void
moveobject(struct vinum_ioctl_msg *msg)
{
struct _ioctl_reply *reply = (struct _ioctl_reply *) msg;
struct drive *drive;
struct sd *sd;
/* Check that our objects are valid (i.e. they exist) */
drive = validdrive(msg->index, (struct _ioctl_reply *) msg);
if (drive == NULL)
return;
sd = validsd(msg->otherobject, (struct _ioctl_reply *) msg);
if (sd == NULL)
return;
if (sd->driveno == msg->index) /* sd already belongs to drive */
return;
if (sd->state > sd_stale)
set_sd_state(sd->sdno, sd_stale, setstate_force); /* make the subdisk stale */
else
sd->state = sd_empty;
if (sd->plexno >= 0) /* part of a plex, */
update_plex_state(sd->plexno); /* update its state */
/* Return the space on the old drive */
if ((sd->driveno >= 0) /* we have a drive, */
&&(sd->sectors > 0)) /* and some space on it */
return_drive_space(sd->driveno, /* return the space */
sd->driveoffset,
sd->sectors);
/* Reassign the old subdisk */
sd->driveno = msg->index;
sd->driveoffset = -1; /* let the drive decide where to put us */
give_sd_to_drive(sd->sdno);
reply->error = 0;
}
/*
* define the low-level requests needed to perform
* a high-level I/O operation for a specific plex
* 'plexno'.
*
* Return 0 if all subdisks involved in the
* request are up, 1 if some subdisks are not up,
* and -1 if the request is at least partially
* outside the bounds of the subdisks.
*
* Modify the pointer *diskstart to point to the
* end address. On read, return on the first bad
* subdisk, so that the caller
* (build_read_request) can try alternatives.
*
* On entry to this routine, the prq structures
* are not assigned. The assignment is performed
* by expandrq(). Strictly speaking, the elements
* rqe->sdno of all entries should be set to -1,
* since 0 (from bzero) is a valid subdisk number.
* We avoid this problem by initializing the ones
* we use, and not looking at the others (index >=
* prq->requests).
*/
enum requeststatus
bre5(struct request *rq,
int plexno,
daddr_t * diskaddr,
daddr_t diskend)
{
struct metrics m; /* most of the information */
struct sd *sd;
struct plex *plex;
struct buf *bp; /* user's bp */
struct rqgroup *rqg; /* the request group that we will create */
struct rqelement *rqe; /* point to this request information */
int rsectors; /* sectors remaining in this stripe */
int mysdno; /* another sd index in loops */
int rqno; /* request number */
rqg = NULL; /* shut up, damn compiler */
m.diskstart = *diskaddr; /* start of transfer */
bp = rq->bp; /* buffer pointer */
plex = &PLEX[plexno]; /* point to the plex */
while (*diskaddr < diskend) { /* until we get it all sorted out */
if (*diskaddr >= plex->length) /* beyond the end of the plex */
return REQUEST_EOF; /* can't continue */
m.badsdno = -1; /* no bad subdisk yet */
/* Part A: Define the request */
/*
* First, calculate some sizes:
* The offset of the start address from
* the start of the stripe.
*/
m.stripeoffset = *diskaddr % (plex->stripesize * (plex->subdisks - 1));
/*
* The plex-relative address of the
* start of the stripe.
*/
m.stripebase = *diskaddr - m.stripeoffset;
/* subdisk containing the parity stripe */
if (plex->organization == plex_raid5)
m.psdno = plex->subdisks - 1
- (*diskaddr / (plex->stripesize * (plex->subdisks - 1)))
% plex->subdisks;
else /* RAID-4 */
m.psdno = plex->subdisks - 1;
/*
* The number of the subdisk in which
* the start is located.
*/
m.firstsdno = m.stripeoffset / plex->stripesize;
if (m.firstsdno >= m.psdno) /* at or past parity sd */
m.firstsdno++; /* increment it */
/*
* The offset from the beginning of
* the stripe on this subdisk.
*/
m.initoffset = m.stripeoffset % plex->stripesize;
/* The offset of the stripe start relative to this subdisk */
m.sdbase = m.stripebase / (plex->subdisks - 1);
m.useroffset = *diskaddr - m.diskstart; /* The offset of the start in the user buffer */
/*
* The number of sectors to transfer in the
* current (first) subdisk.
