1efb80947Sahrens /* 2efb80947Sahrens * CDDL HEADER START 3efb80947Sahrens * 4efb80947Sahrens * The contents of this file are subject to the terms of the 5efb80947Sahrens * Common Development and Distribution License (the "License"). 6efb80947Sahrens * You may not use this file except in compliance with the License. 7efb80947Sahrens * 8efb80947Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9efb80947Sahrens * or http://www.opensolaris.org/os/licensing. 10efb80947Sahrens * See the License for the specific language governing permissions 11efb80947Sahrens * and limitations under the License. 12efb80947Sahrens * 13efb80947Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14efb80947Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15efb80947Sahrens * If applicable, add the following below this CDDL HEADER, with the 16efb80947Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17efb80947Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18efb80947Sahrens * 19efb80947Sahrens * CDDL HEADER END 20efb80947Sahrens */ 21efb80947Sahrens /* 22dc7cd546SMark Shellenbaum * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23ec5cf9d5SAlexander Stetsenko * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 24ca0cc391SMatthew Ahrens * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 25a2afb611SJerry Jelinek * Copyright (c) 2014, Joyent, Inc. All rights reserved. 26e77d42eaSMatthew Ahrens * Copyright 2014 HybridCluster. All rights reserved. 27880094b6SAndrew Stormont * Copyright 2016 RackTop Systems. 28c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 29ec5cf9d5SAlexander Stetsenko */ 30efb80947Sahrens 31efb80947Sahrens #include <sys/dmu.h> 32efb80947Sahrens #include <sys/dmu_impl.h> 33efb80947Sahrens #include <sys/dmu_tx.h> 34efb80947Sahrens #include <sys/dbuf.h> 35efb80947Sahrens #include <sys/dnode.h> 36efb80947Sahrens #include <sys/zfs_context.h> 37efb80947Sahrens #include <sys/dmu_objset.h> 38efb80947Sahrens #include <sys/dmu_traverse.h> 39efb80947Sahrens #include <sys/dsl_dataset.h> 40efb80947Sahrens #include <sys/dsl_dir.h> 4192241e0bSTom Erickson #include <sys/dsl_prop.h> 42efb80947Sahrens #include <sys/dsl_pool.h> 43efb80947Sahrens #include <sys/dsl_synctask.h> 44efb80947Sahrens #include <sys/zfs_ioctl.h> 45efb80947Sahrens #include <sys/zap.h> 46efb80947Sahrens #include <sys/zio_checksum.h> 47dc7cd546SMark Shellenbaum #include <sys/zfs_znode.h> 48cde58dbcSMatthew Ahrens #include <zfs_fletcher.h> 499e69d7d0SLori Alt #include <sys/avl.h> 508e714474SLori Alt #include <sys/ddt.h> 51c99e4bdcSChris Kirby #include <sys/zfs_onexit.h> 523b2aab18SMatthew Ahrens #include <sys/dmu_send.h> 533b2aab18SMatthew Ahrens #include <sys/dsl_destroy.h> 545d7b4d43SMatthew Ahrens #include <sys/blkptr.h> 5578f17100SMatthew Ahrens #include <sys/dsl_bookmark.h> 565d7b4d43SMatthew Ahrens #include <sys/zfeature.h> 57a2cdcdd2SPaul Dagnelie #include <sys/bqueue.h> 58efb80947Sahrens 5919b94df9SMatthew Ahrens /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */ 6019b94df9SMatthew Ahrens int zfs_send_corrupt_data = B_FALSE; 61a2cdcdd2SPaul Dagnelie int zfs_send_queue_length = 16 * 1024 * 1024; 62a2cdcdd2SPaul Dagnelie int zfs_recv_queue_length = 16 * 1024 * 1024; 63880094b6SAndrew Stormont /* Set this tunable to FALSE to disable setting of DRR_FLAG_FREERECORDS */ 64880094b6SAndrew Stormont int zfs_send_set_freerecords_bit = B_TRUE; 6519b94df9SMatthew Ahrens 663cb34c60Sahrens static char *dmu_recv_tag = "dmu_recv_tag"; 679c3fd121SMatthew Ahrens const char *recv_clone_name = "%recv"; 683cb34c60Sahrens 69a2cdcdd2SPaul Dagnelie #define BP_SPAN(datablkszsec, indblkshift, level) \ 70a2cdcdd2SPaul Dagnelie (((uint64_t)datablkszsec) << (SPA_MINBLOCKSHIFT + \ 71a2cdcdd2SPaul Dagnelie (level) * (indblkshift - SPA_BLKPTRSHIFT))) 72a2cdcdd2SPaul Dagnelie 739c3fd121SMatthew Ahrens static void byteswap_record(dmu_replay_record_t *drr); 749c3fd121SMatthew Ahrens 75a2cdcdd2SPaul Dagnelie struct send_thread_arg { 76a2cdcdd2SPaul Dagnelie bqueue_t q; 77a2cdcdd2SPaul Dagnelie dsl_dataset_t *ds; /* Dataset to traverse */ 78a2cdcdd2SPaul Dagnelie uint64_t fromtxg; /* Traverse from this txg */ 79a2cdcdd2SPaul Dagnelie int flags; /* flags to pass to traverse_dataset */ 80a2cdcdd2SPaul Dagnelie int error_code; 81a2cdcdd2SPaul Dagnelie boolean_t cancel; 829c3fd121SMatthew Ahrens zbookmark_phys_t resume; 83a2cdcdd2SPaul Dagnelie }; 84a2cdcdd2SPaul Dagnelie 85a2cdcdd2SPaul Dagnelie struct send_block_record { 86a2cdcdd2SPaul Dagnelie boolean_t eos_marker; /* Marks the end of the stream */ 87a2cdcdd2SPaul Dagnelie blkptr_t bp; 88a2cdcdd2SPaul Dagnelie zbookmark_phys_t zb; 89a2cdcdd2SPaul Dagnelie uint8_t indblkshift; 90a2cdcdd2SPaul Dagnelie uint16_t datablkszsec; 91a2cdcdd2SPaul Dagnelie bqueue_node_t ln; 92a2cdcdd2SPaul Dagnelie }; 93a2cdcdd2SPaul Dagnelie 94efb80947Sahrens static int 954e3c9f44SBill Pijewski dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) 96efb80947Sahrens { 979c3fd121SMatthew Ahrens dsl_dataset_t *ds = dmu_objset_ds(dsp->dsa_os); 98efb80947Sahrens ssize_t resid; /* have to get resid to get detailed errno */ 99c20404ffSEli Rosenthal 100c20404ffSEli Rosenthal /* 101c20404ffSEli Rosenthal * The code does not rely on this (len being a multiple of 8). We keep 102c20404ffSEli Rosenthal * this assertion because of the corresponding assertion in 103c20404ffSEli Rosenthal * receive_read(). Keeping this assertion ensures that we do not 104c20404ffSEli Rosenthal * inadvertently break backwards compatibility (causing the assertion 105c20404ffSEli Rosenthal * in receive_read() to trigger on old software). 106c20404ffSEli Rosenthal * 107c20404ffSEli Rosenthal * Removing the assertions could be rolled into a new feature that uses 108c20404ffSEli Rosenthal * data that isn't 8-byte aligned; if the assertions were removed, a 109c20404ffSEli Rosenthal * feature flag would have to be added. 110c20404ffSEli Rosenthal */ 111c20404ffSEli Rosenthal 112fb09f5aaSMadhav Suresh ASSERT0(len % 8); 113efb80947Sahrens 1144e3c9f44SBill Pijewski dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp, 115efb80947Sahrens (caddr_t)buf, len, 116efb80947Sahrens 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); 1174e3c9f44SBill Pijewski 1184e3c9f44SBill Pijewski mutex_enter(&ds->ds_sendstream_lock); 1194e3c9f44SBill Pijewski *dsp->dsa_off += len; 1204e3c9f44SBill Pijewski mutex_exit(&ds->ds_sendstream_lock); 1214e3c9f44SBill Pijewski 1224e3c9f44SBill Pijewski return (dsp->dsa_err); 123efb80947Sahrens } 124efb80947Sahrens 12598110f08SMatthew Ahrens /* 12698110f08SMatthew Ahrens * For all record types except BEGIN, fill in the checksum (overlaid in 12798110f08SMatthew Ahrens * drr_u.drr_checksum.drr_checksum). The checksum verifies everything 12898110f08SMatthew Ahrens * up to the start of the checksum itself. 12998110f08SMatthew Ahrens */ 13098110f08SMatthew Ahrens static int 13198110f08SMatthew Ahrens dump_record(dmu_sendarg_t *dsp, void *payload, int payload_len) 13298110f08SMatthew Ahrens { 13398110f08SMatthew Ahrens ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 13498110f08SMatthew Ahrens ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 135*770499e1SDan Kimmel (void) fletcher_4_incremental_native(dsp->dsa_drr, 13698110f08SMatthew Ahrens offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 13798110f08SMatthew Ahrens &dsp->dsa_zc); 13812b90ee2SMatt Krantz if (dsp->dsa_drr->drr_type == DRR_BEGIN) { 13912b90ee2SMatt Krantz dsp->dsa_sent_begin = B_TRUE; 14012b90ee2SMatt Krantz } else { 14198110f08SMatthew Ahrens ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp->dsa_drr->drr_u. 14298110f08SMatthew Ahrens drr_checksum.drr_checksum)); 14398110f08SMatthew Ahrens dsp->dsa_drr->drr_u.drr_checksum.drr_checksum = dsp->dsa_zc; 14498110f08SMatthew Ahrens } 14512b90ee2SMatt Krantz if (dsp->dsa_drr->drr_type == DRR_END) { 14612b90ee2SMatt Krantz dsp->dsa_sent_end = B_TRUE; 14712b90ee2SMatt Krantz } 148*770499e1SDan Kimmel (void) fletcher_4_incremental_native(&dsp->dsa_drr-> 14998110f08SMatthew Ahrens drr_u.drr_checksum.drr_checksum, 15098110f08SMatthew Ahrens sizeof (zio_cksum_t), &dsp->dsa_zc); 15198110f08SMatthew Ahrens if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 15298110f08SMatthew Ahrens return (SET_ERROR(EINTR)); 15398110f08SMatthew Ahrens if (payload_len != 0) { 154*770499e1SDan Kimmel (void) fletcher_4_incremental_native(payload, payload_len, 15598110f08SMatthew Ahrens &dsp->dsa_zc); 15698110f08SMatthew Ahrens if (dump_bytes(dsp, payload, payload_len) != 0) 15798110f08SMatthew Ahrens return (SET_ERROR(EINTR)); 15898110f08SMatthew Ahrens } 15998110f08SMatthew Ahrens return (0); 16098110f08SMatthew Ahrens } 16198110f08SMatthew Ahrens 16268ecb2ecSPaul Dagnelie /* 16368ecb2ecSPaul Dagnelie * Fill in the drr_free struct, or perform aggregation if the previous record is 16468ecb2ecSPaul Dagnelie * also a free record, and the two are adjacent. 16568ecb2ecSPaul Dagnelie * 16668ecb2ecSPaul Dagnelie * Note that we send free records even for a full send, because we want to be 16768ecb2ecSPaul Dagnelie * able to receive a full send as a clone, which requires a list of all the free 16868ecb2ecSPaul Dagnelie * and freeobject records that were generated on the source. 16968ecb2ecSPaul Dagnelie */ 170efb80947Sahrens static int 1714e3c9f44SBill Pijewski dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 172efb80947Sahrens uint64_t length) 173efb80947Sahrens { 1744e3c9f44SBill Pijewski struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); 1759e69d7d0SLori Alt 1762f3d8780SMatthew Ahrens /* 1772f3d8780SMatthew Ahrens * When we receive a free record, dbuf_free_range() assumes 1782f3d8780SMatthew Ahrens * that the receiving system doesn't have any dbufs in the range 1792f3d8780SMatthew Ahrens * being freed. This is always true because there is a one-record 1802f3d8780SMatthew Ahrens * constraint: we only send one WRITE record for any given 1819c3fd121SMatthew Ahrens * object,offset. We know that the one-record constraint is 1822f3d8780SMatthew Ahrens * true because we always send data in increasing order by 1832f3d8780SMatthew Ahrens * object,offset. 1842f3d8780SMatthew Ahrens * 1852f3d8780SMatthew Ahrens * If the increasing-order constraint ever changes, we should find 1862f3d8780SMatthew Ahrens * another way to assert that the one-record constraint is still 1872f3d8780SMatthew Ahrens * satisfied. 1882f3d8780SMatthew Ahrens */ 1892f3d8780SMatthew Ahrens ASSERT(object > dsp->dsa_last_data_object || 1902f3d8780SMatthew Ahrens (object == dsp->dsa_last_data_object && 1912f3d8780SMatthew Ahrens offset > dsp->dsa_last_data_offset)); 1922f3d8780SMatthew Ahrens 193534029e5SSimon Klinkert if (length != -1ULL && offset + length < offset) 194534029e5SSimon Klinkert length = -1ULL; 195534029e5SSimon Klinkert 1969e69d7d0SLori Alt /* 1979e69d7d0SLori Alt * If there is a pending op, but it's not PENDING_FREE, push it out, 1989e69d7d0SLori Alt * since free block aggregation can only be done for blocks of the 1999e69d7d0SLori Alt * same type (i.e., DRR_FREE records can only be aggregated with 2009e69d7d0SLori Alt * other DRR_FREE records. DRR_FREEOBJECTS records can only be 2019e69d7d0SLori Alt * aggregated with other DRR_FREEOBJECTS records. 2029e69d7d0SLori Alt */ 2034e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE && 2044e3c9f44SBill Pijewski dsp->dsa_pending_op != PENDING_FREE) { 20598110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 206be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2074e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2089e69d7d0SLori Alt } 2099e69d7d0SLori Alt 2104e3c9f44SBill Pijewski if (dsp->dsa_pending_op == PENDING_FREE) { 2119e69d7d0SLori Alt /* 2129e69d7d0SLori Alt * There should never be a PENDING_FREE if length is -1 2139e69d7d0SLori Alt * (because dump_dnode is the only place where this 2149e69d7d0SLori Alt * function is called with a -1, and only after flushing 2159e69d7d0SLori Alt * any pending record). 2169e69d7d0SLori Alt */ 2179e69d7d0SLori Alt ASSERT(length != -1ULL); 2189e69d7d0SLori Alt /* 2199e69d7d0SLori Alt * Check to see whether this free block can be aggregated 2209e69d7d0SLori Alt * with pending one. 2219e69d7d0SLori Alt */ 2229e69d7d0SLori Alt if (drrf->drr_object == object && drrf->drr_offset + 2239e69d7d0SLori Alt drrf->drr_length == offset) { 2249e69d7d0SLori Alt drrf->drr_length += length; 2259e69d7d0SLori Alt return (0); 2269e69d7d0SLori Alt } else { 2279e69d7d0SLori Alt /* not a continuation. Push out pending record */ 22898110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 229be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2304e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2319e69d7d0SLori Alt } 2329e69d7d0SLori Alt } 2339e69d7d0SLori Alt /* create a FREE record and make it pending */ 2344e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 2354e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_FREE; 2369e69d7d0SLori Alt drrf->drr_object = object; 2379e69d7d0SLori Alt drrf->drr_offset = offset; 2389e69d7d0SLori Alt drrf->drr_length = length; 2394e3c9f44SBill Pijewski drrf->drr_toguid = dsp->dsa_toguid; 2409e69d7d0SLori Alt if (length == -1ULL) { 24198110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 242be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2439e69d7d0SLori Alt } else { 2444e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_FREE; 2459e69d7d0SLori Alt } 246efb80947Sahrens 247efb80947Sahrens return (0); 248efb80947Sahrens } 249efb80947Sahrens 250efb80947Sahrens static int 2515d7b4d43SMatthew Ahrens dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, 2525602294fSDan Kimmel uint64_t object, uint64_t offset, int lsize, int psize, const blkptr_t *bp, 2535602294fSDan Kimmel void *data) 254efb80947Sahrens { 2555602294fSDan Kimmel uint64_t payload_size; 2564e3c9f44SBill Pijewski struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); 2579e69d7d0SLori Alt 2582f3d8780SMatthew Ahrens /* 2592f3d8780SMatthew Ahrens * We send data in increasing object, offset order. 2602f3d8780SMatthew Ahrens * See comment in dump_free() for details. 2612f3d8780SMatthew Ahrens */ 2622f3d8780SMatthew Ahrens ASSERT(object > dsp->dsa_last_data_object || 2632f3d8780SMatthew Ahrens (object == dsp->dsa_last_data_object && 2642f3d8780SMatthew Ahrens offset > dsp->dsa_last_data_offset)); 2652f3d8780SMatthew Ahrens dsp->dsa_last_data_object = object; 2665602294fSDan Kimmel dsp->dsa_last_data_offset = offset + lsize - 1; 2678e714474SLori Alt 2689e69d7d0SLori Alt /* 2699e69d7d0SLori Alt * If there is any kind of pending aggregation (currently either 2709e69d7d0SLori Alt * a grouping of free objects or free blocks), push it out to 2719e69d7d0SLori Alt * the stream, since aggregation can't be done across operations 2729e69d7d0SLori Alt * of different types. 2739e69d7d0SLori Alt */ 2744e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 27598110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 276be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2774e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2789e69d7d0SLori Alt } 27998110f08SMatthew Ahrens /* write a WRITE record */ 2804e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 2814e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_WRITE; 2829e69d7d0SLori Alt drrw->drr_object = object; 2839e69d7d0SLori Alt drrw->drr_type = type; 2849e69d7d0SLori Alt drrw->drr_offset = offset; 2854e3c9f44SBill Pijewski drrw->drr_toguid = dsp->dsa_toguid; 2865602294fSDan Kimmel drrw->drr_logical_size = lsize; 2875602294fSDan Kimmel 2885602294fSDan Kimmel /* only set the compression fields if the buf is compressed */ 2895602294fSDan Kimmel if (lsize != psize) { 2905602294fSDan Kimmel ASSERT(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED); 2915602294fSDan Kimmel ASSERT(!BP_IS_EMBEDDED(bp)); 2925602294fSDan Kimmel ASSERT(!BP_SHOULD_BYTESWAP(bp)); 2935602294fSDan Kimmel ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp))); 2945602294fSDan Kimmel ASSERT3U(BP_GET_COMPRESS(bp), !=, ZIO_COMPRESS_OFF); 2955602294fSDan Kimmel ASSERT3S(psize, >, 0); 2965602294fSDan Kimmel ASSERT3S(lsize, >=, psize); 2975602294fSDan Kimmel 2985602294fSDan Kimmel drrw->drr_compressiontype = BP_GET_COMPRESS(bp); 2995602294fSDan Kimmel drrw->drr_compressed_size = psize; 3005602294fSDan Kimmel payload_size = drrw->drr_compressed_size; 3015602294fSDan Kimmel } else { 3025602294fSDan Kimmel payload_size = drrw->drr_logical_size; 3035602294fSDan Kimmel } 3045602294fSDan Kimmel 305b5152584SMatthew Ahrens if (bp == NULL || BP_IS_EMBEDDED(bp)) { 3065d7b4d43SMatthew Ahrens /* 307b5152584SMatthew Ahrens * There's no pre-computed checksum for partial-block 308b5152584SMatthew Ahrens * writes or embedded BP's, so (like 309b5152584SMatthew Ahrens * fletcher4-checkummed blocks) userland will have to 310b5152584SMatthew Ahrens * compute a dedup-capable checksum itself. 3115d7b4d43SMatthew Ahrens */ 3125d7b4d43SMatthew Ahrens drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; 3135d7b4d43SMatthew Ahrens } else { 3145d7b4d43SMatthew Ahrens drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); 31545818ee1SMatthew Ahrens if (zio_checksum_table[drrw->drr_checksumtype].ci_flags & 31645818ee1SMatthew Ahrens ZCHECKSUM_FLAG_DEDUP) 3175d7b4d43SMatthew Ahrens drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP; 3185d7b4d43SMatthew Ahrens DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); 3195d7b4d43SMatthew Ahrens DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); 3205d7b4d43SMatthew Ahrens DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); 3215d7b4d43SMatthew Ahrens drrw->drr_key.ddk_cksum = bp->blk_cksum; 3225d7b4d43SMatthew Ahrens } 323efb80947Sahrens 3245602294fSDan Kimmel if (dump_record(dsp, data, payload_size) != 0) 325be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 326efb80947Sahrens return (0); 327efb80947Sahrens } 328efb80947Sahrens 3295d7b4d43SMatthew Ahrens static int 3305d7b4d43SMatthew Ahrens dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 3315d7b4d43SMatthew Ahrens int blksz, const blkptr_t *bp) 3325d7b4d43SMatthew Ahrens { 3335d7b4d43SMatthew Ahrens char buf[BPE_PAYLOAD_SIZE]; 3345d7b4d43SMatthew Ahrens struct drr_write_embedded *drrw = 3355d7b4d43SMatthew Ahrens &(dsp->dsa_drr->drr_u.drr_write_embedded); 3365d7b4d43SMatthew Ahrens 3375d7b4d43SMatthew Ahrens if (dsp->dsa_pending_op != PENDING_NONE) { 33898110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 3395d7b4d43SMatthew Ahrens return (EINTR); 3405d7b4d43SMatthew Ahrens dsp->dsa_pending_op = PENDING_NONE; 3415d7b4d43SMatthew Ahrens } 3425d7b4d43SMatthew Ahrens 3435d7b4d43SMatthew Ahrens ASSERT(BP_IS_EMBEDDED(bp)); 3445d7b4d43SMatthew Ahrens 3455d7b4d43SMatthew Ahrens bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 3465d7b4d43SMatthew Ahrens dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; 3475d7b4d43SMatthew Ahrens drrw->drr_object = object; 3485d7b4d43SMatthew Ahrens drrw->drr_offset = offset; 3495d7b4d43SMatthew Ahrens drrw->drr_length = blksz; 3505d7b4d43SMatthew Ahrens drrw->drr_toguid = dsp->dsa_toguid; 3515d7b4d43SMatthew Ahrens drrw->drr_compression = BP_GET_COMPRESS(bp); 3525d7b4d43SMatthew Ahrens drrw->drr_etype = BPE_GET_ETYPE(bp); 3535d7b4d43SMatthew Ahrens drrw->drr_lsize = BPE_GET_LSIZE(bp); 3545d7b4d43SMatthew Ahrens drrw->drr_psize = BPE_GET_PSIZE(bp); 3555d7b4d43SMatthew Ahrens 3565d7b4d43SMatthew Ahrens decode_embedded_bp_compressed(bp, buf); 3575d7b4d43SMatthew Ahrens 35898110f08SMatthew Ahrens if (dump_record(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) 3595d7b4d43SMatthew Ahrens return (EINTR); 3605d7b4d43SMatthew Ahrens return (0); 3615d7b4d43SMatthew Ahrens } 3625d7b4d43SMatthew Ahrens 3630a586ceaSMark Shellenbaum static int 3644e3c9f44SBill Pijewski dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data) 3650a586ceaSMark Shellenbaum { 3664e3c9f44SBill Pijewski struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); 3670a586ceaSMark Shellenbaum 3684e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 36998110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 370be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3714e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 3720a586ceaSMark Shellenbaum } 3730a586ceaSMark Shellenbaum 3740a586ceaSMark Shellenbaum /* write a SPILL record */ 3754e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 3764e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_SPILL; 3770a586ceaSMark Shellenbaum drrs->drr_object = object; 3780a586ceaSMark Shellenbaum drrs->drr_length = blksz; 3794e3c9f44SBill Pijewski drrs->drr_toguid = dsp->dsa_toguid; 3800a586ceaSMark Shellenbaum 38198110f08SMatthew Ahrens if (dump_record(dsp, data, blksz) != 0) 382be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3830a586ceaSMark Shellenbaum return (0); 3840a586ceaSMark Shellenbaum } 3850a586ceaSMark Shellenbaum 386efb80947Sahrens static int 3874e3c9f44SBill Pijewski dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) 388efb80947Sahrens { 3894e3c9f44SBill Pijewski struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); 3909e69d7d0SLori Alt 3919e69d7d0SLori Alt /* 3929e69d7d0SLori Alt * If there is a pending op, but it's not PENDING_FREEOBJECTS, 3939e69d7d0SLori Alt * push it out, since free block aggregation can only be done for 3949e69d7d0SLori Alt * blocks of the same type (i.e., DRR_FREE records can only be 3959e69d7d0SLori Alt * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records 3969e69d7d0SLori Alt * can only be aggregated with other DRR_FREEOBJECTS records. 