rf_kintf.c

RAIDFrame是个非常好的磁盘阵列RAID仿真工具

/*
 * Copyright (c) 1995 Carnegie-Mellon University.
 * All rights reserved.
 *
 * Authors: Mark Holland, Jim Zelenka
 *
 * Permission to use, copy, modify and distribute this software and
 * its documentation is hereby granted, provided that both the copyright
 * notice and this permission notice appear in all copies of the
 * software, derivative works or modified versions, and any portions
 * thereof, and that both notices appear in supporting documentation.
 *
 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
 * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
 *
 * Carnegie Mellon requests users of this software to return to
 *
 *  Software Distribution Coordinator  or  [e-mail]
 *  School of Computer Science
 *  Carnegie Mellon University
 *  Pittsburgh PA 15213-3890
 *
 * any improvements or extensions that they make and grant Carnegie the
 * rights to redistribute these changes.
 */

/***********************************************************
 *
 * rf_kintf.c -- the kernel interface routines for RAIDframe
 *
 ***********************************************************/
/*
 * $Locker:  $
 * $Log: rf_kintf.c,v $
 * Revision 1.57  1996/07/19  16:12:20  jimz
 * remove addition of protectedSectors in InitBP- it's already
 * done in the diskqueue code
 *
 * Revision 1.56  1996/07/17  21:00:58  jimz
 * clean up timer interface, tracing
 *
 * Revision 1.55  1996/06/17  03:00:54  jimz
 * Change RAIDFRAME_GET_INFO interface to do its own copyout()
 * (because size of device config structure now exceeds 8k)
 *
 * Revision 1.54  1996/06/09  02:36:46  jimz
 * lots of little crufty cleanup- fixup whitespace
 * issues, comment #ifdefs, improve typing in some
 * places (esp size-related)
 *
 * Revision 1.53  1996/06/07  21:33:04  jimz
 * begin using consistent types for sector numbers,
 * stripe numbers, row+col numbers, recon unit numbers
 *
 * Revision 1.52  1996/06/06  17:28:08  jimz
 * track sector number of last I/O dequeued
 *
 * Revision 1.51  1996/06/05  18:06:02  jimz
 * Major code cleanup. The Great Renaming is now done.
 * Better modularity. Better typing. Fixed a bunch of
 * synchronization bugs. Made a lot of global stuff
 * per-desc or per-array. Removed dead code.
 *
 * Revision 1.50  1996/06/03  23:28:26  jimz
 * more bugfixes
 * check in tree to sync for IPDS runs with current bugfixes
 * there still may be a problem with threads in the script test
 * getting I/Os stuck- not trivially reproducible (runs ~50 times
 * in a row without getting stuck)
 *
 * Revision 1.49  1996/06/02  17:31:48  jimz
 * Moved a lot of global stuff into array structure, where it belongs.
 * Fixed up paritylogging, pss modules in this manner. Some general
 * code cleanup. Removed lots of dead code, some dead files.
 *
 * Revision 1.48  1996/05/31  22:26:54  jimz
 * fix a lot of mapping problems, memory allocation problems
 * found some weird lock issues, fixed 'em
 * more code cleanup
 *
 * Revision 1.47  1996/05/30  12:59:18  jimz
 * make etimer happier, more portable
 *
 * Revision 1.46  1996/05/30  11:29:41  jimz
 * Numerous bug fixes. Stripe lock release code disagreed with the taking code
 * about when stripes should be locked (I made it consistent: no parity, no lock)
 * There was a lot of extra serialization of I/Os which I've removed- a lot of
 * it was to calculate values for the cache code, which is no longer with us.
 * More types, function, macro cleanup. Added code to properly quiesce the array
 * on shutdown. Made a lot of stuff array-specific which was (bogusly) general
 * before. Fixed memory allocation, freeing bugs.
 *
 * Revision 1.45  1996/05/27  18:56:37  jimz
 * more code cleanup
 * better typing