*/
m.initlen = min(diskend - *diskaddr, /* the amount remaining to transfer */
plex->stripesize - m.initoffset); /* and the amount left in this block */
/*
* The number of sectors to transfer in this stripe
* is the minumum of the amount remaining to transfer
* and the amount left in this stripe.
*/
m.stripesectors = min(diskend - *diskaddr,
plex->stripesize * (plex->subdisks - 1) - m.stripeoffset);
/* The number of data subdisks involved in this request */
m.sdcount = (m.stripesectors + m.initoffset + plex->stripesize - 1) / plex->stripesize;
/* Part B: decide what kind of transfer this will be.
* start and end addresses of the transfer in
* the current block.
*
* There are a number of different kinds of
* transfer, each of which relates to a
* specific subdisk:
*
* 1. Normal read. All participating subdisks
* are up, and the transfer can be made
* directly to the user buffer. The bounds
* of the transfer are described by
* m.dataoffset and m.datalen. We have
* already calculated m.initoffset and
* m.initlen, which define the parameters
* for the first data block.
*
* 2. Recovery read. One participating
* subdisk is down. To recover data, all
* the other subdisks, including the parity
* subdisk, must be read. The data is
* recovered by exclusive-oring all the
* other blocks. The bounds of the
* transfer are described by m.groupoffset
* and m.grouplen.
*
* 3. A read request may request reading both
* available data (normal read) and
* non-available data (recovery read).
* This can be a problem if the address
* ranges of the two reads do not coincide:
* in this case, the normal read needs to
* be extended to cover the address range
* of the recovery read, and must thus be
* performed out of malloced memory.
*
* 4. Normal write. All the participating
* subdisks are up. The bounds of the
* transfer are described by m.dataoffset
* and m.datalen. Since these values
* differ for each block, we calculate the
* bounds for the parity block
* independently as the maximum of the
* individual blocks and store these values
* in m.writeoffset and m.writelen. This
* write proceeds in four phases:
*
* i. Read the old contents of each block
* and the parity block.
* ii. ``Remove'' the old contents from
* the parity block with exclusive or.
* iii. ``Insert'' the new contents of the
* block in the parity block, again
* with exclusive or.
*
* iv. Write the new contents of the data
* blocks and the parity block. The data
* block transfers can be made directly from
* the user buffer.
*
* 5. Degraded write where the data block is
* not available. The bounds of the
* transfer are described by m.groupoffset
* and m.grouplen. This requires the
* following steps:
*
* i. Read in all the other data blocks,
* excluding the parity block.
*
* ii. Recreate the parity block from the
* other data blocks and the data to be
* written.
*
* iii. Write the parity block.
*
* 6. Parityless write, a write where the
* parity block is not available. This is
* in fact the simplest: just write the
* data blocks. This can proceed directly
* from the user buffer. The bounds of the
* transfer are described by m.dataoffset
* and m.datalen.
*
* 7. Combination of degraded data block write
* and normal write. In this case the
* address ranges of the reads may also
* need to be extended to cover all
* participating blocks.
*
* All requests in a group transfer transfer
* the same address range relative to their
* subdisk. The individual transfers may
* vary, but since our group of requests is
* all in a single slice, we can define a
* range in which they all fall.
*
* In the following code section, we determine
* which kind of transfer we will perform. If
* there is a group transfer, we also decide
* its bounds relative to the subdisks. At
* the end, we have the following values:
*
* m.flags indicates the kinds of transfers
* we will perform.
* m.initoffset indicates the offset of the
* beginning of any data operation relative
* to the beginning of the stripe base.
* m.initlen specifies the length of any data
* operation.
* m.dataoffset contains the same value as
* m.initoffset.
* m.datalen contains the same value as
* m.initlen. Initially dataoffset and
* datalen describe the parameters for the
* first data block; while building the data
* block requests, they are updated for each
* block.
* m.groupoffset indicates the offset of any
* group operation relative to the beginning
* of the stripe base.
* m.grouplen specifies the length of any
* group operation.
* m.writeoffset indicates the offset of a
* normal write relative to the beginning of
* the stripe base. This value differs from
* m.dataoffset in that it applies to the
* entire operation, and not just the first
* block.