3979e69d7d0SLori Alt */ 3984e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE && 3994e3c9f44SBill Pijewski dsp->dsa_pending_op != PENDING_FREEOBJECTS) { 40098110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 401be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4024e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4039e69d7d0SLori Alt } 4044e3c9f44SBill Pijewski if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { 4059e69d7d0SLori Alt /* 4069e69d7d0SLori Alt * See whether this free object array can be aggregated 4079e69d7d0SLori Alt * with pending one 4089e69d7d0SLori Alt */ 4099e69d7d0SLori Alt if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { 4109e69d7d0SLori Alt drrfo->drr_numobjs += numobjs; 4119e69d7d0SLori Alt return (0); 4129e69d7d0SLori Alt } else { 4139e69d7d0SLori Alt /* can't be aggregated. Push out pending record */ 41498110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 415be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4164e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4179e69d7d0SLori Alt } 4189e69d7d0SLori Alt } 4199e69d7d0SLori Alt 420efb80947Sahrens /* write a FREEOBJECTS record */ 4214e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 4224e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; 4239e69d7d0SLori Alt drrfo->drr_firstobj = firstobj; 4249e69d7d0SLori Alt drrfo->drr_numobjs = numobjs; 4254e3c9f44SBill Pijewski drrfo->drr_toguid = dsp->dsa_toguid; 4269e69d7d0SLori Alt 4274e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_FREEOBJECTS; 428efb80947Sahrens 429efb80947Sahrens return (0); 430efb80947Sahrens } 431efb80947Sahrens 432efb80947Sahrens static int 4334e3c9f44SBill Pijewski dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) 434efb80947Sahrens { 4354e3c9f44SBill Pijewski struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); 4369e69d7d0SLori Alt 4379c3fd121SMatthew Ahrens if (object < dsp->dsa_resume_object) { 4389c3fd121SMatthew Ahrens /* 4399c3fd121SMatthew Ahrens * Note: when resuming, we will visit all the dnodes in 4409c3fd121SMatthew Ahrens * the block of dnodes that we are resuming from. In 4419c3fd121SMatthew Ahrens * this case it's unnecessary to send the dnodes prior to 4429c3fd121SMatthew Ahrens * the one we are resuming from. We should be at most one 4439c3fd121SMatthew Ahrens * block's worth of dnodes behind the resume point. 4449c3fd121SMatthew Ahrens */ 4459c3fd121SMatthew Ahrens ASSERT3U(dsp->dsa_resume_object - object, <, 4469c3fd121SMatthew Ahrens 1 << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)); 4479c3fd121SMatthew Ahrens return (0); 4489c3fd121SMatthew Ahrens } 4499c3fd121SMatthew Ahrens 450efb80947Sahrens if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) 4514e3c9f44SBill Pijewski return (dump_freeobjects(dsp, object, 1)); 452efb80947Sahrens 4534e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 45498110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 455be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4564e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4579e69d7d0SLori Alt } 4589e69d7d0SLori Alt 459efb80947Sahrens /* write an OBJECT record */ 4604e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 4614e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_OBJECT; 4629e69d7d0SLori Alt drro->drr_object = object; 4639e69d7d0SLori Alt drro->drr_type = dnp->dn_type; 4649e69d7d0SLori Alt drro->drr_bonustype = dnp->dn_bonustype; 4659e69d7d0SLori Alt drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 4669e69d7d0SLori Alt drro->drr_bonuslen = dnp->dn_bonuslen; 4679e69d7d0SLori Alt drro->drr_checksumtype = dnp->dn_checksum; 4689e69d7d0SLori Alt drro->drr_compress = dnp->dn_compress; 4694e3c9f44SBill Pijewski drro->drr_toguid = dsp->dsa_toguid; 4709e69d7d0SLori Alt 471b5152584SMatthew Ahrens if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 472b5152584SMatthew Ahrens drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) 473b5152584SMatthew Ahrens drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; 474b5152584SMatthew Ahrens 47598110f08SMatthew Ahrens if (dump_record(dsp, DN_BONUS(dnp), 47698110f08SMatthew Ahrens P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0) { 477be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 47898110f08SMatthew Ahrens } 479efb80947Sahrens 4802f3d8780SMatthew Ahrens /* Free anything past the end of the file. */ 4814e3c9f44SBill Pijewski if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * 4822f3d8780SMatthew Ahrens (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0) 483be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4843b2aab18SMatthew Ahrens if (dsp->dsa_err != 0) 485be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 486efb80947Sahrens return (0); 487efb80947Sahrens } 488efb80947Sahrens 4895d7b4d43SMatthew Ahrens static boolean_t 4905d7b4d43SMatthew Ahrens backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) 4915d7b4d43SMatthew Ahrens { 4925d7b4d43SMatthew Ahrens if (!BP_IS_EMBEDDED(bp)) 4935d7b4d43SMatthew Ahrens return (B_FALSE); 4945d7b4d43SMatthew Ahrens 4955d7b4d43SMatthew Ahrens /* 4965d7b4d43SMatthew Ahrens * Compression function must be legacy, or explicitly enabled. 4975d7b4d43SMatthew Ahrens */ 4985d7b4d43SMatthew Ahrens if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && 4995602294fSDan Kimmel !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LZ4))) 5005d7b4d43SMatthew Ahrens return (B_FALSE); 5015d7b4d43SMatthew Ahrens 5025d7b4d43SMatthew Ahrens /* 5035d7b4d43SMatthew Ahrens * Embed type must be explicitly enabled. 5045d7b4d43SMatthew Ahrens */ 5055d7b4d43SMatthew Ahrens switch (BPE_GET_ETYPE(bp)) { 5065d7b4d43SMatthew Ahrens case BP_EMBEDDED_TYPE_DATA: 5075d7b4d43SMatthew Ahrens if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) 5085d7b4d43SMatthew Ahrens return (B_TRUE); 5095d7b4d43SMatthew Ahrens break; 5105d7b4d43SMatthew Ahrens default: 5115d7b4d43SMatthew Ahrens return (B_FALSE); 5125d7b4d43SMatthew Ahrens } 5135d7b4d43SMatthew Ahrens return (B_FALSE); 5145d7b4d43SMatthew Ahrens } 5155d7b4d43SMatthew Ahrens 516a2cdcdd2SPaul Dagnelie /* 517a2cdcdd2SPaul Dagnelie * This is the callback function to traverse_dataset that acts as the worker 518a2cdcdd2SPaul Dagnelie * thread for dmu_send_impl. 519a2cdcdd2SPaul Dagnelie */ 520a2cdcdd2SPaul Dagnelie /*ARGSUSED*/ 521a2cdcdd2SPaul Dagnelie static int 522a2cdcdd2SPaul Dagnelie send_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 523a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb, const struct dnode_phys *dnp, void *arg) 524a2cdcdd2SPaul Dagnelie { 525a2cdcdd2SPaul Dagnelie struct send_thread_arg *sta = arg; 526a2cdcdd2SPaul Dagnelie struct send_block_record *record; 527a2cdcdd2SPaul Dagnelie uint64_t record_size; 528a2cdcdd2SPaul Dagnelie int err = 0; 529efb80947Sahrens 5309c3fd121SMatthew Ahrens ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || 5319c3fd121SMatthew Ahrens zb->zb_object >= sta->resume.zb_object); 5329c3fd121SMatthew Ahrens 533a2cdcdd2SPaul Dagnelie if (sta->cancel) 534a2cdcdd2SPaul Dagnelie return (SET_ERROR(EINTR)); 535a2cdcdd2SPaul Dagnelie 536a2cdcdd2SPaul Dagnelie if (bp == NULL) { 537a2cdcdd2SPaul Dagnelie ASSERT3U(zb->zb_level, ==, ZB_DNODE_LEVEL); 538a2cdcdd2SPaul Dagnelie return (0); 539a2cdcdd2SPaul Dagnelie } else if (zb->zb_level < 0) { 540a2cdcdd2SPaul Dagnelie return (0); 541a2cdcdd2SPaul Dagnelie } 542a2cdcdd2SPaul Dagnelie 543a2cdcdd2SPaul Dagnelie record = kmem_zalloc(sizeof (struct send_block_record), KM_SLEEP); 544a2cdcdd2SPaul Dagnelie record->eos_marker = B_FALSE; 545a2cdcdd2SPaul Dagnelie record->bp = *bp; 546a2cdcdd2SPaul Dagnelie record->zb = *zb; 547a2cdcdd2SPaul Dagnelie record->indblkshift = dnp->dn_indblkshift; 548a2cdcdd2SPaul Dagnelie record->datablkszsec = dnp->dn_datablkszsec; 549a2cdcdd2SPaul Dagnelie record_size = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 550a2cdcdd2SPaul Dagnelie bqueue_enqueue(&sta->q, record, record_size); 551a2cdcdd2SPaul Dagnelie 552a2cdcdd2SPaul Dagnelie return (err); 553a2cdcdd2SPaul Dagnelie } 554a2cdcdd2SPaul Dagnelie 555a2cdcdd2SPaul Dagnelie /* 556a2cdcdd2SPaul Dagnelie * This function kicks off the traverse_dataset. It also handles setting the 557a2cdcdd2SPaul Dagnelie * error code of the thread in case something goes wrong, and pushes the End of 558a2cdcdd2SPaul Dagnelie * Stream record when the traverse_dataset call has finished. If there is no 559a2cdcdd2SPaul Dagnelie * dataset to traverse, the thread immediately pushes End of Stream marker. 560a2cdcdd2SPaul Dagnelie */ 561a2cdcdd2SPaul Dagnelie static void 562a2cdcdd2SPaul Dagnelie send_traverse_thread(void *arg) 563a2cdcdd2SPaul Dagnelie { 564a2cdcdd2SPaul Dagnelie struct send_thread_arg *st_arg = arg; 565a2cdcdd2SPaul Dagnelie int err; 566a2cdcdd2SPaul Dagnelie struct send_block_record *data; 567a2cdcdd2SPaul Dagnelie 568a2cdcdd2SPaul Dagnelie if (st_arg->ds != NULL) { 5699c3fd121SMatthew Ahrens err = traverse_dataset_resume(st_arg->ds, 5709c3fd121SMatthew Ahrens st_arg->fromtxg, &st_arg->resume, 5719c3fd121SMatthew Ahrens st_arg->flags, send_cb, st_arg); 5729c3fd121SMatthew Ahrens 573a2cdcdd2SPaul Dagnelie if (err != EINTR) 574a2cdcdd2SPaul Dagnelie st_arg->error_code = err; 575a2cdcdd2SPaul Dagnelie } 576a2cdcdd2SPaul Dagnelie data = kmem_zalloc(sizeof (*data), KM_SLEEP); 577a2cdcdd2SPaul Dagnelie data->eos_marker = B_TRUE; 578a2cdcdd2SPaul Dagnelie bqueue_enqueue(&st_arg->q, data, 1); 579f2c1e9bcSJorgen Lundman thread_exit(); 580a2cdcdd2SPaul Dagnelie } 581a2cdcdd2SPaul Dagnelie 582a2cdcdd2SPaul Dagnelie /* 583a2cdcdd2SPaul Dagnelie * This function actually handles figuring out what kind of record needs to be 584a2cdcdd2SPaul Dagnelie * dumped, reading the data (which has hopefully been prefetched), and calling 585a2cdcdd2SPaul Dagnelie * the appropriate helper function. 586a2cdcdd2SPaul Dagnelie */ 587efb80947Sahrens static int 588a2cdcdd2SPaul Dagnelie do_dump(dmu_sendarg_t *dsa, struct send_block_record *data) 589efb80947Sahrens { 590a2cdcdd2SPaul Dagnelie dsl_dataset_t *ds = dmu_objset_ds(dsa->dsa_os); 591a2cdcdd2SPaul Dagnelie const blkptr_t *bp = &data->bp; 592a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb = &data->zb; 593a2cdcdd2SPaul Dagnelie uint8_t indblkshift = data->indblkshift; 594a2cdcdd2SPaul Dagnelie uint16_t dblkszsec = data->datablkszsec; 595a2cdcdd2SPaul Dagnelie spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 596efb80947Sahrens dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; 597efb80947Sahrens int err = 0; 598efb80947Sahrens 599a2cdcdd2SPaul Dagnelie ASSERT3U(zb->zb_level, >=, 0); 600efb80947Sahrens 6019c3fd121SMatthew Ahrens ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || 6029c3fd121SMatthew Ahrens zb->zb_object >= dsa->dsa_resume_object); 6039c3fd121SMatthew Ahrens 604b24ab676SJeff Bonwick if (zb->zb_object != DMU_META_DNODE_OBJECT && 605b24ab676SJeff Bonwick DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { 60614843421SMatthew Ahrens return (0); 60743466aaeSMax Grossman } else if (BP_IS_HOLE(bp) && 60843466aaeSMax Grossman zb->zb_object == DMU_META_DNODE_OBJECT) { 609a2cdcdd2SPaul Dagnelie uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); 61088b7b0f2SMatthew Ahrens uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; 611a2cdcdd2SPaul Dagnelie err = dump_freeobjects(dsa, dnobj, span >> DNODE_SHIFT); 61243466aaeSMax Grossman } else if (BP_IS_HOLE(bp)) { 613a2cdcdd2SPaul Dagnelie uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); 614a2cdcdd2SPaul Dagnelie uint64_t offset = zb->zb_blkid * span; 615a2cdcdd2SPaul Dagnelie err = dump_free(dsa, zb->zb_object, offset, span); 61688b7b0f2SMatthew Ahrens } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { 61788b7b0f2SMatthew Ahrens return (0); 61888b7b0f2SMatthew Ahrens } else if (type == DMU_OT_DNODE) { 619efb80947Sahrens int blksz = BP_GET_LSIZE(bp); 6207adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 62188b7b0f2SMatthew Ahrens arc_buf_t *abuf; 622efb80947Sahrens 623a2cdcdd2SPaul Dagnelie ASSERT0(zb->zb_level); 624a2cdcdd2SPaul Dagnelie 6251b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 6261b912ec7SGeorge Wilson ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 6271b912ec7SGeorge Wilson &aflags, zb) != 0) 628be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 62988b7b0f2SMatthew Ahrens 630a2cdcdd2SPaul Dagnelie dnode_phys_t *blk = abuf->b_data; 631a2cdcdd2SPaul Dagnelie uint64_t dnobj = zb->zb_blkid * (blksz >> DNODE_SHIFT); 632a2cdcdd2SPaul Dagnelie for (int i = 0; i < blksz >> DNODE_SHIFT; i++) { 633a2cdcdd2SPaul Dagnelie err = dump_dnode(dsa, dnobj + i, blk + i); 6343b2aab18SMatthew Ahrens if (err != 0) 635efb80947Sahrens break; 636efb80947Sahrens } 637dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 6380a586ceaSMark Shellenbaum } else if (type == DMU_OT_SA) { 6397adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 6400a586ceaSMark Shellenbaum arc_buf_t *abuf; 6410a586ceaSMark Shellenbaum int blksz = BP_GET_LSIZE(bp); 6420a586ceaSMark Shellenbaum 6431b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 6441b912ec7SGeorge Wilson ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 6451b912ec7SGeorge Wilson &aflags, zb) != 0) 646be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 6470a586ceaSMark Shellenbaum 648a2cdcdd2SPaul Dagnelie err = dump_spill(dsa, zb->zb_object, blksz, abuf->b_data); 649dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 650a2cdcdd2SPaul Dagnelie } else if (backup_do_embed(dsa, bp)) { 6515d7b4d43SMatthew Ahrens /* it's an embedded level-0 block of a regular object */ 652a2cdcdd2SPaul Dagnelie int blksz = dblkszsec << SPA_MINBLOCKSHIFT; 653a2cdcdd2SPaul Dagnelie ASSERT0(zb->zb_level); 654a2cdcdd2SPaul Dagnelie err = dump_write_embedded(dsa, zb->zb_object, 6555d7b4d43SMatthew Ahrens zb->zb_blkid * blksz, blksz, bp); 656a2cdcdd2SPaul Dagnelie } else { 657a2cdcdd2SPaul Dagnelie /* it's a level-0 block of a regular object */ 6587adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 65988b7b0f2SMatthew Ahrens arc_buf_t *abuf; 660a2cdcdd2SPaul Dagnelie int blksz = dblkszsec << SPA_MINBLOCKSHIFT; 661b5152584SMatthew Ahrens uint64_t offset; 66288b7b0f2SMatthew Ahrens 6635602294fSDan Kimmel /* 6645602294fSDan Kimmel * If we have large blocks stored on disk but the send flags 6655602294fSDan Kimmel * don't allow us to send large blocks, we split the data from 6665602294fSDan Kimmel * the arc buf into chunks. 6675602294fSDan Kimmel */ 6685602294fSDan Kimmel boolean_t split_large_blocks = blksz > SPA_OLD_MAXBLOCKSIZE && 6695602294fSDan Kimmel !(dsa->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS); 6705602294fSDan Kimmel /* 6715602294fSDan Kimmel * We should only request compressed data from the ARC if all 6725602294fSDan Kimmel * the following are true: 6735602294fSDan Kimmel * - stream compression was requested 6745602294fSDan Kimmel * - we aren't splitting large blocks into smaller chunks 6755602294fSDan Kimmel * - the data won't need to be byteswapped before sending 6765602294fSDan Kimmel * - this isn't an embedded block 6775602294fSDan Kimmel * - this isn't metadata (if receiving on a different endian 6785602294fSDan Kimmel * system it can be byteswapped more easily) 6795602294fSDan Kimmel */ 6805602294fSDan Kimmel boolean_t request_compressed = 6815602294fSDan Kimmel (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED) && 6825602294fSDan Kimmel !split_large_blocks && !BP_SHOULD_BYTESWAP(bp) && 6835602294fSDan Kimmel !BP_IS_EMBEDDED(bp) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp)); 6845602294fSDan Kimmel 6855602294fSDan Kimmel ASSERT0(zb->zb_level); 6865602294fSDan Kimmel ASSERT(zb->zb_object > dsa->dsa_resume_object || 6875602294fSDan Kimmel (zb->zb_object == dsa->dsa_resume_object && 6885602294fSDan Kimmel zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); 6895602294fSDan Kimmel 69078f17100SMatthew Ahrens ASSERT0(zb->zb_level); 6919c3fd121SMatthew Ahrens ASSERT(zb->zb_object > dsa->dsa_resume_object || 6929c3fd121SMatthew Ahrens (zb->zb_object == dsa->dsa_resume_object && 6939c3fd121SMatthew Ahrens zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); 6949c3fd121SMatthew Ahrens 6955602294fSDan Kimmel ASSERT3U(blksz, ==, BP_GET_LSIZE(bp)); 6965602294fSDan Kimmel 6975602294fSDan Kimmel enum zio_flag zioflags = ZIO_FLAG_CANFAIL; 6985602294fSDan Kimmel if (request_compressed) 6995602294fSDan Kimmel zioflags |= ZIO_FLAG_RAW; 7001b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 7015602294fSDan Kimmel ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) { 70219b94df9SMatthew Ahrens if (zfs_send_corrupt_data) { 70319b94df9SMatthew Ahrens /* Send a block filled with 0x"zfs badd bloc" */ 7045602294fSDan Kimmel abuf = arc_alloc_buf(spa, &abuf, ARC_BUFC_DATA, 7055602294fSDan Kimmel blksz); 70619b94df9SMatthew Ahrens uint64_t *ptr; 70719b94df9SMatthew Ahrens for (ptr = abuf->b_data; 70819b94df9SMatthew Ahrens (char *)ptr < (char *)abuf->b_data + blksz; 70919b94df9SMatthew Ahrens ptr++) 7108c76e076SBrian Behlendorf *ptr = 0x2f5baddb10cULL; 71119b94df9SMatthew Ahrens } else { 712be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 71319b94df9SMatthew Ahrens } 71419b94df9SMatthew Ahrens } 71588b7b0f2SMatthew Ahrens 716b5152584SMatthew Ahrens offset = zb->zb_blkid * blksz; 717b5152584SMatthew Ahrens 7185602294fSDan Kimmel if (split_large_blocks) { 7195602294fSDan Kimmel ASSERT3U(arc_get_compression(abuf), ==, 7205602294fSDan Kimmel ZIO_COMPRESS_OFF); 721b5152584SMatthew Ahrens char *buf = abuf->b_data; 722b5152584SMatthew Ahrens while (blksz > 0 && err == 0) { 723b5152584SMatthew Ahrens int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); 724a2cdcdd2SPaul Dagnelie err = dump_write(dsa, type, zb->zb_object, 7255602294fSDan Kimmel offset, n, n, NULL, buf); 726b5152584SMatthew Ahrens offset += n; 727b5152584SMatthew Ahrens buf += n; 728b5152584SMatthew Ahrens blksz -= n; 729b5152584SMatthew Ahrens } 730b5152584SMatthew Ahrens } else { 7315602294fSDan Kimmel err = dump_write(dsa, type, zb->zb_object, offset, 7325602294fSDan Kimmel blksz, arc_buf_size(abuf), bp, abuf->b_data); 733b5152584SMatthew Ahrens } 734dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 735efb80947Sahrens } 736efb80947Sahrens 737efb80947Sahrens ASSERT(err == 0 || err == EINTR); 738efb80947Sahrens return (err); 739efb80947Sahrens } 740efb80947Sahrens 7414445fffbSMatthew Ahrens /* 742a2cdcdd2SPaul Dagnelie * Pop the new data off the queue, and free the old data. 743a2cdcdd2SPaul Dagnelie */ 744a2cdcdd2SPaul Dagnelie static struct send_block_record * 745a2cdcdd2SPaul Dagnelie get_next_record(bqueue_t *bq, struct send_block_record *data) 746a2cdcdd2SPaul Dagnelie { 747a2cdcdd2SPaul Dagnelie struct send_block_record *tmp = bqueue_dequeue(bq); 748a2cdcdd2SPaul Dagnelie kmem_free(data, sizeof (*data)); 749a2cdcdd2SPaul Dagnelie return (tmp); 750a2cdcdd2SPaul Dagnelie } 751a2cdcdd2SPaul Dagnelie 752a2cdcdd2SPaul Dagnelie /* 753a2cdcdd2SPaul Dagnelie * Actually do the bulk of the work in a zfs send. 754a2cdcdd2SPaul Dagnelie * 755a2cdcdd2SPaul Dagnelie * Note: Releases dp using the specified tag. 7564445fffbSMatthew Ahrens */ 7573b2aab18SMatthew Ahrens static int 758a2cdcdd2SPaul Dagnelie dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *to_ds, 7595602294fSDan Kimmel zfs_bookmark_phys_t *ancestor_zb, boolean_t is_clone, 7605602294fSDan Kimmel boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, 7615602294fSDan Kimmel int outfd, uint64_t resumeobj, uint64_t resumeoff, 7629c3fd121SMatthew Ahrens vnode_t *vp, offset_t *off) 763efb80947Sahrens { 7643b2aab18SMatthew Ahrens objset_t *os; 765efb80947Sahrens dmu_replay_record_t *drr; 7664e3c9f44SBill Pijewski dmu_sendarg_t *dsp; 767efb80947Sahrens int err; 7683cb34c60Sahrens uint64_t fromtxg = 0; 7695d7b4d43SMatthew Ahrens uint64_t featureflags = 0; 7709c3fd121SMatthew Ahrens struct send_thread_arg to_arg = { 0 }; 771efb80947Sahrens 772a2cdcdd2SPaul Dagnelie err = dmu_objset_from_ds(to_ds, &os); 7733b2aab18SMatthew Ahrens if (err != 0) { 7743b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 7753b2aab18SMatthew Ahrens return (err); 7763b2aab18SMatthew Ahrens } 777efb80947Sahrens 778efb80947Sahrens drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 779efb80947Sahrens drr->drr_type = DRR_BEGIN; 780efb80947Sahrens drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 7819e69d7d0SLori Alt DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, 7829e69d7d0SLori Alt DMU_SUBSTREAM); 783dc7cd546SMark Shellenbaum 784dc7cd546SMark Shellenbaum #ifdef _KERNEL 7853b2aab18SMatthew Ahrens if (dmu_objset_type(os) == DMU_OST_ZFS) { 786dc7cd546SMark Shellenbaum uint64_t version; 7873b2aab18SMatthew Ahrens if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { 7884e3c9f44SBill Pijewski kmem_free(drr, sizeof (dmu_replay_record_t)); 7893b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 790be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 7914e3c9f44SBill Pijewski } 7923b2aab18SMatthew Ahrens if (version >= ZPL_VERSION_SA) { 7935d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 794dc7cd546SMark Shellenbaum } 795dc7cd546SMark Shellenbaum } 796dc7cd546SMark Shellenbaum #endif 797dc7cd546SMark Shellenbaum 798ca0cc391SMatthew Ahrens if (large_block_ok && to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS]) 799b5152584SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; 8005d7b4d43SMatthew Ahrens if (embedok && 8015d7b4d43SMatthew Ahrens spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { 8025d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; 8035d7b4d43SMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 8045602294fSDan Kimmel featureflags |= DMU_BACKUP_FEATURE_LZ4; 8055602294fSDan Kimmel } 8065602294fSDan Kimmel if (compressok) { 8075602294fSDan Kimmel featureflags |= DMU_BACKUP_FEATURE_COMPRESSED; 8085602294fSDan Kimmel } 8095602294fSDan Kimmel if ((featureflags & 8105602294fSDan Kimmel (DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_COMPRESSED)) != 