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 69*df477c0aSPaul Dagnelie /* 70*df477c0aSPaul Dagnelie * Use this to override the recordsize calculation for fast zfs send estimates. 71*df477c0aSPaul Dagnelie */ 72*df477c0aSPaul Dagnelie uint64_t zfs_override_estimate_recordsize = 0; 73*df477c0aSPaul Dagnelie 74a2cdcdd2SPaul Dagnelie #define BP_SPAN(datablkszsec, indblkshift, level) \ 75a2cdcdd2SPaul Dagnelie (((uint64_t)datablkszsec) << (SPA_MINBLOCKSHIFT + \ 76a2cdcdd2SPaul Dagnelie (level) * (indblkshift - SPA_BLKPTRSHIFT))) 77a2cdcdd2SPaul Dagnelie 789c3fd121SMatthew Ahrens static void byteswap_record(dmu_replay_record_t *drr); 799c3fd121SMatthew Ahrens 80a2cdcdd2SPaul Dagnelie struct send_thread_arg { 81a2cdcdd2SPaul Dagnelie bqueue_t q; 82a2cdcdd2SPaul Dagnelie dsl_dataset_t *ds; /* Dataset to traverse */ 83a2cdcdd2SPaul Dagnelie uint64_t fromtxg; /* Traverse from this txg */ 84a2cdcdd2SPaul Dagnelie int flags; /* flags to pass to traverse_dataset */ 85a2cdcdd2SPaul Dagnelie int error_code; 86a2cdcdd2SPaul Dagnelie boolean_t cancel; 879c3fd121SMatthew Ahrens zbookmark_phys_t resume; 88a2cdcdd2SPaul Dagnelie }; 89a2cdcdd2SPaul Dagnelie 90a2cdcdd2SPaul Dagnelie struct send_block_record { 91a2cdcdd2SPaul Dagnelie boolean_t eos_marker; /* Marks the end of the stream */ 92a2cdcdd2SPaul Dagnelie blkptr_t bp; 93a2cdcdd2SPaul Dagnelie zbookmark_phys_t zb; 94a2cdcdd2SPaul Dagnelie uint8_t indblkshift; 95a2cdcdd2SPaul Dagnelie uint16_t datablkszsec; 96a2cdcdd2SPaul Dagnelie bqueue_node_t ln; 97a2cdcdd2SPaul Dagnelie }; 98a2cdcdd2SPaul Dagnelie 99efb80947Sahrens static int 1004e3c9f44SBill Pijewski dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) 101efb80947Sahrens { 1029c3fd121SMatthew Ahrens dsl_dataset_t *ds = dmu_objset_ds(dsp->dsa_os); 103efb80947Sahrens ssize_t resid; /* have to get resid to get detailed errno */ 104c20404ffSEli Rosenthal 105c20404ffSEli Rosenthal /* 106c20404ffSEli Rosenthal * The code does not rely on this (len being a multiple of 8). We keep 107c20404ffSEli Rosenthal * this assertion because of the corresponding assertion in 108c20404ffSEli Rosenthal * receive_read(). Keeping this assertion ensures that we do not 109c20404ffSEli Rosenthal * inadvertently break backwards compatibility (causing the assertion 110c20404ffSEli Rosenthal * in receive_read() to trigger on old software). 111c20404ffSEli Rosenthal * 112c20404ffSEli Rosenthal * Removing the assertions could be rolled into a new feature that uses 113c20404ffSEli Rosenthal * data that isn't 8-byte aligned; if the assertions were removed, a 114c20404ffSEli Rosenthal * feature flag would have to be added. 115c20404ffSEli Rosenthal */ 116c20404ffSEli Rosenthal 117fb09f5aaSMadhav Suresh ASSERT0(len % 8); 118efb80947Sahrens 1194e3c9f44SBill Pijewski dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp, 120efb80947Sahrens (caddr_t)buf, len, 121efb80947Sahrens 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); 1224e3c9f44SBill Pijewski 1234e3c9f44SBill Pijewski mutex_enter(&ds->ds_sendstream_lock); 1244e3c9f44SBill Pijewski *dsp->dsa_off += len; 1254e3c9f44SBill Pijewski mutex_exit(&ds->ds_sendstream_lock); 1264e3c9f44SBill Pijewski 1274e3c9f44SBill Pijewski return (dsp->dsa_err); 128efb80947Sahrens } 129efb80947Sahrens 13098110f08SMatthew Ahrens /* 13198110f08SMatthew Ahrens * For all record types except BEGIN, fill in the checksum (overlaid in 13298110f08SMatthew Ahrens * drr_u.drr_checksum.drr_checksum). The checksum verifies everything 13398110f08SMatthew Ahrens * up to the start of the checksum itself. 13498110f08SMatthew Ahrens */ 13598110f08SMatthew Ahrens static int 13698110f08SMatthew Ahrens dump_record(dmu_sendarg_t *dsp, void *payload, int payload_len) 13798110f08SMatthew Ahrens { 13898110f08SMatthew Ahrens ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 13998110f08SMatthew Ahrens ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 140770499e1SDan Kimmel (void) fletcher_4_incremental_native(dsp->dsa_drr, 14198110f08SMatthew Ahrens offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 14298110f08SMatthew Ahrens &dsp->dsa_zc); 14312b90ee2SMatt Krantz if (dsp->dsa_drr->drr_type == DRR_BEGIN) { 14412b90ee2SMatt Krantz dsp->dsa_sent_begin = B_TRUE; 14512b90ee2SMatt Krantz } else { 14698110f08SMatthew Ahrens ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp->dsa_drr->drr_u. 14798110f08SMatthew Ahrens drr_checksum.drr_checksum)); 14898110f08SMatthew Ahrens dsp->dsa_drr->drr_u.drr_checksum.drr_checksum = dsp->dsa_zc; 14998110f08SMatthew Ahrens } 15012b90ee2SMatt Krantz if (dsp->dsa_drr->drr_type == DRR_END) { 15112b90ee2SMatt Krantz dsp->dsa_sent_end = B_TRUE; 15212b90ee2SMatt Krantz } 153770499e1SDan Kimmel (void) fletcher_4_incremental_native(&dsp->dsa_drr-> 15498110f08SMatthew Ahrens drr_u.drr_checksum.drr_checksum, 15598110f08SMatthew Ahrens sizeof (zio_cksum_t), &dsp->dsa_zc); 15698110f08SMatthew Ahrens if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 15798110f08SMatthew Ahrens return (SET_ERROR(EINTR)); 15898110f08SMatthew Ahrens if (payload_len != 0) { 159770499e1SDan Kimmel (void) fletcher_4_incremental_native(payload, payload_len, 16098110f08SMatthew Ahrens &dsp->dsa_zc); 16198110f08SMatthew Ahrens if (dump_bytes(dsp, payload, payload_len) != 0) 16298110f08SMatthew Ahrens return (SET_ERROR(EINTR)); 16398110f08SMatthew Ahrens } 16498110f08SMatthew Ahrens return (0); 16598110f08SMatthew Ahrens } 16698110f08SMatthew Ahrens 16768ecb2ecSPaul Dagnelie /* 16868ecb2ecSPaul Dagnelie * Fill in the drr_free struct, or perform aggregation if the previous record is 16968ecb2ecSPaul Dagnelie * also a free record, and the two are adjacent. 17068ecb2ecSPaul Dagnelie * 17168ecb2ecSPaul Dagnelie * Note that we send free records even for a full send, because we want to be 17268ecb2ecSPaul Dagnelie * able to receive a full send as a clone, which requires a list of all the free 17368ecb2ecSPaul Dagnelie * and freeobject records that were generated on the source. 17468ecb2ecSPaul Dagnelie */ 175efb80947Sahrens static int 1764e3c9f44SBill Pijewski dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 177efb80947Sahrens uint64_t length) 178efb80947Sahrens { 1794e3c9f44SBill Pijewski struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); 1809e69d7d0SLori Alt 1812f3d8780SMatthew Ahrens /* 1822f3d8780SMatthew Ahrens * When we receive a free record, dbuf_free_range() assumes 1832f3d8780SMatthew Ahrens * that the receiving system doesn't have any dbufs in the range 1842f3d8780SMatthew Ahrens * being freed. This is always true because there is a one-record 1852f3d8780SMatthew Ahrens * constraint: we only send one WRITE record for any given 1869c3fd121SMatthew Ahrens * object,offset. We know that the one-record constraint is 1872f3d8780SMatthew Ahrens * true because we always send data in increasing order by 1882f3d8780SMatthew Ahrens * object,offset. 1892f3d8780SMatthew Ahrens * 1902f3d8780SMatthew Ahrens * If the increasing-order constraint ever changes, we should find 1912f3d8780SMatthew Ahrens * another way to assert that the one-record constraint is still 1922f3d8780SMatthew Ahrens * satisfied. 1932f3d8780SMatthew Ahrens */ 1942f3d8780SMatthew Ahrens ASSERT(object > dsp->dsa_last_data_object || 1952f3d8780SMatthew Ahrens (object == dsp->dsa_last_data_object && 1962f3d8780SMatthew Ahrens offset > dsp->dsa_last_data_offset)); 1972f3d8780SMatthew Ahrens 198534029e5SSimon Klinkert if (length != -1ULL && offset + length < offset) 199534029e5SSimon Klinkert length = -1ULL; 200534029e5SSimon Klinkert 2019e69d7d0SLori Alt /* 2029e69d7d0SLori Alt * If there is a pending op, but it's not PENDING_FREE, push it out, 2039e69d7d0SLori Alt * since free block aggregation can only be done for blocks of the 2049e69d7d0SLori Alt * same type (i.e., DRR_FREE records can only be aggregated with 2059e69d7d0SLori Alt * other DRR_FREE records. DRR_FREEOBJECTS records can only be 2069e69d7d0SLori Alt * aggregated with other DRR_FREEOBJECTS records. 2079e69d7d0SLori Alt */ 2084e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE && 2094e3c9f44SBill Pijewski dsp->dsa_pending_op != PENDING_FREE) { 21098110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 211be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2124e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2139e69d7d0SLori Alt } 2149e69d7d0SLori Alt 2154e3c9f44SBill Pijewski if (dsp->dsa_pending_op == PENDING_FREE) { 2169e69d7d0SLori Alt /* 2179e69d7d0SLori Alt * There should never be a PENDING_FREE if length is -1 2189e69d7d0SLori Alt * (because dump_dnode is the only place where this 2199e69d7d0SLori Alt * function is called with a -1, and only after flushing 2209e69d7d0SLori Alt * any pending record). 2219e69d7d0SLori Alt */ 2229e69d7d0SLori Alt ASSERT(length != -1ULL); 2239e69d7d0SLori Alt /* 2249e69d7d0SLori Alt * Check to see whether this free block can be aggregated 2259e69d7d0SLori Alt * with pending one. 2269e69d7d0SLori Alt */ 2279e69d7d0SLori Alt if (drrf->drr_object == object && drrf->drr_offset + 2289e69d7d0SLori Alt drrf->drr_length == offset) { 2299e69d7d0SLori Alt drrf->drr_length += length; 2309e69d7d0SLori Alt return (0); 2319e69d7d0SLori Alt } else { 2329e69d7d0SLori Alt /* not a continuation. Push out pending record */ 23398110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 234be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2354e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2369e69d7d0SLori Alt } 2379e69d7d0SLori Alt } 2389e69d7d0SLori Alt /* create a FREE record and make it pending */ 2394e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 2404e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_FREE; 2419e69d7d0SLori Alt drrf->drr_object = object; 2429e69d7d0SLori Alt drrf->drr_offset = offset; 2439e69d7d0SLori Alt drrf->drr_length = length; 2444e3c9f44SBill Pijewski drrf->drr_toguid = dsp->dsa_toguid; 2459e69d7d0SLori Alt if (length == -1ULL) { 24698110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 247be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2489e69d7d0SLori Alt } else { 2494e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_FREE; 2509e69d7d0SLori Alt } 251efb80947Sahrens 252efb80947Sahrens return (0); 253efb80947Sahrens } 254efb80947Sahrens 255efb80947Sahrens static int 2565d7b4d43SMatthew Ahrens dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, 2575602294fSDan Kimmel uint64_t object, uint64_t offset, int lsize, int psize, const blkptr_t *bp, 2585602294fSDan Kimmel void *data) 259efb80947Sahrens { 2605602294fSDan Kimmel uint64_t payload_size; 2614e3c9f44SBill Pijewski struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); 2629e69d7d0SLori Alt 2632f3d8780SMatthew Ahrens /* 2642f3d8780SMatthew Ahrens * We send data in increasing object, offset order. 2652f3d8780SMatthew Ahrens * See comment in dump_free() for details. 2662f3d8780SMatthew Ahrens */ 2672f3d8780SMatthew Ahrens ASSERT(object > dsp->dsa_last_data_object || 2682f3d8780SMatthew Ahrens (object == dsp->dsa_last_data_object && 2692f3d8780SMatthew Ahrens offset > dsp->dsa_last_data_offset)); 2702f3d8780SMatthew Ahrens dsp->dsa_last_data_object = object; 2715602294fSDan Kimmel dsp->dsa_last_data_offset = offset + lsize - 1; 2728e714474SLori Alt 2739e69d7d0SLori Alt /* 2749e69d7d0SLori Alt * If there is any kind of pending aggregation (currently either 2759e69d7d0SLori Alt * a grouping of free objects or free blocks), push it out to 2769e69d7d0SLori Alt * the stream, since aggregation can't be done across operations 2779e69d7d0SLori Alt * of different types. 2789e69d7d0SLori Alt */ 2794e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 28098110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 281be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2824e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2839e69d7d0SLori Alt } 28498110f08SMatthew Ahrens /* write a WRITE record */ 2854e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 2864e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_WRITE; 2879e69d7d0SLori Alt drrw->drr_object = object; 2889e69d7d0SLori Alt drrw->drr_type = type; 2899e69d7d0SLori Alt drrw->drr_offset = offset; 2904e3c9f44SBill Pijewski drrw->drr_toguid = dsp->dsa_toguid; 2915602294fSDan Kimmel drrw->drr_logical_size = lsize; 2925602294fSDan Kimmel 2935602294fSDan Kimmel /* only set the compression fields if the buf is compressed */ 2945602294fSDan Kimmel if (lsize != psize) { 2955602294fSDan Kimmel ASSERT(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED); 2965602294fSDan Kimmel ASSERT(!BP_IS_EMBEDDED(bp)); 2975602294fSDan Kimmel ASSERT(!BP_SHOULD_BYTESWAP(bp)); 2985602294fSDan Kimmel ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp))); 2995602294fSDan Kimmel ASSERT3U(BP_GET_COMPRESS(bp), !=, ZIO_COMPRESS_OFF); 3005602294fSDan Kimmel ASSERT3S(psize, >, 0); 3015602294fSDan Kimmel ASSERT3S(lsize, >=, psize); 3025602294fSDan Kimmel 3035602294fSDan Kimmel drrw->drr_compressiontype = BP_GET_COMPRESS(bp); 3045602294fSDan Kimmel drrw->drr_compressed_size = psize; 3055602294fSDan Kimmel payload_size = drrw->drr_compressed_size; 3065602294fSDan Kimmel } else { 3075602294fSDan Kimmel payload_size = drrw->drr_logical_size; 3085602294fSDan Kimmel } 3095602294fSDan Kimmel 310b5152584SMatthew Ahrens if (bp == NULL || BP_IS_EMBEDDED(bp)) { 3115d7b4d43SMatthew Ahrens /* 312b5152584SMatthew Ahrens * There's no pre-computed checksum for partial-block 313b5152584SMatthew Ahrens * writes or embedded BP's, so (like 314b5152584SMatthew Ahrens * fletcher4-checkummed blocks) userland will have to 315b5152584SMatthew Ahrens * compute a dedup-capable checksum itself. 3165d7b4d43SMatthew Ahrens */ 3175d7b4d43SMatthew Ahrens drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; 3185d7b4d43SMatthew Ahrens } else { 3195d7b4d43SMatthew Ahrens drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); 32045818ee1SMatthew Ahrens if (zio_checksum_table[drrw->drr_checksumtype].ci_flags & 32145818ee1SMatthew Ahrens ZCHECKSUM_FLAG_DEDUP) 3225d7b4d43SMatthew Ahrens drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP; 3235d7b4d43SMatthew Ahrens DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); 3245d7b4d43SMatthew Ahrens DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); 3255d7b4d43SMatthew Ahrens DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); 3265d7b4d43SMatthew Ahrens drrw->drr_key.ddk_cksum = bp->blk_cksum; 3275d7b4d43SMatthew Ahrens } 328efb80947Sahrens 3295602294fSDan Kimmel if (dump_record(dsp, data, payload_size) != 0) 330be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 331efb80947Sahrens return (0); 332efb80947Sahrens } 333efb80947Sahrens 3345d7b4d43SMatthew Ahrens static int 3355d7b4d43SMatthew Ahrens dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 3365d7b4d43SMatthew Ahrens int blksz, const blkptr_t *bp) 3375d7b4d43SMatthew Ahrens { 3385d7b4d43SMatthew Ahrens char buf[BPE_PAYLOAD_SIZE]; 3395d7b4d43SMatthew Ahrens struct drr_write_embedded *drrw = 3405d7b4d43SMatthew Ahrens &(dsp->dsa_drr->drr_u.drr_write_embedded); 3415d7b4d43SMatthew Ahrens 3425d7b4d43SMatthew Ahrens if (dsp->dsa_pending_op != PENDING_NONE) { 34398110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 3445d7b4d43SMatthew Ahrens return (EINTR); 3455d7b4d43SMatthew Ahrens dsp->dsa_pending_op = PENDING_NONE; 3465d7b4d43SMatthew Ahrens } 3475d7b4d43SMatthew Ahrens 3485d7b4d43SMatthew Ahrens ASSERT(BP_IS_EMBEDDED(bp)); 3495d7b4d43SMatthew Ahrens 3505d7b4d43SMatthew Ahrens bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 3515d7b4d43SMatthew Ahrens dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; 3525d7b4d43SMatthew Ahrens drrw->drr_object = object; 3535d7b4d43SMatthew Ahrens drrw->drr_offset = offset; 3545d7b4d43SMatthew Ahrens drrw->drr_length = blksz; 3555d7b4d43SMatthew Ahrens drrw->drr_toguid = dsp->dsa_toguid; 3565d7b4d43SMatthew Ahrens drrw->drr_compression = BP_GET_COMPRESS(bp); 3575d7b4d43SMatthew Ahrens drrw->drr_etype = BPE_GET_ETYPE(bp); 3585d7b4d43SMatthew Ahrens drrw->drr_lsize = BPE_GET_LSIZE(bp); 3595d7b4d43SMatthew Ahrens drrw->drr_psize = BPE_GET_PSIZE(bp); 3605d7b4d43SMatthew Ahrens 3615d7b4d43SMatthew Ahrens decode_embedded_bp_compressed(bp, buf); 3625d7b4d43SMatthew Ahrens 36398110f08SMatthew Ahrens if (dump_record(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) 3645d7b4d43SMatthew Ahrens return (EINTR); 3655d7b4d43SMatthew Ahrens return (0); 3665d7b4d43SMatthew Ahrens } 3675d7b4d43SMatthew Ahrens 3680a586ceaSMark Shellenbaum static int 3694e3c9f44SBill Pijewski dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data) 3700a586ceaSMark Shellenbaum { 3714e3c9f44SBill Pijewski struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); 3720a586ceaSMark Shellenbaum 3734e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 37498110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 375be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3764e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 3770a586ceaSMark Shellenbaum } 3780a586ceaSMark Shellenbaum 3790a586ceaSMark Shellenbaum /* write a SPILL record */ 3804e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 3814e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_SPILL; 3820a586ceaSMark Shellenbaum drrs->drr_object = object; 3830a586ceaSMark Shellenbaum drrs->drr_length = blksz; 3844e3c9f44SBill Pijewski drrs->drr_toguid = dsp->dsa_toguid; 3850a586ceaSMark Shellenbaum 38698110f08SMatthew Ahrens if (dump_record(dsp, data, blksz) != 0) 387be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3880a586ceaSMark Shellenbaum return (0); 3890a586ceaSMark Shellenbaum } 3900a586ceaSMark Shellenbaum 391efb80947Sahrens static int 3924e3c9f44SBill Pijewski dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) 393efb80947Sahrens { 3944e3c9f44SBill Pijewski struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); 3959e69d7d0SLori Alt 3969e69d7d0SLori Alt /* 3979e69d7d0SLori Alt * If there is a pending op, but it's not PENDING_FREEOBJECTS, 3989e69d7d0SLori Alt * push it out, since free block aggregation can only be done for 3999e69d7d0SLori Alt * blocks of the same type (i.e., DRR_FREE records can only be 4009e69d7d0SLori Alt * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records 4019e69d7d0SLori Alt * can only be aggregated with other DRR_FREEOBJECTS records. 4029e69d7d0SLori Alt */ 4034e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE && 4044e3c9f44SBill Pijewski dsp->dsa_pending_op != PENDING_FREEOBJECTS) { 40598110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 406be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4074e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4089e69d7d0SLori Alt } 4094e3c9f44SBill Pijewski if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { 4109e69d7d0SLori Alt /* 4119e69d7d0SLori Alt * See whether this free object array can be aggregated 4129e69d7d0SLori Alt * with pending one 4139e69d7d0SLori Alt */ 4149e69d7d0SLori Alt if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { 4159e69d7d0SLori Alt drrfo->drr_numobjs += numobjs; 4169e69d7d0SLori Alt return (0); 4179e69d7d0SLori Alt } else { 4189e69d7d0SLori Alt /* can't be aggregated. Push out pending record */ 41998110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 420be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4214e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4229e69d7d0SLori Alt } 4239e69d7d0SLori Alt } 4249e69d7d0SLori Alt 425efb80947Sahrens /* write a FREEOBJECTS record */ 4264e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 4274e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; 4289e69d7d0SLori Alt drrfo->drr_firstobj = firstobj; 4299e69d7d0SLori Alt drrfo->drr_numobjs = numobjs; 4304e3c9f44SBill Pijewski drrfo->drr_toguid = dsp->dsa_toguid; 4319e69d7d0SLori Alt 4324e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_FREEOBJECTS; 433efb80947Sahrens 434efb80947Sahrens return (0); 435efb80947Sahrens } 436efb80947Sahrens 437efb80947Sahrens static int 4384e3c9f44SBill Pijewski dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) 439efb80947Sahrens { 4404e3c9f44SBill Pijewski struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); 4419e69d7d0SLori Alt 4429c3fd121SMatthew Ahrens if (object < dsp->dsa_resume_object) { 4439c3fd121SMatthew Ahrens /* 4449c3fd121SMatthew Ahrens * Note: when resuming, we will visit all the dnodes in 4459c3fd121SMatthew Ahrens * the block of dnodes that we are resuming from. In 4469c3fd121SMatthew Ahrens * this case it's unnecessary to send the dnodes prior to 4479c3fd121SMatthew Ahrens * the one we are resuming from. We should be at most one 4489c3fd121SMatthew Ahrens * block's worth of dnodes behind the resume point. 