 * compiles in all 3 environments
 *
 * Revision 1.44  1996/05/24  22:17:04  jimz
 * continue code + namespace cleanup
 * typed a bunch of flags
 *
 * Revision 1.43  1996/05/24  01:59:45  jimz
 * another checkpoint in code cleanup for release
 * time to sync kernel tree
 *
 * Revision 1.42  1996/05/23  22:17:54  jimz
 * fix sector size hardcoding problems
 *
 * Revision 1.41  1996/05/23  21:46:35  jimz
 * checkpoint in code cleanup (release prep)
 * lots of types, function names have been fixed
 *
 * Revision 1.40  1996/05/23  13:18:07  jimz
 * tracing_mutex -> rf_tracing_mutex
 *
 * Revision 1.39  1996/05/23  00:33:23  jimz
 * code cleanup: move all debug decls to rf_options.c, all extern
 * debug decls to rf_options.h, all debug vars preceded by rf_
 *
 * Revision 1.38  1996/05/20  16:15:32  jimz
 * switch to rf_{mutex,cond}_{init,destroy}
 *
 * Revision 1.37  1996/05/10  16:23:47  jimz
 * RF_offset -> RF_Offset
 *
 * Revision 1.36  1996/05/08  21:01:24  jimz
 * fixed up enum type names that were conflicting with other
 * enums and function names (ie, "panic")
 * future naming trends will be towards RF_ and rf_ for
 * everything raidframe-related
 *
 * Revision 1.35  1996/05/03  19:10:48  jimz
 * change sanity checking for bogus I/Os to return more appropriate
 * values (to make some user-level utilities happer with RAIDframe)
 *
 * Revision 1.34  1996/05/02  22:17:00  jimz
 * When using DKUSAGE, send a bogus IO after configuring to let DKUSAGE know
 * that we exist. This will let user-level programs doing group stats on the
 * RF device function without error before RF gets its first IO
 *
 * Changed rf_device_config devs and spares fields to RF_RaidDisk_t
 *
 * Inc numOutstanding for the disk queue in rf_DispatchKernelIO if
 * type is IO_TYPE_NOP. I'm not sure this is right, but it seems to be,
 * because the disk IO completion routine wants to dec it, and doesn't
 * care if there was no such IO.
 *
 * Revision 1.33  1996/05/02  15:05:44  jimz
 * for now, rf_DoAccessKernel will reject non-sector-sized I/Os
 * eventually, it should do something more clever...
 * (and do it in DoAccess(), not just DoAccessKernel())
 *
 * Revision 1.32  1996/05/01  16:28:39  jimz
 * get rid of uses of ccmn_ functions
 *
 * Revision 1.31  1996/05/01  15:42:17  jimz
 * ccmn_* memory management is on the way out. This is an archival checkpoint-
 * both the old and new code are in place (all the ccmn_ calls are #if 0). After
 * this, the ccmn_ code will no longer appear.
 *
 * Revision 1.30  1996/04/22  15:53:13  jimz
 * MAX_RAIDS -> NRAIDFRAME
 *
 * Revision 1.29  1995/12/12  18:10:06  jimz
 * MIN -> RF_MIN, MAX -> RF_MAX, ASSERT -> RF_ASSERT
 * fix 80-column brain damage in comments
 *
 * Revision 1.28  1995/12/01  19:11:01  root
 * added copyright info
 *
 * Revision 1.27  1995/11/28  18:56:40  wvcii
 * disabled buffer copy in rf_write
 *
 * Revision 1.26  1995/10/06  16:37:08  jimz
 * get struct bufs from ubc, not cam
 * copy all write data, and operate on copy
 * (temporary hack to get around dags in PQ that want
 * to Xor into user write buffers)
 *
 * Revision 1.25  1995/09/30  22:23:08  jimz
 * do not require raid to be active to perform ACCTOTAL ioctl
 *
 * Revision 1.24  1995/09/30  20:39:08  jimz
 * added new ioctls:
 *   RAIDFRAME_RESET_ACCTOTALS
 *   RAIDFRAME_GET_ACCTOTALS
 *   RAIDFRAME_KEEP_ACCTOTALS
 *
 * Revision 1.23  1995/09/20  21:11:59  jimz
 * include dfstrace.h in KERNEL block
 * (even though it's a kernel-only file, this makes the depend process
 * at user-level happy. Why the user-level Makefile wants to depend
 * kintf.c is less clear, but this is a workaround).
 *
 * Revision 1.22  1995/09/19  23:19:03  jimz
 * added DKUSAGE support
 *
 */

