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. 23b77b9231SDan McDonald * 24b77b9231SDan McDonald * Portions Copyright 2010 Robert Milkowski 25b77b9231SDan McDonald * 26b77b9231SDan McDonald * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 27*7802d7bfSMatthew Ahrens * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 28810e43b2SBill Pijewski * Copyright (c) 2013, Joyent, Inc. All rights reserved. 29fa9e4066Sahrens */ 30fa9e4066Sahrens 31fa9e4066Sahrens /* 32fa9e4066Sahrens * ZFS volume emulation driver. 33fa9e4066Sahrens * 34fa9e4066Sahrens * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 35fa9e4066Sahrens * Volumes are accessed through the symbolic links named: 36fa9e4066Sahrens * 37fa9e4066Sahrens * /dev/zvol/dsk/<pool_name>/<dataset_name> 38fa9e4066Sahrens * /dev/zvol/rdsk/<pool_name>/<dataset_name> 39fa9e4066Sahrens * 40681d9761SEric Taylor * These links are created by the /dev filesystem (sdev_zvolops.c). 41fa9e4066Sahrens * Volumes are persistent through reboot. No user command needs to be 42fa9e4066Sahrens * run before opening and using a device. 43fa9e4066Sahrens */ 44fa9e4066Sahrens 45fa9e4066Sahrens #include <sys/types.h> 46fa9e4066Sahrens #include <sys/param.h> 47fa9e4066Sahrens #include <sys/errno.h> 48fa9e4066Sahrens #include <sys/uio.h> 49fa9e4066Sahrens #include <sys/buf.h> 50fa9e4066Sahrens #include <sys/modctl.h> 51fa9e4066Sahrens #include <sys/open.h> 52fa9e4066Sahrens #include <sys/kmem.h> 53fa9e4066Sahrens #include <sys/conf.h> 54fa9e4066Sahrens #include <sys/cmn_err.h> 55fa9e4066Sahrens #include <sys/stat.h> 56fa9e4066Sahrens #include <sys/zap.h> 57fa9e4066Sahrens #include <sys/spa.h> 58810e43b2SBill Pijewski #include <sys/spa_impl.h> 59fa9e4066Sahrens #include <sys/zio.h> 60e7cbe64fSgw #include <sys/dmu_traverse.h> 61e7cbe64fSgw #include <sys/dnode.h> 62e7cbe64fSgw #include <sys/dsl_dataset.h> 63fa9e4066Sahrens #include <sys/dsl_prop.h> 64fa9e4066Sahrens #include <sys/dkio.h> 65fa9e4066Sahrens #include <sys/efi_partition.h> 66fa9e4066Sahrens #include <sys/byteorder.h> 67fa9e4066Sahrens #include <sys/pathname.h> 68fa9e4066Sahrens #include <sys/ddi.h> 69fa9e4066Sahrens #include <sys/sunddi.h> 70fa9e4066Sahrens #include <sys/crc32.h> 71fa9e4066Sahrens #include <sys/dirent.h> 72fa9e4066Sahrens #include <sys/policy.h> 73fa9e4066Sahrens #include <sys/fs/zfs.h> 74fa9e4066Sahrens #include <sys/zfs_ioctl.h> 75fa9e4066Sahrens #include <sys/mkdev.h> 7622ac5be4Sperrin #include <sys/zil.h> 77c5c6ffa0Smaybee #include <sys/refcount.h> 78c2e6a7d6Sperrin #include <sys/zfs_znode.h> 79c2e6a7d6Sperrin #include <sys/zfs_rlock.h> 80e7cbe64fSgw #include <sys/vdev_disk.h> 81e7cbe64fSgw #include <sys/vdev_impl.h> 82810e43b2SBill Pijewski #include <sys/vdev_raidz.h> 83e7cbe64fSgw #include <sys/zvol.h> 84e7cbe64fSgw #include <sys/dumphdr.h> 851209a471SNeil Perrin #include <sys/zil_impl.h> 8680901aeaSGeorge Wilson #include <sys/dbuf.h> 87810e43b2SBill Pijewski #include <sys/dmu_tx.h> 88810e43b2SBill Pijewski #include <sys/zfeature.h> 89810e43b2SBill Pijewski #include <sys/zio_checksum.h> 90fa9e4066Sahrens 91fa9e4066Sahrens #include "zfs_namecheck.h" 92fa9e4066Sahrens 93c99e4bdcSChris Kirby void *zfsdev_state; 94503ad85cSMatthew Ahrens static char *zvol_tag = "zvol_tag"; 95fa9e4066Sahrens 96e7cbe64fSgw #define ZVOL_DUMPSIZE "dumpsize" 97e7cbe64fSgw 98fa9e4066Sahrens /* 99c99e4bdcSChris Kirby * This lock protects the zfsdev_state structure from being modified 100fa9e4066Sahrens * while it's being used, e.g. an open that comes in before a create 101fa9e4066Sahrens * finishes. It also protects temporary opens of the dataset so that, 102fa9e4066Sahrens * e.g., an open doesn't get a spurious EBUSY. 103fa9e4066Sahrens */ 104c99e4bdcSChris Kirby kmutex_t zfsdev_state_lock; 105fa9e4066Sahrens static uint32_t zvol_minors; 106fa9e4066Sahrens 107e7cbe64fSgw typedef struct zvol_extent { 10888b7b0f2SMatthew Ahrens list_node_t ze_node; 109e7cbe64fSgw dva_t ze_dva; /* dva associated with this extent */ 11088b7b0f2SMatthew Ahrens uint64_t ze_nblks; /* number of blocks in extent */ 111e7cbe64fSgw } zvol_extent_t; 112e7cbe64fSgw 113fa9e4066Sahrens /* 114fa9e4066Sahrens * The in-core state of each volume. 115fa9e4066Sahrens */ 116fa9e4066Sahrens typedef struct zvol_state { 117fa9e4066Sahrens char zv_name[MAXPATHLEN]; /* pool/dd name */ 118fa9e4066Sahrens uint64_t zv_volsize; /* amount of space we advertise */ 11967bd71c6Sperrin uint64_t zv_volblocksize; /* volume block size */ 120fa9e4066Sahrens minor_t zv_minor; /* minor number */ 121fa9e4066Sahrens uint8_t zv_min_bs; /* minimum addressable block shift */ 122701f66c4SEric Taylor uint8_t zv_flags; /* readonly, dumpified, etc. */ 123fa9e4066Sahrens objset_t *zv_objset; /* objset handle */ 124fa9e4066Sahrens uint32_t zv_open_count[OTYPCNT]; /* open counts */ 125fa9e4066Sahrens uint32_t zv_total_opens; /* total open count */ 12622ac5be4Sperrin zilog_t *zv_zilog; /* ZIL handle */ 12788b7b0f2SMatthew Ahrens list_t zv_extents; /* List of extents for dump */ 128c2e6a7d6Sperrin znode_t zv_znode; /* for range locking */ 12994d1a210STim Haley dmu_buf_t *zv_dbuf; /* bonus handle */ 130fa9e4066Sahrens } zvol_state_t; 131fa9e4066Sahrens 132e7cbe64fSgw /* 133e7cbe64fSgw * zvol specific flags 134e7cbe64fSgw */ 135e7cbe64fSgw #define ZVOL_RDONLY 0x1 136e7cbe64fSgw #define ZVOL_DUMPIFIED 0x2 137c7f714e2SEric Taylor #define ZVOL_EXCL 0x4 138701f66c4SEric Taylor #define ZVOL_WCE 0x8 139e7cbe64fSgw 14067bd71c6Sperrin /* 14167bd71c6Sperrin * zvol maximum transfer in one DMU tx. 14267bd71c6Sperrin */ 14367bd71c6Sperrin int zvol_maxphys = DMU_MAX_ACCESS/2; 14467bd71c6Sperrin 14592241e0bSTom Erickson extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 1464445fffbSMatthew Ahrens nvlist_t *, nvlist_t *); 147681d9761SEric Taylor static int zvol_remove_zv(zvol_state_t *); 148feb08c6bSbillm static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 149e7cbe64fSgw static int zvol_dumpify(zvol_state_t *zv); 150e7cbe64fSgw static int zvol_dump_fini(zvol_state_t *zv); 151e7cbe64fSgw static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 15267bd71c6Sperrin 153fa9e4066Sahrens static void 154c61ea566SGeorge Wilson zvol_size_changed(zvol_state_t *zv, uint64_t volsize) 155fa9e4066Sahrens { 156c61ea566SGeorge Wilson dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor); 157fa9e4066Sahrens 158c61ea566SGeorge Wilson zv->zv_volsize = volsize; 159fa9e4066Sahrens VERIFY(ddi_prop_update_int64(dev, zfs_dip, 160681d9761SEric Taylor "Size", volsize) == DDI_SUCCESS); 161fa9e4066Sahrens VERIFY(ddi_prop_update_int64(dev, zfs_dip, 162681d9761SEric Taylor "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 163e7cbe64fSgw 164e7cbe64fSgw /* Notify specfs to invalidate the cached size */ 165e7cbe64fSgw spec_size_invalidate(dev, VBLK); 166e7cbe64fSgw spec_size_invalidate(dev, VCHR); 167fa9e4066Sahrens } 168fa9e4066Sahrens 169fa9e4066Sahrens int 170e9dbad6fSeschrock zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 171fa9e4066Sahrens { 172e9dbad6fSeschrock if (volsize == 0) 173be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 174fa9e4066Sahrens 175e9dbad6fSeschrock if (volsize % blocksize != 0) 176be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1775c5460e9Seschrock 178fa9e4066Sahrens #ifdef _ILP32 179e9dbad6fSeschrock if (volsize - 1 > SPEC_MAXOFFSET_T) 180be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 181fa9e4066Sahrens #endif 182fa9e4066Sahrens return (0); 183fa9e4066Sahrens } 184fa9e4066Sahrens 185fa9e4066Sahrens int 186e9dbad6fSeschrock zvol_check_volblocksize(uint64_t volblocksize) 187fa9e4066Sahrens { 188e9dbad6fSeschrock if (volblocksize < SPA_MINBLOCKSIZE || 189e9dbad6fSeschrock volblocksize > SPA_MAXBLOCKSIZE || 190e9dbad6fSeschrock !ISP2(volblocksize)) 191be6fd75aSMatthew Ahrens return (SET_ERROR(EDOM)); 192fa9e4066Sahrens 193fa9e4066Sahrens return (0); 194fa9e4066Sahrens } 195fa9e4066Sahrens 196fa9e4066Sahrens int 197a2eea2e1Sahrens zvol_get_stats(objset_t *os, nvlist_t *nv) 198fa9e4066Sahrens { 199fa9e4066Sahrens int error; 200fa9e4066Sahrens dmu_object_info_t doi; 201a2eea2e1Sahrens uint64_t val; 202fa9e4066Sahrens 203a2eea2e1Sahrens error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 204fa9e4066Sahrens if (error) 205fa9e4066Sahrens return (error); 206fa9e4066Sahrens 207a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 208a2eea2e1Sahrens 209fa9e4066Sahrens error = dmu_object_info(os, ZVOL_OBJ, &doi); 210fa9e4066Sahrens 211a2eea2e1Sahrens if (error == 0) { 212a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 213a2eea2e1Sahrens doi.doi_data_block_size); 214a2eea2e1Sahrens } 215fa9e4066Sahrens 216fa9e4066Sahrens return (error); 217fa9e4066Sahrens } 218fa9e4066Sahrens 219fa9e4066Sahrens static zvol_state_t * 220e9dbad6fSeschrock zvol_minor_lookup(const char *name) 221fa9e4066Sahrens { 222fa9e4066Sahrens minor_t minor; 223fa9e4066Sahrens zvol_state_t *zv; 224fa9e4066Sahrens 225c99e4bdcSChris Kirby ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 226fa9e4066Sahrens 227c99e4bdcSChris Kirby for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 228c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 229fa9e4066Sahrens if (zv == NULL) 230fa9e4066Sahrens continue; 231fa9e4066Sahrens if (strcmp(zv->zv_name, name) == 0) 232f80ce222SChris Kirby return (zv); 233fa9e4066Sahrens } 234fa9e4066Sahrens 235f80ce222SChris Kirby return (NULL); 236fa9e4066Sahrens } 237fa9e4066Sahrens 238e7cbe64fSgw /* extent mapping arg */ 239e7cbe64fSgw struct maparg { 24088b7b0f2SMatthew Ahrens zvol_state_t *ma_zv; 24188b7b0f2SMatthew Ahrens uint64_t ma_blks; 242e7cbe64fSgw }; 243e7cbe64fSgw 244e7cbe64fSgw /*ARGSUSED*/ 245e7cbe64fSgw static int 2461b912ec7SGeorge Wilson zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 247*7802d7bfSMatthew Ahrens const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 248e7cbe64fSgw { 24988b7b0f2SMatthew Ahrens struct maparg *ma = arg; 25088b7b0f2SMatthew Ahrens zvol_extent_t *ze; 25188b7b0f2SMatthew Ahrens int bs = ma->ma_zv->zv_volblocksize; 252e7cbe64fSgw 25343466aaeSMax Grossman if (BP_IS_HOLE(bp) || 25443466aaeSMax Grossman zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 25588b7b0f2SMatthew Ahrens return (0); 256e7cbe64fSgw 2575d7b4d43SMatthew Ahrens VERIFY(!BP_IS_EMBEDDED(bp)); 2585d7b4d43SMatthew Ahrens 25988b7b0f2SMatthew Ahrens VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 26088b7b0f2SMatthew Ahrens ma->ma_blks++; 261e7cbe64fSgw 26288b7b0f2SMatthew Ahrens /* Abort immediately if we have encountered gang blocks */ 26388b7b0f2SMatthew Ahrens if (BP_IS_GANG(bp)) 264be6fd75aSMatthew Ahrens return (SET_ERROR(EFRAGS)); 265e7cbe64fSgw 26688b7b0f2SMatthew Ahrens /* 26788b7b0f2SMatthew Ahrens * See if the block is at the end of the previous extent. 