1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5bef6b7d2Swebaker * Common Development and Distribution License (the "License"). 6bef6b7d2Swebaker * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 22f13665b7Sbo zhou - Sun Microsystems - Beijing China * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23be6fd75aSMatthew Ahrens * Copyright (c) 2013 by Delphix. All rights reserved. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #include <sys/zfs_context.h> 27dcba9f3fSGeorge Wilson #include <sys/spa_impl.h> 28e7cbe64fSgw #include <sys/refcount.h> 29fa9e4066Sahrens #include <sys/vdev_disk.h> 30fa9e4066Sahrens #include <sys/vdev_impl.h> 31fa9e4066Sahrens #include <sys/fs/zfs.h> 32fa9e4066Sahrens #include <sys/zio.h> 33afefbcddSeschrock #include <sys/sunldi.h> 344263d13fSGeorge Wilson #include <sys/efi_partition.h> 3551ece835Seschrock #include <sys/fm/fs/zfs.h> 36fa9e4066Sahrens 37fa9e4066Sahrens /* 38fa9e4066Sahrens * Virtual device vector for disks. 39fa9e4066Sahrens */ 40fa9e4066Sahrens 41fa9e4066Sahrens extern ldi_ident_t zfs_li; 42fa9e4066Sahrens 43dcba9f3fSGeorge Wilson static void 44dcba9f3fSGeorge Wilson vdev_disk_hold(vdev_t *vd) 45dcba9f3fSGeorge Wilson { 46dcba9f3fSGeorge Wilson ddi_devid_t devid; 47dcba9f3fSGeorge Wilson char *minor; 48dcba9f3fSGeorge Wilson 49dcba9f3fSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER)); 50dcba9f3fSGeorge Wilson 51dcba9f3fSGeorge Wilson /* 52dcba9f3fSGeorge Wilson * We must have a pathname, and it must be absolute. 53dcba9f3fSGeorge Wilson */ 54dcba9f3fSGeorge Wilson if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') 55dcba9f3fSGeorge Wilson return; 56dcba9f3fSGeorge Wilson 57dcba9f3fSGeorge Wilson /* 58dcba9f3fSGeorge Wilson * Only prefetch path and devid info if the device has 59dcba9f3fSGeorge Wilson * never been opened. 60dcba9f3fSGeorge Wilson */ 61dcba9f3fSGeorge Wilson if (vd->vdev_tsd != NULL) 62dcba9f3fSGeorge Wilson return; 63dcba9f3fSGeorge Wilson 64dcba9f3fSGeorge Wilson if (vd->vdev_wholedisk == -1ULL) { 65dcba9f3fSGeorge Wilson size_t len = strlen(vd->vdev_path) + 3; 66dcba9f3fSGeorge Wilson char *buf = kmem_alloc(len, KM_SLEEP); 67dcba9f3fSGeorge Wilson 68dcba9f3fSGeorge Wilson (void) snprintf(buf, len, "%ss0", vd->vdev_path); 69dcba9f3fSGeorge Wilson 70dcba9f3fSGeorge Wilson (void) ldi_vp_from_name(buf, &vd->vdev_name_vp); 71dcba9f3fSGeorge Wilson kmem_free(buf, len); 72dcba9f3fSGeorge Wilson } 73dcba9f3fSGeorge Wilson 74dcba9f3fSGeorge Wilson if (vd->vdev_name_vp == NULL) 75dcba9f3fSGeorge Wilson (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp); 76dcba9f3fSGeorge Wilson 77dcba9f3fSGeorge Wilson if (vd->vdev_devid != NULL && 78dcba9f3fSGeorge Wilson ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) { 79dcba9f3fSGeorge Wilson (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp); 80dcba9f3fSGeorge Wilson ddi_devid_str_free(minor); 81dcba9f3fSGeorge Wilson ddi_devid_free(devid); 82dcba9f3fSGeorge Wilson } 83dcba9f3fSGeorge Wilson } 84dcba9f3fSGeorge Wilson 85dcba9f3fSGeorge Wilson static void 86dcba9f3fSGeorge Wilson vdev_disk_rele(vdev_t *vd) 87dcba9f3fSGeorge Wilson { 88dcba9f3fSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER)); 89dcba9f3fSGeorge Wilson 90dcba9f3fSGeorge Wilson if (vd->vdev_name_vp) { 91dcba9f3fSGeorge Wilson VN_RELE_ASYNC(vd->vdev_name_vp, 92dcba9f3fSGeorge Wilson dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool)); 93dcba9f3fSGeorge Wilson vd->vdev_name_vp = NULL; 94dcba9f3fSGeorge Wilson } 95dcba9f3fSGeorge Wilson if (vd->vdev_devid_vp) { 96dcba9f3fSGeorge Wilson VN_RELE_ASYNC(vd->vdev_devid_vp, 97dcba9f3fSGeorge Wilson dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool)); 