freebsd-nq/sys/dev/vinum/vinuminterrupt.c
Greg Lehey bd131be7d1 complete_rqe: Remove a race condition in RAID-4 and RAID-5 where a
request could be deallocated before the top half had finished
	issuing it.  The problem seems only to happen with IDE drives
	and vn devices, but theoretically it could happen with any
	drive.  This is the most important part of a possible series
	of fixes designed to remove race conditions without locking
	out interrupts for longer than absolutely necessary.

Reported-by:	       sos
Fix-supplied-by:       dillon
2000-04-06 03:03:31 +00:00

419 lines
17 KiB
C

/* vinuminterrupt.c: bottom half of the driver */
/*-
* Copyright (c) 1997, 1998, 1999
* Nan Yang Computer Services Limited. All rights reserved.
*
* Parts copyright (c) 1997, 1998 Cybernet Corporation, NetMAX project.
*
* Written by Greg Lehey
*
* This software is distributed under the so-called ``Berkeley
* License'':
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Nan Yang Computer
* Services Limited.
* 4. Neither the name of the Company nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* This software is provided ``as is'', and any express or implied
* warranties, including, but not limited to, the implied warranties of
* merchantability and fitness for a particular purpose are disclaimed.
* In no event shall the company or contributors be liable for any
* direct, indirect, incidental, special, exemplary, or consequential
* damages (including, but not limited to, procurement of substitute
* goods or services; loss of use, data, or profits; or business
* interruption) however caused and on any theory of liability, whether
* in contract, strict liability, or tort (including negligence or
* otherwise) arising in any way out of the use of this software, even if
* advised of the possibility of such damage.
*
* $Id: vinuminterrupt.c,v 1.9 2000/02/16 01:59:02 grog Exp grog $
* $FreeBSD$
*/
#include <dev/vinum/vinumhdr.h>
#include <dev/vinum/request.h>
#include <sys/resourcevar.h>
void complete_raid5_write(struct rqelement *);
void complete_rqe(struct buf *bp);
void sdio_done(struct buf *bp);
/*
* Take a completed buffer, transfer the data back if
* it's a read, and complete the high-level request
* if this is the last subrequest.
*
* The bp parameter is in fact a struct rqelement, which
* includes a couple of extras at the end.
*/
void
complete_rqe(struct buf *bp)
{
struct rqelement *rqe;
struct request *rq;
struct rqgroup *rqg;
struct buf *ubp; /* user buffer */
struct drive *drive;
rqe = (struct rqelement *) bp; /* point to the element element that completed */
rqg = rqe->rqg; /* and the request group */
rq = rqg->rq; /* and the complete request */
ubp = rq->bp; /* user buffer */
#ifdef VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_iodone, (union rqinfou) rqe, ubp);
#endif
drive = &DRIVE[rqe->driveno];
drive->active--; /* one less outstanding I/O on this drive */
vinum_conf.active--; /* one less outstanding I/O globally */
if ((drive->active == (DRIVE_MAXACTIVE - 1)) /* we were at the drive limit */
||(vinum_conf.active == VINUM_MAXACTIVE)) /* or the global limit */
wakeup(&launch_requests); /* let another one at it */
if ((bp->b_ioflags & BIO_ERROR) != 0) { /* transfer in error */
if (bp->b_error != 0) /* did it return a number? */
rq->error = bp->b_error; /* yes, put it in. */
else if (rq->error == 0) /* no: do we have one already? */
rq->error = EIO; /* no: catchall "I/O error" */
SD[rqe->sdno].lasterror = rq->error;
if (bp->b_iocmd == BIO_READ) {
log(LOG_ERR, "%s: fatal read I/O error\n", SD[rqe->sdno].name);
set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* subdisk is crashed */
} else { /* write operation */
log(LOG_ERR, "%s: fatal write I/O error\n", SD[rqe->sdno].name);
set_sd_state(rqe->sdno, sd_stale, setstate_force); /* subdisk is stale */
}
if (rq->error == ENXIO) { /* the drive's down too */