26888b7b0f2SMatthew Ahrens */ 26988b7b0f2SMatthew Ahrens ze = list_tail(&ma->ma_zv->zv_extents); 27088b7b0f2SMatthew Ahrens if (ze && 27188b7b0f2SMatthew Ahrens DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 27288b7b0f2SMatthew Ahrens DVA_GET_OFFSET(BP_IDENTITY(bp)) == 27388b7b0f2SMatthew Ahrens DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 27488b7b0f2SMatthew Ahrens ze->ze_nblks++; 27588b7b0f2SMatthew Ahrens return (0); 276e7cbe64fSgw } 277e7cbe64fSgw 27888b7b0f2SMatthew Ahrens dprintf_bp(bp, "%s", "next blkptr:"); 279e7cbe64fSgw 28088b7b0f2SMatthew Ahrens /* start a new extent */ 28188b7b0f2SMatthew Ahrens ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 28288b7b0f2SMatthew Ahrens ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 28388b7b0f2SMatthew Ahrens ze->ze_nblks = 1; 28488b7b0f2SMatthew Ahrens list_insert_tail(&ma->ma_zv->zv_extents, ze); 28588b7b0f2SMatthew Ahrens return (0); 28688b7b0f2SMatthew Ahrens } 287e7cbe64fSgw 28888b7b0f2SMatthew Ahrens static void 28988b7b0f2SMatthew Ahrens zvol_free_extents(zvol_state_t *zv) 29088b7b0f2SMatthew Ahrens { 29188b7b0f2SMatthew Ahrens zvol_extent_t *ze; 292e7cbe64fSgw 29388b7b0f2SMatthew Ahrens while (ze = list_head(&zv->zv_extents)) { 29488b7b0f2SMatthew Ahrens list_remove(&zv->zv_extents, ze); 29588b7b0f2SMatthew Ahrens kmem_free(ze, sizeof (zvol_extent_t)); 296e7cbe64fSgw } 29788b7b0f2SMatthew Ahrens } 298e7cbe64fSgw 29988b7b0f2SMatthew Ahrens static int 30088b7b0f2SMatthew Ahrens zvol_get_lbas(zvol_state_t *zv) 30188b7b0f2SMatthew Ahrens { 3023adc9019SEric Taylor objset_t *os = zv->zv_objset; 30388b7b0f2SMatthew Ahrens struct maparg ma; 30488b7b0f2SMatthew Ahrens int err; 30588b7b0f2SMatthew Ahrens 30688b7b0f2SMatthew Ahrens ma.ma_zv = zv; 30788b7b0f2SMatthew Ahrens ma.ma_blks = 0; 30888b7b0f2SMatthew Ahrens zvol_free_extents(zv); 30988b7b0f2SMatthew Ahrens 3103adc9019SEric Taylor /* commit any in-flight changes before traversing the dataset */ 3113adc9019SEric Taylor txg_wait_synced(dmu_objset_pool(os), 0); 3123adc9019SEric Taylor err = traverse_dataset(dmu_objset_ds(os), 0, 31388b7b0f2SMatthew Ahrens TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 31488b7b0f2SMatthew Ahrens if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 31588b7b0f2SMatthew Ahrens zvol_free_extents(zv); 31688b7b0f2SMatthew Ahrens return (err ? err : EIO); 317e7cbe64fSgw } 31888b7b0f2SMatthew Ahrens 319e7cbe64fSgw return (0); 320e7cbe64fSgw } 321e7cbe64fSgw 322ecd6cf80Smarks /* ARGSUSED */ 323fa9e4066Sahrens void 324ecd6cf80Smarks zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 325fa9e4066Sahrens { 326da6c28aaSamw zfs_creat_t *zct = arg; 327da6c28aaSamw nvlist_t *nvprops = zct->zct_props; 328fa9e4066Sahrens int error; 329e9dbad6fSeschrock uint64_t volblocksize, volsize; 330fa9e4066Sahrens 331ecd6cf80Smarks VERIFY(nvlist_lookup_uint64(nvprops, 332e9dbad6fSeschrock zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 333ecd6cf80Smarks if (nvlist_lookup_uint64(nvprops, 334e9dbad6fSeschrock zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 335e9dbad6fSeschrock volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 336e9dbad6fSeschrock 337e9dbad6fSeschrock /* 338e7cbe64fSgw * These properties must be removed from the list so the generic 339e9dbad6fSeschrock * property setting step won't apply to them. 340e9dbad6fSeschrock */ 341ecd6cf80Smarks VERIFY(nvlist_remove_all(nvprops, 342e9dbad6fSeschrock zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 343ecd6cf80Smarks (void) nvlist_remove_all(nvprops, 344e9dbad6fSeschrock zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 345e9dbad6fSeschrock 346e9dbad6fSeschrock error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 347fa9e4066Sahrens DMU_OT_NONE, 0, tx); 348fa9e4066Sahrens ASSERT(error == 0); 349fa9e4066Sahrens 350fa9e4066Sahrens error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 351fa9e4066Sahrens DMU_OT_NONE, 0, tx); 352fa9e4066Sahrens ASSERT(error == 0); 353fa9e4066Sahrens 354e9dbad6fSeschrock error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 355fa9e4066Sahrens ASSERT(error == 0); 356fa9e4066Sahrens } 357fa9e4066Sahrens 358b77b9231SDan McDonald /* 359b77b9231SDan McDonald * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 360b77b9231SDan McDonald * implement DKIOCFREE/free-long-range. 361b77b9231SDan McDonald */ 362b77b9231SDan McDonald static int 363b77b9231SDan McDonald zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap) 364b77b9231SDan McDonald { 365b77b9231SDan McDonald uint64_t offset, length; 366b77b9231SDan McDonald 367b77b9231SDan McDonald if (byteswap) 368b77b9231SDan McDonald byteswap_uint64_array(lr, sizeof (*lr)); 369b77b9231SDan McDonald 370b77b9231SDan McDonald offset = lr->lr_offset; 371b77b9231SDan McDonald length = lr->lr_length; 372b77b9231SDan McDonald 373b77b9231SDan McDonald return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); 374b77b9231SDan McDonald } 375b77b9231SDan McDonald 37622ac5be4Sperrin /* 37722ac5be4Sperrin * Replay a TX_WRITE ZIL transaction that didn't get committed 37822ac5be4Sperrin * after a system failure 37922ac5be4Sperrin */ 38022ac5be4Sperrin static int 38122ac5be4Sperrin zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 38222ac5be4Sperrin { 38322ac5be4Sperrin objset_t *os = zv->zv_objset; 38422ac5be4Sperrin char *data = (char *)(lr + 1); /* data follows lr_write_t */ 385b24ab676SJeff Bonwick uint64_t offset, length; 38622ac5be4Sperrin dmu_tx_t *tx; 38722ac5be4Sperrin int error; 38822ac5be4Sperrin 38922ac5be4Sperrin if (byteswap) 39022ac5be4Sperrin byteswap_uint64_array(lr, sizeof (*lr)); 39122ac5be4Sperrin 392b24ab676SJeff Bonwick offset = lr->lr_offset; 393b24ab676SJeff Bonwick length = lr->lr_length; 394b24ab676SJeff Bonwick 395b24ab676SJeff Bonwick /* If it's a dmu_sync() block, write the whole block */ 396b24ab676SJeff Bonwick if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 397b24ab676SJeff Bonwick uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 398b24ab676SJeff Bonwick if (length < blocksize) { 399b24ab676SJeff Bonwick offset -= offset % blocksize; 400b24ab676SJeff Bonwick length = blocksize; 401b24ab676SJeff Bonwick } 402b24ab676SJeff Bonwick } 403975c32a0SNeil Perrin 40422ac5be4Sperrin tx = dmu_tx_create(os); 405b24ab676SJeff Bonwick dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 4061209a471SNeil Perrin error = dmu_tx_assign(tx, TXG_WAIT); 40722ac5be4Sperrin if (error) { 40822ac5be4Sperrin dmu_tx_abort(tx); 40922ac5be4Sperrin } else { 410b24ab676SJeff Bonwick dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 41122ac5be4Sperrin dmu_tx_commit(tx); 41222ac5be4Sperrin } 41322ac5be4Sperrin 41422ac5be4Sperrin return (error); 41522ac5be4Sperrin } 41622ac5be4Sperrin 41722ac5be4Sperrin /* ARGSUSED */ 41822ac5be4Sperrin static int 41922ac5be4Sperrin zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 42022ac5be4Sperrin { 421be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 42222ac5be4Sperrin } 42322ac5be4Sperrin 42422ac5be4Sperrin /* 42522ac5be4Sperrin * Callback vectors for replaying records. 426b77b9231SDan McDonald * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 42722ac5be4Sperrin */ 42822ac5be4Sperrin zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 42922ac5be4Sperrin zvol_replay_err, /* 0 no such transaction type */ 43022ac5be4Sperrin zvol_replay_err, /* TX_CREATE */ 43122ac5be4Sperrin zvol_replay_err, /* TX_MKDIR */ 43222ac5be4Sperrin zvol_replay_err, /* TX_MKXATTR */ 43322ac5be4Sperrin zvol_replay_err, /* TX_SYMLINK */ 43422ac5be4Sperrin zvol_replay_err, /* TX_REMOVE */ 43522ac5be4Sperrin zvol_replay_err, /* TX_RMDIR */ 43622ac5be4Sperrin zvol_replay_err, /* TX_LINK */ 43722ac5be4Sperrin zvol_replay_err, /* TX_RENAME */ 43822ac5be4Sperrin zvol_replay_write, /* TX_WRITE */ 439b77b9231SDan McDonald zvol_replay_truncate, /* TX_TRUNCATE */ 44022ac5be4Sperrin zvol_replay_err, /* TX_SETATTR */ 44122ac5be4Sperrin zvol_replay_err, /* TX_ACL */ 442975c32a0SNeil Perrin zvol_replay_err, /* TX_CREATE_ACL */ 443975c32a0SNeil Perrin zvol_replay_err, /* TX_CREATE_ATTR */ 444975c32a0SNeil Perrin zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 445975c32a0SNeil Perrin zvol_replay_err, /* TX_MKDIR_ACL */ 446975c32a0SNeil Perrin zvol_replay_err, /* TX_MKDIR_ATTR */ 447975c32a0SNeil Perrin zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 448975c32a0SNeil Perrin zvol_replay_err, /* TX_WRITE2 */ 44922ac5be4Sperrin }; 45022ac5be4Sperrin 451681d9761SEric Taylor int 452681d9761SEric Taylor zvol_name2minor(const char *name, minor_t *minor) 453681d9761SEric Taylor { 454681d9761SEric Taylor zvol_state_t *zv; 455681d9761SEric Taylor 456c99e4bdcSChris Kirby mutex_enter(&zfsdev_state_lock); 457681d9761SEric Taylor zv = zvol_minor_lookup(name); 458681d9761SEric Taylor if (minor && zv) 459681d9761SEric Taylor *minor = zv->zv_minor; 460c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 461681d9761SEric Taylor return (zv ? 0 : -1); 462681d9761SEric Taylor } 463681d9761SEric Taylor 464e7cbe64fSgw /* 465e7cbe64fSgw * Create a minor node (plus a whole lot more) for the specified volume. 466fa9e4066Sahrens */ 467fa9e4066Sahrens int 468681d9761SEric Taylor zvol_create_minor(const char *name) 469fa9e4066Sahrens { 470c99e4bdcSChris Kirby zfs_soft_state_t *zs; 471fa9e4066Sahrens zvol_state_t *zv; 472fa9e4066Sahrens objset_t *os; 47367bd71c6Sperrin dmu_object_info_t doi; 474fa9e4066Sahrens minor_t minor = 0; 475fa9e4066Sahrens char chrbuf[30], blkbuf[30]; 476fa9e4066Sahrens int error; 477fa9e4066Sahrens 478c99e4bdcSChris Kirby mutex_enter(&zfsdev_state_lock); 479fa9e4066Sahrens 4801195e687SMark J Musante if (zvol_minor_lookup(name) != NULL) { 481c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 482be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 483fa9e4066Sahrens } 484fa9e4066Sahrens 485503ad85cSMatthew Ahrens /* lie and say we're read-only */ 4866e0cbcaaSMatthew Ahrens error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 487fa9e4066Sahrens 488fa9e4066Sahrens if (error) { 489c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 490fa9e4066Sahrens return (error); 491fa9e4066Sahrens } 492fa9e4066Sahrens 493c99e4bdcSChris Kirby if ((minor = zfsdev_minor_alloc()) == 0) { 4946e0cbcaaSMatthew Ahrens dmu_objset_disown(os, FTAG); 495c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 496be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 497fa9e4066Sahrens } 498fa9e4066Sahrens 499c99e4bdcSChris Kirby if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 5006e0cbcaaSMatthew Ahrens dmu_objset_disown(os, FTAG); 501c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 502be6fd75aSMatthew Ahrens return (SET_ERROR(EAGAIN)); 503fa9e4066Sahrens } 504e9dbad6fSeschrock (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 505e9dbad6fSeschrock (char *)name); 506fa9e4066Sahrens 507681d9761SEric Taylor (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 508fa9e4066Sahrens 509fa9e4066Sahrens if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 510fa9e4066Sahrens minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 511c99e4bdcSChris Kirby ddi_soft_state_free(zfsdev_state, minor); 5126e0cbcaaSMatthew Ahrens dmu_objset_disown(os, FTAG); 513c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 514be6fd75aSMatthew Ahrens return (SET_ERROR(EAGAIN)); 515fa9e4066Sahrens } 516fa9e4066Sahrens 517681d9761SEric Taylor (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 518fa9e4066Sahrens 519fa9e4066Sahrens if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 520fa9e4066Sahrens minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 521fa9e4066Sahrens ddi_remove_minor_node(zfs_dip, chrbuf); 522c99e4bdcSChris Kirby ddi_soft_state_free(zfsdev_state, minor); 5236e0cbcaaSMatthew Ahrens dmu_objset_disown(os, FTAG); 524c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 525be6fd75aSMatthew Ahrens return (SET_ERROR(EAGAIN)); 526fa9e4066Sahrens } 527fa9e4066Sahrens 528c99e4bdcSChris Kirby zs = ddi_get_soft_state(zfsdev_state, minor); 529c99e4bdcSChris Kirby zs->zss_type = ZSST_ZVOL; 530c99e4bdcSChris Kirby zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 531681d9761SEric Taylor (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 532fa9e4066Sahrens zv->zv_min_bs = DEV_BSHIFT; 533fa9e4066Sahrens zv->zv_minor = minor; 534fa9e4066Sahrens zv->zv_objset = os; 535f9af39baSGeorge Wilson if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 536681d9761SEric Taylor zv->zv_flags |= ZVOL_RDONLY; 537c2e6a7d6Sperrin mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 538c2e6a7d6Sperrin avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 539c2e6a7d6Sperrin sizeof (rl_t), offsetof(rl_t, r_node)); 54088b7b0f2SMatthew Ahrens list_create(&zv->zv_extents, sizeof (zvol_extent_t), 54188b7b0f2SMatthew Ahrens offsetof(zvol_extent_t, ze_node)); 54267bd71c6Sperrin /* get and cache the blocksize */ 54367bd71c6Sperrin error = dmu_object_info(os, ZVOL_OBJ, &doi); 54467bd71c6Sperrin ASSERT(error == 0); 54567bd71c6Sperrin zv->zv_volblocksize = doi.doi_data_block_size; 54622ac5be4Sperrin 547f9af39baSGeorge Wilson if (spa_writeable(dmu_objset_spa(os))) { 548f9af39baSGeorge Wilson if (zil_replay_disable) 549f9af39baSGeorge Wilson zil_destroy(dmu_objset_zil(os), B_FALSE); 550f9af39baSGeorge Wilson else 551f9af39baSGeorge Wilson zil_replay(os, zv, zvol_replay_vector); 552f9af39baSGeorge Wilson } 5536e0cbcaaSMatthew Ahrens dmu_objset_disown(os, FTAG); 554681d9761SEric Taylor zv->zv_objset = NULL; 555fa9e4066Sahrens 556fa9e4066Sahrens zvol_minors++; 557fa9e4066Sahrens 558c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 559fa9e4066Sahrens 560fa9e4066Sahrens return (0); 561fa9e4066Sahrens } 562fa9e4066Sahrens 563fa9e4066Sahrens /* 564fa9e4066Sahrens * Remove minor node for the specified volume. 