1 /* vinuminterrupt.c: bottom half of the driver */ 2 3 /*- 4 * Copyright (c) 1997, 1998, 1999 5 * Nan Yang Computer Services Limited. All rights reserved. 6 * 7 * Parts copyright (c) 1997, 1998 Cybernet Corporation, NetMAX project. 8 * 9 * Written by Greg Lehey 10 * 11 * This software is distributed under the so-called ``Berkeley 12 * License'': 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 3. All advertising materials mentioning features or use of this software 23 * must display the following acknowledgement: 24 * This product includes software developed by Nan Yang Computer 25 * Services Limited. 26 * 4. Neither the name of the Company nor the names of its contributors 27 * may be used to endorse or promote products derived from this software 28 * without specific prior written permission. 29 * 30 * This software is provided ``as is'', and any express or implied 31 * warranties, including, but not limited to, the implied warranties of 32 * merchantability and fitness for a particular purpose are disclaimed. 33 * In no event shall the company or contributors be liable for any 34 * direct, indirect, incidental, special, exemplary, or consequential 35 * damages (including, but not limited to, procurement of substitute 36 * goods or services; loss of use, data, or profits; or business 37 * interruption) however caused and on any theory of liability, whether 38 * in contract, strict liability, or tort (including negligence or 39 * otherwise) arising in any way out of the use of this software, even if 40 * advised of the possibility of such damage. 41 * 42 * $Id: vinuminterrupt.c,v 1.12 2000/11/24 03:41:42 grog Exp grog $ 43 * $FreeBSD: src/sys/dev/vinum/vinuminterrupt.c,v 1.25.2.3 2001/05/28 05:56:27 grog Exp $ 44 */ 45 46 #include "vinumhdr.h" 47 #include "request.h" 48 #include <sys/resourcevar.h> 49 50 void complete_raid5_write(struct rqelement *); 51 void complete_rqe(struct bio *bio); 52 void sdio_done(struct bio *bio); 53 54 /* 55 * Take a completed buffer, transfer the data back if 56 * it's a read, and complete the high-level request 57 * if this is the last subrequest. 58 * 59 * The bp parameter is in fact a struct rqelement, which 60 * includes a couple of extras at the end. 61 */ 62 void 63 complete_rqe(struct bio *bio) 64 { 65 union daemoninfo di; 66 struct buf *bp = bio->bio_buf; 67 struct rqelement *rqe; 68 struct request *rq; 69 struct rqgroup *rqg; 70 struct bio *ubio; /* user buffer */ 71 struct drive *drive; 72 struct sd *sd; 73 char *gravity; /* for error messages */ 74 75 get_mplock(); 76 77 rqe = (struct rqelement *) bp; /* point to the element that completed */ 78 rqg = rqe->rqg; /* and the request group */ 79 rq = rqg->rq; /* and the complete request */ 80 ubio = rq->bio; /* user buffer */ 81 82 #ifdef VINUMDEBUG 83 if (debug & DEBUG_LASTREQS) 84 logrq(loginfo_iodone, (union rqinfou) rqe, ubio); 85 #endif 86 drive = &DRIVE[rqe->driveno]; 87 drive->active--; /* one less outstanding I/O on this drive */ 88 vinum_conf.active--; /* one less outstanding I/O globally */ 89 if ((drive->active == (DRIVE_MAXACTIVE - 1)) /* we were at the drive limit */ 90 ||(vinum_conf.active == VINUM_MAXACTIVE)) /* or the global limit */ 91 wakeup(&launch_requests); /* let another one at it */ 92 if ((bp->b_flags & B_ERROR) != 0) { /* transfer in error */ 93 gravity = ""; 94 sd = &SD[rqe->sdno]; 95 96 if (bp->b_error != 0) /* did it return a number? */ 97 rq->error = bp->b_error; /* yes, put it in. */ 98 else if (rq->error == 0) /* no: do we have one already? */ 99 rq->error = EIO; /* no: catchall "I/O error" */ 100 sd->lasterror = rq->error; 101 if (bp->b_cmd == BUF_CMD_READ) { 102 if ((rq->error == ENXIO) || (sd->flags & VF_RETRYERRORS) == 0) { 103 gravity = " fatal"; 104 set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* subdisk is crashed */ 105 } 106 log(LOG_ERR, 107 "%s:%s read error, offset %lld for %d bytes\n", 108 gravity, 109 sd->name, 110 (long long)bio->bio_offset, 111 bp->b_bcount); 112 } else { /* write operation */ 113 if ((rq->error == ENXIO) || (sd->flags & VF_RETRYERRORS) == 0) { 114 gravity = "fatal "; 115 set_sd_state(rqe->sdno, sd_stale, setstate_force); /* subdisk is stale */ 116 } 117 log(LOG_ERR, 118 "%s:%s write error, offset %lld for %d bytes\n", 119 gravity, 120 sd->name, 121 (long long)bio->bio_offset, 122 bp->b_bcount); 123 } 124 log(LOG_ERR, 125 "%s: user buffer offset %lld for %d bytes\n", 126 sd->name, 127 (long long)ubio->bio_offset, 128 ubio->bio_buf->b_bcount); 129 if (rq->error == ENXIO) { /* the drive's down too */ 130 log(LOG_ERR, 131 "%s: fatal drive I/O error, offset %lld for %d bytes\n", 132 DRIVE[rqe->driveno].label.name, 133 (long long)bio->bio_offset, 134 bp->b_bcount); 135 DRIVE[rqe->driveno].lasterror = rq->error; 136 set_drive_state(rqe->driveno, /* take the drive down */ 137 drive_down, 138 setstate_force); 139 } 140 } 141 /* Now update the statistics */ 142 if (bp->b_cmd == BUF_CMD_READ) { /* read operation */ 143 DRIVE[rqe->driveno].reads++; 144 DRIVE[rqe->driveno].bytes_read += bp->b_bcount; 145 SD[rqe->sdno].reads++; 146 SD[rqe->sdno].bytes_read += bp->b_bcount; 147 PLEX[rqe->rqg->plexno].reads++; 148 PLEX[rqe->rqg->plexno].bytes_read += bp->b_bcount; 149 if (PLEX[rqe->rqg->plexno].volno >= 0) { /* volume I/O, not plex */ 150 VOL[PLEX[rqe->rqg->plexno].volno].reads++; 151 VOL[PLEX[rqe->rqg->plexno].volno].bytes_read += bp->b_bcount; 152 } 153 } else { /* write operation */ 154 DRIVE[rqe->driveno].writes++; 155 DRIVE[rqe->driveno].bytes_written += bp->b_bcount; 156 SD[rqe->sdno].writes++; 157 SD[rqe->sdno].bytes_written += bp->b_bcount; 158 PLEX[rqe->rqg->plexno].writes++; 159 PLEX[rqe->rqg->plexno].bytes_written += bp->b_bcount; 160 if (PLEX[rqe->rqg->plexno].volno >= 0) { /* volume I/O, not plex */ 161 VOL[PLEX[rqe->rqg->plexno].volno].writes++; 162 VOL[PLEX[rqe->rqg->plexno].volno].bytes_written += bp->b_bcount; 163 } 164 } 165 if (rqg->flags & XFR_RECOVERY_READ) { /* recovery read, */ 166 int *sdata; /* source */ 167 int *data; /* and group data */ 168 int length; /* and count involved */ 169 int count; /* loop counter */ 170 struct rqelement *urqe = &rqg->rqe[rqg->badsdno]; /* rqe of the bad subdisk */ 171 172 /* XOR destination is the user data */ 173 sdata = (int *) &rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]; /* old data contents */ 174 data = (int *) &urqe->b.b_data[urqe->groupoffset << DEV_BSHIFT]; /* destination */ 175 length = urqe->grouplen * (DEV_BSIZE / sizeof(int)); /* and number of ints */ 176 177 for (count = 0; count < length; count++) 178 data[count] ^= sdata[count]; 179 180 /* 181 * In a normal read, we will normally read directly 182 * into the user buffer. This doesn't work if 183 * we're also doing a recovery, so we have to 184 * copy it 185 */ 186 if (rqe->flags & XFR_NORMAL_READ) { /* normal read as well, */ 187 char *src = &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* read data is here */ 188 char *dst; 189 190 dst = (char *) ubio->bio_buf->b_data + (rqe->useroffset << DEV_BSHIFT); /* where to put it in user buffer */ 191 length = rqe->datalen << DEV_BSHIFT; /* and count involved */ 192 bcopy(src, dst, length); /* move it */ 193 } 194 } else if ((rqg->flags & (XFR_NORMAL_WRITE | XFR_DEGRADED_WRITE)) /* RAID 4/5 group write operation */ 195 &&(rqg->active == 1)) /* and this is the last active request */ 196 complete_raid5_write(rqe); 197 /* 198 * This is the earliest place where we can be 199 * sure that the request has really finished, 200 * since complete_raid5_write can issue new 201 * requests. 