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; 61*6ccda740Sloli int zfs_send_queue_length = SPA_MAXBLOCKSIZE; 62880094b6SAndrew Stormont /* Set this tunable to FALSE to disable setting of DRR_FLAG_FREERECORDS */ 63880094b6SAndrew Stormont int zfs_send_set_freerecords_bit = B_TRUE; 64eb633035STom Caputi /* Set this tunable to FALSE is disable sending unmodified spill blocks. */ 65eb633035STom Caputi int zfs_send_unmodified_spill_blocks = B_TRUE; 6619b94df9SMatthew Ahrens 67df477c0aSPaul Dagnelie /* 68df477c0aSPaul Dagnelie * Use this to override the recordsize calculation for fast zfs send estimates. 69df477c0aSPaul Dagnelie */ 70df477c0aSPaul Dagnelie uint64_t zfs_override_estimate_recordsize = 0; 71df477c0aSPaul Dagnelie 72a2cdcdd2SPaul Dagnelie #define BP_SPAN(datablkszsec, indblkshift, level) \ 73a2cdcdd2SPaul Dagnelie (((uint64_t)datablkszsec) << (SPA_MINBLOCKSHIFT + \ 74a2cdcdd2SPaul Dagnelie (level) * (indblkshift - SPA_BLKPTRSHIFT))) 75a2cdcdd2SPaul Dagnelie 76a2cdcdd2SPaul Dagnelie struct send_thread_arg { 77a2cdcdd2SPaul Dagnelie bqueue_t q; 78a2cdcdd2SPaul Dagnelie dsl_dataset_t *ds; /* Dataset to traverse */ 79a2cdcdd2SPaul Dagnelie uint64_t fromtxg; /* Traverse from this txg */ 80a2cdcdd2SPaul Dagnelie int flags; /* flags to pass to traverse_dataset */ 81a2cdcdd2SPaul Dagnelie int error_code; 82a2cdcdd2SPaul Dagnelie boolean_t cancel; 839c3fd121SMatthew Ahrens zbookmark_phys_t resume; 84a2cdcdd2SPaul Dagnelie }; 85a2cdcdd2SPaul Dagnelie 86a2cdcdd2SPaul Dagnelie struct send_block_record { 87a2cdcdd2SPaul Dagnelie boolean_t eos_marker; /* Marks the end of the stream */ 88a2cdcdd2SPaul Dagnelie blkptr_t bp; 89a2cdcdd2SPaul Dagnelie zbookmark_phys_t zb; 90a2cdcdd2SPaul Dagnelie uint8_t indblkshift; 91a2cdcdd2SPaul Dagnelie uint16_t datablkszsec; 92a2cdcdd2SPaul Dagnelie bqueue_node_t ln; 93a2cdcdd2SPaul Dagnelie }; 94a2cdcdd2SPaul Dagnelie 95eb633035STom Caputi static int do_dump(dmu_sendarg_t *dsa, struct send_block_record *data); 96eb633035STom Caputi 97efb80947Sahrens static int 984e3c9f44SBill Pijewski dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) 99efb80947Sahrens { 1009c3fd121SMatthew Ahrens dsl_dataset_t *ds = dmu_objset_ds(dsp->dsa_os); 101efb80947Sahrens ssize_t resid; /* have to get resid to get detailed errno */ 102c20404ffSEli Rosenthal 103c20404ffSEli Rosenthal /* 104eb633035STom Caputi * The code does not rely on len being a multiple of 8. We keep 105c20404ffSEli Rosenthal * this assertion because of the corresponding assertion in 106c20404ffSEli Rosenthal * receive_read(). Keeping this assertion ensures that we do not 107c20404ffSEli Rosenthal * inadvertently break backwards compatibility (causing the assertion 108eb633035STom Caputi * in receive_read() to trigger on old software). Newer feature flags 109eb633035STom Caputi * (such as raw send) may break this assertion since they were 110eb633035STom Caputi * introduced after the requirement was made obsolete. 111c20404ffSEli Rosenthal */ 112c20404ffSEli Rosenthal 113eb633035STom Caputi ASSERT(len % 8 == 0 || 114eb633035STom Caputi (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) != 0); 115efb80947Sahrens 1164e3c9f44SBill Pijewski dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp, 117efb80947Sahrens (caddr_t)buf, len, 118efb80947Sahrens 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); 1194e3c9f44SBill Pijewski 1204e3c9f44SBill Pijewski mutex_enter(&ds->ds_sendstream_lock); 1214e3c9f44SBill Pijewski *dsp->dsa_off += len; 1224e3c9f44SBill Pijewski mutex_exit(&ds->ds_sendstream_lock); 1234e3c9f44SBill Pijewski 1244e3c9f44SBill Pijewski return (dsp->dsa_err); 125efb80947Sahrens } 126efb80947Sahrens 12798110f08SMatthew Ahrens /* 12898110f08SMatthew Ahrens * For all record types except BEGIN, fill in the checksum (overlaid in 12998110f08SMatthew Ahrens * drr_u.drr_checksum.drr_checksum). The checksum verifies everything 13098110f08SMatthew Ahrens * up to the start of the checksum itself. 13198110f08SMatthew Ahrens */ 13298110f08SMatthew Ahrens static int 13398110f08SMatthew Ahrens dump_record(dmu_sendarg_t *dsp, void *payload, int payload_len) 13498110f08SMatthew Ahrens { 13598110f08SMatthew Ahrens ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 13698110f08SMatthew Ahrens ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 137770499e1SDan Kimmel (void) fletcher_4_incremental_native(dsp->dsa_drr, 13898110f08SMatthew Ahrens offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 13998110f08SMatthew Ahrens &dsp->dsa_zc); 14012b90ee2SMatt Krantz if (dsp->dsa_drr->drr_type == DRR_BEGIN) { 14112b90ee2SMatt Krantz dsp->dsa_sent_begin = B_TRUE; 14212b90ee2SMatt Krantz } else { 14398110f08SMatthew Ahrens ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp->dsa_drr->drr_u. 14498110f08SMatthew Ahrens drr_checksum.drr_checksum)); 14598110f08SMatthew Ahrens dsp->dsa_drr->drr_u.drr_checksum.drr_checksum = dsp->dsa_zc; 14698110f08SMatthew Ahrens } 14712b90ee2SMatt Krantz if (dsp->dsa_drr->drr_type == DRR_END) { 14812b90ee2SMatt Krantz dsp->dsa_sent_end = B_TRUE; 14912b90ee2SMatt Krantz } 150770499e1SDan Kimmel (void) fletcher_4_incremental_native(&dsp->dsa_drr-> 15198110f08SMatthew Ahrens drr_u.drr_checksum.drr_checksum, 15298110f08SMatthew Ahrens sizeof (zio_cksum_t), &dsp->dsa_zc); 15398110f08SMatthew Ahrens if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 15498110f08SMatthew Ahrens return (SET_ERROR(EINTR)); 15598110f08SMatthew Ahrens if (payload_len != 0) { 156770499e1SDan Kimmel (void) fletcher_4_incremental_native(payload, payload_len, 15798110f08SMatthew Ahrens &dsp->dsa_zc); 15898110f08SMatthew Ahrens if (dump_bytes(dsp, payload, payload_len) != 0) 15998110f08SMatthew Ahrens return (SET_ERROR(EINTR)); 16098110f08SMatthew Ahrens } 16198110f08SMatthew Ahrens return (0); 16298110f08SMatthew Ahrens } 16398110f08SMatthew Ahrens 16468ecb2ecSPaul Dagnelie /* 16568ecb2ecSPaul Dagnelie * Fill in the drr_free struct, or perform aggregation if the previous record is 16668ecb2ecSPaul Dagnelie * also a free record, and the two are adjacent. 16768ecb2ecSPaul Dagnelie * 16868ecb2ecSPaul Dagnelie * Note that we send free records even for a full send, because we want to be 16968ecb2ecSPaul Dagnelie * able to receive a full send as a clone, which requires a list of all the free 17068ecb2ecSPaul Dagnelie * and freeobject records that were generated on the source. 17168ecb2ecSPaul Dagnelie */ 172efb80947Sahrens static int 1734e3c9f44SBill Pijewski dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 174efb80947Sahrens uint64_t length) 175efb80947Sahrens { 1764e3c9f44SBill Pijewski struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); 1779e69d7d0SLori Alt 1782f3d8780SMatthew Ahrens /* 1792f3d8780SMatthew Ahrens * When we receive a free record, dbuf_free_range() assumes 1802f3d8780SMatthew Ahrens * that the receiving system doesn't have any dbufs in the range 1812f3d8780SMatthew Ahrens * being freed. This is always true because there is a one-record 1822f3d8780SMatthew Ahrens * constraint: we only send one WRITE record for any given 1839c3fd121SMatthew Ahrens * object,offset. We know that the one-record constraint is 1842f3d8780SMatthew Ahrens * true because we always send data in increasing order by 1852f3d8780SMatthew Ahrens * object,offset. 1862f3d8780SMatthew Ahrens * 1872f3d8780SMatthew Ahrens * If the increasing-order constraint ever changes, we should find 1882f3d8780SMatthew Ahrens * another way to assert that the one-record constraint is still 1892f3d8780SMatthew Ahrens * satisfied. 1902f3d8780SMatthew Ahrens */ 1912f3d8780SMatthew Ahrens ASSERT(object > dsp->dsa_last_data_object || 1922f3d8780SMatthew Ahrens (object == dsp->dsa_last_data_object && 1932f3d8780SMatthew Ahrens offset > dsp->dsa_last_data_offset)); 1942f3d8780SMatthew Ahrens 1959e69d7d0SLori Alt /* 1969e69d7d0SLori Alt * If there is a pending op, but it's not PENDING_FREE, push it out, 1979e69d7d0SLori Alt * since free block aggregation can only be done for blocks of the 1989e69d7d0SLori Alt * same type (i.e., DRR_FREE records can only be aggregated with 1999e69d7d0SLori Alt * other DRR_FREE records. DRR_FREEOBJECTS records can only be 2009e69d7d0SLori Alt * aggregated with other DRR_FREEOBJECTS records. 2019e69d7d0SLori Alt */ 2024e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE && 2034e3c9f44SBill Pijewski dsp->dsa_pending_op != PENDING_FREE) { 20498110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 205be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2064e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2079e69d7d0SLori Alt } 2089e69d7d0SLori Alt 2094e3c9f44SBill Pijewski if (dsp->dsa_pending_op == PENDING_FREE) { 2109e69d7d0SLori Alt /* 211eb633035STom Caputi * There should never be a PENDING_FREE if length is 212eb633035STom Caputi * DMU_OBJECT_END (because dump_dnode is the only place where 213eb633035STom Caputi * this function is called with a DMU_OBJECT_END, and only after 214eb633035STom Caputi * flushing any pending record). 