* m.writelen specifies the total span of a
* normal write operation. writeoffset and
* writelen are used to define the parity
* block.
*/
m.groupoffset = 0; /* assume no group... */
m.grouplen = 0; /* until we know we have one */
m.writeoffset = m.initoffset; /* start offset of transfer */
m.writelen = 0; /* nothing to write yet */
m.flags = 0; /* no flags yet */
rsectors = m.stripesectors; /* remaining sectors to examine */
m.dataoffset = m.initoffset; /* start at the beginning of the transfer */
m.datalen = m.initlen;
if (m.sdcount > 1) {
plex->multiblock++; /* more than one block for the request */
/*
* If we have two transfers that don't overlap,
* (one at the end of the first block, the other
* at the beginning of the second block),
* it's cheaper to split them.
*/
if (rsectors < plex->stripesize) {
m.sdcount = 1; /* just one subdisk */
m.stripesectors = m.initlen; /* and just this many sectors */
rsectors = m.initlen; /* and in the loop counter */
}
}
if (SD[plex->sdnos[m.psdno]].state < sd_reborn) /* is our parity subdisk down? */
m.badsdno = m.psdno; /* note that it's down */
if (bp->b_flags & B_READ) { /* read operation */
for (mysdno = m.firstsdno; rsectors > 0; mysdno++) {
if (mysdno == m.psdno) /* ignore parity on read */
mysdno++;
if (mysdno == plex->subdisks) /* wraparound */
mysdno = 0;
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
if (SD[plex->sdnos[mysdno]].state < sd_reborn) { /* got a bad subdisk, */
if (m.badsdno >= 0) /* we had one already, */
return REQUEST_DOWN; /* we can't take a second */
m.badsdno = mysdno; /* got the first */
m.groupoffset = m.dataoffset; /* define the bounds */
m.grouplen = m.datalen;
m.flags |= XFR_RECOVERY_READ; /* we need recovery */
plex->recovered_reads++; /* count another one */
} else
m.flags |= XFR_NORMAL_READ; /* normal read */
/* Update the pointers for the next block */
m.dataoffset = 0; /* back to the start of the stripe */
rsectors -= m.datalen; /* remaining sectors to examine */
m.datalen = min(rsectors, plex->stripesize); /* amount that will fit in this block */
}
} else { /* write operation */
for (mysdno = m.firstsdno; rsectors > 0; mysdno++) {
if (mysdno == m.psdno) /* parity stripe, we've dealt with that */
mysdno++;
if (mysdno == plex->subdisks) /* wraparound */
mysdno = 0;
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
sd = &SD[plex->sdnos[mysdno]];
if (sd->state != sd_up) {
enum requeststatus s;
s = checksdstate(sd, rq, *diskaddr, diskend); /* do we need to change state? */
if (s && (m.badsdno >= 0)) { /* second bad disk, */
int sdno;
/*
* If the parity disk is down, there's
* no recovery. We make all involved
* subdisks stale. Otherwise, we
* should be able to recover, but it's
* like pulling teeth. Fix it later.
