1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * 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 /* 22f80ce222SChris Kirby * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23fa9e4066Sahrens */ 24fa9e4066Sahrens 25fa9e4066Sahrens /* 26fa9e4066Sahrens * ZFS volume emulation driver. 27fa9e4066Sahrens * 28fa9e4066Sahrens * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 29fa9e4066Sahrens * Volumes are accessed through the symbolic links named: 30fa9e4066Sahrens * 31fa9e4066Sahrens * /dev/zvol/dsk/<pool_name>/<dataset_name> 32fa9e4066Sahrens * /dev/zvol/rdsk/<pool_name>/<dataset_name> 33fa9e4066Sahrens * 34681d9761SEric Taylor * These links are created by the /dev filesystem (sdev_zvolops.c). 35fa9e4066Sahrens * Volumes are persistent through reboot. No user command needs to be 36fa9e4066Sahrens * run before opening and using a device. 37fa9e4066Sahrens */ 38fa9e4066Sahrens 39fa9e4066Sahrens #include <sys/types.h> 40fa9e4066Sahrens #include <sys/param.h> 41fa9e4066Sahrens #include <sys/errno.h> 42fa9e4066Sahrens #include <sys/uio.h> 43fa9e4066Sahrens #include <sys/buf.h> 44fa9e4066Sahrens #include <sys/modctl.h> 45fa9e4066Sahrens #include <sys/open.h> 46fa9e4066Sahrens #include <sys/kmem.h> 47fa9e4066Sahrens #include <sys/conf.h> 48fa9e4066Sahrens #include <sys/cmn_err.h> 49fa9e4066Sahrens #include <sys/stat.h> 50fa9e4066Sahrens #include <sys/zap.h> 51fa9e4066Sahrens #include <sys/spa.h> 52fa9e4066Sahrens #include <sys/zio.h> 53e7cbe64fSgw #include <sys/dmu_traverse.h> 54e7cbe64fSgw #include <sys/dnode.h> 55e7cbe64fSgw #include <sys/dsl_dataset.h> 56fa9e4066Sahrens #include <sys/dsl_prop.h> 57fa9e4066Sahrens #include <sys/dkio.h> 58fa9e4066Sahrens #include <sys/efi_partition.h> 59fa9e4066Sahrens #include <sys/byteorder.h> 60fa9e4066Sahrens #include <sys/pathname.h> 61fa9e4066Sahrens #include <sys/ddi.h> 62fa9e4066Sahrens #include <sys/sunddi.h> 63fa9e4066Sahrens #include <sys/crc32.h> 64fa9e4066Sahrens #include <sys/dirent.h> 65fa9e4066Sahrens #include <sys/policy.h> 66fa9e4066Sahrens #include <sys/fs/zfs.h> 67fa9e4066Sahrens #include <sys/zfs_ioctl.h> 68fa9e4066Sahrens #include <sys/mkdev.h> 6922ac5be4Sperrin #include <sys/zil.h> 70c5c6ffa0Smaybee #include <sys/refcount.h> 71c2e6a7d6Sperrin #include <sys/zfs_znode.h> 72c2e6a7d6Sperrin #include <sys/zfs_rlock.h> 73e7cbe64fSgw #include <sys/vdev_disk.h> 74e7cbe64fSgw #include <sys/vdev_impl.h> 75e7cbe64fSgw #include <sys/zvol.h> 76e7cbe64fSgw #include <sys/dumphdr.h> 771209a471SNeil Perrin #include <sys/zil_impl.h> 78fa9e4066Sahrens 79fa9e4066Sahrens #include "zfs_namecheck.h" 80fa9e4066Sahrens 81fa9e4066Sahrens static void *zvol_state; 82503ad85cSMatthew Ahrens static char *zvol_tag = "zvol_tag"; 83fa9e4066Sahrens 84e7cbe64fSgw #define ZVOL_DUMPSIZE "dumpsize" 85e7cbe64fSgw 86fa9e4066Sahrens /* 87fa9e4066Sahrens * This lock protects the zvol_state structure from being modified 88fa9e4066Sahrens * while it's being used, e.g. an open that comes in before a create 89fa9e4066Sahrens * finishes. It also protects temporary opens of the dataset so that, 90fa9e4066Sahrens * e.g., an open doesn't get a spurious EBUSY. 91fa9e4066Sahrens */ 92fa9e4066Sahrens static kmutex_t zvol_state_lock; 93fa9e4066Sahrens static uint32_t zvol_minors; 94fa9e4066Sahrens 95e7cbe64fSgw typedef struct zvol_extent { 9688b7b0f2SMatthew Ahrens list_node_t ze_node; 97e7cbe64fSgw dva_t ze_dva; /* dva associated with this extent */ 9888b7b0f2SMatthew Ahrens uint64_t ze_nblks; /* number of blocks in extent */ 99e7cbe64fSgw } zvol_extent_t; 100e7cbe64fSgw 101fa9e4066Sahrens /* 102fa9e4066Sahrens * The in-core state of each volume. 103fa9e4066Sahrens */ 104fa9e4066Sahrens typedef struct zvol_state { 105fa9e4066Sahrens char zv_name[MAXPATHLEN]; /* pool/dd name */ 106fa9e4066Sahrens uint64_t zv_volsize; /* amount of space we advertise */ 10767bd71c6Sperrin uint64_t zv_volblocksize; /* volume block size */ 108fa9e4066Sahrens minor_t zv_minor; /* minor number */ 109fa9e4066Sahrens uint8_t zv_min_bs; /* minimum addressable block shift */ 110701f66c4SEric Taylor uint8_t zv_flags; /* readonly, dumpified, etc. */ 111fa9e4066Sahrens objset_t *zv_objset; /* objset handle */ 112fa9e4066Sahrens uint32_t zv_open_count[OTYPCNT]; /* open counts */ 113fa9e4066Sahrens uint32_t zv_total_opens; /* total open count */ 11422ac5be4Sperrin zilog_t *zv_zilog; /* ZIL handle */ 11588b7b0f2SMatthew Ahrens list_t zv_extents; /* List of extents for dump */ 116c2e6a7d6Sperrin znode_t zv_znode; /* for range locking */ 117*94d1a210STim Haley dmu_buf_t *zv_dbuf; /* bonus handle */ 118fa9e4066Sahrens } zvol_state_t; 119fa9e4066Sahrens 120e7cbe64fSgw /* 121e7cbe64fSgw * zvol specific flags 122e7cbe64fSgw */ 123e7cbe64fSgw #define ZVOL_RDONLY 0x1 124e7cbe64fSgw #define ZVOL_DUMPIFIED 0x2 125c7f714e2SEric Taylor #define ZVOL_EXCL 0x4 126701f66c4SEric Taylor #define ZVOL_WCE 0x8 127e7cbe64fSgw 12867bd71c6Sperrin /* 12967bd71c6Sperrin * zvol maximum transfer in one DMU tx. 13067bd71c6Sperrin */ 13167bd71c6Sperrin int zvol_maxphys = DMU_MAX_ACCESS/2; 13267bd71c6Sperrin 13392241e0bSTom Erickson extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 13492241e0bSTom Erickson nvlist_t *, nvlist_t **); 135681d9761SEric Taylor static int zvol_remove_zv(zvol_state_t *); 136feb08c6bSbillm static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 137e7cbe64fSgw static int zvol_dumpify(zvol_state_t *zv); 138e7cbe64fSgw static int zvol_dump_fini(zvol_state_t *zv); 139e7cbe64fSgw static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 14067bd71c6Sperrin 141fa9e4066Sahrens static void 142681d9761SEric Taylor zvol_size_changed(uint64_t volsize, major_t maj, minor_t min) 143fa9e4066Sahrens { 144681d9761SEric Taylor dev_t dev = makedevice(maj, min); 145fa9e4066Sahrens 146fa9e4066Sahrens VERIFY(ddi_prop_update_int64(dev, zfs_dip, 147681d9761SEric Taylor "Size", volsize) == DDI_SUCCESS); 148fa9e4066Sahrens VERIFY(ddi_prop_update_int64(dev, zfs_dip, 149681d9761SEric Taylor "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 150e7cbe64fSgw 151e7cbe64fSgw /* Notify specfs to invalidate the cached size */ 152e7cbe64fSgw spec_size_invalidate(dev, VBLK); 153e7cbe64fSgw spec_size_invalidate(dev, VCHR); 154fa9e4066Sahrens } 155fa9e4066Sahrens 156fa9e4066Sahrens int 157e9dbad6fSeschrock zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 158fa9e4066Sahrens { 159e9dbad6fSeschrock if (volsize == 0) 160fa9e4066Sahrens return (EINVAL); 161fa9e4066Sahrens 162e9dbad6fSeschrock if (volsize % blocksize != 0) 1635c5460e9Seschrock return (EINVAL); 1645c5460e9Seschrock 165fa9e4066Sahrens #ifdef _ILP32 166e9dbad6fSeschrock if (volsize - 1 > SPEC_MAXOFFSET_T) 167fa9e4066Sahrens return (EOVERFLOW); 168fa9e4066Sahrens #endif 169fa9e4066Sahrens return (0); 170fa9e4066Sahrens } 171fa9e4066Sahrens 172fa9e4066Sahrens int 173e9dbad6fSeschrock zvol_check_volblocksize(uint64_t volblocksize) 174fa9e4066Sahrens { 175e9dbad6fSeschrock if (volblocksize < SPA_MINBLOCKSIZE || 176e9dbad6fSeschrock volblocksize > SPA_MAXBLOCKSIZE || 177e9dbad6fSeschrock !ISP2(volblocksize)) 178fa9e4066Sahrens return (EDOM); 179fa9e4066Sahrens 180fa9e4066Sahrens return (0); 181fa9e4066Sahrens } 182fa9e4066Sahrens 183fa9e4066Sahrens int 184a2eea2e1Sahrens zvol_get_stats(objset_t *os, nvlist_t *nv) 185fa9e4066Sahrens { 186fa9e4066Sahrens int error; 187fa9e4066Sahrens dmu_object_info_t doi; 188a2eea2e1Sahrens uint64_t val; 189fa9e4066Sahrens 190a2eea2e1Sahrens error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 191fa9e4066Sahrens if (error) 192fa9e4066Sahrens return (error); 193fa9e4066Sahrens 194a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 195a2eea2e1Sahrens 196fa9e4066Sahrens error = dmu_object_info(os, ZVOL_OBJ, &doi); 197fa9e4066Sahrens 198a2eea2e1Sahrens if (error == 0) { 199a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 200a2eea2e1Sahrens doi.doi_data_block_size); 201a2eea2e1Sahrens } 202fa9e4066Sahrens 203fa9e4066Sahrens return (error); 204fa9e4066Sahrens } 205fa9e4066Sahrens 206fa9e4066Sahrens /* 207fa9e4066Sahrens * Find a free minor number. 208fa9e4066Sahrens */ 209fa9e4066Sahrens static minor_t 210fa9e4066Sahrens zvol_minor_alloc(void) 211fa9e4066Sahrens { 212fa9e4066Sahrens minor_t minor; 213fa9e4066Sahrens 214fa9e4066Sahrens ASSERT(MUTEX_HELD(&zvol_state_lock)); 215fa9e4066Sahrens 216fa9e4066Sahrens for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) 217fa9e4066Sahrens if (ddi_get_soft_state(zvol_state, minor) == NULL) 218fa9e4066Sahrens return (minor); 219fa9e4066Sahrens 220fa9e4066Sahrens return (0); 221fa9e4066Sahrens } 222fa9e4066Sahrens 223fa9e4066Sahrens static zvol_state_t * 224e9dbad6fSeschrock zvol_minor_lookup(const char *name) 225fa9e4066Sahrens { 226fa9e4066Sahrens minor_t minor; 227fa9e4066Sahrens zvol_state_t *zv; 228fa9e4066Sahrens 229fa9e4066Sahrens ASSERT(MUTEX_HELD(&zvol_state_lock)); 230fa9e4066Sahrens 231fa9e4066Sahrens for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) { 232fa9e4066Sahrens zv = ddi_get_soft_state(zvol_state, minor); 233fa9e4066Sahrens if (zv == NULL) 234fa9e4066Sahrens continue; 235fa9e4066Sahrens if (strcmp(zv->zv_name, name) == 0) 236f80ce222SChris Kirby return (zv); 237fa9e4066Sahrens } 238fa9e4066Sahrens 239f80ce222SChris Kirby return (NULL); 240fa9e4066Sahrens } 241fa9e4066Sahrens 242e7cbe64fSgw /* extent mapping arg */ 243e7cbe64fSgw struct maparg { 24488b7b0f2SMatthew Ahrens zvol_state_t *ma_zv; 24588b7b0f2SMatthew Ahrens uint64_t ma_blks; 246e7cbe64fSgw }; 247e7cbe64fSgw 248e7cbe64fSgw /*ARGSUSED*/ 249e7cbe64fSgw static int 250b24ab676SJeff Bonwick zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 251b24ab676SJeff Bonwick const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 252e7cbe64fSgw { 25388b7b0f2SMatthew Ahrens struct maparg *ma = arg; 25488b7b0f2SMatthew Ahrens zvol_extent_t *ze; 25588b7b0f2SMatthew Ahrens int bs = ma->ma_zv->zv_volblocksize; 256e7cbe64fSgw 25788b7b0f2SMatthew Ahrens if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 25888b7b0f2SMatthew Ahrens return (0); 259e7cbe64fSgw 26088b7b0f2SMatthew Ahrens VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 26188b7b0f2SMatthew Ahrens ma->ma_blks++; 262e7cbe64fSgw 26388b7b0f2SMatthew Ahrens /* Abort immediately if we have encountered gang blocks */ 26488b7b0f2SMatthew Ahrens if (BP_IS_GANG(bp)) 26588b7b0f2SMatthew Ahrens return (EFRAGS); 266e7cbe64fSgw 26788b7b0f2SMatthew Ahrens /* 26888b7b0f2SMatthew Ahrens * See if the block is at the end of the previous extent. 