8115602294fSDan Kimmel 0 && spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) { 8125602294fSDan Kimmel featureflags |= DMU_BACKUP_FEATURE_LZ4; 8135d7b4d43SMatthew Ahrens } 8145d7b4d43SMatthew Ahrens 8159c3fd121SMatthew Ahrens if (resumeobj != 0 || resumeoff != 0) { 8169c3fd121SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_RESUMING; 8179c3fd121SMatthew Ahrens } 8189c3fd121SMatthew Ahrens 8195d7b4d43SMatthew Ahrens DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, 8205d7b4d43SMatthew Ahrens featureflags); 8215d7b4d43SMatthew Ahrens 822efb80947Sahrens drr->drr_u.drr_begin.drr_creation_time = 823a2cdcdd2SPaul Dagnelie dsl_dataset_phys(to_ds)->ds_creation_time; 8243b2aab18SMatthew Ahrens drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); 82578f17100SMatthew Ahrens if (is_clone) 8263cb34c60Sahrens drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; 827a2cdcdd2SPaul Dagnelie drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(to_ds)->ds_guid; 828a2cdcdd2SPaul Dagnelie if (dsl_dataset_phys(to_ds)->ds_flags & DS_FLAG_CI_DATASET) 829ab04eb8eStimh drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; 830880094b6SAndrew Stormont if (zfs_send_set_freerecords_bit) 831880094b6SAndrew Stormont drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_FREERECORDS; 832ab04eb8eStimh 833a2cdcdd2SPaul Dagnelie if (ancestor_zb != NULL) { 834a2cdcdd2SPaul Dagnelie drr->drr_u.drr_begin.drr_fromguid = 835a2cdcdd2SPaul Dagnelie ancestor_zb->zbm_guid; 836a2cdcdd2SPaul Dagnelie fromtxg = ancestor_zb->zbm_creation_txg; 83778f17100SMatthew Ahrens } 838a2cdcdd2SPaul Dagnelie dsl_dataset_name(to_ds, drr->drr_u.drr_begin.drr_toname); 839a2cdcdd2SPaul Dagnelie if (!to_ds->ds_is_snapshot) { 84078f17100SMatthew Ahrens (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", 84178f17100SMatthew Ahrens sizeof (drr->drr_u.drr_begin.drr_toname)); 8423b2aab18SMatthew Ahrens } 8433cb34c60Sahrens 8444e3c9f44SBill Pijewski dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); 8454e3c9f44SBill Pijewski 8464e3c9f44SBill Pijewski dsp->dsa_drr = drr; 8474e3c9f44SBill Pijewski dsp->dsa_vp = vp; 8484e3c9f44SBill Pijewski dsp->dsa_outfd = outfd; 8494e3c9f44SBill Pijewski dsp->dsa_proc = curproc; 8503b2aab18SMatthew Ahrens dsp->dsa_os = os; 8514e3c9f44SBill Pijewski dsp->dsa_off = off; 852a2cdcdd2SPaul Dagnelie dsp->dsa_toguid = dsl_dataset_phys(to_ds)->ds_guid; 8534e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 8545d7b4d43SMatthew Ahrens dsp->dsa_featureflags = featureflags; 8559c3fd121SMatthew Ahrens dsp->dsa_resume_object = resumeobj; 8569c3fd121SMatthew Ahrens dsp->dsa_resume_offset = resumeoff; 8574e3c9f44SBill Pijewski 858a2cdcdd2SPaul Dagnelie mutex_enter(&to_ds->ds_sendstream_lock); 859a2cdcdd2SPaul Dagnelie list_insert_head(&to_ds->ds_sendstreams, dsp); 860a2cdcdd2SPaul Dagnelie mutex_exit(&to_ds->ds_sendstream_lock); 8614e3c9f44SBill Pijewski 862a2cdcdd2SPaul Dagnelie dsl_dataset_long_hold(to_ds, FTAG); 863de8d9cffSMatthew Ahrens dsl_pool_rele(dp, tag); 864de8d9cffSMatthew Ahrens 8659c3fd121SMatthew Ahrens void *payload = NULL; 8669c3fd121SMatthew Ahrens size_t payload_len = 0; 8679c3fd121SMatthew Ahrens if (resumeobj != 0 || resumeoff != 0) { 8689c3fd121SMatthew Ahrens dmu_object_info_t to_doi; 8699c3fd121SMatthew Ahrens err = dmu_object_info(os, resumeobj, &to_doi); 8709c3fd121SMatthew Ahrens if (err != 0) 8719c3fd121SMatthew Ahrens goto out; 8729c3fd121SMatthew Ahrens SET_BOOKMARK(&to_arg.resume, to_ds->ds_object, resumeobj, 0, 8739c3fd121SMatthew Ahrens resumeoff / to_doi.doi_data_block_size); 8749c3fd121SMatthew Ahrens 8759c3fd121SMatthew Ahrens nvlist_t *nvl = fnvlist_alloc(); 8769c3fd121SMatthew Ahrens fnvlist_add_uint64(nvl, "resume_object", resumeobj); 8779c3fd121SMatthew Ahrens fnvlist_add_uint64(nvl, "resume_offset", resumeoff); 8789c3fd121SMatthew Ahrens payload = fnvlist_pack(nvl, &payload_len); 8799c3fd121SMatthew Ahrens drr->drr_payloadlen = payload_len; 8809c3fd121SMatthew Ahrens fnvlist_free(nvl); 8819c3fd121SMatthew Ahrens } 8829c3fd121SMatthew Ahrens 8839c3fd121SMatthew Ahrens err = dump_record(dsp, payload, payload_len); 8849c3fd121SMatthew Ahrens fnvlist_pack_free(payload, payload_len); 8859c3fd121SMatthew Ahrens if (err != 0) { 8864e3c9f44SBill Pijewski err = dsp->dsa_err; 8874e3c9f44SBill Pijewski goto out; 888efb80947Sahrens } 889efb80947Sahrens 890a2cdcdd2SPaul Dagnelie err = bqueue_init(&to_arg.q, zfs_send_queue_length, 891a2cdcdd2SPaul Dagnelie offsetof(struct send_block_record, ln)); 892a2cdcdd2SPaul Dagnelie to_arg.error_code = 0; 893a2cdcdd2SPaul Dagnelie to_arg.cancel = B_FALSE; 894a2cdcdd2SPaul Dagnelie to_arg.ds = to_ds; 895a2cdcdd2SPaul Dagnelie to_arg.fromtxg = fromtxg; 896a2cdcdd2SPaul Dagnelie to_arg.flags = TRAVERSE_PRE | TRAVERSE_PREFETCH; 897a2cdcdd2SPaul Dagnelie (void) thread_create(NULL, 0, send_traverse_thread, &to_arg, 0, curproc, 898a2cdcdd2SPaul Dagnelie TS_RUN, minclsyspri); 899a2cdcdd2SPaul Dagnelie 900a2cdcdd2SPaul Dagnelie struct send_block_record *to_data; 901a2cdcdd2SPaul Dagnelie to_data = bqueue_dequeue(&to_arg.q); 902a2cdcdd2SPaul Dagnelie 903a2cdcdd2SPaul Dagnelie while (!to_data->eos_marker && err == 0) { 904a2cdcdd2SPaul Dagnelie err = do_dump(dsp, to_data); 905a2cdcdd2SPaul Dagnelie to_data = get_next_record(&to_arg.q, to_data); 906a2cdcdd2SPaul Dagnelie if (issig(JUSTLOOKING) && issig(FORREAL)) 907a2cdcdd2SPaul Dagnelie err = EINTR; 908a2cdcdd2SPaul Dagnelie } 909a2cdcdd2SPaul Dagnelie 910a2cdcdd2SPaul Dagnelie if (err != 0) { 911a2cdcdd2SPaul Dagnelie to_arg.cancel = B_TRUE; 912a2cdcdd2SPaul Dagnelie while (!to_data->eos_marker) { 913a2cdcdd2SPaul Dagnelie to_data = get_next_record(&to_arg.q, to_data); 914a2cdcdd2SPaul Dagnelie } 915a2cdcdd2SPaul Dagnelie } 916a2cdcdd2SPaul Dagnelie kmem_free(to_data, sizeof (*to_data)); 917a2cdcdd2SPaul Dagnelie 918a2cdcdd2SPaul Dagnelie bqueue_destroy(&to_arg.q); 919a2cdcdd2SPaul Dagnelie 920a2cdcdd2SPaul Dagnelie if (err == 0 && to_arg.error_code != 0) 921a2cdcdd2SPaul Dagnelie err = to_arg.error_code; 922a2cdcdd2SPaul Dagnelie 923a2cdcdd2SPaul Dagnelie if (err != 0) 924a2cdcdd2SPaul Dagnelie goto out; 925efb80947Sahrens 9264e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) 92798110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 928be6fd75aSMatthew Ahrens err = SET_ERROR(EINTR); 9299e69d7d0SLori Alt 9303b2aab18SMatthew Ahrens if (err != 0) { 9313b2aab18SMatthew Ahrens if (err == EINTR && dsp->dsa_err != 0) 9324e3c9f44SBill Pijewski err = dsp->dsa_err; 9334e3c9f44SBill Pijewski goto out; 934efb80947Sahrens } 935efb80947Sahrens 936efb80947Sahrens bzero(drr, sizeof (dmu_replay_record_t)); 937efb80947Sahrens drr->drr_type = DRR_END; 9384e3c9f44SBill Pijewski drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; 9394e3c9f44SBill Pijewski drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; 940efb80947Sahrens 941a2cdcdd2SPaul Dagnelie if (dump_record(dsp, NULL, 0) != 0) 9424e3c9f44SBill Pijewski err = dsp->dsa_err; 943efb80947Sahrens 9444e3c9f44SBill Pijewski out: 945a2cdcdd2SPaul Dagnelie mutex_enter(&to_ds->ds_sendstream_lock); 946a2cdcdd2SPaul Dagnelie list_remove(&to_ds->ds_sendstreams, dsp); 947a2cdcdd2SPaul Dagnelie mutex_exit(&to_ds->ds_sendstream_lock); 9484e3c9f44SBill Pijewski 94912b90ee2SMatt Krantz VERIFY(err != 0 || (dsp->dsa_sent_begin && dsp->dsa_sent_end)); 95012b90ee2SMatt Krantz 951efb80947Sahrens kmem_free(drr, sizeof (dmu_replay_record_t)); 9524e3c9f44SBill Pijewski kmem_free(dsp, sizeof (dmu_sendarg_t)); 953efb80947Sahrens 954a2cdcdd2SPaul Dagnelie dsl_dataset_long_rele(to_ds, FTAG); 9553b2aab18SMatthew Ahrens 9564e3c9f44SBill Pijewski return (err); 957efb80947Sahrens } 958efb80947Sahrens 95919b94df9SMatthew Ahrens int 9603b2aab18SMatthew Ahrens dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, 9615602294fSDan Kimmel boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, 962b5152584SMatthew Ahrens int outfd, vnode_t *vp, offset_t *off) 9633b2aab18SMatthew Ahrens { 9643b2aab18SMatthew Ahrens dsl_pool_t *dp; 9653b2aab18SMatthew Ahrens dsl_dataset_t *ds; 9663b2aab18SMatthew Ahrens dsl_dataset_t *fromds = NULL; 9673b2aab18SMatthew Ahrens int err; 9683b2aab18SMatthew Ahrens 9693b2aab18SMatthew Ahrens err = dsl_pool_hold(pool, FTAG, &dp); 9703b2aab18SMatthew Ahrens if (err != 0) 9713b2aab18SMatthew Ahrens return (err); 9723b2aab18SMatthew Ahrens 9733b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); 9743b2aab18SMatthew Ahrens if (err != 0) { 9753b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 9763b2aab18SMatthew Ahrens return (err); 9773b2aab18SMatthew Ahrens } 9783b2aab18SMatthew Ahrens 9793b2aab18SMatthew Ahrens if (fromsnap != 0) { 98078f17100SMatthew Ahrens zfs_bookmark_phys_t zb; 98178f17100SMatthew Ahrens boolean_t is_clone; 98278f17100SMatthew Ahrens 9833b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); 9843b2aab18SMatthew Ahrens if (err != 0) { 9853b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 9863b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 9873b2aab18SMatthew Ahrens return (err); 9883b2aab18SMatthew Ahrens } 98978f17100SMatthew Ahrens if (!dsl_dataset_is_before(ds, fromds, 0)) 99078f17100SMatthew Ahrens err = SET_ERROR(EXDEV); 991c1379625SJustin T. Gibbs zb.zbm_creation_time = 992c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_time; 993c1379625SJustin T. Gibbs zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; 994c1379625SJustin T. Gibbs zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 99578f17100SMatthew Ahrens is_clone = (fromds->ds_dir != ds->ds_dir); 99678f17100SMatthew Ahrens dsl_dataset_rele(fromds, FTAG); 997b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 9985602294fSDan Kimmel embedok, large_block_ok, compressok, outfd, 0, 0, vp, off); 99978f17100SMatthew Ahrens } else { 1000b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 10015602294fSDan Kimmel embedok, large_block_ok, compressok, outfd, 0, 0, vp, off); 10023b2aab18SMatthew Ahrens } 100378f17100SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 100478f17100SMatthew Ahrens return (err); 10053b2aab18SMatthew Ahrens } 10063b2aab18SMatthew Ahrens 10073b2aab18SMatthew Ahrens int 10089c3fd121SMatthew Ahrens dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, 10095602294fSDan Kimmel boolean_t large_block_ok, boolean_t compressok, int outfd, 10105602294fSDan Kimmel uint64_t resumeobj, uint64_t resumeoff, 10119c3fd121SMatthew Ahrens vnode_t *vp, offset_t *off) 10123b2aab18SMatthew Ahrens { 10133b2aab18SMatthew Ahrens dsl_pool_t *dp; 10143b2aab18SMatthew Ahrens dsl_dataset_t *ds; 10153b2aab18SMatthew Ahrens int err; 101678f17100SMatthew Ahrens boolean_t owned = B_FALSE; 10173b2aab18SMatthew Ahrens 101878f17100SMatthew Ahrens if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) 1019be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 10203b2aab18SMatthew Ahrens 10213b2aab18SMatthew Ahrens err = dsl_pool_hold(tosnap, FTAG, &dp); 10223b2aab18SMatthew Ahrens if (err != 0) 10233b2aab18SMatthew Ahrens return (err); 10243b2aab18SMatthew Ahrens 102578f17100SMatthew Ahrens if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { 102678f17100SMatthew Ahrens /* 102778f17100SMatthew Ahrens * We are sending a filesystem or volume. Ensure 102878f17100SMatthew Ahrens * that it doesn't change by owning the dataset. 102978f17100SMatthew Ahrens */ 103078f17100SMatthew Ahrens err = dsl_dataset_own(dp, tosnap, FTAG, &ds); 103178f17100SMatthew Ahrens owned = B_TRUE; 103278f17100SMatthew Ahrens } else { 103378f17100SMatthew Ahrens err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); 103478f17100SMatthew Ahrens } 10353b2aab18SMatthew Ahrens if (err != 0) { 10363b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 10373b2aab18SMatthew Ahrens return (err); 10383b2aab18SMatthew Ahrens } 10393b2aab18SMatthew Ahrens 10403b2aab18SMatthew Ahrens if (fromsnap != NULL) { 104178f17100SMatthew Ahrens zfs_bookmark_phys_t zb; 104278f17100SMatthew Ahrens boolean_t is_clone = B_FALSE; 104378f17100SMatthew Ahrens int fsnamelen = strchr(tosnap, '@') - tosnap; 104478f17100SMatthew Ahrens 104578f17100SMatthew Ahrens /* 104678f17100SMatthew Ahrens * If the fromsnap is in a different filesystem, then 104778f17100SMatthew Ahrens * mark the send stream as a clone. 104878f17100SMatthew Ahrens */ 104978f17100SMatthew Ahrens if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || 105078f17100SMatthew Ahrens (fromsnap[fsnamelen] != '@' && 105178f17100SMatthew Ahrens fromsnap[fsnamelen] != '#')) { 105278f17100SMatthew Ahrens is_clone = B_TRUE; 105378f17100SMatthew Ahrens } 105478f17100SMatthew Ahrens 105578f17100SMatthew Ahrens if (strchr(fromsnap, '@')) { 105678f17100SMatthew Ahrens dsl_dataset_t *fromds; 105778f17100SMatthew Ahrens err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); 105878f17100SMatthew Ahrens if (err == 0) { 105978f17100SMatthew Ahrens if (!dsl_dataset_is_before(ds, fromds, 0)) 106078f17100SMatthew Ahrens err = SET_ERROR(EXDEV); 106178f17100SMatthew Ahrens zb.zbm_creation_time = 1062c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_time; 106378f17100SMatthew Ahrens zb.zbm_creation_txg = 1064c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_txg; 1065c1379625SJustin T. Gibbs zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 106678f17100SMatthew Ahrens is_clone = (ds->ds_dir != fromds->ds_dir); 106778f17100SMatthew Ahrens dsl_dataset_rele(fromds, FTAG); 106878f17100SMatthew Ahrens } 106978f17100SMatthew Ahrens } else { 107078f17100SMatthew Ahrens err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); 107178f17100SMatthew Ahrens } 10723b2aab18SMatthew Ahrens if (err != 0) { 10733b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 10743b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 10753b2aab18SMatthew Ahrens return (err); 10763b2aab18SMatthew Ahrens } 1077b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 10785602294fSDan Kimmel embedok, large_block_ok, compressok, 10799c3fd121SMatthew Ahrens outfd, resumeobj, resumeoff, vp, off); 108078f17100SMatthew Ahrens } else { 1081b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 10825602294fSDan Kimmel embedok, large_block_ok, compressok, 10839c3fd121SMatthew Ahrens outfd, resumeobj, resumeoff, vp, off); 10843b2aab18SMatthew Ahrens } 108578f17100SMatthew Ahrens if (owned) 108678f17100SMatthew Ahrens dsl_dataset_disown(ds, FTAG); 108778f17100SMatthew Ahrens else 108878f17100SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 108978f17100SMatthew Ahrens return (err); 10903b2aab18SMatthew Ahrens } 10913b2aab18SMatthew Ahrens 1092643da460SMax Grossman static int 10935602294fSDan Kimmel dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t uncompressed, 10945602294fSDan Kimmel uint64_t compressed, boolean_t stream_compressed, uint64_t *sizep) 1095643da460SMax Grossman { 1096643da460SMax Grossman int err; 10975602294fSDan Kimmel uint64_t size; 1098643da460SMax Grossman /* 1099643da460SMax Grossman * Assume that space (both on-disk and in-stream) is dominated by 1100643da460SMax Grossman * data. We will adjust for indirect blocks and the copies property, 1101643da460SMax Grossman * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). 1102643da460SMax Grossman */ 11035602294fSDan Kimmel uint64_t recordsize; 11045602294fSDan Kimmel uint64_t record_count; 11055602294fSDan Kimmel 11065602294fSDan Kimmel /* Assume all (uncompressed) blocks are recordsize. */ 11075602294fSDan Kimmel err = dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 11085602294fSDan Kimmel &recordsize); 11095602294fSDan Kimmel if (err != 0) 11105602294fSDan Kimmel return (err); 11115602294fSDan Kimmel record_count = uncompressed / recordsize; 11125602294fSDan Kimmel 11135602294fSDan Kimmel /* 11145602294fSDan Kimmel * If we're estimating a send size for a compressed stream, use the 11155602294fSDan Kimmel * compressed data size to estimate the stream size. Otherwise, use the 11165602294fSDan Kimmel * uncompressed data size. 11175602294fSDan Kimmel */ 11185602294fSDan Kimmel size = stream_compressed ? compressed : uncompressed; 1119643da460SMax Grossman 1120643da460SMax Grossman /* 1121643da460SMax Grossman * Subtract out approximate space used by indirect blocks. 1122643da460SMax Grossman * Assume most space is used by data blocks (non-indirect, non-dnode). 11235602294fSDan Kimmel * Assume no ditto blocks or internal fragmentation. 1124643da460SMax Grossman * 1125643da460SMax Grossman * Therefore, space used by indirect blocks is sizeof(blkptr_t) per 11265602294fSDan Kimmel * block. 1127643da460SMax Grossman */ 11285602294fSDan Kimmel size -= record_count * sizeof (blkptr_t); 1129643da460SMax Grossman 1130643da460SMax Grossman /* Add in the space for the record associated with each block. */ 11315602294fSDan Kimmel size += record_count * sizeof (dmu_replay_record_t); 1132643da460SMax Grossman 1133643da460SMax Grossman *sizep = size; 1134643da460SMax Grossman 1135643da460SMax Grossman return (0); 1136643da460SMax Grossman } 1137643da460SMax Grossman 11383b2aab18SMatthew Ahrens int 11395602294fSDan Kimmel dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, 11405602294fSDan Kimmel boolean_t stream_compressed, uint64_t *sizep) 114119b94df9SMatthew Ahrens { 114219b94df9SMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 114319b94df9SMatthew Ahrens int err; 11445602294fSDan Kimmel uint64_t uncomp, comp; 114519b94df9SMatthew Ahrens 11463b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dp)); 11473b2aab18SMatthew Ahrens 114819b94df9SMatthew Ahrens /* tosnap must be a snapshot */ 1149bc9014e6SJustin Gibbs if (!ds->ds_is_snapshot) 1150be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 115119b94df9SMatthew Ahrens 115224218bebSAndriy Gapon /* fromsnap, if provided, must be a snapshot */ 115324218bebSAndriy Gapon if (fromds != NULL && !fromds->ds_is_snapshot) 115424218bebSAndriy Gapon return (SET_ERROR(EINVAL)); 115524218bebSAndriy Gapon 11564445fffbSMatthew Ahrens /* 11574445fffbSMatthew Ahrens * fromsnap must be an earlier snapshot from the same fs as tosnap, 11584445fffbSMatthew Ahrens * or the origin's fs. 11594445fffbSMatthew Ahrens */ 116078f17100SMatthew Ahrens if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) 1161be6fd75aSMatthew Ahrens return (SET_ERROR(EXDEV)); 116219b94df9SMatthew Ahrens 11635602294fSDan Kimmel /* Get compressed and uncompressed size estimates of changed data. */ 116419b94df9SMatthew Ahrens if (fromds == NULL) { 11655602294fSDan Kimmel uncomp = dsl_dataset_phys(ds)->ds_uncompressed_bytes; 11665602294fSDan Kimmel comp = dsl_dataset_phys(ds)->ds_compressed_bytes; 116719b94df9SMatthew Ahrens } else { 11685602294fSDan Kimmel uint64_t used; 116919b94df9SMatthew Ahrens err = dsl_dataset_space_written(fromds, ds, 11705602294fSDan Kimmel &used, &comp, &uncomp); 11713b2aab18SMatthew Ahrens if (err != 0) 117219b94df9SMatthew Ahrens return (err); 117319b94df9SMatthew Ahrens } 117419b94df9SMatthew Ahrens 11755602294fSDan Kimmel err = dmu_adjust_send_estimate_for_indirects(ds, uncomp, comp, 11765602294fSDan Kimmel stream_compressed, sizep); 1177643da460SMax Grossman return (err); 1178643da460SMax Grossman } 1179643da460SMax Grossman 11805602294fSDan Kimmel struct calculate_send_arg { 11815602294fSDan Kimmel uint64_t uncompressed; 11825602294fSDan Kimmel uint64_t compressed; 11835602294fSDan Kimmel }; 11845602294fSDan Kimmel 1185643da460SMax Grossman /* 1186643da460SMax Grossman * Simple callback used to traverse the blocks of a snapshot and sum their 11875602294fSDan Kimmel * uncompressed and compressed sizes. 1188643da460SMax Grossman */ 1189643da460SMax Grossman /* ARGSUSED */ 1190643da460SMax Grossman static int 1191643da460SMax Grossman dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1192643da460SMax Grossman const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1193643da460SMax Grossman { 11945602294fSDan Kimmel struct calculate_send_arg *space = arg; 1195643da460SMax Grossman if (bp != NULL && !BP_IS_HOLE(bp)) { 11965602294fSDan Kimmel space->uncompressed += BP_GET_UCSIZE(bp); 11975602294fSDan Kimmel space->compressed += BP_GET_PSIZE(bp); 1198643da460SMax Grossman } 1199643da460SMax Grossman return (0); 1200643da460SMax Grossman } 1201643da460SMax Grossman 1202643da460SMax Grossman /* 1203643da460SMax Grossman * Given a desination snapshot and a TXG, calculate the approximate size of a 1204643da460SMax Grossman * send stream sent from that TXG. from_txg may be zero, indicating that the 1205643da460SMax Grossman * whole snapshot will be sent. 1206643da460SMax Grossman */ 1207643da460SMax Grossman int 1208643da460SMax Grossman dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg, 12095602294fSDan Kimmel boolean_t stream_compressed, uint64_t *sizep) 1210643da460SMax Grossman { 1211643da460SMax Grossman dsl_pool_t *dp = ds->ds_dir->dd_pool; 1212643da460SMax Grossman int err; 12135602294fSDan Kimmel struct calculate_send_arg size = { 0 }; 1214643da460SMax Grossman 1215643da460SMax Grossman ASSERT(dsl_pool_config_held(dp)); 1216643da460SMax Grossman 1217643da460SMax Grossman /* tosnap must be a snapshot */ 12185602294fSDan Kimmel if (!ds->ds_is_snapshot) 1219643da460SMax Grossman return (SET_ERROR(EINVAL)); 1220643da460SMax Grossman 1221643da460SMax Grossman /* verify that from_txg is before the provided snapshot was taken */ 1222643da460SMax Grossman if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) { 1223643da460SMax Grossman return (SET_ERROR(EXDEV)); 1224643da460SMax Grossman } 122519b94df9SMatthew Ahrens 122619b94df9SMatthew Ahrens /* 1227643da460SMax Grossman * traverse the blocks of the snapshot with birth times after 1228643da460SMax Grossman * from_txg, summing their uncompressed size 122919b94df9SMatthew Ahrens */ 1230643da460SMax Grossman err = traverse_dataset(ds, from_txg, TRAVERSE_POST, 1231643da460SMax Grossman dmu_calculate_send_traversal, &size); 1232643da460SMax Grossman if (err) 123319b94df9SMatthew Ahrens return (err); 123419b94df9SMatthew Ahrens 12355602294fSDan Kimmel err = dmu_adjust_send_estimate_for_indirects(ds, size.uncompressed, 12365602294fSDan Kimmel size.compressed, stream_compressed, sizep); 1237643da460SMax Grossman return (err); 123819b94df9SMatthew Ahrens } 123919b94df9SMatthew Ahrens 12403b2aab18SMatthew Ahrens typedef struct dmu_recv_begin_arg { 12413b2aab18SMatthew Ahrens const char *drba_origin; 12423b2aab18SMatthew Ahrens dmu_recv_cookie_t *drba_cookie; 12433b2aab18SMatthew Ahrens cred_t *drba_cred; 124434f2f8cfSMatthew Ahrens uint64_t drba_snapobj; 12453b2aab18SMatthew Ahrens } dmu_recv_begin_arg_t; 1246f18faf3fSek 1247f18faf3fSek static int 12483b2aab18SMatthew Ahrens recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, 12493b2aab18SMatthew Ahrens uint64_t fromguid) 1250f18faf3fSek { 12513cb34c60Sahrens uint64_t val; 12523b2aab18SMatthew Ahrens int error; 12533b2aab18SMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 1254f18faf3fSek 12553b2aab18SMatthew Ahrens /* temporary clone name must not exist */ 12563b2aab18SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, 1257c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, 12583b2aab18SMatthew Ahrens 8, 1, &val); 12593b2aab18SMatthew Ahrens if (error != ENOENT) 12603b2aab18SMatthew Ahrens return (error == 0 ? EBUSY : error); 12613b2aab18SMatthew Ahrens 1262feaa74e4SMark Maybee /* new snapshot name must not exist */ 12633b2aab18SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, 1264c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1265c1379625SJustin T. Gibbs drba->drba_cookie->drc_tosnap, 8, 1, &val); 12663b2aab18SMatthew Ahrens if (error != ENOENT) 12673b2aab18SMatthew Ahrens return (error == 0 ? EEXIST : error); 1268feaa74e4SMark Maybee 1269a2afb611SJerry Jelinek /* 1270a2afb611SJerry Jelinek * Check snapshot limit before receiving. We'll recheck again at the 1271a2afb611SJerry Jelinek * end, but might as well abort before receiving if we're already over 1272a2afb611SJerry Jelinek * the limit. 