4499c3fd121SMatthew Ahrens */ 4509c3fd121SMatthew Ahrens ASSERT3U(dsp->dsa_resume_object - object, <, 4519c3fd121SMatthew Ahrens 1 << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)); 4529c3fd121SMatthew Ahrens return (0); 4539c3fd121SMatthew Ahrens } 4549c3fd121SMatthew Ahrens 455efb80947Sahrens if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) 4564e3c9f44SBill Pijewski return (dump_freeobjects(dsp, object, 1)); 457efb80947Sahrens 4584e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 45998110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 460be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4614e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4629e69d7d0SLori Alt } 4639e69d7d0SLori Alt 464efb80947Sahrens /* write an OBJECT record */ 4654e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 4664e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_OBJECT; 4679e69d7d0SLori Alt drro->drr_object = object; 4689e69d7d0SLori Alt drro->drr_type = dnp->dn_type; 4699e69d7d0SLori Alt drro->drr_bonustype = dnp->dn_bonustype; 4709e69d7d0SLori Alt drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 4719e69d7d0SLori Alt drro->drr_bonuslen = dnp->dn_bonuslen; 4729e69d7d0SLori Alt drro->drr_checksumtype = dnp->dn_checksum; 4739e69d7d0SLori Alt drro->drr_compress = dnp->dn_compress; 4744e3c9f44SBill Pijewski drro->drr_toguid = dsp->dsa_toguid; 4759e69d7d0SLori Alt 476b5152584SMatthew Ahrens if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 477b5152584SMatthew Ahrens drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) 478b5152584SMatthew Ahrens drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; 479b5152584SMatthew Ahrens 48098110f08SMatthew Ahrens if (dump_record(dsp, DN_BONUS(dnp), 48198110f08SMatthew Ahrens P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0) { 482be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 48398110f08SMatthew Ahrens } 484efb80947Sahrens 4852f3d8780SMatthew Ahrens /* Free anything past the end of the file. */ 4864e3c9f44SBill Pijewski if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * 4872f3d8780SMatthew Ahrens (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0) 488be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4893b2aab18SMatthew Ahrens if (dsp->dsa_err != 0) 490be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 491efb80947Sahrens return (0); 492efb80947Sahrens } 493efb80947Sahrens 4945d7b4d43SMatthew Ahrens static boolean_t 4955d7b4d43SMatthew Ahrens backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) 4965d7b4d43SMatthew Ahrens { 4975d7b4d43SMatthew Ahrens if (!BP_IS_EMBEDDED(bp)) 4985d7b4d43SMatthew Ahrens return (B_FALSE); 4995d7b4d43SMatthew Ahrens 5005d7b4d43SMatthew Ahrens /* 5015d7b4d43SMatthew Ahrens * Compression function must be legacy, or explicitly enabled. 5025d7b4d43SMatthew Ahrens */ 5035d7b4d43SMatthew Ahrens if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && 5045602294fSDan Kimmel !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LZ4))) 5055d7b4d43SMatthew Ahrens return (B_FALSE); 5065d7b4d43SMatthew Ahrens 5075d7b4d43SMatthew Ahrens /* 5085d7b4d43SMatthew Ahrens * Embed type must be explicitly enabled. 5095d7b4d43SMatthew Ahrens */ 5105d7b4d43SMatthew Ahrens switch (BPE_GET_ETYPE(bp)) { 5115d7b4d43SMatthew Ahrens case BP_EMBEDDED_TYPE_DATA: 5125d7b4d43SMatthew Ahrens if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) 5135d7b4d43SMatthew Ahrens return (B_TRUE); 5145d7b4d43SMatthew Ahrens break; 5155d7b4d43SMatthew Ahrens default: 5165d7b4d43SMatthew Ahrens return (B_FALSE); 5175d7b4d43SMatthew Ahrens } 5185d7b4d43SMatthew Ahrens return (B_FALSE); 5195d7b4d43SMatthew Ahrens } 5205d7b4d43SMatthew Ahrens 521a2cdcdd2SPaul Dagnelie /* 522a2cdcdd2SPaul Dagnelie * This is the callback function to traverse_dataset that acts as the worker 523a2cdcdd2SPaul Dagnelie * thread for dmu_send_impl. 524a2cdcdd2SPaul Dagnelie */ 525a2cdcdd2SPaul Dagnelie /*ARGSUSED*/ 526a2cdcdd2SPaul Dagnelie static int 527a2cdcdd2SPaul Dagnelie send_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 528a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb, const struct dnode_phys *dnp, void *arg) 529a2cdcdd2SPaul Dagnelie { 530a2cdcdd2SPaul Dagnelie struct send_thread_arg *sta = arg; 531a2cdcdd2SPaul Dagnelie struct send_block_record *record; 532a2cdcdd2SPaul Dagnelie uint64_t record_size; 533a2cdcdd2SPaul Dagnelie int err = 0; 534efb80947Sahrens 5359c3fd121SMatthew Ahrens ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || 5369c3fd121SMatthew Ahrens zb->zb_object >= sta->resume.zb_object); 5379c3fd121SMatthew Ahrens 538a2cdcdd2SPaul Dagnelie if (sta->cancel) 539a2cdcdd2SPaul Dagnelie return (SET_ERROR(EINTR)); 540a2cdcdd2SPaul Dagnelie 541a2cdcdd2SPaul Dagnelie if (bp == NULL) { 542a2cdcdd2SPaul Dagnelie ASSERT3U(zb->zb_level, ==, ZB_DNODE_LEVEL); 543a2cdcdd2SPaul Dagnelie return (0); 544a2cdcdd2SPaul Dagnelie } else if (zb->zb_level < 0) { 545a2cdcdd2SPaul Dagnelie return (0); 546a2cdcdd2SPaul Dagnelie } 547a2cdcdd2SPaul Dagnelie 548a2cdcdd2SPaul Dagnelie record = kmem_zalloc(sizeof (struct send_block_record), KM_SLEEP); 549a2cdcdd2SPaul Dagnelie record->eos_marker = B_FALSE; 550a2cdcdd2SPaul Dagnelie record->bp = *bp; 551a2cdcdd2SPaul Dagnelie record->zb = *zb; 552a2cdcdd2SPaul Dagnelie record->indblkshift = dnp->dn_indblkshift; 553a2cdcdd2SPaul Dagnelie record->datablkszsec = dnp->dn_datablkszsec; 554a2cdcdd2SPaul Dagnelie record_size = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 555a2cdcdd2SPaul Dagnelie bqueue_enqueue(&sta->q, record, record_size); 556a2cdcdd2SPaul Dagnelie 557a2cdcdd2SPaul Dagnelie return (err); 558a2cdcdd2SPaul Dagnelie } 559a2cdcdd2SPaul Dagnelie 560a2cdcdd2SPaul Dagnelie /* 561a2cdcdd2SPaul Dagnelie * This function kicks off the traverse_dataset. It also handles setting the 562a2cdcdd2SPaul Dagnelie * error code of the thread in case something goes wrong, and pushes the End of 563a2cdcdd2SPaul Dagnelie * Stream record when the traverse_dataset call has finished. If there is no 564a2cdcdd2SPaul Dagnelie * dataset to traverse, the thread immediately pushes End of Stream marker. 565a2cdcdd2SPaul Dagnelie */ 566a2cdcdd2SPaul Dagnelie static void 567a2cdcdd2SPaul Dagnelie send_traverse_thread(void *arg) 568a2cdcdd2SPaul Dagnelie { 569a2cdcdd2SPaul Dagnelie struct send_thread_arg *st_arg = arg; 570a2cdcdd2SPaul Dagnelie int err; 571a2cdcdd2SPaul Dagnelie struct send_block_record *data; 572a2cdcdd2SPaul Dagnelie 573a2cdcdd2SPaul Dagnelie if (st_arg->ds != NULL) { 5749c3fd121SMatthew Ahrens err = traverse_dataset_resume(st_arg->ds, 5759c3fd121SMatthew Ahrens st_arg->fromtxg, &st_arg->resume, 5769c3fd121SMatthew Ahrens st_arg->flags, send_cb, st_arg); 5779c3fd121SMatthew Ahrens 578a2cdcdd2SPaul Dagnelie if (err != EINTR) 579a2cdcdd2SPaul Dagnelie st_arg->error_code = err; 580a2cdcdd2SPaul Dagnelie } 581a2cdcdd2SPaul Dagnelie data = kmem_zalloc(sizeof (*data), KM_SLEEP); 582a2cdcdd2SPaul Dagnelie data->eos_marker = B_TRUE; 583a2cdcdd2SPaul Dagnelie bqueue_enqueue(&st_arg->q, data, 1); 584f2c1e9bcSJorgen Lundman thread_exit(); 585a2cdcdd2SPaul Dagnelie } 586a2cdcdd2SPaul Dagnelie 587a2cdcdd2SPaul Dagnelie /* 588a2cdcdd2SPaul Dagnelie * This function actually handles figuring out what kind of record needs to be 589a2cdcdd2SPaul Dagnelie * dumped, reading the data (which has hopefully been prefetched), and calling 590a2cdcdd2SPaul Dagnelie * the appropriate helper function. 591a2cdcdd2SPaul Dagnelie */ 592efb80947Sahrens static int 593a2cdcdd2SPaul Dagnelie do_dump(dmu_sendarg_t *dsa, struct send_block_record *data) 594efb80947Sahrens { 595a2cdcdd2SPaul Dagnelie dsl_dataset_t *ds = dmu_objset_ds(dsa->dsa_os); 596a2cdcdd2SPaul Dagnelie const blkptr_t *bp = &data->bp; 597a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb = &data->zb; 598a2cdcdd2SPaul Dagnelie uint8_t indblkshift = data->indblkshift; 599a2cdcdd2SPaul Dagnelie uint16_t dblkszsec = data->datablkszsec; 600a2cdcdd2SPaul Dagnelie spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 601efb80947Sahrens dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; 602efb80947Sahrens int err = 0; 603efb80947Sahrens 604a2cdcdd2SPaul Dagnelie ASSERT3U(zb->zb_level, >=, 0); 605efb80947Sahrens 6069c3fd121SMatthew Ahrens ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || 6079c3fd121SMatthew Ahrens zb->zb_object >= dsa->dsa_resume_object); 6089c3fd121SMatthew Ahrens 609b24ab676SJeff Bonwick if (zb->zb_object != DMU_META_DNODE_OBJECT && 610b24ab676SJeff Bonwick DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { 61114843421SMatthew Ahrens return (0); 61243466aaeSMax Grossman } else if (BP_IS_HOLE(bp) && 61343466aaeSMax Grossman zb->zb_object == DMU_META_DNODE_OBJECT) { 614a2cdcdd2SPaul Dagnelie uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); 61588b7b0f2SMatthew Ahrens uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; 616a2cdcdd2SPaul Dagnelie err = dump_freeobjects(dsa, dnobj, span >> DNODE_SHIFT); 61743466aaeSMax Grossman } else if (BP_IS_HOLE(bp)) { 618a2cdcdd2SPaul Dagnelie uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); 619a2cdcdd2SPaul Dagnelie uint64_t offset = zb->zb_blkid * span; 620a2cdcdd2SPaul Dagnelie err = dump_free(dsa, zb->zb_object, offset, span); 62188b7b0f2SMatthew Ahrens } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { 62288b7b0f2SMatthew Ahrens return (0); 62388b7b0f2SMatthew Ahrens } else if (type == DMU_OT_DNODE) { 624efb80947Sahrens int blksz = BP_GET_LSIZE(bp); 6257adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 62688b7b0f2SMatthew Ahrens arc_buf_t *abuf; 627efb80947Sahrens 628a2cdcdd2SPaul Dagnelie ASSERT0(zb->zb_level); 629a2cdcdd2SPaul Dagnelie 6301b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 6311b912ec7SGeorge Wilson ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 6321b912ec7SGeorge Wilson &aflags, zb) != 0) 633be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 63488b7b0f2SMatthew Ahrens 635a2cdcdd2SPaul Dagnelie dnode_phys_t *blk = abuf->b_data; 636a2cdcdd2SPaul Dagnelie uint64_t dnobj = zb->zb_blkid * (blksz >> DNODE_SHIFT); 637a2cdcdd2SPaul Dagnelie for (int i = 0; i < blksz >> DNODE_SHIFT; i++) { 638a2cdcdd2SPaul Dagnelie err = dump_dnode(dsa, dnobj + i, blk + i); 6393b2aab18SMatthew Ahrens if (err != 0) 640efb80947Sahrens break; 641efb80947Sahrens } 642dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 6430a586ceaSMark Shellenbaum } else if (type == DMU_OT_SA) { 6447adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 6450a586ceaSMark Shellenbaum arc_buf_t *abuf; 6460a586ceaSMark Shellenbaum int blksz = BP_GET_LSIZE(bp); 6470a586ceaSMark Shellenbaum 6481b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 6491b912ec7SGeorge Wilson ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 6501b912ec7SGeorge Wilson &aflags, zb) != 0) 651be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 6520a586ceaSMark Shellenbaum 653a2cdcdd2SPaul Dagnelie err = dump_spill(dsa, zb->zb_object, blksz, abuf->b_data); 654dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 655a2cdcdd2SPaul Dagnelie } else if (backup_do_embed(dsa, bp)) { 6565d7b4d43SMatthew Ahrens /* it's an embedded level-0 block of a regular object */ 657a2cdcdd2SPaul Dagnelie int blksz = dblkszsec << SPA_MINBLOCKSHIFT; 658a2cdcdd2SPaul Dagnelie ASSERT0(zb->zb_level); 659a2cdcdd2SPaul Dagnelie err = dump_write_embedded(dsa, zb->zb_object, 6605d7b4d43SMatthew Ahrens zb->zb_blkid * blksz, blksz, bp); 661a2cdcdd2SPaul Dagnelie } else { 662a2cdcdd2SPaul Dagnelie /* it's a level-0 block of a regular object */ 6637adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 66488b7b0f2SMatthew Ahrens arc_buf_t *abuf; 665a2cdcdd2SPaul Dagnelie int blksz = dblkszsec << SPA_MINBLOCKSHIFT; 666b5152584SMatthew Ahrens uint64_t offset; 66788b7b0f2SMatthew Ahrens 6685602294fSDan Kimmel /* 6695602294fSDan Kimmel * If we have large blocks stored on disk but the send flags 6705602294fSDan Kimmel * don't allow us to send large blocks, we split the data from 6715602294fSDan Kimmel * the arc buf into chunks. 6725602294fSDan Kimmel */ 6735602294fSDan Kimmel boolean_t split_large_blocks = blksz > SPA_OLD_MAXBLOCKSIZE && 6745602294fSDan Kimmel !(dsa->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS); 6755602294fSDan Kimmel /* 6765602294fSDan Kimmel * We should only request compressed data from the ARC if all 6775602294fSDan Kimmel * the following are true: 6785602294fSDan Kimmel * - stream compression was requested 6795602294fSDan Kimmel * - we aren't splitting large blocks into smaller chunks 6805602294fSDan Kimmel * - the data won't need to be byteswapped before sending 6815602294fSDan Kimmel * - this isn't an embedded block 6825602294fSDan Kimmel * - this isn't metadata (if receiving on a different endian 6835602294fSDan Kimmel * system it can be byteswapped more easily) 6845602294fSDan Kimmel */ 6855602294fSDan Kimmel boolean_t request_compressed = 6865602294fSDan Kimmel (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED) && 6875602294fSDan Kimmel !split_large_blocks && !BP_SHOULD_BYTESWAP(bp) && 6885602294fSDan Kimmel !BP_IS_EMBEDDED(bp) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp)); 6895602294fSDan Kimmel 6905602294fSDan Kimmel ASSERT0(zb->zb_level); 6915602294fSDan Kimmel ASSERT(zb->zb_object > dsa->dsa_resume_object || 6925602294fSDan Kimmel (zb->zb_object == dsa->dsa_resume_object && 6935602294fSDan Kimmel zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); 6945602294fSDan Kimmel 69578f17100SMatthew Ahrens ASSERT0(zb->zb_level); 6969c3fd121SMatthew Ahrens ASSERT(zb->zb_object > dsa->dsa_resume_object || 6979c3fd121SMatthew Ahrens (zb->zb_object == dsa->dsa_resume_object && 6989c3fd121SMatthew Ahrens zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); 6999c3fd121SMatthew Ahrens 7005602294fSDan Kimmel ASSERT3U(blksz, ==, BP_GET_LSIZE(bp)); 7015602294fSDan Kimmel 7025602294fSDan Kimmel enum zio_flag zioflags = ZIO_FLAG_CANFAIL; 7035602294fSDan Kimmel if (request_compressed) 7045602294fSDan Kimmel zioflags |= ZIO_FLAG_RAW; 7051b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 7065602294fSDan Kimmel ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) { 70719b94df9SMatthew Ahrens if (zfs_send_corrupt_data) { 70819b94df9SMatthew Ahrens /* Send a block filled with 0x"zfs badd bloc" */ 7095602294fSDan Kimmel abuf = arc_alloc_buf(spa, &abuf, ARC_BUFC_DATA, 7105602294fSDan Kimmel blksz); 71119b94df9SMatthew Ahrens uint64_t *ptr; 71219b94df9SMatthew Ahrens for (ptr = abuf->b_data; 71319b94df9SMatthew Ahrens (char *)ptr < (char *)abuf->b_data + blksz; 71419b94df9SMatthew Ahrens ptr++) 7158c76e076SBrian Behlendorf *ptr = 0x2f5baddb10cULL; 71619b94df9SMatthew Ahrens } else { 717be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 71819b94df9SMatthew Ahrens } 71919b94df9SMatthew Ahrens } 72088b7b0f2SMatthew Ahrens 721b5152584SMatthew Ahrens offset = zb->zb_blkid * blksz; 722b5152584SMatthew Ahrens 7235602294fSDan Kimmel if (split_large_blocks) { 7245602294fSDan Kimmel ASSERT3U(arc_get_compression(abuf), ==, 7255602294fSDan Kimmel ZIO_COMPRESS_OFF); 726b5152584SMatthew Ahrens char *buf = abuf->b_data; 727b5152584SMatthew Ahrens while (blksz > 0 && err == 0) { 728b5152584SMatthew Ahrens int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); 729a2cdcdd2SPaul Dagnelie err = dump_write(dsa, type, zb->zb_object, 7305602294fSDan Kimmel offset, n, n, NULL, buf); 731b5152584SMatthew Ahrens offset += n; 732b5152584SMatthew Ahrens buf += n; 733b5152584SMatthew Ahrens blksz -= n; 734b5152584SMatthew Ahrens } 735b5152584SMatthew Ahrens } else { 7365602294fSDan Kimmel err = dump_write(dsa, type, zb->zb_object, offset, 7375602294fSDan Kimmel blksz, arc_buf_size(abuf), bp, abuf->b_data); 738b5152584SMatthew Ahrens } 739dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 740efb80947Sahrens } 741efb80947Sahrens 742efb80947Sahrens ASSERT(err == 0 || err == EINTR); 743efb80947Sahrens return (err); 744efb80947Sahrens } 745efb80947Sahrens 7464445fffbSMatthew Ahrens /* 747a2cdcdd2SPaul Dagnelie * Pop the new data off the queue, and free the old data. 748a2cdcdd2SPaul Dagnelie */ 749a2cdcdd2SPaul Dagnelie static struct send_block_record * 750a2cdcdd2SPaul Dagnelie get_next_record(bqueue_t *bq, struct send_block_record *data) 751a2cdcdd2SPaul Dagnelie { 752a2cdcdd2SPaul Dagnelie struct send_block_record *tmp = bqueue_dequeue(bq); 753a2cdcdd2SPaul Dagnelie kmem_free(data, sizeof (*data)); 754a2cdcdd2SPaul Dagnelie return (tmp); 755a2cdcdd2SPaul Dagnelie } 756a2cdcdd2SPaul Dagnelie 757a2cdcdd2SPaul Dagnelie /* 758a2cdcdd2SPaul Dagnelie * Actually do the bulk of the work in a zfs send. 759a2cdcdd2SPaul Dagnelie * 760a2cdcdd2SPaul Dagnelie * Note: Releases dp using the specified tag. 7614445fffbSMatthew Ahrens */ 7623b2aab18SMatthew Ahrens static int 763a2cdcdd2SPaul Dagnelie dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *to_ds, 7645602294fSDan Kimmel zfs_bookmark_phys_t *ancestor_zb, boolean_t is_clone, 7655602294fSDan Kimmel boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, 7665602294fSDan Kimmel int outfd, uint64_t resumeobj, uint64_t resumeoff, 7679c3fd121SMatthew Ahrens vnode_t *vp, offset_t *off) 768efb80947Sahrens { 7693b2aab18SMatthew Ahrens objset_t *os; 770efb80947Sahrens dmu_replay_record_t *drr; 7714e3c9f44SBill Pijewski dmu_sendarg_t *dsp; 772efb80947Sahrens int err; 7733cb34c60Sahrens uint64_t fromtxg = 0; 7745d7b4d43SMatthew Ahrens uint64_t featureflags = 0; 7759c3fd121SMatthew Ahrens struct send_thread_arg to_arg = { 0 }; 776efb80947Sahrens 777a2cdcdd2SPaul Dagnelie err = dmu_objset_from_ds(to_ds, &os); 7783b2aab18SMatthew Ahrens if (err != 0) { 7793b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 7803b2aab18SMatthew Ahrens return (err); 7813b2aab18SMatthew Ahrens } 782efb80947Sahrens 783efb80947Sahrens drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 784efb80947Sahrens drr->drr_type = DRR_BEGIN; 785efb80947Sahrens drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 7869e69d7d0SLori Alt DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, 7879e69d7d0SLori Alt DMU_SUBSTREAM); 788dc7cd546SMark Shellenbaum 789dc7cd546SMark Shellenbaum #ifdef _KERNEL 7903b2aab18SMatthew Ahrens if (dmu_objset_type(os) == DMU_OST_ZFS) { 791dc7cd546SMark Shellenbaum uint64_t version; 7923b2aab18SMatthew Ahrens if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { 7934e3c9f44SBill Pijewski kmem_free(drr, sizeof (dmu_replay_record_t)); 7943b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 795be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 7964e3c9f44SBill Pijewski } 7973b2aab18SMatthew Ahrens if (version >= ZPL_VERSION_SA) { 7985d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 799dc7cd546SMark Shellenbaum } 800dc7cd546SMark Shellenbaum } 801dc7cd546SMark Shellenbaum #endif 802dc7cd546SMark Shellenbaum 803ca0cc391SMatthew Ahrens if (large_block_ok && to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS]) 804b5152584SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; 8055d7b4d43SMatthew Ahrens if (embedok && 8065d7b4d43SMatthew Ahrens spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { 8075d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; 8085d7b4d43SMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 8095602294fSDan Kimmel featureflags |= DMU_BACKUP_FEATURE_LZ4; 8105602294fSDan Kimmel } 8115602294fSDan Kimmel if (compressok) { 8125602294fSDan Kimmel featureflags |= DMU_BACKUP_FEATURE_COMPRESSED; 8135602294fSDan Kimmel } 8145602294fSDan Kimmel if ((featureflags & 8155602294fSDan Kimmel (DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_COMPRESSED)) != 8165602294fSDan Kimmel 0 && spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) { 8175602294fSDan Kimmel featureflags |= DMU_BACKUP_FEATURE_LZ4; 8185d7b4d43SMatthew Ahrens } 8195d7b4d43SMatthew Ahrens 8209c3fd121SMatthew Ahrens if (resumeobj != 0 || resumeoff != 0) { 8219c3fd121SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_RESUMING; 8229c3fd121SMatthew Ahrens } 8239c3fd121SMatthew Ahrens 8245d7b4d43SMatthew Ahrens DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, 8255d7b4d43SMatthew Ahrens featureflags); 8265d7b4d43SMatthew Ahrens 827efb80947Sahrens drr->drr_u.drr_begin.drr_creation_time = 828a2cdcdd2SPaul Dagnelie dsl_dataset_phys(to_ds)->ds_creation_time; 8293b2aab18SMatthew Ahrens drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); 83078f17100SMatthew Ahrens if (is_clone) 8313cb34c60Sahrens drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; 832a2cdcdd2SPaul Dagnelie drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(to_ds)->ds_guid; 833a2cdcdd2SPaul Dagnelie if (dsl_dataset_phys(to_ds)->ds_flags & DS_FLAG_CI_DATASET) 834ab04eb8eStimh drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; 835880094b6SAndrew Stormont if (zfs_send_set_freerecords_bit) 836880094b6SAndrew Stormont drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_FREERECORDS; 837ab04eb8eStimh 838a2cdcdd2SPaul Dagnelie if (ancestor_zb != NULL) { 839a2cdcdd2SPaul Dagnelie drr->drr_u.drr_begin.drr_fromguid = 840a2cdcdd2SPaul Dagnelie ancestor_zb->zbm_guid; 841a2cdcdd2SPaul Dagnelie fromtxg = ancestor_zb->zbm_creation_txg; 84278f17100SMatthew Ahrens } 843a2cdcdd2SPaul Dagnelie dsl_dataset_name(to_ds, drr->drr_u.drr_begin.drr_toname); 844a2cdcdd2SPaul Dagnelie if (!to_ds->ds_is_snapshot) { 84578f17100SMatthew Ahrens (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", 84678f17100SMatthew Ahrens sizeof (drr->drr_u.drr_begin.drr_toname)); 8473b2aab18SMatthew Ahrens } 8483cb34c60Sahrens 8494e3c9f44SBill Pijewski dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); 8504e3c9f44SBill Pijewski 8514e3c9f44SBill Pijewski dsp->dsa_drr = drr; 8524e3c9f44SBill Pijewski dsp->dsa_vp = vp; 8534e3c9f44SBill Pijewski dsp->dsa_outfd = outfd; 8544e3c9f44SBill Pijewski dsp->dsa_proc = curproc; 8553b2aab18SMatthew Ahrens dsp->dsa_os = os; 8564e3c9f44SBill Pijewski dsp->dsa_off = off; 857a2cdcdd2SPaul Dagnelie dsp->dsa_toguid = dsl_dataset_phys(to_ds)->ds_guid; 8584e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 8595d7b4d43SMatthew Ahrens dsp->dsa_featureflags = featureflags; 8609c3fd121SMatthew Ahrens dsp->dsa_resume_object = resumeobj; 8619c3fd121SMatthew