26988b7b0f2SMatthew Ahrens */ 27088b7b0f2SMatthew Ahrens ze = list_tail(&ma->ma_zv->zv_extents); 27188b7b0f2SMatthew Ahrens if (ze && 27288b7b0f2SMatthew Ahrens DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 27388b7b0f2SMatthew Ahrens DVA_GET_OFFSET(BP_IDENTITY(bp)) == 27488b7b0f2SMatthew Ahrens DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 27588b7b0f2SMatthew Ahrens ze->ze_nblks++; 27688b7b0f2SMatthew Ahrens return (0); 277e7cbe64fSgw } 278e7cbe64fSgw 27988b7b0f2SMatthew Ahrens dprintf_bp(bp, "%s", "next blkptr:"); 280e7cbe64fSgw 28188b7b0f2SMatthew Ahrens /* start a new extent */ 28288b7b0f2SMatthew Ahrens ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 28388b7b0f2SMatthew Ahrens ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 28488b7b0f2SMatthew Ahrens ze->ze_nblks = 1; 28588b7b0f2SMatthew Ahrens list_insert_tail(&ma->ma_zv->zv_extents, ze); 28688b7b0f2SMatthew Ahrens return (0); 28788b7b0f2SMatthew Ahrens } 288e7cbe64fSgw 28988b7b0f2SMatthew Ahrens static void 29088b7b0f2SMatthew Ahrens zvol_free_extents(zvol_state_t *zv) 29188b7b0f2SMatthew Ahrens { 29288b7b0f2SMatthew Ahrens zvol_extent_t *ze; 293e7cbe64fSgw 29488b7b0f2SMatthew Ahrens while (ze = list_head(&zv->zv_extents)) { 29588b7b0f2SMatthew Ahrens list_remove(&zv->zv_extents, ze); 29688b7b0f2SMatthew Ahrens kmem_free(ze, sizeof (zvol_extent_t)); 297e7cbe64fSgw } 29888b7b0f2SMatthew Ahrens } 299e7cbe64fSgw 30088b7b0f2SMatthew Ahrens static int 30188b7b0f2SMatthew Ahrens zvol_get_lbas(zvol_state_t *zv) 30288b7b0f2SMatthew Ahrens { 3033adc9019SEric Taylor objset_t *os = zv->zv_objset; 30488b7b0f2SMatthew Ahrens struct maparg ma; 30588b7b0f2SMatthew Ahrens int err; 30688b7b0f2SMatthew Ahrens 30788b7b0f2SMatthew Ahrens ma.ma_zv = zv; 30888b7b0f2SMatthew Ahrens ma.ma_blks = 0; 30988b7b0f2SMatthew Ahrens zvol_free_extents(zv); 31088b7b0f2SMatthew Ahrens 3113adc9019SEric Taylor /* commit any in-flight changes before traversing the dataset */ 3123adc9019SEric Taylor txg_wait_synced(dmu_objset_pool(os), 0); 3133adc9019SEric Taylor err = traverse_dataset(dmu_objset_ds(os), 0, 31488b7b0f2SMatthew Ahrens TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 31588b7b0f2SMatthew Ahrens if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 31688b7b0f2SMatthew Ahrens zvol_free_extents(zv); 31788b7b0f2SMatthew Ahrens return (err ? err : EIO); 318e7cbe64fSgw } 31988b7b0f2SMatthew Ahrens 320e7cbe64fSgw return (0); 321e7cbe64fSgw } 322e7cbe64fSgw 323ecd6cf80Smarks /* ARGSUSED */ 324fa9e4066Sahrens void 325ecd6cf80Smarks zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 326fa9e4066Sahrens { 327da6c28aaSamw zfs_creat_t *zct = arg; 328da6c28aaSamw nvlist_t *nvprops = zct->zct_props; 329fa9e4066Sahrens int error; 330e9dbad6fSeschrock uint64_t volblocksize, volsize; 331fa9e4066Sahrens 332ecd6cf80Smarks VERIFY(nvlist_lookup_uint64(nvprops, 333e9dbad6fSeschrock zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 334ecd6cf80Smarks if (nvlist_lookup_uint64(nvprops, 335e9dbad6fSeschrock zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 336e9dbad6fSeschrock volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 337e9dbad6fSeschrock 338e9dbad6fSeschrock /* 339e7cbe64fSgw * These properties must be removed from the list so the generic 340e9dbad6fSeschrock * property setting step won't apply to them. 341e9dbad6fSeschrock */ 342ecd6cf80Smarks VERIFY(nvlist_remove_all(nvprops, 343e9dbad6fSeschrock zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 344ecd6cf80Smarks (void) nvlist_remove_all(nvprops, 345e9dbad6fSeschrock zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 346e9dbad6fSeschrock 347e9dbad6fSeschrock error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 348fa9e4066Sahrens DMU_OT_NONE, 0, tx); 349fa9e4066Sahrens ASSERT(error == 0); 350fa9e4066Sahrens 351fa9e4066Sahrens error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 352fa9e4066Sahrens DMU_OT_NONE, 0, tx); 353fa9e4066Sahrens ASSERT(error == 0); 354fa9e4066Sahrens 355e9dbad6fSeschrock error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 356fa9e4066Sahrens ASSERT(error == 0); 357fa9e4066Sahrens } 358fa9e4066Sahrens 35922ac5be4Sperrin /* 36022ac5be4Sperrin * Replay a TX_WRITE ZIL transaction that didn't get committed 36122ac5be4Sperrin * after a system failure 36222ac5be4Sperrin */ 36322ac5be4Sperrin static int 36422ac5be4Sperrin zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 36522ac5be4Sperrin { 36622ac5be4Sperrin objset_t *os = zv->zv_objset; 36722ac5be4Sperrin char *data = (char *)(lr + 1); /* data follows lr_write_t */ 368b24ab676SJeff Bonwick uint64_t offset, length; 36922ac5be4Sperrin dmu_tx_t *tx; 37022ac5be4Sperrin int error; 37122ac5be4Sperrin 37222ac5be4Sperrin if (byteswap) 37322ac5be4Sperrin byteswap_uint64_array(lr, sizeof (*lr)); 37422ac5be4Sperrin 375b24ab676SJeff Bonwick offset = lr->lr_offset; 376b24ab676SJeff Bonwick length = lr->lr_length; 377b24ab676SJeff Bonwick 378b24ab676SJeff Bonwick /* If it's a dmu_sync() block, write the whole block */ 379b24ab676SJeff Bonwick if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 380b24ab676SJeff Bonwick uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 381b24ab676SJeff Bonwick if (length < blocksize) { 382b24ab676SJeff Bonwick offset -= offset % blocksize; 383b24ab676SJeff Bonwick length = blocksize; 384b24ab676SJeff Bonwick } 385b24ab676SJeff Bonwick } 386975c32a0SNeil Perrin 38722ac5be4Sperrin tx = dmu_tx_create(os); 388b24ab676SJeff Bonwick dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 3891209a471SNeil Perrin error = dmu_tx_assign(tx, TXG_WAIT); 39022ac5be4Sperrin if (error) { 39122ac5be4Sperrin dmu_tx_abort(tx); 39222ac5be4Sperrin } else { 393b24ab676SJeff Bonwick dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 39422ac5be4Sperrin dmu_tx_commit(tx); 39522ac5be4Sperrin } 39622ac5be4Sperrin 39722ac5be4Sperrin return (error); 39822ac5be4Sperrin } 39922ac5be4Sperrin 40022ac5be4Sperrin /* ARGSUSED */ 40122ac5be4Sperrin static int 40222ac5be4Sperrin zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 40322ac5be4Sperrin { 40422ac5be4Sperrin return (ENOTSUP); 40522ac5be4Sperrin } 40622ac5be4Sperrin 40722ac5be4Sperrin /* 40822ac5be4Sperrin * Callback vectors for replaying records. 40922ac5be4Sperrin * Only TX_WRITE is needed for zvol. 41022ac5be4Sperrin */ 41122ac5be4Sperrin zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 41222ac5be4Sperrin zvol_replay_err, /* 0 no such transaction type */ 41322ac5be4Sperrin zvol_replay_err, /* TX_CREATE */ 41422ac5be4Sperrin zvol_replay_err, /* TX_MKDIR */ 41522ac5be4Sperrin zvol_replay_err, /* TX_MKXATTR */ 41622ac5be4Sperrin zvol_replay_err, /* TX_SYMLINK */ 41722ac5be4Sperrin zvol_replay_err, /* TX_REMOVE */ 41822ac5be4Sperrin zvol_replay_err, /* TX_RMDIR */ 41922ac5be4Sperrin zvol_replay_err, /* TX_LINK */ 42022ac5be4Sperrin zvol_replay_err, /* TX_RENAME */ 42122ac5be4Sperrin zvol_replay_write, /* TX_WRITE */ 42222ac5be4Sperrin zvol_replay_err, /* TX_TRUNCATE */ 42322ac5be4Sperrin zvol_replay_err, /* TX_SETATTR */ 42422ac5be4Sperrin zvol_replay_err, /* TX_ACL */ 425975c32a0SNeil Perrin zvol_replay_err, /* TX_CREATE_ACL */ 426975c32a0SNeil Perrin zvol_replay_err, /* TX_CREATE_ATTR */ 427975c32a0SNeil Perrin zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 428975c32a0SNeil Perrin zvol_replay_err, /* TX_MKDIR_ACL */ 429975c32a0SNeil Perrin zvol_replay_err, /* TX_MKDIR_ATTR */ 430975c32a0SNeil Perrin zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 431975c32a0SNeil Perrin zvol_replay_err, /* TX_WRITE2 */ 43222ac5be4Sperrin }; 43322ac5be4Sperrin 434681d9761SEric Taylor int 435681d9761SEric Taylor zvol_name2minor(const char *name, minor_t *minor) 436681d9761SEric Taylor { 437681d9761SEric Taylor zvol_state_t *zv; 438681d9761SEric Taylor 439681d9761SEric Taylor mutex_enter(&zvol_state_lock); 440681d9761SEric Taylor zv = zvol_minor_lookup(name); 441681d9761SEric Taylor if (minor && zv) 442681d9761SEric Taylor *minor = zv->zv_minor; 443681d9761SEric Taylor mutex_exit(&zvol_state_lock); 444681d9761SEric Taylor return (zv ? 0 : -1); 445681d9761SEric Taylor } 446681d9761SEric Taylor 447e7cbe64fSgw /* 448e7cbe64fSgw * Create a minor node (plus a whole lot more) for the specified volume. 