98dcba9f3fSGeorge Wilson vd->vdev_devid_vp = NULL; 99dcba9f3fSGeorge Wilson } 100dcba9f3fSGeorge Wilson } 101dcba9f3fSGeorge Wilson 1024263d13fSGeorge Wilson static uint64_t 1034263d13fSGeorge Wilson vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz) 1044263d13fSGeorge Wilson { 1054263d13fSGeorge Wilson ASSERT(vd->vdev_wholedisk); 1064263d13fSGeorge Wilson 1074263d13fSGeorge Wilson vdev_disk_t *dvd = vd->vdev_tsd; 1084263d13fSGeorge Wilson dk_efi_t dk_ioc; 1094263d13fSGeorge Wilson efi_gpt_t *efi; 1104263d13fSGeorge Wilson uint64_t avail_space = 0; 1114263d13fSGeorge Wilson int efisize = EFI_LABEL_SIZE * 2; 1124263d13fSGeorge Wilson 1134263d13fSGeorge Wilson dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP); 1144263d13fSGeorge Wilson dk_ioc.dki_lba = 1; 1154263d13fSGeorge Wilson dk_ioc.dki_length = efisize; 1164263d13fSGeorge Wilson dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data; 1174263d13fSGeorge Wilson efi = dk_ioc.dki_data; 1184263d13fSGeorge Wilson 1194263d13fSGeorge Wilson if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc, 1204263d13fSGeorge Wilson FKIOCTL, kcred, NULL) == 0) { 1214263d13fSGeorge Wilson uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA); 1224263d13fSGeorge Wilson 1234263d13fSGeorge Wilson zfs_dbgmsg("vdev %s, capacity %llu, altern lba %llu", 1244263d13fSGeorge Wilson vd->vdev_path, capacity, efi_altern_lba); 1254263d13fSGeorge Wilson if (capacity > efi_altern_lba) 1264263d13fSGeorge Wilson avail_space = (capacity - efi_altern_lba) * blksz; 1274263d13fSGeorge Wilson } 1284263d13fSGeorge Wilson kmem_free(dk_ioc.dki_data, efisize); 1294263d13fSGeorge Wilson return (avail_space); 1304263d13fSGeorge Wilson } 1314263d13fSGeorge Wilson 132fa9e4066Sahrens static int 1334263d13fSGeorge Wilson vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize, 1344263d13fSGeorge Wilson uint64_t *ashift) 135fa9e4066Sahrens { 1368ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 137fa9e4066Sahrens vdev_disk_t *dvd; 138f13665b7Sbo zhou - Sun Microsystems - Beijing China struct dk_minfo_ext dkmext; 1390a4e9518Sgw int error; 140e14bb325SJeff Bonwick dev_t dev; 141e14bb325SJeff Bonwick int otyp; 142*fb02ae02SGeorge Wilson boolean_t validate_devid = B_FALSE; 143*fb02ae02SGeorge Wilson ddi_devid_t devid; 144fa9e4066Sahrens 145fa9e4066Sahrens /* 146fa9e4066Sahrens * We must have a pathname, and it must be absolute. 147fa9e4066Sahrens */ 148fa9e4066Sahrens if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') { 149fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; 150be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 151fa9e4066Sahrens } 152fa9e4066Sahrens 153095bcd66SGeorge Wilson /* 154095bcd66SGeorge Wilson * Reopen the device if it's not currently open. Otherwise, 155095bcd66SGeorge Wilson * just update the physical size of the device. 156095bcd66SGeorge Wilson */ 157095bcd66SGeorge Wilson if (vd->vdev_tsd != NULL) { 158095bcd66SGeorge Wilson ASSERT(vd->vdev_reopening); 159095bcd66SGeorge Wilson dvd = vd->vdev_tsd; 160095bcd66SGeorge Wilson goto skip_open; 161095bcd66SGeorge Wilson } 162095bcd66SGeorge Wilson 163fa9e4066Sahrens dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP); 164fa9e4066Sahrens 165fa9e4066Sahrens /* 166fa9e4066Sahrens * When opening a disk device, we want to preserve the user's original 167fa9e4066Sahrens * intent. We always want to open the device by the path the user gave 1681724dc7bSJoshua M. Clulow * us, even if it is one of multiple paths to the same device. But we 169fa9e4066Sahrens * also want to be able to survive disks being removed/recabled. 