log(LOG_ERR, "%s: fatal drive I/O error\n", DRIVE[rqe->driveno].label.name);
DRIVE[rqe->driveno].lasterror = rq->error;
set_drive_state(rqe->driveno, /* take the drive down */
drive_down,
setstate_force);
}
}
/* Now update the statistics */
if (bp->b_iocmd == BIO_READ) { /* read operation */
DRIVE[rqe->driveno].reads++;
DRIVE[rqe->driveno].bytes_read += bp->b_bcount;
SD[rqe->sdno].reads++;
SD[rqe->sdno].bytes_read += bp->b_bcount;
PLEX[rqe->rqg->plexno].reads++;
PLEX[rqe->rqg->plexno].bytes_read += bp->b_bcount;
if (PLEX[rqe->rqg->plexno].volno >= 0)
VOL[PLEX[rqe->rqg->plexno].volno].bytes_read += bp->b_bcount;
} else { /* write operation */
DRIVE[rqe->driveno].writes++;
DRIVE[rqe->driveno].bytes_written += bp->b_bcount;
SD[rqe->sdno].writes++;
SD[rqe->sdno].bytes_written += bp->b_bcount;
PLEX[rqe->rqg->plexno].writes++;
PLEX[rqe->rqg->plexno].bytes_written += bp->b_bcount;
if (PLEX[rqe->rqg->plexno].volno >= 0)
VOL[PLEX[rqe->rqg->plexno].volno].bytes_written += bp->b_bcount;
}
if (rqg->flags & XFR_RECOVERY_READ) { /* recovery read, */
int *sdata; /* source */
int *data; /* and group data */
int length; /* and count involved */
int count; /* loop counter */
struct rqelement *urqe = &rqg->rqe[rqg->badsdno]; /* rqe of the bad subdisk */
/* XOR destination is the user data */
sdata = (int *) &rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]; /* old data contents */
data = (int *) &urqe->b.b_data[urqe->groupoffset << DEV_BSHIFT]; /* destination */
length = urqe->grouplen << (DEV_BSHIFT - 2); /* and count involved */
for (count = 0; count < length; count++)
data[count] ^= sdata[count];
/*
* In a normal read, we will normally read directly
* into the user buffer. This doesn't work if
* we're also doing a recovery, so we have to
* copy it
*/
if (rqe->flags & XFR_NORMAL_READ) { /* normal read as well, */
char *src = &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* read data is here */
char *dst;
dst = (char *) ubp->b_data + (rqe->useroffset << DEV_BSHIFT); /* where to put it in user buffer */
length = rqe->datalen << DEV_BSHIFT; /* and count involved */
bcopy(src, dst, length); /* move it */
}
} else if ((rqg->flags & (XFR_NORMAL_WRITE | XFR_DEGRADED_WRITE)) /* RAID 4/5 group write operation */
&&(rqg->active == 1)) /* and this is the last rq of phase 1 */
complete_raid5_write(rqe);
rqg->active--; /* one less request active */
if (rqg->active == 0) { /* request group finished, */
rq->active--; /* one less */
if (rqg->lock) { /* got a lock? */
unlockrange(rqg->plexno, rqg->lock); /* yes, free it */
rqg->lock = 0;
}
}
if (rq->active == 0) { /* request finished, */
#if VINUMDEBUG
if (debug & DEBUG_RESID) {
if (ubp->b_resid != 0) /* still something to transfer? */
Debugger("resid");
}
#endif
if (rq->error) { /* did we have an error? */
if (rq->isplex) { /* plex operation, */
ubp->b_ioflags |= BIO_ERROR; /* yes, propagate to user */
ubp->b_error = rq->error;
} else /* try to recover */
queue_daemon_request(daemonrq_ioerror, (union daemoninfo) rq); /* let the daemon complete */
} else {
ubp->b_resid = 0; /* completed our transfer */
if (rq->isplex == 0) /* volume request, */
VOL[rq->volplex.volno].active--; /* another request finished */
biodone(ubp); /* top level buffer completed */
freerq(rq); /* return the request storage */
}
}
}
/* Free a request block and anything hanging off it */
void
freerq(struct request *rq)
{
struct rqgroup *rqg;
struct rqgroup *nrqg; /* next in chain */
int rqno;
for (rqg = rq->rqg; rqg != NULL; rqg = nrqg) { /* through the whole request chain */
if (rqg->lock) /* got a lock? */
unlockrange(rqg->plexno, rqg->lock); /* yes, free it */
for (rqno = 0; rqno < rqg->count; rqno++)
if ((rqg->rqe[rqno].flags & XFR_MALLOCED) /* data buffer was malloced, */
&&rqg->rqe[rqno].b.b_data) /* and the allocation succeeded */
Free(rqg->rqe[rqno].b.b_data); /* free it */