449fa9e4066Sahrens */ 450fa9e4066Sahrens int 451681d9761SEric Taylor zvol_create_minor(const char *name) 452fa9e4066Sahrens { 453fa9e4066Sahrens zvol_state_t *zv; 454fa9e4066Sahrens objset_t *os; 45567bd71c6Sperrin dmu_object_info_t doi; 456fa9e4066Sahrens minor_t minor = 0; 457fa9e4066Sahrens char chrbuf[30], blkbuf[30]; 458fa9e4066Sahrens int error; 459fa9e4066Sahrens 460fa9e4066Sahrens mutex_enter(&zvol_state_lock); 461fa9e4066Sahrens 4621195e687SMark J Musante if (zvol_minor_lookup(name) != NULL) { 463fa9e4066Sahrens mutex_exit(&zvol_state_lock); 464fa9e4066Sahrens return (EEXIST); 465fa9e4066Sahrens } 466fa9e4066Sahrens 467503ad85cSMatthew Ahrens /* lie and say we're read-only */ 468503ad85cSMatthew Ahrens error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os); 469fa9e4066Sahrens 470fa9e4066Sahrens if (error) { 471fa9e4066Sahrens mutex_exit(&zvol_state_lock); 472fa9e4066Sahrens return (error); 473fa9e4066Sahrens } 474fa9e4066Sahrens 475681d9761SEric Taylor if ((minor = zvol_minor_alloc()) == 0) { 476503ad85cSMatthew Ahrens dmu_objset_disown(os, zvol_tag); 477fa9e4066Sahrens mutex_exit(&zvol_state_lock); 478fa9e4066Sahrens return (ENXIO); 479fa9e4066Sahrens } 480fa9e4066Sahrens 481fa9e4066Sahrens if (ddi_soft_state_zalloc(zvol_state, minor) != DDI_SUCCESS) { 482503ad85cSMatthew Ahrens dmu_objset_disown(os, zvol_tag); 483fa9e4066Sahrens mutex_exit(&zvol_state_lock); 484fa9e4066Sahrens return (EAGAIN); 485fa9e4066Sahrens } 486e9dbad6fSeschrock (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 487e9dbad6fSeschrock (char *)name); 488fa9e4066Sahrens 489681d9761SEric Taylor (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 490fa9e4066Sahrens 491fa9e4066Sahrens if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 492fa9e4066Sahrens minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 493fa9e4066Sahrens ddi_soft_state_free(zvol_state, minor); 494503ad85cSMatthew Ahrens dmu_objset_disown(os, zvol_tag); 495fa9e4066Sahrens mutex_exit(&zvol_state_lock); 496fa9e4066Sahrens return (EAGAIN); 497fa9e4066Sahrens } 498fa9e4066Sahrens 499681d9761SEric Taylor (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 500fa9e4066Sahrens 501fa9e4066Sahrens if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 502fa9e4066Sahrens minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 503fa9e4066Sahrens ddi_remove_minor_node(zfs_dip, chrbuf); 504fa9e4066Sahrens ddi_soft_state_free(zvol_state, minor); 505503ad85cSMatthew Ahrens dmu_objset_disown(os, zvol_tag); 506fa9e4066Sahrens mutex_exit(&zvol_state_lock); 507fa9e4066Sahrens return (EAGAIN); 508fa9e4066Sahrens } 509fa9e4066Sahrens 510fa9e4066Sahrens zv = ddi_get_soft_state(zvol_state, minor); 511fa9e4066Sahrens 512681d9761SEric Taylor (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 513fa9e4066Sahrens zv->zv_min_bs = DEV_BSHIFT; 514fa9e4066Sahrens zv->zv_minor = minor; 515fa9e4066Sahrens zv->zv_objset = os; 516681d9761SEric Taylor if (dmu_objset_is_snapshot(os)) 517681d9761SEric Taylor zv->zv_flags |= ZVOL_RDONLY; 518c2e6a7d6Sperrin mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 519c2e6a7d6Sperrin avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 520c2e6a7d6Sperrin sizeof (rl_t), offsetof(rl_t, r_node)); 52188b7b0f2SMatthew Ahrens list_create(&zv->zv_extents, sizeof (zvol_extent_t), 52288b7b0f2SMatthew Ahrens offsetof(zvol_extent_t, ze_node)); 52367bd71c6Sperrin /* get and cache the blocksize */ 52467bd71c6Sperrin error = dmu_object_info(os, ZVOL_OBJ, &doi); 52567bd71c6Sperrin ASSERT(error == 0); 52667bd71c6Sperrin zv->zv_volblocksize = doi.doi_data_block_size; 52722ac5be4Sperrin 5281209a471SNeil Perrin zil_replay(os, zv, zvol_replay_vector); 529681d9761SEric Taylor dmu_objset_disown(os, zvol_tag); 530681d9761SEric Taylor zv->zv_objset = NULL; 531fa9e4066Sahrens 532fa9e4066Sahrens zvol_minors++; 533fa9e4066Sahrens 534fa9e4066Sahrens mutex_exit(&zvol_state_lock); 535fa9e4066Sahrens 536fa9e4066Sahrens return (0); 537fa9e4066Sahrens } 538fa9e4066Sahrens 539fa9e4066Sahrens /* 540fa9e4066Sahrens * Remove minor node for the specified volume. 541fa9e4066Sahrens */ 542681d9761SEric Taylor static int 543681d9761SEric Taylor zvol_remove_zv(zvol_state_t *zv) 544681d9761SEric Taylor { 545681d9761SEric Taylor char nmbuf[20]; 546681d9761SEric Taylor 547681d9761SEric Taylor ASSERT(MUTEX_HELD(&zvol_state_lock)); 548681d9761SEric Taylor if (zv->zv_total_opens != 0) 549681d9761SEric Taylor return (EBUSY); 550681d9761SEric Taylor 551681d9761SEric Taylor (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", zv->zv_minor); 552681d9761SEric Taylor ddi_remove_minor_node(zfs_dip, nmbuf); 553681d9761SEric Taylor 554681d9761SEric Taylor (void) snprintf(nmbuf, sizeof (nmbuf), "%u", zv->zv_minor); 555681d9761SEric Taylor ddi_remove_minor_node(zfs_dip, nmbuf); 556681d9761SEric Taylor 557681d9761SEric Taylor avl_destroy(&zv->zv_znode.z_range_avl); 558681d9761SEric Taylor mutex_destroy(&zv->zv_znode.z_range_lock); 559681d9761SEric Taylor 560681d9761SEric Taylor ddi_soft_state_free(zvol_state, zv->zv_minor); 561681d9761SEric Taylor 562681d9761SEric Taylor zvol_minors--; 563681d9761SEric Taylor return (0); 564681d9761SEric Taylor } 565681d9761SEric Taylor 566fa9e4066Sahrens int 567e9dbad6fSeschrock zvol_remove_minor(const char *name) 568fa9e4066Sahrens { 569fa9e4066Sahrens zvol_state_t *zv; 570681d9761SEric Taylor int rc; 571fa9e4066Sahrens 572fa9e4066Sahrens mutex_enter(&zvol_state_lock); 573e9dbad6fSeschrock if ((zv = zvol_minor_lookup(name)) == NULL) { 574fa9e4066Sahrens mutex_exit(&zvol_state_lock); 575fa9e4066Sahrens return (ENXIO); 576fa9e4066Sahrens } 577681d9761SEric Taylor rc = zvol_remove_zv(zv); 578681d9761SEric Taylor mutex_exit(&zvol_state_lock); 579681d9761SEric Taylor return (rc); 580681d9761SEric Taylor } 581fa9e4066Sahrens 582681d9761SEric Taylor int 583681d9761SEric Taylor zvol_first_open(zvol_state_t *zv) 584681d9761SEric Taylor { 585681d9761SEric Taylor objset_t *os; 586681d9761SEric Taylor uint64_t volsize; 587681d9761SEric Taylor int error; 588681d9761SEric Taylor uint64_t readonly; 589fa9e4066Sahrens 590681d9761SEric Taylor /* lie and say we're read-only */ 591681d9761SEric Taylor error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 592681d9761SEric Taylor zvol_tag, &os); 593681d9761SEric Taylor if (error) 594681d9761SEric Taylor return (error); 595fa9e4066Sahrens 596681d9761SEric Taylor error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 597681d9761SEric Taylor if (error) { 598681d9761SEric Taylor ASSERT(error == 0); 599681d9761SEric Taylor dmu_objset_disown(os, zvol_tag); 600681d9761SEric Taylor return (error); 601681d9761SEric Taylor } 602681d9761SEric Taylor zv->zv_objset = os; 603*94d1a210STim Haley error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 604*94d1a210STim Haley if (error) { 605*94d1a210STim Haley dmu_objset_disown(os, zvol_tag); 606*94d1a210STim Haley return (error); 607*94d1a210STim Haley } 608681d9761SEric Taylor zv->zv_volsize = volsize; 609681d9761SEric Taylor zv->zv_zilog = zil_open(os, zvol_get_data); 610681d9761SEric Taylor zvol_size_changed(zv->zv_volsize, ddi_driver_major(zfs_dip), 611681d9761SEric Taylor zv->zv_minor); 612fa9e4066Sahrens 613681d9761SEric Taylor VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 614681d9761SEric Taylor NULL) == 0); 615681d9761SEric Taylor if (readonly || dmu_objset_is_snapshot(os)) 616681d9761SEric Taylor zv->zv_flags |= ZVOL_RDONLY; 617681d9761SEric Taylor else 618681d9761SEric Taylor zv->zv_flags &= ~ZVOL_RDONLY; 619681d9761SEric Taylor return (error); 620681d9761SEric Taylor } 621fa9e4066Sahrens 622681d9761SEric Taylor void 623681d9761SEric Taylor zvol_last_close(zvol_state_t *zv) 624681d9761SEric Taylor { 62522ac5be4Sperrin zil_close(zv->zv_zilog); 62622ac5be4Sperrin zv->zv_zilog = NULL; 627*94d1a210STim Haley dmu_buf_rele(zv->zv_dbuf, zvol_tag); 628*94d1a210STim Haley zv->zv_dbuf = NULL; 629503ad85cSMatthew Ahrens dmu_objset_disown(zv->zv_objset, zvol_tag); 630fa9e4066Sahrens zv->zv_objset = NULL; 631fa9e4066Sahrens } 632fa9e4066Sahrens 633e7cbe64fSgw int 634e7cbe64fSgw zvol_prealloc(zvol_state_t *zv) 635e7cbe64fSgw { 636e7cbe64fSgw objset_t *os = zv->zv_objset; 637e7cbe64fSgw dmu_tx_t *tx; 638e7cbe64fSgw uint64_t refd, avail, usedobjs, availobjs; 639e7cbe64fSgw uint64_t resid = zv->zv_volsize; 640e7cbe64fSgw uint64_t off = 0; 641e7cbe64fSgw 642e7cbe64fSgw /* Check the space usage before attempting to allocate the space */ 643e7cbe64fSgw dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 644e7cbe64fSgw if (avail < zv->zv_volsize) 645e7cbe64fSgw return (ENOSPC); 646e7cbe64fSgw 647e7cbe64fSgw /* Free old extents if they exist */ 648e7cbe64fSgw zvol_free_extents(zv); 649e7cbe64fSgw 650e7cbe64fSgw while (resid != 0) { 651e7cbe64fSgw int error; 652e7cbe64fSgw uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 653e7cbe64fSgw 654e7cbe64fSgw tx = dmu_tx_create(os); 655e7cbe64fSgw dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 656e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 657e7cbe64fSgw if (error) { 658e7cbe64fSgw dmu_tx_abort(tx); 659cdb0ab79Smaybee (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 660e7cbe64fSgw return (error); 661e7cbe64fSgw } 66282c9918fSTim Haley dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 663e7cbe64fSgw dmu_tx_commit(tx); 664e7cbe64fSgw off += bytes; 665e7cbe64fSgw resid -= bytes; 666e7cbe64fSgw } 667e7cbe64fSgw txg_wait_synced(dmu_objset_pool(os), 0); 668e7cbe64fSgw 669e7cbe64fSgw return (0); 670e7cbe64fSgw } 671e7cbe64fSgw 672e7cbe64fSgw int 673681d9761SEric Taylor zvol_update_volsize(objset_t *os, uint64_t volsize) 674e7cbe64fSgw { 675e7cbe64fSgw dmu_tx_t *tx; 676e7cbe64fSgw int error; 677e7cbe64fSgw 678e7cbe64fSgw ASSERT(MUTEX_HELD(&zvol_state_lock)); 679e7cbe64fSgw 680681d9761SEric Taylor tx = dmu_tx_create(os); 681e7cbe64fSgw dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 682e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 683e7cbe64fSgw if (error) { 684e7cbe64fSgw dmu_tx_abort(tx); 685e7cbe64fSgw return (error); 686e7cbe64fSgw } 687e7cbe64fSgw 688681d9761SEric Taylor error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 689e7cbe64fSgw &volsize, tx); 690e7cbe64fSgw dmu_tx_commit(tx); 691e7cbe64fSgw 692e7cbe64fSgw if (error == 0) 693681d9761SEric Taylor error = dmu_free_long_range(os, 694cdb0ab79Smaybee ZVOL_OBJ, volsize, DMU_OBJECT_END); 695681d9761SEric Taylor return (error); 696681d9761SEric Taylor } 697e7cbe64fSgw 698681d9761SEric Taylor void 699681d9761SEric Taylor zvol_remove_minors(const char *name) 700681d9761SEric Taylor { 701681d9761SEric Taylor zvol_state_t *zv; 702681d9761SEric Taylor char *namebuf; 703681d9761SEric Taylor minor_t minor; 704681d9761SEric Taylor 705681d9761SEric Taylor namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP); 