geom_disk.c revision 115214
192108Sphk/*- 292108Sphk * Copyright (c) 2002 Poul-Henning Kamp 392108Sphk * Copyright (c) 2002 Networks Associates Technology, Inc. 492108Sphk * All rights reserved. 592108Sphk * 692108Sphk * This software was developed for the FreeBSD Project by Poul-Henning Kamp 792108Sphk * and NAI Labs, the Security Research Division of Network Associates, Inc. 892108Sphk * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 992108Sphk * DARPA CHATS research program. 1092108Sphk * 1192108Sphk * Redistribution and use in source and binary forms, with or without 1292108Sphk * modification, are permitted provided that the following conditions 1392108Sphk * are met: 1492108Sphk * 1. Redistributions of source code must retain the above copyright 1592108Sphk * notice, this list of conditions and the following disclaimer. 1692108Sphk * 2. Redistributions in binary form must reproduce the above copyright 1792108Sphk * notice, this list of conditions and the following disclaimer in the 1892108Sphk * documentation and/or other materials provided with the distribution. 1992108Sphk * 3. The names of the authors may not be used to endorse or promote 2092108Sphk * products derived from this software without specific prior written 2192108Sphk * permission. 2292108Sphk * 2392108Sphk * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 2492108Sphk * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2592108Sphk * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2692108Sphk * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 2792108Sphk * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2892108Sphk * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2992108Sphk * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 3092108Sphk * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 3192108Sphk * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 3292108Sphk * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3392108Sphk * SUCH DAMAGE. 3492108Sphk * 3592108Sphk * $FreeBSD: head/sys/geom/geom_disk.c 115214 2003-05-21 18:52:29Z phk $ 3692108Sphk */ 3792108Sphk 38104519Sphk#include "opt_geom.h" 39104519Sphk 4092108Sphk#include <sys/param.h> 4192108Sphk#include <sys/systm.h> 4292108Sphk#include <sys/kernel.h> 4392108Sphk#include <sys/sysctl.h> 4492108Sphk#include <sys/bio.h> 4592108Sphk#include <sys/conf.h> 46110119Sphk#include <sys/fcntl.h> 4792108Sphk#include <sys/malloc.h> 4892108Sphk#include <sys/sysctl.h> 49111979Sphk#include <sys/devicestat.h> 5092108Sphk#include <machine/md_var.h> 5192108Sphk 5292108Sphk#include <sys/lock.h> 5392108Sphk#include <sys/mutex.h> 5492108Sphk#include <geom/geom.h> 55112952Sphk#include <geom/geom_disk.h> 56112476Sphk#include <geom/geom_int.h> 5792108Sphk 58110720Sphkstatic struct mtx g_disk_done_mtx; 59110720Sphk 6092108Sphkstatic g_access_t g_disk_access; 6192108Sphk 6293248Sphkstruct g_class g_disk_class = { 63112552Sphk .name = "DISK", 6498066Sphk G_CLASS_INITIALIZER 6592108Sphk}; 6692108Sphk 67110720Sphkstatic void 68110720Sphkg_disk_init(void) 69110720Sphk{ 70110720Sphk mtx_unlock(&Giant); 71110720Sphk g_add_class(&g_disk_class); 72110720Sphk mtx_init(&g_disk_done_mtx, "g_disk_done", MTX_DEF, 0); 73110720Sphk mtx_lock(&Giant); 74110720Sphk} 75104936Sphk 76110720SphkDECLARE_GEOM_CLASS_INIT(g_disk_class, g_disk, g_disk_init); 77110720Sphk 78110052Sphkstatic void __inline 79110052Sphkg_disk_lock_giant(struct disk *dp) 80110052Sphk{ 81110119Sphk if (dp->d_flags & DISKFLAG_NOGIANT) 82110119Sphk return; 83110119Sphk mtx_lock(&Giant); 84110052Sphk} 85110052Sphk 86110052Sphkstatic void __inline 87110052Sphkg_disk_unlock_giant(struct disk *dp) 88110052Sphk{ 89110119Sphk if (dp->d_flags & DISKFLAG_NOGIANT) 90110119Sphk return; 91110119Sphk mtx_unlock(&Giant); 92110052Sphk} 93110052Sphk 9492108Sphkstatic int 9592108Sphkg_disk_access(struct g_provider *pp, int r, int w, int e) 9692108Sphk{ 9792108Sphk struct disk *dp; 9892108Sphk int error; 9992108Sphk 10092108Sphk g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)", 10192108Sphk pp->name, r, w, e); 10292108Sphk g_topology_assert(); 10392108Sphk r += pp->acr; 10492108Sphk w += pp->acw; 10592108Sphk e += pp->ace; 10692108Sphk dp = pp->geom->softc; 107115214Sphk if (dp == NULL) 108115214Sphk return (ENXIO); 109110119Sphk error = 0; 11092108Sphk if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 111111668Sphk if (dp->d_open != NULL) { 112110119Sphk g_disk_lock_giant(dp); 113111668Sphk error = dp->d_open(dp); 114110119Sphk if (error != 0) 115110119Sphk printf("Opened disk %s -> %d\n", 116110119Sphk pp->name, error); 117110119Sphk g_disk_unlock_giant(dp); 118110119Sphk } 119105551Sphk pp->mediasize = dp->d_mediasize; 120105551Sphk pp->sectorsize = dp->d_sectorsize; 121110119Sphk dp->d_flags |= DISKFLAG_OPEN; 122110727Sphk if (dp->d_maxsize == 0) { 123110727Sphk printf("WARNING: Disk drive %s%d has no d_maxsize\n", 124110727Sphk dp->d_name, dp->d_unit); 125110727Sphk dp->d_maxsize = DFLTPHYS; 126110727Sphk } 12792108Sphk } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 128111668Sphk if (dp->d_close != NULL) { 129110119Sphk g_disk_lock_giant(dp); 130111668Sphk error = dp->d_close(dp); 131110119Sphk if (error != 0) 132110119Sphk printf("Closed disk %s -> %d\n", 133110119Sphk pp->name, error); 134110119Sphk g_disk_unlock_giant(dp); 135110119Sphk } 136110119Sphk dp->d_flags &= ~DISKFLAG_OPEN; 13792108Sphk } 13892108Sphk return (error); 13992108Sphk} 14092108Sphk 14192108Sphkstatic void 14295038Sphkg_disk_kerneldump(struct bio *bp, struct disk *dp) 14395038Sphk{ 14495038Sphk int error; 14595038Sphk struct g_kerneldump *gkd; 14695038Sphk struct dumperinfo di; 147104450Sphk struct g_geom *gp; 14895038Sphk 14995038Sphk gkd = (struct g_kerneldump*)bp->bio_data; 150104450Sphk gp = bp->bio_to->geom; 151104450Sphk g_trace(G_T_TOPOLOGY, "g_disk_kernedump(%s, %jd, %jd)", 152104450Sphk gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 153110119Sphk di.dumper = dp->d_dump; 154111220Sphk di.priv = dp; 155103714Sphk di.blocksize = dp->d_sectorsize; 15695038Sphk di.mediaoffset = gkd->offset; 15795038Sphk di.mediasize = gkd->length; 15895038Sphk error = set_dumper(&di); 159104195Sphk g_io_deliver(bp, error); 16095038Sphk} 16195038Sphk 16295038Sphkstatic void 16392108Sphkg_disk_done(struct bio *bp) 16492108Sphk{ 165111979Sphk struct bio *bp2; 166111979Sphk struct disk *dp; 16792108Sphk 168110720Sphk /* See "notes" for why we need a mutex here */ 169110720Sphk /* XXX: will witness accept a mix of Giant/unGiant drivers here ? */ 170110720Sphk mtx_lock(&g_disk_done_mtx); 17192108Sphk bp->bio_completed = bp->bio_length - bp->bio_resid; 172111979Sphk 173111979Sphk bp2 = bp->bio_parent; 174111979Sphk dp = bp2->bio_to->geom->softc; 175111979Sphk if (bp2->bio_error == 0) 176111979Sphk bp2->bio_error = bp->bio_error; 177111979Sphk bp2->bio_completed += bp->bio_completed; 178111979Sphk g_destroy_bio(bp); 179111979Sphk bp2->bio_inbed++; 180111979Sphk if (bp2->bio_children == bp2->bio_inbed) { 181112259Sphk bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed; 182112259Sphk devstat_end_transaction_bio(dp->d_devstat, bp2); 183111979Sphk g_io_deliver(bp2, bp2->bio_error); 184111979Sphk } 185110720Sphk mtx_unlock(&g_disk_done_mtx); 18692108Sphk} 18792108Sphk 18892108Sphkstatic void 18992108Sphkg_disk_start(struct bio *bp) 19092108Sphk{ 191110720Sphk struct bio *bp2, *bp3; 19292108Sphk struct disk *dp; 19392403Sphk struct g_ioctl *gio; 19492403Sphk int error; 195110720Sphk off_t off; 19692108Sphk 19792108Sphk dp = bp->bio_to->geom->softc; 198115214Sphk if (dp == NULL) 199115214Sphk g_io_deliver(bp, ENXIO); 200104609Sphk error = EJUSTRETURN; 20192108Sphk switch(bp->bio_cmd) { 202104609Sphk case BIO_DELETE: 203110119Sphk if (!