565fa9e4066Sahrens */ 566681d9761SEric Taylor static int 567681d9761SEric Taylor zvol_remove_zv(zvol_state_t *zv) 568681d9761SEric Taylor { 569681d9761SEric Taylor char nmbuf[20]; 570c99e4bdcSChris Kirby minor_t minor = zv->zv_minor; 571681d9761SEric Taylor 572c99e4bdcSChris Kirby ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 573681d9761SEric Taylor if (zv->zv_total_opens != 0) 574be6fd75aSMatthew Ahrens return (SET_ERROR(EBUSY)); 575681d9761SEric Taylor 576c99e4bdcSChris Kirby (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 577681d9761SEric Taylor ddi_remove_minor_node(zfs_dip, nmbuf); 578681d9761SEric Taylor 579c99e4bdcSChris Kirby (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor); 580681d9761SEric Taylor ddi_remove_minor_node(zfs_dip, nmbuf); 581681d9761SEric Taylor 582681d9761SEric Taylor avl_destroy(&zv->zv_znode.z_range_avl); 583681d9761SEric Taylor mutex_destroy(&zv->zv_znode.z_range_lock); 584681d9761SEric Taylor 585c99e4bdcSChris Kirby kmem_free(zv, sizeof (zvol_state_t)); 586c99e4bdcSChris Kirby 587c99e4bdcSChris Kirby ddi_soft_state_free(zfsdev_state, minor); 588681d9761SEric Taylor 589681d9761SEric Taylor zvol_minors--; 590681d9761SEric Taylor return (0); 591681d9761SEric Taylor } 592681d9761SEric Taylor 593fa9e4066Sahrens int 594e9dbad6fSeschrock zvol_remove_minor(const char *name) 595fa9e4066Sahrens { 596fa9e4066Sahrens zvol_state_t *zv; 597681d9761SEric Taylor int rc; 598fa9e4066Sahrens 599c99e4bdcSChris Kirby mutex_enter(&zfsdev_state_lock); 600e9dbad6fSeschrock if ((zv = zvol_minor_lookup(name)) == NULL) { 601c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 602be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 603fa9e4066Sahrens } 604681d9761SEric Taylor rc = zvol_remove_zv(zv); 605c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 606681d9761SEric Taylor return (rc); 607681d9761SEric Taylor } 608fa9e4066Sahrens 609681d9761SEric Taylor int 610681d9761SEric Taylor zvol_first_open(zvol_state_t *zv) 611681d9761SEric Taylor { 612681d9761SEric Taylor objset_t *os; 613681d9761SEric Taylor uint64_t volsize; 614681d9761SEric Taylor int error; 615681d9761SEric Taylor uint64_t readonly; 616fa9e4066Sahrens 617681d9761SEric Taylor /* lie and say we're read-only */ 618681d9761SEric Taylor error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 619681d9761SEric Taylor zvol_tag, &os); 620681d9761SEric Taylor if (error) 621681d9761SEric Taylor return (error); 622fa9e4066Sahrens 623c61ea566SGeorge Wilson zv->zv_objset = os; 624681d9761SEric Taylor error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 625681d9761SEric Taylor if (error) { 626681d9761SEric Taylor ASSERT(error == 0); 627681d9761SEric Taylor dmu_objset_disown(os, zvol_tag); 628681d9761SEric Taylor return (error); 629681d9761SEric Taylor } 630c61ea566SGeorge Wilson 63194d1a210STim Haley error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 63294d1a210STim Haley if (error) { 63394d1a210STim Haley dmu_objset_disown(os, zvol_tag); 63494d1a210STim Haley return (error); 63594d1a210STim Haley } 636c61ea566SGeorge Wilson 637c61ea566SGeorge Wilson zvol_size_changed(zv, volsize); 638681d9761SEric Taylor zv->zv_zilog = zil_open(os, zvol_get_data); 639fa9e4066Sahrens 640681d9761SEric Taylor VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 641681d9761SEric Taylor NULL) == 0); 642f9af39baSGeorge Wilson if (readonly || dmu_objset_is_snapshot(os) || 643f9af39baSGeorge Wilson !spa_writeable(dmu_objset_spa(os))) 644681d9761SEric Taylor zv->zv_flags |= ZVOL_RDONLY; 645681d9761SEric Taylor else 646681d9761SEric Taylor zv->zv_flags &= ~ZVOL_RDONLY; 647681d9761SEric Taylor return (error); 648681d9761SEric Taylor } 649fa9e4066Sahrens 650681d9761SEric Taylor void 651681d9761SEric Taylor zvol_last_close(zvol_state_t *zv) 652681d9761SEric Taylor { 65322ac5be4Sperrin zil_close(zv->zv_zilog); 65422ac5be4Sperrin zv->zv_zilog = NULL; 6552e2c1355SMatthew Ahrens 65694d1a210STim Haley dmu_buf_rele(zv->zv_dbuf, zvol_tag); 65794d1a210STim Haley zv->zv_dbuf = NULL; 6582e2c1355SMatthew Ahrens 6592e2c1355SMatthew Ahrens /* 6602e2c1355SMatthew Ahrens * Evict cached data 6612e2c1355SMatthew Ahrens */ 6622e2c1355SMatthew Ahrens if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 6632e2c1355SMatthew Ahrens !(zv->zv_flags & ZVOL_RDONLY)) 6642e2c1355SMatthew Ahrens txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 6653b2aab18SMatthew Ahrens dmu_objset_evict_dbufs(zv->zv_objset); 6662e2c1355SMatthew Ahrens 667503ad85cSMatthew Ahrens dmu_objset_disown(zv->zv_objset, zvol_tag); 668fa9e4066Sahrens zv->zv_objset = NULL; 669fa9e4066Sahrens } 670fa9e4066Sahrens 671e7cbe64fSgw int 672e7cbe64fSgw zvol_prealloc(zvol_state_t *zv) 673e7cbe64fSgw { 674e7cbe64fSgw objset_t *os = zv->zv_objset; 675e7cbe64fSgw dmu_tx_t *tx; 676e7cbe64fSgw uint64_t refd, avail, usedobjs, availobjs; 677e7cbe64fSgw uint64_t resid = zv->zv_volsize; 678e7cbe64fSgw uint64_t off = 0; 679e7cbe64fSgw 680e7cbe64fSgw /* Check the space usage before attempting to allocate the space */ 681e7cbe64fSgw dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 682e7cbe64fSgw if (avail < zv->zv_volsize) 683be6fd75aSMatthew Ahrens return (SET_ERROR(ENOSPC)); 684e7cbe64fSgw 685e7cbe64fSgw /* Free old extents if they exist */ 686e7cbe64fSgw zvol_free_extents(zv); 687e7cbe64fSgw 688e7cbe64fSgw while (resid != 0) { 689e7cbe64fSgw int error; 690e7cbe64fSgw uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 691e7cbe64fSgw 692e7cbe64fSgw tx = dmu_tx_create(os); 693e7cbe64fSgw dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 694e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 695e7cbe64fSgw if (error) { 696e7cbe64fSgw dmu_tx_abort(tx); 697cdb0ab79Smaybee (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 698e7cbe64fSgw return (error); 699e7cbe64fSgw } 70082c9918fSTim Haley dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 701e7cbe64fSgw dmu_tx_commit(tx); 702e7cbe64fSgw off += bytes; 703e7cbe64fSgw resid -= bytes; 704e7cbe64fSgw } 705e7cbe64fSgw txg_wait_synced(dmu_objset_pool(os), 0); 706e7cbe64fSgw 707e7cbe64fSgw return (0); 708e7cbe64fSgw } 709e7cbe64fSgw 7103b2aab18SMatthew Ahrens static int 711681d9761SEric Taylor zvol_update_volsize(objset_t *os, uint64_t volsize) 712e7cbe64fSgw { 713e7cbe64fSgw dmu_tx_t *tx; 714e7cbe64fSgw int error; 715e7cbe64fSgw 716c99e4bdcSChris Kirby ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 717e7cbe64fSgw 718681d9761SEric Taylor tx = dmu_tx_create(os); 719e7cbe64fSgw dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 720e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 721e7cbe64fSgw if (error) { 722e7cbe64fSgw dmu_tx_abort(tx); 723e7cbe64fSgw return (error); 724e7cbe64fSgw } 725e7cbe64fSgw 726681d9761SEric Taylor error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 727e7cbe64fSgw &volsize, tx); 728e7cbe64fSgw dmu_tx_commit(tx); 729e7cbe64fSgw 730e7cbe64fSgw if (error == 0) 731681d9761SEric Taylor error = dmu_free_long_range(os, 732cdb0ab79Smaybee ZVOL_OBJ, volsize, DMU_OBJECT_END); 733681d9761SEric Taylor return (error); 734681d9761SEric Taylor } 735e7cbe64fSgw 736681d9761SEric Taylor void 737681d9761SEric Taylor zvol_remove_minors(const char *name) 738681d9761SEric Taylor { 739681d9761SEric Taylor zvol_state_t *zv; 740681d9761SEric Taylor char *namebuf; 741681d9761SEric Taylor minor_t minor; 742681d9761SEric Taylor 743681d9761SEric Taylor namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP); 744681d9761SEric Taylor (void) strncpy(namebuf, name, strlen(name)); 745681d9761SEric Taylor (void) strcat(namebuf, "/"); 746c99e4bdcSChris Kirby mutex_enter(&zfsdev_state_lock); 747c99e4bdcSChris Kirby for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 748681d9761SEric Taylor 749c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 750681d9761SEric Taylor if (zv == NULL) 751681d9761SEric Taylor continue; 752681d9761SEric Taylor if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0) 753681d9761SEric Taylor (void) zvol_remove_zv(zv); 754e7cbe64fSgw } 755681d9761SEric Taylor kmem_free(namebuf, strlen(name) + 2); 756681d9761SEric Taylor 757c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 758e7cbe64fSgw } 759e7cbe64fSgw 760c61ea566SGeorge Wilson static int 7613b2aab18SMatthew Ahrens zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize) 762fa9e4066Sahrens { 763e7cbe64fSgw uint64_t old_volsize = 0ULL; 7643b2aab18SMatthew Ahrens int error = 0; 765fa9e4066Sahrens 766c61ea566SGeorge Wilson ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 767c61ea566SGeorge Wilson 768e7cbe64fSgw /* 769e7cbe64fSgw * Reinitialize the dump area to the new size. If we 770681d9761SEric Taylor * failed to resize the dump area then restore it back to 771c61ea566SGeorge Wilson * its original size. We must set the new volsize prior 772c61ea566SGeorge Wilson * to calling dumpvp_resize() to ensure that the devices' 773c61ea566SGeorge Wilson * size(9P) is not visible by the dump subsystem. 774e7cbe64fSgw */ 7753b2aab18SMatthew Ahrens old_volsize = zv->zv_volsize; 7763b2aab18SMatthew Ahrens zvol_size_changed(zv, volsize); 7773b2aab18SMatthew Ahrens 7783b2aab18SMatthew Ahrens if (zv->zv_flags & ZVOL_DUMPIFIED) { 7793b2aab18SMatthew Ahrens if ((error = zvol_dumpify(zv)) != 0 || 7803b2aab18SMatthew Ahrens (error = dumpvp_resize()) != 0) { 7813b2aab18SMatthew Ahrens int dumpify_error; 7823b2aab18SMatthew Ahrens 7833b2aab18SMatthew Ahrens (void) zvol_update_volsize(zv->zv_objset, old_volsize); 7843b2aab18SMatthew Ahrens zvol_size_changed(zv, old_volsize); 7853b2aab18SMatthew Ahrens dumpify_error = zvol_dumpify(zv); 7863b2aab18SMatthew Ahrens error = dumpify_error ? dumpify_error : error; 787681d9761SEric Taylor } 788fa9e4066Sahrens } 789fa9e4066Sahrens 790573ca77eSGeorge Wilson /* 791573ca77eSGeorge Wilson * Generate a LUN expansion event. 