2159e69d7d0SLori Alt */ 216eb633035STom Caputi ASSERT(length != DMU_OBJECT_END); 2179e69d7d0SLori Alt /* 2189e69d7d0SLori Alt * Check to see whether this free block can be aggregated 2199e69d7d0SLori Alt * with pending one. 2209e69d7d0SLori Alt */ 2219e69d7d0SLori Alt if (drrf->drr_object == object && drrf->drr_offset + 2229e69d7d0SLori Alt drrf->drr_length == offset) { 223eb633035STom Caputi if (offset + length < offset) 224eb633035STom Caputi drrf->drr_length = DMU_OBJECT_END; 225eb633035STom Caputi else 226eb633035STom Caputi drrf->drr_length += length; 2279e69d7d0SLori Alt return (0); 2289e69d7d0SLori Alt } else { 2299e69d7d0SLori Alt /* not a continuation. Push out pending record */ 23098110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 231be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2324e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2339e69d7d0SLori Alt } 2349e69d7d0SLori Alt } 2359e69d7d0SLori Alt /* create a FREE record and make it pending */ 2364e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 2374e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_FREE; 2389e69d7d0SLori Alt drrf->drr_object = object; 2399e69d7d0SLori Alt drrf->drr_offset = offset; 240eb633035STom Caputi if (offset + length < offset) 241eb633035STom Caputi drrf->drr_length = DMU_OBJECT_END; 242eb633035STom Caputi else 243eb633035STom Caputi drrf->drr_length = length; 2444e3c9f44SBill Pijewski drrf->drr_toguid = dsp->dsa_toguid; 245eb633035STom Caputi if (length == DMU_OBJECT_END) { 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 256eb633035STom Caputi dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, uint64_t object, 257eb633035STom Caputi uint64_t offset, int lsize, int psize, const blkptr_t *bp, void *data) 258efb80947Sahrens { 2595602294fSDan Kimmel uint64_t payload_size; 260eb633035STom Caputi boolean_t raw = (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW); 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 293eb633035STom Caputi /* only set the compression fields if the buf is compressed or raw */ 294eb633035STom Caputi if (raw || lsize != psize) { 2955602294fSDan Kimmel ASSERT(!BP_IS_EMBEDDED(bp)); 2965602294fSDan Kimmel ASSERT3S(psize, >, 0); 2975602294fSDan Kimmel 298eb633035STom Caputi if (raw) { 299eb633035STom Caputi ASSERT(BP_IS_PROTECTED(bp)); 300eb633035STom Caputi 301eb633035STom Caputi /* 302eb633035STom Caputi * This is a raw protected block so we need to pass 303eb633035STom Caputi * along everything the receiving side will need to 304eb633035STom Caputi * interpret this block, including the byteswap, salt, 305eb633035STom Caputi * IV, and MAC. 306eb633035STom Caputi */ 307eb633035STom Caputi if (BP_SHOULD_BYTESWAP(bp)) 308eb633035STom Caputi drrw->drr_flags |= DRR_RAW_BYTESWAP; 309eb633035STom Caputi zio_crypt_decode_params_bp(bp, drrw->drr_salt, 310eb633035STom Caputi drrw->drr_iv); 311eb633035STom Caputi zio_crypt_decode_mac_bp(bp, drrw->drr_mac); 312eb633035STom Caputi } else { 313eb633035STom Caputi /* this is a compressed block */ 314eb633035STom Caputi ASSERT(dsp->dsa_featureflags & 315eb633035STom Caputi DMU_BACKUP_FEATURE_COMPRESSED); 316eb633035STom Caputi ASSERT(!BP_SHOULD_BYTESWAP(bp)); 317eb633035STom Caputi ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp))); 318eb633035STom Caputi ASSERT3U(BP_GET_COMPRESS(bp), !=, ZIO_COMPRESS_OFF); 319eb633035STom Caputi ASSERT3S(lsize, >=, psize); 320eb633035STom Caputi } 321eb633035STom Caputi 322eb633035STom Caputi /* set fields common to compressed and raw sends */ 3235602294fSDan Kimmel drrw->drr_compressiontype = BP_GET_COMPRESS(bp); 3245602294fSDan Kimmel drrw->drr_compressed_size = psize; 3255602294fSDan Kimmel payload_size = drrw->drr_compressed_size; 3265602294fSDan Kimmel } else { 3275602294fSDan Kimmel payload_size = drrw->drr_logical_size; 3285602294fSDan Kimmel } 3295602294fSDan Kimmel 330eb633035STom Caputi if (bp == NULL || BP_IS_EMBEDDED(bp) || (BP_IS_PROTECTED(bp) && !raw)) { 3315d7b4d43SMatthew Ahrens /* 332eb633035STom Caputi * There's no pre-computed checksum for partial-block writes, 333eb633035STom Caputi * embedded BP's, or encrypted BP's that are being sent as 334eb633035STom Caputi * plaintext, so (like fletcher4-checkummed blocks) userland 335eb633035STom Caputi * will have to compute a dedup-capable checksum itself. 3365d7b4d43SMatthew Ahrens */ 3375d7b4d43SMatthew Ahrens drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; 3385d7b4d43SMatthew Ahrens } else { 3395d7b4d43SMatthew Ahrens drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); 34045818ee1SMatthew Ahrens if (zio_checksum_table[drrw->drr_checksumtype].ci_flags & 34145818ee1SMatthew Ahrens ZCHECKSUM_FLAG_DEDUP) 342eb633035STom Caputi drrw->drr_flags |= DRR_CHECKSUM_DEDUP; 3435d7b4d43SMatthew Ahrens DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); 3445d7b4d43SMatthew Ahrens DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); 3455d7b4d43SMatthew Ahrens DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); 346eb633035STom Caputi DDK_SET_CRYPT(&drrw->drr_key, BP_IS_PROTECTED(bp)); 3475d7b4d43SMatthew Ahrens drrw->drr_key.ddk_cksum = bp->blk_cksum; 3485d7b4d43SMatthew Ahrens } 349efb80947Sahrens 3505602294fSDan Kimmel if (dump_record(dsp, data, payload_size) != 0) 351be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 352efb80947Sahrens return (0); 353efb80947Sahrens } 354efb80947Sahrens 3555d7b4d43SMatthew Ahrens static int 3565d7b4d43SMatthew Ahrens dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 3575d7b4d43SMatthew Ahrens int blksz, const blkptr_t *bp) 3585d7b4d43SMatthew Ahrens { 3595d7b4d43SMatthew Ahrens char buf[BPE_PAYLOAD_SIZE]; 3605d7b4d43SMatthew Ahrens struct drr_write_embedded *drrw = 3615d7b4d43SMatthew Ahrens &(dsp->dsa_drr->drr_u.drr_write_embedded); 3625d7b4d43SMatthew Ahrens 3635d7b4d43SMatthew Ahrens if (dsp->dsa_pending_op != PENDING_NONE) { 36498110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 3655d7b4d43SMatthew Ahrens return (EINTR); 3665d7b4d43SMatthew Ahrens dsp->dsa_pending_op = PENDING_NONE; 3675d7b4d43SMatthew Ahrens } 3685d7b4d43SMatthew Ahrens 3695d7b4d43SMatthew Ahrens ASSERT(BP_IS_EMBEDDED(bp)); 3705d7b4d43SMatthew Ahrens 3715d7b4d43SMatthew Ahrens bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 3725d7b4d43SMatthew Ahrens dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; 3735d7b4d43SMatthew Ahrens drrw->drr_object = object; 3745d7b4d43SMatthew Ahrens drrw->drr_offset = offset; 3755d7b4d43SMatthew Ahrens drrw->drr_length = blksz; 3765d7b4d43SMatthew Ahrens drrw->drr_toguid = dsp->dsa_toguid; 3775d7b4d43SMatthew Ahrens drrw->drr_compression = BP_GET_COMPRESS(bp); 3785d7b4d43SMatthew Ahrens drrw->drr_etype = BPE_GET_ETYPE(bp); 3795d7b4d43SMatthew Ahrens drrw->drr_lsize = BPE_GET_LSIZE(bp); 3805d7b4d43SMatthew Ahrens drrw->drr_psize = BPE_GET_PSIZE(bp); 3815d7b4d43SMatthew Ahrens 3825d7b4d43SMatthew Ahrens decode_embedded_bp_compressed(bp, buf); 3835d7b4d43SMatthew Ahrens 38498110f08SMatthew Ahrens if (dump_record(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) 3855d7b4d43SMatthew Ahrens return (EINTR); 3865d7b4d43SMatthew Ahrens return (0); 3875d7b4d43SMatthew Ahrens } 3885d7b4d43SMatthew Ahrens 3890a586ceaSMark Shellenbaum static int 390eb633035STom Caputi dump_spill(dmu_sendarg_t *dsp, const blkptr_t *bp, uint64_t object, void *data) 3910a586ceaSMark Shellenbaum { 3924e3c9f44SBill Pijewski struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); 393eb633035STom Caputi uint64_t blksz = BP_GET_LSIZE(bp); 394eb633035STom Caputi uint64_t payload_size = blksz; 3950a586ceaSMark Shellenbaum 3964e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 39798110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 398be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3994e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4000a586ceaSMark Shellenbaum } 4010a586ceaSMark Shellenbaum 4020a586ceaSMark Shellenbaum /* write a SPILL record */ 4034e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 4044e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_SPILL; 4050a586ceaSMark Shellenbaum drrs->drr_object = object; 4060a586ceaSMark Shellenbaum drrs->drr_length = blksz; 4074e3c9f44SBill Pijewski drrs->drr_toguid = dsp->dsa_toguid; 4080a586ceaSMark Shellenbaum 409eb633035STom Caputi /* See comment in dump_dnode() for full details */ 410eb633035STom Caputi if (zfs_send_unmodified_spill_blocks && 411eb633035STom Caputi (bp->blk_birth <= dsp->dsa_fromtxg)) { 412eb633035STom Caputi drrs->drr_flags |= DRR_SPILL_UNMODIFIED; 413eb633035STom Caputi } 414eb633035STom Caputi 415eb633035STom Caputi /* handle raw send fields */ 416eb633035STom Caputi if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) { 417eb633035STom Caputi ASSERT(BP_IS_PROTECTED(bp)); 418eb633035STom Caputi 419eb633035STom Caputi if (BP_SHOULD_BYTESWAP(bp)) 420eb633035STom Caputi drrs->drr_flags |= DRR_RAW_BYTESWAP; 421eb633035STom Caputi drrs->drr_compressiontype = BP_GET_COMPRESS(bp); 422eb633035STom Caputi drrs->drr_compressed_size = BP_GET_PSIZE(bp); 423eb633035STom Caputi zio_crypt_decode_params_bp(bp, drrs->drr_salt, drrs->drr_iv); 424eb633035STom Caputi zio_crypt_decode_mac_bp(bp, drrs->drr_mac); 425eb633035STom Caputi payload_size = drrs->drr_compressed_size; 426eb633035STom Caputi } 427eb633035STom Caputi 428eb633035STom Caputi if (dump_record(dsp, data, payload_size) != 0) 429be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4300a586ceaSMark Shellenbaum return (0); 4310a586ceaSMark Shellenbaum } 4320a586ceaSMark Shellenbaum 433efb80947Sahrens static int 4344e3c9f44SBill Pijewski dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) 435efb80947Sahrens { 4364e3c9f44SBill Pijewski struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); 4379e69d7d0SLori Alt 4389e69d7d0SLori Alt /* 4399e69d7d0SLori Alt * If there is a pending op, but it's not PENDING_FREEOBJECTS, 4409e69d7d0SLori Alt * push it out, since free block aggregation can only be done for 4419e69d7d0SLori Alt * blocks of the same type (i.e., DRR_FREE records can only be 4429e69d7d0SLori Alt * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records 4439e69d7d0SLori Alt * can only be aggregated with other DRR_FREEOBJECTS records. 