1273a2afb611SJerry Jelinek * 1274a2afb611SJerry Jelinek * Note that we do not check the file system limit with 1275a2afb611SJerry Jelinek * dsl_dir_fscount_check because the temporary %clones don't count 1276a2afb611SJerry Jelinek * against that limit. 1277a2afb611SJerry Jelinek */ 1278a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, 1279a2afb611SJerry Jelinek NULL, drba->drba_cred); 1280a2afb611SJerry Jelinek if (error != 0) 1281a2afb611SJerry Jelinek return (error); 1282a2afb611SJerry Jelinek 12833b2aab18SMatthew Ahrens if (fromguid != 0) { 128434f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 1285c1379625SJustin T. Gibbs uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 128634f2f8cfSMatthew Ahrens 128734f2f8cfSMatthew Ahrens /* Find snapshot in this dir that matches fromguid. */ 128834f2f8cfSMatthew Ahrens while (obj != 0) { 128934f2f8cfSMatthew Ahrens error = dsl_dataset_hold_obj(dp, obj, FTAG, 129034f2f8cfSMatthew Ahrens &snap); 129134f2f8cfSMatthew Ahrens if (error != 0) 129234f2f8cfSMatthew Ahrens return (SET_ERROR(ENODEV)); 129334f2f8cfSMatthew Ahrens if (snap->ds_dir != ds->ds_dir) { 129434f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 129534f2f8cfSMatthew Ahrens return (SET_ERROR(ENODEV)); 129634f2f8cfSMatthew Ahrens } 1297c1379625SJustin T. Gibbs if (dsl_dataset_phys(snap)->ds_guid == fromguid) 129834f2f8cfSMatthew Ahrens break; 1299c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 130034f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 130134f2f8cfSMatthew Ahrens } 130234f2f8cfSMatthew Ahrens if (obj == 0) 1303be6fd75aSMatthew Ahrens return (SET_ERROR(ENODEV)); 130492241e0bSTom Erickson 130534f2f8cfSMatthew Ahrens if (drba->drba_cookie->drc_force) { 130634f2f8cfSMatthew Ahrens drba->drba_snapobj = obj; 130734f2f8cfSMatthew Ahrens } else { 130834f2f8cfSMatthew Ahrens /* 130934f2f8cfSMatthew Ahrens * If we are not forcing, there must be no 131034f2f8cfSMatthew Ahrens * changes since fromsnap. 131134f2f8cfSMatthew Ahrens */ 131234f2f8cfSMatthew Ahrens if (dsl_dataset_modified_since_snap(ds, snap)) { 131392241e0bSTom Erickson dsl_dataset_rele(snap, FTAG); 131434f2f8cfSMatthew Ahrens return (SET_ERROR(ETXTBSY)); 131592241e0bSTom Erickson } 131634f2f8cfSMatthew Ahrens drba->drba_snapobj = ds->ds_prev->ds_object; 131792241e0bSTom Erickson } 131834f2f8cfSMatthew Ahrens 131934f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 1320ae46e4c7SMatthew Ahrens } else { 13215bae108fSAndriy Gapon /* if full, then must be forced */ 13225bae108fSAndriy Gapon if (!drba->drba_cookie->drc_force) 13235bae108fSAndriy Gapon return (SET_ERROR(EEXIST)); 13245bae108fSAndriy Gapon /* start from $ORIGIN@$ORIGIN, if supported */ 13255bae108fSAndriy Gapon drba->drba_snapobj = dp->dp_origin_snap != NULL ? 13265bae108fSAndriy Gapon dp->dp_origin_snap->ds_object : 0; 1327ae46e4c7SMatthew Ahrens } 13283cb34c60Sahrens 13293cb34c60Sahrens return (0); 13303b2aab18SMatthew Ahrens 13313b2aab18SMatthew Ahrens } 13323b2aab18SMatthew Ahrens 13333b2aab18SMatthew Ahrens static int 13343b2aab18SMatthew Ahrens dmu_recv_begin_check(void *arg, dmu_tx_t *tx) 13353b2aab18SMatthew Ahrens { 13363b2aab18SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 13373b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 13383b2aab18SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 13393b2aab18SMatthew Ahrens uint64_t fromguid = drrb->drr_fromguid; 13403b2aab18SMatthew Ahrens int flags = drrb->drr_flags; 13413b2aab18SMatthew Ahrens int error; 13425d7b4d43SMatthew Ahrens uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 13433b2aab18SMatthew Ahrens dsl_dataset_t *ds; 13443b2aab18SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 13453b2aab18SMatthew Ahrens 13463b2aab18SMatthew Ahrens /* already checked */ 13473b2aab18SMatthew Ahrens ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 13489c3fd121SMatthew Ahrens ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); 13493b2aab18SMatthew Ahrens 13503b2aab18SMatthew Ahrens if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 13513b2aab18SMatthew Ahrens DMU_COMPOUNDSTREAM || 13523b2aab18SMatthew Ahrens drrb->drr_type >= DMU_OST_NUMTYPES || 13533b2aab18SMatthew Ahrens ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) 1354be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 13553b2aab18SMatthew Ahrens 13563b2aab18SMatthew Ahrens /* Verify pool version supports SA if SA_SPILL feature set */ 13575d7b4d43SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 13585d7b4d43SMatthew Ahrens spa_version(dp->dp_spa) < SPA_VERSION_SA) 13595d7b4d43SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 13605d7b4d43SMatthew Ahrens 13619c3fd121SMatthew Ahrens if (drba->drba_cookie->drc_resumable && 13629c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) 13639c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 13649c3fd121SMatthew Ahrens 13655d7b4d43SMatthew Ahrens /* 13665d7b4d43SMatthew Ahrens * The receiving code doesn't know how to translate a WRITE_EMBEDDED 13675602294fSDan Kimmel * record to a plain WRITE record, so the pool must have the 13685d7b4d43SMatthew Ahrens * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 13695d7b4d43SMatthew Ahrens * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 13705d7b4d43SMatthew Ahrens */ 13715d7b4d43SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 13725d7b4d43SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 13735d7b4d43SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 13745602294fSDan Kimmel if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && 13755d7b4d43SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 1376be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 13773b2aab18SMatthew Ahrens 1378b5152584SMatthew Ahrens /* 1379b5152584SMatthew Ahrens * The receiving code doesn't know how to translate large blocks 1380b5152584SMatthew Ahrens * to smaller ones, so the pool must have the LARGE_BLOCKS 1381b5152584SMatthew Ahrens * feature enabled if the stream has LARGE_BLOCKS. 1382b5152584SMatthew Ahrens */ 1383b5152584SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 1384b5152584SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 1385b5152584SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 1386b5152584SMatthew Ahrens 13873b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 13883b2aab18SMatthew Ahrens if (error == 0) { 13893b2aab18SMatthew Ahrens /* target fs already exists; recv into temp clone */ 13903b2aab18SMatthew Ahrens 13913b2aab18SMatthew Ahrens /* Can't recv a clone into an existing fs */ 139268ecb2ecSPaul Dagnelie if (flags & DRR_FLAG_CLONE || drba->drba_origin) { 13933b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1394be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 13953b2aab18SMatthew Ahrens } 13963b2aab18SMatthew Ahrens 13973b2aab18SMatthew Ahrens error = recv_begin_check_existing_impl(drba, ds, fromguid); 13983b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 13993b2aab18SMatthew Ahrens } else if (error == ENOENT) { 14003b2aab18SMatthew Ahrens /* target fs does not exist; must be a full backup or clone */ 14019adfa60dSMatthew Ahrens char buf[ZFS_MAX_DATASET_NAME_LEN]; 14023b2aab18SMatthew Ahrens 14033b2aab18SMatthew Ahrens /* 14043b2aab18SMatthew Ahrens * If it's a non-clone incremental, we are missing the 14053b2aab18SMatthew Ahrens * target fs, so fail the recv. 14063b2aab18SMatthew Ahrens */ 1407a2cdcdd2SPaul Dagnelie if (fromguid != 0 && !(flags & DRR_FLAG_CLONE || 1408a2cdcdd2SPaul Dagnelie drba->drba_origin)) 1409be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 14103b2aab18SMatthew Ahrens 141168ecb2ecSPaul Dagnelie /* 141268ecb2ecSPaul Dagnelie * If we're receiving a full send as a clone, and it doesn't 141368ecb2ecSPaul Dagnelie * contain all the necessary free records and freeobject 141468ecb2ecSPaul Dagnelie * records, reject it. 141568ecb2ecSPaul Dagnelie */ 141668ecb2ecSPaul Dagnelie if (fromguid == 0 && drba->drba_origin && 141768ecb2ecSPaul Dagnelie !(flags & DRR_FLAG_FREERECORDS)) 141868ecb2ecSPaul Dagnelie return (SET_ERROR(EINVAL)); 141968ecb2ecSPaul Dagnelie 14203b2aab18SMatthew Ahrens /* Open the parent of tofs */ 14219adfa60dSMatthew Ahrens ASSERT3U(strlen(tofs), <, sizeof (buf)); 14223b2aab18SMatthew Ahrens (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); 14233b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, buf, FTAG, &ds); 14243b2aab18SMatthew Ahrens if (error != 0) 14253b2aab18SMatthew Ahrens return (error); 14263b2aab18SMatthew Ahrens 1427a2afb611SJerry Jelinek /* 1428a2afb611SJerry Jelinek * Check filesystem and snapshot limits before receiving. We'll 1429a2afb611SJerry Jelinek * recheck snapshot limits again at the end (we create the 1430a2afb611SJerry Jelinek * filesystems and increment those counts during begin_sync). 1431a2afb611SJerry Jelinek */ 1432a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 1433a2afb611SJerry Jelinek ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); 1434a2afb611SJerry Jelinek if (error != 0) { 1435a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 1436a2afb611SJerry Jelinek return (error); 1437a2afb611SJerry Jelinek } 1438a2afb611SJerry Jelinek 1439a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 1440a2afb611SJerry Jelinek ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); 1441a2afb611SJerry Jelinek if (error != 0) { 1442a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 1443a2afb611SJerry Jelinek return (error); 1444a2afb611SJerry Jelinek } 1445a2afb611SJerry Jelinek 14463b2aab18SMatthew Ahrens if (drba->drba_origin != NULL) { 14473b2aab18SMatthew Ahrens dsl_dataset_t *origin; 14483b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, drba->drba_origin, 14493b2aab18SMatthew Ahrens FTAG, &origin); 14503b2aab18SMatthew Ahrens if (error != 0) { 14513b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14523b2aab18SMatthew Ahrens return (error); 14533b2aab18SMatthew Ahrens } 1454bc9014e6SJustin Gibbs if (!origin->ds_is_snapshot) { 14553b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 14563b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1457be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 14583b2aab18SMatthew Ahrens } 145968ecb2ecSPaul Dagnelie if (dsl_dataset_phys(origin)->ds_guid != fromguid && 146068ecb2ecSPaul Dagnelie fromguid != 0) { 14613b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 14623b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1463be6fd75aSMatthew Ahrens return (SET_ERROR(ENODEV)); 14643b2aab18SMatthew Ahrens } 14653b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 14663b2aab18SMatthew Ahrens } 14673b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14683b2aab18SMatthew Ahrens error = 0; 14693b2aab18SMatthew Ahrens } 14703b2aab18SMatthew Ahrens return (error); 1471f18faf3fSek } 1472f18faf3fSek 1473f18faf3fSek static void 14743b2aab18SMatthew Ahrens dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) 1475f18faf3fSek { 14763b2aab18SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 14773b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 14789c3fd121SMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 14793b2aab18SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 14803b2aab18SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 14813b2aab18SMatthew Ahrens dsl_dataset_t *ds, *newds; 1482f18faf3fSek uint64_t dsobj; 14833b2aab18SMatthew Ahrens int error; 14849c3fd121SMatthew Ahrens uint64_t crflags = 0; 14853b2aab18SMatthew Ahrens 14869c3fd121SMatthew Ahrens if (drrb->drr_flags & DRR_FLAG_CI_DATA) 14879c3fd121SMatthew Ahrens crflags |= DS_FLAG_CI_DATASET; 1488f18faf3fSek 14893b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 14903b2aab18SMatthew Ahrens if (error == 0) { 14913b2aab18SMatthew Ahrens /* create temporary clone */ 149234f2f8cfSMatthew Ahrens dsl_dataset_t *snap = NULL; 149334f2f8cfSMatthew Ahrens if (drba->drba_snapobj != 0) { 149434f2f8cfSMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 149534f2f8cfSMatthew Ahrens drba->drba_snapobj, FTAG, &snap)); 149634f2f8cfSMatthew Ahrens } 14973b2aab18SMatthew Ahrens dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, 149834f2f8cfSMatthew Ahrens snap, crflags, drba->drba_cred, tx); 1499f40b29ceSPaul Dagnelie if (drba->drba_snapobj != 0) 1500f40b29ceSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 15013b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15023b2aab18SMatthew Ahrens } else { 15033b2aab18SMatthew Ahrens dsl_dir_t *dd; 15043b2aab18SMatthew Ahrens const char *tail; 15053b2aab18SMatthew Ahrens dsl_dataset_t *origin = NULL; 15063b2aab18SMatthew Ahrens 15073b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); 15083b2aab18SMatthew Ahrens 15093b2aab18SMatthew Ahrens if (drba->drba_origin != NULL) { 15103b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, 15113b2aab18SMatthew Ahrens FTAG, &origin)); 15123b2aab18SMatthew Ahrens } 15133b2aab18SMatthew Ahrens 15143b2aab18SMatthew Ahrens /* Create new dataset. */ 15153b2aab18SMatthew Ahrens dsobj = dsl_dataset_create_sync(dd, 15163b2aab18SMatthew Ahrens strrchr(tofs, '/') + 1, 15173b2aab18SMatthew Ahrens origin, crflags, drba->drba_cred, tx); 15183b2aab18SMatthew Ahrens if (origin != NULL) 15193b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 15203b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 15213b2aab18SMatthew Ahrens drba->drba_cookie->drc_newfs = B_TRUE; 15223b2aab18SMatthew Ahrens } 15233b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); 15243b2aab18SMatthew Ahrens 15259c3fd121SMatthew Ahrens if (drba->drba_cookie->drc_resumable) { 15269c3fd121SMatthew Ahrens dsl_dataset_zapify(newds, tx); 15279c3fd121SMatthew Ahrens if (drrb->drr_fromguid != 0) { 15289c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, 15299c3fd121SMatthew Ahrens 8, 1, &drrb->drr_fromguid, tx)); 15309c3fd121SMatthew Ahrens } 15319c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, 15329c3fd121SMatthew Ahrens 8, 1, &drrb->drr_toguid, tx)); 15339c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, 15349c3fd121SMatthew Ahrens 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); 15359c3fd121SMatthew Ahrens uint64_t one = 1; 15369c3fd121SMatthew Ahrens uint64_t zero = 0; 15379c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, 15389c3fd121SMatthew Ahrens 8, 1, &one, tx)); 15399c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, 15409c3fd121SMatthew Ahrens 8, 1, &zero, tx)); 15419c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, 15429c3fd121SMatthew Ahrens 8, 1, &zero, tx)); 15435602294fSDan Kimmel if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 15445602294fSDan Kimmel DMU_BACKUP_FEATURE_LARGE_BLOCKS) { 15455602294fSDan Kimmel VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_LARGEBLOCK, 15465602294fSDan Kimmel 8, 1, &one, tx)); 15475602294fSDan Kimmel } 15489c3fd121SMatthew Ahrens if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 15499c3fd121SMatthew Ahrens DMU_BACKUP_FEATURE_EMBED_DATA) { 15509c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, 15519c3fd121SMatthew Ahrens 8, 1, &one, tx)); 15529c3fd121SMatthew Ahrens } 15535602294fSDan Kimmel if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 15545602294fSDan Kimmel DMU_BACKUP_FEATURE_COMPRESSED) { 15555602294fSDan Kimmel VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_COMPRESSOK, 15565602294fSDan Kimmel 8, 1, &one, tx)); 15575602294fSDan Kimmel } 15589c3fd121SMatthew Ahrens } 15599c3fd121SMatthew Ahrens 15603b2aab18SMatthew Ahrens dmu_buf_will_dirty(newds->ds_dbuf, tx); 1561c1379625SJustin T. Gibbs dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; 15623cb34c60Sahrens 1563ae46e4c7SMatthew Ahrens /* 1564ae46e4c7SMatthew Ahrens * If we actually created a non-clone, we need to create the 1565ae46e4c7SMatthew Ahrens * objset in our new dataset. 1566ae46e4c7SMatthew Ahrens */ 1567c166b69dSPaul Dagnelie rrw_enter(&newds->ds_bp_rwlock, RW_READER, FTAG); 15683b2aab18SMatthew Ahrens if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { 1569ae46e4c7SMatthew Ahrens (void) dmu_objset_create_impl(dp->dp_spa, 15703b2aab18SMatthew Ahrens newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); 1571ae46e4c7SMatthew Ahrens } 1572c166b69dSPaul Dagnelie rrw_exit(&newds->ds_bp_rwlock, FTAG); 1573ae46e4c7SMatthew Ahrens 15743b2aab18SMatthew Ahrens drba->drba_cookie->drc_ds = newds; 1575dc7cd546SMark Shellenbaum 15763b2aab18SMatthew Ahrens spa_history_log_internal_ds(newds, "receive", tx, ""); 1577dc7cd546SMark Shellenbaum } 1578dc7cd546SMark Shellenbaum 15799c3fd121SMatthew Ahrens static int 15809c3fd121SMatthew Ahrens dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) 15819c3fd121SMatthew Ahrens { 15829c3fd121SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 15839c3fd121SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 15849c3fd121SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 15859c3fd121SMatthew Ahrens int error; 15869c3fd121SMatthew Ahrens uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 15879c3fd121SMatthew Ahrens dsl_dataset_t *ds; 15889c3fd121SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 15899c3fd121SMatthew Ahrens 15909c3fd121SMatthew Ahrens /* already checked */ 15919c3fd121SMatthew Ahrens ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 15929c3fd121SMatthew Ahrens ASSERT(featureflags & DMU_BACKUP_FEATURE_RESUMING); 15939c3fd121SMatthew Ahrens 15949c3fd121SMatthew Ahrens if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 15959c3fd121SMatthew Ahrens DMU_COMPOUNDSTREAM || 15969c3fd121SMatthew Ahrens drrb->drr_type >= DMU_OST_NUMTYPES) 15979c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 15989c3fd121SMatthew Ahrens 15999c3fd121SMatthew Ahrens /* Verify pool version supports SA if SA_SPILL feature set */ 16009c3fd121SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 16019c3fd121SMatthew Ahrens spa_version(dp->dp_spa) < SPA_VERSION_SA) 16029c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 16039c3fd121SMatthew Ahrens 16049c3fd121SMatthew Ahrens /* 16059c3fd121SMatthew Ahrens * The receiving code doesn't know how to translate a WRITE_EMBEDDED 16069c3fd121SMatthew Ahrens * record to a plain WRITE record, so the pool must have the 16079c3fd121SMatthew Ahrens * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 16089c3fd121SMatthew Ahrens * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 16099c3fd121SMatthew Ahrens */ 16109c3fd121SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 16119c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 16129c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 16135602294fSDan Kimmel if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && 16149c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 16159c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 16169c3fd121SMatthew Ahrens 16179adfa60dSMatthew Ahrens /* 6 extra bytes for /%recv */ 16189adfa60dSMatthew Ahrens char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 16199c3fd121SMatthew Ahrens 16209c3fd121SMatthew Ahrens (void) snprintf(recvname, sizeof (recvname), "%s/%s", 16219c3fd121SMatthew Ahrens tofs, recv_clone_name); 16229c3fd121SMatthew Ahrens 16239c3fd121SMatthew Ahrens if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { 16249c3fd121SMatthew Ahrens /* %recv does not exist; continue in tofs */ 16259c3fd121SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 16269c3fd121SMatthew Ahrens if (error != 0) 16279c3fd121SMatthew Ahrens return (error); 16289c3fd121SMatthew Ahrens } 16299c3fd121SMatthew Ahrens 16309c3fd121SMatthew Ahrens /* check that ds is marked inconsistent */ 16319c3fd121SMatthew Ahrens if (!DS_IS_INCONSISTENT(ds)) { 16329c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16339c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16349c3fd121SMatthew Ahrens } 16359c3fd121SMatthew Ahrens 16369c3fd121SMatthew Ahrens /* check that there is resuming data, and that the toguid matches */ 16379c3fd121SMatthew Ahrens if (!dsl_dataset_is_zapified(ds)) { 16389c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16399c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16409c3fd121SMatthew Ahrens } 16419c3fd121SMatthew Ahrens uint64_t val; 16429c3fd121SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, ds->ds_object, 16439c3fd121SMatthew Ahrens DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); 16449c3fd121SMatthew Ahrens if (error != 0 || drrb->drr_toguid != val) { 16459c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16469c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16479c3fd121SMatthew Ahrens } 16489c3fd121SMatthew Ahrens 16499c3fd121SMatthew Ahrens /* 16509c3fd121SMatthew Ahrens * Check if the receive is still running. If so, it will be owned. 