*/
for (sdno = 0; sdno < m.sdcount; sdno++) {
struct sd *sd = &SD[plex->sdnos[sdno]];
if (sd->state >= sd_reborn) /* sort of up, */
set_sd_state(sd->sdno, sd_stale, setstate_force); /* make it stale */
}
return s; /* and crap out */
}
m.badsdno = mysdno; /* note which one is bad */
m.flags |= XFR_DEGRADED_WRITE; /* we need recovery */
plex->degraded_writes++; /* count another one */
m.groupoffset = m.dataoffset; /* define the bounds */
m.grouplen = m.datalen;
} else {
m.flags |= XFR_NORMAL_WRITE; /* normal write operation */
if (m.writeoffset > m.dataoffset) { /* move write operation lower */
m.writelen = max(m.writeoffset + m.writelen,
m.dataoffset + m.datalen)
- m.dataoffset;
m.writeoffset = m.dataoffset;
} else
m.writelen = max(m.writeoffset + m.writelen,
m.dataoffset + m.datalen)
- m.writeoffset;
}
/* Update the pointers for the next block */
m.dataoffset = 0; /* back to the start of the stripe */
rsectors -= m.datalen; /* remaining sectors to examine */
m.datalen = min(rsectors, plex->stripesize); /* amount that will fit in this block */
}
if (m.badsdno == m.psdno) { /* got a bad parity block, */
struct sd *psd = &SD[plex->sdnos[m.psdno]];
if (psd->state == sd_down)
set_sd_state(psd->sdno, sd_obsolete, setstate_force); /* it's obsolete now */
else if (psd->state == sd_crashed)
set_sd_state(psd->sdno, sd_stale, setstate_force); /* it's stale now */
m.flags &= ~XFR_NORMAL_WRITE; /* this write isn't normal, */
m.flags |= XFR_PARITYLESS_WRITE; /* it's parityless */
plex->parityless_writes++; /* count another one */
}
}
/* reset the initial transfer values */
m.dataoffset = m.initoffset; /* start at the beginning of the transfer */
m.datalen = m.initlen;
/* decide how many requests we need */
if (m.flags & (XFR_RECOVERY_READ | XFR_DEGRADED_WRITE))
/* doing a recovery read or degraded write, */
m.rqcount = plex->subdisks; /* all subdisks */
else if (m.flags & XFR_NORMAL_WRITE) /* normal write, */
m.rqcount = m.sdcount + 1; /* all data blocks and the parity block */
else /* parityless write or normal read */
m.rqcount = m.sdcount; /* just the data blocks */
/* Part C: build the requests */
rqg = allocrqg(rq, m.rqcount); /* get a request group */
if (rqg == NULL) { /* malloc failed */
bp->b_flags |= B_ERROR;
bp->b_error = ENOMEM;
biodone(bp);
return REQUEST_ENOMEM;
}
rqg->plexno = plexno;
rqg->flags = m.flags;
rqno = 0; /* index in the request group */
/* 1: PARITY BLOCK */
/*
* Are we performing an operation which requires parity? In that case,
* work out the parameters and define the parity block.
* XFR_PARITYOP is XFR_NORMAL_WRITE | XFR_RECOVERY_READ | XFR_DEGRADED_WRITE
*/
if (m.flags & XFR_PARITYOP) { /* need parity */
rqe = &rqg->rqe[rqno]; /* point to element */
sd = &SD[plex->sdnos[m.psdno]]; /* the subdisk in question */
rqe->rqg = rqg; /* point back to group */
rqe->flags = (m.flags | XFR_PARITY_BLOCK | XFR_MALLOCED) /* always malloc parity block */
&~(XFR_NORMAL_READ | XFR_PARITYLESS_WRITE); /* transfer flags without data op stuf */
setrqebounds(rqe, &m); /* set up the bounds of the transfer */
rqe->sdno = sd->sdno; /* subdisk number */
rqe->driveno = sd->driveno;
if (build_rq_buffer(rqe, plex)) /* build the buffer */
return REQUEST_ENOMEM; /* can't do it */
rqe->b.b_flags |= B_READ; /* we must read first */
m.sdcount++; /* adjust the subdisk count */
rqno++; /* and point to the next request */
}
/*
* 2: DATA BLOCKS
* Now build up requests for the blocks required
* for individual transfers
*/
for (mysdno = m.firstsdno; rqno < m.sdcount; mysdno++, rqno++) {
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
if (mysdno == plex->subdisks) /* got to the end, */
mysdno = 0; /* wrap around */
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
rqe = &rqg->rqe[rqno]; /* point to element */
sd = &SD[plex->sdnos[mysdno]]; /* the subdisk in question */
rqe->rqg = rqg; /* point to group */
if (m.flags & XFR_NEEDS_MALLOC) /* we need a malloced buffer first */
rqe->flags = m.flags | XFR_DATA_BLOCK | XFR_MALLOCED; /* transfer flags */
else
rqe->flags = m.flags | XFR_DATA_BLOCK; /* transfer flags */
if (mysdno == m.badsdno) { /* this is the bad subdisk */
rqg->badsdno = rqno; /* note which one */
rqe->flags |= XFR_BAD_SUBDISK; /* note that it's dead */
/*
* we can't read or write from/to it,
* but we don't need to malloc
*/
rqe->flags &= ~(XFR_MALLOCED | XFR_NORMAL_READ | XFR_NORMAL_WRITE);
}
setrqebounds(rqe, &m); /* set up the bounds of the transfer */
rqe->useroffset = m.useroffset; /* offset in user buffer */
rqe->sdno = sd->sdno; /* subdisk number */
rqe->driveno = sd->driveno;
if (build_rq_buffer(rqe, plex)) /* build the buffer */
return REQUEST_ENOMEM; /* can't do it */
if ((m.flags & XFR_PARITYOP) /* parity operation, */
&&((m.flags & XFR_BAD_SUBDISK) == 0)) /* and not the bad subdisk, */
rqe->b.b_flags |= B_READ; /* we must read first */
/* Now update pointers for the next block */
*diskaddr += m.datalen; /* skip past what we've done */
m.stripesectors -= m.datalen; /* deduct from what's left */
m.useroffset += m.datalen; /* and move on in the user buffer */
m.datalen = min(m.stripesectors, plex->stripesize); /* and recalculate */
m.dataoffset = 0; /* start at the beginning of next block */
}
/*
* 3: REMAINING BLOCKS FOR RECOVERY
* Finally, if we have a recovery operation, build
* up transfers for the other subdisks. Follow the
* subdisks around until we get to where we started.
