1 /* 2 * Copyright (c) 1982, 1986, 1988 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)ufs_disksubr.c 7.14 (Berkeley) 06/28/90 8 */ 9 10 #include "param.h" 11 #include "systm.h" 12 #include "buf.h" 13 #include "disklabel.h" 14 #include "syslog.h" 15 #include "user.h" 16 17 /* 18 * Seek sort for disks. We depend on the driver 19 * which calls us using b_resid as the current cylinder number. 20 * 21 * The argument dp structure holds a b_actf activity chain pointer 22 * on which we keep two queues, sorted in ascending cylinder order. 23 * The first queue holds those requests which are positioned after 24 * the current cylinder (in the first request); the second holds 25 * requests which came in after their cylinder number was passed. 26 * Thus we implement a one way scan, retracting after reaching the 27 * end of the drive to the first request on the second queue, 28 * at which time it becomes the first queue. 29 * 30 * A one-way scan is natural because of the way UNIX read-ahead 31 * blocks are allocated. 32 */ 33 34 #define b_cylin b_resid 35 36 disksort(dp, bp) 37 register struct buf *dp, *bp; 38 { 39 register struct buf *ap; 40 41 /* 42 * If nothing on the activity queue, then 43 * we become the only thing. 44 */ 45 ap = dp->b_actf; 46 if(ap == NULL) { 47 dp->b_actf = bp; 48 dp->b_actl = bp; 49 bp->av_forw = NULL; 50 return; 51 } 52 /* 53 * If we lie after the first (currently active) 54 * request, then we must locate the second request list 55 * and add ourselves to it. 56 */ 57 if (bp->b_cylin < ap->b_cylin) { 58 while (ap->av_forw) { 59 /* 60 * Check for an ``inversion'' in the 61 * normally ascending cylinder numbers, 62 * indicating the start of the second request list. 63 */ 64 if (ap->av_forw->b_cylin < ap->b_cylin) { 65 /* 66 * Search the second request list 67 * for the first request at a larger 68 * cylinder number. We go before that; 69 * if there is no such request, we go at end. 70 */ 71 do { 72 if (bp->b_cylin < ap->av_forw->b_cylin) 73 goto insert; 74 if (bp->b_cylin == ap->av_forw->b_cylin && 75 bp->b_blkno < ap->av_forw->b_blkno) 76 goto insert; 77 ap = ap->av_forw; 78 } while (ap->av_forw); 79 goto insert; /* after last */ 80 } 81 ap = ap->av_forw; 82 } 83 /* 84 * No inversions... we will go after the last, and 85 * be the first request in the second request list. 86 */ 87 goto insert; 88 } 89 /* 90 * Request is at/after the current request... 91 * sort in the first request list. 92 */ 93 while (ap->av_forw) { 94 /* 95 * We want to go after the current request 96 * if there is an inversion after it (i.e. it is 97 * the end of the first request list), or if 98 * the next request is a larger cylinder than our request. 99 */ 100 if (ap->av_forw->b_cylin < ap->b_cylin || 101 bp->b_cylin < ap->av_forw->b_cylin || 102 (bp->b_cylin == ap->av_forw->b_cylin && 103 bp->b_blkno < ap->av_forw->b_blkno)) 104 goto insert; 105 ap = ap->av_forw; 106 } 107 /* 108 * Neither a second list nor a larger 109 * request... we go at the end of the first list, 110 * which is the same as the end of the whole schebang. 111 */ 112 insert: 113 bp->av_forw = ap->av_forw; 114 ap->av_forw = bp; 115 if (ap == dp->b_actl) 116 dp->b_actl = bp; 117 } 118 119 /* 120 * Attempt to read a disk label from a device 121 * using the indicated stategy routine. 122 * The label must be partly set up before this: 123 * secpercyl and anything required in the strategy routine 124 * (e.g., sector size) must be filled in before calling us. 125 * Returns null on success and an error string on failure. 126 */ 127 char * 128 readdisklabel(dev, strat, lp) 129 dev_t dev; 130 int (*strat)(); 131 register struct disklabel *lp; 132 { 133 register struct buf *bp; 134 struct disklabel *dlp; 135 char *msg = NULL; 136 137 if (lp->d_secperunit == 0) 138 lp->d_secperunit = 0x1fffffff; 139 lp->d_npartitions = 1; 140 if (lp->d_partitions[0].p_size == 0) 141 lp->d_partitions[0].p_size = 0x1fffffff; 142 lp->d_partitions[0].p_offset = 0; 143 144 bp = geteblk((int)lp->d_secsize); 145 bp->b_dev = dev; 146 bp->b_blkno = LABELSECTOR; 147 bp->b_bcount = lp->d_secsize; 148 bp->b_flags = B_BUSY | B_READ; 149 bp->b_cylin = LABELSECTOR / lp->d_secpercyl; 150 (*strat)(bp); 151 if (biowait(bp)) { 152 msg = "I/O error"; 153 } else for (dlp = (struct disklabel *)bp->b_un.b_addr; 154 dlp <= (struct disklabel *)(bp->b_un.b_addr+DEV_BSIZE-sizeof(*dlp)); 155 dlp = (struct disklabel *)((char *)dlp + sizeof(long))) { 156 if (dlp->d_magic != DISKMAGIC || dlp->d_magic2 != DISKMAGIC) { 157 if (msg == NULL) 158 msg = "no disk label"; 159 } else if (dlp->d_npartitions > MAXPARTITIONS || 160 dkcksum(dlp) != 0) 161 msg = "disk label corrupted"; 162 else { 163 *lp = *dlp; 164 msg = NULL; 165 break; 166 } 167 } 168 