170fa9e4066Sahrens * Therefore the sequence of opening devices is: 171fa9e4066Sahrens * 172afefbcddSeschrock * 1. Try opening the device by path. For legacy pools without the 173afefbcddSeschrock * 'whole_disk' property, attempt to fix the path by appending 's0'. 174fa9e4066Sahrens * 175fa9e4066Sahrens * 2. If the devid of the device matches the stored value, return 176fa9e4066Sahrens * success. 177fa9e4066Sahrens * 178fa9e4066Sahrens * 3. Otherwise, the device may have moved. Try opening the device 179fa9e4066Sahrens * by the devid instead. 180fa9e4066Sahrens */ 181fa9e4066Sahrens if (vd->vdev_devid != NULL) { 182fa9e4066Sahrens if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid, 183fa9e4066Sahrens &dvd->vd_minor) != 0) { 184fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; 185be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 186fa9e4066Sahrens } 187fa9e4066Sahrens } 188fa9e4066Sahrens 189fa9e4066Sahrens error = EINVAL; /* presume failure */ 190fa9e4066Sahrens 191095bcd66SGeorge Wilson if (vd->vdev_path != NULL) { 192fa9e4066Sahrens 193afefbcddSeschrock if (vd->vdev_wholedisk == -1ULL) { 194afefbcddSeschrock size_t len = strlen(vd->vdev_path) + 3; 195afefbcddSeschrock char *buf = kmem_alloc(len, KM_SLEEP); 196afefbcddSeschrock ldi_handle_t lh; 197afefbcddSeschrock 198afefbcddSeschrock (void) snprintf(buf, len, "%ss0", vd->vdev_path); 199afefbcddSeschrock 2008ad4d6ddSJeff Bonwick if (ldi_open_by_name(buf, spa_mode(spa), kcred, 201afefbcddSeschrock &lh, zfs_li) == 0) { 202afefbcddSeschrock spa_strfree(vd->vdev_path); 203afefbcddSeschrock vd->vdev_path = buf; 204afefbcddSeschrock vd->vdev_wholedisk = 1ULL; 2058ad4d6ddSJeff Bonwick (void) ldi_close(lh, spa_mode(spa), kcred); 206afefbcddSeschrock } else { 207afefbcddSeschrock kmem_free(buf, len); 208afefbcddSeschrock } 209afefbcddSeschrock } 210fa9e4066Sahrens 2118ad4d6ddSJeff Bonwick error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred, 212afefbcddSeschrock &dvd->vd_lh, zfs_li); 213fa9e4066Sahrens 214fa9e4066Sahrens /* 215fa9e4066Sahrens * Compare the devid to the stored value. 216fa9e4066Sahrens */ 217fa9e4066Sahrens if (error == 0 && vd->vdev_devid != NULL && 218fa9e4066Sahrens ldi_get_devid(dvd->vd_lh, &devid) == 0) { 219fa9e4066Sahrens if (ddi_devid_compare(devid, dvd->vd_devid) != 0) { 220be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 2218ad4d6ddSJeff Bonwick (void) ldi_close(dvd->vd_lh, spa_mode(spa), 2228ad4d6ddSJeff Bonwick kcred); 223fa9e4066Sahrens dvd->vd_lh = NULL; 224fa9e4066Sahrens } 225fa9e4066Sahrens ddi_devid_free(devid); 226fa9e4066Sahrens } 227afefbcddSeschrock 228afefbcddSeschrock /* 229afefbcddSeschrock * If we succeeded in opening the device, but 'vdev_wholedisk' 230afefbcddSeschrock * is not yet set, then this must be a slice. 231afefbcddSeschrock */ 232afefbcddSeschrock if (error == 0 && vd->vdev_wholedisk == -1ULL) 233afefbcddSeschrock vd->vdev_wholedisk = 0; 234fa9e4066Sahrens } 235fa9e4066Sahrens 236fa9e4066Sahrens /* 237fa9e4066Sahrens * If we were unable to open by path, or the devid check fails, open by 238fa9e4066Sahrens * devid instead. 239fa9e4066Sahrens */ 240*fb02ae02SGeorge Wilson if (error != 0 && vd->vdev_devid != NULL) { 241fa9e4066Sahrens error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor, 2428ad4d6ddSJeff Bonwick spa_mode(spa), kcred, &dvd->vd_lh, zfs_li); 243*fb02ae02SGeorge Wilson } 244fa9e4066Sahrens 2453d7072f8Seschrock /* 2463d7072f8Seschrock * If all else fails, then try opening by physical path (if available) 2473d7072f8Seschrock * or the logical path (if we failed due to the devid check). While not 2483d7072f8Seschrock * as reliable as the devid, this will give us something, and the higher 2493d7072f8Seschrock * level vdev validation will prevent us from opening the wrong device. 