Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/dev/vinum/vinuminterrupt.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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 1.9 2000/02/16 01:59:02 grog Exp grog $ 43 * $FreeBSD$ 44 */ 45 46 #include <dev/vinum/vinumhdr.h> 47 #include <dev/vinum/request.h> 48 #include <sys/resourcevar.h> 49 50 void complete_raid5_write(struct rqelement *); 51 void complete_rqe(struct buf *bp); 52 void sdio_done(struct buf *bp); 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 buf *bp) 64 { 65 struct rqelement *rqe; 66 struct request *rq; 67 struct rqgroup *rqg; 68 struct buf *ubp; /* user buffer */ 69 struct drive *drive; 70 71 rqe = (struct rqelement *) bp; /* point to the element element that completed */ 72 rqg = rqe->rqg; /* and the request group */ 73 rq = rqg->rq; /* and the complete request */ 74 ubp = rq->bp; /* user buffer */ 75 76 #ifdef VINUMDEBUG 77 if (debug & DEBUG_LASTREQS) 78 logrq(loginfo_iodone, (union rqinfou) rqe, ubp); 79 #endif 80 drive = &DRIVE[rqe->driveno]; 81 drive->active--; /* one less outstanding I/O on this drive */ 82 vinum_conf.active--; /* one less outstanding I/O globally */ 83 if ((drive->active == (DRIVE_MAXACTIVE - 1)) /* we were at the drive limit */ 84 ||(vinum_conf.active == VINUM_MAXACTIVE)) /* or the global limit */ 85 wakeup(&launch_requests); /* let another one at it */ 86 if ((bp->b_flags & B_ERROR) != 0) { /* transfer in error */ 87 if (bp->b_error != 0) /* did it return a number? */ 88 rq->error = bp->b_error; /* yes, put it in. */ 89 else if (rq->error == 0) /* no: do we have one already? */ 90 rq->error = EIO; /* no: catchall "I/O error" */ 91 SD[rqe->sdno].lasterror = rq->error; 92 if (bp->b_flags & B_READ) { 93 log(LOG_ERR, "%s: fatal read I/O error\n", SD[rqe->sdno].name); 94 set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* subdisk is crashed */ 95 } else { /* write operation */ 96 log(LOG_ERR, "%s: fatal write I/O error\n", SD[rqe->sdno].name); 97 set_sd_state(rqe->sdno, sd_stale, setstate_force); /* subdisk is stale */ 98 } 99 if (rq->error == ENXIO) { /* the drive's down too */ 100 log(LOG_ERR, "%s: fatal drive I/O error\n", DRIVE[rqe->driveno].label.name); 101 DRIVE[rqe->driveno].lasterror = rq->error; 102 set_drive_state(rqe->driveno, /* take the drive down */ 103 drive_down, 104 setstate_force); 105 } 106 } 107 /* Now update the statistics */ 108 if (bp->b_flags & B_READ) { /* read operation */ 109 DRIVE[rqe->driveno].reads++; 110 DRIVE[rqe->driveno].bytes_read += bp->b_bcount; 111 SD[rqe->sdno].reads++; 112 SD[rqe->sdno].bytes_read += bp->b_bcount; 113 PLEX[rqe->rqg->plexno].reads++; 114 PLEX[rqe->rqg->plexno].bytes_read += bp->b_bcount; 115 if (PLEX[rqe->rqg->plexno].volno >= 0) { /* volume I/O, not plex */ 116 VOL[PLEX[rqe->rqg->plexno].volno].reads++; 117 VOL[PLEX[rqe->rqg->plexno].volno].bytes_read += bp->b_bcount; 118 } 119 } else { /* write operation */ 120 DRIVE[rqe->driveno].writes++; 121 DRIVE[rqe->driveno].bytes_written += bp->b_bcount; 122 SD[rqe->sdno].writes++; 123 SD[rqe->sdno].bytes_written += bp->b_bcount; 124 PLEX[rqe->rqg->plexno].writes++; 125 PLEX[rqe->rqg->plexno].bytes_written += bp->b_bcount; 126 if (PLEX[rqe->rqg->plexno].volno >= 0) { /* volume I/O, not plex */ 127 VOL[PLEX[rqe->rqg->plexno].volno].writes++; 128 VOL[PLEX[rqe->rqg->plexno].volno].bytes_written += bp->b_bcount; 129 } 130 } 131 if (rqg->flags & XFR_RECOVERY_READ) { /* recovery read, */ 132 int *sdata; /* source */ 133 int *data; /* and group data */ 134 int length; /* and count involved */ 135 int count; /* loop counter */ 136 struct rqelement *urqe = &rqg->rqe[rqg->badsdno]; /* rqe of the bad subdisk */ 137 138 /* XOR destination is the user data */ 139 sdata = (int *) &rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]; /* old data contents */ 140 data = (int *) &urqe->b.b_data[urqe->groupoffset << DEV_BSHIFT]; /* destination */ 141 length = urqe->grouplen * (DEV_BSIZE / sizeof(int)); /* and number of ints */ 142 143 for (count = 0; count < length; count++) 144 data[count] ^= sdata[count]; 145 146 /* 147 * In a normal read, we will normally read directly 148 * into the user buffer. This doesn't work if 149 * we're also doing a recovery, so we have to 150 * copy it 151 */ 152 if (rqe->flags & XFR_NORMAL_READ) { /* normal read as well, */ 153 char *src = &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* read data is here */ 154 char *dst; 155 156 dst = (char *) ubp->b_data + (rqe->useroffset << DEV_BSHIFT); /* where to put it in user buffer */ 157 length = rqe->datalen << DEV_BSHIFT; /* and count involved */ 158 bcopy(src, dst, length); /* move it */ 159 } 160 } else if ((rqg->flags & (XFR_NORMAL_WRITE | XFR_DEGRADED_WRITE)) /* RAID 4/5 group write operation */ 161 &&(rqg->active == 1)) /* and this is the last active request */ 162 complete_raid5_write(rqe); 163 /* 164 * This is the earliest place where we can be 165 * sure that the request has really finished, 166 * since complete_raid5_write can issue new 167 * requests. 168 */ 169 rqg->active--; /* this request now finished */ 170 if (rqg->active == 0) { /* request group finished, */ 171 rq->active--; /* one less */ 172 if (rqg->lock) { /* got a lock? */ 173 unlockrange(rqg->plexno, rqg->lock); /* yes, free it */ 174 rqg->lock = 0; 175 } 176 } 177 if (rq->active == 0) { /* request finished, */ 178 #if VINUMDEBUG 179 if (debug & DEBUG_RESID) { 180 if (ubp->b_resid != 0) /* still something to transfer? */ 181 Debugger("resid"); 182 } 183 #endif 184 185 if (rq->error) { /* did we have an error? */ 186 if (rq->isplex) { /* plex operation, */ 187 ubp->b_flags |= B_ERROR; /* yes, propagate to user */ 188 ubp->b_error = rq->error; 189 } else /* try to recover */ 190 queue_daemon_request(daemonrq_ioerror, (union daemoninfo) rq); /* let the daemon complete */ 191 } else { 192 ubp->b_resid = 0; /* completed our transfer */ 193 if (rq->isplex == 0) /* volume request, */ 194 VOL[rq->volplex.volno].active--; /* another request finished */ 195 biodone(ubp); /* top level buffer completed */ 196 freerq(rq); /* return the request storage */ 197 } 198 } 199 } 200 201 /* Free a request block