4449e69d7d0SLori Alt */ 4454e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE && 4464e3c9f44SBill Pijewski dsp->dsa_pending_op != PENDING_FREEOBJECTS) { 44798110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 448be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4494e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4509e69d7d0SLori Alt } 4514e3c9f44SBill Pijewski if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { 4529e69d7d0SLori Alt /* 4539e69d7d0SLori Alt * See whether this free object array can be aggregated 4549e69d7d0SLori Alt * with pending one 4559e69d7d0SLori Alt */ 4569e69d7d0SLori Alt if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { 4579e69d7d0SLori Alt drrfo->drr_numobjs += numobjs; 4589e69d7d0SLori Alt return (0); 4599e69d7d0SLori Alt } else { 4609e69d7d0SLori Alt /* can't be aggregated. Push out pending record */ 46198110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 462be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4634e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4649e69d7d0SLori Alt } 4659e69d7d0SLori Alt } 4669e69d7d0SLori Alt 467efb80947Sahrens /* write a FREEOBJECTS record */ 4684e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 4694e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; 4709e69d7d0SLori Alt drrfo->drr_firstobj = firstobj; 4719e69d7d0SLori Alt drrfo->drr_numobjs = numobjs; 4724e3c9f44SBill Pijewski drrfo->drr_toguid = dsp->dsa_toguid; 4739e69d7d0SLori Alt 4744e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_FREEOBJECTS; 475efb80947Sahrens 476efb80947Sahrens return (0); 477efb80947Sahrens } 478efb80947Sahrens 479efb80947Sahrens static int 480eb633035STom Caputi dump_dnode(dmu_sendarg_t *dsp, const blkptr_t *bp, uint64_t object, 481eb633035STom Caputi dnode_phys_t *dnp) 482efb80947Sahrens { 4834e3c9f44SBill Pijewski struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); 484eb633035STom Caputi int bonuslen; 4859e69d7d0SLori Alt 4869c3fd121SMatthew Ahrens if (object < dsp->dsa_resume_object) { 4879c3fd121SMatthew Ahrens /* 4889c3fd121SMatthew Ahrens * Note: when resuming, we will visit all the dnodes in 4899c3fd121SMatthew Ahrens * the block of dnodes that we are resuming from. In 4909c3fd121SMatthew Ahrens * this case it's unnecessary to send the dnodes prior to 4919c3fd121SMatthew Ahrens * the one we are resuming from. We should be at most one 4929c3fd121SMatthew Ahrens * block's worth of dnodes behind the resume point. 4939c3fd121SMatthew Ahrens */ 4949c3fd121SMatthew Ahrens ASSERT3U(dsp->dsa_resume_object - object, <, 4959c3fd121SMatthew Ahrens 1 << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)); 4969c3fd121SMatthew Ahrens return (0); 4979c3fd121SMatthew Ahrens } 4989c3fd121SMatthew Ahrens 499efb80947Sahrens if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) 5004e3c9f44SBill Pijewski return (dump_freeobjects(dsp, object, 1)); 501efb80947Sahrens 5024e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 50398110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 504be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 5054e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 5069e69d7d0SLori Alt } 5079e69d7d0SLori Alt 508efb80947Sahrens /* write an OBJECT record */ 5094e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 5104e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_OBJECT; 5119e69d7d0SLori Alt drro->drr_object = object; 5129e69d7d0SLori Alt drro->drr_type = dnp->dn_type; 5139e69d7d0SLori Alt drro->drr_bonustype = dnp->dn_bonustype; 5149e69d7d0SLori Alt drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 5159e69d7d0SLori Alt drro->drr_bonuslen = dnp->dn_bonuslen; 51654811da5SToomas Soome drro->drr_dn_slots = dnp->dn_extra_slots + 1; 5179e69d7d0SLori Alt drro->drr_checksumtype = dnp->dn_checksum; 5189e69d7d0SLori Alt drro->drr_compress = dnp->dn_compress; 5194e3c9f44SBill Pijewski drro->drr_toguid = dsp->dsa_toguid; 5209e69d7d0SLori Alt 521b5152584SMatthew Ahrens if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 522b5152584SMatthew Ahrens drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) 523b5152584SMatthew Ahrens drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; 524b5152584SMatthew Ahrens 525eb633035STom Caputi bonuslen = P2ROUNDUP(dnp->dn_bonuslen, 8); 526eb633035STom Caputi 527eb633035STom Caputi if ((dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW)) { 528eb633035STom Caputi ASSERT(BP_IS_ENCRYPTED(bp)); 529eb633035STom Caputi 530eb633035STom Caputi if (BP_SHOULD_BYTESWAP(bp)) 531eb633035STom Caputi drro->drr_flags |= DRR_RAW_BYTESWAP; 532eb633035STom Caputi 533eb633035STom Caputi /* needed for reconstructing dnp on recv side */ 534eb633035STom Caputi drro->drr_maxblkid = dnp->dn_maxblkid; 535eb633035STom Caputi drro->drr_indblkshift = dnp->dn_indblkshift; 536eb633035STom Caputi drro->drr_nlevels = dnp->dn_nlevels; 537eb633035STom Caputi drro->drr_nblkptr = dnp->dn_nblkptr; 538eb633035STom Caputi 539eb633035STom Caputi /* 540eb633035STom Caputi * Since we encrypt the entire bonus area, the (raw) part 541eb633035STom Caputi * beyond the bonuslen is actually nonzero, so we need 542eb633035STom Caputi * to send it. 543eb633035STom Caputi */ 544eb633035STom Caputi if (bonuslen != 0) { 545eb633035STom Caputi drro->drr_raw_bonuslen = DN_MAX_BONUS_LEN(dnp); 546eb633035STom Caputi bonuslen = drro->drr_raw_bonuslen; 547eb633035STom Caputi } 54898110f08SMatthew Ahrens } 549efb80947Sahrens 550eb633035STom Caputi /* 551eb633035STom Caputi * DRR_OBJECT_SPILL is set for every dnode which references a 552eb633035STom Caputi * spill block. This allows the receiving pool to definitively 553eb633035STom Caputi * determine when a spill block should be kept or freed. 554eb633035STom Caputi */ 555eb633035STom Caputi if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) 556eb633035STom Caputi drro->drr_flags |= DRR_OBJECT_SPILL; 557eb633035STom Caputi 558eb633035STom Caputi if (dump_record(dsp, DN_BONUS(dnp), bonuslen) != 0) 559eb633035STom Caputi return (SET_ERROR(EINTR)); 560eb633035STom Caputi 5612f3d8780SMatthew Ahrens /* Free anything past the end of the file. */ 5624e3c9f44SBill Pijewski if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * 563eb633035STom Caputi (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), DMU_OBJECT_END) != 0) 564be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 565eb633035STom Caputi 566eb633035STom Caputi /* 567eb633035STom Caputi * Send DRR_SPILL records for unmodified spill blocks. This is useful 568eb633035STom Caputi * because changing certain attributes of the object (e.g. blocksize) 569eb633035STom Caputi * can cause old versions of ZFS to incorrectly remove a spill block. 570eb633035STom Caputi * Including these records in the stream forces an up to date version 571eb633035STom Caputi * to always be written ensuring they're never lost. Current versions 572eb633035STom Caputi * of the code which understand the DRR_FLAG_SPILL_BLOCK feature can 573eb633035STom Caputi * ignore these unmodified spill blocks. 574eb633035STom Caputi */ 575eb633035STom Caputi if (zfs_send_unmodified_spill_blocks && 576eb633035STom Caputi (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) && 577eb633035STom Caputi (DN_SPILL_BLKPTR(dnp)->blk_birth <= dsp->dsa_fromtxg)) { 578eb633035STom Caputi struct send_block_record record; 579eb633035STom Caputi 580eb633035STom Caputi bzero(&record, sizeof (struct send_block_record)); 581eb633035STom Caputi record.eos_marker = B_FALSE; 582eb633035STom Caputi record.bp = *DN_SPILL_BLKPTR(dnp); 583eb633035STom Caputi SET_BOOKMARK(&(record.zb), dmu_objset_id(dsp->dsa_os), 584eb633035STom Caputi object, 0, DMU_SPILL_BLKID); 585eb633035STom Caputi 586eb633035STom Caputi if (do_dump(dsp, &record) != 0) 587eb633035STom Caputi return (SET_ERROR(EINTR)); 588eb633035STom Caputi } 589eb633035STom Caputi 5903b2aab18SMatthew Ahrens if (dsp->dsa_err != 0) 591be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 592efb80947Sahrens return (0); 593efb80947Sahrens } 594efb80947Sahrens 595eb633035STom Caputi static int 596eb633035STom Caputi dump_object_range(dmu_sendarg_t *dsp, const blkptr_t *bp, uint64_t firstobj, 597eb633035STom Caputi uint64_t numslots) 598eb633035STom Caputi { 599eb633035STom Caputi struct drr_object_range *drror = 600eb633035STom Caputi &(dsp->dsa_drr->drr_u.drr_object_range); 601eb633035STom Caputi 602eb633035STom Caputi /* we only use this record type for raw sends */ 603eb633035STom Caputi ASSERT(BP_IS_PROTECTED(bp)); 604eb633035STom Caputi ASSERT(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_RAW); 605eb633035STom Caputi ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); 606eb633035STom Caputi ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_DNODE); 607eb633035STom Caputi ASSERT0(BP_GET_LEVEL(bp)); 608eb633035STom Caputi 609eb633035STom Caputi if (dsp->dsa_pending_op != PENDING_NONE) { 610eb633035STom Caputi if (dump_record(dsp, NULL, 0) != 0) 611eb633035STom Caputi return (SET_ERROR(EINTR)); 612eb633035STom Caputi dsp->dsa_pending_op = PENDING_NONE; 613eb633035STom Caputi } 614eb633035STom Caputi 615eb633035STom Caputi bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 616eb633035STom Caputi dsp->dsa_drr->drr_type = DRR_OBJECT_RANGE; 617eb633035STom Caputi drror->drr_firstobj = firstobj; 618eb633035STom Caputi drror->drr_numslots = numslots; 619eb633035STom Caputi drror->drr_toguid = dsp->dsa_toguid; 620eb633035STom Caputi if (BP_SHOULD_BYTESWAP(bp)) 621eb633035STom Caputi drror->drr_flags |= DRR_RAW_BYTESWAP; 622eb633035STom Caputi zio_crypt_decode_params_bp(bp, drror->drr_salt, drror->drr_iv); 623eb633035STom Caputi zio_crypt_decode_mac_bp(bp, drror->drr_mac); 624eb633035STom Caputi 625eb633035STom Caputi if (dump_record(dsp, NULL, 0) != 0) 626eb633035STom Caputi return (SET_ERROR(EINTR)); 627eb633035STom Caputi return (0); 628eb633035STom Caputi } 629eb633035STom Caputi 6305d7b4d43SMatthew Ahrens static boolean_t 6315d7b4d43SMatthew Ahrens backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) 6325d7b4d43SMatthew Ahrens { 6335d7b4d43SMatthew Ahrens if (!BP_IS_EMBEDDED(bp)) 6345d7b4d43SMatthew Ahrens return (B_FALSE); 6355d7b4d43SMatthew Ahrens 6365d7b4d43SMatthew Ahrens /* 6375d7b4d43SMatthew Ahrens * Compression function must be legacy, or explicitly enabled. 