2503d7072f8Seschrock */ 2513d7072f8Seschrock if (error) { 252*fb02ae02SGeorge Wilson if (vd->vdev_devid != NULL) 253*fb02ae02SGeorge Wilson validate_devid = B_TRUE; 254*fb02ae02SGeorge Wilson 2553d7072f8Seschrock if (vd->vdev_physpath != NULL && 256deb8317bSMark J Musante (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV) 2578ad4d6ddSJeff Bonwick error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa), 2583d7072f8Seschrock kcred, &dvd->vd_lh, zfs_li); 2593d7072f8Seschrock 2603d7072f8Seschrock /* 2613d7072f8Seschrock * Note that we don't support the legacy auto-wholedisk support 2623d7072f8Seschrock * as above. This hasn't been used in a very long time and we 2633d7072f8Seschrock * don't need to propagate its oddities to this edge condition. 2643d7072f8Seschrock */ 265095bcd66SGeorge Wilson if (error && vd->vdev_path != NULL) 2668ad4d6ddSJeff Bonwick error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), 2678ad4d6ddSJeff Bonwick kcred, &dvd->vd_lh, zfs_li); 2683d7072f8Seschrock } 2693d7072f8Seschrock 270e14bb325SJeff Bonwick if (error) { 271fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 272fa9e4066Sahrens return (error); 273e14bb325SJeff Bonwick } 274fa9e4066Sahrens 275*fb02ae02SGeorge Wilson /* 276*fb02ae02SGeorge Wilson * Now that the device has been successfully opened, update the devid 277*fb02ae02SGeorge Wilson * if necessary. 278*fb02ae02SGeorge Wilson */ 279*fb02ae02SGeorge Wilson if (validate_devid && spa_writeable(spa) && 280*fb02ae02SGeorge Wilson ldi_get_devid(dvd->vd_lh, &devid) == 0) { 281*fb02ae02SGeorge Wilson if (ddi_devid_compare(devid, dvd->vd_devid) != 0) { 282*fb02ae02SGeorge Wilson char *vd_devid; 283*fb02ae02SGeorge Wilson 284*fb02ae02SGeorge Wilson vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor); 285*fb02ae02SGeorge Wilson zfs_dbgmsg("vdev %s: update devid from %s, " 286*fb02ae02SGeorge Wilson "to %s", vd->vdev_path, vd->vdev_devid, vd_devid); 287*fb02ae02SGeorge Wilson spa_strfree(vd->vdev_devid); 288*fb02ae02SGeorge Wilson vd->vdev_devid = spa_strdup(vd_devid); 289*fb02ae02SGeorge Wilson ddi_devid_str_free(vd_devid); 290*fb02ae02SGeorge Wilson } 291*fb02ae02SGeorge Wilson ddi_devid_free(devid); 292*fb02ae02SGeorge Wilson } 293*fb02ae02SGeorge Wilson 2943d7072f8Seschrock /* 2953d7072f8Seschrock * Once a device is opened, verify that the physical device path (if 2963d7072f8Seschrock * available) is up to date. 2973d7072f8Seschrock */ 2983d7072f8Seschrock if (ldi_get_dev(dvd->vd_lh, &dev) == 0 && 2993d7072f8Seschrock ldi_get_otyp(dvd->vd_lh, &otyp) == 0) { 3000a4e9518Sgw char *physpath, *minorname; 3010a4e9518Sgw 3023d7072f8Seschrock physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3033d7072f8Seschrock minorname = NULL; 3043d7072f8Seschrock if (ddi_dev_pathname(dev, otyp, physpath) == 0 && 3053d7072f8Seschrock ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 && 3063d7072f8Seschrock (vd->vdev_physpath == NULL || 3073d7072f8Seschrock strcmp(vd->vdev_physpath, physpath) != 0)) { 3083d7072f8Seschrock if (vd->vdev_physpath) 3093d7072f8Seschrock spa_strfree(vd->vdev_physpath); 3103d7072f8Seschrock (void) strlcat(physpath, ":", MAXPATHLEN); 3113d7072f8Seschrock (void) strlcat(physpath, minorname, MAXPATHLEN); 3123d7072f8Seschrock vd->vdev_physpath = spa_strdup(physpath); 3133d7072f8Seschrock } 3143d7072f8Seschrock if (minorname) 3153d7072f8Seschrock kmem_free(minorname, strlen(minorname) + 1); 3163d7072f8Seschrock kmem_free(physpath, MAXPATHLEN); 3173d7072f8Seschrock } 3183d7072f8Seschrock 319095bcd66SGeorge Wilson skip_open: 320fa9e4066Sahrens /* 321fa9e4066Sahrens * Determine the actual size of the device. 