706681d9761SEric Taylor (void) strncpy(namebuf, name, strlen(name)); 707681d9761SEric Taylor (void) strcat(namebuf, "/"); 708681d9761SEric Taylor mutex_enter(&zvol_state_lock); 709681d9761SEric Taylor for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) { 710681d9761SEric Taylor 711681d9761SEric Taylor zv = ddi_get_soft_state(zvol_state, minor); 712681d9761SEric Taylor if (zv == NULL) 713681d9761SEric Taylor continue; 714681d9761SEric Taylor if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0) 715681d9761SEric Taylor (void) zvol_remove_zv(zv); 716e7cbe64fSgw } 717681d9761SEric Taylor kmem_free(namebuf, strlen(name) + 2); 718681d9761SEric Taylor 719681d9761SEric Taylor mutex_exit(&zvol_state_lock); 720e7cbe64fSgw } 721e7cbe64fSgw 722fa9e4066Sahrens int 72391ebeef5Sahrens zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 724fa9e4066Sahrens { 725681d9761SEric Taylor zvol_state_t *zv = NULL; 726681d9761SEric Taylor objset_t *os; 727fa9e4066Sahrens int error; 7285c5460e9Seschrock dmu_object_info_t doi; 729e7cbe64fSgw uint64_t old_volsize = 0ULL; 730681d9761SEric Taylor uint64_t readonly; 731fa9e4066Sahrens 732fa9e4066Sahrens mutex_enter(&zvol_state_lock); 733681d9761SEric Taylor zv = zvol_minor_lookup(name); 734681d9761SEric Taylor if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 735681d9761SEric Taylor mutex_exit(&zvol_state_lock); 736681d9761SEric Taylor return (error); 737fa9e4066Sahrens } 738fa9e4066Sahrens 739681d9761SEric Taylor if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 740e9dbad6fSeschrock (error = zvol_check_volsize(volsize, 741bb0ade09Sahrens doi.doi_data_block_size)) != 0) 742bb0ade09Sahrens goto out; 7435c5460e9Seschrock 744681d9761SEric Taylor VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 745681d9761SEric Taylor NULL) == 0); 746681d9761SEric Taylor if (readonly) { 747bb0ade09Sahrens error = EROFS; 748bb0ade09Sahrens goto out; 749fa9e4066Sahrens } 750fa9e4066Sahrens 751681d9761SEric Taylor error = zvol_update_volsize(os, volsize); 752e7cbe64fSgw /* 753e7cbe64fSgw * Reinitialize the dump area to the new size. If we 754681d9761SEric Taylor * failed to resize the dump area then restore it back to 755681d9761SEric Taylor * its original size. 756e7cbe64fSgw */ 757681d9761SEric Taylor if (zv && error == 0) { 758681d9761SEric Taylor if (zv->zv_flags & ZVOL_DUMPIFIED) { 759681d9761SEric Taylor old_volsize = zv->zv_volsize; 760681d9761SEric Taylor zv->zv_volsize = volsize; 761681d9761SEric Taylor if ((error = zvol_dumpify(zv)) != 0 || 762681d9761SEric Taylor (error = dumpvp_resize()) != 0) { 763681d9761SEric Taylor (void) zvol_update_volsize(os, old_volsize); 764681d9761SEric Taylor zv->zv_volsize = old_volsize; 765681d9761SEric Taylor error = zvol_dumpify(zv); 766681d9761SEric Taylor } 767681d9761SEric Taylor } 768681d9761SEric Taylor if (error == 0) { 769681d9761SEric Taylor zv->zv_volsize = volsize; 770681d9761SEric Taylor zvol_size_changed(volsize, maj, zv->zv_minor); 771e7cbe64fSgw } 772fa9e4066Sahrens } 773fa9e4066Sahrens 774573ca77eSGeorge Wilson /* 775573ca77eSGeorge Wilson * Generate a LUN expansion event. 776573ca77eSGeorge Wilson */ 777681d9761SEric Taylor if (zv && error == 0) { 778573ca77eSGeorge Wilson sysevent_id_t eid; 779573ca77eSGeorge Wilson nvlist_t *attr; 780573ca77eSGeorge Wilson char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 781573ca77eSGeorge Wilson 782681d9761SEric Taylor (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 783573ca77eSGeorge Wilson zv->zv_minor); 784573ca77eSGeorge Wilson 785573ca77eSGeorge Wilson VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 786573ca77eSGeorge Wilson VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 787573ca77eSGeorge Wilson 788573ca77eSGeorge Wilson (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 789573ca77eSGeorge Wilson ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 790573ca77eSGeorge Wilson 791573ca77eSGeorge Wilson nvlist_free(attr); 792573ca77eSGeorge Wilson kmem_free(physpath, MAXPATHLEN); 793573ca77eSGeorge Wilson } 794573ca77eSGeorge Wilson 795bb0ade09Sahrens out: 796681d9761SEric Taylor dmu_objset_rele(os, FTAG); 797bb0ade09Sahrens 798fa9e4066Sahrens mutex_exit(&zvol_state_lock); 799fa9e4066Sahrens 800fa9e4066Sahrens return (error); 801fa9e4066Sahrens } 802fa9e4066Sahrens 803fa9e4066Sahrens /*ARGSUSED*/ 804fa9e4066Sahrens int 805fa9e4066Sahrens zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 806fa9e4066Sahrens { 807fa9e4066Sahrens minor_t minor = getminor(*devp); 808fa9e4066Sahrens zvol_state_t *zv; 809681d9761SEric Taylor int err = 0; 810fa9e4066Sahrens 811fa9e4066Sahrens if (minor == 0) /* This is the control device */ 812fa9e4066Sahrens return (0); 813fa9e4066Sahrens 814fa9e4066Sahrens mutex_enter(&zvol_state_lock); 815fa9e4066Sahrens 816fa9e4066Sahrens zv = ddi_get_soft_state(zvol_state, minor); 817fa9e4066Sahrens if (zv == NULL) { 818fa9e4066Sahrens mutex_exit(&zvol_state_lock); 819fa9e4066Sahrens return (ENXIO); 820fa9e4066Sahrens } 821fa9e4066Sahrens 822681d9761SEric Taylor if (zv->zv_total_opens == 0) 823681d9761SEric Taylor err = zvol_first_open(zv); 824681d9761SEric Taylor if (err) { 825fa9e4066Sahrens mutex_exit(&zvol_state_lock); 826681d9761SEric Taylor return (err); 827681d9761SEric Taylor } 828681d9761SEric Taylor if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 829681d9761SEric Taylor err = EROFS; 830681d9761SEric Taylor goto out; 831fa9e4066Sahrens } 832c7f714e2SEric Taylor if (zv->zv_flags & ZVOL_EXCL) { 833681d9761SEric Taylor err = EBUSY; 834681d9761SEric Taylor goto out; 835c7f714e2SEric Taylor } 836c7f714e2SEric Taylor if (flag & FEXCL) { 837c7f714e2SEric Taylor if (zv->zv_total_opens != 0) { 838681d9761SEric Taylor err = EBUSY; 839681d9761SEric Taylor goto out; 840c7f714e2SEric Taylor } 841c7f714e2SEric Taylor zv->zv_flags |= ZVOL_EXCL; 842c7f714e2SEric Taylor } 843fa9e4066Sahrens 844fa9e4066Sahrens if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 845fa9e4066Sahrens zv->zv_open_count[otyp]++; 846fa9e4066Sahrens zv->zv_total_opens++; 847fa9e4066Sahrens } 848fa9e4066Sahrens mutex_exit(&zvol_state_lock); 849fa9e4066Sahrens 850681d9761SEric Taylor return (err); 851681d9761SEric Taylor out: 852681d9761SEric Taylor if (zv->zv_total_opens == 0) 853681d9761SEric Taylor zvol_last_close(zv); 854681d9761SEric Taylor mutex_exit(&zvol_state_lock); 855681d9761SEric Taylor return (err); 856fa9e4066Sahrens } 857fa9e4066Sahrens 858fa9e4066Sahrens /*ARGSUSED*/ 859fa9e4066Sahrens int 860fa9e4066Sahrens zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 861fa9e4066Sahrens { 862fa9e4066Sahrens minor_t minor = getminor(dev); 863fa9e4066Sahrens zvol_state_t *zv; 864681d9761SEric Taylor int error = 0; 865fa9e4066Sahrens 866fa9e4066Sahrens if (minor == 0) /* This is the control device */ 867fa9e4066Sahrens return (0); 868fa9e4066Sahrens 869fa9e4066Sahrens mutex_enter(&zvol_state_lock); 870fa9e4066Sahrens 871fa9e4066Sahrens zv = ddi_get_soft_state(zvol_state, minor); 872fa9e4066Sahrens if (zv == NULL) { 873fa9e4066Sahrens mutex_exit(&zvol_state_lock); 874fa9e4066Sahrens return (ENXIO); 875fa9e4066Sahrens } 876fa9e4066Sahrens 877c7f714e2SEric Taylor if (zv->zv_flags & ZVOL_EXCL) { 878c7f714e2SEric Taylor ASSERT(zv->zv_total_opens == 1); 879c7f714e2SEric Taylor zv->zv_flags &= ~ZVOL_EXCL; 880fa9e4066Sahrens } 881fa9e4066Sahrens 882fa9e4066Sahrens /* 883fa9e4066Sahrens * If the open count is zero, this is a spurious close. 884fa9e4066Sahrens * That indicates a bug in the kernel / DDI framework. 885fa9e4066Sahrens */ 886fa9e4066Sahrens ASSERT(zv->zv_open_count[otyp] != 0); 887fa9e4066Sahrens ASSERT(zv->zv_total_opens != 0); 888fa9e4066Sahrens 889fa9e4066Sahrens /* 890fa9e4066Sahrens * You may get multiple opens, but only one close. 891fa9e4066Sahrens */ 892fa9e4066Sahrens zv->zv_open_count[otyp]--; 893fa9e4066Sahrens zv->zv_total_opens--; 894fa9e4066Sahrens 895681d9761SEric Taylor if (zv->zv_total_opens == 0) 896681d9761SEric Taylor zvol_last_close(zv); 897fa9e4066Sahrens 898681d9761SEric Taylor mutex_exit(&zvol_state_lock); 899681d9761SEric Taylor return (error); 900fa9e4066Sahrens } 901fa9e4066Sahrens 902feb08c6bSbillm static void 903b24ab676SJeff Bonwick zvol_get_done(zgd_t *zgd, int error) 90467bd71c6Sperrin { 905b24ab676SJeff Bonwick if (zgd->zgd_db) 906b24ab676SJeff Bonwick dmu_buf_rele(zgd->zgd_db, zgd); 907b24ab676SJeff Bonwick 908b24ab676SJeff Bonwick zfs_range_unlock(zgd->zgd_rl); 909b24ab676SJeff Bonwick 910b24ab676SJeff Bonwick if (error == 0 && zgd->zgd_bp) 911b24ab676SJeff Bonwick zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 91267bd71c6Sperrin 91367bd71c6Sperrin kmem_free(zgd, sizeof (zgd_t)); 91467bd71c6Sperrin } 91567bd71c6Sperrin 91667bd71c6Sperrin /* 91767bd71c6Sperrin * Get data to generate a TX_WRITE intent log record. 91867bd71c6Sperrin */ 919feb08c6bSbillm static int 92067bd71c6Sperrin zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 92167bd71c6Sperrin { 92267bd71c6Sperrin zvol_state_t *zv = arg; 92367bd71c6Sperrin objset_t *os = zv->zv_objset; 924b24ab676SJeff Bonwick uint64_t object = ZVOL_OBJ; 925b24ab676SJeff Bonwick uint64_t offset = lr->lr_offset; 926b24ab676SJeff Bonwick uint64_t size = lr->lr_length; /* length of user data */ 927b24ab676SJeff Bonwick blkptr_t *bp = &lr->lr_blkptr; 92867bd71c6Sperrin dmu_buf_t *db; 92967bd71c6Sperrin zgd_t *zgd; 93067bd71c6Sperrin int error; 93167bd71c6Sperrin 932b24ab676SJeff Bonwick ASSERT(zio != NULL); 933b24ab676SJeff Bonwick ASSERT(size != 0); 934b24ab676SJeff Bonwick 935b24ab676SJeff Bonwick zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 936b24ab676SJeff Bonwick zgd->zgd_zilog = zv->zv_zilog; 937b24ab676SJeff Bonwick zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 938feb08c6bSbillm 939c2e6a7d6Sperrin /* 940c2e6a7d6Sperrin * Write records come in two flavors: immediate and indirect. 