Ahrens dsp->dsa_resume_offset = resumeoff; 8624e3c9f44SBill Pijewski 863a2cdcdd2SPaul Dagnelie mutex_enter(&to_ds->ds_sendstream_lock); 864a2cdcdd2SPaul Dagnelie list_insert_head(&to_ds->ds_sendstreams, dsp); 865a2cdcdd2SPaul Dagnelie mutex_exit(&to_ds->ds_sendstream_lock); 8664e3c9f44SBill Pijewski 867a2cdcdd2SPaul Dagnelie dsl_dataset_long_hold(to_ds, FTAG); 868de8d9cffSMatthew Ahrens dsl_pool_rele(dp, tag); 869de8d9cffSMatthew Ahrens 8709c3fd121SMatthew Ahrens void *payload = NULL; 8719c3fd121SMatthew Ahrens size_t payload_len = 0; 8729c3fd121SMatthew Ahrens if (resumeobj != 0 || resumeoff != 0) { 8739c3fd121SMatthew Ahrens dmu_object_info_t to_doi; 8749c3fd121SMatthew Ahrens err = dmu_object_info(os, resumeobj, &to_doi); 8759c3fd121SMatthew Ahrens if (err != 0) 8769c3fd121SMatthew Ahrens goto out; 8779c3fd121SMatthew Ahrens SET_BOOKMARK(&to_arg.resume, to_ds->ds_object, resumeobj, 0, 8789c3fd121SMatthew Ahrens resumeoff / to_doi.doi_data_block_size); 8799c3fd121SMatthew Ahrens 8809c3fd121SMatthew Ahrens nvlist_t *nvl = fnvlist_alloc(); 8819c3fd121SMatthew Ahrens fnvlist_add_uint64(nvl, "resume_object", resumeobj); 8829c3fd121SMatthew Ahrens fnvlist_add_uint64(nvl, "resume_offset", resumeoff); 8839c3fd121SMatthew Ahrens payload = fnvlist_pack(nvl, &payload_len); 8849c3fd121SMatthew Ahrens drr->drr_payloadlen = payload_len; 8859c3fd121SMatthew Ahrens fnvlist_free(nvl); 8869c3fd121SMatthew Ahrens } 8879c3fd121SMatthew Ahrens 8889c3fd121SMatthew Ahrens err = dump_record(dsp, payload, payload_len); 8899c3fd121SMatthew Ahrens fnvlist_pack_free(payload, payload_len); 8909c3fd121SMatthew Ahrens if (err != 0) { 8914e3c9f44SBill Pijewski err = dsp->dsa_err; 8924e3c9f44SBill Pijewski goto out; 893efb80947Sahrens } 894efb80947Sahrens 895a2cdcdd2SPaul Dagnelie err = bqueue_init(&to_arg.q, zfs_send_queue_length, 896a2cdcdd2SPaul Dagnelie offsetof(struct send_block_record, ln)); 897a2cdcdd2SPaul Dagnelie to_arg.error_code = 0; 898a2cdcdd2SPaul Dagnelie to_arg.cancel = B_FALSE; 899a2cdcdd2SPaul Dagnelie to_arg.ds = to_ds; 900a2cdcdd2SPaul Dagnelie to_arg.fromtxg = fromtxg; 901a2cdcdd2SPaul Dagnelie to_arg.flags = TRAVERSE_PRE | TRAVERSE_PREFETCH; 902a2cdcdd2SPaul Dagnelie (void) thread_create(NULL, 0, send_traverse_thread, &to_arg, 0, curproc, 903a2cdcdd2SPaul Dagnelie TS_RUN, minclsyspri); 904a2cdcdd2SPaul Dagnelie 905a2cdcdd2SPaul Dagnelie struct send_block_record *to_data; 906a2cdcdd2SPaul Dagnelie to_data = bqueue_dequeue(&to_arg.q); 907a2cdcdd2SPaul Dagnelie 908a2cdcdd2SPaul Dagnelie while (!to_data->eos_marker && err == 0) { 909a2cdcdd2SPaul Dagnelie err = do_dump(dsp, to_data); 910a2cdcdd2SPaul Dagnelie to_data = get_next_record(&to_arg.q, to_data); 911a2cdcdd2SPaul Dagnelie if (issig(JUSTLOOKING) && issig(FORREAL)) 912a2cdcdd2SPaul Dagnelie err = EINTR; 913a2cdcdd2SPaul Dagnelie } 914a2cdcdd2SPaul Dagnelie 915a2cdcdd2SPaul Dagnelie if (err != 0) { 916a2cdcdd2SPaul Dagnelie to_arg.cancel = B_TRUE; 917a2cdcdd2SPaul Dagnelie while (!to_data->eos_marker) { 918a2cdcdd2SPaul Dagnelie to_data = get_next_record(&to_arg.q, to_data); 919a2cdcdd2SPaul Dagnelie } 920a2cdcdd2SPaul Dagnelie } 921a2cdcdd2SPaul Dagnelie kmem_free(to_data, sizeof (*to_data)); 922a2cdcdd2SPaul Dagnelie 923a2cdcdd2SPaul Dagnelie bqueue_destroy(&to_arg.q); 924a2cdcdd2SPaul Dagnelie 925a2cdcdd2SPaul Dagnelie if (err == 0 && to_arg.error_code != 0) 926a2cdcdd2SPaul Dagnelie err = to_arg.error_code; 927a2cdcdd2SPaul Dagnelie 928a2cdcdd2SPaul Dagnelie if (err != 0) 929a2cdcdd2SPaul Dagnelie goto out; 930efb80947Sahrens 9314e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) 93298110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 933be6fd75aSMatthew Ahrens err = SET_ERROR(EINTR); 9349e69d7d0SLori Alt 9353b2aab18SMatthew Ahrens if (err != 0) { 9363b2aab18SMatthew Ahrens if (err == EINTR && dsp->dsa_err != 0) 9374e3c9f44SBill Pijewski err = dsp->dsa_err; 9384e3c9f44SBill Pijewski goto out; 939efb80947Sahrens } 940efb80947Sahrens 941efb80947Sahrens bzero(drr, sizeof (dmu_replay_record_t)); 942efb80947Sahrens drr->drr_type = DRR_END; 9434e3c9f44SBill Pijewski drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; 9444e3c9f44SBill Pijewski drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; 945efb80947Sahrens 946a2cdcdd2SPaul Dagnelie if (dump_record(dsp, NULL, 0) != 0) 9474e3c9f44SBill Pijewski err = dsp->dsa_err; 948efb80947Sahrens 9494e3c9f44SBill Pijewski out: 950a2cdcdd2SPaul Dagnelie mutex_enter(&to_ds->ds_sendstream_lock); 951a2cdcdd2SPaul Dagnelie list_remove(&to_ds->ds_sendstreams, dsp); 952a2cdcdd2SPaul Dagnelie mutex_exit(&to_ds->ds_sendstream_lock); 9534e3c9f44SBill Pijewski 95412b90ee2SMatt Krantz VERIFY(err != 0 || (dsp->dsa_sent_begin && dsp->dsa_sent_end)); 95512b90ee2SMatt Krantz 956efb80947Sahrens kmem_free(drr, sizeof (dmu_replay_record_t)); 9574e3c9f44SBill Pijewski kmem_free(dsp, sizeof (dmu_sendarg_t)); 958efb80947Sahrens 959a2cdcdd2SPaul Dagnelie dsl_dataset_long_rele(to_ds, FTAG); 9603b2aab18SMatthew Ahrens 9614e3c9f44SBill Pijewski return (err); 962efb80947Sahrens } 963efb80947Sahrens 96419b94df9SMatthew Ahrens int 9653b2aab18SMatthew Ahrens dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, 9665602294fSDan Kimmel boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, 967b5152584SMatthew Ahrens int outfd, vnode_t *vp, offset_t *off) 9683b2aab18SMatthew Ahrens { 9693b2aab18SMatthew Ahrens dsl_pool_t *dp; 9703b2aab18SMatthew Ahrens dsl_dataset_t *ds; 9713b2aab18SMatthew Ahrens dsl_dataset_t *fromds = NULL; 9723b2aab18SMatthew Ahrens int err; 9733b2aab18SMatthew Ahrens 9743b2aab18SMatthew Ahrens err = dsl_pool_hold(pool, FTAG, &dp); 9753b2aab18SMatthew Ahrens if (err != 0) 9763b2aab18SMatthew Ahrens return (err); 9773b2aab18SMatthew Ahrens 9783b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); 9793b2aab18SMatthew Ahrens if (err != 0) { 9803b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 9813b2aab18SMatthew Ahrens return (err); 9823b2aab18SMatthew Ahrens } 9833b2aab18SMatthew Ahrens 9843b2aab18SMatthew Ahrens if (fromsnap != 0) { 98578f17100SMatthew Ahrens zfs_bookmark_phys_t zb; 98678f17100SMatthew Ahrens boolean_t is_clone; 98778f17100SMatthew Ahrens 9883b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); 9893b2aab18SMatthew Ahrens if (err != 0) { 9903b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 9913b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 9923b2aab18SMatthew Ahrens return (err); 9933b2aab18SMatthew Ahrens } 99478f17100SMatthew Ahrens if (!dsl_dataset_is_before(ds, fromds, 0)) 99578f17100SMatthew Ahrens err = SET_ERROR(EXDEV); 996c1379625SJustin T. Gibbs zb.zbm_creation_time = 997c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_time; 998c1379625SJustin T. Gibbs zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; 999c1379625SJustin T. Gibbs zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 100078f17100SMatthew Ahrens is_clone = (fromds->ds_dir != ds->ds_dir); 100178f17100SMatthew Ahrens dsl_dataset_rele(fromds, FTAG); 1002b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 10035602294fSDan Kimmel embedok, large_block_ok, compressok, outfd, 0, 0, vp, off); 100478f17100SMatthew Ahrens } else { 1005b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 10065602294fSDan Kimmel embedok, large_block_ok, compressok, outfd, 0, 0, vp, off); 10073b2aab18SMatthew Ahrens } 100878f17100SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 100978f17100SMatthew Ahrens return (err); 10103b2aab18SMatthew Ahrens } 10113b2aab18SMatthew Ahrens 10123b2aab18SMatthew Ahrens int 10139c3fd121SMatthew Ahrens dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, 10145602294fSDan Kimmel boolean_t large_block_ok, boolean_t compressok, int outfd, 10155602294fSDan Kimmel uint64_t resumeobj, uint64_t resumeoff, 10169c3fd121SMatthew Ahrens vnode_t *vp, offset_t *off) 10173b2aab18SMatthew Ahrens { 10183b2aab18SMatthew Ahrens dsl_pool_t *dp; 10193b2aab18SMatthew Ahrens dsl_dataset_t *ds; 10203b2aab18SMatthew Ahrens int err; 102178f17100SMatthew Ahrens boolean_t owned = B_FALSE; 10223b2aab18SMatthew Ahrens 102378f17100SMatthew Ahrens if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) 1024be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 10253b2aab18SMatthew Ahrens 10263b2aab18SMatthew Ahrens err = dsl_pool_hold(tosnap, FTAG, &dp); 10273b2aab18SMatthew Ahrens if (err != 0) 10283b2aab18SMatthew Ahrens return (err); 10293b2aab18SMatthew Ahrens 103078f17100SMatthew Ahrens if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { 103178f17100SMatthew Ahrens /* 103278f17100SMatthew Ahrens * We are sending a filesystem or volume. Ensure 103378f17100SMatthew Ahrens * that it doesn't change by owning the dataset. 103478f17100SMatthew Ahrens */ 103578f17100SMatthew Ahrens err = dsl_dataset_own(dp, tosnap, FTAG, &ds); 103678f17100SMatthew Ahrens owned = B_TRUE; 103778f17100SMatthew Ahrens } else { 103878f17100SMatthew Ahrens err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); 103978f17100SMatthew Ahrens } 10403b2aab18SMatthew Ahrens if (err != 0) { 10413b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 10423b2aab18SMatthew Ahrens return (err); 10433b2aab18SMatthew Ahrens } 10443b2aab18SMatthew Ahrens 10453b2aab18SMatthew Ahrens if (fromsnap != NULL) { 104678f17100SMatthew Ahrens zfs_bookmark_phys_t zb; 104778f17100SMatthew Ahrens boolean_t is_clone = B_FALSE; 104878f17100SMatthew Ahrens int fsnamelen = strchr(tosnap, '@') - tosnap; 104978f17100SMatthew Ahrens 105078f17100SMatthew Ahrens /* 105178f17100SMatthew Ahrens * If the fromsnap is in a different filesystem, then 105278f17100SMatthew Ahrens * mark the send stream as a clone. 105378f17100SMatthew Ahrens */ 105478f17100SMatthew Ahrens if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || 105578f17100SMatthew Ahrens (fromsnap[fsnamelen] != '@' && 105678f17100SMatthew Ahrens fromsnap[fsnamelen] != '#')) { 105778f17100SMatthew Ahrens is_clone = B_TRUE; 105878f17100SMatthew Ahrens } 105978f17100SMatthew Ahrens 106078f17100SMatthew Ahrens if (strchr(fromsnap, '@')) { 106178f17100SMatthew Ahrens dsl_dataset_t *fromds; 106278f17100SMatthew Ahrens err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); 106378f17100SMatthew Ahrens if (err == 0) { 106478f17100SMatthew Ahrens if (!dsl_dataset_is_before(ds, fromds, 0)) 106578f17100SMatthew Ahrens err = SET_ERROR(EXDEV); 106678f17100SMatthew Ahrens zb.zbm_creation_time = 1067c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_time; 106878f17100SMatthew Ahrens zb.zbm_creation_txg = 1069c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_txg; 1070c1379625SJustin T. Gibbs zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 107178f17100SMatthew Ahrens is_clone = (ds->ds_dir != fromds->ds_dir); 107278f17100SMatthew Ahrens dsl_dataset_rele(fromds, FTAG); 107378f17100SMatthew Ahrens } 107478f17100SMatthew Ahrens } else { 107578f17100SMatthew Ahrens err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); 107678f17100SMatthew Ahrens } 10773b2aab18SMatthew Ahrens if (err != 0) { 10783b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 10793b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 10803b2aab18SMatthew Ahrens return (err); 10813b2aab18SMatthew Ahrens } 1082b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 10835602294fSDan Kimmel embedok, large_block_ok, compressok, 10849c3fd121SMatthew Ahrens outfd, resumeobj, resumeoff, vp, off); 108578f17100SMatthew Ahrens } else { 1086b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 10875602294fSDan Kimmel embedok, large_block_ok, compressok, 10889c3fd121SMatthew Ahrens outfd, resumeobj, resumeoff, vp, off); 10893b2aab18SMatthew Ahrens } 109078f17100SMatthew Ahrens if (owned) 109178f17100SMatthew Ahrens dsl_dataset_disown(ds, FTAG); 109278f17100SMatthew Ahrens else 109378f17100SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 109478f17100SMatthew Ahrens return (err); 10953b2aab18SMatthew Ahrens } 10963b2aab18SMatthew Ahrens 1097643da460SMax Grossman static int 10985602294fSDan Kimmel dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t uncompressed, 10995602294fSDan Kimmel uint64_t compressed, boolean_t stream_compressed, uint64_t *sizep) 1100643da460SMax Grossman { 1101*df477c0aSPaul Dagnelie int err = 0; 11025602294fSDan Kimmel uint64_t size; 1103643da460SMax Grossman /* 1104643da460SMax Grossman * Assume that space (both on-disk and in-stream) is dominated by 1105643da460SMax Grossman * data. We will adjust for indirect blocks and the copies property, 1106643da460SMax Grossman * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). 1107643da460SMax Grossman */ 11085602294fSDan Kimmel uint64_t recordsize; 11095602294fSDan Kimmel uint64_t record_count; 11100255edccSPaul Dagnelie objset_t *os; 11110255edccSPaul Dagnelie VERIFY0(dmu_objset_from_ds(ds, &os)); 11125602294fSDan Kimmel 11135602294fSDan Kimmel /* Assume all (uncompressed) blocks are recordsize. */ 1114*df477c0aSPaul Dagnelie if (zfs_override_estimate_recordsize != 0) { 1115*df477c0aSPaul Dagnelie recordsize = zfs_override_estimate_recordsize; 1116*df477c0aSPaul Dagnelie } else if (os->os_phys->os_type == DMU_OST_ZVOL) { 11170255edccSPaul Dagnelie err = dsl_prop_get_int_ds(ds, 11180255edccSPaul Dagnelie zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &recordsize); 11190255edccSPaul Dagnelie } else { 11200255edccSPaul Dagnelie err = dsl_prop_get_int_ds(ds, 11210255edccSPaul Dagnelie zfs_prop_to_name(ZFS_PROP_RECORDSIZE), &recordsize); 11220255edccSPaul Dagnelie } 11235602294fSDan Kimmel if (err != 0) 11245602294fSDan Kimmel return (err); 11255602294fSDan Kimmel record_count = uncompressed / recordsize; 11265602294fSDan Kimmel 11275602294fSDan Kimmel /* 11285602294fSDan Kimmel * If we're estimating a send size for a compressed stream, use the 11295602294fSDan Kimmel * compressed data size to estimate the stream size. Otherwise, use the 11305602294fSDan Kimmel * uncompressed data size. 11315602294fSDan Kimmel */ 11325602294fSDan Kimmel size = stream_compressed ? compressed : uncompressed; 1133643da460SMax Grossman 1134643da460SMax Grossman /* 1135643da460SMax Grossman * Subtract out approximate space used by indirect blocks. 1136643da460SMax Grossman * Assume most space is used by data blocks (non-indirect, non-dnode). 11375602294fSDan Kimmel * Assume no ditto blocks or internal fragmentation. 1138643da460SMax Grossman * 1139643da460SMax Grossman * Therefore, space used by indirect blocks is sizeof(blkptr_t) per 11405602294fSDan Kimmel * block. 1141643da460SMax Grossman */ 11425602294fSDan Kimmel size -= record_count * sizeof (blkptr_t); 1143643da460SMax Grossman 1144643da460SMax Grossman /* Add in the space for the record associated with each block. */ 11455602294fSDan Kimmel size += record_count * sizeof (dmu_replay_record_t); 1146643da460SMax Grossman 1147643da460SMax Grossman *sizep = size; 1148643da460SMax Grossman 1149643da460SMax Grossman return (0); 1150643da460SMax Grossman } 1151643da460SMax Grossman 11523b2aab18SMatthew Ahrens int 11535602294fSDan Kimmel dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, 11545602294fSDan Kimmel boolean_t stream_compressed, uint64_t *sizep) 115519b94df9SMatthew Ahrens { 115619b94df9SMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 115719b94df9SMatthew Ahrens int err; 11585602294fSDan Kimmel uint64_t uncomp, comp; 115919b94df9SMatthew Ahrens 11603b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dp)); 11613b2aab18SMatthew Ahrens 116219b94df9SMatthew Ahrens /* tosnap must be a snapshot */ 1163bc9014e6SJustin Gibbs if (!ds->ds_is_snapshot) 1164be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 116519b94df9SMatthew Ahrens 116624218bebSAndriy Gapon /* fromsnap, if provided, must be a snapshot */ 116724218bebSAndriy Gapon if (fromds != NULL && !fromds->ds_is_snapshot) 116824218bebSAndriy Gapon return (SET_ERROR(EINVAL)); 116924218bebSAndriy Gapon 11704445fffbSMatthew Ahrens /* 11714445fffbSMatthew Ahrens * fromsnap must be an earlier snapshot from the same fs as tosnap, 11724445fffbSMatthew Ahrens * or the origin's fs. 11734445fffbSMatthew Ahrens */ 117478f17100SMatthew Ahrens if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) 1175be6fd75aSMatthew Ahrens return (SET_ERROR(EXDEV)); 117619b94df9SMatthew Ahrens 11775602294fSDan Kimmel /* Get compressed and uncompressed size estimates of changed data. */ 117819b94df9SMatthew Ahrens if (fromds == NULL) { 11795602294fSDan Kimmel uncomp = dsl_dataset_phys(ds)->ds_uncompressed_bytes; 11805602294fSDan Kimmel comp = dsl_dataset_phys(ds)->ds_compressed_bytes; 118119b94df9SMatthew Ahrens } else { 11825602294fSDan Kimmel uint64_t used; 118319b94df9SMatthew Ahrens err = dsl_dataset_space_written(fromds, ds, 11845602294fSDan Kimmel &used, &comp, &uncomp); 11853b2aab18SMatthew Ahrens if (err != 0) 118619b94df9SMatthew Ahrens return (err); 118719b94df9SMatthew Ahrens } 118819b94df9SMatthew Ahrens 11895602294fSDan Kimmel err = dmu_adjust_send_estimate_for_indirects(ds, uncomp, comp, 11905602294fSDan Kimmel stream_compressed, sizep); 11910255edccSPaul Dagnelie /* 11920255edccSPaul Dagnelie * Add the size of the BEGIN and END records to the estimate. 11930255edccSPaul Dagnelie */ 11940255edccSPaul Dagnelie *sizep += 2 * sizeof (dmu_replay_record_t); 1195643da460SMax Grossman return (err); 1196643da460SMax Grossman } 1197643da460SMax Grossman 11985602294fSDan Kimmel struct calculate_send_arg { 11995602294fSDan Kimmel uint64_t uncompressed; 12005602294fSDan Kimmel uint64_t compressed; 12015602294fSDan Kimmel }; 12025602294fSDan Kimmel 1203643da460SMax Grossman /* 1204643da460SMax Grossman * Simple callback used to traverse the blocks of a snapshot and sum their 12055602294fSDan Kimmel * uncompressed and compressed sizes. 1206643da460SMax Grossman */ 1207643da460SMax Grossman /* ARGSUSED */ 1208643da460SMax Grossman static int 1209643da460SMax Grossman dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1210643da460SMax Grossman const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1211643da460SMax Grossman { 12125602294fSDan Kimmel struct calculate_send_arg *space = arg; 1213643da460SMax Grossman if (bp != NULL && !BP_IS_HOLE(bp)) { 12145602294fSDan Kimmel space->uncompressed += BP_GET_UCSIZE(bp); 12155602294fSDan Kimmel space->compressed += BP_GET_PSIZE(bp); 1216643da460SMax Grossman } 1217643da460SMax Grossman return (0); 1218643da460SMax Grossman } 1219643da460SMax Grossman 1220643da460SMax Grossman /* 1221643da460SMax Grossman * Given a desination snapshot and a TXG, calculate the approximate size of a 1222643da460SMax Grossman * send stream sent from that TXG. from_txg may be zero, indicating that the 1223643da460SMax Grossman * whole snapshot will be sent. 1224643da460SMax Grossman */ 1225643da460SMax Grossman int 1226643da460SMax Grossman dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg, 12275602294fSDan Kimmel boolean_t stream_compressed, uint64_t *sizep) 1228643da460SMax Grossman { 1229643da460SMax Grossman dsl_pool_t *dp = ds->ds_dir->dd_pool; 1230643da460SMax Grossman int err; 12315602294fSDan Kimmel struct calculate_send_arg size = { 0 }; 1232643da460SMax Grossman 1233643da460SMax Grossman ASSERT(dsl_pool_config_held(dp)); 1234643da460SMax Grossman 1235643da460SMax Grossman /* tosnap must be a snapshot */ 12365602294fSDan Kimmel if (!ds->ds_is_snapshot) 1237643da460SMax Grossman return (SET_ERROR(EINVAL)); 1238643da460SMax Grossman 1239643da460SMax Grossman /* verify that from_txg is before the provided snapshot was taken */ 1240643da460SMax Grossman if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) { 1241643da460SMax Grossman return (SET_ERROR(EXDEV)); 1242643da460SMax Grossman } 124319b94df9SMatthew Ahrens 124419b94df9SMatthew Ahrens /* 1245643da460SMax Grossman * traverse the blocks of the snapshot with birth times after 1246643da460SMax Grossman * from_txg, summing their uncompressed size 124719b94df9SMatthew Ahrens */ 1248643da460SMax Grossman err = traverse_dataset(ds, from_txg, TRAVERSE_POST, 1249643da460SMax Grossman dmu_calculate_send_traversal, &size); 1250643da460SMax Grossman if (err) 125119b94df9SMatthew Ahrens return (err); 125219b94df9SMatthew Ahrens 12535602294fSDan Kimmel err = dmu_adjust_send_estimate_for_indirects(ds, size.uncompressed, 12545602294fSDan Kimmel size.compressed, stream_compressed, sizep); 1255643da460SMax Grossman return (err); 125619b94df9SMatthew Ahrens } 125719b94df9SMatthew Ahrens 12583b2aab18SMatthew Ahrens typedef struct dmu_recv_begin_arg { 12593b2aab18SMatthew Ahrens const char *drba_origin; 12603b2aab18SMatthew Ahrens dmu_recv_cookie_t *drba_cookie; 12613b2aab18SMatthew Ahrens cred_t *drba_cred; 126234f2f8cfSMatthew Ahrens uint64_t drba_snapobj; 12633b2aab18SMatthew Ahrens } dmu_recv_begin_arg_t; 1264f18faf3fSek 1265f18faf3fSek static int 12663b2aab18SMatthew Ahrens recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, 12673b2aab18SMatthew Ahrens uint64_t fromguid) 1268f18faf3fSek { 12693cb34c60Sahrens uint64_t val; 12703b2aab18SMatthew Ahrens int error; 12713b2aab18SMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 1272f18faf3fSek 12733b2aab18SMatthew Ahrens /* temporary clone name must not exist */ 12743b2aab18SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, 1275c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, 12763b2aab18SMatthew Ahrens 8, 1, &val); 12773b2aab18SMatthew Ahrens if (error != ENOENT) 12783b2aab18SMatthew Ahrens return (error == 0 ? EBUSY : error); 12793b2aab18SMatthew Ahrens 1280feaa74e4SMark Maybee /* new snapshot name must not exist */ 12813b2aab18SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, 1282c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1283c1379625SJustin T. Gibbs drba->drba_cookie->drc_tosnap, 8, 1, &val); 12843b2aab18SMatthew Ahrens if (error != ENOENT) 12853b2aab18SMatthew Ahrens return (error == 0 ? EEXIST : error); 1286feaa74e4SMark Maybee 1287a2afb611SJerry Jelinek /* 1288a2afb611SJerry Jelinek * Check snapshot limit before receiving. We'll recheck again at the 1289a2afb611SJerry Jelinek * end, but might as well abort before receiving if we're already over 1290a2afb611SJerry Jelinek * the limit. 