6385d7b4d43SMatthew Ahrens */ 6395d7b4d43SMatthew Ahrens if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && 6405602294fSDan Kimmel !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LZ4))) 6415d7b4d43SMatthew Ahrens return (B_FALSE); 6425d7b4d43SMatthew Ahrens 6435d7b4d43SMatthew Ahrens /* 6445d7b4d43SMatthew Ahrens * Embed type must be explicitly enabled. 6455d7b4d43SMatthew Ahrens */ 6465d7b4d43SMatthew Ahrens switch (BPE_GET_ETYPE(bp)) { 6475d7b4d43SMatthew Ahrens case BP_EMBEDDED_TYPE_DATA: 6485d7b4d43SMatthew Ahrens if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) 6495d7b4d43SMatthew Ahrens return (B_TRUE); 6505d7b4d43SMatthew Ahrens break; 6515d7b4d43SMatthew Ahrens default: 6525d7b4d43SMatthew Ahrens return (B_FALSE); 6535d7b4d43SMatthew Ahrens } 6545d7b4d43SMatthew Ahrens return (B_FALSE); 6555d7b4d43SMatthew Ahrens } 6565d7b4d43SMatthew Ahrens 657a2cdcdd2SPaul Dagnelie /* 658a2cdcdd2SPaul Dagnelie * This is the callback function to traverse_dataset that acts as the worker 659a2cdcdd2SPaul Dagnelie * thread for dmu_send_impl. 660a2cdcdd2SPaul Dagnelie */ 661a2cdcdd2SPaul Dagnelie /*ARGSUSED*/ 662a2cdcdd2SPaul Dagnelie static int 663a2cdcdd2SPaul Dagnelie send_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 664a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb, const struct dnode_phys *dnp, void *arg) 665a2cdcdd2SPaul Dagnelie { 666a2cdcdd2SPaul Dagnelie struct send_thread_arg *sta = arg; 667a2cdcdd2SPaul Dagnelie struct send_block_record *record; 668a2cdcdd2SPaul Dagnelie uint64_t record_size; 669a2cdcdd2SPaul Dagnelie int err = 0; 670efb80947Sahrens 6719c3fd121SMatthew Ahrens ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || 6729c3fd121SMatthew Ahrens zb->zb_object >= sta->resume.zb_object); 673eb633035STom Caputi ASSERT3P(sta->ds, !=, NULL); 6749c3fd121SMatthew Ahrens 675a2cdcdd2SPaul Dagnelie if (sta->cancel) 676a2cdcdd2SPaul Dagnelie return (SET_ERROR(EINTR)); 677a2cdcdd2SPaul Dagnelie 678a2cdcdd2SPaul Dagnelie if (bp == NULL) { 679a2cdcdd2SPaul Dagnelie ASSERT3U(zb->zb_level, ==, ZB_DNODE_LEVEL); 680a2cdcdd2SPaul Dagnelie return (0); 681a2cdcdd2SPaul Dagnelie } else if (zb->zb_level < 0) { 682a2cdcdd2SPaul Dagnelie return (0); 683a2cdcdd2SPaul Dagnelie } 684a2cdcdd2SPaul Dagnelie 685a2cdcdd2SPaul Dagnelie record = kmem_zalloc(sizeof (struct send_block_record), KM_SLEEP); 686a2cdcdd2SPaul Dagnelie record->eos_marker = B_FALSE; 687a2cdcdd2SPaul Dagnelie record->bp = *bp; 688a2cdcdd2SPaul Dagnelie record->zb = *zb; 689a2cdcdd2SPaul Dagnelie record->indblkshift = dnp->dn_indblkshift; 690a2cdcdd2SPaul Dagnelie record->datablkszsec = dnp->dn_datablkszsec; 691a2cdcdd2SPaul Dagnelie record_size = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 692a2cdcdd2SPaul Dagnelie bqueue_enqueue(&sta->q, record, record_size); 693a2cdcdd2SPaul Dagnelie 694a2cdcdd2SPaul Dagnelie return (err); 695a2cdcdd2SPaul Dagnelie } 696a2cdcdd2SPaul Dagnelie 697a2cdcdd2SPaul Dagnelie /* 698a2cdcdd2SPaul Dagnelie * This function kicks off the traverse_dataset. It also handles setting the 699a2cdcdd2SPaul Dagnelie * error code of the thread in case something goes wrong, and pushes the End of 700a2cdcdd2SPaul Dagnelie * Stream record when the traverse_dataset call has finished. If there is no 701a2cdcdd2SPaul Dagnelie * dataset to traverse, the thread immediately pushes End of Stream marker. 702a2cdcdd2SPaul Dagnelie */ 703a2cdcdd2SPaul Dagnelie static void 704a2cdcdd2SPaul Dagnelie send_traverse_thread(void *arg) 705a2cdcdd2SPaul Dagnelie { 706a2cdcdd2SPaul Dagnelie struct send_thread_arg *st_arg = arg; 707a2cdcdd2SPaul Dagnelie int err; 708a2cdcdd2SPaul Dagnelie struct send_block_record *data; 709a2cdcdd2SPaul Dagnelie 710a2cdcdd2SPaul Dagnelie if (st_arg->ds != NULL) { 7119c3fd121SMatthew Ahrens err = traverse_dataset_resume(st_arg->ds, 7129c3fd121SMatthew Ahrens st_arg->fromtxg, &st_arg->resume, 7139c3fd121SMatthew Ahrens st_arg->flags, send_cb, st_arg); 7149c3fd121SMatthew Ahrens 715a2cdcdd2SPaul Dagnelie if (err != EINTR) 716a2cdcdd2SPaul Dagnelie st_arg->error_code = err; 717a2cdcdd2SPaul Dagnelie } 718a2cdcdd2SPaul Dagnelie data = kmem_zalloc(sizeof (*data), KM_SLEEP); 719a2cdcdd2SPaul Dagnelie data->eos_marker = B_TRUE; 720a2cdcdd2SPaul Dagnelie bqueue_enqueue(&st_arg->q, data, 1); 721f2c1e9bcSJorgen Lundman thread_exit(); 722a2cdcdd2SPaul Dagnelie } 723a2cdcdd2SPaul Dagnelie 724a2cdcdd2SPaul Dagnelie /* 725a2cdcdd2SPaul Dagnelie * This function actually handles figuring out what kind of record needs to be 726a2cdcdd2SPaul Dagnelie * dumped, reading the data (which has hopefully been prefetched), and calling 727a2cdcdd2SPaul Dagnelie * the appropriate helper function. 728a2cdcdd2SPaul Dagnelie */ 729efb80947Sahrens static int 730a2cdcdd2SPaul Dagnelie do_dump(dmu_sendarg_t *dsa, struct send_block_record *data) 731efb80947Sahrens { 732a2cdcdd2SPaul Dagnelie dsl_dataset_t *ds = dmu_objset_ds(dsa->dsa_os); 733a2cdcdd2SPaul Dagnelie const blkptr_t *bp = &data->bp; 734a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb = &data->zb; 735a2cdcdd2SPaul Dagnelie uint8_t indblkshift = data->indblkshift; 736a2cdcdd2SPaul Dagnelie uint16_t dblkszsec = data->datablkszsec; 737a2cdcdd2SPaul Dagnelie spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 738efb80947Sahrens dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; 739efb80947Sahrens int err = 0; 740efb80947Sahrens 741a2cdcdd2SPaul Dagnelie ASSERT3U(zb->zb_level, >=, 0); 742efb80947Sahrens 7439c3fd121SMatthew Ahrens ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || 7449c3fd121SMatthew Ahrens zb->zb_object >= dsa->dsa_resume_object); 7459c3fd121SMatthew Ahrens 746eb633035STom Caputi /* 747eb633035STom Caputi * All bps of an encrypted os should have the encryption bit set. 748eb633035STom Caputi * If this is not true it indicates tampering and we report an error. 749eb633035STom Caputi */ 750eb633035STom Caputi if (dsa->dsa_os->os_encrypted && 751eb633035STom Caputi !BP_IS_HOLE(bp) && !BP_USES_CRYPT(bp)) { 752eb633035STom Caputi spa_log_error(spa, zb); 753eb633035STom Caputi zfs_panic_recover("unencrypted block in encrypted " 754eb633035STom Caputi "object set %llu", ds->ds_object); 755eb633035STom Caputi return (SET_ERROR(EIO)); 756eb633035STom Caputi } 757eb633035STom Caputi 758b24ab676SJeff Bonwick if (zb->zb_object != DMU_META_DNODE_OBJECT && 759b24ab676SJeff Bonwick DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { 76014843421SMatthew Ahrens return (0); 76143466aaeSMax Grossman } else if (BP_IS_HOLE(bp) && 76243466aaeSMax Grossman zb->zb_object == DMU_META_DNODE_OBJECT) { 763a2cdcdd2SPaul Dagnelie uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); 76488b7b0f2SMatthew Ahrens uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; 765a2cdcdd2SPaul Dagnelie err = dump_freeobjects(dsa, dnobj, span >> DNODE_SHIFT); 76643466aaeSMax Grossman } else if (BP_IS_HOLE(bp)) { 767a2cdcdd2SPaul Dagnelie uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); 768a2cdcdd2SPaul Dagnelie uint64_t offset = zb->zb_blkid * span; 769eb633035STom Caputi /* Don't dump free records for offsets > DMU_OBJECT_END */ 770eb633035STom Caputi if (zb->zb_blkid == 0 || span <= DMU_OBJECT_END / zb->zb_blkid) 771eb633035STom Caputi err = dump_free(dsa, zb->zb_object, offset, span); 77288b7b0f2SMatthew Ahrens } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { 77388b7b0f2SMatthew Ahrens return (0); 77488b7b0f2SMatthew Ahrens } else if (type == DMU_OT_DNODE) { 77554811da5SToomas Soome int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; 7767adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 77788b7b0f2SMatthew Ahrens arc_buf_t *abuf; 778eb633035STom Caputi enum zio_flag zioflags = ZIO_FLAG_CANFAIL; 779eb633035STom Caputi 780eb633035STom Caputi if (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) { 781eb633035STom Caputi ASSERT(BP_IS_ENCRYPTED(bp)); 782eb633035STom Caputi ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); 783eb633035STom Caputi zioflags |= ZIO_FLAG_RAW; 784eb633035STom Caputi } 785efb80947Sahrens 786a2cdcdd2SPaul Dagnelie ASSERT0(zb->zb_level); 787a2cdcdd2SPaul Dagnelie 7881b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 789eb633035STom Caputi ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) 790be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 79188b7b0f2SMatthew Ahrens 792a2cdcdd2SPaul Dagnelie dnode_phys_t *blk = abuf->b_data; 79354811da5SToomas Soome uint64_t dnobj = zb->zb_blkid * epb; 794eb633035STom Caputi 795eb633035STom Caputi /* 796eb633035STom Caputi * Raw sends require sending encryption parameters for the 797eb633035STom Caputi * block of dnodes. Regular sends do not need to send this 798eb633035STom Caputi * info. 799eb633035STom Caputi */ 800eb633035STom Caputi if (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) { 