792573ca77eSGeorge Wilson */ 7933b2aab18SMatthew Ahrens if (error == 0) { 794573ca77eSGeorge Wilson sysevent_id_t eid; 795573ca77eSGeorge Wilson nvlist_t *attr; 796573ca77eSGeorge Wilson char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 797573ca77eSGeorge Wilson 798681d9761SEric Taylor (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 799573ca77eSGeorge Wilson zv->zv_minor); 800573ca77eSGeorge Wilson 801573ca77eSGeorge Wilson VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 802573ca77eSGeorge Wilson VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 803573ca77eSGeorge Wilson 804573ca77eSGeorge Wilson (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 805573ca77eSGeorge Wilson ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 806573ca77eSGeorge Wilson 807573ca77eSGeorge Wilson nvlist_free(attr); 808573ca77eSGeorge Wilson kmem_free(physpath, MAXPATHLEN); 809573ca77eSGeorge Wilson } 810c61ea566SGeorge Wilson return (error); 811c61ea566SGeorge Wilson } 812573ca77eSGeorge Wilson 813c61ea566SGeorge Wilson int 814c61ea566SGeorge Wilson zvol_set_volsize(const char *name, uint64_t volsize) 815c61ea566SGeorge Wilson { 816c61ea566SGeorge Wilson zvol_state_t *zv = NULL; 817c61ea566SGeorge Wilson objset_t *os; 818c61ea566SGeorge Wilson int error; 819c61ea566SGeorge Wilson dmu_object_info_t doi; 820c61ea566SGeorge Wilson uint64_t readonly; 8213b2aab18SMatthew Ahrens boolean_t owned = B_FALSE; 8223b2aab18SMatthew Ahrens 8233b2aab18SMatthew Ahrens error = dsl_prop_get_integer(name, 8243b2aab18SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); 8253b2aab18SMatthew Ahrens if (error != 0) 8263b2aab18SMatthew Ahrens return (error); 8273b2aab18SMatthew Ahrens if (readonly) 828be6fd75aSMatthew Ahrens return (SET_ERROR(EROFS)); 829c61ea566SGeorge Wilson 830c61ea566SGeorge Wilson mutex_enter(&zfsdev_state_lock); 831c61ea566SGeorge Wilson zv = zvol_minor_lookup(name); 8323b2aab18SMatthew Ahrens 8333b2aab18SMatthew Ahrens if (zv == NULL || zv->zv_objset == NULL) { 8343b2aab18SMatthew Ahrens if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, 8353b2aab18SMatthew Ahrens FTAG, &os)) != 0) { 8363b2aab18SMatthew Ahrens mutex_exit(&zfsdev_state_lock); 8373b2aab18SMatthew Ahrens return (error); 8383b2aab18SMatthew Ahrens } 8393b2aab18SMatthew Ahrens owned = B_TRUE; 8403b2aab18SMatthew Ahrens if (zv != NULL) 8413b2aab18SMatthew Ahrens zv->zv_objset = os; 8423b2aab18SMatthew Ahrens } else { 8433b2aab18SMatthew Ahrens os = zv->zv_objset; 844c61ea566SGeorge Wilson } 845c61ea566SGeorge Wilson 846c61ea566SGeorge Wilson if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 8473b2aab18SMatthew Ahrens (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0) 848c61ea566SGeorge Wilson goto out; 849c61ea566SGeorge Wilson 8503b2aab18SMatthew Ahrens error = zvol_update_volsize(os, volsize); 851c61ea566SGeorge Wilson 8523b2aab18SMatthew Ahrens if (error == 0 && zv != NULL) 8533b2aab18SMatthew Ahrens error = zvol_update_live_volsize(zv, volsize); 854bb0ade09Sahrens out: 8553b2aab18SMatthew Ahrens if (owned) { 8563b2aab18SMatthew Ahrens dmu_objset_disown(os, FTAG); 8573b2aab18SMatthew Ahrens if (zv != NULL) 8583b2aab18SMatthew Ahrens zv->zv_objset = NULL; 8593b2aab18SMatthew Ahrens } 860c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 861fa9e4066Sahrens return (error); 862fa9e4066Sahrens } 863fa9e4066Sahrens 864fa9e4066Sahrens /*ARGSUSED*/ 865fa9e4066Sahrens int 866fa9e4066Sahrens zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 867fa9e4066Sahrens { 868fa9e4066Sahrens zvol_state_t *zv; 869681d9761SEric Taylor int err = 0; 870fa9e4066Sahrens 871c99e4bdcSChris Kirby mutex_enter(&zfsdev_state_lock); 872fa9e4066Sahrens 873c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL); 874fa9e4066Sahrens if (zv == NULL) { 875c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 876be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 877fa9e4066Sahrens } 878fa9e4066Sahrens 879681d9761SEric Taylor if (zv->zv_total_opens == 0) 880681d9761SEric Taylor err = zvol_first_open(zv); 881681d9761SEric Taylor if (err) { 882c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 883681d9761SEric Taylor return (err); 884681d9761SEric Taylor } 885681d9761SEric Taylor if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 886be6fd75aSMatthew Ahrens err = SET_ERROR(EROFS); 887681d9761SEric Taylor goto out; 888fa9e4066Sahrens } 889c7f714e2SEric Taylor if (zv->zv_flags & ZVOL_EXCL) { 890be6fd75aSMatthew Ahrens err = SET_ERROR(EBUSY); 891681d9761SEric Taylor goto out; 892c7f714e2SEric Taylor } 893c7f714e2SEric Taylor if (flag & FEXCL) { 894c7f714e2SEric Taylor if (zv->zv_total_opens != 0) { 895be6fd75aSMatthew Ahrens err = SET_ERROR(EBUSY); 896681d9761SEric Taylor goto out; 897c7f714e2SEric Taylor } 898c7f714e2SEric Taylor zv->zv_flags |= ZVOL_EXCL; 899c7f714e2SEric Taylor } 900fa9e4066Sahrens 901fa9e4066Sahrens if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 902fa9e4066Sahrens zv->zv_open_count[otyp]++; 903fa9e4066Sahrens zv->zv_total_opens++; 904fa9e4066Sahrens } 905c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 906fa9e4066Sahrens 907681d9761SEric Taylor return (err); 908681d9761SEric Taylor out: 909681d9761SEric Taylor if (zv->zv_total_opens == 0) 910681d9761SEric Taylor zvol_last_close(zv); 911c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 912681d9761SEric Taylor return (err); 913fa9e4066Sahrens } 914fa9e4066Sahrens 915fa9e4066Sahrens /*ARGSUSED*/ 916fa9e4066Sahrens int 917fa9e4066Sahrens zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 918fa9e4066Sahrens { 919fa9e4066Sahrens minor_t minor = getminor(dev); 920fa9e4066Sahrens zvol_state_t *zv; 921681d9761SEric Taylor int error = 0; 922fa9e4066Sahrens 923c99e4bdcSChris Kirby mutex_enter(&zfsdev_state_lock); 924fa9e4066Sahrens 925c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 926fa9e4066Sahrens if (zv == NULL) { 927c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 928be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 929fa9e4066Sahrens } 930fa9e4066Sahrens 931c7f714e2SEric Taylor if (zv->zv_flags & ZVOL_EXCL) { 932c7f714e2SEric Taylor ASSERT(zv->zv_total_opens == 1); 933c7f714e2SEric Taylor zv->zv_flags &= ~ZVOL_EXCL; 934fa9e4066Sahrens } 935fa9e4066Sahrens 936fa9e4066Sahrens /* 937fa9e4066Sahrens * If the open count is zero, this is a spurious close. 938fa9e4066Sahrens * That indicates a bug in the kernel / DDI framework. 939fa9e4066Sahrens */ 940fa9e4066Sahrens ASSERT(zv->zv_open_count[otyp] != 0); 941fa9e4066Sahrens ASSERT(zv->zv_total_opens != 0); 942fa9e4066Sahrens 943fa9e4066Sahrens /* 944fa9e4066Sahrens * You may get multiple opens, but only one close. 945fa9e4066Sahrens */ 946fa9e4066Sahrens zv->zv_open_count[otyp]--; 947fa9e4066Sahrens zv->zv_total_opens--; 948fa9e4066Sahrens 949681d9761SEric Taylor if (zv->zv_total_opens == 0) 950681d9761SEric Taylor zvol_last_close(zv); 951fa9e4066Sahrens 952c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 953681d9761SEric Taylor return (error); 954fa9e4066Sahrens } 955fa9e4066Sahrens 956feb08c6bSbillm static void 957b24ab676SJeff Bonwick zvol_get_done(zgd_t *zgd, int error) 95867bd71c6Sperrin { 959b24ab676SJeff Bonwick if (zgd->zgd_db) 960b24ab676SJeff Bonwick dmu_buf_rele(zgd->zgd_db, zgd); 961b24ab676SJeff Bonwick 962b24ab676SJeff Bonwick zfs_range_unlock(zgd->zgd_rl); 963b24ab676SJeff Bonwick 964b24ab676SJeff Bonwick if (error == 0 && zgd->zgd_bp) 965b24ab676SJeff Bonwick zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 96667bd71c6Sperrin 96767bd71c6Sperrin kmem_free(zgd, sizeof (zgd_t)); 96867bd71c6Sperrin } 96967bd71c6Sperrin 97067bd71c6Sperrin /* 97167bd71c6Sperrin * Get data to generate a TX_WRITE intent log record. 97267bd71c6Sperrin */ 973feb08c6bSbillm static int 97467bd71c6Sperrin zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 97567bd71c6Sperrin { 97667bd71c6Sperrin zvol_state_t *zv = arg; 97767bd71c6Sperrin objset_t *os = zv->zv_objset; 978b24ab676SJeff Bonwick uint64_t object = ZVOL_OBJ; 979b24ab676SJeff Bonwick uint64_t offset = lr->lr_offset; 980b24ab676SJeff Bonwick uint64_t size = lr->lr_length; /* length of user data */ 981b24ab676SJeff Bonwick blkptr_t *bp = &lr->lr_blkptr; 98267bd71c6Sperrin dmu_buf_t *db; 98367bd71c6Sperrin zgd_t *zgd; 98467bd71c6Sperrin int error; 98567bd71c6Sperrin 986b24ab676SJeff Bonwick ASSERT(zio != NULL); 987b24ab676SJeff Bonwick ASSERT(size != 0); 988b24ab676SJeff Bonwick 989b24ab676SJeff Bonwick zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 990b24ab676SJeff Bonwick zgd->zgd_zilog = zv->zv_zilog; 991b24ab676SJeff Bonwick zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 992feb08c6bSbillm 993c2e6a7d6Sperrin /* 994c2e6a7d6Sperrin * Write records come in two flavors: immediate and indirect. 995c2e6a7d6Sperrin * For small writes it's cheaper to store the data with the 996c2e6a7d6Sperrin * log record (immediate); for large writes it's cheaper to 997c2e6a7d6Sperrin * sync the data and get a pointer to it (indirect) so that 998c2e6a7d6Sperrin * we don't have to write the data twice. 999c2e6a7d6Sperrin */ 1000b24ab676SJeff Bonwick if (buf != NULL) { /* immediate write */ 1001b24ab676SJeff Bonwick error = dmu_read(os, object, offset, size, buf, 1002b24ab676SJeff Bonwick DMU_READ_NO_PREFETCH); 1003b24ab676SJeff Bonwick } else { 1004b24ab676SJeff Bonwick size = zv->zv_volblocksize; 1005b24ab676SJeff Bonwick offset = P2ALIGN(offset, size); 100647cb52daSJeff Bonwick error = dmu_buf_hold(os, object, offset, zgd, &db, 100747cb52daSJeff Bonwick DMU_READ_NO_PREFETCH); 1008b24ab676SJeff Bonwick if (error == 0) { 100980901aeaSGeorge Wilson blkptr_t *obp = dmu_buf_get_blkptr(db); 101080901aeaSGeorge Wilson if (obp) { 101180901aeaSGeorge Wilson ASSERT(BP_IS_HOLE(bp)); 101280901aeaSGeorge Wilson *bp = *obp; 101380901aeaSGeorge Wilson } 101480901aeaSGeorge Wilson 1015b24ab676SJeff Bonwick zgd->zgd_db = db; 1016b24ab676SJeff Bonwick zgd->zgd_bp = bp; 101767bd71c6Sperrin 1018b24ab676SJeff Bonwick ASSERT(db->db_offset == offset); 1019b24ab676SJeff Bonwick ASSERT(db->db_size == size); 102067bd71c6Sperrin 1021b24ab676SJeff Bonwick error = dmu_sync(zio, lr->lr_common.lrc_txg, 1022b24ab676SJeff Bonwick zvol_get_done, zgd); 1023975c32a0SNeil Perrin 1024b24ab676SJeff Bonwick if (error == 0) 1025b24ab676SJeff Bonwick return (0); 1026b24ab676SJeff Bonwick } 1027975c32a0SNeil Perrin } 1028975c32a0SNeil Perrin 1029b24ab676SJeff Bonwick zvol_get_done(zgd, error); 1030b24ab676SJeff Bonwick 103167bd71c6Sperrin return (error); 103267bd71c6Sperrin } 103367bd71c6Sperrin 1034a24e15ceSperrin /* 1035a24e15ceSperrin * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 103622ac5be4Sperrin * 103722ac5be4Sperrin * We store data in the log buffers if it's small enough. 103867bd71c6Sperrin * Otherwise we will later flush the data out via dmu_sync(). 103922ac5be4Sperrin */ 104067bd71c6Sperrin ssize_t zvol_immediate_write_sz = 32768; 104122ac5be4Sperrin 1042feb08c6bSbillm static void 1043510b6c0eSNeil Perrin zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1044510b6c0eSNeil Perrin boolean_t sync) 104522ac5be4Sperrin { 1046feb08c6bSbillm uint32_t blocksize = zv->zv_volblocksize; 10471209a471SNeil Perrin zilog_t *zilog = zv->zv_zilog; 1048510b6c0eSNeil Perrin boolean_t slogging; 1049e09fa4daSNeil Perrin ssize_t immediate_write_sz; 1050510b6c0eSNeil Perrin 1051b24ab676SJeff Bonwick if (zil_replaying(zilog, tx)) 10521209a471SNeil Perrin return; 10531209a471SNeil Perrin 1054e09fa4daSNeil Perrin immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1055e09fa4daSNeil Perrin ? 0 : zvol_immediate_write_sz; 1056e09fa4daSNeil Perrin 1057e09fa4daSNeil Perrin slogging = spa_has_slogs(zilog->zl_spa) && 1058e09fa4daSNeil Perrin (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1059feb08c6bSbillm 1060510b6c0eSNeil Perrin while (resid) { 1061510b6c0eSNeil Perrin itx_t *itx; 1062510b6c0eSNeil Perrin lr_write_t *lr; 1063510b6c0eSNeil Perrin ssize_t len; 1064510b6c0eSNeil Perrin itx_wr_state_t write_state; 1065510b6c0eSNeil Perrin 1066510b6c0eSNeil Perrin /* 1067510b6c0eSNeil Perrin * Unlike zfs_log_write() we can be called with 1068510b6c0eSNeil Perrin * upto DMU_MAX_ACCESS/2 (5MB) writes. 