nrqg = rqg->next; /* note the next one */
Free(rqg); /* and free this one */
}
Free(rq); /* free the request itself */
}
/* I/O on subdisk completed */
void
sdio_done(struct buf *bp)
{
struct sdbuf *sbp;
sbp = (struct sdbuf *) bp;
if (sbp->b.b_ioflags & BIO_ERROR) { /* had an error */
sbp->bp->b_ioflags |= BIO_ERROR; /* propagate upwards */
sbp->bp->b_error = sbp->b.b_error;
}
#ifdef VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_sdiodone, (union rqinfou) bp, bp);
#endif
sbp->bp->b_resid = sbp->b.b_resid; /* copy the resid field */
/* Now update the statistics */
if (bp->b_iocmd == BIO_READ) { /* read operation */
DRIVE[sbp->driveno].reads++;
DRIVE[sbp->driveno].bytes_read += sbp->b.b_bcount;
SD[sbp->sdno].reads++;
SD[sbp->sdno].bytes_read += sbp->b.b_bcount;
} else { /* write operation */
DRIVE[sbp->driveno].writes++;
DRIVE[sbp->driveno].bytes_written += sbp->b.b_bcount;
SD[sbp->sdno].writes++;
SD[sbp->sdno].bytes_written += sbp->b.b_bcount;
}
biodone(sbp->bp); /* complete the caller's I/O */
Free(sbp);
}
/* Start the second phase of a RAID-4 or RAID-5 group write operation. */
void
complete_raid5_write(struct rqelement *rqe)
{
int *sdata; /* source */
int *pdata; /* and parity block data */
int length; /* and count involved */
int count; /* loop counter */
int rqno; /* request index */
int rqoffset; /* offset of request data from parity data */
struct buf *bp; /* user buffer header */
struct request *rq; /* pointer to our request */
struct rqgroup *rqg; /* and to the request group */
struct rqelement *prqe; /* point to the parity block */
struct drive *drive; /* drive to access */
rqg = rqe->rqg; /* and to our request group */
rq = rqg->rq; /* point to our request */
bp = rq->bp; /* user's buffer header */
prqe = &rqg->rqe[0]; /* point to the parity block */
/*
* If we get to this function, we have normal or
* degraded writes, or a combination of both. We do
* the same thing in each case: we perform an
* exclusive or to the parity block. The only
* difference is the origin of the data and the
* address range.
*/
if (rqe->flags & XFR_DEGRADED_WRITE) { /* do the degraded write stuff */
pdata = (int *) (&prqe->b.b_data[(prqe->groupoffset) << DEV_BSHIFT]); /* parity data pointer */
bzero(pdata, prqe->grouplen << DEV_BSHIFT); /* start with nothing in the parity block */
/* Now get what data we need from each block */
for (rqno = 1; rqno < rqg->count; rqno++) { /* for all the data blocks */
/*
* This can do with improvement. If we're doing
* both a degraded and a normal write, we don't
* need to xor (nor to read) the part of the block
* that we're going to overwrite. FIXME XXX
*/
rqe = &rqg->rqe[rqno]; /* this request */
sdata = (int *) (&rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]); /* old data */
length = rqe->grouplen << (DEV_BSHIFT - 2); /* and count involved */
/*
* add the data block to the parity block. Before
* we started the request, we zeroed the parity
* block, so the result of adding all the other
* blocks and the block we want to write will be
* the correct parity block.
*/
for (count = 0; count < length; count++)
pdata[count] ^= sdata[count];
if ((rqe->flags & XFR_MALLOCED) /* the buffer was malloced, */
&&((rqg->flags & XFR_NORMAL_WRITE) == 0)) { /* and we have no normal write, */
Free(rqe->b.b_data); /* free it now */
rqe->flags &= ~XFR_MALLOCED;
}
}
}
if (rqg->flags & XFR_NORMAL_WRITE) { /* do normal write stuff */
/* Get what data we need from each block */
for (rqno = 1; rqno < rqg->count; rqno++) { /* for all the data blocks */
rqe = &rqg->rqe[rqno]; /* this request */
if ((rqe->flags & (XFR_DATA_BLOCK | XFR_BAD_SUBDISK | XFR_NORMAL_WRITE))
== (XFR_DATA_BLOCK | XFR_NORMAL_WRITE)) { /* good data block to write */
sdata = (int *) &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* old data contents */
rqoffset = rqe->dataoffset + rqe->sdoffset - prqe->sdoffset; /* corresponding parity block offset */