16519c3fd121SMatthew Ahrens * Note that nothing else can own the dataset (e.g. after the receive 16529c3fd121SMatthew Ahrens * fails) because it will be marked inconsistent. 16539c3fd121SMatthew Ahrens */ 16549c3fd121SMatthew Ahrens if (dsl_dataset_has_owner(ds)) { 16559c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16569c3fd121SMatthew Ahrens return (SET_ERROR(EBUSY)); 16579c3fd121SMatthew Ahrens } 16589c3fd121SMatthew Ahrens 16599c3fd121SMatthew Ahrens /* There should not be any snapshots of this fs yet. */ 16609c3fd121SMatthew Ahrens if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { 16619c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16629c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16639c3fd121SMatthew Ahrens } 16649c3fd121SMatthew Ahrens 16659c3fd121SMatthew Ahrens /* 16669c3fd121SMatthew Ahrens * Note: resume point will be checked when we process the first WRITE 16679c3fd121SMatthew Ahrens * record. 16689c3fd121SMatthew Ahrens */ 16699c3fd121SMatthew Ahrens 16709c3fd121SMatthew Ahrens /* check that the origin matches */ 16719c3fd121SMatthew Ahrens val = 0; 16729c3fd121SMatthew Ahrens (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, 16739c3fd121SMatthew Ahrens DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); 16749c3fd121SMatthew Ahrens if (drrb->drr_fromguid != val) { 16759c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16769c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16779c3fd121SMatthew Ahrens } 16789c3fd121SMatthew Ahrens 16799c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16809c3fd121SMatthew Ahrens return (0); 16819c3fd121SMatthew Ahrens } 16829c3fd121SMatthew Ahrens 16839c3fd121SMatthew Ahrens static void 16849c3fd121SMatthew Ahrens dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) 16859c3fd121SMatthew Ahrens { 16869c3fd121SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 16879c3fd121SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 16889c3fd121SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 16899c3fd121SMatthew Ahrens dsl_dataset_t *ds; 16909c3fd121SMatthew Ahrens uint64_t dsobj; 16919adfa60dSMatthew Ahrens /* 6 extra bytes for /%recv */ 16929adfa60dSMatthew Ahrens char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 16939c3fd121SMatthew Ahrens 16949c3fd121SMatthew Ahrens (void) snprintf(recvname, sizeof (recvname), "%s/%s", 16959c3fd121SMatthew Ahrens tofs, recv_clone_name); 16969c3fd121SMatthew Ahrens 16979c3fd121SMatthew Ahrens if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { 16989c3fd121SMatthew Ahrens /* %recv does not exist; continue in tofs */ 16999c3fd121SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, tofs, FTAG, &ds)); 17009c3fd121SMatthew Ahrens drba->drba_cookie->drc_newfs = B_TRUE; 17019c3fd121SMatthew Ahrens } 17029c3fd121SMatthew Ahrens 17039c3fd121SMatthew Ahrens /* clear the inconsistent flag so that we can own it */ 17049c3fd121SMatthew Ahrens ASSERT(DS_IS_INCONSISTENT(ds)); 17059c3fd121SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 17069c3fd121SMatthew Ahrens dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 17079c3fd121SMatthew Ahrens dsobj = ds->ds_object; 17089c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 17099c3fd121SMatthew Ahrens 17109c3fd121SMatthew Ahrens VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &ds)); 17119c3fd121SMatthew Ahrens 17129c3fd121SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 17139c3fd121SMatthew Ahrens dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT; 17149c3fd121SMatthew Ahrens 1715c166b69dSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 17169c3fd121SMatthew Ahrens ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds))); 1717c166b69dSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 17189c3fd121SMatthew Ahrens 17199c3fd121SMatthew Ahrens drba->drba_cookie->drc_ds = ds; 17209c3fd121SMatthew Ahrens 17219c3fd121SMatthew Ahrens spa_history_log_internal_ds(ds, "resume receive", tx, ""); 17229c3fd121SMatthew Ahrens } 17239c3fd121SMatthew Ahrens 17243cb34c60Sahrens /* 17253cb34c60Sahrens * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() 17263cb34c60Sahrens * succeeds; otherwise we will leak the holds on the datasets. 17273cb34c60Sahrens */ 17283cb34c60Sahrens int 17299c3fd121SMatthew Ahrens dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, 17309c3fd121SMatthew Ahrens boolean_t force, boolean_t resumable, char *origin, dmu_recv_cookie_t *drc) 1731efb80947Sahrens { 17323b2aab18SMatthew Ahrens dmu_recv_begin_arg_t drba = { 0 }; 1733ab04eb8eStimh 17343cb34c60Sahrens bzero(drc, sizeof (dmu_recv_cookie_t)); 17359c3fd121SMatthew Ahrens drc->drc_drr_begin = drr_begin; 17369c3fd121SMatthew Ahrens drc->drc_drrb = &drr_begin->drr_u.drr_begin; 17373cb34c60Sahrens drc->drc_tosnap = tosnap; 17383b2aab18SMatthew Ahrens drc->drc_tofs = tofs; 17393cb34c60Sahrens drc->drc_force = force; 17409c3fd121SMatthew Ahrens drc->drc_resumable = resumable; 1741a2afb611SJerry Jelinek drc->drc_cred = CRED(); 1742efb80947Sahrens 17439c3fd121SMatthew Ahrens if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { 17443b2aab18SMatthew Ahrens drc->drc_byteswap = B_TRUE; 1745*770499e1SDan Kimmel (void) fletcher_4_incremental_byteswap(drr_begin, 17463b2aab18SMatthew Ahrens sizeof (dmu_replay_record_t), &drc->drc_cksum); 17479c3fd121SMatthew Ahrens byteswap_record(drr_begin); 17489c3fd121SMatthew Ahrens } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { 1749*770499e1SDan Kimmel (void) fletcher_4_incremental_native(drr_begin, 17503b2aab18SMatthew Ahrens sizeof (dmu_replay_record_t), &drc->drc_cksum); 17519c3fd121SMatthew Ahrens } else { 17529c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 17533cb34c60Sahrens } 17543cb34c60Sahrens 17553b2aab18SMatthew Ahrens drba.drba_origin = origin; 17563b2aab18SMatthew Ahrens drba.drba_cookie = drc; 17573b2aab18SMatthew Ahrens drba.drba_cred = CRED(); 17583b2aab18SMatthew Ahrens 17599c3fd121SMatthew Ahrens if (DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & 17609c3fd121SMatthew Ahrens DMU_BACKUP_FEATURE_RESUMING) { 17619c3fd121SMatthew Ahrens return (dsl_sync_task(tofs, 17629c3fd121SMatthew Ahrens dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, 17639c3fd121SMatthew Ahrens &drba, 5, ZFS_SPACE_CHECK_NORMAL)); 17649c3fd121SMatthew Ahrens } else { 17659c3fd121SMatthew Ahrens return (dsl_sync_task(tofs, 17669c3fd121SMatthew Ahrens dmu_recv_begin_check, dmu_recv_begin_sync, 17679c3fd121SMatthew Ahrens &drba, 5, ZFS_SPACE_CHECK_NORMAL)); 17689c3fd121SMatthew Ahrens } 1769efb80947Sahrens } 1770efb80947Sahrens 1771a2cdcdd2SPaul Dagnelie struct receive_record_arg { 1772a2cdcdd2SPaul Dagnelie dmu_replay_record_t header; 1773a2cdcdd2SPaul Dagnelie void *payload; /* Pointer to a buffer containing the payload */ 1774a2cdcdd2SPaul Dagnelie /* 1775a2cdcdd2SPaul Dagnelie * If the record is a write, pointer to the arc_buf_t containing the 1776a2cdcdd2SPaul Dagnelie * payload. 1777a2cdcdd2SPaul Dagnelie */ 1778a2cdcdd2SPaul Dagnelie arc_buf_t *write_buf; 1779a2cdcdd2SPaul Dagnelie int payload_size; 17809c3fd121SMatthew Ahrens uint64_t bytes_read; /* bytes read from stream when record created */ 1781a2cdcdd2SPaul Dagnelie boolean_t eos_marker; /* Marks the end of the stream */ 1782a2cdcdd2SPaul Dagnelie bqueue_node_t node; 1783a2cdcdd2SPaul Dagnelie }; 1784a2cdcdd2SPaul Dagnelie 1785a2cdcdd2SPaul Dagnelie struct receive_writer_arg { 178698110f08SMatthew Ahrens objset_t *os; 17873b2aab18SMatthew Ahrens boolean_t byteswap; 1788a2cdcdd2SPaul Dagnelie bqueue_t q; 17899c3fd121SMatthew Ahrens 1790a2cdcdd2SPaul Dagnelie /* 1791a2cdcdd2SPaul Dagnelie * These three args are used to signal to the main thread that we're 1792a2cdcdd2SPaul Dagnelie * done. 1793a2cdcdd2SPaul Dagnelie */ 1794a2cdcdd2SPaul Dagnelie kmutex_t mutex; 1795a2cdcdd2SPaul Dagnelie kcondvar_t cv; 1796a2cdcdd2SPaul Dagnelie boolean_t done; 17979c3fd121SMatthew Ahrens 1798a2cdcdd2SPaul Dagnelie int err; 1799a2cdcdd2SPaul Dagnelie /* A map from guid to dataset to help handle dedup'd streams. */ 1800a2cdcdd2SPaul Dagnelie avl_tree_t *guid_to_ds_map; 18019c3fd121SMatthew Ahrens boolean_t resumable; 18029c3fd121SMatthew Ahrens uint64_t last_object, last_offset; 18039c3fd121SMatthew Ahrens uint64_t bytes_read; /* bytes read when current record created */ 1804a2cdcdd2SPaul Dagnelie }; 180598110f08SMatthew Ahrens 180668ecb2ecSPaul Dagnelie struct objlist { 180768ecb2ecSPaul Dagnelie list_t list; /* List of struct receive_objnode. */ 180868ecb2ecSPaul Dagnelie /* 180968ecb2ecSPaul Dagnelie * Last object looked up. Used to assert that objects are being looked 181068ecb2ecSPaul Dagnelie * up in ascending order. 181168ecb2ecSPaul Dagnelie */ 181268ecb2ecSPaul Dagnelie uint64_t last_lookup; 181368ecb2ecSPaul Dagnelie }; 181468ecb2ecSPaul Dagnelie 181568ecb2ecSPaul Dagnelie struct receive_objnode { 181668ecb2ecSPaul Dagnelie list_node_t node; 181768ecb2ecSPaul Dagnelie uint64_t object; 181868ecb2ecSPaul Dagnelie }; 181968ecb2ecSPaul Dagnelie 18205602294fSDan Kimmel struct receive_arg { 1821a2cdcdd2SPaul Dagnelie objset_t *os; 1822a2cdcdd2SPaul Dagnelie vnode_t *vp; /* The vnode to read the stream from */ 1823a2cdcdd2SPaul Dagnelie uint64_t voff; /* The current offset in the stream */ 18249c3fd121SMatthew Ahrens uint64_t bytes_read; 1825a2cdcdd2SPaul Dagnelie /* 1826a2cdcdd2SPaul Dagnelie * A record that has had its payload read in, but hasn't yet been handed 1827a2cdcdd2SPaul Dagnelie * off to the worker thread. 1828a2cdcdd2SPaul Dagnelie */ 1829a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd; 1830a2cdcdd2SPaul Dagnelie /* A record that has had its header read in, but not its payload. */ 1831a2cdcdd2SPaul Dagnelie struct receive_record_arg *next_rrd; 18323cb34c60Sahrens zio_cksum_t cksum; 183398110f08SMatthew Ahrens zio_cksum_t prev_cksum; 1834a2cdcdd2SPaul Dagnelie int err; 1835a2cdcdd2SPaul Dagnelie boolean_t byteswap; 1836a2cdcdd2SPaul Dagnelie /* Sorted list of objects not to issue prefetches for. */ 183768ecb2ecSPaul Dagnelie struct objlist ignore_objlist; 18383cb34c60Sahrens }; 18393cb34c60Sahrens 18409e69d7d0SLori Alt typedef struct guid_map_entry { 18419e69d7d0SLori Alt uint64_t guid; 18429e69d7d0SLori Alt dsl_dataset_t *gme_ds; 18439e69d7d0SLori Alt avl_node_t avlnode; 18449e69d7d0SLori Alt } guid_map_entry_t; 18459e69d7d0SLori Alt 18469e69d7d0SLori Alt static int 18479e69d7d0SLori Alt guid_compare(const void *arg1, const void *arg2) 18489e69d7d0SLori Alt { 18499e69d7d0SLori Alt const guid_map_entry_t *gmep1 = arg1; 18509e69d7d0SLori Alt const guid_map_entry_t *gmep2 = arg2; 18519e69d7d0SLori Alt 18529e69d7d0SLori Alt if (gmep1->guid < gmep2->guid) 18539e69d7d0SLori Alt return (-1); 18549e69d7d0SLori Alt else if (gmep1->guid > gmep2->guid) 18559e69d7d0SLori Alt return (1); 18569e69d7d0SLori Alt return (0); 18579e69d7d0SLori Alt } 18589e69d7d0SLori Alt 1859c99e4bdcSChris Kirby static void 1860c99e4bdcSChris Kirby free_guid_map_onexit(void *arg) 1861c99e4bdcSChris Kirby { 1862c99e4bdcSChris Kirby avl_tree_t *ca = arg; 1863c99e4bdcSChris Kirby void *cookie = NULL; 1864c99e4bdcSChris Kirby guid_map_entry_t *gmep; 1865c99e4bdcSChris Kirby 1866c99e4bdcSChris Kirby while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { 18673b2aab18SMatthew Ahrens dsl_dataset_long_rele(gmep->gme_ds, gmep); 1868de8d9cffSMatthew Ahrens dsl_dataset_rele(gmep->gme_ds, gmep); 1869c99e4bdcSChris Kirby kmem_free(gmep, sizeof (guid_map_entry_t)); 1870c99e4bdcSChris Kirby } 1871c99e4bdcSChris Kirby avl_destroy(ca); 1872c99e4bdcSChris Kirby kmem_free(ca, sizeof (avl_tree_t)); 1873c99e4bdcSChris Kirby } 1874c99e4bdcSChris Kirby 187598110f08SMatthew Ahrens static int 1876a2cdcdd2SPaul Dagnelie receive_read(struct receive_arg *ra, int len, void *buf) 1877efb80947Sahrens { 18783cb34c60Sahrens int done = 0; 1879efb80947Sahrens 1880c20404ffSEli Rosenthal /* 1881c20404ffSEli Rosenthal * The code doesn't rely on this (lengths being multiples of 8). See 1882c20404ffSEli Rosenthal * comment in dump_bytes. 1883c20404ffSEli Rosenthal */ 1884fb09f5aaSMadhav Suresh ASSERT0(len % 8); 1885efb80947Sahrens 18863cb34c60Sahrens while (done < len) { 1887efb80947Sahrens ssize_t resid; 1888efb80947Sahrens 1889efb80947Sahrens ra->err = vn_rdwr(UIO_READ, ra->vp, 189098110f08SMatthew Ahrens (char *)buf + done, len - done, 1891efb80947Sahrens ra->voff, UIO_SYSSPACE, FAPPEND, 1892efb80947Sahrens RLIM64_INFINITY, CRED(), &resid); 1893efb80947Sahrens 18949c3fd121SMatthew Ahrens if (resid == len - done) { 18959c3fd121SMatthew Ahrens /* 18969c3fd121SMatthew Ahrens * Note: ECKSUM indicates that the receive 18979c3fd121SMatthew Ahrens * was interrupted and can potentially be resumed. 18989c3fd121SMatthew Ahrens */ 18999c3fd121SMatthew Ahrens ra->err = SET_ERROR(ECKSUM); 19009c3fd121SMatthew Ahrens } 19013cb34c60Sahrens ra->voff += len - done - resid; 19023cb34c60Sahrens done = len - resid; 19033b2aab18SMatthew Ahrens if (ra->err != 0) 190498110f08SMatthew Ahrens return (ra->err); 1905efb80947Sahrens } 1906efb80947Sahrens 19079c3fd121SMatthew Ahrens ra->bytes_read += len; 19089c3fd121SMatthew Ahrens 19093cb34c60Sahrens ASSERT3U(done, ==, len); 191098110f08SMatthew Ahrens return (0); 1911efb80947Sahrens } 1912efb80947Sahrens 1913efb80947Sahrens static void 191498110f08SMatthew Ahrens byteswap_record(dmu_replay_record_t *drr) 1915efb80947Sahrens { 1916efb80947Sahrens #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) 1917efb80947Sahrens #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) 1918efb80947Sahrens drr->drr_type = BSWAP_32(drr->drr_type); 19193cb34c60Sahrens drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); 192098110f08SMatthew Ahrens 1921efb80947Sahrens switch (drr->drr_type) { 1922efb80947Sahrens case DRR_BEGIN: 1923efb80947Sahrens DO64(drr_begin.drr_magic); 19249e69d7d0SLori Alt DO64(drr_begin.drr_versioninfo); 1925efb80947Sahrens DO64(drr_begin.drr_creation_time); 1926efb80947Sahrens DO32(drr_begin.drr_type); 19273cb34c60Sahrens DO32(drr_begin.drr_flags); 1928efb80947Sahrens DO64(drr_begin.drr_toguid); 1929efb80947Sahrens DO64(drr_begin.drr_fromguid); 1930efb80947Sahrens break; 1931efb80947Sahrens case DRR_OBJECT: 1932efb80947Sahrens DO64(drr_object.drr_object); 1933efb80947Sahrens DO32(drr_object.drr_type); 1934efb80947Sahrens DO32(drr_object.drr_bonustype); 1935efb80947Sahrens DO32(drr_object.drr_blksz); 1936efb80947Sahrens DO32(drr_object.drr_bonuslen); 19379e69d7d0SLori Alt DO64(drr_object.drr_toguid); 1938efb80947Sahrens break; 1939efb80947Sahrens case DRR_FREEOBJECTS: 1940efb80947Sahrens DO64(drr_freeobjects.drr_firstobj); 1941efb80947Sahrens DO64(drr_freeobjects.drr_numobjs); 19429e69d7d0SLori Alt DO64(drr_freeobjects.drr_toguid); 1943efb80947Sahrens break; 1944efb80947Sahrens case DRR_WRITE: 1945efb80947Sahrens DO64(drr_write.drr_object); 1946efb80947Sahrens DO32(drr_write.drr_type); 1947efb80947Sahrens DO64(drr_write.drr_offset); 19485602294fSDan Kimmel DO64(drr_write.drr_logical_size); 19499e69d7d0SLori Alt DO64(drr_write.drr_toguid); 195098110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); 19518e714474SLori Alt DO64(drr_write.drr_key.ddk_prop); 19525602294fSDan Kimmel DO64(drr_write.drr_compressed_size); 19539e69d7d0SLori Alt break; 19549e69d7d0SLori Alt case DRR_WRITE_BYREF: 19559e69d7d0SLori Alt DO64(drr_write_byref.drr_object); 19569e69d7d0SLori Alt DO64(drr_write_byref.drr_offset); 19579e69d7d0SLori Alt DO64(drr_write_byref.drr_length); 19589e69d7d0SLori Alt DO64(drr_write_byref.drr_toguid); 19599e69d7d0SLori Alt DO64(drr_write_byref.drr_refguid); 19609e69d7d0SLori Alt DO64(drr_write_byref.drr_refobject); 19619e69d7d0SLori Alt DO64(drr_write_byref.drr_refoffset); 196298110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write_byref. 196398110f08SMatthew Ahrens drr_key.ddk_cksum); 19648e714474SLori Alt DO64(drr_write_byref.drr_key.ddk_prop); 1965efb80947Sahrens break; 19665d7b4d43SMatthew Ahrens case DRR_WRITE_EMBEDDED: 19675d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_object); 19685d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_offset); 19695d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_length); 19705d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_toguid); 19715d7b4d43SMatthew Ahrens DO32(drr_write_embedded.drr_lsize); 19725d7b4d43SMatthew Ahrens DO32(drr_write_embedded.drr_psize); 19735d7b4d43SMatthew Ahrens break; 1974efb80947Sahrens case DRR_FREE: 1975efb80947Sahrens DO64(drr_free.drr_object); 1976efb80947Sahrens DO64(drr_free.drr_offset); 1977efb80947Sahrens DO64(drr_free.drr_length); 19789e69d7d0SLori Alt DO64(drr_free.drr_toguid); 1979efb80947Sahrens break; 19800a586ceaSMark Shellenbaum case DRR_SPILL: 19810a586ceaSMark Shellenbaum DO64(drr_spill.drr_object); 19820a586ceaSMark Shellenbaum DO64(drr_spill.drr_length); 19830a586ceaSMark Shellenbaum DO64(drr_spill.drr_toguid); 19840a586ceaSMark Shellenbaum break; 1985efb80947Sahrens case DRR_END: 19869e69d7d0SLori Alt DO64(drr_end.drr_toguid); 198798110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); 1988efb80947Sahrens break; 1989efb80947Sahrens } 199098110f08SMatthew Ahrens 199198110f08SMatthew Ahrens if (drr->drr_type != DRR_BEGIN) { 199298110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); 199398110f08SMatthew Ahrens } 199498110f08SMatthew Ahrens 1995efb80947Sahrens #undef DO64 1996efb80947Sahrens #undef DO32 1997efb80947Sahrens } 1998efb80947Sahrens 1999e77d42eaSMatthew Ahrens static inline uint8_t 2000e77d42eaSMatthew Ahrens deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) 2001e77d42eaSMatthew Ahrens { 2002e77d42eaSMatthew Ahrens if (bonus_type == DMU_OT_SA) { 2003e77d42eaSMatthew Ahrens return (1); 2004e77d42eaSMatthew Ahrens } else { 2005e77d42eaSMatthew Ahrens return (1 + 2006e77d42eaSMatthew Ahrens ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT)); 2007e77d42eaSMatthew Ahrens } 2008e77d42eaSMatthew Ahrens } 2009e77d42eaSMatthew Ahrens 20109c3fd121SMatthew Ahrens static void 20119c3fd121SMatthew Ahrens save_resume_state(struct receive_writer_arg *rwa, 20129c3fd121SMatthew Ahrens uint64_t object, uint64_t offset, dmu_tx_t *tx) 20139c3fd121SMatthew Ahrens { 20149c3fd121SMatthew Ahrens int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; 20159c3fd121SMatthew Ahrens 20169c3fd121SMatthew Ahrens if (!rwa->resumable) 20179c3fd121SMatthew Ahrens return; 20189c3fd121SMatthew Ahrens 20199c3fd121SMatthew Ahrens /* 20209c3fd121SMatthew Ahrens * We use ds_resume_bytes[] != 0 to indicate that we need to 20219c3fd121SMatthew Ahrens * update this on disk, so it must not be 0. 20229c3fd121SMatthew Ahrens */ 20239c3fd121SMatthew Ahrens ASSERT(rwa->bytes_read != 0); 20249c3fd121SMatthew Ahrens 20259c3fd121SMatthew Ahrens /* 20269c3fd121SMatthew Ahrens * We only resume from write records, which have a valid 20279c3fd121SMatthew Ahrens * (non-meta-dnode) object number. 20289c3fd121SMatthew Ahrens */ 20299c3fd121SMatthew Ahrens ASSERT(object != 0); 20309c3fd121SMatthew Ahrens 20319c3fd121SMatthew Ahrens /* 20329c3fd121SMatthew Ahrens * For resuming to work correctly, we must receive records in order, 20339c3fd121SMatthew Ahrens * sorted by object,offset. This is checked by the callers, but 20349c3fd121SMatthew Ahrens * assert it here for good measure. 20359c3fd121SMatthew Ahrens */ 20369c3fd121SMatthew Ahrens ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); 20379c3fd121SMatthew Ahrens ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || 20389c3fd121SMatthew Ahrens offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); 20399c3fd121SMatthew Ahrens ASSERT3U(rwa->bytes_read, >=, 20409c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); 20419c3fd121SMatthew Ahrens 20429c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; 20439c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; 20449c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; 20459c3fd121SMatthew Ahrens } 20469c3fd121SMatthew Ahrens 2047efb80947Sahrens static int 2048a2cdcdd2SPaul Dagnelie receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, 2049a2cdcdd2SPaul Dagnelie void *data) 2050efb80947Sahrens { 2051e77d42eaSMatthew Ahrens dmu_object_info_t doi; 2052efb80947Sahrens dmu_tx_t *tx; 2053e77d42eaSMatthew Ahrens uint64_t object; 2054e77d42eaSMatthew Ahrens int err; 2055efb80947Sahrens 2056efb80947Sahrens if (drro->drr_type == DMU_OT_NONE || 2057ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drro->drr_type) || 2058ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drro->drr_bonustype) || 20599e69d7d0SLori Alt drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || 2060efb80947Sahrens drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || 2061efb80947Sahrens P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || 2062efb80947Sahrens drro->drr_blksz < SPA_MINBLOCKSIZE || 2063a2cdcdd2SPaul Dagnelie drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || 2064efb80947Sahrens drro->drr_bonuslen > DN_MAX_BONUSLEN) { 2065be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2066efb80947Sahrens } 2067efb80947Sahrens 2068a2cdcdd2SPaul Dagnelie err = dmu_object_info(rwa->os, drro->drr_object, &doi); 20692bf405a2SMark Maybee 20702bf405a2SMark Maybee if (err != 0 && err != ENOENT) 2071be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2072e77d42eaSMatthew Ahrens object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT; 20732bf405a2SMark Maybee 2074e77d42eaSMatthew Ahrens /* 2075e77d42eaSMatthew Ahrens * If we are losing blkptrs or changing the block size this must 2076e77d42eaSMatthew Ahrens * be a new file instance. We must clear out the previous file 2077e77d42eaSMatthew Ahrens * contents before we can change this type of metadata in the dnode. 