#ifdef KERNEL
#include <dkusage.h>
#include <dfstrace.h>
#include <raidframe.h>
#include <raidframe_recon.h>

#include <sys/errno.h>
#include <kern/task.h>
#include <kern/thread.h>
#include <io/common/iotypes.h>
#include <io/cam/cam_debug.h>
#include <io/cam/cam.h>
#include <io/cam/dec_cam.h>
#include <io/cam/uagt.h>
#include <sys/conf.h>
#include <io/common/devdriver.h>
#include <io/cam/pdrv.h>
#include <io/common/pt.h>
#include <sys/disklabel.h>
#include <io/cam/cam_disk.h>
#include <io/common/devdriver.h>
#include <sys/buf.h>
#include <sys/user.h>
#include "rf_raid.h"
#include "rf_raidframe.h"
#include "rf_dag.h"
#include "rf_dagflags.h"
#include "rf_diskqueue.h"
#include "rf_acctrace.h"
#include "rf_etimer.h"
#include "rf_general.h"
#include "rf_debugMem.h"
#include "rf_kintf.h"
#include "rf_options.h"

#if DKUSAGE > 0
#include <sys/dkusage.h>
#include <io/common/iotypes.h>
#include <io/cam/dec_cam.h>
#include <io/cam/cam.h>
#include <io/cam/pdrv.h>
#endif /* DKUSAGE > 0 */

#define DEBUG

static int rf_kdebug_level = 0;

#define RFK_BOOT_NONE 0
#define RFK_BOOT_GOOD 1
#define RFK_BOOT_BAD  2
static int rf_kbooted = RFK_BOOT_NONE;

extern struct buf *ubc_bufget();

#ifdef DEBUG
#define db0_printf(a) printf a
#define db_printf(a) if (rf_kdebug_level > 0) printf a
#define db1_printf(a) if (rf_kdebug_level > 0) printf a
#define db2_printf(a) if (rf_kdebug_level > 1) printf a
#define db3_printf(a) if (rf_kdebug_level > 2) printf a
#define db4_printf(a) if (rf_kdebug_level > 3) printf a
#define db5_printf(a) if (rf_kdebug_level > 4) printf a
#else /* DEBUG */
#define db0_printf(a) printf a
#define db1_printf(a) { }
#define db2_printf(a) { }
#define db3_printf(a) { }
#define db4_printf(a) { }
#define db5_printf(a) { }
#endif /* DEBUG */

static RF_Raid_t *raidPtrs[NRAIDFRAME]; /* global raid device descriptors */

static int rf_pending_testaccs;

RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex)
RF_DECLARE_STATIC_MUTEX(rf_async_done_q_mutex)
static RF_SparetWait_t *rf_sparet_wait_queue;      /* requests to install a spare table */
static RF_SparetWait_t *rf_sparet_resp_queue;      /* responses from installation process */
static struct rf_test_acc    *rf_async_done_qh, *rf_async_done_qt;

static struct rf_recon_req *recon_queue = NULL;          /* used to communicate reconstruction requests */
decl_simple_lock_data(,recon_queue_mutex)
#define LOCK_RECON_Q_MUTEX() simple_lock(&recon_queue_mutex)
#define UNLOCK_RECON_Q_MUTEX() simple_unlock(&recon_queue_mutex)

/* prototypes */
static int rf_WrapUpTestAcc(struct rf_test_acc *ta, RF_Raid_t *raidPtr);
static struct rf_test_acc *rf_DupTestAccDesc(struct rf_test_acc *ta);
static void rf_AsyncTestAccCallbackFunc(struct rf_test_acc *ta);
static void KernelWakeupFunc(struct buf  *bp);
static void InitBP(struct buf *bp, unsigned rw_flag, dev_t dev,
	RF_SectorNum_t startSect, RF_SectorCount_t numSect, caddr_t buf,
	void (*cbFunc)(), void *cbArg, int logBytesPerSector,
	struct proc *b_proc);