801eb633035STom Caputi ASSERT(arc_is_encrypted(abuf)); 802eb633035STom Caputi err = dump_object_range(dsa, bp, dnobj, epb); 803eb633035STom Caputi } 804eb633035STom Caputi 805eb633035STom Caputi if (err == 0) { 806eb633035STom Caputi for (int i = 0; i < epb; 807eb633035STom Caputi i += blk[i].dn_extra_slots + 1) { 808eb633035STom Caputi err = dump_dnode(dsa, bp, dnobj + i, blk + i); 809eb633035STom Caputi if (err != 0) 810eb633035STom Caputi break; 811eb633035STom Caputi } 812efb80947Sahrens } 813dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 8140a586ceaSMark Shellenbaum } else if (type == DMU_OT_SA) { 8157adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 8160a586ceaSMark Shellenbaum arc_buf_t *abuf; 817eb633035STom Caputi enum zio_flag zioflags = ZIO_FLAG_CANFAIL; 818eb633035STom Caputi 819eb633035STom Caputi if (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) { 820eb633035STom Caputi ASSERT(BP_IS_PROTECTED(bp)); 821eb633035STom Caputi zioflags |= ZIO_FLAG_RAW; 822eb633035STom Caputi } 8230a586ceaSMark Shellenbaum 8241b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 825eb633035STom Caputi ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) 826be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 8270a586ceaSMark Shellenbaum 828eb633035STom Caputi err = dump_spill(dsa, bp, zb->zb_object, abuf->b_data); 829dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 830a2cdcdd2SPaul Dagnelie } else if (backup_do_embed(dsa, bp)) { 8315d7b4d43SMatthew Ahrens /* it's an embedded level-0 block of a regular object */ 832a2cdcdd2SPaul Dagnelie int blksz = dblkszsec << SPA_MINBLOCKSHIFT; 833a2cdcdd2SPaul Dagnelie ASSERT0(zb->zb_level); 834a2cdcdd2SPaul Dagnelie err = dump_write_embedded(dsa, zb->zb_object, 8355d7b4d43SMatthew Ahrens zb->zb_blkid * blksz, blksz, bp); 836a2cdcdd2SPaul Dagnelie } else { 837a2cdcdd2SPaul Dagnelie /* it's a level-0 block of a regular object */ 8387adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 83988b7b0f2SMatthew Ahrens arc_buf_t *abuf; 840a2cdcdd2SPaul Dagnelie int blksz = dblkszsec << SPA_MINBLOCKSHIFT; 841b5152584SMatthew Ahrens uint64_t offset; 84288b7b0f2SMatthew Ahrens 8435602294fSDan Kimmel /* 8445602294fSDan Kimmel * If we have large blocks stored on disk but the send flags 8455602294fSDan Kimmel * don't allow us to send large blocks, we split the data from 8465602294fSDan Kimmel * the arc buf into chunks. 8475602294fSDan Kimmel */ 8485602294fSDan Kimmel boolean_t split_large_blocks = blksz > SPA_OLD_MAXBLOCKSIZE && 8495602294fSDan Kimmel !(dsa->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS); 850eb633035STom Caputi 851eb633035STom Caputi /* 852eb633035STom Caputi * Raw sends require that we always get raw data as it exists 853eb633035STom Caputi * on disk, so we assert that we are not splitting blocks here. 854eb633035STom Caputi */ 855eb633035STom Caputi boolean_t request_raw = 856eb633035STom Caputi (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_RAW) != 0; 857eb633035STom Caputi 8585602294fSDan Kimmel /* 8595602294fSDan Kimmel * We should only request compressed data from the ARC if all 8605602294fSDan Kimmel * the following are true: 8615602294fSDan Kimmel * - stream compression was requested 8625602294fSDan Kimmel * - we aren't splitting large blocks into smaller chunks 8635602294fSDan Kimmel * - the data won't need to be byteswapped before sending 8645602294fSDan Kimmel * - this isn't an embedded block 8655602294fSDan Kimmel * - this isn't metadata (if receiving on a different endian 8665602294fSDan Kimmel * system it can be byteswapped more easily) 8675602294fSDan Kimmel */ 8685602294fSDan Kimmel boolean_t request_compressed = 8695602294fSDan Kimmel (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED) && 8705602294fSDan Kimmel !split_large_blocks && !BP_SHOULD_BYTESWAP(bp) && 8715602294fSDan Kimmel !BP_IS_EMBEDDED(bp) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp)); 8725602294fSDan Kimmel 873eb633035STom Caputi IMPLY(request_raw, !split_large_blocks); 874eb633035STom Caputi IMPLY(request_raw, BP_IS_PROTECTED(bp)); 8755602294fSDan Kimmel ASSERT0(zb->zb_level); 8765602294fSDan Kimmel ASSERT(zb->zb_object > dsa->dsa_resume_object || 8775602294fSDan Kimmel (zb->zb_object == dsa->dsa_resume_object && 8785602294fSDan Kimmel zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); 8795602294fSDan Kimmel 88078f17100SMatthew Ahrens ASSERT0(zb->zb_level); 8819c3fd121SMatthew Ahrens ASSERT(zb->zb_object > dsa->dsa_resume_object || 8829c3fd121SMatthew Ahrens (zb->zb_object == dsa->dsa_resume_object && 8839c3fd121SMatthew Ahrens zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); 8849c3fd121SMatthew Ahrens 8855602294fSDan Kimmel ASSERT3U(blksz, ==, BP_GET_LSIZE(bp)); 8865602294fSDan Kimmel 8875602294fSDan Kimmel enum zio_flag zioflags = ZIO_FLAG_CANFAIL; 888eb633035STom Caputi if (request_raw) 8895602294fSDan Kimmel zioflags |= ZIO_FLAG_RAW; 890eb633035STom Caputi else if (request_compressed) 891eb633035STom Caputi zioflags |= ZIO_FLAG_RAW_COMPRESS; 892eb633035STom Caputi 8931b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 8945602294fSDan Kimmel ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) { 89519b94df9SMatthew Ahrens if (zfs_send_corrupt_data) { 89619b94df9SMatthew Ahrens /* Send a block filled with 0x"zfs badd bloc" */ 8975602294fSDan Kimmel abuf = arc_alloc_buf(spa, &abuf, ARC_BUFC_DATA, 8985602294fSDan Kimmel blksz); 89919b94df9SMatthew Ahrens uint64_t *ptr; 90019b94df9SMatthew Ahrens for (ptr = abuf->b_data; 90119b94df9SMatthew Ahrens (char *)ptr < (char *)abuf->b_data + blksz; 90219b94df9SMatthew Ahrens ptr++) 9038c76e076SBrian Behlendorf *ptr = 0x2f5baddb10cULL; 90419b94df9SMatthew Ahrens } else { 905be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 90619b94df9SMatthew Ahrens } 90719b94df9SMatthew Ahrens } 90888b7b0f2SMatthew Ahrens 909b5152584SMatthew Ahrens offset = zb->zb_blkid * blksz; 910b5152584SMatthew Ahrens 9115602294fSDan Kimmel if (split_large_blocks) { 912eb633035STom Caputi ASSERT0(arc_is_encrypted(abuf)); 9135602294fSDan Kimmel ASSERT3U(arc_get_compression(abuf), ==, 9145602294fSDan Kimmel ZIO_COMPRESS_OFF); 915b5152584SMatthew Ahrens char *buf = abuf->b_data; 916b5152584SMatthew Ahrens while (blksz > 0 && err == 0) { 917b5152584SMatthew Ahrens int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); 918a2cdcdd2SPaul Dagnelie err = dump_write(dsa, type, zb->zb_object, 9195602294fSDan Kimmel offset, n, n, NULL, buf); 920b5152584SMatthew Ahrens offset += n; 921b5152584SMatthew Ahrens buf += n; 922b5152584SMatthew Ahrens blksz -= n; 923b5152584SMatthew Ahrens } 924b5152584SMatthew Ahrens } else { 9255602294fSDan Kimmel err = dump_write(dsa, type, zb->zb_object, offset, 9265602294fSDan Kimmel blksz, arc_buf_size(abuf), bp, abuf->b_data); 927b5152584SMatthew Ahrens } 928dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 929efb80947Sahrens } 930efb80947Sahrens 931efb80947Sahrens ASSERT(err == 0 || err == EINTR); 932efb80947Sahrens return (err); 933efb80947Sahrens } 934efb80947Sahrens 9354445fffbSMatthew Ahrens /* 936a2cdcdd2SPaul Dagnelie * Pop the new data off the queue, and free the old data. 937a2cdcdd2SPaul Dagnelie */ 938a2cdcdd2SPaul Dagnelie static struct send_block_record * 939a2cdcdd2SPaul Dagnelie get_next_record(bqueue_t *bq, struct send_block_record *data) 940a2cdcdd2SPaul Dagnelie { 941a2cdcdd2SPaul Dagnelie struct send_block_record *tmp = bqueue_dequeue(bq); 942a2cdcdd2SPaul Dagnelie kmem_free(data, sizeof (*data)); 943a2cdcdd2SPaul Dagnelie return (tmp); 944a2cdcdd2SPaul Dagnelie } 945a2cdcdd2SPaul Dagnelie 946a2cdcdd2SPaul Dagnelie /* 947a2cdcdd2SPaul Dagnelie * Actually do the bulk of the work in a zfs send. 948a2cdcdd2SPaul Dagnelie * 949a2cdcdd2SPaul Dagnelie * Note: Releases dp using the specified tag. 9504445fffbSMatthew Ahrens */ 9513b2aab18SMatthew Ahrens static int 952a2cdcdd2SPaul Dagnelie dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *to_ds, 9535602294fSDan Kimmel zfs_bookmark_phys_t *ancestor_zb, boolean_t is_clone, 9545602294fSDan Kimmel boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, 955eb633035STom Caputi boolean_t rawok, int outfd, uint64_t resumeobj, uint64_t resumeoff, 9569c3fd121SMatthew Ahrens vnode_t *vp, offset_t *off) 957efb80947Sahrens { 9583b2aab18SMatthew Ahrens objset_t *os; 959efb80947Sahrens dmu_replay_record_t *drr; 9604e3c9f44SBill Pijewski dmu_sendarg_t *dsp; 961efb80947Sahrens int err; 9623cb34c60Sahrens uint64_t fromtxg = 0; 9635d7b4d43SMatthew Ahrens uint64_t featureflags = 0; 9649c3fd121SMatthew Ahrens struct send_thread_arg to_arg = { 0 }; 965efb80947Sahrens 966a2cdcdd2SPaul Dagnelie err = dmu_objset_from_ds(to_ds, &os); 9673b2aab18SMatthew Ahrens if (err != 0) { 9683b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 9693b2aab18SMatthew Ahrens return (err); 9703b2aab18SMatthew Ahrens } 971efb80947Sahrens 972eb633035STom Caputi /* 973eb633035STom Caputi * If this is a non-raw send of an encrypted ds, we can ensure that 974eb633035STom Caputi * the objset_phys_t is authenticated. This is safe because this is 975eb633035STom Caputi * either a snapshot or we have owned the dataset, ensuring that 976eb633035STom Caputi * it can't be modified. 