322fa9e4066Sahrens */ 323fa9e4066Sahrens if (ldi_get_size(dvd->vd_lh, psize) != 0) { 324fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 325be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 326fa9e4066Sahrens } 327fa9e4066Sahrens 328ecc2d604Sbonwick /* 329ecc2d604Sbonwick * Determine the device's minimum transfer size. 330ecc2d604Sbonwick * If the ioctl isn't supported, assume DEV_BSIZE. 331ecc2d604Sbonwick */ 332f13665b7Sbo zhou - Sun Microsystems - Beijing China if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext, 333ecc2d604Sbonwick FKIOCTL, kcred, NULL) != 0) 334f13665b7Sbo zhou - Sun Microsystems - Beijing China dkmext.dki_pbsize = DEV_BSIZE; 335bef6b7d2Swebaker 336f13665b7Sbo zhou - Sun Microsystems - Beijing China *ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1; 337bef6b7d2Swebaker 3384263d13fSGeorge Wilson if (vd->vdev_wholedisk == 1) { 3394263d13fSGeorge Wilson uint64_t capacity = dkmext.dki_capacity - 1; 3404263d13fSGeorge Wilson uint64_t blksz = dkmext.dki_lbsize; 3414263d13fSGeorge Wilson int wce = 1; 3424263d13fSGeorge Wilson 3434263d13fSGeorge Wilson /* 3444263d13fSGeorge Wilson * If we own the whole disk, try to enable disk write caching. 3454263d13fSGeorge Wilson * We ignore errors because it's OK if we can't do it. 3464263d13fSGeorge Wilson */ 3474263d13fSGeorge Wilson (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce, 3484263d13fSGeorge Wilson FKIOCTL, kcred, NULL); 3494263d13fSGeorge Wilson 3504263d13fSGeorge Wilson *max_psize = *psize + vdev_disk_get_space(vd, capacity, blksz); 3514263d13fSGeorge Wilson zfs_dbgmsg("capacity change: vdev %s, psize %llu, " 3524263d13fSGeorge Wilson "max_psize %llu", vd->vdev_path, *psize, *max_psize); 3534263d13fSGeorge Wilson } else { 3544263d13fSGeorge Wilson *max_psize = *psize; 3554263d13fSGeorge Wilson } 3564263d13fSGeorge Wilson 357b468a217Seschrock /* 358b468a217Seschrock * Clear the nowritecache bit, so that on a vdev_reopen() we will 359b468a217Seschrock * try again. 360b468a217Seschrock */ 361b468a217Seschrock vd->vdev_nowritecache = B_FALSE; 362b468a217Seschrock 363fa9e4066Sahrens return (0); 364fa9e4066Sahrens } 365fa9e4066Sahrens 366fa9e4066Sahrens static void 367fa9e4066Sahrens vdev_disk_close(vdev_t *vd) 368fa9e4066Sahrens { 369fa9e4066Sahrens vdev_disk_t *dvd = vd->vdev_tsd; 370fa9e4066Sahrens 371095bcd66SGeorge Wilson if (vd->vdev_reopening || dvd == NULL) 372fa9e4066Sahrens return; 373fa9e4066Sahrens 374fa9e4066Sahrens if (dvd->vd_minor != NULL) 375fa9e4066Sahrens ddi_devid_str_free(dvd->vd_minor); 376fa9e4066Sahrens 377fa9e4066Sahrens if (dvd->vd_devid != NULL) 378fa9e4066Sahrens ddi_devid_free(dvd->vd_devid); 379fa9e4066Sahrens 380fa9e4066Sahrens if (dvd->vd_lh != NULL) 3818ad4d6ddSJeff Bonwick (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred); 382fa9e4066Sahrens 38398d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 384fa9e4066Sahrens kmem_free(dvd, sizeof (vdev_disk_t)); 385fa9e4066Sahrens vd->vdev_tsd = NULL; 386fa9e4066Sahrens } 387fa9e4066Sahrens 388e7cbe64fSgw int 389e7cbe64fSgw vdev_disk_physio(ldi_handle_t vd_lh, caddr_t data, size_t size, 390e7cbe64fSgw uint64_t offset, int flags) 391e7cbe64fSgw { 392e7cbe64fSgw buf_t *bp; 393e7cbe64fSgw int error = 0; 394e7cbe64fSgw 395e7cbe64fSgw if (vd_lh == NULL) 396be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 397e7cbe64fSgw 398e7cbe64fSgw ASSERT(flags & B_READ || flags & B_WRITE); 399e7cbe64fSgw 400e7cbe64fSgw bp = getrbuf(KM_SLEEP); 401e7cbe64fSgw bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST; 402e7cbe64fSgw bp->b_bcount = size; 403e7cbe64fSgw bp->b_un.b_addr = (void *)data; 404e7cbe64fSgw bp->b_lblkno = lbtodb(offset); 405e7cbe64fSgw bp->b_bufsize = size; 406e7cbe64fSgw 407e7cbe64fSgw error = ldi_strategy(vd_lh, bp); 408e7cbe64fSgw ASSERT(error == 0); 409e7cbe64fSgw if ((error = biowait(bp)) == 0 && bp->b_resid != 0) 410be6fd75aSMatthew Ahrens error = SET_ERROR(EIO); 411e7cbe64fSgw freerbuf(bp); 412e7cbe64fSgw 413e7cbe64fSgw return (error); 414e7cbe64fSgw } 415e7cbe64fSgw 416fa9e4066Sahrens static void 417fa9e4066Sahrens vdev_disk_io_intr(buf_t *bp) 418fa9e4066Sahrens { 41931d7e8faSGeorge Wilson vdev_buf_t *vb = (vdev_buf_t *)bp; 42031d7e8faSGeorge Wilson zio_t *zio = vb->vb_io; 421fa9e4066Sahrens 42251ece835Seschrock /* 42351ece835Seschrock * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO. 42451ece835Seschrock * Rather than teach the rest of the stack about other error 42551ece835Seschrock * possibilities (EFAULT, etc), we normalize the error value here. 42651ece835Seschrock */ 42751ece835Seschrock zio->io_error = (geterror(bp) != 0 ? EIO : 0); 42851ece835Seschrock 42951ece835Seschrock if (zio->io_error == 0 && bp->b_resid != 0) 430be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(EIO); 431fa9e4066Sahrens 43231d7e8faSGeorge Wilson kmem_free(vb, sizeof (vdev_buf_t)); 433fa9e4066Sahrens 434e05725b1Sbonwick zio_interrupt(zio); 435fa9e4066Sahrens } 436fa9e4066Sahrens 437f4a72450SJeff Bonwick static void 438f4a72450SJeff Bonwick vdev_disk_ioctl_free(zio_t *zio) 439f4a72450SJeff Bonwick { 440f4a72450SJeff Bonwick kmem_free(zio->io_vsd, sizeof (struct dk_callback)); 441f4a72450SJeff Bonwick } 442f4a72450SJeff Bonwick 44322fe2c88SJonathan Adams static const zio_vsd_ops_t vdev_disk_vsd_ops = { 44422fe2c88SJonathan Adams vdev_disk_ioctl_free, 44522fe2c88SJonathan Adams zio_vsd_default_cksum_report 44622fe2c88SJonathan Adams }; 44722fe2c88SJonathan Adams 448fa9e4066Sahrens static void 449fa9e4066Sahrens vdev_disk_ioctl_done(void *zio_arg, int error) 450fa9e4066Sahrens { 451fa9e4066Sahrens zio_t *zio = zio_arg; 452fa9e4066Sahrens 453fa9e4066Sahrens zio->io_error = error; 454fa9e4066Sahrens 455e05725b1Sbonwick zio_interrupt(zio); 456fa9e4066Sahrens } 457fa9e4066Sahrens 458e05725b1Sbonwick static int 459fa9e4066Sahrens vdev_disk_io_start(zio_t *zio) 460fa9e4066Sahrens { 461fa9e4066Sahrens vdev_t *vd = zio->io_vd; 462fa9e4066Sahrens vdev_disk_t *dvd = vd->vdev_tsd; 46331d7e8faSGeorge Wilson vdev_buf_t *vb; 464e14bb325SJeff Bonwick struct dk_callback *dkc; 465fa9e4066Sahrens buf_t *bp; 466e14bb325SJeff Bonwick int error; 467fa9e4066Sahrens 468fa9e4066Sahrens if (zio->io_type == ZIO_TYPE_IOCTL) { 469fa9e4066Sahrens /* XXPOLICY */ 4700a4e9518Sgw if (!vdev_readable(vd)) { 471be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 472e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 473fa9e4066Sahrens } 474fa9e4066Sahrens 475fa9e4066Sahrens switch (zio->io_cmd) { 476fa9e4066Sahrens 477fa9e4066Sahrens case DKIOCFLUSHWRITECACHE: 478fa9e4066Sahrens 479a2eea2e1Sahrens if (zfs_nocacheflush) 480a2eea2e1Sahrens break; 481a2eea2e1Sahrens 482b468a217Seschrock if (vd->vdev_nowritecache) { 483be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENOTSUP); 484b468a217Seschrock break; 485b468a217Seschrock } 486b468a217Seschrock 487e14bb325SJeff Bonwick zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP); 48822fe2c88SJonathan Adams zio->io_vsd_ops = &vdev_disk_vsd_ops; 489e14bb325SJeff Bonwick 490e14bb325SJeff Bonwick dkc->dkc_callback = vdev_disk_ioctl_done; 491e14bb325SJeff Bonwick dkc->dkc_flag = FLUSH_VOLATILE; 492e14bb325SJeff Bonwick dkc->dkc_cookie = zio; 493fa9e4066Sahrens 494fa9e4066Sahrens error = ldi_ioctl(dvd->vd_lh, zio->io_cmd, 495e14bb325SJeff Bonwick (uintptr_t)dkc, FKIOCTL, kcred, NULL); 496fa9e4066Sahrens 497fa9e4066Sahrens if (error == 0) { 498fa9e4066Sahrens /* 499fa9e4066Sahrens * The ioctl will be done asychronously, 500fa9e4066Sahrens * and will call vdev_disk_ioctl_done() 501fa9e4066Sahrens * upon completion. 