1069510b6c0eSNeil Perrin */ 1070e09fa4daSNeil Perrin if (blocksize > immediate_write_sz && !slogging && 1071510b6c0eSNeil Perrin resid >= blocksize && off % blocksize == 0) { 1072510b6c0eSNeil Perrin write_state = WR_INDIRECT; /* uses dmu_sync */ 1073510b6c0eSNeil Perrin len = blocksize; 1074510b6c0eSNeil Perrin } else if (sync) { 1075510b6c0eSNeil Perrin write_state = WR_COPIED; 1076510b6c0eSNeil Perrin len = MIN(ZIL_MAX_LOG_DATA, resid); 1077510b6c0eSNeil Perrin } else { 1078510b6c0eSNeil Perrin write_state = WR_NEED_COPY; 1079510b6c0eSNeil Perrin len = MIN(ZIL_MAX_LOG_DATA, resid); 1080510b6c0eSNeil Perrin } 1081510b6c0eSNeil Perrin 1082510b6c0eSNeil Perrin itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1083510b6c0eSNeil Perrin (write_state == WR_COPIED ? len : 0)); 1084feb08c6bSbillm lr = (lr_write_t *)&itx->itx_lr; 1085510b6c0eSNeil Perrin if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 10867bfdf011SNeil Perrin ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1087b24ab676SJeff Bonwick zil_itx_destroy(itx); 1088510b6c0eSNeil Perrin itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1089510b6c0eSNeil Perrin lr = (lr_write_t *)&itx->itx_lr; 1090510b6c0eSNeil Perrin write_state = WR_NEED_COPY; 1091510b6c0eSNeil Perrin } 1092510b6c0eSNeil Perrin 1093510b6c0eSNeil Perrin itx->itx_wr_state = write_state; 1094510b6c0eSNeil Perrin if (write_state == WR_NEED_COPY) 1095510b6c0eSNeil Perrin itx->itx_sod += len; 1096feb08c6bSbillm lr->lr_foid = ZVOL_OBJ; 1097feb08c6bSbillm lr->lr_offset = off; 1098510b6c0eSNeil Perrin lr->lr_length = len; 1099b24ab676SJeff Bonwick lr->lr_blkoff = 0; 1100feb08c6bSbillm BP_ZERO(&lr->lr_blkptr); 1101feb08c6bSbillm 1102510b6c0eSNeil Perrin itx->itx_private = zv; 1103510b6c0eSNeil Perrin itx->itx_sync = sync; 1104510b6c0eSNeil Perrin 11055002558fSNeil Perrin zil_itx_assign(zilog, itx, tx); 1106510b6c0eSNeil Perrin 1107510b6c0eSNeil Perrin off += len; 1108510b6c0eSNeil Perrin resid -= len; 110922ac5be4Sperrin } 111022ac5be4Sperrin } 111122ac5be4Sperrin 111288b7b0f2SMatthew Ahrens static int 1113810e43b2SBill Pijewski zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1114810e43b2SBill Pijewski uint64_t size, boolean_t doread, boolean_t isdump) 1115e7cbe64fSgw { 1116e7cbe64fSgw vdev_disk_t *dvd; 1117e7cbe64fSgw int c; 1118e7cbe64fSgw int numerrors = 0; 1119e7cbe64fSgw 1120810e43b2SBill Pijewski if (vd->vdev_ops == &vdev_mirror_ops || 1121810e43b2SBill Pijewski vd->vdev_ops == &vdev_replacing_ops || 1122810e43b2SBill Pijewski vd->vdev_ops == &vdev_spare_ops) { 1123810e43b2SBill Pijewski for (c = 0; c < vd->vdev_children; c++) { 1124810e43b2SBill Pijewski int err = zvol_dumpio_vdev(vd->vdev_child[c], 1125810e43b2SBill Pijewski addr, offset, origoffset, size, doread, isdump); 1126810e43b2SBill Pijewski if (err != 0) { 1127810e43b2SBill Pijewski numerrors++; 1128810e43b2SBill Pijewski } else if (doread) { 1129810e43b2SBill Pijewski break; 1130810e43b2SBill Pijewski } 1131e7cbe64fSgw } 1132e7cbe64fSgw } 1133e7cbe64fSgw 1134810e43b2SBill Pijewski if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1135e7cbe64fSgw return (numerrors < vd->vdev_children ? 0 : EIO); 1136e7cbe64fSgw 1137dc0bb255SEric Taylor if (doread && !vdev_readable(vd)) 1138be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 1139dc0bb255SEric Taylor else if (!doread && !vdev_writeable(vd)) 1140be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 1141e7cbe64fSgw 1142810e43b2SBill Pijewski if (vd->vdev_ops == &vdev_raidz_ops) { 1143810e43b2SBill Pijewski return (vdev_raidz_physio(vd, 1144810e43b2SBill Pijewski addr, size, offset, origoffset, doread, isdump)); 1145810e43b2SBill Pijewski } 1146810e43b2SBill Pijewski 1147e7cbe64fSgw offset += VDEV_LABEL_START_SIZE; 1148e7cbe64fSgw 1149e7cbe64fSgw if (ddi_in_panic() || isdump) { 115088b7b0f2SMatthew Ahrens ASSERT(!doread); 115188b7b0f2SMatthew Ahrens if (doread) 1152be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 1153810e43b2SBill Pijewski dvd = vd->vdev_tsd; 1154810e43b2SBill Pijewski ASSERT3P(dvd, !=, NULL); 1155e7cbe64fSgw return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1156e7cbe64fSgw lbtodb(size))); 1157e7cbe64fSgw } else { 1158810e43b2SBill Pijewski dvd = vd->vdev_tsd; 1159810e43b2SBill Pijewski ASSERT3P(dvd, !=, NULL); 1160810e43b2SBill Pijewski return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1161810e43b2SBill Pijewski offset, doread ? B_READ : B_WRITE)); 1162e7cbe64fSgw } 1163e7cbe64fSgw } 1164e7cbe64fSgw 116588b7b0f2SMatthew Ahrens static int 116688b7b0f2SMatthew Ahrens zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 116788b7b0f2SMatthew Ahrens boolean_t doread, boolean_t isdump) 1168e7cbe64fSgw { 1169e7cbe64fSgw vdev_t *vd; 1170e7cbe64fSgw int error; 117188b7b0f2SMatthew Ahrens zvol_extent_t *ze; 1172e7cbe64fSgw spa_t *spa = dmu_objset_spa(zv->zv_objset); 1173e7cbe64fSgw 117488b7b0f2SMatthew Ahrens /* Must be sector aligned, and not stradle a block boundary. */ 117588b7b0f2SMatthew Ahrens if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 117688b7b0f2SMatthew Ahrens P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1177be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 117888b7b0f2SMatthew Ahrens } 117988b7b0f2SMatthew Ahrens ASSERT(size <= zv->zv_volblocksize); 1180e7cbe64fSgw 118188b7b0f2SMatthew Ahrens /* Locate the extent this belongs to */ 118288b7b0f2SMatthew Ahrens ze = list_head(&zv->zv_extents); 118388b7b0f2SMatthew Ahrens while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 118488b7b0f2SMatthew Ahrens offset -= ze->ze_nblks * zv->zv_volblocksize; 118588b7b0f2SMatthew Ahrens ze = list_next(&zv->zv_extents, ze); 118688b7b0f2SMatthew Ahrens } 118724cc0e1cSGeorge Wilson 11883b2aab18SMatthew Ahrens if (ze == NULL) 1189be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 11903b2aab18SMatthew Ahrens 119124cc0e1cSGeorge Wilson if (!ddi_in_panic()) 119224cc0e1cSGeorge Wilson spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 119324cc0e1cSGeorge Wilson 119488b7b0f2SMatthew Ahrens vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 119588b7b0f2SMatthew Ahrens offset += DVA_GET_OFFSET(&ze->ze_dva); 1196810e43b2SBill Pijewski error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1197810e43b2SBill Pijewski size, doread, isdump); 119824cc0e1cSGeorge Wilson 119924cc0e1cSGeorge Wilson if (!ddi_in_panic()) 120024cc0e1cSGeorge Wilson spa_config_exit(spa, SCL_STATE, FTAG); 120124cc0e1cSGeorge Wilson 1202e7cbe64fSgw return (error); 1203e7cbe64fSgw } 1204e7cbe64fSgw 1205fa9e4066Sahrens int 1206fa9e4066Sahrens zvol_strategy(buf_t *bp) 1207fa9e4066Sahrens { 1208c99e4bdcSChris Kirby zfs_soft_state_t *zs = NULL; 1209c99e4bdcSChris Kirby zvol_state_t *zv; 1210fa9e4066Sahrens uint64_t off, volsize; 121188b7b0f2SMatthew Ahrens size_t resid; 1212fa9e4066Sahrens char *addr; 121322ac5be4Sperrin objset_t *os; 1214c2e6a7d6Sperrin rl_t *rl; 1215fa9e4066Sahrens int error = 0; 121688b7b0f2SMatthew Ahrens boolean_t doread = bp->b_flags & B_READ; 1217810e43b2SBill Pijewski boolean_t is_dumpified; 1218510b6c0eSNeil Perrin boolean_t sync; 1219fa9e4066Sahrens 1220c99e4bdcSChris Kirby if (getminor(bp->b_edev) == 0) { 1221be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 1222c99e4bdcSChris Kirby } else { 1223c99e4bdcSChris Kirby zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev)); 1224c99e4bdcSChris Kirby if (zs == NULL) 1225be6fd75aSMatthew Ahrens error = SET_ERROR(ENXIO); 1226c99e4bdcSChris Kirby else if (zs->zss_type != ZSST_ZVOL) 1227be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 1228fa9e4066Sahrens } 1229fa9e4066Sahrens 1230c99e4bdcSChris Kirby if (error) { 1231c99e4bdcSChris Kirby bioerror(bp, error); 1232fa9e4066Sahrens biodone(bp); 1233fa9e4066Sahrens return (0); 1234fa9e4066Sahrens } 1235fa9e4066Sahrens 1236c99e4bdcSChris Kirby zv = zs->zss_data; 1237c99e4bdcSChris Kirby 1238681d9761SEric Taylor if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) { 1239fa9e4066Sahrens bioerror(bp, EROFS); 1240fa9e4066Sahrens biodone(bp); 1241fa9e4066Sahrens return (0); 1242fa9e4066Sahrens } 1243fa9e4066Sahrens 1244fa9e4066Sahrens off = ldbtob(bp->b_blkno); 1245fa9e4066Sahrens volsize = zv->zv_volsize; 1246fa9e4066Sahrens 124722ac5be4Sperrin os = zv->zv_objset; 124822ac5be4Sperrin ASSERT(os != NULL); 1249fa9e4066Sahrens 1250fa9e4066Sahrens bp_mapin(bp); 1251fa9e4066Sahrens addr = bp->b_un.b_addr; 1252fa9e4066Sahrens resid = bp->b_bcount; 1253fa9e4066Sahrens 125488b7b0f2SMatthew Ahrens if (resid > 0 && (off < 0 || off >= volsize)) { 125588b7b0f2SMatthew Ahrens bioerror(bp, EIO); 125688b7b0f2SMatthew Ahrens biodone(bp); 125788b7b0f2SMatthew Ahrens return (0); 125888b7b0f2SMatthew Ahrens } 125973ec3d9cSgw 1260810e43b2SBill Pijewski is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 126155da60b9SMark J Musante sync = ((!(bp->b_flags & B_ASYNC) && 126255da60b9SMark J Musante !(zv->zv_flags & ZVOL_WCE)) || 126355da60b9SMark J Musante (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) && 1264810e43b2SBill Pijewski !doread && !is_dumpified; 1265510b6c0eSNeil Perrin 1266a24e15ceSperrin /* 1267a24e15ceSperrin * There must be no buffer changes when doing a dmu_sync() because 1268a24e15ceSperrin * we can't change the data whilst calculating the checksum. 1269a24e15ceSperrin */ 1270c2e6a7d6Sperrin rl = zfs_range_lock(&zv->zv_znode, off, resid, 127188b7b0f2SMatthew Ahrens doread ? RL_READER : RL_WRITER); 1272fa9e4066Sahrens 1273e7cbe64fSgw while (resid != 0 && off < volsize) { 127488b7b0f2SMatthew Ahrens size_t size = MIN(resid, zvol_maxphys); 1275810e43b2SBill Pijewski if (is_dumpified) { 1276e7cbe64fSgw size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 127788b7b0f2SMatthew Ahrens error = zvol_dumpio(zv, addr, off, size, 127888b7b0f2SMatthew Ahrens doread, B_FALSE); 127988b7b0f2SMatthew Ahrens } else if (doread) { 12807bfdf011SNeil Perrin error = dmu_read(os, ZVOL_OBJ, off, size, addr, 12817bfdf011SNeil Perrin DMU_READ_PREFETCH); 1282fa9e4066Sahrens } else { 128322ac5be4Sperrin dmu_tx_t *tx = dmu_tx_create(os); 1284fa9e4066Sahrens dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1285fa9e4066Sahrens error = dmu_tx_assign(tx, TXG_WAIT); 1286fa9e4066Sahrens if (error) { 1287fa9e4066Sahrens dmu_tx_abort(tx); 1288fa9e4066Sahrens } else { 128922ac5be4Sperrin dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1290510b6c0eSNeil Perrin zvol_log_write(zv, tx, off, size, sync); 1291fa9e4066Sahrens dmu_tx_commit(tx); 1292fa9e4066Sahrens } 1293fa9e4066Sahrens } 1294b87f3af3Sperrin if (error) { 1295b87f3af3Sperrin /* convert checksum errors into IO errors */ 1296b87f3af3Sperrin if (error == ECKSUM) 1297be6fd75aSMatthew Ahrens error = SET_ERROR(EIO); 1298fa9e4066Sahrens break; 1299b87f3af3Sperrin } 1300fa9e4066Sahrens off += size; 1301fa9e4066Sahrens addr += size; 1302fa9e4066Sahrens resid -= size; 1303fa9e4066Sahrens } 1304c2e6a7d6Sperrin zfs_range_unlock(rl); 1305fa9e4066Sahrens 1306fa9e4066Sahrens if ((bp->b_resid = resid) == bp->b_bcount) 1307fa9e4066Sahrens bioerror(bp, off > volsize ? EINVAL : error); 1308fa9e4066Sahrens 1309510b6c0eSNeil Perrin if (sync) 13105002558fSNeil Perrin zil_commit(zv->zv_zilog, ZVOL_OBJ); 1311feb08c6bSbillm biodone(bp); 131222ac5be4Sperrin 1313fa9e4066Sahrens return (0); 1314fa9e4066Sahrens } 1315fa9e4066Sahrens 131667bd71c6Sperrin /* 131767bd71c6Sperrin * Set the buffer count to the zvol maximum transfer. 131867bd71c6Sperrin * Using our own routine instead of the default minphys() 131967bd71c6Sperrin * means that for larger writes we write bigger buffers on X86 132067bd71c6Sperrin * (128K instead of 56K) and flush the disk write cache less often 132167bd71c6Sperrin * (every zvol_maxphys - currently 1MB) instead of minphys (currently 132267bd71c6Sperrin * 56K on X86 and 128K on sparc). 