2078e77d42eaSMatthew Ahrens */ 2079e77d42eaSMatthew Ahrens if (err == 0) { 2080e77d42eaSMatthew Ahrens int nblkptr; 2081e77d42eaSMatthew Ahrens 2082e77d42eaSMatthew Ahrens nblkptr = deduce_nblkptr(drro->drr_bonustype, 2083e77d42eaSMatthew Ahrens drro->drr_bonuslen); 2084e77d42eaSMatthew Ahrens 2085e77d42eaSMatthew Ahrens if (drro->drr_blksz != doi.doi_data_block_size || 2086e77d42eaSMatthew Ahrens nblkptr < doi.doi_nblkptr) { 2087a2cdcdd2SPaul Dagnelie err = dmu_free_long_range(rwa->os, drro->drr_object, 2088e77d42eaSMatthew Ahrens 0, DMU_OBJECT_END); 2089e77d42eaSMatthew Ahrens if (err != 0) 2090e77d42eaSMatthew Ahrens return (SET_ERROR(EINVAL)); 2091efb80947Sahrens } 2092e77d42eaSMatthew Ahrens } 2093e77d42eaSMatthew Ahrens 2094a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 2095e77d42eaSMatthew Ahrens dmu_tx_hold_bonus(tx, object); 2096e77d42eaSMatthew Ahrens err = dmu_tx_assign(tx, TXG_WAIT); 2097e77d42eaSMatthew Ahrens if (err != 0) { 2098e77d42eaSMatthew Ahrens dmu_tx_abort(tx); 2099e77d42eaSMatthew Ahrens return (err); 2100e77d42eaSMatthew Ahrens } 2101e77d42eaSMatthew Ahrens 2102e77d42eaSMatthew Ahrens if (object == DMU_NEW_OBJECT) { 2103e77d42eaSMatthew Ahrens /* currently free, want to be allocated */ 2104a2cdcdd2SPaul Dagnelie err = dmu_object_claim(rwa->os, drro->drr_object, 2105efb80947Sahrens drro->drr_type, drro->drr_blksz, 2106efb80947Sahrens drro->drr_bonustype, drro->drr_bonuslen, tx); 2107e77d42eaSMatthew Ahrens } else if (drro->drr_type != doi.doi_type || 2108e77d42eaSMatthew Ahrens drro->drr_blksz != doi.doi_data_block_size || 2109e77d42eaSMatthew Ahrens drro->drr_bonustype != doi.doi_bonus_type || 2110e77d42eaSMatthew Ahrens drro->drr_bonuslen != doi.doi_bonus_size) { 2111e77d42eaSMatthew Ahrens /* currently allocated, but with different properties */ 2112a2cdcdd2SPaul Dagnelie err = dmu_object_reclaim(rwa->os, drro->drr_object, 2113efb80947Sahrens drro->drr_type, drro->drr_blksz, 2114e77d42eaSMatthew Ahrens drro->drr_bonustype, drro->drr_bonuslen, tx); 2115efb80947Sahrens } 21163b2aab18SMatthew Ahrens if (err != 0) { 2117e77d42eaSMatthew Ahrens dmu_tx_commit(tx); 2118be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 21190a586ceaSMark Shellenbaum } 21202bf405a2SMark Maybee 2121a2cdcdd2SPaul Dagnelie dmu_object_set_checksum(rwa->os, drro->drr_object, 212298110f08SMatthew Ahrens drro->drr_checksumtype, tx); 2123a2cdcdd2SPaul Dagnelie dmu_object_set_compress(rwa->os, drro->drr_object, 212498110f08SMatthew Ahrens drro->drr_compress, tx); 2125efb80947Sahrens 2126adee0b6fSTim Haley if (data != NULL) { 2127efb80947Sahrens dmu_buf_t *db; 2128adee0b6fSTim Haley 2129a2cdcdd2SPaul Dagnelie VERIFY0(dmu_bonus_hold(rwa->os, drro->drr_object, FTAG, &db)); 2130efb80947Sahrens dmu_buf_will_dirty(db, tx); 2131efb80947Sahrens 21321934e92fSmaybee ASSERT3U(db->db_size, >=, drro->drr_bonuslen); 21331934e92fSmaybee bcopy(data, db->db_data, drro->drr_bonuslen); 2134a2cdcdd2SPaul Dagnelie if (rwa->byteswap) { 2135ad135b5dSChristopher Siden dmu_object_byteswap_t byteswap = 2136ad135b5dSChristopher Siden DMU_OT_BYTESWAP(drro->drr_bonustype); 2137ad135b5dSChristopher Siden dmu_ot_byteswap[byteswap].ob_func(db->db_data, 2138efb80947Sahrens drro->drr_bonuslen); 2139efb80947Sahrens } 2140efb80947Sahrens dmu_buf_rele(db, FTAG); 2141efb80947Sahrens } 2142efb80947Sahrens dmu_tx_commit(tx); 21439c3fd121SMatthew Ahrens 2144efb80947Sahrens return (0); 2145efb80947Sahrens } 2146efb80947Sahrens 2147efb80947Sahrens /* ARGSUSED */ 2148efb80947Sahrens static int 2149a2cdcdd2SPaul Dagnelie receive_freeobjects(struct receive_writer_arg *rwa, 2150efb80947Sahrens struct drr_freeobjects *drrfo) 2151efb80947Sahrens { 2152efb80947Sahrens uint64_t obj; 215368ecb2ecSPaul Dagnelie int next_err = 0; 2154efb80947Sahrens 2155efb80947Sahrens if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) 2156be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2157efb80947Sahrens 2158efb80947Sahrens for (obj = drrfo->drr_firstobj; 215968ecb2ecSPaul Dagnelie obj < drrfo->drr_firstobj + drrfo->drr_numobjs && next_err == 0; 216068ecb2ecSPaul Dagnelie next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { 2161efb80947Sahrens int err; 2162efb80947Sahrens 2163a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, obj, NULL) != 0) 2164efb80947Sahrens continue; 2165efb80947Sahrens 2166a2cdcdd2SPaul Dagnelie err = dmu_free_long_object(rwa->os, obj); 21673b2aab18SMatthew Ahrens if (err != 0) 2168efb80947Sahrens return (err); 2169efb80947Sahrens } 217068ecb2ecSPaul Dagnelie if (next_err != ESRCH) 217168ecb2ecSPaul Dagnelie return (next_err); 2172efb80947Sahrens return (0); 2173efb80947Sahrens } 2174efb80947Sahrens 2175efb80947Sahrens static int 2176a2cdcdd2SPaul Dagnelie receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw, 2177a2cdcdd2SPaul Dagnelie arc_buf_t *abuf) 2178efb80947Sahrens { 2179efb80947Sahrens dmu_tx_t *tx; 2180efb80947Sahrens int err; 2181efb80947Sahrens 21825602294fSDan Kimmel if (drrw->drr_offset + drrw->drr_logical_size < drrw->drr_offset || 2183ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drrw->drr_type)) 2184be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2185efb80947Sahrens 21869c3fd121SMatthew Ahrens /* 21879c3fd121SMatthew Ahrens * For resuming to work, records must be in increasing order 21889c3fd121SMatthew Ahrens * by (object, offset). 21899c3fd121SMatthew Ahrens */ 21909c3fd121SMatthew Ahrens if (drrw->drr_object < rwa->last_object || 21919c3fd121SMatthew Ahrens (drrw->drr_object == rwa->last_object && 21929c3fd121SMatthew Ahrens drrw->drr_offset < rwa->last_offset)) { 21939c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 21949c3fd121SMatthew Ahrens } 21959c3fd121SMatthew Ahrens rwa->last_object = drrw->drr_object; 21969c3fd121SMatthew Ahrens rwa->last_offset = drrw->drr_offset; 21979c3fd121SMatthew Ahrens 2198a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrw->drr_object, NULL) != 0) 21998a904709SMatthew Ahrens return (SET_ERROR(EINVAL)); 22008a904709SMatthew Ahrens 2201a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 2202efb80947Sahrens 2203efb80947Sahrens dmu_tx_hold_write(tx, drrw->drr_object, 22045602294fSDan Kimmel drrw->drr_offset, drrw->drr_logical_size); 2205efb80947Sahrens err = dmu_tx_assign(tx, TXG_WAIT); 22063b2aab18SMatthew Ahrens if (err != 0) { 2207efb80947Sahrens dmu_tx_abort(tx); 2208efb80947Sahrens return (err); 2209efb80947Sahrens } 2210a2cdcdd2SPaul Dagnelie if (rwa->byteswap) { 2211ad135b5dSChristopher Siden dmu_object_byteswap_t byteswap = 2212ad135b5dSChristopher Siden DMU_OT_BYTESWAP(drrw->drr_type); 221398110f08SMatthew Ahrens dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, 22145602294fSDan Kimmel DRR_WRITE_PAYLOAD_SIZE(drrw)); 2215ad135b5dSChristopher Siden } 221698110f08SMatthew Ahrens 22175602294fSDan Kimmel /* use the bonus buf to look up the dnode in dmu_assign_arcbuf */ 221898110f08SMatthew Ahrens dmu_buf_t *bonus; 2219a2cdcdd2SPaul Dagnelie if (dmu_bonus_hold(rwa->os, drrw->drr_object, FTAG, &bonus) != 0) 222098110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 22218a904709SMatthew Ahrens dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx); 22229c3fd121SMatthew Ahrens 22239c3fd121SMatthew Ahrens /* 22249c3fd121SMatthew Ahrens * Note: If the receive fails, we want the resume stream to start 22259c3fd121SMatthew Ahrens * with the same record that we last successfully received (as opposed 22269c3fd121SMatthew Ahrens * to the next record), so that we can verify that we are 22279c3fd121SMatthew Ahrens * resuming from the correct location. 22289c3fd121SMatthew Ahrens */ 22299c3fd121SMatthew Ahrens save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); 2230efb80947Sahrens dmu_tx_commit(tx); 22318a904709SMatthew Ahrens dmu_buf_rele(bonus, FTAG); 22329c3fd121SMatthew Ahrens 2233efb80947Sahrens return (0); 2234efb80947Sahrens } 2235efb80947Sahrens 22369e69d7d0SLori Alt /* 22379e69d7d0SLori Alt * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed 22389e69d7d0SLori Alt * streams to refer to a copy of the data that is already on the 22399e69d7d0SLori Alt * system because it came in earlier in the stream. This function 22409e69d7d0SLori Alt * finds the earlier copy of the data, and uses that copy instead of 22419e69d7d0SLori Alt * data from the stream to fulfill this write. 22429e69d7d0SLori Alt */ 22439e69d7d0SLori Alt static int 2244a2cdcdd2SPaul Dagnelie receive_write_byref(struct receive_writer_arg *rwa, 2245a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwbr) 22469e69d7d0SLori Alt { 22479e69d7d0SLori Alt dmu_tx_t *tx; 22489e69d7d0SLori Alt int err; 22499e69d7d0SLori Alt guid_map_entry_t gmesrch; 22509e69d7d0SLori Alt guid_map_entry_t *gmep; 22515d7b4d43SMatthew Ahrens avl_index_t where; 22529e69d7d0SLori Alt objset_t *ref_os = NULL; 22539e69d7d0SLori Alt dmu_buf_t *dbp; 22549e69d7d0SLori Alt 22559e69d7d0SLori Alt if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) 2256be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 22579e69d7d0SLori Alt 22589e69d7d0SLori Alt /* 22599e69d7d0SLori Alt * If the GUID of the referenced dataset is different from the 22609e69d7d0SLori Alt * GUID of the target dataset, find the referenced dataset. 22619e69d7d0SLori Alt */ 22629e69d7d0SLori Alt if (drrwbr->drr_toguid != drrwbr->drr_refguid) { 22639e69d7d0SLori Alt gmesrch.guid = drrwbr->drr_refguid; 2264a2cdcdd2SPaul Dagnelie if ((gmep = avl_find(rwa->guid_to_ds_map, &gmesrch, 22659e69d7d0SLori Alt &where)) == NULL) { 2266be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 22679e69d7d0SLori Alt } 22689e69d7d0SLori Alt if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) 2269be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 22709e69d7d0SLori Alt } else { 2271a2cdcdd2SPaul Dagnelie ref_os = rwa->os; 22729e69d7d0SLori Alt } 22739e69d7d0SLori Alt 22745d7b4d43SMatthew Ahrens err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, 22755d7b4d43SMatthew Ahrens drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH); 22765d7b4d43SMatthew Ahrens if (err != 0) 22779e69d7d0SLori Alt return (err); 22789e69d7d0SLori Alt 2279a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 22809e69d7d0SLori Alt 22819e69d7d0SLori Alt dmu_tx_hold_write(tx, drrwbr->drr_object, 22829e69d7d0SLori Alt drrwbr->drr_offset, drrwbr->drr_length); 22839e69d7d0SLori Alt err = dmu_tx_assign(tx, TXG_WAIT); 22843b2aab18SMatthew Ahrens if (err != 0) { 22859e69d7d0SLori Alt dmu_tx_abort(tx); 22869e69d7d0SLori Alt return (err); 22879e69d7d0SLori Alt } 2288a2cdcdd2SPaul Dagnelie dmu_write(rwa->os, drrwbr->drr_object, 22899e69d7d0SLori Alt drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); 22909e69d7d0SLori Alt dmu_buf_rele(dbp, FTAG); 22919c3fd121SMatthew Ahrens 22929c3fd121SMatthew Ahrens /* See comment in restore_write. */ 22939c3fd121SMatthew Ahrens save_resume_state(rwa, drrwbr->drr_object, drrwbr->drr_offset, tx); 22949e69d7d0SLori Alt dmu_tx_commit(tx); 22959e69d7d0SLori Alt return (0); 22969e69d7d0SLori Alt } 22979e69d7d0SLori Alt 22985d7b4d43SMatthew Ahrens static int 2299a2cdcdd2SPaul Dagnelie receive_write_embedded(struct receive_writer_arg *rwa, 23009c3fd121SMatthew Ahrens struct drr_write_embedded *drrwe, void *data) 23015d7b4d43SMatthew Ahrens { 23025d7b4d43SMatthew Ahrens dmu_tx_t *tx; 23035d7b4d43SMatthew Ahrens int err; 23045d7b4d43SMatthew Ahrens 23059c3fd121SMatthew Ahrens if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) 23065d7b4d43SMatthew Ahrens return (EINVAL); 23075d7b4d43SMatthew Ahrens 23089c3fd121SMatthew Ahrens if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) 23095d7b4d43SMatthew Ahrens return (EINVAL); 23105d7b4d43SMatthew Ahrens 23119c3fd121SMatthew Ahrens if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) 23125d7b4d43SMatthew Ahrens return (EINVAL); 23139c3fd121SMatthew Ahrens if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) 23145d7b4d43SMatthew Ahrens return (EINVAL); 23155d7b4d43SMatthew Ahrens 2316a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 23175d7b4d43SMatthew Ahrens 23189c3fd121SMatthew Ahrens dmu_tx_hold_write(tx, drrwe->drr_object, 23199c3fd121SMatthew Ahrens drrwe->drr_offset, drrwe->drr_length); 23205d7b4d43SMatthew Ahrens err = dmu_tx_assign(tx, TXG_WAIT); 23215d7b4d43SMatthew Ahrens if (err != 0) { 23225d7b4d43SMatthew Ahrens dmu_tx_abort(tx); 23235d7b4d43SMatthew Ahrens return (err); 23245d7b4d43SMatthew Ahrens } 23255d7b4d43SMatthew Ahrens 23269c3fd121SMatthew Ahrens dmu_write_embedded(rwa->os, drrwe->drr_object, 23279c3fd121SMatthew Ahrens drrwe->drr_offset, data, drrwe->drr_etype, 23289c3fd121SMatthew Ahrens drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, 2329a2cdcdd2SPaul Dagnelie rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); 23305d7b4d43SMatthew Ahrens 23319c3fd121SMatthew Ahrens /* See comment in restore_write. */ 23329c3fd121SMatthew Ahrens save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); 23335d7b4d43SMatthew Ahrens dmu_tx_commit(tx); 23345d7b4d43SMatthew Ahrens return (0); 23355d7b4d43SMatthew Ahrens } 23365d7b4d43SMatthew Ahrens 23370a586ceaSMark Shellenbaum static int 2338a2cdcdd2SPaul Dagnelie receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, 2339a2cdcdd2SPaul Dagnelie void *data) 23400a586ceaSMark Shellenbaum { 23410a586ceaSMark Shellenbaum dmu_tx_t *tx; 23420a586ceaSMark Shellenbaum dmu_buf_t *db, *db_spill; 23430a586ceaSMark Shellenbaum int err; 23440a586ceaSMark Shellenbaum 23450a586ceaSMark Shellenbaum if (drrs->drr_length < SPA_MINBLOCKSIZE || 2346a2cdcdd2SPaul Dagnelie drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) 2347be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 23480a586ceaSMark Shellenbaum 2349a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) 2350be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 23510a586ceaSMark Shellenbaum 2352a2cdcdd2SPaul Dagnelie VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); 23530a586ceaSMark Shellenbaum if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) { 23540a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 23550a586ceaSMark Shellenbaum return (err); 23560a586ceaSMark Shellenbaum } 23570a586ceaSMark Shellenbaum 2358a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 23590a586ceaSMark Shellenbaum 23600a586ceaSMark Shellenbaum dmu_tx_hold_spill(tx, db->db_object); 23610a586ceaSMark Shellenbaum 23620a586ceaSMark Shellenbaum err = dmu_tx_assign(tx, TXG_WAIT); 23633b2aab18SMatthew Ahrens if (err != 0) { 23640a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 23650a586ceaSMark Shellenbaum dmu_buf_rele(db_spill, FTAG); 23660a586ceaSMark Shellenbaum dmu_tx_abort(tx); 23670a586ceaSMark Shellenbaum return (err); 23680a586ceaSMark Shellenbaum } 23690a586ceaSMark Shellenbaum dmu_buf_will_dirty(db_spill, tx); 23700a586ceaSMark Shellenbaum 23710a586ceaSMark Shellenbaum if (db_spill->db_size < drrs->drr_length) 23720a586ceaSMark Shellenbaum VERIFY(0 == dbuf_spill_set_blksz(db_spill, 23730a586ceaSMark Shellenbaum drrs->drr_length, tx)); 23740a586ceaSMark Shellenbaum bcopy(data, db_spill->db_data, drrs->drr_length); 23750a586ceaSMark Shellenbaum 23760a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 23770a586ceaSMark Shellenbaum dmu_buf_rele(db_spill, FTAG); 23780a586ceaSMark Shellenbaum 23790a586ceaSMark Shellenbaum dmu_tx_commit(tx); 23800a586ceaSMark Shellenbaum return (0); 23810a586ceaSMark Shellenbaum } 23820a586ceaSMark Shellenbaum 2383efb80947Sahrens /* ARGSUSED */ 2384efb80947Sahrens static int 2385a2cdcdd2SPaul Dagnelie receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) 2386efb80947Sahrens { 2387efb80947Sahrens int err; 2388efb80947Sahrens 2389efb80947Sahrens if (drrf->drr_length != -1ULL && 2390efb80947Sahrens drrf->drr_offset + drrf->drr_length < drrf->drr_offset) 2391be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2392efb80947Sahrens 2393a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) 2394be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2395efb80947Sahrens 2396a2cdcdd2SPaul Dagnelie err = dmu_free_long_range(rwa->os, drrf->drr_object, 2397efb80947Sahrens drrf->drr_offset, drrf->drr_length); 2398a2cdcdd2SPaul Dagnelie 2399efb80947Sahrens return (err); 2400efb80947Sahrens } 2401efb80947Sahrens 24023b2aab18SMatthew Ahrens /* used to destroy the drc_ds on error */ 24033b2aab18SMatthew Ahrens static void 24043b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) 24053b2aab18SMatthew Ahrens { 24069c3fd121SMatthew Ahrens if (drc->drc_resumable) { 24079c3fd121SMatthew Ahrens /* wait for our resume state to be written to disk */ 24089c3fd121SMatthew Ahrens txg_wait_synced(drc->drc_ds->ds_dir->dd_pool, 0); 24099c3fd121SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 24109c3fd121SMatthew Ahrens } else { 24119adfa60dSMatthew Ahrens char name[ZFS_MAX_DATASET_NAME_LEN]; 24129c3fd121SMatthew Ahrens dsl_dataset_name(drc->drc_ds, name); 24139c3fd121SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 24149c3fd121SMatthew Ahrens (void) dsl_destroy_head(name); 24159c3fd121SMatthew Ahrens } 24163b2aab18SMatthew Ahrens } 24173b2aab18SMatthew Ahrens 241898110f08SMatthew Ahrens static void 2419a2cdcdd2SPaul Dagnelie receive_cksum(struct receive_arg *ra, int len, void *buf) 242098110f08SMatthew Ahrens { 242198110f08SMatthew Ahrens if (ra->byteswap) { 2422*770499e1SDan Kimmel (void) fletcher_4_incremental_byteswap(buf, len, &ra->cksum); 242398110f08SMatthew Ahrens } else { 2424*770499e1SDan Kimmel (void) fletcher_4_incremental_native(buf, len, &ra->cksum); 242598110f08SMatthew Ahrens } 242698110f08SMatthew Ahrens } 242798110f08SMatthew Ahrens 242898110f08SMatthew Ahrens /* 2429a2cdcdd2SPaul Dagnelie * Read the payload into a buffer of size len, and update the current record's 2430a2cdcdd2SPaul Dagnelie * payload field. 2431a2cdcdd2SPaul Dagnelie * Allocate ra->next_rrd and read the next record's header into 2432a2cdcdd2SPaul Dagnelie * ra->next_rrd->header. 243398110f08SMatthew Ahrens * Verify checksum of payload and next record. 243498110f08SMatthew Ahrens */ 243598110f08SMatthew Ahrens static int 2436a2cdcdd2SPaul Dagnelie receive_read_payload_and_next_header(struct receive_arg *ra, int len, void *buf) 243798110f08SMatthew Ahrens { 243898110f08SMatthew Ahrens int err; 243998110f08SMatthew Ahrens 244098110f08SMatthew Ahrens if (len != 0) { 2441a2cdcdd2SPaul Dagnelie ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); 24429c3fd121SMatthew Ahrens err = receive_read(ra, len, buf); 244398110f08SMatthew Ahrens if (err != 0) 244498110f08SMatthew Ahrens return (err); 24459c3fd121SMatthew Ahrens receive_cksum(ra, len, buf); 24469c3fd121SMatthew Ahrens 24479c3fd121SMatthew Ahrens /* note: rrd is NULL when reading the begin record's payload */ 24489c3fd121SMatthew Ahrens if (ra->rrd != NULL) { 24499c3fd121SMatthew Ahrens ra->rrd->payload = buf; 24509c3fd121SMatthew Ahrens ra->rrd->payload_size = len; 24519c3fd121SMatthew Ahrens ra->rrd->bytes_read = ra->bytes_read; 24529c3fd121SMatthew Ahrens } 245398110f08SMatthew Ahrens } 245498110f08SMatthew Ahrens 245598110f08SMatthew Ahrens ra->prev_cksum = ra->cksum; 245698110f08SMatthew Ahrens 2457a2cdcdd2SPaul Dagnelie ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); 2458a2cdcdd2SPaul Dagnelie err = receive_read(ra, sizeof (ra->next_rrd->header), 2459a2cdcdd2SPaul Dagnelie &ra->next_rrd->header); 24609c3fd121SMatthew Ahrens ra->next_rrd->bytes_read = ra->bytes_read; 2461a2cdcdd2SPaul Dagnelie if (err != 0) { 2462a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2463a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 246498110f08SMatthew Ahrens return (err); 2465a2cdcdd2SPaul Dagnelie } 2466a2cdcdd2SPaul Dagnelie if (ra->next_rrd->header.drr_type == DRR_BEGIN) { 2467a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2468a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 246998110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 2470a2cdcdd2SPaul Dagnelie } 247198110f08SMatthew Ahrens 247298110f08SMatthew Ahrens /* 247398110f08SMatthew Ahrens * Note: checksum is of everything up to but not including the 247498110f08SMatthew Ahrens * checksum itself. 