977eb633035STom Caputi */ 978eb633035STom Caputi if (!rawok && os->os_encrypted && 979eb633035STom Caputi arc_is_unauthenticated(os->os_phys_buf)) { 980eb633035STom Caputi zbookmark_phys_t zb; 981eb633035STom Caputi 982eb633035STom Caputi SET_BOOKMARK(&zb, to_ds->ds_object, ZB_ROOT_OBJECT, 983eb633035STom Caputi ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 984eb633035STom Caputi err = arc_untransform(os->os_phys_buf, os->os_spa, 985eb633035STom Caputi &zb, B_FALSE); 986eb633035STom Caputi if (err != 0) { 987eb633035STom Caputi dsl_pool_rele(dp, tag); 988eb633035STom Caputi return (err); 989eb633035STom Caputi } 990eb633035STom Caputi 991eb633035STom Caputi ASSERT0(arc_is_unauthenticated(os->os_phys_buf)); 992eb633035STom Caputi } 993eb633035STom Caputi 994efb80947Sahrens drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 995efb80947Sahrens drr->drr_type = DRR_BEGIN; 996efb80947Sahrens drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 9979e69d7d0SLori Alt DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, 9989e69d7d0SLori Alt DMU_SUBSTREAM); 999dc7cd546SMark Shellenbaum 1000dc7cd546SMark Shellenbaum #ifdef _KERNEL 10013b2aab18SMatthew Ahrens if (dmu_objset_type(os) == DMU_OST_ZFS) { 1002dc7cd546SMark Shellenbaum uint64_t version; 10033b2aab18SMatthew Ahrens if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { 10044e3c9f44SBill Pijewski kmem_free(drr, sizeof (dmu_replay_record_t)); 10053b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 1006be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 10074e3c9f44SBill Pijewski } 10083b2aab18SMatthew Ahrens if (version >= ZPL_VERSION_SA) { 10095d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 1010dc7cd546SMark Shellenbaum } 1011dc7cd546SMark Shellenbaum } 1012dc7cd546SMark Shellenbaum #endif 1013dc7cd546SMark Shellenbaum 1014eb633035STom Caputi /* raw sends imply large_block_ok */ 1015eb633035STom Caputi if ((large_block_ok || rawok) && 1016eb633035STom Caputi to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS]) 1017b5152584SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; 101854811da5SToomas Soome if (to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_DNODE]) 101954811da5SToomas Soome featureflags |= DMU_BACKUP_FEATURE_LARGE_DNODE; 1020eb633035STom Caputi 1021eb633035STom Caputi /* encrypted datasets will not have embedded blocks */ 1022eb633035STom Caputi if ((embedok || rawok) && !os->os_encrypted && 10235d7b4d43SMatthew Ahrens spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { 10245d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; 10255602294fSDan Kimmel } 1026eb633035STom Caputi 1027eb633035STom Caputi /* raw send implies compressok */ 1028eb633035STom Caputi if (compressok || rawok) 10295602294fSDan Kimmel featureflags |= DMU_BACKUP_FEATURE_COMPRESSED; 1030eb633035STom Caputi if (rawok && os->os_encrypted) 1031eb633035STom Caputi featureflags |= DMU_BACKUP_FEATURE_RAW; 1032eb633035STom Caputi 10335602294fSDan Kimmel if ((featureflags & 1034eb633035STom Caputi (DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_COMPRESSED | 1035eb633035STom Caputi DMU_BACKUP_FEATURE_RAW)) != 0 && 1036eb633035STom Caputi spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) { 10375602294fSDan Kimmel featureflags |= DMU_BACKUP_FEATURE_LZ4; 10385d7b4d43SMatthew Ahrens } 10395d7b4d43SMatthew Ahrens 10409c3fd121SMatthew Ahrens if (resumeobj != 0 || resumeoff != 0) { 10419c3fd121SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_RESUMING; 10429c3fd121SMatthew Ahrens } 10439c3fd121SMatthew Ahrens 10445d7b4d43SMatthew Ahrens DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, 10455d7b4d43SMatthew Ahrens featureflags); 10465d7b4d43SMatthew Ahrens 1047efb80947Sahrens drr->drr_u.drr_begin.drr_creation_time = 1048a2cdcdd2SPaul Dagnelie dsl_dataset_phys(to_ds)->ds_creation_time; 10493b2aab18SMatthew Ahrens drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); 105078f17100SMatthew Ahrens if (is_clone) 10513cb34c60Sahrens drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; 1052a2cdcdd2SPaul Dagnelie drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(to_ds)->ds_guid; 1053a2cdcdd2SPaul Dagnelie if (dsl_dataset_phys(to_ds)->ds_flags & DS_FLAG_CI_DATASET) 1054ab04eb8eStimh drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; 1055880094b6SAndrew Stormont if (zfs_send_set_freerecords_bit) 1056880094b6SAndrew Stormont drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_FREERECORDS; 1057ab04eb8eStimh 1058eb633035STom Caputi drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_SPILL_BLOCK; 1059eb633035STom Caputi 1060a2cdcdd2SPaul Dagnelie if (ancestor_zb != NULL) { 1061a2cdcdd2SPaul Dagnelie drr->drr_u.drr_begin.drr_fromguid = 1062a2cdcdd2SPaul Dagnelie ancestor_zb->zbm_guid; 1063a2cdcdd2SPaul Dagnelie fromtxg = ancestor_zb->zbm_creation_txg; 106478f17100SMatthew Ahrens } 1065a2cdcdd2SPaul Dagnelie dsl_dataset_name(to_ds, drr->drr_u.drr_begin.drr_toname); 1066a2cdcdd2SPaul Dagnelie if (!to_ds->ds_is_snapshot) { 106778f17100SMatthew Ahrens (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", 106878f17100SMatthew Ahrens sizeof (drr->drr_u.drr_begin.drr_toname)); 10693b2aab18SMatthew Ahrens } 10703cb34c60Sahrens 10714e3c9f44SBill Pijewski dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); 10724e3c9f44SBill Pijewski 10734e3c9f44SBill Pijewski dsp->dsa_drr = drr; 10744e3c9f44SBill Pijewski dsp->dsa_vp = vp; 10754e3c9f44SBill Pijewski dsp->dsa_outfd = outfd; 10764e3c9f44SBill Pijewski dsp->dsa_proc = curproc; 10773b2aab18SMatthew Ahrens dsp->dsa_os = os; 10784e3c9f44SBill Pijewski dsp->dsa_off = off; 1079a2cdcdd2SPaul Dagnelie dsp->dsa_toguid = dsl_dataset_phys(to_ds)->ds_guid; 1080eb633035STom Caputi dsp->dsa_fromtxg = fromtxg; 10814e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 10825d7b4d43SMatthew Ahrens dsp->dsa_featureflags = featureflags; 10839c3fd121SMatthew Ahrens dsp->dsa_resume_object = resumeobj; 10849c3fd121SMatthew Ahrens dsp->dsa_resume_offset = resumeoff; 10854e3c9f44SBill Pijewski 1086a2cdcdd2SPaul Dagnelie mutex_enter(&to_ds->ds_sendstream_lock); 1087a2cdcdd2SPaul Dagnelie list_insert_head(&to_ds->ds_sendstreams, dsp); 1088a2cdcdd2SPaul Dagnelie mutex_exit(&to_ds->ds_sendstream_lock); 10894e3c9f44SBill Pijewski 1090a2cdcdd2SPaul Dagnelie dsl_dataset_long_hold(to_ds, FTAG); 1091de8d9cffSMatthew Ahrens dsl_pool_rele(dp, tag); 1092de8d9cffSMatthew Ahrens 10939c3fd121SMatthew Ahrens void *payload = NULL; 10949c3fd121SMatthew Ahrens size_t payload_len = 0; 1095eb633035STom Caputi /* handle features that require a DRR_BEGIN payload */ 1096eb633035STom Caputi if (featureflags & 1097eb633035STom Caputi (DMU_BACKUP_FEATURE_RESUMING | DMU_BACKUP_FEATURE_RAW)) { 1098eb633035STom Caputi nvlist_t *keynvl = NULL; 10999c3fd121SMatthew Ahrens nvlist_t *nvl = fnvlist_alloc(); 1100eb633035STom Caputi 1101eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { 1102eb633035STom Caputi dmu_object_info_t to_doi; 1103eb633035STom Caputi err = dmu_object_info(os, resumeobj, &to_doi); 1104eb633035STom Caputi if (err != 0) { 1105eb633035STom Caputi fnvlist_free(nvl); 1106eb633035STom Caputi goto out; 1107eb633035STom Caputi } 1108eb633035STom Caputi 1109eb633035STom Caputi SET_BOOKMARK(&to_arg.resume, to_ds->ds_object, 1110eb633035STom Caputi resumeobj, 0, 1111eb633035STom Caputi resumeoff / to_doi.doi_data_block_size); 1112eb633035STom Caputi 1113eb633035STom Caputi fnvlist_add_uint64(nvl, "resume_object", resumeobj); 1114eb633035STom Caputi fnvlist_add_uint64(nvl, "resume_offset", resumeoff); 1115eb633035STom Caputi } 1116eb633035STom Caputi 1117eb633035STom Caputi if (featureflags & DMU_BACKUP_FEATURE_RAW) { 1118eb633035STom Caputi uint64_t ivset_guid = (ancestor_zb != NULL) ? 1119eb633035STom Caputi ancestor_zb->zbm_ivset_guid : 0; 1120eb633035STom Caputi 1121eb633035STom Caputi ASSERT(os->os_encrypted); 1122eb633035STom Caputi 1123eb633035STom Caputi err = dsl_crypto_populate_key_nvlist(to_ds, 1124eb633035STom Caputi ivset_guid, &keynvl); 1125eb633035STom Caputi if (err != 0) { 1126eb633035STom Caputi fnvlist_free(nvl); 1127eb633035STom Caputi goto out; 1128eb633035STom Caputi } 1129eb633035STom Caputi 1130eb633035STom Caputi fnvlist_add_nvlist(nvl, "crypt_keydata", keynvl); 1131eb633035STom Caputi } 1132eb633035STom Caputi 11339c3fd121SMatthew Ahrens payload = fnvlist_pack(nvl, &payload_len); 11349c3fd121SMatthew Ahrens drr->drr_payloadlen = payload_len; 1135eb633035STom Caputi fnvlist_free(keynvl); 11369c3fd121SMatthew Ahrens fnvlist_free(nvl); 11379c3fd121SMatthew Ahrens } 11389c3fd121SMatthew Ahrens 11399c3fd121SMatthew Ahrens err = dump_record(dsp, payload, payload_len); 11409c3fd121SMatthew Ahrens fnvlist_pack_free(payload, payload_len); 11419c3fd121SMatthew Ahrens if (err != 0) { 11424e3c9f44SBill Pijewski err = dsp->dsa_err; 11434e3c9f44SBill Pijewski goto out; 1144efb80947Sahrens } 1145efb80947Sahrens 1146*6ccda740Sloli err = bqueue_init(&to_arg.q, 1147*6ccda740Sloli MAX(zfs_send_queue_length, 2 * zfs_max_recordsize), 1148a2cdcdd2SPaul Dagnelie offsetof(struct send_block_record, ln)); 1149a2cdcdd2SPaul Dagnelie to_arg.error_code = 0; 1150a2cdcdd2SPaul Dagnelie to_arg.cancel = B_FALSE; 1151a2cdcdd2SPaul Dagnelie to_arg.ds = to_ds; 1152a2cdcdd2SPaul Dagnelie to_arg.fromtxg = fromtxg; 1153a2cdcdd2SPaul Dagnelie to_arg.flags = TRAVERSE_PRE | TRAVERSE_PREFETCH; 1154eb633035STom Caputi if (rawok) 1155eb633035STom Caputi to_arg.flags |= TRAVERSE_NO_DECRYPT; 1156a2cdcdd2SPaul Dagnelie (void) thread_create(NULL, 0, send_traverse_thread, &to_arg, 0, curproc, 1157a2cdcdd2SPaul Dagnelie TS_RUN, minclsyspri); 1158a2cdcdd2SPaul Dagnelie 1159a2cdcdd2SPaul Dagnelie struct send_block_record *to_data; 1160a2cdcdd2SPaul Dagnelie to_data = bqueue_dequeue(&to_arg.q); 1161a2cdcdd2SPaul Dagnelie 1162a2cdcdd2SPaul Dagnelie while (!to_data->eos_marker && err == 0) { 1163a2cdcdd2SPaul Dagnelie err = do_dump(dsp, to_data); 1164a2cdcdd2SPaul Dagnelie