132367bd71c6Sperrin */ 132467bd71c6Sperrin void 132567bd71c6Sperrin zvol_minphys(struct buf *bp) 132667bd71c6Sperrin { 132767bd71c6Sperrin if (bp->b_bcount > zvol_maxphys) 132867bd71c6Sperrin bp->b_bcount = zvol_maxphys; 132967bd71c6Sperrin } 133067bd71c6Sperrin 1331e7cbe64fSgw int 1332e7cbe64fSgw zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1333e7cbe64fSgw { 1334e7cbe64fSgw minor_t minor = getminor(dev); 1335e7cbe64fSgw zvol_state_t *zv; 1336e7cbe64fSgw int error = 0; 1337e7cbe64fSgw uint64_t size; 1338e7cbe64fSgw uint64_t boff; 1339e7cbe64fSgw uint64_t resid; 1340e7cbe64fSgw 1341c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1342e7cbe64fSgw if (zv == NULL) 1343be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1344e7cbe64fSgw 13453b2aab18SMatthew Ahrens if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1346be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 13473b2aab18SMatthew Ahrens 1348e7cbe64fSgw boff = ldbtob(blkno); 1349e7cbe64fSgw resid = ldbtob(nblocks); 135088b7b0f2SMatthew Ahrens 135188b7b0f2SMatthew Ahrens VERIFY3U(boff + resid, <=, zv->zv_volsize); 135288b7b0f2SMatthew Ahrens 1353e7cbe64fSgw while (resid) { 1354e7cbe64fSgw size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 135588b7b0f2SMatthew Ahrens error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1356e7cbe64fSgw if (error) 1357e7cbe64fSgw break; 1358e7cbe64fSgw boff += size; 1359e7cbe64fSgw addr += size; 1360e7cbe64fSgw resid -= size; 1361e7cbe64fSgw } 1362e7cbe64fSgw 1363e7cbe64fSgw return (error); 1364e7cbe64fSgw } 1365e7cbe64fSgw 1366fa9e4066Sahrens /*ARGSUSED*/ 1367fa9e4066Sahrens int 1368feb08c6bSbillm zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1369fa9e4066Sahrens { 1370c7ca1008Sgw minor_t minor = getminor(dev); 1371c7ca1008Sgw zvol_state_t *zv; 137273ec3d9cSgw uint64_t volsize; 1373c2e6a7d6Sperrin rl_t *rl; 1374feb08c6bSbillm int error = 0; 1375fa9e4066Sahrens 1376c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1377c7ca1008Sgw if (zv == NULL) 1378be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1379c7ca1008Sgw 138073ec3d9cSgw volsize = zv->zv_volsize; 138173ec3d9cSgw if (uio->uio_resid > 0 && 138273ec3d9cSgw (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1383be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 138473ec3d9cSgw 138588b7b0f2SMatthew Ahrens if (zv->zv_flags & ZVOL_DUMPIFIED) { 138688b7b0f2SMatthew Ahrens error = physio(zvol_strategy, NULL, dev, B_READ, 138788b7b0f2SMatthew Ahrens zvol_minphys, uio); 138888b7b0f2SMatthew Ahrens return (error); 138988b7b0f2SMatthew Ahrens } 139088b7b0f2SMatthew Ahrens 1391c2e6a7d6Sperrin rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1392c2e6a7d6Sperrin RL_READER); 139373ec3d9cSgw while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1394feb08c6bSbillm uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1395fa9e4066Sahrens 139673ec3d9cSgw /* don't read past the end */ 139773ec3d9cSgw if (bytes > volsize - uio->uio_loffset) 139873ec3d9cSgw bytes = volsize - uio->uio_loffset; 139973ec3d9cSgw 1400feb08c6bSbillm error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1401b87f3af3Sperrin if (error) { 1402b87f3af3Sperrin /* convert checksum errors into IO errors */ 1403b87f3af3Sperrin if (error == ECKSUM) 1404be6fd75aSMatthew Ahrens error = SET_ERROR(EIO); 1405feb08c6bSbillm break; 1406b87f3af3Sperrin } 1407feb08c6bSbillm } 1408c2e6a7d6Sperrin zfs_range_unlock(rl); 1409feb08c6bSbillm return (error); 1410fa9e4066Sahrens } 1411fa9e4066Sahrens 1412fa9e4066Sahrens /*ARGSUSED*/ 1413fa9e4066Sahrens int 1414feb08c6bSbillm zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1415fa9e4066Sahrens { 1416c7ca1008Sgw minor_t minor = getminor(dev); 1417c7ca1008Sgw zvol_state_t *zv; 141873ec3d9cSgw uint64_t volsize; 1419c2e6a7d6Sperrin rl_t *rl; 1420feb08c6bSbillm int error = 0; 1421510b6c0eSNeil Perrin boolean_t sync; 1422feb08c6bSbillm 1423c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1424c7ca1008Sgw if (zv == NULL) 1425be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1426c7ca1008Sgw 142773ec3d9cSgw volsize = zv->zv_volsize; 142873ec3d9cSgw if (uio->uio_resid > 0 && 142973ec3d9cSgw (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1430be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 143173ec3d9cSgw 1432e7cbe64fSgw if (zv->zv_flags & ZVOL_DUMPIFIED) { 1433e7cbe64fSgw error = physio(zvol_strategy, NULL, dev, B_WRITE, 1434e7cbe64fSgw zvol_minphys, uio); 1435e7cbe64fSgw return (error); 1436e7cbe64fSgw } 1437e7cbe64fSgw 143855da60b9SMark J Musante sync = !(zv->zv_flags & ZVOL_WCE) || 143955da60b9SMark J Musante (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1440510b6c0eSNeil Perrin 1441c2e6a7d6Sperrin rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1442c2e6a7d6Sperrin RL_WRITER); 144373ec3d9cSgw while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1444feb08c6bSbillm uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1445feb08c6bSbillm uint64_t off = uio->uio_loffset; 1446feb08c6bSbillm dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 144773ec3d9cSgw 144873ec3d9cSgw if (bytes > volsize - off) /* don't write past the end */ 144973ec3d9cSgw bytes = volsize - off; 145073ec3d9cSgw 1451feb08c6bSbillm dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1452feb08c6bSbillm error = dmu_tx_assign(tx, TXG_WAIT); 1453feb08c6bSbillm if (error) { 1454feb08c6bSbillm dmu_tx_abort(tx); 1455feb08c6bSbillm break; 1456feb08c6bSbillm } 145794d1a210STim Haley error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1458feb08c6bSbillm if (error == 0) 1459510b6c0eSNeil Perrin zvol_log_write(zv, tx, off, bytes, sync); 1460feb08c6bSbillm dmu_tx_commit(tx); 1461feb08c6bSbillm 1462feb08c6bSbillm if (error) 1463feb08c6bSbillm break; 1464feb08c6bSbillm } 1465c2e6a7d6Sperrin zfs_range_unlock(rl); 1466510b6c0eSNeil Perrin if (sync) 14675002558fSNeil Perrin zil_commit(zv->zv_zilog, ZVOL_OBJ); 1468feb08c6bSbillm return (error); 1469fa9e4066Sahrens } 1470fa9e4066Sahrens 1471c7f714e2SEric Taylor int 1472c7f714e2SEric Taylor zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1473c7f714e2SEric Taylor { 1474c7f714e2SEric Taylor struct uuid uuid = EFI_RESERVED; 1475c7f714e2SEric Taylor efi_gpe_t gpe = { 0 }; 1476c7f714e2SEric Taylor uint32_t crc; 1477c7f714e2SEric Taylor dk_efi_t efi; 1478c7f714e2SEric Taylor int length; 1479c7f714e2SEric Taylor char *ptr; 1480c7f714e2SEric Taylor 1481c7f714e2SEric Taylor if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1482be6fd75aSMatthew Ahrens return (SET_ERROR(EFAULT)); 1483c7f714e2SEric Taylor ptr = (char *)(uintptr_t)efi.dki_data_64; 1484c7f714e2SEric Taylor length = efi.dki_length; 1485c7f714e2SEric Taylor /* 1486c7f714e2SEric Taylor * Some clients may attempt to request a PMBR for the 1487c7f714e2SEric Taylor * zvol. Currently this interface will return EINVAL to 1488c7f714e2SEric Taylor * such requests. These requests could be supported by 1489c7f714e2SEric Taylor * adding a check for lba == 0 and consing up an appropriate 1490c7f714e2SEric Taylor * PMBR. 1491c7f714e2SEric Taylor */ 1492c7f714e2SEric Taylor if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1493be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1494c7f714e2SEric Taylor 1495c7f714e2SEric Taylor gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1496c7f714e2SEric Taylor gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1497c7f714e2SEric Taylor UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1498c7f714e2SEric Taylor 1499c7f714e2SEric Taylor if (efi.dki_lba == 1) { 1500c7f714e2SEric Taylor efi_gpt_t gpt = { 0 }; 1501c7f714e2SEric Taylor 1502c7f714e2SEric Taylor gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1503c7f714e2SEric Taylor gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1504c7f714e2SEric Taylor gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1505c7f714e2SEric Taylor gpt.efi_gpt_MyLBA = LE_64(1ULL); 1506c7f714e2SEric Taylor gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1507c7f714e2SEric Taylor gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1508c7f714e2SEric Taylor gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1509c7f714e2SEric Taylor gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1510c7f714e2SEric Taylor gpt.efi_gpt_SizeOfPartitionEntry = 1511c7f714e2SEric Taylor LE_32(sizeof (efi_gpe_t)); 1512c7f714e2SEric Taylor CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1513c7f714e2SEric Taylor gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1514c7f714e2SEric Taylor CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1515c7f714e2SEric Taylor gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1516c7f714e2SEric Taylor if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1517c7f714e2SEric Taylor flag)) 1518be6fd75aSMatthew Ahrens return (SET_ERROR(EFAULT)); 1519c7f714e2SEric Taylor ptr += sizeof (gpt); 1520c7f714e2SEric Taylor length -= sizeof (gpt); 1521c7f714e2SEric Taylor } 1522c7f714e2SEric Taylor if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1523c7f714e2SEric Taylor length), flag)) 1524be6fd75aSMatthew Ahrens return (SET_ERROR(EFAULT)); 1525c7f714e2SEric Taylor return (0); 1526c7f714e2SEric Taylor } 1527c7f714e2SEric Taylor 15283fb517f7SJames Moore /* 15293fb517f7SJames Moore * BEGIN entry points to allow external callers access to the volume. 15303fb517f7SJames Moore */ 15313fb517f7SJames Moore /* 15323fb517f7SJames Moore * Return the volume parameters needed for access from an external caller. 15333fb517f7SJames Moore * These values are invariant as long as the volume is held open. 15343fb517f7SJames Moore */ 15353fb517f7SJames Moore int 15363fb517f7SJames Moore zvol_get_volume_params(minor_t minor, uint64_t *blksize, 15373fb517f7SJames Moore uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 15383fb517f7SJames Moore void **rl_hdl, void **bonus_hdl) 15393fb517f7SJames Moore { 15403fb517f7SJames Moore zvol_state_t *zv; 15413fb517f7SJames Moore 1542c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1543c99e4bdcSChris Kirby if (zv == NULL) 1544be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 15453fb517f7SJames Moore if (zv->zv_flags & ZVOL_DUMPIFIED) 1546be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 15473fb517f7SJames Moore 15483fb517f7SJames Moore ASSERT(blksize && max_xfer_len && minor_hdl && 15493fb517f7SJames Moore objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 15503fb517f7SJames Moore 15513fb517f7SJames Moore *blksize = zv->zv_volblocksize; 15523fb517f7SJames Moore *max_xfer_len = (uint64_t)zvol_maxphys; 15533fb517f7SJames Moore *minor_hdl = zv; 15543fb517f7SJames Moore *objset_hdl = zv->zv_objset; 15553fb517f7SJames Moore *zil_hdl = zv->zv_zilog; 15563fb517f7SJames Moore *rl_hdl = &zv->zv_znode; 15573fb517f7SJames Moore *bonus_hdl = zv->zv_dbuf; 15583fb517f7SJames Moore return (0); 15593fb517f7SJames Moore } 15603fb517f7SJames Moore 15613fb517f7SJames Moore /* 15623fb517f7SJames Moore * Return the current volume size to an external caller. 15633fb517f7SJames Moore * The size can change while the volume is open. 15643fb517f7SJames Moore */ 15653fb517f7SJames Moore uint64_t 15663fb517f7SJames Moore zvol_get_volume_size(void *minor_hdl) 15673fb517f7SJames Moore { 15683fb517f7SJames Moore zvol_state_t *zv = minor_hdl; 15693fb517f7SJames Moore 15703fb517f7SJames Moore return (zv->zv_volsize); 15713fb517f7SJames Moore } 15723fb517f7SJames Moore 15733fb517f7SJames Moore /* 15743fb517f7SJames Moore * Return the current WCE setting to an external caller. 15753fb517f7SJames Moore * The WCE setting can change while the volume is open. 15763fb517f7SJames Moore */ 15773fb517f7SJames Moore int 15783fb517f7SJames Moore zvol_get_volume_wce(void *minor_hdl) 15793fb517f7SJames Moore { 15803fb517f7SJames Moore zvol_state_t *zv = minor_hdl; 15813fb517f7SJames Moore 15823fb517f7SJames Moore return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 15833fb517f7SJames Moore } 15843fb517f7SJames Moore 15853fb517f7SJames Moore /* 15863fb517f7SJames Moore * Entry point for external callers to zvol_log_write 15873fb517f7SJames Moore */ 15883fb517f7SJames Moore void 15893fb517f7SJames Moore zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 15903fb517f7SJames Moore boolean_t sync) 15913fb517f7SJames Moore { 15923fb517f7SJames Moore zvol_state_t *zv = minor_hdl; 15933fb517f7SJames Moore 15943fb517f7SJames Moore zvol_log_write(zv, tx, off, resid, sync); 15953fb517f7SJames Moore } 15963fb517f7SJames Moore /* 15973fb517f7SJames Moore * END entry points to allow external callers access to the volume. 15983fb517f7SJames Moore */ 15993fb517f7SJames Moore 1600b77b9231SDan McDonald /* 1601b77b9231SDan McDonald * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 1602b77b9231SDan McDonald */ 1603b77b9231SDan McDonald static void 1604b77b9231SDan McDonald zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 1605b77b9231SDan McDonald boolean_t sync) 1606b77b9231SDan McDonald { 1607b77b9231SDan McDonald itx_t *itx; 1608b77b9231SDan McDonald lr_truncate_t *lr; 1609b77b9231SDan McDonald zilog_t *zilog = zv->zv_zilog; 1610b77b9231SDan McDonald 1611b77b9231SDan McDonald if (zil_replaying(zilog, tx)) 1612b77b9231SDan McDonald return; 1613b77b9231SDan McDonald 1614b77b9231SDan McDonald itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1615b77b9231SDan McDonald lr = (lr_truncate_t *)&itx->itx_lr; 1616b77b9231SDan McDonald lr->lr_foid = ZVOL_OBJ; 1617b77b9231SDan McDonald lr->lr_offset = off; 1618b77b9231SDan McDonald lr->lr_length = len; 1619b77b9231SDan McDonald 1620b77b9231SDan McDonald itx->itx_sync = sync; 1621b77b9231SDan McDonald zil_itx_assign(zilog, itx, tx); 1622b77b9231SDan McDonald } 1623b77b9231SDan McDonald 1624fa9e4066Sahrens /* 1625fa9e4066Sahrens * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1626b77b9231SDan McDonald * Also a dirtbag dkio ioctl for unmap/free-block functionality. 