* These requests use only the group parameters.
*/
if ((rqno < m.rqcount) /* haven't done them all already */
&&(m.flags & (XFR_RECOVERY_READ | XFR_DEGRADED_WRITE))) {
for (; rqno < m.rqcount; rqno++, mysdno++) {
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
if (mysdno == plex->subdisks) /* got to the end, */
mysdno = 0; /* wrap around */
if (mysdno == m.psdno) /* parity, */
mysdno++; /* we've given already */
rqe = &rqg->rqe[rqno]; /* point to element */
sd = &SD[plex->sdnos[mysdno]]; /* the subdisk in question */
rqe->rqg = rqg; /* point to group */
rqe->sdoffset = m.sdbase + m.groupoffset; /* start of transfer */
rqe->dataoffset = 0; /* for tidiness' sake */
rqe->groupoffset = 0; /* group starts at the beginining */
rqe->datalen = 0;
rqe->grouplen = m.grouplen;
rqe->buflen = m.grouplen;
rqe->flags = (m.flags | XFR_MALLOCED) /* transfer flags without data op stuf */
&~XFR_DATAOP;
rqe->sdno = sd->sdno; /* subdisk number */
rqe->driveno = sd->driveno;
if (build_rq_buffer(rqe, plex)) /* build the buffer */
return REQUEST_ENOMEM; /* can't do it */
rqe->b.b_flags |= B_READ; /* we must read first */
}
}
/*
* We need to lock the address range before
* doing anything. We don't have to be
* performing a recovery operation: somebody
* else could be doing so, and the results could
* influence us. Note the fact here, we'll perform
* the lock in launch_requests.
*/
rqg->lockbase = m.stripebase;
if (*diskaddr < diskend) /* didn't finish the request on this stripe */
plex->multistripe++; /* count another one */
}
return REQUEST_OK;
}
/*
* 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 bio *bio)
{
union daemoninfo di;
struct buf *bp = bio->bio_buf;
struct rqelement *rqe;
struct request *rq;
struct rqgroup *rqg;
struct bio *ubio; /* user buffer */
struct drive *drive;
struct sd *sd;
char *gravity; /* for error messages */
get_mplock();
rqe = (struct rqelement *) bp; /* point to the element that completed */
rqg = rqe->rqg; /* and the request group */
rq = rqg->rq; /* and the complete request */
ubio = rq->bio; /* user buffer */
#ifdef VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_iodone, (union rqinfou) rqe, ubio);
#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_flags & B_ERROR) != 0) { /* transfer in error */
gravity = "";
sd = &SD[rqe->sdno];
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->lasterror = rq->error;
if (bp->b_cmd == BUF_CMD_READ) {
if ((rq->error == ENXIO) || (sd->flags & VF_RETRYERRORS) == 0) {
gravity = " fatal";
set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* subdisk is crashed */
}
log(LOG_ERR,
"%s:%s read error, offset %lld for %d bytes\n",
gravity,
sd->name,
(long long)bio->bio_offset,
bp->b_bcount);
} else { /* write operation */
if ((rq->error == ENXIO) || (sd->flags & VF_RETRYERRORS) == 0) {
gravity = "fatal ";
set_sd_state(rqe->sdno, sd_stale, setstate_force); /* subdisk is stale */
}
log(LOG_ERR,
"%s:%s write error, offset %lld for %d bytes\n",
gravity,
sd->name,
(long long)bio->bio_offset,
bp->b_bcount);
}
log(LOG_ERR,
"%s: user buffer offset %lld for %d bytes\n",
sd->name,
(long long)ubio->bio_offset,
ubio->bio_buf->b_bcount);
if (rq->error == ENXIO) { /* the drive's down too */
log(LOG_ERR,
"%s: fatal drive I/O error, offset %lld for %d bytes\n",
DRIVE[rqe->driveno].label.name,
(long long)bio->bio_offset,
bp->b_bcount);
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_cmd == BUF_CMD_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) { /* volume I/O, not plex */
VOL[PLEX[rqe->rqg->plexno].volno].reads++;
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) { /* volume I/O, not plex */
VOL[PLEX[rqe->rqg->plexno].volno].writes++;
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_BSIZE / sizeof(int)); /* and number of ints */
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 *) ubio->bio_buf->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 active request */
complete_raid5_write(rqe);
/*
* This is the earliest place where we can be
* sure that the request has really finished,
* since complete_raid5_write can issue new
* requests.