247598110f08SMatthew Ahrens */ 247698110f08SMatthew Ahrens ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 247798110f08SMatthew Ahrens ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 2478a2cdcdd2SPaul Dagnelie receive_cksum(ra, 247998110f08SMatthew Ahrens offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 2480a2cdcdd2SPaul Dagnelie &ra->next_rrd->header); 248198110f08SMatthew Ahrens 2482a2cdcdd2SPaul Dagnelie zio_cksum_t cksum_orig = 2483a2cdcdd2SPaul Dagnelie ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; 2484a2cdcdd2SPaul Dagnelie zio_cksum_t *cksump = 2485a2cdcdd2SPaul Dagnelie &ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; 248698110f08SMatthew Ahrens 248798110f08SMatthew Ahrens if (ra->byteswap) 2488a2cdcdd2SPaul Dagnelie byteswap_record(&ra->next_rrd->header); 248998110f08SMatthew Ahrens 249098110f08SMatthew Ahrens if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && 2491a2cdcdd2SPaul Dagnelie !ZIO_CHECKSUM_EQUAL(ra->cksum, *cksump)) { 2492a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2493a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 249498110f08SMatthew Ahrens return (SET_ERROR(ECKSUM)); 2495a2cdcdd2SPaul Dagnelie } 249698110f08SMatthew Ahrens 2497a2cdcdd2SPaul Dagnelie receive_cksum(ra, sizeof (cksum_orig), &cksum_orig); 249898110f08SMatthew Ahrens 249998110f08SMatthew Ahrens return (0); 250098110f08SMatthew Ahrens } 250198110f08SMatthew Ahrens 250268ecb2ecSPaul Dagnelie static void 250368ecb2ecSPaul Dagnelie objlist_create(struct objlist *list) 250468ecb2ecSPaul Dagnelie { 250568ecb2ecSPaul Dagnelie list_create(&list->list, sizeof (struct receive_objnode), 250668ecb2ecSPaul Dagnelie offsetof(struct receive_objnode, node)); 250768ecb2ecSPaul Dagnelie list->last_lookup = 0; 250868ecb2ecSPaul Dagnelie } 250968ecb2ecSPaul Dagnelie 251068ecb2ecSPaul Dagnelie static void 251168ecb2ecSPaul Dagnelie objlist_destroy(struct objlist *list) 251268ecb2ecSPaul Dagnelie { 251368ecb2ecSPaul Dagnelie for (struct receive_objnode *n = list_remove_head(&list->list); 251468ecb2ecSPaul Dagnelie n != NULL; n = list_remove_head(&list->list)) { 251568ecb2ecSPaul Dagnelie kmem_free(n, sizeof (*n)); 251668ecb2ecSPaul Dagnelie } 251768ecb2ecSPaul Dagnelie list_destroy(&list->list); 251868ecb2ecSPaul Dagnelie } 251968ecb2ecSPaul Dagnelie 252068ecb2ecSPaul Dagnelie /* 252168ecb2ecSPaul Dagnelie * This function looks through the objlist to see if the specified object number 252268ecb2ecSPaul Dagnelie * is contained in the objlist. In the process, it will remove all object 252368ecb2ecSPaul Dagnelie * numbers in the list that are smaller than the specified object number. Thus, 252468ecb2ecSPaul Dagnelie * any lookup of an object number smaller than a previously looked up object 252568ecb2ecSPaul Dagnelie * number will always return false; therefore, all lookups should be done in 252668ecb2ecSPaul Dagnelie * ascending order. 252768ecb2ecSPaul Dagnelie */ 252868ecb2ecSPaul Dagnelie static boolean_t 252968ecb2ecSPaul Dagnelie objlist_exists(struct objlist *list, uint64_t object) 253068ecb2ecSPaul Dagnelie { 253168ecb2ecSPaul Dagnelie struct receive_objnode *node = list_head(&list->list); 253268ecb2ecSPaul Dagnelie ASSERT3U(object, >=, list->last_lookup); 253368ecb2ecSPaul Dagnelie list->last_lookup = object; 253468ecb2ecSPaul Dagnelie while (node != NULL && node->object < object) { 253568ecb2ecSPaul Dagnelie VERIFY3P(node, ==, list_remove_head(&list->list)); 253668ecb2ecSPaul Dagnelie kmem_free(node, sizeof (*node)); 253768ecb2ecSPaul Dagnelie node = list_head(&list->list); 253868ecb2ecSPaul Dagnelie } 253968ecb2ecSPaul Dagnelie return (node != NULL && node->object == object); 254068ecb2ecSPaul Dagnelie } 254168ecb2ecSPaul Dagnelie 254268ecb2ecSPaul Dagnelie /* 254368ecb2ecSPaul Dagnelie * The objlist is a list of object numbers stored in ascending order. However, 254468ecb2ecSPaul Dagnelie * the insertion of new object numbers does not seek out the correct location to 254568ecb2ecSPaul Dagnelie * store a new object number; instead, it appends it to the list for simplicity. 254668ecb2ecSPaul Dagnelie * Thus, any users must take care to only insert new object numbers in ascending 254768ecb2ecSPaul Dagnelie * order. 254868ecb2ecSPaul Dagnelie */ 254968ecb2ecSPaul Dagnelie static void 255068ecb2ecSPaul Dagnelie objlist_insert(struct objlist *list, uint64_t object) 255168ecb2ecSPaul Dagnelie { 255268ecb2ecSPaul Dagnelie struct receive_objnode *node = kmem_zalloc(sizeof (*node), KM_SLEEP); 255368ecb2ecSPaul Dagnelie node->object = object; 255468ecb2ecSPaul Dagnelie #ifdef ZFS_DEBUG 255568ecb2ecSPaul Dagnelie struct receive_objnode *last_object = list_tail(&list->list); 255668ecb2ecSPaul Dagnelie uint64_t last_objnum = (last_object != NULL ? last_object->object : 0); 255768ecb2ecSPaul Dagnelie ASSERT3U(node->object, >, last_objnum); 255868ecb2ecSPaul Dagnelie #endif 255968ecb2ecSPaul Dagnelie list_insert_tail(&list->list, node); 256068ecb2ecSPaul Dagnelie } 256168ecb2ecSPaul Dagnelie 2562a2cdcdd2SPaul Dagnelie /* 2563a2cdcdd2SPaul Dagnelie * Issue the prefetch reads for any necessary indirect blocks. 2564a2cdcdd2SPaul Dagnelie * 2565a2cdcdd2SPaul Dagnelie * We use the object ignore list to tell us whether or not to issue prefetches 2566a2cdcdd2SPaul Dagnelie * for a given object. We do this for both correctness (in case the blocksize 2567a2cdcdd2SPaul Dagnelie * of an object has changed) and performance (if the object doesn't exist, don't 2568a2cdcdd2SPaul Dagnelie * needlessly try to issue prefetches). We also trim the list as we go through 2569a2cdcdd2SPaul Dagnelie * the stream to prevent it from growing to an unbounded size. 2570a2cdcdd2SPaul Dagnelie * 2571a2cdcdd2SPaul Dagnelie * The object numbers within will always be in sorted order, and any write 2572a2cdcdd2SPaul Dagnelie * records we see will also be in sorted order, but they're not sorted with 2573a2cdcdd2SPaul Dagnelie * respect to each other (i.e. we can get several object records before 2574a2cdcdd2SPaul Dagnelie * receiving each object's write records). As a result, once we've reached a 2575a2cdcdd2SPaul Dagnelie * given object number, we can safely remove any reference to lower object 2576a2cdcdd2SPaul Dagnelie * numbers in the ignore list. In practice, we receive up to 32 object records 2577a2cdcdd2SPaul Dagnelie * before receiving write records, so the list can have up to 32 nodes in it. 2578a2cdcdd2SPaul Dagnelie */ 2579a2cdcdd2SPaul Dagnelie /* ARGSUSED */ 2580a2cdcdd2SPaul Dagnelie static void 2581a2cdcdd2SPaul Dagnelie receive_read_prefetch(struct receive_arg *ra, 2582a2cdcdd2SPaul Dagnelie uint64_t object, uint64_t offset, uint64_t length) 2583a2cdcdd2SPaul Dagnelie { 258468ecb2ecSPaul Dagnelie if (!objlist_exists(&ra->ignore_objlist, object)) { 2585a2cdcdd2SPaul Dagnelie dmu_prefetch(ra->os, object, 1, offset, length, 2586a2cdcdd2SPaul Dagnelie ZIO_PRIORITY_SYNC_READ); 2587a2cdcdd2SPaul Dagnelie } 2588a2cdcdd2SPaul Dagnelie } 2589a2cdcdd2SPaul Dagnelie 2590a2cdcdd2SPaul Dagnelie /* 2591a2cdcdd2SPaul Dagnelie * Read records off the stream, issuing any necessary prefetches. 2592a2cdcdd2SPaul Dagnelie */ 259398110f08SMatthew Ahrens static int 2594a2cdcdd2SPaul Dagnelie receive_read_record(struct receive_arg *ra) 259598110f08SMatthew Ahrens { 259698110f08SMatthew Ahrens int err; 259798110f08SMatthew Ahrens 2598a2cdcdd2SPaul Dagnelie switch (ra->rrd->header.drr_type) { 259998110f08SMatthew Ahrens case DRR_OBJECT: 260098110f08SMatthew Ahrens { 2601a2cdcdd2SPaul Dagnelie struct drr_object *drro = &ra->rrd->header.drr_u.drr_object; 2602a2cdcdd2SPaul Dagnelie uint32_t size = P2ROUNDUP(drro->drr_bonuslen, 8); 2603a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(size, KM_SLEEP); 2604a2cdcdd2SPaul Dagnelie dmu_object_info_t doi; 2605a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, size, buf); 2606a2cdcdd2SPaul Dagnelie if (err != 0) { 2607a2cdcdd2SPaul Dagnelie kmem_free(buf, size); 260898110f08SMatthew Ahrens return (err); 2609a2cdcdd2SPaul Dagnelie } 2610a2cdcdd2SPaul Dagnelie err = dmu_object_info(ra->os, drro->drr_object, &doi); 2611a2cdcdd2SPaul Dagnelie /* 2612a2cdcdd2SPaul Dagnelie * See receive_read_prefetch for an explanation why we're 2613a2cdcdd2SPaul Dagnelie * storing this object in the ignore_obj_list. 2614a2cdcdd2SPaul Dagnelie */ 2615a2cdcdd2SPaul Dagnelie if (err == ENOENT || 2616a2cdcdd2SPaul Dagnelie (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { 261768ecb2ecSPaul Dagnelie objlist_insert(&ra->ignore_objlist, drro->drr_object); 2618a2cdcdd2SPaul Dagnelie err = 0; 2619a2cdcdd2SPaul Dagnelie } 2620a2cdcdd2SPaul Dagnelie return (err); 262198110f08SMatthew Ahrens } 262298110f08SMatthew Ahrens case DRR_FREEOBJECTS: 262398110f08SMatthew Ahrens { 2624a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2625a2cdcdd2SPaul Dagnelie return (err); 262698110f08SMatthew Ahrens } 262798110f08SMatthew Ahrens case DRR_WRITE: 262898110f08SMatthew Ahrens { 2629a2cdcdd2SPaul Dagnelie struct drr_write *drrw = &ra->rrd->header.drr_u.drr_write; 26305602294fSDan Kimmel arc_buf_t *abuf; 26315602294fSDan Kimmel boolean_t is_meta = DMU_OT_IS_METADATA(drrw->drr_type); 26325602294fSDan Kimmel if (DRR_WRITE_COMPRESSED(drrw)) { 26335602294fSDan Kimmel ASSERT3U(drrw->drr_compressed_size, >, 0); 26345602294fSDan Kimmel ASSERT3U(drrw->drr_logical_size, >=, 26355602294fSDan Kimmel drrw->drr_compressed_size); 26365602294fSDan Kimmel ASSERT(!is_meta); 26375602294fSDan Kimmel abuf = arc_loan_compressed_buf( 26385602294fSDan Kimmel dmu_objset_spa(ra->os), 26395602294fSDan Kimmel drrw->drr_compressed_size, drrw->drr_logical_size, 26405602294fSDan Kimmel drrw->drr_compressiontype); 26415602294fSDan Kimmel } else { 26425602294fSDan Kimmel abuf = arc_loan_buf(dmu_objset_spa(ra->os), 26435602294fSDan Kimmel is_meta, drrw->drr_logical_size); 26445602294fSDan Kimmel } 264598110f08SMatthew Ahrens 2646a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 26475602294fSDan Kimmel DRR_WRITE_PAYLOAD_SIZE(drrw), abuf->b_data); 2648a2cdcdd2SPaul Dagnelie if (err != 0) { 264998110f08SMatthew Ahrens dmu_return_arcbuf(abuf); 2650a2cdcdd2SPaul Dagnelie return (err); 2651a2cdcdd2SPaul Dagnelie } 2652a2cdcdd2SPaul Dagnelie ra->rrd->write_buf = abuf; 2653a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrw->drr_object, drrw->drr_offset, 26545602294fSDan Kimmel drrw->drr_logical_size); 265598110f08SMatthew Ahrens return (err); 265698110f08SMatthew Ahrens } 265798110f08SMatthew Ahrens case DRR_WRITE_BYREF: 265898110f08SMatthew Ahrens { 2659a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwb = 2660a2cdcdd2SPaul Dagnelie &ra->rrd->header.drr_u.drr_write_byref; 2661a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2662a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrwb->drr_object, drrwb->drr_offset, 2663a2cdcdd2SPaul Dagnelie drrwb->drr_length); 2664a2cdcdd2SPaul Dagnelie return (err); 266598110f08SMatthew Ahrens } 266698110f08SMatthew Ahrens case DRR_WRITE_EMBEDDED: 266798110f08SMatthew Ahrens { 266898110f08SMatthew Ahrens struct drr_write_embedded *drrwe = 2669a2cdcdd2SPaul Dagnelie &ra->rrd->header.drr_u.drr_write_embedded; 2670a2cdcdd2SPaul Dagnelie uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); 2671a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(size, KM_SLEEP); 2672a2cdcdd2SPaul Dagnelie 2673a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, size, buf); 2674a2cdcdd2SPaul Dagnelie if (err != 0) { 2675a2cdcdd2SPaul Dagnelie kmem_free(buf, size); 267698110f08SMatthew Ahrens return (err); 2677a2cdcdd2SPaul Dagnelie } 2678a2cdcdd2SPaul Dagnelie 2679a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrwe->drr_object, drrwe->drr_offset, 2680a2cdcdd2SPaul Dagnelie drrwe->drr_length); 2681a2cdcdd2SPaul Dagnelie return (err); 268298110f08SMatthew Ahrens } 268398110f08SMatthew Ahrens case DRR_FREE: 268498110f08SMatthew Ahrens { 2685a2cdcdd2SPaul Dagnelie /* 2686a2cdcdd2SPaul Dagnelie * It might be beneficial to prefetch indirect blocks here, but 2687a2cdcdd2SPaul Dagnelie * we don't really have the data to decide for sure. 2688a2cdcdd2SPaul Dagnelie */ 2689a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2690a2cdcdd2SPaul Dagnelie return (err); 269198110f08SMatthew Ahrens } 269298110f08SMatthew Ahrens case DRR_END: 269398110f08SMatthew Ahrens { 2694a2cdcdd2SPaul Dagnelie struct drr_end *drre = &ra->rrd->header.drr_u.drr_end; 269598110f08SMatthew Ahrens if (!ZIO_CHECKSUM_EQUAL(ra->prev_cksum, drre->drr_checksum)) 26969c3fd121SMatthew Ahrens return (SET_ERROR(ECKSUM)); 269798110f08SMatthew Ahrens return (0); 269898110f08SMatthew Ahrens } 269998110f08SMatthew Ahrens case DRR_SPILL: 270098110f08SMatthew Ahrens { 2701a2cdcdd2SPaul Dagnelie struct drr_spill *drrs = &ra->rrd->header.drr_u.drr_spill; 2702a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(drrs->drr_length, KM_SLEEP); 2703a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, drrs->drr_length, 2704a2cdcdd2SPaul Dagnelie buf); 270598110f08SMatthew Ahrens if (err != 0) 2706a2cdcdd2SPaul Dagnelie kmem_free(buf, drrs->drr_length); 2707a2cdcdd2SPaul Dagnelie return (err); 270898110f08SMatthew Ahrens } 270998110f08SMatthew Ahrens default: 271098110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 271198110f08SMatthew Ahrens } 271298110f08SMatthew Ahrens } 271398110f08SMatthew Ahrens 27143cb34c60Sahrens /* 2715a2cdcdd2SPaul Dagnelie * Commit the records to the pool. 2716a2cdcdd2SPaul Dagnelie */ 2717a2cdcdd2SPaul Dagnelie static int 2718a2cdcdd2SPaul Dagnelie receive_process_record(struct receive_writer_arg *rwa, 2719a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd) 2720a2cdcdd2SPaul Dagnelie { 2721a2cdcdd2SPaul Dagnelie int err; 2722a2cdcdd2SPaul Dagnelie 27239c3fd121SMatthew Ahrens /* Processing in order, therefore bytes_read should be increasing. */ 27249c3fd121SMatthew Ahrens ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); 27259c3fd121SMatthew Ahrens rwa->bytes_read = rrd->bytes_read; 27269c3fd121SMatthew Ahrens 2727a2cdcdd2SPaul Dagnelie switch (rrd->header.drr_type) { 2728a2cdcdd2SPaul Dagnelie case DRR_OBJECT: 2729a2cdcdd2SPaul Dagnelie { 2730a2cdcdd2SPaul Dagnelie struct drr_object *drro = &rrd->header.drr_u.drr_object; 2731a2cdcdd2SPaul Dagnelie err = receive_object(rwa, drro, rrd->payload); 2732a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2733a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2734a2cdcdd2SPaul Dagnelie return (err); 2735a2cdcdd2SPaul Dagnelie } 2736a2cdcdd2SPaul Dagnelie case DRR_FREEOBJECTS: 2737a2cdcdd2SPaul Dagnelie { 2738a2cdcdd2SPaul Dagnelie struct drr_freeobjects *drrfo = 2739a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_freeobjects; 2740a2cdcdd2SPaul Dagnelie return (receive_freeobjects(rwa, drrfo)); 2741a2cdcdd2SPaul Dagnelie } 2742a2cdcdd2SPaul Dagnelie case DRR_WRITE: 2743a2cdcdd2SPaul Dagnelie { 2744a2cdcdd2SPaul Dagnelie struct drr_write *drrw = &rrd->header.drr_u.drr_write; 2745a2cdcdd2SPaul Dagnelie err = receive_write(rwa, drrw, rrd->write_buf); 2746a2cdcdd2SPaul Dagnelie /* if receive_write() is successful, it consumes the arc_buf */ 2747a2cdcdd2SPaul Dagnelie if (err != 0) 2748a2cdcdd2SPaul Dagnelie dmu_return_arcbuf(rrd->write_buf); 2749a2cdcdd2SPaul Dagnelie rrd->write_buf = NULL; 2750a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2751a2cdcdd2SPaul Dagnelie return (err); 2752a2cdcdd2SPaul Dagnelie } 2753a2cdcdd2SPaul Dagnelie case DRR_WRITE_BYREF: 2754a2cdcdd2SPaul Dagnelie { 2755a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwbr = 2756a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_write_byref; 2757a2cdcdd2SPaul Dagnelie return (receive_write_byref(rwa, drrwbr)); 2758a2cdcdd2SPaul Dagnelie } 2759a2cdcdd2SPaul Dagnelie case DRR_WRITE_EMBEDDED: 2760a2cdcdd2SPaul Dagnelie { 2761a2cdcdd2SPaul Dagnelie struct drr_write_embedded *drrwe = 2762a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_write_embedded; 2763a2cdcdd2SPaul Dagnelie err = receive_write_embedded(rwa, drrwe, rrd->payload); 2764a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2765a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2766a2cdcdd2SPaul Dagnelie return (err); 2767a2cdcdd2SPaul Dagnelie } 2768a2cdcdd2SPaul Dagnelie case DRR_FREE: 2769a2cdcdd2SPaul Dagnelie { 2770a2cdcdd2SPaul Dagnelie struct drr_free *drrf = &rrd->header.drr_u.drr_free; 2771a2cdcdd2SPaul Dagnelie return (receive_free(rwa, drrf)); 2772a2cdcdd2SPaul Dagnelie } 2773a2cdcdd2SPaul Dagnelie case DRR_SPILL: 2774a2cdcdd2SPaul Dagnelie { 2775a2cdcdd2SPaul Dagnelie struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; 2776a2cdcdd2SPaul Dagnelie err = receive_spill(rwa, drrs, rrd->payload); 2777a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2778a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2779a2cdcdd2SPaul Dagnelie return (err); 2780a2cdcdd2SPaul Dagnelie } 2781a2cdcdd2SPaul Dagnelie default: 2782a2cdcdd2SPaul Dagnelie return (SET_ERROR(EINVAL)); 2783a2cdcdd2SPaul Dagnelie } 2784a2cdcdd2SPaul Dagnelie } 2785a2cdcdd2SPaul Dagnelie 2786a2cdcdd2SPaul Dagnelie /* 2787a2cdcdd2SPaul Dagnelie * dmu_recv_stream's worker thread; pull records off the queue, and then call 2788a2cdcdd2SPaul Dagnelie * receive_process_record When we're done, signal the main thread and exit. 2789a2cdcdd2SPaul Dagnelie */ 2790a2cdcdd2SPaul Dagnelie static void 2791a2cdcdd2SPaul Dagnelie receive_writer_thread(void *arg) 2792a2cdcdd2SPaul Dagnelie { 2793a2cdcdd2SPaul Dagnelie struct receive_writer_arg *rwa = arg; 2794a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd; 2795a2cdcdd2SPaul Dagnelie for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; 2796a2cdcdd2SPaul Dagnelie rrd = bqueue_dequeue(&rwa->q)) { 2797a2cdcdd2SPaul Dagnelie /* 2798a2cdcdd2SPaul Dagnelie * If there's an error, the main thread will stop putting things 2799a2cdcdd2SPaul Dagnelie * on the queue, but we need to clear everything in it before we 2800a2cdcdd2SPaul Dagnelie * can exit. 2801a2cdcdd2SPaul Dagnelie */ 2802a2cdcdd2SPaul Dagnelie if (rwa->err == 0) { 2803a2cdcdd2SPaul Dagnelie rwa->err = receive_process_record(rwa, rrd); 2804a2cdcdd2SPaul Dagnelie } else if (rrd->write_buf != NULL) { 2805a2cdcdd2SPaul Dagnelie dmu_return_arcbuf(rrd->write_buf); 2806a2cdcdd2SPaul Dagnelie rrd->write_buf = NULL; 2807a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2808a2cdcdd2SPaul Dagnelie } else if (rrd->payload != NULL) { 2809a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2810a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2811a2cdcdd2SPaul Dagnelie } 2812a2cdcdd2SPaul Dagnelie kmem_free(rrd, sizeof (*rrd)); 2813a2cdcdd2SPaul Dagnelie } 2814a2cdcdd2SPaul Dagnelie kmem_free(rrd, sizeof (*rrd)); 2815a2cdcdd2SPaul Dagnelie mutex_enter(&rwa->mutex); 2816a2cdcdd2SPaul Dagnelie rwa->done = B_TRUE; 2817a2cdcdd2SPaul Dagnelie cv_signal(&rwa->cv); 2818a2cdcdd2SPaul Dagnelie mutex_exit(&rwa->mutex); 2819f2c1e9bcSJorgen Lundman thread_exit(); 2820a2cdcdd2SPaul Dagnelie } 2821a2cdcdd2SPaul Dagnelie 28229c3fd121SMatthew Ahrens static int 28239c3fd121SMatthew Ahrens resume_check(struct receive_arg *ra, nvlist_t *begin_nvl) 28249c3fd121SMatthew Ahrens { 28259c3fd121SMatthew Ahrens uint64_t val; 28269c3fd121SMatthew Ahrens objset_t *mos = dmu_objset_pool(ra->os)->dp_meta_objset; 28279c3fd121SMatthew Ahrens uint64_t dsobj = dmu_objset_id(ra->os); 28289c3fd121SMatthew Ahrens uint64_t resume_obj, resume_off; 28299c3fd121SMatthew Ahrens 28309c3fd121SMatthew Ahrens if (nvlist_lookup_uint64(begin_nvl, 28319c3fd121SMatthew Ahrens "resume_object", &resume_obj) != 0 || 28329c3fd121SMatthew Ahrens nvlist_lookup_uint64(begin_nvl, 28339c3fd121SMatthew Ahrens "resume_offset", &resume_off) != 0) { 28349c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 28359c3fd121SMatthew Ahrens } 28369c3fd121SMatthew Ahrens VERIFY0(zap_lookup(mos, dsobj, 28379c3fd121SMatthew Ahrens DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); 28389c3fd121SMatthew Ahrens if (resume_obj != val) 28399c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 28409c3fd121SMatthew Ahrens VERIFY0(zap_lookup(mos, dsobj, 28419c3fd121SMatthew Ahrens DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); 28429c3fd121SMatthew Ahrens if (resume_off != val) 28439c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 28449c3fd121SMatthew Ahrens 28459c3fd121SMatthew Ahrens return (0); 28469c3fd121SMatthew Ahrens } 28479c3fd121SMatthew Ahrens 2848a2cdcdd2SPaul Dagnelie /* 2849a2cdcdd2SPaul Dagnelie * Read in the stream's records, one by one, and apply them to the pool. There 2850a2cdcdd2SPaul Dagnelie * are two threads involved; the thread that calls this function will spin up a 2851a2cdcdd2SPaul Dagnelie * worker thread, read the records off the stream one by one, and issue 2852a2cdcdd2SPaul Dagnelie * prefetches for any necessary indirect blocks. It will then push the records 2853a2cdcdd2SPaul Dagnelie * onto an internal blocking queue. The worker thread will pull the records off 2854a2cdcdd2SPaul Dagnelie * the queue, and actually write the data into the DMU. This way, the worker 2855a2cdcdd2SPaul Dagnelie * thread doesn't have to wait for reads to complete, since everything it needs 2856a2cdcdd2SPaul Dagnelie * (the indirect blocks) will be prefetched. 2857a2cdcdd2SPaul Dagnelie * 28583cb34c60Sahrens * NB: callers *must* call dmu_recv_end() if this succeeds. 28593cb34c60Sahrens */ 2860efb80947Sahrens int 2861c99e4bdcSChris Kirby dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, 2862c99e4bdcSChris Kirby int cleanup_fd, uint64_t *action_handlep) 2863efb80947Sahrens { 286498110f08SMatthew Ahrens int err = 0; 2865a2cdcdd2SPaul Dagnelie struct receive_arg ra = { 0 }; 2866a2cdcdd2SPaul Dagnelie struct receive_writer_arg rwa = { 0 }; 28679e69d7d0SLori Alt int featureflags; 28689c3fd121SMatthew Ahrens nvlist_t *begin_nvl = NULL; 2869efb80947Sahrens 28703b2aab18SMatthew Ahrens ra.byteswap = drc->drc_byteswap; 28713b2aab18SMatthew Ahrens ra.cksum = drc->drc_cksum; 28723cb34c60Sahrens ra.vp = vp; 28733cb34c60Sahrens ra.voff = *voffp; 28749c3fd121SMatthew Ahrens 28759c3fd121SMatthew Ahrens if (dsl_dataset_is_zapified(drc->drc_ds)) { 28769c3fd121SMatthew Ahrens (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, 28779c3fd121SMatthew Ahrens drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, 28789c3fd121SMatthew Ahrens sizeof (ra.bytes_read), 1, &ra.bytes_read); 28799c3fd121SMatthew Ahrens } 28809c3fd121SMatthew Ahrens 288168ecb2ecSPaul Dagnelie objlist_create(&ra.ignore_objlist); 2882efb80947Sahrens 28833cb34c60Sahrens /* these were verified in dmu_recv_begin */ 28843b2aab18SMatthew Ahrens ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, 28859e69d7d0SLori Alt DMU_SUBSTREAM); 28863b2aab18SMatthew Ahrens ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); 2887efb80947Sahrens 2888efb80947Sahrens /* 2889efb80947Sahrens * Open the objset we are modifying. 