1627fa9e4066Sahrens */ 1628fa9e4066Sahrens /*ARGSUSED*/ 1629fa9e4066Sahrens int 1630fa9e4066Sahrens zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1631fa9e4066Sahrens { 1632fa9e4066Sahrens zvol_state_t *zv; 1633af2c4821Smaybee struct dk_callback *dkc; 1634fa9e4066Sahrens int error = 0; 1635e7cbe64fSgw rl_t *rl; 1636fa9e4066Sahrens 1637c99e4bdcSChris Kirby mutex_enter(&zfsdev_state_lock); 1638fa9e4066Sahrens 1639c99e4bdcSChris Kirby zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1640fa9e4066Sahrens 1641fa9e4066Sahrens if (zv == NULL) { 1642c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 1643be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1644fa9e4066Sahrens } 1645701f66c4SEric Taylor ASSERT(zv->zv_total_opens > 0); 1646fa9e4066Sahrens 1647fa9e4066Sahrens switch (cmd) { 1648fa9e4066Sahrens 1649fa9e4066Sahrens case DKIOCINFO: 1650a0b60564SGeorge Wilson { 1651a0b60564SGeorge Wilson struct dk_cinfo dki; 1652a0b60564SGeorge Wilson 1653af2c4821Smaybee bzero(&dki, sizeof (dki)); 1654af2c4821Smaybee (void) strcpy(dki.dki_cname, "zvol"); 1655af2c4821Smaybee (void) strcpy(dki.dki_dname, "zvol"); 1656af2c4821Smaybee dki.dki_ctype = DKC_UNKNOWN; 16573adc9019SEric Taylor dki.dki_unit = getminor(dev); 1658af2c4821Smaybee dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1659c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 1660af2c4821Smaybee if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1661be6fd75aSMatthew Ahrens error = SET_ERROR(EFAULT); 1662fa9e4066Sahrens return (error); 1663a0b60564SGeorge Wilson } 1664fa9e4066Sahrens 1665fa9e4066Sahrens case DKIOCGMEDIAINFO: 1666a0b60564SGeorge Wilson { 1667a0b60564SGeorge Wilson struct dk_minfo dkm; 1668a0b60564SGeorge Wilson 1669fa9e4066Sahrens bzero(&dkm, sizeof (dkm)); 1670fa9e4066Sahrens dkm.dki_lbsize = 1U << zv->zv_min_bs; 1671fa9e4066Sahrens dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1672fa9e4066Sahrens dkm.dki_media_type = DK_UNKNOWN; 1673c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 1674fa9e4066Sahrens if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1675be6fd75aSMatthew Ahrens error = SET_ERROR(EFAULT); 1676fa9e4066Sahrens return (error); 1677a0b60564SGeorge Wilson } 1678a0b60564SGeorge Wilson 1679a0b60564SGeorge Wilson case DKIOCGMEDIAINFOEXT: 1680a0b60564SGeorge Wilson { 1681a0b60564SGeorge Wilson struct dk_minfo_ext dkmext; 1682a0b60564SGeorge Wilson 1683a0b60564SGeorge Wilson bzero(&dkmext, sizeof (dkmext)); 1684a0b60564SGeorge Wilson dkmext.dki_lbsize = 1U << zv->zv_min_bs; 1685a0b60564SGeorge Wilson dkmext.dki_pbsize = zv->zv_volblocksize; 1686a0b60564SGeorge Wilson dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1687a0b60564SGeorge Wilson dkmext.dki_media_type = DK_UNKNOWN; 1688a0b60564SGeorge Wilson mutex_exit(&zfsdev_state_lock); 1689a0b60564SGeorge Wilson if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) 1690a0b60564SGeorge Wilson error = SET_ERROR(EFAULT); 1691a0b60564SGeorge Wilson return (error); 1692a0b60564SGeorge Wilson } 1693fa9e4066Sahrens 1694fa9e4066Sahrens case DKIOCGETEFI: 1695a0b60564SGeorge Wilson { 1696a0b60564SGeorge Wilson uint64_t vs = zv->zv_volsize; 1697a0b60564SGeorge Wilson uint8_t bs = zv->zv_min_bs; 1698fa9e4066Sahrens 1699a0b60564SGeorge Wilson mutex_exit(&zfsdev_state_lock); 1700a0b60564SGeorge Wilson error = zvol_getefi((void *)arg, flag, vs, bs); 1701a0b60564SGeorge Wilson return (error); 1702a0b60564SGeorge Wilson } 1703fa9e4066Sahrens 1704feb08c6bSbillm case DKIOCFLUSHWRITECACHE: 1705af2c4821Smaybee dkc = (struct dk_callback *)arg; 1706c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 17075002558fSNeil Perrin zil_commit(zv->zv_zilog, ZVOL_OBJ); 1708af2c4821Smaybee if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1709af2c4821Smaybee (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1710af2c4821Smaybee error = 0; 1711af2c4821Smaybee } 1712701f66c4SEric Taylor return (error); 1713701f66c4SEric Taylor 1714701f66c4SEric Taylor case DKIOCGETWCE: 1715a0b60564SGeorge Wilson { 1716a0b60564SGeorge Wilson int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1717a0b60564SGeorge Wilson if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1718a0b60564SGeorge Wilson flag)) 1719a0b60564SGeorge Wilson error = SET_ERROR(EFAULT); 1720a0b60564SGeorge Wilson break; 1721a0b60564SGeorge Wilson } 1722a0b60564SGeorge Wilson case DKIOCSETWCE: 1723a0b60564SGeorge Wilson { 1724a0b60564SGeorge Wilson int wce; 1725a0b60564SGeorge Wilson if (ddi_copyin((void *)arg, &wce, sizeof (int), 1726a0b60564SGeorge Wilson flag)) { 1727a0b60564SGeorge Wilson error = SET_ERROR(EFAULT); 1728701f66c4SEric Taylor break; 1729701f66c4SEric Taylor } 1730a0b60564SGeorge Wilson if (wce) { 1731a0b60564SGeorge Wilson zv->zv_flags |= ZVOL_WCE; 1732a0b60564SGeorge Wilson mutex_exit(&zfsdev_state_lock); 1733a0b60564SGeorge Wilson } else { 1734a0b60564SGeorge Wilson zv->zv_flags &= ~ZVOL_WCE; 1735a0b60564SGeorge Wilson mutex_exit(&zfsdev_state_lock); 1736a0b60564SGeorge Wilson zil_commit(zv->zv_zilog, ZVOL_OBJ); 1737701f66c4SEric Taylor } 1738a0b60564SGeorge Wilson return (0); 1739a0b60564SGeorge Wilson } 1740feb08c6bSbillm 1741b6130eadSmaybee case DKIOCGGEOM: 1742b6130eadSmaybee case DKIOCGVTOC: 1743e7cbe64fSgw /* 1744e7cbe64fSgw * commands using these (like prtvtoc) expect ENOTSUP 1745e7cbe64fSgw * since we're emulating an EFI label 1746e7cbe64fSgw */ 1747be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 1748b6130eadSmaybee break; 1749b6130eadSmaybee 1750e7cbe64fSgw case DKIOCDUMPINIT: 1751e7cbe64fSgw rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1752e7cbe64fSgw RL_WRITER); 1753e7cbe64fSgw error = zvol_dumpify(zv); 1754e7cbe64fSgw zfs_range_unlock(rl); 1755e7cbe64fSgw break; 1756e7cbe64fSgw 1757e7cbe64fSgw case DKIOCDUMPFINI: 175806d5ae10SEric Taylor if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 175906d5ae10SEric Taylor break; 1760e7cbe64fSgw rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1761e7cbe64fSgw RL_WRITER); 1762e7cbe64fSgw error = zvol_dump_fini(zv); 1763e7cbe64fSgw zfs_range_unlock(rl); 1764e7cbe64fSgw break; 1765e7cbe64fSgw 1766b77b9231SDan McDonald case DKIOCFREE: 1767b77b9231SDan McDonald { 1768b77b9231SDan McDonald dkioc_free_t df; 1769b77b9231SDan McDonald dmu_tx_t *tx; 1770b77b9231SDan McDonald 1771b77b9231SDan McDonald if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 1772be6fd75aSMatthew Ahrens error = SET_ERROR(EFAULT); 1773b77b9231SDan McDonald break; 1774b77b9231SDan McDonald } 1775b77b9231SDan McDonald 1776b77b9231SDan McDonald /* 1777b77b9231SDan McDonald * Apply Postel's Law to length-checking. If they overshoot, 1778b77b9231SDan McDonald * just blank out until the end, if there's a need to blank 1779b77b9231SDan McDonald * out anything. 1780b77b9231SDan McDonald */ 1781b77b9231SDan McDonald if (df.df_start >= zv->zv_volsize) 1782b77b9231SDan McDonald break; /* No need to do anything... */ 1783b77b9231SDan McDonald if (df.df_start + df.df_length > zv->zv_volsize) 1784b77b9231SDan McDonald df.df_length = DMU_OBJECT_END; 1785b77b9231SDan McDonald 1786b77b9231SDan McDonald rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 1787b77b9231SDan McDonald RL_WRITER); 1788b77b9231SDan McDonald tx = dmu_tx_create(zv->zv_objset); 1789b77b9231SDan McDonald error = dmu_tx_assign(tx, TXG_WAIT); 1790b77b9231SDan McDonald if (error != 0) { 1791b77b9231SDan McDonald dmu_tx_abort(tx); 1792b77b9231SDan McDonald } else { 1793b77b9231SDan McDonald zvol_log_truncate(zv, tx, df.df_start, 1794b77b9231SDan McDonald df.df_length, B_TRUE); 1795c08b1637SJosef 'Jeff' Sipek dmu_tx_commit(tx); 1796b77b9231SDan McDonald error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1797b77b9231SDan McDonald df.df_start, df.df_length); 1798b77b9231SDan McDonald } 1799b77b9231SDan McDonald 1800b77b9231SDan McDonald zfs_range_unlock(rl); 1801b77b9231SDan McDonald 1802b77b9231SDan McDonald if (error == 0) { 1803b77b9231SDan McDonald /* 1804b77b9231SDan McDonald * If the write-cache is disabled or 'sync' property 1805b77b9231SDan McDonald * is set to 'always' then treat this as a synchronous 1806b77b9231SDan McDonald * operation (i.e. commit to zil). 1807b77b9231SDan McDonald */ 1808b77b9231SDan McDonald if (!(zv->zv_flags & ZVOL_WCE) || 1809b77b9231SDan McDonald (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 1810b77b9231SDan McDonald zil_commit(zv->zv_zilog, ZVOL_OBJ); 1811b77b9231SDan McDonald 1812b77b9231SDan McDonald /* 1813b77b9231SDan McDonald * If the caller really wants synchronous writes, and 1814b77b9231SDan McDonald * can't wait for them, don't return until the write 1815b77b9231SDan McDonald * is done. 1816b77b9231SDan McDonald */ 1817b77b9231SDan McDonald if (df.df_flags & DF_WAIT_SYNC) { 1818b77b9231SDan McDonald txg_wait_synced( 1819b77b9231SDan McDonald dmu_objset_pool(zv->zv_objset), 0); 1820b77b9231SDan McDonald } 1821b77b9231SDan McDonald } 1822b77b9231SDan McDonald break; 1823b77b9231SDan McDonald } 1824b77b9231SDan McDonald 1825fa9e4066Sahrens default: 1826be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTTY); 1827fa9e4066Sahrens break; 1828fa9e4066Sahrens 1829fa9e4066Sahrens } 1830c99e4bdcSChris Kirby mutex_exit(&zfsdev_state_lock); 1831fa9e4066Sahrens return (error); 1832fa9e4066Sahrens } 1833fa9e4066Sahrens 1834fa9e4066Sahrens int 1835fa9e4066Sahrens zvol_busy(void) 1836fa9e4066Sahrens { 1837fa9e4066Sahrens return (zvol_minors != 0); 1838fa9e4066Sahrens } 1839fa9e4066Sahrens 1840fa9e4066Sahrens void 1841fa9e4066Sahrens zvol_init(void) 1842fa9e4066Sahrens { 1843c99e4bdcSChris Kirby VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1844c99e4bdcSChris Kirby 1) == 0); 1845c99e4bdcSChris Kirby mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL); 1846fa9e4066Sahrens } 1847fa9e4066Sahrens 1848fa9e4066Sahrens void 1849fa9e4066Sahrens zvol_fini(void) 1850fa9e4066Sahrens { 1851c99e4bdcSChris Kirby mutex_destroy(&zfsdev_state_lock); 1852c99e4bdcSChris Kirby ddi_soft_state_fini(&zfsdev_state); 1853fa9e4066Sahrens } 1854e7cbe64fSgw 1855810e43b2SBill Pijewski /*ARGSUSED*/ 1856810e43b2SBill Pijewski static int 1857810e43b2SBill Pijewski zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 1858810e43b2SBill Pijewski { 1859810e43b2SBill Pijewski spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1860810e43b2SBill Pijewski 18612acef22dSMatthew Ahrens if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1862810e43b2SBill Pijewski return (1); 1863810e43b2SBill Pijewski return (0); 1864810e43b2SBill Pijewski } 1865810e43b2SBill Pijewski 1866810e43b2SBill Pijewski /*ARGSUSED*/ 1867810e43b2SBill Pijewski static void 1868810e43b2SBill Pijewski zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 1869810e43b2SBill Pijewski { 1870810e43b2SBill Pijewski spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1871810e43b2SBill Pijewski 18722acef22dSMatthew Ahrens spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 1873810e43b2SBill Pijewski } 1874810e43b2SBill Pijewski 1875e7cbe64fSgw static int 1876e7cbe64fSgw zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1877e7cbe64fSgw { 1878e7cbe64fSgw dmu_tx_t *tx; 1879810e43b2SBill Pijewski int error; 1880e7cbe64fSgw objset_t *os = zv->zv_objset; 1881810e43b2SBill Pijewski spa_t *spa = dmu_objset_spa(os); 1882810e43b2SBill Pijewski vdev_t *vd = spa->spa_root_vdev; 1883e7cbe64fSgw nvlist_t *nv = NULL; 1884810e43b2SBill Pijewski uint64_t version = spa_version(spa); 1885810e43b2SBill Pijewski enum zio_checksum checksum; 1886e7cbe64fSgw 1887c99e4bdcSChris Kirby ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 1888810e43b2SBill Pijewski ASSERT(vd->vdev_ops == &vdev_root_ops); 1889810e43b2SBill Pijewski 1890681d9761SEric Taylor error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1891681d9761SEric Taylor DMU_OBJECT_END); 1892681d9761SEric Taylor /* wait for dmu_free_long_range to actually free the blocks */ 1893681d9761SEric Taylor txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1894e7cbe64fSgw 1895810e43b2SBill Pijewski /* 1896810e43b2SBill Pijewski * If the pool on which the dump device is being initialized has more 1897810e43b2SBill Pijewski * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 1898810e43b2SBill Pijewski * enabled. If so, bump that feature's counter to indicate that the 1899810e43b2SBill Pijewski * feature is active. We also check the vdev type to handle the 1900810e43b2SBill Pijewski * following case: 1901810e43b2SBill Pijewski * # zpool create test raidz disk1 disk2 disk3 1902810e43b2SBill Pijewski * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 1903810e43b2SBill Pijewski * the raidz vdev itself has 3 children. 