1291a2afb611SJerry Jelinek * 1292a2afb611SJerry Jelinek * Note that we do not check the file system limit with 1293a2afb611SJerry Jelinek * dsl_dir_fscount_check because the temporary %clones don't count 1294a2afb611SJerry Jelinek * against that limit. 1295a2afb611SJerry Jelinek */ 1296a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, 1297a2afb611SJerry Jelinek NULL, drba->drba_cred); 1298a2afb611SJerry Jelinek if (error != 0) 1299a2afb611SJerry Jelinek return (error); 1300a2afb611SJerry Jelinek 13013b2aab18SMatthew Ahrens if (fromguid != 0) { 130234f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 1303c1379625SJustin T. Gibbs uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 130434f2f8cfSMatthew Ahrens 130534f2f8cfSMatthew Ahrens /* Find snapshot in this dir that matches fromguid. */ 130634f2f8cfSMatthew Ahrens while (obj != 0) { 130734f2f8cfSMatthew Ahrens error = dsl_dataset_hold_obj(dp, obj, FTAG, 130834f2f8cfSMatthew Ahrens &snap); 130934f2f8cfSMatthew Ahrens if (error != 0) 131034f2f8cfSMatthew Ahrens return (SET_ERROR(ENODEV)); 131134f2f8cfSMatthew Ahrens if (snap->ds_dir != ds->ds_dir) { 131234f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 131334f2f8cfSMatthew Ahrens return (SET_ERROR(ENODEV)); 131434f2f8cfSMatthew Ahrens } 1315c1379625SJustin T. Gibbs if (dsl_dataset_phys(snap)->ds_guid == fromguid) 131634f2f8cfSMatthew Ahrens break; 1317c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 131834f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 131934f2f8cfSMatthew Ahrens } 132034f2f8cfSMatthew Ahrens if (obj == 0) 1321be6fd75aSMatthew Ahrens return (SET_ERROR(ENODEV)); 132292241e0bSTom Erickson 132334f2f8cfSMatthew Ahrens if (drba->drba_cookie->drc_force) { 132434f2f8cfSMatthew Ahrens drba->drba_snapobj = obj; 132534f2f8cfSMatthew Ahrens } else { 132634f2f8cfSMatthew Ahrens /* 132734f2f8cfSMatthew Ahrens * If we are not forcing, there must be no 132834f2f8cfSMatthew Ahrens * changes since fromsnap. 132934f2f8cfSMatthew Ahrens */ 133034f2f8cfSMatthew Ahrens if (dsl_dataset_modified_since_snap(ds, snap)) { 133192241e0bSTom Erickson dsl_dataset_rele(snap, FTAG); 133234f2f8cfSMatthew Ahrens return (SET_ERROR(ETXTBSY)); 133392241e0bSTom Erickson } 133434f2f8cfSMatthew Ahrens drba->drba_snapobj = ds->ds_prev->ds_object; 133592241e0bSTom Erickson } 133634f2f8cfSMatthew Ahrens 133734f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 1338ae46e4c7SMatthew Ahrens } else { 13395bae108fSAndriy Gapon /* if full, then must be forced */ 13405bae108fSAndriy Gapon if (!drba->drba_cookie->drc_force) 13415bae108fSAndriy Gapon return (SET_ERROR(EEXIST)); 13425bae108fSAndriy Gapon /* start from $ORIGIN@$ORIGIN, if supported */ 13435bae108fSAndriy Gapon drba->drba_snapobj = dp->dp_origin_snap != NULL ? 13445bae108fSAndriy Gapon dp->dp_origin_snap->ds_object : 0; 1345ae46e4c7SMatthew Ahrens } 13463cb34c60Sahrens 13473cb34c60Sahrens return (0); 13483b2aab18SMatthew Ahrens 13493b2aab18SMatthew Ahrens } 13503b2aab18SMatthew Ahrens 13513b2aab18SMatthew Ahrens static int 13523b2aab18SMatthew Ahrens dmu_recv_begin_check(void *arg, dmu_tx_t *tx) 13533b2aab18SMatthew Ahrens { 13543b2aab18SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 13553b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 13563b2aab18SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 13573b2aab18SMatthew Ahrens uint64_t fromguid = drrb->drr_fromguid; 13583b2aab18SMatthew Ahrens int flags = drrb->drr_flags; 13593b2aab18SMatthew Ahrens int error; 13605d7b4d43SMatthew Ahrens uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 13613b2aab18SMatthew Ahrens dsl_dataset_t *ds; 13623b2aab18SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 13633b2aab18SMatthew Ahrens 13643b2aab18SMatthew Ahrens /* already checked */ 13653b2aab18SMatthew Ahrens ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 13669c3fd121SMatthew Ahrens ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); 13673b2aab18SMatthew Ahrens 13683b2aab18SMatthew Ahrens if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 13693b2aab18SMatthew Ahrens DMU_COMPOUNDSTREAM || 13703b2aab18SMatthew Ahrens drrb->drr_type >= DMU_OST_NUMTYPES || 13713b2aab18SMatthew Ahrens ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) 1372be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 13733b2aab18SMatthew Ahrens 13743b2aab18SMatthew Ahrens /* Verify pool version supports SA if SA_SPILL feature set */ 13755d7b4d43SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 13765d7b4d43SMatthew Ahrens spa_version(dp->dp_spa) < SPA_VERSION_SA) 13775d7b4d43SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 13785d7b4d43SMatthew Ahrens 13799c3fd121SMatthew Ahrens if (drba->drba_cookie->drc_resumable && 13809c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) 13819c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 13829c3fd121SMatthew Ahrens 13835d7b4d43SMatthew Ahrens /* 13845d7b4d43SMatthew Ahrens * The receiving code doesn't know how to translate a WRITE_EMBEDDED 13855602294fSDan Kimmel * record to a plain WRITE record, so the pool must have the 13865d7b4d43SMatthew Ahrens * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 13875d7b4d43SMatthew Ahrens * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 13885d7b4d43SMatthew Ahrens */ 13895d7b4d43SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 13905d7b4d43SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 13915d7b4d43SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 13925602294fSDan Kimmel if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && 13935d7b4d43SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 1394be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 13953b2aab18SMatthew Ahrens 1396b5152584SMatthew Ahrens /* 1397b5152584SMatthew Ahrens * The receiving code doesn't know how to translate large blocks 1398b5152584SMatthew Ahrens * to smaller ones, so the pool must have the LARGE_BLOCKS 1399b5152584SMatthew Ahrens * feature enabled if the stream has LARGE_BLOCKS. 1400b5152584SMatthew Ahrens */ 1401b5152584SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 1402b5152584SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 1403b5152584SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 1404b5152584SMatthew Ahrens 14053b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 14063b2aab18SMatthew Ahrens if (error == 0) { 14073b2aab18SMatthew Ahrens /* target fs already exists; recv into temp clone */ 14083b2aab18SMatthew Ahrens 14093b2aab18SMatthew Ahrens /* Can't recv a clone into an existing fs */ 141068ecb2ecSPaul Dagnelie if (flags & DRR_FLAG_CLONE || drba->drba_origin) { 14113b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1412be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 14133b2aab18SMatthew Ahrens } 14143b2aab18SMatthew Ahrens 14153b2aab18SMatthew Ahrens error = recv_begin_check_existing_impl(drba, ds, fromguid); 14163b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14173b2aab18SMatthew Ahrens } else if (error == ENOENT) { 14183b2aab18SMatthew Ahrens /* target fs does not exist; must be a full backup or clone */ 14199adfa60dSMatthew Ahrens char buf[ZFS_MAX_DATASET_NAME_LEN]; 14203b2aab18SMatthew Ahrens 14213b2aab18SMatthew Ahrens /* 14223b2aab18SMatthew Ahrens * If it's a non-clone incremental, we are missing the 14233b2aab18SMatthew Ahrens * target fs, so fail the recv. 14243b2aab18SMatthew Ahrens */ 1425a2cdcdd2SPaul Dagnelie if (fromguid != 0 && !(flags & DRR_FLAG_CLONE || 1426a2cdcdd2SPaul Dagnelie drba->drba_origin)) 1427be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 14283b2aab18SMatthew Ahrens 142968ecb2ecSPaul Dagnelie /* 143068ecb2ecSPaul Dagnelie * If we're receiving a full send as a clone, and it doesn't 143168ecb2ecSPaul Dagnelie * contain all the necessary free records and freeobject 143268ecb2ecSPaul Dagnelie * records, reject it. 143368ecb2ecSPaul Dagnelie */ 143468ecb2ecSPaul Dagnelie if (fromguid == 0 && drba->drba_origin && 143568ecb2ecSPaul Dagnelie !(flags & DRR_FLAG_FREERECORDS)) 143668ecb2ecSPaul Dagnelie return (SET_ERROR(EINVAL)); 143768ecb2ecSPaul Dagnelie 14383b2aab18SMatthew Ahrens /* Open the parent of tofs */ 14399adfa60dSMatthew Ahrens ASSERT3U(strlen(tofs), <, sizeof (buf)); 14403b2aab18SMatthew Ahrens (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); 14413b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, buf, FTAG, &ds); 14423b2aab18SMatthew Ahrens if (error != 0) 14433b2aab18SMatthew Ahrens return (error); 14443b2aab18SMatthew Ahrens 1445a2afb611SJerry Jelinek /* 1446a2afb611SJerry Jelinek * Check filesystem and snapshot limits before receiving. We'll 1447a2afb611SJerry Jelinek * recheck snapshot limits again at the end (we create the 1448a2afb611SJerry Jelinek * filesystems and increment those counts during begin_sync). 1449a2afb611SJerry Jelinek */ 1450a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 1451a2afb611SJerry Jelinek ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); 1452a2afb611SJerry Jelinek if (error != 0) { 1453a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 1454a2afb611SJerry Jelinek return (error); 1455a2afb611SJerry Jelinek } 1456a2afb611SJerry Jelinek 1457a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 1458a2afb611SJerry Jelinek ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); 1459a2afb611SJerry Jelinek if (error != 0) { 1460a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 1461a2afb611SJerry Jelinek return (error); 1462a2afb611SJerry Jelinek } 1463a2afb611SJerry Jelinek 14643b2aab18SMatthew Ahrens if (drba->drba_origin != NULL) { 14653b2aab18SMatthew Ahrens dsl_dataset_t *origin; 14663b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, drba->drba_origin, 14673b2aab18SMatthew Ahrens FTAG, &origin); 14683b2aab18SMatthew Ahrens if (error != 0) { 14693b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14703b2aab18SMatthew Ahrens return (error); 14713b2aab18SMatthew Ahrens } 1472bc9014e6SJustin Gibbs if (!origin->ds_is_snapshot) { 14733b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 14743b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1475be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 14763b2aab18SMatthew Ahrens } 147768ecb2ecSPaul Dagnelie if (dsl_dataset_phys(origin)->ds_guid != fromguid && 147868ecb2ecSPaul Dagnelie fromguid != 0) { 14793b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 14803b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1481be6fd75aSMatthew Ahrens return (SET_ERROR(ENODEV)); 14823b2aab18SMatthew Ahrens } 14833b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 14843b2aab18SMatthew Ahrens } 14853b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14863b2aab18SMatthew Ahrens error = 0; 14873b2aab18SMatthew Ahrens } 14883b2aab18SMatthew Ahrens return (error); 1489f18faf3fSek } 1490f18faf3fSek 1491f18faf3fSek static void 14923b2aab18SMatthew Ahrens dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) 1493f18faf3fSek { 14943b2aab18SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 14953b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 14969c3fd121SMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 14973b2aab18SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 14983b2aab18SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 14993b2aab18SMatthew Ahrens dsl_dataset_t *ds, *newds; 1500f18faf3fSek uint64_t dsobj; 15013b2aab18SMatthew Ahrens int error; 15029c3fd121SMatthew Ahrens uint64_t crflags = 0; 15033b2aab18SMatthew Ahrens 15049c3fd121SMatthew Ahrens if (drrb->drr_flags & DRR_FLAG_CI_DATA) 15059c3fd121SMatthew Ahrens crflags |= DS_FLAG_CI_DATASET; 1506f18faf3fSek 15073b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 15083b2aab18SMatthew Ahrens if (error == 0) { 15093b2aab18SMatthew Ahrens /* create temporary clone */ 151034f2f8cfSMatthew Ahrens dsl_dataset_t *snap = NULL; 151134f2f8cfSMatthew Ahrens if (drba->drba_snapobj != 0) { 151234f2f8cfSMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 151334f2f8cfSMatthew Ahrens drba->drba_snapobj, FTAG, &snap)); 151434f2f8cfSMatthew Ahrens } 15153b2aab18SMatthew Ahrens dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, 151634f2f8cfSMatthew Ahrens snap, crflags, drba->drba_cred, tx); 1517f40b29ceSPaul Dagnelie if (drba->drba_snapobj != 0) 1518f40b29ceSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 15193b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15203b2aab18SMatthew Ahrens } else { 15213b2aab18SMatthew Ahrens dsl_dir_t *dd; 15223b2aab18SMatthew Ahrens const char *tail; 15233b2aab18SMatthew Ahrens dsl_dataset_t *origin = NULL; 15243b2aab18SMatthew Ahrens 15253b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); 15263b2aab18SMatthew Ahrens 15273b2aab18SMatthew Ahrens if (drba->drba_origin != NULL) { 15283b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, 15293b2aab18SMatthew Ahrens FTAG, &origin)); 15303b2aab18SMatthew Ahrens } 15313b2aab18SMatthew Ahrens 15323b2aab18SMatthew Ahrens /* Create new dataset. */ 15333b2aab18SMatthew Ahrens dsobj = dsl_dataset_create_sync(dd, 15343b2aab18SMatthew Ahrens strrchr(tofs, '/') + 1, 15353b2aab18SMatthew Ahrens origin, crflags, drba->drba_cred, tx); 15363b2aab18SMatthew Ahrens if (origin != NULL) 15373b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 15383b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 15393b2aab18SMatthew Ahrens drba->drba_cookie->drc_newfs = B_TRUE; 15403b2aab18SMatthew Ahrens } 15413b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); 15423b2aab18SMatthew Ahrens 15439c3fd121SMatthew Ahrens if (drba->drba_cookie->drc_resumable) { 15449c3fd121SMatthew Ahrens dsl_dataset_zapify(newds, tx); 15459c3fd121SMatthew Ahrens if (drrb->drr_fromguid != 0) { 15469c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, 15479c3fd121SMatthew Ahrens 8, 1, &drrb->drr_fromguid, tx)); 15489c3fd121SMatthew Ahrens } 15499c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, 15509c3fd121SMatthew Ahrens 8, 1, &drrb->drr_toguid, tx)); 15519c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, 15529c3fd121SMatthew Ahrens 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); 15539c3fd121SMatthew Ahrens uint64_t one = 1; 15549c3fd121SMatthew Ahrens uint64_t zero = 0; 15559c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, 15569c3fd121SMatthew Ahrens 8, 1, &one, tx)); 15579c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, 15589c3fd121SMatthew Ahrens 8, 1, &zero, tx)); 15599c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, 15609c3fd121SMatthew Ahrens 8, 1, &zero, tx)); 15615602294fSDan Kimmel if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 15625602294fSDan Kimmel DMU_BACKUP_FEATURE_LARGE_BLOCKS) { 15635602294fSDan Kimmel VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_LARGEBLOCK, 15645602294fSDan Kimmel 8, 1, &one, tx)); 15655602294fSDan Kimmel } 15669c3fd121SMatthew Ahrens if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 15679c3fd121SMatthew Ahrens DMU_BACKUP_FEATURE_EMBED_DATA) { 15689c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, 15699c3fd121SMatthew Ahrens 8, 1, &one, tx)); 15709c3fd121SMatthew Ahrens } 15715602294fSDan Kimmel if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 15725602294fSDan Kimmel DMU_BACKUP_FEATURE_COMPRESSED) { 15735602294fSDan Kimmel VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_COMPRESSOK, 15745602294fSDan Kimmel 8, 1, &one, tx)); 15755602294fSDan Kimmel } 15769c3fd121SMatthew Ahrens } 15779c3fd121SMatthew Ahrens 15783b2aab18SMatthew Ahrens dmu_buf_will_dirty(newds->ds_dbuf, tx); 1579c1379625SJustin T. Gibbs dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; 15803cb34c60Sahrens 1581ae46e4c7SMatthew Ahrens /* 1582ae46e4c7SMatthew Ahrens * If we actually created a non-clone, we need to create the 1583ae46e4c7SMatthew Ahrens * objset in our new dataset. 1584ae46e4c7SMatthew Ahrens */ 1585c166b69dSPaul Dagnelie rrw_enter(&newds->ds_bp_rwlock, RW_READER, FTAG); 15863b2aab18SMatthew Ahrens if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { 1587ae46e4c7SMatthew Ahrens (void) dmu_objset_create_impl(dp->dp_spa, 15883b2aab18SMatthew Ahrens newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); 1589ae46e4c7SMatthew Ahrens } 1590c166b69dSPaul Dagnelie rrw_exit(&newds->ds_bp_rwlock, FTAG); 1591ae46e4c7SMatthew Ahrens 15923b2aab18SMatthew Ahrens drba->drba_cookie->drc_ds = newds; 1593dc7cd546SMark Shellenbaum 15943b2aab18SMatthew Ahrens spa_history_log_internal_ds(newds, "receive", tx, ""); 1595dc7cd546SMark Shellenbaum } 1596dc7cd546SMark Shellenbaum 15979c3fd121SMatthew Ahrens static int 15989c3fd121SMatthew Ahrens dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) 15999c3fd121SMatthew Ahrens { 16009c3fd121SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 16019c3fd121SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 16029c3fd121SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 16039c3fd121SMatthew Ahrens int error; 16049c3fd121SMatthew Ahrens uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 16059c3fd121SMatthew Ahrens dsl_dataset_t *ds; 16069c3fd121SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 16079c3fd121SMatthew Ahrens 16089c3fd121SMatthew Ahrens /* already checked */ 16099c3fd121SMatthew Ahrens ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 16109c3fd121SMatthew Ahrens ASSERT(featureflags & DMU_BACKUP_FEATURE_RESUMING); 16119c3fd121SMatthew Ahrens 16129c3fd121SMatthew Ahrens if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 16139c3fd121SMatthew Ahrens DMU_COMPOUNDSTREAM || 16149c3fd121SMatthew Ahrens drrb->drr_type >= DMU_OST_NUMTYPES) 16159c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16169c3fd121SMatthew Ahrens 16179c3fd121SMatthew Ahrens /* Verify pool version supports SA if SA_SPILL feature set */ 16189c3fd121SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 16199c3fd121SMatthew Ahrens spa_version(dp->dp_spa) < SPA_VERSION_SA) 16209c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 16219c3fd121SMatthew Ahrens 16229c3fd121SMatthew Ahrens /* 16239c3fd121SMatthew Ahrens * The receiving code doesn't know how to translate a WRITE_EMBEDDED 16249c3fd121SMatthew Ahrens * record to a plain WRITE record, so the pool must have the 16259c3fd121SMatthew Ahrens * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 16269c3fd121SMatthew Ahrens * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 16279c3fd121SMatthew Ahrens */ 16289c3fd121SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 16299c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 16309c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 16315602294fSDan Kimmel if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && 16329c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 16339c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 16349c3fd121SMatthew Ahrens 16359adfa60dSMatthew Ahrens /* 6 extra bytes for /%recv */ 16369adfa60dSMatthew Ahrens char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 16379c3fd121SMatthew Ahrens 16389c3fd121SMatthew Ahrens (void) snprintf(recvname, sizeof (recvname), "%s/%s", 16399c3fd121SMatthew Ahrens tofs, recv_clone_name); 16409c3fd121SMatthew Ahrens 16419c3fd121SMatthew Ahrens if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { 16429c3fd121SMatthew Ahrens /* %recv does not exist; continue in tofs */ 16439c3fd121SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 16449c3fd121SMatthew Ahrens if (error != 0) 16459c3fd121SMatthew Ahrens return (error); 16469c3fd121SMatthew Ahrens } 16479c3fd121SMatthew Ahrens 16489c3fd121SMatthew Ahrens /* check that ds is marked inconsistent */ 16499c3fd121SMatthew Ahrens if (!DS_IS_INCONSISTENT(ds)) { 16509c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16519c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16529c3fd121SMatthew Ahrens } 16539c3fd121SMatthew Ahrens 16549c3fd121SMatthew Ahrens /* check that there is resuming data, and that the toguid matches */ 16559c3fd121SMatthew Ahrens if (!dsl_dataset_is_zapified(ds)) { 16569c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16579c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16589c3fd121SMatthew Ahrens } 16599c3fd121SMatthew Ahrens uint64_t val; 16609c3fd121SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, ds->ds_object, 16619c3fd121SMatthew Ahrens DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); 16629c3fd121SMatthew Ahrens if (error != 0 || drrb->drr_toguid != val) { 16639c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16649c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16659c3fd121SMatthew Ahrens } 16669c3fd121SMatthew Ahrens 16679c3fd121SMatthew Ahrens /* 16689c3fd121SMatthew Ahrens * Check if the receive is still running. If so, it will be owned. 