941c2e6a7d6Sperrin * For small writes it's cheaper to store the data with the 942c2e6a7d6Sperrin * log record (immediate); for large writes it's cheaper to 943c2e6a7d6Sperrin * sync the data and get a pointer to it (indirect) so that 944c2e6a7d6Sperrin * we don't have to write the data twice. 945c2e6a7d6Sperrin */ 946b24ab676SJeff Bonwick if (buf != NULL) { /* immediate write */ 947b24ab676SJeff Bonwick error = dmu_read(os, object, offset, size, buf, 948b24ab676SJeff Bonwick DMU_READ_NO_PREFETCH); 949b24ab676SJeff Bonwick } else { 950b24ab676SJeff Bonwick size = zv->zv_volblocksize; 951b24ab676SJeff Bonwick offset = P2ALIGN(offset, size); 952b24ab676SJeff Bonwick error = dmu_buf_hold(os, object, offset, zgd, &db); 953b24ab676SJeff Bonwick if (error == 0) { 954b24ab676SJeff Bonwick zgd->zgd_db = db; 955b24ab676SJeff Bonwick zgd->zgd_bp = bp; 95667bd71c6Sperrin 957b24ab676SJeff Bonwick ASSERT(db->db_offset == offset); 958b24ab676SJeff Bonwick ASSERT(db->db_size == size); 95967bd71c6Sperrin 960b24ab676SJeff Bonwick error = dmu_sync(zio, lr->lr_common.lrc_txg, 961b24ab676SJeff Bonwick zvol_get_done, zgd); 962975c32a0SNeil Perrin 963b24ab676SJeff Bonwick if (error == 0) 964b24ab676SJeff Bonwick return (0); 965b24ab676SJeff Bonwick } 966975c32a0SNeil Perrin } 967975c32a0SNeil Perrin 968b24ab676SJeff Bonwick zvol_get_done(zgd, error); 969b24ab676SJeff Bonwick 97067bd71c6Sperrin return (error); 97167bd71c6Sperrin } 97267bd71c6Sperrin 973a24e15ceSperrin /* 974a24e15ceSperrin * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 97522ac5be4Sperrin * 97622ac5be4Sperrin * We store data in the log buffers if it's small enough. 97767bd71c6Sperrin * Otherwise we will later flush the data out via dmu_sync(). 97822ac5be4Sperrin */ 97967bd71c6Sperrin ssize_t zvol_immediate_write_sz = 32768; 98022ac5be4Sperrin 981feb08c6bSbillm static void 982510b6c0eSNeil Perrin zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 983510b6c0eSNeil Perrin boolean_t sync) 98422ac5be4Sperrin { 985feb08c6bSbillm uint32_t blocksize = zv->zv_volblocksize; 9861209a471SNeil Perrin zilog_t *zilog = zv->zv_zilog; 987510b6c0eSNeil Perrin boolean_t slogging; 988e09fa4daSNeil Perrin ssize_t immediate_write_sz; 989510b6c0eSNeil Perrin 990510b6c0eSNeil Perrin if (zil_disable) 991510b6c0eSNeil Perrin return; 99222ac5be4Sperrin 993b24ab676SJeff Bonwick if (zil_replaying(zilog, tx)) 9941209a471SNeil Perrin return; 9951209a471SNeil Perrin 996e09fa4daSNeil Perrin immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 997e09fa4daSNeil Perrin ? 0 : zvol_immediate_write_sz; 998e09fa4daSNeil Perrin 999e09fa4daSNeil Perrin slogging = spa_has_slogs(zilog->zl_spa) && 1000e09fa4daSNeil Perrin (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1001feb08c6bSbillm 1002510b6c0eSNeil Perrin while (resid) { 1003510b6c0eSNeil Perrin itx_t *itx; 1004510b6c0eSNeil Perrin lr_write_t *lr; 1005510b6c0eSNeil Perrin ssize_t len; 1006510b6c0eSNeil Perrin itx_wr_state_t write_state; 1007510b6c0eSNeil Perrin 1008510b6c0eSNeil Perrin /* 1009510b6c0eSNeil Perrin * Unlike zfs_log_write() we can be called with 1010510b6c0eSNeil Perrin * upto DMU_MAX_ACCESS/2 (5MB) writes. 1011510b6c0eSNeil Perrin */ 1012e09fa4daSNeil Perrin if (blocksize > immediate_write_sz && !slogging && 1013510b6c0eSNeil Perrin resid >= blocksize && off % blocksize == 0) { 1014510b6c0eSNeil Perrin write_state = WR_INDIRECT; /* uses dmu_sync */ 1015510b6c0eSNeil Perrin len = blocksize; 1016510b6c0eSNeil Perrin } else if (sync) { 1017510b6c0eSNeil Perrin write_state = WR_COPIED; 1018510b6c0eSNeil Perrin len = MIN(ZIL_MAX_LOG_DATA, resid); 1019510b6c0eSNeil Perrin } else { 1020510b6c0eSNeil Perrin write_state = WR_NEED_COPY; 1021510b6c0eSNeil Perrin len = MIN(ZIL_MAX_LOG_DATA, resid); 1022510b6c0eSNeil Perrin } 1023510b6c0eSNeil Perrin 1024510b6c0eSNeil Perrin itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1025510b6c0eSNeil Perrin (write_state == WR_COPIED ? len : 0)); 1026feb08c6bSbillm lr = (lr_write_t *)&itx->itx_lr; 1027510b6c0eSNeil Perrin if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 10287bfdf011SNeil Perrin ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1029b24ab676SJeff Bonwick zil_itx_destroy(itx); 1030510b6c0eSNeil Perrin itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1031510b6c0eSNeil Perrin lr = (lr_write_t *)&itx->itx_lr; 1032510b6c0eSNeil Perrin write_state = WR_NEED_COPY; 1033510b6c0eSNeil Perrin } 1034510b6c0eSNeil Perrin 1035510b6c0eSNeil Perrin itx->itx_wr_state = write_state; 1036510b6c0eSNeil Perrin if (write_state == WR_NEED_COPY) 1037510b6c0eSNeil Perrin itx->itx_sod += len; 1038feb08c6bSbillm lr->lr_foid = ZVOL_OBJ; 1039feb08c6bSbillm lr->lr_offset = off; 1040510b6c0eSNeil Perrin lr->lr_length = len; 1041b24ab676SJeff Bonwick lr->lr_blkoff = 0; 1042feb08c6bSbillm BP_ZERO(&lr->lr_blkptr); 1043feb08c6bSbillm 1044510b6c0eSNeil Perrin itx->itx_private = zv; 1045510b6c0eSNeil Perrin itx->itx_sync = sync; 1046510b6c0eSNeil Perrin 10471209a471SNeil Perrin (void) zil_itx_assign(zilog, itx, tx); 1048510b6c0eSNeil Perrin 1049510b6c0eSNeil Perrin off += len; 1050510b6c0eSNeil Perrin resid -= len; 105122ac5be4Sperrin } 105222ac5be4Sperrin } 105322ac5be4Sperrin 105488b7b0f2SMatthew Ahrens static int 105588b7b0f2SMatthew Ahrens zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size, 105688b7b0f2SMatthew Ahrens boolean_t doread, boolean_t isdump) 1057e7cbe64fSgw { 1058e7cbe64fSgw vdev_disk_t *dvd; 1059e7cbe64fSgw int c; 1060e7cbe64fSgw int numerrors = 0; 1061e7cbe64fSgw 1062e7cbe64fSgw for (c = 0; c < vd->vdev_children; c++) { 106321ecdf64SLin Ling ASSERT(vd->vdev_ops == &vdev_mirror_ops || 106421ecdf64SLin Ling vd->vdev_ops == &vdev_replacing_ops || 106521ecdf64SLin Ling vd->vdev_ops == &vdev_spare_ops); 106688b7b0f2SMatthew Ahrens int err = zvol_dumpio_vdev(vd->vdev_child[c], 106788b7b0f2SMatthew Ahrens addr, offset, size, doread, isdump); 106888b7b0f2SMatthew Ahrens if (err != 0) { 1069e7cbe64fSgw numerrors++; 107088b7b0f2SMatthew Ahrens } else if (doread) { 1071e7cbe64fSgw break; 1072e7cbe64fSgw } 1073e7cbe64fSgw } 1074e7cbe64fSgw 1075e7cbe64fSgw if (!vd->vdev_ops->vdev_op_leaf) 1076e7cbe64fSgw return (numerrors < vd->vdev_children ? 0 : EIO); 1077e7cbe64fSgw 1078dc0bb255SEric Taylor if (doread && !vdev_readable(vd)) 1079dc0bb255SEric Taylor return (EIO); 1080dc0bb255SEric Taylor else if (!doread && !vdev_writeable(vd)) 1081e7cbe64fSgw return (EIO); 1082e7cbe64fSgw 1083e7cbe64fSgw dvd = vd->vdev_tsd; 1084e7cbe64fSgw ASSERT3P(dvd, !=, NULL); 1085e7cbe64fSgw offset += VDEV_LABEL_START_SIZE; 1086e7cbe64fSgw 1087e7cbe64fSgw if (ddi_in_panic() || isdump) { 108888b7b0f2SMatthew Ahrens ASSERT(!doread); 108988b7b0f2SMatthew Ahrens if (doread) 1090e7cbe64fSgw return (EIO); 1091e7cbe64fSgw return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1092e7cbe64fSgw lbtodb(size))); 1093e7cbe64fSgw } else { 1094e7cbe64fSgw return (vdev_disk_physio(dvd->vd_lh, addr, size, offset, 109588b7b0f2SMatthew Ahrens doread ? B_READ : B_WRITE)); 1096e7cbe64fSgw } 1097e7cbe64fSgw } 1098e7cbe64fSgw 109988b7b0f2SMatthew Ahrens static int 110088b7b0f2SMatthew Ahrens zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 110188b7b0f2SMatthew Ahrens boolean_t doread, boolean_t isdump) 1102e7cbe64fSgw { 1103e7cbe64fSgw vdev_t *vd; 1104e7cbe64fSgw int error; 110588b7b0f2SMatthew Ahrens zvol_extent_t *ze; 1106e7cbe64fSgw spa_t *spa = dmu_objset_spa(zv->zv_objset); 1107e7cbe64fSgw 110888b7b0f2SMatthew Ahrens /* Must be sector aligned, and not stradle a block boundary. */ 110988b7b0f2SMatthew Ahrens if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 111088b7b0f2SMatthew Ahrens P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1111e7cbe64fSgw return (EINVAL); 111288b7b0f2SMatthew Ahrens } 111388b7b0f2SMatthew Ahrens ASSERT(size <= zv->zv_volblocksize); 1114e7cbe64fSgw 111588b7b0f2SMatthew Ahrens /* Locate the extent this belongs to */ 111688b7b0f2SMatthew Ahrens ze = list_head(&zv->zv_extents); 111788b7b0f2SMatthew Ahrens while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 111888b7b0f2SMatthew Ahrens offset -= ze->ze_nblks * zv->zv_volblocksize; 111988b7b0f2SMatthew Ahrens ze = list_next(&zv->zv_extents, ze); 112088b7b0f2SMatthew Ahrens } 112124cc0e1cSGeorge Wilson 112224cc0e1cSGeorge Wilson if (!ddi_in_panic()) 112324cc0e1cSGeorge Wilson spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 112424cc0e1cSGeorge Wilson 112588b7b0f2SMatthew Ahrens vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 112688b7b0f2SMatthew Ahrens offset += DVA_GET_OFFSET(&ze->ze_dva); 112788b7b0f2SMatthew Ahrens error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump); 112824cc0e1cSGeorge Wilson 112924cc0e1cSGeorge Wilson if (!ddi_in_panic()) 113024cc0e1cSGeorge Wilson spa_config_exit(spa, SCL_STATE, FTAG); 113124cc0e1cSGeorge Wilson 1132e7cbe64fSgw return (error); 1133e7cbe64fSgw } 1134e7cbe64fSgw 1135fa9e4066Sahrens int 1136fa9e4066Sahrens zvol_strategy(buf_t *bp) 1137fa9e4066Sahrens { 1138fa9e4066Sahrens zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev)); 1139fa9e4066Sahrens uint64_t off, volsize; 114088b7b0f2SMatthew Ahrens size_t resid; 1141fa9e4066Sahrens char *addr; 114222ac5be4Sperrin objset_t *os; 1143c2e6a7d6Sperrin rl_t *rl; 1144fa9e4066Sahrens int error = 0; 114588b7b0f2SMatthew Ahrens boolean_t doread = bp->b_flags & B_READ; 1146f80ce222SChris Kirby boolean_t is_dump; 1147510b6c0eSNeil Perrin boolean_t sync; 1148fa9e4066Sahrens 1149fa9e4066Sahrens if (zv == NULL) { 1150fa9e4066Sahrens bioerror(bp, ENXIO); 1151fa9e4066Sahrens biodone(bp); 1152fa9e4066Sahrens return (0); 1153fa9e4066Sahrens } 1154fa9e4066Sahrens 1155fa9e4066Sahrens if (getminor(bp->b_edev) == 0) { 1156fa9e4066Sahrens bioerror(bp, EINVAL); 1157fa9e4066Sahrens biodone(bp); 1158fa9e4066Sahrens return (0); 1159fa9e4066Sahrens } 1160fa9e4066Sahrens 1161681d9761SEric Taylor if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) { 1162fa9e4066Sahrens bioerror(bp, EROFS); 1163fa9e4066Sahrens biodone(bp); 1164fa9e4066Sahrens return (0); 1165fa9e4066Sahrens } 1166fa9e4066Sahrens 1167fa9e4066Sahrens off = ldbtob(bp->b_blkno); 1168fa9e4066Sahrens volsize = zv->zv_volsize; 1169fa9e4066Sahrens 117022ac5be4Sperrin os = zv->zv_objset; 117122ac5be4Sperrin ASSERT(os != NULL); 1172fa9e4066Sahrens 1173fa9e4066Sahrens bp_mapin(bp); 1174fa9e4066Sahrens addr = bp->b_un.b_addr; 1175fa9e4066Sahrens resid = bp->b_bcount; 1176fa9e4066Sahrens 117788b7b0f2SMatthew Ahrens if (resid > 0 && (off < 0 || off >= volsize)) { 117888b7b0f2SMatthew Ahrens bioerror(bp, EIO); 117988b7b0f2SMatthew Ahrens biodone(bp); 118088b7b0f2SMatthew Ahrens return (0); 118188b7b0f2SMatthew Ahrens } 118273ec3d9cSgw 1183f80ce222SChris Kirby is_dump = zv->zv_flags & ZVOL_DUMPIFIED; 1184510b6c0eSNeil Perrin sync = !