to_data = get_next_record(&to_arg.q, to_data); 1165a2cdcdd2SPaul Dagnelie if (issig(JUSTLOOKING) && issig(FORREAL)) 1166a2cdcdd2SPaul Dagnelie err = EINTR; 1167a2cdcdd2SPaul Dagnelie } 1168a2cdcdd2SPaul Dagnelie 1169a2cdcdd2SPaul Dagnelie if (err != 0) { 1170a2cdcdd2SPaul Dagnelie to_arg.cancel = B_TRUE; 1171a2cdcdd2SPaul Dagnelie while (!to_data->eos_marker) { 1172a2cdcdd2SPaul Dagnelie to_data = get_next_record(&to_arg.q, to_data); 1173a2cdcdd2SPaul Dagnelie } 1174a2cdcdd2SPaul Dagnelie } 1175a2cdcdd2SPaul Dagnelie kmem_free(to_data, sizeof (*to_data)); 1176a2cdcdd2SPaul Dagnelie 1177a2cdcdd2SPaul Dagnelie bqueue_destroy(&to_arg.q); 1178a2cdcdd2SPaul Dagnelie 1179a2cdcdd2SPaul Dagnelie if (err == 0 && to_arg.error_code != 0) 1180a2cdcdd2SPaul Dagnelie err = to_arg.error_code; 1181a2cdcdd2SPaul Dagnelie 1182a2cdcdd2SPaul Dagnelie if (err != 0) 1183a2cdcdd2SPaul Dagnelie goto out; 1184efb80947Sahrens 11854e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) 118698110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 1187be6fd75aSMatthew Ahrens err = SET_ERROR(EINTR); 11889e69d7d0SLori Alt 11893b2aab18SMatthew Ahrens if (err != 0) { 11903b2aab18SMatthew Ahrens if (err == EINTR && dsp->dsa_err != 0) 11914e3c9f44SBill Pijewski err = dsp->dsa_err; 11924e3c9f44SBill Pijewski goto out; 1193efb80947Sahrens } 1194efb80947Sahrens 1195efb80947Sahrens bzero(drr, sizeof (dmu_replay_record_t)); 1196efb80947Sahrens drr->drr_type = DRR_END; 11974e3c9f44SBill Pijewski drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; 11984e3c9f44SBill Pijewski drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; 1199efb80947Sahrens 1200a2cdcdd2SPaul Dagnelie if (dump_record(dsp, NULL, 0) != 0) 12014e3c9f44SBill Pijewski err = dsp->dsa_err; 12024e3c9f44SBill Pijewski out: 1203a2cdcdd2SPaul Dagnelie mutex_enter(&to_ds->ds_sendstream_lock); 1204a2cdcdd2SPaul Dagnelie list_remove(&to_ds->ds_sendstreams, dsp); 1205a2cdcdd2SPaul Dagnelie mutex_exit(&to_ds->ds_sendstream_lock); 12064e3c9f44SBill Pijewski 120712b90ee2SMatt Krantz VERIFY(err != 0 || (dsp->dsa_sent_begin && dsp->dsa_sent_end)); 120812b90ee2SMatt Krantz 1209efb80947Sahrens kmem_free(drr, sizeof (dmu_replay_record_t)); 12104e3c9f44SBill Pijewski kmem_free(dsp, sizeof (dmu_sendarg_t)); 1211efb80947Sahrens 1212a2cdcdd2SPaul Dagnelie dsl_dataset_long_rele(to_ds, FTAG); 12133b2aab18SMatthew Ahrens 12144e3c9f44SBill Pijewski return (err); 1215efb80947Sahrens } 1216efb80947Sahrens 121719b94df9SMatthew Ahrens int 12183b2aab18SMatthew Ahrens dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, 12195602294fSDan Kimmel boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, 1220eb633035STom Caputi boolean_t rawok, int outfd, vnode_t *vp, offset_t *off) 12213b2aab18SMatthew Ahrens { 12223b2aab18SMatthew Ahrens dsl_pool_t *dp; 12233b2aab18SMatthew Ahrens dsl_dataset_t *ds; 12243b2aab18SMatthew Ahrens dsl_dataset_t *fromds = NULL; 1225eb633035STom Caputi ds_hold_flags_t dsflags = (rawok) ? 0 : DS_HOLD_FLAG_DECRYPT; 12263b2aab18SMatthew Ahrens int err; 12273b2aab18SMatthew Ahrens 12283b2aab18SMatthew Ahrens err = dsl_pool_hold(pool, FTAG, &dp); 12293b2aab18SMatthew Ahrens if (err != 0) 12303b2aab18SMatthew Ahrens return (err); 12313b2aab18SMatthew Ahrens 1232eb633035STom Caputi err = dsl_dataset_hold_obj_flags(dp, tosnap, dsflags, FTAG, &ds); 12333b2aab18SMatthew Ahrens if (err != 0) { 12343b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 12353b2aab18SMatthew Ahrens return (err); 12363b2aab18SMatthew Ahrens } 12373b2aab18SMatthew Ahrens 12383b2aab18SMatthew Ahrens if (fromsnap != 0) { 1239eb633035STom Caputi zfs_bookmark_phys_t zb = { 0 }; 124078f17100SMatthew Ahrens boolean_t is_clone; 124178f17100SMatthew Ahrens 12423b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); 12433b2aab18SMatthew Ahrens if (err != 0) { 1244eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 12453b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 12463b2aab18SMatthew Ahrens return (err); 12473b2aab18SMatthew Ahrens } 1248eb633035STom Caputi if (!dsl_dataset_is_before(ds, fromds, 0)) { 124978f17100SMatthew Ahrens err = SET_ERROR(EXDEV); 1250eb633035STom Caputi dsl_dataset_rele(fromds, FTAG); 1251eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 1252eb633035STom Caputi dsl_pool_rele(dp, FTAG); 1253eb633035STom Caputi return (err); 1254eb633035STom Caputi } 1255eb633035STom Caputi 1256c1379625SJustin T. Gibbs zb.zbm_creation_time = 1257c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_time; 1258c1379625SJustin T. Gibbs zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; 1259c1379625SJustin T. Gibbs zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 1260eb633035STom Caputi 1261eb633035STom Caputi if (dsl_dataset_is_zapified(fromds)) { 1262eb633035STom Caputi (void) zap_lookup(dp->dp_meta_objset, 1263eb633035STom Caputi fromds->ds_object, DS_FIELD_IVSET_GUID, 8, 1, 1264eb633035STom Caputi &zb.zbm_ivset_guid); 1265eb633035STom Caputi } 1266eb633035STom Caputi 126778f17100SMatthew Ahrens is_clone = (fromds->ds_dir != ds->ds_dir); 126878f17100SMatthew Ahrens dsl_dataset_rele(fromds, FTAG); 1269b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 1270eb633035STom Caputi embedok, large_block_ok, compressok, rawok, outfd, 1271eb633035STom Caputi 0, 0, vp, off); 127278f17100SMatthew Ahrens } else { 1273b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 1274eb633035STom Caputi embedok, large_block_ok, compressok, rawok, outfd, 1275eb633035STom Caputi 0, 0, vp, off); 12763b2aab18SMatthew Ahrens } 1277eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 127878f17100SMatthew Ahrens return (err); 12793b2aab18SMatthew Ahrens } 12803b2aab18SMatthew Ahrens 12813b2aab18SMatthew Ahrens int 12829c3fd121SMatthew Ahrens dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, 1283eb633035STom Caputi boolean_t large_block_ok, boolean_t compressok, boolean_t rawok, 1284eb633035STom Caputi int outfd, uint64_t resumeobj, uint64_t resumeoff, vnode_t *vp, 1285eb633035STom Caputi offset_t *off) 12863b2aab18SMatthew Ahrens { 12873b2aab18SMatthew Ahrens dsl_pool_t *dp; 12883b2aab18SMatthew Ahrens dsl_dataset_t *ds; 12893b2aab18SMatthew Ahrens int err; 1290eb633035STom Caputi ds_hold_flags_t dsflags = (rawok) ? 0 : DS_HOLD_FLAG_DECRYPT; 129178f17100SMatthew Ahrens boolean_t owned = B_FALSE; 12923b2aab18SMatthew Ahrens 129378f17100SMatthew Ahrens if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) 1294be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 12953b2aab18SMatthew Ahrens 12963b2aab18SMatthew Ahrens err = dsl_pool_hold(tosnap, FTAG, &dp); 12973b2aab18SMatthew Ahrens if (err != 0) 12983b2aab18SMatthew Ahrens return (err); 129978f17100SMatthew Ahrens if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { 130078f17100SMatthew Ahrens /* 130178f17100SMatthew Ahrens * We are sending a filesystem or volume. Ensure 130278f17100SMatthew Ahrens * that it doesn't change by owning the dataset. 130378f17100SMatthew Ahrens */ 1304eb633035STom Caputi err = dsl_dataset_own(dp, tosnap, dsflags, FTAG, &ds); 130578f17100SMatthew Ahrens owned = B_TRUE; 130678f17100SMatthew Ahrens } else { 1307eb633035STom Caputi err = dsl_dataset_hold_flags(dp, tosnap, dsflags, FTAG, &ds); 130878f17100SMatthew Ahrens } 13093b2aab18SMatthew Ahrens if (err != 0) { 13103b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 13113b2aab18SMatthew Ahrens return (err); 13123b2aab18SMatthew Ahrens } 13133b2aab18SMatthew Ahrens 13143b2aab18SMatthew Ahrens if (fromsnap != NULL) { 1315eb633035STom Caputi zfs_bookmark_phys_t zb = { 0 }; 131678f17100SMatthew Ahrens boolean_t is_clone = B_FALSE; 131778f17100SMatthew Ahrens int fsnamelen = strchr(tosnap, '@') - tosnap; 131878f17100SMatthew Ahrens 131978f17100SMatthew Ahrens /* 132078f17100SMatthew Ahrens * If the fromsnap is in a different filesystem, then 132178f17100SMatthew Ahrens * mark the send stream as a clone. 132278f17100SMatthew Ahrens */ 132378f17100SMatthew Ahrens if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || 132478f17100SMatthew Ahrens (fromsnap[fsnamelen] != '@' && 132578f17100SMatthew Ahrens fromsnap[fsnamelen] != '#')) { 132678f17100SMatthew Ahrens is_clone = B_TRUE; 132778f17100SMatthew Ahrens } 132878f17100SMatthew Ahrens 132978f17100SMatthew Ahrens if (strchr(fromsnap, '@')) { 133078f17100SMatthew Ahrens dsl_dataset_t *fromds; 133178f17100SMatthew Ahrens err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); 133278f17100SMatthew Ahrens if (err == 0) { 133378f17100SMatthew Ahrens if (!dsl_dataset_is_before(ds, fromds, 0)) 133478f17100SMatthew Ahrens err = SET_ERROR(EXDEV); 133578f17100SMatthew Ahrens zb.zbm_creation_time = 1336c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_time; 133778f17100SMatthew Ahrens zb.zbm_creation_txg = 1338c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_txg; 1339c1379625SJustin T. Gibbs zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 134078f17100SMatthew Ahrens is_clone = (ds->ds_dir != fromds->ds_dir); 1341eb633035STom Caputi 1342eb633035STom Caputi if (dsl_dataset_is_zapified(fromds)) { 1343eb633035STom Caputi (void) zap_lookup(dp->dp_meta_objset, 1344eb633035STom Caputi fromds->ds_object, 1345eb633035STom Caputi DS_FIELD_IVSET_GUID, 8, 1, 1346eb633035STom Caputi &zb.zbm_ivset_guid); 1347eb633035STom Caputi } 134878f17100SMatthew Ahrens dsl_dataset_rele(fromds, FTAG); 134978f17100SMatthew Ahrens } 135078f17100SMatthew Ahrens } else { 135178f17100SMatthew Ahrens err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); 135278f17100SMatthew Ahrens } 13533b2aab18SMatthew Ahrens if (err != 0) { 