16699c3fd121SMatthew Ahrens * Note that nothing else can own the dataset (e.g. after the receive 16709c3fd121SMatthew Ahrens * fails) because it will be marked inconsistent. 16719c3fd121SMatthew Ahrens */ 16729c3fd121SMatthew Ahrens if (dsl_dataset_has_owner(ds)) { 16739c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16749c3fd121SMatthew Ahrens return (SET_ERROR(EBUSY)); 16759c3fd121SMatthew Ahrens } 16769c3fd121SMatthew Ahrens 16779c3fd121SMatthew Ahrens /* There should not be any snapshots of this fs yet. */ 16789c3fd121SMatthew Ahrens if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { 16799c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16809c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16819c3fd121SMatthew Ahrens } 16829c3fd121SMatthew Ahrens 16839c3fd121SMatthew Ahrens /* 16849c3fd121SMatthew Ahrens * Note: resume point will be checked when we process the first WRITE 16859c3fd121SMatthew Ahrens * record. 16869c3fd121SMatthew Ahrens */ 16879c3fd121SMatthew Ahrens 16889c3fd121SMatthew Ahrens /* check that the origin matches */ 16899c3fd121SMatthew Ahrens val = 0; 16909c3fd121SMatthew Ahrens (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, 16919c3fd121SMatthew Ahrens DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); 16929c3fd121SMatthew Ahrens if (drrb->drr_fromguid != val) { 16939c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16949c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16959c3fd121SMatthew Ahrens } 16969c3fd121SMatthew Ahrens 16979c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16989c3fd121SMatthew Ahrens return (0); 16999c3fd121SMatthew Ahrens } 17009c3fd121SMatthew Ahrens 17019c3fd121SMatthew Ahrens static void 17029c3fd121SMatthew Ahrens dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) 17039c3fd121SMatthew Ahrens { 17049c3fd121SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 17059c3fd121SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 17069c3fd121SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 17079c3fd121SMatthew Ahrens dsl_dataset_t *ds; 17089c3fd121SMatthew Ahrens uint64_t dsobj; 17099adfa60dSMatthew Ahrens /* 6 extra bytes for /%recv */ 17109adfa60dSMatthew Ahrens char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 17119c3fd121SMatthew Ahrens 17129c3fd121SMatthew Ahrens (void) snprintf(recvname, sizeof (recvname), "%s/%s", 17139c3fd121SMatthew Ahrens tofs, recv_clone_name); 17149c3fd121SMatthew Ahrens 17159c3fd121SMatthew Ahrens if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { 17169c3fd121SMatthew Ahrens /* %recv does not exist; continue in tofs */ 17179c3fd121SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, tofs, FTAG, &ds)); 17189c3fd121SMatthew Ahrens drba->drba_cookie->drc_newfs = B_TRUE; 17199c3fd121SMatthew Ahrens } 17209c3fd121SMatthew Ahrens 17219c3fd121SMatthew Ahrens /* clear the inconsistent flag so that we can own it */ 17229c3fd121SMatthew Ahrens ASSERT(DS_IS_INCONSISTENT(ds)); 17239c3fd121SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 17249c3fd121SMatthew Ahrens dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 17259c3fd121SMatthew Ahrens dsobj = ds->ds_object; 17269c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 17279c3fd121SMatthew Ahrens 17289c3fd121SMatthew Ahrens VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &ds)); 17299c3fd121SMatthew Ahrens 17309c3fd121SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 17319c3fd121SMatthew Ahrens dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT; 17329c3fd121SMatthew Ahrens 1733c166b69dSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 17349c3fd121SMatthew Ahrens ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds))); 1735c166b69dSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 17369c3fd121SMatthew Ahrens 17379c3fd121SMatthew Ahrens drba->drba_cookie->drc_ds = ds; 17389c3fd121SMatthew Ahrens 17399c3fd121SMatthew Ahrens spa_history_log_internal_ds(ds, "resume receive", tx, ""); 17409c3fd121SMatthew Ahrens } 17419c3fd121SMatthew Ahrens 17423cb34c60Sahrens /* 17433cb34c60Sahrens * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() 17443cb34c60Sahrens * succeeds; otherwise we will leak the holds on the datasets. 17453cb34c60Sahrens */ 17463cb34c60Sahrens int 17479c3fd121SMatthew Ahrens dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, 17489c3fd121SMatthew Ahrens boolean_t force, boolean_t resumable, char *origin, dmu_recv_cookie_t *drc) 1749efb80947Sahrens { 17503b2aab18SMatthew Ahrens dmu_recv_begin_arg_t drba = { 0 }; 1751ab04eb8eStimh 17523cb34c60Sahrens bzero(drc, sizeof (dmu_recv_cookie_t)); 17539c3fd121SMatthew Ahrens drc->drc_drr_begin = drr_begin; 17549c3fd121SMatthew Ahrens drc->drc_drrb = &drr_begin->drr_u.drr_begin; 17553cb34c60Sahrens drc->drc_tosnap = tosnap; 17563b2aab18SMatthew Ahrens drc->drc_tofs = tofs; 17573cb34c60Sahrens drc->drc_force = force; 17589c3fd121SMatthew Ahrens drc->drc_resumable = resumable; 1759a2afb611SJerry Jelinek drc->drc_cred = CRED(); 1760efb80947Sahrens 17619c3fd121SMatthew Ahrens if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { 17623b2aab18SMatthew Ahrens drc->drc_byteswap = B_TRUE; 1763770499e1SDan Kimmel (void) fletcher_4_incremental_byteswap(drr_begin, 17643b2aab18SMatthew Ahrens sizeof (dmu_replay_record_t), &drc->drc_cksum); 17659c3fd121SMatthew Ahrens byteswap_record(drr_begin); 17669c3fd121SMatthew Ahrens } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { 1767770499e1SDan Kimmel (void) fletcher_4_incremental_native(drr_begin, 17683b2aab18SMatthew Ahrens sizeof (dmu_replay_record_t), &drc->drc_cksum); 17699c3fd121SMatthew Ahrens } else { 17709c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 17713cb34c60Sahrens } 17723cb34c60Sahrens 17733b2aab18SMatthew Ahrens drba.drba_origin = origin; 17743b2aab18SMatthew Ahrens drba.drba_cookie = drc; 17753b2aab18SMatthew Ahrens drba.drba_cred = CRED(); 17763b2aab18SMatthew Ahrens 17779c3fd121SMatthew Ahrens if (DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & 17789c3fd121SMatthew Ahrens DMU_BACKUP_FEATURE_RESUMING) { 17799c3fd121SMatthew Ahrens return (dsl_sync_task(tofs, 17809c3fd121SMatthew Ahrens dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, 17819c3fd121SMatthew Ahrens &drba, 5, ZFS_SPACE_CHECK_NORMAL)); 17829c3fd121SMatthew Ahrens } else { 17839c3fd121SMatthew Ahrens return (dsl_sync_task(tofs, 17849c3fd121SMatthew Ahrens dmu_recv_begin_check, dmu_recv_begin_sync, 17859c3fd121SMatthew Ahrens &drba, 5, ZFS_SPACE_CHECK_NORMAL)); 17869c3fd121SMatthew Ahrens } 1787efb80947Sahrens } 1788efb80947Sahrens 1789a2cdcdd2SPaul Dagnelie struct receive_record_arg { 1790a2cdcdd2SPaul Dagnelie dmu_replay_record_t header; 1791a2cdcdd2SPaul Dagnelie void *payload; /* Pointer to a buffer containing the payload */ 1792a2cdcdd2SPaul Dagnelie /* 1793a2cdcdd2SPaul Dagnelie * If the record is a write, pointer to the arc_buf_t containing the 1794a2cdcdd2SPaul Dagnelie * payload. 1795a2cdcdd2SPaul Dagnelie */ 1796a2cdcdd2SPaul Dagnelie arc_buf_t *write_buf; 1797a2cdcdd2SPaul Dagnelie int payload_size; 17989c3fd121SMatthew Ahrens uint64_t bytes_read; /* bytes read from stream when record created */ 1799a2cdcdd2SPaul Dagnelie boolean_t eos_marker; /* Marks the end of the stream */ 1800a2cdcdd2SPaul Dagnelie bqueue_node_t node; 1801a2cdcdd2SPaul Dagnelie }; 1802a2cdcdd2SPaul Dagnelie 1803a2cdcdd2SPaul Dagnelie struct receive_writer_arg { 180498110f08SMatthew Ahrens objset_t *os; 18053b2aab18SMatthew Ahrens boolean_t byteswap; 1806a2cdcdd2SPaul Dagnelie bqueue_t q; 18079c3fd121SMatthew Ahrens 1808a2cdcdd2SPaul Dagnelie /* 1809a2cdcdd2SPaul Dagnelie * These three args are used to signal to the main thread that we're 1810a2cdcdd2SPaul Dagnelie * done. 1811a2cdcdd2SPaul Dagnelie */ 1812a2cdcdd2SPaul Dagnelie kmutex_t mutex; 1813a2cdcdd2SPaul Dagnelie kcondvar_t cv; 1814a2cdcdd2SPaul Dagnelie boolean_t done; 18159c3fd121SMatthew Ahrens 1816a2cdcdd2SPaul Dagnelie int err; 1817a2cdcdd2SPaul Dagnelie /* A map from guid to dataset to help handle dedup'd streams. */ 1818a2cdcdd2SPaul Dagnelie avl_tree_t *guid_to_ds_map; 18199c3fd121SMatthew Ahrens boolean_t resumable; 18209c3fd121SMatthew Ahrens uint64_t last_object, last_offset; 18219c3fd121SMatthew Ahrens uint64_t bytes_read; /* bytes read when current record created */ 1822a2cdcdd2SPaul Dagnelie }; 182398110f08SMatthew Ahrens 182468ecb2ecSPaul Dagnelie struct objlist { 182568ecb2ecSPaul Dagnelie list_t list; /* List of struct receive_objnode. */ 182668ecb2ecSPaul Dagnelie /* 182768ecb2ecSPaul Dagnelie * Last object looked up. Used to assert that objects are being looked 182868ecb2ecSPaul Dagnelie * up in ascending order. 182968ecb2ecSPaul Dagnelie */ 183068ecb2ecSPaul Dagnelie uint64_t last_lookup; 183168ecb2ecSPaul Dagnelie }; 183268ecb2ecSPaul Dagnelie 183368ecb2ecSPaul Dagnelie struct receive_objnode { 183468ecb2ecSPaul Dagnelie list_node_t node; 183568ecb2ecSPaul Dagnelie uint64_t object; 183668ecb2ecSPaul Dagnelie }; 183768ecb2ecSPaul Dagnelie 18385602294fSDan Kimmel struct receive_arg { 1839a2cdcdd2SPaul Dagnelie objset_t *os; 1840a2cdcdd2SPaul Dagnelie vnode_t *vp; /* The vnode to read the stream from */ 1841a2cdcdd2SPaul Dagnelie uint64_t voff; /* The current offset in the stream */ 18429c3fd121SMatthew Ahrens uint64_t bytes_read; 1843a2cdcdd2SPaul Dagnelie /* 1844a2cdcdd2SPaul Dagnelie * A record that has had its payload read in, but hasn't yet been handed 1845a2cdcdd2SPaul Dagnelie * off to the worker thread. 1846a2cdcdd2SPaul Dagnelie */ 1847a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd; 1848a2cdcdd2SPaul Dagnelie /* A record that has had its header read in, but not its payload. */ 1849a2cdcdd2SPaul Dagnelie struct receive_record_arg *next_rrd; 18503cb34c60Sahrens zio_cksum_t cksum; 185198110f08SMatthew Ahrens zio_cksum_t prev_cksum; 1852a2cdcdd2SPaul Dagnelie int err; 1853a2cdcdd2SPaul Dagnelie boolean_t byteswap; 1854a2cdcdd2SPaul Dagnelie /* Sorted list of objects not to issue prefetches for. */ 185568ecb2ecSPaul Dagnelie struct objlist ignore_objlist; 18563cb34c60Sahrens }; 18573cb34c60Sahrens 18589e69d7d0SLori Alt typedef struct guid_map_entry { 18599e69d7d0SLori Alt uint64_t guid; 18609e69d7d0SLori Alt dsl_dataset_t *gme_ds; 18619e69d7d0SLori Alt avl_node_t avlnode; 18629e69d7d0SLori Alt } guid_map_entry_t; 18639e69d7d0SLori Alt 18649e69d7d0SLori Alt static int 18659e69d7d0SLori Alt guid_compare(const void *arg1, const void *arg2) 18669e69d7d0SLori Alt { 18679e69d7d0SLori Alt const guid_map_entry_t *gmep1 = arg1; 18689e69d7d0SLori Alt const guid_map_entry_t *gmep2 = arg2; 18699e69d7d0SLori Alt 18709e69d7d0SLori Alt if (gmep1->guid < gmep2->guid) 18719e69d7d0SLori Alt return (-1); 18729e69d7d0SLori Alt else if (gmep1->guid > gmep2->guid) 18739e69d7d0SLori Alt return (1); 18749e69d7d0SLori Alt return (0); 18759e69d7d0SLori Alt } 18769e69d7d0SLori Alt 1877c99e4bdcSChris Kirby static void 1878c99e4bdcSChris Kirby free_guid_map_onexit(void *arg) 1879c99e4bdcSChris Kirby { 1880c99e4bdcSChris Kirby avl_tree_t *ca = arg; 1881c99e4bdcSChris Kirby void *cookie = NULL; 1882c99e4bdcSChris Kirby guid_map_entry_t *gmep; 1883c99e4bdcSChris Kirby 1884c99e4bdcSChris Kirby while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { 18853b2aab18SMatthew Ahrens dsl_dataset_long_rele(gmep->gme_ds, gmep); 1886de8d9cffSMatthew Ahrens dsl_dataset_rele(gmep->gme_ds, gmep); 1887c99e4bdcSChris Kirby kmem_free(gmep, sizeof (guid_map_entry_t)); 1888c99e4bdcSChris Kirby } 1889c99e4bdcSChris Kirby avl_destroy(ca); 1890c99e4bdcSChris Kirby kmem_free(ca, sizeof (avl_tree_t)); 1891c99e4bdcSChris Kirby } 1892c99e4bdcSChris Kirby 189398110f08SMatthew Ahrens static int 1894a2cdcdd2SPaul Dagnelie receive_read(struct receive_arg *ra, int len, void *buf) 1895efb80947Sahrens { 18963cb34c60Sahrens int done = 0; 1897efb80947Sahrens 1898c20404ffSEli Rosenthal /* 1899c20404ffSEli Rosenthal * The code doesn't rely on this (lengths being multiples of 8). See 1900c20404ffSEli Rosenthal * comment in dump_bytes. 1901c20404ffSEli Rosenthal */ 1902fb09f5aaSMadhav Suresh ASSERT0(len % 8); 1903efb80947Sahrens 19043cb34c60Sahrens while (done < len) { 1905efb80947Sahrens ssize_t resid; 1906efb80947Sahrens 1907efb80947Sahrens ra->err = vn_rdwr(UIO_READ, ra->vp, 190898110f08SMatthew Ahrens (char *)buf + done, len - done, 1909efb80947Sahrens ra->voff, UIO_SYSSPACE, FAPPEND, 1910efb80947Sahrens RLIM64_INFINITY, CRED(), &resid); 1911efb80947Sahrens 19129c3fd121SMatthew Ahrens if (resid == len - done) { 19139c3fd121SMatthew Ahrens /* 19149c3fd121SMatthew Ahrens * Note: ECKSUM indicates that the receive 19159c3fd121SMatthew Ahrens * was interrupted and can potentially be resumed. 19169c3fd121SMatthew Ahrens */ 19179c3fd121SMatthew Ahrens ra->err = SET_ERROR(ECKSUM); 19189c3fd121SMatthew Ahrens } 19193cb34c60Sahrens ra->voff += len - done - resid; 19203cb34c60Sahrens done = len - resid; 19213b2aab18SMatthew Ahrens if (ra->err != 0) 192298110f08SMatthew Ahrens return (ra->err); 1923efb80947Sahrens } 1924efb80947Sahrens 19259c3fd121SMatthew Ahrens ra->bytes_read += len; 19269c3fd121SMatthew Ahrens 19273cb34c60Sahrens ASSERT3U(done, ==, len); 192898110f08SMatthew Ahrens return (0); 1929efb80947Sahrens } 1930efb80947Sahrens 1931efb80947Sahrens static void 193298110f08SMatthew Ahrens byteswap_record(dmu_replay_record_t *drr) 1933efb80947Sahrens { 1934efb80947Sahrens #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) 1935efb80947Sahrens #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) 1936efb80947Sahrens drr->drr_type = BSWAP_32(drr->drr_type); 19373cb34c60Sahrens drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); 193898110f08SMatthew Ahrens 1939efb80947Sahrens switch (drr->drr_type) { 1940efb80947Sahrens case DRR_BEGIN: 1941efb80947Sahrens DO64(drr_begin.drr_magic); 19429e69d7d0SLori Alt DO64(drr_begin.drr_versioninfo); 1943efb80947Sahrens DO64(drr_begin.drr_creation_time); 1944efb80947Sahrens DO32(drr_begin.drr_type); 19453cb34c60Sahrens DO32(drr_begin.drr_flags); 1946efb80947Sahrens DO64(drr_begin.drr_toguid); 1947efb80947Sahrens DO64(drr_begin.drr_fromguid); 1948efb80947Sahrens break; 1949efb80947Sahrens case DRR_OBJECT: 1950efb80947Sahrens DO64(drr_object.drr_object); 1951efb80947Sahrens DO32(drr_object.drr_type); 1952efb80947Sahrens DO32(drr_object.drr_bonustype); 1953efb80947Sahrens DO32(drr_object.drr_blksz); 1954efb80947Sahrens DO32(drr_object.drr_bonuslen); 19559e69d7d0SLori Alt DO64(drr_object.drr_toguid); 1956efb80947Sahrens break; 1957efb80947Sahrens case DRR_FREEOBJECTS: 1958efb80947Sahrens DO64(drr_freeobjects.drr_firstobj); 1959efb80947Sahrens DO64(drr_freeobjects.drr_numobjs); 19609e69d7d0SLori Alt DO64(drr_freeobjects.drr_toguid); 1961efb80947Sahrens break; 1962efb80947Sahrens case DRR_WRITE: 1963efb80947Sahrens DO64(drr_write.drr_object); 1964efb80947Sahrens DO32(drr_write.drr_type); 1965efb80947Sahrens DO64(drr_write.drr_offset); 19665602294fSDan Kimmel DO64(drr_write.drr_logical_size); 19679e69d7d0SLori Alt DO64(drr_write.drr_toguid); 196898110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); 19698e714474SLori Alt DO64(drr_write.drr_key.ddk_prop); 19705602294fSDan Kimmel DO64(drr_write.drr_compressed_size); 19719e69d7d0SLori Alt break; 19729e69d7d0SLori Alt case DRR_WRITE_BYREF: 19739e69d7d0SLori Alt DO64(drr_write_byref.drr_object); 19749e69d7d0SLori Alt DO64(drr_write_byref.drr_offset); 19759e69d7d0SLori Alt DO64(drr_write_byref.drr_length); 19769e69d7d0SLori Alt DO64(drr_write_byref.drr_toguid); 19779e69d7d0SLori Alt DO64(drr_write_byref.drr_refguid); 19789e69d7d0SLori Alt DO64(drr_write_byref.drr_refobject); 19799e69d7d0SLori Alt DO64(drr_write_byref.drr_refoffset); 198098110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write_byref. 198198110f08SMatthew Ahrens drr_key.ddk_cksum); 19828e714474SLori Alt DO64(drr_write_byref.drr_key.ddk_prop); 1983efb80947Sahrens break; 19845d7b4d43SMatthew Ahrens case DRR_WRITE_EMBEDDED: 19855d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_object); 19865d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_offset); 19875d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_length); 19885d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_toguid); 19895d7b4d43SMatthew Ahrens DO32(drr_write_embedded.drr_lsize); 19905d7b4d43SMatthew Ahrens DO32(drr_write_embedded.drr_psize); 19915d7b4d43SMatthew Ahrens break; 1992efb80947Sahrens case DRR_FREE: 1993efb80947Sahrens DO64(drr_free.drr_object); 1994efb80947Sahrens DO64(drr_free.drr_offset); 1995efb80947Sahrens DO64(drr_free.drr_length); 19969e69d7d0SLori Alt DO64(drr_free.drr_toguid); 1997efb80947Sahrens break; 19980a586ceaSMark Shellenbaum case DRR_SPILL: 19990a586ceaSMark Shellenbaum DO64(drr_spill.drr_object); 20000a586ceaSMark Shellenbaum DO64(drr_spill.drr_length); 20010a586ceaSMark Shellenbaum DO64(drr_spill.drr_toguid); 20020a586ceaSMark Shellenbaum break; 2003efb80947Sahrens case DRR_END: 20049e69d7d0SLori Alt DO64(drr_end.drr_toguid); 200598110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); 2006efb80947Sahrens break; 2007efb80947Sahrens } 200898110f08SMatthew Ahrens 200998110f08SMatthew Ahrens if (drr->drr_type != DRR_BEGIN) { 201098110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); 201198110f08SMatthew Ahrens } 201298110f08SMatthew Ahrens 2013efb80947Sahrens #undef DO64 2014efb80947Sahrens #undef DO32 2015efb80947Sahrens } 2016efb80947Sahrens 2017e77d42eaSMatthew Ahrens static inline uint8_t 2018e77d42eaSMatthew Ahrens deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) 2019e77d42eaSMatthew Ahrens { 2020e77d42eaSMatthew Ahrens if (bonus_type == DMU_OT_SA) { 2021e77d42eaSMatthew Ahrens return (1); 2022e77d42eaSMatthew Ahrens } else { 2023e77d42eaSMatthew Ahrens return (1 + 2024e77d42eaSMatthew Ahrens ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT)); 2025e77d42eaSMatthew Ahrens } 2026e77d42eaSMatthew Ahrens } 2027e77d42eaSMatthew Ahrens 20289c3fd121SMatthew Ahrens static void 20299c3fd121SMatthew Ahrens save_resume_state(struct receive_writer_arg *rwa, 20309c3fd121SMatthew Ahrens uint64_t object, uint64_t offset, dmu_tx_t *tx) 20319c3fd121SMatthew Ahrens { 20329c3fd121SMatthew Ahrens int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; 20339c3fd121SMatthew Ahrens 20349c3fd121SMatthew Ahrens if (!rwa->resumable) 20359c3fd121SMatthew Ahrens return; 20369c3fd121SMatthew Ahrens 20379c3fd121SMatthew Ahrens /* 20389c3fd121SMatthew Ahrens * We use ds_resume_bytes[] != 0 to indicate that we need to 20399c3fd121SMatthew Ahrens * update this on disk, so it must not be 0. 20409c3fd121SMatthew Ahrens */ 20419c3fd121SMatthew Ahrens ASSERT(rwa->bytes_read != 0); 20429c3fd121SMatthew Ahrens 20439c3fd121SMatthew Ahrens /* 20449c3fd121SMatthew Ahrens * We only resume from write records, which have a valid 20459c3fd121SMatthew Ahrens * (non-meta-dnode) object number. 20469c3fd121SMatthew Ahrens */ 20479c3fd121SMatthew Ahrens ASSERT(object != 0); 20489c3fd121SMatthew Ahrens 20499c3fd121SMatthew Ahrens /* 20509c3fd121SMatthew Ahrens * For resuming to work correctly, we must receive records in order, 20519c3fd121SMatthew Ahrens * sorted by object,offset. This is checked by the callers, but 20529c3fd121SMatthew Ahrens * assert it here for good measure. 20539c3fd121SMatthew Ahrens */ 20549c3fd121SMatthew Ahrens ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); 20559c3fd121SMatthew Ahrens ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || 20569c3fd121SMatthew Ahrens offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); 20579c3fd121SMatthew Ahrens ASSERT3U(rwa->bytes_read, >=, 20589c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); 20599c3fd121SMatthew Ahrens 20609c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; 20619c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; 20629c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; 20639c3fd121SMatthew Ahrens } 20649c3fd121SMatthew Ahrens 2065efb80947Sahrens static int 2066a2cdcdd2SPaul Dagnelie receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, 2067a2cdcdd2SPaul Dagnelie void *data) 2068efb80947Sahrens { 2069e77d42eaSMatthew Ahrens dmu_object_info_t doi; 2070efb80947Sahrens dmu_tx_t *tx; 2071e77d42eaSMatthew Ahrens uint64_t object; 2072e77d42eaSMatthew Ahrens int err; 2073efb80947Sahrens 2074efb80947Sahrens if (drro->drr_type == DMU_OT_NONE || 2075ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drro->drr_type) || 2076ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drro->drr_bonustype) || 20779e69d7d0SLori Alt drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || 2078efb80947Sahrens drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || 2079efb80947Sahrens P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || 2080efb80947Sahrens drro->drr_blksz < SPA_MINBLOCKSIZE || 2081a2cdcdd2SPaul Dagnelie drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || 2082efb80947Sahrens drro->drr_bonuslen > DN_MAX_BONUSLEN) { 2083be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2084efb80947Sahrens } 2085efb80947Sahrens 2086a2cdcdd2SPaul Dagnelie err = dmu_object_info(rwa->os, drro->drr_object, &doi); 20872bf405a2SMark Maybee 20882bf405a2SMark Maybee if (err != 0 && err != ENOENT) 2089be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2090e77d42eaSMatthew Ahrens object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT; 20912bf405a2SMark Maybee 2092e77d42eaSMatthew Ahrens /* 2093e77d42eaSMatthew Ahrens * If we are losing blkptrs or changing the block size this must 2094e77d42eaSMatthew Ahrens * be a new file instance. We must clear out the previous file 2095e77d42eaSMatthew Ahrens * contents before we can change this type of metadata in the dnode. 