(bp->b_flags & B_ASYNC) && !doread && !is_dump && 1185510b6c0eSNeil Perrin !(zv->zv_flags & ZVOL_WCE) && !zil_disable; 1186510b6c0eSNeil Perrin 1187a24e15ceSperrin /* 1188a24e15ceSperrin * There must be no buffer changes when doing a dmu_sync() because 1189a24e15ceSperrin * we can't change the data whilst calculating the checksum. 1190a24e15ceSperrin */ 1191c2e6a7d6Sperrin rl = zfs_range_lock(&zv->zv_znode, off, resid, 119288b7b0f2SMatthew Ahrens doread ? RL_READER : RL_WRITER); 1193fa9e4066Sahrens 1194e7cbe64fSgw while (resid != 0 && off < volsize) { 119588b7b0f2SMatthew Ahrens size_t size = MIN(resid, zvol_maxphys); 1196e7cbe64fSgw if (is_dump) { 1197e7cbe64fSgw size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 119888b7b0f2SMatthew Ahrens error = zvol_dumpio(zv, addr, off, size, 119988b7b0f2SMatthew Ahrens doread, B_FALSE); 120088b7b0f2SMatthew Ahrens } else if (doread) { 12017bfdf011SNeil Perrin error = dmu_read(os, ZVOL_OBJ, off, size, addr, 12027bfdf011SNeil Perrin DMU_READ_PREFETCH); 1203fa9e4066Sahrens } else { 120422ac5be4Sperrin dmu_tx_t *tx = dmu_tx_create(os); 1205fa9e4066Sahrens dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1206fa9e4066Sahrens error = dmu_tx_assign(tx, TXG_WAIT); 1207fa9e4066Sahrens if (error) { 1208fa9e4066Sahrens dmu_tx_abort(tx); 1209fa9e4066Sahrens } else { 121022ac5be4Sperrin dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1211510b6c0eSNeil Perrin zvol_log_write(zv, tx, off, size, sync); 1212fa9e4066Sahrens dmu_tx_commit(tx); 1213fa9e4066Sahrens } 1214fa9e4066Sahrens } 1215b87f3af3Sperrin if (error) { 1216b87f3af3Sperrin /* convert checksum errors into IO errors */ 1217b87f3af3Sperrin if (error == ECKSUM) 1218b87f3af3Sperrin error = EIO; 1219fa9e4066Sahrens break; 1220b87f3af3Sperrin } 1221fa9e4066Sahrens off += size; 1222fa9e4066Sahrens addr += size; 1223fa9e4066Sahrens resid -= size; 1224fa9e4066Sahrens } 1225c2e6a7d6Sperrin zfs_range_unlock(rl); 1226fa9e4066Sahrens 1227fa9e4066Sahrens if ((bp->b_resid = resid) == bp->b_bcount) 1228fa9e4066Sahrens bioerror(bp, off > volsize ? EINVAL : error); 1229fa9e4066Sahrens 1230510b6c0eSNeil Perrin if (sync) 1231feb08c6bSbillm zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1232feb08c6bSbillm biodone(bp); 123322ac5be4Sperrin 1234fa9e4066Sahrens return (0); 1235fa9e4066Sahrens } 1236fa9e4066Sahrens 123767bd71c6Sperrin /* 123867bd71c6Sperrin * Set the buffer count to the zvol maximum transfer. 123967bd71c6Sperrin * Using our own routine instead of the default minphys() 124067bd71c6Sperrin * means that for larger writes we write bigger buffers on X86 124167bd71c6Sperrin * (128K instead of 56K) and flush the disk write cache less often 124267bd71c6Sperrin * (every zvol_maxphys - currently 1MB) instead of minphys (currently 124367bd71c6Sperrin * 56K on X86 and 128K on sparc). 124467bd71c6Sperrin */ 124567bd71c6Sperrin void 124667bd71c6Sperrin zvol_minphys(struct buf *bp) 124767bd71c6Sperrin { 124867bd71c6Sperrin if (bp->b_bcount > zvol_maxphys) 124967bd71c6Sperrin bp->b_bcount = zvol_maxphys; 125067bd71c6Sperrin } 125167bd71c6Sperrin 1252e7cbe64fSgw int 1253e7cbe64fSgw zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1254e7cbe64fSgw { 1255e7cbe64fSgw minor_t minor = getminor(dev); 1256e7cbe64fSgw zvol_state_t *zv; 1257e7cbe64fSgw int error = 0; 1258e7cbe64fSgw uint64_t size; 1259e7cbe64fSgw uint64_t boff; 1260e7cbe64fSgw uint64_t resid; 1261e7cbe64fSgw 1262e7cbe64fSgw if (minor == 0) /* This is the control device */ 1263e7cbe64fSgw return (ENXIO); 1264e7cbe64fSgw 1265e7cbe64fSgw zv = ddi_get_soft_state(zvol_state, minor); 1266e7cbe64fSgw if (zv == NULL) 1267e7cbe64fSgw return (ENXIO); 1268e7cbe64fSgw 1269e7cbe64fSgw boff = ldbtob(blkno); 1270e7cbe64fSgw resid = ldbtob(nblocks); 127188b7b0f2SMatthew Ahrens 127288b7b0f2SMatthew Ahrens VERIFY3U(boff + resid, <=, zv->zv_volsize); 127388b7b0f2SMatthew Ahrens 1274e7cbe64fSgw while (resid) { 1275e7cbe64fSgw size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 127688b7b0f2SMatthew Ahrens error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1277e7cbe64fSgw if (error) 1278e7cbe64fSgw break; 1279e7cbe64fSgw boff += size; 1280e7cbe64fSgw addr += size; 1281e7cbe64fSgw resid -= size; 1282e7cbe64fSgw } 1283e7cbe64fSgw 1284e7cbe64fSgw return (error); 1285e7cbe64fSgw } 1286e7cbe64fSgw 1287fa9e4066Sahrens /*ARGSUSED*/ 1288fa9e4066Sahrens int 1289feb08c6bSbillm zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1290fa9e4066Sahrens { 1291c7ca1008Sgw minor_t minor = getminor(dev); 1292c7ca1008Sgw zvol_state_t *zv; 129373ec3d9cSgw uint64_t volsize; 1294c2e6a7d6Sperrin rl_t *rl; 1295feb08c6bSbillm int error = 0; 1296fa9e4066Sahrens 1297c7ca1008Sgw if (minor == 0) /* This is the control device */ 1298c7ca1008Sgw return (ENXIO); 1299c7ca1008Sgw 1300c7ca1008Sgw zv = ddi_get_soft_state(zvol_state, minor); 1301c7ca1008Sgw if (zv == NULL) 1302c7ca1008Sgw return (ENXIO); 1303c7ca1008Sgw 130473ec3d9cSgw volsize = zv->zv_volsize; 130573ec3d9cSgw if (uio->uio_resid > 0 && 130673ec3d9cSgw (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 130773ec3d9cSgw return (EIO); 130873ec3d9cSgw 130988b7b0f2SMatthew Ahrens if (zv->zv_flags & ZVOL_DUMPIFIED) { 131088b7b0f2SMatthew Ahrens error = physio(zvol_strategy, NULL, dev, B_READ, 131188b7b0f2SMatthew Ahrens zvol_minphys, uio); 131288b7b0f2SMatthew Ahrens return (error); 131388b7b0f2SMatthew Ahrens } 131488b7b0f2SMatthew Ahrens 1315c2e6a7d6Sperrin rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1316c2e6a7d6Sperrin RL_READER); 131773ec3d9cSgw while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1318feb08c6bSbillm uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1319fa9e4066Sahrens 132073ec3d9cSgw /* don't read past the end */ 132173ec3d9cSgw if (bytes > volsize - uio->uio_loffset) 132273ec3d9cSgw bytes = volsize - uio->uio_loffset; 132373ec3d9cSgw 1324feb08c6bSbillm error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1325b87f3af3Sperrin if (error) { 1326b87f3af3Sperrin /* convert checksum errors into IO errors */ 1327b87f3af3Sperrin if (error == ECKSUM) 1328b87f3af3Sperrin error = EIO; 1329feb08c6bSbillm break; 1330b87f3af3Sperrin } 1331feb08c6bSbillm } 1332c2e6a7d6Sperrin zfs_range_unlock(rl); 1333feb08c6bSbillm return (error); 1334fa9e4066Sahrens } 1335fa9e4066Sahrens 1336fa9e4066Sahrens /*ARGSUSED*/ 1337fa9e4066Sahrens int 1338feb08c6bSbillm zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1339fa9e4066Sahrens { 1340c7ca1008Sgw minor_t minor = getminor(dev); 1341c7ca1008Sgw zvol_state_t *zv; 134273ec3d9cSgw uint64_t volsize; 1343c2e6a7d6Sperrin rl_t *rl; 1344feb08c6bSbillm int error = 0; 1345510b6c0eSNeil Perrin boolean_t sync; 1346feb08c6bSbillm 1347c7ca1008Sgw if (minor == 0) /* This is the control device */ 1348c7ca1008Sgw return (ENXIO); 1349c7ca1008Sgw 1350c7ca1008Sgw zv = ddi_get_soft_state(zvol_state, minor); 1351c7ca1008Sgw if (zv == NULL) 1352c7ca1008Sgw return (ENXIO); 1353c7ca1008Sgw 135473ec3d9cSgw volsize = zv->zv_volsize; 135573ec3d9cSgw if (uio->uio_resid > 0 && 135673ec3d9cSgw (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 135773ec3d9cSgw return (EIO); 135873ec3d9cSgw 1359e7cbe64fSgw if (zv->zv_flags & ZVOL_DUMPIFIED) { 1360e7cbe64fSgw error = physio(zvol_strategy, NULL, dev, B_WRITE, 1361e7cbe64fSgw zvol_minphys, uio); 1362e7cbe64fSgw return (error); 1363e7cbe64fSgw } 1364e7cbe64fSgw 1365510b6c0eSNeil Perrin sync = !(zv->zv_flags & ZVOL_WCE) && !zil_disable; 1366510b6c0eSNeil Perrin 1367c2e6a7d6Sperrin rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1368c2e6a7d6Sperrin RL_WRITER); 136973ec3d9cSgw while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1370feb08c6bSbillm uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1371feb08c6bSbillm uint64_t off = uio->uio_loffset; 1372feb08c6bSbillm dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 137373ec3d9cSgw 137473ec3d9cSgw if (bytes > volsize - off) /* don't write past the end */ 137573ec3d9cSgw bytes = volsize - off; 137673ec3d9cSgw 1377feb08c6bSbillm dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1378feb08c6bSbillm error = dmu_tx_assign(tx, TXG_WAIT); 1379feb08c6bSbillm if (error) { 1380feb08c6bSbillm dmu_tx_abort(tx); 1381feb08c6bSbillm break; 1382feb08c6bSbillm } 1383*94d1a210STim Haley error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1384feb08c6bSbillm if (error == 0) 1385510b6c0eSNeil Perrin zvol_log_write(zv, tx, off, bytes, sync); 1386feb08c6bSbillm dmu_tx_commit(tx); 1387feb08c6bSbillm 1388feb08c6bSbillm if (error) 1389feb08c6bSbillm break; 1390feb08c6bSbillm } 1391c2e6a7d6Sperrin zfs_range_unlock(rl); 1392510b6c0eSNeil Perrin if (sync) 1393e08bf2c6SEric Taylor zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1394feb08c6bSbillm return (error); 1395fa9e4066Sahrens } 1396fa9e4066Sahrens 1397c7f714e2SEric Taylor int 1398c7f714e2SEric Taylor zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1399c7f714e2SEric