1904810e43b2SBill Pijewski */ 1905810e43b2SBill Pijewski if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 1906810e43b2SBill Pijewski if (!spa_feature_is_enabled(spa, 19072acef22dSMatthew Ahrens SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1908810e43b2SBill Pijewski return (SET_ERROR(ENOTSUP)); 1909810e43b2SBill Pijewski (void) dsl_sync_task(spa_name(spa), 1910810e43b2SBill Pijewski zfs_mvdev_dump_feature_check, 1911810e43b2SBill Pijewski zfs_mvdev_dump_activate_feature_sync, NULL, 2); 1912810e43b2SBill Pijewski } 1913810e43b2SBill Pijewski 1914e7cbe64fSgw tx = dmu_tx_create(os); 1915e7cbe64fSgw dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1916681d9761SEric Taylor dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1917e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 1918e7cbe64fSgw if (error) { 1919e7cbe64fSgw dmu_tx_abort(tx); 1920e7cbe64fSgw return (error); 1921e7cbe64fSgw } 1922e7cbe64fSgw 1923810e43b2SBill Pijewski /* 1924810e43b2SBill Pijewski * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 1925810e43b2SBill Pijewski * function. Otherwise, use the old default -- OFF. 1926810e43b2SBill Pijewski */ 1927810e43b2SBill Pijewski checksum = spa_feature_is_active(spa, 19282acef22dSMatthew Ahrens SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 19292acef22dSMatthew Ahrens ZIO_CHECKSUM_OFF; 1930810e43b2SBill Pijewski 1931e7cbe64fSgw /* 1932e7cbe64fSgw * If we are resizing the dump device then we only need to 1933e7cbe64fSgw * update the refreservation to match the newly updated 1934e7cbe64fSgw * zvolsize. Otherwise, we save off the original state of the 1935e7cbe64fSgw * zvol so that we can restore them if the zvol is ever undumpified. 1936e7cbe64fSgw */ 1937e7cbe64fSgw if (resize) { 1938e7cbe64fSgw error = zap_update(os, ZVOL_ZAP_OBJ, 1939e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1940e7cbe64fSgw &zv->zv_volsize, tx); 1941e7cbe64fSgw } else { 1942afee20e4SGeorge Wilson uint64_t checksum, compress, refresrv, vbs, dedup; 194388b7b0f2SMatthew Ahrens 1944e7cbe64fSgw error = dsl_prop_get_integer(zv->zv_name, 1945e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1946e7cbe64fSgw error = error ? error : dsl_prop_get_integer(zv->zv_name, 1947e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 1948e7cbe64fSgw error = error ? error : dsl_prop_get_integer(zv->zv_name, 1949e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 195088b7b0f2SMatthew Ahrens error = error ? error : dsl_prop_get_integer(zv->zv_name, 195188b7b0f2SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 19528d265e66SGeorge Wilson if (version >= SPA_VERSION_DEDUP) { 19538d265e66SGeorge Wilson error = error ? error : 19548d265e66SGeorge Wilson dsl_prop_get_integer(zv->zv_name, 19558d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 19568d265e66SGeorge Wilson } 1957e7cbe64fSgw 1958e7cbe64fSgw error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1959e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1960e7cbe64fSgw &compress, tx); 1961e7cbe64fSgw error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1962e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 1963e7cbe64fSgw error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1964e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1965e7cbe64fSgw &refresrv, tx); 196688b7b0f2SMatthew Ahrens error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 196788b7b0f2SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 196888b7b0f2SMatthew Ahrens &vbs, tx); 1969681d9761SEric Taylor error = error ? error : dmu_object_set_blocksize( 1970681d9761SEric Taylor os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 19718d265e66SGeorge Wilson if (version >= SPA_VERSION_DEDUP) { 19728d265e66SGeorge Wilson error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 19738d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 19748d265e66SGeorge Wilson &dedup, tx); 19758d265e66SGeorge Wilson } 1976681d9761SEric Taylor if (error == 0) 1977681d9761SEric Taylor zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 1978e7cbe64fSgw } 1979e7cbe64fSgw dmu_tx_commit(tx); 1980e7cbe64fSgw 1981e7cbe64fSgw /* 1982e7cbe64fSgw * We only need update the zvol's property if we are initializing 1983e7cbe64fSgw * the dump area for the first time. 1984e7cbe64fSgw */ 1985e7cbe64fSgw if (!resize) { 1986e7cbe64fSgw VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1987e7cbe64fSgw VERIFY(nvlist_add_uint64(nv, 1988e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 1989e7cbe64fSgw VERIFY(nvlist_add_uint64(nv, 1990e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), 1991e7cbe64fSgw ZIO_COMPRESS_OFF) == 0); 1992e7cbe64fSgw VERIFY(nvlist_add_uint64(nv, 1993e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), 1994810e43b2SBill Pijewski checksum) == 0); 19958d265e66SGeorge Wilson if (version >= SPA_VERSION_DEDUP) { 19968d265e66SGeorge Wilson VERIFY(nvlist_add_uint64(nv, 19978d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), 19988d265e66SGeorge Wilson ZIO_CHECKSUM_OFF) == 0); 19998d265e66SGeorge Wilson } 2000e7cbe64fSgw 200192241e0bSTom Erickson error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 200292241e0bSTom Erickson nv, NULL); 2003e7cbe64fSgw nvlist_free(nv); 2004e7cbe64fSgw 2005e7cbe64fSgw if (error) 2006e7cbe64fSgw return (error); 2007e7cbe64fSgw } 2008e7cbe64fSgw 2009e7cbe64fSgw /* Allocate the space for the dump */ 2010e7cbe64fSgw error = zvol_prealloc(zv); 2011e7cbe64fSgw return (error); 2012e7cbe64fSgw } 2013e7cbe64fSgw 2014e7cbe64fSgw static int 2015e7cbe64fSgw zvol_dumpify(zvol_state_t *zv) 2016e7cbe64fSgw { 2017e7cbe64fSgw int error = 0; 2018e7cbe64fSgw uint64_t dumpsize = 0; 2019e7cbe64fSgw dmu_tx_t *tx; 2020e7cbe64fSgw objset_t *os = zv->zv_objset; 2021e7cbe64fSgw 2022681d9761SEric Taylor if (zv->zv_flags & ZVOL_RDONLY) 2023be6fd75aSMatthew Ahrens return (SET_ERROR(EROFS)); 2024e7cbe64fSgw 2025e7cbe64fSgw if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2026e7cbe64fSgw 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 20274445fffbSMatthew Ahrens boolean_t resize = (dumpsize > 0); 2028e7cbe64fSgw 2029e7cbe64fSgw if ((error = zvol_dump_init(zv, resize)) != 0) { 2030e7cbe64fSgw (void) zvol_dump_fini(zv); 2031e7cbe64fSgw return (error); 2032e7cbe64fSgw } 2033e7cbe64fSgw } 2034e7cbe64fSgw 2035e7cbe64fSgw /* 2036e7cbe64fSgw * Build up our lba mapping. 2037e7cbe64fSgw */ 2038e7cbe64fSgw error = zvol_get_lbas(zv); 2039e7cbe64fSgw if (error) { 2040e7cbe64fSgw (void) zvol_dump_fini(zv); 2041e7cbe64fSgw return (error); 2042e7cbe64fSgw } 2043e7cbe64fSgw 2044e7cbe64fSgw tx = dmu_tx_create(os); 2045e7cbe64fSgw dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2046e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 2047e7cbe64fSgw if (error) { 2048e7cbe64fSgw dmu_tx_abort(tx); 2049e7cbe64fSgw (void) zvol_dump_fini(zv); 2050e7cbe64fSgw return (error); 2051e7cbe64fSgw } 2052e7cbe64fSgw 2053e7cbe64fSgw zv->zv_flags |= ZVOL_DUMPIFIED; 2054e7cbe64fSgw error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2055e7cbe64fSgw &zv->zv_volsize, tx); 2056e7cbe64fSgw dmu_tx_commit(tx); 2057e7cbe64fSgw 2058e7cbe64fSgw if (error) { 2059e7cbe64fSgw (void) zvol_dump_fini(zv); 2060e7cbe64fSgw return (error); 2061e7cbe64fSgw } 2062e7cbe64fSgw 2063e7cbe64fSgw txg_wait_synced(dmu_objset_pool(os), 0); 2064e7cbe64fSgw return (0); 2065e7cbe64fSgw } 2066e7cbe64fSgw 2067e7cbe64fSgw static int 2068e7cbe64fSgw zvol_dump_fini(zvol_state_t *zv) 2069e7cbe64fSgw { 2070e7cbe64fSgw dmu_tx_t *tx; 2071e7cbe64fSgw objset_t *os = zv->zv_objset; 2072e7cbe64fSgw nvlist_t *nv; 2073e7cbe64fSgw int error = 0; 2074afee20e4SGeorge Wilson uint64_t checksum, compress, refresrv, vbs, dedup; 20758d265e66SGeorge Wilson uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2076e7cbe64fSgw 2077b7e50089Smaybee /* 2078b7e50089Smaybee * Attempt to restore the zvol back to its pre-dumpified state. 2079b7e50089Smaybee * This is a best-effort attempt as it's possible that not all 2080b7e50089Smaybee * of these properties were initialized during the dumpify process 2081b7e50089Smaybee * (i.e. error during zvol_dump_init). 2082b7e50089Smaybee */ 2083b7e50089Smaybee 2084e7cbe64fSgw tx = dmu_tx_create(os); 2085e7cbe64fSgw dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2086e7cbe64fSgw error = dmu_tx_assign(tx, TXG_WAIT); 2087e7cbe64fSgw if (error) { 2088e7cbe64fSgw dmu_tx_abort(tx); 2089e7cbe64fSgw return (error); 2090e7cbe64fSgw } 2091b7e50089Smaybee (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2092b7e50089Smaybee dmu_tx_commit(tx); 2093e7cbe64fSgw 2094e7cbe64fSgw (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2095e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2096e7cbe64fSgw (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2097e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2098e7cbe64fSgw (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2099e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 210088b7b0f2SMatthew Ahrens (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 210188b7b0f2SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2102e7cbe64fSgw 2103e7cbe64fSgw VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2104e7cbe64fSgw (void) nvlist_add_uint64(nv, 2105e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2106e7cbe64fSgw (void) nvlist_add_uint64(nv, 2107e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2108e7cbe64fSgw (void) nvlist_add_uint64(nv, 2109e7cbe64fSgw zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 21108d265e66SGeorge Wilson if (version >= SPA_VERSION_DEDUP && 21118d265e66SGeorge Wilson zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 21128d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 21138d265e66SGeorge Wilson (void) nvlist_add_uint64(nv, 21148d265e66SGeorge Wilson zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 21158d265e66SGeorge Wilson } 211692241e0bSTom Erickson (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 211792241e0bSTom Erickson nv, NULL); 2118e7cbe64fSgw nvlist_free(nv); 2119e7cbe64fSgw 2120b7e50089Smaybee zvol_free_extents(zv); 2121b7e50089Smaybee zv->zv_flags &= ~ZVOL_DUMPIFIED; 2122b7e50089Smaybee (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2123681d9761SEric Taylor /* wait for dmu_free_long_range to actually free the blocks */ 2124681d9761SEric Taylor txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2125681d9761SEric Taylor tx = dmu_tx_create(os); 2126681d9761SEric Taylor dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2127681d9761SEric Taylor error = dmu_tx_assign(tx, TXG_WAIT); 2128681d9761SEric Taylor if (error) { 2129681d9761SEric Taylor dmu_tx_abort(tx); 2130681d9761SEric Taylor return (error); 2131681d9761SEric Taylor } 2132b24ab676SJeff Bonwick if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2133b24ab676SJeff Bonwick zv->zv_volblocksize = vbs; 2134681d9761SEric Taylor dmu_tx_commit(tx); 2135b7e50089Smaybee 2136e7cbe64fSgw return (0); 2137e7cbe64fSgw } 2138