and anything hanging off it */ 202 void 203 freerq(struct request *rq) 204 { 205 struct rqgroup *rqg; 206 struct rqgroup *nrqg; /* next in chain */ 207 int rqno; 208 209 for (rqg = rq->rqg; rqg != NULL; rqg = nrqg) { /* through the whole request chain */ 210 if (rqg->lock) /* got a lock? */ 211 unlockrange(rqg->plexno, rqg->lock); /* yes, free it */ 212 for (rqno = 0; rqno < rqg->count; rqno++) 213 if ((rqg->rqe[rqno].flags & XFR_MALLOCED) /* data buffer was malloced, */ 214 &&rqg->rqe[rqno].b.b_data) /* and the allocation succeeded */ 215 Free(rqg->rqe[rqno].b.b_data); /* free it */ 216 nrqg = rqg->next; /* note the next one */ 217 Free(rqg); /* and free this one */ 218 } 219 Free(rq); /* free the request itself */ 220 } 221 222 /* I/O on subdisk completed */ 223 void 224 sdio_done(struct buf *bp) 225 { 226 struct sdbuf *sbp; 227 228 sbp = (struct sdbuf *) bp; 229 if (sbp->b.b_flags & B_ERROR) { /* had an error */ 230 sbp->bp->b_flags |= B_ERROR; /* propagate upwards */ 231 sbp->bp->b_error = sbp->b.b_error; 232 } 233 #ifdef VINUMDEBUG 234 if (debug & DEBUG_LASTREQS) 235 logrq(loginfo_sdiodone, (union rqinfou) bp, bp); 236 #endif 237 sbp->bp->b_resid = sbp->b.b_resid; /* copy the resid field */ 238 /* Now update the statistics */ 239 if (bp->b_flags & B_READ) { /* read operation */ 240 DRIVE[sbp->driveno].reads++; 241 DRIVE[sbp->driveno].bytes_read += sbp->b.b_bcount; 242 SD[sbp->sdno].reads++; 243 SD[sbp->sdno].bytes_read += sbp->b.b_bcount; 244 } else { /* write operation */ 245 DRIVE[sbp->driveno].writes++; 246 DRIVE[sbp->driveno].bytes_written += sbp->b.b_bcount; 247 SD[sbp->sdno].writes++; 248 SD[sbp->sdno].bytes_written += sbp->b.b_bcount; 249 } 250 biodone(sbp->bp); /* complete the caller's I/O */ 251 Free(sbp); 252 } 253 254 /* Start the second phase of a RAID-4 or RAID-5 group write operation. */ 255 void 256 complete_raid5_write(struct rqelement *rqe) 257 { 258 int *sdata; /* source */ 259 int *pdata; /* and parity block data */ 260 int length; /* and count involved */ 261 int count; /* loop counter */ 262 int rqno; /* request index */ 263 int rqoffset; /* offset of request data from parity data */ 264 struct buf *ubp; /* user buffer header */ 265 struct request *rq; /* pointer to our request */ 266 struct rqgroup *rqg; /* and to the request group */ 267 struct rqelement *prqe; /* point to the parity block */ 268 struct drive *drive; /* drive to access */ 269 270 rqg = rqe->rqg; /* and to our request group */ 271 rq = rqg->rq; /* point to our request */ 272 ubp = rq->bp; /* user's buffer header */ 273 prqe = &rqg->rqe[0]; /* point to the parity block */ 274 275 /* 276 * If we get to this function, we have normal or 277 * degraded writes, or a combination of both. We do 278 * the same thing in each case: we perform an 279 * exclusive or to the parity block. The only 280 * difference is the origin of the data and the 281 * address range. 