202 */ 203 rqg->active--; /* this request now finished */ 204 if (rqg->active == 0) { /* request group finished, */ 205 rq->active--; /* one less */ 206 if (rqg->lock) { /* got a lock? */ 207 unlockrange(rqg->plexno, rqg->lock); /* yes, free it */ 208 rqg->lock = 0; 209 } 210 } 211 if (rq->active == 0) { /* request finished, */ 212 #ifdef VINUMDEBUG 213 if (debug & DEBUG_RESID) { 214 if (ubio->bio_buf->b_resid != 0) /* still something to transfer? */ 215 Debugger("resid"); 216 } 217 #endif 218 219 if (rq->error) { /* did we have an error? */ 220 if (rq->isplex) { /* plex operation, */ 221 ubio->bio_buf->b_flags |= B_ERROR; /* yes, propagate to user */ 222 ubio->bio_buf->b_error = rq->error; 223 } else { /* try to recover */ 224 di.rq = rq; 225 queue_daemon_request(daemonrq_ioerror, di); /* let the daemon complete */ 226 } 227 } else { 228 ubio->bio_buf->b_resid = 0; /* completed our transfer */ 229 if (rq->isplex == 0) /* volume request, */ 230 VOL[rq->volplex.volno].active--; /* another request finished */ 231 biodone(ubio); /* top level buffer completed */ 232 freerq(rq); /* return the request storage */ 233 } 234 } 235 rel_mplock(); 236 } 237 238 /* Free a request block and anything hanging off it */ 239 void 240 freerq(struct request *rq) 241 { 242 struct rqgroup *rqg; 243 struct rqgroup *nrqg; /* next in chain */ 244 int rqno; 245 246 for (rqg = rq->rqg; rqg != NULL; rqg = nrqg) { /* through the whole request chain */ 247 if (rqg->lock) /* got a lock? */ 248 unlockrange(rqg->plexno, rqg->lock); /* yes, free it */ 249 for (rqno = 0; rqno < rqg->count; rqno++) { 250 if ((rqg->rqe[rqno].flags & XFR_MALLOCED) /* data buffer was malloced, */ 251 &&rqg->rqe[rqno].b.b_data) /* and the allocation succeeded */ 252 Free(rqg->rqe[rqno].b.b_data); /* free it */ 253 if (rqg->rqe[rqno].flags & XFR_BUFLOCKED) { /* locked this buffer, */ 254 BUF_UNLOCK(&rqg->rqe[rqno].b); /* unlock it again */ 255 uninitbufbio(&rqg->rqe[rqno].b); 256 } 257 } 258 nrqg = rqg->next; /* note the next one */ 259 Free(rqg); /* and free this one */ 260 } 261 Free(rq); /* free the request itself */ 262 } 263 264 /* I/O on subdisk completed */ 265 void 266 sdio_done(struct bio *bio) 267 { 268 struct sdbuf *sbp; 269 270 get_mplock(); 271 272 sbp = (struct sdbuf *) bio->bio_buf; 273 if (sbp->b.b_flags & B_ERROR) { /* had an error */ 274 sbp->bio->bio_buf->b_flags |= B_ERROR; /* propagate upwards */ 275 sbp->bio->bio_buf->b_error = sbp->b.b_error; 276 } 277 #ifdef VINUMDEBUG 278 if (debug & DEBUG_LASTREQS) 279 logrq(loginfo_sdiodone, (union rqinfou)bio, bio); 280 #endif 281 sbp->bio->bio_buf->b_resid = sbp->b.b_resid; /* copy the resid field */ 282 /* Now update the statistics */ 283 if (sbp->b.b_cmd == BUF_CMD_READ) { /* read operation */ 284 DRIVE[sbp->driveno].reads++; 285 DRIVE[sbp->driveno].bytes_read += sbp->b.b_bcount; 286 SD[sbp->sdno].reads++; 287 SD[sbp->sdno].bytes_read += sbp->b.b_bcount; 288 } else { /* write operation */ 289 DRIVE[sbp->driveno].writes++; 290 DRIVE[sbp->driveno].bytes_written += sbp->b.b_bcount; 291 SD[sbp->sdno].writes++; 292 SD[sbp->sdno].bytes_written += sbp->b.b_bcount; 293 } 294 biodone_sync(bio); 295 biodone(sbp->bio); /* complete the caller's I/O */ 296 BUF_UNLOCK(&sbp->b); 297 uninitbufbio(&sbp->b); 298 Free(sbp); 299 rel_mplock(); 300 } 301 302 /* Start the second