2890efb80947Sahrens */ 289198110f08SMatthew Ahrens VERIFY0(dmu_objset_from_ds(drc->drc_ds, &ra.os)); 2892efb80947Sahrens 2893c1379625SJustin T. Gibbs ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); 2894efb80947Sahrens 28959e69d7d0SLori Alt featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); 28969e69d7d0SLori Alt 28979e69d7d0SLori Alt /* if this stream is dedup'ed, set up the avl tree for guid mapping */ 28989e69d7d0SLori Alt if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { 2899a7f53a56SChris Kirby minor_t minor; 2900a7f53a56SChris Kirby 2901c99e4bdcSChris Kirby if (cleanup_fd == -1) { 2902be6fd75aSMatthew Ahrens ra.err = SET_ERROR(EBADF); 2903c99e4bdcSChris Kirby goto out; 2904c99e4bdcSChris Kirby } 2905a7f53a56SChris Kirby ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor); 29063b2aab18SMatthew Ahrens if (ra.err != 0) { 2907a7f53a56SChris Kirby cleanup_fd = -1; 2908a7f53a56SChris Kirby goto out; 2909a7f53a56SChris Kirby } 2910a7f53a56SChris Kirby 2911c99e4bdcSChris Kirby if (*action_handlep == 0) { 2912a2cdcdd2SPaul Dagnelie rwa.guid_to_ds_map = 2913c99e4bdcSChris Kirby kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); 2914a2cdcdd2SPaul Dagnelie avl_create(rwa.guid_to_ds_map, guid_compare, 2915c99e4bdcSChris Kirby sizeof (guid_map_entry_t), 2916c99e4bdcSChris Kirby offsetof(guid_map_entry_t, avlnode)); 291798110f08SMatthew Ahrens err = zfs_onexit_add_cb(minor, 2918a2cdcdd2SPaul Dagnelie free_guid_map_onexit, rwa.guid_to_ds_map, 2919c99e4bdcSChris Kirby action_handlep); 29203b2aab18SMatthew Ahrens if (ra.err != 0) 2921c99e4bdcSChris Kirby goto out; 2922c99e4bdcSChris Kirby } else { 292398110f08SMatthew Ahrens err = zfs_onexit_cb_data(minor, *action_handlep, 2924a2cdcdd2SPaul Dagnelie (void **)&rwa.guid_to_ds_map); 29253b2aab18SMatthew Ahrens if (ra.err != 0) 2926c99e4bdcSChris Kirby goto out; 2927c99e4bdcSChris Kirby } 2928ec5cf9d5SAlexander Stetsenko 2929a2cdcdd2SPaul Dagnelie drc->drc_guid_to_ds_map = rwa.guid_to_ds_map; 29309e69d7d0SLori Alt } 29319e69d7d0SLori Alt 29329c3fd121SMatthew Ahrens uint32_t payloadlen = drc->drc_drr_begin->drr_payloadlen; 29339c3fd121SMatthew Ahrens void *payload = NULL; 29349c3fd121SMatthew Ahrens if (payloadlen != 0) 29359c3fd121SMatthew Ahrens payload = kmem_alloc(payloadlen, KM_SLEEP); 29369c3fd121SMatthew Ahrens 29379c3fd121SMatthew Ahrens err = receive_read_payload_and_next_header(&ra, payloadlen, payload); 29389c3fd121SMatthew Ahrens if (err != 0) { 29399c3fd121SMatthew Ahrens if (payloadlen != 0) 29409c3fd121SMatthew Ahrens kmem_free(payload, payloadlen); 294198110f08SMatthew Ahrens goto out; 29429c3fd121SMatthew Ahrens } 29439c3fd121SMatthew Ahrens if (payloadlen != 0) { 29449c3fd121SMatthew Ahrens err = nvlist_unpack(payload, payloadlen, &begin_nvl, KM_SLEEP); 29459c3fd121SMatthew Ahrens kmem_free(payload, payloadlen); 29469c3fd121SMatthew Ahrens if (err != 0) 29479c3fd121SMatthew Ahrens goto out; 29489c3fd121SMatthew Ahrens } 29499c3fd121SMatthew Ahrens 29509c3fd121SMatthew Ahrens if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { 29519c3fd121SMatthew Ahrens err = resume_check(&ra, begin_nvl); 29529c3fd121SMatthew Ahrens if (err != 0) 29539c3fd121SMatthew Ahrens goto out; 29549c3fd121SMatthew Ahrens } 295598110f08SMatthew Ahrens 2956a2cdcdd2SPaul Dagnelie (void) bqueue_init(&rwa.q, zfs_recv_queue_length, 2957a2cdcdd2SPaul Dagnelie offsetof(struct receive_record_arg, node)); 2958a2cdcdd2SPaul Dagnelie cv_init(&rwa.cv, NULL, CV_DEFAULT, NULL); 2959a2cdcdd2SPaul Dagnelie mutex_init(&rwa.mutex, NULL, MUTEX_DEFAULT, NULL); 2960a2cdcdd2SPaul Dagnelie rwa.os = ra.os; 2961a2cdcdd2SPaul Dagnelie rwa.byteswap = drc->drc_byteswap; 29629c3fd121SMatthew Ahrens rwa.resumable = drc->drc_resumable; 2963a2cdcdd2SPaul Dagnelie 2964a2cdcdd2SPaul Dagnelie (void) thread_create(NULL, 0, receive_writer_thread, &rwa, 0, curproc, 2965a2cdcdd2SPaul Dagnelie TS_RUN, minclsyspri); 2966a2cdcdd2SPaul Dagnelie /* 2967a2cdcdd2SPaul Dagnelie * We're reading rwa.err without locks, which is safe since we are the 2968a2cdcdd2SPaul Dagnelie * only reader, and the worker thread is the only writer. It's ok if we 2969a2cdcdd2SPaul Dagnelie * miss a write for an iteration or two of the loop, since the writer 2970a2cdcdd2SPaul Dagnelie * thread will keep freeing records we send it until we send it an eos 2971a2cdcdd2SPaul Dagnelie * marker. 2972a2cdcdd2SPaul Dagnelie * 2973a2cdcdd2SPaul Dagnelie * We can leave this loop in 3 ways: First, if rwa.err is 2974a2cdcdd2SPaul Dagnelie * non-zero. In that case, the writer thread will free the rrd we just 2975a2cdcdd2SPaul Dagnelie * pushed. Second, if we're interrupted; in that case, either it's the 2976a2cdcdd2SPaul Dagnelie * first loop and ra.rrd was never allocated, or it's later, and ra.rrd 2977a2cdcdd2SPaul Dagnelie * has been handed off to the writer thread who will free it. Finally, 2978a2cdcdd2SPaul Dagnelie * if receive_read_record fails or we're at the end of the stream, then 2979a2cdcdd2SPaul Dagnelie * we free ra.rrd and exit. 2980a2cdcdd2SPaul Dagnelie */ 2981a2cdcdd2SPaul Dagnelie while (rwa.err == 0) { 2982efb80947Sahrens if (issig(JUSTLOOKING) && issig(FORREAL)) { 298398110f08SMatthew Ahrens err = SET_ERROR(EINTR); 298498110f08SMatthew Ahrens break; 2985efb80947Sahrens } 2986efb80947Sahrens 2987a2cdcdd2SPaul Dagnelie ASSERT3P(ra.rrd, ==, NULL); 2988a2cdcdd2SPaul Dagnelie ra.rrd = ra.next_rrd; 2989a2cdcdd2SPaul Dagnelie ra.next_rrd = NULL; 2990a2cdcdd2SPaul Dagnelie /* Allocates and loads header into ra.next_rrd */ 2991a2cdcdd2SPaul Dagnelie err = receive_read_record(&ra); 2992efb80947Sahrens 2993a2cdcdd2SPaul Dagnelie if (ra.rrd->header.drr_type == DRR_END || err != 0) { 2994a2cdcdd2SPaul Dagnelie kmem_free(ra.rrd, sizeof (*ra.rrd)); 2995a2cdcdd2SPaul Dagnelie ra.rrd = NULL; 29960a586ceaSMark Shellenbaum break; 2997a2cdcdd2SPaul Dagnelie } 2998a2cdcdd2SPaul Dagnelie 2999a2cdcdd2SPaul Dagnelie bqueue_enqueue(&rwa.q, ra.rrd, 3000a2cdcdd2SPaul Dagnelie sizeof (struct receive_record_arg) + ra.rrd->payload_size); 3001a2cdcdd2SPaul Dagnelie ra.rrd = NULL; 3002a2cdcdd2SPaul Dagnelie } 3003a2cdcdd2SPaul Dagnelie if (ra.next_rrd == NULL) 3004a2cdcdd2SPaul Dagnelie ra.next_rrd = kmem_zalloc(sizeof (*ra.next_rrd), KM_SLEEP); 3005a2cdcdd2SPaul Dagnelie ra.next_rrd->eos_marker = B_TRUE; 3006a2cdcdd2SPaul Dagnelie bqueue_enqueue(&rwa.q, ra.next_rrd, 1); 3007a2cdcdd2SPaul Dagnelie 3008a2cdcdd2SPaul Dagnelie mutex_enter(&rwa.mutex); 3009a2cdcdd2SPaul Dagnelie while (!rwa.done) { 3010a2cdcdd2SPaul Dagnelie cv_wait(&rwa.cv, &rwa.mutex); 3011efb80947Sahrens } 3012a2cdcdd2SPaul Dagnelie mutex_exit(&rwa.mutex); 3013a2cdcdd2SPaul Dagnelie 3014a2cdcdd2SPaul Dagnelie cv_destroy(&rwa.cv); 3015a2cdcdd2SPaul Dagnelie mutex_destroy(&rwa.mutex); 3016a2cdcdd2SPaul Dagnelie bqueue_destroy(&rwa.q); 3017a2cdcdd2SPaul Dagnelie if (err == 0) 3018a2cdcdd2SPaul Dagnelie err = rwa.err; 3019efb80947Sahrens 3020efb80947Sahrens out: 30219c3fd121SMatthew Ahrens nvlist_free(begin_nvl); 3022a7f53a56SChris Kirby if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) 3023a7f53a56SChris Kirby zfs_onexit_fd_rele(cleanup_fd); 3024a7f53a56SChris Kirby 302598110f08SMatthew Ahrens if (err != 0) { 3026efb80947Sahrens /* 30279c3fd121SMatthew Ahrens * Clean up references. If receive is not resumable, 30289c3fd121SMatthew Ahrens * destroy what we created, so we don't leave it in 30299c3fd121SMatthew Ahrens * the inconsistent state. 3030efb80947Sahrens */ 30313b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(drc); 3032efb80947Sahrens } 3033efb80947Sahrens 30343cb34c60Sahrens *voffp = ra.voff; 303568ecb2ecSPaul Dagnelie objlist_destroy(&ra.ignore_objlist); 303698110f08SMatthew Ahrens return (err); 3037efb80947Sahrens } 3038f18faf3fSek 30393cb34c60Sahrens static int 30403b2aab18SMatthew Ahrens dmu_recv_end_check(void *arg, dmu_tx_t *tx) 30413cb34c60Sahrens { 30423b2aab18SMatthew Ahrens dmu_recv_cookie_t *drc = arg; 30433b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 30443b2aab18SMatthew Ahrens int error; 30453b2aab18SMatthew Ahrens 30463b2aab18SMatthew Ahrens ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); 30473cb34c60Sahrens 30483b2aab18SMatthew Ahrens if (!drc->drc_newfs) { 30493b2aab18SMatthew Ahrens dsl_dataset_t *origin_head; 30503b2aab18SMatthew Ahrens 30513b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); 30523b2aab18SMatthew Ahrens if (error != 0) 30533b2aab18SMatthew Ahrens return (error); 305434f2f8cfSMatthew Ahrens if (drc->drc_force) { 305534f2f8cfSMatthew Ahrens /* 305634f2f8cfSMatthew Ahrens * We will destroy any snapshots in tofs (i.e. before 305734f2f8cfSMatthew Ahrens * origin_head) that are after the origin (which is 305834f2f8cfSMatthew Ahrens * the snap before drc_ds, because drc_ds can not 305934f2f8cfSMatthew Ahrens * have any snaps of its own). 306034f2f8cfSMatthew Ahrens */ 3061c1379625SJustin T. Gibbs uint64_t obj; 3062c1379625SJustin T. Gibbs 3063c1379625SJustin T. Gibbs obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 3064c1379625SJustin T. Gibbs while (obj != 3065c1379625SJustin T. Gibbs dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 306634f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 306734f2f8cfSMatthew Ahrens error = dsl_dataset_hold_obj(dp, obj, FTAG, 306834f2f8cfSMatthew Ahrens &snap); 306934f2f8cfSMatthew Ahrens if (error != 0) 3070beddaa9cSAndriy Gapon break; 307134f2f8cfSMatthew Ahrens if (snap->ds_dir != origin_head->ds_dir) 307234f2f8cfSMatthew Ahrens error = SET_ERROR(EINVAL); 307334f2f8cfSMatthew Ahrens if (error == 0) { 307434f2f8cfSMatthew Ahrens error = dsl_destroy_snapshot_check_impl( 307534f2f8cfSMatthew Ahrens snap, B_FALSE); 307634f2f8cfSMatthew Ahrens } 3077c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 307834f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 307934f2f8cfSMatthew Ahrens if (error != 0) 3080beddaa9cSAndriy Gapon break; 3081beddaa9cSAndriy Gapon } 3082beddaa9cSAndriy Gapon if (error != 0) { 3083beddaa9cSAndriy Gapon dsl_dataset_rele(origin_head, FTAG); 3084beddaa9cSAndriy Gapon return (error); 308534f2f8cfSMatthew Ahrens } 308634f2f8cfSMatthew Ahrens } 30873b2aab18SMatthew Ahrens error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, 308891948b51SKeith M Wesolowski origin_head, drc->drc_force, drc->drc_owner, tx); 30893b2aab18SMatthew Ahrens if (error != 0) { 30903b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 30913b2aab18SMatthew Ahrens return (error); 30923b2aab18SMatthew Ahrens } 30933b2aab18SMatthew Ahrens error = dsl_dataset_snapshot_check_impl(origin_head, 3094a2afb611SJerry Jelinek drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 30953b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 30963b2aab18SMatthew Ahrens if (error != 0) 30973b2aab18SMatthew Ahrens return (error); 30983b2aab18SMatthew Ahrens 30993b2aab18SMatthew Ahrens error = dsl_destroy_head_check_impl(drc->drc_ds, 1); 31003b2aab18SMatthew Ahrens } else { 31013b2aab18SMatthew Ahrens error = dsl_dataset_snapshot_check_impl(drc->drc_ds, 3102a2afb611SJerry Jelinek drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 31033b2aab18SMatthew Ahrens } 31043b2aab18SMatthew Ahrens return (error); 31053cb34c60Sahrens } 31063cb34c60Sahrens 31073cb34c60Sahrens static void 31083b2aab18SMatthew Ahrens dmu_recv_end_sync(void *arg, dmu_tx_t *tx) 31093cb34c60Sahrens { 31103b2aab18SMatthew Ahrens dmu_recv_cookie_t *drc = arg; 31113b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 31123b2aab18SMatthew Ahrens 31133b2aab18SMatthew Ahrens spa_history_log_internal_ds(drc->drc_ds, "finish receiving", 31143b2aab18SMatthew Ahrens tx, "snap=%s", drc->drc_tosnap); 31153b2aab18SMatthew Ahrens 31163b2aab18SMatthew Ahrens if (!drc->drc_newfs) { 31173b2aab18SMatthew Ahrens dsl_dataset_t *origin_head; 31183b2aab18SMatthew Ahrens 31193b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, 31203b2aab18SMatthew Ahrens &origin_head)); 312134f2f8cfSMatthew Ahrens 312234f2f8cfSMatthew Ahrens if (drc->drc_force) { 312334f2f8cfSMatthew Ahrens /* 312434f2f8cfSMatthew Ahrens * Destroy any snapshots of drc_tofs (origin_head) 312534f2f8cfSMatthew Ahrens * after the origin (the snap before drc_ds). 312634f2f8cfSMatthew Ahrens */ 3127c1379625SJustin T. Gibbs uint64_t obj; 3128c1379625SJustin T. Gibbs 3129c1379625SJustin T. Gibbs obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 3130c1379625SJustin T. Gibbs while (obj != 3131c1379625SJustin T. Gibbs dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 313234f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 313334f2f8cfSMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, 313434f2f8cfSMatthew Ahrens &snap)); 313534f2f8cfSMatthew Ahrens ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); 3136c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 313734f2f8cfSMatthew Ahrens dsl_destroy_snapshot_sync_impl(snap, 313834f2f8cfSMatthew Ahrens B_FALSE, tx); 313934f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 314034f2f8cfSMatthew Ahrens } 314134f2f8cfSMatthew Ahrens } 314234f2f8cfSMatthew Ahrens VERIFY3P(drc->drc_ds->ds_prev, ==, 314334f2f8cfSMatthew Ahrens origin_head->ds_prev); 314434f2f8cfSMatthew Ahrens 31453b2aab18SMatthew Ahrens dsl_dataset_clone_swap_sync_impl(drc->drc_ds, 31463b2aab18SMatthew Ahrens origin_head, tx); 31473b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(origin_head, 31483b2aab18SMatthew Ahrens drc->drc_tosnap, tx); 31493b2aab18SMatthew Ahrens 31503b2aab18SMatthew Ahrens /* set snapshot's creation time and guid */ 31513b2aab18SMatthew Ahrens dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); 3152c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = 31533b2aab18SMatthew Ahrens drc->drc_drrb->drr_creation_time; 3154c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_guid = 31553b2aab18SMatthew Ahrens drc->drc_drrb->drr_toguid; 3156c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= 31573b2aab18SMatthew Ahrens ~DS_FLAG_INCONSISTENT; 31583b2aab18SMatthew Ahrens 31593b2aab18SMatthew Ahrens dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 3160c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head)->ds_flags &= 3161c1379625SJustin T. Gibbs ~DS_FLAG_INCONSISTENT; 31623b2aab18SMatthew Ahrens 31632ad25b40SChris Williamson drc->drc_newsnapobj = 31642ad25b40SChris Williamson dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 31652ad25b40SChris Williamson 31663b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 31673b2aab18SMatthew Ahrens dsl_destroy_head_sync_impl(drc->drc_ds, tx); 316891948b51SKeith M Wesolowski 316991948b51SKeith M Wesolowski if (drc->drc_owner != NULL) 317091948b51SKeith M Wesolowski VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); 31713b2aab18SMatthew Ahrens } else { 31723b2aab18SMatthew Ahrens dsl_dataset_t *ds = drc->drc_ds; 31733cb34c60Sahrens 31743b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); 31753cb34c60Sahrens 31763b2aab18SMatthew Ahrens /* set snapshot's creation time and guid */ 31773b2aab18SMatthew Ahrens dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 3178c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_creation_time = 31793b2aab18SMatthew Ahrens drc->drc_drrb->drr_creation_time; 3180c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_guid = 3181c1379625SJustin T. Gibbs drc->drc_drrb->drr_toguid; 3182c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_flags &= 3183c1379625SJustin T. Gibbs ~DS_FLAG_INCONSISTENT; 31843cb34c60Sahrens 31853b2aab18SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 3186c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 31879c3fd121SMatthew Ahrens if (dsl_dataset_has_resume_receive_state(ds)) { 31889c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 31899c3fd121SMatthew Ahrens DS_FIELD_RESUME_FROMGUID, tx); 31909c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 31919c3fd121SMatthew Ahrens DS_FIELD_RESUME_OBJECT, tx); 31929c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 31939c3fd121SMatthew Ahrens DS_FIELD_RESUME_OFFSET, tx); 31949c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 31959c3fd121SMatthew Ahrens DS_FIELD_RESUME_BYTES, tx); 31969c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 31979c3fd121SMatthew Ahrens DS_FIELD_RESUME_TOGUID, tx); 31989c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 31999c3fd121SMatthew Ahrens DS_FIELD_RESUME_TONAME, tx); 32009c3fd121SMatthew Ahrens } 32012ad25b40SChris Williamson drc->drc_newsnapobj = 32022ad25b40SChris Williamson dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; 32033b2aab18SMatthew Ahrens } 32043b2aab18SMatthew Ahrens /* 32053b2aab18SMatthew Ahrens * Release the hold from dmu_recv_begin. This must be done before 32063b2aab18SMatthew Ahrens * we return to open context, so that when we free the dataset's dnode, 32073b2aab18SMatthew Ahrens * we can evict its bonus buffer. 32083b2aab18SMatthew Ahrens */ 32093b2aab18SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 32103b2aab18SMatthew Ahrens drc->drc_ds = NULL; 32113cb34c60Sahrens } 32123cb34c60Sahrens 3213ec5cf9d5SAlexander Stetsenko static int 32143b2aab18SMatthew Ahrens add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) 3215ec5cf9d5SAlexander Stetsenko { 32163b2aab18SMatthew Ahrens dsl_pool_t *dp; 3217ec5cf9d5SAlexander Stetsenko dsl_dataset_t *snapds; 3218ec5cf9d5SAlexander Stetsenko guid_map_entry_t *gmep; 3219ec5cf9d5SAlexander Stetsenko int err; 3220ec5cf9d5SAlexander Stetsenko 3221ec5cf9d5SAlexander Stetsenko ASSERT(guid_map != NULL); 3222ec5cf9d5SAlexander Stetsenko 32233b2aab18SMatthew Ahrens err = dsl_pool_hold(name, FTAG, &dp); 32243b2aab18SMatthew Ahrens if (err != 0) 32253b2aab18SMatthew Ahrens return (err); 3226de8d9cffSMatthew Ahrens gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); 3227de8d9cffSMatthew Ahrens err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds); 3228ec5cf9d5SAlexander Stetsenko if (err == 0) { 3229c1379625SJustin T. Gibbs gmep->guid = dsl_dataset_phys(snapds)->ds_guid; 3230ec5cf9d5SAlexander Stetsenko gmep->gme_ds = snapds; 3231ec5cf9d5SAlexander Stetsenko avl_add(guid_map, gmep); 32323b2aab18SMatthew Ahrens dsl_dataset_long_hold(snapds, gmep); 3233de8d9cffSMatthew Ahrens } else { 3234de8d9cffSMatthew Ahrens kmem_free(gmep, sizeof (*gmep)); 3235ec5cf9d5SAlexander Stetsenko } 3236ec5cf9d5SAlexander Stetsenko 32373b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 3238ec5cf9d5SAlexander Stetsenko return (err); 3239ec5cf9d5SAlexander Stetsenko } 3240ec5cf9d5SAlexander Stetsenko 32413b2aab18SMatthew Ahrens static int dmu_recv_end_modified_blocks = 3; 32423b2aab18SMatthew Ahrens 3243ae46e4c7SMatthew Ahrens static int 3244ae46e4c7SMatthew Ahrens dmu_recv_existing_end(dmu_recv_cookie_t *drc) 3245f18faf3fSek { 32463b2aab18SMatthew Ahrens #ifdef _KERNEL 32473b2aab18SMatthew Ahrens /* 32483b2aab18SMatthew Ahrens * We will be destroying the ds; make sure its origin is unmounted if 32493b2aab18SMatthew Ahrens * necessary. 32503b2aab18SMatthew Ahrens */ 32519adfa60dSMatthew Ahrens char name[ZFS_MAX_DATASET_NAME_LEN]; 32523b2aab18SMatthew Ahrens dsl_dataset_name(drc->drc_ds, name); 32533b2aab18SMatthew Ahrens zfs_destroy_unmount_origin(name); 32543b2aab18SMatthew Ahrens #endif 32553cb34c60Sahrens 32562ad25b40SChris Williamson return (dsl_sync_task(drc->drc_tofs, 32573b2aab18SMatthew Ahrens dmu_recv_end_check, dmu_recv_end_sync, drc, 32582ad25b40SChris Williamson dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); 3259f18faf3fSek } 3260ae46e4c7SMatthew Ahrens 3261ae46e4c7SMatthew Ahrens static int 3262ae46e4c7SMatthew Ahrens dmu_recv_new_end(dmu_recv_cookie_t *drc) 32632ad25b40SChris Williamson { 32642ad25b40SChris Williamson return (dsl_sync_task(drc->drc_tofs, 32652ad25b40SChris Williamson dmu_recv_end_check, dmu_recv_end_sync, drc, 32662ad25b40SChris Williamson dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); 32672ad25b40SChris Williamson } 32682ad25b40SChris Williamson 32692ad25b40SChris Williamson int 32702ad25b40SChris Williamson dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) 3271ae46e4c7SMatthew Ahrens { 32723b2aab18SMatthew Ahrens int error; 3273ae46e4c7SMatthew Ahrens 32742ad25b40SChris Williamson drc->drc_owner = owner; 32752ad25b40SChris Williamson 32762ad25b40SChris Williamson if (drc->drc_newfs) 32772ad25b40SChris Williamson error = dmu_recv_new_end(drc); 32782ad25b40SChris Williamson else 32792ad25b40SChris Williamson error = dmu_recv_existing_end(drc); 3280ae46e4c7SMatthew Ahrens 32813b2aab18SMatthew Ahrens if (error != 0) { 32823b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(drc); 32833b2aab18SMatthew Ahrens } else if (drc->drc_guid_to_ds_map != NULL) { 32843b2aab18SMatthew Ahrens (void) add_ds_to_guidmap(drc->drc_tofs, 32853b2aab18SMatthew Ahrens drc->drc_guid_to_ds_map, 32863b2aab18SMatthew Ahrens drc->drc_newsnapobj); 3287ae46e4c7SMatthew Ahrens } 32883b2aab18SMatthew Ahrens return (error); 3289ae46e4c7SMatthew Ahrens } 3290ae46e4c7SMatthew Ahrens 32912f3d8780SMatthew Ahrens /* 32922f3d8780SMatthew Ahrens * Return TRUE if this objset is currently being received into. 32932f3d8780SMatthew Ahrens */ 32942f3d8780SMatthew Ahrens boolean_t 32952f3d8780SMatthew Ahrens dmu_objset_is_receiving(objset_t *os) 32962f3d8780SMatthew Ahrens { 32972f3d8780SMatthew Ahrens return (os->os_dsl_dataset != NULL && 32982f3d8780SMatthew Ahrens os->os_dsl_dataset->ds_owner == dmu_recv_tag); 32992f3d8780SMatthew Ahrens } 3300