(dp->d_flags & DISKFLAG_CANDELETE)) { 204104609Sphk error = 0; 205104609Sphk break; 206104609Sphk } 207104609Sphk /* fall-through */ 20892108Sphk case BIO_READ: 20992108Sphk case BIO_WRITE: 210110720Sphk off = 0; 211110720Sphk bp3 = NULL; 21292108Sphk bp2 = g_clone_bio(bp); 213110477Sphk if (bp2 == NULL) { 214110477Sphk error = ENOMEM; 215110477Sphk break; 216110477Sphk } 217112259Sphk devstat_start_transaction_bio(dp->d_devstat, bp); 218110720Sphk do { 219110720Sphk bp2->bio_offset += off; 220110720Sphk bp2->bio_length -= off; 221110766Stegge bp2->bio_data += off; 222110720Sphk if (bp2->bio_length > dp->d_maxsize) { 223110720Sphk /* 224110720Sphk * XXX: If we have a stripesize we should really 225110720Sphk * use it here. 226110720Sphk */ 227110720Sphk bp2->bio_length = dp->d_maxsize; 228110720Sphk off += dp->d_maxsize; 229110720Sphk /* 230110720Sphk * To avoid a race, we need to grab the next bio 231110720Sphk * before we schedule this one. See "notes". 232110720Sphk */ 233110720Sphk bp3 = g_clone_bio(bp); 234110720Sphk if (bp3 == NULL) 235110720Sphk bp->bio_error = ENOMEM; 236110720Sphk } 237110720Sphk bp2->bio_done = g_disk_done; 238110720Sphk bp2->bio_blkno = bp2->bio_offset >> DEV_BSHIFT; 239110720Sphk bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize; 240110720Sphk bp2->bio_bcount = bp2->bio_length; 241110720Sphk bp2->bio_disk = dp; 242110720Sphk g_disk_lock_giant(dp); 243110720Sphk dp->d_strategy(bp2); 244110720Sphk g_disk_unlock_giant(dp); 245110720Sphk bp2 = bp3; 246110720Sphk bp3 = NULL; 247110720Sphk } while (bp2 != NULL); 24892108Sphk break; 24992108Sphk case BIO_GETATTR: 250105551Sphk if (g_handleattr_int(bp, "GEOM::fwsectors", dp->d_fwsectors)) 25192403Sphk break; 252103714Sphk else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads)) 25392403Sphk break; 25498066Sphk else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0)) 25594287Sphk break; 25695038Sphk else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 25795038Sphk g_disk_kerneldump(bp, dp); 258112476Sphk else if ((g_debugflags & G_F_DISKIOCTL) && 259112476Sphk (dp->d_ioctl != NULL) && 260110119Sphk !strcmp(bp->bio_attribute, "GEOM::ioctl") && 26192403Sphk bp->bio_length == sizeof *gio) { 262104602Sphk gio = (struct g_ioctl *)bp->bio_data; 263111668Sphk gio->dev = dp; 264111668Sphk gio->func = (d_ioctl_t *)(dp->d_ioctl); 265104602Sphk error = EDIRIOCTL; 26692403Sphk } else 26792403Sphk error = ENOIOCTL; 26892403Sphk break; 26992108Sphk default: 27092403Sphk error = EOPNOTSUPP; 27192403Sphk break; 27292403Sphk } 273104609Sphk if (error != EJUSTRETURN) 274104195Sphk g_io_deliver(bp, error); 27592403Sphk return; 27692108Sphk} 27792108Sphk 278104936Sphkstatic void 279107953Sphkg_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp) 280105537Sphk{ 281105537Sphk struct disk *dp; 282105537Sphk 283105537Sphk dp = gp->softc; 284106101Sphk if (indent == NULL) { 285106101Sphk sbuf_printf(sb, " hd %u", dp->d_fwheads); 286106101Sphk sbuf_printf(sb, " sc %u", dp->d_fwsectors); 287106101Sphk return; 288106101Sphk } 289105539Sphk if (pp != NULL) { 290105537Sphk sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", 291105537Sphk indent, dp->d_fwheads); 292105537Sphk sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", 293105537Sphk indent, dp->d_fwsectors); 294105537Sphk } 295105537Sphk} 296105537Sphk 297105537Sphkstatic void 298112988Sphkg_disk_create(void *arg, int flag __unused) 29992108Sphk{ 30092108Sphk struct g_geom *gp; 30192108Sphk struct g_provider *pp; 302110708Sphk struct disk *dp; 30392108Sphk 304104936Sphk g_topology_assert(); 305110708Sphk dp = arg; 306110708Sphk gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit); 307104936Sphk gp->start = g_disk_start; 308104936Sphk gp->access = g_disk_access; 309110708Sphk gp->softc = dp; 310105537Sphk gp->dumpconf = g_disk_dumpconf; 