Taylor { 1400c7f714e2SEric Taylor struct uuid uuid = EFI_RESERVED; 1401c7f714e2SEric Taylor efi_gpe_t gpe = { 0 }; 1402c7f714e2SEric Taylor uint32_t crc; 1403c7f714e2SEric Taylor dk_efi_t efi; 1404c7f714e2SEric Taylor int length; 1405c7f714e2SEric Taylor char *ptr; 1406c7f714e2SEric Taylor 1407c7f714e2SEric Taylor if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1408c7f714e2SEric Taylor return (EFAULT); 1409c7f714e2SEric Taylor ptr = (char *)(uintptr_t)efi.dki_data_64; 1410c7f714e2SEric Taylor length = efi.dki_length; 1411c7f714e2SEric Taylor /* 1412c7f714e2SEric Taylor * Some clients may attempt to request a PMBR for the 1413c7f714e2SEric Taylor * zvol. Currently this interface will return EINVAL to 1414c7f714e2SEric Taylor * such requests. These requests could be supported by 1415c7f714e2SEric Taylor * adding a check for lba == 0 and consing up an appropriate 1416c7f714e2SEric Taylor * PMBR. 1417c7f714e2SEric Taylor */ 1418c7f714e2SEric Taylor if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1419c7f714e2SEric Taylor return (EINVAL); 1420c7f714e2SEric Taylor 1421c7f714e2SEric Taylor gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1422c7f714e2SEric Taylor gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1423c7f714e2SEric Taylor UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1424c7f714e2SEric Taylor 1425c7f714e2SEric Taylor if (efi.dki_lba == 1) { 1426c7f714e2SEric Taylor efi_gpt_t gpt = { 0 }; 1427c7f714e2SEric Taylor 1428c7f714e2SEric Taylor gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1429c7f714e2SEric Taylor gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1430c7f714e2SEric Taylor gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1431c7f714e2SEric Taylor gpt.efi_gpt_MyLBA = LE_64(1ULL); 1432c7f714e2SEric Taylor gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1433c7f714e2SEric Taylor gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1434c7f714e2SEric Taylor gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1435c7f714e2SEric Taylor gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1436c7f714e2SEric Taylor gpt.efi_gpt_SizeOfPartitionEntry = 1437c7f714e2SEric Taylor LE_32(sizeof (efi_gpe_t)); 1438c7f714e2SEric Taylor CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1439c7f714e2SEric Taylor gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1440c7f714e2SEric Taylor CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1441c7f714e2SEric Taylor gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1442c7f714e2SEric Taylor if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1443c7f714e2SEric Taylor flag)) 1444c7f714e2SEric Taylor return (EFAULT); 1445c7f714e2SEric Taylor ptr += sizeof (gpt); 1446c7f714e2SEric Taylor length -= sizeof (gpt); 1447c7f714e2SEric Taylor } 1448c7f714e2SEric Taylor if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1449c7f714e2SEric Taylor length), flag)) 1450c7f714e2SEric Taylor return (EFAULT); 1451c7f714e2SEric Taylor return (0); 1452c7f714e2SEric Taylor } 1453c7f714e2SEric Taylor 1454fa9e4066Sahrens /* 1455fa9e4066Sahrens * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1456fa9e4066Sahrens */ 1457fa9e4066Sahrens /*ARGSUSED*/ 1458fa9e4066Sahrens int 1459fa9e4066Sahrens zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1460fa9e4066Sahrens { 1461fa9e4066Sahrens zvol_state_t *zv; 1462af2c4821Smaybee struct dk_cinfo dki; 1463fa9e4066Sahrens struct dk_minfo dkm; 1464af2c4821Smaybee struct dk_callback *dkc; 1465fa9e4066Sahrens int error = 0; 1466e7cbe64fSgw rl_t *rl; 1467fa9e4066Sahrens 1468fa9e4066Sahrens mutex_enter(&zvol_state_lock); 1469fa9e4066Sahrens 1470fa9e4066Sahrens zv = ddi_get_soft_state(zvol_state, getminor(dev)); 1471fa9e4066Sahrens 1472fa9e4066Sahrens if (zv == NULL) { 1473fa9e4066Sahrens mutex_exit(&zvol_state_lock); 1474fa9e4066Sahrens return (ENXIO); 1475fa9e4066Sahrens } 1476701f66c4SEric Taylor ASSERT(zv->zv_total_opens > 0); 1477fa9e4066Sahrens 1478fa9e4066Sahrens switch (cmd) { 1479fa9e4066Sahrens 1480fa9e4066Sahrens case DKIOCINFO: 1481af2c4821Smaybee bzero(&dki, sizeof (dki)); 1482af2c4821Smaybee (void) strcpy(dki.dki_cname, "zvol"); 1483af2c4821Smaybee (void) strcpy(dki.dki_dname, "zvol"); 1484af2c4821Smaybee dki.dki_ctype = DKC_UNKNOWN; 14853adc9019SEric Taylor dki.dki_unit = getminor(dev); 1486af2c4821Smaybee dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1487fa9e4066Sahrens mutex_exit(&zvol_state_lock); 1488af2c4821Smaybee if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1489fa9e4066Sahrens error = EFAULT; 1490fa9e4066Sahrens return (error); 1491fa9e4066Sahrens 1492fa9e4066Sahrens case DKIOCGMEDIAINFO: 1493fa9e4066Sahrens bzero(&dkm, sizeof (dkm)); 1494fa9e4066Sahrens dkm.dki_lbsize = 1U << zv->zv_min_bs; 1495fa9e4066Sahrens dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1496fa9e4066Sahrens dkm.dki_media_type = DK_UNKNOWN; 1497fa9e4066Sahrens mutex_exit(&zvol_state_lock); 1498fa9e4066Sahrens if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1499fa9e4066Sahrens error = EFAULT; 1500fa9e4066Sahrens return (error); 1501fa9e4066Sahrens 1502fa9e4066Sahrens case DKIOCGETEFI: 1503c7f714e2SEric Taylor { 1504c7f714e2SEric Taylor uint64_t vs = zv->zv_volsize; 1505c7f714e2SEric Taylor uint8_t bs = zv->zv_min_bs; 1506fa9e4066Sahrens 150768a5ac4dSmaybee mutex_exit(&zvol_state_lock); 1508c7f714e2SEric Taylor error = zvol_getefi((void *)arg, flag, vs, bs); 1509c7f714e2SEric Taylor return (error); 151068a5ac4dSmaybee } 1511fa9e4066Sahrens 1512feb08c6bSbillm case DKIOCFLUSHWRITECACHE: 1513af2c4821Smaybee dkc = (struct dk_callback *)arg; 1514701f66c4SEric Taylor mutex_exit(&zvol_state_lock); 1515feb08c6bSbillm zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1516af2c4821Smaybee if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1517af2c4821Smaybee (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1518af2c4821Smaybee error = 0; 1519af2c4821Smaybee } 1520701f66c4SEric Taylor return (error); 1521701f66c4SEric Taylor 1522701f66c4SEric Taylor case DKIOCGETWCE: 1523701f66c4SEric Taylor { 1524701f66c4SEric Taylor int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1525701f66c4SEric Taylor if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1526701f66c4SEric Taylor flag)) 1527701f66c4SEric Taylor error = EFAULT; 1528701f66c4SEric Taylor break; 1529701f66c4SEric Taylor } 1530701f66c4SEric Taylor case DKIOCSETWCE: 1531701f66c4SEric Taylor { 1532701f66c4SEric Taylor int wce; 1533701f66c4SEric Taylor if (ddi_copyin((void *)arg, &wce, sizeof (int), 1534701f66c4SEric Taylor flag)) { 1535701f66c4SEric Taylor error = EFAULT; 1536701f66c4SEric Taylor break; 1537701f66c4SEric Taylor } 1538701f66c4SEric Taylor if (wce) { 1539701f66c4SEric Taylor zv->zv_flags |= ZVOL_WCE; 1540701f66c4SEric Taylor mutex_exit(&zvol_state_lock); 1541701f66c4SEric Taylor } else { 1542701f66c4SEric Taylor zv->zv_flags &= ~ZVOL_WCE; 1543701f66c4SEric Taylor mutex_exit(&zvol_state_lock); 1544701f66c4SEric Taylor zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1545701f66c4SEric Taylor } 1546701f66c4SEric Taylor return (0); 1547701f66c4SEric Taylor } 1548feb08c6bSbillm 1549b6130eadSmaybee case DKIOCGGEOM: 1550b6130eadSmaybee case DKIOCGVTOC: 1551e7cbe64fSgw /* 1552e7cbe64fSgw * commands using these (like prtvtoc) expect ENOTSUP 1553e7cbe64fSgw * since we're emulating an EFI label 1554e7cbe64fSgw */ 1555b6130eadSmaybee error = ENOTSUP; 1556b6130eadSmaybee break; 1557b6130eadSmaybee 1558e7cbe64fSgw case DKIOCDUMPINIT: 1559e7cbe64fSgw rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1560e7cbe64fSgw RL_WRITER); 1561e7cbe64fSgw error = zvol_dumpify(zv); 1562e7cbe64fSgw zfs_range_unlock(rl); 1563e7cbe64fSgw break; 1564e7cbe64fSgw 1565e7cbe64fSgw case DKIOCDUMPFINI: 156606d5ae10SEric Taylor if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 156706d5ae10SEric Taylor break; 1568e7cbe64fSgw rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1569e7cbe64fSgw RL_WRITER); 1570e7cbe64fSgw error = zvol_dump_fini(zv); 1571e7cbe64fSgw zfs_range_unlock(rl); 1572e7cbe64fSgw break; 1573e7cbe64fSgw 1574fa9e4066Sahrens default: 157568a5ac4dSmaybee error = ENOTTY; 1576fa9e4066Sahrens break; 1577fa9e4066Sahrens 1578fa9e4066Sahrens } 1579fa9e4066Sahrens mutex_exit(&zvol_state_lock); 1580fa9e4066Sahrens return (error); 1581fa9e4066Sahrens } 1582fa9e4066Sahrens 1583fa9e4066Sahrens int 1584fa9e4066Sahrens zvol_busy(void) 1585fa9e4066Sahrens { 1586fa9e4066Sahrens return (zvol_minors != 0); 1587fa9e4066Sahrens } 1588fa9e4066Sahrens 1589fa9e4066Sahrens void 1590fa9e4066Sahrens zvol_init(void) 1591fa9e4066Sahrens { 1592fa9e4066Sahrens VERIFY(ddi_soft_state_init(&zvol_state, sizeof (zvol_state_t), 1) == 0); 1593fa9e4066Sahrens mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL); 1594fa9e4066Sahrens } 1595fa9e4066Sahrens 1596fa9e4066Sahrens void 1597fa9e4066Sahrens zvol_fini(void) 1598fa9e4066Sahrens { 1599fa9e4066Sahrens mutex_destroy(&zvol_state_lock); 1600fa9e4066Sahrens ddi_soft_state_fini(&zvol_state); 1601fa9e4066Sahrens } 1602e7cbe64fSgw 1603e7cbe64fSgw static int 1604e7cbe64fSgw zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1605e7cbe64fSgw { 1606e7cbe64fSgw dmu_tx_t *tx; 1607e7cbe64fSgw int error = 0; 1608e7cbe64fSgw objset_t *os = zv->zv_objset; 1609e7cbe64fSgw nvlist_t *nv = NULL; 16108d265e66SGeorge Wilson uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 1611e7cbe64fSgw 1612e7cbe64fSgw ASSERT(MUTEX_HELD(&zvol_state_lock)); 1613681d9761SEric Taylor error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1614681d9761SEric Taylor DMU_OBJECT_END); 1615681d9761SEric Taylor /* wait for dmu_free_long_range to actually free the blocks */ 1616681d9761SEric Taylor txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1617e7cbe64fSgw 1618e7cbe64fSgw tx = dmu_tx_create(os); 1619e7cbe64fSgw dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1620681d9761SEric Taylor dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1621e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 1622e7cbe64fSgw if (error) { 1623e7cbe64fSgw dmu_tx_abort(tx); 1624e7cbe64fSgw return (error); 1625e7cbe64fSgw } 1626e7cbe64fSgw 1627e7cbe64fSgw /* 1628e7cbe64fSgw * If we are resizing the dump device then we only need to 1629e7cbe64fSgw * update the refreservation to match the newly updated 1630e7cbe64fSgw * zvolsize. Otherwise, we save off the original state of the 1631e7cbe64fSgw * zvol so that we can restore them if the zvol is ever undumpified. 