bp->b_flags = B_INVAL | B_AGE; 169 brelse(bp); 170 return (msg); 171 } 172 173 /* 174 * Check new disk label for sensibility 175 * before setting it. 176 */ 177 setdisklabel(olp, nlp, openmask) 178 register struct disklabel *olp, *nlp; 179 u_long openmask; 180 { 181 register i; 182 register struct partition *opp, *npp; 183 184 if (nlp->d_magic != DISKMAGIC || nlp->d_magic2 != DISKMAGIC || 185 dkcksum(nlp) != 0) 186 return (EINVAL); 187 while ((i = ffs((long)openmask)) != 0) { 188 i--; 189 openmask &= ~(1 << i); 190 if (nlp->d_npartitions <= i) 191 return (EBUSY); 192 opp = &olp->d_partitions[i]; 193 npp = &nlp->d_partitions[i]; 194 if (npp->p_offset != opp->p_offset || npp->p_size < opp->p_size) 195 return (EBUSY); 196 /* 197 * Copy internally-set partition information 198 * if new label doesn't include it. XXX 199 */ 200 if (npp->p_fstype == FS_UNUSED && opp->p_fstype != FS_UNUSED) { 201 npp->p_fstype = opp->p_fstype; 202 npp->p_fsize = opp->p_fsize; 203 npp->p_frag = opp->p_frag; 204 npp->p_cpg = opp->p_cpg; 205 } 206 } 207 nlp->d_checksum = 0; 208 nlp->d_checksum = dkcksum(nlp); 209 *olp = *nlp; 210 return (0); 211 } 212 213 /* encoding of disk minor numbers, should be elsewhere... */ 214 #define dkunit(dev) (minor(dev) >> 3) 215 #define dkpart(dev) (minor(dev) & 07) 216 #define dkminor(unit, part) (((unit) << 3) | (part)) 217 218 /* 219 * Write disk label back to device after modification. 220 */ 221 writedisklabel(dev, strat, lp) 222 dev_t dev; 223 int (*strat)(); 224 register struct disklabel *lp; 225 { 226 struct buf *bp; 227 struct disklabel *dlp; 228 int labelpart; 229 int error = 0; 230 231 labelpart = dkpart(dev); 232 if (lp->d_partitions[labelpart].p_offset != 0) { 233 if (lp->d_partitions[0].p_offset != 0) 234 return (EXDEV); /* not quite right */ 235 labelpart = 0; 236 } 237 bp = geteblk((int)lp->d_secsize); 238 bp->b_dev = makedev(major(dev), dkminor(dkunit(dev), labelpart)); 239 bp->b_blkno = LABELSECTOR; 240 bp->b_bcount = lp->d_secsize; 241 bp->b_flags = B_READ; 242 (*strat)(bp); 243 if (error = biowait(bp)) 244 goto done; 245 for (dlp = (struct disklabel *)bp->b_un.b_addr; 246 dlp <= (struct disklabel *) 247 (bp->b_un.b_addr + lp->d_secsize - sizeof(*dlp)); 248 dlp = (struct disklabel *)((char *)dlp + sizeof(long))) { 249 if (dlp->d_magic == DISKMAGIC && dlp->d_magic2 == DISKMAGIC && 250 dkcksum(dlp) == 0) { 251 *dlp = *lp; 252 bp->b_flags = B_WRITE; 253 (*strat)(bp); 254 error = biowait(bp); 255 goto done; 256 } 257 } 258 error = ESRCH; 259 done: 260 brelse(bp); 261 return (error); 262 } 263 264 /* 265 * Compute checksum for disk label. 266 */ 267 dkcksum(lp) 268 register struct disklabel *lp; 269 { 270 register u_short *start, *end; 271 register u_short sum = 0; 272 273 start = (u_short *)lp; 274 end = (u_short *)&lp->d_partitions[lp->d_npartitions]; 275 while (start < end) 276 sum ^= *start++; 277 return (sum); 278 } 279 280 /* 281 * Disk error is the preface to plaintive error messages 282 * about failing disk transfers. It prints messages of the form 283 284 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 285 286 * if the offset of the error in the transfer and a disk label 287 * are both available. blkdone should be -1 if the position of the error 288 * is unknown; the disklabel pointer may be null from drivers that have not 289 * been converted to use them. The message is printed with printf 290 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 291 * The message should be completed (with at least a newline) with printf 292 * or addlog, respectively. There is no trailing space. 293 */ 294 diskerr(bp, dname, what, pri, blkdone, lp) 295 register struct buf *bp; 296 char *dname, *what; 297 int pri, blkdone; 298 register struct disklabel *lp; 299 { 300 int unit = dkunit(bp->b_dev), part = dkpart(bp->b_dev); 301 register int (*pr)(), sn; 302 char partname = 'a' + part; 303 extern printf(), addlog(); 304 305 if (pri != LOG_PRINTF) { 306 log(pri, ""); 307 pr = addlog; 308 } else 309 pr = printf; 310 (*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what, 311 bp->b_flags & B_READ ? "read" : "writ"); 312 sn = bp->b_blkno; 313 if (bp->b_bcount <= DEV_BSIZE) 314 (*pr)("%d", sn); 315 else { 316 if (blkdone >= 0) { 317 sn += blkdone; 318 (*pr)("%d of ", sn); 319 } 320 (*pr)("%d-%d", bp->b_blkno, 321 bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE); 322 } 323 if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) { 324 #ifdef tahoe 325 sn *= DEV_BSIZE / lp->d_secsize; /* XXX */ 326 #endif 327 sn += lp->d_partitions[part].p_offset; 328 (*pr)(" (%s%d bn %d; cn %d", dname, unit, sn, 329 sn / lp->d_secpercyl); 330 sn %= lp->d_secpercyl; 331 (*pr)(" tn %d sn %d)", sn / lp->d_nsectors, sn % lp->d_nsectors); 332 } 333 } 334