311104936Sphk pp = g_new_providerf(gp, "%s", gp->name); 312110708Sphk pp->mediasize = dp->d_mediasize; 313110708Sphk pp->sectorsize = dp->d_sectorsize; 314110708Sphk if (dp->d_flags & DISKFLAG_CANDELETE) 315110708Sphk pp->flags |= G_PF_CANDELETE; 316110710Sphk pp->stripeoffset = dp->d_stripeoffset; 317110710Sphk pp->stripesize = dp->d_stripesize; 318105957Sphk if (bootverbose) 319105957Sphk printf("GEOM: new disk %s\n", gp->name); 320112989Sphk dp->d_geom = gp; 321112989Sphk g_error_provider(pp, 0); 322104936Sphk} 323104936Sphk 324104936Sphk 325104936Sphk 326111668Sphkvoid 327111668Sphkdisk_create(int unit, struct disk *dp, int flags, void *unused __unused, void * unused2 __unused) 328104936Sphk{ 329104936Sphk 330110708Sphk dp->d_unit = unit; 331110119Sphk dp->d_flags = flags; 332110119Sphk KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy")); 333110119Sphk KASSERT(dp->d_name != NULL, ("disk_create need d_name")); 334110119Sphk KASSERT(*dp->d_name != 0, ("disk_create need d_name")); 335110317Sphk KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long")); 336112002Sphk dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit, 337111979Sphk dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED, 338111979Sphk DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 339112989Sphk dp->d_geom = NULL; 340113937Sphk g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL); 34192108Sphk} 34292108Sphk 343112989Sphk/* 344112989Sphk * XXX: There is a race if disk_destroy() is called while the g_disk_create() 345112989Sphk * XXX: event is running. I belive the current result is that disk_destroy() 346112989Sphk * XXX: actually doesn't do anything. Considering that the driver owns the 347112989Sphk * XXX: struct disk and is likely to free it in a few moments, this can 348112989Sphk * XXX: hardly be said to be optimal. To what extent we can sleep in 349112989Sphk * XXX: disk_create() and disk_destroy() is currently undefined (but generally 350112989Sphk * XXX: undesirable) so any solution seems to involve an intrusive decision. 351112989Sphk */ 352114958Sphk 353114958Sphkstatic void 354114958Sphkdisk_destroy_event(void *ptr, int flag) 355114958Sphk{ 356114958Sphk 357114958Sphk g_topology_assert(); 358114958Sphk g_wither_geom(ptr, ENXIO); 359114958Sphk} 360114958Sphk 36192108Sphkvoid 362111216Sphkdisk_destroy(struct disk *dp) 36392108Sphk{ 36492108Sphk struct g_geom *gp; 36592108Sphk 366112989Sphk g_cancel_event(dp); 367110419Sphk gp = dp->d_geom; 368112989Sphk if (gp == NULL) 369112989Sphk return; 370110119Sphk gp->softc = NULL; 371111979Sphk devstat_remove_entry(dp->d_devstat); 372114958Sphk g_post_event(disk_destroy_event, gp, M_WAITOK, NULL, NULL); 37392108Sphk} 37492108Sphk 375104451Sphkstatic void 376112988Sphkg_kern_disks(void *p, int flag __unused) 377104451Sphk{ 378104451Sphk struct sbuf *sb; 379104451Sphk struct g_geom *gp; 380104451Sphk char *sp; 381104451Sphk 382104451Sphk sb = p; 383104451Sphk sp = ""; 384104451Sphk g_topology_assert(); 385104451Sphk LIST_FOREACH(gp, &g_disk_class.geom, geom) { 386104451Sphk sbuf_printf(sb, "%s%s", sp, gp->name); 387104451Sphk sp = " "; 388104451Sphk } 389104451Sphk sbuf_finish(sb); 390104451Sphk} 391104451Sphk 392104451Sphkstatic int 393104451Sphksysctl_disks(SYSCTL_HANDLER_ARGS) 394104451Sphk{ 395104451Sphk int error; 396104451Sphk struct sbuf *sb; 397104451Sphk 398104451Sphk sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 399104451Sphk sbuf_clear(sb); 400113940Sphk g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL); 401113937Sphk error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1); 402104451Sphk sbuf_delete(sb); 403104451Sphk return error; 404104451Sphk} 405104451Sphk 406104451SphkSYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NOLOCK, 0, 0, 407104451Sphk sysctl_disks, "A", "names of available disks"); 408104519Sphk 409