1354eb633035STom Caputi if (owned) 1355eb633035STom Caputi dsl_dataset_disown(ds, dsflags, FTAG); 1356eb633035STom Caputi else 1357eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 1358eb633035STom Caputi 13593b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 13603b2aab18SMatthew Ahrens return (err); 13613b2aab18SMatthew Ahrens } 1362b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 1363eb633035STom Caputi embedok, large_block_ok, compressok, rawok, 13649c3fd121SMatthew Ahrens outfd, resumeobj, resumeoff, vp, off); 136578f17100SMatthew Ahrens } else { 1366b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 1367eb633035STom Caputi embedok, large_block_ok, compressok, rawok, 13689c3fd121SMatthew Ahrens outfd, resumeobj, resumeoff, vp, off); 13693b2aab18SMatthew Ahrens } 137078f17100SMatthew Ahrens if (owned) 1371eb633035STom Caputi dsl_dataset_disown(ds, dsflags, FTAG); 137278f17100SMatthew Ahrens else 1373eb633035STom Caputi dsl_dataset_rele_flags(ds, dsflags, FTAG); 1374eb633035STom Caputi 137578f17100SMatthew Ahrens return (err); 13763b2aab18SMatthew Ahrens } 13773b2aab18SMatthew Ahrens 1378643da460SMax Grossman static int 13795602294fSDan Kimmel dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t uncompressed, 13805602294fSDan Kimmel uint64_t compressed, boolean_t stream_compressed, uint64_t *sizep) 1381643da460SMax Grossman { 1382df477c0aSPaul Dagnelie int err = 0; 13835602294fSDan Kimmel uint64_t size; 1384643da460SMax Grossman /* 1385643da460SMax Grossman * Assume that space (both on-disk and in-stream) is dominated by 1386643da460SMax Grossman * data. We will adjust for indirect blocks and the copies property, 1387643da460SMax Grossman * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). 1388643da460SMax Grossman */ 13895602294fSDan Kimmel uint64_t recordsize; 13905602294fSDan Kimmel uint64_t record_count; 13910255edccSPaul Dagnelie objset_t *os; 13920255edccSPaul Dagnelie VERIFY0(dmu_objset_from_ds(ds, &os)); 13935602294fSDan Kimmel 13945602294fSDan Kimmel /* Assume all (uncompressed) blocks are recordsize. */ 1395df477c0aSPaul Dagnelie if (zfs_override_estimate_recordsize != 0) { 1396df477c0aSPaul Dagnelie recordsize = zfs_override_estimate_recordsize; 1397df477c0aSPaul Dagnelie } else if (os->os_phys->os_type == DMU_OST_ZVOL) { 13980255edccSPaul Dagnelie err = dsl_prop_get_int_ds(ds, 13990255edccSPaul Dagnelie zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &recordsize); 14000255edccSPaul Dagnelie } else { 14010255edccSPaul Dagnelie err = dsl_prop_get_int_ds(ds, 14020255edccSPaul Dagnelie zfs_prop_to_name(ZFS_PROP_RECORDSIZE), &recordsize); 14030255edccSPaul Dagnelie } 14045602294fSDan Kimmel if (err != 0) 14055602294fSDan Kimmel return (err); 14065602294fSDan Kimmel record_count = uncompressed / recordsize; 14075602294fSDan Kimmel 14085602294fSDan Kimmel /* 14095602294fSDan Kimmel * If we're estimating a send size for a compressed stream, use the 14105602294fSDan Kimmel * compressed data size to estimate the stream size. Otherwise, use the 14115602294fSDan Kimmel * uncompressed data size. 14125602294fSDan Kimmel */ 14135602294fSDan Kimmel size = stream_compressed ? compressed : uncompressed; 1414643da460SMax Grossman 1415643da460SMax Grossman /* 1416643da460SMax Grossman * Subtract out approximate space used by indirect blocks. 1417643da460SMax Grossman * Assume most space is used by data blocks (non-indirect, non-dnode). 14185602294fSDan Kimmel * Assume no ditto blocks or internal fragmentation. 1419643da460SMax Grossman * 1420643da460SMax Grossman * Therefore, space used by indirect blocks is sizeof(blkptr_t) per 14215602294fSDan Kimmel * block. 1422643da460SMax Grossman */ 14235602294fSDan Kimmel size -= record_count * sizeof (blkptr_t); 1424643da460SMax Grossman 1425643da460SMax Grossman /* Add in the space for the record associated with each block. */ 14265602294fSDan Kimmel size += record_count * sizeof (dmu_replay_record_t); 1427643da460SMax Grossman 1428643da460SMax Grossman *sizep = size; 1429643da460SMax Grossman 1430643da460SMax Grossman return (0); 1431643da460SMax Grossman } 1432643da460SMax Grossman 14333b2aab18SMatthew Ahrens int 14345602294fSDan Kimmel dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, 14355602294fSDan Kimmel boolean_t stream_compressed, uint64_t *sizep) 143619b94df9SMatthew Ahrens { 143719b94df9SMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 143819b94df9SMatthew Ahrens int err; 14395602294fSDan Kimmel uint64_t uncomp, comp; 144019b94df9SMatthew Ahrens 14413b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dp)); 14423b2aab18SMatthew Ahrens 144319b94df9SMatthew Ahrens /* tosnap must be a snapshot */ 1444bc9014e6SJustin Gibbs if (!ds->ds_is_snapshot) 1445be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 144619b94df9SMatthew Ahrens 144724218bebSAndriy Gapon /* fromsnap, if provided, must be a snapshot */ 144824218bebSAndriy Gapon if (fromds != NULL && !fromds->ds_is_snapshot) 144924218bebSAndriy Gapon return (SET_ERROR(EINVAL)); 145024218bebSAndriy Gapon 14514445fffbSMatthew Ahrens /* 14524445fffbSMatthew Ahrens * fromsnap must be an earlier snapshot from the same fs as tosnap, 14534445fffbSMatthew Ahrens * or the origin's fs. 14544445fffbSMatthew Ahrens */ 145578f17100SMatthew Ahrens if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) 1456be6fd75aSMatthew Ahrens return (SET_ERROR(EXDEV)); 145719b94df9SMatthew Ahrens 14585602294fSDan Kimmel /* Get compressed and uncompressed size estimates of changed data. */ 145919b94df9SMatthew Ahrens if (fromds == NULL) { 14605602294fSDan Kimmel uncomp = dsl_dataset_phys(ds)->ds_uncompressed_bytes; 14615602294fSDan Kimmel comp = dsl_dataset_phys(ds)->ds_compressed_bytes; 146219b94df9SMatthew Ahrens } else { 14635602294fSDan Kimmel uint64_t used; 146419b94df9SMatthew Ahrens err = dsl_dataset_space_written(fromds, ds, 14655602294fSDan Kimmel &used, &comp, &uncomp); 14663b2aab18SMatthew Ahrens if (err != 0) 146719b94df9SMatthew Ahrens return (err); 146819b94df9SMatthew Ahrens } 146919b94df9SMatthew Ahrens 14705602294fSDan Kimmel err = dmu_adjust_send_estimate_for_indirects(ds, uncomp, comp, 14715602294fSDan Kimmel stream_compressed, sizep); 14720255edccSPaul Dagnelie /* 14730255edccSPaul Dagnelie * Add the size of the BEGIN and END records to the estimate. 14740255edccSPaul Dagnelie */ 14750255edccSPaul Dagnelie *sizep += 2 * sizeof (dmu_replay_record_t); 1476643da460SMax Grossman return (err); 1477643da460SMax Grossman } 1478643da460SMax Grossman 14795602294fSDan Kimmel struct calculate_send_arg { 14805602294fSDan Kimmel uint64_t uncompressed; 14815602294fSDan Kimmel uint64_t compressed; 14825602294fSDan Kimmel }; 14835602294fSDan Kimmel 1484643da460SMax Grossman /* 1485643da460SMax Grossman * Simple callback used to traverse the blocks of a snapshot and sum their 14865602294fSDan Kimmel * uncompressed and compressed sizes. 1487643da460SMax Grossman */ 1488643da460SMax Grossman /* ARGSUSED */ 1489643da460SMax Grossman static int 1490643da460SMax Grossman dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1491643da460SMax Grossman const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1492643da460SMax Grossman { 14935602294fSDan Kimmel struct calculate_send_arg *space = arg; 1494643da460SMax Grossman if (bp != NULL && !BP_IS_HOLE(bp)) { 14955602294fSDan Kimmel space->uncompressed += BP_GET_UCSIZE(bp); 14965602294fSDan Kimmel space->compressed += BP_GET_PSIZE(bp); 1497643da460SMax Grossman } 1498643da460SMax Grossman return (0); 1499643da460SMax Grossman } 1500643da460SMax Grossman 1501643da460SMax Grossman /* 1502643da460SMax Grossman * Given a desination snapshot and a TXG, calculate the approximate size of a 1503643da460SMax Grossman * send stream sent from that TXG. from_txg may be zero, indicating that the 1504643da460SMax Grossman * whole snapshot will be sent. 1505643da460SMax Grossman */ 1506643da460SMax Grossman int 1507643da460SMax Grossman dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg, 15085602294fSDan Kimmel boolean_t stream_compressed, uint64_t *sizep) 1509643da460SMax Grossman { 1510643da460SMax Grossman dsl_pool_t *dp = ds->ds_dir->dd_pool; 1511643da460SMax Grossman int err; 15125602294fSDan Kimmel struct calculate_send_arg size = { 0 }; 1513643da460SMax Grossman 1514643da460SMax Grossman ASSERT(dsl_pool_config_held(dp)); 1515643da460SMax Grossman 1516643da460SMax Grossman /* tosnap must be a snapshot */ 15175602294fSDan Kimmel if (!ds->ds_is_snapshot) 1518643da460SMax Grossman return (SET_ERROR(EINVAL)); 1519643da460SMax Grossman 1520643da460SMax Grossman /* verify that from_txg is before the provided snapshot was taken */ 1521643da460SMax Grossman if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) { 1522643da460SMax Grossman return (SET_ERROR(EXDEV)); 1523643da460SMax Grossman } 152419b94df9SMatthew Ahrens 152519b94df9SMatthew Ahrens /* 1526643da460SMax Grossman * traverse the blocks of the snapshot with birth times after 1527643da460SMax Grossman * from_txg, summing their uncompressed size 152819b94df9SMatthew Ahrens */ 1529eb633035STom Caputi err = traverse_dataset(ds, from_txg, 1530eb633035STom Caputi TRAVERSE_POST | TRAVERSE_NO_DECRYPT, 1531643da460SMax Grossman dmu_calculate_send_traversal, &size); 1532643da460SMax Grossman if (err) 153319b94df9SMatthew Ahrens return (err); 153419b94df9SMatthew Ahrens 15355602294fSDan Kimmel err = dmu_adjust_send_estimate_for_indirects(ds, size.uncompressed, 15365602294fSDan Kimmel size.compressed, stream_compressed, sizep); 1537643da460SMax Grossman return (err); 153819b94df9SMatthew Ahrens } 1539