2096e77d42eaSMatthew Ahrens */ 2097e77d42eaSMatthew Ahrens if (err == 0) { 2098e77d42eaSMatthew Ahrens int nblkptr; 2099e77d42eaSMatthew Ahrens 2100e77d42eaSMatthew Ahrens nblkptr = deduce_nblkptr(drro->drr_bonustype, 2101e77d42eaSMatthew Ahrens drro->drr_bonuslen); 2102e77d42eaSMatthew Ahrens 2103e77d42eaSMatthew Ahrens if (drro->drr_blksz != doi.doi_data_block_size || 2104e77d42eaSMatthew Ahrens nblkptr < doi.doi_nblkptr) { 2105a2cdcdd2SPaul Dagnelie err = dmu_free_long_range(rwa->os, drro->drr_object, 2106e77d42eaSMatthew Ahrens 0, DMU_OBJECT_END); 2107e77d42eaSMatthew Ahrens if (err != 0) 2108e77d42eaSMatthew Ahrens return (SET_ERROR(EINVAL)); 2109efb80947Sahrens } 2110e77d42eaSMatthew Ahrens } 2111e77d42eaSMatthew Ahrens 2112a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 2113e77d42eaSMatthew Ahrens dmu_tx_hold_bonus(tx, object); 2114e77d42eaSMatthew Ahrens err = dmu_tx_assign(tx, TXG_WAIT); 2115e77d42eaSMatthew Ahrens if (err != 0) { 2116e77d42eaSMatthew Ahrens dmu_tx_abort(tx); 2117e77d42eaSMatthew Ahrens return (err); 2118e77d42eaSMatthew Ahrens } 2119e77d42eaSMatthew Ahrens 2120e77d42eaSMatthew Ahrens if (object == DMU_NEW_OBJECT) { 2121e77d42eaSMatthew Ahrens /* currently free, want to be allocated */ 2122a2cdcdd2SPaul Dagnelie err = dmu_object_claim(rwa->os, drro->drr_object, 2123efb80947Sahrens drro->drr_type, drro->drr_blksz, 2124efb80947Sahrens drro->drr_bonustype, drro->drr_bonuslen, tx); 2125e77d42eaSMatthew Ahrens } else if (drro->drr_type != doi.doi_type || 2126e77d42eaSMatthew Ahrens drro->drr_blksz != doi.doi_data_block_size || 2127e77d42eaSMatthew Ahrens drro->drr_bonustype != doi.doi_bonus_type || 2128e77d42eaSMatthew Ahrens drro->drr_bonuslen != doi.doi_bonus_size) { 2129e77d42eaSMatthew Ahrens /* currently allocated, but with different properties */ 2130a2cdcdd2SPaul Dagnelie err = dmu_object_reclaim(rwa->os, drro->drr_object, 2131efb80947Sahrens drro->drr_type, drro->drr_blksz, 2132e77d42eaSMatthew Ahrens drro->drr_bonustype, drro->drr_bonuslen, tx); 2133efb80947Sahrens } 21343b2aab18SMatthew Ahrens if (err != 0) { 2135e77d42eaSMatthew Ahrens dmu_tx_commit(tx); 2136be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 21370a586ceaSMark Shellenbaum } 21382bf405a2SMark Maybee 2139a2cdcdd2SPaul Dagnelie dmu_object_set_checksum(rwa->os, drro->drr_object, 214098110f08SMatthew Ahrens drro->drr_checksumtype, tx); 2141a2cdcdd2SPaul Dagnelie dmu_object_set_compress(rwa->os, drro->drr_object, 214298110f08SMatthew Ahrens drro->drr_compress, tx); 2143efb80947Sahrens 2144adee0b6fSTim Haley if (data != NULL) { 2145efb80947Sahrens dmu_buf_t *db; 2146adee0b6fSTim Haley 2147a2cdcdd2SPaul Dagnelie VERIFY0(dmu_bonus_hold(rwa->os, drro->drr_object, FTAG, &db)); 2148efb80947Sahrens dmu_buf_will_dirty(db, tx); 2149efb80947Sahrens 21501934e92fSmaybee ASSERT3U(db->db_size, >=, drro->drr_bonuslen); 21511934e92fSmaybee bcopy(data, db->db_data, drro->drr_bonuslen); 2152a2cdcdd2SPaul Dagnelie if (rwa->byteswap) { 2153ad135b5dSChristopher Siden dmu_object_byteswap_t byteswap = 2154ad135b5dSChristopher Siden DMU_OT_BYTESWAP(drro->drr_bonustype); 2155ad135b5dSChristopher Siden dmu_ot_byteswap[byteswap].ob_func(db->db_data, 2156efb80947Sahrens drro->drr_bonuslen); 2157efb80947Sahrens } 2158efb80947Sahrens dmu_buf_rele(db, FTAG); 2159efb80947Sahrens } 2160efb80947Sahrens dmu_tx_commit(tx); 21619c3fd121SMatthew Ahrens 2162efb80947Sahrens return (0); 2163efb80947Sahrens } 2164efb80947Sahrens 2165efb80947Sahrens /* ARGSUSED */ 2166efb80947Sahrens static int 2167a2cdcdd2SPaul Dagnelie receive_freeobjects(struct receive_writer_arg *rwa, 2168efb80947Sahrens struct drr_freeobjects *drrfo) 2169efb80947Sahrens { 2170efb80947Sahrens uint64_t obj; 217168ecb2ecSPaul Dagnelie int next_err = 0; 2172efb80947Sahrens 2173efb80947Sahrens if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) 2174be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2175efb80947Sahrens 2176efb80947Sahrens for (obj = drrfo->drr_firstobj; 217768ecb2ecSPaul Dagnelie obj < drrfo->drr_firstobj + drrfo->drr_numobjs && next_err == 0; 217868ecb2ecSPaul Dagnelie next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { 2179efb80947Sahrens int err; 2180efb80947Sahrens 2181a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, obj, NULL) != 0) 2182efb80947Sahrens continue; 2183efb80947Sahrens 2184a2cdcdd2SPaul Dagnelie err = dmu_free_long_object(rwa->os, obj); 21853b2aab18SMatthew Ahrens if (err != 0) 2186efb80947Sahrens return (err); 2187efb80947Sahrens } 218868ecb2ecSPaul Dagnelie if (next_err != ESRCH) 218968ecb2ecSPaul Dagnelie return (next_err); 2190efb80947Sahrens return (0); 2191efb80947Sahrens } 2192efb80947Sahrens 2193efb80947Sahrens static int 2194a2cdcdd2SPaul Dagnelie receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw, 2195a2cdcdd2SPaul Dagnelie arc_buf_t *abuf) 2196efb80947Sahrens { 2197efb80947Sahrens dmu_tx_t *tx; 2198efb80947Sahrens int err; 2199efb80947Sahrens 22005602294fSDan Kimmel if (drrw->drr_offset + drrw->drr_logical_size < drrw->drr_offset || 2201ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drrw->drr_type)) 2202be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2203efb80947Sahrens 22049c3fd121SMatthew Ahrens /* 22059c3fd121SMatthew Ahrens * For resuming to work, records must be in increasing order 22069c3fd121SMatthew Ahrens * by (object, offset). 22079c3fd121SMatthew Ahrens */ 22089c3fd121SMatthew Ahrens if (drrw->drr_object < rwa->last_object || 22099c3fd121SMatthew Ahrens (drrw->drr_object == rwa->last_object && 22109c3fd121SMatthew Ahrens drrw->drr_offset < rwa->last_offset)) { 22119c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 22129c3fd121SMatthew Ahrens } 22139c3fd121SMatthew Ahrens rwa->last_object = drrw->drr_object; 22149c3fd121SMatthew Ahrens rwa->last_offset = drrw->drr_offset; 22159c3fd121SMatthew Ahrens 2216a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrw->drr_object, NULL) != 0) 22178a904709SMatthew Ahrens return (SET_ERROR(EINVAL)); 22188a904709SMatthew Ahrens 2219a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 2220efb80947Sahrens 2221efb80947Sahrens dmu_tx_hold_write(tx, drrw->drr_object, 22225602294fSDan Kimmel drrw->drr_offset, drrw->drr_logical_size); 2223efb80947Sahrens err = dmu_tx_assign(tx, TXG_WAIT); 22243b2aab18SMatthew Ahrens if (err != 0) { 2225efb80947Sahrens dmu_tx_abort(tx); 2226efb80947Sahrens return (err); 2227efb80947Sahrens } 2228a2cdcdd2SPaul Dagnelie if (rwa->byteswap) { 2229ad135b5dSChristopher Siden dmu_object_byteswap_t byteswap = 2230ad135b5dSChristopher Siden DMU_OT_BYTESWAP(drrw->drr_type); 223198110f08SMatthew Ahrens dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, 22325602294fSDan Kimmel DRR_WRITE_PAYLOAD_SIZE(drrw)); 2233ad135b5dSChristopher Siden } 223498110f08SMatthew Ahrens 22355602294fSDan Kimmel /* use the bonus buf to look up the dnode in dmu_assign_arcbuf */ 223698110f08SMatthew Ahrens dmu_buf_t *bonus; 2237a2cdcdd2SPaul Dagnelie if (dmu_bonus_hold(rwa->os, drrw->drr_object, FTAG, &bonus) != 0) 223898110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 22398a904709SMatthew Ahrens dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx); 22409c3fd121SMatthew Ahrens 22419c3fd121SMatthew Ahrens /* 22429c3fd121SMatthew Ahrens * Note: If the receive fails, we want the resume stream to start 22439c3fd121SMatthew Ahrens * with the same record that we last successfully received (as opposed 22449c3fd121SMatthew Ahrens * to the next record), so that we can verify that we are 22459c3fd121SMatthew Ahrens * resuming from the correct location. 22469c3fd121SMatthew Ahrens */ 22479c3fd121SMatthew Ahrens save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); 2248efb80947Sahrens dmu_tx_commit(tx); 22498a904709SMatthew Ahrens dmu_buf_rele(bonus, FTAG); 22509c3fd121SMatthew Ahrens 2251efb80947Sahrens return (0); 2252efb80947Sahrens } 2253efb80947Sahrens 22549e69d7d0SLori Alt /* 22559e69d7d0SLori Alt * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed 22569e69d7d0SLori Alt * streams to refer to a copy of the data that is already on the 22579e69d7d0SLori Alt * system because it came in earlier in the stream. This function 22589e69d7d0SLori Alt * finds the earlier copy of the data, and uses that copy instead of 22599e69d7d0SLori Alt * data from the stream to fulfill this write. 22609e69d7d0SLori Alt */ 22619e69d7d0SLori Alt static int 2262a2cdcdd2SPaul Dagnelie receive_write_byref(struct receive_writer_arg *rwa, 2263a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwbr) 22649e69d7d0SLori Alt { 22659e69d7d0SLori Alt dmu_tx_t *tx; 22669e69d7d0SLori Alt int err; 22679e69d7d0SLori Alt guid_map_entry_t gmesrch; 22689e69d7d0SLori Alt guid_map_entry_t *gmep; 22695d7b4d43SMatthew Ahrens avl_index_t where; 22709e69d7d0SLori Alt objset_t *ref_os = NULL; 22719e69d7d0SLori Alt dmu_buf_t *dbp; 22729e69d7d0SLori Alt 22739e69d7d0SLori Alt if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) 2274be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 22759e69d7d0SLori Alt 22769e69d7d0SLori Alt /* 22779e69d7d0SLori Alt * If the GUID of the referenced dataset is different from the 22789e69d7d0SLori Alt * GUID of the target dataset, find the referenced dataset. 22799e69d7d0SLori Alt */ 22809e69d7d0SLori Alt if (drrwbr->drr_toguid != drrwbr->drr_refguid) { 22819e69d7d0SLori Alt gmesrch.guid = drrwbr->drr_refguid; 2282a2cdcdd2SPaul Dagnelie if ((gmep = avl_find(rwa->guid_to_ds_map, &gmesrch, 22839e69d7d0SLori Alt &where)) == NULL) { 2284be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 22859e69d7d0SLori Alt } 22869e69d7d0SLori Alt if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) 2287be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 22889e69d7d0SLori Alt } else { 2289a2cdcdd2SPaul Dagnelie ref_os = rwa->os; 22909e69d7d0SLori Alt } 22919e69d7d0SLori Alt 22925d7b4d43SMatthew Ahrens err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, 22935d7b4d43SMatthew Ahrens drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH); 22945d7b4d43SMatthew Ahrens if (err != 0) 22959e69d7d0SLori Alt return (err); 22969e69d7d0SLori Alt 2297a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 22989e69d7d0SLori Alt 22999e69d7d0SLori Alt dmu_tx_hold_write(tx, drrwbr->drr_object, 23009e69d7d0SLori Alt drrwbr->drr_offset, drrwbr->drr_length); 23019e69d7d0SLori Alt err = dmu_tx_assign(tx, TXG_WAIT); 23023b2aab18SMatthew Ahrens if (err != 0) { 23039e69d7d0SLori Alt dmu_tx_abort(tx); 23049e69d7d0SLori Alt return (err); 23059e69d7d0SLori Alt } 2306a2cdcdd2SPaul Dagnelie dmu_write(rwa->os, drrwbr->drr_object, 23079e69d7d0SLori Alt drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); 23089e69d7d0SLori Alt dmu_buf_rele(dbp, FTAG); 23099c3fd121SMatthew Ahrens 23109c3fd121SMatthew Ahrens /* See comment in restore_write. */ 23119c3fd121SMatthew Ahrens save_resume_state(rwa, drrwbr->drr_object, drrwbr->drr_offset, tx); 23129e69d7d0SLori Alt dmu_tx_commit(tx); 23139e69d7d0SLori Alt return (0); 23149e69d7d0SLori Alt } 23159e69d7d0SLori Alt 23165d7b4d43SMatthew Ahrens static int 2317a2cdcdd2SPaul Dagnelie receive_write_embedded(struct receive_writer_arg *rwa, 23189c3fd121SMatthew Ahrens struct drr_write_embedded *drrwe, void *data) 23195d7b4d43SMatthew Ahrens { 23205d7b4d43SMatthew Ahrens dmu_tx_t *tx; 23215d7b4d43SMatthew Ahrens int err; 23225d7b4d43SMatthew Ahrens 23239c3fd121SMatthew Ahrens if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) 23245d7b4d43SMatthew Ahrens return (EINVAL); 23255d7b4d43SMatthew Ahrens 23269c3fd121SMatthew Ahrens if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) 23275d7b4d43SMatthew Ahrens return (EINVAL); 23285d7b4d43SMatthew Ahrens 23299c3fd121SMatthew Ahrens if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) 23305d7b4d43SMatthew Ahrens return (EINVAL); 23319c3fd121SMatthew Ahrens if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) 23325d7b4d43SMatthew Ahrens return (EINVAL); 23335d7b4d43SMatthew Ahrens 2334a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 23355d7b4d43SMatthew Ahrens 23369c3fd121SMatthew Ahrens dmu_tx_hold_write(tx, drrwe->drr_object, 23379c3fd121SMatthew Ahrens drrwe->drr_offset, drrwe->drr_length); 23385d7b4d43SMatthew Ahrens err = dmu_tx_assign(tx, TXG_WAIT); 23395d7b4d43SMatthew Ahrens if (err != 0) { 23405d7b4d43SMatthew Ahrens dmu_tx_abort(tx); 23415d7b4d43SMatthew Ahrens return (err); 23425d7b4d43SMatthew Ahrens } 23435d7b4d43SMatthew Ahrens 23449c3fd121SMatthew Ahrens dmu_write_embedded(rwa->os, drrwe->drr_object, 23459c3fd121SMatthew Ahrens drrwe->drr_offset, data, drrwe->drr_etype, 23469c3fd121SMatthew Ahrens drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, 2347a2cdcdd2SPaul Dagnelie rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); 23485d7b4d43SMatthew Ahrens 23499c3fd121SMatthew Ahrens /* See comment in restore_write. */ 23509c3fd121SMatthew Ahrens save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); 23515d7b4d43SMatthew Ahrens dmu_tx_commit(tx); 23525d7b4d43SMatthew Ahrens return (0); 23535d7b4d43SMatthew Ahrens } 23545d7b4d43SMatthew Ahrens 23550a586ceaSMark Shellenbaum static int 2356a2cdcdd2SPaul Dagnelie receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, 2357a2cdcdd2SPaul Dagnelie void *data) 23580a586ceaSMark Shellenbaum { 23590a586ceaSMark Shellenbaum dmu_tx_t *tx; 23600a586ceaSMark Shellenbaum dmu_buf_t *db, *db_spill; 23610a586ceaSMark Shellenbaum int err; 23620a586ceaSMark Shellenbaum 23630a586ceaSMark Shellenbaum if (drrs->drr_length < SPA_MINBLOCKSIZE || 2364a2cdcdd2SPaul Dagnelie drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) 2365be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 23660a586ceaSMark Shellenbaum 2367a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) 2368be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 23690a586ceaSMark Shellenbaum 2370a2cdcdd2SPaul Dagnelie VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); 23710a586ceaSMark Shellenbaum if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) { 23720a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 23730a586ceaSMark Shellenbaum return (err); 23740a586ceaSMark Shellenbaum } 23750a586ceaSMark Shellenbaum 2376a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 23770a586ceaSMark Shellenbaum 23780a586ceaSMark Shellenbaum dmu_tx_hold_spill(tx, db->db_object); 23790a586ceaSMark Shellenbaum 23800a586ceaSMark Shellenbaum err = dmu_tx_assign(tx, TXG_WAIT); 23813b2aab18SMatthew Ahrens if (err != 0) { 23820a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 23830a586ceaSMark Shellenbaum dmu_buf_rele(db_spill, FTAG); 23840a586ceaSMark Shellenbaum dmu_tx_abort(tx); 23850a586ceaSMark Shellenbaum return (err); 23860a586ceaSMark Shellenbaum } 23870a586ceaSMark Shellenbaum dmu_buf_will_dirty(db_spill, tx); 23880a586ceaSMark Shellenbaum 23890a586ceaSMark Shellenbaum if (db_spill->db_size < drrs->drr_length) 23900a586ceaSMark Shellenbaum VERIFY(0 == dbuf_spill_set_blksz(db_spill, 23910a586ceaSMark Shellenbaum drrs->drr_length, tx)); 23920a586ceaSMark Shellenbaum bcopy(data, db_spill->db_data, drrs->drr_length); 23930a586ceaSMark Shellenbaum 23940a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 23950a586ceaSMark Shellenbaum dmu_buf_rele(db_spill, FTAG); 23960a586ceaSMark Shellenbaum 23970a586ceaSMark Shellenbaum dmu_tx_commit(tx); 23980a586ceaSMark Shellenbaum return (0); 23990a586ceaSMark Shellenbaum } 24000a586ceaSMark Shellenbaum 2401efb80947Sahrens /* ARGSUSED */ 2402efb80947Sahrens static int 2403a2cdcdd2SPaul Dagnelie receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) 2404efb80947Sahrens { 2405efb80947Sahrens int err; 2406efb80947Sahrens 2407efb80947Sahrens if (drrf->drr_length != -1ULL && 2408efb80947Sahrens drrf->drr_offset + drrf->drr_length < drrf->drr_offset) 2409be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2410efb80947Sahrens 2411a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) 2412be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2413efb80947Sahrens 2414a2cdcdd2SPaul Dagnelie err = dmu_free_long_range(rwa->os, drrf->drr_object, 2415efb80947Sahrens drrf->drr_offset, drrf->drr_length); 2416a2cdcdd2SPaul Dagnelie 2417efb80947Sahrens return (err); 2418efb80947Sahrens } 2419efb80947Sahrens 24203b2aab18SMatthew Ahrens /* used to destroy the drc_ds on error */ 24213b2aab18SMatthew Ahrens static void 24223b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) 24233b2aab18SMatthew Ahrens { 24249c3fd121SMatthew Ahrens if (drc->drc_resumable) { 24259c3fd121SMatthew Ahrens /* wait for our resume state to be written to disk */ 24269c3fd121SMatthew Ahrens txg_wait_synced(drc->drc_ds->ds_dir->dd_pool, 0); 24279c3fd121SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 24289c3fd121SMatthew Ahrens } else { 24299adfa60dSMatthew Ahrens char name[ZFS_MAX_DATASET_NAME_LEN]; 24309c3fd121SMatthew Ahrens dsl_dataset_name(drc->drc_ds, name); 24319c3fd121SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 24329c3fd121SMatthew Ahrens (void) dsl_destroy_head(name); 24339c3fd121SMatthew Ahrens } 24343b2aab18SMatthew Ahrens } 24353b2aab18SMatthew Ahrens 243698110f08SMatthew Ahrens static void 2437a2cdcdd2SPaul Dagnelie receive_cksum(struct receive_arg *ra, int len, void *buf) 243898110f08SMatthew Ahrens { 243998110f08SMatthew Ahrens if (ra->byteswap) { 2440770499e1SDan Kimmel (void) fletcher_4_incremental_byteswap(buf, len, &ra->cksum); 244198110f08SMatthew Ahrens } else { 2442770499e1SDan Kimmel (void) fletcher_4_incremental_native(buf, len, &ra->cksum); 244398110f08SMatthew Ahrens } 244498110f08SMatthew Ahrens } 244598110f08SMatthew Ahrens 244698110f08SMatthew Ahrens /* 2447a2cdcdd2SPaul Dagnelie * Read the payload into a buffer of size len, and update the current record's 2448a2cdcdd2SPaul Dagnelie * payload field. 2449a2cdcdd2SPaul Dagnelie * Allocate ra->next_rrd and read the next record's header into 2450a2cdcdd2SPaul Dagnelie * ra->next_rrd->header. 245198110f08SMatthew Ahrens * Verify checksum of payload and next record. 245298110f08SMatthew Ahrens */ 245398110f08SMatthew Ahrens static int 2454a2cdcdd2SPaul Dagnelie receive_read_payload_and_next_header(struct receive_arg *ra, int len, void *buf) 245598110f08SMatthew Ahrens { 245698110f08SMatthew Ahrens int err; 245798110f08SMatthew Ahrens 245898110f08SMatthew Ahrens if (len != 0) { 2459a2cdcdd2SPaul Dagnelie ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); 24609c3fd121SMatthew Ahrens err = receive_read(ra, len, buf); 246198110f08SMatthew Ahrens if (err != 0) 246298110f08SMatthew Ahrens return (err); 24639c3fd121SMatthew Ahrens receive_cksum(ra, len, buf); 24649c3fd121SMatthew Ahrens 24659c3fd121SMatthew Ahrens /* note: rrd is NULL when reading the begin record's payload */ 24669c3fd121SMatthew Ahrens if (ra->rrd != NULL) { 24679c3fd121SMatthew Ahrens ra->rrd->payload = buf; 24689c3fd121SMatthew Ahrens ra->rrd->payload_size = len; 24699c3fd121SMatthew Ahrens ra->rrd->bytes_read = ra->bytes_read; 24709c3fd121SMatthew Ahrens } 247198110f08SMatthew Ahrens } 247298110f08SMatthew Ahrens 247398110f08SMatthew Ahrens ra->prev_cksum = ra->cksum; 247498110f08SMatthew Ahrens 2475a2cdcdd2SPaul Dagnelie ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); 2476a2cdcdd2SPaul Dagnelie err = receive_read(ra, sizeof (ra->next_rrd->header), 2477a2cdcdd2SPaul Dagnelie &ra->next_rrd->header); 24789c3fd121SMatthew Ahrens ra->next_rrd->bytes_read = ra->bytes_read; 2479a2cdcdd2SPaul Dagnelie if (err != 0) { 2480a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2481a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 248298110f08SMatthew Ahrens return (err); 2483a2cdcdd2SPaul Dagnelie } 2484a2cdcdd2SPaul Dagnelie if (ra->next_rrd->header.drr_type == DRR_BEGIN) { 2485a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2486a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 248798110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 2488a2cdcdd2SPaul Dagnelie } 248998110f08SMatthew Ahrens 249098110f08SMatthew Ahrens /* 249198110f08SMatthew Ahrens * Note: checksum is of everything up to but not including the 249298110f08SMatthew Ahrens * checksum itself. 