*/
rqg->active--; /* this request now finished */
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, */
#ifdef VINUMDEBUG
if (debug & DEBUG_RESID) {
if (ubio->bio_buf->b_resid != 0) /* still something to transfer? */
Debugger("resid");
}
#endif
if (rq->error) { /* did we have an error? */
if (rq->isplex) { /* plex operation, */
ubio->bio_buf->b_flags |= B_ERROR; /* yes, propagate to user */
ubio->bio_buf->b_error = rq->error;
} else { /* try to recover */
di.rq = rq;
queue_daemon_request(daemonrq_ioerror, di); /* let the daemon complete */
}
} else {
ubio->bio_buf->b_resid = 0; /* completed our transfer */
if (rq->isplex == 0) /* volume request, */
VOL[rq->volplex.volno].active--; /* another request finished */
biodone(ubio); /* top level buffer completed */
freerq(rq); /* return the request storage */
}
}
rel_mplock();
}
/* attach an object to a superior object */
void
attachobject(struct vinum_ioctl_msg *msg)
{
struct _ioctl_reply *reply = (struct _ioctl_reply *) msg;
int sdno;
struct sd *sd;
struct plex *plex;
struct volume *vol;
switch (msg->type) {
case drive_object: /* you can't attach a drive to anything */
case volume_object: /* nor a volume */
case invalid_object: /* "this can't happen" */
reply->error = EINVAL;
reply->msg[0] = '\0'; /* vinum(8) doesn't do this */
return;
case sd_object:
sd = validsd(msg->index, reply);
if (sd == NULL) /* not a valid subdisk */
return;
plex = validplex(msg->otherobject, reply);
if (plex) {
/*
* We should be more intelligent about this.
* We should be able to reattach a dead
* subdisk, but if we want to increase the total
* number of subdisks, we have a lot of reshuffling
* to do. XXX
*/
if ((plex->organization != plex_concat) /* can't attach to striped and RAID-4/5 */
&&(!msg->force)) { /* without using force */
reply->error = EINVAL; /* no message, the user should check */
strcpy(reply->msg, "Can't attach to this plex organization");
return;
}
if (sd->plexno >= 0) { /* already belong to a plex */
reply->error = EBUSY; /* no message, the user should check */
reply->msg[0] = '\0';
return;
}
sd->plexoffset = msg->offset; /* this is where we want it */
set_sd_state(sd->sdno, sd_stale, setstate_force); /* make sure it's stale */
give_sd_to_plex(plex->plexno, sd->sdno); /* and give it to the plex */
update_sd_config(sd->sdno, 0);
save_config();
}
if (sd->state == sd_reviving)
reply->error = EAGAIN; /* need to revive it */
else
reply->error = 0;
break;
case plex_object:
plex = validplex(msg->index, reply); /* get plex */
if (plex == NULL)
return;
vol = validvol(msg->otherobject, reply); /* and volume information */
if (vol) {
if ((vol->plexes == MAXPLEX) /* we have too many already */
||(plex->volno >= 0)) { /* or the plex has an owner */
reply->error = EINVAL; /* no message, the user should check */
reply->msg[0] = '\0';
return;
}
for (sdno = 0; sdno < plex->subdisks; sdno++) {
sd = &SD[plex->sdnos[sdno]];