502fa9e4066Sahrens */ 503e05725b1Sbonwick return (ZIO_PIPELINE_STOP); 504e05725b1Sbonwick } 505e05725b1Sbonwick 506e05725b1Sbonwick if (error == ENOTSUP || error == ENOTTY) { 507b468a217Seschrock /* 508d5782879Smishra * If we get ENOTSUP or ENOTTY, we know that 509d5782879Smishra * no future attempts will ever succeed. 510d5782879Smishra * In this case we set a persistent bit so 511d5782879Smishra * that we don't bother with the ioctl in the 512d5782879Smishra * future. 513b468a217Seschrock */ 514b468a217Seschrock vd->vdev_nowritecache = B_TRUE; 515fa9e4066Sahrens } 516fa9e4066Sahrens zio->io_error = error; 517b468a217Seschrock 518fa9e4066Sahrens break; 519fa9e4066Sahrens 520fa9e4066Sahrens default: 521be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENOTSUP); 522fa9e4066Sahrens } 523fa9e4066Sahrens 524e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 525fa9e4066Sahrens } 526fa9e4066Sahrens 52731d7e8faSGeorge Wilson vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP); 528fa9e4066Sahrens 52931d7e8faSGeorge Wilson vb->vb_io = zio; 53031d7e8faSGeorge Wilson bp = &vb->vb_buf; 531fa9e4066Sahrens 532fa9e4066Sahrens bioinit(bp); 533e14bb325SJeff Bonwick bp->b_flags = B_BUSY | B_NOCACHE | 5348956713aSEric Schrock (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE); 5358956713aSEric Schrock if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD))) 5368956713aSEric Schrock bp->b_flags |= B_FAILFAST; 537fa9e4066Sahrens bp->b_bcount = zio->io_size; 538fa9e4066Sahrens bp->b_un.b_addr = zio->io_data; 539fa9e4066Sahrens bp->b_lblkno = lbtodb(zio->io_offset); 540fa9e4066Sahrens bp->b_bufsize = zio->io_size; 541fa9e4066Sahrens bp->b_iodone = (int (*)())vdev_disk_io_intr; 542fa9e4066Sahrens 543fa9e4066Sahrens /* ldi_strategy() will return non-zero only on programming errors */ 544e14bb325SJeff Bonwick VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0); 545e05725b1Sbonwick 546e05725b1Sbonwick return (ZIO_PIPELINE_STOP); 547fa9e4066Sahrens } 548fa9e4066Sahrens 549e14bb325SJeff Bonwick static void 550fa9e4066Sahrens vdev_disk_io_done(zio_t *zio) 551fa9e4066Sahrens { 552e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd; 553ea8dc4b6Seschrock 5543d7072f8Seschrock /* 5553d7072f8Seschrock * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if 5563d7072f8Seschrock * the device has been removed. If this is the case, then we trigger an 5570a4e9518Sgw * asynchronous removal of the device. Otherwise, probe the device and 5581f7ad2e1Sgw * make sure it's still accessible. 5593d7072f8Seschrock */ 5601d713200SEric Schrock if (zio->io_error == EIO && !vd->vdev_remove_wanted) { 5610a4e9518Sgw vdev_disk_t *dvd = vd->vdev_tsd; 562e14bb325SJeff Bonwick int state = DKIO_NONE; 5630a4e9518Sgw 564e14bb325SJeff Bonwick if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state, 565e14bb325SJeff Bonwick FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) { 5661d713200SEric Schrock /* 5671d713200SEric Schrock * We post the resource as soon as possible, instead of 5681d713200SEric Schrock * when the async removal actually happens, because the 5691d713200SEric Schrock * DE is using this information to discard previous I/O 5701d713200SEric Schrock * errors. 