249398110f08SMatthew Ahrens */ 249498110f08SMatthew Ahrens ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 249598110f08SMatthew Ahrens ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 2496a2cdcdd2SPaul Dagnelie receive_cksum(ra, 249798110f08SMatthew Ahrens offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 2498a2cdcdd2SPaul Dagnelie &ra->next_rrd->header); 249998110f08SMatthew Ahrens 2500a2cdcdd2SPaul Dagnelie zio_cksum_t cksum_orig = 2501a2cdcdd2SPaul Dagnelie ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; 2502a2cdcdd2SPaul Dagnelie zio_cksum_t *cksump = 2503a2cdcdd2SPaul Dagnelie &ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; 250498110f08SMatthew Ahrens 250598110f08SMatthew Ahrens if (ra->byteswap) 2506a2cdcdd2SPaul Dagnelie byteswap_record(&ra->next_rrd->header); 250798110f08SMatthew Ahrens 250898110f08SMatthew Ahrens if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && 2509a2cdcdd2SPaul Dagnelie !ZIO_CHECKSUM_EQUAL(ra->cksum, *cksump)) { 2510a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2511a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 251298110f08SMatthew Ahrens return (SET_ERROR(ECKSUM)); 2513a2cdcdd2SPaul Dagnelie } 251498110f08SMatthew Ahrens 2515a2cdcdd2SPaul Dagnelie receive_cksum(ra, sizeof (cksum_orig), &cksum_orig); 251698110f08SMatthew Ahrens 251798110f08SMatthew Ahrens return (0); 251898110f08SMatthew Ahrens } 251998110f08SMatthew Ahrens 252068ecb2ecSPaul Dagnelie static void 252168ecb2ecSPaul Dagnelie objlist_create(struct objlist *list) 252268ecb2ecSPaul Dagnelie { 252368ecb2ecSPaul Dagnelie list_create(&list->list, sizeof (struct receive_objnode), 252468ecb2ecSPaul Dagnelie offsetof(struct receive_objnode, node)); 252568ecb2ecSPaul Dagnelie list->last_lookup = 0; 252668ecb2ecSPaul Dagnelie } 252768ecb2ecSPaul Dagnelie 252868ecb2ecSPaul Dagnelie static void 252968ecb2ecSPaul Dagnelie objlist_destroy(struct objlist *list) 253068ecb2ecSPaul Dagnelie { 253168ecb2ecSPaul Dagnelie for (struct receive_objnode *n = list_remove_head(&list->list); 253268ecb2ecSPaul Dagnelie n != NULL; n = list_remove_head(&list->list)) { 253368ecb2ecSPaul Dagnelie kmem_free(n, sizeof (*n)); 253468ecb2ecSPaul Dagnelie } 253568ecb2ecSPaul Dagnelie list_destroy(&list->list); 253668ecb2ecSPaul Dagnelie } 253768ecb2ecSPaul Dagnelie 253868ecb2ecSPaul Dagnelie /* 253968ecb2ecSPaul Dagnelie * This function looks through the objlist to see if the specified object number 254068ecb2ecSPaul Dagnelie * is contained in the objlist. In the process, it will remove all object 254168ecb2ecSPaul Dagnelie * numbers in the list that are smaller than the specified object number. Thus, 254268ecb2ecSPaul Dagnelie * any lookup of an object number smaller than a previously looked up object 254368ecb2ecSPaul Dagnelie * number will always return false; therefore, all lookups should be done in 254468ecb2ecSPaul Dagnelie * ascending order. 254568ecb2ecSPaul Dagnelie */ 254668ecb2ecSPaul Dagnelie static boolean_t 254768ecb2ecSPaul Dagnelie objlist_exists(struct objlist *list, uint64_t object) 254868ecb2ecSPaul Dagnelie { 254968ecb2ecSPaul Dagnelie struct receive_objnode *node = list_head(&list->list); 255068ecb2ecSPaul Dagnelie ASSERT3U(object, >=, list->last_lookup); 255168ecb2ecSPaul Dagnelie list->last_lookup = object; 255268ecb2ecSPaul Dagnelie while (node != NULL && node->object < object) { 255368ecb2ecSPaul Dagnelie VERIFY3P(node, ==, list_remove_head(&list->list)); 255468ecb2ecSPaul Dagnelie kmem_free(node, sizeof (*node)); 255568ecb2ecSPaul Dagnelie node = list_head(&list->list); 255668ecb2ecSPaul Dagnelie } 255768ecb2ecSPaul Dagnelie return (node != NULL && node->object == object); 255868ecb2ecSPaul Dagnelie } 255968ecb2ecSPaul Dagnelie 256068ecb2ecSPaul Dagnelie /* 256168ecb2ecSPaul Dagnelie * The objlist is a list of object numbers stored in ascending order. However, 256268ecb2ecSPaul Dagnelie * the insertion of new object numbers does not seek out the correct location to 256368ecb2ecSPaul Dagnelie * store a new object number; instead, it appends it to the list for simplicity. 256468ecb2ecSPaul Dagnelie * Thus, any users must take care to only insert new object numbers in ascending 256568ecb2ecSPaul Dagnelie * order. 256668ecb2ecSPaul Dagnelie */ 256768ecb2ecSPaul Dagnelie static void 256868ecb2ecSPaul Dagnelie objlist_insert(struct objlist *list, uint64_t object) 256968ecb2ecSPaul Dagnelie { 257068ecb2ecSPaul Dagnelie struct receive_objnode *node = kmem_zalloc(sizeof (*node), KM_SLEEP); 257168ecb2ecSPaul Dagnelie node->object = object; 257268ecb2ecSPaul Dagnelie #ifdef ZFS_DEBUG 257368ecb2ecSPaul Dagnelie struct receive_objnode *last_object = list_tail(&list->list); 257468ecb2ecSPaul Dagnelie uint64_t last_objnum = (last_object != NULL ? last_object->object : 0); 257568ecb2ecSPaul Dagnelie ASSERT3U(node->object, >, last_objnum); 257668ecb2ecSPaul Dagnelie #endif 257768ecb2ecSPaul Dagnelie list_insert_tail(&list->list, node); 257868ecb2ecSPaul Dagnelie } 257968ecb2ecSPaul Dagnelie 2580a2cdcdd2SPaul Dagnelie /* 2581a2cdcdd2SPaul Dagnelie * Issue the prefetch reads for any necessary indirect blocks. 2582a2cdcdd2SPaul Dagnelie * 2583a2cdcdd2SPaul Dagnelie * We use the object ignore list to tell us whether or not to issue prefetches 2584a2cdcdd2SPaul Dagnelie * for a given object. We do this for both correctness (in case the blocksize 2585a2cdcdd2SPaul Dagnelie * of an object has changed) and performance (if the object doesn't exist, don't 2586a2cdcdd2SPaul Dagnelie * needlessly try to issue prefetches). We also trim the list as we go through 2587a2cdcdd2SPaul Dagnelie * the stream to prevent it from growing to an unbounded size. 2588a2cdcdd2SPaul Dagnelie * 2589a2cdcdd2SPaul Dagnelie * The object numbers within will always be in sorted order, and any write 2590a2cdcdd2SPaul Dagnelie * records we see will also be in sorted order, but they're not sorted with 2591a2cdcdd2SPaul Dagnelie * respect to each other (i.e. we can get several object records before 2592a2cdcdd2SPaul Dagnelie * receiving each object's write records). As a result, once we've reached a 2593a2cdcdd2SPaul Dagnelie * given object number, we can safely remove any reference to lower object 2594a2cdcdd2SPaul Dagnelie * numbers in the ignore list. In practice, we receive up to 32 object records 2595a2cdcdd2SPaul Dagnelie * before receiving write records, so the list can have up to 32 nodes in it. 2596a2cdcdd2SPaul Dagnelie */ 2597a2cdcdd2SPaul Dagnelie /* ARGSUSED */ 2598a2cdcdd2SPaul Dagnelie static void 2599a2cdcdd2SPaul Dagnelie receive_read_prefetch(struct receive_arg *ra, 2600a2cdcdd2SPaul Dagnelie uint64_t object, uint64_t offset, uint64_t length) 2601a2cdcdd2SPaul Dagnelie { 260268ecb2ecSPaul Dagnelie if (!objlist_exists(&ra->ignore_objlist, object)) { 2603a2cdcdd2SPaul Dagnelie dmu_prefetch(ra->os, object, 1, offset, length, 2604a2cdcdd2SPaul Dagnelie ZIO_PRIORITY_SYNC_READ); 2605a2cdcdd2SPaul Dagnelie } 2606a2cdcdd2SPaul Dagnelie } 2607a2cdcdd2SPaul Dagnelie 2608a2cdcdd2SPaul Dagnelie /* 2609a2cdcdd2SPaul Dagnelie * Read records off the stream, issuing any necessary prefetches. 2610a2cdcdd2SPaul Dagnelie */ 261198110f08SMatthew Ahrens static int 2612a2cdcdd2SPaul Dagnelie receive_read_record(struct receive_arg *ra) 261398110f08SMatthew Ahrens { 261498110f08SMatthew Ahrens int err; 261598110f08SMatthew Ahrens 2616a2cdcdd2SPaul Dagnelie switch (ra->rrd->header.drr_type) { 261798110f08SMatthew Ahrens case DRR_OBJECT: 261898110f08SMatthew Ahrens { 2619a2cdcdd2SPaul Dagnelie struct drr_object *drro = &ra->rrd->header.drr_u.drr_object; 2620a2cdcdd2SPaul Dagnelie uint32_t size = P2ROUNDUP(drro->drr_bonuslen, 8); 2621a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(size, KM_SLEEP); 2622a2cdcdd2SPaul Dagnelie dmu_object_info_t doi; 2623a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, size, buf); 2624a2cdcdd2SPaul Dagnelie if (err != 0) { 2625a2cdcdd2SPaul Dagnelie kmem_free(buf, size); 262698110f08SMatthew Ahrens return (err); 2627a2cdcdd2SPaul Dagnelie } 2628a2cdcdd2SPaul Dagnelie err = dmu_object_info(ra->os, drro->drr_object, &doi); 2629a2cdcdd2SPaul Dagnelie /* 2630a2cdcdd2SPaul Dagnelie * See receive_read_prefetch for an explanation why we're 2631a2cdcdd2SPaul Dagnelie * storing this object in the ignore_obj_list. 2632a2cdcdd2SPaul Dagnelie */ 2633a2cdcdd2SPaul Dagnelie if (err == ENOENT || 2634a2cdcdd2SPaul Dagnelie (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { 263568ecb2ecSPaul Dagnelie objlist_insert(&ra->ignore_objlist, drro->drr_object); 2636a2cdcdd2SPaul Dagnelie err = 0; 2637a2cdcdd2SPaul Dagnelie } 2638a2cdcdd2SPaul Dagnelie return (err); 263998110f08SMatthew Ahrens } 264098110f08SMatthew Ahrens case DRR_FREEOBJECTS: 264198110f08SMatthew Ahrens { 2642a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2643a2cdcdd2SPaul Dagnelie return (err); 264498110f08SMatthew Ahrens } 264598110f08SMatthew Ahrens case DRR_WRITE: 264698110f08SMatthew Ahrens { 2647a2cdcdd2SPaul Dagnelie struct drr_write *drrw = &ra->rrd->header.drr_u.drr_write; 26485602294fSDan Kimmel arc_buf_t *abuf; 26495602294fSDan Kimmel boolean_t is_meta = DMU_OT_IS_METADATA(drrw->drr_type); 26505602294fSDan Kimmel if (DRR_WRITE_COMPRESSED(drrw)) { 26515602294fSDan Kimmel ASSERT3U(drrw->drr_compressed_size, >, 0); 26525602294fSDan Kimmel ASSERT3U(drrw->drr_logical_size, >=, 26535602294fSDan Kimmel drrw->drr_compressed_size); 26545602294fSDan Kimmel ASSERT(!is_meta); 26555602294fSDan Kimmel abuf = arc_loan_compressed_buf( 26565602294fSDan Kimmel dmu_objset_spa(ra->os), 26575602294fSDan Kimmel drrw->drr_compressed_size, drrw->drr_logical_size, 26585602294fSDan Kimmel drrw->drr_compressiontype); 26595602294fSDan Kimmel } else { 26605602294fSDan Kimmel abuf = arc_loan_buf(dmu_objset_spa(ra->os), 26615602294fSDan Kimmel is_meta, drrw->drr_logical_size); 26625602294fSDan Kimmel } 266398110f08SMatthew Ahrens 2664a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 26655602294fSDan Kimmel DRR_WRITE_PAYLOAD_SIZE(drrw), abuf->b_data); 2666a2cdcdd2SPaul Dagnelie if (err != 0) { 266798110f08SMatthew Ahrens dmu_return_arcbuf(abuf); 2668a2cdcdd2SPaul Dagnelie return (err); 2669a2cdcdd2SPaul Dagnelie } 2670a2cdcdd2SPaul Dagnelie ra->rrd->write_buf = abuf; 2671a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrw->drr_object, drrw->drr_offset, 26725602294fSDan Kimmel drrw->drr_logical_size); 267398110f08SMatthew Ahrens return (err); 267498110f08SMatthew Ahrens } 267598110f08SMatthew Ahrens case DRR_WRITE_BYREF: 267698110f08SMatthew Ahrens { 2677a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwb = 2678a2cdcdd2SPaul Dagnelie &ra->rrd->header.drr_u.drr_write_byref; 2679a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2680a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrwb->drr_object, drrwb->drr_offset, 2681a2cdcdd2SPaul Dagnelie drrwb->drr_length); 2682a2cdcdd2SPaul Dagnelie return (err); 268398110f08SMatthew Ahrens } 268498110f08SMatthew Ahrens case DRR_WRITE_EMBEDDED: 268598110f08SMatthew Ahrens { 268698110f08SMatthew Ahrens struct drr_write_embedded *drrwe = 2687a2cdcdd2SPaul Dagnelie &ra->rrd->header.drr_u.drr_write_embedded; 2688a2cdcdd2SPaul Dagnelie uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); 2689a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(size, KM_SLEEP); 2690a2cdcdd2SPaul Dagnelie 2691a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, size, buf); 2692a2cdcdd2SPaul Dagnelie if (err != 0) { 2693a2cdcdd2SPaul Dagnelie kmem_free(buf, size); 269498110f08SMatthew Ahrens return (err); 2695a2cdcdd2SPaul Dagnelie } 2696a2cdcdd2SPaul Dagnelie 2697a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrwe->drr_object, drrwe->drr_offset, 2698a2cdcdd2SPaul Dagnelie drrwe->drr_length); 2699a2cdcdd2SPaul Dagnelie return (err); 270098110f08SMatthew Ahrens } 270198110f08SMatthew Ahrens case DRR_FREE: 270298110f08SMatthew Ahrens { 2703a2cdcdd2SPaul Dagnelie /* 2704a2cdcdd2SPaul Dagnelie * It might be beneficial to prefetch indirect blocks here, but 2705a2cdcdd2SPaul Dagnelie * we don't really have the data to decide for sure. 2706a2cdcdd2SPaul Dagnelie */ 2707a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2708a2cdcdd2SPaul Dagnelie return (err); 270998110f08SMatthew Ahrens } 271098110f08SMatthew Ahrens case DRR_END: 271198110f08SMatthew Ahrens { 2712a2cdcdd2SPaul Dagnelie struct drr_end *drre = &ra->rrd->header.drr_u.drr_end; 271398110f08SMatthew Ahrens if (!ZIO_CHECKSUM_EQUAL(ra->prev_cksum, drre->drr_checksum)) 27149c3fd121SMatthew Ahrens return (SET_ERROR(ECKSUM)); 271598110f08SMatthew Ahrens return (0); 271698110f08SMatthew Ahrens } 271798110f08SMatthew Ahrens case DRR_SPILL: 271898110f08SMatthew Ahrens { 2719a2cdcdd2SPaul Dagnelie struct drr_spill *drrs = &ra->rrd->header.drr_u.drr_spill; 2720a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(drrs->drr_length, KM_SLEEP); 2721a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, drrs->drr_length, 2722a2cdcdd2SPaul Dagnelie buf); 272398110f08SMatthew Ahrens if (err != 0) 2724a2cdcdd2SPaul Dagnelie kmem_free(buf, drrs->drr_length); 2725a2cdcdd2SPaul Dagnelie return (err); 272698110f08SMatthew Ahrens } 272798110f08SMatthew Ahrens default: 272898110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 272998110f08SMatthew Ahrens } 273098110f08SMatthew Ahrens } 273198110f08SMatthew Ahrens 27323cb34c60Sahrens /* 2733a2cdcdd2SPaul Dagnelie * Commit the records to the pool. 2734a2cdcdd2SPaul Dagnelie */ 2735a2cdcdd2SPaul Dagnelie static int 2736a2cdcdd2SPaul Dagnelie receive_process_record(struct receive_writer_arg *rwa, 2737a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd) 2738a2cdcdd2SPaul Dagnelie { 2739a2cdcdd2SPaul Dagnelie int err; 2740a2cdcdd2SPaul Dagnelie 27419c3fd121SMatthew Ahrens /* Processing in order, therefore bytes_read should be increasing. */ 27429c3fd121SMatthew Ahrens ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); 27439c3fd121SMatthew Ahrens rwa->bytes_read = rrd->bytes_read; 27449c3fd121SMatthew Ahrens 2745a2cdcdd2SPaul Dagnelie switch (rrd->header.drr_type) { 2746a2cdcdd2SPaul Dagnelie case DRR_OBJECT: 2747a2cdcdd2SPaul Dagnelie { 2748a2cdcdd2SPaul Dagnelie struct drr_object *drro = &rrd->header.drr_u.drr_object; 2749a2cdcdd2SPaul Dagnelie err = receive_object(rwa, drro, rrd->payload); 2750a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2751a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2752a2cdcdd2SPaul Dagnelie return (err); 2753a2cdcdd2SPaul Dagnelie } 2754a2cdcdd2SPaul Dagnelie case DRR_FREEOBJECTS: 2755a2cdcdd2SPaul Dagnelie { 2756a2cdcdd2SPaul Dagnelie struct drr_freeobjects *drrfo = 2757a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_freeobjects; 2758a2cdcdd2SPaul Dagnelie return (receive_freeobjects(rwa, drrfo)); 2759a2cdcdd2SPaul Dagnelie } 2760a2cdcdd2SPaul Dagnelie case DRR_WRITE: 2761a2cdcdd2SPaul Dagnelie { 2762a2cdcdd2SPaul Dagnelie struct drr_write *drrw = &rrd->header.drr_u.drr_write; 2763a2cdcdd2SPaul Dagnelie err = receive_write(rwa, drrw, rrd->write_buf); 2764a2cdcdd2SPaul Dagnelie /* if receive_write() is successful, it consumes the arc_buf */ 2765a2cdcdd2SPaul Dagnelie if (err != 0) 2766a2cdcdd2SPaul Dagnelie dmu_return_arcbuf(rrd->write_buf); 2767a2cdcdd2SPaul Dagnelie rrd->write_buf = NULL; 2768a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2769a2cdcdd2SPaul Dagnelie return (err); 2770a2cdcdd2SPaul Dagnelie } 2771a2cdcdd2SPaul Dagnelie case DRR_WRITE_BYREF: 2772a2cdcdd2SPaul Dagnelie { 2773a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwbr = 2774a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_write_byref; 2775a2cdcdd2SPaul Dagnelie return (receive_write_byref(rwa, drrwbr)); 2776a2cdcdd2SPaul Dagnelie } 2777a2cdcdd2SPaul Dagnelie case DRR_WRITE_EMBEDDED: 2778a2cdcdd2SPaul Dagnelie { 2779a2cdcdd2SPaul Dagnelie struct drr_write_embedded *drrwe = 2780a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_write_embedded; 2781a2cdcdd2SPaul Dagnelie err = receive_write_embedded(rwa, drrwe, rrd->payload); 2782a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2783a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2784a2cdcdd2SPaul Dagnelie return (err); 2785a2cdcdd2SPaul Dagnelie } 2786a2cdcdd2SPaul Dagnelie case DRR_FREE: 2787a2cdcdd2SPaul Dagnelie { 2788a2cdcdd2SPaul Dagnelie struct drr_free *drrf = &rrd->header.drr_u.drr_free; 2789a2cdcdd2SPaul Dagnelie return (receive_free(rwa, drrf)); 2790a2cdcdd2SPaul Dagnelie } 2791a2cdcdd2SPaul Dagnelie case DRR_SPILL: 2792a2cdcdd2SPaul Dagnelie { 2793a2cdcdd2SPaul Dagnelie struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; 2794a2cdcdd2SPaul Dagnelie err = receive_spill(rwa, drrs, rrd->payload); 2795a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2796a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2797a2cdcdd2SPaul Dagnelie return (err); 2798a2cdcdd2SPaul Dagnelie } 2799a2cdcdd2SPaul Dagnelie default: 2800a2cdcdd2SPaul Dagnelie return (SET_ERROR(EINVAL)); 2801a2cdcdd2SPaul Dagnelie } 2802a2cdcdd2SPaul Dagnelie } 2803a2cdcdd2SPaul Dagnelie 2804a2cdcdd2SPaul Dagnelie /* 2805a2cdcdd2SPaul Dagnelie * dmu_recv_stream's worker thread; pull records off the queue, and then call 2806a2cdcdd2SPaul Dagnelie * receive_process_record When we're done, signal the main thread and exit. 2807a2cdcdd2SPaul Dagnelie */ 2808a2cdcdd2SPaul Dagnelie static void 2809a2cdcdd2SPaul Dagnelie receive_writer_thread(void *arg) 2810a2cdcdd2SPaul Dagnelie { 2811a2cdcdd2SPaul Dagnelie struct receive_writer_arg *rwa = arg; 2812a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd; 2813a2cdcdd2SPaul Dagnelie for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; 2814a2cdcdd2SPaul Dagnelie rrd = bqueue_dequeue(&rwa->q)) { 2815a2cdcdd2SPaul Dagnelie /* 2816a2cdcdd2SPaul Dagnelie * If there's an error, the main thread will stop putting things 2817a2cdcdd2SPaul Dagnelie * on the queue, but we need to clear everything in it before we 2818a2cdcdd2SPaul Dagnelie * can exit. 2819a2cdcdd2SPaul Dagnelie */ 2820a2cdcdd2SPaul Dagnelie if (rwa->err == 0) { 2821a2cdcdd2SPaul Dagnelie rwa->err = receive_process_record(rwa, rrd); 2822a2cdcdd2SPaul Dagnelie } else if (rrd->write_buf != NULL) { 2823a2cdcdd2SPaul Dagnelie dmu_return_arcbuf(rrd->write_buf); 2824a2cdcdd2SPaul Dagnelie rrd->write_buf = NULL; 2825a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2826a2cdcdd2SPaul Dagnelie } else if (rrd->payload != NULL) { 2827a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2828a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2829a2cdcdd2SPaul Dagnelie } 2830a2cdcdd2SPaul Dagnelie kmem_free(rrd, sizeof (*rrd)); 2831a2cdcdd2SPaul Dagnelie } 2832a2cdcdd2SPaul Dagnelie kmem_free(rrd, sizeof (*rrd)); 2833a2cdcdd2SPaul Dagnelie mutex_enter(&rwa->mutex); 2834a2cdcdd2SPaul Dagnelie rwa->done = B_TRUE; 2835a2cdcdd2SPaul Dagnelie cv_signal(&rwa->cv); 2836a2cdcdd2SPaul Dagnelie mutex_exit(&rwa->mutex); 2837f2c1e9bcSJorgen Lundman thread_exit(); 2838a2cdcdd2SPaul Dagnelie } 2839a2cdcdd2SPaul Dagnelie 28409c3fd121SMatthew Ahrens static int 28419c3fd121SMatthew Ahrens resume_check(struct receive_arg *ra, nvlist_t *begin_nvl) 28429c3fd121SMatthew Ahrens { 28439c3fd121SMatthew Ahrens uint64_t val; 28449c3fd121SMatthew Ahrens objset_t *mos = dmu_objset_pool(ra->os)->dp_meta_objset; 28459c3fd121SMatthew Ahrens uint64_t dsobj = dmu_objset_id(ra->os); 28469c3fd121SMatthew Ahrens uint64_t resume_obj, resume_off; 28479c3fd121SMatthew Ahrens 28489c3fd121SMatthew Ahrens if (nvlist_lookup_uint64(begin_nvl, 28499c3fd121SMatthew Ahrens "resume_object", &resume_obj) != 0 || 28509c3fd121SMatthew Ahrens nvlist_lookup_uint64(begin_nvl, 28519c3fd121SMatthew Ahrens "resume_offset", &resume_off) != 0) { 28529c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 28539c3fd121SMatthew Ahrens } 28549c3fd121SMatthew Ahrens VERIFY0(zap_lookup(mos, dsobj, 28559c3fd121SMatthew Ahrens DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); 28569c3fd121SMatthew Ahrens if (resume_obj != val) 28579c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 28589c3fd121SMatthew Ahrens VERIFY0(zap_lookup(mos, dsobj, 28599c3fd121SMatthew Ahrens DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); 28609c3fd121SMatthew Ahrens if (resume_off != val) 28619c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 28629c3fd121SMatthew Ahrens 28639c3fd121SMatthew Ahrens return (0); 28649c3fd121SMatthew Ahrens } 28659c3fd121SMatthew Ahrens 2866a2cdcdd2SPaul Dagnelie /* 2867a2cdcdd2SPaul Dagnelie * Read in the stream's records, one by one, and apply them to the pool. There 2868a2cdcdd2SPaul Dagnelie * are two threads involved; the thread that calls this function will spin up a 2869a2cdcdd2SPaul Dagnelie * worker thread, read the records off the stream one by one, and issue 2870a2cdcdd2SPaul Dagnelie * prefetches for any necessary indirect blocks. It will then push the records 2871a2cdcdd2SPaul Dagnelie * onto an internal blocking queue. The worker thread will pull the records off 2872a2cdcdd2SPaul Dagnelie * the queue, and actually write the data into the DMU. This way, the worker 2873a2cdcdd2SPaul Dagnelie * thread doesn't have to wait for reads to complete, since everything it needs 2874a2cdcdd2SPaul Dagnelie * (the indirect blocks) will be prefetched. 2875a2cdcdd2SPaul Dagnelie * 28763cb34c60Sahrens * NB: callers *must* call dmu_recv_end() if this succeeds. 28773cb34c60Sahrens */ 2878efb80947Sahrens int 2879c99e4bdcSChris Kirby dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, 2880c99e4bdcSChris Kirby int cleanup_fd, uint64_t *action_handlep) 2881efb80947Sahrens { 288298110f08SMatthew Ahrens int err = 0; 2883a2cdcdd2SPaul Dagnelie struct receive_arg ra = { 0 }; 2884a2cdcdd2SPaul Dagnelie struct receive_writer_arg rwa = { 0 }; 28859e69d7d0SLori Alt int featureflags; 28869c3fd121SMatthew Ahrens nvlist_t *begin_nvl = NULL; 2887efb80947Sahrens 28883b2aab18SMatthew Ahrens ra.byteswap = drc->drc_byteswap; 28893b2aab18SMatthew Ahrens ra.cksum = drc->drc_cksum; 28903cb34c60Sahrens ra.vp = vp; 28913cb34c60Sahrens ra.voff = *voffp; 28929c3fd121SMatthew Ahrens 28939c3fd121SMatthew Ahrens if (dsl_dataset_is_zapified(drc->drc_ds)) { 28949c3fd121SMatthew Ahrens (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, 28959c3fd121SMatthew Ahrens drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, 28969c3fd121SMatthew Ahrens sizeof (ra.bytes_read), 1, &ra.bytes_read); 28979c3fd121SMatthew Ahrens } 28989c3fd121SMatthew Ahrens 289968ecb2ecSPaul Dagnelie objlist_create(&ra.ignore_objlist); 2900efb80947Sahrens 29013cb34c60Sahrens /* these were verified in dmu_recv_begin */ 29023b2aab18SMatthew Ahrens ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, 29039e69d7d0SLori Alt DMU_SUBSTREAM); 29043b2aab18SMatthew Ahrens ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); 2905efb80947Sahrens 2906efb80947Sahrens /* 2907efb80947Sahrens * Open the objset we are modifying. 