phase of a RAID-4 or RAID-5 group write operation. */ 303 void 304 complete_raid5_write(struct rqelement *rqe) 305 { 306 int *sdata; /* source */ 307 int *pdata; /* and parity block data */ 308 int length; /* and count involved */ 309 int count; /* loop counter */ 310 int rqno; /* request index */ 311 int rqoffset; /* offset of request data from parity data */ 312 struct bio *ubio; /* user buffer header */ 313 struct request *rq; /* pointer to our request */ 314 struct rqgroup *rqg; /* and to the request group */ 315 struct rqelement *prqe; /* point to the parity block */ 316 struct drive *drive; /* drive to access */ 317 rqg = rqe->rqg; /* and to our request group */ 318 rq = rqg->rq; /* point to our request */ 319 ubio = rq->bio; /* user's buffer header */ 320 prqe = &rqg->rqe[0]; /* point to the parity block */ 321 322 /* 323 * If we get to this function, we have normal or 324 * degraded writes, or a combination of both. We do 325 * the same thing in each case: we perform an 326 * exclusive or to the parity block. The only 327 * difference is the origin of the data and the 328 * address range. 329 */ 330 if (rqe->flags & XFR_DEGRADED_WRITE) { /* do the degraded write stuff */ 331 pdata = (int *) (&prqe->b.b_data[(prqe->groupoffset) << DEV_BSHIFT]); /* parity data pointer */ 332 bzero(pdata, prqe->grouplen << DEV_BSHIFT); /* start with nothing in the parity block */ 333 334 /* Now get what data we need from each block */ 335 for (rqno = 1; rqno < rqg->count; rqno++) { /* for all the data blocks */ 336 rqe = &rqg->rqe[rqno]; /* this request */ 337 sdata = (int *) (&rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]); /* old data */ 338 length = rqe->grouplen << (DEV_BSHIFT - 2); /* and count involved */ 339 340 /* 341 * Add the data block to the parity block. Before 342 * we started the request, we zeroed the parity 343 * block, so the result of adding all the other 344 * blocks and the block we want to write will be 345 * the correct parity block. 346 */ 347 for (count = 0; count < length; count++) 348 pdata[count] ^= sdata[count]; 349 if ((rqe->flags & XFR_MALLOCED) /* the buffer was malloced, */ 350 &&((rqg->flags & XFR_NORMAL_WRITE) == 0)) { /* and we have no normal write, */ 351 Free(rqe->b.b_data); /* free it now */ 352 rqe->flags &= ~XFR_MALLOCED; 353 } 354 } 355 } 356 if (rqg->flags & XFR_NORMAL_WRITE) { /* do normal write stuff */ 357 /* Get what data we need from each block */ 358 for (rqno = 1; rqno < rqg->count; rqno++) { /* for all the data blocks */ 359 rqe = &rqg->rqe[rqno]; /* this request */ 360 if ((rqe->flags & (XFR_DATA_BLOCK | XFR_BAD_SUBDISK | XFR_NORMAL_WRITE)) 361 == (XFR_DATA_BLOCK | XFR_NORMAL_WRITE)) { /* good data block to write */ 362 sdata = (int *) &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* old data contents */ 363 rqoffset = rqe->dataoffset + rqe->sdoffset - prqe->sdoffset; /* corresponding parity block offset */ 364 pdata = (int *) (&prqe->b.b_data[rqoffset << DEV_BSHIFT]); /* parity data pointer */ 365 length = rqe->datalen * (DEV_BSIZE / sizeof(int)); /* and number of ints */ 366 367 /* 368 * "remove" the old data block 369 * from the parity block 370 */ 371 if ((pdata < ((int *) prqe->b.b_data)) 372 || (&pdata[length] > ((int *) (prqe->b.b_data + prqe->b.b_bcount))) 373 || (sdata < ((int *) rqe->b.b_data)) 374 || (&sdata[length] > ((int *) (rqe->b.b_data + rqe->b.b_bcount)))) 375 panic("complete_raid5_write: bounds overflow"); 376 for (count = 0; count < length; count++) 377 pdata[count] ^= sdata[count]; 378 379 /* "add" the new data block */ 380 sdata = (int *) (&ubio->bio_buf->b_data[rqe->useroffset << DEV_BSHIFT]); /* new data */ 381 if ((sdata < ((int *) ubio->bio_buf->b_data)) 382 || (&sdata[length] > ((int *) (ubio->bio_buf->b_data + ubio->bio_buf->b_bcount)))) 383 panic("complete_raid5_write: bounds overflow"); 384 for (count = 0; count < length; count++) 385 pdata[count] ^= sdata[count]; 386 387 /* Free the malloced buffer */ 388 if (rqe->flags & XFR_MALLOCED) { /* the buffer was malloced, */ 389 Free(rqe->b.b_data); /* free it */ 390 rqe->flags &= ~XFR_MALLOCED; 391 } else 392 panic("complete_raid5_write: malloc conflict"); 393 394 if ((rqe->b.b_cmd == BUF_CMD_READ) /* this was a read */ 395 &&((rqe->flags & XFR_BAD_SUBDISK) == 0)) { /* and we can write this block */ 396 rqe->b.b_cmd = BUF_CMD_WRITE; /* we're writing now */ 397 rqe->b.b_bio1.bio_done = complete_rqe; /* by calling us here */ 398 rqe->flags &= ~XFR_PARITYOP; /* reset flags that brought us here */ 399 rqe->b.b_data = &ubio->bio_buf->b_data[rqe->useroffset << DEV_BSHIFT]; /* point to the user data */ 400 rqe->b.b_bcount = rqe->datalen << DEV_BSHIFT; /* length to write */ 401 rqe->b.b_resid = rqe->b.b_bcount; /* nothing transferred */ 402 rqe->b.b_bio1.bio_offset += (off_t)rqe->dataoffset << DEV_BSHIFT; /* point to the correct block */ 403 drive = &DRIVE[rqe->driveno]; /* drive to access */ 404 rqe->b.b_bio1.bio_driver_info = drive->dev; 405 rqg->active++; /* another active request */ 406 407 /* We can't sleep here, so we just increment the counters. */ 408 drive->active++; 409 if (drive->active >= drive->maxactive) 410 drive->maxactive = drive->active; 411 vinum_conf.active++; 412 if (vinum_conf.active >= vinum_conf.maxactive) 413 vinum_conf.maxactive = vinum_conf.active; 414 #if VINUMDEBUG 415 if (debug & DEBUG_ADDRESSES) 416 log(LOG_DEBUG, 417 " %s dev %s, sd %d, offset 0x%jx, devoffset 0x%jx, length %d\n", 418 (rqe->b.b_cmd == BUF_CMD_READ) ? "Read" : "Write", 419 drive->devicename, 420 rqe->sdno, 421 (uintmax_t)(rqe->b.b_bio1.bio_offset - ((off_t)SD[rqe->sdno].driveoffset << DEV_BSHIFT)), 422 (uintmax_t)rqe->b.b_bio1.bio_offset, 423 rqe->b.b_bcount); 424 if (debug & DEBUG_LASTREQS) 425 logrq(loginfo_raid5_data, (union rqinfou) rqe, ubio); 426 #endif 427 vn_strategy(drive->vp, &rqe->b.b_bio1); 428 } 429 } 430 } 431 } 432 /* Finally, write the parity block */ 433 rqe = &rqg->rqe[0]; 434 rqe->b.b_cmd = BUF_CMD_WRITE; /* we're writing now */ 435 rqe->b.b_bio1.bio_done = complete_rqe; /* by calling us here */ 436 rqg->flags &= ~XFR_PARITYOP; /* reset flags that brought us here */ 437 rqe->b.b_bcount = rqe->buflen << DEV_BSHIFT; /* length to write */ 438 rqe->b.b_resid = rqe->b.b_bcount; /* nothing transferred */ 439 drive = &DRIVE[rqe->driveno]; /* drive to access */ 440 rqe->b.b_bio1.bio_driver_info = drive->dev; 441 rqg->active++; /* another active request */ 442 443 /* We can't sleep here, so we just increment the counters. */ 444 drive->active++; 445 if (drive->active >= drive->maxactive) 446 drive->maxactive = drive->active; 447 vinum_conf.active++; 448 if (vinum_conf.active >= vinum_conf.maxactive) 449 vinum_conf.maxactive = vinum_conf.active; 450 451 #if VINUMDEBUG 452 if (debug & DEBUG_ADDRESSES) 453 log(LOG_DEBUG, 454 " %s dev %s, sd %d, offset 0x%jx, devoffset 0x%jx, length %d\n", 455 (rqe->b.b_cmd == BUF_CMD_READ) ? "Read" : "Write", 456 drive->devicename, 457 rqe->sdno, 458 (uintmax_t)(rqe->b.b_bio1.bio_offset - ((off_t)SD[rqe->sdno].driveoffset << DEV_BSHIFT)), 459 (uintmax_t)rqe->b.b_bio1.bio_offset, 460 rqe->b.b_bcount); 461 if (debug & DEBUG_LASTREQS) 462 logrq(loginfo_raid5_parity, (union rqinfou) rqe, ubio); 463 #endif 464 vn_strategy(drive->vp, &rqe->b.b_bio1); 465 } 466