if (sd->state > sd_down) /* real subdisk, vaguely accessible */
set_sd_state(plex->sdnos[sdno], sd_stale, setstate_force); /* make it stale */
}
set_plex_state(plex->plexno, plex_up, setstate_none); /* update plex state */
give_plex_to_volume(msg->otherobject, msg->index); /* and give it to the volume */
update_plex_config(plex->plexno, 0);
save_config();
reply->error = 0; /* all went well */
}
}
}
/* Perform I/O on a subdisk */
void
sdio(struct buf *bp)
{
int s; /* spl */
struct sd *sd;
struct sdbuf *sbp;
daddr_t endoffset;
struct drive *drive;
#if VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_sdio, (union rqinfou) bp, bp);
#endif
sd = &SD[Sdno(bp->b_dev)]; /* point to the subdisk */
drive = &DRIVE[sd->driveno];
if (drive->state != drive_up) {
if (sd->state >= sd_crashed) {
if (bp->b_flags & B_READ) /* reading, */
set_sd_state(sd->sdno, sd_crashed, setstate_force);
else
set_sd_state(sd->sdno, sd_stale, setstate_force);
}
bp->b_flags |= B_ERROR;
bp->b_error = EIO;
biodone(bp);
return;
}
/*
* We allow access to any kind of subdisk as long as we can expect
* to get the I/O performed.
*/
if (sd->state < sd_empty) { /* nothing to talk to, */
bp->b_flags |= B_ERROR;
bp->b_error = EIO;
biodone(bp);
return;
}
/* Get a buffer */
sbp = (struct sdbuf *) Malloc(sizeof(struct sdbuf));
if (sbp == NULL) {
bp->b_flags |= B_ERROR;
bp->b_error = ENOMEM;
biodone(bp);
return;
}
bzero(sbp, sizeof(struct sdbuf)); /* start with nothing */
sbp->b.b_flags = bp->b_flags | B_CALL; /* inform us when it's done */
sbp->b.b_bufsize = bp->b_bufsize; /* buffer size */
sbp->b.b_bcount = bp->b_bcount; /* number of bytes to transfer */
sbp->b.b_resid = bp->b_resid; /* and amount waiting */
sbp->b.b_dev = DRIVE[sd->driveno].dev; /* device */
sbp->b.b_data = bp->b_data; /* data buffer */
sbp->b.b_blkno = bp->b_blkno + sd->driveoffset;
sbp->b.b_iodone = sdio_done; /* come here on completion */
BUF_LOCKINIT(&sbp->b); /* get a lock for the buffer */
BUF_LOCK(&sbp->b, LK_EXCLUSIVE); /* and lock it */
sbp->bp = bp; /* note the address of the original header */
sbp->sdno = sd->sdno; /* note for statistics */
sbp->driveno = sd->driveno;
endoffset = bp->b_blkno + sbp->b.b_bcount / DEV_BSIZE; /* final sector offset */
if (endoffset > sd->sectors) { /* beyond the end */
sbp->b.b_bcount -= (endoffset - sd->sectors) * DEV_BSIZE; /* trim */
if (sbp->b.b_bcount <= 0) { /* nothing to transfer */
bp->b_resid = bp->b_bcount; /* nothing transferred */
biodone(bp);
Free(sbp);
return;
}
}
#if VINUMDEBUG
if (debug & DEBUG_ADDRESSES)
log(LOG_DEBUG,
" %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n",
sbp->b.b_flags & B_READ ? "Read" : "Write",
major(sbp->b.b_dev),
minor(sbp->b.b_dev),
sbp->sdno,
(u_int) (sbp->b.b_blkno - SD[sbp->sdno].driveoffset),
(int) sbp->b.b_blkno,
sbp->b.b_bcount);
#endif
s = splbio();
#if VINUMDEBUG
if (debug & DEBUG_LASTREQS)
logrq(loginfo_sdiol, (union rqinfou) &sbp->b, &sbp->b);
#endif
BUF_STRATEGY(&sbp->b, 0);
splx(s);
}