5711d713200SEric Schrock */ 5721d713200SEric Schrock zfs_post_remove(zio->io_spa, vd); 5733d7072f8Seschrock vd->vdev_remove_wanted = B_TRUE; 5743d7072f8Seschrock spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE); 57598d1cbfeSGeorge Wilson } else if (!vd->vdev_delayed_close) { 57698d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_TRUE; 5773d7072f8Seschrock } 5783d7072f8Seschrock } 579fa9e4066Sahrens } 580fa9e4066Sahrens 581fa9e4066Sahrens vdev_ops_t vdev_disk_ops = { 582fa9e4066Sahrens vdev_disk_open, 583fa9e4066Sahrens vdev_disk_close, 584fa9e4066Sahrens vdev_default_asize, 585fa9e4066Sahrens vdev_disk_io_start, 586fa9e4066Sahrens vdev_disk_io_done, 587fa9e4066Sahrens NULL, 588dcba9f3fSGeorge Wilson vdev_disk_hold, 589dcba9f3fSGeorge Wilson vdev_disk_rele, 590fa9e4066Sahrens VDEV_TYPE_DISK, /* name of this vdev type */ 591fa9e4066Sahrens B_TRUE /* leaf vdev */ 592fa9e4066Sahrens }; 593e7cbe64fSgw 594e7cbe64fSgw /* 595051aabe6Staylor * Given the root disk device devid or pathname, read the label from 596051aabe6Staylor * the device, and construct a configuration nvlist. 597e7cbe64fSgw */ 598f940fbb1SLin Ling int 599f940fbb1SLin Ling vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config) 600e7cbe64fSgw { 601e7cbe64fSgw ldi_handle_t vd_lh; 602e7cbe64fSgw vdev_label_t *label; 603e7cbe64fSgw uint64_t s, size; 604e7cbe64fSgw int l; 605051aabe6Staylor ddi_devid_t tmpdevid; 606f4565e39SLin Ling int error = -1; 607051aabe6Staylor char *minor_name; 608e7cbe64fSgw 609e7cbe64fSgw /* 610e7cbe64fSgw * Read the device label and build the nvlist. 611e7cbe64fSgw */ 612f4565e39SLin Ling if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid, 613051aabe6Staylor &minor_name) == 0) { 614051aabe6Staylor error = ldi_open_by_devid(tmpdevid, minor_name, 6158ad4d6ddSJeff Bonwick FREAD, kcred, &vd_lh, zfs_li); 616051aabe6Staylor ddi_devid_free(tmpdevid); 617051aabe6Staylor ddi_devid_str_free(minor_name); 618051aabe6Staylor } 619051aabe6Staylor 620f4565e39SLin Ling if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh, 621f4565e39SLin Ling zfs_li))) 622f940fbb1SLin Ling return (error); 623e7cbe64fSgw 624bf82a41bSeschrock if (ldi_get_size(vd_lh, &s)) { 625bf82a41bSeschrock (void) ldi_close(vd_lh, FREAD, kcred); 626be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 627bf82a41bSeschrock } 628e7cbe64fSgw 629e7cbe64fSgw size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t); 630e7cbe64fSgw label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP); 631e7cbe64fSgw 63217f1e64aSEric Taylor *config = NULL; 633e7cbe64fSgw for (l = 0; l < VDEV_LABELS; l++) { 634e7cbe64fSgw uint64_t offset, state, txg = 0; 635e7cbe64fSgw 636e7cbe64fSgw /* read vdev label */ 637e7cbe64fSgw offset = vdev_label_offset(size, l, 0); 638e7cbe64fSgw if (vdev_disk_physio(vd_lh, (caddr_t)label, 6392264ca7fSLin Ling VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0) 640e7cbe64fSgw continue; 641e7cbe64fSgw 642e7cbe64fSgw if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist, 643f940fbb1SLin Ling sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) { 644f940fbb1SLin Ling *config = NULL; 645e7cbe64fSgw continue; 646e7cbe64fSgw } 647e7cbe64fSgw 648f940fbb1SLin Ling if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE, 649e7cbe64fSgw &state) != 0 || state >= POOL_STATE_DESTROYED) { 650f940fbb1SLin Ling nvlist_free(*config); 651f940fbb1SLin Ling *config = NULL; 652e7cbe64fSgw continue; 653e7cbe64fSgw } 654e7cbe64fSgw 655f940fbb1SLin Ling if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG, 656e7cbe64fSgw &txg) != 0 || txg == 0) { 657f940fbb1SLin Ling nvlist_free(*config); 658f940fbb1SLin Ling *config = NULL; 659e7cbe64fSgw continue; 660e7cbe64fSgw } 661e7cbe64fSgw 662e7cbe64fSgw break; 663e7cbe64fSgw } 664e7cbe64fSgw 665e7cbe64fSgw kmem_free(label, sizeof (vdev_label_t)); 666bf82a41bSeschrock (void) ldi_close(vd_lh, FREAD, kcred); 66717f1e64aSEric Taylor if (*config == NULL) 668be6fd75aSMatthew Ahrens error = SET_ERROR(EIDRM); 669bf82a41bSeschrock 670f940fbb1SLin Ling return (error); 671e7cbe64fSgw } 672