#define Dprintf0(s)       if (rf_queueDebug) rf_debug_printf(s,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL)
#define Dprintf1(s,a)     if (rf_queueDebug) rf_debug_printf(s,a,NULL,NULL,NULL,NULL,NULL,NULL,NULL)
#define Dprintf2(s,a,b)   if (rf_queueDebug) rf_debug_printf(s,a,b,NULL,NULL,NULL,NULL,NULL,NULL)
#define Dprintf3(s,a,b,c) if (rf_queueDebug) rf_debug_printf(s,a,b,c,NULL,NULL,NULL,NULL,NULL)

#define rf_FreeTestAccDesc(x) RF_Free((x),sizeof(struct rf_test_acc))

/* this is so that we can compile under 2.0 as well as 3.2 */
#ifndef proc_to_task
#define proc_to_task(x) ((x)->task)
#endif /* !proc_to_task */

/*********************************************************
 *
 * initialization code called at boot time (startup.c)
 *
 ********************************************************/
int rf_boot()
{
  int i, rc;

  rc = rf_mutex_init(&rf_sparet_wait_mutex);
  if (rc) {
    RF_PANIC();
  }
  rc = rf_mutex_init(&rf_async_done_q_mutex);
  if (rc) {
    RF_PANIC();
  }
  rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
  recon_queue = NULL;
  rf_async_done_qh = rf_async_done_qt = NULL;
  for (i=0; i<NRAIDFRAME; i++)
    raidPtrs[i] = NULL;
  rc = rf_BootRaidframe();
  if (rc == 0)
    printf("Kernelized RAIDframe activated\n");
  else
    rf_kbooted = RFK_BOOT_BAD;
  return(rc);
}

/*********************************************************
 *
 * the entries in the block and character device switches
 *
 ********************************************************/

/*
 * We don't check if raidframe is configured on open, because open is called
 * for the ioctl which does the configuration.
 *
 * A RAID minor number consists of the partition number in the low
 * six bits, and the array identifier above that.
 */
int rf_open(dev, flag, fmt)
  dev_t  dev;
  int    flag;
  int    fmt;
{
	unsigned int raidID;
	int rc;

	if (rf_kbooted == RFK_BOOT_NONE) {
		rf_kbooted = RFK_BOOT_GOOD;
		rc = rf_boot();
		if (rc) {
			rf_kbooted = RFK_BOOT_BAD;
			return(rc);
		}
	}
	db2_printf(("rf_open(%lx %d %d)\n", (u_long)dev, flag, fmt));

	raidID = RF_DEV2RAIDID(dev);
	db5_printf(("rf_open: raidID=%d\n", raidID));
	if (raidID >= NRAIDFRAME)
		return(ENODEV);

	db5_printf(("rf_open: raidPtrs[%d]=%lx\n", raidID, raidPtrs[raidID]));
	if (!raidPtrs[raidID]) {
		RF_Calloc(raidPtrs[raidID], 1, sizeof(RF_Raid_t), (RF_Raid_t *));
	}
	if (!raidPtrs[raidID])
		return(ENOMEM);
	db4_printf(("rf_open: raidPtrs[%d]=%lx post-alloc\n", raidID, raidPtrs[raidID]));

	return(0);
}

int rf_close(dev, flag, fmt)
  dev_t  dev;
  int    flag;
  int    fmt;
{
  unsigned int raidID;

  if (rf_kbooted != RFK_BOOT_GOOD) {
    return(EINVAL);
  }
  raidID = RF_DEV2RAIDID(dev);
  if (raidID >= NRAIDFRAME)
    return(ENODEV);
  if (!raidPtrs[raidID])
    return(EINVAL);

  return(0);
}

void rf_strategy(bp)
  struct buf   *bp;
{
	unsigned int raidID = RF_DEV2RAIDID(bp->b_dev);
	RF_Raid_t *raidPtr;

	if (rf_kbooted != RFK_BOOT_GOOD)