282 */ 283 if (rqe->flags & XFR_DEGRADED_WRITE) { /* do the degraded write stuff */ 284 pdata = (int *) (&prqe->b.b_data[(prqe->groupoffset) << DEV_BSHIFT]); /* parity data pointer */ 285 bzero(pdata, prqe->grouplen << DEV_BSHIFT); /* start with nothing in the parity block */ 286 287 /* Now get what data we need from each block */ 288 for (rqno = 1; rqno < rqg->count; rqno++) { /* for all the data blocks */ 289 rqe = &rqg->rqe[rqno]; /* this request */ 290 sdata = (int *) (&rqe->b.b_data[rqe->groupoffset << DEV_BSHIFT]); /* old data */ 291 length = rqe->grouplen << (DEV_BSHIFT - 2); /* and count involved */ 292 293 /* 294 * Add the data block to the parity block. Before 295 * we started the request, we zeroed the parity 296 * block, so the result of adding all the other 297 * blocks and the block we want to write will be 298 * the correct parity block. 299 */ 300 for (count = 0; count < length; count++) 301 pdata[count] ^= sdata[count]; 302 if ((rqe->flags & XFR_MALLOCED) /* the buffer was malloced, */ 303 &&((rqg->flags & XFR_NORMAL_WRITE) == 0)) { /* and we have no normal write, */ 304 Free(rqe->b.b_data); /* free it now */ 305 rqe->flags &= ~XFR_MALLOCED; 306 } 307 } 308 } 309 if (rqg->flags & XFR_NORMAL_WRITE) { /* do normal write stuff */ 310 /* Get what data we need from each block */ 311 for (rqno = 1; rqno < rqg->count; rqno++) { /* for all the data blocks */ 312 rqe = &rqg->rqe[rqno]; /* this request */ 313 if ((rqe->flags & (XFR_DATA_BLOCK | XFR_BAD_SUBDISK | XFR_NORMAL_WRITE)) 314 == (XFR_DATA_BLOCK | XFR_NORMAL_WRITE)) { /* good data block to write */ 315 sdata = (int *) &rqe->b.b_data[rqe->dataoffset << DEV_BSHIFT]; /* old data contents */ 316 rqoffset = rqe->dataoffset + rqe->sdoffset - prqe->sdoffset; /* corresponding parity block offset */ 317 pdata = (int *) (&prqe->b.b_data[rqoffset << DEV_BSHIFT]); /* parity data pointer */ 318 length = rqe->datalen * (DEV_BSIZE / sizeof(int)); /* and number of ints */ 319 320 /* 321 * "remove" the old data block 322 * from the parity block 323 */ 324 if ((pdata < ((int *) prqe->b.b_data)) 325 || (&pdata[length] > ((int *) (prqe->b.b_data + prqe->b.b_bcount))) 326 || (sdata < ((int *) rqe->b.b_data)) 327 || (&sdata[length] > ((int *) (rqe->b.b_data + rqe->b.b_bcount)))) 328 panic("complete_raid5_write: bounds overflow"); 329 for (count = 0; count < length; count++) 330 pdata[count] ^= sdata[count]; 331 332 /* "add" the new data block */ 333 sdata = (int *) (&ubp->b_data[rqe->useroffset << DEV_BSHIFT]); /* new data */ 334 if ((sdata < ((int *) ubp->b_data)) 335 || (&sdata[length] > ((int *) (ubp->b_data + ubp->b_bcount)))) 336 panic("complete_raid5_write: bounds overflow"); 337 for (count = 0; count < length; count++) 338 pdata[count] ^= sdata[count]; 339 340 /* Free the malloced buffer */ 341 if (rqe->flags & XFR_MALLOCED) { /* the buffer was malloced, */ 342 Free(rqe->b.b_data); /* free it */ 343 rqe->flags &= ~XFR_MALLOCED; 344 } else 345 panic("complete_raid5_write: malloc conflict"); 346 347 if ((rqe->b.b_flags & B_READ) /* this was a read */ 348 &&((rqe->flags & XFR_BAD_SUBDISK) == 0)) { /* and we can write this block */ 349 rqe->b.b_flags &= ~(B_READ | B_DONE); /* we're writing now */ 350 rqe->b.b_flags |= B_CALL; /* call us when you're done */ 351 rqe->b.b_iodone = complete_rqe; /* by calling us here */ 352 rqe->flags &= ~XFR_PARITYOP; /* reset flags that brought us here */ 353 rqe->b.b_data = &ubp->b_data[rqe->useroffset << DEV_BSHIFT]; /* point to the user data */ 354 rqe->b.b_bcount = rqe->datalen << DEV_BSHIFT; /* length to write */ 355 rqe->b.b_bufsize = rqe->b.b_bcount; /* don't claim more */ 356 rqe->b.b_resid = rqe->b.b_bcount; /* nothing transferred */ 357 rqe->b.b_blkno += rqe->dataoffset; /* point to the correct block */ 358 rqg->active++; /* another active request */ 359 drive = &DRIVE[rqe->driveno]; /* drive to access */ 360 361 /* We can't sleep here, so we just increment the counters. */ 362 drive->active++; 363 if (drive->active >= drive->maxactive) 364 drive->maxactive = drive->active; 365 vinum_conf.active++; 366 if (vinum_conf.active >= vinum_conf.maxactive) 367 vinum_conf.maxactive = vinum_conf.active; 368 #if VINUMDEBUG 369 if (debug & DEBUG_ADDRESSES) 370 log(LOG_DEBUG, 371 " %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n", 372 rqe->b.b_flags & B_READ ? "Read" : "Write", 373 major(rqe->b.b_dev), 374 minor(rqe->b.b_dev), 375 rqe->sdno, 376 (u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset), 377 rqe->b.b_blkno, 378 rqe->b.b_bcount); 379 if (debug & DEBUG_LASTREQS) 380 logrq(loginfo_raid5_data, (union rqinfou) rqe, ubp); 381 #endif 382 (*bdevsw[major(rqe->b.b_dev)]->d_strategy) (&rqe->b); 383 } 384 } 385 } 386 } 387 /* Finally, write the parity block */ 388 rqe = &rqg->rqe[0]; 389 rqe->b.b_flags &= ~(B_READ | B_DONE); /* we're writing now */ 390 rqe->b.b_flags |= B_CALL; /* tell us when you're done */ 391 rqe->b.b_iodone = complete_rqe; /* by calling us here */ 392 rqg->flags &= ~XFR_PARITYOP; /* reset flags that brought us here */ 393 rqe->b.b_bcount = rqe->buflen << DEV_BSHIFT; /* length to write */ 394 rqe->b.b_bufsize = rqe->b.b_bcount; /* don't claim we have more */ 395 rqe->b.b_resid = rqe->b.b_bcount; /* nothing transferred */ 396 rqg->active++; /* another active request */ 397 drive = &DRIVE[rqe->driveno]; /* drive to access */ 398 399 /* We can't sleep here, so we just increment the counters. */ 400 drive->active++; 401 if (drive->active >= drive->maxactive) 402 drive->maxactive = drive->active; 403 vinum_conf.active++; 404 if (vinum_conf.active >= vinum_conf.maxactive) 405 vinum_conf.maxactive = vinum_conf.active; 406 407 #if VINUMDEBUG 408 if (debug & DEBUG_ADDRESSES) 409 log(LOG_DEBUG, 410 " %s dev %d.%d, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n", 411 rqe->b.b_flags & B_READ ? "Read" : "Write", 412 major(rqe->b.b_dev), 413 minor(rqe->b.b_dev), 414 rqe->sdno, 415 (u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset), 416 rqe->b.b_blkno, 417 rqe->b.b_bcount); 418 if (debug & DEBUG_LASTREQS) 419 logrq(loginfo_raid5_parity, (union rqinfou) rqe, ubp); 420 #endif 421 (*bdevsw[major(rqe->b.b_dev)]->d_strategy) (&rqe->b); 422 } 423 424 /* Local Variables: */ 425 /* fill-column: 50 */ 426 /* End: */
Cache object: f99cbfa1354678f4db3a7faab7740499
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]