692   1.1   reinoud 	pool_destroy(&priv->desc_pool);
    693   1.1   reinoud 
    694  1.11  drochner 	mutex_destroy(&priv->discstrat_mutex);
    695  1.11  drochner 	cv_destroy(&priv->discstrat_cv);
    696  1.11  drochner 
    697   1.1   reinoud 	/* free our private space */
    698   1.1   reinoud 	free(ump->strategy_private, M_UDFTEMP);
    699   1.1   reinoud 	ump->strategy_private = NULL;
    700   1.1   reinoud }
    701   1.1   reinoud 
    702   1.1   reinoud /* --------------------------------------------------------------------- */
    703   1.1   reinoud 
    704   1.1   reinoud struct udf_strategy udf_strat_sequential =
    705   1.1   reinoud {
    706   1.1   reinoud 	udf_create_logvol_dscr_seq,
    707   1.1   reinoud 	udf_free_logvol_dscr_seq,
    708   1.1   reinoud 	udf_read_logvol_dscr_seq,
    709   1.1   reinoud 	udf_write_logvol_dscr_seq,
    710   1.1   reinoud 	udf_queuebuf_seq,
    711  1.15   reinoud 	udf_sync_caches_seq,
    712   1.1   reinoud 	udf_discstrat_init_seq,
    713   1.1   reinoud 	udf_discstrat_finish_seq
    714   1.1   reinoud };
    715  1.18     skrll 
    716   1.1   reinoud 
    717