1632e7cbe64fSgw */ 1633e7cbe64fSgw if (resize) { 1634e7cbe64fSgw error = zap_update(os, ZVOL_ZAP_OBJ, 1635e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1636e7cbe64fSgw &zv->zv_volsize, tx); 1637e7cbe64fSgw } else { 1638afee20e4SGeorge Wilson uint64_t checksum, compress, refresrv, vbs, dedup; 163988b7b0f2SMatthew Ahrens 1640e7cbe64fSgw error = dsl_prop_get_integer(zv->zv_name, 1641e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1642e7cbe64fSgw error = error ? error : dsl_prop_get_integer(zv->zv_name, 1643e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 1644e7cbe64fSgw error = error ? error : dsl_prop_get_integer(zv->zv_name, 1645e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 164688b7b0f2SMatthew Ahrens error = error ? error : dsl_prop_get_integer(zv->zv_name, 164788b7b0f2SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 16488d265e66SGeorge Wilson if (version >= SPA_VERSION_DEDUP) { 16498d265e66SGeorge Wilson error = error ? error : 16508d265e66SGeorge Wilson dsl_prop_get_integer(zv->zv_name, 16518d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 16528d265e66SGeorge Wilson } 1653e7cbe64fSgw 1654e7cbe64fSgw error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1655e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1656e7cbe64fSgw &compress, tx); 1657e7cbe64fSgw error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1658e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 1659e7cbe64fSgw error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1660e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1661e7cbe64fSgw &refresrv, tx); 166288b7b0f2SMatthew Ahrens error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 166388b7b0f2SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 166488b7b0f2SMatthew Ahrens &vbs, tx); 1665681d9761SEric Taylor error = error ? error : dmu_object_set_blocksize( 1666681d9761SEric Taylor os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 16678d265e66SGeorge Wilson if (version >= SPA_VERSION_DEDUP) { 16688d265e66SGeorge Wilson error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 16698d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 16708d265e66SGeorge Wilson &dedup, tx); 16718d265e66SGeorge Wilson } 1672681d9761SEric Taylor if (error == 0) 1673681d9761SEric Taylor zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 1674e7cbe64fSgw } 1675e7cbe64fSgw dmu_tx_commit(tx); 1676e7cbe64fSgw 1677e7cbe64fSgw /* 1678e7cbe64fSgw * We only need update the zvol's property if we are initializing 1679e7cbe64fSgw * the dump area for the first time. 1680e7cbe64fSgw */ 1681e7cbe64fSgw if (!resize) { 1682e7cbe64fSgw VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1683e7cbe64fSgw VERIFY(nvlist_add_uint64(nv, 1684e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 1685e7cbe64fSgw VERIFY(nvlist_add_uint64(nv, 1686e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), 1687e7cbe64fSgw ZIO_COMPRESS_OFF) == 0); 1688e7cbe64fSgw VERIFY(nvlist_add_uint64(nv, 1689e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), 1690e7cbe64fSgw ZIO_CHECKSUM_OFF) == 0); 16918d265e66SGeorge Wilson if (version >= SPA_VERSION_DEDUP) { 16928d265e66SGeorge Wilson VERIFY(nvlist_add_uint64(nv, 16938d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), 16948d265e66SGeorge Wilson ZIO_CHECKSUM_OFF) == 0); 16958d265e66SGeorge Wilson } 1696e7cbe64fSgw 169792241e0bSTom Erickson error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 169892241e0bSTom Erickson nv, NULL); 1699e7cbe64fSgw nvlist_free(nv); 1700e7cbe64fSgw 1701e7cbe64fSgw if (error) 1702e7cbe64fSgw return (error); 1703e7cbe64fSgw } 1704e7cbe64fSgw 1705e7cbe64fSgw /* Allocate the space for the dump */ 1706e7cbe64fSgw error = zvol_prealloc(zv); 1707e7cbe64fSgw return (error); 1708e7cbe64fSgw } 1709e7cbe64fSgw 1710e7cbe64fSgw static int 1711e7cbe64fSgw zvol_dumpify(zvol_state_t *zv) 1712e7cbe64fSgw { 1713e7cbe64fSgw int error = 0; 1714e7cbe64fSgw uint64_t dumpsize = 0; 1715e7cbe64fSgw dmu_tx_t *tx; 1716e7cbe64fSgw objset_t *os = zv->zv_objset; 1717e7cbe64fSgw 1718681d9761SEric Taylor if (zv->zv_flags & ZVOL_RDONLY) 1719e7cbe64fSgw return (EROFS); 1720e7cbe64fSgw 1721e7cbe64fSgw if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 1722e7cbe64fSgw 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 1723e7cbe64fSgw boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE; 1724e7cbe64fSgw 1725e7cbe64fSgw if ((error = zvol_dump_init(zv, resize)) != 0) { 1726e7cbe64fSgw (void) zvol_dump_fini(zv); 1727e7cbe64fSgw return (error); 1728e7cbe64fSgw } 1729e7cbe64fSgw } 1730e7cbe64fSgw 1731e7cbe64fSgw /* 1732e7cbe64fSgw * Build up our lba mapping. 1733e7cbe64fSgw */ 1734e7cbe64fSgw error = zvol_get_lbas(zv); 1735e7cbe64fSgw if (error) { 1736e7cbe64fSgw (void) zvol_dump_fini(zv); 1737e7cbe64fSgw return (error); 1738e7cbe64fSgw } 1739e7cbe64fSgw 1740e7cbe64fSgw tx = dmu_tx_create(os); 1741e7cbe64fSgw dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1742e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 1743e7cbe64fSgw if (error) { 1744e7cbe64fSgw dmu_tx_abort(tx); 1745e7cbe64fSgw (void) zvol_dump_fini(zv); 1746e7cbe64fSgw return (error); 1747e7cbe64fSgw } 1748e7cbe64fSgw 1749e7cbe64fSgw zv->zv_flags |= ZVOL_DUMPIFIED; 1750e7cbe64fSgw error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 1751e7cbe64fSgw &zv->zv_volsize, tx); 1752e7cbe64fSgw dmu_tx_commit(tx); 1753e7cbe64fSgw 1754e7cbe64fSgw if (error) { 1755e7cbe64fSgw (void) zvol_dump_fini(zv); 1756e7cbe64fSgw return (error); 1757e7cbe64fSgw } 1758e7cbe64fSgw 1759e7cbe64fSgw txg_wait_synced(dmu_objset_pool(os), 0); 1760e7cbe64fSgw return (0); 1761e7cbe64fSgw } 1762e7cbe64fSgw 1763e7cbe64fSgw static int 1764e7cbe64fSgw zvol_dump_fini(zvol_state_t *zv) 1765e7cbe64fSgw { 1766e7cbe64fSgw dmu_tx_t *tx; 1767e7cbe64fSgw objset_t *os = zv->zv_objset; 1768e7cbe64fSgw nvlist_t *nv; 1769e7cbe64fSgw int error = 0; 1770afee20e4SGeorge Wilson uint64_t checksum, compress, refresrv, vbs, dedup; 17718d265e66SGeorge Wilson uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 1772e7cbe64fSgw 1773b7e50089Smaybee /* 1774b7e50089Smaybee * Attempt to restore the zvol back to its pre-dumpified state. 1775b7e50089Smaybee * This is a best-effort attempt as it's possible that not all 1776b7e50089Smaybee * of these properties were initialized during the dumpify process 1777b7e50089Smaybee * (i.e. error during zvol_dump_init). 1778b7e50089Smaybee */ 1779b7e50089Smaybee 1780e7cbe64fSgw tx = dmu_tx_create(os); 1781e7cbe64fSgw dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1782e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 1783e7cbe64fSgw if (error) { 1784e7cbe64fSgw dmu_tx_abort(tx); 1785e7cbe64fSgw return (error); 1786e7cbe64fSgw } 1787b7e50089Smaybee (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 1788b7e50089Smaybee dmu_tx_commit(tx); 1789e7cbe64fSgw 1790e7cbe64fSgw (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1791e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 1792e7cbe64fSgw (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1793e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 1794e7cbe64fSgw (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1795e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 179688b7b0f2SMatthew Ahrens (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 179788b7b0f2SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 1798e7cbe64fSgw 1799e7cbe64fSgw VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1800e7cbe64fSgw (void) nvlist_add_uint64(nv, 1801e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 1802e7cbe64fSgw (void) nvlist_add_uint64(nv, 1803e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 1804e7cbe64fSgw (void) nvlist_add_uint64(nv, 1805e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 18068d265e66SGeorge Wilson if (version >= SPA_VERSION_DEDUP && 18078d265e66SGeorge Wilson zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 18088d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 18098d265e66SGeorge Wilson (void) nvlist_add_uint64(nv, 18108d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 18118d265e66SGeorge Wilson } 181292241e0bSTom Erickson (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 181392241e0bSTom Erickson nv, NULL); 1814e7cbe64fSgw nvlist_free(nv); 1815e7cbe64fSgw 1816b7e50089Smaybee zvol_free_extents(zv); 1817b7e50089Smaybee zv->zv_flags &= ~ZVOL_DUMPIFIED; 1818b7e50089Smaybee (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 1819681d9761SEric Taylor /* wait for dmu_free_long_range to actually free the blocks */ 1820681d9761SEric Taylor txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1821681d9761SEric Taylor tx = dmu_tx_create(os); 1822681d9761SEric Taylor dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1823681d9761SEric Taylor error = dmu_tx_assign(tx, TXG_WAIT); 1824681d9761SEric Taylor if (error) { 1825681d9761SEric Taylor dmu_tx_abort(tx); 1826681d9761SEric Taylor return (error); 1827681d9761SEric Taylor } 1828b24ab676SJeff Bonwick if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 1829b24ab676SJeff Bonwick zv->zv_volblocksize = vbs; 1830681d9761SEric Taylor dmu_tx_commit(tx); 1831b7e50089Smaybee 1832e7cbe64fSgw return (0); 1833e7cbe64fSgw } 1834