pdata = (int *) (&prqe->b.b_data[rqoffset << DEV_BSHIFT]); /* parity data pointer */
length = rqe->datalen << (DEV_BSHIFT - 2); /* and count involved */
/*
* "remove" the old data block
* from the parity block
*/
if ((pdata < ((int *) prqe->b.b_data))
|| (&pdata[length] > ((int *) (prqe->b.b_data + prqe->b.b_bcount)))
|| (sdata < ((int *) rqe->b.b_data))
|| (&sdata[length] > ((int *) (rqe->b.b_data + rqe->b.b_bcount))))
panic("complete_raid5_write: bounds overflow");
for (count = 0; count < length; count++)
pdata[count] ^= sdata[count];
/* "add" the new data block */
sdata = (int *) (&bp->b_data[rqe->useroffset << DEV_BSHIFT]); /* new data */
if ((sdata < ((int *) bp->b_data))
|| (&sdata[length] > ((int *) (bp->b_data + bp->b_bcount))))
panic("complete_raid5_write: bounds overflow");
for (count = 0; count < length; count++)
pdata[count] ^= sdata[count];
/* Free the malloced buffer */
if (rqe->flags & XFR_MALLOCED) { /* the buffer was malloced, */
Free(rqe->b.b_data); /* free it */
rqe->flags &= ~XFR_MALLOCED;
} else
panic("complete_raid5_write: malloc conflict");
if ((rqe->b.b_iocmd == BIO_READ) /* this was a read */
&&((rqe->flags & XFR_BAD_SUBDISK) == 0)) { /* and we can write this block */
rqe->b.b_flags &= ~B_DONE; /* we're not done */
rqe->b.b_iocmd = BIO_WRITE; /* we're writing now */
rqe->b.b_iodone = complete_rqe; /* call us here when done */
rqe->flags &= ~XFR_PARITYOP; /* reset flags that brought us here */
rqe->b.b_data = &bp->b_data[rqe->useroffset << DEV_BSHIFT]; /* point to the user data */
rqe->b.b_bcount = rqe->datalen << DEV_BSHIFT; /* length to write */
rqe->b.b_bufsize = rqe->b.b_bcount; /* don't claim more */
rqe->b.b_resid = rqe->b.b_bcount; /* nothing transferred */
rqe->b.b_blkno += rqe->dataoffset; /* point to the correct block */
rqg->active++; /* another active request */
drive = &DRIVE[rqe->driveno]; /* drive to access */
/* We can't sleep here, so we just increment the counters. */
drive->active++;
if (drive->active >= drive->maxactive)
drive->maxactive = drive->active;
vinum_conf.active++;
if (vinum_conf.active >= vinum_conf.maxactive)
vinum_conf.maxactive = vinum_conf.active;
#if VINUMDEBUG
if (debug & DEBUG_ADDRESSES)
log(LOG_DEBUG,
" %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n",
rqe->b.b_iocmd == BIO_READ ? "Read" : "Write",
major(rqe->b.b_dev),
minor(rqe->b.b_dev),
rqe->sdno,
(u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset),
rqe->b.b_blkno,
rqe->b.b_bcount);
if (debug & DEBUG_LASTREQS)
logrq(loginfo_raid5_data, (union rqinfou) rqe, bp);
#endif
DEV_STRATEGY(&rqe->b, 0);
}
}
}
}
/* Finally, write the parity block */
rqe = &rqg->rqe[0];
rqe->b.b_flags &= ~B_DONE; /* we're not done */
rqe->b.b_iocmd = BIO_WRITE; /* we're writing now */
rqe->b.b_iodone = complete_rqe; /* call us here when done */
rqg->flags &= ~XFR_PARITYOP; /* reset flags that brought us here */
rqe->b.b_bcount = rqe->buflen << DEV_BSHIFT; /* length to write */
rqe->b.b_bufsize = rqe->b.b_bcount; /* don't claim we have more */
rqe->b.b_resid = rqe->b.b_bcount; /* nothing transferred */
rqg->active++; /* another active request */
drive = &DRIVE[rqe->driveno]; /* drive to access */
/* We can't sleep here, so we just increment the counters. */
drive->active++;
if (drive->active >= drive->maxactive)
drive->maxactive = drive->active;
vinum_conf.active++;
if (vinum_conf.active >= vinum_conf.maxactive)
vinum_conf.maxactive = vinum_conf.active;
#if VINUMDEBUG
if (debug & DEBUG_ADDRESSES)
log(LOG_DEBUG,
" %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n",
rqe->b.b_iocmd == BIO_READ ? "Read" : "Write",
major(rqe->b.b_dev),
minor(rqe->b.b_dev),
rqe->sdno,
(u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset),
rqe->b.b_blkno,
rqe->b.b_bcount);
if (debug & DEBUG_LASTREQS)
logrq(loginfo_raid5_parity, (union rqinfou) rqe, bp);
#endif
DEV_STRATEGY(&rqe->b, 0);
}