2908efb80947Sahrens */ 290998110f08SMatthew Ahrens VERIFY0(dmu_objset_from_ds(drc->drc_ds, &ra.os)); 2910efb80947Sahrens 2911c1379625SJustin T. Gibbs ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); 2912efb80947Sahrens 29139e69d7d0SLori Alt featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); 29149e69d7d0SLori Alt 29159e69d7d0SLori Alt /* if this stream is dedup'ed, set up the avl tree for guid mapping */ 29169e69d7d0SLori Alt if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { 2917a7f53a56SChris Kirby minor_t minor; 2918a7f53a56SChris Kirby 2919c99e4bdcSChris Kirby if (cleanup_fd == -1) { 2920be6fd75aSMatthew Ahrens ra.err = SET_ERROR(EBADF); 2921c99e4bdcSChris Kirby goto out; 2922c99e4bdcSChris Kirby } 2923a7f53a56SChris Kirby ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor); 29243b2aab18SMatthew Ahrens if (ra.err != 0) { 2925a7f53a56SChris Kirby cleanup_fd = -1; 2926a7f53a56SChris Kirby goto out; 2927a7f53a56SChris Kirby } 2928a7f53a56SChris Kirby 2929c99e4bdcSChris Kirby if (*action_handlep == 0) { 2930a2cdcdd2SPaul Dagnelie rwa.guid_to_ds_map = 2931c99e4bdcSChris Kirby kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); 2932a2cdcdd2SPaul Dagnelie avl_create(rwa.guid_to_ds_map, guid_compare, 2933c99e4bdcSChris Kirby sizeof (guid_map_entry_t), 2934c99e4bdcSChris Kirby offsetof(guid_map_entry_t, avlnode)); 293598110f08SMatthew Ahrens err = zfs_onexit_add_cb(minor, 2936a2cdcdd2SPaul Dagnelie free_guid_map_onexit, rwa.guid_to_ds_map, 2937c99e4bdcSChris Kirby action_handlep); 29383b2aab18SMatthew Ahrens if (ra.err != 0) 2939c99e4bdcSChris Kirby goto out; 2940c99e4bdcSChris Kirby } else { 294198110f08SMatthew Ahrens err = zfs_onexit_cb_data(minor, *action_handlep, 2942a2cdcdd2SPaul Dagnelie (void **)&rwa.guid_to_ds_map); 29433b2aab18SMatthew Ahrens if (ra.err != 0) 2944c99e4bdcSChris Kirby goto out; 2945c99e4bdcSChris Kirby } 2946ec5cf9d5SAlexander Stetsenko 2947a2cdcdd2SPaul Dagnelie drc->drc_guid_to_ds_map = rwa.guid_to_ds_map; 29489e69d7d0SLori Alt } 29499e69d7d0SLori Alt 29509c3fd121SMatthew Ahrens uint32_t payloadlen = drc->drc_drr_begin->drr_payloadlen; 29519c3fd121SMatthew Ahrens void *payload = NULL; 29529c3fd121SMatthew Ahrens if (payloadlen != 0) 29539c3fd121SMatthew Ahrens payload = kmem_alloc(payloadlen, KM_SLEEP); 29549c3fd121SMatthew Ahrens 29559c3fd121SMatthew Ahrens err = receive_read_payload_and_next_header(&ra, payloadlen, payload); 29569c3fd121SMatthew Ahrens if (err != 0) { 29579c3fd121SMatthew Ahrens if (payloadlen != 0) 29589c3fd121SMatthew Ahrens kmem_free(payload, payloadlen); 295998110f08SMatthew Ahrens goto out; 29609c3fd121SMatthew Ahrens } 29619c3fd121SMatthew Ahrens if (payloadlen != 0) { 29629c3fd121SMatthew Ahrens err = nvlist_unpack(payload, payloadlen, &begin_nvl, KM_SLEEP); 29639c3fd121SMatthew Ahrens kmem_free(payload, payloadlen); 29649c3fd121SMatthew Ahrens if (err != 0) 29659c3fd121SMatthew Ahrens goto out; 29669c3fd121SMatthew Ahrens } 29679c3fd121SMatthew Ahrens 29689c3fd121SMatthew Ahrens if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { 29699c3fd121SMatthew Ahrens err = resume_check(&ra, begin_nvl); 29709c3fd121SMatthew Ahrens if (err != 0) 29719c3fd121SMatthew Ahrens goto out; 29729c3fd121SMatthew Ahrens } 297398110f08SMatthew Ahrens 2974a2cdcdd2SPaul Dagnelie (void) bqueue_init(&rwa.q, zfs_recv_queue_length, 2975a2cdcdd2SPaul Dagnelie offsetof(struct receive_record_arg, node)); 2976a2cdcdd2SPaul Dagnelie cv_init(&rwa.cv, NULL, CV_DEFAULT, NULL); 2977a2cdcdd2SPaul Dagnelie mutex_init(&rwa.mutex, NULL, MUTEX_DEFAULT, NULL); 2978a2cdcdd2SPaul Dagnelie rwa.os = ra.os; 2979a2cdcdd2SPaul Dagnelie rwa.byteswap = drc->drc_byteswap; 29809c3fd121SMatthew Ahrens rwa.resumable = drc->drc_resumable; 2981a2cdcdd2SPaul Dagnelie 2982a2cdcdd2SPaul Dagnelie (void) thread_create(NULL, 0, receive_writer_thread, &rwa, 0, curproc, 2983a2cdcdd2SPaul Dagnelie TS_RUN, minclsyspri); 2984a2cdcdd2SPaul Dagnelie /* 2985a2cdcdd2SPaul Dagnelie * We're reading rwa.err without locks, which is safe since we are the 2986a2cdcdd2SPaul Dagnelie * only reader, and the worker thread is the only writer. It's ok if we 2987a2cdcdd2SPaul Dagnelie * miss a write for an iteration or two of the loop, since the writer 2988a2cdcdd2SPaul Dagnelie * thread will keep freeing records we send it until we send it an eos 2989a2cdcdd2SPaul Dagnelie * marker. 2990a2cdcdd2SPaul Dagnelie * 2991a2cdcdd2SPaul Dagnelie * We can leave this loop in 3 ways: First, if rwa.err is 2992a2cdcdd2SPaul Dagnelie * non-zero. In that case, the writer thread will free the rrd we just 2993a2cdcdd2SPaul Dagnelie * pushed. Second, if we're interrupted; in that case, either it's the 2994a2cdcdd2SPaul Dagnelie * first loop and ra.rrd was never allocated, or it's later, and ra.rrd 2995a2cdcdd2SPaul Dagnelie * has been handed off to the writer thread who will free it. Finally, 2996a2cdcdd2SPaul Dagnelie * if receive_read_record fails or we're at the end of the stream, then 2997a2cdcdd2SPaul Dagnelie * we free ra.rrd and exit. 2998a2cdcdd2SPaul Dagnelie */ 2999a2cdcdd2SPaul Dagnelie while (rwa.err == 0) { 3000efb80947Sahrens if (issig(JUSTLOOKING) && issig(FORREAL)) { 300198110f08SMatthew Ahrens err = SET_ERROR(EINTR); 300298110f08SMatthew Ahrens break; 3003efb80947Sahrens } 3004efb80947Sahrens 3005a2cdcdd2SPaul Dagnelie ASSERT3P(ra.rrd, ==, NULL); 3006a2cdcdd2SPaul Dagnelie ra.rrd = ra.next_rrd; 3007a2cdcdd2SPaul Dagnelie ra.next_rrd = NULL; 3008a2cdcdd2SPaul Dagnelie /* Allocates and loads header into ra.next_rrd */ 3009a2cdcdd2SPaul Dagnelie err = receive_read_record(&ra); 3010efb80947Sahrens 3011a2cdcdd2SPaul Dagnelie if (ra.rrd->header.drr_type == DRR_END || err != 0) { 3012a2cdcdd2SPaul Dagnelie kmem_free(ra.rrd, sizeof (*ra.rrd)); 3013a2cdcdd2SPaul Dagnelie ra.rrd = NULL; 30140a586ceaSMark Shellenbaum break; 3015a2cdcdd2SPaul Dagnelie } 3016a2cdcdd2SPaul Dagnelie 3017a2cdcdd2SPaul Dagnelie bqueue_enqueue(&rwa.q, ra.rrd, 3018a2cdcdd2SPaul Dagnelie sizeof (struct receive_record_arg) + ra.rrd->payload_size); 3019a2cdcdd2SPaul Dagnelie ra.rrd = NULL; 3020a2cdcdd2SPaul Dagnelie } 3021a2cdcdd2SPaul Dagnelie if (ra.next_rrd == NULL) 3022a2cdcdd2SPaul Dagnelie ra.next_rrd = kmem_zalloc(sizeof (*ra.next_rrd), KM_SLEEP); 3023a2cdcdd2SPaul Dagnelie ra.next_rrd->eos_marker = B_TRUE; 3024a2cdcdd2SPaul Dagnelie bqueue_enqueue(&rwa.q, ra.next_rrd, 1); 3025a2cdcdd2SPaul Dagnelie 3026a2cdcdd2SPaul Dagnelie mutex_enter(&rwa.mutex); 3027a2cdcdd2SPaul Dagnelie while (!rwa.done) { 3028a2cdcdd2SPaul Dagnelie cv_wait(&rwa.cv, &rwa.mutex); 3029efb80947Sahrens } 3030a2cdcdd2SPaul Dagnelie mutex_exit(&rwa.mutex); 3031a2cdcdd2SPaul Dagnelie 3032a2cdcdd2SPaul Dagnelie cv_destroy(&rwa.cv); 3033a2cdcdd2SPaul Dagnelie mutex_destroy(&rwa.mutex); 3034a2cdcdd2SPaul Dagnelie bqueue_destroy(&rwa.q); 3035a2cdcdd2SPaul Dagnelie if (err == 0) 3036a2cdcdd2SPaul Dagnelie err = rwa.err; 3037efb80947Sahrens 3038efb80947Sahrens out: 30399c3fd121SMatthew Ahrens nvlist_free(begin_nvl); 3040a7f53a56SChris Kirby if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) 3041a7f53a56SChris Kirby zfs_onexit_fd_rele(cleanup_fd); 3042a7f53a56SChris Kirby 304398110f08SMatthew Ahrens if (err != 0) { 3044efb80947Sahrens /* 30459c3fd121SMatthew Ahrens * Clean up references. If receive is not resumable, 30469c3fd121SMatthew Ahrens * destroy what we created, so we don't leave it in 30479c3fd121SMatthew Ahrens * the inconsistent state. 3048efb80947Sahrens */ 30493b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(drc); 3050efb80947Sahrens } 3051efb80947Sahrens 30523cb34c60Sahrens *voffp = ra.voff; 305368ecb2ecSPaul Dagnelie objlist_destroy(&ra.ignore_objlist); 305498110f08SMatthew Ahrens return (err); 3055efb80947Sahrens } 3056f18faf3fSek 30573cb34c60Sahrens static int 30583b2aab18SMatthew Ahrens dmu_recv_end_check(void *arg, dmu_tx_t *tx) 30593cb34c60Sahrens { 30603b2aab18SMatthew Ahrens dmu_recv_cookie_t *drc = arg; 30613b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 30623b2aab18SMatthew Ahrens int error; 30633b2aab18SMatthew Ahrens 30643b2aab18SMatthew Ahrens ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); 30653cb34c60Sahrens 30663b2aab18SMatthew Ahrens if (!drc->drc_newfs) { 30673b2aab18SMatthew Ahrens dsl_dataset_t *origin_head; 30683b2aab18SMatthew Ahrens 30693b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); 30703b2aab18SMatthew Ahrens if (error != 0) 30713b2aab18SMatthew Ahrens return (error); 307234f2f8cfSMatthew Ahrens if (drc->drc_force) { 307334f2f8cfSMatthew Ahrens /* 307434f2f8cfSMatthew Ahrens * We will destroy any snapshots in tofs (i.e. before 307534f2f8cfSMatthew Ahrens * origin_head) that are after the origin (which is 307634f2f8cfSMatthew Ahrens * the snap before drc_ds, because drc_ds can not 307734f2f8cfSMatthew Ahrens * have any snaps of its own). 307834f2f8cfSMatthew Ahrens */ 3079c1379625SJustin T. Gibbs uint64_t obj; 3080c1379625SJustin T. Gibbs 3081c1379625SJustin T. Gibbs obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 3082c1379625SJustin T. Gibbs while (obj != 3083c1379625SJustin T. Gibbs dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 308434f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 308534f2f8cfSMatthew Ahrens error = dsl_dataset_hold_obj(dp, obj, FTAG, 308634f2f8cfSMatthew Ahrens &snap); 308734f2f8cfSMatthew Ahrens if (error != 0) 3088beddaa9cSAndriy Gapon break; 308934f2f8cfSMatthew Ahrens if (snap->ds_dir != origin_head->ds_dir) 309034f2f8cfSMatthew Ahrens error = SET_ERROR(EINVAL); 309134f2f8cfSMatthew Ahrens if (error == 0) { 309234f2f8cfSMatthew Ahrens error = dsl_destroy_snapshot_check_impl( 309334f2f8cfSMatthew Ahrens snap, B_FALSE); 309434f2f8cfSMatthew Ahrens } 3095c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 309634f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 309734f2f8cfSMatthew Ahrens if (error != 0) 3098beddaa9cSAndriy Gapon break; 3099beddaa9cSAndriy Gapon } 3100beddaa9cSAndriy Gapon if (error != 0) { 3101beddaa9cSAndriy Gapon dsl_dataset_rele(origin_head, FTAG); 3102beddaa9cSAndriy Gapon return (error); 310334f2f8cfSMatthew Ahrens } 310434f2f8cfSMatthew Ahrens } 31053b2aab18SMatthew Ahrens error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, 310691948b51SKeith M Wesolowski origin_head, drc->drc_force, drc->drc_owner, tx); 31073b2aab18SMatthew Ahrens if (error != 0) { 31083b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 31093b2aab18SMatthew Ahrens return (error); 31103b2aab18SMatthew Ahrens } 31113b2aab18SMatthew Ahrens error = dsl_dataset_snapshot_check_impl(origin_head, 3112a2afb611SJerry Jelinek drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 31133b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 31143b2aab18SMatthew Ahrens if (error != 0) 31153b2aab18SMatthew Ahrens return (error); 31163b2aab18SMatthew Ahrens 31173b2aab18SMatthew Ahrens error = dsl_destroy_head_check_impl(drc->drc_ds, 1); 31183b2aab18SMatthew Ahrens } else { 31193b2aab18SMatthew Ahrens error = dsl_dataset_snapshot_check_impl(drc->drc_ds, 3120a2afb611SJerry Jelinek drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 31213b2aab18SMatthew Ahrens } 31223b2aab18SMatthew Ahrens return (error); 31233cb34c60Sahrens } 31243cb34c60Sahrens 31253cb34c60Sahrens static void 31263b2aab18SMatthew Ahrens dmu_recv_end_sync(void *arg, dmu_tx_t *tx) 31273cb34c60Sahrens { 31283b2aab18SMatthew Ahrens dmu_recv_cookie_t *drc = arg; 31293b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 31303b2aab18SMatthew Ahrens 31313b2aab18SMatthew Ahrens spa_history_log_internal_ds(drc->drc_ds, "finish receiving", 31323b2aab18SMatthew Ahrens tx, "snap=%s", drc->drc_tosnap); 31333b2aab18SMatthew Ahrens 31343b2aab18SMatthew Ahrens if (!drc->drc_newfs) { 31353b2aab18SMatthew Ahrens dsl_dataset_t *origin_head; 31363b2aab18SMatthew Ahrens 31373b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, 31383b2aab18SMatthew Ahrens &origin_head)); 313934f2f8cfSMatthew Ahrens 314034f2f8cfSMatthew Ahrens if (drc->drc_force) { 314134f2f8cfSMatthew Ahrens /* 314234f2f8cfSMatthew Ahrens * Destroy any snapshots of drc_tofs (origin_head) 314334f2f8cfSMatthew Ahrens * after the origin (the snap before drc_ds). 314434f2f8cfSMatthew Ahrens */ 3145c1379625SJustin T. Gibbs uint64_t obj; 3146c1379625SJustin T. Gibbs 3147c1379625SJustin T. Gibbs obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 3148c1379625SJustin T. Gibbs while (obj != 3149c1379625SJustin T. Gibbs dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 315034f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 315134f2f8cfSMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, 315234f2f8cfSMatthew Ahrens &snap)); 315334f2f8cfSMatthew Ahrens ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); 3154c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 315534f2f8cfSMatthew Ahrens dsl_destroy_snapshot_sync_impl(snap, 315634f2f8cfSMatthew Ahrens B_FALSE, tx); 315734f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 315834f2f8cfSMatthew Ahrens } 315934f2f8cfSMatthew Ahrens } 316034f2f8cfSMatthew Ahrens VERIFY3P(drc->drc_ds->ds_prev, ==, 316134f2f8cfSMatthew Ahrens origin_head->ds_prev); 316234f2f8cfSMatthew Ahrens 31633b2aab18SMatthew Ahrens dsl_dataset_clone_swap_sync_impl(drc->drc_ds, 31643b2aab18SMatthew Ahrens origin_head, tx); 31653b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(origin_head, 31663b2aab18SMatthew Ahrens drc->drc_tosnap, tx); 31673b2aab18SMatthew Ahrens 31683b2aab18SMatthew Ahrens /* set snapshot's creation time and guid */ 31693b2aab18SMatthew Ahrens dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); 3170c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = 31713b2aab18SMatthew Ahrens drc->drc_drrb->drr_creation_time; 3172c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_guid = 31733b2aab18SMatthew Ahrens drc->drc_drrb->drr_toguid; 3174c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= 31753b2aab18SMatthew Ahrens ~DS_FLAG_INCONSISTENT; 31763b2aab18SMatthew Ahrens 31773b2aab18SMatthew Ahrens dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 3178c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head)->ds_flags &= 3179c1379625SJustin T. Gibbs ~DS_FLAG_INCONSISTENT; 31803b2aab18SMatthew Ahrens 31812ad25b40SChris Williamson drc->drc_newsnapobj = 31822ad25b40SChris Williamson dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 31832ad25b40SChris Williamson 31843b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 31853b2aab18SMatthew Ahrens dsl_destroy_head_sync_impl(drc->drc_ds, tx); 318691948b51SKeith M Wesolowski 318791948b51SKeith M Wesolowski if (drc->drc_owner != NULL) 318891948b51SKeith M Wesolowski VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); 31893b2aab18SMatthew Ahrens } else { 31903b2aab18SMatthew Ahrens dsl_dataset_t *ds = drc->drc_ds; 31913cb34c60Sahrens 31923b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); 31933cb34c60Sahrens 31943b2aab18SMatthew Ahrens /* set snapshot's creation time and guid */ 31953b2aab18SMatthew Ahrens dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 3196c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_creation_time = 31973b2aab18SMatthew Ahrens drc->drc_drrb->drr_creation_time; 3198c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_guid = 3199c1379625SJustin T. Gibbs drc->drc_drrb->drr_toguid; 3200c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_flags &= 3201c1379625SJustin T. Gibbs ~DS_FLAG_INCONSISTENT; 32023cb34c60Sahrens 32033b2aab18SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 3204c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 32059c3fd121SMatthew Ahrens if (dsl_dataset_has_resume_receive_state(ds)) { 32069c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 32079c3fd121SMatthew Ahrens DS_FIELD_RESUME_FROMGUID, tx); 32089c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 32099c3fd121SMatthew Ahrens DS_FIELD_RESUME_OBJECT, tx); 32109c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 32119c3fd121SMatthew Ahrens DS_FIELD_RESUME_OFFSET, tx); 32129c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 32139c3fd121SMatthew Ahrens DS_FIELD_RESUME_BYTES, tx); 32149c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 32159c3fd121SMatthew Ahrens DS_FIELD_RESUME_TOGUID, tx); 32169c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 32179c3fd121SMatthew Ahrens DS_FIELD_RESUME_TONAME, tx); 32189c3fd121SMatthew Ahrens } 32192ad25b40SChris Williamson drc->drc_newsnapobj = 32202ad25b40SChris Williamson dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; 32213b2aab18SMatthew Ahrens } 32223b2aab18SMatthew Ahrens /* 32233b2aab18SMatthew Ahrens * Release the hold from dmu_recv_begin. This must be done before 32243b2aab18SMatthew Ahrens * we return to open context, so that when we free the dataset's dnode, 32253b2aab18SMatthew Ahrens * we can evict its bonus buffer. 32263b2aab18SMatthew Ahrens */ 32273b2aab18SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 32283b2aab18SMatthew Ahrens drc->drc_ds = NULL; 32293cb34c60Sahrens } 32303cb34c60Sahrens 3231ec5cf9d5SAlexander Stetsenko static int 32323b2aab18SMatthew Ahrens add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) 3233ec5cf9d5SAlexander Stetsenko { 32343b2aab18SMatthew Ahrens dsl_pool_t *dp; 3235ec5cf9d5SAlexander Stetsenko dsl_dataset_t *snapds; 3236ec5cf9d5SAlexander Stetsenko guid_map_entry_t *gmep; 3237ec5cf9d5SAlexander Stetsenko int err; 3238ec5cf9d5SAlexander Stetsenko 3239ec5cf9d5SAlexander Stetsenko ASSERT(guid_map != NULL); 3240ec5cf9d5SAlexander Stetsenko 32413b2aab18SMatthew Ahrens err = dsl_pool_hold(name, FTAG, &dp); 32423b2aab18SMatthew Ahrens if (err != 0) 32433b2aab18SMatthew Ahrens return (err); 3244de8d9cffSMatthew Ahrens gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); 3245de8d9cffSMatthew Ahrens err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds); 3246ec5cf9d5SAlexander Stetsenko if (err == 0) { 3247c1379625SJustin T. Gibbs gmep->guid = dsl_dataset_phys(snapds)->ds_guid; 3248ec5cf9d5SAlexander Stetsenko gmep->gme_ds = snapds; 3249ec5cf9d5SAlexander Stetsenko avl_add(guid_map, gmep); 32503b2aab18SMatthew Ahrens dsl_dataset_long_hold(snapds, gmep); 3251de8d9cffSMatthew Ahrens } else { 3252de8d9cffSMatthew Ahrens kmem_free(gmep, sizeof (*gmep)); 3253ec5cf9d5SAlexander Stetsenko } 3254ec5cf9d5SAlexander Stetsenko 32553b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 3256ec5cf9d5SAlexander Stetsenko return (err); 3257ec5cf9d5SAlexander Stetsenko } 3258ec5cf9d5SAlexander Stetsenko 32593b2aab18SMatthew Ahrens static int dmu_recv_end_modified_blocks = 3; 32603b2aab18SMatthew Ahrens 3261ae46e4c7SMatthew Ahrens static int 3262ae46e4c7SMatthew Ahrens dmu_recv_existing_end(dmu_recv_cookie_t *drc) 3263f18faf3fSek { 32643b2aab18SMatthew Ahrens #ifdef _KERNEL 32653b2aab18SMatthew Ahrens /* 32663b2aab18SMatthew Ahrens * We will be destroying the ds; make sure its origin is unmounted if 32673b2aab18SMatthew Ahrens * necessary. 32683b2aab18SMatthew Ahrens */ 32699adfa60dSMatthew Ahrens char name[ZFS_MAX_DATASET_NAME_LEN]; 32703b2aab18SMatthew Ahrens dsl_dataset_name(drc->drc_ds, name); 32713b2aab18SMatthew Ahrens zfs_destroy_unmount_origin(name); 32723b2aab18SMatthew Ahrens #endif 32733cb34c60Sahrens 32742ad25b40SChris Williamson return (dsl_sync_task(drc->drc_tofs, 32753b2aab18SMatthew Ahrens dmu_recv_end_check, dmu_recv_end_sync, drc, 32762ad25b40SChris Williamson dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); 3277f18faf3fSek } 3278ae46e4c7SMatthew Ahrens 3279ae46e4c7SMatthew Ahrens static int 3280ae46e4c7SMatthew Ahrens dmu_recv_new_end(dmu_recv_cookie_t *drc) 32812ad25b40SChris Williamson { 32822ad25b40SChris Williamson return (dsl_sync_task(drc->drc_tofs, 32832ad25b40SChris Williamson dmu_recv_end_check, dmu_recv_end_sync, drc, 32842ad25b40SChris Williamson dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); 32852ad25b40SChris Williamson } 32862ad25b40SChris Williamson 32872ad25b40SChris Williamson int 32882ad25b40SChris Williamson dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) 3289ae46e4c7SMatthew Ahrens { 32903b2aab18SMatthew Ahrens int error; 3291ae46e4c7SMatthew Ahrens 32922ad25b40SChris Williamson drc->drc_owner = owner; 32932ad25b40SChris Williamson 32942ad25b40SChris Williamson if (drc->drc_newfs) 32952ad25b40SChris Williamson error = dmu_recv_new_end(drc); 32962ad25b40SChris Williamson else 32972ad25b40SChris Williamson error = dmu_recv_existing_end(drc); 3298ae46e4c7SMatthew Ahrens 32993b2aab18SMatthew Ahrens if (error != 0) { 33003b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(drc); 33013b2aab18SMatthew Ahrens } else if (drc->drc_guid_to_ds_map != NULL) { 33023b2aab18SMatthew Ahrens (void) add_ds_to_guidmap(drc->drc_tofs, 33033b2aab18SMatthew Ahrens drc->drc_guid_to_ds_map, 33043b2aab18SMatthew Ahrens drc->drc_newsnapobj); 3305ae46e4c7SMatthew Ahrens } 33063b2aab18SMatthew Ahrens return (error); 3307ae46e4c7SMatthew Ahrens } 3308ae46e4c7SMatthew Ahrens 33092f3d8780SMatthew Ahrens /* 33102f3d8780SMatthew Ahrens * Return TRUE if this objset is currently being received into. 33112f3d8780SMatthew Ahrens */ 33122f3d8780SMatthew Ahrens boolean_t 33132f3d8780SMatthew Ahrens dmu_objset_is_receiving(objset_t *os) 33142f3d8780SMatthew Ahrens { 33152f3d8780SMatthew Ahrens return (os->os_dsl_dataset != NULL && 33162f3d8780SMatthew Ahrens os->os_dsl_dataset->ds_owner == dmu_recv_tag); 33172f3d8780SMatthew Ahrens } 3318